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>
38 static DECLARE_WAIT_QUEUE_HEAD(rds_message_flush_waitq);
40 static unsigned int rds_exthdr_size[__RDS_EXTHDR_MAX] = {
41 [RDS_EXTHDR_NONE] = 0,
42 [RDS_EXTHDR_VERSION] = sizeof(struct rds_ext_header_version),
43 [RDS_EXTHDR_RDMA] = sizeof(struct rds_ext_header_rdma),
44 [RDS_EXTHDR_RDMA_DEST] = sizeof(struct rds_ext_header_rdma_dest),
48 void rds_message_addref(struct rds_message *rm)
50 rdsdebug("addref rm %p ref %d\n", rm, atomic_read(&rm->m_refcount));
51 atomic_inc(&rm->m_refcount);
53 EXPORT_SYMBOL_GPL(rds_message_addref);
56 * This relies on dma_map_sg() not touching sg[].page during merging.
58 static void rds_message_purge(struct rds_message *rm)
62 if (unlikely(test_bit(RDS_MSG_PAGEVEC, &rm->m_flags)))
65 for (i = 0; i < rm->data.m_nents; i++) {
66 rdsdebug("putting data page %p\n", (void *)sg_page(&rm->data.m_sg[i]));
67 /* XXX will have to put_page for page refs */
68 __free_page(sg_page(&rm->data.m_sg[i]));
72 if (rm->rdma.m_rdma_op.r_active)
73 rds_rdma_free_op(&rm->rdma.m_rdma_op);
74 if (rm->rdma.m_rdma_mr)
75 rds_mr_put(rm->rdma.m_rdma_mr);
78 void rds_message_put(struct rds_message *rm)
80 rdsdebug("put rm %p ref %d\n", rm, atomic_read(&rm->m_refcount));
82 if (atomic_dec_and_test(&rm->m_refcount)) {
83 BUG_ON(!list_empty(&rm->m_sock_item));
84 BUG_ON(!list_empty(&rm->m_conn_item));
85 rds_message_purge(rm);
90 EXPORT_SYMBOL_GPL(rds_message_put);
92 void rds_message_populate_header(struct rds_header *hdr, __be16 sport,
93 __be16 dport, u64 seq)
98 hdr->h_sequence = cpu_to_be64(seq);
99 hdr->h_exthdr[0] = RDS_EXTHDR_NONE;
101 EXPORT_SYMBOL_GPL(rds_message_populate_header);
103 int rds_message_add_extension(struct rds_header *hdr,
104 unsigned int type, const void *data, unsigned int len)
106 unsigned int ext_len = sizeof(u8) + len;
109 /* For now, refuse to add more than one extension header */
110 if (hdr->h_exthdr[0] != RDS_EXTHDR_NONE)
113 if (type >= __RDS_EXTHDR_MAX || len != rds_exthdr_size[type])
116 if (ext_len >= RDS_HEADER_EXT_SPACE)
121 memcpy(dst, data, len);
123 dst[len] = RDS_EXTHDR_NONE;
126 EXPORT_SYMBOL_GPL(rds_message_add_extension);
129 * If a message has extension headers, retrieve them here.
132 * unsigned int pos = 0;
135 * buflen = sizeof(buffer);
136 * type = rds_message_next_extension(hdr, &pos, buffer, &buflen);
137 * if (type == RDS_EXTHDR_NONE)
142 int rds_message_next_extension(struct rds_header *hdr,
143 unsigned int *pos, void *buf, unsigned int *buflen)
145 unsigned int offset, ext_type, ext_len;
146 u8 *src = hdr->h_exthdr;
149 if (offset >= RDS_HEADER_EXT_SPACE)
152 /* Get the extension type and length. For now, the
153 * length is implied by the extension type. */
154 ext_type = src[offset++];
156 if (ext_type == RDS_EXTHDR_NONE || ext_type >= __RDS_EXTHDR_MAX)
158 ext_len = rds_exthdr_size[ext_type];
159 if (offset + ext_len > RDS_HEADER_EXT_SPACE)
162 *pos = offset + ext_len;
163 if (ext_len < *buflen)
165 memcpy(buf, src + offset, *buflen);
169 *pos = RDS_HEADER_EXT_SPACE;
171 return RDS_EXTHDR_NONE;
174 int rds_message_add_version_extension(struct rds_header *hdr, unsigned int version)
176 struct rds_ext_header_version ext_hdr;
178 ext_hdr.h_version = cpu_to_be32(version);
179 return rds_message_add_extension(hdr, RDS_EXTHDR_VERSION, &ext_hdr, sizeof(ext_hdr));
182 int rds_message_get_version_extension(struct rds_header *hdr, unsigned int *version)
184 struct rds_ext_header_version ext_hdr;
185 unsigned int pos = 0, len = sizeof(ext_hdr);
187 /* We assume the version extension is the only one present */
188 if (rds_message_next_extension(hdr, &pos, &ext_hdr, &len) != RDS_EXTHDR_VERSION)
190 *version = be32_to_cpu(ext_hdr.h_version);
194 int rds_message_add_rdma_dest_extension(struct rds_header *hdr, u32 r_key, u32 offset)
196 struct rds_ext_header_rdma_dest ext_hdr;
198 ext_hdr.h_rdma_rkey = cpu_to_be32(r_key);
199 ext_hdr.h_rdma_offset = cpu_to_be32(offset);
200 return rds_message_add_extension(hdr, RDS_EXTHDR_RDMA_DEST, &ext_hdr, sizeof(ext_hdr));
202 EXPORT_SYMBOL_GPL(rds_message_add_rdma_dest_extension);
205 * Each rds_message is allocated with extra space for the scatterlist entries
206 * rds ops will need. This is to minimize memory allocation count. Then, each rds op
207 * can grab SGs when initializing its part of the rds_message.
209 struct rds_message *rds_message_alloc(unsigned int extra_len, gfp_t gfp)
211 struct rds_message *rm;
213 rm = kzalloc(sizeof(struct rds_message) + extra_len, gfp);
218 rm->m_total_sgs = extra_len / sizeof(struct scatterlist);
220 atomic_set(&rm->m_refcount, 1);
221 INIT_LIST_HEAD(&rm->m_sock_item);
222 INIT_LIST_HEAD(&rm->m_conn_item);
223 spin_lock_init(&rm->m_rs_lock);
230 * RDS ops use this to grab SG entries from the rm's sg pool.
232 struct scatterlist *rds_message_alloc_sgs(struct rds_message *rm, int nents)
234 struct scatterlist *sg_first = (struct scatterlist *) &rm[1];
235 struct scatterlist *sg_ret;
237 WARN_ON(rm->m_used_sgs + nents > rm->m_total_sgs);
239 sg_ret = &sg_first[rm->m_used_sgs];
240 sg_init_table(sg_ret, nents);
241 rm->m_used_sgs += nents;
246 struct rds_message *rds_message_map_pages(unsigned long *page_addrs, unsigned int total_len)
248 struct rds_message *rm;
250 int num_sgs = ceil(total_len, PAGE_SIZE);
251 int extra_bytes = num_sgs * sizeof(struct scatterlist);
253 rm = rds_message_alloc(extra_bytes, GFP_KERNEL);
255 return ERR_PTR(-ENOMEM);
257 set_bit(RDS_MSG_PAGEVEC, &rm->m_flags);
258 rm->m_inc.i_hdr.h_len = cpu_to_be32(total_len);
259 rm->data.m_nents = ceil(total_len, PAGE_SIZE);
260 rm->data.m_sg = rds_message_alloc_sgs(rm, num_sgs);
262 for (i = 0; i < rm->data.m_nents; ++i) {
263 sg_set_page(&rm->data.m_sg[i],
264 virt_to_page(page_addrs[i]),
271 int rds_message_copy_from_user(struct rds_message *rm, struct iovec *first_iov,
274 unsigned long to_copy;
275 unsigned long iov_off;
276 unsigned long sg_off;
278 struct scatterlist *sg;
281 rm->m_inc.i_hdr.h_len = cpu_to_be32(total_len);
284 * now allocate and copy in the data payload.
289 sg_off = 0; /* Dear gcc, sg->page will be null from kzalloc. */
293 ret = rds_page_remainder_alloc(sg, total_len,
301 while (iov_off == iov->iov_len) {
306 to_copy = min(iov->iov_len - iov_off, sg->length - sg_off);
307 to_copy = min_t(size_t, to_copy, total_len);
309 rdsdebug("copying %lu bytes from user iov [%p, %zu] + %lu to "
310 "sg [%p, %u, %u] + %lu\n",
311 to_copy, iov->iov_base, iov->iov_len, iov_off,
312 (void *)sg_page(sg), sg->offset, sg->length, sg_off);
314 ret = rds_page_copy_from_user(sg_page(sg), sg->offset + sg_off,
315 iov->iov_base + iov_off,
321 total_len -= to_copy;
324 if (sg_off == sg->length)
332 int rds_message_inc_copy_to_user(struct rds_incoming *inc,
333 struct iovec *first_iov, size_t size)
335 struct rds_message *rm;
337 struct scatterlist *sg;
338 unsigned long to_copy;
339 unsigned long iov_off;
340 unsigned long vec_off;
345 rm = container_of(inc, struct rds_message, m_inc);
346 len = be32_to_cpu(rm->m_inc.i_hdr.h_len);
354 while (copied < size && copied < len) {
355 while (iov_off == iov->iov_len) {
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);
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);
369 ret = rds_page_copy_to_user(sg_page(sg), sg->offset + vec_off,
370 iov->iov_base + iov_off,
381 if (vec_off == sg->length) {
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.
394 void rds_message_wait(struct rds_message *rm)
396 wait_event(rds_message_flush_waitq,
397 !test_bit(RDS_MSG_MAPPED, &rm->m_flags));
400 void rds_message_unmapped(struct rds_message *rm)
402 clear_bit(RDS_MSG_MAPPED, &rm->m_flags);
403 if (waitqueue_active(&rds_message_flush_waitq))
404 wake_up(&rds_message_flush_waitq);
406 EXPORT_SYMBOL_GPL(rds_message_unmapped);