[mirror_ubuntu-bionic-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 "rds.h"

static DECLARE_WAIT_QUEUE_HEAD(rds_message_flush_waitq);

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),
};


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

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

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

	for (i = 0; i < rm->data.m_nents; i++) {
		rdsdebug("putting data page %p\n", (void *)sg_page(&rm->data.m_sg[i]));
		/* XXX will have to put_page for page refs */
		__free_page(sg_page(&rm->data.m_sg[i]));
	}
	rm->data.m_nents = 0;

	if (rm->rdma.m_rdma_op.r_active)
		rds_rdma_free_op(&rm->rdma.m_rdma_op);
	if (rm->rdma.m_rdma_mr)
		rds_mr_put(rm->rdma.m_rdma_mr);
}

void rds_message_put(struct rds_message *rm)
{
	rdsdebug("put rm %p ref %d\n", rm, atomic_read(&rm->m_refcount));

	if (atomic_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_version_extension(struct rds_header *hdr, unsigned int version)
{
	struct rds_ext_header_version ext_hdr;

	ext_hdr.h_version = cpu_to_be32(version);
	return rds_message_add_extension(hdr, RDS_EXTHDR_VERSION, &ext_hdr, sizeof(ext_hdr));
}

int rds_message_get_version_extension(struct rds_header *hdr, unsigned int *version)
{
	struct rds_ext_header_version ext_hdr;
	unsigned int pos = 0, len = sizeof(ext_hdr);

	/* We assume the version extension is the only one present */
	if (rds_message_next_extension(hdr, &pos, &ext_hdr, &len) != RDS_EXTHDR_VERSION)
		return 0;
	*version = be32_to_cpu(ext_hdr.h_version);
	return 1;
}

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;

	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);

	atomic_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);

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);

	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_KERNEL);
	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.m_nents = ceil(total_len, PAGE_SIZE);
	rm->data.m_sg = rds_message_alloc_sgs(rm, num_sgs);

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

	return rm;
}

int rds_message_copy_from_user(struct rds_message *rm, struct iovec *first_iov,
					       size_t total_len)
{
	unsigned long to_copy;
	unsigned long iov_off;
	unsigned long sg_off;
	struct iovec *iov;
	struct scatterlist *sg;
	int ret = 0;

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

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

	while (total_len) {
		if (!sg_page(sg)) {
			ret = rds_page_remainder_alloc(sg, total_len,
						       GFP_HIGHUSER);
			if (ret)
				goto out;
			rm->data.m_nents++;
			sg_off = 0;
		}

		while (iov_off == iov->iov_len) {
			iov_off = 0;
			iov++;
		}

		to_copy = min(iov->iov_len - iov_off, sg->length - sg_off);
		to_copy = min_t(size_t, to_copy, total_len);

		rdsdebug("copying %lu bytes from user iov [%p, %zu] + %lu to "
			 "sg [%p, %u, %u] + %lu\n",
			 to_copy, iov->iov_base, iov->iov_len, iov_off,
			 (void *)sg_page(sg), sg->offset, sg->length, sg_off);

		ret = rds_page_copy_from_user(sg_page(sg), sg->offset + sg_off,
					      iov->iov_base + iov_off,
					      to_copy);
		if (ret)
			goto out;

		iov_off += to_copy;
		total_len -= to_copy;
		sg_off += to_copy;

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

	rm->data.op_active = 1;

out:
	return ret;
}

int rds_message_inc_copy_to_user(struct rds_incoming *inc,
				 struct iovec *first_iov, size_t size)
{
	struct rds_message *rm;
	struct iovec *iov;
	struct scatterlist *sg;
	unsigned long to_copy;
	unsigned long iov_off;
	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);

	iov = first_iov;
	iov_off = 0;
	sg = rm->data.m_sg;
	vec_off = 0;
	copied = 0;

	while (copied < size && copied < len) {
		while (iov_off == iov->iov_len) {
			iov_off = 0;
			iov++;
		}

		to_copy = min(iov->iov_len - iov_off, sg->length - vec_off);
		to_copy = min_t(size_t, to_copy, size - copied);
		to_copy = min_t(unsigned long, to_copy, len - copied);

		rdsdebug("copying %lu bytes to user iov [%p, %zu] + %lu to "
			 "sg [%p, %u, %u] + %lu\n",
			 to_copy, iov->iov_base, iov->iov_len, iov_off,
			 sg_page(sg), sg->offset, sg->length, vec_off);

		ret = rds_page_copy_to_user(sg_page(sg), sg->offset + vec_off,
					    iov->iov_base + iov_off,
					    to_copy);
		if (ret) {
			copied = ret;
			break;
		}

		iov_off += to_copy;
		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(rds_message_flush_waitq,
			!test_bit(RDS_MSG_MAPPED, &rm->m_flags));
}

void rds_message_unmapped(struct rds_message *rm)
{
	clear_bit(RDS_MSG_MAPPED, &rm->m_flags);
	if (waitqueue_active(&rds_message_flush_waitq))
		wake_up(&rds_message_flush_waitq);
}
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>
7875e18e AG	35
7875e18e AG	36	#include "rds.h"
7875e18e AG	37
	38	static DECLARE_WAIT_QUEUE_HEAD(rds_message_flush_waitq);
	39
	40	static unsigned int rds_exthdr_size[__RDS_EXTHDR_MAX] = {
	41	[RDS_EXTHDR_NONE] = 0,
	42	[RDS_EXTHDR_VERSION] = sizeof(struct rds_ext_header_version),
	43	[RDS_EXTHDR_RDMA] = sizeof(struct rds_ext_header_rdma),
	44	[RDS_EXTHDR_RDMA_DEST] = sizeof(struct rds_ext_header_rdma_dest),
	45	};
	46
	47
	48	void rds_message_addref(struct rds_message *rm)
	49	{
	50	rdsdebug("addref rm %p ref %d\n", rm, atomic_read(&rm->m_refcount));
	51	atomic_inc(&rm->m_refcount);
	52	}
616b757a	53	EXPORT_SYMBOL_GPL(rds_message_addref);
7875e18e AG	54
	55	/*
	56	* This relies on dma_map_sg() not touching sg[].page during merging.
	57	*/
	58	static void rds_message_purge(struct rds_message *rm)
	59	{
	60	unsigned long i;
	61
	62	if (unlikely(test_bit(RDS_MSG_PAGEVEC, &rm->m_flags)))
	63	return;
	64
e779137a AG	65	for (i = 0; i < rm->data.m_nents; i++) {
e779137a AG	66	rdsdebug("putting data page %p\n", (void *)sg_page(&rm->data.m_sg[i]));
7875e18e	67	/* XXX will have to put_page for page refs */
e779137a	68	__free_page(sg_page(&rm->data.m_sg[i]));
7875e18e	69	}
e779137a	70	rm->data.m_nents = 0;
7875e18e	71
ff87e97a AG	72	if (rm->rdma.m_rdma_op.r_active)
ff87e97a AG	73	rds_rdma_free_op(&rm->rdma.m_rdma_op);
e779137a AG	74	if (rm->rdma.m_rdma_mr)
e779137a AG	75	rds_mr_put(rm->rdma.m_rdma_mr);
7875e18e	76	}
7875e18e AG	77
	78	void rds_message_put(struct rds_message *rm)
	79	{
	80	rdsdebug("put rm %p ref %d\n", rm, atomic_read(&rm->m_refcount));
	81
	82	if (atomic_dec_and_test(&rm->m_refcount)) {
	83	BUG_ON(!list_empty(&rm->m_sock_item));
	84	BUG_ON(!list_empty(&rm->m_conn_item));
	85	rds_message_purge(rm);
	86
	87	kfree(rm);
	88	}
	89	}
616b757a	90	EXPORT_SYMBOL_GPL(rds_message_put);
7875e18e	91
7875e18e AG	92	void rds_message_populate_header(struct rds_header *hdr, __be16 sport,
	93	__be16 dport, u64 seq)
	94	{
	95	hdr->h_flags = 0;
	96	hdr->h_sport = sport;
	97	hdr->h_dport = dport;
	98	hdr->h_sequence = cpu_to_be64(seq);
	99	hdr->h_exthdr[0] = RDS_EXTHDR_NONE;
	100	}
616b757a	101	EXPORT_SYMBOL_GPL(rds_message_populate_header);
7875e18e AG	102
	103	int rds_message_add_extension(struct rds_header *hdr,
	104	unsigned int type, const void *data, unsigned int len)
	105	{
	106	unsigned int ext_len = sizeof(u8) + len;
	107	unsigned char *dst;
	108
	109	/* For now, refuse to add more than one extension header */
	110	if (hdr->h_exthdr[0] != RDS_EXTHDR_NONE)
	111	return 0;
	112
f64f9e71	113	if (type >= __RDS_EXTHDR_MAX \|\| len != rds_exthdr_size[type])
7875e18e AG	114	return 0;
	115
	116	if (ext_len >= RDS_HEADER_EXT_SPACE)
	117	return 0;
	118	dst = hdr->h_exthdr;
	119
	120	*dst++ = type;
	121	memcpy(dst, data, len);
	122
	123	dst[len] = RDS_EXTHDR_NONE;
	124	return 1;
	125	}
616b757a	126	EXPORT_SYMBOL_GPL(rds_message_add_extension);
7875e18e AG	127
	128	/*
	129	* If a message has extension headers, retrieve them here.
	130	* Call like this:
	131	*
	132	* unsigned int pos = 0;
	133	*
	134	* while (1) {
	135	* buflen = sizeof(buffer);
	136	* type = rds_message_next_extension(hdr, &pos, buffer, &buflen);
	137	* if (type == RDS_EXTHDR_NONE)
	138	* break;
	139	* ...
	140	* }
	141	*/
	142	int rds_message_next_extension(struct rds_header *hdr,
	143	unsigned int pos, void buf, unsigned int *buflen)
	144	{
	145	unsigned int offset, ext_type, ext_len;
	146	u8 *src = hdr->h_exthdr;
	147
	148	offset = *pos;
	149	if (offset >= RDS_HEADER_EXT_SPACE)
	150	goto none;
	151
	152	/* Get the extension type and length. For now, the
	153	* length is implied by the extension type. */
	154	ext_type = src[offset++];
	155
	156	if (ext_type == RDS_EXTHDR_NONE \|\| ext_type >= __RDS_EXTHDR_MAX)
	157	goto none;
	158	ext_len = rds_exthdr_size[ext_type];
	159	if (offset + ext_len > RDS_HEADER_EXT_SPACE)
	160	goto none;
	161
	162	*pos = offset + ext_len;
	163	if (ext_len < *buflen)
	164	*buflen = ext_len;
	165	memcpy(buf, src + offset, *buflen);
	166	return ext_type;
	167
	168	none:
	169	*pos = RDS_HEADER_EXT_SPACE;
	170	*buflen = 0;
	171	return RDS_EXTHDR_NONE;
	172	}
	173
	174	int rds_message_add_version_extension(struct rds_header *hdr, unsigned int version)
	175	{
	176	struct rds_ext_header_version ext_hdr;
	177
	178	ext_hdr.h_version = cpu_to_be32(version);
	179	return rds_message_add_extension(hdr, RDS_EXTHDR_VERSION, &ext_hdr, sizeof(ext_hdr));
	180	}
	181
	182	int rds_message_get_version_extension(struct rds_header hdr, unsigned int version)
	183	{
	184	struct rds_ext_header_version ext_hdr;
	185	unsigned int pos = 0, len = sizeof(ext_hdr);
	186
	187	/* We assume the version extension is the only one present */
	188	if (rds_message_next_extension(hdr, &pos, &ext_hdr, &len) != RDS_EXTHDR_VERSION)
	189	return 0;
	190	*version = be32_to_cpu(ext_hdr.h_version);
191	return 1;
192	}
193
194	int rds_message_add_rdma_dest_extension(struct rds_header *hdr, u32 r_key, u32 offset)
195	{
196	struct rds_ext_header_rdma_dest ext_hdr;
197
198	ext_hdr.h_rdma_rkey = cpu_to_be32(r_key);
199	ext_hdr.h_rdma_offset = cpu_to_be32(offset);
200	return rds_message_add_extension(hdr, RDS_EXTHDR_RDMA_DEST, &ext_hdr, sizeof(ext_hdr));
201	}
616b757a	202	EXPORT_SYMBOL_GPL(rds_message_add_rdma_dest_extension);
7875e18e	203
fc445084 AG	204	/*
	205	* Each rds_message is allocated with extra space for the scatterlist entries
	206	* rds ops will need. This is to minimize memory allocation count. Then, each rds op
	207	* can grab SGs when initializing its part of the rds_message.
	208	*/
	209	struct rds_message *rds_message_alloc(unsigned int extra_len, gfp_t gfp)
7875e18e AG	210	{
	211	struct rds_message *rm;
	212
fc445084	213	rm = kzalloc(sizeof(struct rds_message) + extra_len, gfp);
7875e18e AG	214	if (!rm)
	215	goto out;
	216
fc445084 AG	217	rm->m_used_sgs = 0;
	218	rm->m_total_sgs = extra_len / sizeof(struct scatterlist);
	219
7875e18e AG	220	atomic_set(&rm->m_refcount, 1);
	221	INIT_LIST_HEAD(&rm->m_sock_item);
	222	INIT_LIST_HEAD(&rm->m_conn_item);
	223	spin_lock_init(&rm->m_rs_lock);
	224
	225	out:
	226	return rm;
	227	}
	228
fc445084 AG	229	/*
	230	* RDS ops use this to grab SG entries from the rm's sg pool.
	231	*/
	232	struct scatterlist rds_message_alloc_sgs(struct rds_message rm, int nents)
	233	{
	234	struct scatterlist sg_first = (struct scatterlist ) &rm[1];
	235	struct scatterlist *sg_ret;
	236
	237	WARN_ON(rm->m_used_sgs + nents > rm->m_total_sgs);
	238
	239	sg_ret = &sg_first[rm->m_used_sgs];
f4dd96f7	240	sg_init_table(sg_ret, nents);
fc445084 AG	241	rm->m_used_sgs += nents;
	242
	243	return sg_ret;
	244	}
	245
7875e18e AG	246	struct rds_message rds_message_map_pages(unsigned long page_addrs, unsigned int total_len)
	247	{
	248	struct rds_message *rm;
	249	unsigned int i;
ff87e97a AG	250	int num_sgs = ceil(total_len, PAGE_SIZE);
ff87e97a AG	251	int extra_bytes = num_sgs * sizeof(struct scatterlist);
7875e18e	252
ff87e97a	253	rm = rds_message_alloc(extra_bytes, GFP_KERNEL);
8690bfa1	254	if (!rm)
7875e18e AG	255	return ERR_PTR(-ENOMEM);
	256
	257	set_bit(RDS_MSG_PAGEVEC, &rm->m_flags);
	258	rm->m_inc.i_hdr.h_len = cpu_to_be32(total_len);
e779137a	259	rm->data.m_nents = ceil(total_len, PAGE_SIZE);
ff87e97a	260	rm->data.m_sg = rds_message_alloc_sgs(rm, num_sgs);
7875e18e	261
e779137a AG	262	for (i = 0; i < rm->data.m_nents; ++i) {
e779137a AG	263	sg_set_page(&rm->data.m_sg[i],
7875e18e AG	264	virt_to_page(page_addrs[i]),
	265	PAGE_SIZE, 0);
	266	}
	267
	268	return rm;
	269	}
	270
fc445084	271	int rds_message_copy_from_user(struct rds_message rm, struct iovec first_iov,
7875e18e AG	272	size_t total_len)
	273	{
	274	unsigned long to_copy;
	275	unsigned long iov_off;
	276	unsigned long sg_off;
7875e18e AG	277	struct iovec *iov;
7875e18e AG	278	struct scatterlist *sg;
fc445084	279	int ret = 0;
7875e18e AG	280
	281	rm->m_inc.i_hdr.h_len = cpu_to_be32(total_len);
	282
	283	/*
	284	* now allocate and copy in the data payload.
	285	*/
e779137a	286	sg = rm->data.m_sg;
7875e18e AG	287	iov = first_iov;
	288	iov_off = 0;
	289	sg_off = 0; /* Dear gcc, sg->page will be null from kzalloc. */
	290
	291	while (total_len) {
8690bfa1	292	if (!sg_page(sg)) {
7875e18e AG	293	ret = rds_page_remainder_alloc(sg, total_len,
	294	GFP_HIGHUSER);
	295	if (ret)
	296	goto out;
e779137a	297	rm->data.m_nents++;
7875e18e AG	298	sg_off = 0;
	299	}
	300
	301	while (iov_off == iov->iov_len) {
	302	iov_off = 0;
	303	iov++;
	304	}
	305
	306	to_copy = min(iov->iov_len - iov_off, sg->length - sg_off);
	307	to_copy = min_t(size_t, to_copy, total_len);
	308
	309	rdsdebug("copying %lu bytes from user iov [%p, %zu] + %lu to "
	310	"sg [%p, %u, %u] + %lu\n",
	311	to_copy, iov->iov_base, iov->iov_len, iov_off,
	312	(void *)sg_page(sg), sg->offset, sg->length, sg_off);
	313
	314	ret = rds_page_copy_from_user(sg_page(sg), sg->offset + sg_off,
	315	iov->iov_base + iov_off,
	316	to_copy);
	317	if (ret)
	318	goto out;
	319
	320	iov_off += to_copy;
	321	total_len -= to_copy;
	322	sg_off += to_copy;
	323
	324	if (sg_off == sg->length)
	325	sg++;
	326	}
	327
241eef3e AG	328	rm->data.op_active = 1;
241eef3e AG	329
7875e18e	330	out:
fc445084	331	return ret;
7875e18e AG	332	}
	333
	334	int rds_message_inc_copy_to_user(struct rds_incoming *inc,
	335	struct iovec *first_iov, size_t size)
	336	{
	337	struct rds_message *rm;
	338	struct iovec *iov;
	339	struct scatterlist *sg;
	340	unsigned long to_copy;
	341	unsigned long iov_off;
	342	unsigned long vec_off;
	343	int copied;
	344	int ret;
	345	u32 len;
	346
	347	rm = container_of(inc, struct rds_message, m_inc);
	348	len = be32_to_cpu(rm->m_inc.i_hdr.h_len);
	349
	350	iov = first_iov;
	351	iov_off = 0;
e779137a	352	sg = rm->data.m_sg;
7875e18e AG	353	vec_off = 0;
	354	copied = 0;
	355
	356	while (copied < size && copied < len) {
	357	while (iov_off == iov->iov_len) {
	358	iov_off = 0;
	359	iov++;
	360	}
	361
	362	to_copy = min(iov->iov_len - iov_off, sg->length - vec_off);
	363	to_copy = min_t(size_t, to_copy, size - copied);
	364	to_copy = min_t(unsigned long, to_copy, len - copied);
	365
	366	rdsdebug("copying %lu bytes to user iov [%p, %zu] + %lu to "
	367	"sg [%p, %u, %u] + %lu\n",
	368	to_copy, iov->iov_base, iov->iov_len, iov_off,
	369	sg_page(sg), sg->offset, sg->length, vec_off);
	370
	371	ret = rds_page_copy_to_user(sg_page(sg), sg->offset + vec_off,
	372	iov->iov_base + iov_off,
	373	to_copy);
	374	if (ret) {
	375	copied = ret;
	376	break;
	377	}
	378
	379	iov_off += to_copy;
	380	vec_off += to_copy;
	381	copied += to_copy;
	382
	383	if (vec_off == sg->length) {
	384	vec_off = 0;
	385	sg++;
	386	}
	387	}
	388
	389	return copied;
	390	}
	391
	392	/*
	393	* If the message is still on the send queue, wait until the transport
	394	* is done with it. This is particularly important for RDMA operations.
	395	*/
	396	void rds_message_wait(struct rds_message *rm)
	397	{
	398	wait_event(rds_message_flush_waitq,
	399	!test_bit(RDS_MSG_MAPPED, &rm->m_flags));
	400	}
	401
	402	void rds_message_unmapped(struct rds_message *rm)
	403	{
	404	clear_bit(RDS_MSG_MAPPED, &rm->m_flags);
	405	if (waitqueue_active(&rds_message_flush_waitq))
	406	wake_up(&rds_message_flush_waitq);
	407	}
616b757a	408	EXPORT_SYMBOL_GPL(rds_message_unmapped);
7875e18e	409