debian/patches/backup-add-vma-binary.patch

   1 From 6289a43696ca6f713a5d3bb9f95a5adb608a5e13 Mon Sep 17 00:00:00 2001
   2 From: Dietmar Maurer <dietmar@proxmox.com>
   3 Date: Tue, 13 Nov 2012 11:11:38 +0100
   4 Subject: [PATCH v5 4/6] introduce new vma archive format
   5
   6 This is a very simple archive format, see docs/specs/vma_spec.txt
   7
   8 Signed-off-by: Dietmar Maurer <dietmar@proxmox.com>
   9 ---
  10  Makefile                |    3 +-
  11  Makefile.objs           |    2 +-
  12  blockdev.c              |    6 +-
  13  docs/specs/vma_spec.txt |   24 ++
  14  vma-reader.c            |  799 ++++++++++++++++++++++++++++++++++++++++
  15  vma-writer.c            |  940 +++++++++++++++++++++++++++++++++++++++++++++++
  16  vma.c                   |  561 ++++++++++++++++++++++++++++
  17  vma.h                   |  145 ++++++++
  18  8 files changed, 2476 insertions(+), 4 deletions(-)
  19  create mode 100644 docs/specs/vma_spec.txt
  20  create mode 100644 vma-reader.c
  21  create mode 100644 vma-writer.c
  22  create mode 100644 vma.c
  23  create mode 100644 vma.h
  24
  25 Index: new/Makefile
  26 ===================================================================
  27 --- new.orig/Makefile   2013-11-27 10:51:32.000000000 +0100
  28 +++ new/Makefile        2013-11-28 09:57:02.000000000 +0100
  29 @@ -128,7 +128,7 @@
  30  include $(SRC_PATH)/libcacard/Makefile
  31  endif
  32
  33 -all: $(DOCS) $(TOOLS) $(HELPERS-y) recurse-all
  34 +all: $(DOCS) $(TOOLS) vma$(EXESUF) $(HELPERS-y) recurse-all
  35
  36  config-host.h: config-host.h-timestamp
  37  config-host.h-timestamp: config-host.mak
  38 @@ -194,6 +194,7 @@
  39  qemu-img$(EXESUF): qemu-img.o $(block-obj-y) libqemuutil.a libqemustub.a
  40  qemu-nbd$(EXESUF): qemu-nbd.o $(block-obj-y) libqemuutil.a libqemustub.a
  41  qemu-io$(EXESUF): qemu-io.o $(block-obj-y) libqemuutil.a libqemustub.a
  42 +vma$(EXESUF): vma.o vma-writer.o vma-reader.o $(block-obj-y)  libqemuutil.a libqemustub.a
  43
  44  qemu-bridge-helper$(EXESUF): qemu-bridge-helper.o
  45
  46 Index: new/docs/specs/vma_spec.txt
  47 ===================================================================
  48 --- /dev/null   1970-01-01 00:00:00.000000000 +0000
  49 +++ new/docs/specs/vma_spec.txt 2013-11-28 09:55:29.000000000 +0100
  50 @@ -0,0 +1,24 @@
  51 +=Virtual Machine Archive format (VMA)=
  52 +
  53 +This format contains a header which includes the VM configuration as
  54 +binary blobs, and a list of devices (dev_id, name).
  55 +
  56 +The actual VM image data is stored inside extents. An extent contains
  57 +up to 64 clusters, and start with a 512 byte header containing
  58 +additional information for those clusters.
  59 +
  60 +We use a cluster size of 65536, and use 8 bytes for each
  61 +cluster in the header to store the following information:
  62 +
  63 +* 1 byte dev_id (to identity the drive)
  64 +* 1 byte not used (reserved)
  65 +* 2 bytes zero indicator (mark zero regions (16x4096))
  66 +* 4 bytes cluster number
  67 +
  68 +We only store non-zero blocks (such block is 4096 bytes).
  69 +
  70 +Each archive is marked with a uuid. The archive header and all
  71 +extent headers includes that uuid and a MD5 checksum (over header
  72 +data).
  73 +
  74 +
  75 Index: new/vma-reader.c
  76 ===================================================================
  77 --- /dev/null   1970-01-01 00:00:00.000000000 +0000
  78 +++ new/vma-reader.c    2013-11-28 09:55:29.000000000 +0100
  79 @@ -0,0 +1,799 @@
  80 +/*
  81 + * VMA: Virtual Machine Archive
  82 + *
  83 + * Copyright (C) 2012 Proxmox Server Solutions
  84 + *
  85 + * Authors:
  86 + *  Dietmar Maurer (dietmar@proxmox.com)
  87 + *
  88 + * This work is licensed under the terms of the GNU GPL, version 2 or later.
  89 + * See the COPYING file in the top-level directory.
  90 + *
  91 + */
  92 +
  93 +#include <stdio.h>
  94 +#include <errno.h>
  95 +#include <unistd.h>
  96 +#include <stdio.h>
  97 +#include <string.h>
  98 +#include <sys/types.h>
  99 +#include <sys/stat.h>
 100 +#include <fcntl.h>
 101 +#include <glib.h>
 102 +#include <uuid/uuid.h>
 103 +
 104 +#include "qemu-common.h"
 105 +#include "qemu/timer.h"
 106 +#include "qemu/ratelimit.h"
 107 +#include "vma.h"
 108 +#include "block/block.h"
 109 +
 110 +#define BITS_PER_LONG  (sizeof(unsigned long) * CHAR_BIT)
 111 +
 112 +static unsigned char zero_vma_block[VMA_BLOCK_SIZE];
 113 +
 114 +typedef struct VmaRestoreState {
 115 +    BlockDriverState *bs;
 116 +    bool write_zeroes;
 117 +    unsigned long *bitmap;
 118 +    int bitmap_size;
 119 +}  VmaRestoreState;
 120 +
 121 +struct VmaReader {
 122 +    int fd;
 123 +    GChecksum *md5csum;
 124 +    GHashTable *blob_hash;
 125 +    unsigned char *head_data;
 126 +    VmaDeviceInfo devinfo[256];
 127 +    VmaRestoreState rstate[256];
 128 +    GList *cdata_list;
 129 +    guint8 vmstate_stream;
 130 +    uint32_t vmstate_clusters;
 131 +    /* to show restore percentage if run with -v */
 132 +    time_t start_time;
 133 +    int64_t cluster_count;
 134 +    int64_t clusters_read;
 135 +    int clusters_read_per;
 136 +};
 137 +
 138 +static guint
 139 +g_int32_hash(gconstpointer v)
 140 +{
 141 +    return *(const uint32_t *)v;
 142 +}
 143 +
 144 +static gboolean
 145 +g_int32_equal(gconstpointer v1, gconstpointer v2)
 146 +{
 147 +    return *((const uint32_t *)v1) == *((const uint32_t *)v2);
 148 +}
 149 +
 150 +static int vma_reader_get_bitmap(VmaRestoreState *rstate, int64_t cluster_num)
 151 +{
 152 +    assert(rstate);
 153 +    assert(rstate->bitmap);
 154 +
 155 +    unsigned long val, idx, bit;
 156 +
 157 +    idx = cluster_num / BITS_PER_LONG;
 158 +
 159 +    assert(rstate->bitmap_size > idx);
 160 +
 161 +    bit = cluster_num % BITS_PER_LONG;
 162 +    val = rstate->bitmap[idx];
 163 +
 164 +    return !!(val & (1UL << bit));
 165 +}
 166 +
 167 +static void vma_reader_set_bitmap(VmaRestoreState *rstate, int64_t cluster_num,
 168 +                                  int dirty)
 169 +{
 170 +    assert(rstate);
 171 +    assert(rstate->bitmap);
 172 +
 173 +    unsigned long val, idx, bit;
 174 +
 175 +    idx = cluster_num / BITS_PER_LONG;
 176 +
 177 +    assert(rstate->bitmap_size > idx);
 178 +
 179 +    bit = cluster_num % BITS_PER_LONG;
 180 +    val = rstate->bitmap[idx];
 181 +    if (dirty) {
 182 +        if (!(val & (1UL << bit))) {
 183 +            val |= 1UL << bit;
 184 +        }
 185 +    } else {
 186 +        if (val & (1UL << bit)) {
 187 +            val &= ~(1UL << bit);
 188 +        }
 189 +    }
 190 +    rstate->bitmap[idx] = val;
 191 +}
 192 +
 193 +typedef struct VmaBlob {
 194 +    uint32_t start;
 195 +    uint32_t len;
 196 +    void *data;
 197 +} VmaBlob;
 198 +
 199 +static const VmaBlob *get_header_blob(VmaReader *vmar, uint32_t pos)
 200 +{
 201 +    assert(vmar);
 202 +    assert(vmar->blob_hash);
 203 +
 204 +    return g_hash_table_lookup(vmar->blob_hash, &pos);
 205 +}
 206 +
 207 +static const char *get_header_str(VmaReader *vmar, uint32_t pos)
 208 +{
 209 +    const VmaBlob *blob = get_header_blob(vmar, pos);
 210 +    if (!blob) {
 211 +        return NULL;
 212 +    }
 213 +    const char *res = (char *)blob->data;
 214 +    if (res[blob->len-1] != '\0') {
 215 +        return NULL;
 216 +    }
 217 +    return res;
 218 +}
 219 +
 220 +static ssize_t
 221 +safe_read(int fd, unsigned char *buf, size_t count)
 222 +{
 223 +    ssize_t n;
 224 +
 225 +    do {
 226 +        n = read(fd, buf, count);
 227 +    } while (n < 0 && errno == EINTR);
 228 +
 229 +    return n;
 230 +}
 231 +
 232 +static ssize_t
 233 +full_read(int fd, unsigned char *buf, size_t len)
 234 +{
 235 +    ssize_t n;
 236 +    size_t total;
 237 +
 238 +    total = 0;
 239 +
 240 +    while (len > 0) {
 241 +        n = safe_read(fd, buf, len);
 242 +
 243 +        if (n == 0) {
 244 +            return total;
 245 +        }
 246 +
 247 +        if (n <= 0) {
 248 +            break;
 249 +        }
 250 +
 251 +        buf += n;
 252 +        total += n;
 253 +        len -= n;
 254 +    }
 255 +
 256 +    if (len) {
 257 +        return -1;
 258 +    }
 259 +
 260 +    return total;
 261 +}
 262 +
 263 +void vma_reader_destroy(VmaReader *vmar)
 264 +{
 265 +    assert(vmar);
 266 +
 267 +    if (vmar->fd >= 0) {
 268 +        close(vmar->fd);
 269 +    }
 270 +
 271 +    if (vmar->cdata_list) {
 272 +        g_list_free(vmar->cdata_list);
 273 +    }
 274 +
 275 +    int i;
 276 +    for (i = 1; i < 256; i++) {
 277 +        if (vmar->rstate[i].bitmap) {
 278 +            g_free(vmar->rstate[i].bitmap);
 279 +        }
 280 +    }
 281 +
 282 +    if (vmar->md5csum) {
 283 +        g_checksum_free(vmar->md5csum);
 284 +    }
 285 +
 286 +    if (vmar->blob_hash) {
 287 +        g_hash_table_destroy(vmar->blob_hash);
 288 +    }
 289 +
 290 +    if (vmar->head_data) {
 291 +        g_free(vmar->head_data);
 292 +    }
 293 +
 294 +    g_free(vmar);
 295 +
 296 +};
 297 +
 298 +static int vma_reader_read_head(VmaReader *vmar, Error **errp)
 299 +{
 300 +    assert(vmar);
 301 +    assert(errp);
 302 +    assert(*errp == NULL);
 303 +
 304 +    unsigned char md5sum[16];
 305 +    int i;
 306 +    int ret = 0;
 307 +
 308 +    vmar->head_data = g_malloc(sizeof(VmaHeader));
 309 +
 310 +    if (full_read(vmar->fd, vmar->head_data, sizeof(VmaHeader)) !=
 311 +        sizeof(VmaHeader)) {
 312 +        error_setg(errp, "can't read vma header - %s",
 313 +                   errno ? g_strerror(errno) : "got EOF");
 314 +        return -1;
 315 +    }
 316 +
 317 +    VmaHeader *h = (VmaHeader *)vmar->head_data;
 318 +
 319 +    if (h->magic != VMA_MAGIC) {
 320 +        error_setg(errp, "not a vma file - wrong magic number");
 321 +        return -1;
 322 +    }
 323 +
 324 +    uint32_t header_size = GUINT32_FROM_BE(h->header_size);
 325 +    int need = header_size - sizeof(VmaHeader);
 326 +    if (need <= 0) {
 327 +        error_setg(errp, "wrong vma header size %d", header_size);
 328 +        return -1;
 329 +    }
 330 +
 331 +    vmar->head_data = g_realloc(vmar->head_data, header_size);
 332 +    h = (VmaHeader *)vmar->head_data;
 333 +
 334 +    if (full_read(vmar->fd, vmar->head_data + sizeof(VmaHeader), need) !=
 335 +        need) {
 336 +        error_setg(errp, "can't read vma header data - %s",
 337 +                   errno ? g_strerror(errno) : "got EOF");
 338 +        return -1;
 339 +    }
 340 +
 341 +    memcpy(md5sum, h->md5sum, 16);
 342 +    memset(h->md5sum, 0, 16);
 343 +
 344 +    g_checksum_reset(vmar->md5csum);
 345 +    g_checksum_update(vmar->md5csum, vmar->head_data, header_size);
 346 +    gsize csize = 16;
 347 +    g_checksum_get_digest(vmar->md5csum, (guint8 *)(h->md5sum), &csize);
 348 +
 349 +    if (memcmp(md5sum, h->md5sum, 16) != 0) {
 350 +        error_setg(errp, "wrong vma header chechsum");
 351 +        return -1;
 352 +    }
 353 +
 354 +    /* we can modify header data after checksum verify */
 355 +    h->header_size = header_size;
 356 +
 357 +    h->version = GUINT32_FROM_BE(h->version);
 358 +    if (h->version != 1) {
 359 +        error_setg(errp, "wrong vma version %d", h->version);
 360 +        return -1;
 361 +    }
 362 +
 363 +    h->ctime = GUINT64_FROM_BE(h->ctime);
 364 +    h->blob_buffer_offset = GUINT32_FROM_BE(h->blob_buffer_offset);
 365 +    h->blob_buffer_size = GUINT32_FROM_BE(h->blob_buffer_size);
 366 +
 367 +    uint32_t bstart = h->blob_buffer_offset + 1;
 368 +    uint32_t bend = h->blob_buffer_offset + h->blob_buffer_size;
 369 +
 370 +    if (bstart <= sizeof(VmaHeader)) {
 371 +        error_setg(errp, "wrong vma blob buffer offset %d",
 372 +                   h->blob_buffer_offset);
 373 +        return -1;
 374 +    }
 375 +
 376 +    if (bend > header_size) {
 377 +        error_setg(errp, "wrong vma blob buffer size %d/%d",
 378 +                   h->blob_buffer_offset, h->blob_buffer_size);
 379 +        return -1;
 380 +    }
 381 +
 382 +    while ((bstart + 2) <= bend) {
 383 +        uint32_t size = vmar->head_data[bstart] +
 384 +            (vmar->head_data[bstart+1] << 8);
 385 +        if ((bstart + size + 2) <= bend) {
 386 +            VmaBlob *blob = g_new0(VmaBlob, 1);
 387 +            blob->start = bstart - h->blob_buffer_offset;
 388 +            blob->len = size;
 389 +            blob->data = vmar->head_data + bstart + 2;
 390 +            g_hash_table_insert(vmar->blob_hash, &blob->start, blob);
 391 +        }
 392 +        bstart += size + 2;
 393 +    }
 394 +
 395 +
 396 +    int count = 0;
 397 +    for (i = 1; i < 256; i++) {
 398 +        VmaDeviceInfoHeader *dih = &h->dev_info[i];
 399 +        uint32_t devname_ptr = GUINT32_FROM_BE(dih->devname_ptr);
 400 +        uint64_t size = GUINT64_FROM_BE(dih->size);
 401 +        const char *devname =  get_header_str(vmar, devname_ptr);
 402 +
 403 +        if (size && devname) {
 404 +            count++;
 405 +            vmar->devinfo[i].size = size;
 406 +            vmar->devinfo[i].devname = devname;
 407 +
 408 +            if (strcmp(devname, "vmstate") == 0) {
 409 +                vmar->vmstate_stream = i;
 410 +            }
 411 +        }
 412 +    }
 413 +
 414 +    if (!count) {
 415 +        error_setg(errp, "vma does not contain data");
 416 +        return -1;
 417 +    }
 418 +
 419 +    for (i = 0; i < VMA_MAX_CONFIGS; i++) {
 420 +        uint32_t name_ptr = GUINT32_FROM_BE(h->config_names[i]);
 421 +        uint32_t data_ptr = GUINT32_FROM_BE(h->config_data[i]);
 422 +
 423 +        if (!(name_ptr && data_ptr)) {
 424 +            continue;
 425 +        }
 426 +        const char *name =  get_header_str(vmar, name_ptr);
 427 +        const VmaBlob *blob = get_header_blob(vmar, data_ptr);
 428 +
 429 +        if (!(name && blob)) {
 430 +            error_setg(errp, "vma contains invalid data pointers");
 431 +            return -1;
 432 +        }
 433 +
 434 +        VmaConfigData *cdata = g_new0(VmaConfigData, 1);
 435 +        cdata->name = name;
 436 +        cdata->data = blob->data;
 437 +        cdata->len = blob->len;
 438 +
 439 +        vmar->cdata_list = g_list_append(vmar->cdata_list, cdata);
 440 +    }
 441 +
 442 +    return ret;
 443 +};
 444 +
 445 +VmaReader *vma_reader_create(const char *filename, Error **errp)
 446 +{
 447 +    assert(filename);
 448 +    assert(errp);
 449 +
 450 +    VmaReader *vmar = g_new0(VmaReader, 1);
 451 +
 452 +    if (strcmp(filename, "-") == 0) {
 453 +        vmar->fd = dup(0);
 454 +    } else {
 455 +        vmar->fd = open(filename, O_RDONLY);
 456 +    }
 457 +
 458 +    if (vmar->fd < 0) {
 459 +        error_setg(errp, "can't open file %s - %s\n", filename,
 460 +                   g_strerror(errno));
 461 +        goto err;
 462 +    }
 463 +
 464 +    vmar->md5csum = g_checksum_new(G_CHECKSUM_MD5);
 465 +    if (!vmar->md5csum) {
 466 +        error_setg(errp, "can't allocate cmsum\n");
 467 +        goto err;
 468 +    }
 469 +
 470 +    vmar->blob_hash = g_hash_table_new_full(g_int32_hash, g_int32_equal,
 471 +                                            NULL, g_free);
 472 +
 473 +    if (vma_reader_read_head(vmar, errp) < 0) {
 474 +        goto err;
 475 +    }
 476 +
 477 +    return vmar;
 478 +
 479 +err:
 480 +    if (vmar) {
 481 +        vma_reader_destroy(vmar);
 482 +    }
 483 +
 484 +    return NULL;
 485 +}
 486 +
 487 +VmaHeader *vma_reader_get_header(VmaReader *vmar)
 488 +{
 489 +    assert(vmar);
 490 +    assert(vmar->head_data);
 491 +
 492 +    return (VmaHeader *)(vmar->head_data);
 493 +}
 494 +
 495 +GList *vma_reader_get_config_data(VmaReader *vmar)
 496 +{
 497 +    assert(vmar);
 498 +    assert(vmar->head_data);
 499 +
 500 +    return vmar->cdata_list;
 501 +}
 502 +
 503 +VmaDeviceInfo *vma_reader_get_device_info(VmaReader *vmar, guint8 dev_id)
 504 +{
 505 +    assert(vmar);
 506 +    assert(dev_id);
 507 +
 508 +    if (vmar->devinfo[dev_id].size && vmar->devinfo[dev_id].devname) {
 509 +        return &vmar->devinfo[dev_id];
 510 +    }
 511 +
 512 +    return NULL;
 513 +}
 514 +
 515 +int vma_reader_register_bs(VmaReader *vmar, guint8 dev_id, BlockDriverState *bs,
 516 +                           bool write_zeroes, Error **errp)
 517 +{
 518 +    assert(vmar);
 519 +    assert(bs != NULL);
 520 +    assert(dev_id);
 521 +    assert(vmar->rstate[dev_id].bs == NULL);
 522 +
 523 +    int64_t size = bdrv_getlength(bs);
 524 +    if (size != vmar->devinfo[dev_id].size) {
 525 +        error_setg(errp, "vma_reader_register_bs for stream %s failed - "
 526 +                   "unexpected size %zd != %zd", vmar->devinfo[dev_id].devname,
 527 +                   size, vmar->devinfo[dev_id].size);
 528 +        return -1;
 529 +    }
 530 +
 531 +    vmar->rstate[dev_id].bs = bs;
 532 +    vmar->rstate[dev_id].write_zeroes = write_zeroes;
 533 +
 534 +    int64_t bitmap_size = (size/BDRV_SECTOR_SIZE) +
 535 +        (VMA_CLUSTER_SIZE/BDRV_SECTOR_SIZE) * BITS_PER_LONG - 1;
 536 +    bitmap_size /= (VMA_CLUSTER_SIZE/BDRV_SECTOR_SIZE) * BITS_PER_LONG;
 537 +
 538 +    vmar->rstate[dev_id].bitmap_size = bitmap_size;
 539 +    vmar->rstate[dev_id].bitmap = g_new0(unsigned long, bitmap_size);
 540 +
 541 +    vmar->cluster_count += size/VMA_CLUSTER_SIZE;
 542 +
 543 +    return 0;
 544 +}
 545 +
 546 +static ssize_t safe_write(int fd, void *buf, size_t count)
 547 +{
 548 +    ssize_t n;
 549 +
 550 +    do {
 551 +        n = write(fd, buf, count);
 552 +    } while (n < 0 && errno == EINTR);
 553 +
 554 +    return n;
 555 +}
 556 +
 557 +static size_t full_write(int fd, void *buf, size_t len)
 558 +{
 559 +    ssize_t n;
 560 +    size_t total;
 561 +
 562 +    total = 0;
 563 +
 564 +    while (len > 0) {
 565 +        n = safe_write(fd, buf, len);
 566 +        if (n < 0) {
 567 +            return n;
 568 +        }
 569 +        buf += n;
 570 +        total += n;
 571 +        len -= n;
 572 +    }
 573 +
 574 +    if (len) {
 575 +        /* incomplete write ? */
 576 +        return -1;
 577 +    }
 578 +
 579 +    return total;
 580 +}
 581 +
 582 +static int restore_write_data(VmaReader *vmar, guint8 dev_id,
 583 +                              BlockDriverState *bs, int vmstate_fd,
 584 +                              unsigned char *buf, int64_t sector_num,
 585 +                              int nb_sectors, Error **errp)
 586 +{
 587 +    assert(vmar);
 588 +
 589 +    if (dev_id == vmar->vmstate_stream) {
 590 +        if (vmstate_fd >= 0) {
 591 +            int len = nb_sectors * BDRV_SECTOR_SIZE;
 592 +            int res = full_write(vmstate_fd, buf, len);
 593 +            if (res < 0) {
 594 +                error_setg(errp, "write vmstate failed %d", res);
 595 +                return -1;
 596 +            }
 597 +        }
 598 +    } else {
 599 +        int res = bdrv_write(bs, sector_num, buf, nb_sectors);
 600 +        if (res < 0) {
 601 +            error_setg(errp, "bdrv_write to %s failed (%d)",
 602 +                       bdrv_get_device_name(bs), res);
 603 +            return -1;
 604 +        }
 605 +    }
 606 +    return 0;
 607 +}
 608 +static int restore_extent(VmaReader *vmar, unsigned char *buf,
 609 +                          int extent_size, int vmstate_fd,
 610 +                          bool verbose, Error **errp)
 611 +{
 612 +    assert(vmar);
 613 +    assert(buf);
 614 +
 615 +    VmaExtentHeader *ehead = (VmaExtentHeader *)buf;
 616 +    int start = VMA_EXTENT_HEADER_SIZE;
 617 +    int i;
 618 +
 619 +    for (i = 0; i < VMA_BLOCKS_PER_EXTENT; i++) {
 620 +        uint64_t block_info = GUINT64_FROM_BE(ehead->blockinfo[i]);
 621 +        uint64_t cluster_num = block_info & 0xffffffff;
 622 +        uint8_t dev_id = (block_info >> 32) & 0xff;
 623 +        uint16_t mask = block_info >> (32+16);
 624 +        int64_t max_sector;
 625 +
 626 +        if (!dev_id) {
 627 +            continue;
 628 +        }
 629 +
 630 +        VmaRestoreState *rstate = &vmar->rstate[dev_id];
 631 +        BlockDriverState *bs = NULL;
 632 +
 633 +        if (dev_id != vmar->vmstate_stream) {
 634 +            bs = rstate->bs;
 635 +            if (!bs) {
 636 +                error_setg(errp, "got wrong dev id %d", dev_id);
 637 +                return -1;
 638 +            }
 639 +
 640 +            if (vma_reader_get_bitmap(rstate, cluster_num)) {
 641 +                error_setg(errp, "found duplicated cluster %zd for stream %s",
 642 +                          cluster_num, vmar->devinfo[dev_id].devname);
 643 +                return -1;
 644 +            }
 645 +            vma_reader_set_bitmap(rstate, cluster_num, 1);
 646 +
 647 +            max_sector = vmar->devinfo[dev_id].size/BDRV_SECTOR_SIZE;
 648 +        } else {
 649 +            max_sector = G_MAXINT64;
 650 +            if (cluster_num != vmar->vmstate_clusters) {
 651 +                error_setg(errp, "found out of order vmstate data");
 652 +                return -1;
 653 +            }
 654 +            vmar->vmstate_clusters++;
 655 +        }
 656 +
 657 +        vmar->clusters_read++;
 658 +
 659 +        if (verbose) {
 660 +            time_t duration = time(NULL) - vmar->start_time;
 661 +            int percent = (vmar->clusters_read*100)/vmar->cluster_count;
 662 +            if (percent != vmar->clusters_read_per) {
 663 +                printf("progress %d%% (read %zd bytes, duration %zd sec)\n",
 664 +                       percent, vmar->clusters_read*VMA_CLUSTER_SIZE,
 665 +                       duration);
 666 +                fflush(stdout);
 667 +                vmar->clusters_read_per = percent;
 668 +            }
 669 +        }
 670 +
 671 +        /* try to write whole clusters to speedup restore */
 672 +        if (mask == 0xffff) {
 673 +            if ((start + VMA_CLUSTER_SIZE) > extent_size) {
 674 +                error_setg(errp, "short vma extent - too many blocks");
 675 +                return -1;
 676 +            }
 677 +            int64_t sector_num = (cluster_num * VMA_CLUSTER_SIZE) /
 678 +                BDRV_SECTOR_SIZE;
 679 +            int64_t end_sector = sector_num +
 680 +                VMA_CLUSTER_SIZE/BDRV_SECTOR_SIZE;
 681 +
 682 +            if (end_sector > max_sector) {
 683 +                end_sector = max_sector;
 684 +            }
 685 +
 686 +            if (end_sector <= sector_num) {
 687 +                error_setg(errp, "got wrong block address - write bejond end");
 688 +                return -1;
 689 +            }
 690 +
 691 +            int nb_sectors = end_sector - sector_num;
 692 +            if (restore_write_data(vmar, dev_id, bs, vmstate_fd, buf + start,
 693 +                                   sector_num, nb_sectors, errp) < 0) {
 694 +                return -1;
 695 +            }
 696 +
 697 +            start += VMA_CLUSTER_SIZE;
 698 +        } else {
 699 +            int j;
 700 +            int bit = 1;
 701 +
 702 +            for (j = 0; j < 16; j++) {
 703 +                int64_t sector_num = (cluster_num*VMA_CLUSTER_SIZE +
 704 +                                      j*VMA_BLOCK_SIZE)/BDRV_SECTOR_SIZE;
 705 +
 706 +                int64_t end_sector = sector_num +
 707 +                    VMA_BLOCK_SIZE/BDRV_SECTOR_SIZE;
 708 +                if (end_sector > max_sector) {
 709 +                    end_sector = max_sector;
 710 +                }
 711 +
 712 +                if (mask & bit) {
 713 +                    if ((start + VMA_BLOCK_SIZE) > extent_size) {
 714 +                        error_setg(errp, "short vma extent - too many blocks");
 715 +                        return -1;
 716 +                    }
 717 +
 718 +                    if (end_sector <= sector_num) {
 719 +                        error_setg(errp, "got wrong block address - "
 720 +                                   "write bejond end");
 721 +                        return -1;
 722 +                    }
 723 +
 724 +                    int nb_sectors = end_sector - sector_num;
 725 +                    if (restore_write_data(vmar, dev_id, bs, vmstate_fd,
 726 +                                           buf + start, sector_num,
 727 +                                           nb_sectors, errp) < 0) {
 728 +                        return -1;
 729 +                    }
 730 +
 731 +                    start += VMA_BLOCK_SIZE;
 732 +
 733 +                } else {
 734 +
 735 +                    if (rstate->write_zeroes && (end_sector > sector_num)) {
 736 +                        /* Todo: use bdrv_co_write_zeroes (but that need to
 737 +                         * be run inside coroutine?)
 738 +                         */
 739 +                        int nb_sectors = end_sector - sector_num;
 740 +                        if (restore_write_data(vmar, dev_id, bs, vmstate_fd,
 741 +                                              zero_vma_block, sector_num,
 742 +                                               nb_sectors, errp) < 0) {
 743 +                            return -1;
 744 +                        }
 745 +                    }
 746 +                }
 747 +
 748 +                bit = bit << 1;
 749 +            }
 750 +        }
 751 +    }
 752 +
 753 +    if (start != extent_size) {
 754 +        error_setg(errp, "vma extent error - missing blocks");
 755 +        return -1;
 756 +    }
 757 +
 758 +    return 0;
 759 +}
 760 +
 761 +int vma_reader_restore(VmaReader *vmar, int vmstate_fd, bool verbose,
 762 +                       Error **errp)
 763 +{
 764 +    assert(vmar);
 765 +    assert(vmar->head_data);
 766 +
 767 +    int ret = 0;
 768 +    unsigned char buf[VMA_MAX_EXTENT_SIZE];
 769 +    int buf_pos = 0;
 770 +    unsigned char md5sum[16];
 771 +    VmaHeader *h = (VmaHeader *)vmar->head_data;
 772 +
 773 +    vmar->start_time = time(NULL);
 774 +
 775 +    while (1) {
 776 +        int bytes = full_read(vmar->fd, buf + buf_pos, sizeof(buf) - buf_pos);
 777 +        if (bytes < 0) {
 778 +            error_setg(errp, "read failed - %s", g_strerror(errno));
 779 +            return -1;
 780 +        }
 781 +
 782 +        buf_pos += bytes;
 783 +
 784 +        if (!buf_pos) {
 785 +            break; /* EOF */
 786 +        }
 787 +
 788 +        if (buf_pos < VMA_EXTENT_HEADER_SIZE) {
 789 +            error_setg(errp, "read short extent (%d bytes)", buf_pos);
 790 +            return -1;
 791 +        }
 792 +
 793 +        VmaExtentHeader *ehead = (VmaExtentHeader *)buf;
 794 +
 795 +        /* extract md5sum */
 796 +        memcpy(md5sum, ehead->md5sum, sizeof(ehead->md5sum));
 797 +        memset(ehead->md5sum, 0, sizeof(ehead->md5sum));
 798 +
 799 +        g_checksum_reset(vmar->md5csum);
 800 +        g_checksum_update(vmar->md5csum, buf, VMA_EXTENT_HEADER_SIZE);
 801 +        gsize csize = 16;
 802 +        g_checksum_get_digest(vmar->md5csum, ehead->md5sum, &csize);
 803 +
 804 +        if (memcmp(md5sum, ehead->md5sum, 16) != 0) {
 805 +            error_setg(errp, "wrong vma extent header chechsum");
 806 +            return -1;
 807 +        }
 808 +
 809 +        if (memcmp(h->uuid, ehead->uuid, sizeof(ehead->uuid)) != 0) {
 810 +            error_setg(errp, "wrong vma extent uuid");
 811 +            return -1;
 812 +        }
 813 +
 814 +        if (ehead->magic != VMA_EXTENT_MAGIC || ehead->reserved1 != 0) {
 815 +            error_setg(errp, "wrong vma extent header magic");
 816 +            return -1;
 817 +        }
 818 +
 819 +        int block_count = GUINT16_FROM_BE(ehead->block_count);
 820 +        int extent_size = VMA_EXTENT_HEADER_SIZE + block_count*VMA_BLOCK_SIZE;
 821 +
 822 +        if (buf_pos < extent_size) {
 823 +            error_setg(errp, "short vma extent (%d < %d)", buf_pos,
 824 +                       extent_size);
 825 +            return -1;
 826 +        }
 827 +
 828 +        if (restore_extent(vmar, buf, extent_size, vmstate_fd, verbose,
 829 +                           errp) < 0) {
 830 +            return -1;
 831 +        }
 832 +
 833 +        if (buf_pos > extent_size) {
 834 +            memmove(buf, buf + extent_size, buf_pos - extent_size);
 835 +            buf_pos = buf_pos - extent_size;
 836 +        } else {
 837 +            buf_pos = 0;
 838 +        }
 839 +    }
 840 +
 841 +    bdrv_drain_all();
 842 +
 843 +    int i;
 844 +    for (i = 1; i < 256; i++) {
 845 +        VmaRestoreState *rstate = &vmar->rstate[i];
 846 +        if (!rstate->bs) {
 847 +            continue;
 848 +        }
 849 +
 850 +        if (bdrv_flush(rstate->bs) < 0) {
 851 +            error_setg(errp, "vma bdrv_flush %s failed",
 852 +                       vmar->devinfo[i].devname);
 853 +            return -1;
 854 +        }
 855 +
 856 +        if (vmar->devinfo[i].size &&
 857 +            (strcmp(vmar->devinfo[i].devname, "vmstate") != 0)) {
 858 +            assert(rstate->bitmap);
 859 +
 860 +            int64_t cluster_num, end;
 861 +
 862 +            end = (vmar->devinfo[i].size + VMA_CLUSTER_SIZE - 1) /
 863 +                VMA_CLUSTER_SIZE;
 864 +
 865 +            for (cluster_num = 0; cluster_num < end; cluster_num++) {
 866 +                if (!vma_reader_get_bitmap(rstate, cluster_num)) {
 867 +                    error_setg(errp, "detected missing cluster %zd "
 868 +                               "for stream %s", cluster_num,
 869 +                               vmar->devinfo[i].devname);
 870 +                    return -1;
 871 +                }
 872 +            }
 873 +        }
 874 +    }
 875 +
 876 +    return ret;
 877 +}
 878 +
 879 Index: new/vma-writer.c
 880 ===================================================================
 881 --- /dev/null   1970-01-01 00:00:00.000000000 +0000
 882 +++ new/vma-writer.c    2013-11-28 11:03:52.000000000 +0100
 883 @@ -0,0 +1,869 @@
 884 +/*
 885 + * VMA: Virtual Machine Archive
 886 + *
 887 + * Copyright (C) 2012 Proxmox Server Solutions
 888 + *
 889 + * Authors:
 890 + *  Dietmar Maurer (dietmar@proxmox.com)
 891 + *
 892 + * This work is licensed under the terms of the GNU GPL, version 2 or later.
 893 + * See the COPYING file in the top-level directory.
 894 + *
 895 + */
 896 +
 897 +#include <stdio.h>
 898 +#include <errno.h>
 899 +#include <unistd.h>
 900 +#include <stdio.h>
 901 +#include <string.h>
 902 +#include <sys/types.h>
 903 +#include <sys/stat.h>
 904 +#include <fcntl.h>
 905 +#include <glib.h>
 906 +#include <uuid/uuid.h>
 907 +
 908 +#include "qemu-common.h"
 909 +#include "vma.h"
 910 +#include "block/block.h"
 911 +#include "monitor/monitor.h"
 912 +
 913 +#define DEBUG_VMA 0
 914 +
 915 +#define DPRINTF(fmt, ...)\
 916 +    do { if (DEBUG_VMA) { printf("vma: " fmt, ## __VA_ARGS__); } } while (0)
 917 +
 918 +#define WRITE_BUFFERS 5
 919 +
 920 +typedef struct VmaAIOCB VmaAIOCB;
 921 +struct VmaAIOCB {
 922 +    unsigned char buffer[VMA_MAX_EXTENT_SIZE];
 923 +    VmaWriter *vmaw;
 924 +    size_t bytes;
 925 +    Coroutine *co;
 926 +};
 927 +
 928 +struct VmaWriter {
 929 +    int fd;
 930 +    FILE *cmd;
 931 +    int status;
 932 +    char errmsg[8192];
 933 +    uuid_t uuid;
 934 +    bool header_written;
 935 +    bool closed;
 936 +
 937 +    /* we always write extents */
 938 +    unsigned char outbuf[VMA_MAX_EXTENT_SIZE];
 939 +    int outbuf_pos; /* in bytes */
 940 +    int outbuf_count; /* in VMA_BLOCKS */
 941 +    uint64_t outbuf_block_info[VMA_BLOCKS_PER_EXTENT];
 942 +
 943 +    VmaAIOCB *aiocbs[WRITE_BUFFERS];
 944 +    CoQueue wqueue;
 945 +
 946 +    GChecksum *md5csum;
 947 +    CoMutex writer_lock;
 948 +    CoMutex flush_lock;
 949 +    Coroutine *co_writer;
 950 +
 951 +    /* drive informations */
 952 +    VmaStreamInfo stream_info[256];
 953 +    guint stream_count;
 954 +
 955 +    guint8 vmstate_stream;
 956 +    uint32_t vmstate_clusters;
 957 +
 958 +    /* header blob table */
 959 +    char *header_blob_table;
 960 +    uint32_t header_blob_table_size;
 961 +    uint32_t header_blob_table_pos;
 962 +
 963 +    /* store for config blobs */
 964 +    uint32_t config_names[VMA_MAX_CONFIGS]; /* offset into blob_buffer table */
 965 +    uint32_t config_data[VMA_MAX_CONFIGS];  /* offset into blob_buffer table */
 966 +    uint32_t config_count;
 967 +};
 968 +
 969 +void vma_writer_set_error(VmaWriter *vmaw, const char *fmt, ...)
 970 +{
 971 +    va_list ap;
 972 +
 973 +    if (vmaw->status < 0) {
 974 +        return;
 975 +    }
 976 +
 977 +    vmaw->status = -1;
 978 +
 979 +    va_start(ap, fmt);
 980 +    g_vsnprintf(vmaw->errmsg, sizeof(vmaw->errmsg), fmt, ap);
 981 +    va_end(ap);
 982 +
 983 +    DPRINTF("vma_writer_set_error: %s\n", vmaw->errmsg);
 984 +}
 985 +
 986 +static uint32_t allocate_header_blob(VmaWriter *vmaw, const char *data,
 987 +                                     size_t len)
 988 +{
 989 +    if (len > 65535) {
 990 +        return 0;
 991 +    }
 992 +
 993 +    if (!vmaw->header_blob_table ||
 994 +        (vmaw->header_blob_table_size <
 995 +         (vmaw->header_blob_table_pos + len + 2))) {
 996 +        int newsize = vmaw->header_blob_table_size + ((len + 2 + 511)/512)*512;
 997 +
 998 +        vmaw->header_blob_table = g_realloc(vmaw->header_blob_table, newsize);
 999 +        memset(vmaw->header_blob_table + vmaw->header_blob_table_size,
1000 +               0, newsize - vmaw->header_blob_table_size);
1001 +        vmaw->header_blob_table_size = newsize;
1002 +    }
1003 +
1004 +    uint32_t cpos = vmaw->header_blob_table_pos;
1005 +    vmaw->header_blob_table[cpos] = len & 255;
1006 +    vmaw->header_blob_table[cpos+1] = (len >> 8) & 255;
1007 +    memcpy(vmaw->header_blob_table + cpos + 2, data, len);
1008 +    vmaw->header_blob_table_pos += len + 2;
1009 +    return cpos;
1010 +}
1011 +
1012 +static uint32_t allocate_header_string(VmaWriter *vmaw, const char *str)
1013 +{
1014 +    assert(vmaw);
1015 +
1016 +    size_t len = strlen(str) + 1;
1017 +
1018 +    return allocate_header_blob(vmaw, str, len);
1019 +}
1020 +
1021 +int vma_writer_add_config(VmaWriter *vmaw, const char *name, gpointer data,
1022 +                          gsize len)
1023 +{
1024 +    assert(vmaw);
1025 +    assert(!vmaw->header_written);
1026 +    assert(vmaw->config_count < VMA_MAX_CONFIGS);
1027 +    assert(name);
1028 +    assert(data);
1029 +    assert(len);
1030 +
1031 +    uint32_t name_ptr = allocate_header_string(vmaw, name);
1032 +    if (!name_ptr) {
1033 +        return -1;
1034 +    }
1035 +
1036 +    uint32_t data_ptr = allocate_header_blob(vmaw, data, len);
1037 +    if (!data_ptr) {
1038 +        return -1;
1039 +    }
1040 +
1041 +    vmaw->config_names[vmaw->config_count] = name_ptr;
1042 +    vmaw->config_data[vmaw->config_count] = data_ptr;
1043 +
1044 +    vmaw->config_count++;
1045 +
1046 +    return 0;
1047 +}
1048 +
1049 +int vma_writer_register_stream(VmaWriter *vmaw, const char *devname,
1050 +                               size_t size)
1051 +{
1052 +    assert(vmaw);
1053 +    assert(devname);
1054 +    assert(!vmaw->status);
1055 +
1056 +    if (vmaw->header_written) {
1057 +        vma_writer_set_error(vmaw, "vma_writer_register_stream: header "
1058 +                             "already written");
1059 +        return -1;
1060 +    }
1061 +
1062 +    guint n = vmaw->stream_count + 1;
1063 +
1064 +    /* we can have dev_ids form 1 to 255 (0 reserved)
1065 +     * 255(-1) reseverd for safety
1066 +     */
1067 +    if (n > 254) {
1068 +        vma_writer_set_error(vmaw, "vma_writer_register_stream: "
1069 +                             "too many drives");
1070 +        return -1;
1071 +    }
1072 +
1073 +    if (size <= 0) {
1074 +        vma_writer_set_error(vmaw, "vma_writer_register_stream: "
1075 +                             "got strange size %zd", size);
1076 +        return -1;
1077 +    }
1078 +
1079 +    DPRINTF("vma_writer_register_stream %s %zu %d\n", devname, size, n);
1080 +
1081 +    vmaw->stream_info[n].devname = g_strdup(devname);
1082 +    vmaw->stream_info[n].size = size;
1083 +
1084 +    vmaw->stream_info[n].cluster_count = (size + VMA_CLUSTER_SIZE - 1) /
1085 +        VMA_CLUSTER_SIZE;
1086 +
1087 +    vmaw->stream_count = n;
1088 +
1089 +    if (strcmp(devname, "vmstate") == 0) {
1090 +        vmaw->vmstate_stream = n;
1091 +    }
1092 +
1093 +    return n;
1094 +}
1095 +
1096 +static void vma_co_continue_write(void *opaque)
1097 +{
1098 +    VmaWriter *vmaw = opaque;
1099 +
1100 +    DPRINTF("vma_co_continue_write\n");
1101 +    qemu_coroutine_enter(vmaw->co_writer, NULL);
1102 +}
1103 +
1104 +static ssize_t coroutine_fn
1105 +vma_co_write(VmaWriter *vmaw, const void *buf, size_t bytes)
1106 +{
1107 +    size_t done = 0;
1108 +    ssize_t ret;
1109 +
1110 +    /* atomic writes (we cannot interleave writes) */
1111 +    qemu_co_mutex_lock(&vmaw->writer_lock);
1112 +
1113 +    DPRINTF("vma_co_write enter %zd\n", bytes);
1114 +
1115 +    assert(vmaw->co_writer == NULL);
1116 +
1117 +    vmaw->co_writer = qemu_coroutine_self();
1118 +
1119 +    qemu_aio_set_fd_handler(vmaw->fd, NULL, vma_co_continue_write, vmaw);
1120 +
1121 +    DPRINTF("vma_co_write wait until writable\n");
1122 +    qemu_coroutine_yield();
1123 +    DPRINTF("vma_co_write starting %zd\n", bytes);
1124 +
1125 +    while (done < bytes) {
1126 +        ret = write(vmaw->fd, buf + done, bytes - done);
1127 +        if (ret > 0) {
1128 +            done += ret;
1129 +            DPRINTF("vma_co_write written %zd %zd\n", done, ret);
1130 +        } else if (ret < 0) {
1131 +            if (errno == EAGAIN || errno == EWOULDBLOCK) {
1132 +                DPRINTF("vma_co_write yield %zd\n", done);
1133 +                qemu_coroutine_yield();
1134 +                DPRINTF("vma_co_write restart %zd\n", done);
1135 +            } else {
1136 +                vma_writer_set_error(vmaw, "vma_co_write write error - %s",
1137 +                                     g_strerror(errno));
1138 +                done = -1; /* always return failure for partial writes */
1139 +                break;
1140 +            }
1141 +        } else if (ret == 0) {
1142 +            /* should not happen - simply try again */
1143 +        }
1144 +    }
1145 +
1146 +    qemu_aio_set_fd_handler(vmaw->fd, NULL, NULL, NULL);
1147 +
1148 +    vmaw->co_writer = NULL;
1149 +
1150 +    qemu_co_mutex_unlock(&vmaw->writer_lock);
1151 +
1152 +    DPRINTF("vma_co_write leave %zd\n", done);
1153 +    return done;
1154 +}
1155 +
1156 +static void coroutine_fn vma_co_writer_task(void *opaque)
1157 +{
1158 +    VmaAIOCB *cb = opaque;
1159 +
1160 +    DPRINTF("vma_co_writer_task start\n");
1161 +
1162 +    int64_t done = vma_co_write(cb->vmaw, cb->buffer, cb->bytes);
1163 +    DPRINTF("vma_co_writer_task write done %zd\n", done);
1164 +
1165 +    if (done != cb->bytes) {
1166 +        DPRINTF("vma_co_writer_task failed write %zd %zd", cb->bytes, done);
1167 +        vma_writer_set_error(cb->vmaw, "vma_co_writer_task failed write %zd",
1168 +                             done);
1169 +    }
1170 +
1171 +    cb->bytes = 0;
1172 +
1173 +    qemu_co_queue_next(&cb->vmaw->wqueue);
1174 +
1175 +    DPRINTF("vma_co_writer_task end\n");
1176 +}
1177 +
1178 +static void coroutine_fn vma_queue_flush(VmaWriter *vmaw)
1179 +{
1180 +    DPRINTF("vma_queue_flush enter\n");
1181 +
1182 +    assert(vmaw);
1183 +
1184 +    while (1) {
1185 +        int i;
1186 +        VmaAIOCB *cb = NULL;
1187 +        for (i = 0; i < WRITE_BUFFERS; i++) {
1188 +            if (vmaw->aiocbs[i]->bytes) {
1189 +                cb = vmaw->aiocbs[i];
1190 +                DPRINTF("FOUND USED AIO BUFFER %d %zd\n", i,
1191 +                        vmaw->aiocbs[i]->bytes);
1192 +                break;
1193 +            }
1194 +        }
1195 +        if (!cb) {
1196 +            break;
1197 +        }
1198 +        qemu_co_queue_wait(&vmaw->wqueue);
1199 +    }
1200 +
1201 +    DPRINTF("vma_queue_flush leave\n");
1202 +}
1203 +
1204 +/**
1205 + * NOTE: pipe buffer size in only 4096 bytes on linux (see 'ulimit -a')
1206 + * So we need to create a coroutione to allow 'parallel' execution.
1207 + */
1208 +static ssize_t coroutine_fn
1209 +vma_queue_write(VmaWriter *vmaw, const void *buf, size_t bytes)
1210 +{
1211 +    DPRINTF("vma_queue_write enter %zd\n", bytes);
1212 +
1213 +    assert(vmaw);
1214 +    assert(buf);
1215 +    assert(bytes <= VMA_MAX_EXTENT_SIZE);
1216 +
1217 +    VmaAIOCB *cb = NULL;
1218 +    while (!cb) {
1219 +        int i;
1220 +        for (i = 0; i < WRITE_BUFFERS; i++) {
1221 +            if (!vmaw->aiocbs[i]->bytes) {
1222 +                cb = vmaw->aiocbs[i];
1223 +                break;
1224 +            }
1225 +        }
1226 +        if (!cb) {
1227 +            qemu_co_queue_wait(&vmaw->wqueue);
1228 +        }
1229 +    }
1230 +
1231 +    memcpy(cb->buffer, buf, bytes);
1232 +    cb->bytes = bytes;
1233 +    cb->vmaw = vmaw;
1234 +
1235 +    DPRINTF("vma_queue_write start %zd\n", bytes);
1236 +    cb->co = qemu_coroutine_create(vma_co_writer_task);
1237 +    qemu_coroutine_enter(cb->co, cb);
1238 +
1239 +    DPRINTF("vma_queue_write leave\n");
1240 +
1241 +    return bytes;
1242 +}
1243 +
1244 +VmaWriter *vma_writer_create(const char *filename, uuid_t uuid, Error **errp)
1245 +{
1246 +    const char *p;
1247 +
1248 +    assert(sizeof(VmaHeader) == (4096 + 8192));
1249 +    assert(sizeof(VmaExtentHeader) == 512);
1250 +
1251 +    VmaWriter *vmaw = g_new0(VmaWriter, 1);
1252 +    vmaw->fd = -1;
1253 +
1254 +    vmaw->md5csum = g_checksum_new(G_CHECKSUM_MD5);
1255 +    if (!vmaw->md5csum) {
1256 +        error_setg(errp, "can't allocate cmsum\n");
1257 +        goto err;
1258 +    }
1259 +
1260 +    if (strstart(filename, "exec:", &p)) {
1261 +        vmaw->cmd = popen(p, "w");
1262 +        if (vmaw->cmd == NULL) {
1263 +            error_setg(errp, "can't popen command '%s' - %s\n", p,
1264 +                       g_strerror(errno));
1265 +            goto err;
1266 +        }
1267 +        vmaw->fd = fileno(vmaw->cmd);
1268 +
1269 +        /* try to use O_NONBLOCK and O_DIRECT */
1270 +        fcntl(vmaw->fd, F_SETFL, fcntl(vmaw->fd, F_GETFL)|O_NONBLOCK);
1271 +        fcntl(vmaw->fd, F_SETFL, fcntl(vmaw->fd, F_GETFL)|O_DIRECT);
1272 +
1273 +    } else {
1274 +        struct stat st;
1275 +        int oflags;
1276 +        const char *tmp_id_str;
1277 +
1278 +        if ((stat(filename, &st) == 0) && S_ISFIFO(st.st_mode)) {
1279 +            oflags = O_NONBLOCK|O_DIRECT|O_WRONLY;
1280 +            vmaw->fd = qemu_open(filename, oflags, 0644);
1281 +        } else if (strstart(filename, "/dev/fdset/", &tmp_id_str)) {
1282 +            oflags = O_NONBLOCK|O_DIRECT|O_WRONLY;
1283 +            vmaw->fd = qemu_open(filename, oflags, 0644);
1284 +        } else if (strstart(filename, "/dev/fdname/", &tmp_id_str)) {
1285 +            vmaw->fd = monitor_get_fd(cur_mon, tmp_id_str, errp);
1286 +            if (vmaw->fd < 0) {
1287 +                goto err;
1288 +            }
1289 +            /* try to use O_NONBLOCK and O_DIRECT */
1290 +            fcntl(vmaw->fd, F_SETFL, fcntl(vmaw->fd, F_GETFL)|O_NONBLOCK);
1291 +            fcntl(vmaw->fd, F_SETFL, fcntl(vmaw->fd, F_GETFL)|O_DIRECT);
1292 +        } else  {
1293 +            oflags = O_NONBLOCK|O_DIRECT|O_WRONLY|O_CREAT|O_EXCL;
1294 +            vmaw->fd = qemu_open(filename, oflags, 0644);
1295 +        }
1296 +
1297 +        if (vmaw->fd < 0) {
1298 +            error_setg(errp, "can't open file %s - %s\n", filename,
1299 +                       g_strerror(errno));
1300 +            goto err;
1301 +        }
1302 +    }
1303 +
1304 +    /* we use O_DIRECT, so we need to align IO buffers */
1305 +    int i;
1306 +    for (i = 0; i < WRITE_BUFFERS; i++) {
1307 +        vmaw->aiocbs[i] = qemu_memalign(512, sizeof(VmaAIOCB));
1308 +        memset(vmaw->aiocbs[i], 0, sizeof(VmaAIOCB));
1309 +    }
1310 +
1311 +    vmaw->outbuf_count = 0;
1312 +    vmaw->outbuf_pos = VMA_EXTENT_HEADER_SIZE;
1313 +
1314 +    vmaw->header_blob_table_pos = 1; /* start at pos 1 */
1315 +
1316 +    qemu_co_mutex_init(&vmaw->writer_lock);
1317 +    qemu_co_mutex_init(&vmaw->flush_lock);
1318 +    qemu_co_queue_init(&vmaw->wqueue);
1319 +
1320 +    uuid_copy(vmaw->uuid, uuid);
1321 +
1322 +    return vmaw;
1323 +
1324 +err:
1325 +    if (vmaw) {
1326 +        if (vmaw->cmd) {
1327 +            pclose(vmaw->cmd);
1328 +        } else if (vmaw->fd >= 0) {
1329 +            close(vmaw->fd);
1330 +        }
1331 +
1332 +        if (vmaw->md5csum) {
1333 +            g_checksum_free(vmaw->md5csum);
1334 +        }
1335 +
1336 +        g_free(vmaw);
1337 +    }
1338 +
1339 +    return NULL;
1340 +}
1341 +
1342 +static int coroutine_fn vma_write_header(VmaWriter *vmaw)
1343 +{
1344 +    assert(vmaw);
1345 +    int header_clusters = 8;
1346 +    char buf[65536*header_clusters];
1347 +    VmaHeader *head = (VmaHeader *)buf;
1348 +
1349 +    int i;
1350 +
1351 +    DPRINTF("VMA WRITE HEADER\n");
1352 +
1353 +    if (vmaw->status < 0) {
1354 +        return vmaw->status;
1355 +    }
1356 +
1357 +    memset(buf, 0, sizeof(buf));
1358 +
1359 +    head->magic = VMA_MAGIC;
1360 +    head->version = GUINT32_TO_BE(1); /* v1 */
1361 +    memcpy(head->uuid, vmaw->uuid, 16);
1362 +
1363 +    time_t ctime = time(NULL);
1364 +    head->ctime = GUINT64_TO_BE(ctime);
1365 +
1366 +    if (!vmaw->stream_count) {
1367 +        return -1;
1368 +    }
1369 +
1370 +    for (i = 0; i < VMA_MAX_CONFIGS; i++) {
1371 +        head->config_names[i] = GUINT32_TO_BE(vmaw->config_names[i]);
1372 +        head->config_data[i] = GUINT32_TO_BE(vmaw->config_data[i]);
1373 +    }
1374 +
1375 +    /* 32 bytes per device (12 used currently) = 8192 bytes max */
1376 +    for (i = 1; i <= 254; i++) {
1377 +        VmaStreamInfo *si = &vmaw->stream_info[i];
1378 +        if (si->size) {
1379 +            assert(si->devname);
1380 +            uint32_t devname_ptr = allocate_header_string(vmaw, si->devname);
1381 +            if (!devname_ptr) {
1382 +                return -1;
1383 +            }
1384 +            head->dev_info[i].devname_ptr = GUINT32_TO_BE(devname_ptr);
1385 +            head->dev_info[i].size = GUINT64_TO_BE(si->size);
1386 +        }
1387 +    }
1388 +
1389 +    uint32_t header_size = sizeof(VmaHeader) + vmaw->header_blob_table_size;
1390 +    head->header_size = GUINT32_TO_BE(header_size);
1391 +
1392 +    if (header_size > sizeof(buf)) {
1393 +        return -1; /* just to be sure */
1394 +    }
1395 +
1396 +    uint32_t blob_buffer_offset = sizeof(VmaHeader);
1397 +    memcpy(buf + blob_buffer_offset, vmaw->header_blob_table,
1398 +           vmaw->header_blob_table_size);
1399 +    head->blob_buffer_offset = GUINT32_TO_BE(blob_buffer_offset);
1400 +    head->blob_buffer_size = GUINT32_TO_BE(vmaw->header_blob_table_pos);
1401 +
1402 +    g_checksum_reset(vmaw->md5csum);
1403 +    g_checksum_update(vmaw->md5csum, (const guchar *)buf, header_size);
1404 +    gsize csize = 16;
1405 +    g_checksum_get_digest(vmaw->md5csum, (guint8 *)(head->md5sum), &csize);
1406 +
1407 +    return vma_queue_write(vmaw, buf, header_size);
1408 +}
1409 +
1410 +static int coroutine_fn vma_writer_flush(VmaWriter *vmaw)
1411 +{
1412 +    assert(vmaw);
1413 +
1414 +    int ret;
1415 +    int i;
1416 +
1417 +    if (vmaw->status < 0) {
1418 +        return vmaw->status;
1419 +    }
1420 +
1421 +    if (!vmaw->header_written) {
1422 +        vmaw->header_written = true;
1423 +        ret = vma_write_header(vmaw);
1424 +        if (ret < 0) {
1425 +            vma_writer_set_error(vmaw, "vma_writer_flush: write header failed");
1426 +            return ret;
1427 +        }
1428 +    }
1429 +
1430 +    DPRINTF("VMA WRITE FLUSH %d %d\n", vmaw->outbuf_count, vmaw->outbuf_pos);
1431 +
1432 +
1433 +    VmaExtentHeader *ehead = (VmaExtentHeader *)vmaw->outbuf;
1434 +
1435 +    ehead->magic = VMA_EXTENT_MAGIC;
1436 +    ehead->reserved1 = 0;
1437 +
1438 +    for (i = 0; i < VMA_BLOCKS_PER_EXTENT; i++) {
1439 +        ehead->blockinfo[i] = GUINT64_TO_BE(vmaw->outbuf_block_info[i]);
1440 +    }
1441 +
1442 +    guint16 block_count = (vmaw->outbuf_pos - VMA_EXTENT_HEADER_SIZE) /
1443 +        VMA_BLOCK_SIZE;
1444 +
1445 +    ehead->block_count = GUINT16_TO_BE(block_count);
1446 +
1447 +    memcpy(ehead->uuid, vmaw->uuid, sizeof(ehead->uuid));
1448 +    memset(ehead->md5sum, 0, sizeof(ehead->md5sum));
1449 +
1450 +    g_checksum_reset(vmaw->md5csum);
1451 +    g_checksum_update(vmaw->md5csum, vmaw->outbuf, VMA_EXTENT_HEADER_SIZE);
1452 +    gsize csize = 16;
1453 +    g_checksum_get_digest(vmaw->md5csum, ehead->md5sum, &csize);
1454 +
1455 +    int bytes = vmaw->outbuf_pos;
1456 +    ret = vma_queue_write(vmaw, vmaw->outbuf, bytes);
1457 +    if (ret != bytes) {
1458 +        vma_writer_set_error(vmaw, "vma_writer_flush: failed write");
1459 +    }
1460 +
1461 +    vmaw->outbuf_count = 0;
1462 +    vmaw->outbuf_pos = VMA_EXTENT_HEADER_SIZE;
1463 +
1464 +    for (i = 0; i < VMA_BLOCKS_PER_EXTENT; i++) {
1465 +        vmaw->outbuf_block_info[i] = 0;
1466 +    }
1467 +
1468 +    return vmaw->status;
1469 +}
1470 +
1471 +static int vma_count_open_streams(VmaWriter *vmaw)
1472 +{
1473 +    g_assert(vmaw != NULL);
1474 +
1475 +    int i;
1476 +    int open_drives = 0;
1477 +    for (i = 0; i <= 255; i++) {
1478 +        if (vmaw->stream_info[i].size && !vmaw->stream_info[i].finished) {
1479 +            open_drives++;
1480 +        }
1481 +    }
1482 +
1483 +    return open_drives;
1484 +}
1485 +
1486 +/**
1487 + * all jobs should call this when there is no more data
1488 + * Returns: number of remaining stream (0 ==> finished)
1489 + */
1490 +int coroutine_fn
1491 +vma_writer_close_stream(VmaWriter *vmaw, uint8_t dev_id)
1492 +{
1493 +    g_assert(vmaw != NULL);
1494 +
1495 +    DPRINTF("vma_writer_set_status %d\n", dev_id);
1496 +    if (!vmaw->stream_info[dev_id].size) {
1497 +        vma_writer_set_error(vmaw, "vma_writer_close_stream: "
1498 +                             "no such stream %d", dev_id);
1499 +        return -1;
1500 +    }
1501 +    if (vmaw->stream_info[dev_id].finished) {
1502 +        vma_writer_set_error(vmaw, "vma_writer_close_stream: "
1503 +                             "stream already closed %d", dev_id);
1504 +        return -1;
1505 +    }
1506 +
1507 +    vmaw->stream_info[dev_id].finished = true;
1508 +
1509 +    int open_drives = vma_count_open_streams(vmaw);
1510 +
1511 +    if (open_drives <= 0) {
1512 +        DPRINTF("vma_writer_set_status all drives completed\n");
1513 +        qemu_co_mutex_lock(&vmaw->flush_lock);
1514 +        int ret = vma_writer_flush(vmaw);
1515 +        qemu_co_mutex_unlock(&vmaw->flush_lock);
1516 +        if (ret < 0) {
1517 +            vma_writer_set_error(vmaw, "vma_writer_close_stream: flush failed");
1518 +        }
1519 +    }
1520 +
1521 +    return open_drives;
1522 +}
1523 +
1524 +int vma_writer_get_status(VmaWriter *vmaw, VmaStatus *status)
1525 +{
1526 +    int i;
1527 +
1528 +    g_assert(vmaw != NULL);
1529 +
1530 +    if (status) {
1531 +        status->status = vmaw->status;
1532 +        g_strlcpy(status->errmsg, vmaw->errmsg, sizeof(status->errmsg));
1533 +        for (i = 0; i <= 255; i++) {
1534 +            status->stream_info[i] = vmaw->stream_info[i];
1535 +        }
1536 +
1537 +        uuid_unparse_lower(vmaw->uuid, status->uuid_str);
1538 +    }
1539 +
1540 +    status->closed = vmaw->closed;
1541 +
1542 +    return vmaw->status;
1543 +}
1544 +
1545 +static int vma_writer_get_buffer(VmaWriter *vmaw)
1546 +{
1547 +    int ret = 0;
1548 +
1549 +    qemu_co_mutex_lock(&vmaw->flush_lock);
1550 +
1551 +    /* wait until buffer is available */
1552 +    while (vmaw->outbuf_count >= (VMA_BLOCKS_PER_EXTENT - 1)) {
1553 +        ret = vma_writer_flush(vmaw);
1554 +        if (ret < 0) {
1555 +            vma_writer_set_error(vmaw, "vma_writer_get_buffer: flush failed");
1556 +            break;
1557 +        }
1558 +    }
1559 +
1560 +    qemu_co_mutex_unlock(&vmaw->flush_lock);
1561 +
1562 +    return ret;
1563 +}
1564 +
1565 +
1566 +int64_t coroutine_fn
1567 +vma_writer_write(VmaWriter *vmaw, uint8_t dev_id, int64_t cluster_num,
1568 +                 unsigned char *buf, size_t *zero_bytes)
1569 +{
1570 +    g_assert(vmaw != NULL);
1571 +    g_assert(zero_bytes != NULL);
1572 +
1573 +    *zero_bytes = 0;
1574 +
1575 +    if (vmaw->status < 0) {
1576 +        return vmaw->status;
1577 +    }
1578 +
1579 +    if (!dev_id || !vmaw->stream_info[dev_id].size) {
1580 +        vma_writer_set_error(vmaw, "vma_writer_write: "
1581 +                             "no such stream %d", dev_id);
1582 +        return -1;
1583 +    }
1584 +
1585 +    if (vmaw->stream_info[dev_id].finished) {
1586 +        vma_writer_set_error(vmaw, "vma_writer_write: "
1587 +                             "stream already closed %d", dev_id);
1588 +        return -1;
1589 +    }
1590 +
1591 +
1592 +    if (cluster_num >= (((uint64_t)1)<<32)) {
1593 +        vma_writer_set_error(vmaw, "vma_writer_write: "
1594 +                             "cluster number out of range");
1595 +        return -1;
1596 +    }
1597 +
1598 +    if (dev_id == vmaw->vmstate_stream) {
1599 +        if (cluster_num != vmaw->vmstate_clusters) {
1600 +            vma_writer_set_error(vmaw, "vma_writer_write: "
1601 +                                 "non sequential vmstate write");
1602 +        }
1603 +        vmaw->vmstate_clusters++;
1604 +    } else if (cluster_num >= vmaw->stream_info[dev_id].cluster_count) {
1605 +        vma_writer_set_error(vmaw, "vma_writer_write: cluster number too big");
1606 +        return -1;
1607 +    }
1608 +
1609 +    /* wait until buffer is available */
1610 +    if (vma_writer_get_buffer(vmaw) < 0) {
1611 +        vma_writer_set_error(vmaw, "vma_writer_write: "
1612 +                             "vma_writer_get_buffer failed");
1613 +        return -1;
1614 +    }
1615 +
1616 +    DPRINTF("VMA WRITE %d %zd\n", dev_id, cluster_num);
1617 +
1618 +    uint16_t mask = 0;
1619 +
1620 +    if (buf) {
1621 +        int i;
1622 +        int bit = 1;
1623 +        for (i = 0; i < 16; i++) {
1624 +            unsigned char *vmablock = buf + (i*VMA_BLOCK_SIZE);
1625 +            if (!buffer_is_zero(vmablock, VMA_BLOCK_SIZE)) {
1626 +                mask |= bit;
1627 +                memcpy(vmaw->outbuf + vmaw->outbuf_pos, vmablock,
1628 +                       VMA_BLOCK_SIZE);
1629 +                vmaw->outbuf_pos += VMA_BLOCK_SIZE;
1630 +            } else {
1631 +                DPRINTF("VMA WRITE %zd ZERO BLOCK %d\n", cluster_num, i);
1632 +                vmaw->stream_info[dev_id].zero_bytes += VMA_BLOCK_SIZE;
1633 +                *zero_bytes += VMA_BLOCK_SIZE;
1634 +            }
1635 +
1636 +            bit = bit << 1;
1637 +        }
1638 +    } else {
1639 +        DPRINTF("VMA WRITE %zd ZERO CLUSTER\n", cluster_num);
1640 +        vmaw->stream_info[dev_id].zero_bytes += VMA_CLUSTER_SIZE;
1641 +        *zero_bytes += VMA_CLUSTER_SIZE;
1642 +    }
1643 +
1644 +    uint64_t block_info = ((uint64_t)mask) << (32+16);
1645 +    block_info |= ((uint64_t)dev_id) << 32;
1646 +    block_info |= (cluster_num & 0xffffffff);
1647 +    vmaw->outbuf_block_info[vmaw->outbuf_count] = block_info;
1648 +
1649 +    DPRINTF("VMA WRITE MASK %zd %zx\n", cluster_num, block_info);
1650 +
1651 +    vmaw->outbuf_count++;
1652 +
1653 +    /** NOTE: We allways write whole clusters, but we correctly set
1654 +     * transferred bytes. So transferred == size when when everything
1655 +     * went OK.
1656 +     */
1657 +    size_t transferred = VMA_CLUSTER_SIZE;
1658 +
1659 +    if (dev_id != vmaw->vmstate_stream) {
1660 +        uint64_t last = (cluster_num + 1) * VMA_CLUSTER_SIZE;
1661 +        if (last > vmaw->stream_info[dev_id].size) {
1662 +            uint64_t diff = last - vmaw->stream_info[dev_id].size;
1663 +            if (diff >= VMA_CLUSTER_SIZE) {
1664 +                vma_writer_set_error(vmaw, "vma_writer_write: "
1665 +                                     "read after last cluster");
1666 +                return -1;
1667 +            }
1668 +            transferred -= diff;
1669 +        }
1670 +    }
1671 +
1672 +    vmaw->stream_info[dev_id].transferred += transferred;
1673 +
1674 +    return transferred;
1675 +}
1676 +
1677 +int vma_writer_close(VmaWriter *vmaw, Error **errp)
1678 +{
1679 +    g_assert(vmaw != NULL);
1680 +
1681 +    int i;
1682 +
1683 +    vma_queue_flush(vmaw);
1684 +
1685 +    /* this should not happen - just to be sure */
1686 +    while (!qemu_co_queue_empty(&vmaw->wqueue)) {
1687 +        DPRINTF("vma_writer_close wait\n");
1688 +        co_sleep_ns(QEMU_CLOCK_REALTIME, 1000000);
1689 +    }
1690 +
1691 +    if (vmaw->cmd) {
1692 +        if (pclose(vmaw->cmd) < 0) {
1693 +            vma_writer_set_error(vmaw, "vma_writer_close: "
1694 +                                 "pclose failed - %s", g_strerror(errno));
1695 +        }
1696 +    } else {
1697 +        if (close(vmaw->fd) < 0) {
1698 +            vma_writer_set_error(vmaw, "vma_writer_close: "
1699 +                                 "close failed - %s", g_strerror(errno));
1700 +        }
1701 +    }
1702 +
1703 +    for (i = 0; i <= 255; i++) {
1704 +        VmaStreamInfo *si = &vmaw->stream_info[i];
1705 +        if (si->size) {
1706 +            if (!si->finished) {
1707 +                vma_writer_set_error(vmaw, "vma_writer_close: "
1708 +                                     "detected open stream '%s'", si->devname);
1709 +            } else if ((si->transferred != si->size) &&
1710 +                       (i != vmaw->vmstate_stream)) {
1711 +                vma_writer_set_error(vmaw, "vma_writer_close: "
1712 +                                     "incomplete stream '%s' (%zd != %zd)",
1713 +                                     si->devname, si->transferred, si->size);
1714 +            }
1715 +        }
1716 +    }
1717 +
1718 +    for (i = 0; i <= 255; i++) {
1719 +        vmaw->stream_info[i].finished = 1; /* mark as closed */
1720 +    }
1721 +
1722 +    vmaw->closed = 1;
1723 +
1724 +    if (vmaw->status < 0 && *errp == NULL) {
1725 +        error_setg(errp, "%s", vmaw->errmsg);
1726 +    }
1727 +
1728 +    return vmaw->status;
1729 +}
1730 +
1731 +void vma_writer_destroy(VmaWriter *vmaw)
1732 +{
1733 +    assert(vmaw);
1734 +
1735 +    int i;
1736 +
1737 +    for (i = 0; i <= 255; i++) {
1738 +        if (vmaw->stream_info[i].devname) {
1739 +            g_free(vmaw->stream_info[i].devname);
1740 +        }
1741 +    }
1742 +
1743 +    if (vmaw->md5csum) {
1744 +        g_checksum_free(vmaw->md5csum);
1745 +    }
1746 +
1747 +    for (i = 0; i < WRITE_BUFFERS; i++) {
1748 +        free(vmaw->aiocbs[i]);
1749 +    }
1750 +
1751 +    g_free(vmaw);
1752 +}
1753 Index: new/vma.c
1754 ===================================================================
1755 --- /dev/null   1970-01-01 00:00:00.000000000 +0000
1756 +++ new/vma.c   2013-11-28 12:08:39.000000000 +0100
1757 @@ -0,0 +1,623 @@
1758 +/*
1759 + * VMA: Virtual Machine Archive
1760 + *
1761 + * Copyright (C) 2012-2013 Proxmox Server Solutions
1762 + *
1763 + * Authors:
1764 + *  Dietmar Maurer (dietmar@proxmox.com)
1765 + *
1766 + * This work is licensed under the terms of the GNU GPL, version 2 or later.
1767 + * See the COPYING file in the top-level directory.
1768 + *
1769 + */
1770 +
1771 +#include <stdio.h>
1772 +#include <errno.h>
1773 +#include <unistd.h>
1774 +#include <stdio.h>
1775 +#include <string.h>
1776 +#include <sys/types.h>
1777 +#include <sys/stat.h>
1778 +#include <fcntl.h>
1779 +#include <glib.h>
1780 +
1781 +#include "vma.h"
1782 +#include "qemu-common.h"
1783 +#include "qemu/error-report.h"
1784 +#include "qemu/main-loop.h"
1785 +
1786 +static void help(void)
1787 +{
1788 +    const char *help_msg =
1789 +        "usage: vma command [command options]\n"
1790 +        "\n"
1791 +        "vma list <filename>\n"
1792 +        "vma create <filename> [-c config] <archive> pathname ...\n"
1793 +        "vma extract <filename> [-r] <targetdir>\n"
1794 +        ;
1795 +
1796 +    printf("%s", help_msg);
1797 +    exit(1);
1798 +}
1799 +
1800 +static const char *extract_devname(const char *path, char **devname, int index)
1801 +{
1802 +    assert(path);
1803 +
1804 +    const char *sep = strchr(path, '=');
1805 +
1806 +    if (sep) {
1807 +        *devname = g_strndup(path, sep - path);
1808 +        path = sep + 1;
1809 +    } else {
1810 +        if (index >= 0) {
1811 +            *devname = g_strdup_printf("disk%d", index);
1812 +        } else {
1813 +            *devname = NULL;
1814 +        }
1815 +    }
1816 +
1817 +    return path;
1818 +}
1819 +
1820 +static void print_content(VmaReader *vmar)
1821 +{
1822 +    assert(vmar);
1823 +
1824 +    VmaHeader *head = vma_reader_get_header(vmar);
1825 +
1826 +    GList *l = vma_reader_get_config_data(vmar);
1827 +    while (l && l->data) {
1828 +        VmaConfigData *cdata = (VmaConfigData *)l->data;
1829 +        l = g_list_next(l);
1830 +        printf("CFG: size: %d name: %s\n", cdata->len, cdata->name);
1831 +    }
1832 +
1833 +    int i;
1834 +    VmaDeviceInfo *di;
1835 +    for (i = 1; i < 255; i++) {
1836 +        di = vma_reader_get_device_info(vmar, i);
1837 +        if (di) {
1838 +            if (strcmp(di->devname, "vmstate") == 0) {
1839 +                printf("VMSTATE: dev_id=%d memory: %zd\n", i, di->size);
1840 +            } else {
1841 +                printf("DEV: dev_id=%d size: %zd devname: %s\n",
1842 +                       i, di->size, di->devname);
1843 +            }
1844 +        }
1845 +    }
1846 +    /* ctime is the last entry we print */
1847 +    printf("CTIME: %s", ctime(&head->ctime));
1848 +    fflush(stdout);
1849 +}
1850 +
1851 +static int list_content(int argc, char **argv)
1852 +{
1853 +    int c, ret = 0;
1854 +    const char *filename;
1855 +
1856 +    for (;;) {
1857 +        c = getopt(argc, argv, "h");
1858 +        if (c == -1) {
1859 +            break;
1860 +        }
1861 +        switch (c) {
1862 +        case '?':
1863 +        case 'h':
1864 +            help();
1865 +            break;
1866 +        default:
1867 +            g_assert_not_reached();
1868 +        }
1869 +    }
1870 +
1871 +    /* Get the filename */
1872 +    if ((optind + 1) != argc) {
1873 +        help();
1874 +    }
1875 +    filename = argv[optind++];
1876 +
1877 +    Error *errp = NULL;
1878 +    VmaReader *vmar = vma_reader_create(filename, &errp);
1879 +
1880 +    if (!vmar) {
1881 +        g_error("%s", error_get_pretty(errp));
1882 +    }
1883 +
1884 +    print_content(vmar);
1885 +
1886 +    vma_reader_destroy(vmar);
1887 +
1888 +    return ret;
1889 +}
1890 +
1891 +typedef struct RestoreMap {
1892 +    char *devname;
1893 +    char *path;
1894 +    bool write_zero;
1895 +} RestoreMap;
1896 +
1897 +static int extract_content(int argc, char **argv)
1898 +{
1899 +    int c, ret = 0;
1900 +    int verbose = 0;
1901 +    const char *filename;
1902 +    const char *dirname;
1903 +    const char *readmap = NULL;
1904 +
1905 +    for (;;) {
1906 +        c = getopt(argc, argv, "hvr:");
1907 +        if (c == -1) {
1908 +            break;
1909 +        }
1910 +        switch (c) {
1911 +        case '?':
1912 +        case 'h':
1913 +            help();
1914 +            break;
1915 +        case 'r':
1916 +            readmap = optarg;
1917 +            break;
1918 +        case 'v':
1919 +            verbose = 1;
1920 +            break;
1921 +        default:
1922 +            help();
1923 +        }
1924 +    }
1925 +
1926 +    /* Get the filename */
1927 +    if ((optind + 2) != argc) {
1928 +        help();
1929 +    }
1930 +    filename = argv[optind++];
1931 +    dirname = argv[optind++];
1932 +
1933 +    Error *errp = NULL;
1934 +    VmaReader *vmar = vma_reader_create(filename, &errp);
1935 +
1936 +    if (!vmar) {
1937 +        g_error("%s", error_get_pretty(errp));
1938 +    }
1939 +
1940 +    if (mkdir(dirname, 0777) < 0) {
1941 +        g_error("unable to create target directory %s - %s",
1942 +                dirname, g_strerror(errno));
1943 +    }
1944 +
1945 +    GList *l = vma_reader_get_config_data(vmar);
1946 +    while (l && l->data) {
1947 +        VmaConfigData *cdata = (VmaConfigData *)l->data;
1948 +        l = g_list_next(l);
1949 +        char *cfgfn = g_strdup_printf("%s/%s", dirname, cdata->name);
1950 +        GError *err = NULL;
1951 +        if (!g_file_set_contents(cfgfn, (gchar *)cdata->data, cdata->len,
1952 +                                 &err)) {
1953 +            g_error("unable to write file: %s", err->message);
1954 +        }
1955 +    }
1956 +
1957 +    GHashTable *devmap = g_hash_table_new(g_str_hash, g_str_equal);
1958 +
1959 +    if (readmap) {
1960 +        print_content(vmar);
1961 +
1962 +        FILE *map = fopen(readmap, "r");
1963 +        if (!map) {
1964 +            g_error("unable to open fifo %s - %s", readmap, g_strerror(errno));
1965 +        }
1966 +
1967 +        while (1) {
1968 +            char inbuf[8192];
1969 +            char *line = fgets(inbuf, sizeof(inbuf), map);
1970 +            if (!line || line[0] == '\0' || !strcmp(line, "done\n")) {
1971 +                break;
1972 +            }
1973 +            int len = strlen(line);
1974 +            if (line[len - 1] == '\n') {
1975 +                line[len - 1] = '\0';
1976 +                if (len == 1) {
1977 +                    break;
1978 +                }
1979 +            }
1980 +
1981 +            const char *path;
1982 +            bool write_zero;
1983 +            if (line[0] == '0' && line[1] == ':') {
1984 +                path = inbuf + 2;
1985 +                write_zero = false;
1986 +            } else if (line[0] == '1' && line[1] == ':') {
1987 +                path = inbuf + 2;
1988 +                write_zero = true;
1989 +            } else {
1990 +                g_error("read map failed - parse error ('%s')", inbuf);
1991 +            }
1992 +
1993 +            char *devname = NULL;
1994 +            path = extract_devname(path, &devname, -1);
1995 +            if (!devname) {
1996 +                g_error("read map failed - no dev name specified ('%s')",
1997 +                        inbuf);
1998 +            }
1999 +
2000 +            RestoreMap *map = g_new0(RestoreMap, 1);
2001 +            map->devname = g_strdup(devname);
2002 +            map->path = g_strdup(path);
2003 +            map->write_zero = write_zero;
2004 +
2005 +            g_hash_table_insert(devmap, map->devname, map);
2006 +
2007 +        };
2008 +    }
2009 +
2010 +    int i;
2011 +    int vmstate_fd = -1;
2012 +    guint8 vmstate_stream = 0;
2013 +
2014 +    for (i = 1; i < 255; i++) {
2015 +        VmaDeviceInfo *di = vma_reader_get_device_info(vmar, i);
2016 +        if (di && (strcmp(di->devname, "vmstate") == 0)) {
2017 +            vmstate_stream = i;
2018 +            char *statefn = g_strdup_printf("%s/vmstate.bin", dirname);
2019 +            vmstate_fd = open(statefn, O_WRONLY|O_CREAT|O_EXCL, 0644);
2020 +            if (vmstate_fd < 0) {
2021 +                g_error("create vmstate file '%s' failed - %s", statefn,
2022 +                        g_strerror(errno));
2023 +            }
2024 +            g_free(statefn);
2025 +        } else if (di) {
2026 +            char *devfn = NULL;
2027 +            int flags = BDRV_O_RDWR|BDRV_O_CACHE_WB;
2028 +            bool write_zero = true;
2029 +
2030 +            if (readmap) {
2031 +                RestoreMap *map;
2032 +                map = (RestoreMap *)g_hash_table_lookup(devmap, di->devname);
2033 +                if (map == NULL) {
2034 +                    g_error("no device name mapping for %s", di->devname);
2035 +                }
2036 +                devfn = map->path;
2037 +                write_zero = map->write_zero;
2038 +            } else {
2039 +                devfn = g_strdup_printf("%s/tmp-disk-%s.raw",
2040 +                                        dirname, di->devname);
2041 +                printf("DEVINFO %s %zd\n", devfn, di->size);
2042 +
2043 +                bdrv_img_create(devfn, "raw", NULL, NULL, NULL, di->size,
2044 +                                flags, &errp, 0);
2045 +                if (error_is_set(&errp)) {
2046 +                    g_error("can't create file %s: %s", devfn,
2047 +                            error_get_pretty(errp));
2048 +                }
2049 +
2050 +                /* Note: we created an empty file above, so there is no
2051 +                 * need to write zeroes (so we generate a sparse file)
2052 +                 */
2053 +                write_zero = false;
2054 +            }
2055 +
2056 +            BlockDriverState *bs = bdrv_new(di->devname);
2057 +            if (bdrv_open(bs, devfn, NULL, flags, NULL, &errp)) {
2058 +                g_error("can't open file %s - %s", devfn,
2059 +                        error_get_pretty(errp));
2060 +            }
2061 +            if (vma_reader_register_bs(vmar, i, bs, write_zero, &errp) < 0) {
2062 +                g_error("%s", error_get_pretty(errp));
2063 +            }
2064 +
2065 +            if (!readmap) {
2066 +                g_free(devfn);
2067 +            }
2068 +        }
2069 +    }
2070 +
2071 +    if (vma_reader_restore(vmar, vmstate_fd, verbose, &errp) < 0) {
2072 +        g_error("restore failed - %s", error_get_pretty(errp));
2073 +    }
2074 +
2075 +    if (!readmap) {
2076 +        for (i = 1; i < 255; i++) {
2077 +            VmaDeviceInfo *di = vma_reader_get_device_info(vmar, i);
2078 +            if (di && (i != vmstate_stream)) {
2079 +                char *tmpfn = g_strdup_printf("%s/tmp-disk-%s.raw",
2080 +                                              dirname, di->devname);
2081 +                char *fn = g_strdup_printf("%s/disk-%s.raw",
2082 +                                           dirname, di->devname);
2083 +                if (rename(tmpfn, fn) != 0) {
2084 +                    g_error("rename %s to %s failed - %s",
2085 +                            tmpfn, fn, g_strerror(errno));
2086 +                }
2087 +            }
2088 +        }
2089 +    }
2090 +
2091 +    vma_reader_destroy(vmar);
2092 +
2093 +    bdrv_close_all();
2094 +
2095 +    return ret;
2096 +}
2097 +
2098 +typedef struct BackupJob {
2099 +    BlockDriverState *bs;
2100 +    int64_t len;
2101 +    VmaWriter *vmaw;
2102 +    uint8_t dev_id;
2103 +} BackupJob;
2104 +
2105 +/*
2106 +static int backup_dump_cb(void *opaque, BlockDriverState *bs,
2107 +                          int64_t cluster_num, unsigned char *buf)
2108 +{
2109 +    BackupCB *bcb = opaque;
2110 +    size_t zb = 0;
2111 +    if (vma_writer_write(bcb->vmaw, bcb->dev_id, cluster_num, buf, &zb) < 0) {
2112 +        g_warning("backup_dump_cb vma_writer_write failed");
2113 +        return -1;
2114 +    }
2115 +
2116 +    return 0;
2117 +}
2118 +
2119 +static void backup_complete_cb(void *opaque, int ret)
2120 +{
2121 +    BackupCB *bcb = opaque;
2122 +
2123 +    if (ret < 0) {
2124 +        vma_writer_set_error(bcb->vmaw, "backup_complete_cb %d", ret);
2125 +    }
2126 +
2127 +    if (vma_writer_close_stream(bcb->vmaw, bcb->dev_id) <= 0) {
2128 +        Error *err = NULL;
2129 +        if (vma_writer_close(bcb->vmaw, &err) != 0) {
2130 +            g_warning("vma_writer_close failed %s", error_get_pretty(err));
2131 +        }
2132 +    }
2133 +}
2134 +*/
2135 +
2136 +#define BACKUP_SECTORS_PER_CLUSTER (VMA_CLUSTER_SIZE / BDRV_SECTOR_SIZE)
2137 +
2138 +static void coroutine_fn backup_run(void *opaque)
2139 +{
2140 +    BackupJob *job = (BackupJob *)opaque;
2141 +    struct iovec iov;
2142 +    QEMUIOVector qiov;
2143 +
2144 +    int64_t start, end;
2145 +    int ret = 0;
2146 +
2147 +    unsigned char *buf = qemu_blockalign(job->bs, VMA_CLUSTER_SIZE);
2148 +
2149 +    start = 0;
2150 +    end = DIV_ROUND_UP(job->len / BDRV_SECTOR_SIZE,
2151 +                       BACKUP_SECTORS_PER_CLUSTER);
2152 +
2153 +    for (; start < end; start++) {
2154 +        iov.iov_base = buf;
2155 +        iov.iov_len = VMA_CLUSTER_SIZE;
2156 +        qemu_iovec_init_external(&qiov, &iov, 1);
2157 +
2158 +        ret = bdrv_co_readv(job->bs, start * BACKUP_SECTORS_PER_CLUSTER,
2159 +                            BACKUP_SECTORS_PER_CLUSTER, &qiov);
2160 +        if (ret < 0) {
2161 +            vma_writer_set_error(job->vmaw, "read error", -1);
2162 +            goto out;
2163 +        }
2164 +
2165 +        size_t zb = 0;
2166 +        if (vma_writer_write(job->vmaw, job->dev_id, start, buf, &zb) < 0) {
2167 +            vma_writer_set_error(job->vmaw, "backup_dump_cb vma_writer_write failed", -1);
2168 +            goto out;
2169 +        }
2170 +    }
2171 +
2172 +
2173 +out:
2174 +    if (vma_writer_close_stream(job->vmaw, job->dev_id) <= 0) {
2175 +        Error *err = NULL;
2176 +        if (vma_writer_close(job->vmaw, &err) != 0) {
2177 +            g_warning("vma_writer_close failed %s", error_get_pretty(err));
2178 +        }
2179 +    }
2180 +}
2181 +
2182 +static int create_archive(int argc, char **argv)
2183 +{
2184 +    int i, c, res;
2185 +    int verbose = 0;
2186 +    const char *archivename;
2187 +    GList *config_files = NULL;
2188 +
2189 +    for (;;) {
2190 +        c = getopt(argc, argv, "hvc:");
2191 +        if (c == -1) {
2192 +            break;
2193 +        }
2194 +        switch (c) {
2195 +        case '?':
2196 +        case 'h':
2197 +            help();
2198 +            break;
2199 +        case 'c':
2200 +            config_files = g_list_append(config_files, optarg);
2201 +            break;
2202 +        case 'v':
2203 +            verbose = 1;
2204 +            break;
2205 +        default:
2206 +            g_assert_not_reached();
2207 +        }
2208 +    }
2209 +
2210 +
2211 +    /* make sure we have archive name and at least one path */
2212 +    if ((optind + 2) > argc) {
2213 +        help();
2214 +    }
2215 +
2216 +    archivename = argv[optind++];
2217 +
2218 +    uuid_t uuid;
2219 +    uuid_generate(uuid);
2220 +
2221 +    Error *local_err = NULL;
2222 +    VmaWriter *vmaw = vma_writer_create(archivename, uuid, &local_err);
2223 +
2224 +    if (vmaw == NULL) {
2225 +        g_error("%s", error_get_pretty(local_err));
2226 +    }
2227 +
2228 +    GList *l = config_files;
2229 +    while (l && l->data) {
2230 +        char *name = l->data;
2231 +        char *cdata = NULL;
2232 +        gsize clen = 0;
2233 +        GError *err = NULL;
2234 +        if (!g_file_get_contents(name, &cdata, &clen, &err)) {
2235 +            unlink(archivename);
2236 +            g_error("Unable to read file: %s", err->message);
2237 +        }
2238 +
2239 +        if (vma_writer_add_config(vmaw, name, cdata, clen) != 0) {
2240 +            unlink(archivename);
2241 +            g_error("Unable to append config data %s (len = %zd)",
2242 +                    name, clen);
2243 +        }
2244 +        l = g_list_next(l);
2245 +    }
2246 +
2247 +    int ind = 0;
2248 +    while (optind < argc) {
2249 +        const char *path = argv[optind++];
2250 +        char *devname = NULL;
2251 +        path = extract_devname(path, &devname, ind++);
2252 +
2253 +        BlockDriver *drv = NULL;
2254 +        BlockDriverState *bs = bdrv_new(devname);
2255 +        Error *errp = NULL;
2256 +
2257 +        res = bdrv_open(bs, path, NULL, BDRV_O_CACHE_WB , drv, &errp);
2258 +        if (res < 0) {
2259 +            unlink(archivename);
2260 +            g_error("bdrv_open '%s' failed - %s", path, error_get_pretty(errp));
2261 +        }
2262 +        int64_t size = bdrv_getlength(bs);
2263 +        int dev_id = vma_writer_register_stream(vmaw, devname, size);
2264 +        if (dev_id <= 0) {
2265 +            unlink(archivename);
2266 +            g_error("vma_writer_register_stream '%s' failed", devname);
2267 +        }
2268 +
2269 +        BackupJob *job = g_new0(BackupJob, 1);
2270 +        job->len = size;
2271 +        job->bs = bs;
2272 +        job->vmaw = vmaw;
2273 +        job->dev_id = dev_id;
2274 +
2275 +        Coroutine *co = qemu_coroutine_create(backup_run);
2276 +        qemu_coroutine_enter(co, job);
2277 +
2278 +       //BlockDriverState *target = NULL;
2279 +        //backup_start(bs, target, 0, MIRROR_SYNC_MODE_FULL,
2280 +        //             BLOCKDEV_ON_ERROR_STOP, BLOCKDEV_ON_ERROR_STOP,
2281 +        //             NULL, bcb, &errp);
2282 +
2283 +        //if (backup_job_create(bs, backup_dump_cb, backup_complete_cb,
2284 +        //                     bcb, 0) < 0) {
2285 +        //    unlink(archivename);
2286 +        //    g_error("backup_job_start failed");
2287 +        //} else {
2288 +        //    backup_job_start(bs, false);
2289 +        // }
2290 +    }
2291 +
2292 +    VmaStatus vmastat;
2293 +    int percent = 0;
2294 +    int last_percent = -1;
2295 +
2296 +    while (1) {
2297 +        main_loop_wait(false);
2298 +        vma_writer_get_status(vmaw, &vmastat);
2299 +
2300 +        if (verbose) {
2301 +
2302 +            uint64_t total = 0;
2303 +            uint64_t transferred = 0;
2304 +            uint64_t zero_bytes = 0;
2305 +
2306 +            int i;
2307 +            for (i = 0; i < 256; i++) {
2308 +                if (vmastat.stream_info[i].size) {
2309 +                    total += vmastat.stream_info[i].size;
2310 +                    transferred += vmastat.stream_info[i].transferred;
2311 +                    zero_bytes += vmastat.stream_info[i].zero_bytes;
2312 +                }
2313 +            }
2314 +            percent = (transferred*100)/total;
2315 +            if (percent != last_percent) {
2316 +                fprintf(stderr, "progress %d%% %zd/%zd %zd\n", percent,
2317 +                        transferred, total, zero_bytes);
2318 +                fflush(stderr);
2319 +
2320 +                last_percent = percent;
2321 +            }
2322 +        }
2323 +
2324 +        if (vmastat.closed) {
2325 +            break;
2326 +        }
2327 +    }
2328 +
2329 +    bdrv_drain_all();
2330 +
2331 +    vma_writer_get_status(vmaw, &vmastat);
2332 +
2333 +    if (verbose) {
2334 +        for (i = 0; i < 256; i++) {
2335 +            VmaStreamInfo *si = &vmastat.stream_info[i];
2336 +            if (si->size) {
2337 +                fprintf(stderr, "image %s: size=%zd zeros=%zd saved=%zd\n",
2338 +                        si->devname, si->size, si->zero_bytes,
2339 +                        si->size - si->zero_bytes);
2340 +            }
2341 +        }
2342 +    }
2343 +
2344 +    if (vmastat.status < 0) {
2345 +        unlink(archivename);
2346 +        g_error("creating vma archive failed");
2347 +    }
2348 +
2349 +    return 0;
2350 +}
2351 +
2352 +int main(int argc, char **argv)
2353 +{
2354 +    const char *cmdname;
2355 +
2356 +    error_set_progname(argv[0]);
2357 +
2358 +    qemu_init_main_loop();
2359 +
2360 +    bdrv_init();
2361 +
2362 +    if (argc < 2) {
2363 +        help();
2364 +    }
2365 +
2366 +    cmdname = argv[1];
2367 +    argc--; argv++;
2368 +
2369 +
2370 +    if (!strcmp(cmdname, "list")) {
2371 +        return list_content(argc, argv);
2372 +    } else if (!strcmp(cmdname, "create")) {
2373 +        return create_archive(argc, argv);
2374 +    } else if (!strcmp(cmdname, "extract")) {
2375 +        return extract_content(argc, argv);
2376 +    }
2377 +
2378 +    help();
2379 +    return 0;
2380 +}
2381 Index: new/vma.h
2382 ===================================================================
2383 --- /dev/null   1970-01-01 00:00:00.000000000 +0000
2384 +++ new/vma.h   2013-11-28 12:04:49.000000000 +0100
2385 @@ -0,0 +1,144 @@
2386 +/*
2387 + * VMA: Virtual Machine Archive
2388 + *
2389 + * Copyright (C) Proxmox Server Solutions
2390 + *
2391 + * Authors:
2392 + *  Dietmar Maurer (dietmar@proxmox.com)
2393 + *
2394 + * This work is licensed under the terms of the GNU GPL, version 2 or later.
2395 + * See the COPYING file in the top-level directory.
2396 + *
2397 + */
2398 +
2399 +#ifndef BACKUP_VMA_H
2400 +#define BACKUP_VMA_H
2401 +
2402 +#include <uuid/uuid.h>
2403 +#include "qapi/error.h"
2404 +#include "block/block.h"
2405 +
2406 +#define VMA_BLOCK_BITS 12
2407 +#define VMA_BLOCK_SIZE (1<<VMA_BLOCK_BITS)
2408 +#define VMA_CLUSTER_BITS (VMA_BLOCK_BITS+4)
2409 +#define VMA_CLUSTER_SIZE (1<<VMA_CLUSTER_BITS)
2410 +
2411 +#if VMA_CLUSTER_SIZE != 65536
2412 +#error unexpected cluster size
2413 +#endif
2414 +
2415 +#define VMA_EXTENT_HEADER_SIZE 512
2416 +#define VMA_BLOCKS_PER_EXTENT 59
2417 +#define VMA_MAX_CONFIGS 256
2418 +
2419 +#define VMA_MAX_EXTENT_SIZE \
2420 +    (VMA_EXTENT_HEADER_SIZE+VMA_CLUSTER_SIZE*VMA_BLOCKS_PER_EXTENT)
2421 +#if VMA_MAX_EXTENT_SIZE != 3867136
2422 +#error unexpected VMA_EXTENT_SIZE
2423 +#endif
2424 +
2425 +/* File Format Definitions */
2426 +
2427 +#define VMA_MAGIC (GUINT32_TO_BE(('V'<<24)|('M'<<16)|('A'<<8)|0x00))
2428 +#define VMA_EXTENT_MAGIC (GUINT32_TO_BE(('V'<<24)|('M'<<16)|('A'<<8)|'E'))
2429 +
2430 +typedef struct VmaDeviceInfoHeader {
2431 +    uint32_t devname_ptr; /* offset into blob_buffer table */
2432 +    uint32_t reserved0;
2433 +    uint64_t size; /* device size in bytes */
2434 +    uint64_t reserved1;
2435 +    uint64_t reserved2;
2436 +} VmaDeviceInfoHeader;
2437 +
2438 +typedef struct VmaHeader {
2439 +    uint32_t magic;
2440 +    uint32_t version;
2441 +    unsigned char uuid[16];
2442 +    int64_t ctime;
2443 +    unsigned char md5sum[16];
2444 +
2445 +    uint32_t blob_buffer_offset;
2446 +    uint32_t blob_buffer_size;
2447 +    uint32_t header_size;
2448 +
2449 +    unsigned char reserved[1984];
2450 +
2451 +    uint32_t config_names[VMA_MAX_CONFIGS]; /* offset into blob_buffer table */
2452 +    uint32_t config_data[VMA_MAX_CONFIGS];  /* offset into blob_buffer table */
2453 +
2454 +    VmaDeviceInfoHeader dev_info[256];
2455 +} VmaHeader;
2456 +
2457 +typedef struct VmaExtentHeader {
2458 +    uint32_t magic;
2459 +    uint16_t reserved1;
2460 +    uint16_t block_count;
2461 +    unsigned char uuid[16];
2462 +    unsigned char md5sum[16];
2463 +    uint64_t blockinfo[VMA_BLOCKS_PER_EXTENT];
2464 +} VmaExtentHeader;
2465 +
2466 +/* functions/definitions to read/write vma files */
2467 +
2468 +typedef struct VmaReader VmaReader;
2469 +
2470 +typedef struct VmaWriter VmaWriter;
2471 +
2472 +typedef struct VmaConfigData {
2473 +    const char *name;
2474 +    const void *data;
2475 +    uint32_t len;
2476 +} VmaConfigData;
2477 +
2478 +typedef struct VmaStreamInfo {
2479 +    uint64_t size;
2480 +    uint64_t cluster_count;
2481 +    uint64_t transferred;
2482 +    uint64_t zero_bytes;
2483 +    int finished;
2484 +    char *devname;
2485 +} VmaStreamInfo;
2486 +
2487 +typedef struct VmaStatus {
2488 +    int status;
2489 +    bool closed;
2490 +    char errmsg[8192];
2491 +    char uuid_str[37];
2492 +    VmaStreamInfo stream_info[256];
2493 +} VmaStatus;
2494 +
2495 +typedef struct VmaDeviceInfo {
2496 +    uint64_t size; /* device size in bytes */
2497 +    const char *devname;
2498 +} VmaDeviceInfo;
2499 +
2500 +VmaWriter *vma_writer_create(const char *filename, uuid_t uuid, Error **errp);
2501 +int vma_writer_close(VmaWriter *vmaw, Error **errp);
2502 +void vma_writer_destroy(VmaWriter *vmaw);
2503 +int vma_writer_add_config(VmaWriter *vmaw, const char *name, gpointer data,
2504 +                          size_t len);
2505 +int vma_writer_register_stream(VmaWriter *vmaw, const char *devname,
2506 +                               size_t size);
2507 +
2508 +int64_t coroutine_fn vma_writer_write(VmaWriter *vmaw, uint8_t dev_id,
2509 +                                      int64_t cluster_num, unsigned char *buf,
2510 +                                      size_t *zero_bytes);
2511 +
2512 +int coroutine_fn vma_writer_close_stream(VmaWriter *vmaw, uint8_t dev_id);
2513 +
2514 +int vma_writer_get_status(VmaWriter *vmaw, VmaStatus *status);
2515 +void vma_writer_set_error(VmaWriter *vmaw, const char *fmt, ...);
2516 +
2517 +
2518 +VmaReader *vma_reader_create(const char *filename, Error **errp);
2519 +void vma_reader_destroy(VmaReader *vmar);
2520 +VmaHeader *vma_reader_get_header(VmaReader *vmar);
2521 +GList *vma_reader_get_config_data(VmaReader *vmar);
2522 +VmaDeviceInfo *vma_reader_get_device_info(VmaReader *vmar, guint8 dev_id);
2523 +int vma_reader_register_bs(VmaReader *vmar, guint8 dev_id,
2524 +                           BlockDriverState *bs, bool write_zeroes,
2525 +                           Error **errp);
2526 +int vma_reader_restore(VmaReader *vmar, int vmstate_fd, bool verbose,
2527 +                       Error **errp);
2528 +
2529 +#endif /* BACKUP_VMA_H */