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> | |
5a0e3ad6 | 34 | #include <linux/slab.h> |
7875e18e AG |
35 | |
36 | #include "rds.h" | |
7875e18e AG |
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 | ||
6c7cc6e4 AG |
65 | for (i = 0; i < rm->data.op_nents; i++) { |
66 | rdsdebug("putting data page %p\n", (void *)sg_page(&rm->data.op_sg[i])); | |
7875e18e | 67 | /* XXX will have to put_page for page refs */ |
6c7cc6e4 | 68 | __free_page(sg_page(&rm->data.op_sg[i])); |
7875e18e | 69 | } |
6c7cc6e4 | 70 | rm->data.op_nents = 0; |
7875e18e | 71 | |
f8b3aaf2 AG |
72 | if (rm->rdma.op_active) |
73 | rds_rdma_free_op(&rm->rdma); | |
74 | if (rm->rdma.op_rdma_mr) | |
75 | rds_mr_put(rm->rdma.op_rdma_mr); | |
d0ab25a8 AG |
76 | |
77 | if (rm->atomic.op_active) | |
78 | rds_atomic_free_op(&rm->atomic); | |
79 | if (rm->atomic.op_rdma_mr) | |
80 | rds_mr_put(rm->atomic.op_rdma_mr); | |
7875e18e | 81 | } |
7875e18e AG |
82 | |
83 | void rds_message_put(struct rds_message *rm) | |
84 | { | |
85 | rdsdebug("put rm %p ref %d\n", rm, atomic_read(&rm->m_refcount)); | |
86 | ||
87 | if (atomic_dec_and_test(&rm->m_refcount)) { | |
88 | BUG_ON(!list_empty(&rm->m_sock_item)); | |
89 | BUG_ON(!list_empty(&rm->m_conn_item)); | |
90 | rds_message_purge(rm); | |
91 | ||
92 | kfree(rm); | |
93 | } | |
94 | } | |
616b757a | 95 | EXPORT_SYMBOL_GPL(rds_message_put); |
7875e18e | 96 | |
7875e18e AG |
97 | void rds_message_populate_header(struct rds_header *hdr, __be16 sport, |
98 | __be16 dport, u64 seq) | |
99 | { | |
100 | hdr->h_flags = 0; | |
101 | hdr->h_sport = sport; | |
102 | hdr->h_dport = dport; | |
103 | hdr->h_sequence = cpu_to_be64(seq); | |
104 | hdr->h_exthdr[0] = RDS_EXTHDR_NONE; | |
105 | } | |
616b757a | 106 | EXPORT_SYMBOL_GPL(rds_message_populate_header); |
7875e18e AG |
107 | |
108 | int rds_message_add_extension(struct rds_header *hdr, | |
109 | unsigned int type, const void *data, unsigned int len) | |
110 | { | |
111 | unsigned int ext_len = sizeof(u8) + len; | |
112 | unsigned char *dst; | |
113 | ||
114 | /* For now, refuse to add more than one extension header */ | |
115 | if (hdr->h_exthdr[0] != RDS_EXTHDR_NONE) | |
116 | return 0; | |
117 | ||
f64f9e71 | 118 | if (type >= __RDS_EXTHDR_MAX || len != rds_exthdr_size[type]) |
7875e18e AG |
119 | return 0; |
120 | ||
121 | if (ext_len >= RDS_HEADER_EXT_SPACE) | |
122 | return 0; | |
123 | dst = hdr->h_exthdr; | |
124 | ||
125 | *dst++ = type; | |
126 | memcpy(dst, data, len); | |
127 | ||
128 | dst[len] = RDS_EXTHDR_NONE; | |
129 | return 1; | |
130 | } | |
616b757a | 131 | EXPORT_SYMBOL_GPL(rds_message_add_extension); |
7875e18e AG |
132 | |
133 | /* | |
134 | * If a message has extension headers, retrieve them here. | |
135 | * Call like this: | |
136 | * | |
137 | * unsigned int pos = 0; | |
138 | * | |
139 | * while (1) { | |
140 | * buflen = sizeof(buffer); | |
141 | * type = rds_message_next_extension(hdr, &pos, buffer, &buflen); | |
142 | * if (type == RDS_EXTHDR_NONE) | |
143 | * break; | |
144 | * ... | |
145 | * } | |
146 | */ | |
147 | int rds_message_next_extension(struct rds_header *hdr, | |
148 | unsigned int *pos, void *buf, unsigned int *buflen) | |
149 | { | |
150 | unsigned int offset, ext_type, ext_len; | |
151 | u8 *src = hdr->h_exthdr; | |
152 | ||
153 | offset = *pos; | |
154 | if (offset >= RDS_HEADER_EXT_SPACE) | |
155 | goto none; | |
156 | ||
157 | /* Get the extension type and length. For now, the | |
158 | * length is implied by the extension type. */ | |
159 | ext_type = src[offset++]; | |
160 | ||
161 | if (ext_type == RDS_EXTHDR_NONE || ext_type >= __RDS_EXTHDR_MAX) | |
162 | goto none; | |
163 | ext_len = rds_exthdr_size[ext_type]; | |
164 | if (offset + ext_len > RDS_HEADER_EXT_SPACE) | |
165 | goto none; | |
166 | ||
167 | *pos = offset + ext_len; | |
168 | if (ext_len < *buflen) | |
169 | *buflen = ext_len; | |
170 | memcpy(buf, src + offset, *buflen); | |
171 | return ext_type; | |
172 | ||
173 | none: | |
174 | *pos = RDS_HEADER_EXT_SPACE; | |
175 | *buflen = 0; | |
176 | return RDS_EXTHDR_NONE; | |
177 | } | |
178 | ||
179 | int rds_message_add_version_extension(struct rds_header *hdr, unsigned int version) | |
180 | { | |
181 | struct rds_ext_header_version ext_hdr; | |
182 | ||
183 | ext_hdr.h_version = cpu_to_be32(version); | |
184 | return rds_message_add_extension(hdr, RDS_EXTHDR_VERSION, &ext_hdr, sizeof(ext_hdr)); | |
185 | } | |
186 | ||
187 | int rds_message_get_version_extension(struct rds_header *hdr, unsigned int *version) | |
188 | { | |
189 | struct rds_ext_header_version ext_hdr; | |
190 | unsigned int pos = 0, len = sizeof(ext_hdr); | |
191 | ||
192 | /* We assume the version extension is the only one present */ | |
193 | if (rds_message_next_extension(hdr, &pos, &ext_hdr, &len) != RDS_EXTHDR_VERSION) | |
194 | return 0; | |
195 | *version = be32_to_cpu(ext_hdr.h_version); | |
196 | return 1; | |
197 | } | |
198 | ||
199 | int rds_message_add_rdma_dest_extension(struct rds_header *hdr, u32 r_key, u32 offset) | |
200 | { | |
201 | struct rds_ext_header_rdma_dest ext_hdr; | |
202 | ||
203 | ext_hdr.h_rdma_rkey = cpu_to_be32(r_key); | |
204 | ext_hdr.h_rdma_offset = cpu_to_be32(offset); | |
205 | return rds_message_add_extension(hdr, RDS_EXTHDR_RDMA_DEST, &ext_hdr, sizeof(ext_hdr)); | |
206 | } | |
616b757a | 207 | EXPORT_SYMBOL_GPL(rds_message_add_rdma_dest_extension); |
7875e18e | 208 | |
fc445084 AG |
209 | /* |
210 | * Each rds_message is allocated with extra space for the scatterlist entries | |
211 | * rds ops will need. This is to minimize memory allocation count. Then, each rds op | |
212 | * can grab SGs when initializing its part of the rds_message. | |
213 | */ | |
214 | struct rds_message *rds_message_alloc(unsigned int extra_len, gfp_t gfp) | |
7875e18e AG |
215 | { |
216 | struct rds_message *rm; | |
217 | ||
fc445084 | 218 | rm = kzalloc(sizeof(struct rds_message) + extra_len, gfp); |
7875e18e AG |
219 | if (!rm) |
220 | goto out; | |
221 | ||
fc445084 AG |
222 | rm->m_used_sgs = 0; |
223 | rm->m_total_sgs = extra_len / sizeof(struct scatterlist); | |
224 | ||
7875e18e AG |
225 | atomic_set(&rm->m_refcount, 1); |
226 | INIT_LIST_HEAD(&rm->m_sock_item); | |
227 | INIT_LIST_HEAD(&rm->m_conn_item); | |
228 | spin_lock_init(&rm->m_rs_lock); | |
229 | ||
230 | out: | |
231 | return rm; | |
232 | } | |
233 | ||
fc445084 AG |
234 | /* |
235 | * RDS ops use this to grab SG entries from the rm's sg pool. | |
236 | */ | |
237 | struct scatterlist *rds_message_alloc_sgs(struct rds_message *rm, int nents) | |
238 | { | |
239 | struct scatterlist *sg_first = (struct scatterlist *) &rm[1]; | |
240 | struct scatterlist *sg_ret; | |
241 | ||
242 | WARN_ON(rm->m_used_sgs + nents > rm->m_total_sgs); | |
243 | ||
244 | sg_ret = &sg_first[rm->m_used_sgs]; | |
f4dd96f7 | 245 | sg_init_table(sg_ret, nents); |
fc445084 AG |
246 | rm->m_used_sgs += nents; |
247 | ||
248 | return sg_ret; | |
249 | } | |
250 | ||
7875e18e AG |
251 | struct rds_message *rds_message_map_pages(unsigned long *page_addrs, unsigned int total_len) |
252 | { | |
253 | struct rds_message *rm; | |
254 | unsigned int i; | |
ff87e97a AG |
255 | int num_sgs = ceil(total_len, PAGE_SIZE); |
256 | int extra_bytes = num_sgs * sizeof(struct scatterlist); | |
7875e18e | 257 | |
ff87e97a | 258 | rm = rds_message_alloc(extra_bytes, GFP_KERNEL); |
8690bfa1 | 259 | if (!rm) |
7875e18e AG |
260 | return ERR_PTR(-ENOMEM); |
261 | ||
262 | set_bit(RDS_MSG_PAGEVEC, &rm->m_flags); | |
263 | rm->m_inc.i_hdr.h_len = cpu_to_be32(total_len); | |
6c7cc6e4 AG |
264 | rm->data.op_nents = ceil(total_len, PAGE_SIZE); |
265 | rm->data.op_sg = rds_message_alloc_sgs(rm, num_sgs); | |
7875e18e | 266 | |
6c7cc6e4 AG |
267 | for (i = 0; i < rm->data.op_nents; ++i) { |
268 | sg_set_page(&rm->data.op_sg[i], | |
7875e18e AG |
269 | virt_to_page(page_addrs[i]), |
270 | PAGE_SIZE, 0); | |
271 | } | |
272 | ||
273 | return rm; | |
274 | } | |
275 | ||
fc445084 | 276 | int rds_message_copy_from_user(struct rds_message *rm, struct iovec *first_iov, |
7875e18e AG |
277 | size_t total_len) |
278 | { | |
279 | unsigned long to_copy; | |
280 | unsigned long iov_off; | |
281 | unsigned long sg_off; | |
7875e18e AG |
282 | struct iovec *iov; |
283 | struct scatterlist *sg; | |
fc445084 | 284 | int ret = 0; |
7875e18e AG |
285 | |
286 | rm->m_inc.i_hdr.h_len = cpu_to_be32(total_len); | |
287 | ||
288 | /* | |
289 | * now allocate and copy in the data payload. | |
290 | */ | |
6c7cc6e4 | 291 | sg = rm->data.op_sg; |
7875e18e AG |
292 | iov = first_iov; |
293 | iov_off = 0; | |
294 | sg_off = 0; /* Dear gcc, sg->page will be null from kzalloc. */ | |
295 | ||
296 | while (total_len) { | |
8690bfa1 | 297 | if (!sg_page(sg)) { |
7875e18e AG |
298 | ret = rds_page_remainder_alloc(sg, total_len, |
299 | GFP_HIGHUSER); | |
300 | if (ret) | |
301 | goto out; | |
6c7cc6e4 | 302 | rm->data.op_nents++; |
7875e18e AG |
303 | sg_off = 0; |
304 | } | |
305 | ||
306 | while (iov_off == iov->iov_len) { | |
307 | iov_off = 0; | |
308 | iov++; | |
309 | } | |
310 | ||
311 | to_copy = min(iov->iov_len - iov_off, sg->length - sg_off); | |
312 | to_copy = min_t(size_t, to_copy, total_len); | |
313 | ||
314 | rdsdebug("copying %lu bytes from user iov [%p, %zu] + %lu to " | |
315 | "sg [%p, %u, %u] + %lu\n", | |
316 | to_copy, iov->iov_base, iov->iov_len, iov_off, | |
317 | (void *)sg_page(sg), sg->offset, sg->length, sg_off); | |
318 | ||
319 | ret = rds_page_copy_from_user(sg_page(sg), sg->offset + sg_off, | |
320 | iov->iov_base + iov_off, | |
321 | to_copy); | |
322 | if (ret) | |
323 | goto out; | |
324 | ||
325 | iov_off += to_copy; | |
326 | total_len -= to_copy; | |
327 | sg_off += to_copy; | |
328 | ||
329 | if (sg_off == sg->length) | |
330 | sg++; | |
331 | } | |
332 | ||
7875e18e | 333 | out: |
fc445084 | 334 | return ret; |
7875e18e AG |
335 | } |
336 | ||
337 | int rds_message_inc_copy_to_user(struct rds_incoming *inc, | |
338 | struct iovec *first_iov, size_t size) | |
339 | { | |
340 | struct rds_message *rm; | |
341 | struct iovec *iov; | |
342 | struct scatterlist *sg; | |
343 | unsigned long to_copy; | |
344 | unsigned long iov_off; | |
345 | unsigned long vec_off; | |
346 | int copied; | |
347 | int ret; | |
348 | u32 len; | |
349 | ||
350 | rm = container_of(inc, struct rds_message, m_inc); | |
351 | len = be32_to_cpu(rm->m_inc.i_hdr.h_len); | |
352 | ||
353 | iov = first_iov; | |
354 | iov_off = 0; | |
6c7cc6e4 | 355 | sg = rm->data.op_sg; |
7875e18e AG |
356 | vec_off = 0; |
357 | copied = 0; | |
358 | ||
359 | while (copied < size && copied < len) { | |
360 | while (iov_off == iov->iov_len) { | |
361 | iov_off = 0; | |
362 | iov++; | |
363 | } | |
364 | ||
365 | to_copy = min(iov->iov_len - iov_off, sg->length - vec_off); | |
366 | to_copy = min_t(size_t, to_copy, size - copied); | |
367 | to_copy = min_t(unsigned long, to_copy, len - copied); | |
368 | ||
369 | rdsdebug("copying %lu bytes to user iov [%p, %zu] + %lu to " | |
370 | "sg [%p, %u, %u] + %lu\n", | |
371 | to_copy, iov->iov_base, iov->iov_len, iov_off, | |
372 | sg_page(sg), sg->offset, sg->length, vec_off); | |
373 | ||
374 | ret = rds_page_copy_to_user(sg_page(sg), sg->offset + vec_off, | |
375 | iov->iov_base + iov_off, | |
376 | to_copy); | |
377 | if (ret) { | |
378 | copied = ret; | |
379 | break; | |
380 | } | |
381 | ||
382 | iov_off += to_copy; | |
383 | vec_off += to_copy; | |
384 | copied += to_copy; | |
385 | ||
386 | if (vec_off == sg->length) { | |
387 | vec_off = 0; | |
388 | sg++; | |
389 | } | |
390 | } | |
391 | ||
392 | return copied; | |
393 | } | |
394 | ||
395 | /* | |
396 | * If the message is still on the send queue, wait until the transport | |
397 | * is done with it. This is particularly important for RDMA operations. | |
398 | */ | |
399 | void rds_message_wait(struct rds_message *rm) | |
400 | { | |
401 | wait_event(rds_message_flush_waitq, | |
402 | !test_bit(RDS_MSG_MAPPED, &rm->m_flags)); | |
403 | } | |
404 | ||
405 | void rds_message_unmapped(struct rds_message *rm) | |
406 | { | |
407 | clear_bit(RDS_MSG_MAPPED, &rm->m_flags); | |
408 | if (waitqueue_active(&rds_message_flush_waitq)) | |
409 | wake_up(&rds_message_flush_waitq); | |
410 | } | |
616b757a | 411 | EXPORT_SYMBOL_GPL(rds_message_unmapped); |
7875e18e | 412 |