NFSv4.1/pNFS: Don't queue up a new commit if the layout segment is invalid

If the layout segment is invalid, then we should not be adding more
write requests to the commit list. Instead, those writes should be
replayed after requesting a new layout.

Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>

7 files changed, 39 insertions, 0 deletions

diff --git a/fs/nfs/callback_proc.c b/fs/nfs/callback_proc.c
index e4dbab5dce6e..2be8b252e3b1 100644
--- a/fs/nfs/callback_proc.c
+++ b/fs/nfs/callback_proc.c
@@ -233,6 +233,8 @@ static u32 initiate_file_draining(struct nfs_client *clp,
 unlock:
 	spin_unlock(&ino->i_lock);
 	pnfs_free_lseg_list(&free_me_list);
+	/* Free all lsegs that are attached to commit buckets */
+	nfs_commit_inode(ino, 0);
 	pnfs_put_layout_hdr(lo);
 	trace_nfs4_cb_layoutrecall_file(clp, &args->cbl_fh, ino,
 			&args->cbl_stateid, -rv);
diff --git a/fs/nfs/direct.c b/fs/nfs/direct.c
index 14f77df79c25..a9a93927fe3e 100644
--- a/fs/nfs/direct.c
+++ b/fs/nfs/direct.c
@@ -721,8 +721,20 @@ static void nfs_direct_commit_complete(struct nfs_commit_data *data)
 		nfs_direct_write_complete(dreq, data->inode);
 }
 
+static void nfs_direct_resched_write(struct nfs_commit_info *cinfo,
+		struct nfs_page *req)
+{
+	struct nfs_direct_req *dreq = cinfo->dreq;
+
+	spin_lock(&dreq->lock);
+	dreq->flags = NFS_ODIRECT_RESCHED_WRITES;
+	spin_unlock(&dreq->lock);
+	nfs_mark_request_commit(req, NULL, cinfo, 0);
+}
+
 static const struct nfs_commit_completion_ops nfs_direct_commit_completion_ops = {
 	.completion = nfs_direct_commit_complete,
+	.resched_write = nfs_direct_resched_write,
 };
 
 static void nfs_direct_commit_schedule(struct nfs_direct_req *dreq)
diff --git a/fs/nfs/pnfs.c b/fs/nfs/pnfs.c
index 360fe95c97b5..6593be7c0129 100644
--- a/fs/nfs/pnfs.c
+++ b/fs/nfs/pnfs.c
@@ -703,6 +703,8 @@ pnfs_layout_free_bulk_destroy_list(struct list_head *layout_list,
 			ret = -EAGAIN;
 		spin_unlock(&inode->i_lock);
 		pnfs_free_lseg_list(&lseg_list);
+		/* Free all lsegs that are attached to commit buckets */
+		nfs_commit_inode(inode, 0);
 		pnfs_put_layout_hdr(lo);
 		iput(inode);
 	}
@@ -1811,6 +1813,7 @@ void pnfs_error_mark_layout_for_return(struct inode *inode,
 	pnfs_mark_matching_lsegs_return(lo, &free_me, &range);
 	spin_unlock(&inode->i_lock);
 	pnfs_free_lseg_list(&free_me);
+	nfs_commit_inode(inode, 0);
 }
 EXPORT_SYMBOL_GPL(pnfs_error_mark_layout_for_return);
 
diff --git a/fs/nfs/pnfs.h b/fs/nfs/pnfs.h
index d93c2ebc0fd3..4bd7faf9ce50 100644
--- a/fs/nfs/pnfs.h
+++ b/fs/nfs/pnfs.h
@@ -412,6 +412,12 @@ pnfs_get_lseg(struct pnfs_layout_segment *lseg)
 	return lseg;
 }
 
+static inline bool
+pnfs_is_valid_lseg(struct pnfs_layout_segment *lseg)
+{
+	return test_bit(NFS_LSEG_VALID, &lseg->pls_flags) != 0;
+}
+
 /* Return true if a layout driver is being used for this mountpoint */
 static inline int pnfs_enabled_sb(struct nfs_server *nfss)
 {
diff --git a/fs/nfs/pnfs_nfs.c b/fs/nfs/pnfs_nfs.c
index 3c8e3a44e6ea..81ac6480f9e7 100644
--- a/fs/nfs/pnfs_nfs.c
+++ b/fs/nfs/pnfs_nfs.c
@@ -868,6 +868,11 @@ pnfs_layout_mark_request_commit(struct nfs_page *req,
 	buckets = cinfo->ds->buckets;
 	list = &buckets[ds_commit_idx].written;
 	if (list_empty(list)) {
+		if (!pnfs_is_valid_lseg(lseg)) {
+			spin_unlock(cinfo->lock);
+			cinfo->completion_ops->resched_write(cinfo, req);
+			return;
+		}
 		/* Non-empty buckets hold a reference on the lseg.  That ref
 		 * is normally transferred to the COMMIT call and released
 		 * there.  It could also be released if the last req is pulled
diff --git a/fs/nfs/write.c b/fs/nfs/write.c
index ae29f082c9c2..0aa8d6f23b4c 100644
--- a/fs/nfs/write.c
+++ b/fs/nfs/write.c
@@ -1676,6 +1676,13 @@ void nfs_retry_commit(struct list_head *page_list,
 }
 EXPORT_SYMBOL_GPL(nfs_retry_commit);
 
+static void
+nfs_commit_resched_write(struct nfs_commit_info *cinfo,
+		struct nfs_page *req)
+{
+	__set_page_dirty_nobuffers(req->wb_page);
+}
+
 /*
  * Commit dirty pages
  */
@@ -1777,6 +1784,7 @@ static const struct rpc_call_ops nfs_commit_ops = {
 
 static const struct nfs_commit_completion_ops nfs_commit_completion_ops = {
 	.completion = nfs_commit_release_pages,
+	.resched_write = nfs_commit_resched_write,
 };
 
 int nfs_generic_commit_list(struct inode *inode, struct list_head *head,
@@ -1823,6 +1831,7 @@ out_mark_dirty:
 	__mark_inode_dirty(inode, I_DIRTY_DATASYNC);
 	return res;
 }
+EXPORT_SYMBOL_GPL(nfs_commit_inode);
 
 int nfs_write_inode(struct inode *inode, struct writeback_control *wbc)
 {
diff --git a/include/linux/nfs_xdr.h b/include/linux/nfs_xdr.h
index a8905b7d4d7f..bee3e60a7006 100644
--- a/include/linux/nfs_xdr.h
+++ b/include/linux/nfs_xdr.h
@@ -1420,10 +1420,12 @@ struct nfs_mds_commit_info {
 	struct list_head	list;
 };
 
+struct nfs_commit_info;
 struct nfs_commit_data;
 struct nfs_inode;
 struct nfs_commit_completion_ops {
 	void (*completion) (struct nfs_commit_data *data);
+	void (*resched_write) (struct nfs_commit_info *, struct nfs_page *);
 };
 
 struct nfs_commit_info {