Merge tag 'nfsd-4.14' of git://linux-nfs.org/~bfields/linux

Pull nfsd updates from Bruce Fields:
 "More RDMA work and some op-structure constification from Chuck Lever,
  and a small cleanup to our xdr encoding"

* tag 'nfsd-4.14' of git://linux-nfs.org/~bfields/linux:
  svcrdma: Estimate Send Queue depth properly
  rdma core: Add rdma_rw_mr_payload()
  svcrdma: Limit RQ depth
  svcrdma: Populate tail iovec when receiving
  nfsd: Incoming xdr_bufs may have content in tail buffer
  svcrdma: Clean up svc_rdma_build_read_chunk()
  sunrpc: Const-ify struct sv_serv_ops
  nfsd: Const-ify NFSv4 encoding and decoding ops arrays
  sunrpc: Const-ify instances of struct svc_xprt_ops
  nfsd4: individual encoders no longer see error cases
  nfsd4: skip encoder in trivial error cases
  nfsd4: define ->op_release for compound ops
  nfsd4: opdesc will be useful outside nfs4proc.c
  nfsd4: move some nfsd4 op definitions to xdr4.h

drivers/infiniband/core/rw.c

@@ -643,6 +643,30 @@ void rdma_rw_ctx_destroy_signature(struct rdma_rw_ctx *ctx, struct ib_qp *qp,
 }
 EXPORT_SYMBOL(rdma_rw_ctx_destroy_signature);
 
+/**
+ * rdma_rw_mr_factor - return number of MRs required for a payload
+ * @device:	device handling the connection
+ * @port_num:	port num to which the connection is bound
+ * @maxpages:	maximum payload pages per rdma_rw_ctx
+ *
+ * Returns the number of MRs the device requires to move @maxpayload
+ * bytes. The returned value is used during transport creation to
+ * compute max_rdma_ctxts and the size of the transport's Send and
+ * Send Completion Queues.
+ */
+unsigned int rdma_rw_mr_factor(struct ib_device *device, u8 port_num,
+			       unsigned int maxpages)
+{
+	unsigned int mr_pages;
+
+	if (rdma_rw_can_use_mr(device, port_num))
+		mr_pages = rdma_rw_fr_page_list_len(device);
+	else
+		mr_pages = device->attrs.max_sge_rd;
+	return DIV_ROUND_UP(maxpages, mr_pages);
+}
+EXPORT_SYMBOL(rdma_rw_mr_factor);
+
 void rdma_rw_init_qp(struct ib_device *dev, struct ib_qp_init_attr *attr)
 {
 	u32 factor;

fs/lockd/svc.c

@@ -396,7 +396,7 @@ static int lockd_start_svc(struct svc_serv *serv)
 	return error;
 }
 
-static struct svc_serv_ops lockd_sv_ops = {
+static const struct svc_serv_ops lockd_sv_ops = {
 	.svo_shutdown		= svc_rpcb_cleanup,
 	.svo_enqueue_xprt	= svc_xprt_do_enqueue,
 };

fs/nfs/callback.c

@@ -226,26 +226,26 @@ static int nfs_callback_up_net(int minorversion, struct svc_serv *serv,
 	return ret;
 }
 
-static struct svc_serv_ops nfs40_cb_sv_ops = {
+static const struct svc_serv_ops nfs40_cb_sv_ops = {
 	.svo_function		= nfs4_callback_svc,
 	.svo_enqueue_xprt	= svc_xprt_do_enqueue,
 	.svo_setup		= svc_set_num_threads_sync,
 	.svo_module		= THIS_MODULE,
 };
 #if defined(CONFIG_NFS_V4_1)
-static struct svc_serv_ops nfs41_cb_sv_ops = {
+static const struct svc_serv_ops nfs41_cb_sv_ops = {
 	.svo_function		= nfs41_callback_svc,
 	.svo_enqueue_xprt	= svc_xprt_do_enqueue,
 	.svo_setup		= svc_set_num_threads_sync,
 	.svo_module		= THIS_MODULE,
 };
 
-static struct svc_serv_ops *nfs4_cb_sv_ops[] = {
+static const struct svc_serv_ops *nfs4_cb_sv_ops[] = {
 	[0] = &nfs40_cb_sv_ops,
 	[1] = &nfs41_cb_sv_ops,
 };
 #else
-static struct svc_serv_ops *nfs4_cb_sv_ops[] = {
+static const struct svc_serv_ops *nfs4_cb_sv_ops[] = {
 	[0] = &nfs40_cb_sv_ops,
 	[1] = NULL,
 };
@@ -254,8 +254,8 @@ static struct svc_serv_ops *nfs4_cb_sv_ops[] = {
 static struct svc_serv *nfs_callback_create_svc(int minorversion)
 {
 	struct nfs_callback_data *cb_info = &nfs_callback_info[minorversion];
+	const struct svc_serv_ops *sv_ops;
 	struct svc_serv *serv;
-	struct svc_serv_ops *sv_ops;
 
 	/*
 	 * Check whether we're already up and running.

fs/nfsd/nfs4proc.c

@@ -784,6 +784,14 @@ nfsd4_read(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
 	return status;
 }
 
+
+static void
+nfsd4_read_release(union nfsd4_op_u *u)
+{
+	if (u->read.rd_filp)
+		fput(u->read.rd_filp);
+}
+
 static __be32
 nfsd4_readdir(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
 	      union nfsd4_op_u *u)
@@ -912,6 +920,13 @@ nfsd4_secinfo_no_name(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstat
 	return nfs_ok;
 }
 
+static void
+nfsd4_secinfo_release(union nfsd4_op_u *u)
+{
+	if (u->secinfo.si_exp)
+		exp_put(u->secinfo.si_exp);
+}
+
 static __be32
 nfsd4_setattr(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
 	      union nfsd4_op_u *u)
@@ -1335,6 +1350,12 @@ nfsd4_getdeviceinfo(struct svc_rqst *rqstp,
 	return nfserr;
 }
 
+static void
+nfsd4_getdeviceinfo_release(union nfsd4_op_u *u)
+{
+	kfree(u->getdeviceinfo.gd_device);
+}
+
 static __be32
 nfsd4_layoutget(struct svc_rqst *rqstp,
 		struct nfsd4_compound_state *cstate, union nfsd4_op_u *u)
@@ -1415,6 +1436,12 @@ nfsd4_layoutget(struct svc_rqst *rqstp,
 	return nfserr;
 }
 
+static void
+nfsd4_layoutget_release(union nfsd4_op_u *u)
+{
+	kfree(u->layoutget.lg_content);
+}
+
 static __be32
 nfsd4_layoutcommit(struct svc_rqst *rqstp,
 		struct nfsd4_compound_state *cstate, union nfsd4_op_u *u)
@@ -1541,49 +1568,6 @@ static inline void nfsd4_increment_op_stats(u32 opnum)
 		nfsdstats.nfs4_opcount[opnum]++;
 }
 
-enum nfsd4_op_flags {
-	ALLOWED_WITHOUT_FH = 1 << 0,	/* No current filehandle required */
-	ALLOWED_ON_ABSENT_FS = 1 << 1,	/* ops processed on absent fs */
-	ALLOWED_AS_FIRST_OP = 1 << 2,	/* ops reqired first in compound */
-	/* For rfc 5661 section 2.6.3.1.1: */
-	OP_HANDLES_WRONGSEC = 1 << 3,
-	OP_IS_PUTFH_LIKE = 1 << 4,
-	/*
-	 * These are the ops whose result size we estimate before
-	 * encoding, to avoid performing an op then not being able to
-	 * respond or cache a response.  This includes writes and setattrs
-	 * as well as the operations usually called "nonidempotent":
-	 */
-	OP_MODIFIES_SOMETHING = 1 << 5,
-	/*
-	 * Cache compounds containing these ops in the xid-based drc:
-	 * 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).
-	 */
-	OP_CACHEME = 1 << 6,
-	/*
-	 * These are ops which clear current state id.
-	 */
-	OP_CLEAR_STATEID = 1 << 7,
-};
-
-struct nfsd4_operation {
-	__be32 (*op_func)(struct svc_rqst *, struct nfsd4_compound_state *,
-			union nfsd4_op_u *);
-	u32 op_flags;
-	char *op_name;
-	/* Try to get response size before operation */
-	u32 (*op_rsize_bop)(struct svc_rqst *, struct nfsd4_op *);
-	void (*op_get_currentstateid)(struct nfsd4_compound_state *,
-			union nfsd4_op_u *);
-	void (*op_set_currentstateid)(struct nfsd4_compound_state *,
-			union nfsd4_op_u *);
-};
-
 static const struct nfsd4_operation nfsd4_ops[];
 
 static const char *nfsd4_op_name(unsigned opnum);
@@ -1621,7 +1605,7 @@ static __be32 nfs41_check_op_ordering(struct nfsd4_compoundargs *args)
 	return nfs_ok;
 }
 
-static inline const struct nfsd4_operation *OPDESC(struct nfsd4_op *op)
+const struct nfsd4_operation *OPDESC(struct nfsd4_op *op)
 {
 	return &nfsd4_ops[op->opnum];
 }
@@ -1694,7 +1678,6 @@ nfsd4_proc_compound(struct svc_rqst *rqstp)
 	struct nfsd4_compoundargs *args = rqstp->rq_argp;
 	struct nfsd4_compoundres *resp = rqstp->rq_resp;
 	struct nfsd4_op	*op;
-	const struct nfsd4_operation *opdesc;
 	struct nfsd4_compound_state *cstate = &resp->cstate;
 	struct svc_fh *current_fh = &cstate->current_fh;
 	struct svc_fh *save_fh = &cstate->save_fh;
@@ -1747,15 +1730,13 @@ nfsd4_proc_compound(struct svc_rqst *rqstp)
 			goto encode_op;
 		}
 
-		opdesc = OPDESC(op);
-
 		if (!current_fh->fh_dentry) {
-			if (!(opdesc->op_flags & ALLOWED_WITHOUT_FH)) {
+			if (!(op->opdesc->op_flags & ALLOWED_WITHOUT_FH)) {
 				op->status = nfserr_nofilehandle;
 				goto encode_op;
 			}
 		} else if (current_fh->fh_export->ex_fslocs.migrated &&
-			  !(opdesc->op_flags & ALLOWED_ON_ABSENT_FS)) {
+			  !(op->opdesc->op_flags & ALLOWED_ON_ABSENT_FS)) {
 			op->status = nfserr_moved;
 			goto encode_op;
 		}
@@ -1763,12 +1744,12 @@ nfsd4_proc_compound(struct svc_rqst *rqstp)
 		fh_clear_wcc(current_fh);
 
 		/* If op is non-idempotent */
-		if (opdesc->op_flags & OP_MODIFIES_SOMETHING) {
+		if (op->opdesc->op_flags & OP_MODIFIES_SOMETHING) {
 			/*
 			 * Don't execute this op if we couldn't encode a
 			 * succesful reply:
 			 */
-			u32 plen = opdesc->op_rsize_bop(rqstp, op);
+			u32 plen = op->opdesc->op_rsize_bop(rqstp, op);
 			/*
 			 * Plus if there's another operation, make sure
 			 * we'll have space to at least encode an error:
@@ -1781,9 +1762,9 @@ nfsd4_proc_compound(struct svc_rqst *rqstp)
 		if (op->status)
 			goto encode_op;
 
-		if (opdesc->op_get_currentstateid)
-			opdesc->op_get_currentstateid(cstate, &op->u);
-		op->status = opdesc->op_func(rqstp, cstate, &op->u);
+		if (op->opdesc->op_get_currentstateid)
+			op->opdesc->op_get_currentstateid(cstate, &op->u);
+		op->status = op->opdesc->op_func(rqstp, cstate, &op->u);
 
 		/* Only from SEQUENCE */
 		if (cstate->status == nfserr_replay_cache) {
@@ -1792,10 +1773,10 @@ nfsd4_proc_compound(struct svc_rqst *rqstp)
 			goto out;
 		}
 		if (!op->status) {
-			if (opdesc->op_set_currentstateid)
-				opdesc->op_set_currentstateid(cstate, &op->u);
+			if (op->opdesc->op_set_currentstateid)
+				op->opdesc->op_set_currentstateid(cstate, &op->u);
 
-			if (opdesc->op_flags & OP_CLEAR_STATEID)
+			if (op->opdesc->op_flags & OP_CLEAR_STATEID)
 				clear_current_stateid(cstate);
 
 			if (need_wrongsec_check(rqstp))
@@ -2160,13 +2141,15 @@ static const struct nfsd4_operation nfsd4_ops[] = {
 	},
 	[OP_LOCK] = {
 		.op_func = nfsd4_lock,
-		.op_flags = OP_MODIFIES_SOMETHING,
+		.op_flags = OP_MODIFIES_SOMETHING |
+				OP_NONTRIVIAL_ERROR_ENCODE,
 		.op_name = "OP_LOCK",
 		.op_rsize_bop = nfsd4_lock_rsize,
 		.op_set_currentstateid = nfsd4_set_lockstateid,
 	},
 	[OP_LOCKT] = {
 		.op_func = nfsd4_lockt,
+		.op_flags = OP_NONTRIVIAL_ERROR_ENCODE,
 		.op_name = "OP_LOCKT",
 		.op_rsize_bop = nfsd4_lock_rsize,
 	},
@@ -2238,6 +2221,7 @@ static const struct nfsd4_operation nfsd4_ops[] = {
 	},
 	[OP_READ] = {
 		.op_func = nfsd4_read,
+		.op_release = nfsd4_read_release,
 		.op_name = "OP_READ",
 		.op_rsize_bop = nfsd4_read_rsize,
 		.op_get_currentstateid = nfsd4_get_readstateid,
@@ -2287,6 +2271,7 @@ static const struct nfsd4_operation nfsd4_ops[] = {
 	},
 	[OP_SECINFO] = {
 		.op_func = nfsd4_secinfo,
+		.op_release = nfsd4_secinfo_release,
 		.op_flags = OP_HANDLES_WRONGSEC,
 		.op_name = "OP_SECINFO",
 		.op_rsize_bop = nfsd4_secinfo_rsize,
@@ -2294,14 +2279,16 @@ static const struct nfsd4_operation nfsd4_ops[] = {
 	[OP_SETATTR] = {
 		.op_func = nfsd4_setattr,
 		.op_name = "OP_SETATTR",
-		.op_flags = OP_MODIFIES_SOMETHING | OP_CACHEME,
+		.op_flags = OP_MODIFIES_SOMETHING | OP_CACHEME
+				| OP_NONTRIVIAL_ERROR_ENCODE,
 		.op_rsize_bop = nfsd4_setattr_rsize,
 		.op_get_currentstateid = nfsd4_get_setattrstateid,
 	},
 	[OP_SETCLIENTID] = {
 		.op_func = nfsd4_setclientid,
 		.op_flags = ALLOWED_WITHOUT_FH | ALLOWED_ON_ABSENT_FS
-				| OP_MODIFIES_SOMETHING | OP_CACHEME,
+				| OP_MODIFIES_SOMETHING | OP_CACHEME
+				| OP_NONTRIVIAL_ERROR_ENCODE,
 		.op_name = "OP_SETCLIENTID",
 		.op_rsize_bop = nfsd4_setclientid_rsize,
 	},
@@ -2388,6 +2375,7 @@ static const struct nfsd4_operation nfsd4_ops[] = {
 	},
 	[OP_SECINFO_NO_NAME] = {
 		.op_func = nfsd4_secinfo_no_name,
+		.op_release = nfsd4_secinfo_release,
 		.op_flags = OP_HANDLES_WRONGSEC,
 		.op_name = "OP_SECINFO_NO_NAME",
 		.op_rsize_bop = nfsd4_secinfo_rsize,
@@ -2408,12 +2396,14 @@ static const struct nfsd4_operation nfsd4_ops[] = {
 #ifdef CONFIG_NFSD_PNFS
 	[OP_GETDEVICEINFO] = {
 		.op_func = nfsd4_getdeviceinfo,
+		.op_release = nfsd4_getdeviceinfo_release,
 		.op_flags = ALLOWED_WITHOUT_FH,
 		.op_name = "OP_GETDEVICEINFO",
 		.op_rsize_bop = nfsd4_getdeviceinfo_rsize,
 	},
 	[OP_LAYOUTGET] = {
 		.op_func = nfsd4_layoutget,
+		.op_release = nfsd4_layoutget_release,
 		.op_flags = OP_MODIFIES_SOMETHING,
 		.op_name = "OP_LAYOUTGET",
 		.op_rsize_bop = nfsd4_layoutget_rsize,

fs/nfsd/nfssvc.c

@@ -475,7 +475,7 @@ static int nfsd_get_default_max_blksize(void)
 	return ret;
 }
 
-static struct svc_serv_ops nfsd_thread_sv_ops = {
+static const struct svc_serv_ops nfsd_thread_sv_ops = {
 	.svo_shutdown		= nfsd_last_thread,
 	.svo_function		= nfsd,
 	.svo_enqueue_xprt	= svc_xprt_do_enqueue,

fs/nfsd/xdr4.h

@@ -538,6 +538,7 @@ struct nfsd4_seek {
 
 struct nfsd4_op {
 	int					opnum;
+	const struct nfsd4_operation *		opdesc;
 	__be32					status;
 	union nfsd4_op_u {
 		struct nfsd4_access		access;
@@ -614,6 +615,7 @@ struct nfsd4_compoundargs {
 	__be32 *			end;
 	struct page **			pagelist;
 	int				pagelen;
+	bool				tail;
 	__be32				tmp[8];
 	__be32 *			tmpp;
 	struct svcxdr_tmpbuf		*to_free;
@@ -661,6 +663,7 @@ static inline bool nfsd4_last_compound_op(struct svc_rqst *rqstp)
 	return argp->opcnt == resp->opcnt;
 }
 
+const struct nfsd4_operation *OPDESC(struct nfsd4_op *op);
 int nfsd4_max_reply(struct svc_rqst *rqstp, struct nfsd4_op *op);
 void warn_on_nonidempotent_op(struct nfsd4_op *op);
 
@@ -748,6 +751,53 @@ extern __be32 nfsd4_free_stateid(struct svc_rqst *rqstp,
 		struct nfsd4_compound_state *, union nfsd4_op_u *);
 extern void nfsd4_bump_seqid(struct nfsd4_compound_state *, __be32 nfserr);
 
+enum nfsd4_op_flags {
+	ALLOWED_WITHOUT_FH = 1 << 0,    /* No current filehandle required */
+	ALLOWED_ON_ABSENT_FS = 1 << 1,  /* ops processed on absent fs */
+	ALLOWED_AS_FIRST_OP = 1 << 2,   /* ops reqired first in compound */
+	/* For rfc 5661 section 2.6.3.1.1: */
+	OP_HANDLES_WRONGSEC = 1 << 3,
+	OP_IS_PUTFH_LIKE = 1 << 4,
+	/*
+	 * These are the ops whose result size we estimate before
+	 * encoding, to avoid performing an op then not being able to
+	 * respond or cache a response.  This includes writes and setattrs
+	 * as well as the operations usually called "nonidempotent":
+	 */
+	OP_MODIFIES_SOMETHING = 1 << 5,
+	/*
+	 * Cache compounds containing these ops in the xid-based drc:
+	 * 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).
+	 */
+	OP_CACHEME = 1 << 6,
+	/*
+	 * These are ops which clear current state id.
+	 */
+	OP_CLEAR_STATEID = 1 << 7,
+	/* Most ops return only an error on failure; some may do more: */
+	OP_NONTRIVIAL_ERROR_ENCODE = 1 << 8,
+};
+
+struct nfsd4_operation {
+	__be32 (*op_func)(struct svc_rqst *, struct nfsd4_compound_state *,
+			union nfsd4_op_u *);
+	void (*op_release)(union nfsd4_op_u *);
+	u32 op_flags;
+	char *op_name;
+	/* Try to get response size before operation */
+	u32 (*op_rsize_bop)(struct svc_rqst *, struct nfsd4_op *);
+	void (*op_get_currentstateid)(struct nfsd4_compound_state *,
+			union nfsd4_op_u *);
+	void (*op_set_currentstateid)(struct nfsd4_compound_state *,
+			union nfsd4_op_u *);
+};
+
+
 #endif
 
 /*

include/linux/sunrpc/svc.h

@@ -99,7 +99,7 @@ struct svc_serv {
 
 	unsigned int		sv_nrpools;	/* number of thread pools */
 	struct svc_pool *	sv_pools;	/* array of thread pools */
-	struct svc_serv_ops	*sv_ops;	/* server operations */
+	const struct svc_serv_ops *sv_ops;	/* server operations */
 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
 	struct list_head	sv_cb_list;	/* queue for callback requests
 						 * that arrive over the same
@@ -465,7 +465,7 @@ int svc_rpcb_setup(struct svc_serv *serv, struct net *net);
 void svc_rpcb_cleanup(struct svc_serv *serv, struct net *net);
 int svc_bind(struct svc_serv *serv, struct net *net);
 struct svc_serv *svc_create(struct svc_program *, unsigned int,
-			    struct svc_serv_ops *);
+			    const struct svc_serv_ops *);
 struct svc_rqst *svc_rqst_alloc(struct svc_serv *serv,
 					struct svc_pool *pool, int node);
 struct svc_rqst *svc_prepare_thread(struct svc_serv *serv,
@@ -475,7 +475,7 @@ void		   svc_exit_thread(struct svc_rqst *);
 unsigned int	   svc_pool_map_get(void);
 void		   svc_pool_map_put(void);
 struct svc_serv *  svc_create_pooled(struct svc_program *, unsigned int,
-			struct svc_serv_ops *);
+			const struct svc_serv_ops *);
 int		   svc_set_num_threads(struct svc_serv *, struct svc_pool *, int);
 int		   svc_set_num_threads_sync(struct svc_serv *, struct svc_pool *, int);
 int		   svc_pool_stats_open(struct svc_serv *serv, struct file *file);

include/linux/sunrpc/svc_xprt.h

@@ -31,7 +31,7 @@ struct svc_xprt_ops {
 struct svc_xprt_class {
 	const char		*xcl_name;
 	struct module		*xcl_owner;
-	struct svc_xprt_ops	*xcl_ops;
+	const struct svc_xprt_ops *xcl_ops;
 	struct list_head	xcl_list;
 	u32			xcl_max_payload;
 	int			xcl_ident;
@@ -49,7 +49,7 @@ struct svc_xpt_user {
 
 struct svc_xprt {
 	struct svc_xprt_class	*xpt_class;
-	struct svc_xprt_ops	*xpt_ops;
+	const struct svc_xprt_ops *xpt_ops;
 	struct kref		xpt_ref;
 	struct list_head	xpt_list;
 	struct list_head	xpt_ready;

include/rdma/rw.h

@@ -81,6 +81,8 @@ struct ib_send_wr *rdma_rw_ctx_wrs(struct rdma_rw_ctx *ctx, struct ib_qp *qp,
 int rdma_rw_ctx_post(struct rdma_rw_ctx *ctx, struct ib_qp *qp, u8 port_num,
 		struct ib_cqe *cqe, struct ib_send_wr *chain_wr);
 
+unsigned int rdma_rw_mr_factor(struct ib_device *device, u8 port_num,
+		unsigned int maxpages);
 void rdma_rw_init_qp(struct ib_device *dev, struct ib_qp_init_attr *attr);
 int rdma_rw_init_mrs(struct ib_qp *qp, struct ib_qp_init_attr *attr);
 void rdma_rw_cleanup_mrs(struct ib_qp *qp);

net/sunrpc/svc.c

@@ -421,7 +421,7 @@ __svc_init_bc(struct svc_serv *serv)
  */
 static struct svc_serv *
 __svc_create(struct svc_program *prog, unsigned int bufsize, int npools,
-	     struct svc_serv_ops *ops)
+	     const struct svc_serv_ops *ops)
 {
 	struct svc_serv	*serv;
 	unsigned int vers;
@@ -486,7 +486,7 @@ __svc_create(struct svc_program *prog, unsigned int bufsize, int npools,
 
 struct svc_serv *
 svc_create(struct svc_program *prog, unsigned int bufsize,
-	   struct svc_serv_ops *ops)
+	   const struct svc_serv_ops *ops)
 {
 	return __svc_create(prog, bufsize, /*npools*/1, ops);
 }
@@ -494,7 +494,7 @@ EXPORT_SYMBOL_GPL(svc_create);
 
 struct svc_serv *
 svc_create_pooled(struct svc_program *prog, unsigned int bufsize,
-		  struct svc_serv_ops *ops)
+		  const struct svc_serv_ops *ops)
 {
 	struct svc_serv *serv;
 	unsigned int npools = svc_pool_map_get();

net/sunrpc/svcsock.c

@@ -693,7 +693,7 @@ static struct svc_xprt *svc_udp_create(struct svc_serv *serv,
 	return svc_create_socket(serv, IPPROTO_UDP, net, sa, salen, flags);
 }
 
-static struct svc_xprt_ops svc_udp_ops = {
+static const struct svc_xprt_ops svc_udp_ops = {
 	.xpo_create = svc_udp_create,
 	.xpo_recvfrom = svc_udp_recvfrom,
 	.xpo_sendto = svc_udp_sendto,
@@ -1241,7 +1241,7 @@ static void svc_bc_tcp_sock_detach(struct svc_xprt *xprt)
 {
 }
 
-static struct svc_xprt_ops svc_tcp_bc_ops = {
+static const struct svc_xprt_ops svc_tcp_bc_ops = {
 	.xpo_create = svc_bc_tcp_create,
 	.xpo_detach = svc_bc_tcp_sock_detach,
 	.xpo_free = svc_bc_sock_free,
@@ -1275,7 +1275,7 @@ static void svc_cleanup_bc_xprt_sock(void)
 }
 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
 
-static struct svc_xprt_ops svc_tcp_ops = {
+static const struct svc_xprt_ops svc_tcp_ops = {
 	.xpo_create = svc_tcp_create,
 	.xpo_recvfrom = svc_tcp_recvfrom,
 	.xpo_sendto = svc_tcp_sendto,

net/sunrpc/xprtrdma/svc_rdma_rw.c

@@ -660,19 +660,21 @@ static int svc_rdma_build_read_segment(struct svc_rdma_read_info *info,
 	return -EIO;
 }
 
+/* Walk the segments in the Read chunk starting at @p and construct
+ * RDMA Read operations to pull the chunk to the server.
+ */
 static int svc_rdma_build_read_chunk(struct svc_rqst *rqstp,
 				     struct svc_rdma_read_info *info,
 				     __be32 *p)
 {
 	int ret;
 
+	ret = -EINVAL;
 	info->ri_chunklen = 0;
-	while (*p++ != xdr_zero) {
+	while (*p++ != xdr_zero && be32_to_cpup(p++) == info->ri_position) {
 		u32 rs_handle, rs_length;
 		u64 rs_offset;
 
-		if (be32_to_cpup(p++) != info->ri_position)
-			break;
 		rs_handle = be32_to_cpup(p++);
 		rs_length = be32_to_cpup(p++);
 		p = xdr_decode_hyper(p, &rs_offset);
@@ -689,78 +691,6 @@ static int svc_rdma_build_read_chunk(struct svc_rqst *rqstp,
 	return ret;
 }
 
-/* If there is inline content following the Read chunk, append it to
- * the page list immediately following the data payload. This has to
- * be done after the reader function has determined how many pages
- * were consumed for RDMA Read.
- *
- * On entry, ri_pageno and ri_pageoff point directly to the end of the
- * page list. On exit, both have been updated to the new "next byte".
- *
- * Assumptions:
- *	- Inline content fits entirely in rq_pages[0]
- *	- Trailing content is only a handful of bytes
- */
-static int svc_rdma_copy_tail(struct svc_rqst *rqstp,
-			      struct svc_rdma_read_info *info)
-{
-	struct svc_rdma_op_ctxt *head = info->ri_readctxt;
-	unsigned int tail_length, remaining;
-	u8 *srcp, *destp;
-
-	/* Assert that all inline content fits in page 0. This is an
-	 * implementation limit, not a protocol limit.
-	 */
-	if (head->arg.head[0].iov_len > PAGE_SIZE) {
-		pr_warn_once("svcrdma: too much trailing inline content\n");
-		return -EINVAL;
-	}
-
-	srcp = head->arg.head[0].iov_base;
-	srcp += info->ri_position;
-	tail_length = head->arg.head[0].iov_len - info->ri_position;
-	remaining = tail_length;
-
-	/* If there is room on the last page in the page list, try to
-	 * fit the trailing content there.
-	 */
-	if (info->ri_pageoff > 0) {
-		unsigned int len;
-
-		len = min_t(unsigned int, remaining,
-			    PAGE_SIZE - info->ri_pageoff);
-		destp = page_address(rqstp->rq_pages[info->ri_pageno]);
-		destp += info->ri_pageoff;
-
-		memcpy(destp, srcp, len);
-		srcp += len;
-		destp += len;
-		info->ri_pageoff += len;
-		remaining -= len;
-
-		if (info->ri_pageoff == PAGE_SIZE) {
-			info->ri_pageno++;
-			info->ri_pageoff = 0;
-		}
-	}
-
-	/* Otherwise, a fresh page is needed. */
-	if (remaining) {
-		head->arg.pages[info->ri_pageno] =
-				rqstp->rq_pages[info->ri_pageno];
-		head->count++;
-
-		destp = page_address(rqstp->rq_pages[info->ri_pageno]);
-		memcpy(destp, srcp, remaining);
-		info->ri_pageoff += remaining;
-	}
-
-	head->arg.page_len += tail_length;
-	head->arg.len += tail_length;
-	head->arg.buflen += tail_length;
-	return 0;
-}
-
 /* Construct RDMA Reads to pull over a normal Read chunk. The chunk
  * data lands in the page list of head->arg.pages.
  *
@@ -785,34 +715,28 @@ static int svc_rdma_build_normal_read_chunk(struct svc_rqst *rqstp,
 	if (ret < 0)
 		goto out;
 
-	/* Read chunk may need XDR round-up (see RFC 5666, s. 3.7).
+	/* Split the Receive buffer between the head and tail
+	 * buffers at Read chunk's position. XDR roundup of the
+	 * chunk is not included in either the pagelist or in
+	 * the tail.
 	 */
-	if (info->ri_chunklen & 3) {
-		u32 padlen = 4 - (info->ri_chunklen & 3);
-
-		info->ri_chunklen += padlen;
+	head->arg.tail[0].iov_base =
+		head->arg.head[0].iov_base + info->ri_position;
+	head->arg.tail[0].iov_len =
+		head->arg.head[0].iov_len - info->ri_position;
+	head->arg.head[0].iov_len = info->ri_position;
 
-		/* NB: data payload always starts on XDR alignment,
-		 * thus the pad can never contain a page boundary.
-		 */
-		info->ri_pageoff += padlen;
-		if (info->ri_pageoff == PAGE_SIZE) {
-			info->ri_pageno++;
-			info->ri_pageoff = 0;
-		}
-	}
+	/* Read chunk may need XDR roundup (see RFC 5666, s. 3.7).
+	 *
+	 * NFSv2/3 write decoders need the length of the tail to
+	 * contain the size of the roundup padding.
+	 */
+	head->arg.tail[0].iov_len += 4 - (info->ri_chunklen & 3);
 
 	head->arg.page_len = info->ri_chunklen;
 	head->arg.len += info->ri_chunklen;
 	head->arg.buflen += info->ri_chunklen;
 
-	if (info->ri_position < head->arg.head[0].iov_len) {
-		ret = svc_rdma_copy_tail(rqstp, info);
-		if (ret < 0)
-			goto out;
-	}
-	head->arg.head[0].iov_len = info->ri_position;
-
 out:
 	return ret;
 }

net/sunrpc/xprtrdma/svc_rdma_transport.c

@@ -51,6 +51,7 @@
 #include <linux/workqueue.h>
 #include <rdma/ib_verbs.h>
 #include <rdma/rdma_cm.h>
+#include <rdma/rw.h>
 #include <linux/sunrpc/svc_rdma.h>
 #include <linux/export.h>
 #include "xprt_rdma.h"
@@ -70,7 +71,7 @@ static int svc_rdma_has_wspace(struct svc_xprt *xprt);
 static int svc_rdma_secure_port(struct svc_rqst *);
 static void svc_rdma_kill_temp_xprt(struct svc_xprt *);
 
-static struct svc_xprt_ops svc_rdma_ops = {
+static const struct svc_xprt_ops svc_rdma_ops = {
 	.xpo_create = svc_rdma_create,
 	.xpo_recvfrom = svc_rdma_recvfrom,
 	.xpo_sendto = svc_rdma_sendto,
@@ -98,7 +99,7 @@ static struct svc_xprt *svc_rdma_bc_create(struct svc_serv *, struct net *,
 static void svc_rdma_bc_detach(struct svc_xprt *);
 static void svc_rdma_bc_free(struct svc_xprt *);
 
-static struct svc_xprt_ops svc_rdma_bc_ops = {
+static const struct svc_xprt_ops svc_rdma_bc_ops = {
 	.xpo_create = svc_rdma_bc_create,
 	.xpo_detach = svc_rdma_bc_detach,
 	.xpo_free = svc_rdma_bc_free,
@@ -167,8 +168,8 @@ static bool svc_rdma_prealloc_ctxts(struct svcxprt_rdma *xprt)
 {
 	unsigned int i;
 
-	/* Each RPC/RDMA credit can consume a number of send
-	 * and receive WQEs. One ctxt is allocated for each.
+	/* Each RPC/RDMA credit can consume one Receive and
+	 * one Send WQE at the same time.
 	 */
 	i = xprt->sc_sq_depth + xprt->sc_rq_depth;
 
@@ -713,7 +714,7 @@ static struct svc_xprt *svc_rdma_accept(struct svc_xprt *xprt)
 	struct ib_qp_init_attr qp_attr;
 	struct ib_device *dev;
 	struct sockaddr *sap;
-	unsigned int i;
+	unsigned int i, ctxts;
 	int ret = 0;
 
 	listen_rdma = container_of(xprt, struct svcxprt_rdma, sc_xprt);
@@ -742,14 +743,26 @@ static struct svc_xprt *svc_rdma_accept(struct svc_xprt *xprt)
 	newxprt->sc_max_sge = min((size_t)dev->attrs.max_sge,
 				  (size_t)RPCSVC_MAXPAGES);
 	newxprt->sc_max_req_size = svcrdma_max_req_size;
-	newxprt->sc_max_requests = min_t(u32, dev->attrs.max_qp_wr,
-					 svcrdma_max_requests);
-	newxprt->sc_fc_credits = cpu_to_be32(newxprt->sc_max_requests);
-	newxprt->sc_max_bc_requests = min_t(u32, dev->attrs.max_qp_wr,
-					    svcrdma_max_bc_requests);
+	newxprt->sc_max_requests = svcrdma_max_requests;
+	newxprt->sc_max_bc_requests = svcrdma_max_bc_requests;
 	newxprt->sc_rq_depth = newxprt->sc_max_requests +
 			       newxprt->sc_max_bc_requests;
-	newxprt->sc_sq_depth = newxprt->sc_rq_depth;
+	if (newxprt->sc_rq_depth > dev->attrs.max_qp_wr) {
+		pr_warn("svcrdma: reducing receive depth to %d\n",
+			dev->attrs.max_qp_wr);
+		newxprt->sc_rq_depth = dev->attrs.max_qp_wr;
+		newxprt->sc_max_requests = newxprt->sc_rq_depth - 2;
+		newxprt->sc_max_bc_requests = 2;
+	}
+	newxprt->sc_fc_credits = cpu_to_be32(newxprt->sc_max_requests);
+	ctxts = rdma_rw_mr_factor(dev, newxprt->sc_port_num, RPCSVC_MAXPAGES);
+	ctxts *= newxprt->sc_max_requests;
+	newxprt->sc_sq_depth = newxprt->sc_rq_depth + ctxts;
+	if (newxprt->sc_sq_depth > dev->attrs.max_qp_wr) {
+		pr_warn("svcrdma: reducing send depth to %d\n",
+			dev->attrs.max_qp_wr);
+		newxprt->sc_sq_depth = dev->attrs.max_qp_wr;
+	}
 	atomic_set(&newxprt->sc_sq_avail, newxprt->sc_sq_depth);
 
 	if (!svc_rdma_prealloc_ctxts(newxprt))
@@ -784,8 +797,8 @@ static struct svc_xprt *svc_rdma_accept(struct svc_xprt *xprt)
 	qp_attr.event_handler = qp_event_handler;
 	qp_attr.qp_context = &newxprt->sc_xprt;
 	qp_attr.port_num = newxprt->sc_port_num;
-	qp_attr.cap.max_rdma_ctxs = newxprt->sc_max_requests;
-	qp_attr.cap.max_send_wr = newxprt->sc_sq_depth;
+	qp_attr.cap.max_rdma_ctxs = ctxts;
+	qp_attr.cap.max_send_wr = newxprt->sc_sq_depth - ctxts;
 	qp_attr.cap.max_recv_wr = newxprt->sc_rq_depth;
 	qp_attr.cap.max_send_sge = newxprt->sc_max_sge;
 	qp_attr.cap.max_recv_sge = newxprt->sc_max_sge;
@@ -853,6 +866,7 @@ static struct svc_xprt *svc_rdma_accept(struct svc_xprt *xprt)
 	dprintk("    remote address  : %pIS:%u\n", sap, rpc_get_port(sap));
 	dprintk("    max_sge         : %d\n", newxprt->sc_max_sge);
 	dprintk("    sq_depth        : %d\n", newxprt->sc_sq_depth);
+	dprintk("    rdma_rw_ctxs    : %d\n", ctxts);
 	dprintk("    max_requests    : %d\n", newxprt->sc_max_requests);
 	dprintk("    ord             : %d\n", newxprt->sc_ord);