Received-SPF: pass (google.com: domain of linux-nfs-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
MIME-Version: 1.0
References: <20210409180519.25405-1-dai.ngo@oracle.com> <20210409180519.25405-2-dai.ngo@oracle.com>
 <CAN-5tyF_B9LK3soGfbrp8hpS3RG4CdmFxK246u2ADVn+2s0JHw@mail.gmail.com> <5d7c26c2-1964-7fec-afb4-5f088830f4a4@oracle.com>
In-Reply-To: <5d7c26c2-1964-7fec-afb4-5f088830f4a4@oracle.com>
From:   Olga Kornievskaia <olga.kornievskaia@gmail.com>
Date:   Thu, 22 Apr 2021 15:55:21 -0400
Message-ID: <CAN-5tyF4U-4Yxa_eJ6PE-MfgFbSzDnnx0Ty9kVAQ5UbZmnZdcQ@mail.gmail.com>
Subject: Re: [PATCH v3 1/2] NFSD: delay unmount source's export after
 inter-server copy completed.
To:     Dai Ngo <dai.ngo@oracle.com>
Cc:     linux-nfs <linux-nfs@vger.kernel.org>,
        Trond Myklebust <trondmy@hammerspace.com>,
        "J. Bruce Fields" <bfields@fieldses.org>,
        Chuck Lever <chuck.lever@oracle.com>
Content-Type: text/plain; charset="UTF-8"
Precedence: bulk

On Thu, Apr 22, 2021 at 2:42 PM <dai.ngo@oracle.com> wrote:
>
> Thank you Olga for reviewing the patch and doing the performance
> testing, greatly appreciated!
>
> On 4/21/21 5:31 PM, Olga Kornievskaia wrote:
> > On Fri, Apr 9, 2021 at 2:07 PM Dai Ngo <dai.ngo@oracle.com> wrote:
> >> Currently the source's export is mounted and unmounted on every
> >> inter-server copy operation. This patch is an enhancement to delay
> >> the unmount of the source export for a certain period of time to
> >> eliminate the mount and unmount overhead on subsequent copy operations.
> >>
> >> After a copy operation completes, a delayed task is scheduled to
> >> unmount the export after a configurable idle time. Each time the
> >> export is being used again, its expire time is extended to allow
> >> the export to remain mounted.
> >>
> >> The unmount task and the mount operation of the copy request are
> >> synced to make sure the export is not unmounted while it's being
> >> used.
> > I have done performance testing of this patch on the hardware used to
> > test initial SSC code and there is improvement on the lower end copy
> > filesize (4MB-16MB) with this. Unfortunately, I can't quantify how
> > much difference this patch makes. Why? Because I'm getting very
> > different results on every set of runs (each set does 10runs to get
> > the mean and compute standard deviation too).  The most varied time
> > from run to run is commit time. However, standard deviation for both
> > types of runs (traditional and ssc) is (most of the time) <1% which
> > should still make the runs acceptable (i did re-runs for others).
> > Runs with filesize 128MB+ have more consistent performance but they
> > also don't show as much improvement with this patch. Note I didn't
> > concentrate on  >1G performance at all this time around.
>
> I think user would appreciate more if copying of large files is
> faster since that reduces the time the user has to wait for the
> copy to complete. Copying of small files happen so quickly (<1 sec)
> it's hard to see the different.

I'm not sure what you are commenting or arguing for here. Smaller
copies is what delayed unmount helps more than large copies. AGain
given that numbers were not stable I can't stay what exact improvement
was for small copies compared to the large(r) copies. but it seemed
like small copies had 10% and large copies had 5% improvement with
this patch. I think this patch is the reason to remove/reduce client's
restriction on the small copies.  I think I'm more comfortable
removing the 16MB restriction given the numbers we have.

> > Also, what I discovered is what also makes a difference is the
> > following COMMIT rpc. The saving seems to come from doing an
> > vfs_fsync() after the copy on the server (just like what is done for
> > the CLONE i might note) and I'm hoping to propose a patch for it some
> > time in the next few days.
>
> I think this is a good enhancement.

Perhaps, the argument against it if SOME implementation wanted to send
a bunch of copies and later on wanted to commit separately then making
each async copy do NFS_FILE_SYNC would hurt their performance. I have
to say this change is very linux-client tailored and thus it's RFC.
Current linux implementation is user-space (eg cp) would do a single
copy_file_range() system call for the whole file and linux client
doesn't break down the request into smaller copies, and after that it
would send a commit, so this optimization makes sense for linux.

> > A few general comments on the patch:
> > 1. Shouldn't the "nsui_refcnt" be a refcount_t structure type?
>
> I will convert to refcount_t in the v2 patch.
>
> > 2. Why is the value of 20HZ to retry was chosen?
>
> This is when a thread doing the mount and either (1) there is another
> thread doing another mount to the same source server or (2) the export
> of that server is being unmounted by the delayed task. In this case
> the thread has to wait for the other thread to be done with the source
> server and try again.
>
> I think we should not have the thread waits forever so I just pick
> 20secs which I think should be long enough for the mount/unmount to
> complete in normal condition. I'm open to suggestion on how to handle
> this case.

I understood the need for the timeout value, I just didn't know why 20
and not say 5? Seems like a long time to have the copy wait? I'm not
sure how to simulate this condition to see what happens.

> > 3. "work" gets allocated but instead of checking the allocation
> > failure and dealing with it then, the code has "if (work)" sections.
> > seems like extra indentation and the need to think why "work" can be
> > null be removed?
>
> There is nothing that we need to do if the allocation of work fails.
> We still have to continue to do the mount and just skip the work of
> setting it up for the delayed unmount.

Ok.

> > Here are some numbers. Comparing pure 5.12-rc6 code with 5.12-rc6 code
> > with Dai's delay umount code and for kicks the last set of numbers is
> > +delay umount +vfs_fsync() on the server. In general over 128MB
> > provides 50% improvement over traditional copy in all cases. With
> > delayed umount, we see better numbers (~30% improvement) with copies
> > 4MB-128MB but still some dips and for copies over 128MB over 50%
> > improvement.
>
> I think the improvement from the delayed mount and commit optimization
> is significant.

I agree that a delayed umount is beneficial and should be added.

> -Dai
>
> >
> > pure 5.12-rc6
> >      INFO: 14:11:16.819155 - Server-side COPY: 0.472602200508 seconds
> > (mean), standard deviation: 0.504127857283
> >      INFO: 14:11:16.819252 - Traditional COPY: 0.247301483154 seconds
> > (mean), standard deviation: 0.0325181306537
> >      FAIL: SSC should outperform traditional copy, performance
> > degradation for a 4MB file: 91%
> >
> >      INFO: 14:12:54.584129 - Server-side COPY: 0.401613616943 seconds
> > (mean), standard deviation: 0.100666012531
> >      INFO: 14:12:54.584223 - Traditional COPY: 0.376980066299 seconds
> > (mean), standard deviation: 0.0149147531532
> >      FAIL: SSC should outperform traditional copy, performance
> > degradation for a 8MB file: 6%
> >
> >      INFO: 14:14:37.430685 - Server-side COPY: 0.727260971069 seconds
> > (mean), standard deviation: 0.24591675709
> >      INFO: 14:14:37.430787 - Traditional COPY: 0.858516097069 seconds
> > (mean), standard deviation: 0.203399239696
> >      PASS: SSC should outperform traditional copy, performance
> > improvement for a 16MB file: 18%
> >
> >      INFO: 14:16:25.318145 - Server-side COPY: 1.08072230816 seconds
> > (mean), standard deviation: 0.168767973425
> >      INFO: 14:16:25.318245 - Traditional COPY: 1.21966302395 seconds
> > (mean), standard deviation: 0.0862109147556
> >      PASS: SSC should outperform traditional copy, performance
> > improvement for a 32MB file: 12%
> >
> >      INFO: 14:18:20.913090 - Server-side COPY: 1.47459042072 seconds
> > (mean), standard deviation: 0.118534392658
> >      INFO: 14:18:20.913186 - Traditional COPY: 1.74257912636 seconds
> > (mean), standard deviation: 0.041726092735
> >      PASS: SSC should outperform traditional copy, performance
> > improvement for a 64MB file: 18%
> >
> >      INFO: 14:20:40.794478 - Server-side COPY: 2.27927930355 seconds
> > (mean), standard deviation: 0.0502558652592
> >      INFO: 14:20:40.794579 - Traditional COPY: 3.3685300827 seconds
> > (mean), standard deviation: 0.0232078152411
> >      PASS: SSC should outperform traditional copy, performance
> > improvement for a 128MB file: 47%
> >
> >      INFO: 14:23:54.480951 - Server-side COPY: 4.36852009296 seconds
> > (mean), standard deviation: 0.0421940712129
> >      INFO: 14:23:54.481059 - Traditional COPY: 6.58469381332 seconds
> > (mean), standard deviation: 0.0534006595486
> >      PASS: SSC should outperform traditional copy, performance
> > improvement for a 256MB file: 50%
> >
> >      INFO: 14:28:57.776217 - Server-side COPY: 8.61689963341 seconds
> > (mean), standard deviation: 0.110919967983
> >      INFO: 14:28:57.776361 - Traditional COPY: 13.0591681957 seconds
> > (mean), standard deviation: 0.0766741218971
> >      PASS: SSC should outperform traditional copy, performance
> > improvement for a 512MB file: 51%
> >
> >      INFO: 14:36:45.693363 - Server-side COPY: 15.330485177 seconds
> > (mean), standard deviation: 1.08275054089
> >      INFO: 14:36:45.693547 - Traditional COPY: 22.6392157316 seconds
> > (mean), standard deviation: 0.62612602097
> >      PASS: SSC should outperform traditional copy, performance
> > improvement for a 1GB file: 47%
> >
> >      (another run for 1g to get a better deviation)
> >      INFO: 15:33:08.457225 - Server-side COPY: 14.7443211555 seconds
> > (mean), standard deviation: 0.19152031268
> >      INFO: 15:33:08.457401 - Traditional COPY: 23.3897125483 seconds
> > (mean), standard deviation: 0.634745610516
> >      PASS: SSC should outperform traditional copy, performance
> > improvement for a 1GB file: 58%
> >
> > This run is with delayed umount:
> >      INFO: 16:57:09.750669 - Server-side COPY: 0.264265751839 seconds
> > (mean), standard deviation: 0.114140867309
> >      INFO: 16:57:09.750767 - Traditional COPY: 0.277241683006 seconds
> > (mean), standard deviation: 0.0429385806753
> >      PASS: SSC should outperform traditional copy, performance
> > improvement for a 4MB file: 4%
> >
> >      INFO: 16:58:48.514208 - Server-side COPY: 0.301608347893 seconds
> > (mean), standard deviation: 0.027625417442
> >      INFO: 16:58:48.514305 - Traditional COPY: 0.376611542702 seconds
> > (mean), standard deviation: 0.0256646541006
> >      PASS: SSC should outperform traditional copy, performance
> > improvement for a 8MB file: 24%
> >
> >      INFO: 17:00:32.726180 - Server-side COPY: 0.804727435112 seconds
> > (mean), standard deviation: 0.282478573802
> >      INFO: 17:00:32.726284 - Traditional COPY: 0.750376915932 seconds
> > (mean), standard deviation: 0.188426980103
> >      FAIL: SSC should outperform traditional copy, performance
> > degradation for a 16MB file: 7%
> >
> > another run for 16MB for better deviation
> >      INFO: 17:26:05.710048 - Server-side COPY: 0.561978292465 seconds
> > (mean), standard deviation: 0.176039123641
> >      INFO: 17:26:05.710155 - Traditional COPY: 0.761225438118 seconds
> > (mean), standard deviation: 0.229373528254
> >      PASS: SSC should outperform traditional copy, performance
> > improvement for a 16MB file: 35%
> >
> >      INFO: 17:02:21.387772 - Server-side COPY: 0.895572972298 seconds
> > (mean), standard deviation: 0.18919321178
> >      INFO: 17:02:21.387866 - Traditional COPY: 1.28307352066 seconds
> > (mean), standard deviation: 0.0755687243879
> >      PASS: SSC should outperform traditional copy, performance
> > improvement for a 32MB file: 43%
> >
> >      INFO: 17:04:18.530852 - Server-side COPY: 1.44412658215 seconds
> > (mean), standard deviation: 0.0680101210746
> >      INFO: 17:04:18.530946 - Traditional COPY: 1.7353479147 seconds
> > (mean), standard deviation: 0.0147682451688
> >      PASS: SSC should outperform traditional copy, performance
> > improvement for a 64MB file: 20%
> >
> >      INFO: 17:06:40.127898 - Server-side COPY: 2.18924453259 seconds
> > (mean), standard deviation: 0.0339712202148
> >      INFO: 17:06:40.127995 - Traditional COPY: 3.39783103466 seconds
> > (mean), standard deviation: 0.0439455560873
> >      PASS: SSC should outperform traditional copy, performance
> > improvement for a 128MB file: 55%
> >
> >      INFO: 17:09:56.186889 - Server-side COPY: 4.2961766243 seconds
> > (mean), standard deviation: 0.0486465355758
> >      INFO: 17:09:56.186988 - Traditional COPY: 6.56641271114 seconds
> > (mean), standard deviation: 0.0259688949944
> >      PASS: SSC should outperform traditional copy, performance
> > improvement for a 256MB file: 52%
> >
> >      INFO: 17:15:02.522089 - Server-side COPY: 8.4736330986 seconds
> > (mean), standard deviation: 0.0896888780618
> >      INFO: 17:15:02.522242 - Traditional COPY: 13.074249053 seconds
> > (mean), standard deviation: 0.0805015369838
> >      PASS: SSC should outperform traditional copy, performance
> > improvement for a 512MB file: 54%
> >
> >      INFO: 17:22:59.745446 - Server-side COPY: 15.9621772766 seconds
> > (mean), standard deviation: 2.70296237945
> >      INFO: 17:22:59.745625 - Traditional COPY: 22.5955899239 seconds
> > (mean), standard deviation: 0.682459618149
> >      PASS: SSC should outperform traditional copy, performance
> > improvement for a 1GB file: 41%
> >
> > another run for 1GB to get a better deviation:
> >      INFO: 17:35:02.846602 - Server-side COPY: 14.6351578712 seconds
> > (mean), standard deviation: 0.19192309658
> >      INFO: 17:35:02.846781 - Traditional COPY: 22.8912801266 seconds
> > (mean), standard deviation: 0.682718216033
> >      PASS: SSC should outperform traditional copy, performance
> > improvement for a 1GB file: 56%
> >
> > And this run is for yet unpublished patch of delayed umount +
> > vfs_fsync after the copy
> >      INFO: 18:28:34.764202 - Server-side COPY: 0.209130239487 seconds
> > (mean), standard deviation: 0.0366357417649
> >      INFO: 18:28:34.764297 - Traditional COPY: 0.275605249405 seconds
> > (mean), standard deviation: 0.030773788804
> >      PASS: SSC should outperform traditional copy, performance
> > improvement for a 4MB file: 31%
> >
> >      INFO: 18:30:12.278362 - Server-side COPY: 0.263212919235 seconds
> > (mean), standard deviation: 0.017764772065
> >      INFO: 18:30:12.278459 - Traditional COPY: 0.373342609406 seconds
> > (mean), standard deviation: 0.0235488678036
> >      PASS: SSC should outperform traditional copy, performance
> > improvement for a 8MB file: 41%
> >
> >      INFO: 18:31:56.740901 - Server-side COPY: 0.653366684914 seconds
> > (mean), standard deviation: 0.201038063193
> >      INFO: 18:31:56.741007 - Traditional COPY: 0.719534230232 seconds
> > (mean), standard deviation: 0.144084877516
> >      PASS: SSC should outperform traditional copy, performance
> > improvement for a 16MB file: 10%
> >
> > another run for 16Mb
> >      INFO: 20:18:09.533128 - Server-side COPY: 0.508768010139 seconds
> > (mean), standard deviation: 0.172117325138
> >      INFO: 20:18:09.533224 - Traditional COPY: 0.712138915062 seconds
> > (mean), standard deviation: 0.162059202594
> >      PASS: SSC should outperform traditional copy, performance
> > improvement for a 16MB file: 39%
> >
> >      INFO: 18:33:44.203037 - Server-side COPY: 0.774861812592 seconds
> > (mean), standard deviation: 0.141729231937
> >      INFO: 18:33:44.203140 - Traditional COPY: 1.24551167488 seconds
> > (mean), standard deviation: 0.0520878630978
> >      PASS: SSC should outperform traditional copy, performance
> > improvement for a 32MB file: 60%
> >
> >      INFO: 18:35:44.104048 - Server-side COPY: 1.50374240875 seconds
> > (mean), standard deviation: 0.132060247445
> >      INFO: 18:35:44.104168 - Traditional COPY: 1.76330254078 seconds
> > (mean), standard deviation: 0.0339954657686
> >      PASS: SSC should outperform traditional copy, performance
> > improvement for a 64MB file: 17%
> >
> > another run for 64MB
> >      INFO: 20:22:39.668006 - Server-side COPY: 1.32221701145 seconds
> > (mean), standard deviation: 0.0595545335408
> >      INFO: 20:22:39.668108 - Traditional COPY: 1.73985309601 seconds
> > (mean), standard deviation: 0.0301114835769
> >      PASS: SSC should outperform traditional copy, performance
> > improvement for a 64MB file: 31%
> >
> >      INFO: 18:38:06.184631 - Server-side COPY: 2.11796813011 seconds
> > (mean), standard deviation: 0.0407036659715
> >      INFO: 18:38:06.184726 - Traditional COPY: 3.36791093349 seconds
> > (mean), standard deviation: 0.029777196608
> >      PASS: SSC should outperform traditional copy, performance
> > improvement for a 128MB file: 59%
> >
> >      INFO: 18:41:20.944541 - Server-side COPY: 4.19641048908 seconds
> > (mean), standard deviation: 0.041610728753
> >      INFO: 18:41:20.944641 - Traditional COPY: 6.569201684 seconds
> > (mean), standard deviation: 0.0398935239857
> >      PASS: SSC should outperform traditional copy, performance
> > improvement for a 256MB file: 56%
> >
> >      INFO: 18:46:23.973364 - Server-side COPY: 8.34293022156 seconds
> > (mean), standard deviation: 0.108239167667
> >      INFO: 18:46:23.973562 - Traditional COPY: 13.0458415985 seconds
> > (mean), standard deviation: 0.0399096580857
> >      PASS: SSC should outperform traditional copy, performance
> > improvement for a 512MB file: 56%
> >
> >      INFO: 18:54:05.741343 - Server-side COPY: 14.4749611855 seconds
> > (mean), standard deviation: 0.122991853502
> >      INFO: 18:54:05.741549 - Traditional COPY: 22.6955966711 seconds
> > (mean), standard deviation: 0.461425661742
> >      PASS: SSC should outperform traditional copy, performance
> > improvement for a 1GB file: 56%
> >
> >
> >
> >> Signed-off-by: Dai Ngo <dai.ngo@oracle.com>
> >> ---
> >>   fs/nfsd/nfs4proc.c      | 171 ++++++++++++++++++++++++++++++++++++++++++++++--
> >>   fs/nfsd/nfsd.h          |   4 ++
> >>   fs/nfsd/nfssvc.c        |   3 +
> >>   include/linux/nfs_ssc.h |  20 ++++++
> >>   4 files changed, 194 insertions(+), 4 deletions(-)
> >>
> >> diff --git a/fs/nfsd/nfs4proc.c b/fs/nfsd/nfs4proc.c
> >> index dd9f38d..66dea2f 100644
> >> --- a/fs/nfsd/nfs4proc.c
> >> +++ b/fs/nfsd/nfs4proc.c
> >> @@ -55,6 +55,81 @@
> >>   MODULE_PARM_DESC(inter_copy_offload_enable,
> >>                   "Enable inter server to server copy offload. Default: false");
> >>
> >> +#ifdef CONFIG_NFSD_V4_2_INTER_SSC
> >> +static int nfsd4_ssc_umount_timeout = 900000;          /* default to 15 mins */
> >> +module_param(nfsd4_ssc_umount_timeout, int, 0644);
> >> +MODULE_PARM_DESC(nfsd4_ssc_umount_timeout,
> >> +               "idle msecs before unmount export from source server");
> >> +
> >> +static void nfsd4_ssc_expire_umount(struct work_struct *work);
> >> +static struct nfsd4_ssc_umount nfsd4_ssc_umount;
> >> +
> >> +/* nfsd4_ssc_umount.nsu_lock must be held */
> >> +static void nfsd4_scc_update_umnt_timo(void)
> >> +{
> >> +       struct nfsd4_ssc_umount_item *ni = 0;
> >> +
> >> +       cancel_delayed_work(&nfsd4_ssc_umount.nsu_umount_work);
> >> +       if (!list_empty(&nfsd4_ssc_umount.nsu_list)) {
> >> +               ni = list_first_entry(&nfsd4_ssc_umount.nsu_list,
> >> +                       struct nfsd4_ssc_umount_item, nsui_list);
> >> +               nfsd4_ssc_umount.nsu_expire = ni->nsui_expire;
> >> +               schedule_delayed_work(&nfsd4_ssc_umount.nsu_umount_work,
> >> +                       ni->nsui_expire - jiffies);
> >> +       } else
> >> +               nfsd4_ssc_umount.nsu_expire = 0;
> >> +}
> >> +
> >> +static void nfsd4_ssc_expire_umount(struct work_struct *work)
> >> +{
> >> +       bool do_wakeup = false;
> >> +       struct nfsd4_ssc_umount_item *ni = 0;
> >> +       struct nfsd4_ssc_umount_item *tmp;
> >> +
> >> +       spin_lock(&nfsd4_ssc_umount.nsu_lock);
> >> +       list_for_each_entry_safe(ni, tmp, &nfsd4_ssc_umount.nsu_list, nsui_list) {
> >> +               if (time_after(jiffies, ni->nsui_expire)) {
> >> +                       if (ni->nsui_refcnt > 0)
> >> +                               continue;
> >> +
> >> +                       /* mark being unmount */
> >> +                       ni->nsui_busy = true;
> >> +                       spin_unlock(&nfsd4_ssc_umount.nsu_lock);
> >> +                       mntput(ni->nsui_vfsmount);
> >> +                       spin_lock(&nfsd4_ssc_umount.nsu_lock);
> >> +
> >> +                       /* waiters need to start from begin of list */
> >> +                       list_del(&ni->nsui_list);
> >> +                       kfree(ni);
> >> +
> >> +                       /* wakeup ssc_connect waiters */
> >> +                       do_wakeup = true;
> >> +                       continue;
> >> +               }
> >> +               break;
> >> +       }
> >> +       nfsd4_scc_update_umnt_timo();
> >> +       if (do_wakeup)
> >> +               wake_up_all(&nfsd4_ssc_umount.nsu_waitq);
> >> +       spin_unlock(&nfsd4_ssc_umount.nsu_lock);
> >> +}
> >> +
> >> +static DECLARE_DELAYED_WORK(nfsd4, nfsd4_ssc_expire_umount);
> >> +
> >> +void nfsd4_ssc_init_umount_work(void)
> >> +{
> >> +       if (nfsd4_ssc_umount.nsu_inited)
> >> +               return;
> >> +       INIT_DELAYED_WORK(&nfsd4_ssc_umount.nsu_umount_work,
> >> +               nfsd4_ssc_expire_umount);
> >> +       INIT_LIST_HEAD(&nfsd4_ssc_umount.nsu_list);
> >> +       spin_lock_init(&nfsd4_ssc_umount.nsu_lock);
> >> +       init_waitqueue_head(&nfsd4_ssc_umount.nsu_waitq);
> >> +       nfsd4_ssc_umount.nsu_inited = true;
> >> +}
> >> +EXPORT_SYMBOL_GPL(nfsd4_ssc_init_umount_work);
> >> +#endif
> >> +
> >>   #ifdef CONFIG_NFSD_V4_SECURITY_LABEL
> >>   #include <linux/security.h>
> >>
> >> @@ -1181,6 +1256,9 @@ extern struct file *nfs42_ssc_open(struct vfsmount *ss_mnt,
> >>          char *ipaddr, *dev_name, *raw_data;
> >>          int len, raw_len;
> >>          __be32 status = nfserr_inval;
> >> +       struct nfsd4_ssc_umount_item *ni = 0;
> >> +       struct nfsd4_ssc_umount_item *work, *tmp;
> >> +       DEFINE_WAIT(wait);
> >>
> >>          naddr = &nss->u.nl4_addr;
> >>          tmp_addrlen = rpc_uaddr2sockaddr(SVC_NET(rqstp), naddr->addr,
> >> @@ -1229,12 +1307,63 @@ extern struct file *nfs42_ssc_open(struct vfsmount *ss_mnt,
> >>                  goto out_free_rawdata;
> >>          snprintf(dev_name, len + 5, "%s%s%s:/", startsep, ipaddr, endsep);
> >>
> >> +       work = kzalloc(sizeof(*work), GFP_KERNEL);
> >> +try_again:
> >> +       spin_lock(&nfsd4_ssc_umount.nsu_lock);
> >> +       list_for_each_entry_safe(ni, tmp, &nfsd4_ssc_umount.nsu_list, nsui_list) {
> >> +               if (strncmp(ni->nsui_ipaddr, ipaddr, sizeof(ni->nsui_ipaddr)))
> >> +                       continue;
> >> +               /* found a match */
> >> +               if (ni->nsui_busy) {
> >> +                       /*  wait - and try again */
> >> +                       prepare_to_wait(&nfsd4_ssc_umount.nsu_waitq, &wait,
> >> +                               TASK_INTERRUPTIBLE);
> >> +                       spin_unlock(&nfsd4_ssc_umount.nsu_lock);
> >> +                       if (signal_pending(current) ||
> >> +                                       (schedule_timeout(20*HZ) == 0)) {
> >> +                               status = nfserr_eagain;
> >> +                               kfree(work);
> >> +                               goto out_free_devname;
> >> +                       }
> >> +                       finish_wait(&nfsd4_ssc_umount.nsu_waitq, &wait);
> >> +                       goto try_again;
> >> +               }
> >> +               ss_mnt = ni->nsui_vfsmount;
> >> +               ni->nsui_refcnt++;
> >> +               spin_unlock(&nfsd4_ssc_umount.nsu_lock);
> >> +               kfree(work);
> >> +               goto out_done;
> >> +       }
> >> +       /* create new entry, set busy, insert list, clear busy after mount */
> >> +       if (work) {
> >> +               strncpy(work->nsui_ipaddr, ipaddr, sizeof(work->nsui_ipaddr));
> >> +               work->nsui_refcnt++;
> >> +               work->nsui_busy = true;
> >> +               list_add_tail(&work->nsui_list, &nfsd4_ssc_umount.nsu_list);
> >> +       }
> >> +       spin_unlock(&nfsd4_ssc_umount.nsu_lock);
> >> +
> >>          /* Use an 'internal' mount: SB_KERNMOUNT -> MNT_INTERNAL */
> >>          ss_mnt = vfs_kern_mount(type, SB_KERNMOUNT, dev_name, raw_data);
> >>          module_put(type->owner);
> >> -       if (IS_ERR(ss_mnt))
> >> +       if (IS_ERR(ss_mnt)) {
> >> +               if (work) {
> >> +                       spin_lock(&nfsd4_ssc_umount.nsu_lock);
> >> +                       list_del(&work->nsui_list);
> >> +                       wake_up_all(&nfsd4_ssc_umount.nsu_waitq);
> >> +                       spin_unlock(&nfsd4_ssc_umount.nsu_lock);
> >> +                       kfree(work);
> >> +               }
> >>                  goto out_free_devname;
> >> -
> >> +       }
> >> +       if (work) {
> >> +               spin_lock(&nfsd4_ssc_umount.nsu_lock);
> >> +               work->nsui_vfsmount = ss_mnt;
> >> +               work->nsui_busy = false;
> >> +               wake_up_all(&nfsd4_ssc_umount.nsu_waitq);
> >> +               spin_unlock(&nfsd4_ssc_umount.nsu_lock);
> >> +       }
> >> +out_done:
> >>          status = 0;
> >>          *mount = ss_mnt;
> >>
> >> @@ -1301,10 +1430,44 @@ extern struct file *nfs42_ssc_open(struct vfsmount *ss_mnt,
> >>   nfsd4_cleanup_inter_ssc(struct vfsmount *ss_mnt, struct nfsd_file *src,
> >>                          struct nfsd_file *dst)
> >>   {
> >> +       bool found = false;
> >> +       long timeout;
> >> +       struct nfsd4_ssc_umount_item *tmp;
> >> +       struct nfsd4_ssc_umount_item *ni = 0;
> >> +
> >>          nfs42_ssc_close(src->nf_file);
> >> -       fput(src->nf_file);
> >>          nfsd_file_put(dst);
> >> -       mntput(ss_mnt);
> >> +       fput(src->nf_file);
> >> +
> >> +       timeout = msecs_to_jiffies(nfsd4_ssc_umount_timeout);
> >> +       spin_lock(&nfsd4_ssc_umount.nsu_lock);
> >> +       list_for_each_entry_safe(ni, tmp, &nfsd4_ssc_umount.nsu_list,
> >> +               nsui_list) {
> >> +               if (ni->nsui_vfsmount->mnt_sb == ss_mnt->mnt_sb) {
> >> +                       list_del(&ni->nsui_list);
> >> +                       /*
> >> +                        * vfsmount can be shared by multiple exports,
> >> +                        * decrement refcnt and schedule delayed task
> >> +                        * if it drops to 0.
> >> +                        */
> >> +                       ni->nsui_refcnt--;
> >> +                       ni->nsui_expire = jiffies + timeout;
> >> +                       list_add_tail(&ni->nsui_list, &nfsd4_ssc_umount.nsu_list);
> >> +                       found = true;
> >> +                       break;
> >> +               }
> >> +       }
> >> +       if (!found) {
> >> +               spin_unlock(&nfsd4_ssc_umount.nsu_lock);
> >> +               mntput(ss_mnt);
> >> +               return;
> >> +       }
> >> +       if (ni->nsui_refcnt == 0 && !nfsd4_ssc_umount.nsu_expire) {
> >> +               nfsd4_ssc_umount.nsu_expire = ni->nsui_expire;
> >> +               schedule_delayed_work(&nfsd4_ssc_umount.nsu_umount_work,
> >> +                               timeout);
> >> +       }
> >> +       spin_unlock(&nfsd4_ssc_umount.nsu_lock);
> >>   }
> >>
> >>   #else /* CONFIG_NFSD_V4_2_INTER_SSC */
> >> diff --git a/fs/nfsd/nfsd.h b/fs/nfsd/nfsd.h
> >> index 8bdc37a..b3bf8a5 100644
> >> --- a/fs/nfsd/nfsd.h
> >> +++ b/fs/nfsd/nfsd.h
> >> @@ -483,6 +483,10 @@ static inline bool nfsd_attrs_supported(u32 minorversion, const u32 *bmval)
> >>   extern int nfsd4_is_junction(struct dentry *dentry);
> >>   extern int register_cld_notifier(void);
> >>   extern void unregister_cld_notifier(void);
> >> +#ifdef CONFIG_NFSD_V4_2_INTER_SSC
> >> +extern void nfsd4_ssc_init_umount_work(void);
> >> +#endif
> >> +
> >>   #else /* CONFIG_NFSD_V4 */
> >>   static inline int nfsd4_is_junction(struct dentry *dentry)
> >>   {
> >> diff --git a/fs/nfsd/nfssvc.c b/fs/nfsd/nfssvc.c
> >> index 6de4063..2558db5 100644
> >> --- a/fs/nfsd/nfssvc.c
> >> +++ b/fs/nfsd/nfssvc.c
> >> @@ -322,6 +322,9 @@ static int nfsd_startup_generic(int nrservs)
> >>          ret = nfs4_state_start();
> >>          if (ret)
> >>                  goto out_file_cache;
> >> +#ifdef CONFIG_NFSD_V4_2_INTER_SSC
> >> +       nfsd4_ssc_init_umount_work();
> >> +#endif
> >>          return 0;
> >>
> >>   out_file_cache:
> >> diff --git a/include/linux/nfs_ssc.h b/include/linux/nfs_ssc.h
> >> index f5ba0fb..bb9ed6f 100644
> >> --- a/include/linux/nfs_ssc.h
> >> +++ b/include/linux/nfs_ssc.h
> >> @@ -8,6 +8,7 @@
> >>    */
> >>
> >>   #include <linux/nfs_fs.h>
> >> +#include <linux/sunrpc/svc.h>
> >>
> >>   extern struct nfs_ssc_client_ops_tbl nfs_ssc_client_tbl;
> >>
> >> @@ -52,6 +53,25 @@ static inline void nfs42_ssc_close(struct file *filep)
> >>          if (nfs_ssc_client_tbl.ssc_nfs4_ops)
> >>                  (*nfs_ssc_client_tbl.ssc_nfs4_ops->sco_close)(filep);
> >>   }
> >> +
> >> +struct nfsd4_ssc_umount_item {
> >> +       struct list_head nsui_list;
> >> +       bool nsui_busy;
> >> +       int nsui_refcnt;
> >> +       unsigned long nsui_expire;
> >> +       struct vfsmount *nsui_vfsmount;
> >> +       char nsui_ipaddr[RPC_MAX_ADDRBUFLEN];
> >> +};
> >> +
> >> +struct nfsd4_ssc_umount {
> >> +       struct list_head nsu_list;
> >> +       struct delayed_work nsu_umount_work;
> >> +       spinlock_t nsu_lock;
> >> +       unsigned long nsu_expire;
> >> +       wait_queue_head_t nsu_waitq;
> >> +       bool nsu_inited;
> >> +};
> >> +
> >>   #endif
> >>
> >>   /*
> >> --
> >> 1.8.3.1
> >>