PCI / iommu / thunderbolt: IOMMU based DMA protection
Subsystems (5)
v3 cover PCI / iommu / thunderbolt: IOMMU based DMA protection
Recent systems with Thunderbolt ports may be utilizing IOMMU to prevent DMA attacks. This is different from the previous security level based scheme because the connected device cannot access system memory outside of the regions allocated for it by the driver.
When enabled the BIOS makes sure no device can do DMA outside of RMRR (Reserved Memory Region Record) regions. This means that during OS boot, before it enables IOMMU, none of the connected devices can bypass DMA protection for instance by overwriting the data structures used by the IOMMU. The BIOS communicates support for this to the OS by setting a new bit in ACPI DMAR table [1].
Because these systems utilize an IOMMU to block possible DMA attacks, typically (but not always) the Thunderbolt security level is set to "none" which means that all PCIe devices are immediately usable. This also means that Linux needs to follow Windows 10 and enable IOMMU automatically when running on such system otherwise connected devices can read/write system memory pretty much without any restrictions.
Since there is a way to identify PCIe root ports that are "external facing" we can put all internal devices to pass through (identity mapping) mode and only external devices need to go through full IOMMU mappings. We add a new flag "untrusted" that is supposed to cover various PCIe devices that may be used to conduct DMA attacks.
We also make sure PCIe ATS (Address Translation Service) is not enabled for devices flagged as untrusted because it could be used to bypass IOMMU completely as explained in the changelog of patch 3/4.
Finally we expose this information to userspace so tools such as bolt can do more accurate decision whether or not authorize the connected device.
[1] https://software.intel.com/sites/default/files/managed/c5/15/vt-directed-io-spec.pdf
Previous version of the patch series can be found here:
v2: https://lkml.org/lkml/2018/11/26/638
v1: https://www.spinics.net/lists/linux-pci/msg77751.html
Changes from v2:
* Rename the flag to "untrusted"
* Simplify setting the flag for root ports
* Dropped loop in set_pcie_untrusted()
* Add comment on top of prp_guids explaining the possible caveat resulting when the new GUIDs are treated as equivalent
* Updated changelogs according to feedback
Changes from v1:
* Reword Documentation/admin-guide/thunderbolt.rst to make the feature time frame/platform oriented as there will be systems shipping with Linux installed by default.
* Rename the flag is_external to is_untrusted so that we could use the same flag to cover all kinds of "untrusted" PCI devices, not just Thunderbolt connected devices. I still parse the _DSD in PCI/ACPI core because that's where we currently handle "HotPlugSupportInD3" as well. Also updated comments in patch [1/4].
* Added tags from Ashok, Joerg and Yehezkel. I'm assuming they still apply because I did not change the code with the exception of few comments and rename of the flag. Let me know if that's not the case anymore.
Lu Baolu (1):
iommu/vt-d: Force IOMMU on for platform opt in hint
Mika Westerberg (3):
PCI / ACPI: Identify untrusted PCI devices
iommu/vt-d: Do not enable ATS for untrusted devices
thunderbolt: Export IOMMU based DMA protection support to userspace
.../ABI/testing/sysfs-bus-thunderbolt | 9 +++ Documentation/admin-guide/thunderbolt.rst | 20 +++++++ drivers/acpi/property.c | 11 ++++ drivers/iommu/dmar.c | 25 +++++++++ drivers/iommu/intel-iommu.c | 56 ++++++++++++++++++- drivers/pci/pci-acpi.c | 19 +++++++ drivers/pci/probe.c | 15 +++++ drivers/thunderbolt/domain.c | 17 ++++++ include/linux/dmar.h | 8 +++ include/linux/pci.h | 8 +++ 10 files changed, 185 insertions(+), 3 deletions(-)
Signature · 1 line
2.19.2
PATCH v3 1/4 PCI / ACPI: Identify untrusted PCI devices
A malicious PCI device may use DMA to attack the system. An external Thunderbolt port is a convenient point to attach such a device. The OS may use IOMMU to defend against DMA attacks.
Recent BIOSes with Thunderbolt ports mark these externally facing root ports with this ACPI _DSD [1]:
Name (_DSD, Package () {
ToUUID ("efcc06cc-73ac-4bc3-bff0-76143807c389"),
Package () {
Package () {"ExternalFacingPort", 1},
Package () {"UID", 0 }
}})
If we find such a root port, mark it and all its children as untrusted. The rest of the OS may use this information to enable DMA protection against malicious devices. For instance the device may be put behind an IOMMU to keep it from accessing memory outside of what the driver has allocated for it.
While at it, add a comment on top of prp_guids array explaining the possible caveat resulting when these GUIDs are treated equivalent.
--- drivers/acpi/property.c | 11 +++++++++++ drivers/pci/pci-acpi.c | 19 +++++++++++++++++++ drivers/pci/probe.c | 15 +++++++++++++++ include/linux/pci.h | 8 ++++++++ 4 files changed, 53 insertions(+)
drivers/acpi/property.c +11 -0
@@ -24,6 +24,14 @@ static int acpi_data_get_property_array(const struct acpi_device_data *data,acpi_object_type type, const union acpi_object **obj); +/* +* The GUIDs here are made equivalent to each other in order to avoid extra +* complexity in the properties handling code, with the caveat that the +* kernel will accept certain combinations of GUID and properties that are +* not defined without a warning. For instance if any of the properties +* from different GUID appear in a property list of another, it will be +* accepted by the kernel. Firmware validation tools should catch these. +*/ static const guid_t prp_guids[] = {/* ACPI _DSD device properties GUID: daffd814-6eba-4d8c-8a91-bc9bbf4aa301 */ GUID_INIT(0xdaffd814, 0x6eba, 0x4d8c,
@@ -31,6 +39,9 @@ static const guid_t prp_guids[] = {/* Hotplug in D3 GUID: 6211e2c0-58a3-4af3-90e1-927a4e0c55a4 */ GUID_INIT (0x6211e2c0 ,0x58a3 ,0x4af3 ,0x90 ,0xe1 ,0x92 ,0x7a ,0x4e ,0x0c ,0x55 ,0xa4 ), +/* External facing port GUID: efcc06cc-73ac-4bc3-bff0-76143807c389 */ +GUID_INIT (0xefcc06cc ,0x73ac ,0x4bc3 , + 0xbf, 0xf0, 0x76, 0x14, 0x38, 0x07, 0xc3, 0x89), }; static const guid_t ads_guid =
drivers/pci/pci-acpi.c +19 -0
@@ -789,6 +789,24 @@ static void pci_acpi_optimize_delay(struct pci_dev *pdev,ACPI_FREE (obj); } +static void pci_acpi_set_untrusted (struct pci_dev *dev) +{ +u8 val; + +if (pci_pcie_type (dev) !=PCI_EXP_TYPE_ROOT_PORT ) +return ; +if (device_property_read_u8 (&dev->dev,"ExternalFacingPort" , &val)) +return ; + +/* +* These root ports expose PCIe (including DMA) outside of the +* system so make sure we treat them and everything behind as +* untrusted. +*/ +if (val) + dev->untrusted =1 ; +} +static void pci_acpi_setup (struct device *dev) {struct pci_dev *pci_dev =to_pci_dev (dev);
@@ -798,6 +816,7 @@ static void pci_acpi_setup(struct device *dev)return ;pci_acpi_optimize_delay (pci_dev, adev->handle); +pci_acpi_set_untrusted (pci_dev);pci_acpi_add_pm_notifier (adev, pci_dev);if (!adev->wakeup.flags.valid)
drivers/pci/probe.c +15 -0
@@ -1378,6 +1378,19 @@ static void set_pcie_thunderbolt(struct pci_dev *dev) } } +static void set_pcie_untrusted (struct pci_dev *dev) +{ +struct pci_dev *parent; + +/* +* If the upstream bridge is untrusted we treat this device +* untrusted as well. +*/ + parent =pci_upstream_bridge (dev); +if (parent && parent->untrusted) + dev->untrusted = true; +} +/** * pci_ext_cfg_is_aliased - Is ext config space just an alias of std config? * @dev: PCI device
@@ -1638,6 +1651,8 @@ int pci_setup_device(struct pci_dev *dev)/* Need to have dev->cfg_size ready */ set_pcie_thunderbolt (dev); +set_pcie_untrusted (dev); +/* "Unknown power state" */ dev->current_state =PCI_UNKNOWN ;
include/linux/pci.h +8 -0
@@ -396,6 +396,14 @@ struct pci_dev { unsigned int is_hotplug_bridge:1; 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 +* 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 __aer_firmware_first_valid:1; unsigned int __aer_firmware_first:1; unsigned int broken_intx_masking:1; /* INTx masking can't be used */
Signature · 1 line
2.19.2
Re: [PATCH v3 1/4] PCI / ACPI: Identify untrusted PCI devices
Quoted 22 lines · Mika Westerberg
A malicious PCI device may use DMA to attack the system. An external Thunderbolt port is a convenient point to attach such a device. The OS may use IOMMU to defend against DMA attacks.
Recent BIOSes with Thunderbolt ports mark these externally facing root ports with this ACPI
_DSD[1]:Name (_DSD, Package () { ToUUID ("efcc06cc-73ac-4bc3-bff0-76143807c389"), Package () { Package () {"ExternalFacingPort", 1}, Package () {"UID", 0 } } })If we find such a root port, mark it and all its children as untrusted. The rest of the OS may use this information to enable DMA protection against malicious devices. For instance the device may be put behind an IOMMU to keep it from accessing memory outside of what the driver has allocated for it.
Quoted 6 lines
While at it, add a comment on top of
prp_guidsarray explaining the possible caveat resulting when these GUIDs are treated equivalent.
Quoted 128 lines
--- drivers/acpi/property.c | 11 +++++++++++ drivers/pci/pci-acpi.c | 19 +++++++++++++++++++ drivers/pci/probe.c | 15 +++++++++++++++ include/linux/pci.h | 8 ++++++++ 4 files changed, 53 insertions(+)drivers/acpi/property.c +11
-0@@ -24,6 +24,14 @@ static int acpi_data_get_property_array(const struct acpi_device_data *data,acpi_object_type type, const union acpi_object **obj); +/* +* The GUIDs here are made equivalent to each other in order to avoid extra +* complexity in the properties handling code, with the caveat that the +* kernel will accept certain combinations of GUID and properties that are +* not defined without a warning. For instance if any of the properties +* from different GUID appear in a property list of another, it will be +* accepted by the kernel. Firmware validation tools should catch these. +*/ static const guid_t prp_guids[] = {/* ACPI _DSD device properties GUID: daffd814-6eba-4d8c-8a91-bc9bbf4aa301 */ GUID_INIT(0xdaffd814, 0x6eba, 0x4d8c,@@ -31,6 +39,9 @@ static const guid_t prp_guids[] = {/* Hotplug in D3 GUID: 6211e2c0-58a3-4af3-90e1-927a4e0c55a4 */ GUID_INIT (0x6211e2c0 ,0x58a3 ,0x4af3 ,0x90 ,0xe1 ,0x92 ,0x7a ,0x4e ,0x0c ,0x55 ,0xa4 ), +/* External facing port GUID: efcc06cc-73ac-4bc3-bff0-76143807c389 */ +GUID_INIT (0xefcc06cc ,0x73ac ,0x4bc3 , + 0xbf, 0xf0, 0x76, 0x14, 0x38, 0x07, 0xc3, 0x89), }; static const guid_t ads_guid =drivers/pci/pci-acpi.c +19
-0@@ -789,6 +789,24 @@ static void pci_acpi_optimize_delay(struct pci_dev *pdev,ACPI_FREE (obj); } +static void pci_acpi_set_untrusted (struct pci_dev *dev) +{ +u8 val; + +if (pci_pcie_type (dev) !=PCI_EXP_TYPE_ROOT_PORT ) +return ; +if (device_property_read_u8 (&dev->dev,"ExternalFacingPort" , &val)) +return ; + +/* +* These root ports expose PCIe (including DMA) outside of the +* system so make sure we treat them and everything behind as +* untrusted. +*/ +if (val) + dev->untrusted =1 ; +} +static void pci_acpi_setup (struct device *dev) {struct pci_dev *pci_dev =to_pci_dev (dev);@@ -798,6 +816,7 @@ static void pci_acpi_setup(struct device *dev)return ;pci_acpi_optimize_delay (pci_dev, adev->handle); +pci_acpi_set_untrusted (pci_dev);pci_acpi_add_pm_notifier (adev, pci_dev);if (!adev->wakeup.flags.valid)drivers/pci/probe.c +15
-0@@ -1378,6 +1378,19 @@ static void set_pcie_thunderbolt(struct pci_dev *dev) } } +static void set_pcie_untrusted (struct pci_dev *dev) +{ +struct pci_dev *parent; + +/* +* If the upstream bridge is untrusted we treat this device +* untrusted as well. +*/ + parent =pci_upstream_bridge (dev); +if (parent && parent->untrusted) + dev->untrusted = true; +} +/** * pci_ext_cfg_is_aliased - Is ext config space just an alias of std config? * @dev: PCI device @@ -1638,6 +1651,8 @@ int pci_setup_device(struct pci_dev *dev)/* Need to have dev->cfg_size ready */ set_pcie_thunderbolt (dev); +set_pcie_untrusted (dev); +/* "Unknown power state" */ dev->current_state =PCI_UNKNOWN ;include/linux/pci.h +8
-1@@ -396,6 +396,14 @@ struct pci_dev { unsigned int is_hotplug_bridge:1; 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 +* 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 __aer_firmware_first_valid:1; unsigned int __aer_firmware_first:1; unsigned int broken_intx_masking:1; /* INTx masking can't be used */ --2.19.2
Re: [PATCH v3 1/4] PCI / ACPI: Identify untrusted PCI devices
In reply to Mika WesterbergQuoted 6 lines · Mika Westerberg
A malicious PCI device may use DMA to attack the system. An external Thunderbolt port is a convenient point to attach such a device. The OS may use IOMMU to defend against DMA attacks.
Recent BIOSes with Thunderbolt ports mark these externally facing root ports with this ACPI
_DSD[1]:
I'm not 100% comfortable with the "Recent BIOSes" wording because that suggests that we can rely on the fact that *all* BIOSes newer than some date X mark these ports.
Since this _DSD usage is Microsoft-specific and not required by either PCIe or ACPI specs, we can't rely on it. A BIOS that doesn't implement it may not be Windows-certified, but it's perfectly spec-compliant otherwise and we have to keep in mind the possibility that ports without this _DSD may still be externally visible and may still be attack vectors.
Quoted 14 lines
Name (_DSD, Package () { ToUUID ("efcc06cc-73ac-4bc3-bff0-76143807c389"), Package () { Package () {"ExternalFacingPort", 1}, Package () {"UID", 0 } }})
If we find such a root port, mark it and all its children as untrusted. The rest of the OS may use this information to enable DMA protection against malicious devices. For instance the device may be put behind an IOMMU to keep it from accessing memory outside of what the driver has allocated for it.
Quoted 6 lines
While at it, add a comment on top of
prp_guidsarray explaining the possible caveat resulting when these GUIDs are treated equivalent.
Quoted 128 lines
--- drivers/acpi/property.c | 11 +++++++++++ drivers/pci/pci-acpi.c | 19 +++++++++++++++++++ drivers/pci/probe.c | 15 +++++++++++++++ include/linux/pci.h | 8 ++++++++ 4 files changed, 53 insertions(+)drivers/acpi/property.c +11
-0@@ -24,6 +24,14 @@ static int acpi_data_get_property_array(const struct acpi_device_data *data,acpi_object_type type, const union acpi_object **obj); +/* +* The GUIDs here are made equivalent to each other in order to avoid extra +* complexity in the properties handling code, with the caveat that the +* kernel will accept certain combinations of GUID and properties that are +* not defined without a warning. For instance if any of the properties +* from different GUID appear in a property list of another, it will be +* accepted by the kernel. Firmware validation tools should catch these. +*/ static const guid_t prp_guids[] = {/* ACPI _DSD device properties GUID: daffd814-6eba-4d8c-8a91-bc9bbf4aa301 */ GUID_INIT(0xdaffd814, 0x6eba, 0x4d8c,@@ -31,6 +39,9 @@ static const guid_t prp_guids[] = {/* Hotplug in D3 GUID: 6211e2c0-58a3-4af3-90e1-927a4e0c55a4 */ GUID_INIT (0x6211e2c0 ,0x58a3 ,0x4af3 ,0x90 ,0xe1 ,0x92 ,0x7a ,0x4e ,0x0c ,0x55 ,0xa4 ), +/* External facing port GUID: efcc06cc-73ac-4bc3-bff0-76143807c389 */ +GUID_INIT (0xefcc06cc ,0x73ac ,0x4bc3 , + 0xbf, 0xf0, 0x76, 0x14, 0x38, 0x07, 0xc3, 0x89), }; static const guid_t ads_guid =drivers/pci/pci-acpi.c +19
-0@@ -789,6 +789,24 @@ static void pci_acpi_optimize_delay(struct pci_dev *pdev,ACPI_FREE (obj); } +static void pci_acpi_set_untrusted (struct pci_dev *dev) +{ +u8 val; + +if (pci_pcie_type (dev) !=PCI_EXP_TYPE_ROOT_PORT ) +return ; +if (device_property_read_u8 (&dev->dev,"ExternalFacingPort" , &val)) +return ; + +/* +* These root ports expose PCIe (including DMA) outside of the +* system so make sure we treat them and everything behind as +* untrusted. +*/ +if (val) + dev->untrusted =1 ; +} +static void pci_acpi_setup (struct device *dev) {struct pci_dev *pci_dev =to_pci_dev (dev);@@ -798,6 +816,7 @@ static void pci_acpi_setup(struct device *dev)return ;pci_acpi_optimize_delay (pci_dev, adev->handle); +pci_acpi_set_untrusted (pci_dev);pci_acpi_add_pm_notifier (adev, pci_dev);if (!adev->wakeup.flags.valid)drivers/pci/probe.c +15
-0@@ -1378,6 +1378,19 @@ static void set_pcie_thunderbolt(struct pci_dev *dev) } } +static void set_pcie_untrusted (struct pci_dev *dev) +{ +struct pci_dev *parent; + +/* +* If the upstream bridge is untrusted we treat this device +* untrusted as well. +*/ + parent =pci_upstream_bridge (dev); +if (parent && parent->untrusted) + dev->untrusted = true; +} +/** * pci_ext_cfg_is_aliased - Is ext config space just an alias of std config? * @dev: PCI device @@ -1638,6 +1651,8 @@ int pci_setup_device(struct pci_dev *dev)/* Need to have dev->cfg_size ready */ set_pcie_thunderbolt (dev); +set_pcie_untrusted (dev); +/* "Unknown power state" */ dev->current_state =PCI_UNKNOWN ;include/linux/pci.h +8
-0@@ -396,6 +396,14 @@ struct pci_dev { unsigned int is_hotplug_bridge:1; 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 +* 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 __aer_firmware_first_valid:1; unsigned int __aer_firmware_first:1; unsigned int broken_intx_masking:1; /* INTx masking can't be used */Signature · 1 line
2.19.2
Re: [PATCH v3 1/4] PCI / ACPI: Identify untrusted PCI devices
Quoted 12 lines · Bjorn Helgaas
Quoted 6 lines · Mika Westerberg
A malicious PCI device may use DMA to attack the system. An external Thunderbolt port is a convenient point to attach such a device. The OS may use IOMMU to defend against DMA attacks.
Recent BIOSes with Thunderbolt ports mark these externally facing root ports with this ACPI
_DSD[1]:I'm not 100% comfortable with the "Recent BIOSes" wording because that suggests that we can rely on the fact that *all* BIOSes newer than some date X mark these ports.
Quoted 6 lines
Since this
_DSDusage is Microsoft-specific and not required by either PCIe or ACPI specs, we can't rely on it. A BIOS that doesn't implement it may not be Windows-certified, but it's perfectly spec-compliant otherwise and we have to keep in mind the possibility that ports without this_DSDmay still be externally visible and may still be attack vectors.
OK.
I will change it to "Some BIOSes .." following what you suggested earlier. That should make it clear not all BIOSes are required to implement this.
Quoted 14 lines
Quoted 14 lines
Name (_DSD, Package () { ToUUID ("efcc06cc-73ac-4bc3-bff0-76143807c389"), Package () { Package () {"ExternalFacingPort", 1}, Package () {"UID", 0 } }})
If we find such a root port, mark it and all its children as untrusted. The rest of the OS may use this information to enable DMA protection against malicious devices. For instance the device may be put behind an IOMMU to keep it from accessing memory outside of what the driver has allocated for it.
Quoted 8 lines
Quoted 6 lines
While at it, add a comment on top of
prp_guidsarray explaining the possible caveat resulting when these GUIDs are treated equivalent.Signed-off-by Mika Westerberg <[email protected]>Acked-by Bjorn Helgaas <[email protected]>
Thanks!
PATCH v3 2/4 iommu/vt-d: Force IOMMU on for platform opt in hint
In reply to Mika WesterbergFrom: Lu Baolu <[email protected]>
Intel VT-d spec added a new DMA_CTRL_PLATFORM_OPT_IN_FLAG flag in DMAR ACPI table [1] for BIOS to report compliance about platform initiated DMA restricted to RMRR ranges when transferring control to the OS. This means that during OS boot, before it enables IOMMU none of the connected devices can bypass DMA protection for instance by overwriting the data structures used by the IOMMU. The OS also treats this as a hint that the IOMMU should be enabled to prevent DMA attacks from possible malicious devices.
A use of this flag is Kernel DMA protection for Thunderbolt [2] which in practice means that IOMMU should be enabled for PCIe devices connected to the Thunderbolt ports. With IOMMU enabled for these devices, all DMA operations are limited in the range reserved for it, thus the DMA attacks are prevented. All these devices are enumerated in the PCI/PCIe module and marked with an untrusted flag.
This forces IOMMU to be enabled if DMA_CTRL_PLATFORM_OPT_IN_FLAG is set in DMAR ACPI table and there are PCIe devices marked as untrusted in the system. This can be turned off by adding "intel_iommu=off" in the kernel command line, if any problems are found.
[1] https://software.intel.com/sites/default/files/managed/c5/15/vt-directed-io-spec.pdf
[2] https://docs.microsoft.com/en-us/windows/security/information-protection/kernel-dma-protection-for-thunderbolt
--- drivers/iommu/dmar.c | 25 +++++++++++++++++ drivers/iommu/intel-iommu.c | 53 +++++++++++++++++++++++++++++++++++-- include/linux/dmar.h | 8 ++++++ 3 files changed, 84 insertions(+), 2 deletions(-)
drivers/iommu/dmar.c +25 -0
@@ -2042,3 +2042,28 @@ int dmar_device_remove(acpi_handle handle) {return dmar_device_hotplug (handle, false); } + +/* +* dmar_platform_optin - Is %DMA_CTRL_PLATFORM_OPT_IN_FLAG set in DMAR table +* +* Returns true if the platform has %DMA_CTRL_PLATFORM_OPT_IN_FLAG set in +* the ACPI DMAR table. This means that the platform boot firmware has made +* sure no device can issue DMA outside of RMRR regions. +*/ +bool dmar_platform_optin (void ) +{ +struct acpi_table_dmar *dmar; +acpi_status status; +bool ret; + + status =acpi_get_table (ACPI_SIG_DMAR ,0 , + (struct acpi_table_header **)&dmar); +if (ACPI_FAILURE (status)) +return false; + + ret = !!(dmar->flags &DMAR_PLATFORM_OPT_IN ); +acpi_put_table ((struct acpi_table_header *)dmar); + +return ret; +} +EXPORT_SYMBOL_GPL (dmar_platform_optin);
drivers/iommu/intel-iommu.c +51 -2
83 lines not on this page · raw message
include/linux/dmar.h +8 -0
@@ -39,6 +39,7 @@ struct acpi_dmar_header;/* DMAR Flags */ #define DMAR_INTR_REMAP 0x1#define DMAR_X2APIC_OPT_OUT 0x2 +#define DMAR_PLATFORM_OPT_IN 0x4struct intel_iommu ;
@@ -170,6 +171,8 @@ static inline int dmar_ir_hotplug(struct dmar_drhd_unit *dmaru, bool insert) {return 0 ; }#endif /* CONFIG_IRQ_REMAP */ +extern bool dmar_platform_optin (void ); +#else /* CONFIG_DMAR_TABLE */ static inline int dmar_device_add (void *handle)
@@ -182,6 +185,11 @@ static inline int dmar_device_remove(void *handle)return 0 ; } +static inline bool dmar_platform_optin (void ) +{ +return false; +} +#endif /* CONFIG_DMAR_TABLE */ struct irte {
Signature · 1 line
2.19.2
PATCH v3 3/4 iommu/vt-d: Do not enable ATS for untrusted devices
In reply to Mika WesterbergCurrently Linux automatically enables ATS (Address Translation Service) for any device that supports it (and IOMMU is turned on). ATS is used to accelerate DMA access as the device can cache translations locally so there is no need to do full translation on IOMMU side. However, as pointed out in [1] ATS can be used to bypass IOMMU based security completely by simply sending PCIe read/write transaction with AT (Address Translation) field set to "translated".
To mitigate this modify the Intel IOMMU code so that it does not enable ATS for any device that is marked as being untrusted. In case this turns out to cause performance issues we may selectively allow ATS based on user decision but currently use big hammer and disable it completely to be on the safe side.
[1] https://www.repository.cam.ac.uk/handle/1810/274352
--- drivers/iommu/intel-iommu.c | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-)
drivers/iommu/intel-iommu.c +2 -1
@@ -1473,7 +1473,8 @@ static void iommu_enable_dev_iotlb(struct device_domain_info *info)if (info->pri_supported && !pci_reset_pri (pdev) && !pci_enable_pri (pdev,32 )) info->pri_enabled =1 ;#endif -if (info->ats_supported && !pci_enable_ats (pdev,VTD_PAGE_SHIFT )) { +if (!pdev->untrusted && info->ats_supported && + !pci_enable_ats (pdev,VTD_PAGE_SHIFT )) { info->ats_enabled =1 ;domain_update_iotlb (info->domain); info->ats_qdep =pci_ats_queue_depth (pdev);
Signature · 1 line
2.19.2
PATCH v3 4/4 thunderbolt: Export IOMMU based DMA protection support to userspace
In reply to Mika WesterbergRecent systems with Thunderbolt ports may support IOMMU natively. In practice this means that Thunderbolt connected devices are placed behind an IOMMU during the whole time it is connected (including during boot) making Thunderbolt security levels redundant. This is called Kernel DMA protection [1] by Microsoft.
Some of these systems still have Thunderbolt security level set to "user" in order to support OS downgrade (the older version of the OS might not support IOMMU based DMA protection so connecting a device still relies on user approval).
Export this information to userspace by introducing a new sysfs attribute (iommu_dma_protection). Based on it userspace tools can make more accurate decision whether or not authorize the connected device.
In addition update Thunderbolt documentation regarding IOMMU based DMA protection.
--- .../ABI/testing/sysfs-bus-thunderbolt | 9 +++++++++ Documentation/admin-guide/thunderbolt.rst | 20 +++++++++++++++++++ drivers/thunderbolt/domain.c | 17 ++++++++++++++++ 3 files changed, 46 insertions(+)
Documentation/ABI/testing/sysfs-bus-thunderbolt +9 -0
@@ -21,6 +21,15 @@ Description: Holds a comma separated list of device unique_ids that If a device is authorized automatically during boot its boot attribute is set to 1. +What: /sys/bus/thunderbolt/devices/.../domainX/iommu_dma_protection +Date: Mar 2019 +KernelVersion: 4.21 +Contact: [email protected] +Description: This attribute tells whether the system uses IOMMU + for DMA protection. Value of 1 means IOMMU is used 0 means + it is not (DMA protection is solely based on Thunderbolt + security levels). + What: /sys/bus/thunderbolt/devices/.../domainX/security Date: Sep 2017 KernelVersion: 4.13
Documentation/admin-guide/thunderbolt.rst +20 -0
@@ -133,6 +133,26 @@ If the user still wants to connect the device they can either approve the device without a key or write a new key and write 1 to the ``authorized`` file to get the new key stored on the device NVM. +DMA protection utilizing IOMMU +------------------------------ +Recent systems from 2018 and forward with Thunderbolt ports may natively +support IOMMU. This means that Thunderbolt security is handled by an IOMMU +so connected devices cannot access memory regions outside of what is +allocated for them by drivers. When Linux is running on such system it +automatically enables IOMMU if not enabled by the user already. These +systems can be identified by reading ``1`` from +``/sys/bus/thunderbolt/devices/domainX/iommu_dma_protection`` attribute. + +The driver does not do anything special in this case but because DMA +protection is handled by the IOMMU, security levels (if set) are +redundant. For this reason some systems ship with security level set to +``none``. Other systems have security level set to ``user`` in order to +support downgrade to older OS, so users who want to automatically +authorize devices when IOMMU DMA protection is enabled can use the +following ``udev`` rule:: + + ACTION=="add", SUBSYSTEM=="thunderbolt", ATTRS{iommu_dma_protection}=="1", ATTR{authorized}=="0", ATTR{authorized}="1" + Upgrading NVM on Thunderbolt device or host ------------------------------------------- Since most of the functionality is handled in firmware running on a
drivers/thunderbolt/domain.c +17 -0
@@ -7,7 +7,9 @@*/ #include <linux/device.h> +#include <linux/dmar.h> #include <linux/idr.h> +#include <linux/iommu.h> #include <linux/module.h> #include <linux/pm_runtime.h> #include <linux/slab.h>
@@ -236,6 +238,20 @@ static ssize_t boot_acl_store(struct device *dev, struct device_attribute *attr, }static DEVICE_ATTR_RW (boot_acl); +static ssize_t iommu_dma_protection_show (struct device *dev, +struct device_attribute *attr, +char *buf) +{ +/* +* Kernel DMA protection is a feature where Thunderbolt security is +* handled natively using IOMMU. It is enabled when IOMMU is +* enabled and ACPI DMAR table has DMAR_PLATFORM_OPT_IN set. +*/ +return sprintf (buf,"%d\n" , +iommu_present (&pci_bus_type) &&dmar_platform_optin ()); +} +static DEVICE_ATTR_RO (iommu_dma_protection); +static ssize_t security_show (struct device *dev,struct device_attribute *attr,char *buf) {
@@ -251,6 +267,7 @@ static DEVICE_ATTR_RO(security);static struct attribute *domain_attrs[] = { &dev_attr_boot_acl.attr, + &dev_attr_iommu_dma_protection.attr, &dev_attr_security.attr,NULL , };
Signature · 1 line
2.19.2
Mika Westerberg
AppliedQuoted 4 lines · Mika Westerberg
Recent systems with Thunderbolt ports may be utilizing IOMMU to prevent DMA attacks. This is different from the previous security level based scheme because the connected device cannot access system memory outside of the regions allocated for it by the driver.
Applied all to thunderbolt.git/next with acks from Bjorn and Rafael.