debian/patches/0004-introduce-new-vma-archive-format.patch

   1 From ada8961d41298b340dc624da571cba760ab39158 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 v3 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            |  776 +++++++++++++++++++++++++++++++++++++++
  15  vma-writer.c            |  920 +++++++++++++++++++++++++++++++++++++++++++++++
  16  vma.c                   |  555 ++++++++++++++++++++++++++++
  17  vma.h                   |  145 ++++++++
  18  8 files changed, 2427 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 diff --git a/Makefile b/Makefile
  26 index 9ecbcbb..30a9268 100644
  27 --- a/Makefile
  28 +++ b/Makefile
  29 @@ -100,7 +100,7 @@ defconfig:
  30
  31  -include config-all-devices.mak
  32
  33 -all: $(DOCS) $(TOOLS) $(HELPERS-y) recurse-all
  34 +all: $(DOCS) $(TOOLS) vma$(EXESUF) $(HELPERS-y) recurse-all
  35
  36  config-host.h: config-host.h-timestamp
  37  config-host.h-timestamp: config-host.mak
  38 @@ -194,6 +194,7 @@ tools-obj-$(CONFIG_POSIX) += compatfd.o
  39  qemu-img$(EXESUF): qemu-img.o $(tools-obj-y) $(block-obj-y) libqemustub.a
  40  qemu-nbd$(EXESUF): qemu-nbd.o $(tools-obj-y) $(block-obj-y) libqemustub.a
  41  qemu-io$(EXESUF): qemu-io.o cmd.o $(tools-obj-y) $(block-obj-y) libqemustub.a
  42 +vma$(EXESUF): vma.o vma-writer.o vma-reader.o $(tools-obj-y) $(block-obj-y) libqemustub.a
  43
  44  qemu-bridge-helper$(EXESUF): qemu-bridge-helper.o
  45
  46 diff --git a/Makefile.objs b/Makefile.objs
  47 index cb46be5..b5732e2 100644
  48 --- a/Makefile.objs
  49 +++ b/Makefile.objs
  50 @@ -48,7 +48,7 @@ coroutine-obj-$(CONFIG_WIN32) += coroutine-win32.o
  51  block-obj-y = iov.o cache-utils.o qemu-option.o module.o async.o
  52  block-obj-y += nbd.o block.o blockjob.o aes.o qemu-config.o
  53  block-obj-y += thread-pool.o qemu-progress.o qemu-sockets.o uri.o notify.o
  54 -block-obj-y += backup.o
  55 +block-obj-y += vma-writer.o backup.o
  56  block-obj-y += $(coroutine-obj-y) $(qobject-obj-y) $(version-obj-y)
  57  block-obj-$(CONFIG_POSIX) += event_notifier-posix.o aio-posix.o
  58  block-obj-$(CONFIG_WIN32) += event_notifier-win32.o aio-win32.o
  59 diff --git a/blockdev.c b/blockdev.c
  60 index 9e85ffe..139e350 100644
  61 --- a/blockdev.c
  62 +++ b/blockdev.c
  63 @@ -21,6 +21,7 @@
  64  #include "trace.h"
  65  #include "arch_init.h"
  66  #include "backup.h"
  67 +#include "vma.h"
  68
  69  static QTAILQ_HEAD(drivelist, DriveInfo) drives = QTAILQ_HEAD_INITIALIZER(drives);
  70
  71 @@ -1483,10 +1484,11 @@ char *qmp_backup(const char *backupfile, bool has_format, const char *format,
  72      /* Todo: try to auto-detect format based on file name */
  73      format = has_format ? format : "vma";
  74
  75 -    /* fixme: find driver for specifued format */
  76      const BackupDriver *driver = NULL;
  77
  78 -    if (!driver) {
  79 +    if (strcmp(format, "vma") == 0) {
  80 +        driver = &backup_vma_driver;
  81 +    } else {
  82          error_set(errp, ERROR_CLASS_GENERIC_ERROR,
  83                    "no backup driver for format '%s'", format);
  84          return NULL;
  85 diff --git a/docs/specs/vma_spec.txt b/docs/specs/vma_spec.txt
  86 new file mode 100644
  87 index 0000000..052c629
  88 --- /dev/null
  89 +++ b/docs/specs/vma_spec.txt
  90 @@ -0,0 +1,24 @@
  91 +=Virtual Machine Archive format (VMA)=
  92 +
  93 +This format contains a header which includes the VM configuration as
  94 +binary blobs, and a list of devices (dev_id, name).
  95 +
  96 +The actual VM image data is stored inside extents. An extent contains
  97 +up to 64 clusters, and start with a 512 byte header containing
  98 +additional information for those clusters.
  99 +
 100 +We use a cluster size of 65536, and use 8 bytes for each
 101 +cluster in the header to store the following information:
 102 +
 103 +* 1 byte dev_id (to identity the drive)
 104 +* 2 bytes zero indicator (mark zero regions (16x4096))
 105 +* 4 bytes cluster number
 106 +* 1 byte not used (reserved)
 107 +
 108 +We only store non-zero blocks (such block is 4096 bytes).
 109 +
 110 +Each archive is marked with a uuid. The archive header and all
 111 +extent headers includes that uuid and a MD5 checksum (over header
 112 +data).
 113 +
 114 +
 115 diff --git a/vma-reader.c b/vma-reader.c
 116 new file mode 100644
 117 index 0000000..b05d7da
 118 --- /dev/null
 119 +++ b/vma-reader.c
 120 @@ -0,0 +1,776 @@
 121 +/*
 122 + * VMA: Virtual Machine Archive
 123 + *
 124 + * Copyright (C) 2012 Proxmox Server Solutions
 125 + *
 126 + * Authors:
 127 + *  Dietmar Maurer (dietmar@proxmox.com)
 128 + *
 129 + * This work is licensed under the terms of the GNU GPL, version 2 or later.
 130 + * See the COPYING file in the top-level directory.
 131 + *
 132 + */
 133 +
 134 +#include <stdio.h>
 135 +#include <errno.h>
 136 +#include <unistd.h>
 137 +#include <stdio.h>
 138 +#include <string.h>
 139 +#include <sys/types.h>
 140 +#include <sys/stat.h>
 141 +#include <fcntl.h>
 142 +#include <glib.h>
 143 +#include <uuid/uuid.h>
 144 +
 145 +#include "qemu-common.h"
 146 +#include "qemu_socket.h"
 147 +#include "qemu-coroutine.h"
 148 +#include "qemu-aio.h"
 149 +#include "qemu/ratelimit.h"
 150 +#include "vma.h"
 151 +#include "block.h"
 152 +
 153 +#define BITS_PER_LONG  (sizeof(unsigned long) * 8)
 154 +
 155 +static unsigned char zero_vma_block[VMA_BLOCK_SIZE];
 156 +
 157 +typedef struct VmaRestoreState {
 158 +    BlockDriverState *bs;
 159 +    bool write_zeroes;
 160 +    unsigned long *bitmap;
 161 +    int bitmap_size;
 162 +}  VmaRestoreState;
 163 +
 164 +struct VmaReader {
 165 +    int fd;
 166 +    GChecksum *md5csum;
 167 +    GHashTable *blob_hash;
 168 +    unsigned char *head_data;
 169 +    VmaDeviceInfo devinfo[256];
 170 +    VmaRestoreState rstate[256];
 171 +    GList *cdata_list;
 172 +    guint8 vmstate_stream;
 173 +    uint32_t vmstate_clusters;
 174 +};
 175 +
 176 +static guint
 177 +g_int32_hash(gconstpointer v)
 178 +{
 179 +    return *(const uint32_t *)v;
 180 +}
 181 +
 182 +static gboolean
 183 +g_int32_equal(gconstpointer v1, gconstpointer v2)
 184 +{
 185 +    return *((const uint32_t *)v1) == *((const uint32_t *)v2);
 186 +}
 187 +
 188 +static int vma_reader_get_bitmap(VmaRestoreState *rstate, int64_t cluster_num)
 189 +{
 190 +    assert(rstate);
 191 +    assert(rstate->bitmap);
 192 +
 193 +    unsigned long val, idx, bit;
 194 +
 195 +    idx = cluster_num / BITS_PER_LONG;
 196 +
 197 +    assert(rstate->bitmap_size > idx);
 198 +
 199 +    bit = cluster_num % BITS_PER_LONG;
 200 +    val = rstate->bitmap[idx];
 201 +
 202 +    return !!(val & (1UL << bit));
 203 +}
 204 +
 205 +static void vma_reader_set_bitmap(VmaRestoreState *rstate, int64_t cluster_num,
 206 +                                  int dirty)
 207 +{
 208 +    assert(rstate);
 209 +    assert(rstate->bitmap);
 210 +
 211 +    unsigned long val, idx, bit;
 212 +
 213 +    idx = cluster_num / BITS_PER_LONG;
 214 +
 215 +    assert(rstate->bitmap_size > idx);
 216 +
 217 +    bit = cluster_num % BITS_PER_LONG;
 218 +    val = rstate->bitmap[idx];
 219 +    if (dirty) {
 220 +        if (!(val & (1UL << bit))) {
 221 +            val |= 1UL << bit;
 222 +        }
 223 +    } else {
 224 +        if (val & (1UL << bit)) {
 225 +            val &= ~(1UL << bit);
 226 +        }
 227 +    }
 228 +    rstate->bitmap[idx] = val;
 229 +}
 230 +
 231 +typedef struct VmaBlob {
 232 +    uint32_t start;
 233 +    uint32_t len;
 234 +    void *data;
 235 +} VmaBlob;
 236 +
 237 +static const VmaBlob *get_header_blob(VmaReader *vmar, uint32_t pos)
 238 +{
 239 +    assert(vmar);
 240 +    assert(vmar->blob_hash);
 241 +
 242 +    return g_hash_table_lookup(vmar->blob_hash, &pos);
 243 +}
 244 +
 245 +static const char *get_header_str(VmaReader *vmar, uint32_t pos)
 246 +{
 247 +    const VmaBlob *blob = get_header_blob(vmar, pos);
 248 +    if (!blob) {
 249 +        return NULL;
 250 +    }
 251 +    const char *res = (char *)blob->data;
 252 +    if (res[blob->len-1] != '\0') {
 253 +        return NULL;
 254 +    }
 255 +    return res;
 256 +}
 257 +
 258 +static ssize_t
 259 +safe_read(int fd, unsigned char *buf, size_t count)
 260 +{
 261 +    ssize_t n;
 262 +
 263 +    do {
 264 +        n = read(fd, buf, count);
 265 +    } while (n < 0 && errno == EINTR);
 266 +
 267 +    return n;
 268 +}
 269 +
 270 +static ssize_t
 271 +full_read(int fd, unsigned char *buf, size_t len)
 272 +{
 273 +    ssize_t n;
 274 +    size_t total;
 275 +
 276 +    total = 0;
 277 +
 278 +    while (len > 0) {
 279 +        n = safe_read(fd, buf, len);
 280 +
 281 +        if (n == 0) {
 282 +            return total;
 283 +        }
 284 +
 285 +        if (n <= 0) {
 286 +            break;
 287 +        }
 288 +
 289 +        buf += n;
 290 +        total += n;
 291 +        len -= n;
 292 +    }
 293 +
 294 +    if (len) {
 295 +        return -1;
 296 +    }
 297 +
 298 +    return total;
 299 +}
 300 +
 301 +void vma_reader_destroy(VmaReader *vmar)
 302 +{
 303 +    assert(vmar);
 304 +
 305 +    if (vmar->fd >= 0) {
 306 +        close(vmar->fd);
 307 +    }
 308 +
 309 +    if (vmar->cdata_list) {
 310 +        g_list_free(vmar->cdata_list);
 311 +    }
 312 +
 313 +    int i;
 314 +    for (i = 1; i < 256; i++) {
 315 +        if (vmar->rstate[i].bitmap) {
 316 +            g_free(vmar->rstate[i].bitmap);
 317 +        }
 318 +    }
 319 +
 320 +    if (vmar->md5csum) {
 321 +        g_checksum_free(vmar->md5csum);
 322 +    }
 323 +
 324 +    if (vmar->blob_hash) {
 325 +        g_hash_table_destroy(vmar->blob_hash);
 326 +    }
 327 +
 328 +    if (vmar->head_data) {
 329 +        g_free(vmar->head_data);
 330 +    }
 331 +
 332 +    g_free(vmar);
 333 +
 334 +};
 335 +
 336 +static int vma_reader_read_head(VmaReader *vmar, Error **errp)
 337 +{
 338 +    assert(vmar);
 339 +    assert(errp);
 340 +    assert(*errp == NULL);
 341 +
 342 +    unsigned char md5sum[16];
 343 +    int i;
 344 +    int ret = 0;
 345 +
 346 +    vmar->head_data = g_malloc(sizeof(VmaHeader));
 347 +
 348 +    if (full_read(vmar->fd, vmar->head_data, sizeof(VmaHeader)) !=
 349 +        sizeof(VmaHeader)) {
 350 +        error_setg(errp, "can't read vma header - %s",
 351 +                   errno ? strerror(errno) : "got EOF");
 352 +        return -1;
 353 +    }
 354 +
 355 +    VmaHeader *h = (VmaHeader *)vmar->head_data;
 356 +
 357 +    if (h->magic != VMA_MAGIC) {
 358 +        error_setg(errp, "not a vma file - wrong magic number");
 359 +        return -1;
 360 +    }
 361 +
 362 +    uint32_t header_size = GUINT32_FROM_BE(h->header_size);
 363 +    int need = header_size - sizeof(VmaHeader);
 364 +    if (need <= 0) {
 365 +        error_setg(errp, "wrong vma header size %d", header_size);
 366 +        return -1;
 367 +    }
 368 +
 369 +    vmar->head_data = g_realloc(vmar->head_data, header_size);
 370 +    h = (VmaHeader *)vmar->head_data;
 371 +
 372 +    if (full_read(vmar->fd, vmar->head_data + sizeof(VmaHeader), need) !=
 373 +        need) {
 374 +        error_setg(errp, "can't read vma header data - %s",
 375 +                   errno ? strerror(errno) : "got EOF");
 376 +        return -1;
 377 +    }
 378 +
 379 +    memcpy(md5sum, h->md5sum, 16);
 380 +    memset(h->md5sum, 0, 16);
 381 +
 382 +    g_checksum_reset(vmar->md5csum);
 383 +    g_checksum_update(vmar->md5csum, vmar->head_data, header_size);
 384 +    gsize csize = 16;
 385 +    g_checksum_get_digest(vmar->md5csum, (guint8 *)(h->md5sum), &csize);
 386 +
 387 +    if (memcmp(md5sum, h->md5sum, 16) != 0) {
 388 +        error_setg(errp, "wrong vma header chechsum");
 389 +        return -1;
 390 +    }
 391 +
 392 +    /* we can modify header data after checksum verify */
 393 +    h->header_size = header_size;
 394 +
 395 +    h->version = GUINT32_FROM_BE(h->version);
 396 +    if (h->version != 1) {
 397 +        error_setg(errp, "wrong vma version %d", h->version);
 398 +        return -1;
 399 +    }
 400 +
 401 +    h->ctime = GUINT64_FROM_BE(h->ctime);
 402 +    h->blob_buffer_offset = GUINT32_FROM_BE(h->blob_buffer_offset);
 403 +    h->blob_buffer_size = GUINT32_FROM_BE(h->blob_buffer_size);
 404 +
 405 +    uint32_t bstart = h->blob_buffer_offset + 1;
 406 +    uint32_t bend = h->blob_buffer_offset + h->blob_buffer_size;
 407 +
 408 +    if (bstart <= sizeof(VmaHeader)) {
 409 +        error_setg(errp, "wrong vma blob buffer offset %d",
 410 +                   h->blob_buffer_offset);
 411 +        return -1;
 412 +    }
 413 +
 414 +    if (bend > header_size) {
 415 +        error_setg(errp, "wrong vma blob buffer size %d/%d",
 416 +                   h->blob_buffer_offset, h->blob_buffer_size);
 417 +        return -1;
 418 +    }
 419 +
 420 +    while ((bstart + 2) <= bend) {
 421 +        uint32_t size = vmar->head_data[bstart] +
 422 +            (vmar->head_data[bstart+1] << 8);
 423 +        if ((bstart + size + 2) <= bend) {
 424 +            VmaBlob *blob = g_new0(VmaBlob, 1);
 425 +            blob->start = bstart - h->blob_buffer_offset;
 426 +            blob->len = size;
 427 +            blob->data = vmar->head_data + bstart + 2;
 428 +            g_hash_table_insert(vmar->blob_hash, &blob->start, blob);
 429 +        }
 430 +        bstart += size + 2;
 431 +    }
 432 +
 433 +
 434 +    int count = 0;
 435 +    for (i = 1; i < 256; i++) {
 436 +        VmaDeviceInfoHeader *dih = &h->dev_info[i];
 437 +        uint32_t devname_ptr = GUINT32_FROM_BE(dih->devname_ptr);
 438 +        uint64_t size = GUINT64_FROM_BE(dih->size);
 439 +        const char *devname =  get_header_str(vmar, devname_ptr);
 440 +
 441 +        if (size && devname) {
 442 +            count++;
 443 +            vmar->devinfo[i].size = size;
 444 +            vmar->devinfo[i].devname = devname;
 445 +
 446 +            if (strcmp(devname, "vmstate") == 0) {
 447 +                vmar->vmstate_stream = i;
 448 +            }
 449 +        }
 450 +    }
 451 +
 452 +    if (!count) {
 453 +        error_setg(errp, "vma does not contain data");
 454 +        return -1;
 455 +    }
 456 +
 457 +    for (i = 0; i < VMA_MAX_CONFIGS; i++) {
 458 +        uint32_t name_ptr = GUINT32_FROM_BE(h->config_names[i]);
 459 +        uint32_t data_ptr = GUINT32_FROM_BE(h->config_data[i]);
 460 +
 461 +        if (!(name_ptr && data_ptr)) {
 462 +            continue;
 463 +        }
 464 +        const char *name =  get_header_str(vmar, name_ptr);
 465 +        const VmaBlob *blob = get_header_blob(vmar, data_ptr);
 466 +
 467 +        if (!(name && blob)) {
 468 +            error_setg(errp, "vma contains invalid data pointers");
 469 +            return -1;
 470 +        }
 471 +
 472 +        VmaConfigData *cdata = g_new0(VmaConfigData, 1);
 473 +        cdata->name = name;
 474 +        cdata->data = blob->data;
 475 +        cdata->len = blob->len;
 476 +
 477 +        vmar->cdata_list = g_list_append(vmar->cdata_list, cdata);
 478 +    }
 479 +
 480 +    return ret;
 481 +};
 482 +
 483 +VmaReader *vma_reader_create(const char *filename, Error **errp)
 484 +{
 485 +    assert(filename);
 486 +    assert(errp);
 487 +
 488 +    VmaReader *vmar = g_new0(VmaReader, 1);
 489 +
 490 +    if (strcmp(filename, "-") == 0) {
 491 +        vmar->fd = dup(0);
 492 +    } else {
 493 +        vmar->fd = open(filename, O_RDONLY);
 494 +    }
 495 +
 496 +    if (vmar->fd < 0) {
 497 +        error_setg(errp, "can't open file %s - %s\n", filename,
 498 +                   strerror(errno));
 499 +        goto err;
 500 +    }
 501 +
 502 +    vmar->md5csum = g_checksum_new(G_CHECKSUM_MD5);
 503 +    if (!vmar->md5csum) {
 504 +        error_setg(errp, "can't allocate cmsum\n");
 505 +        goto err;
 506 +    }
 507 +
 508 +    vmar->blob_hash = g_hash_table_new_full(g_int32_hash, g_int32_equal,
 509 +                                            NULL, g_free);
 510 +
 511 +    if (vma_reader_read_head(vmar, errp) < 0) {
 512 +        goto err;
 513 +    }
 514 +
 515 +    return vmar;
 516 +
 517 +err:
 518 +    if (vmar) {
 519 +        vma_reader_destroy(vmar);
 520 +    }
 521 +
 522 +    return NULL;
 523 +}
 524 +
 525 +VmaHeader *vma_reader_get_header(VmaReader *vmar)
 526 +{
 527 +    assert(vmar);
 528 +    assert(vmar->head_data);
 529 +
 530 +    return (VmaHeader *)(vmar->head_data);
 531 +}
 532 +
 533 +GList *vma_reader_get_config_data(VmaReader *vmar)
 534 +{
 535 +    assert(vmar);
 536 +    assert(vmar->head_data);
 537 +
 538 +    return vmar->cdata_list;
 539 +}
 540 +
 541 +VmaDeviceInfo *vma_reader_get_device_info(VmaReader *vmar, guint8 dev_id)
 542 +{
 543 +    assert(vmar);
 544 +    assert(dev_id);
 545 +
 546 +    if (vmar->devinfo[dev_id].size && vmar->devinfo[dev_id].devname) {
 547 +        return &vmar->devinfo[dev_id];
 548 +    }
 549 +
 550 +    return NULL;
 551 +}
 552 +
 553 +int vma_reader_register_bs(VmaReader *vmar, guint8 dev_id, BlockDriverState *bs,
 554 +                           bool write_zeroes, Error **errp)
 555 +{
 556 +    assert(vmar);
 557 +    assert(bs != NULL);
 558 +    assert(dev_id);
 559 +    assert(vmar->rstate[dev_id].bs == NULL);
 560 +
 561 +    int64_t size = bdrv_getlength(bs);
 562 +    if (size != vmar->devinfo[dev_id].size) {
 563 +        error_setg(errp, "vma_reader_register_bs for stream %s failed - "
 564 +                   "unexpected size %zd != %zd", vmar->devinfo[dev_id].devname,
 565 +                   size, vmar->devinfo[dev_id].size);
 566 +        return -1;
 567 +    }
 568 +
 569 +    vmar->rstate[dev_id].bs = bs;
 570 +    vmar->rstate[dev_id].write_zeroes = write_zeroes;
 571 +
 572 +    int64_t bitmap_size = (size/BDRV_SECTOR_SIZE) +
 573 +        (VMA_CLUSTER_SIZE/BDRV_SECTOR_SIZE) * BITS_PER_LONG - 1;
 574 +    bitmap_size /= (VMA_CLUSTER_SIZE/BDRV_SECTOR_SIZE) * BITS_PER_LONG;
 575 +
 576 +    vmar->rstate[dev_id].bitmap_size = bitmap_size;
 577 +    vmar->rstate[dev_id].bitmap = g_new0(unsigned long, bitmap_size);
 578 +
 579 +    return 0;
 580 +}
 581 +
 582 +static ssize_t safe_write(int fd, void *buf, size_t count)
 583 +{
 584 +    ssize_t n;
 585 +
 586 +    do {
 587 +        n = write(fd, buf, count);
 588 +    } while (n < 0 && errno == EINTR);
 589 +
 590 +    return n;
 591 +}
 592 +
 593 +static size_t full_write(int fd, void *buf, size_t len)
 594 +{
 595 +    ssize_t n;
 596 +    size_t total;
 597 +
 598 +    total = 0;
 599 +
 600 +    while (len > 0) {
 601 +        n = safe_write(fd, buf, len);
 602 +        if (n < 0) {
 603 +            return n;
 604 +        }
 605 +        buf += n;
 606 +        total += n;
 607 +        len -= n;
 608 +    }
 609 +
 610 +    if (len) {
 611 +        /* incomplete write ? */
 612 +        return -1;
 613 +    }
 614 +
 615 +    return total;
 616 +}
 617 +
 618 +static int restore_write_data(VmaReader *vmar, guint8 dev_id,
 619 +                              BlockDriverState *bs, int vmstate_fd,
 620 +                              unsigned char *buf, int64_t sector_num,
 621 +                              int nb_sectors, Error **errp)
 622 +{
 623 +    assert(vmar);
 624 +
 625 +    if (dev_id == vmar->vmstate_stream) {
 626 +        if (vmstate_fd >= 0) {
 627 +            int len = nb_sectors * BDRV_SECTOR_SIZE;
 628 +            int res = full_write(vmstate_fd, buf, len);
 629 +            if (res < 0) {
 630 +                error_setg(errp, "write vmstate failed %d", res);
 631 +                return -1;
 632 +            }
 633 +        }
 634 +    } else {
 635 +        int res = bdrv_write(bs, sector_num, buf, nb_sectors);
 636 +        if (res < 0) {
 637 +            error_setg(errp, "bdrv_write to %s failed (%d)",
 638 +                       bdrv_get_device_name(bs), res);
 639 +            return -1;
 640 +        }
 641 +    }
 642 +    return 0;
 643 +}
 644 +static int restore_extent(VmaReader *vmar, unsigned char *buf,
 645 +                          int extent_size, int vmstate_fd, Error **errp)
 646 +{
 647 +    assert(vmar);
 648 +    assert(buf);
 649 +
 650 +    VmaExtentHeader *ehead = (VmaExtentHeader *)buf;
 651 +    int start = VMA_EXTENT_HEADER_SIZE;
 652 +    int i;
 653 +
 654 +    for (i = 0; i < VMA_BLOCKS_PER_EXTENT; i++) {
 655 +        uint64_t block_info = GUINT64_FROM_BE(ehead->blockinfo[i]);
 656 +        uint32_t cluster_num = block_info &  0xffffffff;
 657 +        uint8_t dev_id = (block_info >> 32) &  0xff;
 658 +        uint16_t mask = block_info >> (32+16);
 659 +        int64_t max_sector;
 660 +
 661 +        if (!dev_id) {
 662 +            continue;
 663 +        }
 664 +
 665 +        VmaRestoreState *rstate = &vmar->rstate[dev_id];
 666 +        BlockDriverState *bs = NULL;
 667 +
 668 +        if (dev_id != vmar->vmstate_stream) {
 669 +            bs = rstate->bs;
 670 +            if (!bs) {
 671 +                error_setg(errp, "got wrong dev id %d", dev_id);
 672 +                return -1;
 673 +            }
 674 +
 675 +            if (vma_reader_get_bitmap(rstate, cluster_num)) {
 676 +                error_setg(errp, "found duplicated cluster %d for stream %s",
 677 +                          cluster_num, vmar->devinfo[dev_id].devname);
 678 +                return -1;
 679 +            }
 680 +            vma_reader_set_bitmap(rstate, cluster_num, 1);
 681 +
 682 +            max_sector = vmar->devinfo[dev_id].size/BDRV_SECTOR_SIZE;
 683 +        } else {
 684 +            max_sector = G_MAXINT64;
 685 +            if (cluster_num != vmar->vmstate_clusters) {
 686 +                error_setg(errp, "found out of order vmstate data");
 687 +                return -1;
 688 +            }
 689 +            vmar->vmstate_clusters++;
 690 +        }
 691 +
 692 +        /* try to write whole clusters to speedup restore */
 693 +        if (mask == 0xffff) {
 694 +            if ((start + VMA_CLUSTER_SIZE) > extent_size) {
 695 +                error_setg(errp, "short vma extent - too many blocks");
 696 +                return -1;
 697 +            }
 698 +            int64_t sector_num = (cluster_num * VMA_CLUSTER_SIZE) /
 699 +                BDRV_SECTOR_SIZE;
 700 +            int64_t end_sector = sector_num +
 701 +                VMA_CLUSTER_SIZE/BDRV_SECTOR_SIZE;
 702 +
 703 +            if (end_sector > max_sector) {
 704 +                end_sector = max_sector;
 705 +            }
 706 +
 707 +            if (end_sector <= sector_num) {
 708 +                error_setg(errp, "got wrong block address - write bejond end");
 709 +                return -1;
 710 +            }
 711 +
 712 +            int nb_sectors = end_sector - sector_num;
 713 +            if (restore_write_data(vmar, dev_id, bs, vmstate_fd, buf + start,
 714 +                                   sector_num, nb_sectors, errp) < 0) {
 715 +                return -1;
 716 +            }
 717 +
 718 +            start += VMA_CLUSTER_SIZE;
 719 +        } else {
 720 +            int j;
 721 +            int bit = 1;
 722 +
 723 +            for (j = 0; j < 16; j++) {
 724 +                int64_t sector_num = (cluster_num*VMA_CLUSTER_SIZE +
 725 +                                      j*VMA_BLOCK_SIZE)/BDRV_SECTOR_SIZE;
 726 +
 727 +                int64_t end_sector = sector_num +
 728 +                    VMA_BLOCK_SIZE/BDRV_SECTOR_SIZE;
 729 +                if (end_sector > max_sector) {
 730 +                    end_sector = max_sector;
 731 +                }
 732 +
 733 +                if (mask & bit) {
 734 +                    if ((start + VMA_BLOCK_SIZE) > extent_size) {
 735 +                        error_setg(errp, "short vma extent - too many blocks");
 736 +                        return -1;
 737 +                    }
 738 +
 739 +                    if (end_sector <= sector_num) {
 740 +                        error_setg(errp, "got wrong block address - "
 741 +                                   "write bejond end");
 742 +                        return -1;
 743 +                    }
 744 +
 745 +                    int nb_sectors = end_sector - sector_num;
 746 +                    if (restore_write_data(vmar, dev_id, bs, vmstate_fd,
 747 +                                           buf + start, sector_num,
 748 +                                           nb_sectors, errp) < 0) {
 749 +                        return -1;
 750 +                    }
 751 +
 752 +                    start += VMA_BLOCK_SIZE;
 753 +
 754 +                } else {
 755 +
 756 +                    if (rstate->write_zeroes & (end_sector > sector_num)) {
 757 +                        /* Todo: use bdrv_co_write_zeroes (but that need to
 758 +                         * be run inside coroutine?)
 759 +                         */
 760 +                        int nb_sectors = end_sector - sector_num;
 761 +                        if (restore_write_data(vmar, dev_id, bs, vmstate_fd,
 762 +                                              zero_vma_block, sector_num,
 763 +                                               nb_sectors, errp) < 0) {
 764 +                            return -1;
 765 +                        }
 766 +                    }
 767 +                }
 768 +
 769 +                bit = bit << 1;
 770 +            }
 771 +        }
 772 +    }
 773 +
 774 +    if (start != extent_size) {
 775 +        error_setg(errp, "vma extent error - missing blocks");
 776 +        return -1;
 777 +    }
 778 +
 779 +    return 0;
 780 +}
 781 +
 782 +int vma_reader_restore(VmaReader *vmar, int vmstate_fd, Error **errp)
 783 +{
 784 +    assert(vmar);
 785 +    assert(vmar->head_data);
 786 +
 787 +    int ret = 0;
 788 +    unsigned char buf[VMA_MAX_EXTENT_SIZE];
 789 +    int buf_pos = 0;
 790 +    unsigned char md5sum[16];
 791 +    VmaHeader *h = (VmaHeader *)vmar->head_data;
 792 +
 793 +
 794 +    while (1) {
 795 +        int bytes = full_read(vmar->fd, buf + buf_pos, sizeof(buf) - buf_pos);
 796 +        if (bytes < 0) {
 797 +            error_setg(errp, "read failed - %s", strerror(errno));
 798 +            return -1;
 799 +        }
 800 +
 801 +        buf_pos += bytes;
 802 +
 803 +        if (!buf_pos) {
 804 +            break; /* EOF */
 805 +        }
 806 +
 807 +        if (buf_pos < VMA_EXTENT_HEADER_SIZE) {
 808 +            error_setg(errp, "read short extent (%d bytes)", buf_pos);
 809 +            return -1;
 810 +        }
 811 +
 812 +        VmaExtentHeader *ehead = (VmaExtentHeader *)buf;
 813 +
 814 +        /* extract md5sum */
 815 +        memcpy(md5sum, ehead->md5sum, sizeof(ehead->md5sum));
 816 +        memset(ehead->md5sum, 0, sizeof(ehead->md5sum));
 817 +
 818 +        g_checksum_reset(vmar->md5csum);
 819 +        g_checksum_update(vmar->md5csum, buf, VMA_EXTENT_HEADER_SIZE);
 820 +        gsize csize = 16;
 821 +        g_checksum_get_digest(vmar->md5csum, ehead->md5sum, &csize);
 822 +
 823 +        if (memcmp(md5sum, ehead->md5sum, 16) != 0) {
 824 +            error_setg(errp, "wrong vma extent header chechsum");
 825 +            return -1;
 826 +        }
 827 +
 828 +        if (memcmp(h->uuid, ehead->uuid, sizeof(ehead->uuid)) != 0) {
 829 +            error_setg(errp, "wrong vma extent uuid");
 830 +            return -1;
 831 +        }
 832 +
 833 +        if (ehead->magic != VMA_EXTENT_MAGIC || ehead->reserved1 != 0) {
 834 +            error_setg(errp, "wrong vma extent header magic");
 835 +            return -1;
 836 +        }
 837 +
 838 +        int block_count = GUINT16_FROM_BE(ehead->block_count);
 839 +        int extent_size = VMA_EXTENT_HEADER_SIZE + block_count*VMA_BLOCK_SIZE;
 840 +
 841 +        if (buf_pos < extent_size) {
 842 +            error_setg(errp, "short vma extent (%d < %d)", buf_pos,
 843 +                       extent_size);
 844 +            return -1;
 845 +        }
 846 +
 847 +        if (restore_extent(vmar, buf, extent_size, vmstate_fd, errp) < 0) {
 848 +            return -1;
 849 +        }
 850 +
 851 +        if (buf_pos > extent_size) {
 852 +            memmove(buf, buf + extent_size, buf_pos - extent_size);
 853 +            buf_pos = buf_pos - extent_size;
 854 +        } else {
 855 +            buf_pos = 0;
 856 +        }
 857 +    }
 858 +
 859 +    bdrv_drain_all();
 860 +
 861 +    int i;
 862 +    for (i = 1; i < 256; i++) {
 863 +        VmaRestoreState *rstate = &vmar->rstate[i];
 864 +        if (!rstate->bs) {
 865 +            continue;
 866 +        }
 867 +
 868 +        if (bdrv_flush(rstate->bs) < 0) {
 869 +            error_setg(errp, "vma bdrv_flush %s failed",
 870 +                       vmar->devinfo[i].devname);
 871 +            return -1;
 872 +        }
 873 +
 874 +        if (vmar->devinfo[i].size &&
 875 +            (strcmp(vmar->devinfo[i].devname, "vmstate") != 0)) {
 876 +            assert(rstate->bitmap);
 877 +
 878 +            int64_t cluster_num, end;
 879 +
 880 +            end = (vmar->devinfo[i].size + VMA_CLUSTER_SIZE - 1) /
 881 +                VMA_CLUSTER_SIZE;
 882 +
 883 +            for (cluster_num = 0; cluster_num < end; cluster_num++) {
 884 +                if (!vma_reader_get_bitmap(rstate, cluster_num)) {
 885 +                    error_setg(errp, "detected missing cluster %zd "
 886 +                               "for stream %s", cluster_num,
 887 +                               vmar->devinfo[i].devname);
 888 +                    return -1;
 889 +                }
 890 +            }
 891 +        }
 892 +    }
 893 +
 894 +    return ret;
 895 +}
 896 +
 897 diff --git a/vma-writer.c b/vma-writer.c
 898 new file mode 100644
 899 index 0000000..c1e1afe
 900 --- /dev/null
 901 +++ b/vma-writer.c
 902 @@ -0,0 +1,920 @@
 903 +/*
 904 + * VMA: Virtual Machine Archive
 905 + *
 906 + * Copyright (C) 2012 Proxmox Server Solutions
 907 + *
 908 + * Authors:
 909 + *  Dietmar Maurer (dietmar@proxmox.com)
 910 + *
 911 + * This work is licensed under the terms of the GNU GPL, version 2 or later.
 912 + * See the COPYING file in the top-level directory.
 913 + *
 914 + */
 915 +
 916 +#include <stdio.h>
 917 +#include <errno.h>
 918 +#include <unistd.h>
 919 +#include <stdio.h>
 920 +#include <string.h>
 921 +#include <sys/types.h>
 922 +#include <sys/stat.h>
 923 +#include <fcntl.h>
 924 +#include <glib.h>
 925 +#include <uuid/uuid.h>
 926 +
 927 +#include "qemu-common.h"
 928 +#include "qemu_socket.h"
 929 +#include "qemu-coroutine.h"
 930 +#include "qemu-aio.h"
 931 +#include "qemu/ratelimit.h"
 932 +#include "vma.h"
 933 +#include "block.h"
 934 +#include "monitor.h"
 935 +
 936 +#define DEBUG_VMA 0
 937 +
 938 +#define DPRINTF(fmt, ...)\
 939 +    do { if (DEBUG_VMA) { printf("vma: " fmt, ## __VA_ARGS__); } } while (0)
 940 +
 941 +#define WRITE_BUFFERS 5
 942 +
 943 +typedef struct VmaAIOCB VmaAIOCB;
 944 +struct VmaAIOCB {
 945 +    VmaWriter *vmaw;
 946 +    unsigned char buffer[VMA_MAX_EXTENT_SIZE];
 947 +    size_t bytes;
 948 +    Coroutine *co;
 949 +};
 950 +
 951 +struct VmaWriter {
 952 +    int fd;
 953 +    FILE *cmd;
 954 +    int status;
 955 +    char errmsg[8192];
 956 +    uuid_t uuid;
 957 +    bool header_written;
 958 +    bool closed;
 959 +
 960 +    /* we always write extents */
 961 +    unsigned char outbuf[VMA_MAX_EXTENT_SIZE];
 962 +    int outbuf_pos; /* in bytes */
 963 +    int outbuf_count; /* in VMA_BLOCKS */
 964 +    uint64_t outbuf_block_info[VMA_BLOCKS_PER_EXTENT];
 965 +
 966 +    VmaAIOCB aiocbs[WRITE_BUFFERS];
 967 +    CoQueue wqueue;
 968 +
 969 +    GChecksum *md5csum;
 970 +    CoMutex writer_lock;
 971 +    CoMutex flush_lock;
 972 +    Coroutine *co_writer;
 973 +    RateLimit limit;
 974 +    uint64_t delay_ns;
 975 +
 976 +    /* drive informations */
 977 +    VmaStreamInfo stream_info[256];
 978 +    guint stream_count;
 979 +
 980 +    guint8 vmstate_stream;
 981 +    uint32_t vmstate_clusters;
 982 +
 983 +    /* header blob table */
 984 +    char *header_blob_table;
 985 +    uint32_t header_blob_table_size;
 986 +    uint32_t header_blob_table_pos;
 987 +
 988 +    /* store for config blobs */
 989 +    uint32_t config_names[VMA_MAX_CONFIGS]; /* offset into blob_buffer table */
 990 +    uint32_t config_data[VMA_MAX_CONFIGS];  /* offset into blob_buffer table */
 991 +    uint32_t config_count;
 992 +};
 993 +
 994 +void vma_writer_set_error(VmaWriter *vmaw, const char *fmt, ...)
 995 +{
 996 +    va_list ap;
 997 +
 998 +    if (vmaw->status < 0) {
 999 +        return;
1000 +    }
1001 +
1002 +    vmaw->status = -1;
1003 +
1004 +    va_start(ap, fmt);
1005 +    g_vsnprintf(vmaw->errmsg, sizeof(vmaw->errmsg), fmt, ap);
1006 +    va_end(ap);
1007 +
1008 +    DPRINTF("vma_writer_set_error: %s\n", vmaw->errmsg);
1009 +}
1010 +
1011 +static uint32_t allocate_header_blob(VmaWriter *vmaw, const char *data,
1012 +                                     size_t len)
1013 +{
1014 +    if (len > 65535) {
1015 +        return 0;
1016 +    }
1017 +
1018 +    if (!vmaw->header_blob_table ||
1019 +        (vmaw->header_blob_table_size <
1020 +         (vmaw->header_blob_table_pos + len + 2))) {
1021 +        int newsize = vmaw->header_blob_table_size + ((len + 2 + 511)/512)*512;
1022 +
1023 +        vmaw->header_blob_table = g_realloc(vmaw->header_blob_table, newsize);
1024 +        memset(vmaw->header_blob_table + vmaw->header_blob_table_size,
1025 +               0, newsize - vmaw->header_blob_table_size);
1026 +        vmaw->header_blob_table_size = newsize;
1027 +    }
1028 +
1029 +    uint32_t cpos = vmaw->header_blob_table_pos;
1030 +    vmaw->header_blob_table[cpos] = len & 255;
1031 +    vmaw->header_blob_table[cpos+1] = (len >> 8) & 255;
1032 +    memcpy(vmaw->header_blob_table + cpos + 2, data, len);
1033 +    vmaw->header_blob_table_pos += len + 2;
1034 +    return cpos;
1035 +}
1036 +
1037 +static uint32_t allocate_header_string(VmaWriter *vmaw, const char *str)
1038 +{
1039 +    assert(vmaw);
1040 +
1041 +    size_t len = strlen(str) + 1;
1042 +
1043 +    return allocate_header_blob(vmaw, str, len);
1044 +}
1045 +
1046 +int vma_writer_add_config(VmaWriter *vmaw, const char *name, gpointer data,
1047 +                          gsize len)
1048 +{
1049 +    assert(vmaw);
1050 +    assert(!vmaw->header_written);
1051 +    assert(vmaw->config_count < VMA_MAX_CONFIGS);
1052 +    assert(name);
1053 +    assert(data);
1054 +    assert(len);
1055 +
1056 +    uint32_t name_ptr = allocate_header_string(vmaw, name);
1057 +    if (!name_ptr) {
1058 +        return -1;
1059 +    }
1060 +
1061 +    uint32_t data_ptr = allocate_header_blob(vmaw, data, len);
1062 +    if (!data_ptr) {
1063 +        return -1;
1064 +    }
1065 +
1066 +    vmaw->config_names[vmaw->config_count] = name_ptr;
1067 +    vmaw->config_data[vmaw->config_count] = data_ptr;
1068 +
1069 +    vmaw->config_count++;
1070 +
1071 +    return 0;
1072 +}
1073 +
1074 +int vma_writer_register_stream(VmaWriter *vmaw, const char *devname,
1075 +                               size_t size)
1076 +{
1077 +    assert(vmaw);
1078 +    assert(devname);
1079 +    assert(!vmaw->status);
1080 +
1081 +    if (vmaw->header_written) {
1082 +        vma_writer_set_error(vmaw, "vma_writer_register_stream: header "
1083 +                             "already written");
1084 +        return -1;
1085 +    }
1086 +
1087 +    guint n = vmaw->stream_count + 1;
1088 +
1089 +    /* we can have dev_ids form 1 to 255 (0 reserved)
1090 +     * 255(-1) reseverd for safety
1091 +     */
1092 +    if (n > 254) {
1093 +        vma_writer_set_error(vmaw, "vma_writer_register_stream: "
1094 +                             "too many drives");
1095 +        return -1;
1096 +    }
1097 +
1098 +    if (size <= 0) {
1099 +        vma_writer_set_error(vmaw, "vma_writer_register_stream: "
1100 +                             "got strange size %zd", size);
1101 +        return -1;
1102 +    }
1103 +
1104 +    DPRINTF("vma_writer_register_stream %s %zu %d\n", devname, size, n);
1105 +
1106 +    vmaw->stream_info[n].devname = g_strdup(devname);
1107 +    vmaw->stream_info[n].size = size;
1108 +
1109 +    vmaw->stream_info[n].cluster_count = (size + VMA_CLUSTER_SIZE - 1) /
1110 +        VMA_CLUSTER_SIZE;
1111 +
1112 +    vmaw->stream_count = n;
1113 +
1114 +    if (strcmp(devname, "vmstate") == 0) {
1115 +        vmaw->vmstate_stream = n;
1116 +    }
1117 +
1118 +    return n;
1119 +}
1120 +
1121 +static void vma_co_continue_write(void *opaque)
1122 +{
1123 +    VmaWriter *vmaw = opaque;
1124 +
1125 +    qemu_aio_set_fd_handler(vmaw->fd, NULL, NULL, NULL, NULL);
1126 +
1127 +    DPRINTF("vma_co_continue_write\n");
1128 +    qemu_coroutine_enter(vmaw->co_writer, NULL);
1129 +}
1130 +
1131 +static ssize_t coroutine_fn
1132 +vma_co_write(VmaWriter *vmaw, const void *buf, size_t bytes)
1133 +{
1134 +    size_t done = 0;
1135 +    ssize_t ret;
1136 +
1137 +    /* atomic writes (we cannot interleave writes) */
1138 +    qemu_co_mutex_lock(&vmaw->writer_lock);
1139 +
1140 +    DPRINTF("vma_co_write enter %zd\n", bytes);
1141 +
1142 +    while (done < bytes) {
1143 +        ret = write(vmaw->fd, buf + done, bytes - done);
1144 +        if (ret > 0) {
1145 +            done += ret;
1146 +            DPRINTF("vma_co_write written %zd %zd\n", done, ret);
1147 +        } else if (ret < 0) {
1148 +            if (errno == EAGAIN || errno == EWOULDBLOCK) {
1149 +                DPRINTF("vma_co_write yield %zd\n", done);
1150 +
1151 +                vmaw->co_writer = qemu_coroutine_self();
1152 +                qemu_aio_set_fd_handler(vmaw->fd, NULL, vma_co_continue_write,
1153 +                                        NULL, vmaw);
1154 +
1155 +                qemu_coroutine_yield();
1156 +                DPRINTF("vma_co_write restart %zd\n", done);
1157 +            } else {
1158 +                vma_writer_set_error(vmaw, "vma_co_write write error - %s",
1159 +                                     strerror(errno));
1160 +                done = -1; /* always return failure for partial writes */
1161 +                break;
1162 +            }
1163 +        } else if (ret == 0) {
1164 +            /* should not happen - simply try again */
1165 +        }
1166 +    }
1167 +
1168 +    qemu_co_mutex_unlock(&vmaw->writer_lock);
1169 +
1170 +    DPRINTF("vma_co_write leave %zd\n", done);
1171 +    return done;
1172 +}
1173 +
1174 +static void coroutine_fn vma_co_writer_task(void *opaque)
1175 +{
1176 +    VmaAIOCB *cb = opaque;
1177 +
1178 +    DPRINTF("vma_co_writer_task start\n");
1179 +
1180 +    int64_t done = vma_co_write(cb->vmaw, cb->buffer, cb->bytes);
1181 +    DPRINTF("vma_co_writer_task write done %zd\n", done);
1182 +
1183 +    if (done != cb->bytes) {
1184 +        DPRINTF("vma_co_writer_task failed write %zd %zd", cb->bytes, done);
1185 +        vma_writer_set_error(cb->vmaw, "vma_co_writer_task failed write %zd",
1186 +                             done);
1187 +    }
1188 +
1189 +    cb->bytes = 0;
1190 +
1191 +    qemu_co_queue_next(&cb->vmaw->wqueue);
1192 +
1193 +    DPRINTF("vma_co_writer_task end\n");
1194 +}
1195 +
1196 +static void coroutine_fn vma_queue_flush(VmaWriter *vmaw)
1197 +{
1198 +    DPRINTF("vma_queue_flush enter\n");
1199 +
1200 +    assert(vmaw);
1201 +
1202 +    while (1) {
1203 +        int i;
1204 +        VmaAIOCB *cb = NULL;
1205 +        for (i = 0; i < WRITE_BUFFERS; i++) {
1206 +            if (vmaw->aiocbs[i].bytes) {
1207 +                cb = &vmaw->aiocbs[i];
1208 +                DPRINTF("FOUND USED AIO BUFFER %d %zd\n", i,
1209 +                        vmaw->aiocbs[i].bytes);
1210 +                break;
1211 +            }
1212 +        }
1213 +        if (!cb) {
1214 +            break;
1215 +        }
1216 +        qemu_co_queue_wait(&vmaw->wqueue);
1217 +    }
1218 +
1219 +    DPRINTF("vma_queue_flush leave\n");
1220 +}
1221 +
1222 +/**
1223 + * NOTE: pipe buffer size in only 4096 bytes on linux (see 'ulimit -a')
1224 + * So we need to create a coroutione to allow 'parallel' execution.
1225 + */
1226 +static ssize_t coroutine_fn
1227 +vma_queue_write(VmaWriter *vmaw, const void *buf, size_t bytes)
1228 +{
1229 +    DPRINTF("vma_queue_write enter %zd\n", bytes);
1230 +
1231 +    assert(vmaw);
1232 +    assert(buf);
1233 +    assert(bytes <= VMA_MAX_EXTENT_SIZE);
1234 +
1235 +    VmaAIOCB *cb = NULL;
1236 +    while (!cb) {
1237 +        int i;
1238 +        for (i = 0; i < WRITE_BUFFERS; i++) {
1239 +            if (!vmaw->aiocbs[i].bytes) {
1240 +                cb = &vmaw->aiocbs[i];
1241 +                break;
1242 +            }
1243 +        }
1244 +        if (!cb) {
1245 +            qemu_co_queue_wait(&vmaw->wqueue);
1246 +        }
1247 +    }
1248 +
1249 +    memcpy(cb->buffer, buf, bytes);
1250 +    cb->bytes = bytes;
1251 +    cb->vmaw = vmaw;
1252 +
1253 +    DPRINTF("vma_queue_write start %zd\n", bytes);
1254 +    cb->co = qemu_coroutine_create(vma_co_writer_task);
1255 +    qemu_coroutine_enter(cb->co, cb);
1256 +
1257 +    DPRINTF("vma_queue_write leave\n");
1258 +
1259 +    return bytes;
1260 +}
1261 +
1262 +VmaWriter *vma_writer_create(const char *filename, uuid_t uuid, int64_t speed,
1263 +                             Error **errp)
1264 +{
1265 +    const char *p;
1266 +
1267 +    assert(sizeof(VmaHeader) == (4096 + 8192));
1268 +    assert(sizeof(VmaExtentHeader) == 512);
1269 +
1270 +    VmaWriter *vmaw = g_new0(VmaWriter, 1);
1271 +    vmaw->fd = -1;
1272 +
1273 +    vmaw->md5csum = g_checksum_new(G_CHECKSUM_MD5);
1274 +    if (!vmaw->md5csum) {
1275 +        error_setg(errp, "can't allocate cmsum\n");
1276 +        goto err;
1277 +    }
1278 +
1279 +    if (strstart(filename, "exec:", &p)) {
1280 +        vmaw->cmd = popen(p, "w");
1281 +        if (vmaw->cmd == NULL) {
1282 +            error_setg(errp, "can't popen command '%s' - %s\n", p,
1283 +                       strerror(errno));
1284 +            goto err;
1285 +        }
1286 +        vmaw->fd = fileno(vmaw->cmd);
1287 +        socket_set_nonblock(vmaw->fd);
1288 +
1289 +    } else {
1290 +        struct stat st;
1291 +        int oflags;
1292 +       const char *tmp_id_str;
1293 +
1294 +        if ((stat(filename, &st) == 0) && S_ISFIFO(st.st_mode)) {
1295 +            oflags = O_NONBLOCK|O_WRONLY;
1296 +           vmaw->fd = qemu_open(filename, oflags, 0644);
1297 +        } else if (strstart(filename, "/dev/fdset/", &tmp_id_str)) {
1298 +            oflags = O_NONBLOCK|O_WRONLY;
1299 +            vmaw->fd = qemu_open(filename, oflags, 0644);
1300 +       } else if (strstart(filename, "/dev/fdname/", &tmp_id_str)) {
1301 +            vmaw->fd = monitor_get_fd(cur_mon, tmp_id_str, errp);
1302 +            if (vmaw->fd < 0) {
1303 +                goto err;
1304 +            }
1305 +       } else  {
1306 +            oflags = O_NONBLOCK|O_WRONLY|O_CREAT|O_EXCL;
1307 +           vmaw->fd = qemu_open(filename, oflags, 0644);
1308 +        }
1309 +
1310 +        if (vmaw->fd < 0) {
1311 +            error_setg(errp, "can't open file %s - %s\n", filename,
1312 +                       strerror(errno));
1313 +            goto err;
1314 +        }
1315 +    }
1316 +
1317 +    vmaw->outbuf_count = 0;
1318 +    vmaw->outbuf_pos = VMA_EXTENT_HEADER_SIZE;
1319 +
1320 +    vmaw->header_blob_table_pos = 1; /* start at pos 1 */
1321 +
1322 +    qemu_co_mutex_init(&vmaw->writer_lock);
1323 +    qemu_co_mutex_init(&vmaw->flush_lock);
1324 +    qemu_co_queue_init(&vmaw->wqueue);
1325 +
1326 +    uuid_copy(vmaw->uuid, uuid);
1327 +
1328 +    if (speed <= 0) {
1329 +        speed = 10*1024*1024*1024LLU; /* default 10GB/s */
1330 +    }
1331 +
1332 +    ratelimit_set_speed(&vmaw->limit, speed, 100000000ULL /* 0.1 sec */);
1333 +
1334 +    return vmaw;
1335 +
1336 +err:
1337 +    if (vmaw) {
1338 +        if (vmaw->cmd) {
1339 +            pclose(vmaw->cmd);
1340 +        } else if (vmaw->fd >= 0) {
1341 +            close(vmaw->fd);
1342 +        }
1343 +
1344 +        if (vmaw->md5csum) {
1345 +            g_checksum_free(vmaw->md5csum);
1346 +        }
1347 +
1348 +        g_free(vmaw);
1349 +    }
1350 +
1351 +    return NULL;
1352 +}
1353 +
1354 +static int coroutine_fn vma_write_header(VmaWriter *vmaw)
1355 +{
1356 +    assert(vmaw);
1357 +    int header_clusters = 8;
1358 +    char buf[65536*header_clusters];
1359 +    VmaHeader *head = (VmaHeader *)buf;
1360 +
1361 +    int i;
1362 +
1363 +    DPRINTF("VMA WRITE HEADER\n");
1364 +
1365 +    if (vmaw->status < 0) {
1366 +        return vmaw->status;
1367 +    }
1368 +
1369 +    memset(buf, 0, sizeof(buf));
1370 +
1371 +    head->magic = VMA_MAGIC;
1372 +    head->version = GUINT32_TO_BE(1); /* v1 */
1373 +    memcpy(head->uuid, vmaw->uuid, 16);
1374 +
1375 +    time_t ctime = time(NULL);
1376 +    head->ctime = GUINT64_TO_BE(ctime);
1377 +
1378 +    if (!vmaw->stream_count) {
1379 +        return -1;
1380 +    }
1381 +
1382 +    for (i = 0; i < VMA_MAX_CONFIGS; i++) {
1383 +        head->config_names[i] = GUINT32_TO_BE(vmaw->config_names[i]);
1384 +        head->config_data[i] = GUINT32_TO_BE(vmaw->config_data[i]);
1385 +    }
1386 +
1387 +    /* 32 bytes per device (12 used currently) = 8192 bytes max */
1388 +    for (i = 1; i <= 254; i++) {
1389 +        VmaStreamInfo *si = &vmaw->stream_info[i];
1390 +        if (si->size) {
1391 +            assert(si->devname);
1392 +            uint32_t devname_ptr = allocate_header_string(vmaw, si->devname);
1393 +            if (!devname_ptr) {
1394 +                return -1;
1395 +            }
1396 +            head->dev_info[i].devname_ptr = GUINT32_TO_BE(devname_ptr);
1397 +            head->dev_info[i].size = GUINT64_TO_BE(si->size);
1398 +        }
1399 +    }
1400 +
1401 +    uint32_t header_size = sizeof(VmaHeader) + vmaw->header_blob_table_size;
1402 +    head->header_size = GUINT32_TO_BE(header_size);
1403 +
1404 +    if (header_size > sizeof(buf)) {
1405 +        return -1; /* just to be sure */
1406 +    }
1407 +
1408 +    uint32_t blob_buffer_offset = sizeof(VmaHeader);
1409 +    memcpy(buf + blob_buffer_offset, vmaw->header_blob_table,
1410 +           vmaw->header_blob_table_size);
1411 +    head->blob_buffer_offset = GUINT32_TO_BE(blob_buffer_offset);
1412 +    head->blob_buffer_size = GUINT32_TO_BE(vmaw->header_blob_table_pos);
1413 +
1414 +    g_checksum_reset(vmaw->md5csum);
1415 +    g_checksum_update(vmaw->md5csum, (const guchar *)buf, header_size);
1416 +    gsize csize = 16;
1417 +    g_checksum_get_digest(vmaw->md5csum, (guint8 *)(head->md5sum), &csize);
1418 +
1419 +    return vma_queue_write(vmaw, buf, header_size);
1420 +}
1421 +
1422 +static int coroutine_fn vma_writer_flush(VmaWriter *vmaw)
1423 +{
1424 +    assert(vmaw);
1425 +
1426 +    int ret;
1427 +    int i;
1428 +
1429 +    if (vmaw->status < 0) {
1430 +        return vmaw->status;
1431 +    }
1432 +
1433 +    if (!vmaw->header_written) {
1434 +        vmaw->header_written = true;
1435 +        ret = vma_write_header(vmaw);
1436 +        if (ret < 0) {
1437 +            vma_writer_set_error(vmaw, "vma_writer_flush: write header failed");
1438 +            return ret;
1439 +        }
1440 +    }
1441 +
1442 +    DPRINTF("VMA WRITE FLUSH %d %d\n", vmaw->outbuf_count, vmaw->outbuf_pos);
1443 +
1444 +
1445 +    VmaExtentHeader *ehead = (VmaExtentHeader *)vmaw->outbuf;
1446 +
1447 +    ehead->magic = VMA_EXTENT_MAGIC;
1448 +    ehead->reserved1 = 0;
1449 +
1450 +    for (i = 0; i < VMA_BLOCKS_PER_EXTENT; i++) {
1451 +        ehead->blockinfo[i] = GUINT64_TO_BE(vmaw->outbuf_block_info[i]);
1452 +    }
1453 +
1454 +    guint16 block_count = (vmaw->outbuf_pos - VMA_EXTENT_HEADER_SIZE) /
1455 +        VMA_BLOCK_SIZE;
1456 +
1457 +    ehead->block_count = GUINT16_TO_BE(block_count);
1458 +
1459 +    memcpy(ehead->uuid, vmaw->uuid, sizeof(ehead->uuid));
1460 +    memset(ehead->md5sum, 0, sizeof(ehead->md5sum));
1461 +
1462 +    g_checksum_reset(vmaw->md5csum);
1463 +    g_checksum_update(vmaw->md5csum, vmaw->outbuf, VMA_EXTENT_HEADER_SIZE);
1464 +    gsize csize = 16;
1465 +    g_checksum_get_digest(vmaw->md5csum, ehead->md5sum, &csize);
1466 +
1467 +    int bytes = vmaw->outbuf_pos;
1468 +    ret = vma_queue_write(vmaw, vmaw->outbuf, bytes);
1469 +    if (ret != bytes) {
1470 +        vma_writer_set_error(vmaw, "vma_writer_flush: failed write");
1471 +    }
1472 +
1473 +    vmaw->outbuf_count = 0;
1474 +    vmaw->outbuf_pos = VMA_EXTENT_HEADER_SIZE;
1475 +
1476 +    for (i = 0; i < VMA_BLOCKS_PER_EXTENT; i++) {
1477 +        vmaw->outbuf_block_info[i] = 0;
1478 +    }
1479 +
1480 +    return vmaw->status;
1481 +}
1482 +
1483 +static int vma_count_open_streams(VmaWriter *vmaw)
1484 +{
1485 +    g_assert(vmaw != NULL);
1486 +
1487 +    int i;
1488 +    int open_drives = 0;
1489 +    for (i = 0; i <= 255; i++) {
1490 +        if (vmaw->stream_info[i].size && !vmaw->stream_info[i].finished) {
1491 +            open_drives++;
1492 +        }
1493 +    }
1494 +
1495 +    return open_drives;
1496 +}
1497 +
1498 +/**
1499 + * all jobs should call this when there is no more data
1500 + * Returns: number of remaining stream (0 ==> finished)
1501 + */
1502 +int coroutine_fn
1503 +vma_writer_close_stream(VmaWriter *vmaw, uint8_t dev_id)
1504 +{
1505 +    g_assert(vmaw != NULL);
1506 +
1507 +    DPRINTF("vma_writer_set_status %d\n", dev_id);
1508 +    if (!vmaw->stream_info[dev_id].size) {
1509 +        vma_writer_set_error(vmaw, "vma_writer_close_stream: "
1510 +                             "no such stream %d", dev_id);
1511 +        return -1;
1512 +    }
1513 +    if (vmaw->stream_info[dev_id].finished) {
1514 +        vma_writer_set_error(vmaw, "vma_writer_close_stream: "
1515 +                             "stream already closed %d", dev_id);
1516 +        return -1;
1517 +    }
1518 +
1519 +    vmaw->stream_info[dev_id].finished = true;
1520 +
1521 +    int open_drives = vma_count_open_streams(vmaw);
1522 +
1523 +    if (open_drives <= 0) {
1524 +        DPRINTF("vma_writer_set_status all drives completed\n");
1525 +        qemu_co_mutex_lock(&vmaw->flush_lock);
1526 +        int ret = vma_writer_flush(vmaw);
1527 +        qemu_co_mutex_unlock(&vmaw->flush_lock);
1528 +        if (ret < 0) {
1529 +            vma_writer_set_error(vmaw, "vma_writer_close_stream: flush failed");
1530 +        }
1531 +    }
1532 +
1533 +    return open_drives;
1534 +}
1535 +
1536 +int vma_writer_get_status(VmaWriter *vmaw, VmaStatus *status)
1537 +{
1538 +    int i;
1539 +
1540 +    g_assert(vmaw != NULL);
1541 +
1542 +    if (status) {
1543 +        status->status = vmaw->status;
1544 +        g_strlcpy(status->errmsg, vmaw->errmsg, sizeof(status->errmsg));
1545 +        for (i = 0; i <= 255; i++) {
1546 +            status->stream_info[i] = vmaw->stream_info[i];
1547 +        }
1548 +
1549 +        uuid_unparse_lower(vmaw->uuid, status->uuid_str);
1550 +    }
1551 +
1552 +    status->closed = vmaw->closed;
1553 +
1554 +    return vmaw->status;
1555 +}
1556 +
1557 +static int vma_writer_get_buffer(VmaWriter *vmaw)
1558 +{
1559 +
1560 +    /* wait until buffer is available */
1561 +    while (vmaw->outbuf_count >= (VMA_BLOCKS_PER_EXTENT - 1)) {
1562 +        int res = 0;
1563 +
1564 +        qemu_co_mutex_lock(&vmaw->flush_lock);
1565 +        res = vma_writer_flush(vmaw);
1566 +        qemu_co_mutex_unlock(&vmaw->flush_lock);
1567 +
1568 +        if (res < 0) {
1569 +            vma_writer_set_error(vmaw, "vma_writer_get_buffer: flush failed");
1570 +            return -1;
1571 +        }
1572 +    }
1573 +
1574 +    return 0;
1575 +}
1576 +
1577 +
1578 +int64_t coroutine_fn
1579 +vma_writer_write(VmaWriter *vmaw, uint8_t dev_id, int64_t cluster_num,
1580 +                 unsigned char *buf, size_t *zero_bytes)
1581 +{
1582 +    g_assert(vmaw != NULL);
1583 +    g_assert(zero_bytes != NULL);
1584 +
1585 +    *zero_bytes = 0;
1586 +
1587 +    if (vmaw->status < 0) {
1588 +        return vmaw->status;
1589 +    }
1590 +
1591 +    if (!dev_id || !vmaw->stream_info[dev_id].size) {
1592 +        vma_writer_set_error(vmaw, "vma_writer_write: "
1593 +                             "no such stream %d", dev_id);
1594 +        return -1;
1595 +    }
1596 +
1597 +    if (vmaw->stream_info[dev_id].finished) {
1598 +        vma_writer_set_error(vmaw, "vma_writer_write: "
1599 +                             "stream already closed %d", dev_id);
1600 +        return -1;
1601 +    }
1602 +
1603 +
1604 +    if (cluster_num >= (((uint64_t)1)<<32)) {
1605 +        vma_writer_set_error(vmaw, "vma_writer_write: "
1606 +                             "cluster number out of range");
1607 +        return -1;
1608 +    }
1609 +
1610 +    if (dev_id == vmaw->vmstate_stream) {
1611 +        if (cluster_num != vmaw->vmstate_clusters) {
1612 +            vma_writer_set_error(vmaw, "vma_writer_write: "
1613 +                                 "non sequential vmstate write");
1614 +        }
1615 +        vmaw->vmstate_clusters++;
1616 +    } else if (cluster_num >= vmaw->stream_info[dev_id].cluster_count) {
1617 +        vma_writer_set_error(vmaw, "vma_writer_write: cluster number too big");
1618 +        return -1;
1619 +    }
1620 +
1621 +    /* wait until buffer is available */
1622 +    if (vma_writer_get_buffer(vmaw) < 0) {
1623 +        vma_writer_set_error(vmaw, "vma_writer_write: "
1624 +                             "vma_writer_get_buffer failed");
1625 +        return -1;
1626 +    }
1627 +
1628 +    DPRINTF("VMA WRITE %zd\n", cluster_num);
1629 +
1630 +    int i;
1631 +    int bit = 1;
1632 +    uint16_t mask = 0;
1633 +    for (i = 0; i < 16; i++) {
1634 +        unsigned char *vmablock = buf + (i*VMA_BLOCK_SIZE);
1635 +        if (buffer_is_zero(vmablock, VMA_BLOCK_SIZE)) {
1636 +            DPRINTF("VMA WRITE %zd ZERO BLOCK %d\n", cluster_num, i);
1637 +            vmaw->stream_info[dev_id].zero_bytes += VMA_BLOCK_SIZE;
1638 +            *zero_bytes += VMA_BLOCK_SIZE;
1639 +        } else {
1640 +            mask |= bit;
1641 +            memcpy(vmaw->outbuf + vmaw->outbuf_pos, vmablock, VMA_BLOCK_SIZE);
1642 +            vmaw->outbuf_pos += VMA_BLOCK_SIZE;
1643 +
1644 +            vmaw->delay_ns = ratelimit_calculate_delay(&vmaw->limit,
1645 +                                                       VMA_BLOCK_SIZE);
1646 +            if (vmaw->delay_ns) {
1647 +                co_sleep_ns(rt_clock, vmaw->delay_ns);
1648 +            }
1649 +        }
1650 +
1651 +        bit = bit << 1;
1652 +    }
1653 +
1654 +    uint64_t block_info = ((uint64_t)mask) << (32+16);
1655 +    block_info |= ((uint64_t)dev_id) << 32;
1656 +    block_info |= (cluster_num & 0xffffffff);
1657 +    vmaw->outbuf_block_info[vmaw->outbuf_count] =  block_info;
1658 +
1659 +    DPRINTF("VMA WRITE MASK %zd %zx\n", cluster_num, block_info);
1660 +
1661 +    vmaw->outbuf_count++;
1662 +
1663 +    /** NOTE: We allways write whole clusters, but we correctly set
1664 +     * transferred bytes. So transferred == size when when everything
1665 +     * went OK.
1666 +     */
1667 +    size_t transferred = VMA_CLUSTER_SIZE;
1668 +
1669 +    if (dev_id != vmaw->vmstate_stream) {
1670 +        uint64_t last = (cluster_num + 1) * VMA_CLUSTER_SIZE;
1671 +        if (last > vmaw->stream_info[dev_id].size) {
1672 +            uint64_t diff = last - vmaw->stream_info[dev_id].size;
1673 +            if (diff >= VMA_CLUSTER_SIZE) {
1674 +                vma_writer_set_error(vmaw, "vma_writer_write: "
1675 +                                     "read after last cluster");
1676 +                return -1;
1677 +            }
1678 +            transferred -= diff;
1679 +        }
1680 +    }
1681 +
1682 +    vmaw->stream_info[dev_id].transferred += transferred;
1683 +
1684 +    return transferred;
1685 +}
1686 +
1687 +int vma_writer_close(VmaWriter *vmaw, Error **errp)
1688 +{
1689 +    g_assert(vmaw != NULL);
1690 +
1691 +    int i;
1692 +
1693 +    vma_queue_flush(vmaw);
1694 +
1695 +    /* this should not happen - just to be sure */
1696 +    while (!qemu_co_queue_empty(&vmaw->wqueue)) {
1697 +        DPRINTF("vma_writer_close wait\n");
1698 +        co_sleep_ns(rt_clock, 1000000);
1699 +    }
1700 +
1701 +    if (vmaw->cmd) {
1702 +        if (pclose(vmaw->cmd) < 0) {
1703 +            vma_writer_set_error(vmaw, "vma_writer_close: "
1704 +                                 "pclose failed - %s", strerror(errno));
1705 +        }
1706 +    } else {
1707 +        if (close(vmaw->fd) < 0) {
1708 +            vma_writer_set_error(vmaw, "vma_writer_close: "
1709 +                                 "close failed - %s", strerror(errno));
1710 +        }
1711 +    }
1712 +
1713 +    for (i = 0; i <= 255; i++) {
1714 +        VmaStreamInfo *si = &vmaw->stream_info[i];
1715 +        if (si->size) {
1716 +            if (!si->finished) {
1717 +                vma_writer_set_error(vmaw, "vma_writer_close: "
1718 +                                     "detected open stream '%s'", si->devname);
1719 +            } else if ((si->transferred != si->size) &&
1720 +                       (i != vmaw->vmstate_stream)) {
1721 +                vma_writer_set_error(vmaw, "vma_writer_close: "
1722 +                                     "incomplete stream '%s' (%zd != %zd)",
1723 +                                     si->devname, si->transferred, si->size);
1724 +            }
1725 +        }
1726 +    }
1727 +
1728 +    for (i = 0; i <= 255; i++) {
1729 +        vmaw->stream_info[i].finished = 1; /* mark as closed */
1730 +    }
1731 +
1732 +    vmaw->closed = 1;
1733 +
1734 +    if (vmaw->status < 0 && *errp == NULL) {
1735 +        error_setg(errp, "%s", vmaw->errmsg);
1736 +    }
1737 +
1738 +    return vmaw->status;
1739 +}
1740 +
1741 +void vma_writer_destroy(VmaWriter *vmaw)
1742 +{
1743 +    assert(vmaw);
1744 +
1745 +    int i;
1746 +
1747 +    for (i = 0; i <= 255; i++) {
1748 +        if (vmaw->stream_info[i].devname) {
1749 +            g_free(vmaw->stream_info[i].devname);
1750 +        }
1751 +    }
1752 +
1753 +    if (vmaw->md5csum) {
1754 +        g_checksum_free(vmaw->md5csum);
1755 +    }
1756 +
1757 +    g_free(vmaw);
1758 +}
1759 +
1760 +/* backup driver plugin */
1761 +
1762 +static int vma_dump_cb(void *opaque, uint8_t dev_id, int64_t cluster_num,
1763 +                       unsigned char *buf, size_t *zero_bytes)
1764 +{
1765 +    VmaWriter *vmaw = opaque;
1766 +
1767 +    return vma_writer_write(vmaw, dev_id, cluster_num, buf, zero_bytes);
1768 +}
1769 +
1770 +static int vma_close_cb(void *opaque, Error **errp)
1771 +{
1772 +    VmaWriter *vmaw = opaque;
1773 +
1774 +    int res = vma_writer_close(vmaw, errp);
1775 +    vma_writer_destroy(vmaw);
1776 +
1777 +    return res;
1778 +}
1779 +
1780 +static int vma_complete_cb(void *opaque, uint8_t dev_id, int ret)
1781 +{
1782 +    VmaWriter *vmaw = opaque;
1783 +
1784 +    if (ret < 0) {
1785 +        vma_writer_set_error(vmaw, "backup_complete_cb %d", ret);
1786 +    }
1787 +
1788 +    return vma_writer_close_stream(vmaw, dev_id);
1789 +}
1790 +
1791 +static int vma_register_stream_cb(void *opaque, const char *devname,
1792 +                                  size_t size)
1793 +{
1794 +    VmaWriter *vmaw = opaque;
1795 +
1796 +    return vma_writer_register_stream(vmaw, devname, size);
1797 +}
1798 +
1799 +static int vma_register_config_cb(void *opaque, const char *name,
1800 +                                  gpointer data, size_t data_len)
1801 +{
1802 +    VmaWriter *vmaw = opaque;
1803 +
1804 +    return vma_writer_add_config(vmaw, name, data, data_len);
1805 +}
1806 +
1807 +static void *vma_open_cb(const char *filename, uuid_t uuid, int64_t speed,
1808 +                         Error **errp)
1809 +{
1810 +    return vma_writer_create(filename, uuid, speed, errp);
1811 +}
1812 +
1813 +const BackupDriver backup_vma_driver = {
1814 +    .format = "vma",
1815 +    .open_cb = vma_open_cb,
1816 +    .close_cb = vma_close_cb,
1817 +    .register_config_cb = vma_register_config_cb,
1818 +    .register_stream_cb = vma_register_stream_cb,
1819 +    .dump_cb = vma_dump_cb,
1820 +    .complete_cb = vma_complete_cb,
1821 +};
1822 +
1823 diff --git a/vma.c b/vma.c
1824 new file mode 100644
1825 index 0000000..7a88d3f
1826 --- /dev/null
1827 +++ b/vma.c
1828 @@ -0,0 +1,555 @@
1829 +/*
1830 + * VMA: Virtual Machine Archive
1831 + *
1832 + * Copyright (C) 2012 Proxmox Server Solutions
1833 + *
1834 + * Authors:
1835 + *  Dietmar Maurer (dietmar@proxmox.com)
1836 + *
1837 + * This work is licensed under the terms of the GNU GPL, version 2 or later.
1838 + * See the COPYING file in the top-level directory.
1839 + *
1840 + */
1841 +
1842 +#include <stdio.h>
1843 +#include <errno.h>
1844 +#include <unistd.h>
1845 +#include <stdio.h>
1846 +#include <string.h>
1847 +#include <sys/types.h>
1848 +#include <sys/stat.h>
1849 +#include <fcntl.h>
1850 +#include <glib.h>
1851 +
1852 +#include "qemu-common.h"
1853 +#include "qemu-option.h"
1854 +#include "qemu-error.h"
1855 +#include "osdep.h"
1856 +#include "sysemu.h"
1857 +#include "block_int.h"
1858 +#include <stdio.h>
1859 +#include "vma.h"
1860 +
1861 +static void help(void)
1862 +{
1863 +    const char *help_msg =
1864 +        "usage: vma command [command options]\n"
1865 +        "\n"
1866 +        "vma list <filename>\n"
1867 +        "vma create <filename> [-c config] <archive> pathname ...\n"
1868 +        "vma extract <filename> [-r] <targetdir>\n"
1869 +        ;
1870 +
1871 +    printf("%s", help_msg);
1872 +    exit(1);
1873 +}
1874 +
1875 +static const char *extract_devname(const char *path, char **devname, int index)
1876 +{
1877 +    assert(path);
1878 +
1879 +    const char *sep = strchr(path, '=');
1880 +
1881 +    if (sep) {
1882 +        *devname = g_strndup(path, sep - path);
1883 +        path = sep + 1;
1884 +    } else {
1885 +        if (index >= 0) {
1886 +            *devname = g_strdup_printf("disk%d", index);
1887 +        } else {
1888 +            *devname = NULL;
1889 +        }
1890 +    }
1891 +
1892 +    return path;
1893 +}
1894 +
1895 +static void print_content(VmaReader *vmar)
1896 +{
1897 +    assert(vmar);
1898 +
1899 +    VmaHeader *head = vma_reader_get_header(vmar);
1900 +
1901 +    printf("CTIME: %s", ctime(&head->ctime));
1902 +
1903 +    GList *l = vma_reader_get_config_data(vmar);
1904 +    while (l && l->data) {
1905 +        VmaConfigData *cdata = (VmaConfigData *)l->data;
1906 +        l = g_list_next(l);
1907 +        printf("CFG: size: %d name: %s\n", cdata->len, cdata->name);
1908 +    }
1909 +
1910 +    int i;
1911 +    VmaDeviceInfo *di;
1912 +    for (i = 1; i < 255; i++) {
1913 +        di = vma_reader_get_device_info(vmar, i);
1914 +        if (di) {
1915 +            if (strcmp(di->devname, "vmstate") == 0) {
1916 +                printf("VMSTATE: dev_id=%d memory: %zd\n", i, di->size);
1917 +            } else {
1918 +                printf("DEV: dev_id=%d size: %zd devname: %s\n",
1919 +                       i, di->size, di->devname);
1920 +            }
1921 +        }
1922 +    }
1923 +}
1924 +
1925 +static int list_content(int argc, char **argv)
1926 +{
1927 +    int c, ret = 0;
1928 +    const char *filename;
1929 +
1930 +    for (;;) {
1931 +        c = getopt(argc, argv, "h");
1932 +        if (c == -1) {
1933 +            break;
1934 +        }
1935 +        switch (c) {
1936 +        case '?':
1937 +        case 'h':
1938 +            help();
1939 +            break;
1940 +        default:
1941 +            g_assert_not_reached();
1942 +        }
1943 +    }
1944 +
1945 +    /* Get the filename */
1946 +    if ((optind + 1) != argc) {
1947 +        help();
1948 +    }
1949 +    filename = argv[optind++];
1950 +
1951 +    Error *errp = NULL;
1952 +    VmaReader *vmar = vma_reader_create(filename, &errp);
1953 +
1954 +    if (!vmar) {
1955 +        g_error("%s", error_get_pretty(errp));
1956 +    }
1957 +
1958 +    print_content(vmar);
1959 +
1960 +    vma_reader_destroy(vmar);
1961 +
1962 +    return ret;
1963 +}
1964 +
1965 +typedef struct RestoreMap {
1966 +    char *devname;
1967 +    char *path;
1968 +    bool write_zero;
1969 +} RestoreMap;
1970 +
1971 +static int extract_content(int argc, char **argv)
1972 +{
1973 +    int c, ret = 0;
1974 +    const char *filename;
1975 +    const char *dirname;
1976 +    const char *readmap = NULL;
1977 +
1978 +    for (;;) {
1979 +        c = getopt(argc, argv, "hr:");
1980 +        if (c == -1) {
1981 +            break;
1982 +        }
1983 +        switch (c) {
1984 +        case '?':
1985 +        case 'h':
1986 +            help();
1987 +            break;
1988 +        case 'r':
1989 +            readmap = optarg;
1990 +            break;
1991 +        default:
1992 +            help();
1993 +        }
1994 +    }
1995 +
1996 +    /* Get the filename */
1997 +    if ((optind + 2) != argc) {
1998 +        help();
1999 +    }
2000 +    filename = argv[optind++];
2001 +    dirname = argv[optind++];
2002 +
2003 +    Error *errp = NULL;
2004 +    VmaReader *vmar = vma_reader_create(filename, &errp);
2005 +
2006 +    if (!vmar) {
2007 +        g_error("%s", error_get_pretty(errp));
2008 +    }
2009 +
2010 +    if (mkdir(dirname, 0777) < 0) {
2011 +        g_error("unable to create target directory %s - %s",
2012 +                dirname, strerror(errno));
2013 +    }
2014 +
2015 +    GList *l = vma_reader_get_config_data(vmar);
2016 +    while (l && l->data) {
2017 +        VmaConfigData *cdata = (VmaConfigData *)l->data;
2018 +        l = g_list_next(l);
2019 +        char *cfgfn = g_strdup_printf("%s/%s", dirname, cdata->name);
2020 +        GError *err = NULL;
2021 +        if (!g_file_set_contents(cfgfn, (gchar *)cdata->data, cdata->len,
2022 +                                 &err)) {
2023 +            g_error("unable to write file: %s", err->message);
2024 +        }
2025 +    }
2026 +
2027 +    GHashTable *devmap = g_hash_table_new(g_str_hash, g_str_equal);
2028 +
2029 +    if (readmap) {
2030 +        print_content(vmar);
2031 +
2032 +        FILE *map = fopen(readmap, "r");
2033 +        if (!map) {
2034 +            g_error("unable to open fifo %s - %s", readmap, strerror(errno));
2035 +        }
2036 +
2037 +        while (1) {
2038 +            char inbuf[8192];
2039 +            char *line = fgets(inbuf, sizeof(inbuf), map);
2040 +            if (!line || line[0] == '\0' || !strcmp(line, "done\n")) {
2041 +                break;
2042 +            }
2043 +            int len = strlen(line);
2044 +            if (line[len - 1] == '\n') {
2045 +                line[len - 1] = '\0';
2046 +                if (len == 1) {
2047 +                    break;
2048 +                }
2049 +            }
2050 +
2051 +            const char *path;
2052 +            bool write_zero;
2053 +            if (line[0] == '0' && line[1] == ':') {
2054 +                path = inbuf + 2;
2055 +                write_zero = false;
2056 +            } else if (line[0] == '1' && line[1] == ':') {
2057 +                path = inbuf + 2;
2058 +                write_zero = true;
2059 +            } else {
2060 +                g_error("read map failed - parse error ('%s')", inbuf);
2061 +            }
2062 +
2063 +            char *devname = NULL;
2064 +            path = extract_devname(path, &devname, -1);
2065 +            if (!devname) {
2066 +                g_error("read map failed - no dev name specified ('%s')",
2067 +                        inbuf);
2068 +            }
2069 +
2070 +            printf("TEST %s %s\n", path, devname);
2071 +
2072 +            RestoreMap *map = g_new0(RestoreMap, 1);
2073 +            map->devname = g_strdup(devname);
2074 +            map->path = g_strdup(path);
2075 +            map->write_zero = write_zero;
2076 +
2077 +            g_hash_table_insert(devmap, map->devname, map);
2078 +
2079 +        };
2080 +    }
2081 +
2082 +    int i;
2083 +    int vmstate_fd = -1;
2084 +    guint8 vmstate_stream = 0;
2085 +
2086 +    for (i = 1; i < 255; i++) {
2087 +        VmaDeviceInfo *di = vma_reader_get_device_info(vmar, i);
2088 +        if (di && (strcmp(di->devname, "vmstate") == 0)) {
2089 +            vmstate_stream = i;
2090 +            char *statefn = g_strdup_printf("%s/vmstate.bin", dirname);
2091 +            vmstate_fd = open(statefn, O_WRONLY|O_CREAT|O_EXCL, 0644);
2092 +            if (vmstate_fd < 0) {
2093 +                g_error("create vmstate file '%s' failed - %s", statefn,
2094 +                        strerror(errno));
2095 +            }
2096 +            g_free(statefn);
2097 +        } else if (di) {
2098 +            char *devfn = NULL;
2099 +            int flags = BDRV_O_RDWR|BDRV_O_CACHE_WB;
2100 +            bool write_zero = true;
2101 +
2102 +            if (readmap) {
2103 +                RestoreMap *map;
2104 +                map = (RestoreMap *)g_hash_table_lookup(devmap, di->devname);
2105 +                if (map == NULL) {
2106 +                    g_error("no device name mapping for %s", di->devname);
2107 +                }
2108 +                devfn = map->path;
2109 +                write_zero = map->write_zero;
2110 +            } else {
2111 +                devfn = g_strdup_printf("%s/tmp-disk-%s.raw",
2112 +                                        dirname, di->devname);
2113 +                printf("DEVINFO %s %zd\n", devfn, di->size);
2114 +
2115 +                if (bdrv_img_create(devfn, "raw", NULL, NULL, NULL,
2116 +                                    di->size, flags)) {
2117 +                    g_error("can't create file %s", devfn);
2118 +                }
2119 +
2120 +                /* Note: we created an empty file above, so there is no
2121 +                 * need to write zeroes (so we generate a sparse file)
2122 +                 */
2123 +                write_zero = false;
2124 +            }
2125 +
2126 +            BlockDriverState *bs = NULL;
2127 +            if (bdrv_file_open(&bs, devfn, flags)) {
2128 +                g_error("can't open file %s", devfn);
2129 +            }
2130 +            if (vma_reader_register_bs(vmar, i, bs, write_zero, &errp) < 0) {
2131 +                g_error("%s", error_get_pretty(errp));
2132 +            }
2133 +
2134 +            if (!readmap) {
2135 +                g_free(devfn);
2136 +            }
2137 +        }
2138 +    }
2139 +
2140 +    if (vma_reader_restore(vmar, vmstate_fd, &errp) < 0) {
2141 +        g_error("restore failed - %s", error_get_pretty(errp));
2142 +    }
2143 +
2144 +    if (!readmap) {
2145 +        for (i = 1; i < 255; i++) {
2146 +            VmaDeviceInfo *di = vma_reader_get_device_info(vmar, i);
2147 +            if (di && (i != vmstate_stream)) {
2148 +                char *tmpfn = g_strdup_printf("%s/tmp-disk-%s.raw",
2149 +                                              dirname, di->devname);
2150 +                char *fn = g_strdup_printf("%s/disk-%s.raw",
2151 +                                           dirname, di->devname);
2152 +                if (rename(tmpfn, fn) != 0) {
2153 +                    g_error("rename %s to %s failed - %s",
2154 +                            tmpfn, fn, strerror(errno));
2155 +                }
2156 +            }
2157 +        }
2158 +    }
2159 +
2160 +    vma_reader_destroy(vmar);
2161 +
2162 +    bdrv_close_all();
2163 +
2164 +    return ret;
2165 +}
2166 +
2167 +typedef struct BackupCB {
2168 +    VmaWriter *vmaw;
2169 +    uint8_t dev_id;
2170 +} BackupCB;
2171 +
2172 +static int backup_dump_cb(void *opaque, BlockDriverState *bs,
2173 +                          int64_t cluster_num, unsigned char *buf)
2174 +{
2175 +    BackupCB *bcb = opaque;
2176 +    size_t zb = 0;
2177 +    if (vma_writer_write(bcb->vmaw, bcb->dev_id, cluster_num, buf, &zb) < 0) {
2178 +        g_warning("backup_dump_cb vma_writer_write failed");
2179 +        return -1;
2180 +    }
2181 +
2182 +    return 0;
2183 +}
2184 +
2185 +static void backup_complete_cb(void *opaque, int ret)
2186 +{
2187 +    BackupCB *bcb = opaque;
2188 +
2189 +    if (ret < 0) {
2190 +        vma_writer_set_error(bcb->vmaw, "backup_complete_cb %d", ret);
2191 +    }
2192 +
2193 +    if (vma_writer_close_stream(bcb->vmaw, bcb->dev_id) <= 0) {
2194 +        Error *err = NULL;
2195 +        if (vma_writer_close(bcb->vmaw, &err) != 0) {
2196 +            g_warning("vma_writer_close failed %s", error_get_pretty(err));
2197 +        }
2198 +    }
2199 +}
2200 +
2201 +static int create_archive(int argc, char **argv)
2202 +{
2203 +    int i, c, res;
2204 +    int verbose = 0;
2205 +    const char *archivename;
2206 +    GList *config_files = NULL;
2207 +
2208 +    for (;;) {
2209 +        c = getopt(argc, argv, "hvc:");
2210 +        if (c == -1) {
2211 +            break;
2212 +        }
2213 +        switch (c) {
2214 +        case '?':
2215 +        case 'h':
2216 +            help();
2217 +            break;
2218 +        case 'c':
2219 +            config_files = g_list_append(config_files, optarg);
2220 +            break;
2221 +        case 'v':
2222 +            verbose = 1;
2223 +            break;
2224 +        default:
2225 +            g_assert_not_reached();
2226 +        }
2227 +    }
2228 +
2229 +
2230 +    /* make sure we have archive name and at least one path */
2231 +    if ((optind + 2) > argc) {
2232 +        help();
2233 +    }
2234 +
2235 +    archivename = argv[optind++];
2236 +
2237 +    uuid_t uuid;
2238 +    uuid_generate(uuid);
2239 +
2240 +    Error *local_err = NULL;
2241 +    VmaWriter *vmaw = vma_writer_create(archivename, uuid, 0, &local_err);
2242 +
2243 +    if (vmaw == NULL) {
2244 +        g_error("%s", error_get_pretty(local_err));
2245 +    }
2246 +
2247 +    GList *l = config_files;
2248 +    while (l && l->data) {
2249 +        char *name = l->data;
2250 +        char *cdata = NULL;
2251 +        gsize clen = 0;
2252 +        GError *err = NULL;
2253 +        if (!g_file_get_contents(name, &cdata, &clen, &err)) {
2254 +            unlink(archivename);
2255 +            g_error("Unable to read file: %s", err->message);
2256 +        }
2257 +
2258 +        if (vma_writer_add_config(vmaw, name, cdata, clen) != 0) {
2259 +            unlink(archivename);
2260 +            g_error("Unable to append config data %s (len = %zd)",
2261 +                    name, clen);
2262 +        }
2263 +        l = g_list_next(l);
2264 +    }
2265 +
2266 +    int ind = 0;
2267 +    while (optind < argc) {
2268 +        const char *path = argv[optind++];
2269 +        char *devname = NULL;
2270 +        path = extract_devname(path, &devname, ind++);
2271 +
2272 +        BlockDriver *drv = NULL;
2273 +        BlockDriverState *bs = bdrv_new(devname);
2274 +
2275 +        res = bdrv_open(bs, path, BDRV_O_CACHE_WB , drv);
2276 +        if (res < 0) {
2277 +            unlink(archivename);
2278 +            g_error("bdrv_open '%s' failed", path);
2279 +        }
2280 +        int64_t size = bdrv_getlength(bs);
2281 +        int dev_id = vma_writer_register_stream(vmaw, devname, size);
2282 +        if (dev_id <= 0) {
2283 +            unlink(archivename);
2284 +            g_error("vma_writer_register_stream '%s' failed", devname);
2285 +        }
2286 +
2287 +        BackupCB *bcb = g_new0(BackupCB, 1);
2288 +        bcb->vmaw = vmaw;
2289 +        bcb->dev_id = dev_id;
2290 +
2291 +        if (backup_job_start(bs, backup_dump_cb, backup_complete_cb, bcb) < 0) {
2292 +            unlink(archivename);
2293 +            g_error("backup_job_start failed");
2294 +        }
2295 +    }
2296 +
2297 +    VmaStatus vmastat;
2298 +    int percent = 0;
2299 +    int last_percent = -1;
2300 +
2301 +    while (1) {
2302 +        main_loop_wait(false);
2303 +        vma_writer_get_status(vmaw, &vmastat);
2304 +
2305 +        if (verbose) {
2306 +
2307 +            uint64_t total = 0;
2308 +            uint64_t transferred = 0;
2309 +            uint64_t zero_bytes = 0;
2310 +
2311 +            int i;
2312 +            for (i = 0; i < 256; i++) {
2313 +                if (vmastat.stream_info[i].size) {
2314 +                    total += vmastat.stream_info[i].size;
2315 +                    transferred += vmastat.stream_info[i].transferred;
2316 +                    zero_bytes += vmastat.stream_info[i].zero_bytes;
2317 +                }
2318 +            }
2319 +            percent = (transferred*100)/total;
2320 +            if (percent != last_percent) {
2321 +                printf("progress %d%% %zd/%zd %zd\n", percent,
2322 +                       transferred, total, zero_bytes);
2323 +
2324 +                last_percent = percent;
2325 +            }
2326 +        }
2327 +
2328 +        if (vmastat.closed) {
2329 +            break;
2330 +        }
2331 +    }
2332 +
2333 +    bdrv_drain_all();
2334 +
2335 +    vma_writer_get_status(vmaw, &vmastat);
2336 +
2337 +    if (verbose) {
2338 +        for (i = 0; i < 256; i++) {
2339 +            VmaStreamInfo *si = &vmastat.stream_info[i];
2340 +            if (si->size) {
2341 +                printf("image %s: size=%zd zeros=%zd saved=%zd\n", si->devname,
2342 +                       si->size, si->zero_bytes, si->size - si->zero_bytes);
2343 +            }
2344 +        }
2345 +    }
2346 +
2347 +    if (vmastat.status < 0) {
2348 +        unlink(archivename);
2349 +        g_error("creating vma archive failed");
2350 +    }
2351 +
2352 +    return 0;
2353 +}
2354 +
2355 +int main(int argc, char **argv)
2356 +{
2357 +    const char *cmdname;
2358 +
2359 +    error_set_progname(argv[0]);
2360 +
2361 +    qemu_init_main_loop();
2362 +
2363 +    bdrv_init();
2364 +
2365 +    if (argc < 2) {
2366 +        help();
2367 +    }
2368 +
2369 +    cmdname = argv[1];
2370 +    argc--; argv++;
2371 +
2372 +
2373 +    if (!strcmp(cmdname, "list")) {
2374 +        return list_content(argc, argv);
2375 +    } else if (!strcmp(cmdname, "create")) {
2376 +        return create_archive(argc, argv);
2377 +    } else if (!strcmp(cmdname, "extract")) {
2378 +        return extract_content(argc, argv);
2379 +    }
2380 +
2381 +    help();
2382 +    return 0;
2383 +}
2384 diff --git a/vma.h b/vma.h
2385 new file mode 100644
2386 index 0000000..10800a1
2387 --- /dev/null
2388 +++ b/vma.h
2389 @@ -0,0 +1,145 @@
2390 +/*
2391 + * VMA: Virtual Machine Archive
2392 + *
2393 + * Copyright (C) Proxmox Server Solutions
2394 + *
2395 + * Authors:
2396 + *  Dietmar Maurer (dietmar@proxmox.com)
2397 + *
2398 + * This work is licensed under the terms of the GNU GPL, version 2 or later.
2399 + * See the COPYING file in the top-level directory.
2400 + *
2401 + */
2402 +
2403 +#ifndef BACKUP_VMA_H
2404 +#define BACKUP_VMA_H
2405 +
2406 +#include "backup.h"
2407 +#include "error.h"
2408 +
2409 +#define VMA_BLOCK_BITS 12
2410 +#define VMA_BLOCK_SIZE (1<<VMA_BLOCK_BITS)
2411 +#define VMA_CLUSTER_BITS (VMA_BLOCK_BITS+4)
2412 +#define VMA_CLUSTER_SIZE (1<<VMA_CLUSTER_BITS)
2413 +
2414 +#if VMA_CLUSTER_SIZE != 65536
2415 +#error unexpected cluster size
2416 +#endif
2417 +
2418 +#define VMA_EXTENT_HEADER_SIZE 512
2419 +#define VMA_BLOCKS_PER_EXTENT 59
2420 +#define VMA_MAX_CONFIGS 256
2421 +
2422 +#define VMA_MAX_EXTENT_SIZE \
2423 +    (VMA_EXTENT_HEADER_SIZE+VMA_CLUSTER_SIZE*VMA_BLOCKS_PER_EXTENT)
2424 +#if VMA_MAX_EXTENT_SIZE != 3867136
2425 +#error unexpected VMA_EXTENT_SIZE
2426 +#endif
2427 +
2428 +/* File Format Definitions */
2429 +
2430 +#define VMA_MAGIC (GUINT32_TO_BE(('V'<<24)|('M'<<16)|('A'<<8)|0x00))
2431 +#define VMA_EXTENT_MAGIC (GUINT32_TO_BE(('V'<<24)|('M'<<16)|('A'<<8)|'E'))
2432 +
2433 +typedef struct VmaDeviceInfoHeader {
2434 +    uint32_t devname_ptr; /* offset into blob_buffer table */
2435 +    uint32_t reserved0;
2436 +    uint64_t size; /* device size in bytes */
2437 +    uint64_t reserved1;
2438 +    uint64_t reserved2;
2439 +} VmaDeviceInfoHeader;
2440 +
2441 +typedef struct VmaHeader {
2442 +    uint32_t magic;
2443 +    uint32_t version;
2444 +    unsigned char uuid[16];
2445 +    int64_t ctime;
2446 +    unsigned char md5sum[16];
2447 +
2448 +    uint32_t blob_buffer_offset;
2449 +    uint32_t blob_buffer_size;
2450 +    uint32_t header_size;
2451 +
2452 +    unsigned char reserved[1984];
2453 +
2454 +    uint32_t config_names[VMA_MAX_CONFIGS]; /* offset into blob_buffer table */
2455 +    uint32_t config_data[VMA_MAX_CONFIGS];  /* offset into blob_buffer table */
2456 +
2457 +    VmaDeviceInfoHeader dev_info[256];
2458 +} VmaHeader;
2459 +
2460 +typedef struct VmaExtentHeader {
2461 +    uint32_t magic;
2462 +    uint16_t reserved1;
2463 +    uint16_t block_count;
2464 +    unsigned char uuid[16];
2465 +    unsigned char md5sum[16];
2466 +    uint64_t blockinfo[VMA_BLOCKS_PER_EXTENT];
2467 +} VmaExtentHeader;
2468 +
2469 +/* functions/definitions to read/write vma files */
2470 +
2471 +typedef struct VmaReader VmaReader;
2472 +
2473 +typedef struct VmaWriter VmaWriter;
2474 +
2475 +typedef struct VmaConfigData {
2476 +    const char *name;
2477 +    const void *data;
2478 +    uint32_t len;
2479 +} VmaConfigData;
2480 +
2481 +typedef struct VmaStreamInfo {
2482 +    uint64_t size;
2483 +    uint64_t cluster_count;
2484 +    uint64_t transferred;
2485 +    uint64_t zero_bytes;
2486 +    int finished;
2487 +    char *devname;
2488 +} VmaStreamInfo;
2489 +
2490 +typedef struct VmaStatus {
2491 +    int status;
2492 +    bool closed;
2493 +    char errmsg[8192];
2494 +    char uuid_str[37];
2495 +    VmaStreamInfo stream_info[256];
2496 +} VmaStatus;
2497 +
2498 +typedef struct VmaDeviceInfo {
2499 +    uint64_t size; /* device size in bytes */
2500 +    const char *devname;
2501 +} VmaDeviceInfo;
2502 +
2503 +extern const BackupDriver backup_vma_driver;
2504 +
2505 +VmaWriter *vma_writer_create(const char *filename, uuid_t uuid, int64_t speed,
2506 +                             Error **errp);
2507 +int vma_writer_close(VmaWriter *vmaw, Error **errp);
2508 +void vma_writer_destroy(VmaWriter *vmaw);
2509 +int vma_writer_add_config(VmaWriter *vmaw, const char *name, gpointer data,
2510 +                          size_t len);
2511 +int vma_writer_register_stream(VmaWriter *vmaw, const char *devname,
2512 +                               size_t size);
2513 +
2514 +int64_t coroutine_fn vma_writer_write(VmaWriter *vmaw, uint8_t dev_id,
2515 +                                      int64_t cluster_num, unsigned char *buf,
2516 +                                      size_t *zero_bytes);
2517 +
2518 +int coroutine_fn vma_writer_close_stream(VmaWriter *vmaw, uint8_t dev_id);
2519 +
2520 +int vma_writer_get_status(VmaWriter *vmaw, VmaStatus *status);
2521 +void vma_writer_set_error(VmaWriter *vmaw, const char *fmt, ...);
2522 +
2523 +
2524 +VmaReader *vma_reader_create(const char *filename, Error **errp);
2525 +void vma_reader_destroy(VmaReader *vmar);
2526 +VmaHeader *vma_reader_get_header(VmaReader *vmar);
2527 +GList *vma_reader_get_config_data(VmaReader *vmar);
2528 +VmaDeviceInfo *vma_reader_get_device_info(VmaReader *vmar, guint8 dev_id);
2529 +int vma_reader_register_bs(VmaReader *vmar, guint8 dev_id,
2530 +                           BlockDriverState *bs, bool write_zeroes,
2531 +                           Error **errp);
2532 +int vma_reader_restore(VmaReader *vmar, int vmstate_fd, Error **errp);
2533 +
2534 +#endif /* BACKUP_VMA_H */
2535 --
2536 1.7.2.5
2537