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
36#include "rds.h"
37#include "rdma.h"
38
39static DECLARE_WAIT_QUEUE_HEAD(rds_message_flush_waitq);
40
41static unsigned int	rds_exthdr_size[__RDS_EXTHDR_MAX] = {
42[RDS_EXTHDR_NONE]	= 0,
43[RDS_EXTHDR_VERSION]	= sizeof(struct rds_ext_header_version),
44[RDS_EXTHDR_RDMA]	= sizeof(struct rds_ext_header_rdma),
45[RDS_EXTHDR_RDMA_DEST]	= sizeof(struct rds_ext_header_rdma_dest),
46};
47
48
49void rds_message_addref(struct rds_message *rm)
50{
51	rdsdebug("addref rm %p ref %d\n", rm, atomic_read(&rm->m_refcount));
52	atomic_inc(&rm->m_refcount);
53}
54EXPORT_SYMBOL_GPL(rds_message_addref);
55
56/*
57 * This relies on dma_map_sg() not touching sg[].page during merging.
58 */
59static void rds_message_purge(struct rds_message *rm)
60{
61	unsigned long i;
62
63	if (unlikely(test_bit(RDS_MSG_PAGEVEC, &rm->m_flags)))
64		return;
65
66	for (i = 0; i < rm->m_nents; i++) {
67		rdsdebug("putting data page %p\n", (void *)sg_page(&rm->m_sg[i]));
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
78void 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
84void 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}
96EXPORT_SYMBOL_GPL(rds_message_put);
97
98void 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
104void 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}
113EXPORT_SYMBOL_GPL(rds_message_populate_header);
114
115int 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
125	if (type >= __RDS_EXTHDR_MAX || len != rds_exthdr_size[type])
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}
138EXPORT_SYMBOL_GPL(rds_message_add_extension);
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 */
154int 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
180none:
181	*pos = RDS_HEADER_EXT_SPACE;
182	*buflen = 0;
183	return RDS_EXTHDR_NONE;
184}
185
186int 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
194int 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
206int 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}
214EXPORT_SYMBOL_GPL(rds_message_add_rdma_dest_extension);
215
216struct 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
232out:
233	return rm;
234}
235
236struct 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
258struct 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;
323out:
324	if (ret) {
325		if (rm)
326			rds_message_put(rm);
327		rm = ERR_PTR(ret);
328	}
329	return rm;
330}
331
332int 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 */
394void 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
400void 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}
406EXPORT_SYMBOL_GPL(rds_message_unmapped);
407