Re: [rfc patch 3/4] splice: remove confirm from pipe_buf_operations

[Miklos Szeredi <miklos@szeredi.hu>]

> > 
> > Or it can only happen if there was an I/O error on reading the page.
> 
> Now, IO errors are something else. They should have the PG_error bit set, 
> and we should just return EIO or something.

Linus, you're right (as always), but see where this is going?  A rare
problem (splice() returning short count because of an invalidated
page) is becoming an even more rare problem (splice() returning
rubbish instead of an error, if ->readpage() failed, and filesystem
forgot to set PG_error).  And it won't show up in any other paths,
because the generic_file_aio_read() path will just check
PageUptodate(), and return -EIO if not.

OK, maybe we should add a WARN_ON(!PageError()) for the
!PageUptodate() case in generic_file_aio_read(), but that could still
leave some filesystems broken for a long time which experience I/O
errors rarely.

So I think the only sane solution here is to remove
ClearPageUptodate().  But that's a VM people's call, I don't have
enough insight into that.

Miklos

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>