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