Skip to content

RFC move layout get/return synchronization to rpc prepare

Benny Halevy linux-nfs 1 version In review

Get series with b4
b4 am [email protected]

Versions newest first

  1. v1 RFC Latest4 patches 4 repliesIn review

Patches in v1 No applies recorded yet

#Subject and filesSizeRepliesApplied
1/4SQUASHME: pnfs-post-submit: remove take_ref and only_valid params from pnfs_has_layoutfs/nfs/pnfs.c+8 -150 replies
2/4pnfs: allow nfs4_proc_layoutget to sleep on invalid lsegsfs/nfs/inode.c fs/nfs/nfs4proc.c fs/nfs/pnfs.c +2 more+45 -224 replies
3/4pnfs: allow nfs4_proc_layoutreturn to sleep on barrierfs/nfs/nfs4proc.c fs/nfs/pnfs.c fs/nfs/pnfs.h+10 -110 replies
4/4SQUASHME: pnfs: get rid of lo_waitqfs/nfs/inode.c fs/nfs/pnfs.c include/linux/nfs_fs.h+14 -200 replies

Reviews trailers per patch

No Reviewed-by, Acked-by or Tested-by trailers for v1 yet.

Read the v1 thread

Cover letter [RFC 0/4] move layout get/return synchronization to rpc prepare

The following patchset introduces a per-inode rpc wait queue to be used to synchronize layout gets and returns during their respective rpc prepare phase.

Essentially, layout gets wait on layout segments marked as invalid, currently meaning there is a layout return in progress. It should be ok to send the layout get anyway but waiting avoids the race rather than having to possibly resolve it on the done path.

In the future layout gets could insert the layout segment they ask for as "invalid" so that parallel layout gets would wait too. However, since the server may return a different segment than requested (bigger or smaller), there's no guarantee that the parallel layoutget will overlap the one in progress.

Layout returns wait on layout segments while they are being used (e.g. for I/O operations).

[PATCH 1/4] SQUASHME: pnfs-post-submit: remove take_ref and only_valid params from pnfs_has_layout [PATCH 2/4] pnfs: allow nfs4_proc_layoutget to sleep on invalid lsegs
[PATCH 3/4] pnfs: allow nfs4_proc_layoutreturn to sleep on barrier
[PATCH 4/4] SQUASHME: pnfs: get rid of lo_waitq