On Sat, Jun 19, 2021 at 12:51 PM Jeff Layton <jlayton@xxxxxxxxxx> wrote:
>
> On Fri, 2021-06-18 at 22:27 +0800, Yejune Deng wrote:
> > The Documentation/core-api/timekeeping.rst recommend that we should use
> > monotonic time ktime_get_ts64(), to avoid glitches with a concurrent
> > settimeofday().
> >
> > Signed-off-by: Yejune Deng <yejune.deng@xxxxxxxxx>
> > ---
> > net/ceph/messenger.c | 2 +-
> > net/ceph/messenger_v1.c | 2 +-
> > net/ceph/messenger_v2.c | 2 +-
> > net/ceph/osd_client.c | 4 ++--
> > 4 files changed, 5 insertions(+), 5 deletions(-)
> >
> > diff --git a/net/ceph/messenger.c b/net/ceph/messenger.c
> > index 57d043b..2d07ab5 100644
> > --- a/net/ceph/messenger.c
> > +++ b/net/ceph/messenger.c
> > @@ -1809,7 +1809,7 @@ bool ceph_con_keepalive_expired(struct ceph_connection *con,
> > (con->peer_features & CEPH_FEATURE_MSGR_KEEPALIVE2)) {
> > struct timespec64 now;
> > struct timespec64 ts;
> > - ktime_get_real_ts64(&now);
> > + ktime_get_ts64(&now);
> > jiffies_to_timespec64(interval, &ts);
> > ts = timespec64_add(con->last_keepalive_ack, ts);
> > return timespec64_compare(&now, &ts) >= 0;
> > diff --git a/net/ceph/messenger_v1.c b/net/ceph/messenger_v1.c
> > index 2cb5ffd..2ec7b1d 100644
> > --- a/net/ceph/messenger_v1.c
> > +++ b/net/ceph/messenger_v1.c
> > @@ -310,7 +310,7 @@ static void prepare_write_keepalive(struct ceph_connection *con)
> > if (con->peer_features & CEPH_FEATURE_MSGR_KEEPALIVE2) {
> > struct timespec64 now;
> >
> > - ktime_get_real_ts64(&now);
> > + ktime_get_ts64(&now);
> > con_out_kvec_add(con, sizeof(tag_keepalive2), &tag_keepalive2);
> > ceph_encode_timespec64(&con->v1.out_temp_keepalive2, &now);
> > con_out_kvec_add(con, sizeof(con->v1.out_temp_keepalive2),
> > diff --git a/net/ceph/messenger_v2.c b/net/ceph/messenger_v2.c
> > index cc40ce4..2125e77 100644
> > --- a/net/ceph/messenger_v2.c
> > +++ b/net/ceph/messenger_v2.c
> > @@ -1439,7 +1439,7 @@ static int prepare_keepalive2(struct ceph_connection *con)
> > struct ceph_timespec *ts = CTRL_BODY(con->v2.out_buf);
> > struct timespec64 now;
> >
> > - ktime_get_real_ts64(&now);
> > + ktime_get_ts64(&now);
> > dout("%s con %p timestamp %lld.%09ld\n", __func__, con, now.tv_sec,
> > now.tv_nsec);
> >
> > diff --git a/net/ceph/osd_client.c b/net/ceph/osd_client.c
> > index ff8624a..5192a8a 100644
> > --- a/net/ceph/osd_client.c
> > +++ b/net/ceph/osd_client.c
> > @@ -4717,7 +4717,7 @@ ceph_osdc_watch(struct ceph_osd_client *osdc,
> > ceph_oid_copy(&lreq->t.base_oid, oid);
> > ceph_oloc_copy(&lreq->t.base_oloc, oloc);
> > lreq->t.flags = CEPH_OSD_FLAG_WRITE;
> > - ktime_get_real_ts64(&lreq->mtime);
> > + ktime_get_ts64(&lreq->mtime);
> >
> > lreq->reg_req = alloc_watch_request(lreq, CEPH_OSD_WATCH_OP_WATCH);
> > if (!lreq->reg_req) {
> > @@ -4767,7 +4767,7 @@ int ceph_osdc_unwatch(struct ceph_osd_client *osdc,
> > ceph_oid_copy(&req->r_base_oid, &lreq->t.base_oid);
> > ceph_oloc_copy(&req->r_base_oloc, &lreq->t.base_oloc);
> > req->r_flags = CEPH_OSD_FLAG_WRITE;
> > - ktime_get_real_ts64(&req->r_mtime);
> > + ktime_get_ts64(&req->r_mtime);
> > osd_req_op_watch_init(req, 0, lreq->linger_id,
> > CEPH_OSD_WATCH_OP_UNWATCH);
> >
>
> I think this is OK. Most of these timestamps get marshalled onto the
> wire to use as identifiers. Those are almost certainly better off with
> CLOCK_MONOTONIC since you could have a clock jump that caused you to
> send (e.g.) 2 keepalives with the same timestamp otherwise.
While I generally agree that for keepalive (i.e. the first free
instances) using the monotonic clock would be better, in userspace
CLOCK_REALTIME is used and I don't want to diverge.
Thanks,
Ilya
