[mirror_ubuntu-bionic-kernel.git] / net / sunrpc / xprtrdma / backchannel.c

/*
 * Copyright (c) 2015 Oracle.  All rights reserved.
 *
 * Support for backward direction RPCs on RPC/RDMA.
 */

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

#include "xprt_rdma.h"

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

#undef RPCRDMA_BACKCHANNEL_DEBUG

static void rpcrdma_bc_free_rqst(struct rpcrdma_xprt *r_xprt,
				 struct rpc_rqst *rqst)
{
	struct rpcrdma_buffer *buf = &r_xprt->rx_buf;
	struct rpcrdma_req *req = rpcr_to_rdmar(rqst);

	spin_lock(&buf->rb_reqslock);
	list_del(&req->rl_all);
	spin_unlock(&buf->rb_reqslock);

	rpcrdma_destroy_req(req);

	kfree(rqst);
}

static int rpcrdma_bc_setup_rqst(struct rpcrdma_xprt *r_xprt,
				 struct rpc_rqst *rqst)
{
	struct rpcrdma_regbuf *rb;
	struct rpcrdma_req *req;
	size_t size;

	req = rpcrdma_create_req(r_xprt);
	if (IS_ERR(req))
		return PTR_ERR(req);
	req->rl_backchannel = true;

	rb = rpcrdma_alloc_regbuf(RPCRDMA_HDRBUF_SIZE,
				  DMA_TO_DEVICE, GFP_KERNEL);
	if (IS_ERR(rb))
		goto out_fail;
	req->rl_rdmabuf = rb;

	size = r_xprt->rx_data.inline_rsize;
	rb = rpcrdma_alloc_regbuf(size, DMA_TO_DEVICE, GFP_KERNEL);
	if (IS_ERR(rb))
		goto out_fail;
	req->rl_sendbuf = rb;
	xdr_buf_init(&rqst->rq_snd_buf, rb->rg_base, size);
	rpcrdma_set_xprtdata(rqst, req);
	return 0;

out_fail:
	rpcrdma_bc_free_rqst(r_xprt, rqst);
	return -ENOMEM;
}

/* Allocate and add receive buffers to the rpcrdma_buffer's
 * existing list of rep's. These are released when the
 * transport is destroyed.
 */
static int rpcrdma_bc_setup_reps(struct rpcrdma_xprt *r_xprt,
				 unsigned int count)
{
	struct rpcrdma_rep *rep;
	int rc = 0;

	while (count--) {
		rep = rpcrdma_create_rep(r_xprt);
		if (IS_ERR(rep)) {
			pr_err("RPC:       %s: reply buffer alloc failed\n",
			       __func__);
			rc = PTR_ERR(rep);
			break;
		}

		rpcrdma_recv_buffer_put(rep);
	}

	return rc;
}

/**
 * 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);
	struct rpcrdma_buffer *buffer = &r_xprt->rx_buf;
	struct rpc_rqst *rqst;
	unsigned int i;
	int rc;

	/* The backchannel reply path returns each rpc_rqst to the
	 * bc_pa_list _after_ the reply is sent. If the server is
	 * faster than the client, it can send another backward
	 * direction request before the rpc_rqst is returned to the
	 * list. The client rejects the request in this case.
	 *
	 * Twice as many rpc_rqsts are prepared to ensure there is
	 * always an rpc_rqst available as soon as a reply is sent.
	 */
	if (reqs > RPCRDMA_BACKWARD_WRS >> 1)
		goto out_err;

	for (i = 0; i < (reqs << 1); i++) {
		rqst = kzalloc(sizeof(*rqst), GFP_KERNEL);
		if (!rqst) {
			pr_err("RPC:       %s: Failed to create bc rpc_rqst\n",
			       __func__);
			goto out_free;
		}
		dprintk("RPC:       %s: new rqst %p\n", __func__, rqst);

		rqst->rq_xprt = &r_xprt->rx_xprt;
		INIT_LIST_HEAD(&rqst->rq_list);
		INIT_LIST_HEAD(&rqst->rq_bc_list);

		if (rpcrdma_bc_setup_rqst(r_xprt, rqst))
			goto out_free;

		spin_lock_bh(&xprt->bc_pa_lock);
		list_add(&rqst->rq_bc_pa_list, &xprt->bc_pa_list);
		spin_unlock_bh(&xprt->bc_pa_lock);
	}

	rc = rpcrdma_bc_setup_reps(r_xprt, reqs);
	if (rc)
		goto out_free;

	rc = rpcrdma_ep_post_extra_recv(r_xprt, reqs);
	if (rc)
		goto out_free;

	buffer->rb_bc_srv_max_requests = reqs;
	request_module("svcrdma");

	return 0;

out_free:
	xprt_rdma_bc_destroy(xprt, reqs);

out_err:
	pr_err("RPC:       %s: setup backchannel transport failed\n", __func__);
	return -ENOMEM;
}

/**
 * xprt_rdma_bc_up - Create transport endpoint for backchannel service
 * @serv: server endpoint
 * @net: network namespace
 *
 * The "xprt" is an implied argument: it supplies the name of the
 * backchannel transport class.
 *
 * Returns zero on success, negative errno on failure
 */
int xprt_rdma_bc_up(struct svc_serv *serv, struct net *net)
{
	int ret;

	ret = svc_create_xprt(serv, "rdma-bc", net, PF_INET, 0, 0);
	if (ret < 0)
		return ret;
	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_create_data_internal *cdata = &r_xprt->rx_data;
	size_t maxmsg;

	maxmsg = min_t(unsigned int, cdata->inline_rsize, cdata->inline_wsize);
	return maxmsg - RPCRDMA_HDRLEN_MIN;
}

/**
 * rpcrdma_bc_marshal_reply - Send backwards direction reply
 * @rqst: buffer containing RPC reply data
 *
 * Returns zero on success.
 */
int rpcrdma_bc_marshal_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);
	struct rpcrdma_msg *headerp;
	size_t rpclen;

	headerp = rdmab_to_msg(req->rl_rdmabuf);
	headerp->rm_xid = rqst->rq_xid;
	headerp->rm_vers = rpcrdma_version;
	headerp->rm_credit =
			cpu_to_be32(r_xprt->rx_buf.rb_bc_srv_max_requests);
	headerp->rm_type = rdma_msg;
	headerp->rm_body.rm_chunks[0] = xdr_zero;
	headerp->rm_body.rm_chunks[1] = xdr_zero;
	headerp->rm_body.rm_chunks[2] = xdr_zero;

	rpclen = rqst->rq_svec[0].iov_len;

#ifdef RPCRDMA_BACKCHANNEL_DEBUG
	pr_info("RPC:       %s: rpclen %zd headerp 0x%p lkey 0x%x\n",
		__func__, rpclen, headerp, rdmab_lkey(req->rl_rdmabuf));
	pr_info("RPC:       %s: RPC/RDMA: %*ph\n",
		__func__, (int)RPCRDMA_HDRLEN_MIN, headerp);
	pr_info("RPC:       %s:      RPC: %*ph\n",
		__func__, (int)rpclen, rqst->rq_svec[0].iov_base);
#endif

	if (!rpcrdma_dma_map_regbuf(&r_xprt->rx_ia, req->rl_rdmabuf))
		goto out_map;
	req->rl_send_iov[0].addr = rdmab_addr(req->rl_rdmabuf);
	req->rl_send_iov[0].length = RPCRDMA_HDRLEN_MIN;
	req->rl_send_iov[0].lkey = rdmab_lkey(req->rl_rdmabuf);

	if (!rpcrdma_dma_map_regbuf(&r_xprt->rx_ia, req->rl_sendbuf))
		goto out_map;
	req->rl_send_iov[1].addr = rdmab_addr(req->rl_sendbuf);
	req->rl_send_iov[1].length = rpclen;
	req->rl_send_iov[1].lkey = rdmab_lkey(req->rl_sendbuf);

	req->rl_send_wr.num_sge = 2;

	return 0;

out_map:
	pr_err("rpcrdma: failed to DMA map a Send buffer\n");
	return -EIO;
}

/**
 * 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 rpcrdma_xprt *r_xprt = rpcx_to_rdmax(xprt);
	struct rpc_rqst *rqst, *tmp;

	spin_lock_bh(&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_bh(&xprt->bc_pa_lock);

		rpcrdma_bc_free_rqst(r_xprt, rqst);

		spin_lock_bh(&xprt->bc_pa_lock);
	}
	spin_unlock_bh(&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 rpc_xprt *xprt = rqst->rq_xprt;

	dprintk("RPC:       %s: freeing rqst %p (req %p)\n",
		__func__, rqst, rpcr_to_rdmar(rqst));

	smp_mb__before_atomic();
	WARN_ON_ONCE(!test_bit(RPC_BC_PA_IN_USE, &rqst->rq_bc_pa_state));
	clear_bit(RPC_BC_PA_IN_USE, &rqst->rq_bc_pa_state);
	smp_mb__after_atomic();

	spin_lock_bh(&xprt->bc_pa_lock);
	list_add_tail(&rqst->rq_bc_pa_list, &xprt->bc_pa_list);
	spin_unlock_bh(&xprt->bc_pa_lock);
}

/**
 * rpcrdma_bc_receive_call - Handle a backward direction call
 * @xprt: transport receiving the call
 * @rep: receive buffer containing the call
 *
 * Called in the RPC reply handler, which runs in a tasklet.
 * Be quick about it.
 *
 * 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 rpcrdma_msg *headerp;
	struct svc_serv *bc_serv;
	struct rpcrdma_req *req;
	struct rpc_rqst *rqst;
	struct xdr_buf *buf;
	size_t size;
	__be32 *p;

	headerp = rdmab_to_msg(rep->rr_rdmabuf);
#ifdef RPCRDMA_BACKCHANNEL_DEBUG
	pr_info("RPC:       %s: callback XID %08x, length=%u\n",
		__func__, be32_to_cpu(headerp->rm_xid), rep->rr_len);
	pr_info("RPC:       %s: %*ph\n", __func__, rep->rr_len, headerp);
#endif

	/* Sanity check:
	 * Need at least enough bytes for RPC/RDMA header, as code
	 * here references the header fields by array offset. Also,
	 * backward calls are always inline, so ensure there
	 * are some bytes beyond the RPC/RDMA header.
	 */
	if (rep->rr_len < RPCRDMA_HDRLEN_MIN + 24)
		goto out_short;
	p = (__be32 *)((unsigned char *)headerp + RPCRDMA_HDRLEN_MIN);
	size = rep->rr_len - RPCRDMA_HDRLEN_MIN;

	/* Grab a free bc rqst */
	spin_lock(&xprt->bc_pa_lock);
	if (list_empty(&xprt->bc_pa_list)) {
		spin_unlock(&xprt->bc_pa_lock);
		goto out_overflow;
	}
	rqst = list_first_entry(&xprt->bc_pa_list,
				struct rpc_rqst, rq_bc_pa_list);
	list_del(&rqst->rq_bc_pa_list);
	spin_unlock(&xprt->bc_pa_lock);
	dprintk("RPC:       %s: using rqst %p\n", __func__, rqst);

	/* Prepare rqst */
	rqst->rq_reply_bytes_recvd = 0;
	rqst->rq_bytes_sent = 0;
	rqst->rq_xid = headerp->rm_xid;

	rqst->rq_private_buf.len = size;
	set_bit(RPC_BC_PA_IN_USE, &rqst->rq_bc_pa_state);

	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 can be reposted after the server is done
	 * parsing it but just before sending the backward
	 * direction reply.
	 */
	req = rpcr_to_rdmar(rqst);
	dprintk("RPC:       %s: attaching rep %p to req %p\n",
		__func__, rep, req);
	req->rl_reply = rep;

	/* Defeat the retransmit detection logic in send_request */
	req->rl_connect_cookie = 0;

	/* 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_disconnect_done(xprt);
	/* This receive buffer gets reposted automatically
	 * when the connection is re-established.
	 */
	return;

out_short:
	pr_warn("RPC/RDMA short backward direction call\n");

	if (rpcrdma_ep_post_recv(&r_xprt->rx_ia, rep))
		xprt_disconnect_done(xprt);
	else
		pr_warn("RPC:       %s: reposting rep %p\n",
			__func__, rep);
}
Commit	Line	Data
f531a5db CL	1	/*
	2	* Copyright (c) 2015 Oracle. All rights reserved.
	3	*
	4	* Support for backward direction RPCs on RPC/RDMA.
	5	*/
	6
	7	#include <linux/module.h>
63cae470 CL	8	#include <linux/sunrpc/xprt.h>
63cae470 CL	9	#include <linux/sunrpc/svc.h>
76566773	10	#include <linux/sunrpc/svc_xprt.h>
f531a5db CL	11
	12	#include "xprt_rdma.h"
	13
	14	#if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
	15	# define RPCDBG_FACILITY RPCDBG_TRANS
	16	#endif
	17
c8bbe0c7	18	#undef RPCRDMA_BACKCHANNEL_DEBUG
63cae470	19
f531a5db CL	20	static void rpcrdma_bc_free_rqst(struct rpcrdma_xprt *r_xprt,
	21	struct rpc_rqst *rqst)
	22	{
	23	struct rpcrdma_buffer *buf = &r_xprt->rx_buf;
	24	struct rpcrdma_req *req = rpcr_to_rdmar(rqst);
	25
	26	spin_lock(&buf->rb_reqslock);
	27	list_del(&req->rl_all);
	28	spin_unlock(&buf->rb_reqslock);
	29
13650c23	30	rpcrdma_destroy_req(req);
f531a5db CL	31
	32	kfree(rqst);
	33	}
	34
	35	static int rpcrdma_bc_setup_rqst(struct rpcrdma_xprt *r_xprt,
	36	struct rpc_rqst *rqst)
	37	{
f531a5db CL	38	struct rpcrdma_regbuf *rb;
f531a5db CL	39	struct rpcrdma_req *req;
f531a5db CL	40	size_t size;
	41
	42	req = rpcrdma_create_req(r_xprt);
abfb6897 DC	43	if (IS_ERR(req))
abfb6897 DC	44	return PTR_ERR(req);
f531a5db CL	45	req->rl_backchannel = true;
f531a5db CL	46
13650c23	47	rb = rpcrdma_alloc_regbuf(RPCRDMA_HDRBUF_SIZE,
99ef4db3	48	DMA_TO_DEVICE, GFP_KERNEL);
f531a5db CL	49	if (IS_ERR(rb))
	50	goto out_fail;
	51	req->rl_rdmabuf = rb;
	52
08cf2efd	53	size = r_xprt->rx_data.inline_rsize;
13650c23	54	rb = rpcrdma_alloc_regbuf(size, DMA_TO_DEVICE, GFP_KERNEL);
f531a5db CL	55	if (IS_ERR(rb))
f531a5db CL	56	goto out_fail;
f531a5db	57	req->rl_sendbuf = rb;
5a6d1db4 CL	58	xdr_buf_init(&rqst->rq_snd_buf, rb->rg_base, size);
5a6d1db4 CL	59	rpcrdma_set_xprtdata(rqst, req);
f531a5db CL	60	return 0;
	61
	62	out_fail:
	63	rpcrdma_bc_free_rqst(r_xprt, rqst);
	64	return -ENOMEM;
	65	}
	66
	67	/* Allocate and add receive buffers to the rpcrdma_buffer's
	68	* existing list of rep's. These are released when the
	69	* transport is destroyed.
	70	*/
	71	static int rpcrdma_bc_setup_reps(struct rpcrdma_xprt *r_xprt,
	72	unsigned int count)
	73	{
f531a5db	74	struct rpcrdma_rep *rep;
f531a5db CL	75	int rc = 0;
	76
	77	while (count--) {
	78	rep = rpcrdma_create_rep(r_xprt);
	79	if (IS_ERR(rep)) {
	80	pr_err("RPC: %s: reply buffer alloc failed\n",
	81	__func__);
	82	rc = PTR_ERR(rep);
	83	break;
	84	}
	85
9b06688b	86	rpcrdma_recv_buffer_put(rep);
f531a5db CL	87	}
	88
	89	return rc;
	90	}
	91
	92	/**
	93	* xprt_rdma_bc_setup - Pre-allocate resources for handling backchannel requests
	94	* @xprt: transport associated with these backchannel resources
	95	* @reqs: number of concurrent incoming requests to expect
	96	*
	97	* Returns 0 on success; otherwise a negative errno
	98	*/
	99	int xprt_rdma_bc_setup(struct rpc_xprt *xprt, unsigned int reqs)
	100	{
	101	struct rpcrdma_xprt *r_xprt = rpcx_to_rdmax(xprt);
	102	struct rpcrdma_buffer *buffer = &r_xprt->rx_buf;
	103	struct rpc_rqst *rqst;
	104	unsigned int i;
	105	int rc;
	106
	107	/* The backchannel reply path returns each rpc_rqst to the
	108	* bc_pa_list _after_ the reply is sent. If the server is
	109	* faster than the client, it can send another backward
	110	* direction request before the rpc_rqst is returned to the
	111	* list. The client rejects the request in this case.
	112	*
	113	* Twice as many rpc_rqsts are prepared to ensure there is
	114	* always an rpc_rqst available as soon as a reply is sent.
	115	*/
124fa17d CL	116	if (reqs > RPCRDMA_BACKWARD_WRS >> 1)
	117	goto out_err;
	118
f531a5db CL	119	for (i = 0; i < (reqs << 1); i++) {
	120	rqst = kzalloc(sizeof(*rqst), GFP_KERNEL);
	121	if (!rqst) {
	122	pr_err("RPC: %s: Failed to create bc rpc_rqst\n",
	123	__func__);
	124	goto out_free;
	125	}
c8bbe0c7	126	dprintk("RPC: %s: new rqst %p\n", __func__, rqst);
f531a5db CL	127
	128	rqst->rq_xprt = &r_xprt->rx_xprt;
	129	INIT_LIST_HEAD(&rqst->rq_list);
	130	INIT_LIST_HEAD(&rqst->rq_bc_list);
	131
	132	if (rpcrdma_bc_setup_rqst(r_xprt, rqst))
	133	goto out_free;
	134
	135	spin_lock_bh(&xprt->bc_pa_lock);
	136	list_add(&rqst->rq_bc_pa_list, &xprt->bc_pa_list);
	137	spin_unlock_bh(&xprt->bc_pa_lock);
	138	}
	139
	140	rc = rpcrdma_bc_setup_reps(r_xprt, reqs);
	141	if (rc)
	142	goto out_free;
	143
	144	rc = rpcrdma_ep_post_extra_recv(r_xprt, reqs);
	145	if (rc)
	146	goto out_free;
	147
	148	buffer->rb_bc_srv_max_requests = reqs;
	149	request_module("svcrdma");
	150
	151	return 0;
	152
	153	out_free:
	154	xprt_rdma_bc_destroy(xprt, reqs);
	155
124fa17d	156	out_err:
f531a5db CL	157	pr_err("RPC: %s: setup backchannel transport failed\n", __func__);
	158	return -ENOMEM;
	159	}
	160
76566773 CL	161	/**
	162	* xprt_rdma_bc_up - Create transport endpoint for backchannel service
	163	* @serv: server endpoint
	164	* @net: network namespace
	165	*
	166	* The "xprt" is an implied argument: it supplies the name of the
	167	* backchannel transport class.
	168	*
	169	* Returns zero on success, negative errno on failure
	170	*/
	171	int xprt_rdma_bc_up(struct svc_serv serv, struct net net)
	172	{
	173	int ret;
	174
	175	ret = svc_create_xprt(serv, "rdma-bc", net, PF_INET, 0, 0);
	176	if (ret < 0)
	177	return ret;
	178	return 0;
	179	}
	180
6b26cc8c CL	181	/**
	182	* xprt_rdma_bc_maxpayload - Return maximum backchannel message size
	183	* @xprt: transport
	184	*
	185	* Returns maximum size, in bytes, of a backchannel message
	186	*/
	187	size_t xprt_rdma_bc_maxpayload(struct rpc_xprt *xprt)
	188	{
	189	struct rpcrdma_xprt *r_xprt = rpcx_to_rdmax(xprt);
	190	struct rpcrdma_create_data_internal *cdata = &r_xprt->rx_data;
	191	size_t maxmsg;
	192
	193	maxmsg = min_t(unsigned int, cdata->inline_rsize, cdata->inline_wsize);
	194	return maxmsg - RPCRDMA_HDRLEN_MIN;
	195	}
	196
83128a60 CL	197	/**
	198	* rpcrdma_bc_marshal_reply - Send backwards direction reply
	199	* @rqst: buffer containing RPC reply data
	200	*
	201	* Returns zero on success.
	202	*/
	203	int rpcrdma_bc_marshal_reply(struct rpc_rqst *rqst)
	204	{
	205	struct rpc_xprt *xprt = rqst->rq_xprt;
	206	struct rpcrdma_xprt *r_xprt = rpcx_to_rdmax(xprt);
	207	struct rpcrdma_req *req = rpcr_to_rdmar(rqst);
	208	struct rpcrdma_msg *headerp;
	209	size_t rpclen;
	210
	211	headerp = rdmab_to_msg(req->rl_rdmabuf);
	212	headerp->rm_xid = rqst->rq_xid;
	213	headerp->rm_vers = rpcrdma_version;
	214	headerp->rm_credit =
	215	cpu_to_be32(r_xprt->rx_buf.rb_bc_srv_max_requests);
	216	headerp->rm_type = rdma_msg;
	217	headerp->rm_body.rm_chunks[0] = xdr_zero;
	218	headerp->rm_body.rm_chunks[1] = xdr_zero;
	219	headerp->rm_body.rm_chunks[2] = xdr_zero;
	220
	221	rpclen = rqst->rq_svec[0].iov_len;
	222
c8bbe0c7	223	#ifdef RPCRDMA_BACKCHANNEL_DEBUG
83128a60 CL	224	pr_info("RPC: %s: rpclen %zd headerp 0x%p lkey 0x%x\n",
	225	__func__, rpclen, headerp, rdmab_lkey(req->rl_rdmabuf));
	226	pr_info("RPC: %s: RPC/RDMA: %*ph\n",
	227	__func__, (int)RPCRDMA_HDRLEN_MIN, headerp);
	228	pr_info("RPC: %s: RPC: %*ph\n",
	229	__func__, (int)rpclen, rqst->rq_svec[0].iov_base);
c8bbe0c7	230	#endif
83128a60	231
54cbd6b0 CL	232	if (!rpcrdma_dma_map_regbuf(&r_xprt->rx_ia, req->rl_rdmabuf))
54cbd6b0 CL	233	goto out_map;
83128a60 CL	234	req->rl_send_iov[0].addr = rdmab_addr(req->rl_rdmabuf);
	235	req->rl_send_iov[0].length = RPCRDMA_HDRLEN_MIN;
	236	req->rl_send_iov[0].lkey = rdmab_lkey(req->rl_rdmabuf);
	237
54cbd6b0 CL	238	if (!rpcrdma_dma_map_regbuf(&r_xprt->rx_ia, req->rl_sendbuf))
54cbd6b0 CL	239	goto out_map;
83128a60 CL	240	req->rl_send_iov[1].addr = rdmab_addr(req->rl_sendbuf);
	241	req->rl_send_iov[1].length = rpclen;
	242	req->rl_send_iov[1].lkey = rdmab_lkey(req->rl_sendbuf);
	243
90aab602 CL	244	req->rl_send_wr.num_sge = 2;
90aab602 CL	245
83128a60	246	return 0;
54cbd6b0 CL	247
	248	out_map:
	249	pr_err("rpcrdma: failed to DMA map a Send buffer\n");
	250	return -EIO;
83128a60 CL	251	}
83128a60 CL	252
f531a5db CL	253	/**
	254	* xprt_rdma_bc_destroy - Release resources for handling backchannel requests
	255	* @xprt: transport associated with these backchannel resources
	256	* @reqs: number of incoming requests to destroy; ignored
	257	*/
	258	void xprt_rdma_bc_destroy(struct rpc_xprt *xprt, unsigned int reqs)
	259	{
	260	struct rpcrdma_xprt *r_xprt = rpcx_to_rdmax(xprt);
	261	struct rpc_rqst rqst, tmp;
	262
	263	spin_lock_bh(&xprt->bc_pa_lock);
	264	list_for_each_entry_safe(rqst, tmp, &xprt->bc_pa_list, rq_bc_pa_list) {
	265	list_del(&rqst->rq_bc_pa_list);
	266	spin_unlock_bh(&xprt->bc_pa_lock);
	267
	268	rpcrdma_bc_free_rqst(r_xprt, rqst);
	269
	270	spin_lock_bh(&xprt->bc_pa_lock);
	271	}
	272	spin_unlock_bh(&xprt->bc_pa_lock);
	273	}
	274
	275	/**
	276	* xprt_rdma_bc_free_rqst - Release a backchannel rqst
	277	* @rqst: request to release
	278	*/
	279	void xprt_rdma_bc_free_rqst(struct rpc_rqst *rqst)
	280	{
	281	struct rpc_xprt *xprt = rqst->rq_xprt;
	282
c8bbe0c7 CL	283	dprintk("RPC: %s: freeing rqst %p (req %p)\n",
	284	__func__, rqst, rpcr_to_rdmar(rqst));
	285
f531a5db CL	286	smp_mb__before_atomic();
	287	WARN_ON_ONCE(!test_bit(RPC_BC_PA_IN_USE, &rqst->rq_bc_pa_state));
	288	clear_bit(RPC_BC_PA_IN_USE, &rqst->rq_bc_pa_state);
	289	smp_mb__after_atomic();
	290
	291	spin_lock_bh(&xprt->bc_pa_lock);
	292	list_add_tail(&rqst->rq_bc_pa_list, &xprt->bc_pa_list);
	293	spin_unlock_bh(&xprt->bc_pa_lock);
	294	}
63cae470 CL	295
	296	/**
	297	* rpcrdma_bc_receive_call - Handle a backward direction call
	298	* @xprt: transport receiving the call
	299	* @rep: receive buffer containing the call
	300	*
	301	* Called in the RPC reply handler, which runs in a tasklet.
	302	* Be quick about it.
	303	*
	304	* Operational assumptions:
	305	* o Backchannel credits are ignored, just as the NFS server
	306	* forechannel currently does
	307	* o The ULP manages a replay cache (eg, NFSv4.1 sessions).
	308	* No replay detection is done at the transport level
	309	*/
	310	void rpcrdma_bc_receive_call(struct rpcrdma_xprt *r_xprt,
	311	struct rpcrdma_rep *rep)
	312	{
	313	struct rpc_xprt *xprt = &r_xprt->rx_xprt;
	314	struct rpcrdma_msg *headerp;
	315	struct svc_serv *bc_serv;
	316	struct rpcrdma_req *req;
	317	struct rpc_rqst *rqst;
	318	struct xdr_buf *buf;
	319	size_t size;
	320	__be32 *p;
	321
	322	headerp = rdmab_to_msg(rep->rr_rdmabuf);
	323	#ifdef RPCRDMA_BACKCHANNEL_DEBUG
	324	pr_info("RPC: %s: callback XID %08x, length=%u\n",
	325	__func__, be32_to_cpu(headerp->rm_xid), rep->rr_len);
	326	pr_info("RPC: %s: %*ph\n", __func__, rep->rr_len, headerp);
	327	#endif
	328
	329	/* Sanity check:
	330	* Need at least enough bytes for RPC/RDMA header, as code
	331	* here references the header fields by array offset. Also,
	332	* backward calls are always inline, so ensure there
	333	* are some bytes beyond the RPC/RDMA header.
	334	*/
	335	if (rep->rr_len < RPCRDMA_HDRLEN_MIN + 24)
	336	goto out_short;
	337	p = (__be32 )((unsigned char )headerp + RPCRDMA_HDRLEN_MIN);
	338	size = rep->rr_len - RPCRDMA_HDRLEN_MIN;
	339
	340	/* Grab a free bc rqst */
	341	spin_lock(&xprt->bc_pa_lock);
	342	if (list_empty(&xprt->bc_pa_list)) {
	343	spin_unlock(&xprt->bc_pa_lock);
	344	goto out_overflow;
	345	}
	346	rqst = list_first_entry(&xprt->bc_pa_list,
	347	struct rpc_rqst, rq_bc_pa_list);
	348	list_del(&rqst->rq_bc_pa_list);
	349	spin_unlock(&xprt->bc_pa_lock);
c8bbe0c7	350	dprintk("RPC: %s: using rqst %p\n", __func__, rqst);
63cae470 CL	351
	352	/* Prepare rqst */
	353	rqst->rq_reply_bytes_recvd = 0;
	354	rqst->rq_bytes_sent = 0;
	355	rqst->rq_xid = headerp->rm_xid;
9f74660b CL	356
9f74660b CL	357	rqst->rq_private_buf.len = size;
63cae470 CL	358	set_bit(RPC_BC_PA_IN_USE, &rqst->rq_bc_pa_state);
	359
	360	buf = &rqst->rq_rcv_buf;
	361	memset(buf, 0, sizeof(*buf));
	362	buf->head[0].iov_base = p;
	363	buf->head[0].iov_len = size;
	364	buf->len = size;
	365
	366	/* The receive buffer has to be hooked to the rpcrdma_req
	367	* so that it can be reposted after the server is done
	368	* parsing it but just before sending the backward
	369	* direction reply.
	370	*/
	371	req = rpcr_to_rdmar(rqst);
c8bbe0c7	372	dprintk("RPC: %s: attaching rep %p to req %p\n",
63cae470	373	__func__, rep, req);
63cae470 CL	374	req->rl_reply = rep;
	375
	376	/* Defeat the retransmit detection logic in send_request */
	377	req->rl_connect_cookie = 0;
	378
	379	/* Queue rqst for ULP's callback service */
	380	bc_serv = xprt->bc_serv;
	381	spin_lock(&bc_serv->sv_cb_lock);
	382	list_add(&rqst->rq_bc_list, &bc_serv->sv_cb_list);
	383	spin_unlock(&bc_serv->sv_cb_lock);
	384
	385	wake_up(&bc_serv->sv_cb_waitq);
	386
	387	r_xprt->rx_stats.bcall_count++;
	388	return;
	389
	390	out_overflow:
	391	pr_warn("RPC/RDMA backchannel overflow\n");
	392	xprt_disconnect_done(xprt);
	393	/* This receive buffer gets reposted automatically
	394	* when the connection is re-established.
	395	*/
	396	return;
	397
	398	out_short:
	399	pr_warn("RPC/RDMA short backward direction call\n");
	400
b157380a	401	if (rpcrdma_ep_post_recv(&r_xprt->rx_ia, rep))
63cae470 CL	402	xprt_disconnect_done(xprt);
	403	else
	404	pr_warn("RPC: %s: reposting rep %p\n",
	405	__func__, rep);
	406	}