2 * Copyright (c) 2006 Oracle. All rights reserved.
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:
10 * Redistribution and use in source and binary forms, with or
11 * without modification, are permitted provided that the following
14 * - Redistributions of source code must retain the above
15 * copyright notice, this list of conditions and the following
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.
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
33 #include <linux/kernel.h>
34 #include <linux/slab.h>
39 static DECLARE_WAIT_QUEUE_HEAD(rds_message_flush_waitq
);
41 static 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
),
49 void rds_message_addref(struct rds_message
*rm
)
51 rdsdebug("addref rm %p ref %d\n", rm
, atomic_read(&rm
->m_refcount
));
52 atomic_inc(&rm
->m_refcount
);
54 EXPORT_SYMBOL_GPL(rds_message_addref
);
57 * This relies on dma_map_sg() not touching sg[].page during merging.
59 static void rds_message_purge(struct rds_message
*rm
)
63 if (unlikely(test_bit(RDS_MSG_PAGEVEC
, &rm
->m_flags
)))
66 for (i
= 0; i
< rm
->m_nents
; i
++) {
67 rdsdebug("putting data page %p\n", (void *)sg_page(&rm
->m_sg
[i
]));
68 /* XXX will have to put_page for page refs */
69 __free_page(sg_page(&rm
->m_sg
[i
]));
74 rds_rdma_free_op(rm
->m_rdma_op
);
76 rds_mr_put(rm
->m_rdma_mr
);
79 void rds_message_inc_purge(struct rds_incoming
*inc
)
81 struct rds_message
*rm
= container_of(inc
, struct rds_message
, m_inc
);
82 rds_message_purge(rm
);
85 void rds_message_put(struct rds_message
*rm
)
87 rdsdebug("put rm %p ref %d\n", rm
, atomic_read(&rm
->m_refcount
));
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
);
97 EXPORT_SYMBOL_GPL(rds_message_put
);
99 void rds_message_inc_free(struct rds_incoming
*inc
)
101 struct rds_message
*rm
= container_of(inc
, struct rds_message
, m_inc
);
105 void rds_message_populate_header(struct rds_header
*hdr
, __be16 sport
,
106 __be16 dport
, u64 seq
)
109 hdr
->h_sport
= sport
;
110 hdr
->h_dport
= dport
;
111 hdr
->h_sequence
= cpu_to_be64(seq
);
112 hdr
->h_exthdr
[0] = RDS_EXTHDR_NONE
;
114 EXPORT_SYMBOL_GPL(rds_message_populate_header
);
116 int rds_message_add_extension(struct rds_header
*hdr
,
117 unsigned int type
, const void *data
, unsigned int len
)
119 unsigned int ext_len
= sizeof(u8
) + len
;
122 /* For now, refuse to add more than one extension header */
123 if (hdr
->h_exthdr
[0] != RDS_EXTHDR_NONE
)
126 if (type
>= __RDS_EXTHDR_MAX
|| len
!= rds_exthdr_size
[type
])
129 if (ext_len
>= RDS_HEADER_EXT_SPACE
)
134 memcpy(dst
, data
, len
);
136 dst
[len
] = RDS_EXTHDR_NONE
;
139 EXPORT_SYMBOL_GPL(rds_message_add_extension
);
142 * If a message has extension headers, retrieve them here.
145 * unsigned int pos = 0;
148 * buflen = sizeof(buffer);
149 * type = rds_message_next_extension(hdr, &pos, buffer, &buflen);
150 * if (type == RDS_EXTHDR_NONE)
155 int rds_message_next_extension(struct rds_header
*hdr
,
156 unsigned int *pos
, void *buf
, unsigned int *buflen
)
158 unsigned int offset
, ext_type
, ext_len
;
159 u8
*src
= hdr
->h_exthdr
;
162 if (offset
>= RDS_HEADER_EXT_SPACE
)
165 /* Get the extension type and length. For now, the
166 * length is implied by the extension type. */
167 ext_type
= src
[offset
++];
169 if (ext_type
== RDS_EXTHDR_NONE
|| ext_type
>= __RDS_EXTHDR_MAX
)
171 ext_len
= rds_exthdr_size
[ext_type
];
172 if (offset
+ ext_len
> RDS_HEADER_EXT_SPACE
)
175 *pos
= offset
+ ext_len
;
176 if (ext_len
< *buflen
)
178 memcpy(buf
, src
+ offset
, *buflen
);
182 *pos
= RDS_HEADER_EXT_SPACE
;
184 return RDS_EXTHDR_NONE
;
187 int rds_message_add_version_extension(struct rds_header
*hdr
, unsigned int version
)
189 struct rds_ext_header_version ext_hdr
;
191 ext_hdr
.h_version
= cpu_to_be32(version
);
192 return rds_message_add_extension(hdr
, RDS_EXTHDR_VERSION
, &ext_hdr
, sizeof(ext_hdr
));
195 int rds_message_get_version_extension(struct rds_header
*hdr
, unsigned int *version
)
197 struct rds_ext_header_version ext_hdr
;
198 unsigned int pos
= 0, len
= sizeof(ext_hdr
);
200 /* We assume the version extension is the only one present */
201 if (rds_message_next_extension(hdr
, &pos
, &ext_hdr
, &len
) != RDS_EXTHDR_VERSION
)
203 *version
= be32_to_cpu(ext_hdr
.h_version
);
207 int rds_message_add_rdma_dest_extension(struct rds_header
*hdr
, u32 r_key
, u32 offset
)
209 struct rds_ext_header_rdma_dest ext_hdr
;
211 ext_hdr
.h_rdma_rkey
= cpu_to_be32(r_key
);
212 ext_hdr
.h_rdma_offset
= cpu_to_be32(offset
);
213 return rds_message_add_extension(hdr
, RDS_EXTHDR_RDMA_DEST
, &ext_hdr
, sizeof(ext_hdr
));
215 EXPORT_SYMBOL_GPL(rds_message_add_rdma_dest_extension
);
217 struct rds_message
*rds_message_alloc(unsigned int nents
, gfp_t gfp
)
219 struct rds_message
*rm
;
221 rm
= kzalloc(sizeof(struct rds_message
) +
222 (nents
* sizeof(struct scatterlist
)), gfp
);
227 sg_init_table(rm
->m_sg
, nents
);
228 atomic_set(&rm
->m_refcount
, 1);
229 INIT_LIST_HEAD(&rm
->m_sock_item
);
230 INIT_LIST_HEAD(&rm
->m_conn_item
);
231 spin_lock_init(&rm
->m_rs_lock
);
237 struct rds_message
*rds_message_map_pages(unsigned long *page_addrs
, unsigned int total_len
)
239 struct rds_message
*rm
;
242 rm
= rds_message_alloc(ceil(total_len
, PAGE_SIZE
), GFP_KERNEL
);
244 return ERR_PTR(-ENOMEM
);
246 set_bit(RDS_MSG_PAGEVEC
, &rm
->m_flags
);
247 rm
->m_inc
.i_hdr
.h_len
= cpu_to_be32(total_len
);
248 rm
->m_nents
= ceil(total_len
, PAGE_SIZE
);
250 for (i
= 0; i
< rm
->m_nents
; ++i
) {
251 sg_set_page(&rm
->m_sg
[i
],
252 virt_to_page(page_addrs
[i
]),
259 struct rds_message
*rds_message_copy_from_user(struct iovec
*first_iov
,
262 unsigned long to_copy
;
263 unsigned long iov_off
;
264 unsigned long sg_off
;
265 struct rds_message
*rm
;
267 struct scatterlist
*sg
;
270 rm
= rds_message_alloc(ceil(total_len
, PAGE_SIZE
), GFP_KERNEL
);
276 rm
->m_inc
.i_hdr
.h_len
= cpu_to_be32(total_len
);
279 * now allocate and copy in the data payload.
284 sg_off
= 0; /* Dear gcc, sg->page will be null from kzalloc. */
287 if (sg_page(sg
) == NULL
) {
288 ret
= rds_page_remainder_alloc(sg
, total_len
,
296 while (iov_off
== iov
->iov_len
) {
301 to_copy
= min(iov
->iov_len
- iov_off
, sg
->length
- sg_off
);
302 to_copy
= min_t(size_t, to_copy
, total_len
);
304 rdsdebug("copying %lu bytes from user iov [%p, %zu] + %lu to "
305 "sg [%p, %u, %u] + %lu\n",
306 to_copy
, iov
->iov_base
, iov
->iov_len
, iov_off
,
307 (void *)sg_page(sg
), sg
->offset
, sg
->length
, sg_off
);
309 ret
= rds_page_copy_from_user(sg_page(sg
), sg
->offset
+ sg_off
,
310 iov
->iov_base
+ iov_off
,
316 total_len
-= to_copy
;
319 if (sg_off
== sg
->length
)
333 int rds_message_inc_copy_to_user(struct rds_incoming
*inc
,
334 struct iovec
*first_iov
, size_t size
)
336 struct rds_message
*rm
;
338 struct scatterlist
*sg
;
339 unsigned long to_copy
;
340 unsigned long iov_off
;
341 unsigned long vec_off
;
346 rm
= container_of(inc
, struct rds_message
, m_inc
);
347 len
= be32_to_cpu(rm
->m_inc
.i_hdr
.h_len
);
355 while (copied
< size
&& copied
< len
) {
356 while (iov_off
== iov
->iov_len
) {
361 to_copy
= min(iov
->iov_len
- iov_off
, sg
->length
- vec_off
);
362 to_copy
= min_t(size_t, to_copy
, size
- copied
);
363 to_copy
= min_t(unsigned long, to_copy
, len
- copied
);
365 rdsdebug("copying %lu bytes to user iov [%p, %zu] + %lu to "
366 "sg [%p, %u, %u] + %lu\n",
367 to_copy
, iov
->iov_base
, iov
->iov_len
, iov_off
,
368 sg_page(sg
), sg
->offset
, sg
->length
, vec_off
);
370 ret
= rds_page_copy_to_user(sg_page(sg
), sg
->offset
+ vec_off
,
371 iov
->iov_base
+ iov_off
,
382 if (vec_off
== sg
->length
) {
392 * If the message is still on the send queue, wait until the transport
393 * is done with it. This is particularly important for RDMA operations.
395 void rds_message_wait(struct rds_message
*rm
)
397 wait_event(rds_message_flush_waitq
,
398 !test_bit(RDS_MSG_MAPPED
, &rm
->m_flags
));
401 void rds_message_unmapped(struct rds_message
*rm
)
403 clear_bit(RDS_MSG_MAPPED
, &rm
->m_flags
);
404 if (waitqueue_active(&rds_message_flush_waitq
))
405 wake_up(&rds_message_flush_waitq
);
407 EXPORT_SYMBOL_GPL(rds_message_unmapped
);