nfs41: serialize first layoutget of a file

Per RFC 5661 Errata 3208:
| A client MAY always forget its layout state and associated
| layout stateid at any time (See also section 12.5.5.1).
| In such case, the client MUST use a non-layout stateid for the next
| LAYOUTGET operation. This will signal the server that the client has
| no more layouts on the file and its respective layout state can be
| released before issuing a new layout in response to LAYOUTGET.

In order to make such a signal unique to server, client needs to serialize
all layoutgets using non-layout stateid. We implement this by serializing
layoutgets when client has no layout segments at hand.

Signed-off-by: Peng Tao <tao.peng@primarydata.com>
Signed-off-by: Tom Haynes <Thomas.Haynes@primarydata.com>

2 files changed, 32 insertions, 4 deletions

diff --git a/fs/nfs/pnfs.c b/fs/nfs/pnfs.c
index fa00b56f176a..7e1bac189d1c 100644
--- a/fs/nfs/pnfs.c
+++ b/fs/nfs/pnfs.c
@@ -1288,6 +1288,7 @@ pnfs_update_layout(struct inode *ino,
         struct nfs_client *clp = server->nfs_client;
         struct pnfs_layout_hdr *lo;
         struct pnfs_layout_segment *lseg = NULL;
+        bool first;
 
         if (!pnfs_enabled_sb(NFS_SERVER(ino)))
                 goto out;
@@ -1295,6 +1296,8 @@ pnfs_update_layout(struct inode *ino,
         if (pnfs_within_mdsthreshold(ctx, ino, iomode))
                 goto out;
 
+lookup_again:
+        first = false;
         spin_lock(&ino->i_lock);
         lo = pnfs_find_alloc_layout(ino, ctx, gfp_flags);
         if (lo == NULL) {
@@ -1312,10 +1315,27 @@ pnfs_update_layout(struct inode *ino,
         if (pnfs_layout_io_test_failed(lo, iomode))
                 goto out_unlock;
 
-        /* Check to see if the layout for the given range already exists */
-        lseg = pnfs_find_lseg(lo, &arg);
-        if (lseg)
-                goto out_unlock;
+        first = list_empty(&lo->plh_segs);
+        if (first) {
+                /* The first layoutget for the file. Need to serialize per
+                 * RFC 5661 Errata 3208.
+                 */
+                if (test_and_set_bit(NFS_LAYOUT_FIRST_LAYOUTGET,
+                                     &lo->plh_flags)) {
+                        spin_unlock(&ino->i_lock);
+                        wait_on_bit(&lo->plh_flags, NFS_LAYOUT_FIRST_LAYOUTGET,
+                                    TASK_UNINTERRUPTIBLE);
+                        pnfs_put_layout_hdr(lo);
+                        goto lookup_again;
+                }
+        } else {
+                /* Check to see if the layout for the given range
+                 * already exists
+                 */
+                lseg = pnfs_find_lseg(lo, &arg);
+                if (lseg)
+                        goto out_unlock;
+        }
 
         if (pnfs_layoutgets_blocked(lo, 0))
                 goto out_unlock;
@@ -1343,6 +1363,13 @@ pnfs_update_layout(struct inode *ino,
         lseg = send_layoutget(lo, ctx, &arg, gfp_flags);
         atomic_dec(&lo->plh_outstanding);
 out_put_layout_hdr:
+        if (first) {
+                unsigned long *bitlock = &lo->plh_flags;
+
+                clear_bit_unlock(NFS_LAYOUT_FIRST_LAYOUTGET, bitlock);
+                smp_mb__after_atomic();
+                wake_up_bit(bitlock, NFS_LAYOUT_FIRST_LAYOUTGET);
+        }
         pnfs_put_layout_hdr(lo);
 out:
         dprintk("%s: inode %s/%llu pNFS layout segment %s for "
diff --git a/fs/nfs/pnfs.h b/fs/nfs/pnfs.h
index c39882191651..4cf0d54e14c3 100644
--- a/fs/nfs/pnfs.h
+++ b/fs/nfs/pnfs.h
@@ -95,6 +95,7 @@ enum {
         NFS_LAYOUT_ROC,                 /* some lseg had roc bit set */
         NFS_LAYOUT_RETURN,              /* Return this layout ASAP */
         NFS_LAYOUT_INVALID_STID,        /* layout stateid id is invalid */
+        NFS_LAYOUT_FIRST_LAYOUTGET,     /* Serialize first layoutget */
 };
 
 enum layoutdriver_policy_flags {