linux-kernel - Re: [PATCH] ceph: using POOL FULL flag instead of OSDMAP FULL flag

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <CAOi1vP-BKfaL-d2GMWDHf7tD=LpDLEug0-NY9dgT=qEi00gpLQ@mail.gmail.com>
Date:   Mon, 2 Mar 2020 11:08:58 +0100
From:   Ilya Dryomov <idryomov@...il.com>
To:     Yanhu Cao <gmayyyha@...il.com>
Cc:     Jeff Layton <jlayton@...nel.org>, Sage Weil <sage@...hat.com>,
        "David S. Miller" <davem@...emloft.net>, kuba@...nel.org,
        Ceph Development <ceph-devel@...r.kernel.org>,
        LKML <linux-kernel@...r.kernel.org>,
        netdev <netdev@...r.kernel.org>
Subject: Re: [PATCH] ceph: using POOL FULL flag instead of OSDMAP FULL flag

On Mon, Mar 2, 2020 at 3:30 AM Yanhu Cao <gmayyyha@...il.com> wrote:
>
> On Fri, Feb 28, 2020 at 10:02 PM Ilya Dryomov <idryomov@...il.com> wrote:
> >
> > On Fri, Feb 28, 2020 at 12:41 PM Yanhu Cao <gmayyyha@...il.com> wrote:
> > >
> > > On Fri, Feb 28, 2020 at 6:23 PM Ilya Dryomov <idryomov@...il.com> wrote:
> > > >
> > > > On Fri, Feb 28, 2020 at 5:45 AM Yanhu Cao <gmayyyha@...il.com> wrote:
> > > > >
> > > > > OSDMAP_FULL and OSDMAP_NEARFULL are deprecated since mimic.
> > > > >
> > > > > Signed-off-by: Yanhu Cao <gmayyyha@...il.com>
> > > > > ---
> > > > >  fs/ceph/file.c                  |  6 ++++--
> > > > >  include/linux/ceph/osd_client.h |  2 ++
> > > > >  include/linux/ceph/osdmap.h     |  3 ++-
> > > > >  net/ceph/osd_client.c           | 23 +++++++++++++----------
> > > > >  4 files changed, 21 insertions(+), 13 deletions(-)
> > > > >
> > > > > diff --git a/fs/ceph/file.c b/fs/ceph/file.c
> > > > > index 7e0190b1f821..60ea1eed1b84 100644
> > > > > --- a/fs/ceph/file.c
> > > > > +++ b/fs/ceph/file.c
> > > > > @@ -1482,7 +1482,8 @@ static ssize_t ceph_write_iter(struct kiocb *iocb, struct iov_iter *from)
> > > > >         }
> > > > >
> > > > >         /* FIXME: not complete since it doesn't account for being at quota */
> > > > > -       if (ceph_osdmap_flag(&fsc->client->osdc, CEPH_OSDMAP_FULL)) {
> > > > > +       if (pool_flag(&fsc->client->osdc, ci->i_layout.pool_id,
> > > > > +                               CEPH_POOL_FLAG_FULL)) {
> > > > >                 err = -ENOSPC;
> > > > >                 goto out;
> > > > >         }
> > > > > @@ -1575,7 +1576,8 @@ static ssize_t ceph_write_iter(struct kiocb *iocb, struct iov_iter *from)
> > > > >         }
> > > > >
> > > > >         if (written >= 0) {
> > > > > -               if (ceph_osdmap_flag(&fsc->client->osdc, CEPH_OSDMAP_NEARFULL))
> > > > > +               if (pool_flag(&fsc->client->osdc, ci->i_layout.pool_id,
> > > > > +                                       CEPH_POOL_FLAG_NEARFULL))
> > > >
> > > > Hi Yanhu,
> > > >
> > > > Have you considered pre-mimic clusters here?  They are still supported
> > > > (and will continue to be supported for the foreseeable future).
> > > >
> > > > Thanks,
> > > >
> > > >                 Ilya
> > >
> > > I have tested it work on Luminous, I think it work too since
> > > ceph-v0.80(https://github.com/ceph/ceph/blob/b78644e7dee100e48dfeca32c9270a6b210d3003/src/osd/osd_types.h#L815)
> > > alread have pool FLAG_FULL.
> >
> > But not FLAG_NEARFULL, which appeared in mimic.
> FLAG_NEARFULL appeared in Luminous.

Well, it appeared in mimic in v13.0.1 and was backported to luminous
in v12.2.2.  So technically, some luminous releases don't have it.

>
> >
> > >
> > > CephFS doesn't write synchronously even if CEPH_OSDMAP_NEARFULL is
> > > used, then should fixed by CEPH_POOL_FLAG_NEARFULL.
> >
> > I'm not sure I follow.
> >
> > -    if (ceph_osdmap_flag(&fsc->client->osdc, CEPH_OSDMAP_NEARFULL))
> > +    if (pool_flag(&fsc->client->osdc, ci->i_layout.pool_id,
> > +                            CEPH_POOL_FLAG_NEARFULL))
> >
> > AFAICT this change would effectively disable this branch for pre-mimic
> > clusters.  Are you saying this branch is already broken?
> >
> > Thanks,
> >
> >                 Ilya
> CEPH_OSDMAP_NEARFULL is not set in Jewel, so it has no effect. And in
> Luminous version, this flag is cleared as a legacy and has no effect
> too.

Are you sure?  What about this code in OSDMonitor::tick() that showed
up in kraken in v11.0.1 and was backported to jewel in v10.2.4?

  if (!mon->pgmon()->pg_map.nearfull_osds.empty()) {
    ...
    add_flag(CEPH_OSDMAP_NEARFULL);
  } else if (osdmap.test_flag(CEPH_OSDMAP_NEARFULL)){
    ...
    remove_flag(CEPH_OSDMAP_NEARFULL);
  }
  if (pending_inc.new_flags != -1 &&
     (pending_inc.new_flags ^ osdmap.flags) & (CEPH_OSDMAP_FULL |
                                               CEPH_OSDMAP_NEARFULL)) {
    ...
    do_propose = true;

It's there in v10.2.11 (the final jewel release).  It's also there
in hammer since v0.94.10...

Thanks,

                Ilya