| Message ID | 20191122175400.4001-1-andrew.cooper3@citrix.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | [for-4.13] x86/vvmx: Fix livelock with XSA-304 fix | expand |
On Fri, Nov 22, 2019 at 5:55 PM Andrew Cooper <andrew.cooper3@citrix.com> wrote: > > It turns out that the XSA-304 / CVE-2018-12207 fix of disabling executable > superpages doesn't work well with the nested p2m code. > > Nested virt is experimental and not security supported, but is useful for > development purposes. In order to not regress the status quo, disable the > XSA-304 workaround until the nested p2m code can be improved. > > Introduce a per-domain exec_sp control and set it based on the current > opt_ept_exec_sp setting. Take the oppotunity to omit a PVH hardware domain > from the performance hit, because it is already permitted to DoS the system in > such ways as issuing a reboot. > > When nested virt is enabled on a domain, force it to using executable > superpages and rebuild the p2m. > > Having the setting per-domain involves rearranging the internals of > parse_ept_param_runtime() but it still retains the same overall semantics - > for each applicable domain whose setting needs to change, rebuild the p2m. > > Signed-off-by: Andrew Cooper <andrew.cooper3@citrix.com> > --- > CC: Jan Beulich <JBeulich@suse.com> > CC: Wei Liu <wl@xen.org> > CC: Roger Pau Monné <roger.pau@citrix.com> > CC: Juergen Gross <jgross@suse.com> > --- > xen/arch/x86/hvm/vmx/vmcs.c | 31 +++++++++++++++++++++++-------- > xen/arch/x86/hvm/vmx/vmx.c | 6 ++++++ > xen/arch/x86/hvm/vmx/vvmx.c | 13 +++++++++++++ > xen/arch/x86/mm/p2m-ept.c | 2 +- > xen/include/asm-x86/hvm/vmx/vmcs.h | 6 ++++++ > 5 files changed, 49 insertions(+), 9 deletions(-) > > diff --git a/xen/arch/x86/hvm/vmx/vmcs.c b/xen/arch/x86/hvm/vmx/vmcs.c > index 477c968409..f10f6b78ec 100644 > --- a/xen/arch/x86/hvm/vmx/vmcs.c > +++ b/xen/arch/x86/hvm/vmx/vmcs.c > @@ -31,6 +31,7 @@ > #include <asm/xstate.h> > #include <asm/hvm/hvm.h> > #include <asm/hvm/io.h> > +#include <asm/hvm/nestedhvm.h> > #include <asm/hvm/support.h> > #include <asm/hvm/vmx/vmx.h> > #include <asm/hvm/vmx/vvmx.h> > @@ -97,6 +98,7 @@ custom_param("ept", parse_ept_param); > > static int parse_ept_param_runtime(const char *s) > { > + struct domain *d; > int val; > > if ( !cpu_has_vmx_ept || !hvm_funcs.hap_supported || > @@ -110,18 +112,31 @@ static int parse_ept_param_runtime(const char *s) > if ( (val = parse_boolean("exec-sp", s, NULL)) < 0 ) > return -EINVAL; > > - if ( val != opt_ept_exec_sp ) > + opt_ept_exec_sp = val; > + > + rcu_read_lock(&domlist_read_lock); > + for_each_domain ( d ) > { > - struct domain *d; > + /* PV, or HVM Shadow domain? Not applicable. */ > + if ( !paging_mode_hap(d) ) > + continue; > > - opt_ept_exec_sp = val; > + /* Hardware domain? Not applicable. */ > + if ( is_hardware_domain(d) ) > + continue; > > - rcu_read_lock(&domlist_read_lock); > - for_each_domain ( d ) > - if ( paging_mode_hap(d) ) > - p2m_change_entry_type_global(d, p2m_ram_rw, p2m_ram_rw); > - rcu_read_unlock(&domlist_read_lock); > + /* Nested Virt? Broken and exec_sp forced on to avoid livelocks. */ > + if ( nestedhvm_enabled(d) ) > + continue; > + > + /* Setting already matches? No need to rebuild the p2m. */ > + if ( d->arch.hvm.vmx.exec_sp == val ) > + continue; > + > + d->arch.hvm.vmx.exec_sp = val; > + p2m_change_entry_type_global(d, p2m_ram_rw, p2m_ram_rw); > } > + rcu_read_unlock(&domlist_read_lock); > > printk("VMX: EPT executable superpages %sabled\n", > val ? "en" : "dis"); > diff --git a/xen/arch/x86/hvm/vmx/vmx.c b/xen/arch/x86/hvm/vmx/vmx.c > index 6a5eeb5c13..a71df71bc1 100644 > --- a/xen/arch/x86/hvm/vmx/vmx.c > +++ b/xen/arch/x86/hvm/vmx/vmx.c > @@ -404,6 +404,12 @@ static int vmx_domain_initialise(struct domain *d) > > d->arch.ctxt_switch = &csw; > > + /* > + * Work around CVE-2018-12207? The hardware domain is already permitted > + * to reboot the system, so doesn't need mitigating against DoS's. > + */ > + d->arch.hvm.vmx.exec_sp = is_hardware_domain(d) || opt_ept_exec_sp; > + > if ( !has_vlapic(d) ) > return 0; > > diff --git a/xen/arch/x86/hvm/vmx/vvmx.c b/xen/arch/x86/hvm/vmx/vvmx.c > index 6696bd6240..5dd00e11b5 100644 > --- a/xen/arch/x86/hvm/vmx/vvmx.c > +++ b/xen/arch/x86/hvm/vmx/vvmx.c > @@ -63,10 +63,23 @@ void nvmx_cpu_dead(unsigned int cpu) > > int nvmx_vcpu_initialise(struct vcpu *v) > { > + struct domain *d = v->domain; > struct nestedvmx *nvmx = &vcpu_2_nvmx(v); > struct nestedvcpu *nvcpu = &vcpu_nestedhvm(v); > struct page_info *pg = alloc_domheap_page(NULL, 0); > > + /* > + * Gross bodge. The nested p2m logic can't cope with the CVE-2018-12207 > + * workaround of using NX EPT superpages, and livelocks. Nested HVM isn't > + * security supported, so disable the workaround until the nested p2m > + * logic can be improved. > + */ > + if ( !d->arch.hvm.vmx.exec_sp ) > + { > + d->arch.hvm.vmx.exec_sp = true; > + p2m_change_entry_type_global(d, p2m_ram_rw, p2m_ram_rw); There wasn't an issue with nested guests having to deal with the changed entry type? Assuming the answer to that is "no": Acked-by: George Dunlap <george.dunlap@citrix.com>
On 22/11/2019 18:08, George Dunlap wrote: > On Fri, Nov 22, 2019 at 5:55 PM Andrew Cooper <andrew.cooper3@citrix.com> wrote: >> It turns out that the XSA-304 / CVE-2018-12207 fix of disabling executable >> superpages doesn't work well with the nested p2m code. >> >> Nested virt is experimental and not security supported, but is useful for >> development purposes. In order to not regress the status quo, disable the >> XSA-304 workaround until the nested p2m code can be improved. >> >> Introduce a per-domain exec_sp control and set it based on the current >> opt_ept_exec_sp setting. Take the oppotunity to omit a PVH hardware domain >> from the performance hit, because it is already permitted to DoS the system in >> such ways as issuing a reboot. >> >> When nested virt is enabled on a domain, force it to using executable >> superpages and rebuild the p2m. >> >> Having the setting per-domain involves rearranging the internals of >> parse_ept_param_runtime() but it still retains the same overall semantics - >> for each applicable domain whose setting needs to change, rebuild the p2m. >> >> Signed-off-by: Andrew Cooper <andrew.cooper3@citrix.com> >> --- >> CC: Jan Beulich <JBeulich@suse.com> >> CC: Wei Liu <wl@xen.org> >> CC: Roger Pau Monné <roger.pau@citrix.com> >> CC: Juergen Gross <jgross@suse.com> >> --- >> xen/arch/x86/hvm/vmx/vmcs.c | 31 +++++++++++++++++++++++-------- >> xen/arch/x86/hvm/vmx/vmx.c | 6 ++++++ >> xen/arch/x86/hvm/vmx/vvmx.c | 13 +++++++++++++ >> xen/arch/x86/mm/p2m-ept.c | 2 +- >> xen/include/asm-x86/hvm/vmx/vmcs.h | 6 ++++++ >> 5 files changed, 49 insertions(+), 9 deletions(-) >> >> diff --git a/xen/arch/x86/hvm/vmx/vmcs.c b/xen/arch/x86/hvm/vmx/vmcs.c >> index 477c968409..f10f6b78ec 100644 >> --- a/xen/arch/x86/hvm/vmx/vmcs.c >> +++ b/xen/arch/x86/hvm/vmx/vmcs.c >> @@ -31,6 +31,7 @@ >> #include <asm/xstate.h> >> #include <asm/hvm/hvm.h> >> #include <asm/hvm/io.h> >> +#include <asm/hvm/nestedhvm.h> >> #include <asm/hvm/support.h> >> #include <asm/hvm/vmx/vmx.h> >> #include <asm/hvm/vmx/vvmx.h> >> @@ -97,6 +98,7 @@ custom_param("ept", parse_ept_param); >> >> static int parse_ept_param_runtime(const char *s) >> { >> + struct domain *d; >> int val; >> >> if ( !cpu_has_vmx_ept || !hvm_funcs.hap_supported || >> @@ -110,18 +112,31 @@ static int parse_ept_param_runtime(const char *s) >> if ( (val = parse_boolean("exec-sp", s, NULL)) < 0 ) >> return -EINVAL; >> >> - if ( val != opt_ept_exec_sp ) >> + opt_ept_exec_sp = val; >> + >> + rcu_read_lock(&domlist_read_lock); >> + for_each_domain ( d ) >> { >> - struct domain *d; >> + /* PV, or HVM Shadow domain? Not applicable. */ >> + if ( !paging_mode_hap(d) ) >> + continue; >> >> - opt_ept_exec_sp = val; >> + /* Hardware domain? Not applicable. */ >> + if ( is_hardware_domain(d) ) >> + continue; >> >> - rcu_read_lock(&domlist_read_lock); >> - for_each_domain ( d ) >> - if ( paging_mode_hap(d) ) >> - p2m_change_entry_type_global(d, p2m_ram_rw, p2m_ram_rw); >> - rcu_read_unlock(&domlist_read_lock); >> + /* Nested Virt? Broken and exec_sp forced on to avoid livelocks. */ >> + if ( nestedhvm_enabled(d) ) >> + continue; >> + >> + /* Setting already matches? No need to rebuild the p2m. */ >> + if ( d->arch.hvm.vmx.exec_sp == val ) >> + continue; >> + >> + d->arch.hvm.vmx.exec_sp = val; >> + p2m_change_entry_type_global(d, p2m_ram_rw, p2m_ram_rw); >> } >> + rcu_read_unlock(&domlist_read_lock); >> >> printk("VMX: EPT executable superpages %sabled\n", >> val ? "en" : "dis"); >> diff --git a/xen/arch/x86/hvm/vmx/vmx.c b/xen/arch/x86/hvm/vmx/vmx.c >> index 6a5eeb5c13..a71df71bc1 100644 >> --- a/xen/arch/x86/hvm/vmx/vmx.c >> +++ b/xen/arch/x86/hvm/vmx/vmx.c >> @@ -404,6 +404,12 @@ static int vmx_domain_initialise(struct domain *d) >> >> d->arch.ctxt_switch = &csw; >> >> + /* >> + * Work around CVE-2018-12207? The hardware domain is already permitted >> + * to reboot the system, so doesn't need mitigating against DoS's. >> + */ >> + d->arch.hvm.vmx.exec_sp = is_hardware_domain(d) || opt_ept_exec_sp; >> + >> if ( !has_vlapic(d) ) >> return 0; >> >> diff --git a/xen/arch/x86/hvm/vmx/vvmx.c b/xen/arch/x86/hvm/vmx/vvmx.c >> index 6696bd6240..5dd00e11b5 100644 >> --- a/xen/arch/x86/hvm/vmx/vvmx.c >> +++ b/xen/arch/x86/hvm/vmx/vvmx.c >> @@ -63,10 +63,23 @@ void nvmx_cpu_dead(unsigned int cpu) >> >> int nvmx_vcpu_initialise(struct vcpu *v) >> { >> + struct domain *d = v->domain; >> struct nestedvmx *nvmx = &vcpu_2_nvmx(v); >> struct nestedvcpu *nvcpu = &vcpu_nestedhvm(v); >> struct page_info *pg = alloc_domheap_page(NULL, 0); >> >> + /* >> + * Gross bodge. The nested p2m logic can't cope with the CVE-2018-12207 >> + * workaround of using NX EPT superpages, and livelocks. Nested HVM isn't >> + * security supported, so disable the workaround until the nested p2m >> + * logic can be improved. >> + */ >> + if ( !d->arch.hvm.vmx.exec_sp ) >> + { >> + d->arch.hvm.vmx.exec_sp = true; >> + p2m_change_entry_type_global(d, p2m_ram_rw, p2m_ram_rw); > There wasn't an issue with nested guests having to deal with the > changed entry type? > > Assuming the answer to that is "no": > > Acked-by: George Dunlap <george.dunlap@citrix.com> That is the issue we're working around by not letting the L01 walk encounter an NX superpage to begin with. ~Andrew
On 22.11.19 18:54, Andrew Cooper wrote: > It turns out that the XSA-304 / CVE-2018-12207 fix of disabling executable > superpages doesn't work well with the nested p2m code. > > Nested virt is experimental and not security supported, but is useful for > development purposes. In order to not regress the status quo, disable the > XSA-304 workaround until the nested p2m code can be improved. > > Introduce a per-domain exec_sp control and set it based on the current > opt_ept_exec_sp setting. Take the oppotunity to omit a PVH hardware domain > from the performance hit, because it is already permitted to DoS the system in > such ways as issuing a reboot. > > When nested virt is enabled on a domain, force it to using executable > superpages and rebuild the p2m. > > Having the setting per-domain involves rearranging the internals of > parse_ept_param_runtime() but it still retains the same overall semantics - > for each applicable domain whose setting needs to change, rebuild the p2m. > > Signed-off-by: Andrew Cooper <andrew.cooper3@citrix.com> Release-acked-by: Juergen Gross <jgross@suse.com> Juergen
On 23/11/2019 05:56, Jürgen Groß wrote: > On 22.11.19 18:54, Andrew Cooper wrote: >> It turns out that the XSA-304 / CVE-2018-12207 fix of disabling >> executable >> superpages doesn't work well with the nested p2m code. >> >> Nested virt is experimental and not security supported, but is useful >> for >> development purposes. In order to not regress the status quo, >> disable the >> XSA-304 workaround until the nested p2m code can be improved. >> >> Introduce a per-domain exec_sp control and set it based on the current >> opt_ept_exec_sp setting. Take the oppotunity to omit a PVH hardware >> domain >> from the performance hit, because it is already permitted to DoS the >> system in >> such ways as issuing a reboot. >> >> When nested virt is enabled on a domain, force it to using executable >> superpages and rebuild the p2m. >> >> Having the setting per-domain involves rearranging the internals of >> parse_ept_param_runtime() but it still retains the same overall >> semantics - >> for each applicable domain whose setting needs to change, rebuild the >> p2m. >> >> Signed-off-by: Andrew Cooper <andrew.cooper3@citrix.com> > > Release-acked-by: Juergen Gross <jgross@suse.com> Thanks. I've pushed this to staging to hopefully unblock OSSTest for RC3. Any further delta to this patch can be done incrementally. ~Andrew
diff --git a/xen/arch/x86/hvm/vmx/vmcs.c b/xen/arch/x86/hvm/vmx/vmcs.c index 477c968409..f10f6b78ec 100644 --- a/xen/arch/x86/hvm/vmx/vmcs.c +++ b/xen/arch/x86/hvm/vmx/vmcs.c @@ -31,6 +31,7 @@ #include <asm/xstate.h> #include <asm/hvm/hvm.h> #include <asm/hvm/io.h> +#include <asm/hvm/nestedhvm.h> #include <asm/hvm/support.h> #include <asm/hvm/vmx/vmx.h> #include <asm/hvm/vmx/vvmx.h> @@ -97,6 +98,7 @@ custom_param("ept", parse_ept_param); static int parse_ept_param_runtime(const char *s) { + struct domain *d; int val; if ( !cpu_has_vmx_ept || !hvm_funcs.hap_supported || @@ -110,18 +112,31 @@ static int parse_ept_param_runtime(const char *s) if ( (val = parse_boolean("exec-sp", s, NULL)) < 0 ) return -EINVAL; - if ( val != opt_ept_exec_sp ) + opt_ept_exec_sp = val; + + rcu_read_lock(&domlist_read_lock); + for_each_domain ( d ) { - struct domain *d; + /* PV, or HVM Shadow domain? Not applicable. */ + if ( !paging_mode_hap(d) ) + continue; - opt_ept_exec_sp = val; + /* Hardware domain? Not applicable. */ + if ( is_hardware_domain(d) ) + continue; - rcu_read_lock(&domlist_read_lock); - for_each_domain ( d ) - if ( paging_mode_hap(d) ) - p2m_change_entry_type_global(d, p2m_ram_rw, p2m_ram_rw); - rcu_read_unlock(&domlist_read_lock); + /* Nested Virt? Broken and exec_sp forced on to avoid livelocks. */ + if ( nestedhvm_enabled(d) ) + continue; + + /* Setting already matches? No need to rebuild the p2m. */ + if ( d->arch.hvm.vmx.exec_sp == val ) + continue; + + d->arch.hvm.vmx.exec_sp = val; + p2m_change_entry_type_global(d, p2m_ram_rw, p2m_ram_rw); } + rcu_read_unlock(&domlist_read_lock); printk("VMX: EPT executable superpages %sabled\n", val ? "en" : "dis"); diff --git a/xen/arch/x86/hvm/vmx/vmx.c b/xen/arch/x86/hvm/vmx/vmx.c index 6a5eeb5c13..a71df71bc1 100644 --- a/xen/arch/x86/hvm/vmx/vmx.c +++ b/xen/arch/x86/hvm/vmx/vmx.c @@ -404,6 +404,12 @@ static int vmx_domain_initialise(struct domain *d) d->arch.ctxt_switch = &csw; + /* + * Work around CVE-2018-12207? The hardware domain is already permitted + * to reboot the system, so doesn't need mitigating against DoS's. + */ + d->arch.hvm.vmx.exec_sp = is_hardware_domain(d) || opt_ept_exec_sp; + if ( !has_vlapic(d) ) return 0; diff --git a/xen/arch/x86/hvm/vmx/vvmx.c b/xen/arch/x86/hvm/vmx/vvmx.c index 6696bd6240..5dd00e11b5 100644 --- a/xen/arch/x86/hvm/vmx/vvmx.c +++ b/xen/arch/x86/hvm/vmx/vvmx.c @@ -63,10 +63,23 @@ void nvmx_cpu_dead(unsigned int cpu) int nvmx_vcpu_initialise(struct vcpu *v) { + struct domain *d = v->domain; struct nestedvmx *nvmx = &vcpu_2_nvmx(v); struct nestedvcpu *nvcpu = &vcpu_nestedhvm(v); struct page_info *pg = alloc_domheap_page(NULL, 0); + /* + * Gross bodge. The nested p2m logic can't cope with the CVE-2018-12207 + * workaround of using NX EPT superpages, and livelocks. Nested HVM isn't + * security supported, so disable the workaround until the nested p2m + * logic can be improved. + */ + if ( !d->arch.hvm.vmx.exec_sp ) + { + d->arch.hvm.vmx.exec_sp = true; + p2m_change_entry_type_global(d, p2m_ram_rw, p2m_ram_rw); + } + if ( !pg ) { gdprintk(XENLOG_ERR, "nest: allocation for shadow vmcs failed\n"); diff --git a/xen/arch/x86/mm/p2m-ept.c b/xen/arch/x86/mm/p2m-ept.c index f06e51904a..b5517769c9 100644 --- a/xen/arch/x86/mm/p2m-ept.c +++ b/xen/arch/x86/mm/p2m-ept.c @@ -178,7 +178,7 @@ static void ept_p2m_type_to_flags(struct p2m_domain *p2m, ept_entry_t *entry, * Don't create executable superpages if we need to shatter them to * protect against CVE-2018-12207. */ - if ( !opt_ept_exec_sp && is_epte_superpage(entry) ) + if ( !p2m->domain->arch.hvm.vmx.exec_sp && is_epte_superpage(entry) ) entry->x = 0; } diff --git a/xen/include/asm-x86/hvm/vmx/vmcs.h b/xen/include/asm-x86/hvm/vmx/vmcs.h index b3e800138e..2713da74db 100644 --- a/xen/include/asm-x86/hvm/vmx/vmcs.h +++ b/xen/include/asm-x86/hvm/vmx/vmcs.h @@ -63,6 +63,12 @@ struct vmx_domain { unsigned long apic_access_mfn; /* VMX_DOMAIN_* */ unsigned int status; + + /* + * Domain permitted to use Executable EPT Superpages? Cleared to work + * around CVE-2018-12207 as appropriate. + */ + bool exec_sp; }; /*
It turns out that the XSA-304 / CVE-2018-12207 fix of disabling executable superpages doesn't work well with the nested p2m code. Nested virt is experimental and not security supported, but is useful for development purposes. In order to not regress the status quo, disable the XSA-304 workaround until the nested p2m code can be improved. Introduce a per-domain exec_sp control and set it based on the current opt_ept_exec_sp setting. Take the oppotunity to omit a PVH hardware domain from the performance hit, because it is already permitted to DoS the system in such ways as issuing a reboot. When nested virt is enabled on a domain, force it to using executable superpages and rebuild the p2m. Having the setting per-domain involves rearranging the internals of parse_ept_param_runtime() but it still retains the same overall semantics - for each applicable domain whose setting needs to change, rebuild the p2m. Signed-off-by: Andrew Cooper <andrew.cooper3@citrix.com> --- CC: Jan Beulich <JBeulich@suse.com> CC: Wei Liu <wl@xen.org> CC: Roger Pau Monné <roger.pau@citrix.com> CC: Juergen Gross <jgross@suse.com> --- xen/arch/x86/hvm/vmx/vmcs.c | 31 +++++++++++++++++++++++-------- xen/arch/x86/hvm/vmx/vmx.c | 6 ++++++ xen/arch/x86/hvm/vmx/vvmx.c | 13 +++++++++++++ xen/arch/x86/mm/p2m-ept.c | 2 +- xen/include/asm-x86/hvm/vmx/vmcs.h | 6 ++++++ 5 files changed, 49 insertions(+), 9 deletions(-)