Commit | Line | Data |
---|---|---|
7875e18e AG |
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 | ||
35 | #include "rds.h" | |
36 | #include "rdma.h" | |
37 | ||
38 | static DECLARE_WAIT_QUEUE_HEAD(rds_message_flush_waitq); | |
39 | ||
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), | |
45 | }; | |
46 | ||
47 | ||
48 | void rds_message_addref(struct rds_message *rm) | |
49 | { | |
50 | rdsdebug("addref rm %p ref %d\n", rm, atomic_read(&rm->m_refcount)); | |
51 | atomic_inc(&rm->m_refcount); | |
52 | } | |
616b757a | 53 | EXPORT_SYMBOL_GPL(rds_message_addref); |
7875e18e AG |
54 | |
55 | /* | |
56 | * This relies on dma_map_sg() not touching sg[].page during merging. | |
57 | */ | |
58 | static void rds_message_purge(struct rds_message *rm) | |
59 | { | |
60 | unsigned long i; | |
61 | ||
62 | if (unlikely(test_bit(RDS_MSG_PAGEVEC, &rm->m_flags))) | |
63 | return; | |
64 | ||
65 | for (i = 0; i < rm->m_nents; i++) { | |
66 | rdsdebug("putting data page %p\n", (void *)sg_page(&rm->m_sg[i])); | |
67 | /* XXX will have to put_page for page refs */ | |
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 | ||
78 | void 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 | ||
84 | void 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 | } | |
616b757a | 96 | EXPORT_SYMBOL_GPL(rds_message_put); |
7875e18e AG |
97 | |
98 | void 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 | ||
104 | void 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 | } | |
616b757a | 113 | EXPORT_SYMBOL_GPL(rds_message_populate_header); |
7875e18e AG |
114 | |
115 | int 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 | |
126 | || len != rds_exthdr_size[type]) | |
127 | return 0; | |
128 | ||
129 | if (ext_len >= RDS_HEADER_EXT_SPACE) | |
130 | return 0; | |
131 | dst = hdr->h_exthdr; | |
132 | ||
133 | *dst++ = type; | |
134 | memcpy(dst, data, len); | |
135 | ||
136 | dst[len] = RDS_EXTHDR_NONE; | |
137 | return 1; | |
138 | } | |
616b757a | 139 | EXPORT_SYMBOL_GPL(rds_message_add_extension); |
7875e18e AG |
140 | |
141 | /* | |
142 | * If a message has extension headers, retrieve them here. | |
143 | * Call like this: | |
144 | * | |
145 | * unsigned int pos = 0; | |
146 | * | |
147 | * while (1) { | |
148 | * buflen = sizeof(buffer); | |
149 | * type = rds_message_next_extension(hdr, &pos, buffer, &buflen); | |
150 | * if (type == RDS_EXTHDR_NONE) | |
151 | * break; | |
152 | * ... | |
153 | * } | |
154 | */ | |
155 | int rds_message_next_extension(struct rds_header *hdr, | |
156 | unsigned int *pos, void *buf, unsigned int *buflen) | |
157 | { | |
158 | unsigned int offset, ext_type, ext_len; | |
159 | u8 *src = hdr->h_exthdr; | |
160 | ||
161 | offset = *pos; | |
162 | if (offset >= RDS_HEADER_EXT_SPACE) | |
163 | goto none; | |
164 | ||
165 | /* Get the extension type and length. For now, the | |
166 | * length is implied by the extension type. */ | |
167 | ext_type = src[offset++]; | |
168 | ||
169 | if (ext_type == RDS_EXTHDR_NONE || ext_type >= __RDS_EXTHDR_MAX) | |
170 | goto none; | |
171 | ext_len = rds_exthdr_size[ext_type]; | |
172 | if (offset + ext_len > RDS_HEADER_EXT_SPACE) | |
173 | goto none; | |
174 | ||
175 | *pos = offset + ext_len; | |
176 | if (ext_len < *buflen) | |
177 | *buflen = ext_len; | |
178 | memcpy(buf, src + offset, *buflen); | |
179 | return ext_type; | |
180 | ||
181 | none: | |
182 | *pos = RDS_HEADER_EXT_SPACE; | |
183 | *buflen = 0; | |
184 | return RDS_EXTHDR_NONE; | |
185 | } | |
186 | ||
187 | int rds_message_add_version_extension(struct rds_header *hdr, unsigned int version) | |
188 | { | |
189 | struct rds_ext_header_version ext_hdr; | |
190 | ||
191 | ext_hdr.h_version = cpu_to_be32(version); | |
192 | return rds_message_add_extension(hdr, RDS_EXTHDR_VERSION, &ext_hdr, sizeof(ext_hdr)); | |
193 | } | |
194 | ||
195 | int rds_message_get_version_extension(struct rds_header *hdr, unsigned int *version) | |
196 | { | |
197 | struct rds_ext_header_version ext_hdr; | |
198 | unsigned int pos = 0, len = sizeof(ext_hdr); | |
199 | ||
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) | |
202 | return 0; | |
203 | *version = be32_to_cpu(ext_hdr.h_version); | |
204 | return 1; | |
205 | } | |
206 | ||
207 | int rds_message_add_rdma_dest_extension(struct rds_header *hdr, u32 r_key, u32 offset) | |
208 | { | |
209 | struct rds_ext_header_rdma_dest ext_hdr; | |
210 | ||
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)); | |
214 | } | |
616b757a | 215 | EXPORT_SYMBOL_GPL(rds_message_add_rdma_dest_extension); |
7875e18e AG |
216 | |
217 | struct rds_message *rds_message_alloc(unsigned int nents, gfp_t gfp) | |
218 | { | |
219 | struct rds_message *rm; | |
220 | ||
221 | rm = kzalloc(sizeof(struct rds_message) + | |
222 | (nents * sizeof(struct scatterlist)), gfp); | |
223 | if (!rm) | |
224 | goto out; | |
225 | ||
226 | if (nents) | |
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); | |
232 | ||
233 | out: | |
234 | return rm; | |
235 | } | |
236 | ||
237 | struct rds_message *rds_message_map_pages(unsigned long *page_addrs, unsigned int total_len) | |
238 | { | |
239 | struct rds_message *rm; | |
240 | unsigned int i; | |
241 | ||
242 | rm = rds_message_alloc(ceil(total_len, PAGE_SIZE), GFP_KERNEL); | |
243 | if (rm == NULL) | |
244 | return ERR_PTR(-ENOMEM); | |
245 | ||
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); | |
249 | ||
250 | for (i = 0; i < rm->m_nents; ++i) { | |
251 | sg_set_page(&rm->m_sg[i], | |
252 | virt_to_page(page_addrs[i]), | |
253 | PAGE_SIZE, 0); | |
254 | } | |
255 | ||
256 | return rm; | |
257 | } | |
258 | ||
259 | struct rds_message *rds_message_copy_from_user(struct iovec *first_iov, | |
260 | size_t total_len) | |
261 | { | |
262 | unsigned long to_copy; | |
263 | unsigned long iov_off; | |
264 | unsigned long sg_off; | |
265 | struct rds_message *rm; | |
266 | struct iovec *iov; | |
267 | struct scatterlist *sg; | |
268 | int ret; | |
269 | ||
270 | rm = rds_message_alloc(ceil(total_len, PAGE_SIZE), GFP_KERNEL); | |
271 | if (rm == NULL) { | |
272 | ret = -ENOMEM; | |
273 | goto out; | |
274 | } | |
275 | ||
276 | rm->m_inc.i_hdr.h_len = cpu_to_be32(total_len); | |
277 | ||
278 | /* | |
279 | * now allocate and copy in the data payload. | |
280 | */ | |
281 | sg = rm->m_sg; | |
282 | iov = first_iov; | |
283 | iov_off = 0; | |
284 | sg_off = 0; /* Dear gcc, sg->page will be null from kzalloc. */ | |
285 | ||
286 | while (total_len) { | |
287 | if (sg_page(sg) == NULL) { | |
288 | ret = rds_page_remainder_alloc(sg, total_len, | |
289 | GFP_HIGHUSER); | |
290 | if (ret) | |
291 | goto out; | |
292 | rm->m_nents++; | |
293 | sg_off = 0; | |
294 | } | |
295 | ||
296 | while (iov_off == iov->iov_len) { | |
297 | iov_off = 0; | |
298 | iov++; | |
299 | } | |
300 | ||
301 | to_copy = min(iov->iov_len - iov_off, sg->length - sg_off); | |
302 | to_copy = min_t(size_t, to_copy, total_len); | |
303 | ||
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); | |
308 | ||
309 | ret = rds_page_copy_from_user(sg_page(sg), sg->offset + sg_off, | |
310 | iov->iov_base + iov_off, | |
311 | to_copy); | |
312 | if (ret) | |
313 | goto out; | |
314 | ||
315 | iov_off += to_copy; | |
316 | total_len -= to_copy; | |
317 | sg_off += to_copy; | |
318 | ||
319 | if (sg_off == sg->length) | |
320 | sg++; | |
321 | } | |
322 | ||
323 | ret = 0; | |
324 | out: | |
325 | if (ret) { | |
326 | if (rm) | |
327 | rds_message_put(rm); | |
328 | rm = ERR_PTR(ret); | |
329 | } | |
330 | return rm; | |
331 | } | |
332 | ||
333 | int rds_message_inc_copy_to_user(struct rds_incoming *inc, | |
334 | struct iovec *first_iov, size_t size) | |
335 | { | |
336 | struct rds_message *rm; | |
337 | struct iovec *iov; | |
338 | struct scatterlist *sg; | |
339 | unsigned long to_copy; | |
340 | unsigned long iov_off; | |
341 | unsigned long vec_off; | |
342 | int copied; | |
343 | int ret; | |
344 | u32 len; | |
345 | ||
346 | rm = container_of(inc, struct rds_message, m_inc); | |
347 | len = be32_to_cpu(rm->m_inc.i_hdr.h_len); | |
348 | ||
349 | iov = first_iov; | |
350 | iov_off = 0; | |
351 | sg = rm->m_sg; | |
352 | vec_off = 0; | |
353 | copied = 0; | |
354 | ||
355 | while (copied < size && copied < len) { | |
356 | while (iov_off == iov->iov_len) { | |
357 | iov_off = 0; | |
358 | iov++; | |
359 | } | |
360 | ||
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); | |
364 | ||
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); | |
369 | ||
370 | ret = rds_page_copy_to_user(sg_page(sg), sg->offset + vec_off, | |
371 | iov->iov_base + iov_off, | |
372 | to_copy); | |
373 | if (ret) { | |
374 | copied = ret; | |
375 | break; | |
376 | } | |
377 | ||
378 | iov_off += to_copy; | |
379 | vec_off += to_copy; | |
380 | copied += to_copy; | |
381 | ||
382 | if (vec_off == sg->length) { | |
383 | vec_off = 0; | |
384 | sg++; | |
385 | } | |
386 | } | |
387 | ||
388 | return copied; | |
389 | } | |
390 | ||
391 | /* | |
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. | |
394 | */ | |
395 | void rds_message_wait(struct rds_message *rm) | |
396 | { | |
397 | wait_event(rds_message_flush_waitq, | |
398 | !test_bit(RDS_MSG_MAPPED, &rm->m_flags)); | |
399 | } | |
400 | ||
401 | void rds_message_unmapped(struct rds_message *rm) | |
402 | { | |
403 | clear_bit(RDS_MSG_MAPPED, &rm->m_flags); | |
404 | if (waitqueue_active(&rds_message_flush_waitq)) | |
405 | wake_up(&rds_message_flush_waitq); | |
406 | } | |
616b757a | 407 | EXPORT_SYMBOL_GPL(rds_message_unmapped); |
7875e18e | 408 |