[08/11] NFSv4/pNFS: Retry the layout return later in case of a timeout or reboot

Commit Message

Trond Myklebust June 13, 2024, 5 a.m. UTC

From: Trond Myklebust <[e-mail]>

If the layout return failed due to a timeout or reboot, then leave the
layout segments on the list so that the layout return gets replayed
later.
The exception would be if we're freeing the inode.

Signed-off-by: Trond Myklebust <[e-mail]>
---
 fs/nfs/nfs4proc.c | 21 ++++++++++++++++++++-
 fs/nfs/pnfs.c     | 12 ++++++++++++
 fs/nfs/pnfs.h     |  3 +++
 3 files changed, 35 insertions(+), 1 deletion(-)

Patch

diff --git a/fs/nfs/nfs4proc.c b/fs/nfs/nfs4proc.c
index ae835d14ac75..952d1e930185 100644
--- a/fs/nfs/nfs4proc.c
+++ b/fs/nfs/nfs4proc.c
@@ -9944,6 +9944,11 @@  static void nfs4_layoutreturn_done(struct rpc_task *task, void *calldata)
 	if (!nfs41_sequence_process(task, &lrp->res.seq_res))
 		return;
 
+	if (task->tk_rpc_status == -ETIMEDOUT) {
+		lrp->rpc_status = -EAGAIN;
+		lrp->res.lrs_present = 0;
+		return;
+	}
 	/*
 	 * Was there an RPC level error? Assume the call succeeded,
 	 * and that we need to release the layout
@@ -9966,6 +9971,15 @@  static void nfs4_layoutreturn_done(struct rpc_task *task, void *calldata)
 		fallthrough;
 	case 0:
 		break;
+	case -NFS4ERR_BADSESSION:
+	case -NFS4ERR_DEADSESSION:
+	case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
+		nfs4_schedule_session_recovery(server->nfs_client->cl_session,
+					       task->tk_status);
+		lrp->res.lrs_present = 0;
+		lrp->rpc_status = -EAGAIN;
+		task->tk_status = 0;
+		break;
 	case -NFS4ERR_DELAY:
 		if (nfs4_async_handle_error(task, server, NULL, NULL) != -EAGAIN)
 			break;
@@ -9983,8 +9997,13 @@  static void nfs4_layoutreturn_release(void *calldata)
 	struct nfs4_layoutreturn *lrp = calldata;
 	struct pnfs_layout_hdr *lo = lrp->args.layout;
 
-	pnfs_layoutreturn_free_lsegs(lo, &lrp->args.stateid, &lrp->args.range,
+	if (lrp->rpc_status == 0 || !lrp->inode)
+		pnfs_layoutreturn_free_lsegs(
+			lo, &lrp->args.stateid, &lrp->args.range,
 			lrp->res.lrs_present ? &lrp->res.stateid : NULL);
+	else
+		pnfs_layoutreturn_retry_later(lo, &lrp->args.stateid,
+					      &lrp->args.range);
 	nfs4_sequence_free_slot(&lrp->res.seq_res);
 	if (lrp->ld_private.ops && lrp->ld_private.ops->free)
 		lrp->ld_private.ops->free(&lrp->ld_private);
diff --git a/fs/nfs/pnfs.c b/fs/nfs/pnfs.c
index c8b1be1810e2..04a52fa3d28c 100644
--- a/fs/nfs/pnfs.c
+++ b/fs/nfs/pnfs.c
@@ -1191,6 +1191,18 @@  pnfs_layoutreturn_retry_later_locked(struct pnfs_layout_hdr *lo,
 	}
 }
 
+void pnfs_layoutreturn_retry_later(struct pnfs_layout_hdr *lo,
+				   const nfs4_stateid *arg_stateid,
+				   const struct pnfs_layout_range *range)
+{
+	struct inode *inode = lo->plh_inode;
+
+	spin_lock(&inode->i_lock);
+	pnfs_layoutreturn_retry_later_locked(lo, arg_stateid, range);
+	pnfs_clear_layoutreturn_waitbit(lo);
+	spin_unlock(&inode->i_lock);
+}
+
 void pnfs_layoutreturn_free_lsegs(struct pnfs_layout_hdr *lo,
 		const nfs4_stateid *arg_stateid,
 		const struct pnfs_layout_range *range,
diff --git a/fs/nfs/pnfs.h b/fs/nfs/pnfs.h
index 8fa0f152ed19..cd23a38eac75 100644
--- a/fs/nfs/pnfs.h
+++ b/fs/nfs/pnfs.h
@@ -328,6 +328,9 @@  struct pnfs_layout_segment *pnfs_update_layout(struct inode *ino,
 					       enum pnfs_iomode iomode,
 					       bool strict_iomode,
 					       gfp_t gfp_flags);
+void pnfs_layoutreturn_retry_later(struct pnfs_layout_hdr *lo,
+				   const nfs4_stateid *arg_stateid,
+				   const struct pnfs_layout_range *range);
 void pnfs_layoutreturn_free_lsegs(struct pnfs_layout_hdr *lo,
 		const nfs4_stateid *arg_stateid,
 		const struct pnfs_layout_range *range,