qemu/util/iov.c
<<
>>
Prefs
   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
  22size_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
  41size_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
  60size_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
  79size_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() */
  92static ssize_t
  93do_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
 136ssize_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
 212void 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
 229unsigned 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
 254void 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
 262void 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
 274void 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 */
 298void 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 */
 332void 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
 338void qemu_iovec_destroy(QEMUIOVector *qiov)
 339{
 340    assert(qiov->nalloc != -1);
 341
 342    qemu_iovec_reset(qiov);
 343    g_free(qiov->iov);
 344    qiov->nalloc = 0;
 345    qiov->iov = NULL;
 346}
 347
 348void qemu_iovec_reset(QEMUIOVector *qiov)
 349{
 350    assert(qiov->nalloc != -1);
 351
 352    qiov->niov = 0;
 353    qiov->size = 0;
 354}
 355
 356size_t qemu_iovec_to_buf(QEMUIOVector *qiov, size_t offset,
 357                         void *buf, size_t bytes)
 358{
 359    return iov_to_buf(qiov->iov, qiov->niov, offset, buf, bytes);
 360}
 361
 362size_t qemu_iovec_from_buf(QEMUIOVector *qiov, size_t offset,
 363                           const void *buf, size_t bytes)
 364{
 365    return iov_from_buf(qiov->iov, qiov->niov, offset, buf, bytes);
 366}
 367
 368size_t qemu_iovec_memset(QEMUIOVector *qiov, size_t offset,
 369                         int fillc, size_t bytes)
 370{
 371    return iov_memset(qiov->iov, qiov->niov, offset, fillc, bytes);
 372}
 373
 374/**
 375 * Check that I/O vector contents are identical
 376 *
 377 * The IO vectors must have the same structure (same length of all parts).
 378 * A typical usage is to compare vectors created with qemu_iovec_clone().
 379 *
 380 * @a:          I/O vector
 381 * @b:          I/O vector
 382 * @ret:        Offset to first mismatching byte or -1 if match
 383 */
 384ssize_t qemu_iovec_compare(QEMUIOVector *a, QEMUIOVector *b)
 385{
 386    int i;
 387    ssize_t offset = 0;
 388
 389    assert(a->niov == b->niov);
 390    for (i = 0; i < a->niov; i++) {
 391        size_t len = 0;
 392        uint8_t *p = (uint8_t *)a->iov[i].iov_base;
 393        uint8_t *q = (uint8_t *)b->iov[i].iov_base;
 394
 395        assert(a->iov[i].iov_len == b->iov[i].iov_len);
 396        while (len < a->iov[i].iov_len && *p++ == *q++) {
 397            len++;
 398        }
 399
 400        offset += len;
 401
 402        if (len != a->iov[i].iov_len) {
 403            return offset;
 404        }
 405    }
 406    return -1;
 407}
 408
 409typedef struct {
 410    int src_index;
 411    struct iovec *src_iov;
 412    void *dest_base;
 413} IOVectorSortElem;
 414
 415static int sortelem_cmp_src_base(const void *a, const void *b)
 416{
 417    const IOVectorSortElem *elem_a = a;
 418    const IOVectorSortElem *elem_b = b;
 419
 420    /* Don't overflow */
 421    if (elem_a->src_iov->iov_base < elem_b->src_iov->iov_base) {
 422        return -1;
 423    } else if (elem_a->src_iov->iov_base > elem_b->src_iov->iov_base) {
 424        return 1;
 425    } else {
 426        return 0;
 427    }
 428}
 429
 430static int sortelem_cmp_src_index(const void *a, const void *b)
 431{
 432    const IOVectorSortElem *elem_a = a;
 433    const IOVectorSortElem *elem_b = b;
 434
 435    return elem_a->src_index - elem_b->src_index;
 436}
 437
 438/**
 439 * Copy contents of I/O vector
 440 *
 441 * The relative relationships of overlapping iovecs are preserved.  This is
 442 * necessary to ensure identical semantics in the cloned I/O vector.
 443 */
 444void qemu_iovec_clone(QEMUIOVector *dest, const QEMUIOVector *src, void *buf)
 445{
 446    IOVectorSortElem sortelems[src->niov];
 447    void *last_end;
 448    int i;
 449
 450    /* Sort by source iovecs by base address */
 451    for (i = 0; i < src->niov; i++) {
 452        sortelems[i].src_index = i;
 453        sortelems[i].src_iov = &src->iov[i];
 454    }
 455    qsort(sortelems, src->niov, sizeof(sortelems[0]), sortelem_cmp_src_base);
 456
 457    /* Allocate buffer space taking into account overlapping iovecs */
 458    last_end = NULL;
 459    for (i = 0; i < src->niov; i++) {
 460        struct iovec *cur = sortelems[i].src_iov;
 461        ptrdiff_t rewind = 0;
 462
 463        /* Detect overlap */
 464        if (last_end && last_end > cur->iov_base) {
 465            rewind = last_end - cur->iov_base;
 466        }
 467
 468        sortelems[i].dest_base = buf - rewind;
 469        buf += cur->iov_len - MIN(rewind, cur->iov_len);
 470        last_end = MAX(cur->iov_base + cur->iov_len, last_end);
 471    }
 472
 473    /* Sort by source iovec index and build destination iovec */
 474    qsort(sortelems, src->niov, sizeof(sortelems[0]), sortelem_cmp_src_index);
 475    for (i = 0; i < src->niov; i++) {
 476        qemu_iovec_add(dest, sortelems[i].dest_base, src->iov[i].iov_len);
 477    }
 478}
 479
 480size_t iov_discard_front(struct iovec **iov, unsigned int *iov_cnt,
 481                         size_t bytes)
 482{
 483    size_t total = 0;
 484    struct iovec *cur;
 485
 486    for (cur = *iov; *iov_cnt > 0; cur++) {
 487        if (cur->iov_len > bytes) {
 488            cur->iov_base += bytes;
 489            cur->iov_len -= bytes;
 490            total += bytes;
 491            break;
 492        }
 493
 494        bytes -= cur->iov_len;
 495        total += cur->iov_len;
 496        *iov_cnt -= 1;
 497    }
 498
 499    *iov = cur;
 500    return total;
 501}
 502
 503size_t iov_discard_back(struct iovec *iov, unsigned int *iov_cnt,
 504                        size_t bytes)
 505{
 506    size_t total = 0;
 507    struct iovec *cur;
 508
 509    if (*iov_cnt == 0) {
 510        return 0;
 511    }
 512
 513    cur = iov + (*iov_cnt - 1);
 514
 515    while (*iov_cnt > 0) {
 516        if (cur->iov_len > bytes) {
 517            cur->iov_len -= bytes;
 518            total += bytes;
 519            break;
 520        }
 521
 522        bytes -= cur->iov_len;
 523        total += cur->iov_len;
 524        cur--;
 525        *iov_cnt -= 1;
 526    }
 527
 528    return total;
 529}
 530