On 12/11/24 1:36 AM, Nitesh Shetty wrote:
Block layer can allocate a buffer and send this as part of copy
operation.
The block layer can only do that if it knows how large the buffer should
be. Or in other words, if knowledge of a SCSI buffer size is embedded in
the block layer. That doesn't sound ideal to me.
This scheme will require sequential submission of SRC and DST
bio's. This might increase in latency, but allows to have simpler design.
Main use case for copy is GC, which is mostly a background operation.
I still prefer a single REQ_OP_COPY operation instead of separate
REQ_OP_COPY_SRC and REQ_OP_COPY_DST operations. While this will require
additional work in the SCSI disk (sd) driver (implementation of a state
machine), it prevents that any details about the SCSI copy offloading
approach have to be known by the block layer. Even if copy offloading
would be implemented as two operations (REQ_OP_COPY_SRC and
REQ_OP_COPY_DST), a state machine is required anyway in the SCSI disk
driver because REQ_OP_COPY_SRC would have to be translated into two SCSI
commands (POPULATE TOKEN + RECEIVE ROD TOKEN INFORMATION).
Thanks,
Bart.
