William Lee Irwin III <wli@xxxxxxxxxxxxxx> wrote:
>
> sizeof(struct kiocb) is dangerously large for a structure commonly
> allocated on-stack. This patch converts the 24*sizeof(long) field,
> ->private, to a void pointer for use by file_operations entrypoints.
> A ->dtor() method is added to the kiocb in order to support the release
> of dynamically allocated structures referred to by ->private.
>
> The sole in-tree users of ->private are async network read/write,
> which are not, in fact, async, and so need not handle preallocated
> ->private as they would need to if ->ki_retry were ever used. The sole
> truly async operations are direct IO pread()/pwrite() which do not
> now use ->ki_retry(). All they would need to do in that case is to
> check for ->private already being allocated for async kiocbs.
>
> This rips 88B off the stack on 32-bit in the common case.
>
> int sock_sendmsg(struct socket *sock, struct msghdr *msg, size_t size)
> {
> struct kiocb iocb;
> + struct sock_iocb siocb;
> int ret;
>
> init_sync_kiocb(&iocb, NULL);
> + iocb.private = &siocb;
> ret = __sock_sendmsg(&iocb, sock, msg, size);
> if (-EIOCBQUEUED == ret)
> ret = wait_on_sync_kiocb(&iocb);
That's so much better than what we had before it ain't funny.
Was this runtime tested?
|