[linux.git] / net / rds / message.c

/*
 * Copyright (c) 2006, 2020 Oracle and/or its affiliates.
 *
 * This software is available to you under a choice of one of two
 * licenses.  You may choose to be licensed under the terms of the GNU
 * General Public License (GPL) Version 2, available from the file
 * COPYING in the main directory of this source tree, or the
 * OpenIB.org BSD license below:
 *
 *     Redistribution and use in source and binary forms, with or
 *     without modification, are permitted provided that the following
 *     conditions are met:
 *
 *      - Redistributions of source code must retain the above
 *        copyright notice, this list of conditions and the following
 *        disclaimer.
 *
 *      - Redistributions in binary form must reproduce the above
 *        copyright notice, this list of conditions and the following
 *        disclaimer in the documentation and/or other materials
 *        provided with the distribution.
 *
 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
 * SOFTWARE.
 *
 */
#include <linux/kernel.h>
#include <linux/slab.h>
#include <linux/export.h>
#include <linux/skbuff.h>
#include <linux/list.h>
#include <linux/errqueue.h>

#include "rds.h"

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

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

static inline bool rds_zcookie_add(struct rds_msg_zcopy_info *info, u32 cookie)
{
	struct rds_zcopy_cookies *ck = &info->zcookies;
	int ncookies = ck->num;

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

static struct rds_msg_zcopy_info *rds_info_from_znotifier(struct rds_znotifier *znotif)
{
	return container_of(znotif, struct rds_msg_zcopy_info, znotif);
}

void rds_notify_msg_zcopy_purge(struct rds_msg_zcopy_queue *q)
{
	unsigned long flags;
	LIST_HEAD(copy);
	struct rds_msg_zcopy_info *info, *tmp;

	spin_lock_irqsave(&q->lock, flags);
	list_splice(&q->zcookie_head, &copy);
	INIT_LIST_HEAD(&q->zcookie_head);
	spin_unlock_irqrestore(&q->lock, flags);

	list_for_each_entry_safe(info, tmp, &copy, rs_zcookie_next) {
		list_del(&info->rs_zcookie_next);
		kfree(info);
	}
}

static void rds_rm_zerocopy_callback(struct rds_sock *rs,
				     struct rds_znotifier *znotif)
{
	struct rds_msg_zcopy_info *info;
	struct rds_msg_zcopy_queue *q;
	u32 cookie = znotif->z_cookie;
	struct rds_zcopy_cookies *ck;
	struct list_head *head;
	unsigned long flags;

	mm_unaccount_pinned_pages(&znotif->z_mmp);
	q = &rs->rs_zcookie_queue;
	spin_lock_irqsave(&q->lock, flags);
	head = &q->zcookie_head;
	if (!list_empty(head)) {
		info = list_entry(head, struct rds_msg_zcopy_info,
				  rs_zcookie_next);
		if (info && rds_zcookie_add(info, cookie)) {
			spin_unlock_irqrestore(&q->lock, flags);
			kfree(rds_info_from_znotifier(znotif));
			/* caller invokes rds_wake_sk_sleep() */
			return;
		}
	}

	info = rds_info_from_znotifier(znotif);
	ck = &info->zcookies;
	memset(ck, 0, sizeof(*ck));
	WARN_ON(!rds_zcookie_add(info, cookie));
	list_add_tail(&q->zcookie_head, &info->rs_zcookie_next);

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

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

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

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

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

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

	if (rm->rdma.op_active)
		rds_rdma_free_op(&rm->rdma);
	if (rm->rdma.op_rdma_mr)
		kref_put(&rm->rdma.op_rdma_mr->r_kref, __rds_put_mr_final);

	if (rm->atomic.op_active)
		rds_atomic_free_op(&rm->atomic);
	if (rm->atomic.op_rdma_mr)
		kref_put(&rm->atomic.op_rdma_mr->r_kref, __rds_put_mr_final);
}

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

out:
	return rm;
}

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

	if (nents <= 0) {
		pr_warn("rds: alloc sgs failed! nents <= 0\n");
		return ERR_PTR(-EINVAL);
	}

	if (rm->m_used_sgs + nents > rm->m_total_sgs) {
		pr_warn("rds: alloc sgs failed! total %d used %d nents %d\n",
			rm->m_total_sgs, rm->m_used_sgs, nents);
		return ERR_PTR(-ENOMEM);
	}

	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 = DIV_ROUND_UP(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 = DIV_ROUND_UP(total_len, PAGE_SIZE);
	rm->data.op_sg = rds_message_alloc_sgs(rm, num_sgs);
	if (IS_ERR(rm->data.op_sg)) {
		rds_message_put(rm);
		return ERR_CAST(rm->data.op_sg);
	}

	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;
}

static int rds_message_zcopy_from_user(struct rds_message *rm, struct iov_iter *from)
{
	struct scatterlist *sg;
	int ret = 0;
	int length = iov_iter_count(from);
	int total_copied = 0;
	struct rds_msg_zcopy_info *info;

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

	/*
	 * now allocate and copy in the data payload.
	 */
	sg = rm->data.op_sg;

	info = kzalloc(sizeof(*info), GFP_KERNEL);
	if (!info)
		return -ENOMEM;
	INIT_LIST_HEAD(&info->rs_zcookie_next);
	rm->data.op_mmp_znotifier = &info->znotif;
	if (mm_account_pinned_pages(&rm->data.op_mmp_znotifier->z_mmp,
				    length)) {
		ret = -ENOMEM;
		goto err;
	}
	while (iov_iter_count(from)) {
		struct page *pages;
		size_t start;
		ssize_t copied;

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

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

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

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

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

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

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

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

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

		sg_off += to_copy;

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

	return ret;
}

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

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

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

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

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

		vec_off += to_copy;
		copied += to_copy;

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

	return copied;
}

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

void rds_message_unmapped(struct rds_message *rm)
{
	clear_bit(RDS_MSG_MAPPED, &rm->m_flags);
	wake_up_interruptible(&rm->m_flush_wait);
}
EXPORT_SYMBOL_GPL(rds_message_unmapped);
Commit	Line	Data
7875e18e	1	/*
e228a5d0	2	* Copyright (c) 2006, 2020 Oracle and/or its affiliates.
7875e18e AG	3	*
	4	* This software is available to you under a choice of one of two
	5	* licenses. You may choose to be licensed under the terms of the GNU
	6	* General Public License (GPL) Version 2, available from the file
	7	* COPYING in the main directory of this source tree, or the
	8	* OpenIB.org BSD license below:
	9	*
	10	* Redistribution and use in source and binary forms, with or
	11	* without modification, are permitted provided that the following
	12	* conditions are met:
	13	*
	14	* - Redistributions of source code must retain the above
	15	* copyright notice, this list of conditions and the following
	16	* disclaimer.
	17	*
	18	* - Redistributions in binary form must reproduce the above
	19	* copyright notice, this list of conditions and the following
	20	* disclaimer in the documentation and/or other materials
	21	* provided with the distribution.
	22	*
	23	* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
	24	* EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
	25	* MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
	26	* NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
	27	* BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
	28	* ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
	29	* CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
	30	* SOFTWARE.
	31	*
	32	*/
	33	#include <linux/kernel.h>
5a0e3ad6	34	#include <linux/slab.h>
bc3b2d7f	35	#include <linux/export.h>
01883eda SV	36	#include <linux/skbuff.h>
	37	#include <linux/list.h>
	38	#include <linux/errqueue.h>
7875e18e AG	39
7875e18e AG	40	#include "rds.h"
7875e18e	41
7875e18e AG	42	static unsigned int rds_exthdr_size[__RDS_EXTHDR_MAX] = {
	43	[RDS_EXTHDR_NONE] = 0,
	44	[RDS_EXTHDR_VERSION] = sizeof(struct rds_ext_header_version),
	45	[RDS_EXTHDR_RDMA] = sizeof(struct rds_ext_header_rdma),
	46	[RDS_EXTHDR_RDMA_DEST] = sizeof(struct rds_ext_header_rdma_dest),
5916e2c1	47	[RDS_EXTHDR_NPATHS] = sizeof(u16),
905dd418	48	[RDS_EXTHDR_GEN_NUM] = sizeof(u32),
7875e18e AG	49	};
7875e18e AG	50
7875e18e AG	51	void rds_message_addref(struct rds_message *rm)
7875e18e AG	52	{
6c5a1c4a RE	53	rdsdebug("addref rm %p ref %d\n", rm, refcount_read(&rm->m_refcount));
6c5a1c4a RE	54	refcount_inc(&rm->m_refcount);
7875e18e	55	}
616b757a	56	EXPORT_SYMBOL_GPL(rds_message_addref);
7875e18e	57
9426bbc6	58	static inline bool rds_zcookie_add(struct rds_msg_zcopy_info *info, u32 cookie)
01883eda	59	{
9426bbc6	60	struct rds_zcopy_cookies *ck = &info->zcookies;
401910db	61	int ncookies = ck->num;
01883eda	62
401910db	63	if (ncookies == RDS_MAX_ZCOOKIES)
01883eda	64	return false;
401910db SV	65	ck->cookies[ncookies] = cookie;
401910db SV	66	ck->num = ++ncookies;
01883eda SV	67	return true;
	68	}
	69
571e6776	70	static struct rds_msg_zcopy_info rds_info_from_znotifier(struct rds_znotifier znotif)
9426bbc6 SV	71	{
	72	return container_of(znotif, struct rds_msg_zcopy_info, znotif);
	73	}
	74
	75	void rds_notify_msg_zcopy_purge(struct rds_msg_zcopy_queue *q)
	76	{
	77	unsigned long flags;
	78	LIST_HEAD(copy);
	79	struct rds_msg_zcopy_info info, tmp;
	80
	81	spin_lock_irqsave(&q->lock, flags);
	82	list_splice(&q->zcookie_head, &copy);
	83	INIT_LIST_HEAD(&q->zcookie_head);
	84	spin_unlock_irqrestore(&q->lock, flags);
	85
	86	list_for_each_entry_safe(info, tmp, &copy, rs_zcookie_next) {
	87	list_del(&info->rs_zcookie_next);
	88	kfree(info);
	89	}
	90	}
	91
01883eda SV	92	static void rds_rm_zerocopy_callback(struct rds_sock *rs,
	93	struct rds_znotifier *znotif)
	94	{
9426bbc6 SV	95	struct rds_msg_zcopy_info *info;
9426bbc6 SV	96	struct rds_msg_zcopy_queue *q;
01883eda	97	u32 cookie = znotif->z_cookie;
401910db	98	struct rds_zcopy_cookies *ck;
9426bbc6 SV	99	struct list_head *head;
9426bbc6 SV	100	unsigned long flags;
01883eda	101
9426bbc6	102	mm_unaccount_pinned_pages(&znotif->z_mmp);
401910db	103	q = &rs->rs_zcookie_queue;
01883eda	104	spin_lock_irqsave(&q->lock, flags);
9426bbc6 SV	105	head = &q->zcookie_head;
	106	if (!list_empty(head)) {
	107	info = list_entry(head, struct rds_msg_zcopy_info,
	108	rs_zcookie_next);
	109	if (info && rds_zcookie_add(info, cookie)) {
	110	spin_unlock_irqrestore(&q->lock, flags);
	111	kfree(rds_info_from_znotifier(znotif));
	112	/* caller invokes rds_wake_sk_sleep() */
	113	return;
	114	}
01883eda SV	115	}
01883eda SV	116
9426bbc6 SV	117	info = rds_info_from_znotifier(znotif);
9426bbc6 SV	118	ck = &info->zcookies;
401910db	119	memset(ck, 0, sizeof(*ck));
9426bbc6 SV	120	WARN_ON(!rds_zcookie_add(info, cookie));
9426bbc6 SV	121	list_add_tail(&q->zcookie_head, &info->rs_zcookie_next);
01883eda SV	122
01883eda SV	123	spin_unlock_irqrestore(&q->lock, flags);
401910db	124	/* caller invokes rds_wake_sk_sleep() */
01883eda SV	125	}
01883eda SV	126
7875e18e AG	127	/*
	128	* This relies on dma_map_sg() not touching sg[].page during merging.
	129	*/
	130	static void rds_message_purge(struct rds_message *rm)
	131	{
ea8994cb	132	unsigned long i, flags;
01883eda	133	bool zcopy = false;
7875e18e AG	134
	135	if (unlikely(test_bit(RDS_MSG_PAGEVEC, &rm->m_flags)))
	136	return;
	137
ea8994cb SV	138	spin_lock_irqsave(&rm->m_rs_lock, flags);
ea8994cb SV	139	if (rm->m_rs) {
01883eda SV	140	struct rds_sock *rs = rm->m_rs;
	141
	142	if (rm->data.op_mmp_znotifier) {
	143	zcopy = true;
	144	rds_rm_zerocopy_callback(rs, rm->data.op_mmp_znotifier);
401910db	145	rds_wake_sk_sleep(rs);
01883eda SV	146	rm->data.op_mmp_znotifier = NULL;
	147	}
	148	sock_put(rds_rs_to_sk(rs));
ea8994cb SV	149	rm->m_rs = NULL;
	150	}
	151	spin_unlock_irqrestore(&rm->m_rs_lock, flags);
7875e18e	152
01883eda SV	153	for (i = 0; i < rm->data.op_nents; i++) {
	154	/* XXX will have to put_page for page refs */
	155	if (!zcopy)
	156	__free_page(sg_page(&rm->data.op_sg[i]));
	157	else
	158	put_page(sg_page(&rm->data.op_sg[i]));
	159	}
	160	rm->data.op_nents = 0;
	161
f8b3aaf2 AG	162	if (rm->rdma.op_active)
	163	rds_rdma_free_op(&rm->rdma);
	164	if (rm->rdma.op_rdma_mr)
e228a5d0	165	kref_put(&rm->rdma.op_rdma_mr->r_kref, __rds_put_mr_final);
d0ab25a8 AG	166
	167	if (rm->atomic.op_active)
	168	rds_atomic_free_op(&rm->atomic);
	169	if (rm->atomic.op_rdma_mr)
e228a5d0	170	kref_put(&rm->atomic.op_rdma_mr->r_kref, __rds_put_mr_final);
7875e18e	171	}
7875e18e AG	172
	173	void rds_message_put(struct rds_message *rm)
	174	{
6c5a1c4a RE	175	rdsdebug("put rm %p ref %d\n", rm, refcount_read(&rm->m_refcount));
	176	WARN(!refcount_read(&rm->m_refcount), "danger refcount zero on %p\n", rm);
	177	if (refcount_dec_and_test(&rm->m_refcount)) {
7875e18e AG	178	BUG_ON(!list_empty(&rm->m_sock_item));
	179	BUG_ON(!list_empty(&rm->m_conn_item));
	180	rds_message_purge(rm);
	181
	182	kfree(rm);
	183	}
	184	}
616b757a	185	EXPORT_SYMBOL_GPL(rds_message_put);
7875e18e	186
7875e18e AG	187	void rds_message_populate_header(struct rds_header *hdr, __be16 sport,
	188	__be16 dport, u64 seq)
	189	{
	190	hdr->h_flags = 0;
	191	hdr->h_sport = sport;
	192	hdr->h_dport = dport;
	193	hdr->h_sequence = cpu_to_be64(seq);
	194	hdr->h_exthdr[0] = RDS_EXTHDR_NONE;
	195	}
616b757a	196	EXPORT_SYMBOL_GPL(rds_message_populate_header);
7875e18e	197
ff51bf84	198	int rds_message_add_extension(struct rds_header *hdr, unsigned int type,
ff51bf84	199	const void *data, unsigned int len)
7875e18e AG	200	{
	201	unsigned int ext_len = sizeof(u8) + len;
	202	unsigned char *dst;
	203
	204	/* For now, refuse to add more than one extension header */
	205	if (hdr->h_exthdr[0] != RDS_EXTHDR_NONE)
	206	return 0;
	207
f64f9e71	208	if (type >= __RDS_EXTHDR_MAX \|\| len != rds_exthdr_size[type])
7875e18e AG	209	return 0;
	210
	211	if (ext_len >= RDS_HEADER_EXT_SPACE)
	212	return 0;
	213	dst = hdr->h_exthdr;
	214
	215	*dst++ = type;
	216	memcpy(dst, data, len);
	217
	218	dst[len] = RDS_EXTHDR_NONE;
	219	return 1;
	220	}
616b757a	221	EXPORT_SYMBOL_GPL(rds_message_add_extension);
7875e18e AG	222
	223	/*
	224	* If a message has extension headers, retrieve them here.
	225	* Call like this:
	226	*
	227	* unsigned int pos = 0;
	228	*
	229	* while (1) {
	230	* buflen = sizeof(buffer);
	231	* type = rds_message_next_extension(hdr, &pos, buffer, &buflen);
	232	* if (type == RDS_EXTHDR_NONE)
	233	* break;
	234	* ...
	235	* }
	236	*/
	237	int rds_message_next_extension(struct rds_header *hdr,
	238	unsigned int pos, void buf, unsigned int *buflen)
	239	{
	240	unsigned int offset, ext_type, ext_len;
	241	u8 *src = hdr->h_exthdr;
	242
	243	offset = *pos;
	244	if (offset >= RDS_HEADER_EXT_SPACE)
	245	goto none;
	246
	247	/* Get the extension type and length. For now, the
	248	* length is implied by the extension type. */
	249	ext_type = src[offset++];
	250
	251	if (ext_type == RDS_EXTHDR_NONE \|\| ext_type >= __RDS_EXTHDR_MAX)
	252	goto none;
	253	ext_len = rds_exthdr_size[ext_type];
	254	if (offset + ext_len > RDS_HEADER_EXT_SPACE)
	255	goto none;
	256
	257	*pos = offset + ext_len;
	258	if (ext_len < *buflen)
	259	*buflen = ext_len;
	260	memcpy(buf, src + offset, *buflen);
	261	return ext_type;
	262
	263	none:
	264	*pos = RDS_HEADER_EXT_SPACE;
	265	*buflen = 0;
	266	return RDS_EXTHDR_NONE;
	267	}
	268
7875e18e AG	269	int rds_message_add_rdma_dest_extension(struct rds_header *hdr, u32 r_key, u32 offset)
	270	{
	271	struct rds_ext_header_rdma_dest ext_hdr;
	272
	273	ext_hdr.h_rdma_rkey = cpu_to_be32(r_key);
	274	ext_hdr.h_rdma_offset = cpu_to_be32(offset);
	275	return rds_message_add_extension(hdr, RDS_EXTHDR_RDMA_DEST, &ext_hdr, sizeof(ext_hdr));
	276	}
616b757a	277	EXPORT_SYMBOL_GPL(rds_message_add_rdma_dest_extension);
7875e18e	278
fc445084 AG	279	/*
	280	* Each rds_message is allocated with extra space for the scatterlist entries
	281	* rds ops will need. This is to minimize memory allocation count. Then, each rds op
	282	* can grab SGs when initializing its part of the rds_message.
	283	*/
	284	struct rds_message *rds_message_alloc(unsigned int extra_len, gfp_t gfp)
7875e18e AG	285	{
	286	struct rds_message *rm;
	287
ece6b0a2 CW	288	if (extra_len > KMALLOC_MAX_SIZE - sizeof(struct rds_message))
	289	return NULL;
	290
fc445084	291	rm = kzalloc(sizeof(struct rds_message) + extra_len, gfp);
7875e18e AG	292	if (!rm)
	293	goto out;
	294
fc445084 AG	295	rm->m_used_sgs = 0;
	296	rm->m_total_sgs = extra_len / sizeof(struct scatterlist);
	297
6c5a1c4a	298	refcount_set(&rm->m_refcount, 1);
7875e18e AG	299	INIT_LIST_HEAD(&rm->m_sock_item);
	300	INIT_LIST_HEAD(&rm->m_conn_item);
	301	spin_lock_init(&rm->m_rs_lock);
c83188dc	302	init_waitqueue_head(&rm->m_flush_wait);
7875e18e AG	303
	304	out:
	305	return rm;
	306	}
	307
fc445084 AG	308	/*
	309	* RDS ops use this to grab SG entries from the rm's sg pool.
	310	*/
7dba9203	311	struct scatterlist rds_message_alloc_sgs(struct rds_message rm, int nents)
fc445084 AG	312	{
	313	struct scatterlist sg_first = (struct scatterlist ) &rm[1];
	314	struct scatterlist *sg_ret;
	315
c75ab8a5	316	if (nents <= 0) {
c75ab8a5	317	pr_warn("rds: alloc sgs failed! nents <= 0\n");
7dba9203	318	return ERR_PTR(-EINVAL);
c75ab8a5	319	}
fc445084	320
c75ab8a5	321	if (rm->m_used_sgs + nents > rm->m_total_sgs) {
	322	pr_warn("rds: alloc sgs failed! total %d used %d nents %d\n",
	323	rm->m_total_sgs, rm->m_used_sgs, nents);
7dba9203	324	return ERR_PTR(-ENOMEM);
c75ab8a5	325	}
d139ff09	326
fc445084	327	sg_ret = &sg_first[rm->m_used_sgs];
f4dd96f7	328	sg_init_table(sg_ret, nents);
fc445084 AG	329	rm->m_used_sgs += nents;
	330
	331	return sg_ret;
	332	}
	333
7875e18e AG	334	struct rds_message rds_message_map_pages(unsigned long page_addrs, unsigned int total_len)
	335	{
	336	struct rds_message *rm;
	337	unsigned int i;
eeb2c4fb	338	int num_sgs = DIV_ROUND_UP(total_len, PAGE_SIZE);
ff87e97a	339	int extra_bytes = num_sgs * sizeof(struct scatterlist);
7875e18e	340
f2ec76f2	341	rm = rds_message_alloc(extra_bytes, GFP_NOWAIT);
8690bfa1	342	if (!rm)
7875e18e AG	343	return ERR_PTR(-ENOMEM);
	344
	345	set_bit(RDS_MSG_PAGEVEC, &rm->m_flags);
	346	rm->m_inc.i_hdr.h_len = cpu_to_be32(total_len);
eeb2c4fb	347	rm->data.op_nents = DIV_ROUND_UP(total_len, PAGE_SIZE);
7dba9203 JG	348	rm->data.op_sg = rds_message_alloc_sgs(rm, num_sgs);
7dba9203 JG	349	if (IS_ERR(rm->data.op_sg)) {
aa58163a	350	rds_message_put(rm);
7dba9203	351	return ERR_CAST(rm->data.op_sg);
aa58163a	352	}
7875e18e	353
6c7cc6e4 AG	354	for (i = 0; i < rm->data.op_nents; ++i) {
6c7cc6e4 AG	355	sg_set_page(&rm->data.op_sg[i],
7875e18e AG	356	virt_to_page(page_addrs[i]),
	357	PAGE_SIZE, 0);
	358	}
	359
	360	return rm;
	361	}
	362
496c7f3c	363	static int rds_message_zcopy_from_user(struct rds_message rm, struct iov_iter from)
7875e18e	364	{
7875e18e	365	struct scatterlist *sg;
fc445084	366	int ret = 0;
0cebacce	367	int length = iov_iter_count(from);
d40a126b	368	int total_copied = 0;
9426bbc6	369	struct rds_msg_zcopy_info *info;
7875e18e	370
083735f4	371	rm->m_inc.i_hdr.h_len = cpu_to_be32(iov_iter_count(from));
7875e18e AG	372
	373	/*
	374	* now allocate and copy in the data payload.
	375	*/
6c7cc6e4	376	sg = rm->data.op_sg;
7875e18e	377
9426bbc6 SV	378	info = kzalloc(sizeof(*info), GFP_KERNEL);
9426bbc6 SV	379	if (!info)
d40a126b	380	return -ENOMEM;
9426bbc6 SV	381	INIT_LIST_HEAD(&info->rs_zcookie_next);
9426bbc6 SV	382	rm->data.op_mmp_znotifier = &info->znotif;
d40a126b SV	383	if (mm_account_pinned_pages(&rm->data.op_mmp_znotifier->z_mmp,
	384	length)) {
	385	ret = -ENOMEM;
	386	goto err;
	387	}
	388	while (iov_iter_count(from)) {
	389	struct page *pages;
	390	size_t start;
	391	ssize_t copied;
	392
	393	copied = iov_iter_get_pages(from, &pages, PAGE_SIZE,
	394	1, &start);
	395	if (copied < 0) {
	396	struct mmpin *mmp;
	397	int i;
	398
	399	for (i = 0; i < rm->data.op_nents; i++)
	400	put_page(sg_page(&rm->data.op_sg[i]));
	401	mmp = &rm->data.op_mmp_znotifier->z_mmp;
	402	mm_unaccount_pinned_pages(mmp);
	403	ret = -EFAULT;
0cebacce SV	404	goto err;
0cebacce SV	405	}
d40a126b SV	406	total_copied += copied;
	407	iov_iter_advance(from, copied);
	408	length -= copied;
	409	sg_set_page(sg, pages, copied, start);
	410	rm->data.op_nents++;
	411	sg++;
	412	}
	413	WARN_ON_ONCE(length != 0);
	414	return ret;
0cebacce	415	err:
9426bbc6	416	kfree(info);
d40a126b SV	417	rm->data.op_mmp_znotifier = NULL;
	418	return ret;
	419	}
	420
	421	int rds_message_copy_from_user(struct rds_message rm, struct iov_iter from,
	422	bool zcopy)
	423	{
	424	unsigned long to_copy, nbytes;
	425	unsigned long sg_off;
	426	struct scatterlist *sg;
	427	int ret = 0;
	428
	429	rm->m_inc.i_hdr.h_len = cpu_to_be32(iov_iter_count(from));
	430
	431	/* now allocate and copy in the data payload. */
	432	sg = rm->data.op_sg;
	433	sg_off = 0; /* Dear gcc, sg->page will be null from kzalloc. */
	434
	435	if (zcopy)
	436	return rds_message_zcopy_from_user(rm, from);
0cebacce	437
083735f4	438	while (iov_iter_count(from)) {
8690bfa1	439	if (!sg_page(sg)) {
083735f4	440	ret = rds_page_remainder_alloc(sg, iov_iter_count(from),
7875e18e AG	441	GFP_HIGHUSER);
7875e18e AG	442	if (ret)
083735f4	443	return ret;
6c7cc6e4	444	rm->data.op_nents++;
7875e18e AG	445	sg_off = 0;
	446	}
	447
083735f4 AV	448	to_copy = min_t(unsigned long, iov_iter_count(from),
083735f4 AV	449	sg->length - sg_off);
7875e18e	450
083735f4	451	rds_stats_add(s_copy_from_user, to_copy);
d0a47d32 SV	452	nbytes = copy_page_from_iter(sg_page(sg), sg->offset + sg_off,
	453	to_copy, from);
	454	if (nbytes != to_copy)
083735f4	455	return -EFAULT;
7875e18e	456
7875e18e AG	457	sg_off += to_copy;
	458
	459	if (sg_off == sg->length)
	460	sg++;
	461	}
	462
fc445084	463	return ret;
7875e18e AG	464	}
7875e18e AG	465
c310e72c	466	int rds_message_inc_copy_to_user(struct rds_incoming inc, struct iov_iter to)
7875e18e AG	467	{
7875e18e AG	468	struct rds_message *rm;
7875e18e AG	469	struct scatterlist *sg;
7875e18e AG	470	unsigned long to_copy;
7875e18e AG	471	unsigned long vec_off;
	472	int copied;
	473	int ret;
	474	u32 len;
	475
	476	rm = container_of(inc, struct rds_message, m_inc);
	477	len = be32_to_cpu(rm->m_inc.i_hdr.h_len);
	478
6c7cc6e4	479	sg = rm->data.op_sg;
7875e18e AG	480	vec_off = 0;
	481	copied = 0;
	482
c310e72c	483	while (iov_iter_count(to) && copied < len) {
6ff4a8ad GU	484	to_copy = min_t(unsigned long, iov_iter_count(to),
6ff4a8ad GU	485	sg->length - vec_off);
7875e18e AG	486	to_copy = min_t(unsigned long, to_copy, len - copied);
7875e18e AG	487
c310e72c AV	488	rds_stats_add(s_copy_to_user, to_copy);
	489	ret = copy_page_to_iter(sg_page(sg), sg->offset + vec_off,
	490	to_copy, to);
	491	if (ret != to_copy)
	492	return -EFAULT;
7875e18e	493
7875e18e AG	494	vec_off += to_copy;
	495	copied += to_copy;
	496
	497	if (vec_off == sg->length) {
	498	vec_off = 0;
	499	sg++;
	500	}
	501	}
	502
	503	return copied;
	504	}
	505
	506	/*
	507	* If the message is still on the send queue, wait until the transport
	508	* is done with it. This is particularly important for RDMA operations.
	509	*/
	510	void rds_message_wait(struct rds_message *rm)
	511	{
c83188dc	512	wait_event_interruptible(rm->m_flush_wait,
7875e18e AG	513	!test_bit(RDS_MSG_MAPPED, &rm->m_flags));
	514	}
	515
	516	void rds_message_unmapped(struct rds_message *rm)
	517	{
	518	clear_bit(RDS_MSG_MAPPED, &rm->m_flags);
c83188dc	519	wake_up_interruptible(&rm->m_flush_wait);
7875e18e	520	}
616b757a	521	EXPORT_SYMBOL_GPL(rds_message_unmapped);