[linux-2.6-block.git] / net / sunrpc / xprtrdma / backchannel.c

// SPDX-License-Identifier: GPL-2.0
/*
 * Copyright (c) 2015 Oracle.  All rights reserved.
 *
 * Support for backward direction RPCs on RPC/RDMA.
 */

#include <linux/sunrpc/xprt.h>
#include <linux/sunrpc/svc.h>
#include <linux/sunrpc/svc_xprt.h>
#include <linux/sunrpc/svc_rdma.h>

#include "xprt_rdma.h"
#include <trace/events/rpcrdma.h>

#if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
# define RPCDBG_FACILITY	RPCDBG_TRANS
#endif

#undef RPCRDMA_BACKCHANNEL_DEBUG

/**
 * xprt_rdma_bc_setup - Pre-allocate resources for handling backchannel requests
 * @xprt: transport associated with these backchannel resources
 * @reqs: number of concurrent incoming requests to expect
 *
 * Returns 0 on success; otherwise a negative errno
 */
int xprt_rdma_bc_setup(struct rpc_xprt *xprt, unsigned int reqs)
{
	struct rpcrdma_xprt *r_xprt = rpcx_to_rdmax(xprt);

	r_xprt->rx_buf.rb_bc_srv_max_requests = RPCRDMA_BACKWARD_WRS >> 1;
	trace_xprtrdma_cb_setup(r_xprt, reqs);
	return 0;
}

/**
 * xprt_rdma_bc_maxpayload - Return maximum backchannel message size
 * @xprt: transport
 *
 * Returns maximum size, in bytes, of a backchannel message
 */
size_t xprt_rdma_bc_maxpayload(struct rpc_xprt *xprt)
{
	struct rpcrdma_xprt *r_xprt = rpcx_to_rdmax(xprt);
	struct rpcrdma_ep *ep = &r_xprt->rx_ep;
	size_t maxmsg;

	maxmsg = min_t(unsigned int, ep->rep_inline_send, ep->rep_inline_recv);
	maxmsg = min_t(unsigned int, maxmsg, PAGE_SIZE);
	return maxmsg - RPCRDMA_HDRLEN_MIN;
}

static int rpcrdma_bc_marshal_reply(struct rpc_rqst *rqst)
{
	struct rpcrdma_xprt *r_xprt = rpcx_to_rdmax(rqst->rq_xprt);
	struct rpcrdma_req *req = rpcr_to_rdmar(rqst);
	__be32 *p;

	rpcrdma_set_xdrlen(&req->rl_hdrbuf, 0);
	xdr_init_encode(&req->rl_stream, &req->rl_hdrbuf,
			rdmab_data(req->rl_rdmabuf), rqst);

	p = xdr_reserve_space(&req->rl_stream, 28);
	if (unlikely(!p))
		return -EIO;
	*p++ = rqst->rq_xid;
	*p++ = rpcrdma_version;
	*p++ = cpu_to_be32(r_xprt->rx_buf.rb_bc_srv_max_requests);
	*p++ = rdma_msg;
	*p++ = xdr_zero;
	*p++ = xdr_zero;
	*p = xdr_zero;

	if (rpcrdma_prepare_send_sges(r_xprt, req, RPCRDMA_HDRLEN_MIN,
				      &rqst->rq_snd_buf, rpcrdma_noch))
		return -EIO;

	trace_xprtrdma_cb_reply(rqst);
	return 0;
}

/**
 * xprt_rdma_bc_send_reply - marshal and send a backchannel reply
 * @rqst: RPC rqst with a backchannel RPC reply in rq_snd_buf
 *
 * Caller holds the transport's write lock.
 *
 * Returns:
 *	%0 if the RPC message has been sent
 *	%-ENOTCONN if the caller should reconnect and call again
 *	%-EIO if a permanent error occurred and the request was not
 *		sent. Do not try to send this message again.
 */
int xprt_rdma_bc_send_reply(struct rpc_rqst *rqst)
{
	struct rpc_xprt *xprt = rqst->rq_xprt;
	struct rpcrdma_xprt *r_xprt = rpcx_to_rdmax(xprt);
	struct rpcrdma_req *req = rpcr_to_rdmar(rqst);
	int rc;

	if (!xprt_connected(xprt))
		return -ENOTCONN;

	if (!xprt_request_get_cong(xprt, rqst))
		return -EBADSLT;

	rc = rpcrdma_bc_marshal_reply(rqst);
	if (rc < 0)
		goto failed_marshal;

	if (rpcrdma_ep_post(&r_xprt->rx_ia, &r_xprt->rx_ep, req))
		goto drop_connection;
	return 0;

failed_marshal:
	if (rc != -ENOTCONN)
		return rc;
drop_connection:
	xprt_rdma_close(xprt);
	return -ENOTCONN;
}

/**
 * xprt_rdma_bc_destroy - Release resources for handling backchannel requests
 * @xprt: transport associated with these backchannel resources
 * @reqs: number of incoming requests to destroy; ignored
 */
void xprt_rdma_bc_destroy(struct rpc_xprt *xprt, unsigned int reqs)
{
	struct rpc_rqst *rqst, *tmp;

	spin_lock(&xprt->bc_pa_lock);
	list_for_each_entry_safe(rqst, tmp, &xprt->bc_pa_list, rq_bc_pa_list) {
		list_del(&rqst->rq_bc_pa_list);
		spin_unlock(&xprt->bc_pa_lock);

		rpcrdma_req_destroy(rpcr_to_rdmar(rqst));

		spin_lock(&xprt->bc_pa_lock);
	}
	spin_unlock(&xprt->bc_pa_lock);
}

/**
 * xprt_rdma_bc_free_rqst - Release a backchannel rqst
 * @rqst: request to release
 */
void xprt_rdma_bc_free_rqst(struct rpc_rqst *rqst)
{
	struct rpcrdma_req *req = rpcr_to_rdmar(rqst);
	struct rpc_xprt *xprt = rqst->rq_xprt;

	rpcrdma_recv_buffer_put(req->rl_reply);
	req->rl_reply = NULL;

	spin_lock(&xprt->bc_pa_lock);
	list_add_tail(&rqst->rq_bc_pa_list, &xprt->bc_pa_list);
	spin_unlock(&xprt->bc_pa_lock);
}

static struct rpc_rqst *rpcrdma_bc_rqst_get(struct rpcrdma_xprt *r_xprt)
{
	struct rpc_xprt *xprt = &r_xprt->rx_xprt;
	struct rpcrdma_req *req;
	struct rpc_rqst *rqst;
	size_t size;

	spin_lock(&xprt->bc_pa_lock);
	rqst = list_first_entry_or_null(&xprt->bc_pa_list, struct rpc_rqst,
					rq_bc_pa_list);
	if (!rqst)
		goto create_req;
	list_del(&rqst->rq_bc_pa_list);
	spin_unlock(&xprt->bc_pa_lock);
	return rqst;

create_req:
	spin_unlock(&xprt->bc_pa_lock);

	/* Set a limit to prevent a remote from overrunning our resources.
	 */
	if (xprt->bc_alloc_count >= RPCRDMA_BACKWARD_WRS)
		return NULL;

	size = min_t(size_t, r_xprt->rx_ep.rep_inline_recv, PAGE_SIZE);
	req = rpcrdma_req_create(r_xprt, size, GFP_KERNEL);
	if (!req)
		return NULL;

	xprt->bc_alloc_count++;
	rqst = &req->rl_slot;
	rqst->rq_xprt = xprt;
	__set_bit(RPC_BC_PA_IN_USE, &rqst->rq_bc_pa_state);
	xdr_buf_init(&rqst->rq_snd_buf, rdmab_data(req->rl_sendbuf), size);
	return rqst;
}

/**
 * rpcrdma_bc_receive_call - Handle a backward direction call
 * @r_xprt: transport receiving the call
 * @rep: receive buffer containing the call
 *
 * Operational assumptions:
 *    o Backchannel credits are ignored, just as the NFS server
 *      forechannel currently does
 *    o The ULP manages a replay cache (eg, NFSv4.1 sessions).
 *      No replay detection is done at the transport level
 */
void rpcrdma_bc_receive_call(struct rpcrdma_xprt *r_xprt,
			     struct rpcrdma_rep *rep)
{
	struct rpc_xprt *xprt = &r_xprt->rx_xprt;
	struct svc_serv *bc_serv;
	struct rpcrdma_req *req;
	struct rpc_rqst *rqst;
	struct xdr_buf *buf;
	size_t size;
	__be32 *p;

	p = xdr_inline_decode(&rep->rr_stream, 0);
	size = xdr_stream_remaining(&rep->rr_stream);

#ifdef RPCRDMA_BACKCHANNEL_DEBUG
	pr_info("RPC:       %s: callback XID %08x, length=%u\n",
		__func__, be32_to_cpup(p), size);
	pr_info("RPC:       %s: %*ph\n", __func__, size, p);
#endif

	rqst = rpcrdma_bc_rqst_get(r_xprt);
	if (!rqst)
		goto out_overflow;

	rqst->rq_reply_bytes_recvd = 0;
	rqst->rq_xid = *p;

	rqst->rq_private_buf.len = size;

	buf = &rqst->rq_rcv_buf;
	memset(buf, 0, sizeof(*buf));
	buf->head[0].iov_base = p;
	buf->head[0].iov_len = size;
	buf->len = size;

	/* The receive buffer has to be hooked to the rpcrdma_req
	 * so that it is not released while the req is pointing
	 * to its buffer, and so that it can be reposted after
	 * the Upper Layer is done decoding it.
	 */
	req = rpcr_to_rdmar(rqst);
	req->rl_reply = rep;
	trace_xprtrdma_cb_call(rqst);

	/* Queue rqst for ULP's callback service */
	bc_serv = xprt->bc_serv;
	spin_lock(&bc_serv->sv_cb_lock);
	list_add(&rqst->rq_bc_list, &bc_serv->sv_cb_list);
	spin_unlock(&bc_serv->sv_cb_lock);

	wake_up(&bc_serv->sv_cb_waitq);

	r_xprt->rx_stats.bcall_count++;
	return;

out_overflow:
	pr_warn("RPC/RDMA backchannel overflow\n");
	xprt_force_disconnect(xprt);
	/* This receive buffer gets reposted automatically
	 * when the connection is re-established.
	 */
	return;
}
Commit	Line	Data
b2441318	1	// SPDX-License-Identifier: GPL-2.0
f531a5db CL	2	/*
	3	* Copyright (c) 2015 Oracle. All rights reserved.
	4	*
	5	* Support for backward direction RPCs on RPC/RDMA.
	6	*/
	7
63cae470 CL	8	#include <linux/sunrpc/xprt.h>
63cae470 CL	9	#include <linux/sunrpc/svc.h>
76566773	10	#include <linux/sunrpc/svc_xprt.h>
bd2abef3	11	#include <linux/sunrpc/svc_rdma.h>
f531a5db CL	12
f531a5db CL	13	#include "xprt_rdma.h"
b6e717cb	14	#include <trace/events/rpcrdma.h>
f531a5db CL	15
	16	#if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
	17	# define RPCDBG_FACILITY RPCDBG_TRANS
	18	#endif
	19
c8bbe0c7	20	#undef RPCRDMA_BACKCHANNEL_DEBUG
63cae470	21
f531a5db CL	22	/**
	23	* xprt_rdma_bc_setup - Pre-allocate resources for handling backchannel requests
	24	* @xprt: transport associated with these backchannel resources
	25	* @reqs: number of concurrent incoming requests to expect
	26	*
	27	* Returns 0 on success; otherwise a negative errno
	28	*/
	29	int xprt_rdma_bc_setup(struct rpc_xprt *xprt, unsigned int reqs)
	30	{
	31	struct rpcrdma_xprt *r_xprt = rpcx_to_rdmax(xprt);
f531a5db	32
3f9c7e76	33	r_xprt->rx_buf.rb_bc_srv_max_requests = RPCRDMA_BACKWARD_WRS >> 1;
fc1eb807	34	trace_xprtrdma_cb_setup(r_xprt, reqs);
f531a5db	35	return 0;
f531a5db CL	36	}
f531a5db CL	37
6b26cc8c CL	38	/**
	39	* xprt_rdma_bc_maxpayload - Return maximum backchannel message size
	40	* @xprt: transport
	41	*
	42	* Returns maximum size, in bytes, of a backchannel message
	43	*/
	44	size_t xprt_rdma_bc_maxpayload(struct rpc_xprt *xprt)
	45	{
	46	struct rpcrdma_xprt *r_xprt = rpcx_to_rdmax(xprt);
94087e97	47	struct rpcrdma_ep *ep = &r_xprt->rx_ep;
6b26cc8c CL	48	size_t maxmsg;
6b26cc8c CL	49
94087e97	50	maxmsg = min_t(unsigned int, ep->rep_inline_send, ep->rep_inline_recv);
62aee0e3	51	maxmsg = min_t(unsigned int, maxmsg, PAGE_SIZE);
6b26cc8c CL	52	return maxmsg - RPCRDMA_HDRLEN_MIN;
	53	}
	54
cf73daf5	55	static int rpcrdma_bc_marshal_reply(struct rpc_rqst *rqst)
83128a60	56	{
7ec910e7	57	struct rpcrdma_xprt *r_xprt = rpcx_to_rdmax(rqst->rq_xprt);
83128a60	58	struct rpcrdma_req *req = rpcr_to_rdmar(rqst);
7ec910e7 CL	59	__be32 *p;
	60
	61	rpcrdma_set_xdrlen(&req->rl_hdrbuf, 0);
	62	xdr_init_encode(&req->rl_stream, &req->rl_hdrbuf,
8cec3dba	63	rdmab_data(req->rl_rdmabuf), rqst);
7ec910e7 CL	64
	65	p = xdr_reserve_space(&req->rl_stream, 28);
	66	if (unlikely(!p))
	67	return -EIO;
	68	*p++ = rqst->rq_xid;
	69	*p++ = rpcrdma_version;
	70	*p++ = cpu_to_be32(r_xprt->rx_buf.rb_bc_srv_max_requests);
	71	*p++ = rdma_msg;
	72	*p++ = xdr_zero;
	73	*p++ = xdr_zero;
	74	*p = xdr_zero;
83128a60	75
857f9aca CL	76	if (rpcrdma_prepare_send_sges(r_xprt, req, RPCRDMA_HDRLEN_MIN,
857f9aca CL	77	&rqst->rq_snd_buf, rpcrdma_noch))
655fec69	78	return -EIO;
fc1eb807 CL	79
fc1eb807 CL	80	trace_xprtrdma_cb_reply(rqst);
83128a60 CL	81	return 0;
	82	}
	83
cf73daf5 CL	84	/**
	85	* xprt_rdma_bc_send_reply - marshal and send a backchannel reply
	86	* @rqst: RPC rqst with a backchannel RPC reply in rq_snd_buf
	87	*
	88	* Caller holds the transport's write lock.
	89	*
	90	* Returns:
	91	* %0 if the RPC message has been sent
	92	* %-ENOTCONN if the caller should reconnect and call again
	93	* %-EIO if a permanent error occurred and the request was not
	94	* sent. Do not try to send this message again.
	95	*/
	96	int xprt_rdma_bc_send_reply(struct rpc_rqst *rqst)
	97	{
0c0829bc CL	98	struct rpc_xprt *xprt = rqst->rq_xprt;
0c0829bc CL	99	struct rpcrdma_xprt *r_xprt = rpcx_to_rdmax(xprt);
cf73daf5 CL	100	struct rpcrdma_req *req = rpcr_to_rdmar(rqst);
	101	int rc;
	102
0c0829bc CL	103	if (!xprt_connected(xprt))
0c0829bc CL	104	return -ENOTCONN;
cf73daf5	105
0c0829bc	106	if (!xprt_request_get_cong(xprt, rqst))
75891f50 TM	107	return -EBADSLT;
75891f50 TM	108
cf73daf5 CL	109	rc = rpcrdma_bc_marshal_reply(rqst);
	110	if (rc < 0)
	111	goto failed_marshal;
	112
	113	if (rpcrdma_ep_post(&r_xprt->rx_ia, &r_xprt->rx_ep, req))
	114	goto drop_connection;
	115	return 0;
	116
	117	failed_marshal:
	118	if (rc != -ENOTCONN)
	119	return rc;
	120	drop_connection:
0c0829bc	121	xprt_rdma_close(xprt);
cf73daf5 CL	122	return -ENOTCONN;
	123	}
	124
f531a5db CL	125	/**
	126	* xprt_rdma_bc_destroy - Release resources for handling backchannel requests
	127	* @xprt: transport associated with these backchannel resources
	128	* @reqs: number of incoming requests to destroy; ignored
	129	*/
	130	void xprt_rdma_bc_destroy(struct rpc_xprt *xprt, unsigned int reqs)
	131	{
f531a5db CL	132	struct rpc_rqst rqst, tmp;
f531a5db CL	133
f7d46681	134	spin_lock(&xprt->bc_pa_lock);
f531a5db CL	135	list_for_each_entry_safe(rqst, tmp, &xprt->bc_pa_list, rq_bc_pa_list) {
f531a5db CL	136	list_del(&rqst->rq_bc_pa_list);
f7d46681	137	spin_unlock(&xprt->bc_pa_lock);
f531a5db	138
92f4433e	139	rpcrdma_req_destroy(rpcr_to_rdmar(rqst));
f531a5db	140
f7d46681	141	spin_lock(&xprt->bc_pa_lock);
f531a5db	142	}
f7d46681	143	spin_unlock(&xprt->bc_pa_lock);
f531a5db CL	144	}
	145
	146	/**
	147	* xprt_rdma_bc_free_rqst - Release a backchannel rqst
	148	* @rqst: request to release
	149	*/
	150	void xprt_rdma_bc_free_rqst(struct rpc_rqst *rqst)
	151	{
7c8d9e7c	152	struct rpcrdma_req *req = rpcr_to_rdmar(rqst);
f531a5db CL	153	struct rpc_xprt *xprt = rqst->rq_xprt;
f531a5db CL	154
7c8d9e7c CL	155	rpcrdma_recv_buffer_put(req->rl_reply);
7c8d9e7c CL	156	req->rl_reply = NULL;
c8bbe0c7	157
f7d46681	158	spin_lock(&xprt->bc_pa_lock);
f531a5db	159	list_add_tail(&rqst->rq_bc_pa_list, &xprt->bc_pa_list);
f7d46681	160	spin_unlock(&xprt->bc_pa_lock);
f531a5db	161	}
63cae470	162
3f9c7e76 CL	163	static struct rpc_rqst rpcrdma_bc_rqst_get(struct rpcrdma_xprt r_xprt)
	164	{
	165	struct rpc_xprt *xprt = &r_xprt->rx_xprt;
	166	struct rpcrdma_req *req;
	167	struct rpc_rqst *rqst;
	168	size_t size;
	169
	170	spin_lock(&xprt->bc_pa_lock);
	171	rqst = list_first_entry_or_null(&xprt->bc_pa_list, struct rpc_rqst,
	172	rq_bc_pa_list);
	173	if (!rqst)
	174	goto create_req;
	175	list_del(&rqst->rq_bc_pa_list);
	176	spin_unlock(&xprt->bc_pa_lock);
	177	return rqst;
	178
	179	create_req:
	180	spin_unlock(&xprt->bc_pa_lock);
	181
	182	/* Set a limit to prevent a remote from overrunning our resources.
	183	*/
	184	if (xprt->bc_alloc_count >= RPCRDMA_BACKWARD_WRS)
	185	return NULL;
	186
94087e97	187	size = min_t(size_t, r_xprt->rx_ep.rep_inline_recv, PAGE_SIZE);
3f9c7e76 CL	188	req = rpcrdma_req_create(r_xprt, size, GFP_KERNEL);
	189	if (!req)
	190	return NULL;
	191
	192	xprt->bc_alloc_count++;
	193	rqst = &req->rl_slot;
	194	rqst->rq_xprt = xprt;
	195	__set_bit(RPC_BC_PA_IN_USE, &rqst->rq_bc_pa_state);
	196	xdr_buf_init(&rqst->rq_snd_buf, rdmab_data(req->rl_sendbuf), size);
	197	return rqst;
	198	}
	199
63cae470 CL	200	/**
63cae470 CL	201	* rpcrdma_bc_receive_call - Handle a backward direction call
9ab6d89e	202	* @r_xprt: transport receiving the call
63cae470 CL	203	* @rep: receive buffer containing the call
63cae470 CL	204	*
63cae470 CL	205	* Operational assumptions:
	206	* o Backchannel credits are ignored, just as the NFS server
	207	* forechannel currently does
	208	* o The ULP manages a replay cache (eg, NFSv4.1 sessions).
	209	* No replay detection is done at the transport level
	210	*/
	211	void rpcrdma_bc_receive_call(struct rpcrdma_xprt *r_xprt,
	212	struct rpcrdma_rep *rep)
	213	{
	214	struct rpc_xprt *xprt = &r_xprt->rx_xprt;
63cae470 CL	215	struct svc_serv *bc_serv;
	216	struct rpcrdma_req *req;
	217	struct rpc_rqst *rqst;
	218	struct xdr_buf *buf;
	219	size_t size;
	220	__be32 *p;
	221
41c8f70f CL	222	p = xdr_inline_decode(&rep->rr_stream, 0);
	223	size = xdr_stream_remaining(&rep->rr_stream);
	224
63cae470 CL	225	#ifdef RPCRDMA_BACKCHANNEL_DEBUG
63cae470 CL	226	pr_info("RPC: %s: callback XID %08x, length=%u\n",
41c8f70f CL	227	__func__, be32_to_cpup(p), size);
41c8f70f CL	228	pr_info("RPC: %s: %*ph\n", __func__, size, p);
63cae470 CL	229	#endif
63cae470 CL	230
3f9c7e76 CL	231	rqst = rpcrdma_bc_rqst_get(r_xprt);
3f9c7e76 CL	232	if (!rqst)
63cae470	233	goto out_overflow;
63cae470	234
63cae470	235	rqst->rq_reply_bytes_recvd = 0;
41c8f70f	236	rqst->rq_xid = *p;
9f74660b CL	237
9f74660b CL	238	rqst->rq_private_buf.len = size;
63cae470 CL	239
	240	buf = &rqst->rq_rcv_buf;
	241	memset(buf, 0, sizeof(*buf));
	242	buf->head[0].iov_base = p;
	243	buf->head[0].iov_len = size;
	244	buf->len = size;
	245
	246	/* The receive buffer has to be hooked to the rpcrdma_req
41c8f70f CL	247	* so that it is not released while the req is pointing
	248	* to its buffer, and so that it can be reposted after
	249	* the Upper Layer is done decoding it.
63cae470 CL	250	*/
63cae470 CL	251	req = rpcr_to_rdmar(rqst);
63cae470	252	req->rl_reply = rep;
fc1eb807	253	trace_xprtrdma_cb_call(rqst);
63cae470	254
63cae470 CL	255	/* Queue rqst for ULP's callback service */
	256	bc_serv = xprt->bc_serv;
	257	spin_lock(&bc_serv->sv_cb_lock);
	258	list_add(&rqst->rq_bc_list, &bc_serv->sv_cb_list);
	259	spin_unlock(&bc_serv->sv_cb_lock);
	260
	261	wake_up(&bc_serv->sv_cb_waitq);
	262
	263	r_xprt->rx_stats.bcall_count++;
	264	return;
	265
	266	out_overflow:
	267	pr_warn("RPC/RDMA backchannel overflow\n");
0c0829bc	268	xprt_force_disconnect(xprt);
63cae470 CL	269	/* This receive buffer gets reposted automatically
	270	* when the connection is re-established.
	271	*/
	272	return;
63cae470	273	}