On 19/05/2023 17:54, Bart Van Assche wrote:
On 5/19/23 09:06, John Garry wrote:
Sure, what I describe is ideal,
* not ideal
To be clear, I mean something like:
struct scsi_exec_args { unsigned char **sense; }
scsi_execute_cmd() { ... *args->sense = kmemdup(scsi_cmd->sense_buffer); ... }
some_func() { unsigned char *sense = NULL; struct scsi_exec_args = { .sense = &sense, };
ret = scsi_execute_cmd(); if (ret < 0) return ret; kfree(sense); }
But not perfect as we need a separate small buffer for sensehdr and we need to always kfree those buffers.
If only we could pass the actual scsi_cmnd sense buffer to the caller...
but I still just dislike passing both
sensebuf and hdr into scsi_execute_cmd(). The semantics of how scsi_execute_cmd() treats them is vague.
Is this something that can be addressed by improving the scsi_execute_cmd() documentation?
Hmmm, I'm not sure documentation helps too much avoiding all programming errors and better make the code foolproof.
Anyway, if we fix up the callers of scsi_execute_cmd() to properly check for errors then if is not such a big deal.
Thanks, John