nfsd: turn on reply cache for NFSv4

It's sort of ridiculous that we've never had a working reply cache for NFSv4. On the other hand, we may still not: our current reply cache is likely not very good, especially in the TCP case (which is the only case that matters for v4). What we really need here is some serious testing. Anyway, here's a start. Signed-off-by: J. Bruce Fields <bfields@redhat.com>

nfsd: turn on reply cache for NFSv4
It's sort of ridiculous that we've never had a working reply cache for NFSv4. On the other hand, we may still not: our current reply cache is likely not very good, especially in the TCP case (which is the only case that matters for v4). What we really need here is some serious testing. Anyway, here's a start. Signed-off-by: J. Bruce Fields <bfields@redhat.com>
J. Bruce Fields
1 parent 3e98abffd1
Showing 7 changed files with 55 additions and 23 deletions Side-by-side Diff
fs/nfsd/cache.h
fs/nfsd/nfs4proc.c
fs/nfsd/nfs4xdr.c
fs/nfsd/nfscache.c
fs/nfsd/nfssvc.c
fs/nfsd/xdr4.h
include/linux/sunrpc/svc.h
@@ -69,7 +69,7 @@
  
 int	nfsd_reply_cache_init(void);
 void	nfsd_reply_cache_shutdown(void);
-int	nfsd_cache_lookup(struct svc_rqst *, int);
+int	nfsd_cache_lookup(struct svc_rqst *);
 void	nfsd_cache_update(struct svc_rqst *, int, __be32 *);
  
 #ifdef CONFIG_NFSD_V4
@@ -1007,6 +1007,15 @@
 	nfsd4op_func op_func;
 	u32 op_flags;
 	char *op_name;
+	/*
+	 * We use the DRC for compounds containing non-idempotent
+	 * operations, *except* those that are 4.1-specific (since
+	 * sessions provide their own EOS), and except for stateful
+	 * operations other than setclientid and setclientid_confirm
+	 * (since sequence numbers provide EOS for open, lock, etc in
+	 * the v4.0 case).
+	 */
+	bool op_cacheresult;
 };
  
 static struct nfsd4_operation nfsd4_ops[];
@@ -1051,6 +1060,11 @@
 	return &nfsd4_ops[op->opnum];
 }
  
+bool nfsd4_cache_this_op(struct nfsd4_op *op)
+{
+	return OPDESC(op)->op_cacheresult;
+}
+
 static bool need_wrongsec_check(struct svc_rqst *rqstp)
 {
 	struct nfsd4_compoundres *resp = rqstp->rq_resp;
@@ -1240,6 +1254,7 @@
 	[OP_CREATE] = {
 		.op_func = (nfsd4op_func)nfsd4_create,
 		.op_name = "OP_CREATE",
+		.op_cacheresult = true,
 	},
 	[OP_DELEGRETURN] = {
 		.op_func = (nfsd4op_func)nfsd4_delegreturn,
@@ -1257,6 +1272,7 @@
 	[OP_LINK] = {
 		.op_func = (nfsd4op_func)nfsd4_link,
 		.op_name = "OP_LINK",
+		.op_cacheresult = true,
 	},
 	[OP_LOCK] = {
 		.op_func = (nfsd4op_func)nfsd4_lock,
  
@@ -1330,10 +1346,12 @@
 	[OP_REMOVE] = {
 		.op_func = (nfsd4op_func)nfsd4_remove,
 		.op_name = "OP_REMOVE",
+		.op_cacheresult = true,
 	},
 	[OP_RENAME] = {
 		.op_name = "OP_RENAME",
 		.op_func = (nfsd4op_func)nfsd4_rename,
+		.op_cacheresult = true,
 	},
 	[OP_RENEW] = {
 		.op_func = (nfsd4op_func)nfsd4_renew,
  
  
@@ -1359,16 +1377,19 @@
 	[OP_SETATTR] = {
 		.op_func = (nfsd4op_func)nfsd4_setattr,
 		.op_name = "OP_SETATTR",
+		.op_cacheresult = true,
 	},
 	[OP_SETCLIENTID] = {
 		.op_func = (nfsd4op_func)nfsd4_setclientid,
 		.op_flags = ALLOWED_WITHOUT_FH | ALLOWED_ON_ABSENT_FS,
 		.op_name = "OP_SETCLIENTID",
+		.op_cacheresult = true,
 	},
 	[OP_SETCLIENTID_CONFIRM] = {
 		.op_func = (nfsd4op_func)nfsd4_setclientid_confirm,
 		.op_flags = ALLOWED_WITHOUT_FH | ALLOWED_ON_ABSENT_FS,
 		.op_name = "OP_SETCLIENTID_CONFIRM",
+		.op_cacheresult = true,
 	},
 	[OP_VERIFY] = {
 		.op_func = (nfsd4op_func)nfsd4_verify,
@@ -1377,6 +1398,7 @@
 	[OP_WRITE] = {
 		.op_func = (nfsd4op_func)nfsd4_write,
 		.op_name = "OP_WRITE",
+		.op_cacheresult = true,
 	},
 	[OP_RELEASE_LOCKOWNER] = {
 		.op_func = (nfsd4op_func)nfsd4_release_lockowner,
@@ -1447,16 +1469,6 @@
 #define nfsd4_voidres			nfsd4_voidargs
 struct nfsd4_voidargs { int dummy; };
  
-/*
- * TODO: At the present time, the NFSv4 server does not do XID caching
- * of requests.  Implementing XID caching would not be a serious problem,
- * although it would require a mild change in interfaces since one
- * doesn't know whether an NFSv4 request is idempotent until after the
- * XDR decode.  However, XID caching totally confuses pynfs (Peter
- * Astrand's regression testsuite for NFSv4 servers), which reuses
- * XID's liberally, so I've left it unimplemented until pynfs generates
- * better XID's.
- */
 static struct svc_procedure		nfsd_procedures4[2] = {
 	[NFSPROC4_NULL] = {
 		.pc_func = (svc_procfunc) nfsd4_proc_null,
@@ -52,6 +52,7 @@
 #include "xdr4.h"
 #include "vfs.h"
 #include "state.h"
+#include "cache.h"
  
 #define NFSDDBG_FACILITY		NFSDDBG_XDR
  
@@ -1466,6 +1467,7 @@
 	DECODE_HEAD;
 	struct nfsd4_op *op;
 	struct nfsd4_minorversion_ops *ops;
+	bool cachethis = false;
 	int i;
  
 	/*
  
@@ -1547,7 +1549,16 @@
 			argp->opcnt = i+1;
 			break;
 		}
+		/*
+		 * We'll try to cache the result in the DRC if any one
+		 * op in the compound wants to be cached:
+		 */
+		cachethis |= nfsd4_cache_this_op(op);
 	}
+	/* Sessions make the DRC unnecessary: */
+	if (argp->minorversion)
+		cachethis = false;
+	argp->rqstp->rq_cachetype = cachethis ? RC_REPLBUFF : RC_NOCACHE;
  
 	DECODE_TAIL;
 }
@@ -118,7 +118,7 @@
  * Note that no operation within the loop may sleep.
  */
 int
-nfsd_cache_lookup(struct svc_rqst *rqstp, int type)
+nfsd_cache_lookup(struct svc_rqst *rqstp)
 {
 	struct hlist_node	*hn;
 	struct hlist_head 	*rh;
@@ -128,6 +128,7 @@
 				vers = rqstp->rq_vers,
 				proc = rqstp->rq_proc;
 	unsigned long		age;
+	int type = rqstp->rq_cachetype;
 	int rtn;
  
 	rqstp->rq_cacherep = NULL;
@@ -570,8 +570,22 @@
 				rqstp->rq_vers, rqstp->rq_proc);
 	proc = rqstp->rq_procinfo;
  
+	/*
+	 * Give the xdr decoder a chance to change this if it wants
+	 * (necessary in the NFSv4.0 compound case)
+	 */
+	rqstp->rq_cachetype = proc->pc_cachetype;
+	/* Decode arguments */
+	xdr = proc->pc_decode;
+	if (xdr && !xdr(rqstp, (__be32*)rqstp->rq_arg.head[0].iov_base,
+			rqstp->rq_argp)) {
+		dprintk("nfsd: failed to decode arguments!\n");
+		*statp = rpc_garbage_args;
+		return 1;
+	}
+
 	/* Check whether we have this call in the cache. */
-	switch (nfsd_cache_lookup(rqstp, proc->pc_cachetype)) {
+	switch (nfsd_cache_lookup(rqstp)) {
 	case RC_INTR:
 	case RC_DROPIT:
 		return 0;
@@ -579,16 +593,6 @@
 		return 1;
 	case RC_DOIT:;
 		/* do it */
-	}
-
-	/* Decode arguments */
-	xdr = proc->pc_decode;
-	if (xdr && !xdr(rqstp, (__be32*)rqstp->rq_arg.head[0].iov_base,
-			rqstp->rq_argp)) {
-		dprintk("nfsd: failed to decode arguments!\n");
-		nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
-		*statp = rpc_garbage_args;
-		return 1;
 	}
  
 	/* need to grab the location to store the status, as
@@ -457,6 +457,8 @@
 	struct nfs4_replay *			replay;
 };
  
+bool nfsd4_cache_this_op(struct nfsd4_op *);
+
 struct nfsd4_compoundargs {
 	/* scratch variables for XDR decode */
 	__be32 *			p;
@@ -479,6 +481,7 @@
 	u32				opcnt;
 	struct nfsd4_op			*ops;
 	struct nfsd4_op			iops[8];
+	int				cachetype;
 };
  
 struct nfsd4_compoundres {
@@ -273,6 +273,7 @@
 	/* Catering to nfsd */
 	struct auth_domain *	rq_client;	/* RPC peer info */
 	struct auth_domain *	rq_gssclient;	/* "gss/"-style peer info */
+	int			rq_cachetype;
 	struct svc_cacherep *	rq_cacherep;	/* cache info */
 	int			rq_splice_ok;   /* turned off in gss privacy
 						 * to prevent encrypting page
...	...	@@ -69,7 +69,7 @@
69	69
70	70	int nfsd_reply_cache_init(void);
71	71	void nfsd_reply_cache_shutdown(void);
72		-int nfsd_cache_lookup(struct svc_rqst *, int);
	72	+int nfsd_cache_lookup(struct svc_rqst *);
73	73	void nfsd_cache_update(struct svc_rqst , int, __be32 );
74	74
75	75	#ifdef CONFIG_NFSD_V4
...	...	@@ -1007,6 +1007,15 @@
1007	1007	nfsd4op_func op_func;
1008	1008	u32 op_flags;
1009	1009	char *op_name;
	1010	+ /*
	1011	+ * We use the DRC for compounds containing non-idempotent
	1012	+ * operations, except those that are 4.1-specific (since
	1013	+ * sessions provide their own EOS), and except for stateful
	1014	+ * operations other than setclientid and setclientid_confirm
	1015	+ * (since sequence numbers provide EOS for open, lock, etc in
	1016	+ * the v4.0 case).
	1017	+ */
	1018	+ bool op_cacheresult;
1010	1019	};
1011	1020
1012	1021	static struct nfsd4_operation nfsd4_ops[];
...	...	@@ -1051,6 +1060,11 @@
1051	1060	return &nfsd4_ops[op->opnum];
1052	1061	}
1053	1062
	1063	+bool nfsd4_cache_this_op(struct nfsd4_op *op)
	1064	+{
	1065	+ return OPDESC(op)->op_cacheresult;
	1066	+}
	1067	+
1054	1068	static bool need_wrongsec_check(struct svc_rqst *rqstp)
1055	1069	{
1056	1070	struct nfsd4_compoundres *resp = rqstp->rq_resp;
...	...	@@ -1240,6 +1254,7 @@
1240	1254	[OP_CREATE] = {
1241	1255	.op_func = (nfsd4op_func)nfsd4_create,
1242	1256	.op_name = "OP_CREATE",
	1257	+ .op_cacheresult = true,
1243	1258	},
1244	1259	[OP_DELEGRETURN] = {
1245	1260	.op_func = (nfsd4op_func)nfsd4_delegreturn,
...	...	@@ -1257,6 +1272,7 @@
1257	1272	[OP_LINK] = {
1258	1273	.op_func = (nfsd4op_func)nfsd4_link,
1259	1274	.op_name = "OP_LINK",
	1275	+ .op_cacheresult = true,
1260	1276	},
1261	1277	[OP_LOCK] = {
1262	1278	.op_func = (nfsd4op_func)nfsd4_lock,
1263	1279
...	...	@@ -1330,10 +1346,12 @@
1330	1346	[OP_REMOVE] = {
1331	1347	.op_func = (nfsd4op_func)nfsd4_remove,
1332	1348	.op_name = "OP_REMOVE",
	1349	+ .op_cacheresult = true,
1333	1350	},
1334	1351	[OP_RENAME] = {
1335	1352	.op_name = "OP_RENAME",
1336	1353	.op_func = (nfsd4op_func)nfsd4_rename,
	1354	+ .op_cacheresult = true,
1337	1355	},
1338	1356	[OP_RENEW] = {
1339	1357	.op_func = (nfsd4op_func)nfsd4_renew,
1340	1358
1341	1359
...	...	@@ -1359,16 +1377,19 @@
1359	1377	[OP_SETATTR] = {
1360	1378	.op_func = (nfsd4op_func)nfsd4_setattr,
1361	1379	.op_name = "OP_SETATTR",
	1380	+ .op_cacheresult = true,
1362	1381	},
1363	1382	[OP_SETCLIENTID] = {
1364	1383	.op_func = (nfsd4op_func)nfsd4_setclientid,
1365	1384	.op_flags = ALLOWED_WITHOUT_FH \| ALLOWED_ON_ABSENT_FS,
1366	1385	.op_name = "OP_SETCLIENTID",
	1386	+ .op_cacheresult = true,
1367	1387	},
1368	1388	[OP_SETCLIENTID_CONFIRM] = {
1369	1389	.op_func = (nfsd4op_func)nfsd4_setclientid_confirm,
1370	1390	.op_flags = ALLOWED_WITHOUT_FH \| ALLOWED_ON_ABSENT_FS,
1371	1391	.op_name = "OP_SETCLIENTID_CONFIRM",
	1392	+ .op_cacheresult = true,
1372	1393	},
1373	1394	[OP_VERIFY] = {
1374	1395	.op_func = (nfsd4op_func)nfsd4_verify,
...	...	@@ -1377,6 +1398,7 @@
1377	1398	[OP_WRITE] = {
1378	1399	.op_func = (nfsd4op_func)nfsd4_write,
1379	1400	.op_name = "OP_WRITE",
	1401	+ .op_cacheresult = true,
1380	1402	},
1381	1403	[OP_RELEASE_LOCKOWNER] = {
1382	1404	.op_func = (nfsd4op_func)nfsd4_release_lockowner,
...	...	@@ -1447,16 +1469,6 @@
1447	1469	#define nfsd4_voidres nfsd4_voidargs
1448	1470	struct nfsd4_voidargs { int dummy; };
1449	1471
1450		-/*
1451		- * TODO: At the present time, the NFSv4 server does not do XID caching
1452		- * of requests. Implementing XID caching would not be a serious problem,
1453		- * although it would require a mild change in interfaces since one
1454		- * doesn't know whether an NFSv4 request is idempotent until after the
1455		- * XDR decode. However, XID caching totally confuses pynfs (Peter
1456		- * Astrand's regression testsuite for NFSv4 servers), which reuses
1457		- * XID's liberally, so I've left it unimplemented until pynfs generates
1458		- * better XID's.
1459		- */
1460	1472	static struct svc_procedure nfsd_procedures4[2] = {
1461	1473	[NFSPROC4_NULL] = {
1462	1474	.pc_func = (svc_procfunc) nfsd4_proc_null,
...	...	@@ -52,6 +52,7 @@
52	52	#include "xdr4.h"
53	53	#include "vfs.h"
54	54	#include "state.h"
	55	+#include "cache.h"
55	56
56	57	#define NFSDDBG_FACILITY NFSDDBG_XDR
57	58
...	...	@@ -1466,6 +1467,7 @@
1466	1467	DECODE_HEAD;
1467	1468	struct nfsd4_op *op;
1468	1469	struct nfsd4_minorversion_ops *ops;
	1470	+ bool cachethis = false;
1469	1471	int i;
1470	1472
1471	1473	/*
1472	1474
...	...	@@ -1547,7 +1549,16 @@
1547	1549	argp->opcnt = i+1;
1548	1550	break;
1549	1551	}
	1552	+ /*
	1553	+ * We'll try to cache the result in the DRC if any one
	1554	+ * op in the compound wants to be cached:
	1555	+ */
	1556	+ cachethis \|= nfsd4_cache_this_op(op);
1550	1557	}
	1558	+ /* Sessions make the DRC unnecessary: */
	1559	+ if (argp->minorversion)
	1560	+ cachethis = false;
	1561	+ argp->rqstp->rq_cachetype = cachethis ? RC_REPLBUFF : RC_NOCACHE;
1551	1562
1552	1563	DECODE_TAIL;
1553	1564	}
...	...	@@ -118,7 +118,7 @@
118	118	* Note that no operation within the loop may sleep.
119	119	*/
120	120	int
121		-nfsd_cache_lookup(struct svc_rqst *rqstp, int type)
	121	+nfsd_cache_lookup(struct svc_rqst *rqstp)
122	122	{
123	123	struct hlist_node *hn;
124	124	struct hlist_head *rh;
...	...	@@ -128,6 +128,7 @@
128	128	vers = rqstp->rq_vers,
129	129	proc = rqstp->rq_proc;
130	130	unsigned long age;
	131	+ int type = rqstp->rq_cachetype;
131	132	int rtn;
132	133
133	134	rqstp->rq_cacherep = NULL;
...	...	@@ -570,8 +570,22 @@
570	570	rqstp->rq_vers, rqstp->rq_proc);
571	571	proc = rqstp->rq_procinfo;
572	572
	573	+ /*
	574	+ * Give the xdr decoder a chance to change this if it wants
	575	+ * (necessary in the NFSv4.0 compound case)
	576	+ */
	577	+ rqstp->rq_cachetype = proc->pc_cachetype;
	578	+ /* Decode arguments */
	579	+ xdr = proc->pc_decode;
	580	+ if (xdr && !xdr(rqstp, (__be32*)rqstp->rq_arg.head[0].iov_base,
	581	+ rqstp->rq_argp)) {
	582	+ dprintk("nfsd: failed to decode arguments!\n");
	583	+ *statp = rpc_garbage_args;
	584	+ return 1;
	585	+ }
	586	+
573	587	/* Check whether we have this call in the cache. */
574		- switch (nfsd_cache_lookup(rqstp, proc->pc_cachetype)) {
	588	+ switch (nfsd_cache_lookup(rqstp)) {
575	589	case RC_INTR:
576	590	case RC_DROPIT:
577	591	return 0;
...	...	@@ -579,16 +593,6 @@
579	593	return 1;
580	594	case RC_DOIT:;
581	595	/* do it */
582		- }
583		-
584		- /* Decode arguments */
585		- xdr = proc->pc_decode;
586		- if (xdr && !xdr(rqstp, (__be32*)rqstp->rq_arg.head[0].iov_base,
587		- rqstp->rq_argp)) {
588		- dprintk("nfsd: failed to decode arguments!\n");
589		- nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
590		- *statp = rpc_garbage_args;
591		- return 1;
592	596	}
593	597
594	598	/* need to grab the location to store the status, as
...	...	@@ -457,6 +457,8 @@
457	457	struct nfs4_replay * replay;
458	458	};
459	459
	460	+bool nfsd4_cache_this_op(struct nfsd4_op *);
	461	+
460	462	struct nfsd4_compoundargs {
461	463	/* scratch variables for XDR decode */
462	464	__be32 * p;
...	...	@@ -479,6 +481,7 @@
479	481	u32 opcnt;
480	482	struct nfsd4_op *ops;
481	483	struct nfsd4_op iops[8];
	484	+ int cachetype;
482	485	};
483	486
484	487	struct nfsd4_compoundres {
...	...	@@ -273,6 +273,7 @@
273	273	/* Catering to nfsd */
274	274	struct auth_domain * rq_client; /* RPC peer info */
275	275	struct auth_domain * rq_gssclient; /* "gss/"-style peer info */
	276	+ int rq_cachetype;
276	277	struct svc_cacherep * rq_cacherep; /* cache info */
277	278	int rq_splice_ok; /* turned off in gss privacy
278	279	* to prevent encrypting page