Received-SPF: pass (google.com: domain of linux-nfs+bounces-3600-linux.lists.archive=gmail.com@vger.kernel.org designates 147.75.48.161 as permitted sender) client-ip=147.75.48.161;
From: Mike Snitzer <snitzer@kernel.org>
To: linux-nfs@vger.kernel.org
Cc: Jeff Layton <jlayton@kernel.org>,
	Chuck Lever <chuck.lever@oracle.com>,
	Trond Myklebust <trondmy@hammerspace.com>,
	snitzer@hammerspace.com
Subject: [for-6.11 PATCH 15/29] NFS: Don't call filesystem write() routines directly
Date: Fri,  7 Jun 2024 10:26:32 -0400
Message-ID: <20240607142646.20924-16-snitzer@kernel.org>
In-Reply-To: <20240607142646.20924-1-snitzer@kernel.org>
References: <20240607142646.20924-1-snitzer@kernel.org>
Precedence: bulk
MIME-Version: 1.0
Content-Transfer-Encoding: 8bit

From: Trond Myklebust <trond.myklebust@hammerspace.com>

Some filesystem writeback routines can end up taking up a lot of stack
space (particularly xfs). Instead of risking running over due to the
extra overhead from the NFS stack, we should just call these routines
from a workqueue job.

Signed-off-by: Trond Myklebust <trond.myklebust@hammerspace.com>
Signed-off-by: Mike Snitzer <snitzer@kernel.org>
---
 fs/nfs/localio.c | 51 ++++++++++++++++++++++++++++++++++++------------
 1 file changed, 38 insertions(+), 13 deletions(-)

diff --git a/fs/nfs/localio.c b/fs/nfs/localio.c
index 5939ca2216be..2c6811b20dcf 100644
--- a/fs/nfs/localio.c
+++ b/fs/nfs/localio.c
@@ -63,6 +63,12 @@ struct nfs_local_fsync_ctx {
 };
 static void nfs_local_fsync_work(struct work_struct *work);
 
+struct nfs_local_io_args {
+	struct nfs_local_kiocb *iocb;
+	struct work_struct work;
+	struct completion *done;
+};
+
 /*
  * We need to translate between nfs status return values and
  * the local errno values which may not be the same.
@@ -597,14 +603,35 @@ nfs_local_write_aio_complete(struct kiocb *kiocb, long ret)
 	nfs_local_pgio_complete(iocb);
 }
 
-static int
-nfs_do_local_write(struct nfs_pgio_header *hdr, struct file *filp,
-		const struct rpc_call_ops *call_ops)
+static void nfs_local_call_write(struct work_struct *work)
 {
-	struct nfs_local_kiocb *iocb;
+	struct nfs_local_io_args *args =
+		container_of(work, struct nfs_local_io_args, work);
+	struct nfs_local_kiocb *iocb = args->iocb;
+	struct file *filp = iocb->kiocb.ki_filp;
 	struct iov_iter iter;
 	ssize_t status;
 
+	nfs_local_iter_init(&iter, iocb, WRITE);
+
+	file_start_write(filp);
+	status = filp->f_op->write_iter(&iocb->kiocb, &iter);
+	file_end_write(filp);
+	if (status != -EIOCBQUEUED) {
+		nfs_local_write_done(iocb, status);
+		nfs_get_vfs_attr(filp, iocb->hdr->res.fattr);
+		nfs_local_pgio_release(iocb);
+	}
+	complete(args->done);
+}
+
+static int nfs_do_local_write(struct nfs_pgio_header *hdr, struct file *filp,
+			      const struct rpc_call_ops *call_ops)
+{
+	struct nfs_local_io_args args;
+	DECLARE_COMPLETION_ONSTACK(done);
+	struct nfs_local_kiocb *iocb;
+
 	dprintk("%s: vfs_write count=%u pos=%llu %s\n",
 		__func__, hdr->args.count, hdr->args.offset,
 		(hdr->args.stable == NFS_UNSTABLE) ?  "unstable" : "stable");
@@ -612,7 +639,6 @@ nfs_do_local_write(struct nfs_pgio_header *hdr, struct file *filp,
 	iocb = nfs_local_iocb_alloc(hdr, filp, GFP_NOIO);
 	if (iocb == NULL)
 		return -ENOMEM;
-	nfs_local_iter_init(&iter, iocb, WRITE);
 
 	switch (hdr->args.stable) {
 	default:
@@ -632,14 +658,13 @@ nfs_do_local_write(struct nfs_pgio_header *hdr, struct file *filp,
 
 	nfs_set_local_verifier(hdr->inode, hdr->res.verf, hdr->args.stable);
 
-	file_start_write(filp);
-	status = filp->f_op->write_iter(&iocb->kiocb, &iter);
-	file_end_write(filp);
-	if (status != -EIOCBQUEUED) {
-		nfs_local_write_done(iocb, status);
-		nfs_get_vfs_attr(filp, hdr->res.fattr);
-		nfs_local_pgio_release(iocb);
-	}
+	args.iocb = iocb;
+	args.done = &done;
+	INIT_WORK_ONSTACK(&args.work, nfs_local_call_write);
+
+	queue_work(nfsiod_workqueue, &args.work);
+	wait_for_completion(&done);
+	destroy_work_on_stack(&args.work);
 	return 0;
 }
 
-- 
2.44.0