On 2011-06-10 13:32, Trond Myklebust wrote: > On Fri, 2011-06-10 at 09:57 -0700, Boaz Harrosh wrote: >> On 06/10/2011 09:28 AM, Trond Myklebust wrote: >>>> >>>> That makes it even more complicated for a do nothing function. We dont do >>>> a different function for each different parameter. We can just do a >>>> "bool write" and unify the dam thing >>> >>> Right now, the nfs_pageio_descriptor doesn't know about reads vs writes. >>> It just knows about 'coalesce requests' and 'perform i/o'. I'd prefer to >>> keep that abstraction, as it makes things cleaner, particularly when you >>> get to patch 5 (NFSv4.1: Fall back to ordinary i/o through the mds if we >>> have no layout segment). Why add more 'if' statements when you don't >>> need to... >>> >> >> OK It's fine. I'm convinced. Do you have this on a git tree? I want to test >> it out. > > I've added it to the 'nfs-for-bakeathon' branch. > I've also merged it into git://git.linux-nfs.org/~bhalevy/linux-pnfs.git nfs-upstream and rebased everything on top of it. >> What was the disposition of desc->pg_bsize do I need to adjust it for the >> pnfs_case in objlayout? > > You might need to adjust it. Please check... > As long the the MDS [rw]size are larger or equal to PAGE_SIZE I believe you should be OK. > I'm still working on the 'fallback to write through mds' case to ensure > that we re-coalesce if the call to pnfs_try_to_read_data() and > pnfs_try_to_write_data(). Once that is done, I think that the objects > code will always do the right thing and I anticipate that the blocks > code can reuse the same code... Right. Thanks for picking this up! Benny > > Cheers > Trond > -- To unsubscribe from this list: send the line "unsubscribe linux-nfs" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html