NFS: Fix error handling for O_DIRECT write scheduling

[ Upstream commit 954998b ]

If we fail to schedule a request for transmission, there are 2
possibilities:
1) Either we hit a fatal error, and we just want to drop the remaining
   requests on the floor.
2) We were asked to try again, in which case we should allow the
   outstanding RPC calls to complete, so that we can recoalesce requests
   and try again.

Fixes: d600ad1 ("NFS41: pop some layoutget errors to application")
Signed-off-by: Trond Myklebust <trond.myklebust@hammerspace.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>

fs/nfs/direct.c

@@ -530,10 +530,9 @@ nfs_direct_write_scan_commit_list(struct inode *inode,
 static void nfs_direct_write_reschedule(struct nfs_direct_req *dreq)
 {
 	struct nfs_pageio_descriptor desc;
-	struct nfs_page *req, *tmp;
+	struct nfs_page *req;
 	LIST_HEAD(reqs);
 	struct nfs_commit_info cinfo;
-	LIST_HEAD(failed);
 
 	nfs_init_cinfo_from_dreq(&cinfo, dreq);
 	nfs_direct_write_scan_commit_list(dreq->inode, &reqs, &cinfo);
@@ -551,27 +550,36 @@ static void nfs_direct_write_reschedule(struct nfs_direct_req *dreq)
 			      &nfs_direct_write_completion_ops);
 	desc.pg_dreq = dreq;
 
-	list_for_each_entry_safe(req, tmp, &reqs, wb_list) {
+	while (!list_empty(&reqs)) {
+		req = nfs_list_entry(reqs.next);
 		/* Bump the transmission count */
 		req->wb_nio++;
 		if (!nfs_pageio_add_request(&desc, req)) {
-			nfs_list_move_request(req, &failed);
 			spin_lock(&cinfo.inode->i_lock);
-			dreq->flags = 0;
-			if (desc.pg_error < 0)
+			if (dreq->error < 0) {
+				desc.pg_error = dreq->error;
+			} else if (desc.pg_error != -EAGAIN) {
+				dreq->flags = 0;
+				if (!desc.pg_error)
+					desc.pg_error = -EIO;
 				dreq->error = desc.pg_error;
-			else
-				dreq->error = -EIO;
+			} else
+				dreq->flags = NFS_ODIRECT_RESCHED_WRITES;
 			spin_unlock(&cinfo.inode->i_lock);
+			break;
 		}
 		nfs_release_request(req);
 	}
 	nfs_pageio_complete(&desc);
 
-	while (!list_empty(&failed)) {
-		req = nfs_list_entry(failed.next);
+	while (!list_empty(&reqs)) {
+		req = nfs_list_entry(reqs.next);
 		nfs_list_remove_request(req);
 		nfs_unlock_and_release_request(req);
+		if (desc.pg_error == -EAGAIN)
+			nfs_mark_request_commit(req, NULL, &cinfo, 0);
+		else
+			nfs_release_request(req);
 	}
 
 	if (put_dreq(dreq))
@@ -796,9 +804,11 @@ static ssize_t nfs_direct_write_schedule_iovec(struct nfs_direct_req *dreq,
 {
 	struct nfs_pageio_descriptor desc;
 	struct inode *inode = dreq->inode;
+	struct nfs_commit_info cinfo;
 	ssize_t result = 0;
 	size_t requested_bytes = 0;
 	size_t wsize = max_t(size_t, NFS_SERVER(inode)->wsize, PAGE_SIZE);
+	bool defer = false;
 
 	trace_nfs_direct_write_schedule_iovec(dreq);
 
@@ -839,19 +849,39 @@ static ssize_t nfs_direct_write_schedule_iovec(struct nfs_direct_req *dreq,
 				break;
 			}
 
+			pgbase = 0;
+			bytes -= req_len;
+			requested_bytes += req_len;
+			pos += req_len;
+			dreq->bytes_left -= req_len;
+
+			if (defer) {
+				nfs_mark_request_commit(req, NULL, &cinfo, 0);
+				continue;
+			}
+
 			nfs_lock_request(req);
 			req->wb_index = pos >> PAGE_SHIFT;
 			req->wb_offset = pos & ~PAGE_MASK;
-			if (!nfs_pageio_add_request(&desc, req)) {
+			if (nfs_pageio_add_request(&desc, req))
+				continue;
+
+			/* Exit on hard errors */
+			if (desc.pg_error < 0 && desc.pg_error != -EAGAIN) {
 				result = desc.pg_error;
 				nfs_unlock_and_release_request(req);
 				break;
 			}
-			pgbase = 0;
-			bytes -= req_len;
-			requested_bytes += req_len;
-			pos += req_len;
-			dreq->bytes_left -= req_len;
+
+			/* If the error is soft, defer remaining requests */
+			nfs_init_cinfo_from_dreq(&cinfo, dreq);
+			spin_lock(&cinfo.inode->i_lock);
+			dreq->flags = NFS_ODIRECT_RESCHED_WRITES;
+			spin_unlock(&cinfo.inode->i_lock);
+			nfs_unlock_request(req);
+			nfs_mark_request_commit(req, NULL, &cinfo, 0);
+			desc.pg_error = 0;
+			defer = true;
 		}
 		nfs_direct_release_pages(pagevec, npages);
 		kvfree(pagevec);