Archipelago » qemu

/*

 * Command line utility to exercise the QEMU I/O path.

 *

 * Copyright (C) 2009 Red Hat, Inc.

 * Copyright (c) 2003-2005 Silicon Graphics, Inc.

 *

 * This work is licensed under the terms of the GNU GPL, version 2 or later.

 * See the COPYING file in the top-level directory.

 */

#include "qemu-io.h"

#include "block/block_int.h"

#include "block/qapi.h"

#include "qemu/main-loop.h"

#define CMD_NOFILE_OK   0x01

int qemuio_misalign;

static cmdinfo_t *cmdtab;

static int ncmds;

static int compare_cmdname(const void *a, const void *b)

{

    return strcmp(((const cmdinfo_t *)a)->name,

                  ((const cmdinfo_t *)b)->name);

}

void qemuio_add_command(const cmdinfo_t *ci)

{

    cmdtab = g_realloc(cmdtab, ++ncmds * sizeof(*cmdtab));

    cmdtab[ncmds - 1] = *ci;

    qsort(cmdtab, ncmds, sizeof(*cmdtab), compare_cmdname);

}

int qemuio_command_usage(const cmdinfo_t *ci)

{

    printf("%s %s -- %s\n", ci->name, ci->args, ci->oneline);

    return 0;

}

static int init_check_command(BlockDriverState *bs, const cmdinfo_t *ct)

{

    if (ct->flags & CMD_FLAG_GLOBAL) {

        return 1;

    }

    if (!(ct->flags & CMD_NOFILE_OK) && !bs) {

        fprintf(stderr, "no file open, try 'help open'\n");

        return 0;

    }

    return 1;

}

static int command(BlockDriverState *bs, const cmdinfo_t *ct, int argc,

                   char **argv)

{

    char *cmd = argv[0];

    if (!init_check_command(bs, ct)) {

        return 0;

    }

    if (argc - 1 < ct->argmin || (ct->argmax != -1 && argc - 1 > ct->argmax)) {

        if (ct->argmax == -1) {

            fprintf(stderr,

                    "bad argument count %d to %s, expected at least %d arguments\n",

                    argc-1, cmd, ct->argmin);

        } else if (ct->argmin == ct->argmax) {

            fprintf(stderr,

                    "bad argument count %d to %s, expected %d arguments\n",

                    argc-1, cmd, ct->argmin);

        } else {

            fprintf(stderr,

                    "bad argument count %d to %s, expected between %d and %d arguments\n",

                    argc-1, cmd, ct->argmin, ct->argmax);

        }

        return 0;

    }

    optind = 0;

    return ct->cfunc(bs, argc, argv);

}

static const cmdinfo_t *find_command(const char *cmd)

{

    cmdinfo_t *ct;

    for (ct = cmdtab; ct < &cmdtab[ncmds]; ct++) {

        if (strcmp(ct->name, cmd) == 0 ||

            (ct->altname && strcmp(ct->altname, cmd) == 0))

        {

            return (const cmdinfo_t *)ct;

        }

    }

    return NULL;

}

/* Invoke fn() for commands with a matching prefix */

void qemuio_complete_command(const char *input,

                             void (*fn)(const char *cmd, void *opaque),

                             void *opaque)

{

    cmdinfo_t *ct;

    size_t input_len = strlen(input);

    for (ct = cmdtab; ct < &cmdtab[ncmds]; ct++) {

        if (strncmp(input, ct->name, input_len) == 0) {

            fn(ct->name, opaque);

        }

    }

}

static char **breakline(char *input, int *count)

{

    int c = 0;

    char *p;

    char **rval = g_malloc0(sizeof(char *));

    char **tmp;

    while (rval && (p = qemu_strsep(&input, " ")) != NULL) {

        if (!*p) {

            continue;

        }

        c++;

        tmp = g_realloc(rval, sizeof(*rval) * (c + 1));

        if (!tmp) {

            g_free(rval);

            rval = NULL;

            c = 0;

            break;

        } else {

            rval = tmp;

        }

        rval[c - 1] = p;

        rval[c] = NULL;

    }

    *count = c;

    return rval;

}

static int64_t cvtnum(const char *s)

{

    char *end;

    return strtosz_suffix(s, &end, STRTOSZ_DEFSUFFIX_B);

}

#define EXABYTES(x)     ((long long)(x) << 60)

#define PETABYTES(x)    ((long long)(x) << 50)

#define TERABYTES(x)    ((long long)(x) << 40)

#define GIGABYTES(x)    ((long long)(x) << 30)

#define MEGABYTES(x)    ((long long)(x) << 20)

#define KILOBYTES(x)    ((long long)(x) << 10)

#define TO_EXABYTES(x)  ((x) / EXABYTES(1))

#define TO_PETABYTES(x) ((x) / PETABYTES(1))

#define TO_TERABYTES(x) ((x) / TERABYTES(1))

#define TO_GIGABYTES(x) ((x) / GIGABYTES(1))

#define TO_MEGABYTES(x) ((x) / MEGABYTES(1))

#define TO_KILOBYTES(x) ((x) / KILOBYTES(1))

static void cvtstr(double value, char *str, size_t size)

{

    char *trim;

    const char *suffix;

    if (value >= EXABYTES(1)) {

        suffix = " EiB";

        snprintf(str, size - 4, "%.3f", TO_EXABYTES(value));

    } else if (value >= PETABYTES(1)) {

        suffix = " PiB";

        snprintf(str, size - 4, "%.3f", TO_PETABYTES(value));

    } else if (value >= TERABYTES(1)) {

        suffix = " TiB";

        snprintf(str, size - 4, "%.3f", TO_TERABYTES(value));

    } else if (value >= GIGABYTES(1)) {

        suffix = " GiB";

        snprintf(str, size - 4, "%.3f", TO_GIGABYTES(value));

    } else if (value >= MEGABYTES(1)) {

        suffix = " MiB";

        snprintf(str, size - 4, "%.3f", TO_MEGABYTES(value));

    } else if (value >= KILOBYTES(1)) {

        suffix = " KiB";

        snprintf(str, size - 4, "%.3f", TO_KILOBYTES(value));

    } else {

        suffix = " bytes";

        snprintf(str, size - 6, "%f", value);

    }

    trim = strstr(str, ".000");

    if (trim) {

        strcpy(trim, suffix);

    } else {

        strcat(str, suffix);

    }

}

static struct timeval tsub(struct timeval t1, struct timeval t2)

{

    t1.tv_usec -= t2.tv_usec;

    if (t1.tv_usec < 0) {

        t1.tv_usec += 1000000;

        t1.tv_sec--;

    }

    t1.tv_sec -= t2.tv_sec;

    return t1;

}

static double tdiv(double value, struct timeval tv)

{

    return value / ((double)tv.tv_sec + ((double)tv.tv_usec / 1000000.0));

}

#define HOURS(sec)      ((sec) / (60 * 60))

#define MINUTES(sec)    (((sec) % (60 * 60)) / 60)

#define SECONDS(sec)    ((sec) % 60)

enum {

    DEFAULT_TIME        = 0x0,

    TERSE_FIXED_TIME    = 0x1,

    VERBOSE_FIXED_TIME  = 0x2,

};

static void timestr(struct timeval *tv, char *ts, size_t size, int format)

{

    double usec = (double)tv->tv_usec / 1000000.0;

    if (format & TERSE_FIXED_TIME) {

        if (!HOURS(tv->tv_sec)) {

            snprintf(ts, size, "%u:%02u.%02u",

                    (unsigned int) MINUTES(tv->tv_sec),

                    (unsigned int) SECONDS(tv->tv_sec),

                    (unsigned int) (usec * 100));

            return;

        }

        format |= VERBOSE_FIXED_TIME; /* fallback if hours needed */

    }

    if ((format & VERBOSE_FIXED_TIME) || tv->tv_sec) {

        snprintf(ts, size, "%u:%02u:%02u.%02u",

                (unsigned int) HOURS(tv->tv_sec),

                (unsigned int) MINUTES(tv->tv_sec),

                (unsigned int) SECONDS(tv->tv_sec),

                (unsigned int) (usec * 100));

    } else {

        snprintf(ts, size, "0.%04u sec", (unsigned int) (usec * 10000));

    }

}

/*

 * Parse the pattern argument to various sub-commands.

 *

 * Because the pattern is used as an argument to memset it must evaluate

 * to an unsigned integer that fits into a single byte.

 */

static int parse_pattern(const char *arg)

{

    char *endptr = NULL;

    long pattern;

    pattern = strtol(arg, &endptr, 0);

    if (pattern < 0 || pattern > UCHAR_MAX || *endptr != '\0') {

        printf("%s is not a valid pattern byte\n", arg);

        return -1;

    }

    return pattern;

}

/*

 * Memory allocation helpers.

 *

 * Make sure memory is aligned by default, or purposefully misaligned if

 * that is specified on the command line.

 */

#define MISALIGN_OFFSET     16

static void *qemu_io_alloc(BlockDriverState *bs, size_t len, int pattern)

{

    void *buf;

    if (qemuio_misalign) {

        len += MISALIGN_OFFSET;

    }

    buf = qemu_blockalign(bs, len);

    memset(buf, pattern, len);

    if (qemuio_misalign) {

        buf += MISALIGN_OFFSET;

    }

    return buf;

}

static void qemu_io_free(void *p)

{

    if (qemuio_misalign) {

        p -= MISALIGN_OFFSET;

    }

    qemu_vfree(p);

}

static void dump_buffer(const void *buffer, int64_t offset, int len)

{

    int i, j;

    const uint8_t *p;

    for (i = 0, p = buffer; i < len; i += 16) {

        const uint8_t *s = p;

        printf("%08" PRIx64 ":  ", offset + i);

        for (j = 0; j < 16 && i + j < len; j++, p++) {

            printf("%02x ", *p);

        }

        printf(" ");

        for (j = 0; j < 16 && i + j < len; j++, s++) {

            if (isalnum(*s)) {

                printf("%c", *s);

            } else {

                printf(".");

            }

        }

        printf("\n");

    }

}

static void print_report(const char *op, struct timeval *t, int64_t offset,

                         int count, int total, int cnt, int Cflag)

{

    char s1[64], s2[64], ts[64];

    timestr(t, ts, sizeof(ts), Cflag ? VERBOSE_FIXED_TIME : 0);

    if (!Cflag) {

        cvtstr((double)total, s1, sizeof(s1));

        cvtstr(tdiv((double)total, *t), s2, sizeof(s2));

        printf("%s %d/%d bytes at offset %" PRId64 "\n",

               op, total, count, offset);

        printf("%s, %d ops; %s (%s/sec and %.4f ops/sec)\n",

               s1, cnt, ts, s2, tdiv((double)cnt, *t));

    } else {/* bytes,ops,time,bytes/sec,ops/sec */

        printf("%d,%d,%s,%.3f,%.3f\n",

            total, cnt, ts,

            tdiv((double)total, *t),

            tdiv((double)cnt, *t));

    }

}

/*

 * Parse multiple length statements for vectored I/O, and construct an I/O

 * vector matching it.

 */

static void *

create_iovec(BlockDriverState *bs, QEMUIOVector *qiov, char **argv, int nr_iov,

             int pattern)

{

    size_t *sizes = g_new0(size_t, nr_iov);

    size_t count = 0;

    void *buf = NULL;

    void *p;

    int i;

    for (i = 0; i < nr_iov; i++) {

        char *arg = argv[i];

        int64_t len;

        len = cvtnum(arg);

        if (len < 0) {

            printf("non-numeric length argument -- %s\n", arg);

            goto fail;

        }

        /* should be SIZE_T_MAX, but that doesn't exist */

        if (len > INT_MAX) {

            printf("too large length argument -- %s\n", arg);

            goto fail;

        }

        if (len & 0x1ff) {

            printf("length argument %" PRId64

                   " is not sector aligned\n", len);

            goto fail;

        }

        sizes[i] = len;

        count += len;

    }

    qemu_iovec_init(qiov, nr_iov);

    buf = p = qemu_io_alloc(bs, count, pattern);

    for (i = 0; i < nr_iov; i++) {

        qemu_iovec_add(qiov, p, sizes[i]);

        p += sizes[i];

    }

fail:

    g_free(sizes);

    return buf;

}

static int do_read(BlockDriverState *bs, char *buf, int64_t offset, int count,

                   int *total)

{

    int ret;

    ret = bdrv_read(bs, offset >> 9, (uint8_t *)buf, count >> 9);

    if (ret < 0) {

        return ret;

    }

    *total = count;

    return 1;

}

static int do_write(BlockDriverState *bs, char *buf, int64_t offset, int count,

                    int *total)

{

    int ret;

    ret = bdrv_write(bs, offset >> 9, (uint8_t *)buf, count >> 9);

    if (ret < 0) {

        return ret;

    }

    *total = count;

    return 1;

}

static int do_pread(BlockDriverState *bs, char *buf, int64_t offset, int count,

                    int *total)

{

    *total = bdrv_pread(bs, offset, (uint8_t *)buf, count);

    if (*total < 0) {

        return *total;

    }

    return 1;

}

static int do_pwrite(BlockDriverState *bs, char *buf, int64_t offset, int count,

                     int *total)

{

    *total = bdrv_pwrite(bs, offset, (uint8_t *)buf, count);

    if (*total < 0) {

        return *total;

    }

    return 1;

}

typedef struct {

    BlockDriverState *bs;

    int64_t offset;

    int count;

    int *total;

    int ret;

    bool done;

} CoWriteZeroes;

static void coroutine_fn co_write_zeroes_entry(void *opaque)

{

    CoWriteZeroes *data = opaque;

    data->ret = bdrv_co_write_zeroes(data->bs, data->offset / BDRV_SECTOR_SIZE,

                                     data->count / BDRV_SECTOR_SIZE, 0);

    data->done = true;

    if (data->ret < 0) {

        *data->total = data->ret;

        return;

    }

    *data->total = data->count;

}

static int do_co_write_zeroes(BlockDriverState *bs, int64_t offset, int count,

                              int *total)

{

    Coroutine *co;

    CoWriteZeroes data = {

        .bs     = bs,

        .offset = offset,

        .count  = count,

        .total  = total,

        .done   = false,

    };

    co = qemu_coroutine_create(co_write_zeroes_entry);

    qemu_coroutine_enter(co, &data);

    while (!data.done) {

        qemu_aio_wait();

    }

    if (data.ret < 0) {

        return data.ret;

    } else {

        return 1;

    }

}

static int do_write_compressed(BlockDriverState *bs, char *buf, int64_t offset,

                               int count, int *total)

{

    int ret;

    ret = bdrv_write_compressed(bs, offset >> 9, (uint8_t *)buf, count >> 9);

    if (ret < 0) {

        return ret;

    }

    *total = count;

    return 1;

}

static int do_load_vmstate(BlockDriverState *bs, char *buf, int64_t offset,

                           int count, int *total)

{

    *total = bdrv_load_vmstate(bs, (uint8_t *)buf, offset, count);

    if (*total < 0) {

        return *total;

    }

    return 1;

}

static int do_save_vmstate(BlockDriverState *bs, char *buf, int64_t offset,

                           int count, int *total)

{

    *total = bdrv_save_vmstate(bs, (uint8_t *)buf, offset, count);

    if (*total < 0) {

        return *total;

    }

    return 1;

}

#define NOT_DONE 0x7fffffff

static void aio_rw_done(void *opaque, int ret)

{

    *(int *)opaque = ret;

}

static int do_aio_readv(BlockDriverState *bs, QEMUIOVector *qiov,

                        int64_t offset, int *total)

{

    int async_ret = NOT_DONE;

    bdrv_aio_readv(bs, offset >> 9, qiov, qiov->size >> 9,

                   aio_rw_done, &async_ret);

    while (async_ret == NOT_DONE) {

        main_loop_wait(false);

    }

    *total = qiov->size;

    return async_ret < 0 ? async_ret : 1;

}

static int do_aio_writev(BlockDriverState *bs, QEMUIOVector *qiov,

                         int64_t offset, int *total)

{

    int async_ret = NOT_DONE;

    bdrv_aio_writev(bs, offset >> 9, qiov, qiov->size >> 9,

                    aio_rw_done, &async_ret);

    while (async_ret == NOT_DONE) {

        main_loop_wait(false);

    }

    *total = qiov->size;

    return async_ret < 0 ? async_ret : 1;

}

struct multiwrite_async_ret {

    int num_done;

    int error;

};

static void multiwrite_cb(void *opaque, int ret)

{

    struct multiwrite_async_ret *async_ret = opaque;

    async_ret->num_done++;

    if (ret < 0) {

        async_ret->error = ret;

    }

}

static int do_aio_multiwrite(BlockDriverState *bs, BlockRequest* reqs,

                             int num_reqs, int *total)

{

    int i, ret;

    struct multiwrite_async_ret async_ret = {

        .num_done = 0,

        .error = 0,

    };

    *total = 0;

    for (i = 0; i < num_reqs; i++) {

        reqs[i].cb = multiwrite_cb;

        reqs[i].opaque = &async_ret;

        *total += reqs[i].qiov->size;

    }

    ret = bdrv_aio_multiwrite(bs, reqs, num_reqs);

    if (ret < 0) {

        return ret;

    }

    while (async_ret.num_done < num_reqs) {

        main_loop_wait(false);

    }

    return async_ret.error < 0 ? async_ret.error : 1;

}

static void read_help(void)

{

    printf(

"\n"

" reads a range of bytes from the given offset\n"

"\n"

" Example:\n"

" 'read -v 512 1k' - dumps 1 kilobyte read from 512 bytes into the file\n"

"\n"

" Reads a segment of the currently open file, optionally dumping it to the\n"

" standard output stream (with -v option) for subsequent inspection.\n"

" -b, -- read from the VM state rather than the virtual disk\n"

" -C, -- report statistics in a machine parsable format\n"

" -l, -- length for pattern verification (only with -P)\n"

" -p, -- use bdrv_pread to read the file\n"

" -P, -- use a pattern to verify read data\n"

" -q, -- quiet mode, do not show I/O statistics\n"

" -s, -- start offset for pattern verification (only with -P)\n"

" -v, -- dump buffer to standard output\n"

"\n");

}

static int read_f(BlockDriverState *bs, int argc, char **argv);

static const cmdinfo_t read_cmd = {

    .name       = "read",

    .altname    = "r",

    .cfunc      = read_f,

    .argmin     = 2,

    .argmax     = -1,

    .args       = "[-abCpqv] [-P pattern [-s off] [-l len]] off len",

    .oneline    = "reads a number of bytes at a specified offset",

    .help       = read_help,

};

static int read_f(BlockDriverState *bs, int argc, char **argv)

{

    struct timeval t1, t2;

    int Cflag = 0, pflag = 0, qflag = 0, vflag = 0;

    int Pflag = 0, sflag = 0, lflag = 0, bflag = 0;

    int c, cnt;

    char *buf;

    int64_t offset;

    int count;

    /* Some compilers get confused and warn if this is not initialized.  */

    int total = 0;

    int pattern = 0, pattern_offset = 0, pattern_count = 0;

    while ((c = getopt(argc, argv, "bCl:pP:qs:v")) != EOF) {

        switch (c) {

        case 'b':

            bflag = 1;

            break;

        case 'C':

            Cflag = 1;

            break;

        case 'l':

            lflag = 1;

            pattern_count = cvtnum(optarg);

            if (pattern_count < 0) {

                printf("non-numeric length argument -- %s\n", optarg);

                return 0;

            }

            break;

        case 'p':

            pflag = 1;

            break;

        case 'P':

            Pflag = 1;

            pattern = parse_pattern(optarg);

            if (pattern < 0) {

                return 0;

            }

            break;

        case 'q':

            qflag = 1;

            break;

        case 's':

            sflag = 1;

            pattern_offset = cvtnum(optarg);

            if (pattern_offset < 0) {

                printf("non-numeric length argument -- %s\n", optarg);

                return 0;

            }

            break;

        case 'v':

            vflag = 1;

            break;

        default:

            return qemuio_command_usage(&read_cmd);

        }

    }

    if (optind != argc - 2) {

        return qemuio_command_usage(&read_cmd);

    }

    if (bflag && pflag) {

        printf("-b and -p cannot be specified at the same time\n");

        return 0;

    }

    offset = cvtnum(argv[optind]);

    if (offset < 0) {

        printf("non-numeric length argument -- %s\n", argv[optind]);

        return 0;

    }

    optind++;

    count = cvtnum(argv[optind]);

    if (count < 0) {

        printf("non-numeric length argument -- %s\n", argv[optind]);

        return 0;

    }

    if (!Pflag && (lflag || sflag)) {

        return qemuio_command_usage(&read_cmd);

    }

    if (!lflag) {

        pattern_count = count - pattern_offset;

    }

    if ((pattern_count < 0) || (pattern_count + pattern_offset > count))  {

        printf("pattern verification range exceeds end of read data\n");

        return 0;

    }

    if (!pflag) {

        if (offset & 0x1ff) {

            printf("offset %" PRId64 " is not sector aligned\n",

                   offset);

            return 0;

        }

        if (count & 0x1ff) {

            printf("count %d is not sector aligned\n",

                   count);

            return 0;

        }

    }

    buf = qemu_io_alloc(bs, count, 0xab);

    gettimeofday(&t1, NULL);

    if (pflag) {

        cnt = do_pread(bs, buf, offset, count, &total);

    } else if (bflag) {

        cnt = do_load_vmstate(bs, buf, offset, count, &total);

    } else {

        cnt = do_read(bs, buf, offset, count, &total);

    }

    gettimeofday(&t2, NULL);

    if (cnt < 0) {

        printf("read failed: %s\n", strerror(-cnt));

        goto out;

    }

    if (Pflag) {

        void *cmp_buf = g_malloc(pattern_count);

        memset(cmp_buf, pattern, pattern_count);

        if (memcmp(buf + pattern_offset, cmp_buf, pattern_count)) {

            printf("Pattern verification failed at offset %"

                   PRId64 ", %d bytes\n",

                   offset + pattern_offset, pattern_count);

        }

        g_free(cmp_buf);

    }

    if (qflag) {

        goto out;

    }

    if (vflag) {

        dump_buffer(buf, offset, count);

    }

    /* Finally, report back -- -C gives a parsable format */

    t2 = tsub(t2, t1);

    print_report("read", &t2, offset, count, total, cnt, Cflag);

out:

    qemu_io_free(buf);

    return 0;

}

static void readv_help(void)

{

    printf(

"\n"

" reads a range of bytes from the given offset into multiple buffers\n"

"\n"

" Example:\n"

" 'readv -v 512 1k 1k ' - dumps 2 kilobytes read from 512 bytes into the file\n"

"\n"

" Reads a segment of the currently open file, optionally dumping it to the\n"

" standard output stream (with -v option) for subsequent inspection.\n"

" Uses multiple iovec buffers if more than one byte range is specified.\n"

" -C, -- report statistics in a machine parsable format\n"

" -P, -- use a pattern to verify read data\n"

" -v, -- dump buffer to standard output\n"

" -q, -- quiet mode, do not show I/O statistics\n"

"\n");

}

static int readv_f(BlockDriverState *bs, int argc, char **argv);

static const cmdinfo_t readv_cmd = {

    .name       = "readv",

    .cfunc      = readv_f,

    .argmin     = 2,

    .argmax     = -1,

    .args       = "[-Cqv] [-P pattern ] off len [len..]",

    .oneline    = "reads a number of bytes at a specified offset",

    .help       = readv_help,

};

static int readv_f(BlockDriverState *bs, int argc, char **argv)

{

    struct timeval t1, t2;

    int Cflag = 0, qflag = 0, vflag = 0;

    int c, cnt;

    char *buf;

    int64_t offset;

    /* Some compilers get confused and warn if this is not initialized.  */

    int total = 0;

    int nr_iov;

    QEMUIOVector qiov;

    int pattern = 0;

    int Pflag = 0;

    while ((c = getopt(argc, argv, "CP:qv")) != EOF) {

        switch (c) {

        case 'C':

            Cflag = 1;

            break;

        case 'P':

            Pflag = 1;

            pattern = parse_pattern(optarg);

            if (pattern < 0) {

                return 0;

            }

            break;