PCI/ACPI: Support Microsoft's "DmaProperty"
Subsystems (6)
v9 cover PCI/ACPI: Support Microsoft's "DmaProperty"
Threat model: An overview of the security implications of non-strict IOMMU is presented at [1]. This change is motivated by “Case 1” where a DMA-capable device is processing untrusted inputs, e.g. network devices.
This patchset proposes using “DMA protection” to mitigate these risks. This has the following effects, currently controlled by the “pci_dev::untrusted” flag.
- Separate IOMMU DMA domains
- Use of SWIOTLB if CONFIG_SWIOTLB
- Disables quirks in Intel IOMMU
- Disables Address Translation Services
The “untrusted” flag was introduced in 2018 in [2]. The motivation for that change was to enable using IOMMU to protect against DMA attacks from externally facing devices such as thunderbolt ports. The patchset introduces the “untrusted” flag which “is supposed to cover various PCIe devices that may be used to conduct DMA attacks.” The patchset originally proposes naming the flag “is_external” but is renamed to “is_untrusted” and then “untrusted” supposing that it could apply to more than just externally facing thunderbolt devices. The fact that “ExternalFacingPort” is not part of any standard is called out during review but also that Windows expecting firmware to identify external facing ports makes it “as good as a formal standard in the Windows world.”
This current patch series was first proposed in January 2022 [3]. It originally proposed a new property “UntrustedDevice” which would cause the untrusted flag to be set. In V1 Greg questions whether the new property is part of the ACPI standard and asks who is making this policy decision. Mika links to Microsoft's documentation of “DmaProperty” and suggests that property should be adopted instead. Greg objects that Linux does not have “dma protection” but Mika says that this is the IOMMU. Today, the term “DMA protection” is used in thunderbolt driver code with the same meaning and in an Intel white paper [4] describing the technique. Mika also observes that Linux has recognized several properties documented by Microsoft but not part of the ACPI standard. There is discussion between Mika, Rafael, and Rajat about seeking to align with Microsoft on the semantics of “DmaProperty” for compatibility with firmware produced for Windows.
V2 of this patch series [5] again proposed an “UntrustedDevice” property which Greg objects to because it is not sufficiently descriptive, not sufficiently documented, and policies about trust don’t belong in the kernel. Rajat describes the “untrusted” flag’s current use, controlling IOMMU and Greg suggests naming the flag “use_iommu” or “able to do DMA.”
V3 of this patch series [6] proposes recognizing “DmaProperty” with slightly altered semantics from Microsoft’s documentation. Greg suggests adhering to Microsoft’s semantics for “DmaProperty” and to introduce a new property with new semantics instead. Greg again states that the flag being named “untrusted” is confusing.
V4 renames “untrusted” to “poses_dma_risk”. Christoph suggests “untrusted_dma” and Rafael agrees.
V5 renames the flag to “untrusted_dma”. Bjorn asks for clarification about whether the semantics of this flag will match Microsoft’s documentation. Rajat responds that Microsoft has agreed to update their documentation to have aligned semantics, in particular “the property is not restricted to identify ‘internal PCIe hierarchies’ (starting at root port), but to "any PCI device". As of today, Microsoft’s documentation does not appear to have been updated.
In V6 Rajat updates a link to Microsoft’s documentation, renames a function to pci_dev_has_dma_property() and uses acpi_dev_get_property() to read “DmaProperty”.
In V7 (Nov 2024) Joshua re-sends and Greg requests a summary of the history of discussion about the name for the “untrusted” flag and justification of the new name.
In V8 Joshua renames the “untrusted” flag to “requires_dma_protection”. Greg requests more information about the threat model, what does this property convey, and why we should use Microsoft’s DmaProperty and its semantics instead of inventing something new.
In V9 Joshua updates the cover letter with more information from previous submissions in this series and the “untrusted” flag’s introduction.
Links:
[1] https://lore.kernel.org/linux-arm-msm/20210624101557.v2.3.Icde6be7601a5939960caf802056c88cd5132eb4e@changeid/
[2] https://lore.kernel.org/lkml/[email protected]/
[3] https://lore.kernel.org/all/[email protected]/
[4] https://www.intel.com/content/dam/develop/external/us/en/documents/intel-whitepaper-using-iommu-for-dma-protection-in-uefi-820238.pdf
[5] https://lore.kernel.org/all/[email protected]/
[6] https://lore.kernel.org/all/[email protected]/
Rajat Jain (2):
PCI/ACPI: Support Microsoft's "DmaProperty"
PCI: Rename pci_dev->untrusted to pci_dev->requires_dma_protection
drivers/acpi/property.c | 3 +++ drivers/iommu/amd/iommu.c | 3 +-- drivers/iommu/dma-iommu.c | 16 ++++++++-------- drivers/iommu/intel/iommu.c | 10 +++++----- drivers/iommu/iommu.c | 5 ++--- drivers/pci/ats.c | 2 +- drivers/pci/pci-acpi.c | 22 ++++++++++++++++++++++ drivers/pci/pci.c | 2 +- drivers/pci/probe.c | 10 +++++----- drivers/pci/quirks.c | 4 ++-- include/linux/pci.h | 7 ++++--- 11 files changed, 54 insertions(+), 30 deletions(-)
base-commit: 0ad2507d5d93f39619fc42372c347d6006b64319
Signature · 1 line
2.48.1.601.g30ceb7b040-goog
PATCH v9 1/2 PCI/ACPI: Support Microsoft's "DmaProperty"
From: Rajat Jain <[email protected]>
The "DmaProperty" is supported and currently documented and used by Microsoft [link 1 below], to flag internal PCIe root ports that need DMA protection [link 2 below]. We have discussed with them and reached a common understanding that they shall change their MSDN documentation to say that the same property can be used to protect any PCI device, and not just internal PCIe root ports (since there is no point introducing yet another property for arbitrary PCI devices). This helps with security from internal devices that offer an attack surface for DMA attacks (e.g. internal network devices).
Support DmaProperty to mark DMA from a PCI device as untrusted.
--- drivers/acpi/property.c | 3 +++ drivers/pci/pci-acpi.c | 22 ++++++++++++++++++++++ 2 files changed, 25 insertions(+)
drivers/acpi/property.c +3 -0
@@ -56,6 +56,9 @@ static const guid_t prp_guids[] = {/* Storage device needs D3 GUID: 5025030f-842f-4ab4-a561-99a5189762d0 */ GUID_INIT (0x5025030f ,0x842f ,0x4ab4 ,0xa5 ,0x61 ,0x99 ,0xa5 ,0x18 ,0x97 ,0x62 ,0xd0 ), +/* DmaProperty for PCI devices GUID: 70d24161-6dd5-4c9e-8070-705531292865 */ +GUID_INIT (0x70d24161 ,0x6dd5 ,0x4c9e , + 0x80, 0x70, 0x70, 0x55, 0x31, 0x29, 0x28, 0x65), };/* ACPI _DSD data subnodes GUID [1]: dbb8e3e6-5886-4ba6-8795-1319f52a966b */
drivers/pci/pci-acpi.c +22 -0
@@ -1438,12 +1438,34 @@ static void pci_acpi_set_external_facing(struct pci_dev *dev) dev->external_facing =1 ; } +static int pci_dev_has_dma_property (struct pci_dev *dev) +{ +struct acpi_device *adev; +const union acpi_object *obj; + + adev =ACPI_COMPANION (&dev->dev); +if (!adev) +return 0 ; + +/* +* Property used by Microsoft Windows to enforce IOMMU DMA +* protection from any device, that the system may not fully trust; +* we'll honour it the same way. +*/ +if (!acpi_dev_get_property (adev,"DmaProperty" ,ACPI_TYPE_INTEGER , + &obj) && obj->integer.value ==1 ) +return 1 ; + +return 0 ; +} +void pci_acpi_setup (struct device *dev,struct acpi_device *adev) {struct pci_dev *pci_dev =to_pci_dev (dev);pci_acpi_optimize_delay (pci_dev, adev->handle);pci_acpi_set_external_facing (pci_dev); + pci_dev->untrusted |=pci_dev_has_dma_property (pci_dev);pci_acpi_add_edr_notifier (pci_dev);pci_acpi_add_pm_notifier (adev, pci_dev);
Signature · 1 line
2.48.1.601.g30ceb7b040-goog
Quoted 20 lines · Joshua Peraza
From: Rajat Jain <[email protected]>
The "DmaProperty" is supported and currently documented and used by Microsoft [link 1 below], to flag internal PCIe root ports that need DMA protection [link 2 below]. We have discussed with them and reached a common understanding that they shall change their MSDN documentation to say that the same property can be used to protect any PCI device, and not just internal PCIe root ports (since there is no point introducing yet another property for arbitrary PCI devices). This helps with security from internal devices that offer an attack surface for DMA attacks (e.g. internal network devices).
Support DmaProperty to mark DMA from a PCI device as untrusted.
Signed-off-by Rajat Jain <[email protected]>Reviewed-by Mika Westerberg <[email protected]>Signed-off-by Joshua Peraza <[email protected]>
Quoted 4 lines
--- drivers/acpi/property.c | 3 +++ drivers/pci/pci-acpi.c | 22 ++++++++++++++++++++++ 2 files changed, 25 insertions(+)
PATCH v9 2/2 PCI: Rename pci_dev->untrusted to pci_dev->requires_dma_protection
In reply to Joshua PerazaFrom: Rajat Jain <[email protected]>
Rename the field to make it more clear, that the device can execute DMA attacks on the system, and thus the system may need protection from such attacks from this device.
No functional change intended.
--- drivers/iommu/amd/iommu.c | 3 +-- drivers/iommu/dma-iommu.c | 16 ++++++++-------- drivers/iommu/intel/iommu.c | 10 +++++----- drivers/iommu/iommu.c | 5 ++--- drivers/pci/ats.c | 2 +- drivers/pci/pci-acpi.c | 2 +- drivers/pci/pci.c | 2 +- drivers/pci/probe.c | 10 +++++----- drivers/pci/quirks.c | 4 ++-- include/linux/pci.h | 7 ++++--- 10 files changed, 30 insertions(+), 31 deletions(-)
drivers/iommu/amd/iommu.c +1 -2
@@ -2952,8 +2952,7 @@ static int amd_iommu_def_domain_type(struct device *dev)if (!dev_data)return 0 ; -/* Always use DMA domain for untrusted device */ -if (dev_is_pci (dev) &&to_pci_dev (dev)->untrusted) +if (dev_is_pci (dev) &&to_pci_dev (dev)->requires_dma_protection)return IOMMU_DOMAIN_DMA ;/*
drivers/iommu/dma-iommu.c +8 -8
@@ -598,16 +598,16 @@ static int iova_reserve_iommu_regions(struct device *dev,return ret; } -static bool dev_is_untrusted (struct device *dev) +static bool dev_requires_dma_protection (struct device *dev) { -return dev_is_pci (dev) &&to_pci_dev (dev)->untrusted; +return dev_is_pci (dev) &&to_pci_dev (dev)->requires_dma_protection; }static bool dev_use_swiotlb (struct device *dev,size_t size,enum dma_data_direction dir) {return IS_ENABLED (CONFIG_SWIOTLB ) && - (dev_is_untrusted (dev) || + (dev_requires_dma_protection (dev) ||dma_kmalloc_needs_bounce (dev, size, dir)); }
@@ -620,7 +620,7 @@ static bool dev_use_sg_swiotlb(struct device *dev, struct scatterlist *sg,if (!IS_ENABLED (CONFIG_SWIOTLB ))return false; -if (dev_is_untrusted (dev)) +if (dev_requires_dma_protection (dev))return true;/*
@@ -1192,12 +1192,12 @@ dma_addr_t iommu_dma_map_page(struct device *dev, struct page *page,return DMA_MAPPING_ERROR ;/* -* Untrusted devices should not see padding areas with random -* leftover kernel data, so zero the pre- and post-padding. +* Zero the pre- and post-padding to prevent exposing kernel data to devices +* requiring DMA protection. * swiotlb_tbl_map_single() has initialized the bounce buffer * proper to the contents of the original memory buffer. */ -if (dev_is_untrusted (dev)) { +if (dev_requires_dma_protection (dev)) {size_t start, virt = (size_t )phys_to_virt (phys);/* Pre-padding */
@@ -1738,7 +1738,7 @@ size_t iommu_dma_opt_mapping_size(void)size_t iommu_dma_max_mapping_size (struct device *dev) { -if (dev_is_untrusted (dev)) +if (dev_requires_dma_protection (dev))return swiotlb_max_mapping_size (dev);return SIZE_MAX ;
drivers/iommu/intel/iommu.c +5 -5
@@ -2985,7 +2985,7 @@ static int __init platform_optin_force_iommu(void)/* * If Intel-IOMMU is disabled by default, we will apply identity -* map for all devices except those marked as being untrusted. +* map for all devices except those marked as requiring DMA protection. */ if (dmar_disabled)iommu_set_default_passthrough (false);
@@ -4036,13 +4036,13 @@ static bool intel_iommu_is_attach_deferred(struct device *dev) }/* -* Check that the device does not live on an external facing PCI port that is -* marked as untrusted. Such devices should not be able to apply quirks and -* thus not be able to bypass the IOMMU restrictions. +* Check that the device does not require DMA protection. Such devices should +* not be able to apply quirks and thus not be able to bypass the IOMMU +* restrictions. */ static bool risky_device (struct pci_dev *pdev) { -if (pdev->untrusted) { +if (pdev->requires_dma_protection) {pci_info (pdev,"Skipping IOMMU quirk for dev [%04X:%04X] on untrusted PCI link\n" , pdev->vendor, pdev->device);
drivers/iommu/iommu.c +2 -3
@@ -1793,10 +1793,9 @@ static int iommu_get_default_domain_type(struct iommu_group *group, driver_type =iommu_get_def_domain_type (group, gdev->dev, driver_type); -if (dev_is_pci (gdev->dev) &&to_pci_dev (gdev->dev)->untrusted) { +if (dev_is_pci (gdev->dev) &&to_pci_dev (gdev->dev)->requires_dma_protection) {/* -* No ARM32 using systems will set untrusted, it cannot -* work. +* ARM32 systems don't support DMA protection. */ if (WARN_ON (IS_ENABLED (CONFIG_ARM_DMA_USE_IOMMU )))return -1 ;
drivers/pci/ats.c +1 -1
@@ -43,7 +43,7 @@ bool pci_ats_supported(struct pci_dev *dev)if (!dev->ats_cap)return false; -return (dev->untrusted ==0 ); +return (dev->requires_dma_protection ==0 ); }EXPORT_SYMBOL_GPL (pci_ats_supported);
drivers/pci/pci-acpi.c +1 -1
@@ -1465,7 +1465,7 @@ void pci_acpi_setup(struct device *dev, struct acpi_device *adev)pci_acpi_optimize_delay (pci_dev, adev->handle);pci_acpi_set_external_facing (pci_dev); - pci_dev->untrusted |=pci_dev_has_dma_property (pci_dev); + pci_dev->requires_dma_protection |=pci_dev_has_dma_property (pci_dev);pci_acpi_add_edr_notifier (pci_dev);pci_acpi_add_pm_notifier (adev, pci_dev);
drivers/pci/pci.c +1 -1
@@ -1055,7 +1055,7 @@ static void pci_std_enable_acs(struct pci_dev *dev, struct pci_acs *caps) caps->ctrl |= (caps->cap &PCI_ACS_UF );/* Enable Translation Blocking for external devices and noats */ -if (pci_ats_disabled () || dev->external_facing || dev->untrusted) +if (pci_ats_disabled () || dev->external_facing || dev->requires_dma_protection) caps->ctrl |= (caps->cap &PCI_ACS_TB ); }
drivers/pci/probe.c +5 -5
@@ -1677,7 +1677,7 @@ static void set_pcie_thunderbolt(struct pci_dev *dev) dev->is_thunderbolt =1 ; } -static void set_pcie_untrusted (struct pci_dev *dev) +static void pci_set_requires_dma_protection (struct pci_dev *dev) {struct pci_dev *parent =pci_upstream_bridge (dev);
@@ -1687,14 +1687,14 @@ static void set_pcie_untrusted(struct pci_dev *dev)* If the upstream bridge is untrusted we treat this device as * untrusted as well. */ -if (parent->untrusted) { - dev->untrusted = true; +if (parent->requires_dma_protection) { + dev->requires_dma_protection = true;return ; }if (arch_pci_dev_is_removable (dev)) {pci_dbg (dev,"marking as untrusted\n" ); - dev->untrusted = true; + dev->requires_dma_protection = true; } }
@@ -2007,7 +2007,7 @@ int pci_setup_device(struct pci_dev *dev)/* Need to have dev->cfg_size ready */ set_pcie_thunderbolt (dev); -set_pcie_untrusted (dev); +pci_set_requires_dma_protection (dev);if (pci_is_pcie (dev)) dev->supported_speeds =pcie_get_supported_speeds (dev);
drivers/pci/quirks.c +2 -2
@@ -5299,7 +5299,7 @@ static void pci_quirk_enable_intel_rp_mpc_acs(struct pci_dev *dev)* PCI_ACS_SV | PCI_ACS_RR | PCI_ACS_CR | PCI_ACS_UF * * TODO: This quirk also needs to do equivalent of PCI_ACS_TB, -* if dev->external_facing || dev->untrusted +* if dev->external_facing || dev->requires_dma_protection */ static int pci_quirk_enable_intel_pch_acs (struct pci_dev *dev) {
@@ -5340,7 +5340,7 @@ static int pci_quirk_enable_intel_spt_pch_acs(struct pci_dev *dev) ctrl |= (cap &PCI_ACS_CR ); ctrl |= (cap &PCI_ACS_UF ); -if (pci_ats_disabled () || dev->external_facing || dev->untrusted) +if (pci_ats_disabled () || dev->external_facing || dev->requires_dma_protection) ctrl |= (cap &PCI_ACS_TB );pci_write_config_dword (dev, pos +INTEL_SPT_ACS_CTRL , ctrl);
include/linux/pci.h +4 -3
@@ -453,13 +453,14 @@ struct pci_dev { unsigned int shpc_managed:1; /* SHPC owned by shpchp */ unsigned int is_thunderbolt:1; /* Thunderbolt controller *//* -* Devices marked being untrusted are the ones that can potentially -* execute DMA attacks and similar. They are typically connected +* Devices marked with requires_dma_protection are the ones that can +* potentially execute DMA attacks and similar. They are typically connected * through external ports such as Thunderbolt but not limited to * that. When an IOMMU is enabled they should be getting full * mappings to make sure they cannot access arbitrary memory. */ - unsigned int untrusted:1; + unsigned int requires_dma_protection:1; +/* * Info from the platform, e.g., ACPI or device tree, may mark a * device as "external-facing". An external-facing device is
Signature · 1 line
2.48.1.601.g30ceb7b040-goog
Re: [v9 PATCH 2/2] PCI: Rename pci_dev->untrusted to pci_dev->requires_dma_protection
Quoted 5 lines · Joshua Peraza
From: Rajat Jain <[email protected]>
Rename the field to make it more clear, that the device can execute DMA attacks on the system, and thus the system may need protection from such attacks from this device.
It's not "may", it is "must" as that is what the kernel code does.
Anyway, no objection from me on this now, it makes more sense overall, thanks for sticking with it.