Hi, On Tue, Oct 21, 2008 at 06:20:17PM +0200, Miklos Szeredi wrote: > On Tue, 21 Oct 2008, steve@xxxxxxxxxx wrote: > > No, I guess it might be possible, but for the time being it is > > its own "glock" plus the page lock dependency. I'd have to > > think quite hard about what the consequences of using the > > inode lock would be. > > > > Of course we do demand the inode lock as well in some cases > > since the vfs has already grabbed it before calling > > into the filesystem when its required. Because of that and > > where we run the glock state machine from, it would be rather > > complicated to make that work I suspect, > > BTW, why do you want strict coherency for memory mappings? It's not > something POSIX mandates. It's not even something that Linux always > did. > Its something that GFS has always done, and so we've tried to keep that feature in GFS2. I think we do (at least I do) try to suggest to people that they shouldn't be relying on this, but we've always tried to make it work anyway, at least on the principle of least surprise. > If I were an application writer, I'd never try to rely on mmap > coherency without the appropriate magic msync() calls. > > Miklos Yes, I'd agree, but I write kernel code, not applications :-) Thanks for the explanation on splice, I'll take a look at that code now and try to understand it in more detail, Steve. -- To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
