Fix truncation of various XENVER_* subops
cover Fix truncation of various XENVER_* subops
See patch 4 for details of the problem. Other patches fix other errors found while investigating.
This is only the hypervisor side of the change for now, because I want agreement before starting to untangle the mess which is libxc's helpers for this.
Also a disaster is Linux's sysfs handling for these. In several places it makes a heap allocation for a pointer (or two) sized object.
Andrew Cooper (4):
public/version: Change xen_feature_info to have a fixed size xen/version: Drop compat/kernel.c
xen/version: Drop bogus return values for XENVER_platform_parameters xen/version: Introduce non-truncating XENVER_* subops
xen/common/Makefile | 2 +- xen/common/compat/kernel.c | 53 --------------------- xen/common/kernel.c | 108 ++++++++++++++++++++++++++++++++++++++----- xen/include/hypercall-defs.c | 2 +- xen/include/public/version.h | 78 +++++++++++++++++++++++++++++-- xen/include/xlat.lst | 4 ++ xen/xsm/flask/hooks.c | 4 ++ 7 files changed, 181 insertions(+), 70 deletions(-) delete mode 100644 xen/common/compat/kernel.c
Signature · 1 line
2.11.0
PATCH 1/4 public/version: Change xen_feature_info to have a fixed size
This is technically an ABI change, but Xen doesn't operate in any environment where "unsigned int" is differnet to uint32_t, so switch to the explicit form. This avoids the need to derive (identical) compat logic for handling the subop.
---
CC: George Dunlap <[email protected]>
CC: Jan Beulich <[email protected]>
CC: Stefano Stabellini <[email protected]>
CC: Wei Liu <[email protected]>
CC: Julien Grall <[email protected]>
--- xen/include/public/version.h | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
xen/include/public/version.h +1 -1
@@ -50,7 +50,7 @@ typedef struct xen_platform_parameters xen_platform_parameters_t;#define XENVER_get_features 6struct xen_feature_info { -unsigned int submap_idx;/* IN: which 32-bit submap to return */ +uint32_t submap_idx;/* IN: which 32-bit submap to return */ uint32_t submap;/* OUT: 32-bit submap */ };typedef struct xen_feature_info xen_feature_info_t ;
Signature · 1 line
2.11.0
Re: [PATCH 1/4] public/version: Change xen_feature_info to have a fixed size
Hi Andrew,
Quoted 2 lines · Andrew Cooper
This is technically an ABI change, but Xen doesn't operate in any environment where "unsigned int" is differnet to
uint32_t, so switch to the explicit form.
typo: s/differnet/different/
Quoted 4 lines
This avoids the need to derive (identical) compat logic for handling the subop.
Signed-off-by Andrew Cooper <[email protected]>
Cheers,
Quoted 23 lines
---
CC: George Dunlap <[email protected]>
CC: Jan Beulich <[email protected]>
CC: Stefano Stabellini <[email protected]>
CC: Wei Liu <[email protected]>
CC: Julien Grall <[email protected]>--- xen/include/public/version.h | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)xen/include/public/version.h +1
-1@@ -50,7 +50,7 @@ typedef struct xen_platform_parameters xen_platform_parameters_t;#define XENVER_get_features 6struct xen_feature_info { -unsigned int submap_idx;/* IN: which 32-bit submap to return */ +uint32_t submap_idx;/* IN: which 32-bit submap to return */ uint32_t submap;/* OUT: 32-bit submap */ };typedef struct xen_feature_info xen_feature_info_t ;
Signature · 1 line
Julien Grall
PATCH 2/4 xen/version: Drop compat/kernel.c
In reply to Andrew Cooperkernel.c is mostly in an #ifndef COMPAT guard, because compat/kernel.c reincludes kernel.c to recompile xen_version() in a compat form.
However, the xen_version hypercall is almost guest-ABI-agnostic; only XENVER_platform_parameters has a compat split. Handle this locally, and do away with the reinclude entirely.
In particular, this removed the final instances of obfucation via the DO() macro.
No functional change. Also saves 2k of of .text in the x86 build.
---
CC: George Dunlap <[email protected]>
CC: Jan Beulich <[email protected]>
CC: Stefano Stabellini <[email protected]>
CC: Wei Liu <[email protected]>
CC: Julien Grall <[email protected]>
--- xen/common/Makefile | 2 +- xen/common/compat/kernel.c | 53 -------------------------------------------- xen/common/kernel.c | 43 ++++++++++++++++++++++------------- xen/include/hypercall-defs.c | 2 +- xen/include/xlat.lst | 3 +++ 5 files changed, 33 insertions(+), 70 deletions(-) delete mode 100644 xen/common/compat/kernel.c
xen/common/Makefile +1 -1
@@ -59,7 +59,7 @@ obj-y += xmalloc_tlsf.o obj-bin-$(CONFIG_X86) += $(foreach n,decompress bunzip2 unxz unlzma lzo unlzo unlz4 unzstd earlycpio,$(n).init.o) -obj-$(CONFIG_COMPAT) += $(addprefix compat/,domain.o kernel.o memory.o multicall.o xlat.o) +obj-$(CONFIG_COMPAT) += $(addprefix compat/,domain.o memory.o multicall.o xlat.o) ifneq ($(CONFIG_PV_SHIM_EXCLUSIVE),y) obj-y += domctl.o
xen/common/compat/kernel.c deleted +0 -53
53 lines not on this page · raw message
xen/common/kernel.c +28 -15
62 lines not on this page · raw message
xen/include/hypercall-defs.c +1 -1
@@ -245,7 +245,7 @@ multicall compat:2 do:2 compat do do update_va_mapping compat do - - - set_timer_op compat do compat do - event_channel_op_compat do do - - dep -xen_version compat do compat do do +xen_version do do do do do console_io do do do do do physdev_op_compat compat do - - dep #if defined(CONFIG_GRANT_TABLE)
xen/include/xlat.lst +3 -0
@@ -169,6 +169,9 @@ ! vcpu_runstate_info vcpu.h ? vcpu_set_periodic_timer vcpu.h ! vcpu_set_singleshot_timer vcpu.h +? compile_info version.h +? feature_info version.h +? build_id version.h ? xenoprof_init xenoprof.h ? xenoprof_passive xenoprof.h ? flask_access xsm/flask_op.h
Signature · 1 line
2.11.0
Re: [PATCH 2/4] xen/version: Drop compat/kernel.c
Quoted 6 lines · Andrew Cooper
kernel.cis mostly in an #ifndef COMPAT guard, becausecompat/kernel.creincludeskernel.cto recompilexen_version()in a compat form.However, the
xen_versionhypercall is almost guest-ABI-agnostic; onlyXENVER_platform_parametershas a compat split. Handle this locally, and do away with the reinclude entirely.
And we henceforth mean to not introduce any interface structures here which would require translation (or we're willing to accept the clutter of handling those "locally" as well). Fine with me, just wanting to mention it.
Quoted 18 lines
xen/common/compat/kernel.c deleted +0
-15@@ -1,53 +0,0 @@ -/****************************************************************************** -* kernel.c -*/ - -EMIT_FILE ; - -#include <xen/init.h> -#include <xen/lib.h> -#include <xen/errno.h> -#include <xen/version.h> -#include <xen/sched.h> -#include <xen/guest_access.h> -#include <asm/current.h> -#include <compat/xen.h> -#include <compat/version.h>
xen/common/compat/kernel.c line 25
Quoted 10 lines
xen/common/compat/kernel.c deleted +0
-10- -extern xen_commandline_t saved_cmdline; - -#define xen_extraversion compat_extraversion -#define xen_extraversion_t compat_extraversion_t - -#define xen_compile_info compat_compile_info -#define xen_compile_info_t compat_compile_info_t - -CHECK_TYPE (capabilities_info);
This and ...
xen/common/compat/kernel.c line 38
Quoted 12 lines
xen/common/compat/kernel.c deleted +0
-12-#define xen_platform_parameters compat_platform_parameters -#define xen_platform_parameters_t compat_platform_parameters_t -#undef HYPERVISOR_VIRT_START -#define HYPERVISOR_VIRT_START HYPERVISOR_COMPAT_VIRT_START(current->domain) - -#define xen_changeset_info compat_changeset_info -#define xen_changeset_info_t compat_changeset_info_t - -#define xen_feature_info compat_feature_info -#define xen_feature_info_t compat_feature_info_t - -CHECK_TYPE (domain_handle);
... this go away without any replacement. Considering they're both char[], that's probably fine, but could do with mentioning in the description.
xen/common/compat/kernel.c DO(xen_version)(int cmd, XEN_GUEST_HANDLE_PARAM(void) arg) · line 525
Quoted 7 lines
xen/common/compat/kernel.c deleted +0
-3@@ -520,12 +518,27 @@ DO(xen_version)(int cmd, XEN_GUEST_HANDLE_PARAM(void) arg)case XENVER_platform_parameters: { -xen_platform_parameters_t params = { - .virt_start =HYPERVISOR_VIRT_START - };
With this gone the oddly (but intentionally) placed braces can then also go away.
Preferably with these minor adjustments
Jan
Re: [PATCH 2/4] xen/version: Drop compat/kernel.c
Quoted 11 lines · Jan Beulich
Quoted 6 lines · Andrew Cooper
kernel.cis mostly in an #ifndef COMPAT guard, becausecompat/kernel.creincludeskernel.cto recompilexen_version()in a compat form.However, the
xen_versionhypercall is almost guest-ABI-agnostic; onlyXENVER_platform_parametershas a compat split. Handle this locally, and do away with the reinclude entirely.And we henceforth mean to not introduce any interface structures here which would require translation (or we're willing to accept the clutter of handling those "locally" as well). Fine with me, just wanting to mention it.
Sure - I'll put a note in the commit message.
In general, we don't want guest-variant interfaces.
Quoted 31 lines
Quoted 19 lines
xen/common/compat/kernel.c deleted +0
-15@@ -1,53 +0,0 @@ -/****************************************************************************** -* kernel.c -*/ - -EMIT_FILE ; - -#include <xen/init.h> -#include <xen/lib.h> -#include <xen/errno.h> -#include <xen/version.h> -#include <xen/sched.h> -#include <xen/guest_access.h> -#include <asm/current.h> -#include <compat/xen.h> -#include <compat/version.h> xen/common/compat/kernel.c line 25
Quoted 10 lines
xen/common/compat/kernel.c deleted +0
-10- -extern xen_commandline_t saved_cmdline; - -#define xen_extraversion compat_extraversion -#define xen_extraversion_t compat_extraversion_t - -#define xen_compile_info compat_compile_info -#define xen_compile_info_t compat_compile_info_t - -CHECK_TYPE (capabilities_info);This and ...
Quoted 15 lines
xen/common/compat/kernel.c line 38
Quoted 12 lines
xen/common/compat/kernel.c deleted +0
-12-#define xen_platform_parameters compat_platform_parameters -#define xen_platform_parameters_t compat_platform_parameters_t -#undef HYPERVISOR_VIRT_START -#define HYPERVISOR_VIRT_START HYPERVISOR_COMPAT_VIRT_START(current->domain) - -#define xen_changeset_info compat_changeset_info -#define xen_changeset_info_t compat_changeset_info_t - -#define xen_feature_info compat_feature_info -#define xen_feature_info_t compat_feature_info_t - -CHECK_TYPE (domain_handle);... this go away without any replacement. Considering they're both char[], that's probably fine, but could do with mentioning in the description.
I did actually mean to ask about these two, because they're incomplete already.
Why do we CHECK_TYPE(capabilities_info) but define identity aliases for compat_extraversion (amongst others) ?
Is there even a point for having a compat alias of a char array?
I'm tempted to just drop them. I don't think the check does anything useful for us.
Quoted 9 lines
xen/common/compat/kernel.c DO(xen_version)(int cmd, XEN_GUEST_HANDLE_PARAM(void) arg) · line 525
Quoted 7 lines
xen/common/compat/kernel.c deleted +0
-3@@ -520,12 +518,27 @@ DO(xen_version)(int cmd, XEN_GUEST_HANDLE_PARAM(void) arg)case XENVER_platform_parameters: { -xen_platform_parameters_t params = { - .virt_start =HYPERVISOR_VIRT_START - };With this gone the oddly (but intentionally) placed braces can then also go away.
In light of how patch 3 ended up, I was considering pulling curr out into a variable.
Quoted 2 lines
Preferably with these minor adjustments
Reviewed-by Jan Beulich <[email protected]>
Thanks,
~Andrew
Re: [PATCH 2/4] xen/version: Drop compat/kernel.c
Quoted 48 lines · Andrew Cooper
Quoted 31 lines · Jan Beulich
Quoted 18 lines · Andrew Cooper
xen/common/compat/kernel.c deleted +0
-15@@ -1,53 +0,0 @@ -/****************************************************************************** -* kernel.c -*/ - -EMIT_FILE ; - -#include <xen/init.h> -#include <xen/lib.h> -#include <xen/errno.h> -#include <xen/version.h> -#include <xen/sched.h> -#include <xen/guest_access.h> -#include <asm/current.h> -#include <compat/xen.h> -#include <compat/version.h> xen/common/compat/kernel.c line 25
Quoted 10 lines
xen/common/compat/kernel.c deleted +0
-10- -extern xen_commandline_t saved_cmdline; - -#define xen_extraversion compat_extraversion -#define xen_extraversion_t compat_extraversion_t - -#define xen_compile_info compat_compile_info -#define xen_compile_info_t compat_compile_info_t - -CHECK_TYPE (capabilities_info);This and ...
Quoted 15 lines
xen/common/compat/kernel.c line 38
Quoted 12 lines
xen/common/compat/kernel.c deleted +0
-12-#define xen_platform_parameters compat_platform_parameters -#define xen_platform_parameters_t compat_platform_parameters_t -#undef HYPERVISOR_VIRT_START -#define HYPERVISOR_VIRT_START HYPERVISOR_COMPAT_VIRT_START(current->domain) - -#define xen_changeset_info compat_changeset_info -#define xen_changeset_info_t compat_changeset_info_t - -#define xen_feature_info compat_feature_info -#define xen_feature_info_t compat_feature_info_t - -CHECK_TYPE (domain_handle);... this go away without any replacement. Considering they're both char[], that's probably fine, but could do with mentioning in the description.
Quoted 7 lines
I did actually mean to ask about these two, because they're incomplete already.
Why do we
CHECK_TYPE(capabilities_info)but define identity aliases forcompat_extraversion(amongst others) ?Is there even a point for having a compat alias of a char array?
To be quite honest, for capabilities_info I don't recall why it wasn't aliased. For domain_handle I think the reason was that it's declared outside of this header, and it could conceivably be a more UUID-like struct.
Quoted 2 lines
I'm tempted to just drop them. I don't think the check does anything useful for us.
As said, I'm pretty much okay with this, but would like you to mention in the description that this is an intentional change.
Jan
Re: [PATCH 2/4] xen/version: Drop compat/kernel.c
In reply to Andrew Cooperxen/include/xlat.lst line 174
Quoted 9 lines · Andrew Cooper
xen/include/xlat.lst +3
-0@@ -169,6 +169,9 @@ ! vcpu_runstate_info vcpu.h ? vcpu_set_periodic_timer vcpu.h ! vcpu_set_singleshot_timer vcpu.h +? compile_info version.h +? feature_info version.h +? build_id version.h
Oh, btw, another minor request: Can you please sort these by name (as secondary criteria after the name of the containing header)?
Thanks, Jan
PATCH 3/4 xen/version: Drop bogus return values for XENVER_platform_parameters
In reply to Andrew CooperA split in virtual address space is only applicable for x86 PV guests. Furthermore, the information returned for x86 64bit PV guests is wrong.
Explain the problem in version.h, stating the other information that PV guests need to know.
For 64bit PV guests, and all non-x86-PV guests, return 0, which is strictly less wrong than the values currently returned.
---
CC: George Dunlap <[email protected]>
CC: Jan Beulich <[email protected]>
CC: Stefano Stabellini <[email protected]>
CC: Wei Liu <[email protected]>
CC: Julien Grall <[email protected]>
--- xen/common/kernel.c | 6 ++++-- xen/include/public/version.h | 20 ++++++++++++++++++++ 2 files changed, 24 insertions(+), 2 deletions(-)
xen/common/kernel.c +4 -2
@@ -522,7 +522,9 @@ long do_xen_version(int cmd, XEN_GUEST_HANDLE_PARAM(void) arg)if ( current->hcall_compat ) {compat_platform_parameters_t params = { - .virt_start =HYPERVISOR_COMPAT_VIRT_START (current->domain), + .virt_start =is_pv_vcpu (current) + ?HYPERVISOR_COMPAT_VIRT_START (current->domain) + :0 , };if (copy_to_guest (arg, ¶ms,1 ) )
@@ -532,7 +534,7 @@ long do_xen_version(int cmd, XEN_GUEST_HANDLE_PARAM(void) arg)#endif {xen_platform_parameters_t params = { - .virt_start =HYPERVISOR_VIRT_START , + .virt_start =0 , };if (copy_to_guest (arg, ¶ms,1 ) )
xen/include/public/version.h +20 -0
@@ -42,6 +42,26 @@ typedef char xen_capabilities_info_t[1024];typedef char xen_changeset_info_t [64 ];#define XEN_CHANGESET_INFO_LEN (sizeof(xen_changeset_info_t)) +/* +* This API is problematic. +* +* It is only applicable to guests which share pagetables with Xen (x86 PV +* guests), and is supposed to identify the virtual address split between +* guest kernel and Xen. +* +* For 32bit PV guests, it mostly does this, but the caller needs to know that +* Xen lives between the split and 4G. +* +* For 64bit PV guests, Xen lives at the bottom of the upper canonical range. +* This previously returned the start of the upper canonical range (which is +* the userspace/Xen split), not the Xen/kernel split (which is 8TB further +* on). This now returns 0 because the old number wasn't correct, and +* changing it to anything else would be even worse. +* +* For all guest types using hardware virt extentions, Xen is not mapped into +* the guest kernel virtual address space. This now return 0, where it +* previously returned unrelated data. +*/ #define XENVER_platform_parameters 5struct xen_platform_parameters {xen_ulong_t virt_start;
Signature · 1 line
2.11.0
Re: [PATCH 3/4] xen/version: Drop bogus return values for XENVER_platform_parameters
Quoted 8 lines · Andrew Cooper
A split in virtual address space is only applicable for x86 PV guests. Furthermore, the information returned for x86 64bit PV guests is wrong.
Explain the problem in
version.h, stating the other information that PV guests need to know.For 64bit PV guests, and all non-x86-PV guests, return 0, which is strictly less wrong than the values currently returned.
I disagree for the 64-bit part of this. Seeing Linux'es exposure of the value in sysfs I even wonder whether we can change this like you do for HVM. Who knows what is being inferred from the value, and by whom.
Quoted 11 lines
xen/include/public/version.h +5
-0@@ -42,6 +42,26 @@ typedef char xen_capabilities_info_t[1024];typedef char xen_changeset_info_t [64 ];#define XEN_CHANGESET_INFO_LEN (sizeof(xen_changeset_info_t)) +/* +* This API is problematic. +* +* It is only applicable to guests which share pagetables with Xen (x86 PV +* guests), and is supposed to identify the virtual address split between
xen/include/public/version.h typedef char xen_capabilities_info_t[1024]; · line 59
Quoted 10 lines
xen/include/public/version.h +10
-0+* guest kernel and Xen. +* +* For 32bit PV guests, it mostly does this, but the caller needs to know that +* Xen lives between the split and 4G. +* +* For 64bit PV guests, Xen lives at the bottom of the upper canonical range. +* This previously returned the start of the upper canonical range (which is +* the userspace/Xen split), not the Xen/kernel split (which is 8TB further +* on). This now returns 0 because the old number wasn't correct, and +* changing it to anything else would be even worse.
Whether the guest runs user mode code in the low or high half (or in yet another way of splitting) isn't really dictated by the PV ABI, is it? So whether the value is "wrong" is entirely unclear. Instead ...
xen/include/public/version.h typedef char xen_capabilities_info_t[1024]; · line 67
Quoted 7 lines
xen/include/public/version.h +4
-0+* For all guest types using hardware virt extentions, Xen is not mapped into +* the guest kernel virtual address space. This now return 0, where it +* previously returned unrelated data. +*/ #define XENVER_platform_parameters 5struct xen_platform_parameters {xen_ulong_t virt_start;
... the field name tells me that all that is being conveyed is the virtual address of where the hypervisor area starts.
Jan
Re: [PATCH 3/4] xen/version: Drop bogus return values for XENVER_platform_parameters
Quoted 4 lines · Jan Beulich
Quoted 3 lines · Andrew Cooper
A split in virtual address space is only applicable for x86 PV guests. Furthermore, the information returned for x86 64bit PV guests is wrong.
Quoted 8 lines
Quoted 5 lines
Explain the problem in
version.h, stating the other information that PV guests need to know.For 64bit PV guests, and all non-x86-PV guests, return 0, which is strictly less wrong than the values currently returned.
I disagree for the 64-bit part of this. Seeing Linux'es exposure of the value in sysfs I even wonder whether we can change this like you do for HVM. Who knows what is being inferred from the value, and by whom.
Linux's sysfs ABI isn't relevant to us here. The sysfs ABI says it reports what the hypervisor presents, not that it will be a nonzero number.
Quoted 15 lines
Quoted 15 lines
xen/include/public/version.h +9
-0@@ -42,6 +42,26 @@ typedef char xen_capabilities_info_t[1024];typedef char xen_changeset_info_t [64 ];#define XEN_CHANGESET_INFO_LEN (sizeof(xen_changeset_info_t)) +/* +* This API is problematic. +* +* It is only applicable to guests which share pagetables with Xen (x86 PV +* guests), and is supposed to identify the virtual address split between +* guest kernel and Xen. +* +* For 32bit PV guests, it mostly does this, but the caller needs to know that +* Xen lives between the split and 4G.
Quoted 8 lines
xen/include/public/version.h typedef char xen_capabilities_info_t[1024]; · line 59
Quoted 6 lines
xen/include/public/version.h +6
-0+* +* For 64bit PV guests, Xen lives at the bottom of the upper canonical range. +* This previously returned the start of the upper canonical range (which is +* the userspace/Xen split), not the Xen/kernel split (which is 8TB further +* on). This now returns 0 because the old number wasn't correct, and +* changing it to anything else would be even worse. Whether the guest runs user mode code in the low or high half (or in yet another way of splitting) isn't really dictated by the PV ABI, is it?
No, but given a choice of reporting the thing which is an architectural boundary, or the one which is the actual split between the two adjacent ranges, reporting the architectural boundary is clearly the unhelpful thing.
Quoted 10 lines
So
whether the value is "wrong" is entirely unclear. Instead ...xen/include/public/version.h typedef char xen_capabilities_info_t[1024]; · line 67
Quoted 7 lines
xen/include/public/version.h +4
-0+* For all guest types using hardware virt extentions, Xen is not mapped into +* the guest kernel virtual address space. This now return 0, where it +* previously returned unrelated data. +*/ #define XENVER_platform_parameters 5struct xen_platform_parameters {xen_ulong_t virt_start;
Quoted 2 lines
... the field name tells me that all that is being conveyed is the virtual address of where the hypervisor area starts.
IMO, it doesn't matter what the name of the field is. It dates from the days when 32bit PV was the only type of guest.
32bit PV guests really do have a variable split, so the guest kernel really does need to get this value from Xen.
The split for 64bit PV guests is compile-time constant, hence why 64bit PV kernels don't care.
For compat HVM, it happens to pick up the -1 from:
#ifdef CONFIG_PV32
HYPERVISOR_COMPAT_VIRT_START(d) =
is_pv_domain(d) ? __HYPERVISOR_COMPAT_VIRT_START : ~0u; #endif
in arch_domain_create(), whereas for non-compat HVM, it gets a number in an address space it has no connection to in the slightest. ARM guests end up getting XEN_VIRT_START (== 2M) handed back, but this absolutely an internal detail that guests have no business knowing.
The only reason I'm not issuing an XSA for this is because we don't have any pretence of KASLR in Xen. Pretty much every other kernel gets CVEs for infoleaks like this.
We feasibly could do KASLR in !PV builds, at which point this would qualify for an XSA.
~Andrew
Re: [PATCH 3/4] xen/version: Drop bogus return values for XENVER_platform_parameters
Quoted 8 lines · Andrew Cooper
Quoted 7 lines · Jan Beulich
Quoted 6 lines · Andrew Cooper
A split in virtual address space is only applicable for x86 PV guests. Furthermore, the information returned for x86 64bit PV guests is wrong.
Explain the problem in
version.h, stating the other information that PV guests need to know.
Quoted 8 lines
Quoted 5 lines
Quoted 2 lines
For 64bit PV guests, and all non-x86-PV guests, return 0, which is strictly less wrong than the values currently returned.
I disagree for the 64-bit part of this. Seeing Linux'es exposure of the value in sysfs I even wonder whether we can change this like you do for HVM. Who knows what is being inferred from the value, and by whom.
Linux's sysfs ABI isn't relevant to us here. The sysfs ABI says it reports what the hypervisor presents, not that it will be a nonzero number.
It effectively reports the hypervisor (virtual) base address there. How can we not care if something (kexec would come to mind) would be using it for whatever purpose. And thinking of it, the tool stack has uses, too. Assuming you audited them, did you consider removing dead uses in a prereq patch (and discuss the effects on live ones in the description)?
Quoted 24 lines
Quoted 15 lines
Quoted 15 lines
xen/include/public/version.h +9
-0@@ -42,6 +42,26 @@ typedef char xen_capabilities_info_t[1024];typedef char xen_changeset_info_t [64 ];#define XEN_CHANGESET_INFO_LEN (sizeof(xen_changeset_info_t)) +/* +* This API is problematic. +* +* It is only applicable to guests which share pagetables with Xen (x86 PV +* guests), and is supposed to identify the virtual address split between +* guest kernel and Xen. +* +* For 32bit PV guests, it mostly does this, but the caller needs to know that +* Xen lives between the split and 4G. Quoted 8 lines
xen/include/public/version.h typedef char xen_capabilities_info_t[1024]; · line 59
Quoted 6 lines
xen/include/public/version.h +6
-0+* +* For 64bit PV guests, Xen lives at the bottom of the upper canonical range. +* This previously returned the start of the upper canonical range (which is +* the userspace/Xen split), not the Xen/kernel split (which is 8TB further +* on). This now returns 0 because the old number wasn't correct, and +* changing it to anything else would be even worse. Whether the guest runs user mode code in the low or high half (or in yet another way of splitting) isn't really dictated by the PV ABI, is it?
Quoted 3 lines
No, but given a choice of reporting the thing which is an architectural boundary, or the one which is the actual split between the two adjacent ranges, reporting the architectural boundary is clearly the unhelpful thing.
Hmm. To properly parallel the 32-bit variant, a [start,end] range would need exposing for 64-bit, rather than exposing nothing. Not the least because ...
Quoted 13 lines
Quoted 10 lines
So
whether the value is "wrong" is entirely unclear. Instead ...xen/include/public/version.h typedef char xen_capabilities_info_t[1024]; · line 67
Quoted 7 lines
xen/include/public/version.h +4
-0+* For all guest types using hardware virt extentions, Xen is not mapped into +* the guest kernel virtual address space. This now return 0, where it +* previously returned unrelated data. +*/ #define XENVER_platform_parameters 5struct xen_platform_parameters {xen_ulong_t virt_start;Quoted 2 lines
... the field name tells me that all that is being conveyed is the virtual address of where the hypervisor area starts.
Quoted 8 lines
IMO, it doesn't matter what the name of the field is. It dates from the days when 32bit PV was the only type of guest.
32bit PV guests really do have a variable split, so the guest kernel really does need to get this value from Xen.
The split for 64bit PV guests is compile-time constant, hence why 64bit PV kernels don't care.
... once we get to run Xen in 5-level mode, 4-level PV guests could also gain a variable split: Like for 32-bit guests now, only the r/o M2P would need to live in that area, and this may well occupy less than the full range presently reserved for the hypervisor.
Quoted 11 lines
For compat HVM, it happens to pick up the -1 from:
#ifdef
CONFIG_PV32
HYPERVISOR_COMPAT_VIRT_START(d)=
is_pv_domain(d)?__HYPERVISOR_COMPAT_VIRT_START: ~0u; #endifin
arch_domain_create(), whereas for non-compat HVM, it gets a number in an address space it has no connection to in the slightest. ARM guests end up gettingXEN_VIRT_START(== 2M) handed back, but this absolutely an internal detail that guests have no business knowing.
Well, okay, this looks to be good enough an argument to make the adjustment you propose for !PV guests.
Quoted 6 lines
The only reason I'm not issuing an XSA for this is because we don't have any pretence of KASLR in Xen. Pretty much every other kernel gets CVEs for infoleaks like this.
We feasibly could do KASLR in !PV builds, at which point this would qualify for an XSA.
I would question that, but I can see your view as one possible one.
Jan
Re: [PATCH 3/4] xen/version: Drop bogus return values for XENVER_platform_parameters
Quoted 11 lines · Jan Beulich
Quoted 10 lines · Andrew Cooper
Quoted 9 lines · Jan Beulich
Quoted 8 lines · Andrew Cooper
A split in virtual address space is only applicable for x86 PV guests. Furthermore, the information returned for x86 64bit PV guests is wrong.
Explain the problem in
version.h, stating the other information that PV guests need to know.For 64bit PV guests, and all non-x86-PV guests, return 0, which is strictly less wrong than the values currently returned.
Quoted 8 lines
Quoted 5 lines
Quoted 3 lines
I disagree for the 64-bit part of this. Seeing Linux'es exposure of the value in sysfs I even wonder whether we can change this like you do for HVM. Who knows what is being inferred from the value, and by whom.
Linux's sysfs ABI isn't relevant to us here. The sysfs ABI says it reports what the hypervisor presents, not that it will be a nonzero number.
It effectively reports the hypervisor (virtual) base address there. How can we not care if something (kexec would come to mind) would be using it for whatever purpose.
What about kexec do you think would care?
The only thing kexec-tools cares about is XENVER_capabilities, but even that's not actually correct for figuring out whether Xen can do kexec transitions to ELF64/32.
Quoted 3 lines
And thinking of it, the tool stack has uses,
too. Assuming you audited them, did you consider removing dead uses in a prereq patch (and discuss the effects on live ones in the description)?
There is only one toolstack use I can spot which is non-informational, and it's broken AFAICT.
xl dump-core writes out a header which includes this metadata, but it takes dom0's value, not domU's. (Not that this is relevant AFAICT, because the M2P is handled specially anyway.)
Most XENVER_* information is global (and by this, I mean invariant and non-caller dependent, outside of livepatching.)
XENVER_guest_handle is caller-variant, but the toolstack has proper interfaces to get/set this value.
XENVER_platform_parameters (and XENVER_get_features for that matter) are caller-variant, and the toolstack has no way to get domU's view of this data.
Every instance (well - this is the only interesting one) of the use of XENVER_platform_parameters I can find is broken, even in the Xen code.
Quoted 23 lines
Quoted 23 lines
Quoted 23 lines
Quoted 11 lines
xen/include/public/version.h +5
-0@@ -42,6 +42,26 @@ typedef char xen_capabilities_info_t[1024];typedef char xen_changeset_info_t [64 ];#define XEN_CHANGESET_INFO_LEN (sizeof(xen_changeset_info_t)) +/* +* This API is problematic. +* +* It is only applicable to guests which share pagetables with Xen (x86 PV +* guests), and is supposed to identify the virtual address split between xen/include/public/version.h typedef char xen_capabilities_info_t[1024]; · line 59
Quoted 10 lines
xen/include/public/version.h +10
-0+* guest kernel and Xen. +* +* For 32bit PV guests, it mostly does this, but the caller needs to know that +* Xen lives between the split and 4G. +* +* For 64bit PV guests, Xen lives at the bottom of the upper canonical range. +* This previously returned the start of the upper canonical range (which is +* the userspace/Xen split), not the Xen/kernel split (which is 8TB further +* on). This now returns 0 because the old number wasn't correct, and +* changing it to anything else would be even worse. Whether the guest runs user mode code in the low or high half (or in yet another way of splitting) isn't really dictated by the PV ABI, is it?
Quoted 5 lines
Quoted 3 lines
No, but given a choice of reporting the thing which is an architectural boundary, or the one which is the actual split between the two adjacent ranges, reporting the architectural boundary is clearly the unhelpful thing.
Hmm. To properly parallel the 32-bit variant, a [start,end] range would need exposing for 64-bit, rather than exposing nothing.
The 32-bit version is a start/end pair, but with end being implicit at the 4G architectural boundary.
If we were doing 64-bit from scratch, then reporting end would have been sensible, because for 64-bit, start is the architectural boundary which can be implicit.
But there is no such thing as a 64bit PV guest with any (useful) idea of a variable split, because this number has been junk for the entire lifetime of 64bit PV guests. In particular, ...
Quoted 15 lines
Quoted 15 lines
Quoted 9 lines
xen/include/public/version.h typedef char xen_capabilities_info_t[1024]; · line 67
Quoted 7 lines
xen/include/public/version.h +4
-0+* For all guest types using hardware virt extentions, Xen is not mapped into +* the guest kernel virtual address space. This now return 0, where it +* previously returned unrelated data. +*/ #define XENVER_platform_parameters 5struct xen_platform_parameters {xen_ulong_t virt_start;... the field name tells me that all that is being conveyed is the virtual address of where the hypervisor area starts.
IMO, it doesn't matter what the name of the field is. It dates from the days when 32bit PV was the only type of guest.
32bit PV guests really do have a variable split, so the guest kernel really does need to get this value from Xen.
Quoted 6 lines
Quoted 2 lines
The split for 64bit PV guests is compile-time constant, hence why 64bit PV kernels don't care.
... once we get to run Xen in 5-level mode, 4-level PV guests could also gain a variable split: Like for 32-bit guests now, only the r/o M2P would need to live in that area, and this may well occupy less than the full range presently reserved for the hypervisor.
... you can't do this, because it only works for guests which have chosen to find the M2P using XENMEM_machphys_mapping (e.g. Linux), and doesn't for e.g. MiniOS which does:
#define machine_to_phys_mapping ((unsigned long *)HYPERVISOR_VIRT_START)
in fact, looking at this, MiniOS is also broken as a 32bit PV dom0, because it hardcodes __MACH2PHYS_VIRT_START in the case where the split really is variable.
Its only PV guests which are LA57 aware which can possibly benefit from a variable position M2P, and only because that will be a new ELFNOTE protocol.
Quoted 8 lines
Quoted 8 lines
For compat HVM, it happens to pick up the -1 from:
#ifdef
CONFIG_PV32
HYPERVISOR_COMPAT_VIRT_START(d)=
is_pv_domain(d)?__HYPERVISOR_COMPAT_VIRT_START: ~0u; #endif
Quoted 6 lines
Quoted 4 lines
in
arch_domain_create(), whereas for non-compat HVM, it gets a number in an address space it has no connection to in the slightest. ARM guests end up gettingXEN_VIRT_START(== 2M) handed back, but this absolutely an internal detail that guests have no business knowing.Well, okay, this looks to be good enough an argument to make the adjustment you propose for !PV guests.
Right, HVM (on all architectures) is very cut and dry.
But it feels wrong to not address the PV64 issue at the same time because it is similar level of broken, despite there being (in theory) a legitimate need for a PV guest kernel to know it.
~Andrew
Re: [PATCH 3/4] xen/version: Drop bogus return values for XENVER_platform_parameters
Quoted 15 lines · Andrew Cooper
Quoted 14 lines · Jan Beulich
Quoted 13 lines · Andrew Cooper
Quoted 12 lines · Jan Beulich
Quoted 8 lines · Andrew Cooper
A split in virtual address space is only applicable for x86 PV guests. Furthermore, the information returned for x86 64bit PV guests is wrong.
Explain the problem in
version.h, stating the other information that PV guests need to know.For 64bit PV guests, and all non-x86-PV guests, return 0, which is strictly less wrong than the values currently returned.
I disagree for the 64-bit part of this. Seeing Linux'es exposure of the value in sysfs I even wonder whether we can change this like you do for HVM. Who knows what is being inferred from the value, and by whom.
Quoted 7 lines
Quoted 5 lines
Quoted 2 lines
Linux's sysfs ABI isn't relevant to us here. The sysfs ABI says it reports what the hypervisor presents, not that it will be a nonzero number.
It effectively reports the hypervisor (virtual) base address there. How can we not care if something (kexec would come to mind) would be using it for whatever purpose.
What about kexec do you think would care?
I didn't think about anything specific, but I could see why it may want to know where in VA space Xen sits.
Quoted 62 lines
Quoted 23 lines
Quoted 23 lines
Quoted 23 lines
Quoted 11 lines
xen/include/public/version.h +5
-0@@ -42,6 +42,26 @@ typedef char xen_capabilities_info_t[1024];typedef char xen_changeset_info_t [64 ];#define XEN_CHANGESET_INFO_LEN (sizeof(xen_changeset_info_t)) +/* +* This API is problematic. +* +* It is only applicable to guests which share pagetables with Xen (x86 PV +* guests), and is supposed to identify the virtual address split between xen/include/public/version.h typedef char xen_capabilities_info_t[1024]; · line 59
Quoted 10 lines
xen/include/public/version.h +10
-0+* guest kernel and Xen. +* +* For 32bit PV guests, it mostly does this, but the caller needs to know that +* Xen lives between the split and 4G. +* +* For 64bit PV guests, Xen lives at the bottom of the upper canonical range. +* This previously returned the start of the upper canonical range (which is +* the userspace/Xen split), not the Xen/kernel split (which is 8TB further +* on). This now returns 0 because the old number wasn't correct, and +* changing it to anything else would be even worse. Whether the guest runs user mode code in the low or high half (or in yet another way of splitting) isn't really dictated by the PV ABI, is it?
Quoted 5 lines
Quoted 3 lines
No, but given a choice of reporting the thing which is an architectural boundary, or the one which is the actual split between the two adjacent ranges, reporting the architectural boundary is clearly the unhelpful thing.
Hmm. To properly parallel the 32-bit variant, a [start,end] range would need exposing for 64-bit, rather than exposing nothing.
The 32-bit version is a start/end pair, but with end being implicit at the 4G architectural boundary.
If we were doing 64-bit from scratch, then reporting end would have been sensible, because for 64-bit, start is the architectural boundary which can be implicit.
But there is no such thing as a 64bit PV guest with any (useful) idea of a variable split, because this number has been junk for the entire lifetime of 64bit PV guests. In particular, ...
Quoted 15 lines
Quoted 15 lines
Quoted 9 lines
xen/include/public/version.h typedef char xen_capabilities_info_t[1024]; · line 67
Quoted 7 lines
xen/include/public/version.h +4
-0+* For all guest types using hardware virt extentions, Xen is not mapped into +* the guest kernel virtual address space. This now return 0, where it +* previously returned unrelated data. +*/ #define XENVER_platform_parameters 5struct xen_platform_parameters {xen_ulong_t virt_start;... the field name tells me that all that is being conveyed is the virtual address of where the hypervisor area starts.
IMO, it doesn't matter what the name of the field is. It dates from the days when 32bit PV was the only type of guest.
32bit PV guests really do have a variable split, so the guest kernel really does need to get this value from Xen.
Quoted 6 lines
Quoted 2 lines
The split for 64bit PV guests is compile-time constant, hence why 64bit PV kernels don't care.
... once we get to run Xen in 5-level mode, 4-level PV guests could also gain a variable split: Like for 32-bit guests now, only the r/o M2P would need to live in that area, and this may well occupy less than the full range presently reserved for the hypervisor.
Quoted 5 lines
... you can't do this, because it only works for guests which have chosen to find the M2P using
XENMEM_machphys_mapping(e.g. Linux), and doesn't for e.g. MiniOS which does:#define
machine_to_phys_mapping((unsigned long *)HYPERVISOR_VIRT_START)
Hmm, looks like a misunderstanding? I certainly wasn't thinking about making the start of that region variable, but rather the end (i.e. not exactly like for 32-bit compat).
Quoted 11 lines
Quoted 11 lines
Quoted 11 lines
For compat HVM, it happens to pick up the -1 from:
#ifdef
CONFIG_PV32
HYPERVISOR_COMPAT_VIRT_START(d)=
is_pv_domain(d)?__HYPERVISOR_COMPAT_VIRT_START: ~0u; #endifin
arch_domain_create(), whereas for non-compat HVM, it gets a number in an address space it has no connection to in the slightest. ARM guests end up gettingXEN_VIRT_START(== 2M) handed back, but this absolutely an internal detail that guests have no business knowing.
Quoted 8 lines
Quoted 2 lines
Well, okay, this looks to be good enough an argument to make the adjustment you propose for !PV guests.
Right, HVM (on all architectures) is very cut and dry.
But it feels wrong to not address the PV64 issue at the same time because it is similar level of broken, despite there being (in theory) a legitimate need for a PV guest kernel to know it.
To me it feels wrong to address the 64-bit PV issue by removing information, when - as you also say - it is actually _missing_ information. To me the proper course of action would be to expose the upper bound as well (such that, down the road, it could become dynamic). There's also no info leak there, as the two (static) bounds are part of the PV ABI anyway.
Jan
Re: [PATCH 3/4] xen/version: Drop bogus return values for XENVER_platform_parameters
Quoted 16 lines · Jan Beulich
Quoted 15 lines · Andrew Cooper
Quoted 14 lines · Jan Beulich
Quoted 13 lines · Andrew Cooper
Quoted 12 lines · Jan Beulich
Quoted 8 lines · Andrew Cooper
A split in virtual address space is only applicable for x86 PV guests. Furthermore, the information returned for x86 64bit PV guests is wrong.
Explain the problem in
version.h, stating the other information that PV guests need to know.For 64bit PV guests, and all non-x86-PV guests, return 0, which is strictly less wrong than the values currently returned.
I disagree for the 64-bit part of this. Seeing Linux'es exposure of the value in sysfs I even wonder whether we can change this like you do for HVM. Who knows what is being inferred from the value, and by whom.
Quoted 8 lines
Quoted 6 lines
Quoted 5 lines
Quoted 2 lines
Linux's sysfs ABI isn't relevant to us here. The sysfs ABI says it reports what the hypervisor presents, not that it will be a nonzero number.
It effectively reports the hypervisor (virtual) base address there. How can we not care if something (kexec would come to mind) would be using it for whatever purpose.
What about kexec do you think would care?
I didn't think about anything specific, but I could see why it may want to know where in VA space Xen sits.
The kexec image doesn't run "under" Xen; it replaces Xen in memory, and transition into the new image is via no paging (32bit) or identity paging (64bit) in the reserved region.
We don't really support kexec load (it's there, but I don't expect anyone has exercised it in anger), but if we were to load a new Xen+dom0, then kexec-tools still has nothing relevant to do with new-Xen+dom0's split.
Quoted 21 lines
Quoted 21 lines
Quoted 21 lines
Quoted 9 lines
Quoted 9 lines
Quoted 7 lines
+ * For all guest types using hardware virt extentions, Xen is not mapped into
+ * the guest kernel virtual address space. This now return 0, where it
+ * previously returned unrelated data.
+ */
#defineXENVER_platform_parameters5struct xen_platform_parameters { xen_ulong_t virt_start;... the field name tells me that all that is being conveyed is the virtual address of where the hypervisor area starts.
Quoted 8 lines
IMO, it doesn't matter what the name of the field is. It dates from the days when 32bit PV was the only type of guest.
32bit PV guests really do have a variable split, so the guest kernel really does need to get this value from Xen.
The split for 64bit PV guests is compile-time constant, hence why 64bit PV kernels don't care.
... once we get to run Xen in 5-level mode, 4-level PV guests could also gain a variable split: Like for 32-bit guests now, only the r/o M2P would need to live in that area, and this may well occupy less than the full range presently reserved for the hypervisor.
Quoted 8 lines
Quoted 5 lines
... you can't do this, because it only works for guests which have chosen to find the M2P using
XENMEM_machphys_mapping(e.g. Linux), and doesn't for e.g. MiniOS which does:#define
machine_to_phys_mapping((unsigned long *)HYPERVISOR_VIRT_START)Hmm, looks like a misunderstanding? I certainly wasn't thinking about making the start of that region variable, but rather the end (i.e. not exactly like for 32-bit compat).
But for what purpose? You can't make 4-level guests have a variable range.
The best you can do is make a 5-level-aware guest running on 4-level Xen have some new semantics, but a 4-level PV guest already owns the overwhelming majority of virtual address space, so being able to hand back a few extra TB is not interesting. And for making that happen...
Quoted 15 lines
Quoted 15 lines
Quoted 7 lines
Quoted 7 lines
For compat HVM, it happens to pick up the -1 from:
#ifdef
CONFIG_PV32
HYPERVISOR_COMPAT_VIRT_START(d)=
is_pv_domain(d)?__HYPERVISOR_COMPAT_VIRT_START: ~0u; #endifQuoted 6 lines
Quoted 4 lines
in
arch_domain_create(), whereas for non-compat HVM, it gets a number in an address space it has no connection to in the slightest. ARM guests end up gettingXEN_VIRT_START(== 2M) handed back, but this absolutely an internal detail that guests have no business knowing.Well, okay, this looks to be good enough an argument to make the adjustment you propose for !PV guests.
Right, HVM (on all architectures) is very cut and dry.
Quoted 8 lines
Quoted 3 lines
But it feels wrong to not address the PV64 issue at the same time because it is similar level of broken, despite there being (in theory) a legitimate need for a PV guest kernel to know it.
To me it feels wrong to address the 64-bit PV issue by removing information, when - as you also say - it is actually _missing_ information. To me the proper course of action would be to expose the upper bound as well (such that, down the road, it could become dynamic). There's also no info leak there, as the two (static) bounds are part of the PV ABI anyway.
... the absolute best you could do here is introduce a new XENVER_platform_parameters2 that has a different structure.
Which still leaves us with XENVER_platform_parameters providing data which is somewhere between useless and actively unhelpful.
~Andrew
Re: [PATCH 3/4] xen/version: Drop bogus return values for XENVER_platform_parameters
Quoted 31 lines · Andrew Cooper
Quoted 30 lines · Jan Beulich
Quoted 29 lines · Andrew Cooper
Quoted 18 lines · Jan Beulich
Quoted 17 lines · Andrew Cooper
Quoted 10 lines · Jan Beulich
Quoted 7 lines · Andrew Cooper
+ * For all guest types using hardware virt extentions, Xen is not mapped into
+ * the guest kernel virtual address space. This now return 0, where it
+ * previously returned unrelated data.
+ */
#defineXENVER_platform_parameters5struct xen_platform_parameters { xen_ulong_t virt_start;... the field name tells me that all that is being conveyed is the virtual address of where the hypervisor area starts.
IMO, it doesn't matter what the name of the field is. It dates from the days when 32bit PV was the only type of guest.
32bit PV guests really do have a variable split, so the guest kernel really does need to get this value from Xen.
Quoted 6 lines
Quoted 2 lines
The split for 64bit PV guests is compile-time constant, hence why 64bit PV kernels don't care.
... once we get to run Xen in 5-level mode, 4-level PV guests could also gain a variable split: Like for 32-bit guests now, only the r/o M2P would need to live in that area, and this may well occupy less than the full range presently reserved for the hypervisor.
... you can't do this, because it only works for guests which have chosen to find the M2P using
XENMEM_machphys_mapping(e.g. Linux), and doesn't for e.g. MiniOS which does:
Quoted 6 lines
Quoted 4 lines
Quoted 1 line
#define
machine_to_phys_mapping((unsigned long *)HYPERVISOR_VIRT_START)Hmm, looks like a misunderstanding? I certainly wasn't thinking about making the start of that region variable, but rather the end (i.e. not exactly like for 32-bit compat).
But for what purpose? You can't make 4-level guests have a variable range.
That entirely depends on the kernel. Linux, for example, could put its direct map lower, ...
Quoted 4 lines
The best you can do is make a 5-level-aware guest running on 4-level Xen have some new semantics, but a 4-level PV guest already owns the overwhelming majority of virtual address space, so being able to hand back a few extra TB is not interesting. And for making that happen...
... allowing it to cover some more space. That's not a whole lot more, sure.
Quoted 18 lines
Quoted 18 lines
Quoted 18 lines
Quoted 7 lines
Quoted 7 lines
For compat HVM, it happens to pick up the -1 from:
#ifdef
CONFIG_PV32
HYPERVISOR_COMPAT_VIRT_START(d)=
is_pv_domain(d)?__HYPERVISOR_COMPAT_VIRT_START: ~0u; #endifQuoted 6 lines
Quoted 4 lines
in
arch_domain_create(), whereas for non-compat HVM, it gets a number in an address space it has no connection to in the slightest. ARM guests end up gettingXEN_VIRT_START(== 2M) handed back, but this absolutely an internal detail that guests have no business knowing.Well, okay, this looks to be good enough an argument to make the adjustment you propose for !PV guests.
Right, HVM (on all architectures) is very cut and dry.
But it feels wrong to not address the PV64 issue at the same time because it is similar level of broken, despite there being (in theory) a legitimate need for a PV guest kernel to know it.
Quoted 8 lines
Quoted 5 lines
To me it feels wrong to address the 64-bit PV issue by removing information, when - as you also say - it is actually _missing_ information. To me the proper course of action would be to expose the upper bound as well (such that, down the road, it could become dynamic). There's also no info leak there, as the two (static) bounds are part of the PV ABI anyway.
... the absolute best you could do here is introduce a new
XENVER_platform_parameters2that has a different structure.
Indeed.
Quoted 2 lines
Which still leaves us with
XENVER_platform_parametersproviding data which is somewhere between useless and actively unhelpful.
If it's useless, there's still no reason to remove / alter the information, as you can never be absolutely certain that no-one uses this in some way. And for the "actively unhelpful" aspect it looks like our views simply differ. Is there no way we can settle on making the change affect HVM only?
Jan
PATCH 4/4 xen/version: Introduce non-truncating XENVER_* subops
In reply to Andrew CooperRecently in XenServer, we have encountered problems caused by both XENVER_extraversion and XENVER_commandline having fixed bounds.
More than just the invariant size, the APIs/ABIs also broken by typedef-ing an array, and using an unqualified 'char' which has implementation-specific signed-ness.
Provide brand new ops, which are capable of expressing variable length strings, and mark the older ops as broken.
This fixes all issues around XENVER_extraversion being longer than 15 chars. More work is required to remove other assumptions about XENVER_commandline being 1023 chars long.
---
CC: George Dunlap <[email protected]>
CC: Jan Beulich <[email protected]>
CC: Stefano Stabellini <[email protected]>
CC: Wei Liu <[email protected]>
CC: Julien Grall <[email protected]>
Tested by forcing XENVER_extraversion to be 20 chars long, and confirming that an untruncated version can be obtained.
API/ABI wise, XENVER_build_id could be merged into xenver_varstring_op(), but the internal infrastructure is awkward.
--- xen/common/kernel.c | 69 ++++++++++++++++++++++++++++++++++++++++++++ xen/include/public/version.h | 56 +++++++++++++++++++++++++++++++++-- xen/include/xlat.lst | 1 + xen/xsm/flask/hooks.c | 4 +++ 4 files changed, 128 insertions(+), 2 deletions(-)
xen/common/kernel.c +69 -0
@@ -24,6 +24,7 @@ CHECK_compile_info; CHECK_feature_info; CHECK_build_id; +CHECK_var_string;#endif enum system_state system_state = SYS_STATE_early_boot;
@@ -469,6 +470,66 @@ static int __init cf_check param_init(void)__initcall (param_init);#endif +static long xenver_varstring_op (int cmd,XEN_GUEST_HANDLE_PARAM (void ) arg) +{ +const char *str =NULL ; +size_t sz =0 ; +union { +xen_capabilities_info_t info; + } u; +struct xen_var_string user_str; + +switch ( cmd ) + { +case XENVER_extraversion2: + str =xen_extra_version (); +break ; + +case XENVER_changeset2: + str =xen_changeset (); +break ; + +case XENVER_commandline2: + str = saved_cmdline; +break ; + +case XENVER_capabilities2: +memset (u.info,0 ,sizeof (u.info)); +arch_get_xen_caps (&u.info); + str = u.info; +break ; + +default : +ASSERT_UNREACHABLE (); +break ; + } + +if ( !str || + !(sz =strlen (str)) ) +return -ENODATA ;/* failsafe */ + +if ( sz >INT32_MAX ) +return -E2BIG ;/* Compat guests. 2G ought to be plenty. */ + +if (guest_handle_is_null (arg) )/* Length request */ +return sz; + +if (copy_from_guest (&user_str, arg,1 ) ) +return -EFAULT ; + +if ( user_str.len ==0 ) +return -EINVAL ; + +if ( sz > user_str.len ) +return -ENOBUFS ; + +if (copy_to_guest_offset (arg, offsetof(struct xen_var_string , buf), + str, sz) ) +return -EFAULT ; + +return sz; +} +long do_xen_version (int cmd,XEN_GUEST_HANDLE_PARAM (void ) arg) {bool_t deny = !!xsm_xen_version (XSM_OTHER , cmd);
@@ -670,6 +731,14 @@ long do_xen_version(int cmd, XEN_GUEST_HANDLE_PARAM(void) arg)return sz; } + +case XENVER_extraversion2: +case XENVER_capabilities2: +case XENVER_changeset2: +case XENVER_commandline2: +if ( deny ) +return -EPERM ; +return xenver_varstring_op (cmd, arg); }return -ENOSYS ;
xen/include/public/version.h +54 -2
88 lines not on this page · raw message
xen/include/xlat.lst +1 -0
@@ -172,6 +172,7 @@ ? compile_info version.h ? feature_info version.h ? build_id version.h +? var_string version.h ? xenoprof_init xenoprof.h ? xenoprof_passive xenoprof.h ? flask_access xsm/flask_op.h
xen/xsm/flask/hooks.c +4 -0
@@ -1777,15 +1777,18 @@ static int cf_check flask_xen_version(uint32_t op)/* These sub-ops ignore the permission checks and return data. */ return 0 ;case XENVER_extraversion: +case XENVER_extraversion2:return avc_has_perm (dsid,SECINITSID_XEN ,SECCLASS_VERSION ,VERSION__XEN_EXTRAVERSION ,NULL );case XENVER_compile_info:return avc_has_perm (dsid,SECINITSID_XEN ,SECCLASS_VERSION ,VERSION__XEN_COMPILE_INFO ,NULL );case XENVER_capabilities: +case XENVER_capabilities2:return avc_has_perm (dsid,SECINITSID_XEN ,SECCLASS_VERSION ,VERSION__XEN_CAPABILITIES ,NULL );case XENVER_changeset: +case XENVER_changeset2:return avc_has_perm (dsid,SECINITSID_XEN ,SECCLASS_VERSION ,VERSION__XEN_CHANGESET ,NULL );case XENVER_pagesize:
@@ -1795,6 +1798,7 @@ static int cf_check flask_xen_version(uint32_t op)return avc_has_perm (dsid,SECINITSID_XEN ,SECCLASS_VERSION ,VERSION__XEN_GUEST_HANDLE ,NULL );case XENVER_commandline: +case XENVER_commandline2:return avc_has_perm (dsid,SECINITSID_XEN ,SECCLASS_VERSION ,VERSION__XEN_COMMANDLINE ,NULL );case XENVER_build_id:
Signature · 1 line
2.11.0
Re: [PATCH 4/4] xen/version: Introduce non-truncating XENVER_* subops
Hi Andrew,
xen/include/public/version.h line 25
Quoted 13 lines · Andrew Cooper
xen/include/public/version.h +4
-1@@ -19,12 +19,20 @@/* arg == NULL; returns major:minor (16:16). */ #define XENVER_version 0 -/* arg == xen_extraversion_t. */ +/* +* arg == xen_extraversion_t. +* +* This API/ABI is broken. Use XENVER_extraversion2 instead.
I read this as newer tools should never try to call XENVER_extraversion. But I don't think this is what you intend to say, correct? If so, I would say that an OS should first try XENVER_extraversion2 and then fallback to XENVER_extraversion if it returns -ENOSYS.
Same goes for the new hypercalls.
Cheers,
Signature · 1 line
Julien Grall
Re: [PATCH 4/4] xen/version: Introduce non-truncating XENVER_* subops
Quoted 22 lines · Julien Grall
Hi Andrew,
Quoted 7 lines · Andrew Cooper
xen/include/public/version.h +0
-0@@ -19,12 +19,20 @@ /* arg == NULL; returns major:minor (16:16). */ #define XENVER_version 0Quoted 5 lines
-/* arg ==
xen_extraversion_t. */
+/*
+ * arg ==xen_extraversion_t.
+ *
+ * This API/ABI is broken. UseXENVER_extraversion2instead.I read this as newer tools should never try to call
XENVER_extraversion. But I don't think this is what you intend to say, correct? If so, I would say that an OS should first tryXENVER_extraversion2and then fallback toXENVER_extraversionif it returns-ENOSYS.
Quoted 1 line
Same goes for the new hypercalls.
Right, but that's sufficiently basic that it goes without saying.
This is not a "my first introduction to writing code" tutorial.
~Andrew
Re: [PATCH 4/4] xen/version: Introduce non-truncating XENVER_* subops
Quoted 21 lines · Andrew Cooper
Quoted 20 lines · Julien Grall
xen/include/public/version.h line 25
Quoted 12 lines · Andrew Cooper
xen/include/public/version.h +4
-0@@ -19,12 +19,20 @@ /* arg == NULL; returns major:minor (16:16). */ #define XENVER_version 0 -/* arg == xen_extraversion_t. */ +/* +* arg == xen_extraversion_t. +* +* This API/ABI is broken. Use XENVER_extraversion2 instead. I read this as newer tools should never try to call
XENVER_extraversion. But I don't think this is what you intend to say, correct? If so, I would say that an OS should first tryXENVER_extraversion2and then fallback toXENVER_extraversionif it returns-ENOSYS.
Quoted 3 lines
Quoted 1 line
Same goes for the new hypercalls.
Right, but that's sufficiently basic that it goes without saying.
It never hurts to be obvious. I couldn't say the same for the other way around.
Quoted 2 lines
This is not a "my first introduction to writing code" tutorial.
That's the first introduction to Xen for an OS developer. Adding a few words (maybe with the first version introducing it) would likely be welcomed by any developer.
This avoids any misinterpretation and/or losing time playing the git blame game...
Cheers,
Signature · 1 line
Julien Grall
Re: [PATCH 4/4] xen/version: Introduce non-truncating XENVER_* subops
In reply to Andrew CooperQuoted 6 lines · Andrew Cooper
Recently in XenServer, we have encountered problems caused by both
XENVER_extraversionandXENVER_commandlinehaving fixed bounds.More than just the invariant size, the APIs/ABIs also broken by typedef-ing an array, and using an unqualified 'char' which has implementation-specific signed-ness.
Which is fine as long as only ASCII is returned. If non-ASCII can be returned, I agree "unsigned char" is better, but then we also need to spell out what encoding the strings use (UTF-8 presumably).
Quoted 2 lines
API/ABI wise,
XENVER_build_idcould be merged intoxenver_varstring_op(), but the internal infrastructure is awkward.
I guess build-id also doesn't fit the rest because of not returning strings, but indeed an array of bytes. You also couldn't use strlen() on the array.
Quoted 31 lines
xen/common/kernel.c +27
-0@@ -469,6 +470,66 @@ static int __init cf_check param_init(void)__initcall (param_init);#endif +static long xenver_varstring_op (int cmd,XEN_GUEST_HANDLE_PARAM (void ) arg) +{ +const char *str =NULL ; +size_t sz =0 ; +union { +xen_capabilities_info_t info; + } u; +struct xen_var_string user_str; + +switch ( cmd ) + { +case XENVER_extraversion2: + str =xen_extra_version (); +break ; + +case XENVER_changeset2: + str =xen_changeset (); +break ; + +case XENVER_commandline2: + str = saved_cmdline; +break ; + +case XENVER_capabilities2: +memset (u.info,0 ,sizeof (u.info)); +arch_get_xen_caps (&u.info); + str = u.info;
xen/common/kernel.c static int __init cf_check param_init(void) · line 509
Quoted 10 lines
xen/common/kernel.c +10
-0+break ; + +default : +ASSERT_UNREACHABLE (); +break ; + } + +if ( !str || + !(sz =strlen (str)) ) +return -ENODATA ;/* failsafe */
Is this really appropriate for e.g. an empty command line?
xen/common/kernel.c static int __init cf_check param_init(void) · line 512
Quoted 2 lines
xen/common/kernel.c +2
-0+if ( sz >INT32_MAX ) +return -E2BIG ;/* Compat guests. 2G ought to be plenty. */
While the comment here and in the public header mention compat guests, the check is uniform. What's the deal?
Quoted 5 lines
xen/common/kernel.c +5
-0+if (guest_handle_is_null (arg) )/* Length request */ +return sz; + +if (copy_from_guest (&user_str, arg,1 ) ) +return -EFAULT ;
xen/common/kernel.c static int __init cf_check param_init(void) · line 528
Quoted 10 lines
xen/common/kernel.c +10
-0+ +if ( user_str.len ==0 ) +return -EINVAL ; + +if ( sz > user_str.len ) +return -ENOBUFS ; + +if (copy_to_guest_offset (arg, offsetof(struct xen_var_string , buf), + str, sz) ) +return -EFAULT ;
Not inserting a nul terminator is going to make this slightly awkward to use.
Quoted 5 lines
xen/include/public/version.h +1
-0@@ -103,6 +126,35 @@ struct xen_build_id { };typedef struct xen_build_id xen_build_id_t ; +/*
xen/include/public/version.h struct xen_build_id { · line 139
Quoted 10 lines
xen/include/public/version.h +10
-0+* Container for an arbitrary variable length string. +*/ +struct xen_var_string { +uint32_t len;/* IN: size of buf[] in bytes. */ +unsigned char buf[XEN_FLEX_ARRAY_DIM ];/* OUT: requested data. */ +}; +typedef struct xen_var_string xen_var_string_t ; + +/* +* arg == xenver_string_t
Nit: xen_var_string_t (also again in the following text).
xen/include/public/version.h struct xen_build_id { · line 145
Quoted 5 lines
xen/include/public/version.h +5
-0+* Equivalent to the original ops, but with a non-truncating API/ABI. +* +* Passing arg == NULL is a request for size. The returned size does not +* include a NUL terminator, and has a practical upper limit of INT32_MAX for +* 32bit guests. This is expected to be plenty for the purpose.
As said above, the limit applies to all guests, which the wording here doesn't suggest.
Jan
Quoted 14 lines
xen/include/public/version.h +11
-0+* Otherwise, the input xenver_string_t provides the size of the following +* buffer. Xen will fill the buffer, and return the number of bytes written +* (e.g. if the input buffer was longer than necessary). +* +* These hypercalls can fail, in which case they'll return -XEN_Exx. +*/ +#define XENVER_extraversion2 11 +#define XENVER_capabilities2 12 +#define XENVER_changeset2 13 +#define XENVER_commandline2 14 +#endif /* __XEN_PUBLIC_VERSION_H__ */ /*
Re: [PATCH 4/4] xen/version: Introduce non-truncating XENVER_* subops
Quoted 10 lines · Jan Beulich
Quoted 6 lines · Andrew Cooper
Recently in XenServer, we have encountered problems caused by both
XENVER_extraversionandXENVER_commandlinehaving fixed bounds.More than just the invariant size, the APIs/ABIs also broken by typedef-ing an array, and using an unqualified 'char' which has implementation-specific signed-ness.
Which is fine as long as only ASCII is returned. If non-ASCII can be returned, I agree "unsigned char" is better, but then we also need to spell out what encoding the strings use (UTF-8 presumably).
Well, it "functions" as far as ASCII is concerned, but I wouldn't say that was fine for an ABI.
The command line can reasonably have non-ASCII characters these days. (as can the compile information, but I intentionally didn't convert them to this new format).
UTF-8 is probably the sensible thing to specify, if we're going to make any statement.
Quoted 5 lines
Quoted 3 lines
API/ABI wise,
XENVER_build_idcould be merged intoxenver_varstring_op(), but the internal infrastructure is awkward.I guess build-id also doesn't fit the rest because of not returning strings, but indeed an array of bytes. You also couldn't use
strlen()on the array.
The format is unspecified, but it is a base64 encoded ASCII string (not NUL terminated).
Quoted 34 lines
Quoted 34 lines
xen/common/kernel.c +30
-0@@ -469,6 +470,66 @@ static int __init cf_check param_init(void)__initcall (param_init);#endif +static long xenver_varstring_op (int cmd,XEN_GUEST_HANDLE_PARAM (void ) arg) +{ +const char *str =NULL ; +size_t sz =0 ; +union { +xen_capabilities_info_t info; + } u; +struct xen_var_string user_str; + +switch ( cmd ) + { +case XENVER_extraversion2: + str =xen_extra_version (); +break ; + +case XENVER_changeset2: + str =xen_changeset (); +break ; + +case XENVER_commandline2: + str = saved_cmdline; +break ; + +case XENVER_capabilities2: +memset (u.info,0 ,sizeof (u.info)); +arch_get_xen_caps (&u.info); + str = u.info; +break ; + +default :
Quoted 8 lines
xen/common/kernel.c static int __init cf_check param_init(void) · line 509
Quoted 7 lines
xen/common/kernel.c +7
-0+ASSERT_UNREACHABLE (); +break ; + } + +if ( !str || + !(sz =strlen (str)) ) +return -ENODATA ;/* failsafe */ Is this really appropriate for e.g. an empty command line?
Hmm. Good point. All can in principle be empty.
In which case I think I'll put the -ENODATA in the default case and have a return 0 for the sz == 0 case.
Quoted 4 lines
xen/common/kernel.c static int __init cf_check param_init(void) · line 512
Quoted 2 lines
xen/common/kernel.c +2
-0+if ( sz >INT32_MAX ) +return -E2BIG ;/* Compat guests. 2G ought to be plenty. */ While the comment here and in the public header mention compat guests, the check is uniform. What's the deal?
Well its either this, or a (comat ? INT32_MAX : INT64_MAX) check, along with the ifdefary and predicates required to make that compile.
But there's not a CPU today which can actually move 2G of data (which is 4G of L1d bandwidth) without suffering the watchdog (especially as we've just read it once for strlen(), so that's 6G of bandwidth), nor do I expect this to change in the forseeable future.
There's some boundary (probably far lower) beyond which we can't use the algorithm here.
There wants to be some limit, and I don't feel it is necessary to make it variable on the guest type.
Quoted 10 lines
Quoted 10 lines
xen/common/kernel.c +9
-0+if (guest_handle_is_null (arg) )/* Length request */ +return sz; + +if (copy_from_guest (&user_str, arg,1 ) ) +return -EFAULT ; + +if ( user_str.len ==0 ) +return -EINVAL ; +
Quoted 8 lines
xen/common/kernel.c static int __init cf_check param_init(void) · line 528
Quoted 6 lines
xen/common/kernel.c +6
-0+if ( sz > user_str.len ) +return -ENOBUFS ; + +if (copy_to_guest_offset (arg, offsetof(struct xen_var_string , buf), + str, sz) ) +return -EFAULT ;Not inserting a nul terminator is going to make this slightly awkward to use.
Not really. It matches how build-id currently works, and forces the caller to use proper tained-string discipline.
To safely printk() it, all you need is:
rc = xen_version(XENVER_$X, &str);
if ( rc < 0 )
return rc;printk("%*.s\n", rc, str.buf);
Quoted 4 lines
Quoted 4 lines
xen/include/public/version.h +0
-0@@ -103,6 +126,35 @@ struct xen_build_id { };typedef struct xen_build_id xen_build_id_t ;
Quoted 12 lines
xen/include/public/version.h struct xen_build_id { · line 139
Quoted 11 lines
xen/include/public/version.h +11
-0+/* +* Container for an arbitrary variable length string. +*/ +struct xen_var_string { +uint32_t len;/* IN: size of buf[] in bytes. */ +unsigned char buf[XEN_FLEX_ARRAY_DIM ];/* OUT: requested data. */ +}; +typedef struct xen_var_string xen_var_string_t ; + +/* +* arg == xenver_string_t Nit:
xen_var_string_t(also again in the following text).
Ah yes. I originally called it xenver_string_t then found a whole load of compat assumptions about xen prefixes on names.
Will fix.
But overall, I'm not seeing a major objection to this change? In which case I'll go ahead and do the tools/ cleanup too for v2.
~Andrew
Re: [PATCH 4/4] xen/version: Introduce non-truncating XENVER_* subops
Quoted 9 lines · Andrew Cooper
Quoted 5 lines · Jan Beulich
Quoted 2 lines · Andrew Cooper
API/ABI wise,
XENVER_build_idcould be merged intoxenver_varstring_op(), but the internal infrastructure is awkward.I guess build-id also doesn't fit the rest because of not returning strings, but indeed an array of bytes. You also couldn't use
strlen()on the array.The format is unspecified, but it is a base64 encoded ASCII string (not NUL terminated).
Where are you taking this base64 encoding from? I see the build ID being pure binary data, which could easily have an embedded nul.
Quoted 13 lines
Quoted 4 lines
xen/common/kernel.c static int __init cf_check param_init(void) · line 512
Quoted 2 lines
xen/common/kernel.c +2
-0+if ( sz >INT32_MAX ) +return -E2BIG ;/* Compat guests. 2G ought to be plenty. */ While the comment here and in the public header mention compat guests, the check is uniform. What's the deal?
Well its either this, or a (comat ?
INT32_MAX:INT64_MAX) check, along with the ifdefary and predicates required to make that compile.But there's not a CPU today which can actually move 2G of data (which is 4G of L1d bandwidth) without suffering the watchdog (especially as we've just read it once for
strlen(), so that's 6G of bandwidth), nor do I expect this to change in the forseeable future.
Quoted 5 lines
There's some boundary (probably far lower) beyond which we can't use the algorithm here.
There wants to be some limit, and I don't feel it is necessary to make it variable on the guest type.
Sure. My question was merely because of the special mentioning of 32-bit / compat guests. I'm fine with the universal limit, and I'd also be fine with a lower (universal) bound. All I'm after is that the (to me at least) confusing comments be adjusted.
Quoted 2 lines
But overall, I'm not seeing a major objection to this change? In which case I'll go ahead and do the
tools/cleanup too for v2.
Well, I'm not entirely convinced of the need for the new subops (we could as well introduce build time checks to make sure no truncation will occur for the existing ones), but I also won't object to their introduction. At least for the command line I can see that we will want (need) to support more than 1k worth of a string, considering how many options we have accumulated.
Jan
Re: [PATCH 4/4] xen/version: Introduce non-truncating XENVER_* subops
Quoted 11 lines · Jan Beulich
Quoted 8 lines · Andrew Cooper
Quoted 5 lines · Jan Beulich
Quoted 2 lines · Andrew Cooper
API/ABI wise,
XENVER_build_idcould be merged intoxenver_varstring_op(), but the internal infrastructure is awkward.I guess build-id also doesn't fit the rest because of not returning strings, but indeed an array of bytes. You also couldn't use
strlen()on the array.The format is unspecified, but it is a base64 encoded ASCII string (not NUL terminated).
Where are you taking this base64 encoding from? I see the build ID being pure binary data, which could easily have an embedded nul.
Oh, so it is.
I'd failed to spot that libxl formats it automatically on behalf of the caller.
Quoted 15 lines
Quoted 15 lines
Quoted 4 lines
xen/common/kernel.c static int __init cf_check param_init(void) · line 512
Quoted 2 lines
xen/common/kernel.c +2
-0+if ( sz >INT32_MAX ) +return -E2BIG ;/* Compat guests. 2G ought to be plenty. */ While the comment here and in the public header mention compat guests, the check is uniform. What's the deal?
Well its either this, or a (comat ?
INT32_MAX:INT64_MAX) check, along with the ifdefary and predicates required to make that compile.But there's not a CPU today which can actually move 2G of data (which is 4G of L1d bandwidth) without suffering the watchdog (especially as we've just read it once for
strlen(), so that's 6G of bandwidth), nor do I expect this to change in the forseeable future.There's some boundary (probably far lower) beyond which we can't use the algorithm here.
Quoted 6 lines
Quoted 2 lines
There wants to be some limit, and I don't feel it is necessary to make it variable on the guest type.
Sure. My question was merely because of the special mentioning of 32-bit / compat guests. I'm fine with the universal limit, and I'd also be fine with a lower (universal) bound. All I'm after is that the (to me at least) confusing comments be adjusted.
How about 16k then?
Quoted 8 lines
Quoted 2 lines
But overall, I'm not seeing a major objection to this change? In which case I'll go ahead and do the
tools/cleanup too for v2.Well, I'm not entirely convinced of the need for the new subops (we could as well introduce build time checks to make sure no truncation will occur for the existing ones), but I also won't object to their introduction. At least for the command line I can see that we will want (need) to support more than 1k worth of a string, considering how many options we have accumulated.
I've legitimate customer bugs caused by the cmdline limit, and real test problems caused by the extraversion limit which I'm unwilling to "fix" by sticking to the current API/ABI.
~Andrew
Re: [PATCH 4/4] xen/version: Introduce non-truncating XENVER_* subops
Quoted 19 lines · Andrew Cooper
Quoted 18 lines · Jan Beulich
Quoted 17 lines · Andrew Cooper
Quoted 5 lines · Jan Beulich
xen/common/kernel.c static int __init cf_check param_init(void) · line 512
Quoted 2 lines · Andrew Cooper
xen/common/kernel.c +2
-0+if ( sz >INT32_MAX ) +return -E2BIG ;/* Compat guests. 2G ought to be plenty. */ While the comment here and in the public header mention compat guests, the check is uniform. What's the deal?
Well its either this, or a (comat ?
INT32_MAX:INT64_MAX) check, along with the ifdefary and predicates required to make that compile.But there's not a CPU today which can actually move 2G of data (which is 4G of L1d bandwidth) without suffering the watchdog (especially as we've just read it once for
strlen(), so that's 6G of bandwidth), nor do I expect this to change in the forseeable future.There's some boundary (probably far lower) beyond which we can't use the algorithm here.
Quoted 8 lines
Quoted 6 lines
Quoted 2 lines
There wants to be some limit, and I don't feel it is necessary to make it variable on the guest type.
Sure. My question was merely because of the special mentioning of 32-bit / compat guests. I'm fine with the universal limit, and I'd also be fine with a lower (universal) bound. All I'm after is that the (to me at least) confusing comments be adjusted.
How about 16k then?
Might be okay. If I was to pick a value, I'd use 64k.
Jan