message.c 10 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402
  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. #include <linux/slab.h>
  35. #include "rds.h"
  36. static unsigned int rds_exthdr_size[__RDS_EXTHDR_MAX] = {
  37. [RDS_EXTHDR_NONE] = 0,
  38. [RDS_EXTHDR_VERSION] = sizeof(struct rds_ext_header_version),
  39. [RDS_EXTHDR_RDMA] = sizeof(struct rds_ext_header_rdma),
  40. [RDS_EXTHDR_RDMA_DEST] = sizeof(struct rds_ext_header_rdma_dest),
  41. };
  42. void rds_message_addref(struct rds_message *rm)
  43. {
  44. rdsdebug("addref rm %p ref %d\n", rm, atomic_read(&rm->m_refcount));
  45. atomic_inc(&rm->m_refcount);
  46. }
  47. EXPORT_SYMBOL_GPL(rds_message_addref);
  48. /*
  49. * This relies on dma_map_sg() not touching sg[].page during merging.
  50. */
  51. static void rds_message_purge(struct rds_message *rm)
  52. {
  53. unsigned long i;
  54. if (unlikely(test_bit(RDS_MSG_PAGEVEC, &rm->m_flags)))
  55. return;
  56. for (i = 0; i < rm->data.op_nents; i++) {
  57. rdsdebug("putting data page %p\n", (void *)sg_page(&rm->data.op_sg[i]));
  58. /* XXX will have to put_page for page refs */
  59. __free_page(sg_page(&rm->data.op_sg[i]));
  60. }
  61. rm->data.op_nents = 0;
  62. if (rm->rdma.op_active)
  63. rds_rdma_free_op(&rm->rdma);
  64. if (rm->rdma.op_rdma_mr)
  65. rds_mr_put(rm->rdma.op_rdma_mr);
  66. if (rm->atomic.op_active)
  67. rds_atomic_free_op(&rm->atomic);
  68. if (rm->atomic.op_rdma_mr)
  69. rds_mr_put(rm->atomic.op_rdma_mr);
  70. }
  71. void rds_message_put(struct rds_message *rm)
  72. {
  73. rdsdebug("put rm %p ref %d\n", rm, atomic_read(&rm->m_refcount));
  74. if (atomic_read(&rm->m_refcount) == 0) {
  75. printk(KERN_CRIT "danger refcount zero on %p\n", rm);
  76. WARN_ON(1);
  77. }
  78. if (atomic_dec_and_test(&rm->m_refcount)) {
  79. BUG_ON(!list_empty(&rm->m_sock_item));
  80. BUG_ON(!list_empty(&rm->m_conn_item));
  81. rds_message_purge(rm);
  82. kfree(rm);
  83. }
  84. }
  85. EXPORT_SYMBOL_GPL(rds_message_put);
  86. void rds_message_populate_header(struct rds_header *hdr, __be16 sport,
  87. __be16 dport, u64 seq)
  88. {
  89. hdr->h_flags = 0;
  90. hdr->h_sport = sport;
  91. hdr->h_dport = dport;
  92. hdr->h_sequence = cpu_to_be64(seq);
  93. hdr->h_exthdr[0] = RDS_EXTHDR_NONE;
  94. }
  95. EXPORT_SYMBOL_GPL(rds_message_populate_header);
  96. int rds_message_add_extension(struct rds_header *hdr, unsigned int type,
  97. const void *data, unsigned int len)
  98. {
  99. unsigned int ext_len = sizeof(u8) + len;
  100. unsigned char *dst;
  101. /* For now, refuse to add more than one extension header */
  102. if (hdr->h_exthdr[0] != RDS_EXTHDR_NONE)
  103. return 0;
  104. if (type >= __RDS_EXTHDR_MAX || len != rds_exthdr_size[type])
  105. return 0;
  106. if (ext_len >= RDS_HEADER_EXT_SPACE)
  107. return 0;
  108. dst = hdr->h_exthdr;
  109. *dst++ = type;
  110. memcpy(dst, data, len);
  111. dst[len] = RDS_EXTHDR_NONE;
  112. return 1;
  113. }
  114. EXPORT_SYMBOL_GPL(rds_message_add_extension);
  115. /*
  116. * If a message has extension headers, retrieve them here.
  117. * Call like this:
  118. *
  119. * unsigned int pos = 0;
  120. *
  121. * while (1) {
  122. * buflen = sizeof(buffer);
  123. * type = rds_message_next_extension(hdr, &pos, buffer, &buflen);
  124. * if (type == RDS_EXTHDR_NONE)
  125. * break;
  126. * ...
  127. * }
  128. */
  129. int rds_message_next_extension(struct rds_header *hdr,
  130. unsigned int *pos, void *buf, unsigned int *buflen)
  131. {
  132. unsigned int offset, ext_type, ext_len;
  133. u8 *src = hdr->h_exthdr;
  134. offset = *pos;
  135. if (offset >= RDS_HEADER_EXT_SPACE)
  136. goto none;
  137. /* Get the extension type and length. For now, the
  138. * length is implied by the extension type. */
  139. ext_type = src[offset++];
  140. if (ext_type == RDS_EXTHDR_NONE || ext_type >= __RDS_EXTHDR_MAX)
  141. goto none;
  142. ext_len = rds_exthdr_size[ext_type];
  143. if (offset + ext_len > RDS_HEADER_EXT_SPACE)
  144. goto none;
  145. *pos = offset + ext_len;
  146. if (ext_len < *buflen)
  147. *buflen = ext_len;
  148. memcpy(buf, src + offset, *buflen);
  149. return ext_type;
  150. none:
  151. *pos = RDS_HEADER_EXT_SPACE;
  152. *buflen = 0;
  153. return RDS_EXTHDR_NONE;
  154. }
  155. int rds_message_add_rdma_dest_extension(struct rds_header *hdr, u32 r_key, u32 offset)
  156. {
  157. struct rds_ext_header_rdma_dest ext_hdr;
  158. ext_hdr.h_rdma_rkey = cpu_to_be32(r_key);
  159. ext_hdr.h_rdma_offset = cpu_to_be32(offset);
  160. return rds_message_add_extension(hdr, RDS_EXTHDR_RDMA_DEST, &ext_hdr, sizeof(ext_hdr));
  161. }
  162. EXPORT_SYMBOL_GPL(rds_message_add_rdma_dest_extension);
  163. /*
  164. * Each rds_message is allocated with extra space for the scatterlist entries
  165. * rds ops will need. This is to minimize memory allocation count. Then, each rds op
  166. * can grab SGs when initializing its part of the rds_message.
  167. */
  168. struct rds_message *rds_message_alloc(unsigned int extra_len, gfp_t gfp)
  169. {
  170. struct rds_message *rm;
  171. rm = kzalloc(sizeof(struct rds_message) + extra_len, gfp);
  172. if (!rm)
  173. goto out;
  174. rm->m_used_sgs = 0;
  175. rm->m_total_sgs = extra_len / sizeof(struct scatterlist);
  176. atomic_set(&rm->m_refcount, 1);
  177. INIT_LIST_HEAD(&rm->m_sock_item);
  178. INIT_LIST_HEAD(&rm->m_conn_item);
  179. spin_lock_init(&rm->m_rs_lock);
  180. init_waitqueue_head(&rm->m_flush_wait);
  181. out:
  182. return rm;
  183. }
  184. /*
  185. * RDS ops use this to grab SG entries from the rm's sg pool.
  186. */
  187. struct scatterlist *rds_message_alloc_sgs(struct rds_message *rm, int nents)
  188. {
  189. struct scatterlist *sg_first = (struct scatterlist *) &rm[1];
  190. struct scatterlist *sg_ret;
  191. WARN_ON(rm->m_used_sgs + nents > rm->m_total_sgs);
  192. WARN_ON(!nents);
  193. if (rm->m_used_sgs + nents > rm->m_total_sgs)
  194. return NULL;
  195. sg_ret = &sg_first[rm->m_used_sgs];
  196. sg_init_table(sg_ret, nents);
  197. rm->m_used_sgs += nents;
  198. return sg_ret;
  199. }
  200. struct rds_message *rds_message_map_pages(unsigned long *page_addrs, unsigned int total_len)
  201. {
  202. struct rds_message *rm;
  203. unsigned int i;
  204. int num_sgs = ceil(total_len, PAGE_SIZE);
  205. int extra_bytes = num_sgs * sizeof(struct scatterlist);
  206. rm = rds_message_alloc(extra_bytes, GFP_NOWAIT);
  207. if (!rm)
  208. return ERR_PTR(-ENOMEM);
  209. set_bit(RDS_MSG_PAGEVEC, &rm->m_flags);
  210. rm->m_inc.i_hdr.h_len = cpu_to_be32(total_len);
  211. rm->data.op_nents = ceil(total_len, PAGE_SIZE);
  212. rm->data.op_sg = rds_message_alloc_sgs(rm, num_sgs);
  213. if (!rm->data.op_sg) {
  214. rds_message_put(rm);
  215. return ERR_PTR(-ENOMEM);
  216. }
  217. for (i = 0; i < rm->data.op_nents; ++i) {
  218. sg_set_page(&rm->data.op_sg[i],
  219. virt_to_page(page_addrs[i]),
  220. PAGE_SIZE, 0);
  221. }
  222. return rm;
  223. }
  224. int rds_message_copy_from_user(struct rds_message *rm, struct iovec *first_iov,
  225. size_t total_len)
  226. {
  227. unsigned long to_copy;
  228. unsigned long iov_off;
  229. unsigned long sg_off;
  230. struct iovec *iov;
  231. struct scatterlist *sg;
  232. int ret = 0;
  233. rm->m_inc.i_hdr.h_len = cpu_to_be32(total_len);
  234. /*
  235. * now allocate and copy in the data payload.
  236. */
  237. sg = rm->data.op_sg;
  238. iov = first_iov;
  239. iov_off = 0;
  240. sg_off = 0; /* Dear gcc, sg->page will be null from kzalloc. */
  241. while (total_len) {
  242. if (!sg_page(sg)) {
  243. ret = rds_page_remainder_alloc(sg, total_len,
  244. GFP_HIGHUSER);
  245. if (ret)
  246. goto out;
  247. rm->data.op_nents++;
  248. sg_off = 0;
  249. }
  250. while (iov_off == iov->iov_len) {
  251. iov_off = 0;
  252. iov++;
  253. }
  254. to_copy = min(iov->iov_len - iov_off, sg->length - sg_off);
  255. to_copy = min_t(size_t, to_copy, total_len);
  256. rdsdebug("copying %lu bytes from user iov [%p, %zu] + %lu to "
  257. "sg [%p, %u, %u] + %lu\n",
  258. to_copy, iov->iov_base, iov->iov_len, iov_off,
  259. (void *)sg_page(sg), sg->offset, sg->length, sg_off);
  260. ret = rds_page_copy_from_user(sg_page(sg), sg->offset + sg_off,
  261. iov->iov_base + iov_off,
  262. to_copy);
  263. if (ret)
  264. goto out;
  265. iov_off += to_copy;
  266. total_len -= to_copy;
  267. sg_off += to_copy;
  268. if (sg_off == sg->length)
  269. sg++;
  270. }
  271. out:
  272. return ret;
  273. }
  274. int rds_message_inc_copy_to_user(struct rds_incoming *inc,
  275. struct iovec *first_iov, size_t size)
  276. {
  277. struct rds_message *rm;
  278. struct iovec *iov;
  279. struct scatterlist *sg;
  280. unsigned long to_copy;
  281. unsigned long iov_off;
  282. unsigned long vec_off;
  283. int copied;
  284. int ret;
  285. u32 len;
  286. rm = container_of(inc, struct rds_message, m_inc);
  287. len = be32_to_cpu(rm->m_inc.i_hdr.h_len);
  288. iov = first_iov;
  289. iov_off = 0;
  290. sg = rm->data.op_sg;
  291. vec_off = 0;
  292. copied = 0;
  293. while (copied < size && copied < len) {
  294. while (iov_off == iov->iov_len) {
  295. iov_off = 0;
  296. iov++;
  297. }
  298. to_copy = min(iov->iov_len - iov_off, sg->length - vec_off);
  299. to_copy = min_t(size_t, to_copy, size - copied);
  300. to_copy = min_t(unsigned long, to_copy, len - copied);
  301. rdsdebug("copying %lu bytes to user iov [%p, %zu] + %lu to "
  302. "sg [%p, %u, %u] + %lu\n",
  303. to_copy, iov->iov_base, iov->iov_len, iov_off,
  304. sg_page(sg), sg->offset, sg->length, vec_off);
  305. ret = rds_page_copy_to_user(sg_page(sg), sg->offset + vec_off,
  306. iov->iov_base + iov_off,
  307. to_copy);
  308. if (ret) {
  309. copied = ret;
  310. break;
  311. }
  312. iov_off += to_copy;
  313. vec_off += to_copy;
  314. copied += to_copy;
  315. if (vec_off == sg->length) {
  316. vec_off = 0;
  317. sg++;
  318. }
  319. }
  320. return copied;
  321. }
  322. /*
  323. * If the message is still on the send queue, wait until the transport
  324. * is done with it. This is particularly important for RDMA operations.
  325. */
  326. void rds_message_wait(struct rds_message *rm)
  327. {
  328. wait_event_interruptible(rm->m_flush_wait,
  329. !test_bit(RDS_MSG_MAPPED, &rm->m_flags));
  330. }
  331. void rds_message_unmapped(struct rds_message *rm)
  332. {
  333. clear_bit(RDS_MSG_MAPPED, &rm->m_flags);
  334. wake_up_interruptible(&rm->m_flush_wait);
  335. }
  336. EXPORT_SYMBOL_GPL(rds_message_unmapped);