XSA-409 fixes
cover XSA-409 fixes
For 4.17, and backport to all stable releases.
Patch 2 doesn't technically need backporting, but it's safe and I'm dis-enclined to forgo testing in older releases.
See patches for changelogs.
Andrew Cooper (4):
xen: Introduce non-broken hypercalls for the paging mempool size tools/tests: Unit test for paging mempool size
xen/arm, libxl: Revert XEN_DOMCTL_shadow_op; use p2m mempool hypercalls xen/arm: Correct the p2m pool size calculations
tools/include/xenctrl.h | 3 + tools/libs/ctrl/xc_domain.c | 29 ++++ tools/libs/light/libxl_arm.c | 14 +- tools/libs/light/libxl_dom.c | 19 +++ tools/libs/light/libxl_internal.h | 3 + tools/libs/light/libxl_x86.c | 15 +- tools/tests/Makefile | 1 + tools/tests/paging-mempool/.gitignore | 1 + tools/tests/paging-mempool/Makefile | 42 ++++++ tools/tests/paging-mempool/test-paging-mempool.c | 181 +++++++++++++++++++++++ xen/arch/arm/domctl.c | 53 ------- xen/arch/arm/include/asm/p2m.h | 1 - xen/arch/arm/p2m.c | 32 ++-- xen/arch/x86/include/asm/hap.h | 1 + xen/arch/x86/include/asm/shadow.h | 4 + xen/arch/x86/mm/hap/hap.c | 11 ++ xen/arch/x86/mm/paging.c | 43 ++++++ xen/arch/x86/mm/shadow/common.c | 11 ++ xen/common/domctl.c | 14 ++ xen/include/public/domctl.h | 24 ++- xen/include/xen/domain.h | 3 + 21 files changed, 415 insertions(+), 90 deletions(-) create mode 100644 tools/tests/paging-mempool/.gitignore create mode 100644 tools/tests/paging-mempool/Makefile create mode 100644 tools/tests/paging-mempool/test-paging-mempool.c
Signature · 1 line
2.11.0
PATCH 1/4 xen: Introduce non-broken hypercalls for the paging mempool size
The existing XEN_DOMCTL_SHADOW_OP_{GET,SET}_ALLOCATION have problems:
* All set_allocation() flavours have an overflow-before-widen bug when calculating "sc->mb << (20 - PAGE_SHIFT)".
* All flavours have a granularity of 1M. This was tolerable when the size of the pool could only be set at the same granularity, but is broken now that ARM has a 16-page stopgap allocation in use.
* All get_allocation() flavours round up, and in particular turn 0 into 1, meaning the get op returns junk before a successful set op.
* The x86 flavours reject the hypercalls before the VM has vCPUs allocated, despite the pool size being a domain property.
* Even the hypercall names are long-obsolete.
Implement a better interface, which can be first used to unit test the behaviour, and subsequently correct a broken implementation. The old interface will be retired in due course.
The unit of bytes (as opposed pages) is a deliberate API/ABI improvement to more easily support multiple page granularities.
This is part of XSA-409 / CVE-2022-33747.
---
CC: Jan Beulich <[email protected]>
CC: Roger Pau Monné <[email protected]>
CC: Wei Liu <[email protected]>
CC: Stefano Stabellini <[email protected]>
CC: Julien Grall <[email protected]>
CC: Volodymyr Babchuk <[email protected]>
CC: Bertrand Marquis <[email protected]>
CC: Henry Wang <[email protected]>
CC: Anthony PERARD <[email protected]>
v2:
* s/p2m/paging/
* Fix overflow-before-widen in ARM's arch_get_p2m_mempool_size()
* Fix overflow-before-widen in both {hap,shadow}_get_allocation_bytes()
* Leave a TODO about x86/PV, drop assertion.
* Check for long->int truncation in x86's arch_set_paging_mempool_size()
Future TODOs:
* x86 shadow still rounds up. This is buggy as it's a simultaneous equation with tot_pages which varies over time with ballooning.
* x86 PV is weird. There are no toolstack interact with the shadow pool size, but the "shadow" pool it does come into existence when logdirty (or pv-l1tf) when first enabled.
* The shadow+hap logic is in desperate need of deduping.
--- tools/include/xenctrl.h | 3 +++ tools/libs/ctrl/xc_domain.c | 29 ++++++++++++++++++++++++++ xen/arch/arm/p2m.c | 26 +++++++++++++++++++++++ xen/arch/x86/include/asm/hap.h | 1 + xen/arch/x86/include/asm/shadow.h | 4 ++++ xen/arch/x86/mm/hap/hap.c | 11 ++++++++++ xen/arch/x86/mm/paging.c | 43 +++++++++++++++++++++++++++++++++++++++ xen/arch/x86/mm/shadow/common.c | 11 ++++++++++ xen/common/domctl.c | 14 +++++++++++++ xen/include/public/domctl.h | 24 +++++++++++++++++++++- xen/include/xen/domain.h | 3 +++ 11 files changed, 168 insertions(+), 1 deletion(-)
tools/include/xenctrl.h +3 -0
@@ -893,6 +893,9 @@ long long xc_logdirty_control(xc_interface *xch,unsigned int mode, xc_shadow_op_stats_t *stats); +int xc_get_paging_mempool_size (xc_interface *xch,uint32_t domid,uint64_t *size); +int xc_set_paging_mempool_size (xc_interface *xch,uint32_t domid,uint64_t size); +int xc_sched_credit_domain_set (xc_interface *xch,uint32_t domid,struct xen_domctl_sched_credit *sdom);
tools/libs/ctrl/xc_domain.c +29 -0
@@ -706,6 +706,35 @@ long long xc_logdirty_control(xc_interface *xch,return (rc ==0 ) ? domctl.u.shadow_op.pages : rc; } +int xc_get_paging_mempool_size (xc_interface *xch,uint32_t domid,uint64_t *size) +{ +int rc; +struct xen_domctl domctl = { + .cmd = XEN_DOMCTL_get_paging_mempool_size, + .domain = domid, + }; + + rc =do_domctl (xch, &domctl); +if ( rc ) +return rc; + + *size = domctl.u.paging_mempool.size; +return 0 ; +} + +int xc_set_paging_mempool_size (xc_interface *xch,uint32_t domid,uint64_t size) +{ +struct xen_domctl domctl = { + .cmd = XEN_DOMCTL_set_paging_mempool_size, + .domain = domid, + .u.paging_mempool = { + .size = size, + }, + }; + +return do_domctl (xch, &domctl); +} +int xc_domain_setmaxmem (xc_interface *xch,uint32_t domid,uint64_t max_memkb)
xen/arch/arm/p2m.c +26 -0
@@ -100,6 +100,13 @@ unsigned int p2m_get_allocation(struct domain *d)return ROUNDUP (nr_pages,1 << (20 -PAGE_SHIFT )) >> (20 -PAGE_SHIFT ); } +/* Return the size of the pool, in bytes. */ +int arch_get_paging_mempool_size (struct domain *d,uint64_t *size) +{ + *size = (uint64_t )ACCESS_ONCE (d->arch.paging.p2m_total_pages) <<PAGE_SHIFT ; +return 0 ; +} +/* * Set the pool of pages to the required number of pages. * Returns 0 for success, non-zero for failure.
@@ -157,6 +164,25 @@ int p2m_set_allocation(struct domain *d, unsigned long pages, bool *preempted)return 0 ; } +int arch_set_paging_mempool_size (struct domain *d,uint64_t size) +{ +unsigned long pages = size >>PAGE_SHIFT ; +bool preempted = false; +int rc; + +if ( (size & ~PAGE_MASK ) ||/* Non page-sized request? */ + pages != (size >>PAGE_SHIFT ) )/* 32-bit overflow? */ +return -EINVAL ; + +spin_lock (&d->arch.paging.lock); + rc =p2m_set_allocation (d, pages, &preempted); +spin_unlock (&d->arch.paging.lock); + +ASSERT (preempted == (rc == -ERESTART )); + +return rc; +} +int p2m_teardown_allocation (struct domain *d) {int ret =0 ;
xen/arch/x86/include/asm/hap.h +1 -0
@@ -47,6 +47,7 @@ int hap_track_dirty_vram(struct domain *d,extern const struct paging_mode *hap_paging_get_mode (struct vcpu *);int hap_set_allocation (struct domain *d,unsigned int pages,bool *preempted);unsigned int hap_get_allocation (struct domain *d); +int hap_get_allocation_bytes (struct domain *d,uint64_t *size);#endif /* XEN_HAP_H */
xen/arch/x86/include/asm/shadow.h +4 -0
@@ -97,6 +97,8 @@ void shadow_blow_tables_per_domain(struct domain *d);int shadow_set_allocation (struct domain *d,unsigned int pages,bool *preempted); +int shadow_get_allocation_bytes (struct domain *d,uint64_t *size); +#else /* !CONFIG_SHADOW_PAGING */ #define shadow_vcpu_teardown (v) ASSERT(is_pv_vcpu(v))
@@ -108,6 +110,8 @@ int shadow_set_allocation(struct domain *d, unsigned int pages, ({ ASSERT_UNREACHABLE(); -EOPNOTSUPP; })#define shadow_set_allocation (d, pages, preempted) \ ({ ASSERT_UNREACHABLE(); -EOPNOTSUPP; }) +#define shadow_get_allocation_bytes (d, size) \ + ({ ASSERT_UNREACHABLE(); -EOPNOTSUPP; })static inline void sh_remove_shadows (struct domain *d,mfn_t gmfn,int fast,int all) {}
xen/arch/x86/mm/hap/hap.c +11 -0
@@ -345,6 +345,17 @@ unsigned int hap_get_allocation(struct domain *d) + ((pg & ((1 << (20 - PAGE_SHIFT)) - 1)) ? 1 : 0)); } +int hap_get_allocation_bytes (struct domain *d,uint64_t *size) +{ +unsigned long pages = d->arch.paging.hap.total_pages; + + pages += d->arch.paging.hap.p2m_pages; + + *size = pages <<PAGE_SHIFT ; + +return 0 ; +} +/* Set the pool of pages to the required number of pages. * Returns 0 for success, non-zero for failure. */ int hap_set_allocation (struct domain *d,unsigned int pages,bool *preempted)
xen/arch/x86/mm/paging.c +43 -0
49 lines not on this page · raw message
xen/arch/x86/mm/shadow/common.c +11 -0
@@ -1427,6 +1427,17 @@ static unsigned int shadow_get_allocation(struct domain *d) + ((pg & ((1 << (20 - PAGE_SHIFT)) - 1)) ? 1 : 0)); } +int shadow_get_allocation_bytes (struct domain *d,uint64_t *size) +{ +unsigned long pages = d->arch.paging.shadow.total_pages; + + pages += d->arch.paging.shadow.p2m_pages; + + *size = pages <<PAGE_SHIFT ; + +return 0 ; +} +/**************************************************************************/ /* Hash table for storing the guest->shadow mappings. * The table itself is an array of pointers to shadows; the shadows are then
xen/common/domctl.c +14 -0
@@ -874,6 +874,20 @@ long do_domctl(XEN_GUEST_HANDLE_PARAM(xen_domctl_t) u_domctl) ret =iommu_do_domctl (op, d, u_domctl);break ; +case XEN_DOMCTL_get_paging_mempool_size: + ret =arch_get_paging_mempool_size (d, &op->u.paging_mempool.size); +if ( !ret ) + copyback =1 ; +break ; + +case XEN_DOMCTL_set_paging_mempool_size: + ret =arch_set_paging_mempool_size (d, op->u.paging_mempool.size); + +if ( ret == -ERESTART ) + ret =hypercall_create_continuation ( + __HYPERVISOR_domctl,"h" , u_domctl); +break ; +default : ret =arch_do_domctl (op, d, u_domctl);break ;
xen/include/public/domctl.h +23 -1
@@ -214,7 +214,10 @@ struct xen_domctl_getpageframeinfo3 {/* Return the bitmap but do not modify internal copy. */ #define XEN_DOMCTL_SHADOW_OP_PEEK 12 -/* Memory allocation accessors. */ +/* +* Memory allocation accessors. These APIs are broken and will be removed. +* Use XEN_DOMCTL_{get,set}_paging_mempool_size instead. +*/ #define XEN_DOMCTL_SHADOW_OP_GET_ALLOCATION 30#define XEN_DOMCTL_SHADOW_OP_SET_ALLOCATION 31
@@ -946,6 +949,22 @@ struct xen_domctl_cacheflush {xen_pfn_t start_pfn, nr_pfns; }; +/* +* XEN_DOMCTL_get_paging_mempool_size / XEN_DOMCTL_set_paging_mempool_size. +* +* Get or set the paging memory pool size. The size is in bytes. +* +* This is a dedicated pool of memory for Xen to use while managing the guest, +* typically containing pagetables. As such, there is an implementation +* specific minimum granularity. +* +* The set operation can fail mid-way through the request (e.g. Xen running +* out of memory, no free memory to reclaim from the pool, etc.). +*/ +struct xen_domctl_paging_mempool { +uint64_aligned_t size;/* IN/OUT. Size in bytes. */ +}; + #if defined(__i386__) || defined(__x86_64__) struct xen_domctl_vcpu_msr { uint32_t index;
@@ -1274,6 +1293,8 @@ struct xen_domctl {#define XEN_DOMCTL_get_cpu_policy 82#define XEN_DOMCTL_set_cpu_policy 83#define XEN_DOMCTL_vmtrace_op 84 +#define XEN_DOMCTL_get_paging_mempool_size 85 +#define XEN_DOMCTL_set_paging_mempool_size 86#define XEN_DOMCTL_gdbsx_guestmemio 1000#define XEN_DOMCTL_gdbsx_pausevcpu 1001#define XEN_DOMCTL_gdbsx_unpausevcpu 1002
@@ -1335,6 +1356,7 @@ struct xen_domctl {struct xen_domctl_psr_alloc psr_alloc;struct xen_domctl_vuart_op vuart_op;struct xen_domctl_vmtrace_op vmtrace_op; +struct xen_domctl_paging_mempool paging_mempool;uint8_t pad[128 ]; } u; };
xen/include/xen/domain.h +3 -0
@@ -98,6 +98,9 @@ void arch_get_info_guest(struct vcpu *, vcpu_guest_context_u);int arch_initialise_vcpu (struct vcpu *v,XEN_GUEST_HANDLE_PARAM (void ) arg);int default_initialise_vcpu (struct vcpu *v,XEN_GUEST_HANDLE_PARAM (void ) arg); +int arch_get_paging_mempool_size (struct domain *d,uint64_t *size/* bytes */ ); +int arch_set_paging_mempool_size (struct domain *d,uint64_t size/* bytes */ ); +int domain_relinquish_resources (struct domain *d);void dump_pageframe_info (struct domain *d);
Signature · 1 line
2.11.0
Re: [PATCH 1/4] xen: Introduce non-broken hypercalls for the paging mempool size
Quoted 17 lines · Andrew Cooper
The existing XEN_DOMCTL_SHADOW_OP_{GET,SET}_ALLOCATION have problems:
* All
set_allocation()flavours have an overflow-before-widen bug when calculating "sc->mb<< (20 -PAGE_SHIFT)".
* All flavours have a granularity of 1M. This was tolerable when the size of the pool could only be set at the same granularity, but is broken now that ARM has a 16-page stopgap allocation in use.
* Allget_allocation()flavours round up, and in particular turn 0 into 1, meaning the get op returns junk before a successful set op.
* The x86 flavours reject the hypercalls before the VM has vCPUs allocated, despite the pool size being a domain property.
* Even the hypercall names are long-obsolete.Implement a better interface, which can be first used to unit test the behaviour, and subsequently correct a broken implementation. The old interface will be retired in due course.
Quoted 7 lines
The unit of bytes (as opposed pages) is a deliberate API/ABI improvement to more easily support multiple page granularities.
This is part of XSA-409 / CVE-2022-33747.
Signed-off-by Andrew Cooper <[email protected]>Release-acked-by Henry Wang <[email protected]>
ARM side:
Quoted 367 lines
---
CC: Jan Beulich <[email protected]>
CC: Roger Pau Monné <[email protected]>
CC: Wei Liu <[email protected]>
CC: Stefano Stabellini <[email protected]>
CC: Julien Grall <[email protected]>
CC: Volodymyr Babchuk <[email protected]>
CC: Bertrand Marquis <[email protected]>
CC: Henry Wang <[email protected]>
CC: Anthony PERARD <[email protected]>v2:
* s/p2m/paging/
* Fix overflow-before-widen in ARM'sarch_get_p2m_mempool_size()
* Fix overflow-before-widen in both {hap,shadow}_get_allocation_bytes()
* Leave a TODO about x86/PV, drop assertion.
* Check forlong->inttruncation in x86'sarch_set_paging_mempool_size()Future TODOs:
* x86 shadow still rounds up. This is buggy as it's a simultaneous equation withtot_pageswhich varies over time with ballooning.
* x86 PV is weird. There are no toolstack interact with the shadow pool size, but the "shadow" pool it does come into existence when logdirty (or pv-l1tf) when first enabled.
* The shadow+hap logic is in desperate need of deduping.--- tools/include/xenctrl.h | 3 +++ tools/libs/ctrl/xc_domain.c | 29 ++++++++++++++++++++++++++ xen/arch/arm/p2m.c | 26 +++++++++++++++++++++++ xen/arch/x86/include/asm/hap.h | 1 + xen/arch/x86/include/asm/shadow.h | 4 ++++ xen/arch/x86/mm/hap/hap.c | 11 ++++++++++ xen/arch/x86/mm/paging.c | 43 +++++++++++++++++++++++++++++++++++++++ xen/arch/x86/mm/shadow/common.c | 11 ++++++++++ xen/common/domctl.c | 14 +++++++++++++ xen/include/public/domctl.h | 24 +++++++++++++++++++++- xen/include/xen/domain.h | 3 +++ 11 files changed, 168 insertions(+), 1 deletion(-)tools/include/xenctrl.h +3
-0@@ -893,6 +893,9 @@ long long xc_logdirty_control(xc_interface *xch,unsigned int mode, xc_shadow_op_stats_t *stats); +int xc_get_paging_mempool_size (xc_interface *xch,uint32_t domid,uint64_t *size); +int xc_set_paging_mempool_size (xc_interface *xch,uint32_t domid,uint64_t size); +int xc_sched_credit_domain_set (xc_interface *xch,uint32_t domid,struct xen_domctl_sched_credit *sdom);tools/libs/ctrl/xc_domain.c +29
-0@@ -706,6 +706,35 @@ long long xc_logdirty_control(xc_interface *xch,return (rc ==0 ) ? domctl.u.shadow_op.pages : rc; } +int xc_get_paging_mempool_size (xc_interface *xch,uint32_t domid,uint64_t *size) +{ +int rc; +struct xen_domctl domctl = { + .cmd = XEN_DOMCTL_get_paging_mempool_size, + .domain = domid, + }; + + rc =do_domctl (xch, &domctl); +if ( rc ) +return rc; + + *size = domctl.u.paging_mempool.size; +return 0 ; +} + +int xc_set_paging_mempool_size (xc_interface *xch,uint32_t domid,uint64_t size) +{ +struct xen_domctl domctl = { + .cmd = XEN_DOMCTL_set_paging_mempool_size, + .domain = domid, + .u.paging_mempool = { + .size = size, + }, + }; + +return do_domctl (xch, &domctl); +} +int xc_domain_setmaxmem (xc_interface *xch,uint32_t domid,uint64_t max_memkb)xen/arch/arm/p2m.c +26
-0@@ -100,6 +100,13 @@ unsigned int p2m_get_allocation(struct domain *d)return ROUNDUP (nr_pages,1 << (20 -PAGE_SHIFT )) >> (20 -PAGE_SHIFT ); } +/* Return the size of the pool, in bytes. */ +int arch_get_paging_mempool_size (struct domain *d,uint64_t *size) +{ + *size = (uint64_t )ACCESS_ONCE (d->arch.paging.p2m_total_pages) <<PAGE_SHIFT ; +return 0 ; +} +/* * Set the pool of pages to the required number of pages. * Returns 0 for success, non-zero for failure. @@ -157,6 +164,25 @@ int p2m_set_allocation(struct domain *d, unsigned long pages, bool *preempted)return 0 ; } +int arch_set_paging_mempool_size (struct domain *d,uint64_t size) +{ +unsigned long pages = size >>PAGE_SHIFT ; +bool preempted = false; +int rc; + +if ( (size & ~PAGE_MASK ) ||/* Non page-sized request? */ + pages != (size >>PAGE_SHIFT ) )/* 32-bit overflow? */ +return -EINVAL ; + +spin_lock (&d->arch.paging.lock); + rc =p2m_set_allocation (d, pages, &preempted); +spin_unlock (&d->arch.paging.lock); + +ASSERT (preempted == (rc == -ERESTART )); + +return rc; +} +int p2m_teardown_allocation (struct domain *d) {int ret =0 ;xen/arch/x86/include/asm/hap.h +1
-0@@ -47,6 +47,7 @@ int hap_track_dirty_vram(struct domain *d,extern const struct paging_mode *hap_paging_get_mode (struct vcpu *);int hap_set_allocation (struct domain *d,unsigned int pages,bool *preempted);unsigned int hap_get_allocation (struct domain *d); +int hap_get_allocation_bytes (struct domain *d,uint64_t *size);#endif /* XEN_HAP_H */ xen/arch/x86/include/asm/shadow.h +4
-0@@ -97,6 +97,8 @@ void shadow_blow_tables_per_domain(struct domain *d);int shadow_set_allocation (struct domain *d,unsigned int pages,bool *preempted); +int shadow_get_allocation_bytes (struct domain *d,uint64_t *size); +#else /* !CONFIG_SHADOW_PAGING */ #define shadow_vcpu_teardown (v) ASSERT(is_pv_vcpu(v))@@ -108,6 +110,8 @@ int shadow_set_allocation(struct domain *d, unsigned int pages, ({ ASSERT_UNREACHABLE(); -EOPNOTSUPP; })#define shadow_set_allocation (d, pages, preempted) \ ({ ASSERT_UNREACHABLE(); -EOPNOTSUPP; }) +#define shadow_get_allocation_bytes (d, size) \ + ({ ASSERT_UNREACHABLE(); -EOPNOTSUPP; })static inline void sh_remove_shadows (struct domain *d,mfn_t gmfn,int fast,int all) {}xen/arch/x86/mm/hap/hap.c +11
-0@@ -345,6 +345,17 @@ unsigned int hap_get_allocation(struct domain *d) + ((pg & ((1 << (20 - PAGE_SHIFT)) - 1)) ? 1 : 0)); } +int hap_get_allocation_bytes (struct domain *d,uint64_t *size) +{ +unsigned long pages = d->arch.paging.hap.total_pages; + + pages += d->arch.paging.hap.p2m_pages; + + *size = pages <<PAGE_SHIFT ; + +return 0 ; +} +/* Set the pool of pages to the required number of pages. * Returns 0 for success, non-zero for failure. */ int hap_set_allocation (struct domain *d,unsigned int pages,bool *preempted)xen/arch/x86/mm/paging.c +43
-0@@ -977,6 +977,49 @@ int __init paging_set_allocation(struct domain *d, unsigned int pages, }#endif +int arch_get_paging_mempool_size (struct domain *d,uint64_t *size) +{ +int rc; + +if (is_pv_domain (d) )/* TODO: Relax in due course */ +return -EOPNOTSUPP ; + +if (hap_enabled (d) ) + rc =hap_get_allocation_bytes (d, size); +else + rc =shadow_get_allocation_bytes (d, size); + +return rc; +} + +int arch_set_paging_mempool_size (struct domain *d,uint64_t size) +{ +unsigned long pages = size >>PAGE_SHIFT ; +bool preempted = false; +int rc; + +if (is_pv_domain (d) )/* TODO: Relax in due course */ +return -EOPNOTSUPP ; + +if ( size & ~PAGE_MASK ||/* Non page-sized request? */ + pages != (unsigned int )pages )/* Overflow $X_set_allocation()? */ +return -EINVAL ; + +paging_lock (d); +if (hap_enabled (d) ) + rc =hap_set_allocation (d, pages, &preempted); +else + rc =shadow_set_allocation (d, pages, &preempted); +paging_unlock (d); + +/* +* TODO: Adjust $X_set_allocation() so this is true. +ASSERT(preempted == (rc == -ERESTART)); +*/ + +return preempted ? -ERESTART : rc; +} +/* * Local variables: * mode: C xen/arch/x86/mm/shadow/common.c +11
-0@@ -1427,6 +1427,17 @@ static unsigned int shadow_get_allocation(struct domain *d) + ((pg & ((1 << (20 - PAGE_SHIFT)) - 1)) ? 1 : 0)); } +int shadow_get_allocation_bytes (struct domain *d,uint64_t *size) +{ +unsigned long pages = d->arch.paging.shadow.total_pages; + + pages += d->arch.paging.shadow.p2m_pages; + + *size = pages <<PAGE_SHIFT ; + +return 0 ; +} +/**************************************************************************/ /* Hash table for storing the guest->shadow mappings. * The table itself is an array of pointers to shadows; the shadows are then xen/common/domctl.c +14
-0@@ -874,6 +874,20 @@ long do_domctl(XEN_GUEST_HANDLE_PARAM(xen_domctl_t) u_domctl) ret =iommu_do_domctl (op, d, u_domctl);break ; +case XEN_DOMCTL_get_paging_mempool_size: + ret =arch_get_paging_mempool_size (d, &op->u.paging_mempool.size); +if ( !ret ) + copyback =1 ; +break ; + +case XEN_DOMCTL_set_paging_mempool_size: + ret =arch_set_paging_mempool_size (d, op->u.paging_mempool.size); + +if ( ret == -ERESTART ) + ret =hypercall_create_continuation ( + __HYPERVISOR_domctl,"h" , u_domctl); +break ; +default : ret =arch_do_domctl (op, d, u_domctl);break ;xen/include/public/domctl.h +23
-1@@ -214,7 +214,10 @@ struct xen_domctl_getpageframeinfo3 {/* Return the bitmap but do not modify internal copy. */ #define XEN_DOMCTL_SHADOW_OP_PEEK 12 -/* Memory allocation accessors. */ +/* +* Memory allocation accessors. These APIs are broken and will be removed. +* Use XEN_DOMCTL_{get,set}_paging_mempool_size instead. +*/ #define XEN_DOMCTL_SHADOW_OP_GET_ALLOCATION 30#define XEN_DOMCTL_SHADOW_OP_SET_ALLOCATION 31@@ -946,6 +949,22 @@ struct xen_domctl_cacheflush {xen_pfn_t start_pfn, nr_pfns; }; +/* +* XEN_DOMCTL_get_paging_mempool_size / XEN_DOMCTL_set_paging_mempool_size. +* +* Get or set the paging memory pool size. The size is in bytes. +* +* This is a dedicated pool of memory for Xen to use while managing the guest, +* typically containing pagetables. As such, there is an implementation +* specific minimum granularity. +* +* The set operation can fail mid-way through the request (e.g. Xen running +* out of memory, no free memory to reclaim from the pool, etc.). +*/ +struct xen_domctl_paging_mempool { +uint64_aligned_t size;/* IN/OUT. Size in bytes. */ +}; + #if defined(__i386__) || defined(__x86_64__) struct xen_domctl_vcpu_msr { uint32_t index;@@ -1274,6 +1293,8 @@ struct xen_domctl {#define XEN_DOMCTL_get_cpu_policy 82#define XEN_DOMCTL_set_cpu_policy 83#define XEN_DOMCTL_vmtrace_op 84 +#define XEN_DOMCTL_get_paging_mempool_size 85 +#define XEN_DOMCTL_set_paging_mempool_size 86#define XEN_DOMCTL_gdbsx_guestmemio 1000#define XEN_DOMCTL_gdbsx_pausevcpu 1001#define XEN_DOMCTL_gdbsx_unpausevcpu 1002@@ -1335,6 +1356,7 @@ struct xen_domctl {struct xen_domctl_psr_alloc psr_alloc;struct xen_domctl_vuart_op vuart_op;struct xen_domctl_vmtrace_op vmtrace_op; +struct xen_domctl_paging_mempool paging_mempool;uint8_t pad[128 ]; } u; };xen/include/xen/domain.h +3
-0@@ -98,6 +98,9 @@ void arch_get_info_guest(struct vcpu *, vcpu_guest_context_u);int arch_initialise_vcpu (struct vcpu *v,XEN_GUEST_HANDLE_PARAM (void ) arg);int default_initialise_vcpu (struct vcpu *v,XEN_GUEST_HANDLE_PARAM (void ) arg); +int arch_get_paging_mempool_size (struct domain *d,uint64_t *size/* bytes */ ); +int arch_set_paging_mempool_size (struct domain *d,uint64_t size/* bytes */ ); +int domain_relinquish_resources (struct domain *d);void dump_pageframe_info (struct domain *d);Signature · 1 line
2.11.0
Re: [PATCH 1/4] xen: Introduce non-broken hypercalls for the paging mempool size
In reply to Andrew CooperQuoted 13 lines · Andrew Cooper
The existing XEN_DOMCTL_SHADOW_OP_{GET,SET}_ALLOCATION have problems:
* All
set_allocation()flavours have an overflow-before-widen bug when calculating "sc->mb<< (20 -PAGE_SHIFT)".
* All flavours have a granularity of 1M. This was tolerable when the size of the pool could only be set at the same granularity, but is broken now that ARM has a 16-page stopgap allocation in use.
* Allget_allocation()flavours round up, and in particular turn 0 into 1, meaning the get op returns junk before a successful set op.
* The x86 flavours reject the hypercalls before the VM has vCPUs allocated, despite the pool size being a domain property.
* Even the hypercall names are long-obsolete.
Quoted 8 lines
Implement a better interface, which can be first used to unit test the behaviour, and subsequently correct a broken implementation. The old interface will be retired in due course.
The unit of bytes (as opposed pages) is a deliberate API/ABI improvement to more easily support multiple page granularities.
This is part of XSA-409 / CVE-2022-33747.
While I'm not convinced of this attribution, ...
Quoted 2 lines
Signed-off-by Andrew Cooper <[email protected]>Release-acked-by Henry Wang <[email protected]>
albeit with remarks:
xen/arch/x86/mm/paging.c int __init paging_set_allocation(struct domain *d, unsigned int pages, · line 985
Quoted 12 lines
xen/arch/x86/mm/paging.c +6
-0@@ -977,6 +977,49 @@ int __init paging_set_allocation(struct domain *d, unsigned int pages, }#endif +int arch_get_paging_mempool_size (struct domain *d,uint64_t *size) +{ +int rc; + +if (is_pv_domain (d) )/* TODO: Relax in due course */ +return -EOPNOTSUPP ;
I guess this is merely for symmetry with ...
xen/arch/x86/mm/paging.c int __init paging_set_allocation(struct domain *d, unsigned int pages, · line 1002
Quoted 8 lines
xen/arch/x86/mm/paging.c +8
-0+int arch_set_paging_mempool_size (struct domain *d,uint64_t size) +{ +unsigned long pages = size >>PAGE_SHIFT ; +bool preempted = false; +int rc; + +if (is_pv_domain (d) )/* TODO: Relax in due course */ +return -EOPNOTSUPP ;
... this, since otherwise "get" ought to be fine for PV?
Quoted 8 lines
xen/arch/x86/mm/paging.c +4
-0@@ -946,6 +949,22 @@ struct xen_domctl_cacheflush {xen_pfn_t start_pfn, nr_pfns; }; +/* +* XEN_DOMCTL_get_paging_mempool_size / XEN_DOMCTL_set_paging_mempool_size. +* +* Get or set the paging memory pool size. The size is in bytes.
xen/arch/x86/mm/paging.c struct xen_domctl_cacheflush { · line 965
Quoted 10 lines
xen/arch/x86/mm/paging.c +10
-0+* +* This is a dedicated pool of memory for Xen to use while managing the guest, +* typically containing pagetables. As such, there is an implementation +* specific minimum granularity. +* +* The set operation can fail mid-way through the request (e.g. Xen running +* out of memory, no free memory to reclaim from the pool, etc.). +*/ +struct xen_domctl_paging_mempool { +uint64_aligned_t size;/* IN/OUT. Size in bytes. */
While likely people will correctly infer what is meant, strictly speaking this is wrong: The field is IN for "set" and OUT for "get".
Jan
Re: [PATCH 1/4] xen: Introduce non-broken hypercalls for the paging mempool size
Quoted 18 lines · Jan Beulich
Quoted 17 lines · Andrew Cooper
The existing XEN_DOMCTL_SHADOW_OP_{GET,SET}_ALLOCATION have problems:
* All
set_allocation()flavours have an overflow-before-widen bug when calculating "sc->mb<< (20 -PAGE_SHIFT)".
* All flavours have a granularity of 1M. This was tolerable when the size of the pool could only be set at the same granularity, but is broken now that ARM has a 16-page stopgap allocation in use.
* Allget_allocation()flavours round up, and in particular turn 0 into 1, meaning the get op returns junk before a successful set op.
* The x86 flavours reject the hypercalls before the VM has vCPUs allocated, despite the pool size being a domain property.
* Even the hypercall names are long-obsolete.Implement a better interface, which can be first used to unit test the behaviour, and subsequently correct a broken implementation. The old interface will be retired in due course.
Quoted 5 lines
Quoted 4 lines
The unit of bytes (as opposed pages) is a deliberate API/ABI improvement to more easily support multiple page granularities.
This is part of XSA-409 / CVE-2022-33747.
While I'm not convinced of this attribution, ...
I think this very much depends on how critical the unit test is deemed.
If this was done the first time around, it would all have had attribution. We're on the 3rd set of fixes, and the unit test is a key justification of the safety of the fix.
Quoted 3 lines
Quoted 2 lines
Signed-off-by Andrew Cooper <[email protected]>Release-acked-by Henry Wang <[email protected]>Reviewed-by Jan Beulich <[email protected]> # hypervisor
Thanks.
Quoted 16 lines
albeit with remarks:
xen/arch/x86/mm/paging.c int __init paging_set_allocation(struct domain *d, unsigned int pages, · line 985
Quoted 12 lines
xen/arch/x86/mm/paging.c +6
-0@@ -977,6 +977,49 @@ int __init paging_set_allocation(struct domain *d, unsigned int pages, }#endif +int arch_get_paging_mempool_size (struct domain *d,uint64_t *size) +{ +int rc; + +if (is_pv_domain (d) )/* TODO: Relax in due course */ +return -EOPNOTSUPP ;I guess this is merely for symmetry with ...
Quoted 9 lines
xen/arch/x86/mm/paging.c int __init paging_set_allocation(struct domain *d, unsigned int pages, · line 1002
Quoted 8 lines
xen/arch/x86/mm/paging.c +8
-0+int arch_set_paging_mempool_size (struct domain *d,uint64_t size) +{ +unsigned long pages = size >>PAGE_SHIFT ; +bool preempted = false; +int rc; + +if (is_pv_domain (d) )/* TODO: Relax in due course */ +return -EOPNOTSUPP ;... this, since otherwise "get" ought to be fine for PV?
Its the safest course of action, given other known issues with PV. There's no need for a working get without a working set.
Quoted 5 lines
Quoted 5 lines
xen/arch/x86/mm/paging.c +0
-0@@ -946,6 +949,22 @@ struct xen_domctl_cacheflush {xen_pfn_t start_pfn, nr_pfns; };
Quoted 16 lines
Quoted 4 lines
xen/arch/x86/mm/paging.c +4
-0+/* +* XEN_DOMCTL_get_paging_mempool_size / XEN_DOMCTL_set_paging_mempool_size. +* +* Get or set the paging memory pool size. The size is in bytes. xen/arch/x86/mm/paging.c struct xen_domctl_cacheflush { · line 965
Quoted 10 lines
xen/arch/x86/mm/paging.c +10
-0+* +* This is a dedicated pool of memory for Xen to use while managing the guest, +* typically containing pagetables. As such, there is an implementation +* specific minimum granularity. +* +* The set operation can fail mid-way through the request (e.g. Xen running +* out of memory, no free memory to reclaim from the pool, etc.). +*/ +struct xen_domctl_paging_mempool { +uint64_aligned_t size;/* IN/OUT. Size in bytes. */ While likely people will correctly infer what is meant, strictly speaking this is wrong: The field is IN for "set" and OUT for "get".
I'll drop them, to reduce any possible confusion. As you say, the meaning is entirely clear.
~Andrew
Re: [PATCH 1/4] xen: Introduce non-broken hypercalls for the paging mempool size
In reply to Andrew CooperQuoted 17 lines · Andrew Cooper
The existing XEN_DOMCTL_SHADOW_OP_{GET,SET}_ALLOCATION have problems:
* All
set_allocation()flavours have an overflow-before-widen bug when calculating "sc->mb<< (20 -PAGE_SHIFT)".
* All flavours have a granularity of 1M. This was tolerable when the size of the pool could only be set at the same granularity, but is broken now that ARM has a 16-page stopgap allocation in use.
* Allget_allocation()flavours round up, and in particular turn 0 into 1, meaning the get op returns junk before a successful set op.
* The x86 flavours reject the hypercalls before the VM has vCPUs allocated, despite the pool size being a domain property.
* Even the hypercall names are long-obsolete.Implement a better interface, which can be first used to unit test the behaviour, and subsequently correct a broken implementation. The old interface will be retired in due course.
Quoted 7 lines
The unit of bytes (as opposed pages) is a deliberate API/ABI improvement to more easily support multiple page granularities.
This is part of XSA-409 / CVE-2022-33747.
Signed-off-by Andrew Cooper <[email protected]>Release-acked-by Henry Wang <[email protected]>
Thanks,
Signature · 1 line
Anthony PERARD
PATCH 2/4 tools/tests: Unit test for paging mempool size
In reply to Andrew CooperExercise some basic functionality of the new
xc_{get,set}_paging_mempool_size() hypercalls.
This passes on x86, but fails currently on ARM. ARM will be fixed up in future patches.
This is part of XSA-409 / CVE-2022-33747.
---
CC: Jan Beulich <[email protected]>
CC: Roger Pau Monné <[email protected]>
CC: Wei Liu <[email protected]>
CC: Stefano Stabellini <[email protected]>
CC: Julien Grall <[email protected]>
CC: Volodymyr Babchuk <[email protected]>
CC: Bertrand Marquis <[email protected]>
CC: Henry Wang <[email protected]>
CC: Anthony PERARD <[email protected]>
x86 Shadow is complicated because of how it behaves for PV guests, and because of how it forms a simultaneous equation with tot_pages. This will require more work to untangle.
v2:
* s/p2m/paging/
* Fix CFLAGS_libxenforeginmemory typo
--- tools/tests/Makefile | 1 + tools/tests/paging-mempool/.gitignore | 1 + tools/tests/paging-mempool/Makefile | 42 ++++++ tools/tests/paging-mempool/test-paging-mempool.c | 181 +++++++++++++++++++++++ 4 files changed, 225 insertions(+) create mode 100644 tools/tests/paging-mempool/.gitignore create mode 100644 tools/tests/paging-mempool/Makefile create mode 100644 tools/tests/paging-mempool/test-paging-mempool.c
tools/tests/Makefile +1 -0
@@ -11,6 +11,7 @@ endif SUBDIRS-y += xenstore SUBDIRS-y += depriv SUBDIRS-y += vpci +SUBDIRS-y += paging-mempool .PHONY: all clean install distclean uninstall all clean distclean install uninstall: %: subdirs-%
tools/tests/paging-mempool/.gitignore new file +1 -0
@@ -0,0 +1 @@ +test-paging-mempool
tools/tests/paging-mempool/Makefile new file +42 -0
42 lines not on this page · raw message
tools/tests/paging-mempool/test-paging-mempool.c new file +181 -0
181 lines not on this page · raw message
Signature · 1 line
2.11.0
Re: [PATCH 2/4] tools/tests: Unit test for paging mempool size
Quoted 10 lines · Andrew Cooper
Exercise some basic functionality of the new
xc_{get,set}_paging_mempool_size() hypercalls.This passes on x86, but fails currently on ARM. ARM will be fixed up in future patches.
This is part of XSA-409 / CVE-2022-33747.
Signed-off-by Andrew Cooper <[email protected]>Release-acked-by Henry Wang <[email protected]>
(if this counts anything, since as it stands the new stuff all falls under tool stack maintainership)
Quoted 17 lines
tools/tests/paging-mempool/test-paging-mempool.c new file +14
-0@@ -0,0 +1,181 @@ +#include <err.h> +#include <errno.h> +#include <inttypes.h> +#include <stdio.h> +#include <string.h> +#include <sys/mman.h> + +#include <xenctrl.h> +#include <xenforeignmemory.h> +#include <xengnttab.h> +#include <xen-tools/libs.h> + +static unsigned int nr_failures; +#define fail (fmt, ...) \
tools/tests/paging-mempool/test-paging-mempool.c line 24
Quoted 10 lines
tools/tests/paging-mempool/test-paging-mempool.c new file +10
-0+({ \ + nr_failures++; \ + (void)printf(fmt, ##__VA_ARGS__); \ +}) + +static xc_interface *xch; +static uint32_t domid; + +static struct xen_domctl_createdomain create = { + .flags = XEN_DOMCTL_CDF_hvm | XEN_DOMCTL_CDF_hap,
I understand that it is accepted that this test will thus fail when run on HAP-incapable hardware (including when run with Xen itself running on top of another hypervisor not surfacing HAP capabilities)? Oh, I notice you're actually translating EINVAL and EOPNOTSUPP failures into "skip". That'll probably do, albeit personally I consider skipping when EINVAL (which we use all over the place) as a overly relaxed.
tools/tests/paging-mempool/test-paging-mempool.c line 45
Quoted 3 lines
tools/tests/paging-mempool/test-paging-mempool.c new file +3
-0+static void run_tests (void ) +{ +xen_pfn_t physmap[] = {0 };
I have to admit that I'm uncertain whether Arm (or other architectures that Xen is being planned to be ported to) have constraints which may cause populating of GFN 0 to fail.
Quoted 7 lines
tools/tests/paging-mempool/test-paging-mempool.c new file +7
-0+uint64_t size_bytes, old_size_bytes; +int rc; + +printf ("Test default mempool size\n" ); + + rc =xc_get_paging_mempool_size (xch, domid, &size_bytes); +if ( rc )
tools/tests/paging-mempool/test-paging-mempool.c line 62
Quoted 10 lines
tools/tests/paging-mempool/test-paging-mempool.c new file +10
-0+return fail (" Fail: get mempool size: %d - %s\n" , + errno,strerror (errno)); + +printf ("mempool size %" PRIu64" bytes (%" PRIu64"kB, %" PRIu64"MB)\n" , + size_bytes, size_bytes >>10 , size_bytes >>20 ); + + +/* +* Check that the domain has the expected default allocation size. This +* will fail if the logic in Xen is altered without an equivelent
Nit: equivalent
Quoted 42 lines
tools/tests/paging-mempool/test-paging-mempool.c new file +42
-0+* adjustment here. +*/ +if ( size_bytes != default_mempool_size_bytes ) +return fail (" Fail: size %" PRIu64" != expected size %" PRIu64"\n" , + size_bytes, default_mempool_size_bytes); + + +printf ("Test that allocate doesn't alter pool size\n" ); + +/* +* Populate the domain with some RAM. This will cause more of the mempool +* to be used. +*/ + old_size_bytes = size_bytes; + + rc =xc_domain_setmaxmem (xch, domid,-1 ); +if ( rc ) +return fail (" Fail: setmaxmem: : %d - %s\n" , + errno,strerror (errno)); + + rc =xc_domain_populate_physmap_exact (xch, domid,1 ,0 ,0 , physmap); +if ( rc ) +return fail (" Fail: populate physmap: %d - %s\n" , + errno,strerror (errno)); + +/* +* Re-get the p2m size. Should not have changed as a consequence of +* populate physmap. +*/ + rc =xc_get_paging_mempool_size (xch, domid, &size_bytes); +if ( rc ) +return fail (" Fail: get mempool size: %d - %s\n" , + errno,strerror (errno)); + +if ( old_size_bytes != size_bytes ) +return fail (" Fail: mempool size changed %" PRIu64" => %" PRIu64"\n" , + old_size_bytes, size_bytes); + + + +printf ("Test bad set size\n" ); +
tools/tests/paging-mempool/test-paging-mempool.c line 114
Quoted 10 lines
tools/tests/paging-mempool/test-paging-mempool.c new file +10
-0+/* +* Check that setting a non-page size results in failure. +*/ + rc =xc_set_paging_mempool_size (xch, domid, size_bytes +1 ); +if ( rc !=-1 || errno !=EINVAL ) +return fail (" Fail: Bad set size: expected -1/EINVAL, got %d/%d - %s\n" , + rc, errno,strerror (errno)); + + +printf ("Test very large set size\n" );
Maybe drop "very", as 64M isn't all that much (and would, in particular, not expose any 32-bit truncation issues)?
Jan
Re: [PATCH 2/4] tools/tests: Unit test for paging mempool size
Quoted 4 lines · Jan Beulich
Quoted 3 lines · Andrew Cooper
Exercise some basic functionality of the new
xc_{get,set}_paging_mempool_size() hypercalls.
Quoted 8 lines
Quoted 7 lines
This passes on x86, but fails currently on ARM. ARM will be fixed up in future patches.
This is part of XSA-409 / CVE-2022-33747.
Signed-off-by Andrew Cooper <[email protected]>Release-acked-by Henry Wang <[email protected]>Acked-by Jan Beulich <[email protected]>
Thanks.
Quoted 2 lines
(if this counts anything, since as it stands the new stuff all falls under tool stack maintainership)
I do intend to give it its own section in due course, but this falls very much into "The Rest" at the moment.
Quoted 25 lines
Quoted 25 lines
tools/tests/paging-mempool/test-paging-mempool.c new file +22
-0@@ -0,0 +1,181 @@ +#include <err.h> +#include <errno.h> +#include <inttypes.h> +#include <stdio.h> +#include <string.h> +#include <sys/mman.h> + +#include <xenctrl.h> +#include <xenforeignmemory.h> +#include <xengnttab.h> +#include <xen-tools/libs.h> + +static unsigned int nr_failures; +#define fail (fmt, ...) \ +({ \ + nr_failures++; \ + (void)printf(fmt, ##__VA_ARGS__); \ +}) + +static xc_interface *xch; +static uint32_t domid; +
Quoted 8 lines
tools/tests/paging-mempool/test-paging-mempool.c line 24
Quoted 2 lines
tools/tests/paging-mempool/test-paging-mempool.c new file +2
-0+static struct xen_domctl_createdomain create = { + .flags = XEN_DOMCTL_CDF_hvm | XEN_DOMCTL_CDF_hap,I understand that it is accepted that this test will thus fail when run on HAP-incapable hardware (including when run with Xen itself running on top of another hypervisor not surfacing HAP capabilities)? Oh, I notice you're actually translating EINVAL and EOPNOTSUPP failures into "skip". That'll probably do, albeit personally I consider skipping when EINVAL (which we use all over the place) as a overly relaxed.
Checking capabilities needs to happen anyway to get PV and HVM Shadow support working.
But this will do for now.
Quoted 6 lines
tools/tests/paging-mempool/test-paging-mempool.c line 45
Quoted 3 lines
tools/tests/paging-mempool/test-paging-mempool.c new file +3
-0+static void run_tests (void ) +{ +xen_pfn_t physmap[] = {0 };I have to admit that I'm uncertain whether Arm (or other architectures that Xen is being planned to be ported to) have constraints which may cause populating of GFN 0 to fail.
Mechanically, no. x86 PV is an entirely arbitrary mapping, while all HVM modes (x86 and ARM) are just filling in a guest physical -> "some RAM" mapping in a pagetable structure.
Logically, I hope that CDF_directmap on ARM checks for an alias to a real RAM block, but this capability isn't even exposed to the toolstack.
And now I've looked at this, ewwww. We've got:
d->options which holds config->flags, DOMCTL_CDF_* d->cdf which holds the local 'flags' parameter, CFD_* d->is_privileged which holds 'flags & CDF_privileged' and was never cleaned up when d->cdf was introduced.
This is unnecessarily confusing to follow. Yet more for the cleanup pile.
Quoted 10 lines
Quoted 10 lines
tools/tests/paging-mempool/test-paging-mempool.c new file +10
-0+uint64_t size_bytes, old_size_bytes; +int rc; + +printf ("Test default mempool size\n" ); + + rc =xc_get_paging_mempool_size (xch, domid, &size_bytes); +if ( rc ) +return fail (" Fail: get mempool size: %d - %s\n" , + errno,strerror (errno)); +
Quoted 8 lines
tools/tests/paging-mempool/test-paging-mempool.c line 62
Quoted 7 lines
tools/tests/paging-mempool/test-paging-mempool.c new file +7
-0+printf ("mempool size %" PRIu64" bytes (%" PRIu64"kB, %" PRIu64"MB)\n" , + size_bytes, size_bytes >>10 , size_bytes >>20 ); + + +/* +* Check that the domain has the expected default allocation size. This +* will fail if the logic in Xen is altered without an equivelent Nit: equivalent
Fixed.
Quoted 47 lines
Quoted 47 lines
tools/tests/paging-mempool/test-paging-mempool.c new file +46
-0+* adjustment here. +*/ +if ( size_bytes != default_mempool_size_bytes ) +return fail (" Fail: size %" PRIu64" != expected size %" PRIu64"\n" , + size_bytes, default_mempool_size_bytes); + + +printf ("Test that allocate doesn't alter pool size\n" ); + +/* +* Populate the domain with some RAM. This will cause more of the mempool +* to be used. +*/ + old_size_bytes = size_bytes; + + rc =xc_domain_setmaxmem (xch, domid,-1 ); +if ( rc ) +return fail (" Fail: setmaxmem: : %d - %s\n" , + errno,strerror (errno)); + + rc =xc_domain_populate_physmap_exact (xch, domid,1 ,0 ,0 , physmap); +if ( rc ) +return fail (" Fail: populate physmap: %d - %s\n" , + errno,strerror (errno)); + +/* +* Re-get the p2m size. Should not have changed as a consequence of +* populate physmap. +*/ + rc =xc_get_paging_mempool_size (xch, domid, &size_bytes); +if ( rc ) +return fail (" Fail: get mempool size: %d - %s\n" , + errno,strerror (errno)); + +if ( old_size_bytes != size_bytes ) +return fail (" Fail: mempool size changed %" PRIu64" => %" PRIu64"\n" , + old_size_bytes, size_bytes); + + + +printf ("Test bad set size\n" ); + +/* +* Check that setting a non-page size results in failure. +*/ + rc =xc_set_paging_mempool_size (xch, domid, size_bytes +1 );
Quoted 8 lines
tools/tests/paging-mempool/test-paging-mempool.c line 114
Quoted 6 lines
tools/tests/paging-mempool/test-paging-mempool.c new file +6
-0+if ( rc !=-1 || errno !=EINVAL ) +return fail (" Fail: Bad set size: expected -1/EINVAL, got %d/%d - %s\n" , + rc, errno,strerror (errno)); + + +printf ("Test very large set size\n" );Maybe drop "very", as 64M isn't all that much (and would, in particular, not expose any 32-bit truncation issues)?
Hmm yeah. That was rather stale.
I've switched to "Test set continuation\n" seeing as that's the purpose of the test. 64M is an internal detail which ought to be small enough to work reliably everywhere, but large enough to cause a continuation.
~Andrew
Re: [PATCH 2/4] tools/tests: Unit test for paging mempool size
In reply to Andrew CooperQuoted 10 lines · Andrew Cooper
Exercise some basic functionality of the new
xc_{get,set}_paging_mempool_size() hypercalls.This passes on x86, but fails currently on ARM. ARM will be fixed up in future patches.
This is part of XSA-409 / CVE-2022-33747.
Signed-off-by Andrew Cooper <[email protected]>Release-acked-by Henry Wang <[email protected]>
Thanks,
Signature · 1 line
Anthony PERARD
PATCH 3/4 xen/arm, libxl: Revert XEN_DOMCTL_shadow_op; use p2m mempool hypercalls
In reply to Andrew CooperThis reverts most of commit cf2a68d2ffbc3ce95e01449d46180bddb10d24a0, and bits of cbea5a1149ca7fd4b7cdbfa3ec2e4f109b601ff7.
First of all, with ARM borrowing x86's implementation, the logic to set the pool size should have been common, not duplicated. Introduce libxl__domain_set_p2m_pool_size() as a shared implementation, and use it from the ARM and x86 paths. It is left as an exercise to the reader to judge how libxl/xl can reasonably function without the ability to query the pool size...
Remove ARM's p2m_domctl() infrastructure now the functioanlity has been replaced with a working and unit tested interface.
This is part of XSA-409 / CVE-2022-33747.
---
CC: Jan Beulich <[email protected]>
CC: Roger Pau Monné <[email protected]>
CC: Wei Liu <[email protected]>
CC: Stefano Stabellini <[email protected]>
CC: Julien Grall <[email protected]>
CC: Volodymyr Babchuk <[email protected]>
CC: Bertrand Marquis <[email protected]>
CC: Henry Wang <[email protected]>
CC: Anthony PERARD <[email protected]>
v2:
* s/p2m/paging/
* Fix get/set typo in libxl__domain_set_p2m_pool_size()
--- tools/libs/light/libxl_arm.c | 14 +---------- tools/libs/light/libxl_dom.c | 19 ++++++++++++++ tools/libs/light/libxl_internal.h | 3 +++ tools/libs/light/libxl_x86.c | 15 ++--------- xen/arch/arm/domctl.c | 53 --------------------------------------- xen/arch/arm/include/asm/p2m.h | 1 - xen/arch/arm/p2m.c | 8 ------ 7 files changed, 25 insertions(+), 88 deletions(-)
tools/libs/light/libxl_arm.c +1 -13
@@ -209,19 +209,7 @@ int libxl__arch_domain_create(libxl__gc *gc,libxl__domain_build_state *state, uint32_t domid) { -libxl_ctx *ctx =libxl__gc_owner (gc); -unsigned int shadow_mb =DIV_ROUNDUP (d_config->b_info.shadow_memkb,1024 ); - -int r =xc_shadow_control (ctx->xch, domid, -XEN_DOMCTL_SHADOW_OP_SET_ALLOCATION , - &shadow_mb,0 ); -if (r) { -LOGED (ERROR , domid, -"Failed to set %u MiB shadow allocation" , shadow_mb); -return ERROR_FAIL ; - } - -return 0 ; + return libxl__domain_set_p2m_pool_size(gc, d_config, domid); } int libxl__arch_extra_memory(libxl__gc *gc,
tools/libs/light/libxl_dom.c +19 -0
@@ -1448,6 +1448,25 @@ int libxl_userdata_unlink(libxl_ctx *ctx, uint32_t domid,return rc; } +int libxl__domain_set_p2m_pool_size ( +libxl__gc *gc,libxl_domain_config *d_config,uint32_t domid) +{ +libxl_ctx *ctx =libxl__gc_owner (gc); +uint64_t shadow_mem; + + shadow_mem = d_config->b_info.shadow_memkb; + shadow_mem <<=10 ; + +int r =xc_set_paging_mempool_size (ctx->xch, domid, shadow_mem); +if (r) { +LOGED (ERROR , domid, +"Failed to set paging mempool size to %" PRIu64"kB" , shadow_mem); +return ERROR_FAIL ; + } + +return 0 ; +} +/* * Local variables: * mode: C
tools/libs/light/libxl_internal.h +3 -0
@@ -4864,6 +4864,9 @@ int libxl__is_domid_recent(libxl__gc *gc, uint32_t domid, bool *recent);/* os-specific implementation of setresuid() */ int libxl__setresuid (uid_t ruid,uid_t euid,uid_t suid); +_hidden int libxl__domain_set_p2m_pool_size( +libxl__gc *gc,libxl_domain_config *d_config,uint32_t domid); +#endif /*
tools/libs/light/libxl_x86.c +2 -13
@@ -538,20 +538,9 @@ int libxl__arch_domain_create(libxl__gc *gc,xc_domain_set_time_offset (ctx->xch, domid, rtc_timeoffset);if (d_config->b_info.type !=LIBXL_DOMAIN_TYPE_PV ) { -unsigned int shadow_mb =DIV_ROUNDUP (d_config->b_info.shadow_memkb, -1024 ); -int r =xc_shadow_control (ctx->xch, domid, -XEN_DOMCTL_SHADOW_OP_SET_ALLOCATION , - &shadow_mb,0 ); - -if (r) { -LOGED (ERROR , domid, -"Failed to set %u MiB %s allocation" , - shadow_mb, -libxl_defbool_val (d_config->c_info.hap) ?"HAP" :"shadow" ); - ret =ERROR_FAIL ; + ret =libxl__domain_set_p2m_pool_size (gc, d_config, domid); +if (ret) goto out; - } } if (d_config->c_info.type == LIBXL_DOMAIN_TYPE_PV &&
xen/arch/arm/domctl.c +0 -53
64 lines not on this page · raw message
xen/arch/arm/include/asm/p2m.h +0 -1
@@ -222,7 +222,6 @@ void p2m_restore_state(struct vcpu *n);/* Print debugging/statistial info about a domain's p2m */ void p2m_dump_info (struct domain *d); -unsigned int p2m_get_allocation (struct domain *d);int p2m_set_allocation (struct domain *d,unsigned long pages,bool *preempted);int p2m_teardown_allocation (struct domain *d);
xen/arch/arm/p2m.c +0 -8
@@ -92,14 +92,6 @@ static void p2m_free_page(struct domain *d, struct page_info *pg)spin_unlock (&d->arch.paging.lock); } -/* Return the size of the pool, rounded up to the nearest MB */ -unsigned int p2m_get_allocation (struct domain *d) -{ -unsigned long nr_pages =ACCESS_ONCE (d->arch.paging.p2m_total_pages); - -return ROUNDUP (nr_pages,1 << (20 -PAGE_SHIFT )) >> (20 -PAGE_SHIFT ); -} -/* Return the size of the pool, in bytes. */ int arch_get_paging_mempool_size (struct domain *d,uint64_t *size) {
Signature · 1 line
2.11.0
Re: [PATCH 3/4] xen/arm, libxl: Revert XEN_DOMCTL_shadow_op; use p2m mempool hypercalls
Quoted 9 lines · Andrew Cooper
This reverts most of commit
cf2a68d2ffbc3ce95e01449d46180bddb10d24a0, and bits ofcbea5a1149ca7fd4b7cdbfa3ec2e4f109b601ff7.First of all, with ARM borrowing x86's implementation, the logic to set the pool size should have been common, not duplicated. Introduce
libxl__domain_set_p2m_pool_size()as a shared implementation, and use it from the ARM and x86 paths. It is left as an exercise to the reader to judge how libxl/xl can reasonably function without the ability to query the pool size...
Quoted 7 lines
Remove ARM's
p2m_domctl()infrastructure now the functioanlity has been replaced with a working and unit tested interface.This is part of XSA-409 / CVE-2022-33747.
Signed-off-by Andrew Cooper <[email protected]>Release-acked-by Henry Wang <[email protected]>
Quoted 223 lines
---
CC: Jan Beulich <[email protected]>
CC: Roger Pau Monné <[email protected]>
CC: Wei Liu <[email protected]>
CC: Stefano Stabellini <[email protected]>
CC: Julien Grall <[email protected]>
CC: Volodymyr Babchuk <[email protected]>
CC: Bertrand Marquis <[email protected]>
CC: Henry Wang <[email protected]>
CC: Anthony PERARD <[email protected]>v2:
* s/p2m/paging/
* Fix get/set typo inlibxl__domain_set_p2m_pool_size()--- tools/libs/light/libxl_arm.c | 14 +---------- tools/libs/light/libxl_dom.c | 19 ++++++++++++++ tools/libs/light/libxl_internal.h | 3 +++ tools/libs/light/libxl_x86.c | 15 ++--------- xen/arch/arm/domctl.c | 53 --------------------------------------- xen/arch/arm/include/asm/p2m.h | 1 - xen/arch/arm/p2m.c | 8 ------ 7 files changed, 25 insertions(+), 88 deletions(-)tools/libs/light/libxl_arm.c +1
-13@@ -209,19 +209,7 @@ int libxl__arch_domain_create(libxl__gc *gc,libxl__domain_build_state *state, uint32_t domid) { -libxl_ctx *ctx =libxl__gc_owner (gc); -unsigned int shadow_mb =DIV_ROUNDUP (d_config->b_info.shadow_memkb,1024 ); - -int r =xc_shadow_control (ctx->xch, domid, -XEN_DOMCTL_SHADOW_OP_SET_ALLOCATION , - &shadow_mb,0 ); -if (r) { -LOGED (ERROR , domid, -"Failed to set %u MiB shadow allocation" , shadow_mb); -return ERROR_FAIL ; - } - -return 0 ; + return libxl__domain_set_p2m_pool_size(gc, d_config, domid); } int libxl__arch_extra_memory(libxl__gc *gc,tools/libs/light/libxl_dom.c +19
-0@@ -1448,6 +1448,25 @@ int libxl_userdata_unlink(libxl_ctx *ctx, uint32_t domid,return rc; } +int libxl__domain_set_p2m_pool_size ( +libxl__gc *gc,libxl_domain_config *d_config,uint32_t domid) +{ +libxl_ctx *ctx =libxl__gc_owner (gc); +uint64_t shadow_mem; + + shadow_mem = d_config->b_info.shadow_memkb; + shadow_mem <<=10 ; + +int r =xc_set_paging_mempool_size (ctx->xch, domid, shadow_mem); +if (r) { +LOGED (ERROR , domid, +"Failed to set paging mempool size to %" PRIu64"kB" , shadow_mem); +return ERROR_FAIL ; + } + +return 0 ; +} +/* * Local variables: * mode: C tools/libs/light/libxl_internal.h +3
-0@@ -4864,6 +4864,9 @@ int libxl__is_domid_recent(libxl__gc *gc, uint32_t domid, bool *recent);/* os-specific implementation of setresuid() */ int libxl__setresuid (uid_t ruid,uid_t euid,uid_t suid); +_hidden int libxl__domain_set_p2m_pool_size( +libxl__gc *gc,libxl_domain_config *d_config,uint32_t domid); +#endif /* tools/libs/light/libxl_x86.c +2
-13@@ -538,20 +538,9 @@ int libxl__arch_domain_create(libxl__gc *gc,xc_domain_set_time_offset (ctx->xch, domid, rtc_timeoffset);if (d_config->b_info.type !=LIBXL_DOMAIN_TYPE_PV ) { -unsigned int shadow_mb =DIV_ROUNDUP (d_config->b_info.shadow_memkb, -1024 ); -int r =xc_shadow_control (ctx->xch, domid, -XEN_DOMCTL_SHADOW_OP_SET_ALLOCATION , - &shadow_mb,0 ); - -if (r) { -LOGED (ERROR , domid, -"Failed to set %u MiB %s allocation" , - shadow_mb, -libxl_defbool_val (d_config->c_info.hap) ?"HAP" :"shadow" ); - ret =ERROR_FAIL ; + ret =libxl__domain_set_p2m_pool_size (gc, d_config, domid); +if (ret) goto out; - } } if (d_config->c_info.type == LIBXL_DOMAIN_TYPE_PV &&xen/arch/arm/domctl.c +0
-53@@ -47,64 +47,11 @@ static int handle_vuart_init(struct domain *d,return rc; } -static long p2m_domctl (struct domain *d,struct xen_domctl_shadow_op *sc, -XEN_GUEST_HANDLE_PARAM (xen_domctl_t) u_domctl) -{ -long rc; -bool preempted = false; - -if (unlikely (d == current->domain) ) - { -printk (XENLOG_ERR "Tried to do a p2m domctl op on itself.\n" ); -return -EINVAL ; - } - -if (unlikely (d->is_dying) ) - { -printk (XENLOG_ERR "Tried to do a p2m domctl op on dying domain %u\n" , - d->domain_id); -return -EINVAL ; - } - -switch ( sc->op ) - { -case XEN_DOMCTL_SHADOW_OP_SET_ALLOCATION : - { -/* Allow and handle preemption */ -spin_lock (&d->arch.paging.lock); - rc =p2m_set_allocation (d, sc->mb << (20 -PAGE_SHIFT ), &preempted); -spin_unlock (&d->arch.paging.lock); - -if ( preempted ) -/* Not finished. Set up to re-run the call. */ - rc =hypercall_create_continuation (__HYPERVISOR_domctl,"h" , - u_domctl); -else -/* Finished. Return the new allocation. */ - sc->mb =p2m_get_allocation (d); - -return rc; - } -case XEN_DOMCTL_SHADOW_OP_GET_ALLOCATION : - { - sc->mb =p2m_get_allocation (d); -return 0 ; - } -default : - { -printk (XENLOG_ERR "Bad p2m domctl op %u\n" , sc->op); -return -EINVAL ; - } - } -} -long arch_do_domctl (struct xen_domctl *domctl,struct domain *d,XEN_GUEST_HANDLE_PARAM (xen_domctl_t) u_domctl) {switch ( domctl->cmd ) { -case XEN_DOMCTL_shadow_op: -return p2m_domctl (d, &domctl->u.shadow_op, u_domctl);case XEN_DOMCTL_cacheflush: {gfn_t s =_gfn (domctl->u.cacheflush.start_pfn);xen/arch/arm/include/asm/p2m.h +0
-1@@ -222,7 +222,6 @@ void p2m_restore_state(struct vcpu *n);/* Print debugging/statistial info about a domain's p2m */ void p2m_dump_info (struct domain *d); -unsigned int p2m_get_allocation (struct domain *d);int p2m_set_allocation (struct domain *d,unsigned long pages,bool *preempted);int p2m_teardown_allocation (struct domain *d);xen/arch/arm/p2m.c +0
-8@@ -92,14 +92,6 @@ static void p2m_free_page(struct domain *d, struct page_info *pg)spin_unlock (&d->arch.paging.lock); } -/* Return the size of the pool, rounded up to the nearest MB */ -unsigned int p2m_get_allocation (struct domain *d) -{ -unsigned long nr_pages =ACCESS_ONCE (d->arch.paging.p2m_total_pages); - -return ROUNDUP (nr_pages,1 << (20 -PAGE_SHIFT )) >> (20 -PAGE_SHIFT ); -} -/* Return the size of the pool, in bytes. */ int arch_get_paging_mempool_size (struct domain *d,uint64_t *size) {Signature · 1 line
2.11.0
Re: [PATCH 3/4] xen/arm, libxl: Revert XEN_DOMCTL_shadow_op; use p2m mempool hypercalls
In reply to Andrew CooperQuoted 8 lines · Andrew Cooper
tools/libs/light/libxl_dom.c +0
-0@@ -1448,6 +1448,25 @@ int libxl_userdata_unlink(libxl_ctx *ctx, uint32_t domid,return rc; }
tools/libs/light/libxl_dom.c int libxl_userdata_unlink(libxl_ctx *ctx, uint32_t domid, · line 1460
Quoted 10 lines
tools/libs/light/libxl_dom.c +10
-0+int libxl__domain_set_p2m_pool_size ( +libxl__gc *gc,libxl_domain_config *d_config,uint32_t domid) +{ +libxl_ctx *ctx =libxl__gc_owner (gc); +uint64_t shadow_mem; + + shadow_mem = d_config->b_info.shadow_memkb; + shadow_mem <<=10 ; + +int r =xc_set_paging_mempool_size (ctx->xch, domid, shadow_mem);
Nit: 'ctx' could be replace by using the macro 'CTX', without having to declare a local ctx variable.
With or without this changed, patch looks fine:
Thanks,
Signature · 1 line
Anthony PERARD
PATCH 4/4 xen/arm: Correct the p2m pool size calculations
In reply to Andrew CooperAllocating or freeing p2m pages doesn't alter the size of the mempool; only the split between free and used pages.
Right now, the hypercalls operate on the free subset of the pool, meaning that XEN_DOMCTL_get_paging_mempool_size varies with time as the guest shuffles its physmap, and XEN_DOMCTL_set_paging_mempool_size ignores the used subset of the pool and lets the guest grow unbounded.
This fixes test-pagign-mempool on ARM so that the behaviour matches x86.
This is part of XSA-409 / CVE-2022-33747.
---
CC: Jan Beulich <[email protected]>
CC: Roger Pau Monné <[email protected]>
CC: Wei Liu <[email protected]>
CC: Stefano Stabellini <[email protected]>
CC: Julien Grall <[email protected]>
CC: Volodymyr Babchuk <[email protected]>
CC: Bertrand Marquis <[email protected]>
CC: Henry Wang <[email protected]>
CC: Anthony PERARD <[email protected]>
--- xen/arch/arm/p2m.c | 4 ---- 1 file changed, 4 deletions(-)
xen/arch/arm/p2m.c +0 -4
@@ -72,7 +72,6 @@ static struct page_info *p2m_alloc_page(struct domain *d)spin_unlock (&d->arch.paging.lock);return NULL ; } - d->arch.paging.p2m_total_pages--; }spin_unlock (&d->arch.paging.lock);
@@ -85,10 +84,7 @@ static void p2m_free_page(struct domain *d, struct page_info *pg)if (is_hardware_domain (d) )free_domheap_page (pg);else - { - d->arch.paging.p2m_total_pages++;page_list_add_tail (pg, &d->arch.paging.p2m_freelist); - }spin_unlock (&d->arch.paging.lock); }
Signature · 1 line
2.11.0