On Thu, Feb 13, 2020 at 9:20 PM Jeff Layton <jlayton@xxxxxxxxxx> wrote: > > On Thu, 2020-02-13 at 21:05 +0800, Yan, Zheng wrote: > > On Thu, Feb 13, 2020 at 1:29 AM Jeff Layton <jlayton@xxxxxxxxxx> wrote: > > > I've dropped the async unlink patch from testing branch and am > > > resubmitting it here along with the rest of the create patches. > > > > > > Zheng had pointed out that DIR_* caps should be cleared when the session > > > is reconnected. The underlying submission code needed changes to > > > handle that so it needed a bit of rework (along with the create code). > > > > > > Since v3: > > > - rework async request submission to never queue the request when the > > > session isn't open > > > - clean out DIR_* caps, layouts and delegated inodes when session goes down > > > - better ordering for dependent requests > > > - new mount options (wsync/nowsync) instead of module option > > > - more comprehensive error handling > > > > > > Jeff Layton (9): > > > ceph: add flag to designate that a request is asynchronous > > > ceph: perform asynchronous unlink if we have sufficient caps > > > ceph: make ceph_fill_inode non-static > > > ceph: make __take_cap_refs non-static > > > ceph: decode interval_sets for delegated inos > > > ceph: add infrastructure for waiting for async create to complete > > > ceph: add new MDS req field to hold delegated inode number > > > ceph: cache layout in parent dir on first sync create > > > ceph: attempt to do async create when possible > > > > > > fs/ceph/caps.c | 73 +++++++--- > > > fs/ceph/dir.c | 101 +++++++++++++- > > > fs/ceph/file.c | 253 +++++++++++++++++++++++++++++++++-- > > > fs/ceph/inode.c | 58 ++++---- > > > fs/ceph/mds_client.c | 156 +++++++++++++++++++-- > > > fs/ceph/mds_client.h | 17 ++- > > > fs/ceph/super.c | 20 +++ > > > fs/ceph/super.h | 21 ++- > > > include/linux/ceph/ceph_fs.h | 17 ++- > > > 9 files changed, 637 insertions(+), 79 deletions(-) > > > > > > > Please implement something like > > https://github.com/ceph/ceph/pull/32576/commits/e9aa5ec062fab8324e13020ff2f583537e326a0b. > > MDS may revoke Fx when replaying unsafe/async requests. Make mds not > > do this is quite complex. > > > > I added this in reconnect_caps_cb in the latest set: > > /* These are lost when the session goes away */ > if (S_ISDIR(inode->i_mode)) { > if (cap->issued & CEPH_CAP_DIR_CREATE) { > ceph_put_string(rcu_dereference_raw(ci->i_cached_layout.pool_ns)); > memset(&ci->i_cached_layout, 0, sizeof(ci->i_cached_layout)); > } > cap->issued &= ~(CEPH_CAP_DIR_CREATE|CEPH_CAP_DIR_UNLINK); > } > It's not enough. for async create/unlink, we need to call ceph_put_cap_refs(..., CEPH_CAP_FILE_EXCL | CEPH_CAP_DIR_FOO) to release caps > Basically, wipe out the layout and Duc caps when we reconnect the > session. Outstanding references to the caps will be put when the call > completes. Is that not sufficient? > -- > Jeff Layton <jlayton@xxxxxxxxxx> >
