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 <linux/export.h> 36 37 #include "rds.h" 38 39 static unsigned int rds_exthdr_size[__RDS_EXTHDR_MAX] = { 40 [RDS_EXTHDR_NONE] = 0, 41 [RDS_EXTHDR_VERSION] = sizeof(struct rds_ext_header_version), 42 [RDS_EXTHDR_RDMA] = sizeof(struct rds_ext_header_rdma), 43 [RDS_EXTHDR_RDMA_DEST] = sizeof(struct rds_ext_header_rdma_dest), 44 [RDS_EXTHDR_NPATHS] = sizeof(u16), 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 } 53 EXPORT_SYMBOL_GPL(rds_message_addref); 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->data.op_nents; i++) { 66 rdsdebug("putting data page %p\n", (void *)sg_page(&rm->data.op_sg[i])); 67 /* XXX will have to put_page for page refs */ 68 __free_page(sg_page(&rm->data.op_sg[i])); 69 } 70 rm->data.op_nents = 0; 71 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); 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); 81 } 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 WARN(!atomic_read(&rm->m_refcount), "danger refcount zero on %p\n", rm); 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 } 95 EXPORT_SYMBOL_GPL(rds_message_put); 96 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 } 106 EXPORT_SYMBOL_GPL(rds_message_populate_header); 107 108 int rds_message_add_extension(struct rds_header *hdr, unsigned int type, 109 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; 117 118 if (type >= __RDS_EXTHDR_MAX || len != rds_exthdr_size[type]) 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 } 131 EXPORT_SYMBOL_GPL(rds_message_add_extension); 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_rdma_dest_extension(struct rds_header *hdr, u32 r_key, u32 offset) 180 { 181 struct rds_ext_header_rdma_dest ext_hdr; 182 183 ext_hdr.h_rdma_rkey = cpu_to_be32(r_key); 184 ext_hdr.h_rdma_offset = cpu_to_be32(offset); 185 return rds_message_add_extension(hdr, RDS_EXTHDR_RDMA_DEST, &ext_hdr, sizeof(ext_hdr)); 186 } 187 EXPORT_SYMBOL_GPL(rds_message_add_rdma_dest_extension); 188 189 /* 190 * Each rds_message is allocated with extra space for the scatterlist entries 191 * rds ops will need. This is to minimize memory allocation count. Then, each rds op 192 * can grab SGs when initializing its part of the rds_message. 193 */ 194 struct rds_message *rds_message_alloc(unsigned int extra_len, gfp_t gfp) 195 { 196 struct rds_message *rm; 197 198 if (extra_len > KMALLOC_MAX_SIZE - sizeof(struct rds_message)) 199 return NULL; 200 201 rm = kzalloc(sizeof(struct rds_message) + extra_len, gfp); 202 if (!rm) 203 goto out; 204 205 rm->m_used_sgs = 0; 206 rm->m_total_sgs = extra_len / sizeof(struct scatterlist); 207 208 atomic_set(&rm->m_refcount, 1); 209 INIT_LIST_HEAD(&rm->m_sock_item); 210 INIT_LIST_HEAD(&rm->m_conn_item); 211 spin_lock_init(&rm->m_rs_lock); 212 init_waitqueue_head(&rm->m_flush_wait); 213 214 out: 215 return rm; 216 } 217 218 /* 219 * RDS ops use this to grab SG entries from the rm's sg pool. 220 */ 221 struct scatterlist *rds_message_alloc_sgs(struct rds_message *rm, int nents) 222 { 223 struct scatterlist *sg_first = (struct scatterlist *) &rm[1]; 224 struct scatterlist *sg_ret; 225 226 WARN_ON(rm->m_used_sgs + nents > rm->m_total_sgs); 227 WARN_ON(!nents); 228 229 if (rm->m_used_sgs + nents > rm->m_total_sgs) 230 return NULL; 231 232 sg_ret = &sg_first[rm->m_used_sgs]; 233 sg_init_table(sg_ret, nents); 234 rm->m_used_sgs += nents; 235 236 return sg_ret; 237 } 238 239 struct rds_message *rds_message_map_pages(unsigned long *page_addrs, unsigned int total_len) 240 { 241 struct rds_message *rm; 242 unsigned int i; 243 int num_sgs = ceil(total_len, PAGE_SIZE); 244 int extra_bytes = num_sgs * sizeof(struct scatterlist); 245 246 rm = rds_message_alloc(extra_bytes, GFP_NOWAIT); 247 if (!rm) 248 return ERR_PTR(-ENOMEM); 249 250 set_bit(RDS_MSG_PAGEVEC, &rm->m_flags); 251 rm->m_inc.i_hdr.h_len = cpu_to_be32(total_len); 252 rm->data.op_nents = ceil(total_len, PAGE_SIZE); 253 rm->data.op_sg = rds_message_alloc_sgs(rm, num_sgs); 254 if (!rm->data.op_sg) { 255 rds_message_put(rm); 256 return ERR_PTR(-ENOMEM); 257 } 258 259 for (i = 0; i < rm->data.op_nents; ++i) { 260 sg_set_page(&rm->data.op_sg[i], 261 virt_to_page(page_addrs[i]), 262 PAGE_SIZE, 0); 263 } 264 265 return rm; 266 } 267 268 int rds_message_copy_from_user(struct rds_message *rm, struct iov_iter *from) 269 { 270 unsigned long to_copy, nbytes; 271 unsigned long sg_off; 272 struct scatterlist *sg; 273 int ret = 0; 274 275 rm->m_inc.i_hdr.h_len = cpu_to_be32(iov_iter_count(from)); 276 277 /* 278 * now allocate and copy in the data payload. 279 */ 280 sg = rm->data.op_sg; 281 sg_off = 0; /* Dear gcc, sg->page will be null from kzalloc. */ 282 283 while (iov_iter_count(from)) { 284 if (!sg_page(sg)) { 285 ret = rds_page_remainder_alloc(sg, iov_iter_count(from), 286 GFP_HIGHUSER); 287 if (ret) 288 return ret; 289 rm->data.op_nents++; 290 sg_off = 0; 291 } 292 293 to_copy = min_t(unsigned long, iov_iter_count(from), 294 sg->length - sg_off); 295 296 rds_stats_add(s_copy_from_user, to_copy); 297 nbytes = copy_page_from_iter(sg_page(sg), sg->offset + sg_off, 298 to_copy, from); 299 if (nbytes != to_copy) 300 return -EFAULT; 301 302 sg_off += to_copy; 303 304 if (sg_off == sg->length) 305 sg++; 306 } 307 308 return ret; 309 } 310 311 int rds_message_inc_copy_to_user(struct rds_incoming *inc, struct iov_iter *to) 312 { 313 struct rds_message *rm; 314 struct scatterlist *sg; 315 unsigned long to_copy; 316 unsigned long vec_off; 317 int copied; 318 int ret; 319 u32 len; 320 321 rm = container_of(inc, struct rds_message, m_inc); 322 len = be32_to_cpu(rm->m_inc.i_hdr.h_len); 323 324 sg = rm->data.op_sg; 325 vec_off = 0; 326 copied = 0; 327 328 while (iov_iter_count(to) && copied < len) { 329 to_copy = min_t(unsigned long, iov_iter_count(to), 330 sg->length - vec_off); 331 to_copy = min_t(unsigned long, to_copy, len - copied); 332 333 rds_stats_add(s_copy_to_user, to_copy); 334 ret = copy_page_to_iter(sg_page(sg), sg->offset + vec_off, 335 to_copy, to); 336 if (ret != to_copy) 337 return -EFAULT; 338 339 vec_off += to_copy; 340 copied += to_copy; 341 342 if (vec_off == sg->length) { 343 vec_off = 0; 344 sg++; 345 } 346 } 347 348 return copied; 349 } 350 351 /* 352 * If the message is still on the send queue, wait until the transport 353 * is done with it. This is particularly important for RDMA operations. 354 */ 355 void rds_message_wait(struct rds_message *rm) 356 { 357 wait_event_interruptible(rm->m_flush_wait, 358 !test_bit(RDS_MSG_MAPPED, &rm->m_flags)); 359 } 360 361 void rds_message_unmapped(struct rds_message *rm) 362 { 363 clear_bit(RDS_MSG_MAPPED, &rm->m_flags); 364 wake_up_interruptible(&rm->m_flush_wait); 365 } 366 EXPORT_SYMBOL_GPL(rds_message_unmapped); 367 368