[linux.git] / net / rds / tcp_recv.c

/*
 * Copyright (c) 2006, 2017 Oracle and/or its affiliates. All rights reserved.
 *
 * This software is available to you under a choice of one of two
 * licenses.  You may choose to be licensed under the terms of the GNU
 * General Public License (GPL) Version 2, available from the file
 * COPYING in the main directory of this source tree, or the
 * OpenIB.org BSD license below:
 *
 *     Redistribution and use in source and binary forms, with or
 *     without modification, are permitted provided that the following
 *     conditions are met:
 *
 *      - Redistributions of source code must retain the above
 *        copyright notice, this list of conditions and the following
 *        disclaimer.
 *
 *      - Redistributions in binary form must reproduce the above
 *        copyright notice, this list of conditions and the following
 *        disclaimer in the documentation and/or other materials
 *        provided with the distribution.
 *
 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
 * SOFTWARE.
 *
 */
#include <linux/kernel.h>
#include <linux/slab.h>
#include <net/tcp.h>
#include <trace/events/sock.h>

#include "rds.h"
#include "tcp.h"

static struct kmem_cache *rds_tcp_incoming_slab;

static void rds_tcp_inc_purge(struct rds_incoming *inc)
{
	struct rds_tcp_incoming *tinc;
	tinc = container_of(inc, struct rds_tcp_incoming, ti_inc);
	rdsdebug("purging tinc %p inc %p\n", tinc, inc);
	skb_queue_purge(&tinc->ti_skb_list);
}

void rds_tcp_inc_free(struct rds_incoming *inc)
{
	struct rds_tcp_incoming *tinc;
	tinc = container_of(inc, struct rds_tcp_incoming, ti_inc);
	rds_tcp_inc_purge(inc);
	rdsdebug("freeing tinc %p inc %p\n", tinc, inc);
	kmem_cache_free(rds_tcp_incoming_slab, tinc);
}

/*
 * this is pretty lame, but, whatever.
 */
int rds_tcp_inc_copy_to_user(struct rds_incoming *inc, struct iov_iter *to)
{
	struct rds_tcp_incoming *tinc;
	struct sk_buff *skb;
	int ret = 0;

	if (!iov_iter_count(to))
		goto out;

	tinc = container_of(inc, struct rds_tcp_incoming, ti_inc);

	skb_queue_walk(&tinc->ti_skb_list, skb) {
		unsigned long to_copy, skb_off;
		for (skb_off = 0; skb_off < skb->len; skb_off += to_copy) {
			to_copy = iov_iter_count(to);
			to_copy = min(to_copy, skb->len - skb_off);

			if (skb_copy_datagram_iter(skb, skb_off, to, to_copy))
				return -EFAULT;

			rds_stats_add(s_copy_to_user, to_copy);
			ret += to_copy;

			if (!iov_iter_count(to))
				goto out;
		}
	}
out:
	return ret;
}

/*
 * We have a series of skbs that have fragmented pieces of the congestion
 * bitmap.  They must add up to the exact size of the congestion bitmap.  We
 * use the skb helpers to copy those into the pages that make up the in-memory
 * congestion bitmap for the remote address of this connection.  We then tell
 * the congestion core that the bitmap has been changed so that it can wake up
 * sleepers.
 *
 * This is racing with sending paths which are using test_bit to see if the
 * bitmap indicates that their recipient is congested.
 */

static void rds_tcp_cong_recv(struct rds_connection *conn,
			      struct rds_tcp_incoming *tinc)
{
	struct sk_buff *skb;
	unsigned int to_copy, skb_off;
	unsigned int map_off;
	unsigned int map_page;
	struct rds_cong_map *map;
	int ret;

	/* catch completely corrupt packets */
	if (be32_to_cpu(tinc->ti_inc.i_hdr.h_len) != RDS_CONG_MAP_BYTES)
		return;

	map_page = 0;
	map_off = 0;
	map = conn->c_fcong;

	skb_queue_walk(&tinc->ti_skb_list, skb) {
		skb_off = 0;
		while (skb_off < skb->len) {
			to_copy = min_t(unsigned int, PAGE_SIZE - map_off,
					skb->len - skb_off);

			BUG_ON(map_page >= RDS_CONG_MAP_PAGES);

			/* only returns 0 or -error */
			ret = skb_copy_bits(skb, skb_off,
				(void *)map->m_page_addrs[map_page] + map_off,
				to_copy);
			BUG_ON(ret != 0);

			skb_off += to_copy;
			map_off += to_copy;
			if (map_off == PAGE_SIZE) {
				map_off = 0;
				map_page++;
			}
		}
	}

	rds_cong_map_updated(map, ~(u64) 0);
}

struct rds_tcp_desc_arg {
	struct rds_conn_path *conn_path;
	gfp_t gfp;
};

static int rds_tcp_data_recv(read_descriptor_t *desc, struct sk_buff *skb,
			     unsigned int offset, size_t len)
{
	struct rds_tcp_desc_arg *arg = desc->arg.data;
	struct rds_conn_path *cp = arg->conn_path;
	struct rds_tcp_connection *tc = cp->cp_transport_data;
	struct rds_tcp_incoming *tinc = tc->t_tinc;
	struct sk_buff *clone;
	size_t left = len, to_copy;

	rdsdebug("tcp data tc %p skb %p offset %u len %zu\n", tc, skb, offset,
		 len);

	/*
	 * tcp_read_sock() interprets partial progress as an indication to stop
	 * processing.
	 */
	while (left) {
		if (!tinc) {
			tinc = kmem_cache_alloc(rds_tcp_incoming_slab,
						arg->gfp);
			if (!tinc) {
				desc->error = -ENOMEM;
				goto out;
			}
			tc->t_tinc = tinc;
			rdsdebug("allocated tinc %p\n", tinc);
			rds_inc_path_init(&tinc->ti_inc, cp,
					  &cp->cp_conn->c_faddr);
			tinc->ti_inc.i_rx_lat_trace[RDS_MSG_RX_HDR] =
					local_clock();

			/*
			 * XXX * we might be able to use the __ variants when
			 * we've already serialized at a higher level.
			 */
			skb_queue_head_init(&tinc->ti_skb_list);
		}

		if (left && tc->t_tinc_hdr_rem) {
			to_copy = min(tc->t_tinc_hdr_rem, left);
			rdsdebug("copying %zu header from skb %p\n", to_copy,
				 skb);
			skb_copy_bits(skb, offset,
				      (char *)&tinc->ti_inc.i_hdr +
						sizeof(struct rds_header) -
						tc->t_tinc_hdr_rem,
				      to_copy);
			tc->t_tinc_hdr_rem -= to_copy;
			left -= to_copy;
			offset += to_copy;

			if (tc->t_tinc_hdr_rem == 0) {
				/* could be 0 for a 0 len message */
				tc->t_tinc_data_rem =
					be32_to_cpu(tinc->ti_inc.i_hdr.h_len);
				tinc->ti_inc.i_rx_lat_trace[RDS_MSG_RX_START] =
					local_clock();
			}
		}

		if (left && tc->t_tinc_data_rem) {
			to_copy = min(tc->t_tinc_data_rem, left);

			clone = pskb_extract(skb, offset, to_copy, arg->gfp);
			if (!clone) {
				desc->error = -ENOMEM;
				goto out;
			}

			skb_queue_tail(&tinc->ti_skb_list, clone);

			rdsdebug("skb %p data %p len %d off %u to_copy %zu -> "
				 "clone %p data %p len %d\n",
				 skb, skb->data, skb->len, offset, to_copy,
				 clone, clone->data, clone->len);

			tc->t_tinc_data_rem -= to_copy;
			left -= to_copy;
			offset += to_copy;
		}

		if (tc->t_tinc_hdr_rem == 0 && tc->t_tinc_data_rem == 0) {
			struct rds_connection *conn = cp->cp_conn;

			if (tinc->ti_inc.i_hdr.h_flags == RDS_FLAG_CONG_BITMAP)
				rds_tcp_cong_recv(conn, tinc);
			else
				rds_recv_incoming(conn, &conn->c_faddr,
						  &conn->c_laddr,
						  &tinc->ti_inc,
						  arg->gfp);

			tc->t_tinc_hdr_rem = sizeof(struct rds_header);
			tc->t_tinc_data_rem = 0;
			tc->t_tinc = NULL;
			rds_inc_put(&tinc->ti_inc);
			tinc = NULL;
		}
	}
out:
	rdsdebug("returning len %zu left %zu skb len %d rx queue depth %d\n",
		 len, left, skb->len,
		 skb_queue_len(&tc->t_sock->sk->sk_receive_queue));
	return len - left;
}

/* the caller has to hold the sock lock */
static int rds_tcp_read_sock(struct rds_conn_path *cp, gfp_t gfp)
{
	struct rds_tcp_connection *tc = cp->cp_transport_data;
	struct socket *sock = tc->t_sock;
	read_descriptor_t desc;
	struct rds_tcp_desc_arg arg;

	/* It's like glib in the kernel! */
	arg.conn_path = cp;
	arg.gfp = gfp;
	desc.arg.data = &arg;
	desc.error = 0;
	desc.count = 1; /* give more than one skb per call */

	tcp_read_sock(sock->sk, &desc, rds_tcp_data_recv);
	rdsdebug("tcp_read_sock for tc %p gfp 0x%x returned %d\n", tc, gfp,
		 desc.error);

	return desc.error;
}

/*
 * We hold the sock lock to serialize our rds_tcp_recv->tcp_read_sock from
 * data_ready.
 *
 * if we fail to allocate we're in trouble.. blindly wait some time before
 * trying again to see if the VM can free up something for us.
 */
int rds_tcp_recv_path(struct rds_conn_path *cp)
{
	struct rds_tcp_connection *tc = cp->cp_transport_data;
	struct socket *sock = tc->t_sock;
	int ret = 0;

	rdsdebug("recv worker path [%d] tc %p sock %p\n",
		 cp->cp_index, tc, sock);

	lock_sock(sock->sk);
	ret = rds_tcp_read_sock(cp, GFP_KERNEL);
	release_sock(sock->sk);

	return ret;
}

void rds_tcp_data_ready(struct sock *sk)
{
	void (*ready)(struct sock *sk);
	struct rds_conn_path *cp;
	struct rds_tcp_connection *tc;

	trace_sk_data_ready(sk);
	rdsdebug("data ready sk %p\n", sk);

	read_lock_bh(&sk->sk_callback_lock);
	cp = sk->sk_user_data;
	if (!cp) { /* check for teardown race */
		ready = sk->sk_data_ready;
		goto out;
	}

	tc = cp->cp_transport_data;
	ready = tc->t_orig_data_ready;
	rds_tcp_stats_inc(s_tcp_data_ready_calls);

	if (rds_tcp_read_sock(cp, GFP_ATOMIC) == -ENOMEM) {
		rcu_read_lock();
		if (!rds_destroy_pending(cp->cp_conn))
			queue_delayed_work(rds_wq, &cp->cp_recv_w, 0);
		rcu_read_unlock();
	}
out:
	read_unlock_bh(&sk->sk_callback_lock);
	ready(sk);
}

int rds_tcp_recv_init(void)
{
	rds_tcp_incoming_slab = kmem_cache_create("rds_tcp_incoming",
					sizeof(struct rds_tcp_incoming),
					0, 0, NULL);
	if (!rds_tcp_incoming_slab)
		return -ENOMEM;
	return 0;
}

void rds_tcp_recv_exit(void)
{
	kmem_cache_destroy(rds_tcp_incoming_slab);
}
Commit	Line	Data
70041088	1	/*
eee2fa6a	2	* Copyright (c) 2006, 2017 Oracle and/or its affiliates. All rights reserved.
70041088 AG	3	*
	4	* This software is available to you under a choice of one of two
	5	* licenses. You may choose to be licensed under the terms of the GNU
	6	* General Public License (GPL) Version 2, available from the file
	7	* COPYING in the main directory of this source tree, or the
	8	* OpenIB.org BSD license below:
	9	*
	10	* Redistribution and use in source and binary forms, with or
	11	* without modification, are permitted provided that the following
	12	* conditions are met:
	13	*
	14	* - Redistributions of source code must retain the above
	15	* copyright notice, this list of conditions and the following
	16	* disclaimer.
	17	*
	18	* - Redistributions in binary form must reproduce the above
	19	* copyright notice, this list of conditions and the following
	20	* disclaimer in the documentation and/or other materials
	21	* provided with the distribution.
	22	*
	23	* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
	24	* EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
	25	* MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
	26	* NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
	27	* BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
	28	* ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
	29	* CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
	30	* SOFTWARE.
	31	*
	32	*/
	33	#include <linux/kernel.h>
5a0e3ad6	34	#include <linux/slab.h>
70041088	35	#include <net/tcp.h>
40e0b090	36	#include <trace/events/sock.h>
70041088 AG	37
	38	#include "rds.h"
	39	#include "tcp.h"
	40
	41	static struct kmem_cache *rds_tcp_incoming_slab;
	42
809fa148	43	static void rds_tcp_inc_purge(struct rds_incoming *inc)
70041088 AG	44	{
	45	struct rds_tcp_incoming *tinc;
	46	tinc = container_of(inc, struct rds_tcp_incoming, ti_inc);
	47	rdsdebug("purging tinc %p inc %p\n", tinc, inc);
	48	skb_queue_purge(&tinc->ti_skb_list);
	49	}
	50
	51	void rds_tcp_inc_free(struct rds_incoming *inc)
	52	{
	53	struct rds_tcp_incoming *tinc;
	54	tinc = container_of(inc, struct rds_tcp_incoming, ti_inc);
	55	rds_tcp_inc_purge(inc);
	56	rdsdebug("freeing tinc %p inc %p\n", tinc, inc);
	57	kmem_cache_free(rds_tcp_incoming_slab, tinc);
	58	}
	59
	60	/*
	61	* this is pretty lame, but, whatever.
	62	*/
c310e72c	63	int rds_tcp_inc_copy_to_user(struct rds_incoming inc, struct iov_iter to)
70041088 AG	64	{
70041088 AG	65	struct rds_tcp_incoming *tinc;
70041088	66	struct sk_buff *skb;
70041088 AG	67	int ret = 0;
70041088 AG	68
c310e72c	69	if (!iov_iter_count(to))
70041088 AG	70	goto out;
	71
	72	tinc = container_of(inc, struct rds_tcp_incoming, ti_inc);
70041088 AG	73
70041088 AG	74	skb_queue_walk(&tinc->ti_skb_list, skb) {
c310e72c AV	75	unsigned long to_copy, skb_off;
	76	for (skb_off = 0; skb_off < skb->len; skb_off += to_copy) {
	77	to_copy = iov_iter_count(to);
70041088 AG	78	to_copy = min(to_copy, skb->len - skb_off);
70041088 AG	79
c310e72c AV	80	if (skb_copy_datagram_iter(skb, skb_off, to, to_copy))
c310e72c AV	81	return -EFAULT;
70041088	82
b075cfdb	83	rds_stats_add(s_copy_to_user, to_copy);
70041088	84	ret += to_copy;
c310e72c AV	85
c310e72c AV	86	if (!iov_iter_count(to))
70041088 AG	87	goto out;
	88	}
	89	}
	90	out:
	91	return ret;
	92	}
	93
	94	/*
	95	* We have a series of skbs that have fragmented pieces of the congestion
	96	* bitmap. They must add up to the exact size of the congestion bitmap. We
	97	* use the skb helpers to copy those into the pages that make up the in-memory
	98	* congestion bitmap for the remote address of this connection. We then tell
	99	* the congestion core that the bitmap has been changed so that it can wake up
	100	* sleepers.
	101	*
	102	* This is racing with sending paths which are using test_bit to see if the
	103	* bitmap indicates that their recipient is congested.
	104	*/
	105
	106	static void rds_tcp_cong_recv(struct rds_connection *conn,
	107	struct rds_tcp_incoming *tinc)
	108	{
	109	struct sk_buff *skb;
	110	unsigned int to_copy, skb_off;
	111	unsigned int map_off;
	112	unsigned int map_page;
	113	struct rds_cong_map *map;
	114	int ret;
	115
	116	/* catch completely corrupt packets */
	117	if (be32_to_cpu(tinc->ti_inc.i_hdr.h_len) != RDS_CONG_MAP_BYTES)
	118	return;
	119
	120	map_page = 0;
	121	map_off = 0;
	122	map = conn->c_fcong;
	123
	124	skb_queue_walk(&tinc->ti_skb_list, skb) {
	125	skb_off = 0;
	126	while (skb_off < skb->len) {
	127	to_copy = min_t(unsigned int, PAGE_SIZE - map_off,
	128	skb->len - skb_off);
	129
	130	BUG_ON(map_page >= RDS_CONG_MAP_PAGES);
	131
	132	/* only returns 0 or -error */
	133	ret = skb_copy_bits(skb, skb_off,
	134	(void *)map->m_page_addrs[map_page] + map_off,
	135	to_copy);
	136	BUG_ON(ret != 0);
	137
	138	skb_off += to_copy;
	139	map_off += to_copy;
	140	if (map_off == PAGE_SIZE) {
	141	map_off = 0;
	142	map_page++;
	143	}
	144	}
	145	}
	146
	147	rds_cong_map_updated(map, ~(u64) 0);
	148	}
	149
	150	struct rds_tcp_desc_arg {
2da43c4a	151	struct rds_conn_path *conn_path;
70041088	152	gfp_t gfp;
70041088 AG	153	};
	154
	155	static int rds_tcp_data_recv(read_descriptor_t desc, struct sk_buff skb,
	156	unsigned int offset, size_t len)
	157	{
	158	struct rds_tcp_desc_arg *arg = desc->arg.data;
2da43c4a SV	159	struct rds_conn_path *cp = arg->conn_path;
2da43c4a SV	160	struct rds_tcp_connection *tc = cp->cp_transport_data;
70041088 AG	161	struct rds_tcp_incoming *tinc = tc->t_tinc;
	162	struct sk_buff *clone;
	163	size_t left = len, to_copy;
	164
	165	rdsdebug("tcp data tc %p skb %p offset %u len %zu\n", tc, skb, offset,
	166	len);
	167
	168	/*
	169	* tcp_read_sock() interprets partial progress as an indication to stop
	170	* processing.
	171	*/
	172	while (left) {
8690bfa1	173	if (!tinc) {
70041088	174	tinc = kmem_cache_alloc(rds_tcp_incoming_slab,
5c3da57d	175	arg->gfp);
8690bfa1	176	if (!tinc) {
70041088 AG	177	desc->error = -ENOMEM;
	178	goto out;
	179	}
	180	tc->t_tinc = tinc;
379aecbc	181	rdsdebug("allocated tinc %p\n", tinc);
2da43c4a	182	rds_inc_path_init(&tinc->ti_inc, cp,
eee2fa6a	183	&cp->cp_conn->c_faddr);
3289025a SS	184	tinc->ti_inc.i_rx_lat_trace[RDS_MSG_RX_HDR] =
	185	local_clock();
	186
70041088 AG	187	/*
	188	* XXX * we might be able to use the __ variants when
	189	* we've already serialized at a higher level.
	190	*/
	191	skb_queue_head_init(&tinc->ti_skb_list);
	192	}
	193
	194	if (left && tc->t_tinc_hdr_rem) {
	195	to_copy = min(tc->t_tinc_hdr_rem, left);
	196	rdsdebug("copying %zu header from skb %p\n", to_copy,
	197	skb);
	198	skb_copy_bits(skb, offset,
	199	(char *)&tinc->ti_inc.i_hdr +
	200	sizeof(struct rds_header) -
	201	tc->t_tinc_hdr_rem,
	202	to_copy);
	203	tc->t_tinc_hdr_rem -= to_copy;
	204	left -= to_copy;
	205	offset += to_copy;
	206
	207	if (tc->t_tinc_hdr_rem == 0) {
	208	/* could be 0 for a 0 len message */
	209	tc->t_tinc_data_rem =
	210	be32_to_cpu(tinc->ti_inc.i_hdr.h_len);
3289025a SS	211	tinc->ti_inc.i_rx_lat_trace[RDS_MSG_RX_START] =
3289025a SS	212	local_clock();
70041088 AG	213	}
	214	}
	215
	216	if (left && tc->t_tinc_data_rem) {
947d2756 SV	217	to_copy = min(tc->t_tinc_data_rem, left);
	218
	219	clone = pskb_extract(skb, offset, to_copy, arg->gfp);
8690bfa1	220	if (!clone) {
70041088 AG	221	desc->error = -ENOMEM;
	222	goto out;
	223	}
	224
70041088 AG	225	skb_queue_tail(&tinc->ti_skb_list, clone);
	226
	227	rdsdebug("skb %p data %p len %d off %u to_copy %zu -> "
	228	"clone %p data %p len %d\n",
	229	skb, skb->data, skb->len, offset, to_copy,
	230	clone, clone->data, clone->len);
	231
	232	tc->t_tinc_data_rem -= to_copy;
	233	left -= to_copy;
	234	offset += to_copy;
	235	}
	236
	237	if (tc->t_tinc_hdr_rem == 0 && tc->t_tinc_data_rem == 0) {
2da43c4a SV	238	struct rds_connection *conn = cp->cp_conn;
2da43c4a SV	239
70041088 AG	240	if (tinc->ti_inc.i_hdr.h_flags == RDS_FLAG_CONG_BITMAP)
	241	rds_tcp_cong_recv(conn, tinc);
	242	else
eee2fa6a KCP	243	rds_recv_incoming(conn, &conn->c_faddr,
	244	&conn->c_laddr,
	245	&tinc->ti_inc,
6114eab5	246	arg->gfp);
70041088 AG	247
	248	tc->t_tinc_hdr_rem = sizeof(struct rds_header);
	249	tc->t_tinc_data_rem = 0;
	250	tc->t_tinc = NULL;
	251	rds_inc_put(&tinc->ti_inc);
	252	tinc = NULL;
	253	}
	254	}
	255	out:
	256	rdsdebug("returning len %zu left %zu skb len %d rx queue depth %d\n",
	257	len, left, skb->len,
	258	skb_queue_len(&tc->t_sock->sk->sk_receive_queue));
	259	return len - left;
	260	}
	261
	262	/* the caller has to hold the sock lock */
2da43c4a	263	static int rds_tcp_read_sock(struct rds_conn_path *cp, gfp_t gfp)
70041088	264	{
2da43c4a	265	struct rds_tcp_connection *tc = cp->cp_transport_data;
70041088 AG	266	struct socket *sock = tc->t_sock;
	267	read_descriptor_t desc;
	268	struct rds_tcp_desc_arg arg;
	269
	270	/* It's like glib in the kernel! */
2da43c4a	271	arg.conn_path = cp;
70041088	272	arg.gfp = gfp;
70041088 AG	273	desc.arg.data = &arg;
	274	desc.error = 0;
	275	desc.count = 1; /* give more than one skb per call */
	276
	277	tcp_read_sock(sock->sk, &desc, rds_tcp_data_recv);
	278	rdsdebug("tcp_read_sock for tc %p gfp 0x%x returned %d\n", tc, gfp,
	279	desc.error);
	280
	281	return desc.error;
	282	}
	283
	284	/*
	285	* We hold the sock lock to serialize our rds_tcp_recv->tcp_read_sock from
	286	* data_ready.
	287	*
	288	* if we fail to allocate we're in trouble.. blindly wait some time before
	289	* trying again to see if the VM can free up something for us.
	290	*/
2da43c4a	291	int rds_tcp_recv_path(struct rds_conn_path *cp)
70041088	292	{
2da43c4a	293	struct rds_tcp_connection *tc = cp->cp_transport_data;
70041088 AG	294	struct socket *sock = tc->t_sock;
	295	int ret = 0;
	296
2da43c4a SV	297	rdsdebug("recv worker path [%d] tc %p sock %p\n",
2da43c4a SV	298	cp->cp_index, tc, sock);
70041088 AG	299
70041088 AG	300	lock_sock(sock->sk);
2da43c4a	301	ret = rds_tcp_read_sock(cp, GFP_KERNEL);
70041088 AG	302	release_sock(sock->sk);
	303
	304	return ret;
	305	}
	306
676d2369	307	void rds_tcp_data_ready(struct sock *sk)
70041088	308	{
676d2369	309	void (ready)(struct sock sk);
ea3b1ea5	310	struct rds_conn_path *cp;
70041088 AG	311	struct rds_tcp_connection *tc;
70041088 AG	312
40e0b090	313	trace_sk_data_ready(sk);
676d2369	314	rdsdebug("data ready sk %p\n", sk);
70041088	315
38036629	316	read_lock_bh(&sk->sk_callback_lock);
ea3b1ea5 SV	317	cp = sk->sk_user_data;
ea3b1ea5 SV	318	if (!cp) { /* check for teardown race */
70041088 AG	319	ready = sk->sk_data_ready;
	320	goto out;
	321	}
	322
ea3b1ea5	323	tc = cp->cp_transport_data;
70041088 AG	324	ready = tc->t_orig_data_ready;
	325	rds_tcp_stats_inc(s_tcp_data_ready_calls);
	326
3db6e0d1 SV	327	if (rds_tcp_read_sock(cp, GFP_ATOMIC) == -ENOMEM) {
3db6e0d1 SV	328	rcu_read_lock();
ebeeb1ad	329	if (!rds_destroy_pending(cp->cp_conn))
3db6e0d1 SV	330	queue_delayed_work(rds_wq, &cp->cp_recv_w, 0);
	331	rcu_read_unlock();
	332	}
70041088	333	out:
38036629	334	read_unlock_bh(&sk->sk_callback_lock);
676d2369	335	ready(sk);
70041088 AG	336	}
70041088 AG	337
ef87b7ea	338	int rds_tcp_recv_init(void)
70041088 AG	339	{
	340	rds_tcp_incoming_slab = kmem_cache_create("rds_tcp_incoming",
	341	sizeof(struct rds_tcp_incoming),
	342	0, 0, NULL);
8690bfa1	343	if (!rds_tcp_incoming_slab)
70041088 AG	344	return -ENOMEM;
	345	return 0;
	346	}
	347
	348	void rds_tcp_recv_exit(void)
	349	{
	350	kmem_cache_destroy(rds_tcp_incoming_slab);
	351	}