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 unsigned int rds_exthdr_size[__RDS_EXTHDR_MAX] = {
39 [RDS_EXTHDR_NONE] = 0,
40 [RDS_EXTHDR_VERSION] = sizeof(struct rds_ext_header_version),
41 [RDS_EXTHDR_RDMA] = sizeof(struct rds_ext_header_rdma),
42 [RDS_EXTHDR_RDMA_DEST] = sizeof(struct rds_ext_header_rdma_dest),
46 void rds_message_addref(struct rds_message *rm)
48 rdsdebug("addref rm %p ref %d\n", rm, atomic_read(&rm->m_refcount));
49 atomic_inc(&rm->m_refcount);
51 EXPORT_SYMBOL_GPL(rds_message_addref);
54 * This relies on dma_map_sg() not touching sg[].page during merging.
56 static void rds_message_purge(struct rds_message *rm)
60 if (unlikely(test_bit(RDS_MSG_PAGEVEC, &rm->m_flags)))
63 for (i = 0; i < rm->data.op_nents; i++) {
64 rdsdebug("putting data page %p\n", (void *)sg_page(&rm->data.op_sg[i]));
65 /* XXX will have to put_page for page refs */
66 __free_page(sg_page(&rm->data.op_sg[i]));
68 rm->data.op_nents = 0;
70 if (rm->rdma.op_active)
71 rds_rdma_free_op(&rm->rdma);
72 if (rm->rdma.op_rdma_mr)
73 rds_mr_put(rm->rdma.op_rdma_mr);
75 if (rm->atomic.op_active)
76 rds_atomic_free_op(&rm->atomic);
77 if (rm->atomic.op_rdma_mr)
78 rds_mr_put(rm->atomic.op_rdma_mr);
81 void rds_message_put(struct rds_message *rm)
83 rdsdebug("put rm %p ref %d\n", rm, atomic_read(&rm->m_refcount));
84 if (atomic_read(&rm->m_refcount) == 0) {
85 printk(KERN_CRIT "danger refcount zero on %p\n", rm);
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);
96 EXPORT_SYMBOL_GPL(rds_message_put);
98 void rds_message_populate_header(struct rds_header *hdr, __be16 sport,
99 __be16 dport, u64 seq)
102 hdr->h_sport = sport;
103 hdr->h_dport = dport;
104 hdr->h_sequence = cpu_to_be64(seq);
105 hdr->h_exthdr[0] = RDS_EXTHDR_NONE;
107 EXPORT_SYMBOL_GPL(rds_message_populate_header);
109 int rds_message_add_extension(struct rds_header *hdr, unsigned int type,
110 const void *data, unsigned int len)
112 unsigned int ext_len = sizeof(u8) + len;
115 /* For now, refuse to add more than one extension header */
116 if (hdr->h_exthdr[0] != RDS_EXTHDR_NONE)
119 if (type >= __RDS_EXTHDR_MAX || len != rds_exthdr_size[type])
122 if (ext_len >= RDS_HEADER_EXT_SPACE)
127 memcpy(dst, data, len);
129 dst[len] = RDS_EXTHDR_NONE;
132 EXPORT_SYMBOL_GPL(rds_message_add_extension);
135 * If a message has extension headers, retrieve them here.
138 * unsigned int pos = 0;
141 * buflen = sizeof(buffer);
142 * type = rds_message_next_extension(hdr, &pos, buffer, &buflen);
143 * if (type == RDS_EXTHDR_NONE)
148 int rds_message_next_extension(struct rds_header *hdr,
149 unsigned int *pos, void *buf, unsigned int *buflen)
151 unsigned int offset, ext_type, ext_len;
152 u8 *src = hdr->h_exthdr;
155 if (offset >= RDS_HEADER_EXT_SPACE)
158 /* Get the extension type and length. For now, the
159 * length is implied by the extension type. */
160 ext_type = src[offset++];
162 if (ext_type == RDS_EXTHDR_NONE || ext_type >= __RDS_EXTHDR_MAX)
164 ext_len = rds_exthdr_size[ext_type];
165 if (offset + ext_len > RDS_HEADER_EXT_SPACE)
168 *pos = offset + ext_len;
169 if (ext_len < *buflen)
171 memcpy(buf, src + offset, *buflen);
175 *pos = RDS_HEADER_EXT_SPACE;
177 return RDS_EXTHDR_NONE;
180 int rds_message_add_rdma_dest_extension(struct rds_header *hdr, u32 r_key, u32 offset)
182 struct rds_ext_header_rdma_dest ext_hdr;
184 ext_hdr.h_rdma_rkey = cpu_to_be32(r_key);
185 ext_hdr.h_rdma_offset = cpu_to_be32(offset);
186 return rds_message_add_extension(hdr, RDS_EXTHDR_RDMA_DEST, &ext_hdr, sizeof(ext_hdr));
188 EXPORT_SYMBOL_GPL(rds_message_add_rdma_dest_extension);
191 * Each rds_message is allocated with extra space for the scatterlist entries
192 * rds ops will need. This is to minimize memory allocation count. Then, each rds op
193 * can grab SGs when initializing its part of the rds_message.
195 struct rds_message *rds_message_alloc(unsigned int extra_len, gfp_t gfp)
197 struct rds_message *rm;
199 rm = kzalloc(sizeof(struct rds_message) + extra_len, gfp);
204 rm->m_total_sgs = extra_len / sizeof(struct scatterlist);
206 atomic_set(&rm->m_refcount, 1);
207 INIT_LIST_HEAD(&rm->m_sock_item);
208 INIT_LIST_HEAD(&rm->m_conn_item);
209 spin_lock_init(&rm->m_rs_lock);
210 init_waitqueue_head(&rm->m_flush_wait);
217 * RDS ops use this to grab SG entries from the rm's sg pool.
219 struct scatterlist *rds_message_alloc_sgs(struct rds_message *rm, int nents)
221 struct scatterlist *sg_first = (struct scatterlist *) &rm[1];
222 struct scatterlist *sg_ret;
224 WARN_ON(rm->m_used_sgs + nents > rm->m_total_sgs);
227 if (rm->m_used_sgs + nents > rm->m_total_sgs)
230 sg_ret = &sg_first[rm->m_used_sgs];
231 sg_init_table(sg_ret, nents);
232 rm->m_used_sgs += nents;
237 struct rds_message *rds_message_map_pages(unsigned long *page_addrs, unsigned int total_len)
239 struct rds_message *rm;
241 int num_sgs = ceil(total_len, PAGE_SIZE);
242 int extra_bytes = num_sgs * sizeof(struct scatterlist);
244 rm = rds_message_alloc(extra_bytes, GFP_NOWAIT);
246 return ERR_PTR(-ENOMEM);
248 set_bit(RDS_MSG_PAGEVEC, &rm->m_flags);
249 rm->m_inc.i_hdr.h_len = cpu_to_be32(total_len);
250 rm->data.op_nents = ceil(total_len, PAGE_SIZE);
251 rm->data.op_sg = rds_message_alloc_sgs(rm, num_sgs);
252 if (!rm->data.op_sg) {
254 return ERR_PTR(-ENOMEM);
257 for (i = 0; i < rm->data.op_nents; ++i) {
258 sg_set_page(&rm->data.op_sg[i],
259 virt_to_page(page_addrs[i]),
266 int rds_message_copy_from_user(struct rds_message *rm, struct iovec *first_iov,
269 unsigned long to_copy;
270 unsigned long iov_off;
271 unsigned long sg_off;
273 struct scatterlist *sg;
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. */
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)
327 int rds_message_inc_copy_to_user(struct rds_incoming *inc,
328 struct iovec *first_iov, size_t size)
330 struct rds_message *rm;
332 struct scatterlist *sg;
333 unsigned long to_copy;
334 unsigned long iov_off;
335 unsigned long vec_off;
340 rm = container_of(inc, struct rds_message, m_inc);
341 len = be32_to_cpu(rm->m_inc.i_hdr.h_len);
349 while (copied < size && copied < len) {
350 while (iov_off == iov->iov_len) {
355 to_copy = min(iov->iov_len - iov_off, sg->length - vec_off);
356 to_copy = min_t(size_t, to_copy, size - copied);
357 to_copy = min_t(unsigned long, to_copy, len - copied);
359 rdsdebug("copying %lu bytes to user iov [%p, %zu] + %lu to "
360 "sg [%p, %u, %u] + %lu\n",
361 to_copy, iov->iov_base, iov->iov_len, iov_off,
362 sg_page(sg), sg->offset, sg->length, vec_off);
364 ret = rds_page_copy_to_user(sg_page(sg), sg->offset + vec_off,
365 iov->iov_base + iov_off,
376 if (vec_off == sg->length) {
386 * If the message is still on the send queue, wait until the transport
387 * is done with it. This is particularly important for RDMA operations.
389 void rds_message_wait(struct rds_message *rm)
391 wait_event_interruptible(rm->m_flush_wait,
392 !test_bit(RDS_MSG_MAPPED, &rm->m_flags));
395 void rds_message_unmapped(struct rds_message *rm)
397 clear_bit(RDS_MSG_MAPPED, &rm->m_flags);
398 wake_up_interruptible(&rm->m_flush_wait);
400 EXPORT_SYMBOL_GPL(rds_message_unmapped);