Received-SPF: pass (google.com: domain of linux-nfs-owner@vger.kernel.org designates 2620:137:e000::1:18 as permitted sender) client-ip=2620:137:e000::1:18;
Subject: [PATCH RFC] NFSD: Instantiate a filecache entry when creating a file
From:   Chuck Lever <chuck.lever@oracle.com>
To:     bfields@fieldses.org
Cc:     jlayton@kernel.org, linux-nfs@vger.kernel.org
Date:   Thu, 24 Mar 2022 18:08:01 -0400
Message-ID: <164815968129.1809.12154191330176123202.stgit@manet.1015granger.net>
User-Agent: StGit/1.5
MIME-Version: 1.0
Content-Type: text/plain; charset="utf-8"
Content-Transfer-Encoding: 7bit
Precedence: bulk

There have been reports of races that cause NFSv4 OPEN(CREATE) to
return an error even though the requested file was created. NFSv4
does not seem to provide a status code for this case.

There are plenty of things that can go wrong between the
vfs_create() call in do_nfsd_create() and nfs4_vfs_get_file(), but
one of them is another client looking up and modifying the file's
mode bits in that window. If that happens, the creator might lose
access to the file before it can be opened.

Instead of opening the newly created file in nfsd4_process_open2(),
open it as soon as the file is created, and leave it sitting in the
file cache.

This patch is not optimal, of course - we should replace the
vfs_create() call in do_nfsd_create() with a call that creates the
file and returns a "struct file *" that can be planted immediately
in nf->nf_file.

But first, I would like to hear opinions about the approach
suggested above.

BugLink: https://bugzilla.linux-nfs.org/show_bug.cgi?id=382
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
---
 fs/nfsd/vfs.c |    9 +++++++++
 1 file changed, 9 insertions(+)

diff --git a/fs/nfsd/vfs.c b/fs/nfsd/vfs.c
index 02a544645b52..80b568fa12f1 100644
--- a/fs/nfsd/vfs.c
+++ b/fs/nfsd/vfs.c
@@ -1410,6 +1410,7 @@ do_nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp,
 	__be32		err;
 	int		host_err;
 	__u32		v_mtime=0, v_atime=0;
+	struct nfsd_file *nf;
 
 	/* XXX: Shouldn't this be nfserr_inval? */
 	err = nfserr_perm;
@@ -1535,6 +1536,14 @@ do_nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp,
 		iap->ia_atime.tv_nsec = 0;
 	}
 
+	/* Attempt to instantiate a filecache entry while we still hold
+	 * the parent's inode mutex. */
+	err = nfsd_file_acquire(rqstp, resfhp, NFSD_MAY_WRITE, &nf);
+	if (err)
+		/* This would be bad */
+		goto out;
+	nfsd_file_put(nf);
+
  set_attr:
 	err = nfsd_create_setattr(rqstp, resfhp, iap);