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>
35 #include <linux/export.h>
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
),
47 void rds_message_addref(struct rds_message
*rm
)
49 rdsdebug("addref rm %p ref %d\n", rm
, atomic_read(&rm
->m_refcount
));
50 atomic_inc(&rm
->m_refcount
);
52 EXPORT_SYMBOL_GPL(rds_message_addref
);
55 * This relies on dma_map_sg() not touching sg[].page during merging.
57 static void rds_message_purge(struct rds_message
*rm
)
61 if (unlikely(test_bit(RDS_MSG_PAGEVEC
, &rm
->m_flags
)))
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
]));
69 rm
->data
.op_nents
= 0;
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
);
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
);
82 void rds_message_put(struct rds_message
*rm
)
84 rdsdebug("put rm %p ref %d\n", rm
, atomic_read(&rm
->m_refcount
));
85 if (atomic_read(&rm
->m_refcount
) == 0) {
86 printk(KERN_CRIT
"danger refcount zero on %p\n", rm
);
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_populate_header(struct rds_header
*hdr
, __be16 sport
,
100 __be16 dport
, u64 seq
)
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
;
108 EXPORT_SYMBOL_GPL(rds_message_populate_header
);
110 int rds_message_add_extension(struct rds_header
*hdr
, unsigned int type
,
111 const void *data
, unsigned int len
)
113 unsigned int ext_len
= sizeof(u8
) + len
;
116 /* For now, refuse to add more than one extension header */
117 if (hdr
->h_exthdr
[0] != RDS_EXTHDR_NONE
)
120 if (type
>= __RDS_EXTHDR_MAX
|| len
!= rds_exthdr_size
[type
])
123 if (ext_len
>= RDS_HEADER_EXT_SPACE
)
128 memcpy(dst
, data
, len
);
130 dst
[len
] = RDS_EXTHDR_NONE
;
133 EXPORT_SYMBOL_GPL(rds_message_add_extension
);
136 * If a message has extension headers, retrieve them here.
139 * unsigned int pos = 0;
142 * buflen = sizeof(buffer);
143 * type = rds_message_next_extension(hdr, &pos, buffer, &buflen);
144 * if (type == RDS_EXTHDR_NONE)
149 int rds_message_next_extension(struct rds_header
*hdr
,
150 unsigned int *pos
, void *buf
, unsigned int *buflen
)
152 unsigned int offset
, ext_type
, ext_len
;
153 u8
*src
= hdr
->h_exthdr
;
156 if (offset
>= RDS_HEADER_EXT_SPACE
)
159 /* Get the extension type and length. For now, the
160 * length is implied by the extension type. */
161 ext_type
= src
[offset
++];
163 if (ext_type
== RDS_EXTHDR_NONE
|| ext_type
>= __RDS_EXTHDR_MAX
)
165 ext_len
= rds_exthdr_size
[ext_type
];
166 if (offset
+ ext_len
> RDS_HEADER_EXT_SPACE
)
169 *pos
= offset
+ ext_len
;
170 if (ext_len
< *buflen
)
172 memcpy(buf
, src
+ offset
, *buflen
);
176 *pos
= RDS_HEADER_EXT_SPACE
;
178 return RDS_EXTHDR_NONE
;
181 int rds_message_add_rdma_dest_extension(struct rds_header
*hdr
, u32 r_key
, u32 offset
)
183 struct rds_ext_header_rdma_dest ext_hdr
;
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
));
189 EXPORT_SYMBOL_GPL(rds_message_add_rdma_dest_extension
);
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.
196 struct rds_message
*rds_message_alloc(unsigned int extra_len
, gfp_t gfp
)
198 struct rds_message
*rm
;
200 rm
= kzalloc(sizeof(struct rds_message
) + extra_len
, gfp
);
205 rm
->m_total_sgs
= extra_len
/ sizeof(struct scatterlist
);
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
);
218 * RDS ops use this to grab SG entries from the rm's sg pool.
220 struct scatterlist
*rds_message_alloc_sgs(struct rds_message
*rm
, int nents
)
222 struct scatterlist
*sg_first
= (struct scatterlist
*) &rm
[1];
223 struct scatterlist
*sg_ret
;
225 WARN_ON(rm
->m_used_sgs
+ nents
> rm
->m_total_sgs
);
228 if (rm
->m_used_sgs
+ nents
> rm
->m_total_sgs
)
231 sg_ret
= &sg_first
[rm
->m_used_sgs
];
232 sg_init_table(sg_ret
, nents
);
233 rm
->m_used_sgs
+= nents
;
238 struct rds_message
*rds_message_map_pages(unsigned long *page_addrs
, unsigned int total_len
)
240 struct rds_message
*rm
;
242 int num_sgs
= ceil(total_len
, PAGE_SIZE
);
243 int extra_bytes
= num_sgs
* sizeof(struct scatterlist
);
245 rm
= rds_message_alloc(extra_bytes
, GFP_NOWAIT
);
247 return ERR_PTR(-ENOMEM
);
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
) {
255 return ERR_PTR(-ENOMEM
);
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
]),
267 int rds_message_copy_from_user(struct rds_message
*rm
, struct iovec
*first_iov
,
270 unsigned long to_copy
;
271 unsigned long iov_off
;
272 unsigned long sg_off
;
274 struct scatterlist
*sg
;
277 rm
->m_inc
.i_hdr
.h_len
= cpu_to_be32(total_len
);
280 * now allocate and copy in the data payload.
285 sg_off
= 0; /* Dear gcc, sg->page will be null from kzalloc. */
289 ret
= rds_page_remainder_alloc(sg
, total_len
,
297 while (iov_off
== iov
->iov_len
) {
302 to_copy
= min(iov
->iov_len
- iov_off
, sg
->length
- sg_off
);
303 to_copy
= min_t(size_t, to_copy
, total_len
);
305 rdsdebug("copying %lu bytes from user iov [%p, %zu] + %lu to "
306 "sg [%p, %u, %u] + %lu\n",
307 to_copy
, iov
->iov_base
, iov
->iov_len
, iov_off
,
308 (void *)sg_page(sg
), sg
->offset
, sg
->length
, sg_off
);
310 ret
= rds_page_copy_from_user(sg_page(sg
), sg
->offset
+ sg_off
,
311 iov
->iov_base
+ iov_off
,
317 total_len
-= to_copy
;
320 if (sg_off
== sg
->length
)
328 int rds_message_inc_copy_to_user(struct rds_incoming
*inc
,
329 struct iovec
*first_iov
, size_t size
)
331 struct rds_message
*rm
;
333 struct scatterlist
*sg
;
334 unsigned long to_copy
;
335 unsigned long iov_off
;
336 unsigned long vec_off
;
341 rm
= container_of(inc
, struct rds_message
, m_inc
);
342 len
= be32_to_cpu(rm
->m_inc
.i_hdr
.h_len
);
350 while (copied
< size
&& copied
< len
) {
351 while (iov_off
== iov
->iov_len
) {
356 to_copy
= min(iov
->iov_len
- iov_off
, sg
->length
- vec_off
);
357 to_copy
= min_t(size_t, to_copy
, size
- copied
);
358 to_copy
= min_t(unsigned long, to_copy
, len
- copied
);
360 rdsdebug("copying %lu bytes to user iov [%p, %zu] + %lu to "
361 "sg [%p, %u, %u] + %lu\n",
362 to_copy
, iov
->iov_base
, iov
->iov_len
, iov_off
,
363 sg_page(sg
), sg
->offset
, sg
->length
, vec_off
);
365 ret
= rds_page_copy_to_user(sg_page(sg
), sg
->offset
+ vec_off
,
366 iov
->iov_base
+ iov_off
,
377 if (vec_off
== sg
->length
) {
387 * If the message is still on the send queue, wait until the transport
388 * is done with it. This is particularly important for RDMA operations.
390 void rds_message_wait(struct rds_message
*rm
)
392 wait_event_interruptible(rm
->m_flush_wait
,
393 !test_bit(RDS_MSG_MAPPED
, &rm
->m_flags
));
396 void rds_message_unmapped(struct rds_message
*rm
)
398 clear_bit(RDS_MSG_MAPPED
, &rm
->m_flags
);
399 wake_up_interruptible(&rm
->m_flush_wait
);
401 EXPORT_SYMBOL_GPL(rds_message_unmapped
);