[GitHub/mt8127/android_kernel_alcatel_ttab.git] / net / rds / tcp_recv.c

/*
 * Copyright (c) 2006 Oracle.  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 "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 iovec *first_iov,
			     size_t size)
{
	struct rds_tcp_incoming *tinc;
	struct iovec *iov, tmp;
	struct sk_buff *skb;
	unsigned long to_copy, skb_off;
	int ret = 0;

	if (size == 0)
		goto out;

	tinc = container_of(inc, struct rds_tcp_incoming, ti_inc);
	iov = first_iov;
	tmp = *iov;

	skb_queue_walk(&tinc->ti_skb_list, skb) {
		skb_off = 0;
		while (skb_off < skb->len) {
			while (tmp.iov_len == 0) {
				iov++;
				tmp = *iov;
			}

			to_copy = min(tmp.iov_len, size);
			to_copy = min(to_copy, skb->len - skb_off);

			rdsdebug("ret %d size %zu skb %p skb_off %lu "
				 "skblen %d iov_base %p iov_len %zu cpy %lu\n",
				 ret, size, skb, skb_off, skb->len,
				 tmp.iov_base, tmp.iov_len, to_copy);

			/* modifies tmp as it copies */
			if (skb_copy_datagram_iovec(skb, skb_off, &tmp,
						    to_copy)) {
				ret = -EFAULT;
				goto out;
			}

			rds_stats_add(s_copy_to_user, to_copy);
			size -= to_copy;
			ret += to_copy;
			skb_off += to_copy;
			if (size == 0)
				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_connection *conn;
	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_connection *conn = arg->conn;
	struct rds_tcp_connection *tc = conn->c_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("alloced tinc %p\n", tinc);
			rds_inc_init(&tinc->ti_inc, conn, conn->c_faddr);
			/*
			 * 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);
			}
		}

		if (left && tc->t_tinc_data_rem) {
			clone = skb_clone(skb, arg->gfp);
			if (!clone) {
				desc->error = -ENOMEM;
				goto out;
			}

			to_copy = min(tc->t_tinc_data_rem, left);
			if (!pskb_pull(clone, offset) ||
			    pskb_trim(clone, to_copy)) {
				pr_warn("rds_tcp_data_recv: pull/trim failed "
					"left %zu data_rem %zu skb_len %d\n",
					left, tc->t_tinc_data_rem, skb->len);
				kfree_skb(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) {
			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_connection *conn, gfp_t gfp)
{
	struct rds_tcp_connection *tc = conn->c_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 = conn;
	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(struct rds_connection *conn)
{
	struct rds_tcp_connection *tc = conn->c_transport_data;
	struct socket *sock = tc->t_sock;
	int ret = 0;

	rdsdebug("recv worker conn %p tc %p sock %p\n", conn, tc, sock);

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

	return ret;
}

void rds_tcp_data_ready(struct sock *sk, int bytes)
{
	void (*ready)(struct sock *sk, int bytes);
	struct rds_connection *conn;
	struct rds_tcp_connection *tc;

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

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

	tc = conn->c_transport_data;
	ready = tc->t_orig_data_ready;
	rds_tcp_stats_inc(s_tcp_data_ready_calls);

	if (rds_tcp_read_sock(conn, GFP_ATOMIC) == -ENOMEM)
		queue_delayed_work(rds_wq, &conn->c_recv_w, 0);
out:
	read_unlock(&sk->sk_callback_lock);
	ready(sk, bytes);
}

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 AG	1	/*
	2	* Copyright (c) 2006 Oracle. All rights reserved.
	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 AG	35	#include <net/tcp.h>
	36
	37	#include "rds.h"
	38	#include "tcp.h"
	39
	40	static struct kmem_cache *rds_tcp_incoming_slab;
	41
809fa148	42	static void rds_tcp_inc_purge(struct rds_incoming *inc)
70041088 AG	43	{
	44	struct rds_tcp_incoming *tinc;
	45	tinc = container_of(inc, struct rds_tcp_incoming, ti_inc);
	46	rdsdebug("purging tinc %p inc %p\n", tinc, inc);
	47	skb_queue_purge(&tinc->ti_skb_list);
	48	}
	49
	50	void rds_tcp_inc_free(struct rds_incoming *inc)
	51	{
	52	struct rds_tcp_incoming *tinc;
	53	tinc = container_of(inc, struct rds_tcp_incoming, ti_inc);
	54	rds_tcp_inc_purge(inc);
	55	rdsdebug("freeing tinc %p inc %p\n", tinc, inc);
	56	kmem_cache_free(rds_tcp_incoming_slab, tinc);
	57	}
	58
	59	/*
	60	* this is pretty lame, but, whatever.
	61	*/
	62	int rds_tcp_inc_copy_to_user(struct rds_incoming inc, struct iovec first_iov,
	63	size_t size)
	64	{
	65	struct rds_tcp_incoming *tinc;
	66	struct iovec *iov, tmp;
	67	struct sk_buff *skb;
	68	unsigned long to_copy, skb_off;
	69	int ret = 0;
	70
	71	if (size == 0)
	72	goto out;
	73
	74	tinc = container_of(inc, struct rds_tcp_incoming, ti_inc);
	75	iov = first_iov;
	76	tmp = *iov;
	77
	78	skb_queue_walk(&tinc->ti_skb_list, skb) {
	79	skb_off = 0;
	80	while (skb_off < skb->len) {
	81	while (tmp.iov_len == 0) {
	82	iov++;
	83	tmp = *iov;
	84	}
	85
	86	to_copy = min(tmp.iov_len, size);
	87	to_copy = min(to_copy, skb->len - skb_off);
	88
	89	rdsdebug("ret %d size %zu skb %p skb_off %lu "
	90	"skblen %d iov_base %p iov_len %zu cpy %lu\n",
	91	ret, size, skb, skb_off, skb->len,
	92	tmp.iov_base, tmp.iov_len, to_copy);
	93
	94	/* modifies tmp as it copies */
	95	if (skb_copy_datagram_iovec(skb, skb_off, &tmp,
	96	to_copy)) {
	97	ret = -EFAULT;
	98	goto out;
	99	}
	100
b075cfdb	101	rds_stats_add(s_copy_to_user, to_copy);
70041088 AG	102	size -= to_copy;
	103	ret += to_copy;
	104	skb_off += to_copy;
	105	if (size == 0)
	106	goto out;
	107	}
	108	}
	109	out:
	110	return ret;
	111	}
	112
	113	/*
	114	* We have a series of skbs that have fragmented pieces of the congestion
	115	* bitmap. They must add up to the exact size of the congestion bitmap. We
	116	* use the skb helpers to copy those into the pages that make up the in-memory
	117	* congestion bitmap for the remote address of this connection. We then tell
	118	* the congestion core that the bitmap has been changed so that it can wake up
	119	* sleepers.
	120	*
	121	* This is racing with sending paths which are using test_bit to see if the
	122	* bitmap indicates that their recipient is congested.
	123	*/
	124
	125	static void rds_tcp_cong_recv(struct rds_connection *conn,
	126	struct rds_tcp_incoming *tinc)
	127	{
	128	struct sk_buff *skb;
	129	unsigned int to_copy, skb_off;
	130	unsigned int map_off;
	131	unsigned int map_page;
	132	struct rds_cong_map *map;
	133	int ret;
	134
	135	/* catch completely corrupt packets */
	136	if (be32_to_cpu(tinc->ti_inc.i_hdr.h_len) != RDS_CONG_MAP_BYTES)
	137	return;
	138
	139	map_page = 0;
	140	map_off = 0;
	141	map = conn->c_fcong;
	142
	143	skb_queue_walk(&tinc->ti_skb_list, skb) {
	144	skb_off = 0;
	145	while (skb_off < skb->len) {
	146	to_copy = min_t(unsigned int, PAGE_SIZE - map_off,
	147	skb->len - skb_off);
	148
	149	BUG_ON(map_page >= RDS_CONG_MAP_PAGES);
	150
	151	/* only returns 0 or -error */
	152	ret = skb_copy_bits(skb, skb_off,
	153	(void *)map->m_page_addrs[map_page] + map_off,
	154	to_copy);
	155	BUG_ON(ret != 0);
	156
	157	skb_off += to_copy;
	158	map_off += to_copy;
	159	if (map_off == PAGE_SIZE) {
	160	map_off = 0;
	161	map_page++;
	162	}
	163	}
	164	}
	165
166	rds_cong_map_updated(map, ~(u64) 0);
167	}
168
169	struct rds_tcp_desc_arg {
170	struct rds_connection *conn;
171	gfp_t gfp;
70041088 AG	172	};
	173
	174	static int rds_tcp_data_recv(read_descriptor_t desc, struct sk_buff skb,
	175	unsigned int offset, size_t len)
	176	{
	177	struct rds_tcp_desc_arg *arg = desc->arg.data;
	178	struct rds_connection *conn = arg->conn;
	179	struct rds_tcp_connection *tc = conn->c_transport_data;
	180	struct rds_tcp_incoming *tinc = tc->t_tinc;
	181	struct sk_buff *clone;
	182	size_t left = len, to_copy;
	183
	184	rdsdebug("tcp data tc %p skb %p offset %u len %zu\n", tc, skb, offset,
	185	len);
	186
	187	/*
	188	* tcp_read_sock() interprets partial progress as an indication to stop
	189	* processing.
	190	*/
	191	while (left) {
8690bfa1	192	if (!tinc) {
70041088 AG	193	tinc = kmem_cache_alloc(rds_tcp_incoming_slab,
70041088 AG	194	arg->gfp);
8690bfa1	195	if (!tinc) {
70041088 AG	196	desc->error = -ENOMEM;
	197	goto out;
	198	}
	199	tc->t_tinc = tinc;
	200	rdsdebug("alloced tinc %p\n", tinc);
	201	rds_inc_init(&tinc->ti_inc, conn, conn->c_faddr);
	202	/*
	203	* XXX * we might be able to use the __ variants when
	204	* we've already serialized at a higher level.
	205	*/
	206	skb_queue_head_init(&tinc->ti_skb_list);
	207	}
	208
	209	if (left && tc->t_tinc_hdr_rem) {
	210	to_copy = min(tc->t_tinc_hdr_rem, left);
	211	rdsdebug("copying %zu header from skb %p\n", to_copy,
	212	skb);
	213	skb_copy_bits(skb, offset,
	214	(char *)&tinc->ti_inc.i_hdr +
	215	sizeof(struct rds_header) -
	216	tc->t_tinc_hdr_rem,
	217	to_copy);
	218	tc->t_tinc_hdr_rem -= to_copy;
	219	left -= to_copy;
	220	offset += to_copy;
	221
	222	if (tc->t_tinc_hdr_rem == 0) {
	223	/* could be 0 for a 0 len message */
	224	tc->t_tinc_data_rem =
	225	be32_to_cpu(tinc->ti_inc.i_hdr.h_len);
	226	}
	227	}
	228
	229	if (left && tc->t_tinc_data_rem) {
	230	clone = skb_clone(skb, arg->gfp);
8690bfa1	231	if (!clone) {
70041088 AG	232	desc->error = -ENOMEM;
	233	goto out;
	234	}
	235
	236	to_copy = min(tc->t_tinc_data_rem, left);
074043fd SV	237	if (!pskb_pull(clone, offset) \|\|
	238	pskb_trim(clone, to_copy)) {
	239	pr_warn("rds_tcp_data_recv: pull/trim failed "
	240	"left %zu data_rem %zu skb_len %d\n",
	241	left, tc->t_tinc_data_rem, skb->len);
	242	kfree_skb(clone);
	243	desc->error = -ENOMEM;
	244	goto out;
	245	}
70041088 AG	246	skb_queue_tail(&tinc->ti_skb_list, clone);
	247
	248	rdsdebug("skb %p data %p len %d off %u to_copy %zu -> "
	249	"clone %p data %p len %d\n",
	250	skb, skb->data, skb->len, offset, to_copy,
	251	clone, clone->data, clone->len);
	252
	253	tc->t_tinc_data_rem -= to_copy;
	254	left -= to_copy;
	255	offset += to_copy;
	256	}
	257
	258	if (tc->t_tinc_hdr_rem == 0 && tc->t_tinc_data_rem == 0) {
	259	if (tinc->ti_inc.i_hdr.h_flags == RDS_FLAG_CONG_BITMAP)
	260	rds_tcp_cong_recv(conn, tinc);
	261	else
	262	rds_recv_incoming(conn, conn->c_faddr,
	263	conn->c_laddr, &tinc->ti_inc,
6114eab5	264	arg->gfp);
70041088 AG	265
	266	tc->t_tinc_hdr_rem = sizeof(struct rds_header);
	267	tc->t_tinc_data_rem = 0;
	268	tc->t_tinc = NULL;
	269	rds_inc_put(&tinc->ti_inc);
	270	tinc = NULL;
	271	}
	272	}
	273	out:
	274	rdsdebug("returning len %zu left %zu skb len %d rx queue depth %d\n",
	275	len, left, skb->len,
	276	skb_queue_len(&tc->t_sock->sk->sk_receive_queue));
	277	return len - left;
	278	}
	279
	280	/* the caller has to hold the sock lock */
6114eab5	281	static int rds_tcp_read_sock(struct rds_connection *conn, gfp_t gfp)
70041088 AG	282	{
	283	struct rds_tcp_connection *tc = conn->c_transport_data;
	284	struct socket *sock = tc->t_sock;
	285	read_descriptor_t desc;
	286	struct rds_tcp_desc_arg arg;
	287
	288	/* It's like glib in the kernel! */
	289	arg.conn = conn;
	290	arg.gfp = gfp;
70041088 AG	291	desc.arg.data = &arg;
	292	desc.error = 0;
	293	desc.count = 1; /* give more than one skb per call */
	294
	295	tcp_read_sock(sock->sk, &desc, rds_tcp_data_recv);
	296	rdsdebug("tcp_read_sock for tc %p gfp 0x%x returned %d\n", tc, gfp,
	297	desc.error);
	298
	299	return desc.error;
	300	}
	301
	302	/*
	303	* We hold the sock lock to serialize our rds_tcp_recv->tcp_read_sock from
	304	* data_ready.
	305	*
	306	* if we fail to allocate we're in trouble.. blindly wait some time before
	307	* trying again to see if the VM can free up something for us.
	308	*/
	309	int rds_tcp_recv(struct rds_connection *conn)
	310	{
	311	struct rds_tcp_connection *tc = conn->c_transport_data;
	312	struct socket *sock = tc->t_sock;
	313	int ret = 0;
	314
	315	rdsdebug("recv worker conn %p tc %p sock %p\n", conn, tc, sock);
	316
	317	lock_sock(sock->sk);
6114eab5	318	ret = rds_tcp_read_sock(conn, GFP_KERNEL);
70041088 AG	319	release_sock(sock->sk);
	320
	321	return ret;
	322	}
	323
	324	void rds_tcp_data_ready(struct sock *sk, int bytes)
	325	{
	326	void (ready)(struct sock sk, int bytes);
	327	struct rds_connection *conn;
	328	struct rds_tcp_connection *tc;
	329
	330	rdsdebug("data ready sk %p bytes %d\n", sk, bytes);
	331
bfdc587c	332	read_lock(&sk->sk_callback_lock);
70041088	333	conn = sk->sk_user_data;
8690bfa1	334	if (!conn) { /* check for teardown race */
70041088 AG	335	ready = sk->sk_data_ready;
	336	goto out;
	337	}
	338
	339	tc = conn->c_transport_data;
	340	ready = tc->t_orig_data_ready;
	341	rds_tcp_stats_inc(s_tcp_data_ready_calls);
	342
6114eab5	343	if (rds_tcp_read_sock(conn, GFP_ATOMIC) == -ENOMEM)
70041088 AG	344	queue_delayed_work(rds_wq, &conn->c_recv_w, 0);
70041088 AG	345	out:
bfdc587c	346	read_unlock(&sk->sk_callback_lock);
70041088 AG	347	ready(sk, bytes);
	348	}
	349
ef87b7ea	350	int rds_tcp_recv_init(void)
70041088 AG	351	{
	352	rds_tcp_incoming_slab = kmem_cache_create("rds_tcp_incoming",
	353	sizeof(struct rds_tcp_incoming),
	354	0, 0, NULL);
8690bfa1	355	if (!rds_tcp_incoming_slab)
70041088 AG	356	return -ENOMEM;
	357	return 0;
	358	}
	359
	360	void rds_tcp_recv_exit(void)
	361	{
	362	kmem_cache_destroy(rds_tcp_incoming_slab);
	363	}