debian/patches/pve/0026-PVE-Backup-add-vma-backup-format-code.patch

   1 From 0000000000000000000000000000000000000000 Mon Sep 17 00:00:00 2001
   2 From: Dietmar Maurer <dietmar@proxmox.com>
   3 Date: Mon, 6 Apr 2020 12:16:57 +0200
   4 Subject: [PATCH] PVE-Backup: add vma backup format code
   5
   6 ---
   7  block/meson.build |   2 +
   8  meson.build       |   5 +
   9  vma-reader.c      | 857 ++++++++++++++++++++++++++++++++++++++++++++++
  10  vma-writer.c      | 790 ++++++++++++++++++++++++++++++++++++++++++
  11  vma.c             | 839 +++++++++++++++++++++++++++++++++++++++++++++
  12  vma.h             | 150 ++++++++
  13  6 files changed, 2643 insertions(+)
  14  create mode 100644 vma-reader.c
  15  create mode 100644 vma-writer.c
  16  create mode 100644 vma.c
  17  create mode 100644 vma.h
  18
  19 diff --git a/block/meson.build b/block/meson.build
  20 index cf9b278fa2..d0a8397edf 100644
  21 --- a/block/meson.build
  22 +++ b/block/meson.build
  23 @@ -43,6 +43,8 @@ block_ss.add(files(
  24    'zeroinit.c',
  25  ), zstd, zlib, gnutls)
  26
  27 +block_ss.add(files('../vma-writer.c'), libuuid)
  28 +
  29  softmmu_ss.add(when: 'CONFIG_TCG', if_true: files('blkreplay.c'))
  30
  31  block_ss.add(when: 'CONFIG_QCOW1', if_true: files('qcow.c'))
  32 diff --git a/meson.build b/meson.build
  33 index c6f4b0cf5e..7db08406bd 100644
  34 --- a/meson.build
  35 +++ b/meson.build
  36 @@ -945,6 +945,8 @@ keyutils = dependency('libkeyutils', required: false,
  37
  38  has_gettid = cc.has_function('gettid')
  39
  40 +libuuid = cc.find_library('uuid', required: true)
  41 +
  42  # Malloc tests
  43
  44  malloc = []
  45 @@ -2344,6 +2346,9 @@ if have_tools
  46    qemu_nbd = executable('qemu-nbd', files('qemu-nbd.c'),
  47                 dependencies: [blockdev, qemuutil, gnutls], install: true)
  48
  49 +  vma = executable('vma', files('vma.c', 'vma-reader.c'),
  50 +                   dependencies: [authz, block, crypto, io, qom], install: true)
  51 +
  52    subdir('storage-daemon')
  53    subdir('contrib/rdmacm-mux')
  54    subdir('contrib/elf2dmp')
  55 diff --git a/vma-reader.c b/vma-reader.c
  56 new file mode 100644
  57 index 0000000000..2b1d1cdab3
  58 --- /dev/null
  59 +++ b/vma-reader.c
  60 @@ -0,0 +1,857 @@
  61 +/*
  62 + * VMA: Virtual Machine Archive
  63 + *
  64 + * Copyright (C) 2012 Proxmox Server Solutions
  65 + *
  66 + * Authors:
  67 + *  Dietmar Maurer (dietmar@proxmox.com)
  68 + *
  69 + * This work is licensed under the terms of the GNU GPL, version 2 or later.
  70 + * See the COPYING file in the top-level directory.
  71 + *
  72 + */
  73 +
  74 +#include "qemu/osdep.h"
  75 +#include <glib.h>
  76 +#include <uuid/uuid.h>
  77 +
  78 +#include "qemu-common.h"
  79 +#include "qemu/timer.h"
  80 +#include "qemu/ratelimit.h"
  81 +#include "vma.h"
  82 +#include "block/block.h"
  83 +#include "sysemu/block-backend.h"
  84 +
  85 +static unsigned char zero_vma_block[VMA_BLOCK_SIZE];
  86 +
  87 +typedef struct VmaRestoreState {
  88 +    BlockBackend *target;
  89 +    bool write_zeroes;
  90 +    unsigned long *bitmap;
  91 +    int bitmap_size;
  92 +}  VmaRestoreState;
  93 +
  94 +struct VmaReader {
  95 +    int fd;
  96 +    GChecksum *md5csum;
  97 +    GHashTable *blob_hash;
  98 +    unsigned char *head_data;
  99 +    VmaDeviceInfo devinfo[256];
 100 +    VmaRestoreState rstate[256];
 101 +    GList *cdata_list;
 102 +    guint8 vmstate_stream;
 103 +    uint32_t vmstate_clusters;
 104 +    /* to show restore percentage if run with -v */
 105 +    time_t start_time;
 106 +    int64_t cluster_count;
 107 +    int64_t clusters_read;
 108 +    int64_t zero_cluster_data;
 109 +    int64_t partial_zero_cluster_data;
 110 +    int clusters_read_per;
 111 +};
 112 +
 113 +static guint
 114 +g_int32_hash(gconstpointer v)
 115 +{
 116 +    return *(const uint32_t *)v;
 117 +}
 118 +
 119 +static gboolean
 120 +g_int32_equal(gconstpointer v1, gconstpointer v2)
 121 +{
 122 +    return *((const uint32_t *)v1) == *((const uint32_t *)v2);
 123 +}
 124 +
 125 +static int vma_reader_get_bitmap(VmaRestoreState *rstate, int64_t cluster_num)
 126 +{
 127 +    assert(rstate);
 128 +    assert(rstate->bitmap);
 129 +
 130 +    unsigned long val, idx, bit;
 131 +
 132 +    idx = cluster_num / BITS_PER_LONG;
 133 +
 134 +    assert(rstate->bitmap_size > idx);
 135 +
 136 +    bit = cluster_num % BITS_PER_LONG;
 137 +    val = rstate->bitmap[idx];
 138 +
 139 +    return !!(val & (1UL << bit));
 140 +}
 141 +
 142 +static void vma_reader_set_bitmap(VmaRestoreState *rstate, int64_t cluster_num,
 143 +                                  int dirty)
 144 +{
 145 +    assert(rstate);
 146 +    assert(rstate->bitmap);
 147 +
 148 +    unsigned long val, idx, bit;
 149 +
 150 +    idx = cluster_num / BITS_PER_LONG;
 151 +
 152 +    assert(rstate->bitmap_size > idx);
 153 +
 154 +    bit = cluster_num % BITS_PER_LONG;
 155 +    val = rstate->bitmap[idx];
 156 +    if (dirty) {
 157 +        if (!(val & (1UL << bit))) {
 158 +            val |= 1UL << bit;
 159 +        }
 160 +    } else {
 161 +        if (val & (1UL << bit)) {
 162 +            val &= ~(1UL << bit);
 163 +        }
 164 +    }
 165 +    rstate->bitmap[idx] = val;
 166 +}
 167 +
 168 +typedef struct VmaBlob {
 169 +    uint32_t start;
 170 +    uint32_t len;
 171 +    void *data;
 172 +} VmaBlob;
 173 +
 174 +static const VmaBlob *get_header_blob(VmaReader *vmar, uint32_t pos)
 175 +{
 176 +    assert(vmar);
 177 +    assert(vmar->blob_hash);
 178 +
 179 +    return g_hash_table_lookup(vmar->blob_hash, &pos);
 180 +}
 181 +
 182 +static const char *get_header_str(VmaReader *vmar, uint32_t pos)
 183 +{
 184 +    const VmaBlob *blob = get_header_blob(vmar, pos);
 185 +    if (!blob) {
 186 +        return NULL;
 187 +    }
 188 +    const char *res = (char *)blob->data;
 189 +    if (res[blob->len-1] != '\0') {
 190 +        return NULL;
 191 +    }
 192 +    return res;
 193 +}
 194 +
 195 +static ssize_t
 196 +safe_read(int fd, unsigned char *buf, size_t count)
 197 +{
 198 +    ssize_t n;
 199 +
 200 +    do {
 201 +        n = read(fd, buf, count);
 202 +    } while (n < 0 && errno == EINTR);
 203 +
 204 +    return n;
 205 +}
 206 +
 207 +static ssize_t
 208 +full_read(int fd, unsigned char *buf, size_t len)
 209 +{
 210 +    ssize_t n;
 211 +    size_t total;
 212 +
 213 +    total = 0;
 214 +
 215 +    while (len > 0) {
 216 +        n = safe_read(fd, buf, len);
 217 +
 218 +        if (n == 0) {
 219 +            return total;
 220 +        }
 221 +
 222 +        if (n <= 0) {
 223 +            break;
 224 +        }
 225 +
 226 +        buf += n;
 227 +        total += n;
 228 +        len -= n;
 229 +    }
 230 +
 231 +    if (len) {
 232 +        return -1;
 233 +    }
 234 +
 235 +    return total;
 236 +}
 237 +
 238 +void vma_reader_destroy(VmaReader *vmar)
 239 +{
 240 +    assert(vmar);
 241 +
 242 +    if (vmar->fd >= 0) {
 243 +        close(vmar->fd);
 244 +    }
 245 +
 246 +    if (vmar->cdata_list) {
 247 +        g_list_free(vmar->cdata_list);
 248 +    }
 249 +
 250 +    int i;
 251 +    for (i = 1; i < 256; i++) {
 252 +        if (vmar->rstate[i].bitmap) {
 253 +            g_free(vmar->rstate[i].bitmap);
 254 +        }
 255 +    }
 256 +
 257 +    if (vmar->md5csum) {
 258 +        g_checksum_free(vmar->md5csum);
 259 +    }
 260 +
 261 +    if (vmar->blob_hash) {
 262 +        g_hash_table_destroy(vmar->blob_hash);
 263 +    }
 264 +
 265 +    if (vmar->head_data) {
 266 +        g_free(vmar->head_data);
 267 +    }
 268 +
 269 +    g_free(vmar);
 270 +
 271 +};
 272 +
 273 +static int vma_reader_read_head(VmaReader *vmar, Error **errp)
 274 +{
 275 +    assert(vmar);
 276 +    assert(errp);
 277 +    assert(*errp == NULL);
 278 +
 279 +    unsigned char md5sum[16];
 280 +    int i;
 281 +    int ret = 0;
 282 +
 283 +    vmar->head_data = g_malloc(sizeof(VmaHeader));
 284 +
 285 +    if (full_read(vmar->fd, vmar->head_data, sizeof(VmaHeader)) !=
 286 +        sizeof(VmaHeader)) {
 287 +        error_setg(errp, "can't read vma header - %s",
 288 +                   errno ? g_strerror(errno) : "got EOF");
 289 +        return -1;
 290 +    }
 291 +
 292 +    VmaHeader *h = (VmaHeader *)vmar->head_data;
 293 +
 294 +    if (h->magic != VMA_MAGIC) {
 295 +        error_setg(errp, "not a vma file - wrong magic number");
 296 +        return -1;
 297 +    }
 298 +
 299 +    uint32_t header_size = GUINT32_FROM_BE(h->header_size);
 300 +    int need = header_size - sizeof(VmaHeader);
 301 +    if (need <= 0) {
 302 +        error_setg(errp, "wrong vma header size %d", header_size);
 303 +        return -1;
 304 +    }
 305 +
 306 +    vmar->head_data = g_realloc(vmar->head_data, header_size);
 307 +    h = (VmaHeader *)vmar->head_data;
 308 +
 309 +    if (full_read(vmar->fd, vmar->head_data + sizeof(VmaHeader), need) !=
 310 +        need) {
 311 +        error_setg(errp, "can't read vma header data - %s",
 312 +                   errno ? g_strerror(errno) : "got EOF");
 313 +        return -1;
 314 +    }
 315 +
 316 +    memcpy(md5sum, h->md5sum, 16);
 317 +    memset(h->md5sum, 0, 16);
 318 +
 319 +    g_checksum_reset(vmar->md5csum);
 320 +    g_checksum_update(vmar->md5csum, vmar->head_data, header_size);
 321 +    gsize csize = 16;
 322 +    g_checksum_get_digest(vmar->md5csum, (guint8 *)(h->md5sum), &csize);
 323 +
 324 +    if (memcmp(md5sum, h->md5sum, 16) != 0) {
 325 +        error_setg(errp, "wrong vma header chechsum");
 326 +        return -1;
 327 +    }
 328 +
 329 +    /* we can modify header data after checksum verify */
 330 +    h->header_size = header_size;
 331 +
 332 +    h->version = GUINT32_FROM_BE(h->version);
 333 +    if (h->version != 1) {
 334 +        error_setg(errp, "wrong vma version %d", h->version);
 335 +        return -1;
 336 +    }
 337 +
 338 +    h->ctime = GUINT64_FROM_BE(h->ctime);
 339 +    h->blob_buffer_offset = GUINT32_FROM_BE(h->blob_buffer_offset);
 340 +    h->blob_buffer_size = GUINT32_FROM_BE(h->blob_buffer_size);
 341 +
 342 +    uint32_t bstart = h->blob_buffer_offset + 1;
 343 +    uint32_t bend = h->blob_buffer_offset + h->blob_buffer_size;
 344 +
 345 +    if (bstart <= sizeof(VmaHeader)) {
 346 +        error_setg(errp, "wrong vma blob buffer offset %d",
 347 +                   h->blob_buffer_offset);
 348 +        return -1;
 349 +    }
 350 +
 351 +    if (bend > header_size) {
 352 +        error_setg(errp, "wrong vma blob buffer size %d/%d",
 353 +                   h->blob_buffer_offset, h->blob_buffer_size);
 354 +        return -1;
 355 +    }
 356 +
 357 +    while ((bstart + 2) <= bend) {
 358 +        uint32_t size = vmar->head_data[bstart] +
 359 +            (vmar->head_data[bstart+1] << 8);
 360 +        if ((bstart + size + 2) <= bend) {
 361 +            VmaBlob *blob = g_new0(VmaBlob, 1);
 362 +            blob->start = bstart - h->blob_buffer_offset;
 363 +            blob->len = size;
 364 +            blob->data = vmar->head_data + bstart + 2;
 365 +            g_hash_table_insert(vmar->blob_hash, &blob->start, blob);
 366 +        }
 367 +        bstart += size + 2;
 368 +    }
 369 +
 370 +
 371 +    int count = 0;
 372 +    for (i = 1; i < 256; i++) {
 373 +        VmaDeviceInfoHeader *dih = &h->dev_info[i];
 374 +        uint32_t devname_ptr = GUINT32_FROM_BE(dih->devname_ptr);
 375 +        uint64_t size = GUINT64_FROM_BE(dih->size);
 376 +        const char *devname =  get_header_str(vmar, devname_ptr);
 377 +
 378 +        if (size && devname) {
 379 +            count++;
 380 +            vmar->devinfo[i].size = size;
 381 +            vmar->devinfo[i].devname = devname;
 382 +
 383 +            if (strcmp(devname, "vmstate") == 0) {
 384 +                vmar->vmstate_stream = i;
 385 +            }
 386 +        }
 387 +    }
 388 +
 389 +    for (i = 0; i < VMA_MAX_CONFIGS; i++) {
 390 +        uint32_t name_ptr = GUINT32_FROM_BE(h->config_names[i]);
 391 +        uint32_t data_ptr = GUINT32_FROM_BE(h->config_data[i]);
 392 +
 393 +        if (!(name_ptr && data_ptr)) {
 394 +            continue;
 395 +        }
 396 +        const char *name =  get_header_str(vmar, name_ptr);
 397 +        const VmaBlob *blob = get_header_blob(vmar, data_ptr);
 398 +
 399 +        if (!(name && blob)) {
 400 +            error_setg(errp, "vma contains invalid data pointers");
 401 +            return -1;
 402 +        }
 403 +
 404 +        VmaConfigData *cdata = g_new0(VmaConfigData, 1);
 405 +        cdata->name = name;
 406 +        cdata->data = blob->data;
 407 +        cdata->len = blob->len;
 408 +
 409 +        vmar->cdata_list = g_list_append(vmar->cdata_list, cdata);
 410 +    }
 411 +
 412 +    return ret;
 413 +};
 414 +
 415 +VmaReader *vma_reader_create(const char *filename, Error **errp)
 416 +{
 417 +    assert(filename);
 418 +    assert(errp);
 419 +
 420 +    VmaReader *vmar = g_new0(VmaReader, 1);
 421 +
 422 +    if (strcmp(filename, "-") == 0) {
 423 +        vmar->fd = dup(0);
 424 +    } else {
 425 +        vmar->fd = open(filename, O_RDONLY);
 426 +    }
 427 +
 428 +    if (vmar->fd < 0) {
 429 +        error_setg(errp, "can't open file %s - %s\n", filename,
 430 +                   g_strerror(errno));
 431 +        goto err;
 432 +    }
 433 +
 434 +    vmar->md5csum = g_checksum_new(G_CHECKSUM_MD5);
 435 +    if (!vmar->md5csum) {
 436 +        error_setg(errp, "can't allocate cmsum\n");
 437 +        goto err;
 438 +    }
 439 +
 440 +    vmar->blob_hash = g_hash_table_new_full(g_int32_hash, g_int32_equal,
 441 +                                            NULL, g_free);
 442 +
 443 +    if (vma_reader_read_head(vmar, errp) < 0) {
 444 +        goto err;
 445 +    }
 446 +
 447 +    return vmar;
 448 +
 449 +err:
 450 +    if (vmar) {
 451 +        vma_reader_destroy(vmar);
 452 +    }
 453 +
 454 +    return NULL;
 455 +}
 456 +
 457 +VmaHeader *vma_reader_get_header(VmaReader *vmar)
 458 +{
 459 +    assert(vmar);
 460 +    assert(vmar->head_data);
 461 +
 462 +    return (VmaHeader *)(vmar->head_data);
 463 +}
 464 +
 465 +GList *vma_reader_get_config_data(VmaReader *vmar)
 466 +{
 467 +    assert(vmar);
 468 +    assert(vmar->head_data);
 469 +
 470 +    return vmar->cdata_list;
 471 +}
 472 +
 473 +VmaDeviceInfo *vma_reader_get_device_info(VmaReader *vmar, guint8 dev_id)
 474 +{
 475 +    assert(vmar);
 476 +    assert(dev_id);
 477 +
 478 +    if (vmar->devinfo[dev_id].size && vmar->devinfo[dev_id].devname) {
 479 +        return &vmar->devinfo[dev_id];
 480 +    }
 481 +
 482 +    return NULL;
 483 +}
 484 +
 485 +static void allocate_rstate(VmaReader *vmar,  guint8 dev_id,
 486 +                            BlockBackend *target, bool write_zeroes)
 487 +{
 488 +    assert(vmar);
 489 +    assert(dev_id);
 490 +
 491 +    vmar->rstate[dev_id].target = target;
 492 +    vmar->rstate[dev_id].write_zeroes = write_zeroes;
 493 +
 494 +    int64_t size = vmar->devinfo[dev_id].size;
 495 +
 496 +    int64_t bitmap_size = (size/BDRV_SECTOR_SIZE) +
 497 +        (VMA_CLUSTER_SIZE/BDRV_SECTOR_SIZE) * BITS_PER_LONG - 1;
 498 +    bitmap_size /= (VMA_CLUSTER_SIZE/BDRV_SECTOR_SIZE) * BITS_PER_LONG;
 499 +
 500 +    vmar->rstate[dev_id].bitmap_size = bitmap_size;
 501 +    vmar->rstate[dev_id].bitmap = g_new0(unsigned long, bitmap_size);
 502 +
 503 +    vmar->cluster_count += size/VMA_CLUSTER_SIZE;
 504 +}
 505 +
 506 +int vma_reader_register_bs(VmaReader *vmar, guint8 dev_id, BlockBackend *target,
 507 +                           bool write_zeroes, Error **errp)
 508 +{
 509 +    assert(vmar);
 510 +    assert(target != NULL);
 511 +    assert(dev_id);
 512 +    assert(vmar->rstate[dev_id].target == NULL);
 513 +
 514 +    int64_t size = blk_getlength(target);
 515 +    int64_t size_diff = size - vmar->devinfo[dev_id].size;
 516 +
 517 +    /* storage types can have different size restrictions, so it
 518 +     * is not always possible to create an image with exact size.
 519 +     * So we tolerate a size difference up to 4MB.
 520 +     */
 521 +    if ((size_diff < 0) || (size_diff > 4*1024*1024)) {
 522 +        error_setg(errp, "vma_reader_register_bs for stream %s failed - "
 523 +                   "unexpected size %zd != %zd", vmar->devinfo[dev_id].devname,
 524 +                   size, vmar->devinfo[dev_id].size);
 525 +        return -1;
 526 +    }
 527 +
 528 +    allocate_rstate(vmar, dev_id, target, write_zeroes);
 529 +
 530 +    return 0;
 531 +}
 532 +
 533 +static ssize_t safe_write(int fd, void *buf, size_t count)
 534 +{
 535 +    ssize_t n;
 536 +
 537 +    do {
 538 +        n = write(fd, buf, count);
 539 +    } while (n < 0 && errno == EINTR);
 540 +
 541 +    return n;
 542 +}
 543 +
 544 +static size_t full_write(int fd, void *buf, size_t len)
 545 +{
 546 +    ssize_t n;
 547 +    size_t total;
 548 +
 549 +    total = 0;
 550 +
 551 +    while (len > 0) {
 552 +        n = safe_write(fd, buf, len);
 553 +        if (n < 0) {
 554 +            return n;
 555 +        }
 556 +        buf += n;
 557 +        total += n;
 558 +        len -= n;
 559 +    }
 560 +
 561 +    if (len) {
 562 +        /* incomplete write ? */
 563 +        return -1;
 564 +    }
 565 +
 566 +    return total;
 567 +}
 568 +
 569 +static int restore_write_data(VmaReader *vmar, guint8 dev_id,
 570 +                              BlockBackend *target, int vmstate_fd,
 571 +                              unsigned char *buf, int64_t sector_num,
 572 +                              int nb_sectors, Error **errp)
 573 +{
 574 +    assert(vmar);
 575 +
 576 +    if (dev_id == vmar->vmstate_stream) {
 577 +        if (vmstate_fd >= 0) {
 578 +            int len = nb_sectors * BDRV_SECTOR_SIZE;
 579 +            int res = full_write(vmstate_fd, buf, len);
 580 +            if (res < 0) {
 581 +                error_setg(errp, "write vmstate failed %d", res);
 582 +                return -1;
 583 +            }
 584 +        }
 585 +    } else {
 586 +        int res = blk_pwrite(target, sector_num * BDRV_SECTOR_SIZE, buf, nb_sectors * BDRV_SECTOR_SIZE, 0);
 587 +        if (res < 0) {
 588 +            error_setg(errp, "blk_pwrite to %s failed (%d)",
 589 +                       bdrv_get_device_name(blk_bs(target)), res);
 590 +            return -1;
 591 +        }
 592 +    }
 593 +    return 0;
 594 +}
 595 +
 596 +static int restore_extent(VmaReader *vmar, unsigned char *buf,
 597 +                          int extent_size, int vmstate_fd,
 598 +                          bool verbose, bool verify, Error **errp)
 599 +{
 600 +    assert(vmar);
 601 +    assert(buf);
 602 +
 603 +    VmaExtentHeader *ehead = (VmaExtentHeader *)buf;
 604 +    int start = VMA_EXTENT_HEADER_SIZE;
 605 +    int i;
 606 +
 607 +    for (i = 0; i < VMA_BLOCKS_PER_EXTENT; i++) {
 608 +        uint64_t block_info = GUINT64_FROM_BE(ehead->blockinfo[i]);
 609 +        uint64_t cluster_num = block_info & 0xffffffff;
 610 +        uint8_t dev_id = (block_info >> 32) & 0xff;
 611 +        uint16_t mask = block_info >> (32+16);
 612 +        int64_t max_sector;
 613 +
 614 +        if (!dev_id) {
 615 +            continue;
 616 +        }
 617 +
 618 +        VmaRestoreState *rstate = &vmar->rstate[dev_id];
 619 +        BlockBackend *target = NULL;
 620 +
 621 +        if (dev_id != vmar->vmstate_stream) {
 622 +            target = rstate->target;
 623 +            if (!verify && !target) {
 624 +                error_setg(errp, "got wrong dev id %d", dev_id);
 625 +                return -1;
 626 +            }
 627 +
 628 +            if (vma_reader_get_bitmap(rstate, cluster_num)) {
 629 +                error_setg(errp, "found duplicated cluster %zd for stream %s",
 630 +                          cluster_num, vmar->devinfo[dev_id].devname);
 631 +                return -1;
 632 +            }
 633 +            vma_reader_set_bitmap(rstate, cluster_num, 1);
 634 +
 635 +            max_sector = vmar->devinfo[dev_id].size/BDRV_SECTOR_SIZE;
 636 +        } else {
 637 +            max_sector = G_MAXINT64;
 638 +            if (cluster_num != vmar->vmstate_clusters) {
 639 +                error_setg(errp, "found out of order vmstate data");
 640 +                return -1;
 641 +            }
 642 +            vmar->vmstate_clusters++;
 643 +        }
 644 +
 645 +        vmar->clusters_read++;
 646 +
 647 +        if (verbose) {
 648 +            time_t duration = time(NULL) - vmar->start_time;
 649 +            int percent = (vmar->clusters_read*100)/vmar->cluster_count;
 650 +            if (percent != vmar->clusters_read_per) {
 651 +                printf("progress %d%% (read %zd bytes, duration %zd sec)\n",
 652 +                       percent, vmar->clusters_read*VMA_CLUSTER_SIZE,
 653 +                       duration);
 654 +                fflush(stdout);
 655 +                vmar->clusters_read_per = percent;
 656 +            }
 657 +        }
 658 +
 659 +        /* try to write whole clusters to speedup restore */
 660 +        if (mask == 0xffff) {
 661 +            if ((start + VMA_CLUSTER_SIZE) > extent_size) {
 662 +                error_setg(errp, "short vma extent - too many blocks");
 663 +                return -1;
 664 +            }
 665 +            int64_t sector_num = (cluster_num * VMA_CLUSTER_SIZE) /
 666 +                BDRV_SECTOR_SIZE;
 667 +            int64_t end_sector = sector_num +
 668 +                VMA_CLUSTER_SIZE/BDRV_SECTOR_SIZE;
 669 +
 670 +            if (end_sector > max_sector) {
 671 +                end_sector = max_sector;
 672 +            }
 673 +
 674 +            if (end_sector <= sector_num) {
 675 +                error_setg(errp, "got wrong block address - write beyond end");
 676 +                return -1;
 677 +            }
 678 +
 679 +            if (!verify) {
 680 +                int nb_sectors = end_sector - sector_num;
 681 +                if (restore_write_data(vmar, dev_id, target, vmstate_fd,
 682 +                                       buf + start, sector_num, nb_sectors,
 683 +                                       errp) < 0) {
 684 +                    return -1;
 685 +                }
 686 +            }
 687 +
 688 +            start += VMA_CLUSTER_SIZE;
 689 +        } else {
 690 +            int j;
 691 +            int bit = 1;
 692 +
 693 +            for (j = 0; j < 16; j++) {
 694 +                int64_t sector_num = (cluster_num*VMA_CLUSTER_SIZE +
 695 +                                      j*VMA_BLOCK_SIZE)/BDRV_SECTOR_SIZE;
 696 +
 697 +                int64_t end_sector = sector_num +
 698 +                    VMA_BLOCK_SIZE/BDRV_SECTOR_SIZE;
 699 +                if (end_sector > max_sector) {
 700 +                    end_sector = max_sector;
 701 +                }
 702 +
 703 +                if (mask & bit) {
 704 +                    if ((start + VMA_BLOCK_SIZE) > extent_size) {
 705 +                        error_setg(errp, "short vma extent - too many blocks");
 706 +                        return -1;
 707 +                    }
 708 +
 709 +                    if (end_sector <= sector_num) {
 710 +                        error_setg(errp, "got wrong block address - "
 711 +                                   "write beyond end");
 712 +                        return -1;
 713 +                    }
 714 +
 715 +                    if (!verify) {
 716 +                        int nb_sectors = end_sector - sector_num;
 717 +                        if (restore_write_data(vmar, dev_id, target, vmstate_fd,
 718 +                                               buf + start, sector_num,
 719 +                                               nb_sectors, errp) < 0) {
 720 +                            return -1;
 721 +                        }
 722 +                    }
 723 +
 724 +                    start += VMA_BLOCK_SIZE;
 725 +
 726 +                } else {
 727 +
 728 +
 729 +                    if (end_sector > sector_num) {
 730 +                        /* Todo: use bdrv_co_write_zeroes (but that need to
 731 +                         * be run inside coroutine?)
 732 +                         */
 733 +                        int nb_sectors = end_sector - sector_num;
 734 +                        int zero_size = BDRV_SECTOR_SIZE*nb_sectors;
 735 +                        vmar->zero_cluster_data += zero_size;
 736 +                        if (mask != 0) {
 737 +                            vmar->partial_zero_cluster_data += zero_size;
 738 +                        }
 739 +
 740 +                        if (rstate->write_zeroes && !verify) {
 741 +                            if (restore_write_data(vmar, dev_id, target, vmstate_fd,
 742 +                                                   zero_vma_block, sector_num,
 743 +                                                   nb_sectors, errp) < 0) {
 744 +                                return -1;
 745 +                            }
 746 +                        }
 747 +                    }
 748 +                }
 749 +
 750 +                bit = bit << 1;
 751 +            }
 752 +        }
 753 +    }
 754 +
 755 +    if (start != extent_size) {
 756 +        error_setg(errp, "vma extent error - missing blocks");
 757 +        return -1;
 758 +    }
 759 +
 760 +    return 0;
 761 +}
 762 +
 763 +static int vma_reader_restore_full(VmaReader *vmar, int vmstate_fd,
 764 +                                   bool verbose, bool verify,
 765 +                                   Error **errp)
 766 +{
 767 +    assert(vmar);
 768 +    assert(vmar->head_data);
 769 +
 770 +    int ret = 0;
 771 +    unsigned char buf[VMA_MAX_EXTENT_SIZE];
 772 +    int buf_pos = 0;
 773 +    unsigned char md5sum[16];
 774 +    VmaHeader *h = (VmaHeader *)vmar->head_data;
 775 +
 776 +    vmar->start_time = time(NULL);
 777 +
 778 +    while (1) {
 779 +        int bytes = full_read(vmar->fd, buf + buf_pos, sizeof(buf) - buf_pos);
 780 +        if (bytes < 0) {
 781 +            error_setg(errp, "read failed - %s", g_strerror(errno));
 782 +            return -1;
 783 +        }
 784 +
 785 +        buf_pos += bytes;
 786 +
 787 +        if (!buf_pos) {
 788 +            break; /* EOF */
 789 +        }
 790 +
 791 +        if (buf_pos < VMA_EXTENT_HEADER_SIZE) {
 792 +            error_setg(errp, "read short extent (%d bytes)", buf_pos);
 793 +            return -1;
 794 +        }
 795 +
 796 +        VmaExtentHeader *ehead = (VmaExtentHeader *)buf;
 797 +
 798 +        /* extract md5sum */
 799 +        memcpy(md5sum, ehead->md5sum, sizeof(ehead->md5sum));
 800 +        memset(ehead->md5sum, 0, sizeof(ehead->md5sum));
 801 +
 802 +        g_checksum_reset(vmar->md5csum);
 803 +        g_checksum_update(vmar->md5csum, buf, VMA_EXTENT_HEADER_SIZE);
 804 +        gsize csize = 16;
 805 +        g_checksum_get_digest(vmar->md5csum, ehead->md5sum, &csize);
 806 +
 807 +        if (memcmp(md5sum, ehead->md5sum, 16) != 0) {
 808 +            error_setg(errp, "wrong vma extent header chechsum");
 809 +            return -1;
 810 +        }
 811 +
 812 +        if (memcmp(h->uuid, ehead->uuid, sizeof(ehead->uuid)) != 0) {
 813 +            error_setg(errp, "wrong vma extent uuid");
 814 +            return -1;
 815 +        }
 816 +
 817 +        if (ehead->magic != VMA_EXTENT_MAGIC || ehead->reserved1 != 0) {
 818 +            error_setg(errp, "wrong vma extent header magic");
 819 +            return -1;
 820 +        }
 821 +
 822 +        int block_count = GUINT16_FROM_BE(ehead->block_count);
 823 +        int extent_size = VMA_EXTENT_HEADER_SIZE + block_count*VMA_BLOCK_SIZE;
 824 +
 825 +        if (buf_pos < extent_size) {
 826 +            error_setg(errp, "short vma extent (%d < %d)", buf_pos,
 827 +                       extent_size);
 828 +            return -1;
 829 +        }
 830 +
 831 +        if (restore_extent(vmar, buf, extent_size, vmstate_fd, verbose,
 832 +                           verify, errp) < 0) {
 833 +            return -1;
 834 +        }
 835 +
 836 +        if (buf_pos > extent_size) {
 837 +            memmove(buf, buf + extent_size, buf_pos - extent_size);
 838 +            buf_pos = buf_pos - extent_size;
 839 +        } else {
 840 +            buf_pos = 0;
 841 +        }
 842 +    }
 843 +
 844 +    bdrv_drain_all();
 845 +
 846 +    int i;
 847 +    for (i = 1; i < 256; i++) {
 848 +        VmaRestoreState *rstate = &vmar->rstate[i];
 849 +        if (!rstate->target) {
 850 +            continue;
 851 +        }
 852 +
 853 +        if (blk_flush(rstate->target) < 0) {
 854 +            error_setg(errp, "vma blk_flush %s failed",
 855 +                       vmar->devinfo[i].devname);
 856 +            return -1;
 857 +        }
 858 +
 859 +        if (vmar->devinfo[i].size &&
 860 +            (strcmp(vmar->devinfo[i].devname, "vmstate") != 0)) {
 861 +            assert(rstate->bitmap);
 862 +
 863 +            int64_t cluster_num, end;
 864 +
 865 +            end = (vmar->devinfo[i].size + VMA_CLUSTER_SIZE - 1) /
 866 +                VMA_CLUSTER_SIZE;
 867 +
 868 +            for (cluster_num = 0; cluster_num < end; cluster_num++) {
 869 +                if (!vma_reader_get_bitmap(rstate, cluster_num)) {
 870 +                    error_setg(errp, "detected missing cluster %zd "
 871 +                               "for stream %s", cluster_num,
 872 +                               vmar->devinfo[i].devname);
 873 +                    return -1;
 874 +                }
 875 +            }
 876 +        }
 877 +    }
 878 +
 879 +    if (verbose) {
 880 +        if (vmar->clusters_read) {
 881 +            printf("total bytes read %zd, sparse bytes %zd (%.3g%%)\n",
 882 +                   vmar->clusters_read*VMA_CLUSTER_SIZE,
 883 +                   vmar->zero_cluster_data,
 884 +                   (double)(100.0*vmar->zero_cluster_data)/
 885 +                   (vmar->clusters_read*VMA_CLUSTER_SIZE));
 886 +
 887 +            int64_t datasize = vmar->clusters_read*VMA_CLUSTER_SIZE-vmar->zero_cluster_data;
 888 +            if (datasize) { // this does not make sense for empty files
 889 +                printf("space reduction due to 4K zero blocks %.3g%%\n",
 890 +                       (double)(100.0*vmar->partial_zero_cluster_data) / datasize);
 891 +            }
 892 +        } else {
 893 +            printf("vma archive contains no image data\n");
 894 +        }
 895 +    }
 896 +    return ret;
 897 +}
 898 +
 899 +int vma_reader_restore(VmaReader *vmar, int vmstate_fd, bool verbose,
 900 +                       Error **errp)
 901 +{
 902 +    return vma_reader_restore_full(vmar, vmstate_fd, verbose, false, errp);
 903 +}
 904 +
 905 +int vma_reader_verify(VmaReader *vmar, bool verbose, Error **errp)
 906 +{
 907 +    guint8 dev_id;
 908 +
 909 +    for (dev_id = 1; dev_id < 255; dev_id++) {
 910 +        if (vma_reader_get_device_info(vmar, dev_id)) {
 911 +            allocate_rstate(vmar, dev_id, NULL, false);
 912 +        }
 913 +    }
 914 +
 915 +    return vma_reader_restore_full(vmar, -1, verbose, true, errp);
 916 +}
 917 +
 918 diff --git a/vma-writer.c b/vma-writer.c
 919 new file mode 100644
 920 index 0000000000..11d8321ffd
 921 --- /dev/null
 922 +++ b/vma-writer.c
 923 @@ -0,0 +1,790 @@
 924 +/*
 925 + * VMA: Virtual Machine Archive
 926 + *
 927 + * Copyright (C) 2012 Proxmox Server Solutions
 928 + *
 929 + * Authors:
 930 + *  Dietmar Maurer (dietmar@proxmox.com)
 931 + *
 932 + * This work is licensed under the terms of the GNU GPL, version 2 or later.
 933 + * See the COPYING file in the top-level directory.
 934 + *
 935 + */
 936 +
 937 +#include "qemu/osdep.h"
 938 +#include <glib.h>
 939 +#include <uuid/uuid.h>
 940 +
 941 +#include "vma.h"
 942 +#include "block/block.h"
 943 +#include "monitor/monitor.h"
 944 +#include "qemu/main-loop.h"
 945 +#include "qemu/coroutine.h"
 946 +#include "qemu/cutils.h"
 947 +
 948 +#define DEBUG_VMA 0
 949 +
 950 +#define DPRINTF(fmt, ...)\
 951 +    do { if (DEBUG_VMA) { printf("vma: " fmt, ## __VA_ARGS__); } } while (0)
 952 +
 953 +#define WRITE_BUFFERS 5
 954 +#define HEADER_CLUSTERS 8
 955 +#define HEADERBUF_SIZE (VMA_CLUSTER_SIZE*HEADER_CLUSTERS)
 956 +
 957 +struct VmaWriter {
 958 +    int fd;
 959 +    FILE *cmd;
 960 +    int status;
 961 +    char errmsg[8192];
 962 +    uuid_t uuid;
 963 +    bool header_written;
 964 +    bool closed;
 965 +
 966 +    /* we always write extents */
 967 +    unsigned char *outbuf;
 968 +    int outbuf_pos; /* in bytes */
 969 +    int outbuf_count; /* in VMA_BLOCKS */
 970 +    uint64_t outbuf_block_info[VMA_BLOCKS_PER_EXTENT];
 971 +
 972 +    unsigned char *headerbuf;
 973 +
 974 +    GChecksum *md5csum;
 975 +    CoMutex flush_lock;
 976 +    Coroutine *co_writer;
 977 +
 978 +    /* drive informations */
 979 +    VmaStreamInfo stream_info[256];
 980 +    guint stream_count;
 981 +
 982 +    guint8 vmstate_stream;
 983 +    uint32_t vmstate_clusters;
 984 +
 985 +    /* header blob table */
 986 +    char *header_blob_table;
 987 +    uint32_t header_blob_table_size;
 988 +    uint32_t header_blob_table_pos;
 989 +
 990 +    /* store for config blobs */
 991 +    uint32_t config_names[VMA_MAX_CONFIGS]; /* offset into blob_buffer table */
 992 +    uint32_t config_data[VMA_MAX_CONFIGS];  /* offset into blob_buffer table */
 993 +    uint32_t config_count;
 994 +};
 995 +
 996 +void vma_writer_set_error(VmaWriter *vmaw, const char *fmt, ...)
 997 +{
 998 +    va_list ap;
 999 +
1000 +    if (vmaw->status < 0) {
1001 +        return;
1002 +    }
1003 +
1004 +    vmaw->status = -1;
1005 +
1006 +    va_start(ap, fmt);
1007 +    g_vsnprintf(vmaw->errmsg, sizeof(vmaw->errmsg), fmt, ap);
1008 +    va_end(ap);
1009 +
1010 +    DPRINTF("vma_writer_set_error: %s\n", vmaw->errmsg);
1011 +}
1012 +
1013 +static uint32_t allocate_header_blob(VmaWriter *vmaw, const char *data,
1014 +                                     size_t len)
1015 +{
1016 +    if (len > 65535) {
1017 +        return 0;
1018 +    }
1019 +
1020 +    if (!vmaw->header_blob_table ||
1021 +        (vmaw->header_blob_table_size <
1022 +         (vmaw->header_blob_table_pos + len + 2))) {
1023 +        int newsize = vmaw->header_blob_table_size + ((len + 2 + 511)/512)*512;
1024 +
1025 +        vmaw->header_blob_table = g_realloc(vmaw->header_blob_table, newsize);
1026 +        memset(vmaw->header_blob_table + vmaw->header_blob_table_size,
1027 +               0, newsize - vmaw->header_blob_table_size);
1028 +        vmaw->header_blob_table_size = newsize;
1029 +    }
1030 +
1031 +    uint32_t cpos = vmaw->header_blob_table_pos;
1032 +    vmaw->header_blob_table[cpos] = len & 255;
1033 +    vmaw->header_blob_table[cpos+1] = (len >> 8) & 255;
1034 +    memcpy(vmaw->header_blob_table + cpos + 2, data, len);
1035 +    vmaw->header_blob_table_pos += len + 2;
1036 +    return cpos;
1037 +}
1038 +
1039 +static uint32_t allocate_header_string(VmaWriter *vmaw, const char *str)
1040 +{
1041 +    assert(vmaw);
1042 +
1043 +    size_t len = strlen(str) + 1;
1044 +
1045 +    return allocate_header_blob(vmaw, str, len);
1046 +}
1047 +
1048 +int vma_writer_add_config(VmaWriter *vmaw, const char *name, gpointer data,
1049 +                          gsize len)
1050 +{
1051 +    assert(vmaw);
1052 +    assert(!vmaw->header_written);
1053 +    assert(vmaw->config_count < VMA_MAX_CONFIGS);
1054 +    assert(name);
1055 +    assert(data);
1056 +
1057 +    gchar *basename = g_path_get_basename(name);
1058 +    uint32_t name_ptr = allocate_header_string(vmaw, basename);
1059 +    g_free(basename);
1060 +
1061 +    if (!name_ptr) {
1062 +        return -1;
1063 +    }
1064 +
1065 +    uint32_t data_ptr = allocate_header_blob(vmaw, data, len);
1066 +    if (!data_ptr) {
1067 +        return -1;
1068 +    }
1069 +
1070 +    vmaw->config_names[vmaw->config_count] = name_ptr;
1071 +    vmaw->config_data[vmaw->config_count] = data_ptr;
1072 +
1073 +    vmaw->config_count++;
1074 +
1075 +    return 0;
1076 +}
1077 +
1078 +int vma_writer_register_stream(VmaWriter *vmaw, const char *devname,
1079 +                               size_t size)
1080 +{
1081 +    assert(vmaw);
1082 +    assert(devname);
1083 +    assert(!vmaw->status);
1084 +
1085 +    if (vmaw->header_written) {
1086 +        vma_writer_set_error(vmaw, "vma_writer_register_stream: header "
1087 +                             "already written");
1088 +        return -1;
1089 +    }
1090 +
1091 +    guint n = vmaw->stream_count + 1;
1092 +
1093 +    /* we can have dev_ids form 1 to 255 (0 reserved)
1094 +     * 255(-1) reseverd for safety
1095 +     */
1096 +    if (n > 254) {
1097 +        vma_writer_set_error(vmaw, "vma_writer_register_stream: "
1098 +                             "too many drives");
1099 +        return -1;
1100 +    }
1101 +
1102 +    if (size <= 0) {
1103 +        vma_writer_set_error(vmaw, "vma_writer_register_stream: "
1104 +                             "got strange size %zd", size);
1105 +        return -1;
1106 +    }
1107 +
1108 +    DPRINTF("vma_writer_register_stream %s %zu %d\n", devname, size, n);
1109 +
1110 +    vmaw->stream_info[n].devname = g_strdup(devname);
1111 +    vmaw->stream_info[n].size = size;
1112 +
1113 +    vmaw->stream_info[n].cluster_count = (size + VMA_CLUSTER_SIZE - 1) /
1114 +        VMA_CLUSTER_SIZE;
1115 +
1116 +    vmaw->stream_count = n;
1117 +
1118 +    if (strcmp(devname, "vmstate") == 0) {
1119 +        vmaw->vmstate_stream = n;
1120 +    }
1121 +
1122 +    return n;
1123 +}
1124 +
1125 +static void coroutine_fn yield_until_fd_writable(int fd)
1126 +{
1127 +    assert(qemu_in_coroutine());
1128 +    AioContext *ctx = qemu_get_current_aio_context();
1129 +    aio_set_fd_handler(ctx, fd, false, NULL, (IOHandler *)qemu_coroutine_enter,
1130 +                       NULL, qemu_coroutine_self());
1131 +    qemu_coroutine_yield();
1132 +    aio_set_fd_handler(ctx, fd, false, NULL, NULL, NULL, NULL);
1133 +}
1134 +
1135 +static ssize_t coroutine_fn
1136 +vma_queue_write(VmaWriter *vmaw, const void *buf, size_t bytes)
1137 +{
1138 +    DPRINTF("vma_queue_write enter %zd\n", bytes);
1139 +
1140 +    assert(vmaw);
1141 +    assert(buf);
1142 +    assert(bytes <= VMA_MAX_EXTENT_SIZE);
1143 +
1144 +    size_t done = 0;
1145 +    ssize_t ret;
1146 +
1147 +    assert(vmaw->co_writer == NULL);
1148 +
1149 +    vmaw->co_writer = qemu_coroutine_self();
1150 +
1151 +    while (done < bytes) {
1152 +        if (vmaw->status < 0) {
1153 +            DPRINTF("vma_queue_write detected canceled backup\n");
1154 +            done = -1;
1155 +            break;
1156 +        }
1157 +        yield_until_fd_writable(vmaw->fd);
1158 +        ret = write(vmaw->fd, buf + done, bytes - done);
1159 +        if (ret > 0) {
1160 +            done += ret;
1161 +            DPRINTF("vma_queue_write written %zd %zd\n", done, ret);
1162 +        } else if (ret < 0) {
1163 +            if (errno == EAGAIN || errno == EWOULDBLOCK) {
1164 +                /* try again */
1165 +           } else {
1166 +                vma_writer_set_error(vmaw, "vma_queue_write: write error - %s",
1167 +                                     g_strerror(errno));
1168 +                done = -1; /* always return failure for partial writes */
1169 +                break;
1170 +            }
1171 +        } else if (ret == 0) {
1172 +            /* should not happen - simply try again */
1173 +        }
1174 +    }
1175 +
1176 +    vmaw->co_writer = NULL;
1177 +
1178 +    return (done == bytes) ? bytes : -1;
1179 +}
1180 +
1181 +VmaWriter *vma_writer_create(const char *filename, uuid_t uuid, Error **errp)
1182 +{
1183 +    const char *p;
1184 +
1185 +    assert(sizeof(VmaHeader) == (4096 + 8192));
1186 +    assert(G_STRUCT_OFFSET(VmaHeader, config_names) == 2044);
1187 +    assert(G_STRUCT_OFFSET(VmaHeader, config_data) == 3068);
1188 +    assert(G_STRUCT_OFFSET(VmaHeader, dev_info) == 4096);
1189 +    assert(sizeof(VmaExtentHeader) == 512);
1190 +
1191 +    VmaWriter *vmaw = g_new0(VmaWriter, 1);
1192 +    vmaw->fd = -1;
1193 +
1194 +    vmaw->md5csum = g_checksum_new(G_CHECKSUM_MD5);
1195 +    if (!vmaw->md5csum) {
1196 +        error_setg(errp, "can't allocate cmsum\n");
1197 +        goto err;
1198 +    }
1199 +
1200 +    if (strstart(filename, "exec:", &p)) {
1201 +        vmaw->cmd = popen(p, "w");
1202 +        if (vmaw->cmd == NULL) {
1203 +            error_setg(errp, "can't popen command '%s' - %s\n", p,
1204 +                       g_strerror(errno));
1205 +            goto err;
1206 +        }
1207 +        vmaw->fd = fileno(vmaw->cmd);
1208 +
1209 +        /* try to use O_NONBLOCK */
1210 +        fcntl(vmaw->fd, F_SETFL, fcntl(vmaw->fd, F_GETFL)|O_NONBLOCK);
1211 +
1212 +    } else {
1213 +        struct stat st;
1214 +        int oflags;
1215 +        const char *tmp_id_str;
1216 +
1217 +        if ((stat(filename, &st) == 0) && S_ISFIFO(st.st_mode)) {
1218 +            oflags = O_NONBLOCK|O_WRONLY;
1219 +            vmaw->fd = qemu_open(filename, oflags, errp);
1220 +        } else if (strstart(filename, "/dev/fdset/", &tmp_id_str)) {
1221 +            oflags = O_NONBLOCK|O_WRONLY;
1222 +            vmaw->fd = qemu_open(filename, oflags, errp);
1223 +        } else if (strstart(filename, "/dev/fdname/", &tmp_id_str)) {
1224 +            vmaw->fd = monitor_get_fd(monitor_cur(), tmp_id_str, errp);
1225 +            if (vmaw->fd < 0) {
1226 +                goto err;
1227 +            }
1228 +            /* try to use O_NONBLOCK */
1229 +            fcntl(vmaw->fd, F_SETFL, fcntl(vmaw->fd, F_GETFL)|O_NONBLOCK);
1230 +        } else  {
1231 +            oflags = O_NONBLOCK|O_DIRECT|O_WRONLY|O_EXCL;
1232 +            vmaw->fd = qemu_create(filename, oflags, 0644, errp);
1233 +        }
1234 +
1235 +        if (vmaw->fd < 0) {
1236 +            error_setg(errp, "can't open file %s - %s\n", filename,
1237 +                       g_strerror(errno));
1238 +            goto err;
1239 +        }
1240 +    }
1241 +
1242 +    /* we use O_DIRECT, so we need to align IO buffers */
1243 +
1244 +    vmaw->outbuf = qemu_memalign(512, VMA_MAX_EXTENT_SIZE);
1245 +    vmaw->headerbuf = qemu_memalign(512, HEADERBUF_SIZE);
1246 +
1247 +    vmaw->outbuf_count = 0;
1248 +    vmaw->outbuf_pos = VMA_EXTENT_HEADER_SIZE;
1249 +
1250 +    vmaw->header_blob_table_pos = 1; /* start at pos 1 */
1251 +
1252 +    qemu_co_mutex_init(&vmaw->flush_lock);
1253 +
1254 +    uuid_copy(vmaw->uuid, uuid);
1255 +
1256 +    return vmaw;
1257 +
1258 +err:
1259 +    if (vmaw) {
1260 +        if (vmaw->cmd) {
1261 +            pclose(vmaw->cmd);
1262 +        } else if (vmaw->fd >= 0) {
1263 +            close(vmaw->fd);
1264 +        }
1265 +
1266 +        if (vmaw->md5csum) {
1267 +            g_checksum_free(vmaw->md5csum);
1268 +        }
1269 +
1270 +        g_free(vmaw);
1271 +    }
1272 +
1273 +    return NULL;
1274 +}
1275 +
1276 +static int coroutine_fn vma_write_header(VmaWriter *vmaw)
1277 +{
1278 +    assert(vmaw);
1279 +    unsigned char *buf = vmaw->headerbuf;
1280 +    VmaHeader *head = (VmaHeader *)buf;
1281 +
1282 +    int i;
1283 +
1284 +    DPRINTF("VMA WRITE HEADER\n");
1285 +
1286 +    if (vmaw->status < 0) {
1287 +        return vmaw->status;
1288 +    }
1289 +
1290 +    memset(buf, 0, HEADERBUF_SIZE);
1291 +
1292 +    head->magic = VMA_MAGIC;
1293 +    head->version = GUINT32_TO_BE(1); /* v1 */
1294 +    memcpy(head->uuid, vmaw->uuid, 16);
1295 +
1296 +    time_t ctime = time(NULL);
1297 +    head->ctime = GUINT64_TO_BE(ctime);
1298 +
1299 +    for (i = 0; i < VMA_MAX_CONFIGS; i++) {
1300 +        head->config_names[i] = GUINT32_TO_BE(vmaw->config_names[i]);
1301 +        head->config_data[i] = GUINT32_TO_BE(vmaw->config_data[i]);
1302 +    }
1303 +
1304 +    /* 32 bytes per device (12 used currently) = 8192 bytes max */
1305 +    for (i = 1; i <= 254; i++) {
1306 +        VmaStreamInfo *si = &vmaw->stream_info[i];
1307 +        if (si->size) {
1308 +            assert(si->devname);
1309 +            uint32_t devname_ptr = allocate_header_string(vmaw, si->devname);
1310 +            if (!devname_ptr) {
1311 +                return -1;
1312 +            }
1313 +            head->dev_info[i].devname_ptr = GUINT32_TO_BE(devname_ptr);
1314 +            head->dev_info[i].size = GUINT64_TO_BE(si->size);
1315 +        }
1316 +    }
1317 +
1318 +    uint32_t header_size = sizeof(VmaHeader) + vmaw->header_blob_table_size;
1319 +    head->header_size = GUINT32_TO_BE(header_size);
1320 +
1321 +    if (header_size > HEADERBUF_SIZE) {
1322 +        return -1; /* just to be sure */
1323 +    }
1324 +
1325 +    uint32_t blob_buffer_offset = sizeof(VmaHeader);
1326 +    memcpy(buf + blob_buffer_offset, vmaw->header_blob_table,
1327 +           vmaw->header_blob_table_size);
1328 +    head->blob_buffer_offset = GUINT32_TO_BE(blob_buffer_offset);
1329 +    head->blob_buffer_size = GUINT32_TO_BE(vmaw->header_blob_table_pos);
1330 +
1331 +    g_checksum_reset(vmaw->md5csum);
1332 +    g_checksum_update(vmaw->md5csum, (const guchar *)buf, header_size);
1333 +    gsize csize = 16;
1334 +    g_checksum_get_digest(vmaw->md5csum, (guint8 *)(head->md5sum), &csize);
1335 +
1336 +    return vma_queue_write(vmaw, buf, header_size);
1337 +}
1338 +
1339 +static int coroutine_fn vma_writer_flush(VmaWriter *vmaw)
1340 +{
1341 +    assert(vmaw);
1342 +
1343 +    int ret;
1344 +    int i;
1345 +
1346 +    if (vmaw->status < 0) {
1347 +        return vmaw->status;
1348 +    }
1349 +
1350 +    if (!vmaw->header_written) {
1351 +        vmaw->header_written = true;
1352 +        ret = vma_write_header(vmaw);
1353 +        if (ret < 0) {
1354 +            vma_writer_set_error(vmaw, "vma_writer_flush: write header failed");
1355 +            return ret;
1356 +        }
1357 +    }
1358 +
1359 +    DPRINTF("VMA WRITE FLUSH %d %d\n", vmaw->outbuf_count, vmaw->outbuf_pos);
1360 +
1361 +
1362 +    VmaExtentHeader *ehead = (VmaExtentHeader *)vmaw->outbuf;
1363 +
1364 +    ehead->magic = VMA_EXTENT_MAGIC;
1365 +    ehead->reserved1 = 0;
1366 +
1367 +    for (i = 0; i < VMA_BLOCKS_PER_EXTENT; i++) {
1368 +        ehead->blockinfo[i] = GUINT64_TO_BE(vmaw->outbuf_block_info[i]);
1369 +    }
1370 +
1371 +    guint16 block_count = (vmaw->outbuf_pos - VMA_EXTENT_HEADER_SIZE) /
1372 +        VMA_BLOCK_SIZE;
1373 +
1374 +    ehead->block_count = GUINT16_TO_BE(block_count);
1375 +
1376 +    memcpy(ehead->uuid, vmaw->uuid, sizeof(ehead->uuid));
1377 +    memset(ehead->md5sum, 0, sizeof(ehead->md5sum));
1378 +
1379 +    g_checksum_reset(vmaw->md5csum);
1380 +    g_checksum_update(vmaw->md5csum, vmaw->outbuf, VMA_EXTENT_HEADER_SIZE);
1381 +    gsize csize = 16;
1382 +    g_checksum_get_digest(vmaw->md5csum, ehead->md5sum, &csize);
1383 +
1384 +    int bytes = vmaw->outbuf_pos;
1385 +    ret = vma_queue_write(vmaw, vmaw->outbuf, bytes);
1386 +    if (ret != bytes) {
1387 +        vma_writer_set_error(vmaw, "vma_writer_flush: failed write");
1388 +    }
1389 +
1390 +    vmaw->outbuf_count = 0;
1391 +    vmaw->outbuf_pos = VMA_EXTENT_HEADER_SIZE;
1392 +
1393 +    for (i = 0; i < VMA_BLOCKS_PER_EXTENT; i++) {
1394 +        vmaw->outbuf_block_info[i] = 0;
1395 +    }
1396 +
1397 +    return vmaw->status;
1398 +}
1399 +
1400 +static int vma_count_open_streams(VmaWriter *vmaw)
1401 +{
1402 +    g_assert(vmaw != NULL);
1403 +
1404 +    int i;
1405 +    int open_drives = 0;
1406 +    for (i = 0; i <= 255; i++) {
1407 +        if (vmaw->stream_info[i].size && !vmaw->stream_info[i].finished) {
1408 +            open_drives++;
1409 +        }
1410 +    }
1411 +
1412 +    return open_drives;
1413 +}
1414 +
1415 +
1416 +/**
1417 + * You need to call this if the vma archive does not contain
1418 + * any data stream.
1419 + */
1420 +int coroutine_fn
1421 +vma_writer_flush_output(VmaWriter *vmaw)
1422 +{
1423 +    qemu_co_mutex_lock(&vmaw->flush_lock);
1424 +    int ret = vma_writer_flush(vmaw);
1425 +    qemu_co_mutex_unlock(&vmaw->flush_lock);
1426 +    if (ret < 0) {
1427 +        vma_writer_set_error(vmaw, "vma_writer_flush_header failed");
1428 +    }
1429 +    return ret;
1430 +}
1431 +
1432 +/**
1433 + * all jobs should call this when there is no more data
1434 + * Returns: number of remaining stream (0 ==> finished)
1435 + */
1436 +int coroutine_fn
1437 +vma_writer_close_stream(VmaWriter *vmaw, uint8_t dev_id)
1438 +{
1439 +    g_assert(vmaw != NULL);
1440 +
1441 +    DPRINTF("vma_writer_set_status %d\n", dev_id);
1442 +    if (!vmaw->stream_info[dev_id].size) {
1443 +        vma_writer_set_error(vmaw, "vma_writer_close_stream: "
1444 +                             "no such stream %d", dev_id);
1445 +        return -1;
1446 +    }
1447 +    if (vmaw->stream_info[dev_id].finished) {
1448 +        vma_writer_set_error(vmaw, "vma_writer_close_stream: "
1449 +                             "stream already closed %d", dev_id);
1450 +        return -1;
1451 +    }
1452 +
1453 +    vmaw->stream_info[dev_id].finished = true;
1454 +
1455 +    int open_drives = vma_count_open_streams(vmaw);
1456 +
1457 +    if (open_drives <= 0) {
1458 +        DPRINTF("vma_writer_set_status all drives completed\n");
1459 +        vma_writer_flush_output(vmaw);
1460 +    }
1461 +
1462 +    return open_drives;
1463 +}
1464 +
1465 +int vma_writer_get_status(VmaWriter *vmaw, VmaStatus *status)
1466 +{
1467 +    int i;
1468 +
1469 +    g_assert(vmaw != NULL);
1470 +
1471 +    if (status) {
1472 +        status->status = vmaw->status;
1473 +        g_strlcpy(status->errmsg, vmaw->errmsg, sizeof(status->errmsg));
1474 +        for (i = 0; i <= 255; i++) {
1475 +            status->stream_info[i] = vmaw->stream_info[i];
1476 +        }
1477 +
1478 +        uuid_unparse_lower(vmaw->uuid, status->uuid_str);
1479 +    }
1480 +
1481 +    status->closed = vmaw->closed;
1482 +
1483 +    return vmaw->status;
1484 +}
1485 +
1486 +static int vma_writer_get_buffer(VmaWriter *vmaw)
1487 +{
1488 +    int ret = 0;
1489 +
1490 +    qemu_co_mutex_lock(&vmaw->flush_lock);
1491 +
1492 +    /* wait until buffer is available */
1493 +    while (vmaw->outbuf_count >= (VMA_BLOCKS_PER_EXTENT - 1)) {
1494 +        ret = vma_writer_flush(vmaw);
1495 +        if (ret < 0) {
1496 +            vma_writer_set_error(vmaw, "vma_writer_get_buffer: flush failed");
1497 +            break;
1498 +        }
1499 +    }
1500 +
1501 +    qemu_co_mutex_unlock(&vmaw->flush_lock);
1502 +
1503 +    return ret;
1504 +}
1505 +
1506 +
1507 +int64_t coroutine_fn
1508 +vma_writer_write(VmaWriter *vmaw, uint8_t dev_id, int64_t cluster_num,
1509 +                 const unsigned char *buf, size_t *zero_bytes)
1510 +{
1511 +    g_assert(vmaw != NULL);
1512 +    g_assert(zero_bytes != NULL);
1513 +
1514 +    *zero_bytes = 0;
1515 +
1516 +    if (vmaw->status < 0) {
1517 +        return vmaw->status;
1518 +    }
1519 +
1520 +    if (!dev_id || !vmaw->stream_info[dev_id].size) {
1521 +        vma_writer_set_error(vmaw, "vma_writer_write: "
1522 +                             "no such stream %d", dev_id);
1523 +        return -1;
1524 +    }
1525 +
1526 +    if (vmaw->stream_info[dev_id].finished) {
1527 +        vma_writer_set_error(vmaw, "vma_writer_write: "
1528 +                             "stream already closed %d", dev_id);
1529 +        return -1;
1530 +    }
1531 +
1532 +
1533 +    if (cluster_num >= (((uint64_t)1)<<32)) {
1534 +        vma_writer_set_error(vmaw, "vma_writer_write: "
1535 +                             "cluster number out of range");
1536 +        return -1;
1537 +    }
1538 +
1539 +    if (dev_id == vmaw->vmstate_stream) {
1540 +        if (cluster_num != vmaw->vmstate_clusters) {
1541 +            vma_writer_set_error(vmaw, "vma_writer_write: "
1542 +                                 "non sequential vmstate write");
1543 +        }
1544 +        vmaw->vmstate_clusters++;
1545 +    } else if (cluster_num >= vmaw->stream_info[dev_id].cluster_count) {
1546 +        vma_writer_set_error(vmaw, "vma_writer_write: cluster number too big");
1547 +        return -1;
1548 +    }
1549 +
1550 +    /* wait until buffer is available */
1551 +    if (vma_writer_get_buffer(vmaw) < 0) {
1552 +        vma_writer_set_error(vmaw, "vma_writer_write: "
1553 +                             "vma_writer_get_buffer failed");
1554 +        return -1;
1555 +    }
1556 +
1557 +    DPRINTF("VMA WRITE %d %zd\n", dev_id, cluster_num);
1558 +
1559 +    uint64_t dev_size = vmaw->stream_info[dev_id].size;
1560 +    uint16_t mask = 0;
1561 +
1562 +    if (buf) {
1563 +        int i;
1564 +        int bit = 1;
1565 +        uint64_t byte_offset = cluster_num * VMA_CLUSTER_SIZE;
1566 +        for (i = 0; i < 16; i++) {
1567 +            const unsigned char *vmablock = buf + (i*VMA_BLOCK_SIZE);
1568 +
1569 +            // Note: If the source is not 64k-aligned, we might reach 4k blocks
1570 +            // after the end of the device. Always mark these as zero in the
1571 +            // mask, so the restore handles them correctly.
1572 +            if (byte_offset < dev_size &&
1573 +                !buffer_is_zero(vmablock, VMA_BLOCK_SIZE))
1574 +            {
1575 +                mask |= bit;
1576 +                memcpy(vmaw->outbuf + vmaw->outbuf_pos, vmablock,
1577 +                       VMA_BLOCK_SIZE);
1578 +
1579 +                // prevent memory leakage on unaligned last block
1580 +                if (byte_offset + VMA_BLOCK_SIZE > dev_size) {
1581 +                    uint64_t real_data_in_block = dev_size - byte_offset;
1582 +                    memset(vmaw->outbuf + vmaw->outbuf_pos + real_data_in_block,
1583 +                           0, VMA_BLOCK_SIZE - real_data_in_block);
1584 +                }
1585 +
1586 +                vmaw->outbuf_pos += VMA_BLOCK_SIZE;
1587 +            } else {
1588 +                DPRINTF("VMA WRITE %zd ZERO BLOCK %d\n", cluster_num, i);
1589 +                vmaw->stream_info[dev_id].zero_bytes += VMA_BLOCK_SIZE;
1590 +                *zero_bytes += VMA_BLOCK_SIZE;
1591 +            }
1592 +
1593 +            byte_offset += VMA_BLOCK_SIZE;
1594 +            bit = bit << 1;
1595 +        }
1596 +    } else {
1597 +        DPRINTF("VMA WRITE %zd ZERO CLUSTER\n", cluster_num);
1598 +        vmaw->stream_info[dev_id].zero_bytes += VMA_CLUSTER_SIZE;
1599 +        *zero_bytes += VMA_CLUSTER_SIZE;
1600 +    }
1601 +
1602 +    uint64_t block_info = ((uint64_t)mask) << (32+16);
1603 +    block_info |= ((uint64_t)dev_id) << 32;
1604 +    block_info |= (cluster_num & 0xffffffff);
1605 +    vmaw->outbuf_block_info[vmaw->outbuf_count] = block_info;
1606 +
1607 +    DPRINTF("VMA WRITE MASK %zd %zx\n", cluster_num, block_info);
1608 +
1609 +    vmaw->outbuf_count++;
1610 +
1611 +    /** NOTE: We allways write whole clusters, but we correctly set
1612 +     * transferred bytes. So transferred == size when when everything
1613 +     * went OK.
1614 +     */
1615 +    size_t transferred = VMA_CLUSTER_SIZE;
1616 +
1617 +    if (dev_id != vmaw->vmstate_stream) {
1618 +        uint64_t last = (cluster_num + 1) * VMA_CLUSTER_SIZE;
1619 +        if (last > dev_size) {
1620 +            uint64_t diff = last - dev_size;
1621 +            if (diff >= VMA_CLUSTER_SIZE) {
1622 +                vma_writer_set_error(vmaw, "vma_writer_write: "
1623 +                                     "read after last cluster");
1624 +                return -1;
1625 +            }
1626 +            transferred -= diff;
1627 +        }
1628 +    }
1629 +
1630 +    vmaw->stream_info[dev_id].transferred += transferred;
1631 +
1632 +    return transferred;
1633 +}
1634 +
1635 +void vma_writer_error_propagate(VmaWriter *vmaw, Error **errp)
1636 +{
1637 +    if (vmaw->status < 0 && *errp == NULL) {
1638 +        error_setg(errp, "%s", vmaw->errmsg);
1639 +    }
1640 +}
1641 +
1642 +int vma_writer_close(VmaWriter *vmaw, Error **errp)
1643 +{
1644 +    g_assert(vmaw != NULL);
1645 +
1646 +    int i;
1647 +
1648 +    qemu_co_mutex_lock(&vmaw->flush_lock); // wait for pending writes
1649 +
1650 +    assert(vmaw->co_writer == NULL);
1651 +
1652 +    if (vmaw->cmd) {
1653 +        if (pclose(vmaw->cmd) < 0) {
1654 +            vma_writer_set_error(vmaw, "vma_writer_close: "
1655 +                                 "pclose failed - %s", g_strerror(errno));
1656 +        }
1657 +    } else {
1658 +        if (close(vmaw->fd) < 0) {
1659 +            vma_writer_set_error(vmaw, "vma_writer_close: "
1660 +                                 "close failed - %s", g_strerror(errno));
1661 +        }
1662 +    }
1663 +
1664 +    for (i = 0; i <= 255; i++) {
1665 +        VmaStreamInfo *si = &vmaw->stream_info[i];
1666 +        if (si->size) {
1667 +            if (!si->finished) {
1668 +                vma_writer_set_error(vmaw, "vma_writer_close: "
1669 +                                     "detected open stream '%s'", si->devname);
1670 +            } else if ((si->transferred != si->size) &&
1671 +                       (i != vmaw->vmstate_stream)) {
1672 +                vma_writer_set_error(vmaw, "vma_writer_close: "
1673 +                                     "incomplete stream '%s' (%zd != %zd)",
1674 +                                     si->devname, si->transferred, si->size);
1675 +            }
1676 +        }
1677 +    }
1678 +
1679 +    for (i = 0; i <= 255; i++) {
1680 +        vmaw->stream_info[i].finished = 1; /* mark as closed */
1681 +    }
1682 +
1683 +    vmaw->closed = 1;
1684 +
1685 +    if (vmaw->status < 0 && *errp == NULL) {
1686 +        error_setg(errp, "%s", vmaw->errmsg);
1687 +    }
1688 +
1689 +    qemu_co_mutex_unlock(&vmaw->flush_lock);
1690 +
1691 +    return vmaw->status;
1692 +}
1693 +
1694 +void vma_writer_destroy(VmaWriter *vmaw)
1695 +{
1696 +    assert(vmaw);
1697 +
1698 +    int i;
1699 +
1700 +    for (i = 0; i <= 255; i++) {
1701 +        if (vmaw->stream_info[i].devname) {
1702 +            g_free(vmaw->stream_info[i].devname);
1703 +        }
1704 +    }
1705 +
1706 +    if (vmaw->md5csum) {
1707 +        g_checksum_free(vmaw->md5csum);
1708 +    }
1709 +
1710 +    qemu_vfree(vmaw->headerbuf);
1711 +    qemu_vfree(vmaw->outbuf);
1712 +    g_free(vmaw);
1713 +}
1714 diff --git a/vma.c b/vma.c
1715 new file mode 100644
1716 index 0000000000..2eea2fc281
1717 --- /dev/null
1718 +++ b/vma.c
1719 @@ -0,0 +1,839 @@
1720 +/*
1721 + * VMA: Virtual Machine Archive
1722 + *
1723 + * Copyright (C) 2012-2013 Proxmox Server Solutions
1724 + *
1725 + * Authors:
1726 + *  Dietmar Maurer (dietmar@proxmox.com)
1727 + *
1728 + * This work is licensed under the terms of the GNU GPL, version 2 or later.
1729 + * See the COPYING file in the top-level directory.
1730 + *
1731 + */
1732 +
1733 +#include "qemu/osdep.h"
1734 +#include <glib.h>
1735 +
1736 +#include "vma.h"
1737 +#include "qemu-common.h"
1738 +#include "qemu/module.h"
1739 +#include "qemu/error-report.h"
1740 +#include "qemu/main-loop.h"
1741 +#include "qemu/cutils.h"
1742 +#include "qapi/qmp/qdict.h"
1743 +#include "sysemu/block-backend.h"
1744 +
1745 +static void help(void)
1746 +{
1747 +    const char *help_msg =
1748 +        "usage: vma command [command options]\n"
1749 +        "\n"
1750 +        "vma list <filename>\n"
1751 +        "vma config <filename> [-c config]\n"
1752 +        "vma create <filename> [-c config] pathname ...\n"
1753 +        "vma extract <filename> [-r <fifo>] <targetdir>\n"
1754 +        "vma verify <filename> [-v]\n"
1755 +        ;
1756 +
1757 +    printf("%s", help_msg);
1758 +    exit(1);
1759 +}
1760 +
1761 +static const char *extract_devname(const char *path, char **devname, int index)
1762 +{
1763 +    assert(path);
1764 +
1765 +    const char *sep = strchr(path, '=');
1766 +
1767 +    if (sep) {
1768 +        *devname = g_strndup(path, sep - path);
1769 +        path = sep + 1;
1770 +    } else {
1771 +        if (index >= 0) {
1772 +            *devname = g_strdup_printf("disk%d", index);
1773 +        } else {
1774 +            *devname = NULL;
1775 +        }
1776 +    }
1777 +
1778 +    return path;
1779 +}
1780 +
1781 +static void print_content(VmaReader *vmar)
1782 +{
1783 +    assert(vmar);
1784 +
1785 +    VmaHeader *head = vma_reader_get_header(vmar);
1786 +
1787 +    GList *l = vma_reader_get_config_data(vmar);
1788 +    while (l && l->data) {
1789 +        VmaConfigData *cdata = (VmaConfigData *)l->data;
1790 +        l = g_list_next(l);
1791 +        printf("CFG: size: %d name: %s\n", cdata->len, cdata->name);
1792 +    }
1793 +
1794 +    int i;
1795 +    VmaDeviceInfo *di;
1796 +    for (i = 1; i < 255; i++) {
1797 +        di = vma_reader_get_device_info(vmar, i);
1798 +        if (di) {
1799 +            if (strcmp(di->devname, "vmstate") == 0) {
1800 +                printf("VMSTATE: dev_id=%d memory: %zd\n", i, di->size);
1801 +            } else {
1802 +                printf("DEV: dev_id=%d size: %zd devname: %s\n",
1803 +                       i, di->size, di->devname);
1804 +            }
1805 +        }
1806 +    }
1807 +    /* ctime is the last entry we print */
1808 +    printf("CTIME: %s", ctime(&head->ctime));
1809 +    fflush(stdout);
1810 +}
1811 +
1812 +static int list_content(int argc, char **argv)
1813 +{
1814 +    int c, ret = 0;
1815 +    const char *filename;
1816 +
1817 +    for (;;) {
1818 +        c = getopt(argc, argv, "h");
1819 +        if (c == -1) {
1820 +            break;
1821 +        }
1822 +        switch (c) {
1823 +        case '?':
1824 +        case 'h':
1825 +            help();
1826 +            break;
1827 +        default:
1828 +            g_assert_not_reached();
1829 +        }
1830 +    }
1831 +
1832 +    /* Get the filename */
1833 +    if ((optind + 1) != argc) {
1834 +        help();
1835 +    }
1836 +    filename = argv[optind++];
1837 +
1838 +    Error *errp = NULL;
1839 +    VmaReader *vmar = vma_reader_create(filename, &errp);
1840 +
1841 +    if (!vmar) {
1842 +        g_error("%s", error_get_pretty(errp));
1843 +    }
1844 +
1845 +    print_content(vmar);
1846 +
1847 +    vma_reader_destroy(vmar);
1848 +
1849 +    return ret;
1850 +}
1851 +
1852 +typedef struct RestoreMap {
1853 +    char *devname;
1854 +    char *path;
1855 +    char *format;
1856 +    uint64_t throttling_bps;
1857 +    char *throttling_group;
1858 +    char *cache;
1859 +    bool write_zero;
1860 +} RestoreMap;
1861 +
1862 +static bool try_parse_option(char **line, const char *optname, char **out, const char *inbuf) {
1863 +    size_t optlen = strlen(optname);
1864 +    if (strncmp(*line, optname, optlen) != 0 || (*line)[optlen] != '=') {
1865 +        return false;
1866 +    }
1867 +    if (*out) {
1868 +        g_error("read map failed - duplicate value for option '%s'", optname);
1869 +    }
1870 +    char *value = (*line) + optlen + 1; /* including a '=' */
1871 +    char *colon = strchr(value, ':');
1872 +    if (!colon) {
1873 +        g_error("read map failed - option '%s' not terminated ('%s')",
1874 +                optname, inbuf);
1875 +    }
1876 +    *line = colon+1;
1877 +    *out = g_strndup(value, colon - value);
1878 +    return true;
1879 +}
1880 +
1881 +static uint64_t verify_u64(const char *text) {
1882 +    uint64_t value;
1883 +    const char *endptr = NULL;
1884 +    if (qemu_strtou64(text, &endptr, 0, &value) != 0 || !endptr || *endptr) {
1885 +        g_error("read map failed - not a number: %s", text);
1886 +    }
1887 +    return value;
1888 +}
1889 +
1890 +static int extract_content(int argc, char **argv)
1891 +{
1892 +    int c, ret = 0;
1893 +    int verbose = 0;
1894 +    const char *filename;
1895 +    const char *dirname;
1896 +    const char *readmap = NULL;
1897 +
1898 +    for (;;) {
1899 +        c = getopt(argc, argv, "hvr:");
1900 +        if (c == -1) {
1901 +            break;
1902 +        }
1903 +        switch (c) {
1904 +        case '?':
1905 +        case 'h':
1906 +            help();
1907 +            break;
1908 +        case 'r':
1909 +            readmap = optarg;
1910 +            break;
1911 +        case 'v':
1912 +            verbose = 1;
1913 +            break;
1914 +        default:
1915 +            help();
1916 +        }
1917 +    }
1918 +
1919 +    /* Get the filename */
1920 +    if ((optind + 2) != argc) {
1921 +        help();
1922 +    }
1923 +    filename = argv[optind++];
1924 +    dirname = argv[optind++];
1925 +
1926 +    Error *errp = NULL;
1927 +    VmaReader *vmar = vma_reader_create(filename, &errp);
1928 +
1929 +    if (!vmar) {
1930 +        g_error("%s", error_get_pretty(errp));
1931 +    }
1932 +
1933 +    if (mkdir(dirname, 0777) < 0) {
1934 +        g_error("unable to create target directory %s - %s",
1935 +                dirname, g_strerror(errno));
1936 +    }
1937 +
1938 +    GList *l = vma_reader_get_config_data(vmar);
1939 +    while (l && l->data) {
1940 +        VmaConfigData *cdata = (VmaConfigData *)l->data;
1941 +        l = g_list_next(l);
1942 +        char *cfgfn = g_strdup_printf("%s/%s", dirname, cdata->name);
1943 +        GError *err = NULL;
1944 +        if (!g_file_set_contents(cfgfn, (gchar *)cdata->data, cdata->len,
1945 +                                 &err)) {
1946 +            g_error("unable to write file: %s", err->message);
1947 +        }
1948 +    }
1949 +
1950 +    GHashTable *devmap = g_hash_table_new(g_str_hash, g_str_equal);
1951 +
1952 +    if (readmap) {
1953 +        print_content(vmar);
1954 +
1955 +        FILE *map = fopen(readmap, "r");
1956 +        if (!map) {
1957 +            g_error("unable to open fifo %s - %s", readmap, g_strerror(errno));
1958 +        }
1959 +
1960 +        while (1) {
1961 +            char inbuf[8192];
1962 +            char *line = fgets(inbuf, sizeof(inbuf), map);
1963 +            char *format = NULL;
1964 +            char *bps = NULL;
1965 +            char *group = NULL;
1966 +            char *cache = NULL;
1967 +            if (!line || line[0] == '\0' || !strcmp(line, "done\n")) {
1968 +                break;
1969 +            }
1970 +            int len = strlen(line);
1971 +            if (line[len - 1] == '\n') {
1972 +                line[len - 1] = '\0';
1973 +                if (len == 1) {
1974 +                    break;
1975 +                }
1976 +            }
1977 +
1978 +            while (1) {
1979 +                if (!try_parse_option(&line, "format", &format, inbuf) &&
1980 +                    !try_parse_option(&line, "throttling.bps", &bps, inbuf) &&
1981 +                    !try_parse_option(&line, "throttling.group", &group, inbuf) &&
1982 +                    !try_parse_option(&line, "cache", &cache, inbuf))
1983 +                {
1984 +                    break;
1985 +                }
1986 +            }
1987 +
1988 +            uint64_t bps_value = 0;
1989 +            if (bps) {
1990 +                bps_value = verify_u64(bps);
1991 +                g_free(bps);
1992 +            }
1993 +
1994 +            const char *path;
1995 +            bool write_zero;
1996 +            if (line[0] == '0' && line[1] == ':') {
1997 +                path = line + 2;
1998 +                write_zero = false;
1999 +            } else if (line[0] == '1' && line[1] == ':') {
2000 +                path = line + 2;
2001 +                write_zero = true;
2002 +            } else {
2003 +                g_error("read map failed - parse error ('%s')", inbuf);
2004 +            }
2005 +
2006 +            char *devname = NULL;
2007 +            path = extract_devname(path, &devname, -1);
2008 +            if (!devname) {
2009 +                g_error("read map failed - no dev name specified ('%s')",
2010 +                        inbuf);
2011 +            }
2012 +
2013 +            RestoreMap *map = g_new0(RestoreMap, 1);
2014 +            map->devname = g_strdup(devname);
2015 +            map->path = g_strdup(path);
2016 +            map->format = format;
2017 +            map->throttling_bps = bps_value;
2018 +            map->throttling_group = group;
2019 +            map->cache = cache;
2020 +            map->write_zero = write_zero;
2021 +
2022 +            g_hash_table_insert(devmap, map->devname, map);
2023 +
2024 +        };
2025 +    }
2026 +
2027 +    int i;
2028 +    int vmstate_fd = -1;
2029 +    guint8 vmstate_stream = 0;
2030 +
2031 +    BlockBackend *blk = NULL;
2032 +
2033 +    for (i = 1; i < 255; i++) {
2034 +        VmaDeviceInfo *di = vma_reader_get_device_info(vmar, i);
2035 +        if (di && (strcmp(di->devname, "vmstate") == 0)) {
2036 +            vmstate_stream = i;
2037 +            char *statefn = g_strdup_printf("%s/vmstate.bin", dirname);
2038 +            vmstate_fd = open(statefn, O_WRONLY|O_CREAT|O_EXCL, 0644);
2039 +            if (vmstate_fd < 0) {
2040 +                g_error("create vmstate file '%s' failed - %s", statefn,
2041 +                        g_strerror(errno));
2042 +            }
2043 +            g_free(statefn);
2044 +        } else if (di) {
2045 +            char *devfn = NULL;
2046 +            const char *format = NULL;
2047 +            uint64_t throttling_bps = 0;
2048 +            const char *throttling_group = NULL;
2049 +            const char *cache = NULL;
2050 +            int flags = BDRV_O_RDWR;
2051 +            bool write_zero = true;
2052 +
2053 +            if (readmap) {
2054 +                RestoreMap *map;
2055 +                map = (RestoreMap *)g_hash_table_lookup(devmap, di->devname);
2056 +                if (map == NULL) {
2057 +                    g_error("no device name mapping for %s", di->devname);
2058 +                }
2059 +                devfn = map->path;
2060 +                format = map->format;
2061 +                throttling_bps = map->throttling_bps;
2062 +                throttling_group = map->throttling_group;
2063 +                cache = map->cache;
2064 +                write_zero = map->write_zero;
2065 +            } else {
2066 +                devfn = g_strdup_printf("%s/tmp-disk-%s.raw",
2067 +                                        dirname, di->devname);
2068 +                printf("DEVINFO %s %zd\n", devfn, di->size);
2069 +
2070 +                bdrv_img_create(devfn, "raw", NULL, NULL, NULL, di->size,
2071 +                                flags, true, &errp);
2072 +                if (errp) {
2073 +                    g_error("can't create file %s: %s", devfn,
2074 +                            error_get_pretty(errp));
2075 +                }
2076 +
2077 +                /* Note: we created an empty file above, so there is no
2078 +                 * need to write zeroes (so we generate a sparse file)
2079 +                 */
2080 +                write_zero = false;
2081 +            }
2082 +
2083 +           size_t devlen = strlen(devfn);
2084 +           QDict *options = NULL;
2085 +            bool writethrough;
2086 +            if (format) {
2087 +                /* explicit format from commandline */
2088 +                options = qdict_new();
2089 +                qdict_put_str(options, "driver", format);
2090 +            } else if ((devlen > 4 && strcmp(devfn+devlen-4, ".raw") == 0) ||
2091 +                      strncmp(devfn, "/dev/", 5) == 0)
2092 +           {
2093 +                /* This part is now deprecated for PVE as well (just as qemu
2094 +                 * deprecated not specifying an explicit raw format, too.
2095 +                 */
2096 +               /* explicit raw format */
2097 +               options = qdict_new();
2098 +               qdict_put_str(options, "driver", "raw");
2099 +           }
2100 +            if (cache && bdrv_parse_cache_mode(cache, &flags, &writethrough)) {
2101 +                g_error("invalid cache option: %s\n", cache);
2102 +            }
2103 +
2104 +           if (errp || !(blk = blk_new_open(devfn, NULL, options, flags, &errp))) {
2105 +                g_error("can't open file %s - %s", devfn,
2106 +                        error_get_pretty(errp));
2107 +            }
2108 +
2109 +            if (cache) {
2110 +                blk_set_enable_write_cache(blk, !writethrough);
2111 +            }
2112 +
2113 +            if (throttling_group) {
2114 +                blk_io_limits_enable(blk, throttling_group);
2115 +            }
2116 +
2117 +            if (throttling_bps) {
2118 +                if (!throttling_group) {
2119 +                    blk_io_limits_enable(blk, devfn);
2120 +                }
2121 +
2122 +                ThrottleConfig cfg;
2123 +                throttle_config_init(&cfg);
2124 +                cfg.buckets[THROTTLE_BPS_WRITE].avg = throttling_bps;
2125 +                Error *err = NULL;
2126 +                if (!throttle_is_valid(&cfg, &err)) {
2127 +                    error_report_err(err);
2128 +                    g_error("failed to apply throttling");
2129 +                }
2130 +                blk_set_io_limits(blk, &cfg);
2131 +            }
2132 +
2133 +            if (vma_reader_register_bs(vmar, i, blk, write_zero, &errp) < 0) {
2134 +                g_error("%s", error_get_pretty(errp));
2135 +            }
2136 +
2137 +            if (!readmap) {
2138 +                g_free(devfn);
2139 +            }
2140 +        }
2141 +    }
2142 +
2143 +    if (vma_reader_restore(vmar, vmstate_fd, verbose, &errp) < 0) {
2144 +        g_error("restore failed - %s", error_get_pretty(errp));
2145 +    }
2146 +
2147 +    if (!readmap) {
2148 +        for (i = 1; i < 255; i++) {
2149 +            VmaDeviceInfo *di = vma_reader_get_device_info(vmar, i);
2150 +            if (di && (i != vmstate_stream)) {
2151 +                char *tmpfn = g_strdup_printf("%s/tmp-disk-%s.raw",
2152 +                                              dirname, di->devname);
2153 +                char *fn = g_strdup_printf("%s/disk-%s.raw",
2154 +                                           dirname, di->devname);
2155 +                if (rename(tmpfn, fn) != 0) {
2156 +                    g_error("rename %s to %s failed - %s",
2157 +                            tmpfn, fn, g_strerror(errno));
2158 +                }
2159 +            }
2160 +        }
2161 +    }
2162 +
2163 +    vma_reader_destroy(vmar);
2164 +
2165 +    blk_unref(blk);
2166 +
2167 +    bdrv_close_all();
2168 +
2169 +    return ret;
2170 +}
2171 +
2172 +static int verify_content(int argc, char **argv)
2173 +{
2174 +    int c, ret = 0;
2175 +    int verbose = 0;
2176 +    const char *filename;
2177 +
2178 +    for (;;) {
2179 +        c = getopt(argc, argv, "hv");
2180 +        if (c == -1) {
2181 +            break;
2182 +        }
2183 +        switch (c) {
2184 +        case '?':
2185 +        case 'h':
2186 +            help();
2187 +            break;
2188 +        case 'v':
2189 +            verbose = 1;
2190 +            break;
2191 +        default:
2192 +            help();
2193 +        }
2194 +    }
2195 +
2196 +    /* Get the filename */
2197 +    if ((optind + 1) != argc) {
2198 +        help();
2199 +    }
2200 +    filename = argv[optind++];
2201 +
2202 +    Error *errp = NULL;
2203 +    VmaReader *vmar = vma_reader_create(filename, &errp);
2204 +
2205 +    if (!vmar) {
2206 +        g_error("%s", error_get_pretty(errp));
2207 +    }
2208 +
2209 +    if (verbose) {
2210 +        print_content(vmar);
2211 +    }
2212 +
2213 +    if (vma_reader_verify(vmar, verbose, &errp) < 0) {
2214 +        g_error("verify failed - %s", error_get_pretty(errp));
2215 +    }
2216 +
2217 +    vma_reader_destroy(vmar);
2218 +
2219 +    bdrv_close_all();
2220 +
2221 +    return ret;
2222 +}
2223 +
2224 +typedef struct BackupJob {
2225 +    BlockBackend *target;
2226 +    int64_t len;
2227 +    VmaWriter *vmaw;
2228 +    uint8_t dev_id;
2229 +} BackupJob;
2230 +
2231 +#define BACKUP_SECTORS_PER_CLUSTER (VMA_CLUSTER_SIZE / BDRV_SECTOR_SIZE)
2232 +
2233 +static void coroutine_fn backup_run_empty(void *opaque)
2234 +{
2235 +    VmaWriter *vmaw = (VmaWriter *)opaque;
2236 +
2237 +    vma_writer_flush_output(vmaw);
2238 +
2239 +    Error *err = NULL;
2240 +    if (vma_writer_close(vmaw, &err) != 0) {
2241 +        g_warning("vma_writer_close failed %s", error_get_pretty(err));
2242 +    }
2243 +}
2244 +
2245 +static void coroutine_fn backup_run(void *opaque)
2246 +{
2247 +    BackupJob *job = (BackupJob *)opaque;
2248 +    struct iovec iov;
2249 +    QEMUIOVector qiov;
2250 +
2251 +    int64_t start, end;
2252 +    int ret = 0;
2253 +
2254 +    unsigned char *buf = blk_blockalign(job->target, VMA_CLUSTER_SIZE);
2255 +
2256 +    start = 0;
2257 +    end = DIV_ROUND_UP(job->len / BDRV_SECTOR_SIZE,
2258 +                       BACKUP_SECTORS_PER_CLUSTER);
2259 +
2260 +    for (; start < end; start++) {
2261 +        iov.iov_base = buf;
2262 +        iov.iov_len = VMA_CLUSTER_SIZE;
2263 +        qemu_iovec_init_external(&qiov, &iov, 1);
2264 +
2265 +        ret = blk_co_preadv(job->target, start * VMA_CLUSTER_SIZE,
2266 +                            VMA_CLUSTER_SIZE, &qiov, 0);
2267 +        if (ret < 0) {
2268 +            vma_writer_set_error(job->vmaw, "read error", -1);
2269 +            goto out;
2270 +        }
2271 +
2272 +        size_t zb = 0;
2273 +        if (vma_writer_write(job->vmaw, job->dev_id, start, buf, &zb) < 0) {
2274 +            vma_writer_set_error(job->vmaw, "backup_dump_cb vma_writer_write failed", -1);
2275 +            goto out;
2276 +        }
2277 +    }
2278 +
2279 +
2280 +out:
2281 +    if (vma_writer_close_stream(job->vmaw, job->dev_id) <= 0) {
2282 +        Error *err = NULL;
2283 +        if (vma_writer_close(job->vmaw, &err) != 0) {
2284 +            g_warning("vma_writer_close failed %s", error_get_pretty(err));
2285 +        }
2286 +    }
2287 +    qemu_vfree(buf);
2288 +}
2289 +
2290 +static int create_archive(int argc, char **argv)
2291 +{
2292 +    int i, c;
2293 +    int verbose = 0;
2294 +    const char *archivename;
2295 +    GList *config_files = NULL;
2296 +
2297 +    for (;;) {
2298 +        c = getopt(argc, argv, "hvc:");
2299 +        if (c == -1) {
2300 +            break;
2301 +        }
2302 +        switch (c) {
2303 +        case '?':
2304 +        case 'h':
2305 +            help();
2306 +            break;
2307 +        case 'c':
2308 +            config_files = g_list_append(config_files, optarg);
2309 +            break;
2310 +        case 'v':
2311 +            verbose = 1;
2312 +            break;
2313 +        default:
2314 +            g_assert_not_reached();
2315 +        }
2316 +    }
2317 +
2318 +
2319 +    /* make sure we an archive name */
2320 +    if ((optind + 1) > argc) {
2321 +        help();
2322 +    }
2323 +
2324 +    archivename = argv[optind++];
2325 +
2326 +    uuid_t uuid;
2327 +    uuid_generate(uuid);
2328 +
2329 +    Error *local_err = NULL;
2330 +    VmaWriter *vmaw = vma_writer_create(archivename, uuid, &local_err);
2331 +
2332 +    if (vmaw == NULL) {
2333 +        g_error("%s", error_get_pretty(local_err));
2334 +    }
2335 +
2336 +    GList *l = config_files;
2337 +    while (l && l->data) {
2338 +        char *name = l->data;
2339 +        char *cdata = NULL;
2340 +        gsize clen = 0;
2341 +        GError *err = NULL;
2342 +        if (!g_file_get_contents(name, &cdata, &clen, &err)) {
2343 +            unlink(archivename);
2344 +            g_error("Unable to read file: %s", err->message);
2345 +        }
2346 +
2347 +        if (vma_writer_add_config(vmaw, name, cdata, clen) != 0) {
2348 +            unlink(archivename);
2349 +            g_error("Unable to append config data %s (len = %zd)",
2350 +                    name, clen);
2351 +        }
2352 +        l = g_list_next(l);
2353 +    }
2354 +
2355 +    int devcount = 0;
2356 +    while (optind < argc) {
2357 +        const char *path = argv[optind++];
2358 +        char *devname = NULL;
2359 +        path = extract_devname(path, &devname, devcount++);
2360 +
2361 +        Error *errp = NULL;
2362 +        BlockBackend *target;
2363 +
2364 +        target = blk_new_open(path, NULL, NULL, 0, &errp);
2365 +        if (!target) {
2366 +            unlink(archivename);
2367 +            g_error("bdrv_open '%s' failed - %s", path, error_get_pretty(errp));
2368 +        }
2369 +        int64_t size = blk_getlength(target);
2370 +        int dev_id = vma_writer_register_stream(vmaw, devname, size);
2371 +        if (dev_id <= 0) {
2372 +            unlink(archivename);
2373 +            g_error("vma_writer_register_stream '%s' failed", devname);
2374 +        }
2375 +
2376 +        BackupJob *job = g_new0(BackupJob, 1);
2377 +        job->len = size;
2378 +        job->target = target;
2379 +        job->vmaw = vmaw;
2380 +        job->dev_id = dev_id;
2381 +
2382 +        Coroutine *co = qemu_coroutine_create(backup_run, job);
2383 +        qemu_coroutine_enter(co);
2384 +    }
2385 +
2386 +    VmaStatus vmastat;
2387 +    int percent = 0;
2388 +    int last_percent = -1;
2389 +
2390 +    if (devcount) {
2391 +        while (1) {
2392 +            main_loop_wait(false);
2393 +            vma_writer_get_status(vmaw, &vmastat);
2394 +
2395 +            if (verbose) {
2396 +
2397 +                uint64_t total = 0;
2398 +                uint64_t transferred = 0;
2399 +                uint64_t zero_bytes = 0;
2400 +
2401 +                int i;
2402 +                for (i = 0; i < 256; i++) {
2403 +                    if (vmastat.stream_info[i].size) {
2404 +                        total += vmastat.stream_info[i].size;
2405 +                        transferred += vmastat.stream_info[i].transferred;
2406 +                        zero_bytes += vmastat.stream_info[i].zero_bytes;
2407 +                    }
2408 +                }
2409 +                percent = (transferred*100)/total;
2410 +                if (percent != last_percent) {
2411 +                    fprintf(stderr, "progress %d%% %zd/%zd %zd\n", percent,
2412 +                            transferred, total, zero_bytes);
2413 +                    fflush(stderr);
2414 +
2415 +                    last_percent = percent;
2416 +                }
2417 +            }
2418 +
2419 +            if (vmastat.closed) {
2420 +                break;
2421 +            }
2422 +        }
2423 +    } else {
2424 +        Coroutine *co = qemu_coroutine_create(backup_run_empty, vmaw);
2425 +        qemu_coroutine_enter(co);
2426 +        while (1) {
2427 +            main_loop_wait(false);
2428 +            vma_writer_get_status(vmaw, &vmastat);
2429 +            if (vmastat.closed) {
2430 +                    break;
2431 +            }
2432 +        }
2433 +    }
2434 +
2435 +    bdrv_drain_all();
2436 +
2437 +    vma_writer_get_status(vmaw, &vmastat);
2438 +
2439 +    if (verbose) {
2440 +        for (i = 0; i < 256; i++) {
2441 +            VmaStreamInfo *si = &vmastat.stream_info[i];
2442 +            if (si->size) {
2443 +                fprintf(stderr, "image %s: size=%zd zeros=%zd saved=%zd\n",
2444 +                        si->devname, si->size, si->zero_bytes,
2445 +                        si->size - si->zero_bytes);
2446 +            }
2447 +        }
2448 +    }
2449 +
2450 +    if (vmastat.status < 0) {
2451 +        unlink(archivename);
2452 +        g_error("creating vma archive failed");
2453 +    }
2454 +
2455 +    vma_writer_destroy(vmaw);
2456 +    return 0;
2457 +}
2458 +
2459 +static int dump_config(int argc, char **argv)
2460 +{
2461 +    int c, ret = 0;
2462 +    const char *filename;
2463 +    const char *config_name = "qemu-server.conf";
2464 +
2465 +    for (;;) {
2466 +        c = getopt(argc, argv, "hc:");
2467 +        if (c == -1) {
2468 +            break;
2469 +        }
2470 +        switch (c) {
2471 +        case '?':
2472 +        case 'h':
2473 +            help();
2474 +            break;
2475 +        case 'c':
2476 +            config_name = optarg;
2477 +            break;
2478 +        default:
2479 +            help();
2480 +        }
2481 +    }
2482 +
2483 +    /* Get the filename */
2484 +    if ((optind + 1) != argc) {
2485 +        help();
2486 +    }
2487 +    filename = argv[optind++];
2488 +
2489 +    Error *errp = NULL;
2490 +    VmaReader *vmar = vma_reader_create(filename, &errp);
2491 +
2492 +    if (!vmar) {
2493 +        g_error("%s", error_get_pretty(errp));
2494 +    }
2495 +
2496 +    int found = 0;
2497 +    GList *l = vma_reader_get_config_data(vmar);
2498 +    while (l && l->data) {
2499 +        VmaConfigData *cdata = (VmaConfigData *)l->data;
2500 +        l = g_list_next(l);
2501 +        if (strcmp(cdata->name, config_name) == 0) {
2502 +            found = 1;
2503 +            fwrite(cdata->data,  cdata->len, 1, stdout);
2504 +            break;
2505 +        }
2506 +    }
2507 +
2508 +    vma_reader_destroy(vmar);
2509 +
2510 +    bdrv_close_all();
2511 +
2512 +    if (!found) {
2513 +        fprintf(stderr, "unable to find configuration data '%s'\n", config_name);
2514 +        return -1;
2515 +    }
2516 +
2517 +    return ret;
2518 +}
2519 +
2520 +int main(int argc, char **argv)
2521 +{
2522 +    const char *cmdname;
2523 +    Error *main_loop_err = NULL;
2524 +
2525 +    error_init(argv[0]);
2526 +    module_call_init(MODULE_INIT_TRACE);
2527 +    qemu_init_exec_dir(argv[0]);
2528 +
2529 +    if (qemu_init_main_loop(&main_loop_err)) {
2530 +        g_error("%s", error_get_pretty(main_loop_err));
2531 +    }
2532 +
2533 +    bdrv_init();
2534 +    module_call_init(MODULE_INIT_QOM);
2535 +
2536 +    if (argc < 2) {
2537 +        help();
2538 +    }
2539 +
2540 +    cmdname = argv[1];
2541 +    argc--; argv++;
2542 +
2543 +
2544 +    if (!strcmp(cmdname, "list")) {
2545 +        return list_content(argc, argv);
2546 +    } else if (!strcmp(cmdname, "create")) {
2547 +        return create_archive(argc, argv);
2548 +    } else if (!strcmp(cmdname, "extract")) {
2549 +        return extract_content(argc, argv);
2550 +    } else if (!strcmp(cmdname, "verify")) {
2551 +        return verify_content(argc, argv);
2552 +    } else if (!strcmp(cmdname, "config")) {
2553 +        return dump_config(argc, argv);
2554 +    }
2555 +
2556 +    help();
2557 +    return 0;
2558 +}
2559 diff --git a/vma.h b/vma.h
2560 new file mode 100644
2561 index 0000000000..c895c97f6d
2562 --- /dev/null
2563 +++ b/vma.h
2564 @@ -0,0 +1,150 @@
2565 +/*
2566 + * VMA: Virtual Machine Archive
2567 + *
2568 + * Copyright (C) Proxmox Server Solutions
2569 + *
2570 + * Authors:
2571 + *  Dietmar Maurer (dietmar@proxmox.com)
2572 + *
2573 + * This work is licensed under the terms of the GNU GPL, version 2 or later.
2574 + * See the COPYING file in the top-level directory.
2575 + *
2576 + */
2577 +
2578 +#ifndef BACKUP_VMA_H
2579 +#define BACKUP_VMA_H
2580 +
2581 +#include <uuid/uuid.h>
2582 +#include "qapi/error.h"
2583 +#include "block/block.h"
2584 +
2585 +#define VMA_BLOCK_BITS 12
2586 +#define VMA_BLOCK_SIZE (1<<VMA_BLOCK_BITS)
2587 +#define VMA_CLUSTER_BITS (VMA_BLOCK_BITS+4)
2588 +#define VMA_CLUSTER_SIZE (1<<VMA_CLUSTER_BITS)
2589 +
2590 +#if VMA_CLUSTER_SIZE != 65536
2591 +#error unexpected cluster size
2592 +#endif
2593 +
2594 +#define VMA_EXTENT_HEADER_SIZE 512
2595 +#define VMA_BLOCKS_PER_EXTENT 59
2596 +#define VMA_MAX_CONFIGS 256
2597 +
2598 +#define VMA_MAX_EXTENT_SIZE \
2599 +    (VMA_EXTENT_HEADER_SIZE+VMA_CLUSTER_SIZE*VMA_BLOCKS_PER_EXTENT)
2600 +#if VMA_MAX_EXTENT_SIZE != 3867136
2601 +#error unexpected VMA_EXTENT_SIZE
2602 +#endif
2603 +
2604 +/* File Format Definitions */
2605 +
2606 +#define VMA_MAGIC (GUINT32_TO_BE(('V'<<24)|('M'<<16)|('A'<<8)|0x00))
2607 +#define VMA_EXTENT_MAGIC (GUINT32_TO_BE(('V'<<24)|('M'<<16)|('A'<<8)|'E'))
2608 +
2609 +typedef struct VmaDeviceInfoHeader {
2610 +    uint32_t devname_ptr; /* offset into blob_buffer table */
2611 +    uint32_t reserved0;
2612 +    uint64_t size; /* device size in bytes */
2613 +    uint64_t reserved1;
2614 +    uint64_t reserved2;
2615 +} VmaDeviceInfoHeader;
2616 +
2617 +typedef struct VmaHeader {
2618 +    uint32_t magic;
2619 +    uint32_t version;
2620 +    unsigned char uuid[16];
2621 +    int64_t ctime;
2622 +    unsigned char md5sum[16];
2623 +
2624 +    uint32_t blob_buffer_offset;
2625 +    uint32_t blob_buffer_size;
2626 +    uint32_t header_size;
2627 +
2628 +    unsigned char reserved[1984];
2629 +
2630 +    uint32_t config_names[VMA_MAX_CONFIGS]; /* offset into blob_buffer table */
2631 +    uint32_t config_data[VMA_MAX_CONFIGS];  /* offset into blob_buffer table */
2632 +
2633 +    uint32_t reserved1;
2634 +
2635 +    VmaDeviceInfoHeader dev_info[256];
2636 +} VmaHeader;
2637 +
2638 +typedef struct VmaExtentHeader {
2639 +    uint32_t magic;
2640 +    uint16_t reserved1;
2641 +    uint16_t block_count;
2642 +    unsigned char uuid[16];
2643 +    unsigned char md5sum[16];
2644 +    uint64_t blockinfo[VMA_BLOCKS_PER_EXTENT];
2645 +} VmaExtentHeader;
2646 +
2647 +/* functions/definitions to read/write vma files */
2648 +
2649 +typedef struct VmaReader VmaReader;
2650 +
2651 +typedef struct VmaWriter VmaWriter;
2652 +
2653 +typedef struct VmaConfigData {
2654 +    const char *name;
2655 +    const void *data;
2656 +    uint32_t len;
2657 +} VmaConfigData;
2658 +
2659 +typedef struct VmaStreamInfo {
2660 +    uint64_t size;
2661 +    uint64_t cluster_count;
2662 +    uint64_t transferred;
2663 +    uint64_t zero_bytes;
2664 +    int finished;
2665 +    char *devname;
2666 +} VmaStreamInfo;
2667 +
2668 +typedef struct VmaStatus {
2669 +    int status;
2670 +    bool closed;
2671 +    char errmsg[8192];
2672 +    char uuid_str[37];
2673 +    VmaStreamInfo stream_info[256];
2674 +} VmaStatus;
2675 +
2676 +typedef struct VmaDeviceInfo {
2677 +    uint64_t size; /* device size in bytes */
2678 +    const char *devname;
2679 +} VmaDeviceInfo;
2680 +
2681 +VmaWriter *vma_writer_create(const char *filename, uuid_t uuid, Error **errp);
2682 +int vma_writer_close(VmaWriter *vmaw, Error **errp);
2683 +void vma_writer_error_propagate(VmaWriter *vmaw, Error **errp);
2684 +void vma_writer_destroy(VmaWriter *vmaw);
2685 +int vma_writer_add_config(VmaWriter *vmaw, const char *name, gpointer data,
2686 +                          size_t len);
2687 +int vma_writer_register_stream(VmaWriter *vmaw, const char *devname,
2688 +                               size_t size);
2689 +
2690 +int64_t coroutine_fn vma_writer_write(VmaWriter *vmaw, uint8_t dev_id,
2691 +                                      int64_t cluster_num,
2692 +                                      const unsigned char *buf,
2693 +                                      size_t *zero_bytes);
2694 +
2695 +int coroutine_fn vma_writer_close_stream(VmaWriter *vmaw, uint8_t dev_id);
2696 +int coroutine_fn vma_writer_flush_output(VmaWriter *vmaw);
2697 +
2698 +int vma_writer_get_status(VmaWriter *vmaw, VmaStatus *status);
2699 +void vma_writer_set_error(VmaWriter *vmaw, const char *fmt, ...);
2700 +
2701 +
2702 +VmaReader *vma_reader_create(const char *filename, Error **errp);
2703 +void vma_reader_destroy(VmaReader *vmar);
2704 +VmaHeader *vma_reader_get_header(VmaReader *vmar);
2705 +GList *vma_reader_get_config_data(VmaReader *vmar);
2706 +VmaDeviceInfo *vma_reader_get_device_info(VmaReader *vmar, guint8 dev_id);
2707 +int vma_reader_register_bs(VmaReader *vmar, guint8 dev_id,
2708 +                           BlockBackend *target, bool write_zeroes,
2709 +                           Error **errp);
2710 +int vma_reader_restore(VmaReader *vmar, int vmstate_fd, bool verbose,
2711 +                       Error **errp);
2712 +int vma_reader_verify(VmaReader *vmar, bool verbose, Error **errp);
2713 +
2714 +#endif /* BACKUP_VMA_H */