Message ID | cover.1719771133.git.lukas@wunner.de |
---|---|
Headers | show |
Series | PCI device authentication | expand |
On 1/7/24 05:35, Lukas Wunner wrote: > PCI device authentication v2 > > Authenticate PCI devices with CMA-SPDM (PCIe r6.2 sec 6.31) and > expose the result in sysfs. What is it based on? I am using https://github.com/l1k/linux.git branch cma_v2 for now but wonder if that's the right one. Thanks, > > Five big changes since v1 (and many smaller ones, full list at end): > > * Certificates presented by a device are now exposed in sysfs > (new patch 12). > > * Per James Bottomley's request at Plumbers, a log of signatures > received from a device is exposed in sysfs (new patches 13-18), > allowing for re-verification by remote attestation services. > Comments welcome whether the proposed ABI makes sense. > > * Per Damien Le Moal's request at Plumbers, sysfs attributes are > now implemented in the SPDM library instead of in the PCI core. > Thereby, ATA and SCSI will be able to re-use them seamlessly. > > * I've dropped a controversial patch to grant guests exclusive control > of authentication of passed-through devices (old patch 12 in v1). > People were more interested in granting the TSM exclusive control > instead of the guest. Dan Williams is driving an effort to negotiate > SPDM control between kernel and TSM. > > * The SPDM library (in patch 7) has undergone significant changes > to enable the above-mentioned sysfs exposure of certificates and > signatures: It retrieves and caches all certificates from a device > and collects all exchanged SPDM messages in a transcript buffer. > To ease future maintenance, the code has been split into multiple > files in lib/spdm/. > > > Link to v1 and subsequent Plumbers discussion: > https://lore.kernel.org/all/cover.1695921656.git.lukas@wunner.de/ > https://lpc.events/event/17/contributions/1558/ > > How to test with qemu: > https://github.com/twilfredo/qemu-spdm-emulation-guide > > > Changes v1 -> v2: > > * [PATCH 01/18] X.509: Make certificate parser public > * Add include guard #ifndef + #define to <keys/x509-parser.h> (Ilpo). > > * [PATCH 02/18] X.509: Parse Subject Alternative Name in certificates > * Return -EBADMSG instead of -EINVAL on duplicate Subject Alternative > Name, drop error message for consistency with existing code. > > * [PATCH 03/18] X.509: Move certificate length retrieval into new helper > * Use ssize_t instead of int (Ilpo). > * Amend commit message to explain why the helper is exported (Dan). > > * [PATCH 06/18] crypto: ecdsa - Support P1363 signature encoding > * Use idiomatic &buffer[keylen] notation. > * Rebase on NIST P521 curve support introduced with v6.10-rc1 > > * [PATCH 07/18] spdm: Introduce library to authenticate devices > New features: > * In preparation for exposure of certificate chains in sysfs, retrieve > the certificates from *all* populated slots instead of stopping on > the first valid slot. Cache certificate chains in struct spdm_state. > * Collect all exchanged messages of an authentication sequence in a > transcript buffer for exposure in sysfs. Compute hash over this > transcript rather than peacemeal over each exchanged message. > * Support NIST P521 curve introduced with v6.10-rc1. > Bugs: > * Amend spdm_validate_cert_chain() to cope with zero length chain. > * Print correct error code returned from x509_cert_parse(). > * Emit error if there are no common supported algorithms. > * Implicitly this causes an error if responder selects algorithms > not supported by requester during NEGOTIATE_ALGORITHMS exchange, > previously this was silently ignored (Jonathan). > * Refine checks of Basic Constraints and Key Usage certificate fields. > * Add code comment explaining those checks (Jonathan). > Usability: > * Log informational message on successful authentication (Tomi Sarvela). > Style: > * Split spdm_requester.c into spdm.h, core.c and req-authenticate.c. > * Use __counted_by() in struct spdm_get_version_rsp (Ilpo). > * Return ssize_t instead of int from spdm_transport (Ilpo). > * Downcase hex characters, vertically align SPDM_REQ macro (Ilpo). > * Upcase spdm_error_code enum, vertically align it (Ilpo). > * Return -ECONNRESET instead of -ERESTART from spdm_err() (Ilpo). > * Access versions with le16_to_cpu() instead of get_unaligned_le16() > in spdm_get_version() because __packed attribute already implies > byte-wise access (Ilpo). > * Add code comment in spdm_start_hash() that shash and desc > allocations are freed by spdm_reset(), thus seemingly leaked (Ilpo). > * Rename "s" and "h" members of struct spdm_state to "sig_len" and > "hash_len" for clarity (Ilpo). > * Use FIELD_GET() in spdm_create_combined_prefix() for clarity (Ilpo). > * Add SPDM_NONCE_SZ macro (Ilpo). > * Reorder error path of spdm_authenticate() for symmetry (Jonathan). > * Fix indentation of Kconfig entry (Jonathan). > * Annotate capabilities introduced with SPDM 1.1 (Jonathan). > * Annotate algorithms introduced with SPDM 1.2 (Jonathan). > * Annotate errors introduced with SPDM 1.1 and 1.2 (Jonathan). > * Amend algorithm #ifdef's to avoid trailing "|" (Jonathan). > * Add code comment explaining that some SPDM messages are enlarged > by fields added in new SPDM versions whereas others use reserved > space for new fields (Jonathan). > * Refine code comments on various fields in SPDM messages (Jonathan). > * Duplicate spdm_get_capabilities_reqrsp into separate structs (Jonathan). > * Document SupportedAlgorithms field at end of spdm_get_capabilities_rsp, > introduced with SPDM 1.3 (Jonathan). > * Use offsetofend() rather than offsetof() to set SPDM message size > based on SPDM version (Jonathan). > * Use cleanup.h to unwind heap allocations (Jonathan). > * In spdm_verify_signature(), change code comment to refer to "SPDM 1.0 > and 1.1" instead of "Until SPDM 1.1" (Jonathan). > * Use namespace "SPDM" for exported symbols (Jonathan). > * Drop __spdm_exchange(). > * In spdm_exchange(), do not return an error on truncation of > spdm_header so that callers can take care of it. > * Rename "SPDM_CAPS" macro to "SPDM_REQ_CAPS" to prepare for later > addition of responder support. > * Rename "SPDM_MIN_CAPS" macro to "SPDM_RSP_MIN_CAPS" and > rename "responder_caps" member of struct spdm_state to "rsp_caps". > * Rename "SPDM_REQUESTER" Kconfig symbol to "SPDM". There is actually > no clear-cut separation between requester and responder code because > mutual authentication will require the responder to invoke requester > functions. > * Rename "slot_mask" member of struct spdm_state to "provisioned_slots" > to follow SPDM 1.3 spec language. > > * [PATCH 08/18] PCI/CMA: Authenticate devices on enumeration > * In pci_cma_init(), check whether pci_cma_keyring is an ERR_PTR > rather than checking whether it's NULL. keyring_alloc() never > returns NULL. > * On failure to allocate keyring, emit "PCI: " and ".cma" as part of > error message for clarity (Bjorn). > * Drop superfluous curly braces around two if-blocks (Jonathan, Bjorn). > * Add code comment explaining why spdm_state is kept despite initial > authentication failure (Jonathan). > * Rename PCI_DOE_PROTOCOL_CMA to PCI_DOE_FEATURE_CMA for DOE r1.1 > compliance. > > * [PATCH 09/18] PCI/CMA: Validate Subject Alternative Name in certificates > * Amend commit message with note on Reference Integrity Manifest (Jonathan). > * Amend commit message and code comment with note on PCIe r6.2 changes. > * Add SPDX identifer and IETF copyright to cma.asn1 per section 4 of: > https://trustee.ietf.org/documents/trust-legal-provisions/tlp-5/ > * Pass slot number to ->validate() callback and emit it in error messages. > * Move all of cma-x509.c into cma.c (Bjorn). > > * [PATCH 10/18] PCI/CMA: Reauthenticate devices on reset and resume > * Drop "cma_capable" bit in struct pci_dev and instead check whether > "spdm_state" is a NULL pointer. Only difference: Devices which > didn't support the minimum set of capabilities on enumeration > are now attempted to be reauthenticated. The rationale being that > they may have gained new capabilities due to a runtime firmware update. > * Add kernel-doc for pci_cma_reauthenticate(). > > * [PATCH 11/18] PCI/CMA: Expose in sysfs whether devices are authenticated > * Change write semantics of sysfs attribute such that reauthentication > is triggered by writing "re" (instead of an arbitrary string). > This allows adding other commands down the road. > * Move sysfs attribute from PCI core to SPDM library for reuse by other > bus types such as SCSI/ATA (Damien). > * If DOE or CMA initialization fails, set pci_dev->spdm_state to ERR_PTR > instead of using additional boolean flags. > * Amend commit message to mention downgrade attack prevention (Ilpo, > Jonathan). > * Amend ABI documentation to mention reauthentication after downloading > firmware to an FPGA device. > > * [PATCH 12/18 to 18/18] are new in v2 > > > Jonathan Cameron (2): > spdm: Introduce library to authenticate devices > PCI/CMA: Authenticate devices on enumeration > > Lukas Wunner (16): > X.509: Make certificate parser public > X.509: Parse Subject Alternative Name in certificates > X.509: Move certificate length retrieval into new helper > certs: Create blacklist keyring earlier > crypto: akcipher - Support more than one signature encoding > crypto: ecdsa - Support P1363 signature encoding > PCI/CMA: Validate Subject Alternative Name in certificates > PCI/CMA: Reauthenticate devices on reset and resume > PCI/CMA: Expose in sysfs whether devices are authenticated > PCI/CMA: Expose certificates in sysfs > sysfs: Allow bin_attributes to be added to groups > sysfs: Allow symlinks to be added between sibling groups > PCI/CMA: Expose a log of received signatures in sysfs > spdm: Limit memory consumed by log of received signatures > spdm: Authenticate devices despite invalid certificate chain > spdm: Allow control of next requester nonce through sysfs > > Documentation/ABI/testing/sysfs-devices-spdm | 247 ++++++ > Documentation/admin-guide/sysctl/index.rst | 2 + > Documentation/admin-guide/sysctl/spdm.rst | 33 + > MAINTAINERS | 14 + > certs/blacklist.c | 4 +- > crypto/akcipher.c | 2 +- > crypto/asymmetric_keys/public_key.c | 44 +- > crypto/asymmetric_keys/x509_cert_parser.c | 9 + > crypto/asymmetric_keys/x509_loader.c | 38 +- > crypto/asymmetric_keys/x509_parser.h | 40 +- > crypto/ecdsa.c | 18 +- > crypto/internal.h | 1 + > crypto/rsa-pkcs1pad.c | 11 +- > crypto/sig.c | 6 +- > crypto/testmgr.c | 8 +- > crypto/testmgr.h | 20 + > drivers/pci/Kconfig | 13 + > drivers/pci/Makefile | 4 + > drivers/pci/cma.asn1 | 41 + > drivers/pci/cma.c | 247 ++++++ > drivers/pci/doe.c | 5 +- > drivers/pci/pci-driver.c | 1 + > drivers/pci/pci-sysfs.c | 5 + > drivers/pci/pci.c | 12 +- > drivers/pci/pci.h | 17 + > drivers/pci/pcie/err.c | 3 + > drivers/pci/probe.c | 3 + > drivers/pci/remove.c | 1 + > fs/sysfs/file.c | 69 +- > fs/sysfs/group.c | 33 + > include/crypto/akcipher.h | 10 +- > include/crypto/sig.h | 6 +- > include/keys/asymmetric-type.h | 2 + > include/keys/x509-parser.h | 55 ++ > include/linux/kernfs.h | 2 + > include/linux/oid_registry.h | 3 + > include/linux/pci-doe.h | 4 + > include/linux/pci.h | 16 + > include/linux/spdm.h | 46 ++ > include/linux/sysfs.h | 29 + > lib/Kconfig | 15 + > lib/Makefile | 2 + > lib/spdm/Makefile | 11 + > lib/spdm/core.c | 442 +++++++++++ > lib/spdm/req-authenticate.c | 765 +++++++++++++++++++ > lib/spdm/req-sysfs.c | 619 +++++++++++++++ > lib/spdm/spdm.h | 560 ++++++++++++++ > 47 files changed, 3436 insertions(+), 102 deletions(-) > create mode 100644 Documentation/ABI/testing/sysfs-devices-spdm > create mode 100644 Documentation/admin-guide/sysctl/spdm.rst > create mode 100644 drivers/pci/cma.asn1 > create mode 100644 drivers/pci/cma.c > create mode 100644 include/keys/x509-parser.h > create mode 100644 include/linux/spdm.h > create mode 100644 lib/spdm/Makefile > create mode 100644 lib/spdm/core.c > create mode 100644 lib/spdm/req-authenticate.c > create mode 100644 lib/spdm/req-sysfs.c > create mode 100644 lib/spdm/spdm.h >
On Mon, Jul 08, 2024 at 07:47:51PM +1000, Alexey Kardashevskiy wrote: > On 1/7/24 05:35, Lukas Wunner wrote: > > PCI device authentication v2 > > > > Authenticate PCI devices with CMA-SPDM (PCIe r6.2 sec 6.31) and > > expose the result in sysfs. > > What is it based on? This series is based on v6.10-rc1. I also successfully cherry-picked the patches onto v6.10-rc6 and linux-next 20240628 (no merge conflicts and no issues reported by 0-day). Older kernels than v6.10-rc1 won't work because they're missing ecdsa-nist-p521 support as well as a few preparatory sysfs patches of mine that went into v6.10-rc1. > I am using https://github.com/l1k/linux.git branch cma_v2 for now but wonder > if that's the right one. Yes that's fine. There's now also a kernel.org repository with a testing branch: https://git.kernel.org/pub/scm/linux/kernel/git/devsec/spdm.git/ Future maintenance of the SPDM library is intended to be happening in that repo. I assumed that Bjorn may not be keen on having to deal with SPDM patches forever, so creating a dedicated repo seemed to make sense. Most patches in this series with a "PCI/CMA: " subject actually only change very few lines in the PCI core. The bulk of the changes is in the SPDM library instead. I used that subject merely to highlight that at least an ack from Bjorn is required. The only patches containing PCI core changes to speak of are patches 8, 9, 10. The devsec group (short for Device Security Alphabet Soup) currently only contains the spdm.git repo. Going forward, further repos may be added below the devsec umbrella, such as tsm.git to deal with a vendor-neutral interface between kernel and Trusted Security Module. Thanks, Lukas