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 36 #include "rds.h" 37 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), 43 }; 44 45 46 void rds_message_addref(struct rds_message *rm) 47 { 48 rdsdebug("addref rm %p ref %d\n", rm, atomic_read(&rm->m_refcount)); 49 atomic_inc(&rm->m_refcount); 50 } 51 EXPORT_SYMBOL_GPL(rds_message_addref); 52 53 /* 54 * This relies on dma_map_sg() not touching sg[].page during merging. 55 */ 56 static void rds_message_purge(struct rds_message *rm) 57 { 58 unsigned long i; 59 60 if (unlikely(test_bit(RDS_MSG_PAGEVEC, &rm->m_flags))) 61 return; 62 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])); 67 } 68 rm->data.op_nents = 0; 69 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); 74 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); 79 } 80 81 void rds_message_put(struct rds_message *rm) 82 { 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); 86 WARN_ON(1); 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 } 96 EXPORT_SYMBOL_GPL(rds_message_put); 97 98 void rds_message_populate_header(struct rds_header *hdr, __be16 sport, 99 __be16 dport, u64 seq) 100 { 101 hdr->h_flags = 0; 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; 106 } 107 EXPORT_SYMBOL_GPL(rds_message_populate_header); 108 109 int rds_message_add_extension(struct rds_header *hdr, unsigned int type, 110 const void *data, unsigned int len) 111 { 112 unsigned int ext_len = sizeof(u8) + len; 113 unsigned char *dst; 114 115 /* For now, refuse to add more than one extension header */ 116 if (hdr->h_exthdr[0] != RDS_EXTHDR_NONE) 117 return 0; 118 119 if (type >= __RDS_EXTHDR_MAX || len != rds_exthdr_size[type]) 120 return 0; 121 122 if (ext_len >= RDS_HEADER_EXT_SPACE) 123 return 0; 124 dst = hdr->h_exthdr; 125 126 *dst++ = type; 127 memcpy(dst, data, len); 128 129 dst[len] = RDS_EXTHDR_NONE; 130 return 1; 131 } 132 EXPORT_SYMBOL_GPL(rds_message_add_extension); 133 134 /* 135 * If a message has extension headers, retrieve them here. 136 * Call like this: 137 * 138 * unsigned int pos = 0; 139 * 140 * while (1) { 141 * buflen = sizeof(buffer); 142 * type = rds_message_next_extension(hdr, &pos, buffer, &buflen); 143 * if (type == RDS_EXTHDR_NONE) 144 * break; 145 * ... 146 * } 147 */ 148 int rds_message_next_extension(struct rds_header *hdr, 149 unsigned int *pos, void *buf, unsigned int *buflen) 150 { 151 unsigned int offset, ext_type, ext_len; 152 u8 *src = hdr->h_exthdr; 153 154 offset = *pos; 155 if (offset >= RDS_HEADER_EXT_SPACE) 156 goto none; 157 158 /* Get the extension type and length. For now, the 159 * length is implied by the extension type. */ 160 ext_type = src[offset++]; 161 162 if (ext_type == RDS_EXTHDR_NONE || ext_type >= __RDS_EXTHDR_MAX) 163 goto none; 164 ext_len = rds_exthdr_size[ext_type]; 165 if (offset + ext_len > RDS_HEADER_EXT_SPACE) 166 goto none; 167 168 *pos = offset + ext_len; 169 if (ext_len < *buflen) 170 *buflen = ext_len; 171 memcpy(buf, src + offset, *buflen); 172 return ext_type; 173 174 none: 175 *pos = RDS_HEADER_EXT_SPACE; 176 *buflen = 0; 177 return RDS_EXTHDR_NONE; 178 } 179 180 int rds_message_add_rdma_dest_extension(struct rds_header *hdr, u32 r_key, u32 offset) 181 { 182 struct rds_ext_header_rdma_dest ext_hdr; 183 184 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)); 187 } 188 EXPORT_SYMBOL_GPL(rds_message_add_rdma_dest_extension); 189 190 /* 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. 194 */ 195 struct rds_message *rds_message_alloc(unsigned int extra_len, gfp_t gfp) 196 { 197 struct rds_message *rm; 198 199 rm = kzalloc(sizeof(struct rds_message) + extra_len, gfp); 200 if (!rm) 201 goto out; 202 203 rm->m_used_sgs = 0; 204 rm->m_total_sgs = extra_len / sizeof(struct scatterlist); 205 206 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); 211 212 out: 213 return rm; 214 } 215 216 /* 217 * RDS ops use this to grab SG entries from the rm's sg pool. 218 */ 219 struct scatterlist *rds_message_alloc_sgs(struct rds_message *rm, int nents) 220 { 221 struct scatterlist *sg_first = (struct scatterlist *) &rm[1]; 222 struct scatterlist *sg_ret; 223 224 WARN_ON(rm->m_used_sgs + nents > rm->m_total_sgs); 225 WARN_ON(!nents); 226 227 sg_ret = &sg_first[rm->m_used_sgs]; 228 sg_init_table(sg_ret, nents); 229 rm->m_used_sgs += nents; 230 231 return sg_ret; 232 } 233 234 struct rds_message *rds_message_map_pages(unsigned long *page_addrs, unsigned int total_len) 235 { 236 struct rds_message *rm; 237 unsigned int i; 238 int num_sgs = ceil(total_len, PAGE_SIZE); 239 int extra_bytes = num_sgs * sizeof(struct scatterlist); 240 241 rm = rds_message_alloc(extra_bytes, GFP_NOWAIT); 242 if (!rm) 243 return ERR_PTR(-ENOMEM); 244 245 set_bit(RDS_MSG_PAGEVEC, &rm->m_flags); 246 rm->m_inc.i_hdr.h_len = cpu_to_be32(total_len); 247 rm->data.op_nents = ceil(total_len, PAGE_SIZE); 248 rm->data.op_sg = rds_message_alloc_sgs(rm, num_sgs); 249 250 for (i = 0; i < rm->data.op_nents; ++i) { 251 sg_set_page(&rm->data.op_sg[i], 252 virt_to_page(page_addrs[i]), 253 PAGE_SIZE, 0); 254 } 255 256 return rm; 257 } 258 259 int rds_message_copy_from_user(struct rds_message *rm, 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 iovec *iov; 266 struct scatterlist *sg; 267 int ret = 0; 268 269 rm->m_inc.i_hdr.h_len = cpu_to_be32(total_len); 270 271 /* 272 * now allocate and copy in the data payload. 273 */ 274 sg = rm->data.op_sg; 275 iov = first_iov; 276 iov_off = 0; 277 sg_off = 0; /* Dear gcc, sg->page will be null from kzalloc. */ 278 279 while (total_len) { 280 if (!sg_page(sg)) { 281 ret = rds_page_remainder_alloc(sg, total_len, 282 GFP_HIGHUSER); 283 if (ret) 284 goto out; 285 rm->data.op_nents++; 286 sg_off = 0; 287 } 288 289 while (iov_off == iov->iov_len) { 290 iov_off = 0; 291 iov++; 292 } 293 294 to_copy = min(iov->iov_len - iov_off, sg->length - sg_off); 295 to_copy = min_t(size_t, to_copy, total_len); 296 297 rdsdebug("copying %lu bytes from user iov [%p, %zu] + %lu to " 298 "sg [%p, %u, %u] + %lu\n", 299 to_copy, iov->iov_base, iov->iov_len, iov_off, 300 (void *)sg_page(sg), sg->offset, sg->length, sg_off); 301 302 ret = rds_page_copy_from_user(sg_page(sg), sg->offset + sg_off, 303 iov->iov_base + iov_off, 304 to_copy); 305 if (ret) 306 goto out; 307 308 iov_off += to_copy; 309 total_len -= to_copy; 310 sg_off += to_copy; 311 312 if (sg_off == sg->length) 313 sg++; 314 } 315 316 out: 317 return ret; 318 } 319 320 int rds_message_inc_copy_to_user(struct rds_incoming *inc, 321 struct iovec *first_iov, size_t size) 322 { 323 struct rds_message *rm; 324 struct iovec *iov; 325 struct scatterlist *sg; 326 unsigned long to_copy; 327 unsigned long iov_off; 328 unsigned long vec_off; 329 int copied; 330 int ret; 331 u32 len; 332 333 rm = container_of(inc, struct rds_message, m_inc); 334 len = be32_to_cpu(rm->m_inc.i_hdr.h_len); 335 336 iov = first_iov; 337 iov_off = 0; 338 sg = rm->data.op_sg; 339 vec_off = 0; 340 copied = 0; 341 342 while (copied < size && copied < len) { 343 while (iov_off == iov->iov_len) { 344 iov_off = 0; 345 iov++; 346 } 347 348 to_copy = min(iov->iov_len - iov_off, sg->length - vec_off); 349 to_copy = min_t(size_t, to_copy, size - copied); 350 to_copy = min_t(unsigned long, to_copy, len - copied); 351 352 rdsdebug("copying %lu bytes to user iov [%p, %zu] + %lu to " 353 "sg [%p, %u, %u] + %lu\n", 354 to_copy, iov->iov_base, iov->iov_len, iov_off, 355 sg_page(sg), sg->offset, sg->length, vec_off); 356 357 ret = rds_page_copy_to_user(sg_page(sg), sg->offset + vec_off, 358 iov->iov_base + iov_off, 359 to_copy); 360 if (ret) { 361 copied = ret; 362 break; 363 } 364 365 iov_off += to_copy; 366 vec_off += to_copy; 367 copied += to_copy; 368 369 if (vec_off == sg->length) { 370 vec_off = 0; 371 sg++; 372 } 373 } 374 375 return copied; 376 } 377 378 /* 379 * If the message is still on the send queue, wait until the transport 380 * is done with it. This is particularly important for RDMA operations. 381 */ 382 void rds_message_wait(struct rds_message *rm) 383 { 384 wait_event_interruptible(rm->m_flush_wait, 385 !test_bit(RDS_MSG_MAPPED, &rm->m_flags)); 386 } 387 388 void rds_message_unmapped(struct rds_message *rm) 389 { 390 clear_bit(RDS_MSG_MAPPED, &rm->m_flags); 391 wake_up_interruptible(&rm->m_flush_wait); 392 } 393 EXPORT_SYMBOL_GPL(rds_message_unmapped); 394 395