[mirror_ubuntu-jammy-kernel.git] / net / rds / message.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 <linux/export.h>
#include <linux/skbuff.h>
#include <linux/list.h>
#include <linux/errqueue.h>

#include "rds.h"

static unsigned int	rds_exthdr_size[__RDS_EXTHDR_MAX] = {
[RDS_EXTHDR_NONE]	= 0,
[RDS_EXTHDR_VERSION]	= sizeof(struct rds_ext_header_version),
[RDS_EXTHDR_RDMA]	= sizeof(struct rds_ext_header_rdma),
[RDS_EXTHDR_RDMA_DEST]	= sizeof(struct rds_ext_header_rdma_dest),
[RDS_EXTHDR_NPATHS]	= sizeof(u16),
[RDS_EXTHDR_GEN_NUM]	= sizeof(u32),
};


void rds_message_addref(struct rds_message *rm)
{
	rdsdebug("addref rm %p ref %d\n", rm, refcount_read(&rm->m_refcount));
	refcount_inc(&rm->m_refcount);
}
EXPORT_SYMBOL_GPL(rds_message_addref);

static inline bool skb_zcookie_add(struct sk_buff *skb, u32 cookie)
{
	struct rds_zcopy_cookies *ck = (struct rds_zcopy_cookies *)skb->cb;
	int ncookies = ck->num;

	if (ncookies == RDS_MAX_ZCOOKIES)
		return false;
	ck->cookies[ncookies] = cookie;
	ck->num =  ++ncookies;
	return true;
}

static void rds_rm_zerocopy_callback(struct rds_sock *rs,
				     struct rds_znotifier *znotif)
{
	struct sk_buff *skb, *tail;
	unsigned long flags;
	struct sk_buff_head *q;
	u32 cookie = znotif->z_cookie;
	struct rds_zcopy_cookies *ck;

	q = &rs->rs_zcookie_queue;
	spin_lock_irqsave(&q->lock, flags);
	tail = skb_peek_tail(q);

	if (tail && skb_zcookie_add(tail, cookie)) {
		spin_unlock_irqrestore(&q->lock, flags);
		mm_unaccount_pinned_pages(&znotif->z_mmp);
		consume_skb(rds_skb_from_znotifier(znotif));
		/* caller invokes rds_wake_sk_sleep() */
		return;
	}

	skb = rds_skb_from_znotifier(znotif);
	ck = (struct rds_zcopy_cookies *)skb->cb;
	memset(ck, 0, sizeof(*ck));
	WARN_ON(!skb_zcookie_add(skb, cookie));

	__skb_queue_tail(q, skb);

	spin_unlock_irqrestore(&q->lock, flags);
	/* caller invokes rds_wake_sk_sleep() */

	mm_unaccount_pinned_pages(&znotif->z_mmp);
}

/*
 * This relies on dma_map_sg() not touching sg[].page during merging.
 */
static void rds_message_purge(struct rds_message *rm)
{
	unsigned long i, flags;
	bool zcopy = false;

	if (unlikely(test_bit(RDS_MSG_PAGEVEC, &rm->m_flags)))
		return;

	spin_lock_irqsave(&rm->m_rs_lock, flags);
	if (rm->m_rs) {
		struct rds_sock *rs = rm->m_rs;

		if (rm->data.op_mmp_znotifier) {
			zcopy = true;
			rds_rm_zerocopy_callback(rs, rm->data.op_mmp_znotifier);
			rds_wake_sk_sleep(rs);
			rm->data.op_mmp_znotifier = NULL;
		}
		sock_put(rds_rs_to_sk(rs));
		rm->m_rs = NULL;
	}
	spin_unlock_irqrestore(&rm->m_rs_lock, flags);

	for (i = 0; i < rm->data.op_nents; i++) {
		/* XXX will have to put_page for page refs */
		if (!zcopy)
			__free_page(sg_page(&rm->data.op_sg[i]));
		else
			put_page(sg_page(&rm->data.op_sg[i]));
	}
	rm->data.op_nents = 0;

	if (rm->rdma.op_active)
		rds_rdma_free_op(&rm->rdma);
	if (rm->rdma.op_rdma_mr)
		rds_mr_put(rm->rdma.op_rdma_mr);

	if (rm->atomic.op_active)
		rds_atomic_free_op(&rm->atomic);
	if (rm->atomic.op_rdma_mr)
		rds_mr_put(rm->atomic.op_rdma_mr);
}

void rds_message_put(struct rds_message *rm)
{
	rdsdebug("put rm %p ref %d\n", rm, refcount_read(&rm->m_refcount));
	WARN(!refcount_read(&rm->m_refcount), "danger refcount zero on %p\n", rm);
	if (refcount_dec_and_test(&rm->m_refcount)) {
		BUG_ON(!list_empty(&rm->m_sock_item));
		BUG_ON(!list_empty(&rm->m_conn_item));
		rds_message_purge(rm);

		kfree(rm);
	}
}
EXPORT_SYMBOL_GPL(rds_message_put);

void rds_message_populate_header(struct rds_header *hdr, __be16 sport,
				 __be16 dport, u64 seq)
{
	hdr->h_flags = 0;
	hdr->h_sport = sport;
	hdr->h_dport = dport;
	hdr->h_sequence = cpu_to_be64(seq);
	hdr->h_exthdr[0] = RDS_EXTHDR_NONE;
}
EXPORT_SYMBOL_GPL(rds_message_populate_header);

int rds_message_add_extension(struct rds_header *hdr, unsigned int type,
			      const void *data, unsigned int len)
{
	unsigned int ext_len = sizeof(u8) + len;
	unsigned char *dst;

	/* For now, refuse to add more than one extension header */
	if (hdr->h_exthdr[0] != RDS_EXTHDR_NONE)
		return 0;

	if (type >= __RDS_EXTHDR_MAX || len != rds_exthdr_size[type])
		return 0;

	if (ext_len >= RDS_HEADER_EXT_SPACE)
		return 0;
	dst = hdr->h_exthdr;

	*dst++ = type;
	memcpy(dst, data, len);

	dst[len] = RDS_EXTHDR_NONE;
	return 1;
}
EXPORT_SYMBOL_GPL(rds_message_add_extension);

/*
 * If a message has extension headers, retrieve them here.
 * Call like this:
 *
 * unsigned int pos = 0;
 *
 * while (1) {
 *	buflen = sizeof(buffer);
 *	type = rds_message_next_extension(hdr, &pos, buffer, &buflen);
 *	if (type == RDS_EXTHDR_NONE)
 *		break;
 *	...
 * }
 */
int rds_message_next_extension(struct rds_header *hdr,
		unsigned int *pos, void *buf, unsigned int *buflen)
{
	unsigned int offset, ext_type, ext_len;
	u8 *src = hdr->h_exthdr;

	offset = *pos;
	if (offset >= RDS_HEADER_EXT_SPACE)
		goto none;

	/* Get the extension type and length. For now, the
	 * length is implied by the extension type. */
	ext_type = src[offset++];

	if (ext_type == RDS_EXTHDR_NONE || ext_type >= __RDS_EXTHDR_MAX)
		goto none;
	ext_len = rds_exthdr_size[ext_type];
	if (offset + ext_len > RDS_HEADER_EXT_SPACE)
		goto none;

	*pos = offset + ext_len;
	if (ext_len < *buflen)
		*buflen = ext_len;
	memcpy(buf, src + offset, *buflen);
	return ext_type;

none:
	*pos = RDS_HEADER_EXT_SPACE;
	*buflen = 0;
	return RDS_EXTHDR_NONE;
}

int rds_message_add_rdma_dest_extension(struct rds_header *hdr, u32 r_key, u32 offset)
{
	struct rds_ext_header_rdma_dest ext_hdr;

	ext_hdr.h_rdma_rkey = cpu_to_be32(r_key);
	ext_hdr.h_rdma_offset = cpu_to_be32(offset);
	return rds_message_add_extension(hdr, RDS_EXTHDR_RDMA_DEST, &ext_hdr, sizeof(ext_hdr));
}
EXPORT_SYMBOL_GPL(rds_message_add_rdma_dest_extension);

/*
 * Each rds_message is allocated with extra space for the scatterlist entries
 * rds ops will need. This is to minimize memory allocation count. Then, each rds op
 * can grab SGs when initializing its part of the rds_message.
 */
struct rds_message *rds_message_alloc(unsigned int extra_len, gfp_t gfp)
{
	struct rds_message *rm;

	if (extra_len > KMALLOC_MAX_SIZE - sizeof(struct rds_message))
		return NULL;

	rm = kzalloc(sizeof(struct rds_message) + extra_len, gfp);
	if (!rm)
		goto out;

	rm->m_used_sgs = 0;
	rm->m_total_sgs = extra_len / sizeof(struct scatterlist);

	refcount_set(&rm->m_refcount, 1);
	INIT_LIST_HEAD(&rm->m_sock_item);
	INIT_LIST_HEAD(&rm->m_conn_item);
	spin_lock_init(&rm->m_rs_lock);
	init_waitqueue_head(&rm->m_flush_wait);

out:
	return rm;
}

/*
 * RDS ops use this to grab SG entries from the rm's sg pool.
 */
struct scatterlist *rds_message_alloc_sgs(struct rds_message *rm, int nents)
{
	struct scatterlist *sg_first = (struct scatterlist *) &rm[1];
	struct scatterlist *sg_ret;

	WARN_ON(rm->m_used_sgs + nents > rm->m_total_sgs);
	WARN_ON(!nents);

	if (rm->m_used_sgs + nents > rm->m_total_sgs)
		return NULL;

	sg_ret = &sg_first[rm->m_used_sgs];
	sg_init_table(sg_ret, nents);
	rm->m_used_sgs += nents;

	return sg_ret;
}

struct rds_message *rds_message_map_pages(unsigned long *page_addrs, unsigned int total_len)
{
	struct rds_message *rm;
	unsigned int i;
	int num_sgs = ceil(total_len, PAGE_SIZE);
	int extra_bytes = num_sgs * sizeof(struct scatterlist);

	rm = rds_message_alloc(extra_bytes, GFP_NOWAIT);
	if (!rm)
		return ERR_PTR(-ENOMEM);

	set_bit(RDS_MSG_PAGEVEC, &rm->m_flags);
	rm->m_inc.i_hdr.h_len = cpu_to_be32(total_len);
	rm->data.op_nents = ceil(total_len, PAGE_SIZE);
	rm->data.op_sg = rds_message_alloc_sgs(rm, num_sgs);
	if (!rm->data.op_sg) {
		rds_message_put(rm);
		return ERR_PTR(-ENOMEM);
	}

	for (i = 0; i < rm->data.op_nents; ++i) {
		sg_set_page(&rm->data.op_sg[i],
				virt_to_page(page_addrs[i]),
				PAGE_SIZE, 0);
	}

	return rm;
}

int rds_message_zcopy_from_user(struct rds_message *rm, struct iov_iter *from)
{
	unsigned long sg_off;
	struct scatterlist *sg;
	int ret = 0;
	int length = iov_iter_count(from);
	int total_copied = 0;
	struct sk_buff *skb;

	rm->m_inc.i_hdr.h_len = cpu_to_be32(iov_iter_count(from));

	/*
	 * now allocate and copy in the data payload.
	 */
	sg = rm->data.op_sg;
	sg_off = 0; /* Dear gcc, sg->page will be null from kzalloc. */

	skb = alloc_skb(0, GFP_KERNEL);
	if (!skb)
		return -ENOMEM;
	BUILD_BUG_ON(sizeof(skb->cb) < max_t(int, sizeof(struct rds_znotifier),
					     sizeof(struct rds_zcopy_cookies)));
	rm->data.op_mmp_znotifier = RDS_ZCOPY_SKB(skb);
	if (mm_account_pinned_pages(&rm->data.op_mmp_znotifier->z_mmp,
				    length)) {
		ret = -ENOMEM;
		goto err;
	}
	while (iov_iter_count(from)) {
		struct page *pages;
		size_t start;
		ssize_t copied;

		copied = iov_iter_get_pages(from, &pages, PAGE_SIZE,
					    1, &start);
		if (copied < 0) {
			struct mmpin *mmp;
			int i;

			for (i = 0; i < rm->data.op_nents; i++)
				put_page(sg_page(&rm->data.op_sg[i]));
			mmp = &rm->data.op_mmp_znotifier->z_mmp;
			mm_unaccount_pinned_pages(mmp);
			ret = -EFAULT;
			goto err;
		}
		total_copied += copied;
		iov_iter_advance(from, copied);
		length -= copied;
		sg_set_page(sg, pages, copied, start);
		rm->data.op_nents++;
		sg++;
	}
	WARN_ON_ONCE(length != 0);
	return ret;
err:
	consume_skb(skb);
	rm->data.op_mmp_znotifier = NULL;
	return ret;
}

int rds_message_copy_from_user(struct rds_message *rm, struct iov_iter *from,
			       bool zcopy)
{
	unsigned long to_copy, nbytes;
	unsigned long sg_off;
	struct scatterlist *sg;
	int ret = 0;

	rm->m_inc.i_hdr.h_len = cpu_to_be32(iov_iter_count(from));

	/* now allocate and copy in the data payload.  */
	sg = rm->data.op_sg;
	sg_off = 0; /* Dear gcc, sg->page will be null from kzalloc. */

	if (zcopy)
		return rds_message_zcopy_from_user(rm, from);

	while (iov_iter_count(from)) {
		if (!sg_page(sg)) {
			ret = rds_page_remainder_alloc(sg, iov_iter_count(from),
						       GFP_HIGHUSER);
			if (ret)
				return ret;
			rm->data.op_nents++;
			sg_off = 0;
		}

		to_copy = min_t(unsigned long, iov_iter_count(from),
				sg->length - sg_off);

		rds_stats_add(s_copy_from_user, to_copy);
		nbytes = copy_page_from_iter(sg_page(sg), sg->offset + sg_off,
					     to_copy, from);
		if (nbytes != to_copy)
			return -EFAULT;

		sg_off += to_copy;

		if (sg_off == sg->length)
			sg++;
	}

	return ret;
}

int rds_message_inc_copy_to_user(struct rds_incoming *inc, struct iov_iter *to)
{
	struct rds_message *rm;
	struct scatterlist *sg;
	unsigned long to_copy;
	unsigned long vec_off;
	int copied;
	int ret;
	u32 len;

	rm = container_of(inc, struct rds_message, m_inc);
	len = be32_to_cpu(rm->m_inc.i_hdr.h_len);

	sg = rm->data.op_sg;
	vec_off = 0;
	copied = 0;

	while (iov_iter_count(to) && copied < len) {
		to_copy = min_t(unsigned long, iov_iter_count(to),
				sg->length - vec_off);
		to_copy = min_t(unsigned long, to_copy, len - copied);

		rds_stats_add(s_copy_to_user, to_copy);
		ret = copy_page_to_iter(sg_page(sg), sg->offset + vec_off,
					to_copy, to);
		if (ret != to_copy)
			return -EFAULT;

		vec_off += to_copy;
		copied += to_copy;

		if (vec_off == sg->length) {
			vec_off = 0;
			sg++;
		}
	}

	return copied;
}

/*
 * If the message is still on the send queue, wait until the transport
 * is done with it. This is particularly important for RDMA operations.
 */
void rds_message_wait(struct rds_message *rm)
{
	wait_event_interruptible(rm->m_flush_wait,
			!test_bit(RDS_MSG_MAPPED, &rm->m_flags));
}

void rds_message_unmapped(struct rds_message *rm)
{
	clear_bit(RDS_MSG_MAPPED, &rm->m_flags);
	wake_up_interruptible(&rm->m_flush_wait);
}
EXPORT_SYMBOL_GPL(rds_message_unmapped);
Commit	Line	Data
7875e18e 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>
bc3b2d7f	35	#include <linux/export.h>
01883eda SV	36	#include <linux/skbuff.h>
	37	#include <linux/list.h>
	38	#include <linux/errqueue.h>
7875e18e AG	39
7875e18e AG	40	#include "rds.h"
7875e18e	41
7875e18e AG	42	static unsigned int rds_exthdr_size[__RDS_EXTHDR_MAX] = {
	43	[RDS_EXTHDR_NONE] = 0,
	44	[RDS_EXTHDR_VERSION] = sizeof(struct rds_ext_header_version),
	45	[RDS_EXTHDR_RDMA] = sizeof(struct rds_ext_header_rdma),
	46	[RDS_EXTHDR_RDMA_DEST] = sizeof(struct rds_ext_header_rdma_dest),
5916e2c1	47	[RDS_EXTHDR_NPATHS] = sizeof(u16),
905dd418	48	[RDS_EXTHDR_GEN_NUM] = sizeof(u32),
7875e18e AG	49	};
	50
	51
	52	void rds_message_addref(struct rds_message *rm)
	53	{
6c5a1c4a RE	54	rdsdebug("addref rm %p ref %d\n", rm, refcount_read(&rm->m_refcount));
6c5a1c4a RE	55	refcount_inc(&rm->m_refcount);
7875e18e	56	}
616b757a	57	EXPORT_SYMBOL_GPL(rds_message_addref);
7875e18e	58
01883eda SV	59	static inline bool skb_zcookie_add(struct sk_buff *skb, u32 cookie)
01883eda SV	60	{
401910db SV	61	struct rds_zcopy_cookies ck = (struct rds_zcopy_cookies )skb->cb;
401910db SV	62	int ncookies = ck->num;
01883eda	63
401910db	64	if (ncookies == RDS_MAX_ZCOOKIES)
01883eda	65	return false;
401910db SV	66	ck->cookies[ncookies] = cookie;
401910db SV	67	ck->num = ++ncookies;
01883eda SV	68	return true;
	69	}
	70
	71	static void rds_rm_zerocopy_callback(struct rds_sock *rs,
	72	struct rds_znotifier *znotif)
	73	{
01883eda	74	struct sk_buff skb, tail;
01883eda SV	75	unsigned long flags;
	76	struct sk_buff_head *q;
	77	u32 cookie = znotif->z_cookie;
401910db	78	struct rds_zcopy_cookies *ck;
01883eda	79
401910db	80	q = &rs->rs_zcookie_queue;
01883eda SV	81	spin_lock_irqsave(&q->lock, flags);
	82	tail = skb_peek_tail(q);
	83
	84	if (tail && skb_zcookie_add(tail, cookie)) {
	85	spin_unlock_irqrestore(&q->lock, flags);
	86	mm_unaccount_pinned_pages(&znotif->z_mmp);
	87	consume_skb(rds_skb_from_znotifier(znotif));
401910db	88	/* caller invokes rds_wake_sk_sleep() */
01883eda SV	89	return;
	90	}
	91
	92	skb = rds_skb_from_znotifier(znotif);
401910db SV	93	ck = (struct rds_zcopy_cookies *)skb->cb;
401910db SV	94	memset(ck, 0, sizeof(*ck));
01883eda SV	95	WARN_ON(!skb_zcookie_add(skb, cookie));
	96
	97	__skb_queue_tail(q, skb);
	98
	99	spin_unlock_irqrestore(&q->lock, flags);
401910db	100	/* caller invokes rds_wake_sk_sleep() */
01883eda SV	101
	102	mm_unaccount_pinned_pages(&znotif->z_mmp);
	103	}
	104
7875e18e AG	105	/*
	106	* This relies on dma_map_sg() not touching sg[].page during merging.
	107	*/
	108	static void rds_message_purge(struct rds_message *rm)
	109	{
ea8994cb	110	unsigned long i, flags;
01883eda	111	bool zcopy = false;
7875e18e AG	112
	113	if (unlikely(test_bit(RDS_MSG_PAGEVEC, &rm->m_flags)))
	114	return;
	115
ea8994cb SV	116	spin_lock_irqsave(&rm->m_rs_lock, flags);
ea8994cb SV	117	if (rm->m_rs) {
01883eda SV	118	struct rds_sock *rs = rm->m_rs;
	119
	120	if (rm->data.op_mmp_znotifier) {
	121	zcopy = true;
	122	rds_rm_zerocopy_callback(rs, rm->data.op_mmp_znotifier);
401910db	123	rds_wake_sk_sleep(rs);
01883eda SV	124	rm->data.op_mmp_znotifier = NULL;
	125	}
	126	sock_put(rds_rs_to_sk(rs));
ea8994cb SV	127	rm->m_rs = NULL;
	128	}
	129	spin_unlock_irqrestore(&rm->m_rs_lock, flags);
7875e18e	130
01883eda SV	131	for (i = 0; i < rm->data.op_nents; i++) {
	132	/* XXX will have to put_page for page refs */
	133	if (!zcopy)
	134	__free_page(sg_page(&rm->data.op_sg[i]));
	135	else
	136	put_page(sg_page(&rm->data.op_sg[i]));
	137	}
	138	rm->data.op_nents = 0;
	139
f8b3aaf2 AG	140	if (rm->rdma.op_active)
	141	rds_rdma_free_op(&rm->rdma);
	142	if (rm->rdma.op_rdma_mr)
	143	rds_mr_put(rm->rdma.op_rdma_mr);
d0ab25a8 AG	144
	145	if (rm->atomic.op_active)
	146	rds_atomic_free_op(&rm->atomic);
	147	if (rm->atomic.op_rdma_mr)
	148	rds_mr_put(rm->atomic.op_rdma_mr);
7875e18e	149	}
7875e18e AG	150
	151	void rds_message_put(struct rds_message *rm)
	152	{
6c5a1c4a RE	153	rdsdebug("put rm %p ref %d\n", rm, refcount_read(&rm->m_refcount));
	154	WARN(!refcount_read(&rm->m_refcount), "danger refcount zero on %p\n", rm);
	155	if (refcount_dec_and_test(&rm->m_refcount)) {
7875e18e AG	156	BUG_ON(!list_empty(&rm->m_sock_item));
	157	BUG_ON(!list_empty(&rm->m_conn_item));
	158	rds_message_purge(rm);
	159
	160	kfree(rm);
	161	}
	162	}
616b757a	163	EXPORT_SYMBOL_GPL(rds_message_put);
7875e18e	164
7875e18e AG	165	void rds_message_populate_header(struct rds_header *hdr, __be16 sport,
	166	__be16 dport, u64 seq)
	167	{
	168	hdr->h_flags = 0;
	169	hdr->h_sport = sport;
	170	hdr->h_dport = dport;
	171	hdr->h_sequence = cpu_to_be64(seq);
	172	hdr->h_exthdr[0] = RDS_EXTHDR_NONE;
	173	}
616b757a	174	EXPORT_SYMBOL_GPL(rds_message_populate_header);
7875e18e	175
ff51bf84	176	int rds_message_add_extension(struct rds_header *hdr, unsigned int type,
ff51bf84	177	const void *data, unsigned int len)
7875e18e AG	178	{
	179	unsigned int ext_len = sizeof(u8) + len;
	180	unsigned char *dst;
	181
	182	/* For now, refuse to add more than one extension header */
	183	if (hdr->h_exthdr[0] != RDS_EXTHDR_NONE)
	184	return 0;
	185
f64f9e71	186	if (type >= __RDS_EXTHDR_MAX \|\| len != rds_exthdr_size[type])
7875e18e AG	187	return 0;
	188
	189	if (ext_len >= RDS_HEADER_EXT_SPACE)
	190	return 0;
	191	dst = hdr->h_exthdr;
	192
	193	*dst++ = type;
	194	memcpy(dst, data, len);
	195
	196	dst[len] = RDS_EXTHDR_NONE;
	197	return 1;
	198	}
616b757a	199	EXPORT_SYMBOL_GPL(rds_message_add_extension);
7875e18e AG	200
	201	/*
	202	* If a message has extension headers, retrieve them here.
	203	* Call like this:
	204	*
	205	* unsigned int pos = 0;
	206	*
	207	* while (1) {
	208	* buflen = sizeof(buffer);
	209	* type = rds_message_next_extension(hdr, &pos, buffer, &buflen);
	210	* if (type == RDS_EXTHDR_NONE)
	211	* break;
	212	* ...
	213	* }
	214	*/
	215	int rds_message_next_extension(struct rds_header *hdr,
	216	unsigned int pos, void buf, unsigned int *buflen)
	217	{
	218	unsigned int offset, ext_type, ext_len;
	219	u8 *src = hdr->h_exthdr;
	220
	221	offset = *pos;
	222	if (offset >= RDS_HEADER_EXT_SPACE)
	223	goto none;
	224
	225	/* Get the extension type and length. For now, the
	226	* length is implied by the extension type. */
	227	ext_type = src[offset++];
	228
	229	if (ext_type == RDS_EXTHDR_NONE \|\| ext_type >= __RDS_EXTHDR_MAX)
	230	goto none;
	231	ext_len = rds_exthdr_size[ext_type];
	232	if (offset + ext_len > RDS_HEADER_EXT_SPACE)
	233	goto none;
	234
	235	*pos = offset + ext_len;
	236	if (ext_len < *buflen)
	237	*buflen = ext_len;
	238	memcpy(buf, src + offset, *buflen);
	239	return ext_type;
	240
	241	none:
	242	*pos = RDS_HEADER_EXT_SPACE;
	243	*buflen = 0;
	244	return RDS_EXTHDR_NONE;
	245	}
	246
7875e18e AG	247	int rds_message_add_rdma_dest_extension(struct rds_header *hdr, u32 r_key, u32 offset)
	248	{
	249	struct rds_ext_header_rdma_dest ext_hdr;
	250
	251	ext_hdr.h_rdma_rkey = cpu_to_be32(r_key);
	252	ext_hdr.h_rdma_offset = cpu_to_be32(offset);
	253	return rds_message_add_extension(hdr, RDS_EXTHDR_RDMA_DEST, &ext_hdr, sizeof(ext_hdr));
	254	}
616b757a	255	EXPORT_SYMBOL_GPL(rds_message_add_rdma_dest_extension);
7875e18e	256
fc445084 AG	257	/*
	258	* Each rds_message is allocated with extra space for the scatterlist entries
	259	* rds ops will need. This is to minimize memory allocation count. Then, each rds op
	260	* can grab SGs when initializing its part of the rds_message.
	261	*/
	262	struct rds_message *rds_message_alloc(unsigned int extra_len, gfp_t gfp)
7875e18e AG	263	{
	264	struct rds_message *rm;
	265
ece6b0a2 CW	266	if (extra_len > KMALLOC_MAX_SIZE - sizeof(struct rds_message))
	267	return NULL;
	268
fc445084	269	rm = kzalloc(sizeof(struct rds_message) + extra_len, gfp);
7875e18e AG	270	if (!rm)
	271	goto out;
	272
fc445084 AG	273	rm->m_used_sgs = 0;
	274	rm->m_total_sgs = extra_len / sizeof(struct scatterlist);
	275
6c5a1c4a	276	refcount_set(&rm->m_refcount, 1);
7875e18e AG	277	INIT_LIST_HEAD(&rm->m_sock_item);
	278	INIT_LIST_HEAD(&rm->m_conn_item);
	279	spin_lock_init(&rm->m_rs_lock);
c83188dc	280	init_waitqueue_head(&rm->m_flush_wait);
7875e18e AG	281
	282	out:
	283	return rm;
	284	}
	285
fc445084 AG	286	/*
	287	* RDS ops use this to grab SG entries from the rm's sg pool.
	288	*/
	289	struct scatterlist rds_message_alloc_sgs(struct rds_message rm, int nents)
	290	{
	291	struct scatterlist sg_first = (struct scatterlist ) &rm[1];
	292	struct scatterlist *sg_ret;
	293
	294	WARN_ON(rm->m_used_sgs + nents > rm->m_total_sgs);
ee4c7b47	295	WARN_ON(!nents);
fc445084	296
d139ff09 AG	297	if (rm->m_used_sgs + nents > rm->m_total_sgs)
	298	return NULL;
	299
fc445084	300	sg_ret = &sg_first[rm->m_used_sgs];
f4dd96f7	301	sg_init_table(sg_ret, nents);
fc445084 AG	302	rm->m_used_sgs += nents;
	303
	304	return sg_ret;
	305	}
	306
7875e18e AG	307	struct rds_message rds_message_map_pages(unsigned long page_addrs, unsigned int total_len)
	308	{
	309	struct rds_message *rm;
	310	unsigned int i;
ff87e97a AG	311	int num_sgs = ceil(total_len, PAGE_SIZE);
ff87e97a AG	312	int extra_bytes = num_sgs * sizeof(struct scatterlist);
7875e18e	313
f2ec76f2	314	rm = rds_message_alloc(extra_bytes, GFP_NOWAIT);
8690bfa1	315	if (!rm)
7875e18e AG	316	return ERR_PTR(-ENOMEM);
	317
	318	set_bit(RDS_MSG_PAGEVEC, &rm->m_flags);
	319	rm->m_inc.i_hdr.h_len = cpu_to_be32(total_len);
6c7cc6e4 AG	320	rm->data.op_nents = ceil(total_len, PAGE_SIZE);
6c7cc6e4 AG	321	rm->data.op_sg = rds_message_alloc_sgs(rm, num_sgs);
aa58163a PE	322	if (!rm->data.op_sg) {
aa58163a PE	323	rds_message_put(rm);
d139ff09	324	return ERR_PTR(-ENOMEM);
aa58163a	325	}
7875e18e	326
6c7cc6e4 AG	327	for (i = 0; i < rm->data.op_nents; ++i) {
6c7cc6e4 AG	328	sg_set_page(&rm->data.op_sg[i],
7875e18e AG	329	virt_to_page(page_addrs[i]),
	330	PAGE_SIZE, 0);
	331	}
	332
	333	return rm;
	334	}
	335
d40a126b	336	int rds_message_zcopy_from_user(struct rds_message rm, struct iov_iter from)
7875e18e	337	{
7875e18e	338	unsigned long sg_off;
7875e18e	339	struct scatterlist *sg;
fc445084	340	int ret = 0;
0cebacce	341	int length = iov_iter_count(from);
d40a126b SV	342	int total_copied = 0;
d40a126b SV	343	struct sk_buff *skb;
7875e18e	344
083735f4	345	rm->m_inc.i_hdr.h_len = cpu_to_be32(iov_iter_count(from));
7875e18e AG	346
	347	/*
	348	* now allocate and copy in the data payload.
	349	*/
6c7cc6e4	350	sg = rm->data.op_sg;
7875e18e AG	351	sg_off = 0; /* Dear gcc, sg->page will be null from kzalloc. */
7875e18e AG	352
d40a126b SV	353	skb = alloc_skb(0, GFP_KERNEL);
	354	if (!skb)
	355	return -ENOMEM;
	356	BUILD_BUG_ON(sizeof(skb->cb) < max_t(int, sizeof(struct rds_znotifier),
	357	sizeof(struct rds_zcopy_cookies)));
	358	rm->data.op_mmp_znotifier = RDS_ZCOPY_SKB(skb);
	359	if (mm_account_pinned_pages(&rm->data.op_mmp_znotifier->z_mmp,
	360	length)) {
	361	ret = -ENOMEM;
	362	goto err;
	363	}
	364	while (iov_iter_count(from)) {
	365	struct page *pages;
	366	size_t start;
	367	ssize_t copied;
	368
	369	copied = iov_iter_get_pages(from, &pages, PAGE_SIZE,
	370	1, &start);
	371	if (copied < 0) {
	372	struct mmpin *mmp;
	373	int i;
	374
	375	for (i = 0; i < rm->data.op_nents; i++)
	376	put_page(sg_page(&rm->data.op_sg[i]));
	377	mmp = &rm->data.op_mmp_znotifier->z_mmp;
	378	mm_unaccount_pinned_pages(mmp);
	379	ret = -EFAULT;
0cebacce SV	380	goto err;
0cebacce SV	381	}
d40a126b SV	382	total_copied += copied;
	383	iov_iter_advance(from, copied);
	384	length -= copied;
	385	sg_set_page(sg, pages, copied, start);
	386	rm->data.op_nents++;
	387	sg++;
	388	}
	389	WARN_ON_ONCE(length != 0);
	390	return ret;
0cebacce	391	err:
d40a126b SV	392	consume_skb(skb);
	393	rm->data.op_mmp_znotifier = NULL;
	394	return ret;
	395	}
	396
	397	int rds_message_copy_from_user(struct rds_message rm, struct iov_iter from,
	398	bool zcopy)
	399	{
	400	unsigned long to_copy, nbytes;
	401	unsigned long sg_off;
	402	struct scatterlist *sg;
	403	int ret = 0;
	404
	405	rm->m_inc.i_hdr.h_len = cpu_to_be32(iov_iter_count(from));
	406
	407	/* now allocate and copy in the data payload. */
	408	sg = rm->data.op_sg;
	409	sg_off = 0; /* Dear gcc, sg->page will be null from kzalloc. */
	410
	411	if (zcopy)
	412	return rds_message_zcopy_from_user(rm, from);
0cebacce	413
083735f4	414	while (iov_iter_count(from)) {
8690bfa1	415	if (!sg_page(sg)) {
083735f4	416	ret = rds_page_remainder_alloc(sg, iov_iter_count(from),
7875e18e AG	417	GFP_HIGHUSER);
7875e18e AG	418	if (ret)
083735f4	419	return ret;
6c7cc6e4	420	rm->data.op_nents++;
7875e18e AG	421	sg_off = 0;
	422	}
	423
083735f4 AV	424	to_copy = min_t(unsigned long, iov_iter_count(from),
083735f4 AV	425	sg->length - sg_off);
7875e18e	426
083735f4	427	rds_stats_add(s_copy_from_user, to_copy);
d0a47d32 SV	428	nbytes = copy_page_from_iter(sg_page(sg), sg->offset + sg_off,
	429	to_copy, from);
	430	if (nbytes != to_copy)
083735f4	431	return -EFAULT;
7875e18e	432
7875e18e AG	433	sg_off += to_copy;
	434
	435	if (sg_off == sg->length)
	436	sg++;
	437	}
	438
fc445084	439	return ret;
7875e18e AG	440	}
7875e18e AG	441
c310e72c	442	int rds_message_inc_copy_to_user(struct rds_incoming inc, struct iov_iter to)
7875e18e AG	443	{
7875e18e AG	444	struct rds_message *rm;
7875e18e AG	445	struct scatterlist *sg;
7875e18e AG	446	unsigned long to_copy;
7875e18e AG	447	unsigned long vec_off;
	448	int copied;
	449	int ret;
	450	u32 len;
	451
	452	rm = container_of(inc, struct rds_message, m_inc);
	453	len = be32_to_cpu(rm->m_inc.i_hdr.h_len);
	454
6c7cc6e4	455	sg = rm->data.op_sg;
7875e18e AG	456	vec_off = 0;
	457	copied = 0;
	458
c310e72c	459	while (iov_iter_count(to) && copied < len) {
6ff4a8ad GU	460	to_copy = min_t(unsigned long, iov_iter_count(to),
6ff4a8ad GU	461	sg->length - vec_off);
7875e18e AG	462	to_copy = min_t(unsigned long, to_copy, len - copied);
7875e18e AG	463
c310e72c AV	464	rds_stats_add(s_copy_to_user, to_copy);
	465	ret = copy_page_to_iter(sg_page(sg), sg->offset + vec_off,
	466	to_copy, to);
	467	if (ret != to_copy)
	468	return -EFAULT;
7875e18e	469
7875e18e AG	470	vec_off += to_copy;
	471	copied += to_copy;
	472
	473	if (vec_off == sg->length) {
	474	vec_off = 0;
	475	sg++;
	476	}
	477	}
	478
	479	return copied;
	480	}
	481
	482	/*
	483	* If the message is still on the send queue, wait until the transport
	484	* is done with it. This is particularly important for RDMA operations.
	485	*/
	486	void rds_message_wait(struct rds_message *rm)
	487	{
c83188dc	488	wait_event_interruptible(rm->m_flush_wait,
7875e18e AG	489	!test_bit(RDS_MSG_MAPPED, &rm->m_flags));
	490	}
	491
	492	void rds_message_unmapped(struct rds_message *rm)
	493	{
	494	clear_bit(RDS_MSG_MAPPED, &rm->m_flags);
c83188dc	495	wake_up_interruptible(&rm->m_flush_wait);
7875e18e	496	}
616b757a	497	EXPORT_SYMBOL_GPL(rds_message_unmapped);
7875e18e	498