| 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 | * Copyright (c) 2024 Seagate Technology LLC and/or its Affiliates |
| 7 | * |
| 8 | * Author(s): |
| 9 | * Anthony Liguori <aliguori@us.ibm.com> |
| 10 | * Amit Shah <amit.shah@redhat.com> |
| 11 | * Michael Tokarev <mjt@tls.msk.ru> |
| 12 | * |
| 13 | * This work is licensed under the terms of the GNU GPL, version 2. See |
| 14 | * the COPYING file in the top-level directory. |
| 15 | * |
| 16 | * Contributions after 2012-01-13 are licensed under the terms of the |
| 17 | * GNU GPL, version 2 or (at your option) any later version. |
| 18 | */ |
| 19 | |
| 20 | #include "qemu/osdep.h" |
| 21 | #include "qemu/iov.h" |
| 22 | #include "qemu/sockets.h" |
| 23 | #include "qemu/cutils.h" |
| 24 | |
| 25 | size_t iov_from_buf_full(const struct iovec *iov, unsigned int iov_cnt, |
| 26 | size_t offset, const void *buf, size_t bytes) |
| 27 | { |
| 28 | size_t done; |
| 29 | unsigned int i; |
| 30 | for (i = 0, done = 0; (offset || done < bytes) && i < iov_cnt; i++) { |
| 31 | if (offset < iov[i].iov_len) { |
| 32 | size_t len = MIN(iov[i].iov_len - offset, bytes - done); |
| 33 | memcpy(iov[i].iov_base + offset, buf + done, len); |
| 34 | done += len; |
| 35 | offset = 0; |
| 36 | } else { |
| 37 | offset -= iov[i].iov_len; |
| 38 | } |
| 39 | } |
| 40 | return done; |
| 41 | } |
| 42 | |
| 43 | size_t iov_to_buf_full(const struct iovec *iov, const unsigned int iov_cnt, |
| 44 | size_t offset, void *buf, size_t bytes) |
| 45 | { |
| 46 | size_t done; |
| 47 | unsigned int i; |
| 48 | for (i = 0, done = 0; (offset || done < bytes) && i < iov_cnt; i++) { |
| 49 | if (offset < iov[i].iov_len) { |
| 50 | size_t len = MIN(iov[i].iov_len - offset, bytes - done); |
| 51 | memcpy(buf + done, iov[i].iov_base + offset, len); |
| 52 | done += len; |
| 53 | offset = 0; |
| 54 | } else { |
| 55 | offset -= iov[i].iov_len; |
| 56 | } |
| 57 | } |
| 58 | return done; |
| 59 | } |
| 60 | |
| 61 | size_t iov_memset(const struct iovec *iov, const unsigned int iov_cnt, |
| 62 | size_t offset, int fillc, size_t bytes) |
| 63 | { |
| 64 | size_t done; |
| 65 | unsigned int i; |
| 66 | for (i = 0, done = 0; (offset || done < bytes) && i < iov_cnt; i++) { |
| 67 | if (offset < iov[i].iov_len) { |
| 68 | size_t len = MIN(iov[i].iov_len - offset, bytes - done); |
| 69 | memset(iov[i].iov_base + offset, fillc, len); |
| 70 | done += len; |
| 71 | offset = 0; |
| 72 | } else { |
| 73 | offset -= iov[i].iov_len; |
| 74 | } |
| 75 | } |
| 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, int flags, struct iovec *iov, unsigned iov_cnt, |
| 94 | bool do_send) |
| 95 | { |
| 96 | #ifdef CONFIG_POSIX |
| 97 | ssize_t ret; |
| 98 | struct msghdr msg; |
| 99 | memset(&msg, 0, sizeof(msg)); |
| 100 | msg.msg_iov = iov; |
| 101 | msg.msg_iovlen = iov_cnt; |
| 102 | do { |
| 103 | ret = do_send |
| 104 | ? sendmsg(sockfd, &msg, flags) |
| 105 | : recvmsg(sockfd, &msg, flags); |
| 106 | } while (ret < 0 && errno == EINTR); |
| 107 | return ret; |
| 108 | #else |
| 109 | /* else send piece-by-piece */ |
| 110 | /*XXX Note: windows has WSASend() and WSARecv() */ |
| 111 | unsigned i = 0; |
| 112 | ssize_t ret = 0; |
| 113 | ssize_t off = 0; |
| 114 | while (i < iov_cnt) { |
| 115 | ssize_t r = do_send |
| 116 | ? send(sockfd, iov[i].iov_base + off, iov[i].iov_len - off, flags) |
| 117 | : recv(sockfd, iov[i].iov_base + off, iov[i].iov_len - off, flags); |
| 118 | if (r > 0) { |
| 119 | ret += r; |
| 120 | off += r; |
| 121 | if (off < iov[i].iov_len) { |
| 122 | continue; |
| 123 | } |
| 124 | } else if (!r) { |
| 125 | break; |
| 126 | } else if (errno == EINTR) { |
| 127 | continue; |
| 128 | } else { |
| 129 | /* else it is some "other" error, |
| 130 | * only return if there was no data processed. */ |
| 131 | if (ret == 0) { |
| 132 | ret = -1; |
| 133 | } |
| 134 | break; |
| 135 | } |
| 136 | off = 0; |
| 137 | i++; |
| 138 | } |
| 139 | return ret; |
| 140 | #endif |
| 141 | } |
| 142 | |
| 143 | ssize_t iov_send_recv(int sockfd, const struct iovec *_iov, unsigned iov_cnt, |
| 144 | size_t offset, size_t bytes, |
| 145 | bool do_send) |
| 146 | { |
| 147 | return iov_send_recv_with_flags(sockfd, 0, _iov, iov_cnt, offset, bytes, |
| 148 | do_send); |
| 149 | } |
| 150 | |
| 151 | ssize_t iov_send_recv_with_flags(int sockfd, int sockflags, |
| 152 | const struct iovec *_iov, |
| 153 | unsigned iov_cnt, size_t offset, |
| 154 | size_t bytes, bool do_send) |
| 155 | { |
| 156 | ssize_t total = 0; |
| 157 | ssize_t ret; |
| 158 | size_t orig_len, tail; |
| 159 | unsigned niov; |
| 160 | struct iovec *local_iov, *iov; |
| 161 | |
| 162 | if (bytes <= 0) { |
| 163 | return 0; |
| 164 | } |
| 165 | |
| 166 | local_iov = g_new0(struct iovec, iov_cnt); |
| 167 | iov_copy(local_iov, iov_cnt, _iov, iov_cnt, offset, bytes); |
| 168 | offset = 0; |
| 169 | iov = local_iov; |
| 170 | |
| 171 | while (bytes > 0) { |
| 172 | /* Find the start position, skipping `offset' bytes: |
| 173 | * first, skip all full-sized vector elements, */ |
| 174 | for (niov = 0; niov < iov_cnt && offset >= iov[niov].iov_len; ++niov) { |
| 175 | offset -= iov[niov].iov_len; |
| 176 | } |
| 177 | |
| 178 | /* niov == iov_cnt would only be valid if bytes == 0, which |
| 179 | * we already ruled out in the loop condition. */ |
| 180 | assert(niov < iov_cnt); |
| 181 | iov += niov; |
| 182 | iov_cnt -= niov; |
| 183 | |
| 184 | if (offset) { |
| 185 | /* second, skip `offset' bytes from the (now) first element, |
| 186 | * undo it on exit */ |
| 187 | iov[0].iov_base += offset; |
| 188 | iov[0].iov_len -= offset; |
| 189 | } |
| 190 | /* Find the end position skipping `bytes' bytes: */ |
| 191 | /* first, skip all full-sized elements */ |
| 192 | tail = bytes; |
| 193 | for (niov = 0; niov < iov_cnt && iov[niov].iov_len <= tail; ++niov) { |
| 194 | tail -= iov[niov].iov_len; |
| 195 | } |
| 196 | if (tail) { |
| 197 | /* second, fixup the last element, and remember the original |
| 198 | * length */ |
| 199 | assert(niov < iov_cnt); |
| 200 | assert(iov[niov].iov_len > tail); |
| 201 | orig_len = iov[niov].iov_len; |
| 202 | iov[niov++].iov_len = tail; |
| 203 | ret = do_send_recv(sockfd, sockflags, iov, niov, do_send); |
| 204 | /* Undo the changes above before checking for errors */ |
| 205 | iov[niov-1].iov_len = orig_len; |
| 206 | } else { |
| 207 | ret = do_send_recv(sockfd, sockflags, iov, niov, do_send); |
| 208 | } |
| 209 | if (offset) { |
| 210 | iov[0].iov_base -= offset; |
| 211 | iov[0].iov_len += offset; |
| 212 | } |
| 213 | |
| 214 | if (ret < 0) { |
| 215 | assert(errno != EINTR); |
| 216 | g_free(local_iov); |
| 217 | if (errno == EAGAIN && total > 0) { |
| 218 | return total; |
| 219 | } |
| 220 | return -1; |
| 221 | } |
| 222 | |
| 223 | if (ret == 0 && !do_send) { |
| 224 | /* recv returns 0 when the peer has performed an orderly |
| 225 | * shutdown. */ |
| 226 | break; |
| 227 | } |
| 228 | |
| 229 | /* Prepare for the next iteration */ |
| 230 | offset += ret; |
| 231 | total += ret; |
| 232 | bytes -= ret; |
| 233 | } |
| 234 | |
| 235 | g_free(local_iov); |
| 236 | return total; |
| 237 | } |
| 238 | |
| 239 | |
| 240 | void iov_hexdump(const struct iovec *iov, const unsigned int iov_cnt, |
| 241 | FILE *fp, const char *prefix, size_t limit) |
| 242 | { |
| 243 | int v; |
| 244 | size_t size = 0; |
| 245 | char *buf; |
| 246 | |
| 247 | for (v = 0; v < iov_cnt; v++) { |
| 248 | size += iov[v].iov_len; |
| 249 | } |
| 250 | size = size > limit ? limit : size; |
| 251 | buf = g_malloc(size); |
| 252 | iov_to_buf(iov, iov_cnt, 0, buf, size); |
| 253 | qemu_hexdump(fp, prefix, buf, size); |
| 254 | g_free(buf); |
| 255 | } |
| 256 | |
| 257 | unsigned iov_copy(struct iovec *dst_iov, unsigned int dst_iov_cnt, |
| 258 | const struct iovec *iov, unsigned int iov_cnt, |
| 259 | size_t offset, size_t bytes) |
| 260 | { |
| 261 | size_t len; |
| 262 | unsigned int i, j; |
| 263 | for (i = 0, j = 0; |
| 264 | i < iov_cnt && j < dst_iov_cnt && (offset || bytes); i++) { |
| 265 | if (offset >= iov[i].iov_len) { |
| 266 | offset -= iov[i].iov_len; |
| 267 | continue; |
| 268 | } |
| 269 | len = MIN(bytes, iov[i].iov_len - offset); |
| 270 | |
| 271 | dst_iov[j].iov_base = iov[i].iov_base + offset; |
| 272 | dst_iov[j].iov_len = len; |
| 273 | j++; |
| 274 | bytes -= len; |
| 275 | offset = 0; |
| 276 | } |
| 277 | return j; |
| 278 | } |
| 279 | |
| 280 | /* io vectors */ |
| 281 | |
| 282 | void qemu_iovec_init(QEMUIOVector *qiov, int alloc_hint) |
| 283 | { |
| 284 | qiov->iov = g_new(struct iovec, alloc_hint); |
| 285 | qiov->niov = 0; |
| 286 | qiov->nalloc = alloc_hint; |
| 287 | qiov->size = 0; |
| 288 | } |
| 289 | |
| 290 | void qemu_iovec_init_external(QEMUIOVector *qiov, struct iovec *iov, int niov) |
| 291 | { |
| 292 | int i; |
| 293 | |
| 294 | qiov->iov = iov; |
| 295 | qiov->niov = niov; |
| 296 | qiov->nalloc = -1; |
| 297 | qiov->size = 0; |
| 298 | for (i = 0; i < niov; i++) |
| 299 | qiov->size += iov[i].iov_len; |
| 300 | } |
| 301 | |
| 302 | void qemu_iovec_add(QEMUIOVector *qiov, void *base, size_t len) |
| 303 | { |
| 304 | assert(qiov->nalloc != -1); |
| 305 | |
| 306 | if (qiov->niov == qiov->nalloc) { |
| 307 | qiov->nalloc = 2 * qiov->nalloc + 1; |
| 308 | qiov->iov = g_renew(struct iovec, qiov->iov, qiov->nalloc); |
| 309 | } |
| 310 | qiov->iov[qiov->niov].iov_base = base; |
| 311 | qiov->iov[qiov->niov].iov_len = len; |
| 312 | qiov->size += len; |
| 313 | ++qiov->niov; |
| 314 | } |
| 315 | |
| 316 | /* |
| 317 | * Concatenates (partial) iovecs from src_iov to the end of dst. |
| 318 | * It starts copying after skipping `soffset' bytes at the |
| 319 | * beginning of src and adds individual vectors from src to |
| 320 | * dst copies up to `sbytes' bytes total, or up to the end |
| 321 | * of src_iov if it comes first. This way, it is okay to specify |
| 322 | * very large value for `sbytes' to indicate "up to the end |
| 323 | * of src". |
| 324 | * Only vector pointers are processed, not the actual data buffers. |
| 325 | */ |
| 326 | size_t qemu_iovec_concat_iov(QEMUIOVector *dst, |
| 327 | struct iovec *src_iov, unsigned int src_cnt, |
| 328 | size_t soffset, size_t sbytes) |
| 329 | { |
| 330 | int i; |
| 331 | size_t done; |
| 332 | |
| 333 | if (!sbytes) { |
| 334 | return 0; |
| 335 | } |
| 336 | assert(dst->nalloc != -1); |
| 337 | for (i = 0, done = 0; done < sbytes && i < src_cnt; i++) { |
| 338 | if (soffset < src_iov[i].iov_len) { |
| 339 | size_t len = MIN(src_iov[i].iov_len - soffset, sbytes - done); |
| 340 | qemu_iovec_add(dst, src_iov[i].iov_base + soffset, len); |
| 341 | done += len; |
| 342 | soffset = 0; |
| 343 | } else { |
| 344 | soffset -= src_iov[i].iov_len; |
| 345 | } |
| 346 | } |
| 347 | |
| 348 | return done; |
| 349 | } |
| 350 | |
| 351 | /* |
| 352 | * Concatenates (partial) iovecs from src to the end of dst. |
| 353 | * It starts copying after skipping `soffset' bytes at the |
| 354 | * beginning of src and adds individual vectors from src to |
| 355 | * dst copies up to `sbytes' bytes total, or up to the end |
| 356 | * of src if it comes first. This way, it is okay to specify |
| 357 | * very large value for `sbytes' to indicate "up to the end |
| 358 | * of src". |
| 359 | * Only vector pointers are processed, not the actual data buffers. |
| 360 | */ |
| 361 | void qemu_iovec_concat(QEMUIOVector *dst, |
| 362 | QEMUIOVector *src, size_t soffset, size_t sbytes) |
| 363 | { |
| 364 | qemu_iovec_concat_iov(dst, src->iov, src->niov, soffset, sbytes); |
| 365 | } |
| 366 | |
| 367 | /* |
| 368 | * qiov_find_iov |
| 369 | * |
| 370 | * Return pointer to iovec structure, where byte at @offset in original vector |
| 371 | * @iov exactly is. |
| 372 | * Set @remaining_offset to be offset inside that iovec to the same byte. |
| 373 | */ |
| 374 | static struct iovec *iov_skip_offset(struct iovec *iov, size_t offset, |
| 375 | size_t *remaining_offset) |
| 376 | { |
| 377 | while (offset > 0 && offset >= iov->iov_len) { |
| 378 | offset -= iov->iov_len; |
| 379 | iov++; |
| 380 | } |
| 381 | *remaining_offset = offset; |
| 382 | |
| 383 | return iov; |
| 384 | } |
| 385 | |
| 386 | /* |
| 387 | * qemu_iovec_slice |
| 388 | * |
| 389 | * Find subarray of iovec's, containing requested range. @head would |
| 390 | * be offset in first iov (returned by the function), @tail would be |
| 391 | * count of extra bytes in last iovec (returned iov + @niov - 1). |
| 392 | */ |
| 393 | struct iovec *qemu_iovec_slice(QEMUIOVector *qiov, |
| 394 | size_t offset, size_t len, |
| 395 | size_t *head, size_t *tail, int *niov) |
| 396 | { |
| 397 | struct iovec *iov, *end_iov; |
| 398 | |
| 399 | assert(offset + len <= qiov->size); |
| 400 | |
| 401 | iov = iov_skip_offset(qiov->iov, offset, head); |
| 402 | end_iov = iov_skip_offset(iov, *head + len, tail); |
| 403 | |
| 404 | if (*tail > 0) { |
| 405 | assert(*tail < end_iov->iov_len); |
| 406 | *tail = end_iov->iov_len - *tail; |
| 407 | end_iov++; |
| 408 | } |
| 409 | |
| 410 | *niov = end_iov - iov; |
| 411 | |
| 412 | return iov; |
| 413 | } |
| 414 | |
| 415 | int qemu_iovec_subvec_niov(QEMUIOVector *qiov, size_t offset, size_t len) |
| 416 | { |
| 417 | size_t head, tail; |
| 418 | int niov; |
| 419 | |
| 420 | qemu_iovec_slice(qiov, offset, len, &head, &tail, &niov); |
| 421 | |
| 422 | return niov; |
| 423 | } |
| 424 | |
| 425 | /* |
| 426 | * Check if the contents of subrange of qiov data is all zeroes. |
| 427 | */ |
| 428 | bool qemu_iovec_is_zero(QEMUIOVector *qiov, size_t offset, size_t bytes) |
| 429 | { |
| 430 | struct iovec *iov; |
| 431 | size_t current_offset; |
| 432 | |
| 433 | assert(offset + bytes <= qiov->size); |
| 434 | |
| 435 | iov = iov_skip_offset(qiov->iov, offset, ¤t_offset); |
| 436 | |
| 437 | while (bytes) { |
| 438 | uint8_t *base = (uint8_t *)iov->iov_base + current_offset; |
| 439 | size_t len = MIN(iov->iov_len - current_offset, bytes); |
| 440 | |
| 441 | if (!buffer_is_zero(base, len)) { |
| 442 | return false; |
| 443 | } |
| 444 | |
| 445 | current_offset = 0; |
| 446 | bytes -= len; |
| 447 | iov++; |
| 448 | } |
| 449 | |
| 450 | return true; |
| 451 | } |
| 452 | |
| 453 | void qemu_iovec_init_slice(QEMUIOVector *qiov, QEMUIOVector *source, |
| 454 | size_t offset, size_t len) |
| 455 | { |
| 456 | struct iovec *slice_iov; |
| 457 | int slice_niov; |
| 458 | size_t slice_head, slice_tail; |
| 459 | |
| 460 | assert(source->size >= len); |
| 461 | assert(source->size - len >= offset); |
| 462 | |
| 463 | slice_iov = qemu_iovec_slice(source, offset, len, |
| 464 | &slice_head, &slice_tail, &slice_niov); |
| 465 | if (slice_niov == 1) { |
| 466 | qemu_iovec_init_buf(qiov, slice_iov[0].iov_base + slice_head, len); |
| 467 | } else { |
| 468 | qemu_iovec_init(qiov, slice_niov); |
| 469 | qemu_iovec_concat_iov(qiov, slice_iov, slice_niov, slice_head, len); |
| 470 | } |
| 471 | } |
| 472 | |
| 473 | void qemu_iovec_destroy(QEMUIOVector *qiov) |
| 474 | { |
| 475 | if (qiov->nalloc != -1) { |
| 476 | g_free(qiov->iov); |
| 477 | } |
| 478 | |
| 479 | memset(qiov, 0, sizeof(*qiov)); |
| 480 | } |
| 481 | |
| 482 | void qemu_iovec_reset(QEMUIOVector *qiov) |
| 483 | { |
| 484 | assert(qiov->nalloc != -1); |
| 485 | |
| 486 | qiov->niov = 0; |
| 487 | qiov->size = 0; |
| 488 | } |
| 489 | |
| 490 | size_t qemu_iovec_to_buf(QEMUIOVector *qiov, size_t offset, |
| 491 | void *buf, size_t bytes) |
| 492 | { |
| 493 | return iov_to_buf(qiov->iov, qiov->niov, offset, buf, bytes); |
| 494 | } |
| 495 | |
| 496 | size_t qemu_iovec_from_buf(QEMUIOVector *qiov, size_t offset, |
| 497 | const void *buf, size_t bytes) |
| 498 | { |
| 499 | return iov_from_buf(qiov->iov, qiov->niov, offset, buf, bytes); |
| 500 | } |
| 501 | |
| 502 | size_t qemu_iovec_memset(QEMUIOVector *qiov, size_t offset, |
| 503 | int fillc, size_t bytes) |
| 504 | { |
| 505 | return iov_memset(qiov->iov, qiov->niov, offset, fillc, bytes); |
| 506 | } |
| 507 | |
| 508 | /** |
| 509 | * Check that I/O vector contents are identical |
| 510 | * |
| 511 | * The IO vectors must have the same structure (same length of all parts). |
| 512 | * A typical usage is to compare vectors created with qemu_iovec_clone(). |
| 513 | * |
| 514 | * @a: I/O vector |
| 515 | * @b: I/O vector |
| 516 | * @ret: Offset to first mismatching byte or -1 if match |
| 517 | */ |
| 518 | ssize_t qemu_iovec_compare(QEMUIOVector *a, QEMUIOVector *b) |
| 519 | { |
| 520 | int i; |
| 521 | ssize_t offset = 0; |
| 522 | |
| 523 | assert(a->niov == b->niov); |
| 524 | for (i = 0; i < a->niov; i++) { |
| 525 | size_t len = 0; |
| 526 | uint8_t *p = (uint8_t *)a->iov[i].iov_base; |
| 527 | uint8_t *q = (uint8_t *)b->iov[i].iov_base; |
| 528 | |
| 529 | assert(a->iov[i].iov_len == b->iov[i].iov_len); |
| 530 | while (len < a->iov[i].iov_len && *p++ == *q++) { |
| 531 | len++; |
| 532 | } |
| 533 | |
| 534 | offset += len; |
| 535 | |
| 536 | if (len != a->iov[i].iov_len) { |
| 537 | return offset; |
| 538 | } |
| 539 | } |
| 540 | return -1; |
| 541 | } |
| 542 | |
| 543 | typedef struct { |
| 544 | int src_index; |
| 545 | struct iovec *src_iov; |
| 546 | void *dest_base; |
| 547 | } IOVectorSortElem; |
| 548 | |
| 549 | static int sortelem_cmp_src_base(const void *a, const void *b) |
| 550 | { |
| 551 | const IOVectorSortElem *elem_a = a; |
| 552 | const IOVectorSortElem *elem_b = b; |
| 553 | |
| 554 | /* Don't overflow */ |
| 555 | if (elem_a->src_iov->iov_base < elem_b->src_iov->iov_base) { |
| 556 | return -1; |
| 557 | } else if (elem_a->src_iov->iov_base > elem_b->src_iov->iov_base) { |
| 558 | return 1; |
| 559 | } else { |
| 560 | return 0; |
| 561 | } |
| 562 | } |
| 563 | |
| 564 | static int sortelem_cmp_src_index(const void *a, const void *b) |
| 565 | { |
| 566 | const IOVectorSortElem *elem_a = a; |
| 567 | const IOVectorSortElem *elem_b = b; |
| 568 | |
| 569 | return elem_a->src_index - elem_b->src_index; |
| 570 | } |
| 571 | |
| 572 | /** |
| 573 | * Copy contents of I/O vector |
| 574 | * |
| 575 | * The relative relationships of overlapping iovecs are preserved. This is |
| 576 | * necessary to ensure identical semantics in the cloned I/O vector. |
| 577 | */ |
| 578 | void qemu_iovec_clone(QEMUIOVector *dest, const QEMUIOVector *src, void *buf) |
| 579 | { |
| 580 | g_autofree IOVectorSortElem *sortelems = g_new(IOVectorSortElem, src->niov); |
| 581 | void *last_end; |
| 582 | int i; |
| 583 | |
| 584 | /* Sort by source iovecs by base address */ |
| 585 | for (i = 0; i < src->niov; i++) { |
| 586 | sortelems[i].src_index = i; |
| 587 | sortelems[i].src_iov = &src->iov[i]; |
| 588 | } |
| 589 | qsort(sortelems, src->niov, sizeof(sortelems[0]), sortelem_cmp_src_base); |
| 590 | |
| 591 | /* Allocate buffer space taking into account overlapping iovecs */ |
| 592 | last_end = NULL; |
| 593 | for (i = 0; i < src->niov; i++) { |
| 594 | struct iovec *cur = sortelems[i].src_iov; |
| 595 | ptrdiff_t rewind = 0; |
| 596 | |
| 597 | /* Detect overlap */ |
| 598 | if (last_end && last_end > cur->iov_base) { |
| 599 | rewind = last_end - cur->iov_base; |
| 600 | } |
| 601 | |
| 602 | sortelems[i].dest_base = buf - rewind; |
| 603 | buf += cur->iov_len - MIN(rewind, cur->iov_len); |
| 604 | last_end = MAX(cur->iov_base + cur->iov_len, last_end); |
| 605 | } |
| 606 | |
| 607 | /* Sort by source iovec index and build destination iovec */ |
| 608 | qsort(sortelems, src->niov, sizeof(sortelems[0]), sortelem_cmp_src_index); |
| 609 | for (i = 0; i < src->niov; i++) { |
| 610 | qemu_iovec_add(dest, sortelems[i].dest_base, src->iov[i].iov_len); |
| 611 | } |
| 612 | } |
| 613 | |
| 614 | void iov_discard_undo(IOVDiscardUndo *undo) |
| 615 | { |
| 616 | /* Restore original iovec if it was modified */ |
| 617 | if (undo->modified_iov) { |
| 618 | *undo->modified_iov = undo->orig; |
| 619 | } |
| 620 | } |
| 621 | |
| 622 | size_t iov_discard_front_undoable(struct iovec **iov, |
| 623 | unsigned int *iov_cnt, |
| 624 | size_t bytes, |
| 625 | IOVDiscardUndo *undo) |
| 626 | { |
| 627 | size_t total = 0; |
| 628 | struct iovec *cur; |
| 629 | |
| 630 | if (undo) { |
| 631 | undo->modified_iov = NULL; |
| 632 | } |
| 633 | |
| 634 | for (cur = *iov; *iov_cnt > 0; cur++) { |
| 635 | if (cur->iov_len > bytes) { |
| 636 | if (undo) { |
| 637 | undo->modified_iov = cur; |
| 638 | undo->orig = *cur; |
| 639 | } |
| 640 | |
| 641 | cur->iov_base += bytes; |
| 642 | cur->iov_len -= bytes; |
| 643 | total += bytes; |
| 644 | break; |
| 645 | } |
| 646 | |
| 647 | bytes -= cur->iov_len; |
| 648 | total += cur->iov_len; |
| 649 | *iov_cnt -= 1; |
| 650 | } |
| 651 | |
| 652 | *iov = cur; |
| 653 | return total; |
| 654 | } |
| 655 | |
| 656 | size_t iov_discard_front(struct iovec **iov, unsigned int *iov_cnt, |
| 657 | size_t bytes) |
| 658 | { |
| 659 | return iov_discard_front_undoable(iov, iov_cnt, bytes, NULL); |
| 660 | } |
| 661 | |
| 662 | size_t iov_discard_back_undoable(struct iovec *iov, |
| 663 | unsigned int *iov_cnt, |
| 664 | size_t bytes, |
| 665 | IOVDiscardUndo *undo) |
| 666 | { |
| 667 | size_t total = 0; |
| 668 | struct iovec *cur; |
| 669 | |
| 670 | if (undo) { |
| 671 | undo->modified_iov = NULL; |
| 672 | } |
| 673 | |
| 674 | if (*iov_cnt == 0) { |
| 675 | return 0; |
| 676 | } |
| 677 | |
| 678 | cur = iov + (*iov_cnt - 1); |
| 679 | |
| 680 | while (*iov_cnt > 0) { |
| 681 | if (cur->iov_len > bytes) { |
| 682 | if (undo) { |
| 683 | undo->modified_iov = cur; |
| 684 | undo->orig = *cur; |
| 685 | } |
| 686 | |
| 687 | cur->iov_len -= bytes; |
| 688 | total += bytes; |
| 689 | break; |
| 690 | } |
| 691 | |
| 692 | bytes -= cur->iov_len; |
| 693 | total += cur->iov_len; |
| 694 | cur--; |
| 695 | *iov_cnt -= 1; |
| 696 | } |
| 697 | |
| 698 | return total; |
| 699 | } |
| 700 | |
| 701 | size_t iov_discard_back(struct iovec *iov, unsigned int *iov_cnt, |
| 702 | size_t bytes) |
| 703 | { |
| 704 | return iov_discard_back_undoable(iov, iov_cnt, bytes, NULL); |
| 705 | } |
| 706 | |
| 707 | void qemu_iovec_discard_back(QEMUIOVector *qiov, size_t bytes) |
| 708 | { |
| 709 | size_t total; |
| 710 | unsigned int niov = qiov->niov; |
| 711 | |
| 712 | assert(qiov->size >= bytes); |
| 713 | total = iov_discard_back(qiov->iov, &niov, bytes); |
| 714 | assert(total == bytes); |
| 715 | |
| 716 | qiov->niov = niov; |
| 717 | qiov->size -= bytes; |
| 718 | } |