[linux-2.6-block.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 "rds.h"
#include "rdma.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->m_nents; i++) {
		rdsdebug("putting data page %p\n", (void *)sg_page(&rm->m_sg[i]));
		/* XXX will have to put_page for page refs */
		__free_page(sg_page(&rm->m_sg[i]));
	}
	rm->m_nents = 0;

	if (rm->m_rdma_op)
		rds_rdma_free_op(rm->m_rdma_op);
	if (rm->m_rdma_mr)
		rds_mr_put(rm->m_rdma_mr);
}

void rds_message_inc_purge(struct rds_incoming *inc)
{
	struct rds_message *rm = container_of(inc, struct rds_message, m_inc);
	rds_message_purge(rm);
}

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_inc_free(struct rds_incoming *inc)
{
	struct rds_message *rm = container_of(inc, struct rds_message, m_inc);
	rds_message_put(rm);
}

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);

struct rds_message *rds_message_alloc(unsigned int nents, gfp_t gfp)
{
	struct rds_message *rm;

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

	if (nents)
		sg_init_table(rm->m_sg, nents);
	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;
}

struct rds_message *rds_message_map_pages(unsigned long *page_addrs, unsigned int total_len)
{
	struct rds_message *rm;
	unsigned int i;

	rm = rds_message_alloc(ceil(total_len, PAGE_SIZE), GFP_KERNEL);
	if (rm == NULL)
		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->m_nents = ceil(total_len, PAGE_SIZE);

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

	return rm;
}

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

	rm = rds_message_alloc(ceil(total_len, PAGE_SIZE), GFP_KERNEL);
	if (rm == NULL) {
		ret = -ENOMEM;
		goto out;
	}

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

	/*
	 * now allocate and copy in the data payload.
	 */
	sg = rm->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) == NULL) {
			ret = rds_page_remainder_alloc(sg, total_len,
						       GFP_HIGHUSER);
			if (ret)
				goto out;
			rm->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++;
	}

	ret = 0;
out:
	if (ret) {
		if (rm)
			rds_message_put(rm);
		rm = ERR_PTR(ret);
	}
	return rm;
}

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->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>
	34
	35	#include "rds.h"
	36	#include "rdma.h"
	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
	65	for (i = 0; i < rm->m_nents; i++) {
	66	rdsdebug("putting data page %p\n", (void *)sg_page(&rm->m_sg[i]));
	67	/* XXX will have to put_page for page refs */
	68	__free_page(sg_page(&rm->m_sg[i]));
	69	}
	70	rm->m_nents = 0;
	71
	72	if (rm->m_rdma_op)
	73	rds_rdma_free_op(rm->m_rdma_op);
	74	if (rm->m_rdma_mr)
	75	rds_mr_put(rm->m_rdma_mr);
	76	}
	77
	78	void rds_message_inc_purge(struct rds_incoming *inc)
	79	{
	80	struct rds_message *rm = container_of(inc, struct rds_message, m_inc);
	81	rds_message_purge(rm);
	82	}
	83
	84	void rds_message_put(struct rds_message *rm)
	85	{
	86	rdsdebug("put rm %p ref %d\n", rm, atomic_read(&rm->m_refcount));
	87
	88	if (atomic_dec_and_test(&rm->m_refcount)) {
	89	BUG_ON(!list_empty(&rm->m_sock_item));
	90	BUG_ON(!list_empty(&rm->m_conn_item));
	91	rds_message_purge(rm);
	92
	93	kfree(rm);
	94	}
	95	}
616b757a	96	EXPORT_SYMBOL_GPL(rds_message_put);
7875e18e AG	97
	98	void rds_message_inc_free(struct rds_incoming *inc)
	99	{
	100	struct rds_message *rm = container_of(inc, struct rds_message, m_inc);
	101	rds_message_put(rm);
	102	}
	103
	104	void rds_message_populate_header(struct rds_header *hdr, __be16 sport,
	105	__be16 dport, u64 seq)
	106	{
	107	hdr->h_flags = 0;
	108	hdr->h_sport = sport;
	109	hdr->h_dport = dport;
	110	hdr->h_sequence = cpu_to_be64(seq);
	111	hdr->h_exthdr[0] = RDS_EXTHDR_NONE;
	112	}
616b757a	113	EXPORT_SYMBOL_GPL(rds_message_populate_header);
7875e18e AG	114
	115	int rds_message_add_extension(struct rds_header *hdr,
	116	unsigned int type, const void *data, unsigned int len)
	117	{
	118	unsigned int ext_len = sizeof(u8) + len;
	119	unsigned char *dst;
	120
	121	/* For now, refuse to add more than one extension header */
	122	if (hdr->h_exthdr[0] != RDS_EXTHDR_NONE)
	123	return 0;
	124
f64f9e71	125	if (type >= __RDS_EXTHDR_MAX \|\| len != rds_exthdr_size[type])
7875e18e AG	126	return 0;
	127
	128	if (ext_len >= RDS_HEADER_EXT_SPACE)
	129	return 0;
	130	dst = hdr->h_exthdr;
	131
	132	*dst++ = type;
	133	memcpy(dst, data, len);
	134
	135	dst[len] = RDS_EXTHDR_NONE;
	136	return 1;
	137	}
616b757a	138	EXPORT_SYMBOL_GPL(rds_message_add_extension);
7875e18e AG	139
	140	/*
	141	* If a message has extension headers, retrieve them here.
	142	* Call like this:
	143	*
	144	* unsigned int pos = 0;
	145	*
	146	* while (1) {
	147	* buflen = sizeof(buffer);
	148	* type = rds_message_next_extension(hdr, &pos, buffer, &buflen);
	149	* if (type == RDS_EXTHDR_NONE)
	150	* break;
	151	* ...
	152	* }
	153	*/
	154	int rds_message_next_extension(struct rds_header *hdr,
	155	unsigned int pos, void buf, unsigned int *buflen)
	156	{
	157	unsigned int offset, ext_type, ext_len;
	158	u8 *src = hdr->h_exthdr;
	159
	160	offset = *pos;
	161	if (offset >= RDS_HEADER_EXT_SPACE)
	162	goto none;
	163
	164	/* Get the extension type and length. For now, the
	165	* length is implied by the extension type. */
	166	ext_type = src[offset++];
	167
	168	if (ext_type == RDS_EXTHDR_NONE \|\| ext_type >= __RDS_EXTHDR_MAX)
	169	goto none;
	170	ext_len = rds_exthdr_size[ext_type];
	171	if (offset + ext_len > RDS_HEADER_EXT_SPACE)
	172	goto none;
	173
	174	*pos = offset + ext_len;
	175	if (ext_len < *buflen)
	176	*buflen = ext_len;
	177	memcpy(buf, src + offset, *buflen);
	178	return ext_type;
	179
	180	none:
	181	*pos = RDS_HEADER_EXT_SPACE;
	182	*buflen = 0;
	183	return RDS_EXTHDR_NONE;
	184	}
	185
	186	int rds_message_add_version_extension(struct rds_header *hdr, unsigned int version)
	187	{
	188	struct rds_ext_header_version ext_hdr;
	189
	190	ext_hdr.h_version = cpu_to_be32(version);
	191	return rds_message_add_extension(hdr, RDS_EXTHDR_VERSION, &ext_hdr, sizeof(ext_hdr));
	192	}
	193
	194	int rds_message_get_version_extension(struct rds_header hdr, unsigned int version)
	195	{
	196	struct rds_ext_header_version ext_hdr;
	197	unsigned int pos = 0, len = sizeof(ext_hdr);
	198
	199	/* We assume the version extension is the only one present */
	200	if (rds_message_next_extension(hdr, &pos, &ext_hdr, &len) != RDS_EXTHDR_VERSION)
	201	return 0;
	202	*version = be32_to_cpu(ext_hdr.h_version);
203	return 1;
204	}
205
206	int rds_message_add_rdma_dest_extension(struct rds_header *hdr, u32 r_key, u32 offset)
207	{
208	struct rds_ext_header_rdma_dest ext_hdr;
209
210	ext_hdr.h_rdma_rkey = cpu_to_be32(r_key);
211	ext_hdr.h_rdma_offset = cpu_to_be32(offset);
212	return rds_message_add_extension(hdr, RDS_EXTHDR_RDMA_DEST, &ext_hdr, sizeof(ext_hdr));
213	}
616b757a	214	EXPORT_SYMBOL_GPL(rds_message_add_rdma_dest_extension);
7875e18e AG	215
	216	struct rds_message *rds_message_alloc(unsigned int nents, gfp_t gfp)
	217	{
	218	struct rds_message *rm;
	219
	220	rm = kzalloc(sizeof(struct rds_message) +
	221	(nents * sizeof(struct scatterlist)), gfp);
	222	if (!rm)
	223	goto out;
	224
	225	if (nents)
	226	sg_init_table(rm->m_sg, nents);
	227	atomic_set(&rm->m_refcount, 1);
	228	INIT_LIST_HEAD(&rm->m_sock_item);
	229	INIT_LIST_HEAD(&rm->m_conn_item);
	230	spin_lock_init(&rm->m_rs_lock);
	231
	232	out:
	233	return rm;
	234	}
	235
	236	struct rds_message rds_message_map_pages(unsigned long page_addrs, unsigned int total_len)
	237	{
	238	struct rds_message *rm;
	239	unsigned int i;
	240
	241	rm = rds_message_alloc(ceil(total_len, PAGE_SIZE), GFP_KERNEL);
	242	if (rm == NULL)
	243	return ERR_PTR(-ENOMEM);
	244
	245	set_bit(RDS_MSG_PAGEVEC, &rm->m_flags);
	246	rm->m_inc.i_hdr.h_len = cpu_to_be32(total_len);
	247	rm->m_nents = ceil(total_len, PAGE_SIZE);
	248
	249	for (i = 0; i < rm->m_nents; ++i) {
	250	sg_set_page(&rm->m_sg[i],
	251	virt_to_page(page_addrs[i]),
	252	PAGE_SIZE, 0);
	253	}
	254
	255	return rm;
	256	}
	257
	258	struct rds_message rds_message_copy_from_user(struct iovec first_iov,
	259	size_t total_len)
	260	{
	261	unsigned long to_copy;
	262	unsigned long iov_off;
	263	unsigned long sg_off;
	264	struct rds_message *rm;
	265	struct iovec *iov;
	266	struct scatterlist *sg;
	267	int ret;
	268
	269	rm = rds_message_alloc(ceil(total_len, PAGE_SIZE), GFP_KERNEL);
	270	if (rm == NULL) {
	271	ret = -ENOMEM;
	272	goto out;
	273	}
	274
	275	rm->m_inc.i_hdr.h_len = cpu_to_be32(total_len);
	276
	277	/*
	278	* now allocate and copy in the data payload.
279	*/
280	sg = rm->m_sg;
281	iov = first_iov;
282	iov_off = 0;
283	sg_off = 0; /* Dear gcc, sg->page will be null from kzalloc. */
284
285	while (total_len) {
286	if (sg_page(sg) == NULL) {
287	ret = rds_page_remainder_alloc(sg, total_len,
288	GFP_HIGHUSER);
289	if (ret)
290	goto out;
291	rm->m_nents++;
292	sg_off = 0;
293	}
294
295	while (iov_off == iov->iov_len) {
296	iov_off = 0;
297	iov++;
298	}
299
300	to_copy = min(iov->iov_len - iov_off, sg->length - sg_off);
301	to_copy = min_t(size_t, to_copy, total_len);
302
303	rdsdebug("copying %lu bytes from user iov [%p, %zu] + %lu to "
304	"sg [%p, %u, %u] + %lu\n",
305	to_copy, iov->iov_base, iov->iov_len, iov_off,
306	(void *)sg_page(sg), sg->offset, sg->length, sg_off);
307
308	ret = rds_page_copy_from_user(sg_page(sg), sg->offset + sg_off,
309	iov->iov_base + iov_off,
310	to_copy);
311	if (ret)
312	goto out;
313
314	iov_off += to_copy;
315	total_len -= to_copy;
316	sg_off += to_copy;
317
318	if (sg_off == sg->length)
319	sg++;
320	}
321
322	ret = 0;
323	out:
324	if (ret) {
325	if (rm)
326	rds_message_put(rm);
327	rm = ERR_PTR(ret);
328	}
329	return rm;
330	}
331
332	int rds_message_inc_copy_to_user(struct rds_incoming *inc,
333	struct iovec *first_iov, size_t size)
334	{
335	struct rds_message *rm;
336	struct iovec *iov;
337	struct scatterlist *sg;
338	unsigned long to_copy;
339	unsigned long iov_off;
340	unsigned long vec_off;
341	int copied;
342	int ret;
343	u32 len;
344
345	rm = container_of(inc, struct rds_message, m_inc);
346	len = be32_to_cpu(rm->m_inc.i_hdr.h_len);
347
348	iov = first_iov;
349	iov_off = 0;
350	sg = rm->m_sg;
351	vec_off = 0;
352	copied = 0;
353
354	while (copied < size && copied < len) {
355	while (iov_off == iov->iov_len) {
356	iov_off = 0;
357	iov++;
358	}
359
360	to_copy = min(iov->iov_len - iov_off, sg->length - vec_off);
361	to_copy = min_t(size_t, to_copy, size - copied);
362	to_copy = min_t(unsigned long, to_copy, len - copied);
363
364	rdsdebug("copying %lu bytes to user iov [%p, %zu] + %lu to "
365	"sg [%p, %u, %u] + %lu\n",
366	to_copy, iov->iov_base, iov->iov_len, iov_off,
367	sg_page(sg), sg->offset, sg->length, vec_off);
368
369	ret = rds_page_copy_to_user(sg_page(sg), sg->offset + vec_off,
370	iov->iov_base + iov_off,
371	to_copy);
372	if (ret) {
373	copied = ret;
374	break;
375	}
376
377	iov_off += to_copy;
378	vec_off += to_copy;
379	copied += to_copy;
380
381	if (vec_off == sg->length) {
382	vec_off = 0;
383	sg++;
384	}
385	}
386
387	return copied;
388	}
389
390	/*
391	* If the message is still on the send queue, wait until the transport
392	* is done with it. This is particularly important for RDMA operations.
393	*/
394	void rds_message_wait(struct rds_message *rm)
395	{
396	wait_event(rds_message_flush_waitq,
397	!test_bit(RDS_MSG_MAPPED, &rm->m_flags));
398	}
399
400	void rds_message_unmapped(struct rds_message *rm)
401	{
402	clear_bit(RDS_MSG_MAPPED, &rm->m_flags);
403	if (waitqueue_active(&rds_message_flush_waitq))
404	wake_up(&rds_message_flush_waitq);
405	}
616b757a	406	EXPORT_SYMBOL_GPL(rds_message_unmapped);
7875e18e	407