On 11/16/2011 04:31 PM, Christoph Hellwig wrote:
> On Wed, Nov 16, 2011 at 04:11:36PM +0800, Liu Yuan wrote:
>> +struct store_driver {
>> + const char *driver_name;
>> + int (*init)(char *path);
>> + int (*open)(uint32_t epoch, uint64_t oid, int flags, int *ret);
>> + ssize_t (*write)(uint64_t oid, int fd, void *buf, int len, off_t offset);
>> + ssize_t (*read)(uint64_t oid, int fd, void *buf, int len, off_t offset);
>> + int (*close)(int fd);
>> +};
>
> Passing an fd round seems to make the interface fairly non-generic.
>
> A void pointer would allow the driver to store whatever it needs
> there (and simple drivers can still cast the fd into them without memory
> allocation).
>
Okay. I don't like fd either. Even more, I don't like open/rw/close
group to interface with a kv-store.
Void pointer is a nice way to return something back if store driver
doesn't care fd.
Anyway, I hope this interface is just provisional.
> Also in the open routine I would:
>
> - return the return value, and the private data in the arguments, that
> makes the code a bit more readable
> - not pass posix open flags, but your own. It seems like the only one
> for now would be the create new file flag.
> - shouldn't the check for short writes, and the error code translation
> side inside the write method?
>
Okay, I'll submit V2.
Thanks,
Yuan
|
|