NFS client fixes for Linux 4.15-rc4

Stable bugfixes:
 - NFS: Avoid a BUG_ON() in nfs_commit_inode() by not waiting for a
        commit in the case that there were no commit requests.
 - SUNRPC: Fix a race in the receive code path
 
 Other fixes:
 - NFS: Fix a deadlock in nfs client initialization
 - xprtrdma: Fix a performance regression for small IOs
 -----BEGIN PGP SIGNATURE-----
 
 iQIzBAABCAAdFiEEnZ5MQTpR7cLU7KEp18tUv7ClQOsFAlo0PdMACgkQ18tUv7Cl
 QOvlUg/+KoXWXNwItHIyyegYgRXcAPpaCtdnCjjOP6R9HEJ+clnLcaqDxdDKVWQ/
 oDvEcQcsBpywbUi7vVrvdar4mofwuyjXPpbcZPlDP1Ru4yyAlyylftwIuQW/nzdd
 vX2tZaVf+B9y1XvSD5NI+2EKWmp7MVrPdNhYxAB39TQZnAAvYDFHhywtZ0UR7vJt
 7YVcZoPtKUhg15jhCOr73eaCT0884/tlgedfd6DkDGR6bCtSQC2PySfqq9Lnnl/1
 ruDzzcgTARzSEzvta/uyBRspOLBHeeBhTdQUp79lMfekC4+68Tx6DFWnydIUttuE
 G7LphN6hfbJLF20U/ENb2H8v10WZsKvGEuxM+fp5PXGcIMSlX4qoJUe/egJFiiSL
 IaikgibvfiKmYSJvwdxTlOcr793X2Ej19HNciNjJQp4pviDOdZixgtGvVVHJBmh6
 LYzE5q9jgbW9wQXwTTeWHp/nyqL80NslX0UARYnS2Ua0B96GRCESXqCUFtxK6tKR
 wbYiHzKc4dOfSxpNlKI+FlX63m5oSAmTEii3ODsWZjObbwYHNX2Zqj2cVFiSLCpv
 ZXgmpNL+tL2zBWxPvn6rzYhpaXo++PqlHK7vv2QVBI6XM2J8ztpj5Wr5zneRoJaE
 ejk8nw/mR43bfdQuUGZRKh/Z+FTqL0/2WbDgJMXl09c+zRz7J2c=
 =XhEC
 -----END PGP SIGNATURE-----

Merge tag 'nfs-for-4.15-3' of git://git.linux-nfs.org/projects/anna/linux-nfs

Pull NFS client fixes from Anna Schumaker:
 "This has two stable bugfixes, one to fix a BUG_ON() when
  nfs_commit_inode() is called with no outstanding commit requests and
  another to fix a race in the SUNRPC receive codepath.

  Additionally, there are also fixes for an NFS client deadlock and an
  xprtrdma performance regression.

  Summary:

  Stable bugfixes:
   - NFS: Avoid a BUG_ON() in nfs_commit_inode() by not waiting for a
     commit in the case that there were no commit requests.
   - SUNRPC: Fix a race in the receive code path

  Other fixes:
   - NFS: Fix a deadlock in nfs client initialization
   - xprtrdma: Fix a performance regression for small IOs"

* tag 'nfs-for-4.15-3' of git://git.linux-nfs.org/projects/anna/linux-nfs:
  SUNRPC: Fix a race in the receive code path
  nfs: don't wait on commit in nfs_commit_inode() if there were no commit requests
  xprtrdma: Spread reply processing over more CPUs
  nfs: fix a deadlock in nfs client initialization

fs/nfs/client.c

@@ -291,12 +291,23 @@ static struct nfs_client *nfs_match_client(const struct nfs_client_initdata *dat
 	const struct sockaddr *sap = data->addr;
 	struct nfs_net *nn = net_generic(data->net, nfs_net_id);
 
+again:
 	list_for_each_entry(clp, &nn->nfs_client_list, cl_share_link) {
 	        const struct sockaddr *clap = (struct sockaddr *)&clp->cl_addr;
 		/* Don't match clients that failed to initialise properly */
 		if (clp->cl_cons_state < 0)
 			continue;
 
+		/* If a client is still initializing then we need to wait */
+		if (clp->cl_cons_state > NFS_CS_READY) {
+			refcount_inc(&clp->cl_count);
+			spin_unlock(&nn->nfs_client_lock);
+			nfs_wait_client_init_complete(clp);
+			nfs_put_client(clp);
+			spin_lock(&nn->nfs_client_lock);
+			goto again;
+		}
+
 		/* Different NFS versions cannot share the same nfs_client */
 		if (clp->rpc_ops != data->nfs_mod->rpc_ops)
 			continue;

fs/nfs/nfs4client.c

@@ -404,15 +404,19 @@ struct nfs_client *nfs4_init_client(struct nfs_client *clp,
 	if (error < 0)
 		goto error;
 
-	if (!nfs4_has_session(clp))
-		nfs_mark_client_ready(clp, NFS_CS_READY);
-
 	error = nfs4_discover_server_trunking(clp, &old);
 	if (error < 0)
 		goto error;
 
-	if (clp != old)
+	if (clp != old) {
 		clp->cl_preserve_clid = true;
+		/*
+		 * Mark the client as having failed initialization so other
+		 * processes walking the nfs_client_list in nfs_match_client()
+		 * won't try to use it.
+		 */
+		nfs_mark_client_ready(clp, -EPERM);
+	}
 	nfs_put_client(clp);
 	clear_bit(NFS_CS_TSM_POSSIBLE, &clp->cl_flags);
 	return old;
@@ -539,6 +543,9 @@ int nfs40_walk_client_list(struct nfs_client *new,
 	spin_lock(&nn->nfs_client_lock);
 	list_for_each_entry(pos, &nn->nfs_client_list, cl_share_link) {
 
+		if (pos == new)
+			goto found;
+
 		status = nfs4_match_client(pos, new, &prev, nn);
 		if (status < 0)
 			goto out_unlock;
@@ -559,6 +566,7 @@ int nfs40_walk_client_list(struct nfs_client *new,
 		 * way that a SETCLIENTID_CONFIRM to pos can succeed is
 		 * if new and pos point to the same server:
 		 */
+found:
 		refcount_inc(&pos->cl_count);
 		spin_unlock(&nn->nfs_client_lock);
 
@@ -572,6 +580,7 @@ int nfs40_walk_client_list(struct nfs_client *new,
 		case 0:
 			nfs4_swap_callback_idents(pos, new);
 			pos->cl_confirm = new->cl_confirm;
+			nfs_mark_client_ready(pos, NFS_CS_READY);
 
 			prev = NULL;
 			*result = pos;

fs/nfs/write.c

@@ -1890,6 +1890,8 @@ int nfs_commit_inode(struct inode *inode, int how)
 	if (res)
 		error = nfs_generic_commit_list(inode, &head, how, &cinfo);
 	nfs_commit_end(cinfo.mds);
+	if (res == 0)
+		return res;
 	if (error < 0)
 		goto out_error;
 	if (!may_wait)

net/sunrpc/xprt.c

@@ -1001,6 +1001,7 @@ void xprt_transmit(struct rpc_task *task)
 {
 	struct rpc_rqst	*req = task->tk_rqstp;
 	struct rpc_xprt	*xprt = req->rq_xprt;
+	unsigned int connect_cookie;
 	int status, numreqs;
 
 	dprintk("RPC: %5u xprt_transmit(%u)\n", task->tk_pid, req->rq_slen);
@@ -1024,6 +1025,7 @@ void xprt_transmit(struct rpc_task *task)
 	} else if (!req->rq_bytes_sent)
 		return;
 
+	connect_cookie = xprt->connect_cookie;
 	req->rq_xtime = ktime_get();
 	status = xprt->ops->send_request(task);
 	trace_xprt_transmit(xprt, req->rq_xid, status);
@@ -1047,20 +1049,28 @@ void xprt_transmit(struct rpc_task *task)
 	xprt->stat.bklog_u += xprt->backlog.qlen;
 	xprt->stat.sending_u += xprt->sending.qlen;
 	xprt->stat.pending_u += xprt->pending.qlen;
-
-	/* Don't race with disconnect */
-	if (!xprt_connected(xprt))
-		task->tk_status = -ENOTCONN;
-	else {
-		/*
-		 * Sleep on the pending queue since
-		 * we're expecting a reply.
-		 */
-		if (!req->rq_reply_bytes_recvd && rpc_reply_expected(task))
-			rpc_sleep_on(&xprt->pending, task, xprt_timer);
-		req->rq_connect_cookie = xprt->connect_cookie;
-	}
 	spin_unlock_bh(&xprt->transport_lock);
+
+	req->rq_connect_cookie = connect_cookie;
+	if (rpc_reply_expected(task) && !READ_ONCE(req->rq_reply_bytes_recvd)) {
+		/*
+		 * Sleep on the pending queue if we're expecting a reply.
+		 * The spinlock ensures atomicity between the test of
+		 * req->rq_reply_bytes_recvd, and the call to rpc_sleep_on().
+		 */
+		spin_lock(&xprt->recv_lock);
+		if (!req->rq_reply_bytes_recvd) {
+			rpc_sleep_on(&xprt->pending, task, xprt_timer);
+			/*
+			 * Send an extra queue wakeup call if the
+			 * connection was dropped in case the call to
+			 * rpc_sleep_on() raced.
+			 */
+			if (!xprt_connected(xprt))
+				xprt_wake_pending_tasks(xprt, -ENOTCONN);
+		}
+		spin_unlock(&xprt->recv_lock);
+	}
 }
 
 static void xprt_add_backlog(struct rpc_xprt *xprt, struct rpc_task *task)

net/sunrpc/xprtrdma/rpc_rdma.c

@@ -1408,11 +1408,7 @@ void rpcrdma_reply_handler(struct rpcrdma_rep *rep)
 	dprintk("RPC:       %s: reply %p completes request %p (xid 0x%08x)\n",
 		__func__, rep, req, be32_to_cpu(rep->rr_xid));
 
-	if (list_empty(&req->rl_registered) &&
-	    !test_bit(RPCRDMA_REQ_F_TX_RESOURCES, &req->rl_flags))
-		rpcrdma_complete_rqst(rep);
-	else
-		queue_work(rpcrdma_receive_wq, &rep->rr_work);
+	queue_work_on(req->rl_cpu, rpcrdma_receive_wq, &rep->rr_work);
 	return;
 
 out_badstatus:

net/sunrpc/xprtrdma/transport.c

@@ -52,6 +52,7 @@
 #include <linux/slab.h>
 #include <linux/seq_file.h>
 #include <linux/sunrpc/addr.h>
+#include <linux/smp.h>
 
 #include "xprt_rdma.h"
 
@@ -656,6 +657,7 @@ xprt_rdma_allocate(struct rpc_task *task)
 		task->tk_pid, __func__, rqst->rq_callsize,
 		rqst->rq_rcvsize, req);
 
+	req->rl_cpu = smp_processor_id();
 	req->rl_connect_cookie = 0;	/* our reserved value */
 	rpcrdma_set_xprtdata(rqst, req);
 	rqst->rq_buffer = req->rl_sendbuf->rg_base;

net/sunrpc/xprtrdma/verbs.c

@@ -83,7 +83,7 @@ rpcrdma_alloc_wq(void)
 	struct workqueue_struct *recv_wq;
 
 	recv_wq = alloc_workqueue("xprtrdma_receive",
-				  WQ_MEM_RECLAIM | WQ_UNBOUND | WQ_HIGHPRI,
+				  WQ_MEM_RECLAIM | WQ_HIGHPRI,
 				  0);
 	if (!recv_wq)
 		return -ENOMEM;

net/sunrpc/xprtrdma/xprt_rdma.h

@@ -342,6 +342,7 @@ enum {
 struct rpcrdma_buffer;
 struct rpcrdma_req {
 	struct list_head	rl_list;
+	int			rl_cpu;
 	unsigned int		rl_connect_cookie;
 	struct rpcrdma_buffer	*rl_buffer;
 	struct rpcrdma_rep	*rl_reply;