[net-next-2.6.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.op_nents; i++) {
		rdsdebug("putting data page %p\n", (void *)sg_page(&rm->data.op_sg[i]));
		/* XXX will have to put_page for page refs */
		__free_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, 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);
	WARN_ON(!nents);

	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);

	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_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.op_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.op_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++;
	}

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.op_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);
	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
6c7cc6e4 AG	65	for (i = 0; i < rm->data.op_nents; i++) {
6c7cc6e4 AG	66	rdsdebug("putting data page %p\n", (void *)sg_page(&rm->data.op_sg[i]));
7875e18e	67	/* XXX will have to put_page for page refs */
6c7cc6e4	68	__free_page(sg_page(&rm->data.op_sg[i]));
7875e18e	69	}
6c7cc6e4	70	rm->data.op_nents = 0;
7875e18e	71
f8b3aaf2 AG	72	if (rm->rdma.op_active)
	73	rds_rdma_free_op(&rm->rdma);
	74	if (rm->rdma.op_rdma_mr)
	75	rds_mr_put(rm->rdma.op_rdma_mr);
d0ab25a8 AG	76
	77	if (rm->atomic.op_active)
	78	rds_atomic_free_op(&rm->atomic);
	79	if (rm->atomic.op_rdma_mr)
	80	rds_mr_put(rm->atomic.op_rdma_mr);
7875e18e	81	}
7875e18e AG	82
	83	void rds_message_put(struct rds_message *rm)
	84	{
	85	rdsdebug("put rm %p ref %d\n", rm, atomic_read(&rm->m_refcount));
	86
	87	if (atomic_dec_and_test(&rm->m_refcount)) {
	88	BUG_ON(!list_empty(&rm->m_sock_item));
	89	BUG_ON(!list_empty(&rm->m_conn_item));
	90	rds_message_purge(rm);
	91
	92	kfree(rm);
	93	}
	94	}
616b757a	95	EXPORT_SYMBOL_GPL(rds_message_put);
7875e18e	96
7875e18e AG	97	void rds_message_populate_header(struct rds_header *hdr, __be16 sport,
	98	__be16 dport, u64 seq)
	99	{
	100	hdr->h_flags = 0;
	101	hdr->h_sport = sport;
	102	hdr->h_dport = dport;
	103	hdr->h_sequence = cpu_to_be64(seq);
	104	hdr->h_exthdr[0] = RDS_EXTHDR_NONE;
	105	}
616b757a	106	EXPORT_SYMBOL_GPL(rds_message_populate_header);
7875e18e AG	107
	108	int rds_message_add_extension(struct rds_header *hdr,
	109	unsigned int type, const void *data, unsigned int len)
	110	{
	111	unsigned int ext_len = sizeof(u8) + len;
	112	unsigned char *dst;
	113
	114	/* For now, refuse to add more than one extension header */
	115	if (hdr->h_exthdr[0] != RDS_EXTHDR_NONE)
	116	return 0;
	117
f64f9e71	118	if (type >= __RDS_EXTHDR_MAX \|\| len != rds_exthdr_size[type])
7875e18e AG	119	return 0;
	120
	121	if (ext_len >= RDS_HEADER_EXT_SPACE)
	122	return 0;
	123	dst = hdr->h_exthdr;
	124
	125	*dst++ = type;
	126	memcpy(dst, data, len);
	127
	128	dst[len] = RDS_EXTHDR_NONE;
	129	return 1;
	130	}
616b757a	131	EXPORT_SYMBOL_GPL(rds_message_add_extension);
7875e18e AG	132
	133	/*
	134	* If a message has extension headers, retrieve them here.
	135	* Call like this:
	136	*
	137	* unsigned int pos = 0;
	138	*
	139	* while (1) {
	140	* buflen = sizeof(buffer);
	141	* type = rds_message_next_extension(hdr, &pos, buffer, &buflen);
	142	* if (type == RDS_EXTHDR_NONE)
	143	* break;
	144	* ...
	145	* }
	146	*/
	147	int rds_message_next_extension(struct rds_header *hdr,
	148	unsigned int pos, void buf, unsigned int *buflen)
	149	{
	150	unsigned int offset, ext_type, ext_len;
	151	u8 *src = hdr->h_exthdr;
	152
	153	offset = *pos;
	154	if (offset >= RDS_HEADER_EXT_SPACE)
	155	goto none;
	156
	157	/* Get the extension type and length. For now, the
	158	* length is implied by the extension type. */
	159	ext_type = src[offset++];
	160
	161	if (ext_type == RDS_EXTHDR_NONE \|\| ext_type >= __RDS_EXTHDR_MAX)
	162	goto none;
	163	ext_len = rds_exthdr_size[ext_type];
	164	if (offset + ext_len > RDS_HEADER_EXT_SPACE)
	165	goto none;
	166
	167	*pos = offset + ext_len;
	168	if (ext_len < *buflen)
	169	*buflen = ext_len;
	170	memcpy(buf, src + offset, *buflen);
	171	return ext_type;
	172
	173	none:
	174	*pos = RDS_HEADER_EXT_SPACE;
	175	*buflen = 0;
	176	return RDS_EXTHDR_NONE;
	177	}
	178
	179	int rds_message_add_version_extension(struct rds_header *hdr, unsigned int version)
	180	{
	181	struct rds_ext_header_version ext_hdr;
	182
	183	ext_hdr.h_version = cpu_to_be32(version);
	184	return rds_message_add_extension(hdr, RDS_EXTHDR_VERSION, &ext_hdr, sizeof(ext_hdr));
	185	}
	186
	187	int rds_message_get_version_extension(struct rds_header hdr, unsigned int version)
	188	{
	189	struct rds_ext_header_version ext_hdr;
	190	unsigned int pos = 0, len = sizeof(ext_hdr);
	191
	192	/* We assume the version extension is the only one present */
	193	if (rds_message_next_extension(hdr, &pos, &ext_hdr, &len) != RDS_EXTHDR_VERSION)
	194	return 0;
	195	*version = be32_to_cpu(ext_hdr.h_version);
196	return 1;
197	}
198
199	int rds_message_add_rdma_dest_extension(struct rds_header *hdr, u32 r_key, u32 offset)
200	{
201	struct rds_ext_header_rdma_dest ext_hdr;
202
203	ext_hdr.h_rdma_rkey = cpu_to_be32(r_key);
204	ext_hdr.h_rdma_offset = cpu_to_be32(offset);
205	return rds_message_add_extension(hdr, RDS_EXTHDR_RDMA_DEST, &ext_hdr, sizeof(ext_hdr));
206	}
616b757a	207	EXPORT_SYMBOL_GPL(rds_message_add_rdma_dest_extension);
7875e18e	208
fc445084 AG	209	/*
	210	* Each rds_message is allocated with extra space for the scatterlist entries
	211	* rds ops will need. This is to minimize memory allocation count. Then, each rds op
	212	* can grab SGs when initializing its part of the rds_message.
	213	*/
	214	struct rds_message *rds_message_alloc(unsigned int extra_len, gfp_t gfp)
7875e18e AG	215	{
	216	struct rds_message *rm;
	217
fc445084	218	rm = kzalloc(sizeof(struct rds_message) + extra_len, gfp);
7875e18e AG	219	if (!rm)
	220	goto out;
	221
fc445084 AG	222	rm->m_used_sgs = 0;
	223	rm->m_total_sgs = extra_len / sizeof(struct scatterlist);
	224
7875e18e AG	225	atomic_set(&rm->m_refcount, 1);
	226	INIT_LIST_HEAD(&rm->m_sock_item);
	227	INIT_LIST_HEAD(&rm->m_conn_item);
	228	spin_lock_init(&rm->m_rs_lock);
	229
	230	out:
	231	return rm;
	232	}
	233
fc445084 AG	234	/*
	235	* RDS ops use this to grab SG entries from the rm's sg pool.
	236	*/
	237	struct scatterlist rds_message_alloc_sgs(struct rds_message rm, int nents)
	238	{
	239	struct scatterlist sg_first = (struct scatterlist ) &rm[1];
	240	struct scatterlist *sg_ret;
	241
	242	WARN_ON(rm->m_used_sgs + nents > rm->m_total_sgs);
ee4c7b47	243	WARN_ON(!nents);
fc445084 AG	244
fc445084 AG	245	sg_ret = &sg_first[rm->m_used_sgs];
f4dd96f7	246	sg_init_table(sg_ret, nents);
fc445084 AG	247	rm->m_used_sgs += nents;
	248
	249	return sg_ret;
	250	}
	251
7875e18e AG	252	struct rds_message rds_message_map_pages(unsigned long page_addrs, unsigned int total_len)
	253	{
	254	struct rds_message *rm;
	255	unsigned int i;
ff87e97a AG	256	int num_sgs = ceil(total_len, PAGE_SIZE);
ff87e97a AG	257	int extra_bytes = num_sgs * sizeof(struct scatterlist);
7875e18e	258
f2ec76f2	259	rm = rds_message_alloc(extra_bytes, GFP_NOWAIT);
8690bfa1	260	if (!rm)
7875e18e AG	261	return ERR_PTR(-ENOMEM);
	262
	263	set_bit(RDS_MSG_PAGEVEC, &rm->m_flags);
	264	rm->m_inc.i_hdr.h_len = cpu_to_be32(total_len);
6c7cc6e4 AG	265	rm->data.op_nents = ceil(total_len, PAGE_SIZE);
6c7cc6e4 AG	266	rm->data.op_sg = rds_message_alloc_sgs(rm, num_sgs);
7875e18e	267
6c7cc6e4 AG	268	for (i = 0; i < rm->data.op_nents; ++i) {
6c7cc6e4 AG	269	sg_set_page(&rm->data.op_sg[i],
7875e18e AG	270	virt_to_page(page_addrs[i]),
	271	PAGE_SIZE, 0);
	272	}
	273
	274	return rm;
	275	}
	276
fc445084	277	int rds_message_copy_from_user(struct rds_message rm, struct iovec first_iov,
7875e18e AG	278	size_t total_len)
	279	{
	280	unsigned long to_copy;
	281	unsigned long iov_off;
	282	unsigned long sg_off;
7875e18e AG	283	struct iovec *iov;
7875e18e AG	284	struct scatterlist *sg;
fc445084	285	int ret = 0;
7875e18e AG	286
	287	rm->m_inc.i_hdr.h_len = cpu_to_be32(total_len);
	288
	289	/*
	290	* now allocate and copy in the data payload.
	291	*/
6c7cc6e4	292	sg = rm->data.op_sg;
7875e18e AG	293	iov = first_iov;
	294	iov_off = 0;
	295	sg_off = 0; /* Dear gcc, sg->page will be null from kzalloc. */
	296
	297	while (total_len) {
8690bfa1	298	if (!sg_page(sg)) {
7875e18e AG	299	ret = rds_page_remainder_alloc(sg, total_len,
	300	GFP_HIGHUSER);
	301	if (ret)
	302	goto out;
6c7cc6e4	303	rm->data.op_nents++;
7875e18e AG	304	sg_off = 0;
	305	}
	306
	307	while (iov_off == iov->iov_len) {
	308	iov_off = 0;
	309	iov++;
	310	}
	311
	312	to_copy = min(iov->iov_len - iov_off, sg->length - sg_off);
	313	to_copy = min_t(size_t, to_copy, total_len);
	314
	315	rdsdebug("copying %lu bytes from user iov [%p, %zu] + %lu to "
	316	"sg [%p, %u, %u] + %lu\n",
	317	to_copy, iov->iov_base, iov->iov_len, iov_off,
	318	(void *)sg_page(sg), sg->offset, sg->length, sg_off);
	319
	320	ret = rds_page_copy_from_user(sg_page(sg), sg->offset + sg_off,
	321	iov->iov_base + iov_off,
	322	to_copy);
	323	if (ret)
	324	goto out;
	325
	326	iov_off += to_copy;
	327	total_len -= to_copy;
	328	sg_off += to_copy;
	329
	330	if (sg_off == sg->length)
	331	sg++;
	332	}
	333
7875e18e	334	out:
fc445084	335	return ret;
7875e18e AG	336	}
	337
	338	int rds_message_inc_copy_to_user(struct rds_incoming *inc,
	339	struct iovec *first_iov, size_t size)
	340	{
	341	struct rds_message *rm;
	342	struct iovec *iov;
	343	struct scatterlist *sg;
	344	unsigned long to_copy;
	345	unsigned long iov_off;
	346	unsigned long vec_off;
	347	int copied;
	348	int ret;
	349	u32 len;
	350
	351	rm = container_of(inc, struct rds_message, m_inc);
	352	len = be32_to_cpu(rm->m_inc.i_hdr.h_len);
	353
	354	iov = first_iov;
	355	iov_off = 0;
6c7cc6e4	356	sg = rm->data.op_sg;
7875e18e AG	357	vec_off = 0;
	358	copied = 0;
	359
	360	while (copied < size && copied < len) {
	361	while (iov_off == iov->iov_len) {
	362	iov_off = 0;
	363	iov++;
	364	}
	365
	366	to_copy = min(iov->iov_len - iov_off, sg->length - vec_off);
	367	to_copy = min_t(size_t, to_copy, size - copied);
	368	to_copy = min_t(unsigned long, to_copy, len - copied);
	369
	370	rdsdebug("copying %lu bytes to user iov [%p, %zu] + %lu to "
	371	"sg [%p, %u, %u] + %lu\n",
	372	to_copy, iov->iov_base, iov->iov_len, iov_off,
	373	sg_page(sg), sg->offset, sg->length, vec_off);
	374
	375	ret = rds_page_copy_to_user(sg_page(sg), sg->offset + vec_off,
	376	iov->iov_base + iov_off,
	377	to_copy);
	378	if (ret) {
	379	copied = ret;
	380	break;
	381	}
	382
	383	iov_off += to_copy;
	384	vec_off += to_copy;
	385	copied += to_copy;
	386
	387	if (vec_off == sg->length) {
	388	vec_off = 0;
	389	sg++;
	390	}
	391	}
	392
	393	return copied;
	394	}
	395
	396	/*
	397	* If the message is still on the send queue, wait until the transport
	398	* is done with it. This is particularly important for RDMA operations.
	399	*/
	400	void rds_message_wait(struct rds_message *rm)
	401	{
	402	wait_event(rds_message_flush_waitq,
	403	!test_bit(RDS_MSG_MAPPED, &rm->m_flags));
	404	}
	405
	406	void rds_message_unmapped(struct rds_message *rm)
	407	{
	408	clear_bit(RDS_MSG_MAPPED, &rm->m_flags);
ab1a6926	409	wake_up(&rds_message_flush_waitq);
7875e18e	410	}
616b757a	411	EXPORT_SYMBOL_GPL(rds_message_unmapped);
7875e18e	412