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 #include "qemu/sockets.h" 21 22 size_t iov_from_buf(const struct iovec *iov, unsigned int iov_cnt, 23 size_t offset, const void *buf, size_t bytes) 24 { 25 size_t done; 26 unsigned int i; 27 for (i = 0, done = 0; (offset || done < bytes) && i < iov_cnt; i++) { 28 if (offset < iov[i].iov_len) { 29 size_t len = MIN(iov[i].iov_len - offset, bytes - done); 30 memcpy(iov[i].iov_base + offset, buf + done, len); 31 done += len; 32 offset = 0; 33 } else { 34 offset -= iov[i].iov_len; 35 } 36 } 37 assert(offset == 0); 38 return done; 39 } 40 41 size_t iov_to_buf(const struct iovec *iov, const unsigned int iov_cnt, 42 size_t offset, void *buf, size_t bytes) 43 { 44 size_t done; 45 unsigned int i; 46 for (i = 0, done = 0; (offset || done < bytes) && i < iov_cnt; i++) { 47 if (offset < iov[i].iov_len) { 48 size_t len = MIN(iov[i].iov_len - offset, bytes - done); 49 memcpy(buf + done, iov[i].iov_base + offset, len); 50 done += len; 51 offset = 0; 52 } else { 53 offset -= iov[i].iov_len; 54 } 55 } 56 assert(offset == 0); 57 return done; 58 } 59 60 size_t iov_memset(const struct iovec *iov, const unsigned int iov_cnt, 61 size_t offset, int fillc, size_t bytes) 62 { 63 size_t done; 64 unsigned int i; 65 for (i = 0, done = 0; (offset || done < bytes) && i < iov_cnt; i++) { 66 if (offset < iov[i].iov_len) { 67 size_t len = MIN(iov[i].iov_len - offset, bytes - done); 68 memset(iov[i].iov_base + offset, fillc, len); 69 done += len; 70 offset = 0; 71 } else { 72 offset -= iov[i].iov_len; 73 } 74 } 75 assert(offset == 0); 76 return done; 77 } 78 79 size_t iov_size(const struct iovec *iov, const unsigned int iov_cnt) 80 { 81 size_t len; 82 unsigned int i; 83 84 len = 0; 85 for (i = 0; i < iov_cnt; i++) { 86 len += iov[i].iov_len; 87 } 88 return len; 89 } 90 91 /* helper function for iov_send_recv() */ 92 static ssize_t 93 do_send_recv(int sockfd, struct iovec *iov, unsigned iov_cnt, bool do_send) 94 { 95 #ifdef CONFIG_POSIX 96 ssize_t ret; 97 struct msghdr msg; 98 memset(&msg, 0, sizeof(msg)); 99 msg.msg_iov = iov; 100 msg.msg_iovlen = iov_cnt; 101 do { 102 ret = do_send 103 ? sendmsg(sockfd, &msg, 0) 104 : recvmsg(sockfd, &msg, 0); 105 } while (ret < 0 && errno == EINTR); 106 return ret; 107 #else 108 /* else send piece-by-piece */ 109 /*XXX Note: windows has WSASend() and WSARecv() */ 110 unsigned i = 0; 111 ssize_t ret = 0; 112 while (i < iov_cnt) { 113 ssize_t r = do_send 114 ? send(sockfd, iov[i].iov_base, iov[i].iov_len, 0) 115 : recv(sockfd, iov[i].iov_base, iov[i].iov_len, 0); 116 if (r > 0) { 117 ret += r; 118 } else if (!r) { 119 break; 120 } else if (errno == EINTR) { 121 continue; 122 } else { 123 /* else it is some "other" error, 124 * only return if there was no data processed. */ 125 if (ret == 0) { 126 ret = -1; 127 } 128 break; 129 } 130 i++; 131 } 132 return ret; 133 #endif 134 } 135 136 ssize_t iov_send_recv(int sockfd, struct iovec *iov, unsigned iov_cnt, 137 size_t offset, size_t bytes, 138 bool do_send) 139 { 140 ssize_t total = 0; 141 ssize_t ret; 142 size_t orig_len, tail; 143 unsigned niov; 144 145 while (bytes > 0) { 146 /* Find the start position, skipping `offset' bytes: 147 * first, skip all full-sized vector elements, */ 148 for (niov = 0; niov < iov_cnt && offset >= iov[niov].iov_len; ++niov) { 149 offset -= iov[niov].iov_len; 150 } 151 152 /* niov == iov_cnt would only be valid if bytes == 0, which 153 * we already ruled out in the loop condition. */ 154 assert(niov < iov_cnt); 155 iov += niov; 156 iov_cnt -= niov; 157 158 if (offset) { 159 /* second, skip `offset' bytes from the (now) first element, 160 * undo it on exit */ 161 iov[0].iov_base += offset; 162 iov[0].iov_len -= offset; 163 } 164 /* Find the end position skipping `bytes' bytes: */ 165 /* first, skip all full-sized elements */ 166 tail = bytes; 167 for (niov = 0; niov < iov_cnt && iov[niov].iov_len <= tail; ++niov) { 168 tail -= iov[niov].iov_len; 169 } 170 if (tail) { 171 /* second, fixup the last element, and remember the original 172 * length */ 173 assert(niov < iov_cnt); 174 assert(iov[niov].iov_len > tail); 175 orig_len = iov[niov].iov_len; 176 iov[niov++].iov_len = tail; 177 ret = do_send_recv(sockfd, iov, niov, do_send); 178 /* Undo the changes above before checking for errors */ 179 iov[niov-1].iov_len = orig_len; 180 } else { 181 ret = do_send_recv(sockfd, iov, niov, do_send); 182 } 183 if (offset) { 184 iov[0].iov_base -= offset; 185 iov[0].iov_len += offset; 186 } 187 188 if (ret < 0) { 189 assert(errno != EINTR); 190 if (errno == EAGAIN && total > 0) { 191 return total; 192 } 193 return -1; 194 } 195 196 if (ret == 0 && !do_send) { 197 /* recv returns 0 when the peer has performed an orderly 198 * shutdown. */ 199 break; 200 } 201 202 /* Prepare for the next iteration */ 203 offset += ret; 204 total += ret; 205 bytes -= ret; 206 } 207 208 return total; 209 } 210 211 212 void iov_hexdump(const struct iovec *iov, const unsigned int iov_cnt, 213 FILE *fp, const char *prefix, size_t limit) 214 { 215 int v; 216 size_t size = 0; 217 char *buf; 218 219 for (v = 0; v < iov_cnt; v++) { 220 size += iov[v].iov_len; 221 } 222 size = size > limit ? limit : size; 223 buf = g_malloc(size); 224 iov_to_buf(iov, iov_cnt, 0, buf, size); 225 qemu_hexdump(buf, fp, prefix, size); 226 g_free(buf); 227 } 228 229 unsigned iov_copy(struct iovec *dst_iov, unsigned int dst_iov_cnt, 230 const struct iovec *iov, unsigned int iov_cnt, 231 size_t offset, size_t bytes) 232 { 233 size_t len; 234 unsigned int i, j; 235 for (i = 0, j = 0; i < iov_cnt && j < dst_iov_cnt && bytes; i++) { 236 if (offset >= iov[i].iov_len) { 237 offset -= iov[i].iov_len; 238 continue; 239 } 240 len = MIN(bytes, iov[i].iov_len - offset); 241 242 dst_iov[j].iov_base = iov[i].iov_base + offset; 243 dst_iov[j].iov_len = len; 244 j++; 245 bytes -= len; 246 offset = 0; 247 } 248 assert(offset == 0); 249 return j; 250 } 251 252 /* io vectors */ 253 254 void qemu_iovec_init(QEMUIOVector *qiov, int alloc_hint) 255 { 256 qiov->iov = g_malloc(alloc_hint * sizeof(struct iovec)); 257 qiov->niov = 0; 258 qiov->nalloc = alloc_hint; 259 qiov->size = 0; 260 } 261 262 void qemu_iovec_init_external(QEMUIOVector *qiov, struct iovec *iov, int niov) 263 { 264 int i; 265 266 qiov->iov = iov; 267 qiov->niov = niov; 268 qiov->nalloc = -1; 269 qiov->size = 0; 270 for (i = 0; i < niov; i++) 271 qiov->size += iov[i].iov_len; 272 } 273 274 void qemu_iovec_add(QEMUIOVector *qiov, void *base, size_t len) 275 { 276 assert(qiov->nalloc != -1); 277 278 if (qiov->niov == qiov->nalloc) { 279 qiov->nalloc = 2 * qiov->nalloc + 1; 280 qiov->iov = g_realloc(qiov->iov, qiov->nalloc * sizeof(struct iovec)); 281 } 282 qiov->iov[qiov->niov].iov_base = base; 283 qiov->iov[qiov->niov].iov_len = len; 284 qiov->size += len; 285 ++qiov->niov; 286 } 287 288 /* 289 * Concatenates (partial) iovecs from src_iov to the end of dst. 290 * It starts copying after skipping `soffset' bytes at the 291 * beginning of src and adds individual vectors from src to 292 * dst copies up to `sbytes' bytes total, or up to the end 293 * of src_iov if it comes first. This way, it is okay to specify 294 * very large value for `sbytes' to indicate "up to the end 295 * of src". 296 * Only vector pointers are processed, not the actual data buffers. 297 */ 298 void qemu_iovec_concat_iov(QEMUIOVector *dst, 299 struct iovec *src_iov, unsigned int src_cnt, 300 size_t soffset, size_t sbytes) 301 { 302 int i; 303 size_t done; 304 305 if (!sbytes) { 306 return; 307 } 308 assert(dst->nalloc != -1); 309 for (i = 0, done = 0; done < sbytes && i < src_cnt; i++) { 310 if (soffset < src_iov[i].iov_len) { 311 size_t len = MIN(src_iov[i].iov_len - soffset, sbytes - done); 312 qemu_iovec_add(dst, src_iov[i].iov_base + soffset, len); 313 done += len; 314 soffset = 0; 315 } else { 316 soffset -= src_iov[i].iov_len; 317 } 318 } 319 assert(soffset == 0); /* offset beyond end of src */ 320 } 321 322 /* 323 * Concatenates (partial) iovecs from src to the end of dst. 324 * It starts copying after skipping `soffset' bytes at the 325 * beginning of src and adds individual vectors from src to 326 * dst copies up to `sbytes' bytes total, or up to the end 327 * of src if it comes first. This way, it is okay to specify 328 * very large value for `sbytes' to indicate "up to the end 329 * of src". 330 * Only vector pointers are processed, not the actual data buffers. 331 */ 332 void qemu_iovec_concat(QEMUIOVector *dst, 333 QEMUIOVector *src, size_t soffset, size_t sbytes) 334 { 335 qemu_iovec_concat_iov(dst, src->iov, src->niov, soffset, sbytes); 336 } 337 338 /* 339 * Check if the contents of the iovecs are all zero 340 */ 341 bool qemu_iovec_is_zero(QEMUIOVector *qiov) 342 { 343 int i; 344 for (i = 0; i < qiov->niov; i++) { 345 size_t offs = QEMU_ALIGN_DOWN(qiov->iov[i].iov_len, 4 * sizeof(long)); 346 uint8_t *ptr = qiov->iov[i].iov_base; 347 if (offs && !buffer_is_zero(qiov->iov[i].iov_base, offs)) { 348 return false; 349 } 350 for (; offs < qiov->iov[i].iov_len; offs++) { 351 if (ptr[offs]) { 352 return false; 353 } 354 } 355 } 356 return true; 357 } 358 359 void qemu_iovec_destroy(QEMUIOVector *qiov) 360 { 361 assert(qiov->nalloc != -1); 362 363 qemu_iovec_reset(qiov); 364 g_free(qiov->iov); 365 qiov->nalloc = 0; 366 qiov->iov = NULL; 367 } 368 369 void qemu_iovec_reset(QEMUIOVector *qiov) 370 { 371 assert(qiov->nalloc != -1); 372 373 qiov->niov = 0; 374 qiov->size = 0; 375 } 376 377 size_t qemu_iovec_to_buf(QEMUIOVector *qiov, size_t offset, 378 void *buf, size_t bytes) 379 { 380 return iov_to_buf(qiov->iov, qiov->niov, offset, buf, bytes); 381 } 382 383 size_t qemu_iovec_from_buf(QEMUIOVector *qiov, size_t offset, 384 const void *buf, size_t bytes) 385 { 386 return iov_from_buf(qiov->iov, qiov->niov, offset, buf, bytes); 387 } 388 389 size_t qemu_iovec_memset(QEMUIOVector *qiov, size_t offset, 390 int fillc, size_t bytes) 391 { 392 return iov_memset(qiov->iov, qiov->niov, offset, fillc, bytes); 393 } 394 395 /** 396 * Check that I/O vector contents are identical 397 * 398 * The IO vectors must have the same structure (same length of all parts). 399 * A typical usage is to compare vectors created with qemu_iovec_clone(). 400 * 401 * @a: I/O vector 402 * @b: I/O vector 403 * @ret: Offset to first mismatching byte or -1 if match 404 */ 405 ssize_t qemu_iovec_compare(QEMUIOVector *a, QEMUIOVector *b) 406 { 407 int i; 408 ssize_t offset = 0; 409 410 assert(a->niov == b->niov); 411 for (i = 0; i < a->niov; i++) { 412 size_t len = 0; 413 uint8_t *p = (uint8_t *)a->iov[i].iov_base; 414 uint8_t *q = (uint8_t *)b->iov[i].iov_base; 415 416 assert(a->iov[i].iov_len == b->iov[i].iov_len); 417 while (len < a->iov[i].iov_len && *p++ == *q++) { 418 len++; 419 } 420 421 offset += len; 422 423 if (len != a->iov[i].iov_len) { 424 return offset; 425 } 426 } 427 return -1; 428 } 429 430 typedef struct { 431 int src_index; 432 struct iovec *src_iov; 433 void *dest_base; 434 } IOVectorSortElem; 435 436 static int sortelem_cmp_src_base(const void *a, const void *b) 437 { 438 const IOVectorSortElem *elem_a = a; 439 const IOVectorSortElem *elem_b = b; 440 441 /* Don't overflow */ 442 if (elem_a->src_iov->iov_base < elem_b->src_iov->iov_base) { 443 return -1; 444 } else if (elem_a->src_iov->iov_base > elem_b->src_iov->iov_base) { 445 return 1; 446 } else { 447 return 0; 448 } 449 } 450 451 static int sortelem_cmp_src_index(const void *a, const void *b) 452 { 453 const IOVectorSortElem *elem_a = a; 454 const IOVectorSortElem *elem_b = b; 455 456 return elem_a->src_index - elem_b->src_index; 457 } 458 459 /** 460 * Copy contents of I/O vector 461 * 462 * The relative relationships of overlapping iovecs are preserved. This is 463 * necessary to ensure identical semantics in the cloned I/O vector. 464 */ 465 void qemu_iovec_clone(QEMUIOVector *dest, const QEMUIOVector *src, void *buf) 466 { 467 IOVectorSortElem sortelems[src->niov]; 468 void *last_end; 469 int i; 470 471 /* Sort by source iovecs by base address */ 472 for (i = 0; i < src->niov; i++) { 473 sortelems[i].src_index = i; 474 sortelems[i].src_iov = &src->iov[i]; 475 } 476 qsort(sortelems, src->niov, sizeof(sortelems[0]), sortelem_cmp_src_base); 477 478 /* Allocate buffer space taking into account overlapping iovecs */ 479 last_end = NULL; 480 for (i = 0; i < src->niov; i++) { 481 struct iovec *cur = sortelems[i].src_iov; 482 ptrdiff_t rewind = 0; 483 484 /* Detect overlap */ 485 if (last_end && last_end > cur->iov_base) { 486 rewind = last_end - cur->iov_base; 487 } 488 489 sortelems[i].dest_base = buf - rewind; 490 buf += cur->iov_len - MIN(rewind, cur->iov_len); 491 last_end = MAX(cur->iov_base + cur->iov_len, last_end); 492 } 493 494 /* Sort by source iovec index and build destination iovec */ 495 qsort(sortelems, src->niov, sizeof(sortelems[0]), sortelem_cmp_src_index); 496 for (i = 0; i < src->niov; i++) { 497 qemu_iovec_add(dest, sortelems[i].dest_base, src->iov[i].iov_len); 498 } 499 } 500 501 size_t iov_discard_front(struct iovec **iov, unsigned int *iov_cnt, 502 size_t bytes) 503 { 504 size_t total = 0; 505 struct iovec *cur; 506 507 for (cur = *iov; *iov_cnt > 0; cur++) { 508 if (cur->iov_len > bytes) { 509 cur->iov_base += bytes; 510 cur->iov_len -= bytes; 511 total += bytes; 512 break; 513 } 514 515 bytes -= cur->iov_len; 516 total += cur->iov_len; 517 *iov_cnt -= 1; 518 } 519 520 *iov = cur; 521 return total; 522 } 523 524 size_t iov_discard_back(struct iovec *iov, unsigned int *iov_cnt, 525 size_t bytes) 526 { 527 size_t total = 0; 528 struct iovec *cur; 529 530 if (*iov_cnt == 0) { 531 return 0; 532 } 533 534 cur = iov + (*iov_cnt - 1); 535 536 while (*iov_cnt > 0) { 537 if (cur->iov_len > bytes) { 538 cur->iov_len -= bytes; 539 total += bytes; 540 break; 541 } 542 543 bytes -= cur->iov_len; 544 total += cur->iov_len; 545 cur--; 546 *iov_cnt -= 1; 547 } 548 549 return total; 550 } 551