1 /* 2 * Helpers for getting linearized buffers from iov / filling buffers into iovs 3 * 4 * Copyright IBM, Corp. 2007, 2008 5 * Copyright (C) 2010 Red Hat, Inc. 6 * 7 * Author(s): 8 * Anthony Liguori <aliguori@us.ibm.com> 9 * Amit Shah <amit.shah@redhat.com> 10 * Michael Tokarev <mjt@tls.msk.ru> 11 * 12 * This work is licensed under the terms of the GNU GPL, version 2. See 13 * the COPYING file in the top-level directory. 14 * 15 * Contributions after 2012-01-13 are licensed under the terms of the 16 * GNU GPL, version 2 or (at your option) any later version. 17 */ 18 19 #include "qemu/iov.h" 20 21 #ifdef _WIN32 22 # include <windows.h> 23 # include <winsock2.h> 24 #else 25 # include <sys/types.h> 26 # include <sys/socket.h> 27 #endif 28 29 size_t iov_from_buf(const struct iovec *iov, unsigned int iov_cnt, 30 size_t offset, const void *buf, size_t bytes) 31 { 32 size_t done; 33 unsigned int i; 34 for (i = 0, done = 0; (offset || done < bytes) && i < iov_cnt; i++) { 35 if (offset < iov[i].iov_len) { 36 size_t len = MIN(iov[i].iov_len - offset, bytes - done); 37 memcpy(iov[i].iov_base + offset, buf + done, len); 38 done += len; 39 offset = 0; 40 } else { 41 offset -= iov[i].iov_len; 42 } 43 } 44 assert(offset == 0); 45 return done; 46 } 47 48 size_t iov_to_buf(const struct iovec *iov, const unsigned int iov_cnt, 49 size_t offset, void *buf, size_t bytes) 50 { 51 size_t done; 52 unsigned int i; 53 for (i = 0, done = 0; (offset || done < bytes) && i < iov_cnt; i++) { 54 if (offset < iov[i].iov_len) { 55 size_t len = MIN(iov[i].iov_len - offset, bytes - done); 56 memcpy(buf + done, iov[i].iov_base + offset, len); 57 done += len; 58 offset = 0; 59 } else { 60 offset -= iov[i].iov_len; 61 } 62 } 63 assert(offset == 0); 64 return done; 65 } 66 67 size_t iov_memset(const struct iovec *iov, const unsigned int iov_cnt, 68 size_t offset, int fillc, size_t bytes) 69 { 70 size_t done; 71 unsigned int i; 72 for (i = 0, done = 0; (offset || done < bytes) && i < iov_cnt; i++) { 73 if (offset < iov[i].iov_len) { 74 size_t len = MIN(iov[i].iov_len - offset, bytes - done); 75 memset(iov[i].iov_base + offset, fillc, len); 76 done += len; 77 offset = 0; 78 } else { 79 offset -= iov[i].iov_len; 80 } 81 } 82 assert(offset == 0); 83 return done; 84 } 85 86 size_t iov_size(const struct iovec *iov, const unsigned int iov_cnt) 87 { 88 size_t len; 89 unsigned int i; 90 91 len = 0; 92 for (i = 0; i < iov_cnt; i++) { 93 len += iov[i].iov_len; 94 } 95 return len; 96 } 97 98 /* helper function for iov_send_recv() */ 99 static ssize_t 100 do_send_recv(int sockfd, struct iovec *iov, unsigned iov_cnt, bool do_send) 101 { 102 #ifdef CONFIG_POSIX 103 ssize_t ret; 104 struct msghdr msg; 105 memset(&msg, 0, sizeof(msg)); 106 msg.msg_iov = iov; 107 msg.msg_iovlen = iov_cnt; 108 do { 109 ret = do_send 110 ? sendmsg(sockfd, &msg, 0) 111 : recvmsg(sockfd, &msg, 0); 112 } while (ret < 0 && errno == EINTR); 113 return ret; 114 #else 115 /* else send piece-by-piece */ 116 /*XXX Note: windows has WSASend() and WSARecv() */ 117 unsigned i = 0; 118 ssize_t ret = 0; 119 while (i < iov_cnt) { 120 ssize_t r = do_send 121 ? send(sockfd, iov[i].iov_base, iov[i].iov_len, 0) 122 : recv(sockfd, iov[i].iov_base, iov[i].iov_len, 0); 123 if (r > 0) { 124 ret += r; 125 } else if (!r) { 126 break; 127 } else if (errno == EINTR) { 128 continue; 129 } else { 130 /* else it is some "other" error, 131 * only return if there was no data processed. */ 132 if (ret == 0) { 133 ret = -1; 134 } 135 break; 136 } 137 i++; 138 } 139 return ret; 140 #endif 141 } 142 143 ssize_t iov_send_recv(int sockfd, struct iovec *iov, unsigned iov_cnt, 144 size_t offset, size_t bytes, 145 bool do_send) 146 { 147 ssize_t total = 0; 148 ssize_t ret; 149 size_t orig_len, tail; 150 unsigned niov; 151 152 while (bytes > 0) { 153 /* Find the start position, skipping `offset' bytes: 154 * first, skip all full-sized vector elements, */ 155 for (niov = 0; niov < iov_cnt && offset >= iov[niov].iov_len; ++niov) { 156 offset -= iov[niov].iov_len; 157 } 158 159 /* niov == iov_cnt would only be valid if bytes == 0, which 160 * we already ruled out in the loop condition. */ 161 assert(niov < iov_cnt); 162 iov += niov; 163 iov_cnt -= niov; 164 165 if (offset) { 166 /* second, skip `offset' bytes from the (now) first element, 167 * undo it on exit */ 168 iov[0].iov_base += offset; 169 iov[0].iov_len -= offset; 170 } 171 /* Find the end position skipping `bytes' bytes: */ 172 /* first, skip all full-sized elements */ 173 tail = bytes; 174 for (niov = 0; niov < iov_cnt && iov[niov].iov_len <= tail; ++niov) { 175 tail -= iov[niov].iov_len; 176 } 177 if (tail) { 178 /* second, fixup the last element, and remember the original 179 * length */ 180 assert(niov < iov_cnt); 181 assert(iov[niov].iov_len > tail); 182 orig_len = iov[niov].iov_len; 183 iov[niov++].iov_len = tail; 184 } 185 186 ret = do_send_recv(sockfd, iov, niov, do_send); 187 188 /* Undo the changes above before checking for errors */ 189 if (tail) { 190 iov[niov-1].iov_len = orig_len; 191 } 192 if (offset) { 193 iov[0].iov_base -= offset; 194 iov[0].iov_len += offset; 195 } 196 197 if (ret < 0) { 198 assert(errno != EINTR); 199 if (errno == EAGAIN && total > 0) { 200 return total; 201 } 202 return -1; 203 } 204 205 if (ret == 0 && !do_send) { 206 /* recv returns 0 when the peer has performed an orderly 207 * shutdown. */ 208 break; 209 } 210 211 /* Prepare for the next iteration */ 212 offset += ret; 213 total += ret; 214 bytes -= ret; 215 } 216 217 return total; 218 } 219 220 221 void iov_hexdump(const struct iovec *iov, const unsigned int iov_cnt, 222 FILE *fp, const char *prefix, size_t limit) 223 { 224 int v; 225 size_t size = 0; 226 char *buf; 227 228 for (v = 0; v < iov_cnt; v++) { 229 size += iov[v].iov_len; 230 } 231 size = size > limit ? limit : size; 232 buf = g_malloc(size); 233 iov_to_buf(iov, iov_cnt, 0, buf, size); 234 qemu_hexdump(buf, fp, prefix, size); 235 g_free(buf); 236 } 237 238 unsigned iov_copy(struct iovec *dst_iov, unsigned int dst_iov_cnt, 239 const struct iovec *iov, unsigned int iov_cnt, 240 size_t offset, size_t bytes) 241 { 242 size_t len; 243 unsigned int i, j; 244 for (i = 0, j = 0; i < iov_cnt && j < dst_iov_cnt && bytes; i++) { 245 if (offset >= iov[i].iov_len) { 246 offset -= iov[i].iov_len; 247 continue; 248 } 249 len = MIN(bytes, iov[i].iov_len - offset); 250 251 dst_iov[j].iov_base = iov[i].iov_base + offset; 252 dst_iov[j].iov_len = len; 253 j++; 254 bytes -= len; 255 offset = 0; 256 } 257 assert(offset == 0); 258 return j; 259 } 260 261 /* io vectors */ 262 263 void qemu_iovec_init(QEMUIOVector *qiov, int alloc_hint) 264 { 265 qiov->iov = g_malloc(alloc_hint * sizeof(struct iovec)); 266 qiov->niov = 0; 267 qiov->nalloc = alloc_hint; 268 qiov->size = 0; 269 } 270 271 void qemu_iovec_init_external(QEMUIOVector *qiov, struct iovec *iov, int niov) 272 { 273 int i; 274 275 qiov->iov = iov; 276 qiov->niov = niov; 277 qiov->nalloc = -1; 278 qiov->size = 0; 279 for (i = 0; i < niov; i++) 280 qiov->size += iov[i].iov_len; 281 } 282 283 void qemu_iovec_add(QEMUIOVector *qiov, void *base, size_t len) 284 { 285 assert(qiov->nalloc != -1); 286 287 if (qiov->niov == qiov->nalloc) { 288 qiov->nalloc = 2 * qiov->nalloc + 1; 289 qiov->iov = g_realloc(qiov->iov, qiov->nalloc * sizeof(struct iovec)); 290 } 291 qiov->iov[qiov->niov].iov_base = base; 292 qiov->iov[qiov->niov].iov_len = len; 293 qiov->size += len; 294 ++qiov->niov; 295 } 296 297 /* 298 * Concatenates (partial) iovecs from src_iov to the end of dst. 299 * It starts copying after skipping `soffset' bytes at the 300 * beginning of src and adds individual vectors from src to 301 * dst copies up to `sbytes' bytes total, or up to the end 302 * of src_iov if it comes first. This way, it is okay to specify 303 * very large value for `sbytes' to indicate "up to the end 304 * of src". 305 * Only vector pointers are processed, not the actual data buffers. 306 */ 307 void qemu_iovec_concat_iov(QEMUIOVector *dst, 308 struct iovec *src_iov, unsigned int src_cnt, 309 size_t soffset, size_t sbytes) 310 { 311 int i; 312 size_t done; 313 314 if (!sbytes) { 315 return; 316 } 317 assert(dst->nalloc != -1); 318 for (i = 0, done = 0; done < sbytes && i < src_cnt; i++) { 319 if (soffset < src_iov[i].iov_len) { 320 size_t len = MIN(src_iov[i].iov_len - soffset, sbytes - done); 321 qemu_iovec_add(dst, src_iov[i].iov_base + soffset, len); 322 done += len; 323 soffset = 0; 324 } else { 325 soffset -= src_iov[i].iov_len; 326 } 327 } 328 assert(soffset == 0); /* offset beyond end of src */ 329 } 330 331 /* 332 * Concatenates (partial) iovecs from src to the end of dst. 333 * It starts copying after skipping `soffset' bytes at the 334 * beginning of src and adds individual vectors from src to 335 * dst copies up to `sbytes' bytes total, or up to the end 336 * of src if it comes first. This way, it is okay to specify 337 * very large value for `sbytes' to indicate "up to the end 338 * of src". 339 * Only vector pointers are processed, not the actual data buffers. 340 */ 341 void qemu_iovec_concat(QEMUIOVector *dst, 342 QEMUIOVector *src, size_t soffset, size_t sbytes) 343 { 344 qemu_iovec_concat_iov(dst, src->iov, src->niov, soffset, sbytes); 345 } 346 347 void qemu_iovec_destroy(QEMUIOVector *qiov) 348 { 349 assert(qiov->nalloc != -1); 350 351 qemu_iovec_reset(qiov); 352 g_free(qiov->iov); 353 qiov->nalloc = 0; 354 qiov->iov = NULL; 355 } 356 357 void qemu_iovec_reset(QEMUIOVector *qiov) 358 { 359 assert(qiov->nalloc != -1); 360 361 qiov->niov = 0; 362 qiov->size = 0; 363 } 364 365 size_t qemu_iovec_to_buf(QEMUIOVector *qiov, size_t offset, 366 void *buf, size_t bytes) 367 { 368 return iov_to_buf(qiov->iov, qiov->niov, offset, buf, bytes); 369 } 370 371 size_t qemu_iovec_from_buf(QEMUIOVector *qiov, size_t offset, 372 const void *buf, size_t bytes) 373 { 374 return iov_from_buf(qiov->iov, qiov->niov, offset, buf, bytes); 375 } 376 377 size_t qemu_iovec_memset(QEMUIOVector *qiov, size_t offset, 378 int fillc, size_t bytes) 379 { 380 return iov_memset(qiov->iov, qiov->niov, offset, fillc, bytes); 381 } 382 383 size_t iov_discard_front(struct iovec **iov, unsigned int *iov_cnt, 384 size_t bytes) 385 { 386 size_t total = 0; 387 struct iovec *cur; 388 389 for (cur = *iov; *iov_cnt > 0; cur++) { 390 if (cur->iov_len > bytes) { 391 cur->iov_base += bytes; 392 cur->iov_len -= bytes; 393 total += bytes; 394 break; 395 } 396 397 bytes -= cur->iov_len; 398 total += cur->iov_len; 399 *iov_cnt -= 1; 400 } 401 402 *iov = cur; 403 return total; 404 } 405 406 size_t iov_discard_back(struct iovec *iov, unsigned int *iov_cnt, 407 size_t bytes) 408 { 409 size_t total = 0; 410 struct iovec *cur; 411 412 if (*iov_cnt == 0) { 413 return 0; 414 } 415 416 cur = iov + (*iov_cnt - 1); 417 418 while (*iov_cnt > 0) { 419 if (cur->iov_len > bytes) { 420 cur->iov_len -= bytes; 421 total += bytes; 422 break; 423 } 424 425 bytes -= cur->iov_len; 426 total += cur->iov_len; 427 cur--; 428 *iov_cnt -= 1; 429 } 430 431 return total; 432 } 433