binarystore: Fix for newer boost

The API for .data() on endian types changes from returning `char *` to
`unsigned char *`. This fixes the library to build against both APIs.

Change-Id: I432769e3a3992678becb6121de89d7c00f7c2d85
Signed-off-by: William A. Kennington III <wak@google.com>
1 file changed
tree: f6bf3331a30e5e28828a13af43cbaa2c12ca4cae
  1. proto/
  2. test/
  3. .clang-format
  4. .gitignore
  5. binarystore.cpp
  6. binarystore.hpp
  7. binarystore_mock.hpp
  8. bootstrap.sh
  9. configure.ac
  10. handler.cpp
  11. handler.hpp
  12. LICENSE
  13. main.cpp
  14. MAINTAINERS
  15. Makefile.am
  16. parse_config.hpp
  17. README.md
  18. sys.cpp
  19. sys.hpp
  20. sys_file.cpp
  21. sys_file.hpp
README.md

BMC Generic Binary Blob Store via OEM IPMI Blob Transport

The BMC generic IPMI blobs binary store, or "binary store" in short, serves a simple purpose: provide a read/write/serialize abstraction layer through IPMI blobs transport layer to allow users to store binary data on persistent locations accessible to the BMC.

Despite its name, the binary blob store cannot be used for everything.

  • It is not a host/BMC mailbox. In general, BMC should reserve the space for blob store and not try to write it due to concurrency concerns. It is expected the only accessors are the IPMI blob store commands.
  • It is not a key store. Because the data stored is accessible to IPMI users and is not encrypted in anyway, extra caution should be used. It is not meant for storing data containing any keys or secrets.
  • The data to be stored should not be too large. Since each IPMI packet is limited in size, trying to send an overly large binary is going to take too long, and the overhead of the IPMI transport layer might make it not worthwhile.

Background and References

Please read the IPMI Blob protocol design as primer here.

Under the hood, the binary blobs are stored as a binary protocol buffer, or "protobuf" in short.

Requirements

  1. BMC image should have phosphor-ipmi-blobs installed.
  2. The host should know the specific blob base id that it intends to operate on. For this design the discovery operations are limited.
  3. The host should only store binary data that is suitable using this transfer mechanism. As mentioned it is not meant for mailbox communication, key store, or large binary data.

Specification

This section describes how the handler phosphor-ipmi-blobs-binarystore defines each handler of the IPMI Blob protocol.

Blob ID Definition

A "blob id" is a unique string that identifies a blob. Binary Store handler may show two kinds of blob ids: "base id" and "file id". They should only contain ASCII alphanumerical characters and forward slashes ('/').

A "base id" begins and ends with a forward slash. It is analagous to a unix directory path. The binary store handler will assign each storage location a unique base id (See next section for details).

A "file id" begins with a forward slash but must not have a slash at the end, and is analagous to a unix file path. Any file id with a valid base id as its longest matching prefix is considered reserved as a binary blob in the storage space.

For example, if /test/ and /test/blob0 are the initial blob ids, we know there is one binary store location with one blob already created. To create a new blob named /test/blob1, one simply calls open with the id and write/commit with the returned session id. Opening invalid ids such as /foo/bar or /test/nested/dir will fail.

Platform Configuration

For the binary store handler, a configuration file provides the base id, which file and which offset in the file to store the data. Optionally a "max_size" param can be specified to indicate the total size of such binary storage should not exceed the limitation. If "max_size" is specified as -1 or not specified, the storage could grow up to what the physical media allows.

base_id: /bmc_store/
sysfile_path: /sys/class/i2c-dev/i2c-1/device/1-0050/eeprom
offset: 256
max_size: 1024

[1] Example Configuration

Binary Store Protobuf Definition

The data is stored as a binary protobuf containing a variable number of binary blobs, each having a unique blob_id string with the base id as a common prefix.

message BinaryBlob {
  optional string blob_id = 1;
  optional bytes data = 2;
}

message BinaryBlobStore {
  optional string blob_base_id = 1;
  repeated BinaryBlob blob = 2;
  optional uint32 max_size = 3;
  optional string sysfile_path = 4;
  optional uint32 offset = 5;
}

Storing data as a protobuf makes the format more flexible and expandable, and allows future modifications to the storage format.

IPMI Blob Transfer Command Primitives

The binary store handler will implement the following primitives:

BmcBlobGetCount/BmcBlobEnumerate

Initially only the base id will appear when enumerating the existing blobs. Once a valid binary blob has been created, its blob id will appear in the list.

BmcBlobOpen

flags can be READ for read-only access or READ|WRITE. blob_id can be any string with a matching prefix. If there is not already a valid binary stored with supplied blob_id, the handler treats it as a request to create such a blob.

The session_id returned should be used by the rest of the session based commands to operate on the blob. If there is already an open session, this call will fail.

NOTE: the newly created blob is not serialized and stored until BmcBlobCommit is called.

BmcBlobRead

Returns bytes with the requested offset and size. If there are not enough bytes the handler will return the bytes that are available.

Note this operation reads from memory. Make sure the stat is 'COMMITTED' which indicates that the memory content matches the data serialized to storage.

BmcBlobWrite

Writes bytes to the requested offset. Return number of bytes written if success, zero if failed. If not all of the bytes can be written, this operation will fail.

BmcBlobCommit

Store the serialized BinaryBlobStore to the associated system file.

BmcBlobClose

Mark the session as closed. Any uncommitted changes to the blob state is lost.

BmcBlobDelete

Delete the binary data associated with blob_id. Must operate on an open blob. Deleting the base_id (the 'directory' level) will fail harmlessly.

BmcBlobStat

size returned equals to length of the data bytes in the protobuf. blob_state will be set with OPEN_R, OPEN_W, and/or COMMITTED as appropriate.

BmcBlobSessionStat/BmcBlobWriteMeta

Not supported.

Example Host Command Flow

No binary data yet, write data

  1. BmcBlobGetCount followed by BmcBlobEnumerate. Since there is no valid blob with binary data stored, BMC handler only populates the base_id per platform configuration. e.g. /bmc_store/.
  2. BmcBlobOpen with blob_id = /bmc_store/blob0, BMC honors the request and returns session_id = 0.
  3. BmcBlobWrite multiple times to write the data into the blob.
  4. BmcBlobCommit. BMC writes data into configured path, e.g. to EEPROM.
  5. BmcBlobClose

Read existing data

  1. BmcBlobGetCount followed by BmcBlobEnumrate shows /bmc_store/ and /bmc_store/blob0.
  2. BmcBlobStat on /bmc_store/blob0 shows non-zero size and COMMITTED state.
  3. BmcBlobOpen with blob_id = /bmc_store/blob0.
  4. BmcBlobRead multiple times to read the data.
  5. BmcBlobClose.

Alternatives Considered

The first alternative considered was to store the data via IPMI FRU commands; as mentioned in the problem description, it is not always viable.

There is a Google OEM I2C-over-IPMI driver that allows the host to read/write I2C devices attached to the BMC. In comparison, the blob store approach proposed offer more abstraction and is more flexible in where to store the data.