Messages in this thread | | | From | (Alan Cox) | Subject | Re: Thread implementations...t | Date | Thu, 25 Jun 1998 21:23:31 +0100 (BST) |
| |
> Unless you need copying small fragments (let's say few pages) from one > file to another. In such cases mmap/munmap is just too much overhead.
In the case where mmap is too much overhead so is the locking for a kernel "copy*" operation quite honestly. What are the semantics of send_file when some other process is DMAing into the buffer cache page we have queued on the socket ?
> > Now the reason to have it is for NFS type systems that support copy > > and copytree operations directly on the remote (eg SMB) > > Or maybe SCSI block copy commands, but it would be hard to implement > for anything else than block devices.
SCSI block copy worries me. Nobody I know uses disk->disk SCSI facilities which makes me suspect if we implement it drive firmware will die left right and centre.
Alan [he of little faith]
- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.rutgers.edu
| |