Alex Burlyga reported a POLA violation for the new NFS client as

compared to the old NFS client via email to the freebsd-fs@ mailing list.
For the new client, when multiple clients attempted to create a symbolic
link concurrently, more that one client would report success instead of
EEXIST. This was caused by code in the new client that mapped EEXIST to
OK assuming it was caused by a retried RPC request.
Since the old client did not do this, the patch defaults to the old
behaviour and permits the new behaviour to be enabled via a sysctl.

Reported by:	alex.burlyga.ietf@gmail.com
Tested by:	alex.burlyga.ietf@gmail.com
MFC after:	2 weeks

1 files changed, 20 insertions, 3 deletions

diff --git a/sys/fs/nfsclient/nfs_clrpcops.c b/sys/fs/nfsclient/nfs_clrpcops.c
index 55eb983..4528855 100644
--- a/sys/fs/nfsclient/nfs_clrpcops.c
+++ b/sys/fs/nfsclient/nfs_clrpcops.c
@@ -46,6 +46,13 @@ __FBSDID("$FreeBSD$");
 #include "opt_inet6.h"
 
 #include <fs/nfs/nfsport.h>
+#include <sys/sysctl.h>
+
+SYSCTL_DECL(_vfs_nfs);
+
+static int	nfsignore_eexist = 0;
+SYSCTL_INT(_vfs_nfs, OID_AUTO, ignore_eexist, CTLFLAG_RW,
+    &nfsignore_eexist, 0, "NFS ignore EEXIST replies for mkdir/symlink");
 
 /*
  * Global variables
@@ -2530,8 +2537,12 @@ nfsrpc_symlink(vnode_t dvp, char *name, int namelen, char *target,
 	mbuf_freem(nd->nd_mrep);
 	/*
 	 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry.
+	 * Only do this if vfs.nfs.ignore_eexist is set.
+	 * Never do this for NFSv4.1 or later minor versions, since sessions
+	 * should guarantee "exactly once" RPC semantics.
 	 */
-	if (error == EEXIST)
+	if (error == EEXIST && nfsignore_eexist != 0 && (!NFSHASNFSV4(nmp) ||
+	    nmp->nm_minorvers == 0))
 		error = 0;
 	return (error);
 }
@@ -2550,10 +2561,12 @@ nfsrpc_mkdir(vnode_t dvp, char *name, int namelen, struct vattr *vap,
 	nfsattrbit_t attrbits;
 	int error = 0;
 	struct nfsfh *fhp;
+	struct nfsmount *nmp;
 
 	*nfhpp = NULL;
 	*attrflagp = 0;
 	*dattrflagp = 0;
+	nmp = VFSTONFS(vnode_mount(dvp));
 	fhp = VTONFS(dvp)->n_fhp;
 	if (namelen > NFS_MAXNAMLEN)
 		return (ENAMETOOLONG);
@@ -2605,9 +2618,13 @@ nfsrpc_mkdir(vnode_t dvp, char *name, int namelen, struct vattr *vap,
 nfsmout:
 	mbuf_freem(nd->nd_mrep);
 	/*
-	 * Kludge: Map EEXIST => 0 assuming that you have a reply to a retry.
+	 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry.
+	 * Only do this if vfs.nfs.ignore_eexist is set.
+	 * Never do this for NFSv4.1 or later minor versions, since sessions
+	 * should guarantee "exactly once" RPC semantics.
 	 */
-	if (error == EEXIST)
+	if (error == EEXIST && nfsignore_eexist != 0 && (!NFSHASNFSV4(nmp) ||
+	    nmp->nm_minorvers == 0))
 		error = 0;
 	return (error);
 }