[deliverable/linux.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 "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),
};


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_read(&rm->m_refcount) == 0) {
printk(KERN_CRIT "danger refcount zero on %p\n", rm);
WARN_ON(1);
	}
	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_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);

	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);
	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_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_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>
7875e18e AG	36
7875e18e AG	37	#include "rds.h"
7875e18e	38
7875e18e AG	39	static unsigned int rds_exthdr_size[__RDS_EXTHDR_MAX] = {
	40	[RDS_EXTHDR_NONE] = 0,
	41	[RDS_EXTHDR_VERSION] = sizeof(struct rds_ext_header_version),
	42	[RDS_EXTHDR_RDMA] = sizeof(struct rds_ext_header_rdma),
	43	[RDS_EXTHDR_RDMA_DEST] = sizeof(struct rds_ext_header_rdma_dest),
	44	};
	45
	46
	47	void rds_message_addref(struct rds_message *rm)
	48	{
	49	rdsdebug("addref rm %p ref %d\n", rm, atomic_read(&rm->m_refcount));
	50	atomic_inc(&rm->m_refcount);
	51	}
616b757a	52	EXPORT_SYMBOL_GPL(rds_message_addref);
7875e18e AG	53
	54	/*
	55	* This relies on dma_map_sg() not touching sg[].page during merging.
	56	*/
	57	static void rds_message_purge(struct rds_message *rm)
	58	{
	59	unsigned long i;
	60
	61	if (unlikely(test_bit(RDS_MSG_PAGEVEC, &rm->m_flags)))
	62	return;
	63
6c7cc6e4 AG	64	for (i = 0; i < rm->data.op_nents; i++) {
6c7cc6e4 AG	65	rdsdebug("putting data page %p\n", (void *)sg_page(&rm->data.op_sg[i]));
7875e18e	66	/* XXX will have to put_page for page refs */
6c7cc6e4	67	__free_page(sg_page(&rm->data.op_sg[i]));
7875e18e	68	}
6c7cc6e4	69	rm->data.op_nents = 0;
7875e18e	70
f8b3aaf2 AG	71	if (rm->rdma.op_active)
	72	rds_rdma_free_op(&rm->rdma);
	73	if (rm->rdma.op_rdma_mr)
	74	rds_mr_put(rm->rdma.op_rdma_mr);
d0ab25a8 AG	75
	76	if (rm->atomic.op_active)
	77	rds_atomic_free_op(&rm->atomic);
	78	if (rm->atomic.op_rdma_mr)
	79	rds_mr_put(rm->atomic.op_rdma_mr);
7875e18e	80	}
7875e18e AG	81
	82	void rds_message_put(struct rds_message *rm)
	83	{
	84	rdsdebug("put rm %p ref %d\n", rm, atomic_read(&rm->m_refcount));
7e3f2952 CM	85	if (atomic_read(&rm->m_refcount) == 0) {
	86	printk(KERN_CRIT "danger refcount zero on %p\n", rm);
	87	WARN_ON(1);
	88	}
7875e18e AG	89	if (atomic_dec_and_test(&rm->m_refcount)) {
	90	BUG_ON(!list_empty(&rm->m_sock_item));
	91	BUG_ON(!list_empty(&rm->m_conn_item));
	92	rds_message_purge(rm);
	93
	94	kfree(rm);
	95	}
	96	}
616b757a	97	EXPORT_SYMBOL_GPL(rds_message_put);
7875e18e	98
7875e18e AG	99	void rds_message_populate_header(struct rds_header *hdr, __be16 sport,
	100	__be16 dport, u64 seq)
	101	{
	102	hdr->h_flags = 0;
	103	hdr->h_sport = sport;
	104	hdr->h_dport = dport;
	105	hdr->h_sequence = cpu_to_be64(seq);
	106	hdr->h_exthdr[0] = RDS_EXTHDR_NONE;
	107	}
616b757a	108	EXPORT_SYMBOL_GPL(rds_message_populate_header);
7875e18e	109
ff51bf84	110	int rds_message_add_extension(struct rds_header *hdr, unsigned int type,
ff51bf84	111	const void *data, unsigned int len)
7875e18e AG	112	{
	113	unsigned int ext_len = sizeof(u8) + len;
	114	unsigned char *dst;
	115
	116	/* For now, refuse to add more than one extension header */
	117	if (hdr->h_exthdr[0] != RDS_EXTHDR_NONE)
	118	return 0;
	119
f64f9e71	120	if (type >= __RDS_EXTHDR_MAX \|\| len != rds_exthdr_size[type])
7875e18e AG	121	return 0;
	122
	123	if (ext_len >= RDS_HEADER_EXT_SPACE)
	124	return 0;
	125	dst = hdr->h_exthdr;
	126
	127	*dst++ = type;
	128	memcpy(dst, data, len);
	129
	130	dst[len] = RDS_EXTHDR_NONE;
	131	return 1;
	132	}
616b757a	133	EXPORT_SYMBOL_GPL(rds_message_add_extension);
7875e18e AG	134
	135	/*
	136	* If a message has extension headers, retrieve them here.
	137	* Call like this:
	138	*
	139	* unsigned int pos = 0;
	140	*
	141	* while (1) {
	142	* buflen = sizeof(buffer);
	143	* type = rds_message_next_extension(hdr, &pos, buffer, &buflen);
	144	* if (type == RDS_EXTHDR_NONE)
	145	* break;
	146	* ...
	147	* }
	148	*/
	149	int rds_message_next_extension(struct rds_header *hdr,
	150	unsigned int pos, void buf, unsigned int *buflen)
	151	{
	152	unsigned int offset, ext_type, ext_len;
	153	u8 *src = hdr->h_exthdr;
	154
	155	offset = *pos;
	156	if (offset >= RDS_HEADER_EXT_SPACE)
	157	goto none;
	158
	159	/* Get the extension type and length. For now, the
	160	* length is implied by the extension type. */
	161	ext_type = src[offset++];
	162
	163	if (ext_type == RDS_EXTHDR_NONE \|\| ext_type >= __RDS_EXTHDR_MAX)
	164	goto none;
	165	ext_len = rds_exthdr_size[ext_type];
	166	if (offset + ext_len > RDS_HEADER_EXT_SPACE)
	167	goto none;
	168
	169	*pos = offset + ext_len;
	170	if (ext_len < *buflen)
	171	*buflen = ext_len;
	172	memcpy(buf, src + offset, *buflen);
	173	return ext_type;
	174
	175	none:
	176	*pos = RDS_HEADER_EXT_SPACE;
	177	*buflen = 0;
	178	return RDS_EXTHDR_NONE;
	179	}
	180
7875e18e AG	181	int rds_message_add_rdma_dest_extension(struct rds_header *hdr, u32 r_key, u32 offset)
	182	{
	183	struct rds_ext_header_rdma_dest ext_hdr;
	184
	185	ext_hdr.h_rdma_rkey = cpu_to_be32(r_key);
	186	ext_hdr.h_rdma_offset = cpu_to_be32(offset);
	187	return rds_message_add_extension(hdr, RDS_EXTHDR_RDMA_DEST, &ext_hdr, sizeof(ext_hdr));
	188	}
616b757a	189	EXPORT_SYMBOL_GPL(rds_message_add_rdma_dest_extension);
7875e18e	190
fc445084 AG	191	/*
	192	* Each rds_message is allocated with extra space for the scatterlist entries
	193	* rds ops will need. This is to minimize memory allocation count. Then, each rds op
	194	* can grab SGs when initializing its part of the rds_message.
	195	*/
	196	struct rds_message *rds_message_alloc(unsigned int extra_len, gfp_t gfp)
7875e18e AG	197	{
	198	struct rds_message *rm;
	199
ece6b0a2 CW	200	if (extra_len > KMALLOC_MAX_SIZE - sizeof(struct rds_message))
	201	return NULL;
	202
fc445084	203	rm = kzalloc(sizeof(struct rds_message) + extra_len, gfp);
7875e18e AG	204	if (!rm)
	205	goto out;
	206
fc445084 AG	207	rm->m_used_sgs = 0;
	208	rm->m_total_sgs = extra_len / sizeof(struct scatterlist);
	209
7875e18e AG	210	atomic_set(&rm->m_refcount, 1);
	211	INIT_LIST_HEAD(&rm->m_sock_item);
	212	INIT_LIST_HEAD(&rm->m_conn_item);
	213	spin_lock_init(&rm->m_rs_lock);
c83188dc	214	init_waitqueue_head(&rm->m_flush_wait);
7875e18e AG	215
	216	out:
	217	return rm;
	218	}
	219
fc445084 AG	220	/*
	221	* RDS ops use this to grab SG entries from the rm's sg pool.
	222	*/
	223	struct scatterlist rds_message_alloc_sgs(struct rds_message rm, int nents)
	224	{
	225	struct scatterlist sg_first = (struct scatterlist ) &rm[1];
	226	struct scatterlist *sg_ret;
	227
	228	WARN_ON(rm->m_used_sgs + nents > rm->m_total_sgs);
ee4c7b47	229	WARN_ON(!nents);
fc445084	230
d139ff09 AG	231	if (rm->m_used_sgs + nents > rm->m_total_sgs)
	232	return NULL;
	233
fc445084	234	sg_ret = &sg_first[rm->m_used_sgs];
f4dd96f7	235	sg_init_table(sg_ret, nents);
fc445084 AG	236	rm->m_used_sgs += nents;
	237
	238	return sg_ret;
	239	}
	240
7875e18e AG	241	struct rds_message rds_message_map_pages(unsigned long page_addrs, unsigned int total_len)
	242	{
	243	struct rds_message *rm;
	244	unsigned int i;
ff87e97a AG	245	int num_sgs = ceil(total_len, PAGE_SIZE);
ff87e97a AG	246	int extra_bytes = num_sgs * sizeof(struct scatterlist);
7875e18e	247
f2ec76f2	248	rm = rds_message_alloc(extra_bytes, GFP_NOWAIT);
8690bfa1	249	if (!rm)
7875e18e AG	250	return ERR_PTR(-ENOMEM);
	251
	252	set_bit(RDS_MSG_PAGEVEC, &rm->m_flags);
	253	rm->m_inc.i_hdr.h_len = cpu_to_be32(total_len);
6c7cc6e4 AG	254	rm->data.op_nents = ceil(total_len, PAGE_SIZE);
6c7cc6e4 AG	255	rm->data.op_sg = rds_message_alloc_sgs(rm, num_sgs);
aa58163a PE	256	if (!rm->data.op_sg) {
aa58163a PE	257	rds_message_put(rm);
d139ff09	258	return ERR_PTR(-ENOMEM);
aa58163a	259	}
7875e18e	260
6c7cc6e4 AG	261	for (i = 0; i < rm->data.op_nents; ++i) {
6c7cc6e4 AG	262	sg_set_page(&rm->data.op_sg[i],
7875e18e AG	263	virt_to_page(page_addrs[i]),
	264	PAGE_SIZE, 0);
	265	}
	266
	267	return rm;
	268	}
	269
fc445084	270	int rds_message_copy_from_user(struct rds_message rm, struct iovec first_iov,
7875e18e AG	271	size_t total_len)
	272	{
	273	unsigned long to_copy;
	274	unsigned long iov_off;
	275	unsigned long sg_off;
7875e18e AG	276	struct iovec *iov;
7875e18e AG	277	struct scatterlist *sg;
fc445084	278	int ret = 0;
7875e18e AG	279
	280	rm->m_inc.i_hdr.h_len = cpu_to_be32(total_len);
	281
	282	/*
	283	* now allocate and copy in the data payload.
	284	*/
6c7cc6e4	285	sg = rm->data.op_sg;
7875e18e AG	286	iov = first_iov;
	287	iov_off = 0;
	288	sg_off = 0; /* Dear gcc, sg->page will be null from kzalloc. */
	289
	290	while (total_len) {
8690bfa1	291	if (!sg_page(sg)) {
7875e18e AG	292	ret = rds_page_remainder_alloc(sg, total_len,
	293	GFP_HIGHUSER);
	294	if (ret)
	295	goto out;
6c7cc6e4	296	rm->data.op_nents++;
7875e18e AG	297	sg_off = 0;
	298	}
	299
	300	while (iov_off == iov->iov_len) {
	301	iov_off = 0;
	302	iov++;
	303	}
	304
	305	to_copy = min(iov->iov_len - iov_off, sg->length - sg_off);
	306	to_copy = min_t(size_t, to_copy, total_len);
	307
	308	rdsdebug("copying %lu bytes from user iov [%p, %zu] + %lu to "
	309	"sg [%p, %u, %u] + %lu\n",
	310	to_copy, iov->iov_base, iov->iov_len, iov_off,
	311	(void *)sg_page(sg), sg->offset, sg->length, sg_off);
	312
	313	ret = rds_page_copy_from_user(sg_page(sg), sg->offset + sg_off,
	314	iov->iov_base + iov_off,
	315	to_copy);
	316	if (ret)
	317	goto out;
	318
	319	iov_off += to_copy;
	320	total_len -= to_copy;
	321	sg_off += to_copy;
	322
	323	if (sg_off == sg->length)
	324	sg++;
	325	}
	326
7875e18e	327	out:
fc445084	328	return ret;
7875e18e AG	329	}
	330
	331	int rds_message_inc_copy_to_user(struct rds_incoming *inc,
	332	struct iovec *first_iov, size_t size)
	333	{
	334	struct rds_message *rm;
	335	struct iovec *iov;
	336	struct scatterlist *sg;
	337	unsigned long to_copy;
	338	unsigned long iov_off;
	339	unsigned long vec_off;
	340	int copied;
	341	int ret;
	342	u32 len;
	343
	344	rm = container_of(inc, struct rds_message, m_inc);
	345	len = be32_to_cpu(rm->m_inc.i_hdr.h_len);
	346
	347	iov = first_iov;
	348	iov_off = 0;
6c7cc6e4	349	sg = rm->data.op_sg;
7875e18e AG	350	vec_off = 0;
	351	copied = 0;
	352
	353	while (copied < size && copied < len) {
	354	while (iov_off == iov->iov_len) {
	355	iov_off = 0;
	356	iov++;
	357	}
	358
	359	to_copy = min(iov->iov_len - iov_off, sg->length - vec_off);
	360	to_copy = min_t(size_t, to_copy, size - copied);
	361	to_copy = min_t(unsigned long, to_copy, len - copied);
	362
	363	rdsdebug("copying %lu bytes to user iov [%p, %zu] + %lu to "
	364	"sg [%p, %u, %u] + %lu\n",
	365	to_copy, iov->iov_base, iov->iov_len, iov_off,
	366	sg_page(sg), sg->offset, sg->length, vec_off);
	367
	368	ret = rds_page_copy_to_user(sg_page(sg), sg->offset + vec_off,
	369	iov->iov_base + iov_off,
	370	to_copy);
	371	if (ret) {
	372	copied = ret;
	373	break;
	374	}
	375
	376	iov_off += to_copy;
	377	vec_off += to_copy;
	378	copied += to_copy;
	379
	380	if (vec_off == sg->length) {
	381	vec_off = 0;
	382	sg++;
	383	}
	384	}
	385
	386	return copied;
	387	}
	388
	389	/*
	390	* If the message is still on the send queue, wait until the transport
	391	* is done with it. This is particularly important for RDMA operations.
	392	*/
	393	void rds_message_wait(struct rds_message *rm)
	394	{
c83188dc	395	wait_event_interruptible(rm->m_flush_wait,
7875e18e AG	396	!test_bit(RDS_MSG_MAPPED, &rm->m_flags));
	397	}
	398
	399	void rds_message_unmapped(struct rds_message *rm)
	400	{
	401	clear_bit(RDS_MSG_MAPPED, &rm->m_flags);
c83188dc	402	wake_up_interruptible(&rm->m_flush_wait);
7875e18e	403	}
616b757a	404	EXPORT_SYMBOL_GPL(rds_message_unmapped);
7875e18e	405