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

   1 From 5476ae43806488e74cd293bbaa17f130aa53d402 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 v4 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  backup.h                |    1 +
  13  blockdev.c              |    6 +-
  14  docs/specs/vma_spec.txt |   24 ++
  15  vma-reader.c            |  799 ++++++++++++++++++++++++++++++++++++++++
  16  vma-writer.c            |  932 +++++++++++++++++++++++++++++++++++++++++++++++
  17  vma.c                   |  559 ++++++++++++++++++++++++++++
  18  vma.h                   |  145 ++++++++
  19  9 files changed, 2467 insertions(+), 4 deletions(-)
  20  create mode 100644 docs/specs/vma_spec.txt
  21  create mode 100644 vma-reader.c
  22  create mode 100644 vma-writer.c
  23  create mode 100644 vma.c
  24  create mode 100644 vma.h
  25
  26 diff --git a/Makefile b/Makefile
  27 index 0d9099a..16f1c25 100644
  28 --- a/Makefile
  29 +++ b/Makefile
  30 @@ -115,7 +115,7 @@ ifeq ($(CONFIG_SMARTCARD_NSS),y)
  31  include $(SRC_PATH)/libcacard/Makefile
  32  endif
  33
  34 -all: $(DOCS) $(TOOLS) $(HELPERS-y) recurse-all
  35 +all: $(DOCS) $(TOOLS) vma$(EXESUF) $(HELPERS-y) recurse-all
  36
  37  config-host.h: config-host.h-timestamp
  38  config-host.h-timestamp: config-host.mak
  39 @@ -167,6 +167,7 @@ qemu-img.o: qemu-img-cmds.h
  40  qemu-img$(EXESUF): qemu-img.o $(block-obj-y) libqemuutil.a libqemustub.a
  41  qemu-nbd$(EXESUF): qemu-nbd.o $(block-obj-y) libqemuutil.a libqemustub.a
  42  qemu-io$(EXESUF): qemu-io.o cmd.o $(block-obj-y) libqemuutil.a libqemustub.a
  43 +vma$(EXESUF): vma.o vma-writer.o vma-reader.o $(block-obj-y)  libqemuutil.a libqemustub.a
  44
  45  qemu-bridge-helper$(EXESUF): qemu-bridge-helper.o
  46
  47 diff --git a/Makefile.objs b/Makefile.objs
  48 index df64f70..91f133b 100644
  49 --- a/Makefile.objs
  50 +++ b/Makefile.objs
  51 @@ -13,7 +13,7 @@ block-obj-$(CONFIG_POSIX) += aio-posix.o
  52  block-obj-$(CONFIG_WIN32) += aio-win32.o
  53  block-obj-y += block/
  54  block-obj-y += qapi-types.o qapi-visit.o
  55 -block-obj-y += backup.o
  56 +block-obj-y += vma-writer.o backup.o
  57
  58  block-obj-y += qemu-coroutine.o qemu-coroutine-lock.o qemu-coroutine-io.o
  59  block-obj-y += qemu-coroutine-sleep.o
  60 diff --git a/backup.h b/backup.h
  61 index c8ba153..406f011 100644
  62 --- a/backup.h
  63 +++ b/backup.h
  64 @@ -15,6 +15,7 @@
  65  #define QEMU_BACKUP_H
  66
  67  #include <uuid/uuid.h>
  68 +#include "block/block.h"
  69
  70  #define BACKUP_CLUSTER_BITS 16
  71  #define BACKUP_CLUSTER_SIZE (1<<BACKUP_CLUSTER_BITS)
  72 diff --git a/blockdev.c b/blockdev.c
  73 index c340fde..1cfc780 100644
  74 --- a/blockdev.c
  75 +++ b/blockdev.c
  76 @@ -21,6 +21,7 @@
  77  #include "trace.h"
  78  #include "sysemu/arch_init.h"
  79  #include "backup.h"
  80 +#include "vma.h"
  81
  82  static QTAILQ_HEAD(drivelist, DriveInfo) drives = QTAILQ_HEAD_INITIALIZER(drives);
  83
  84 @@ -1530,10 +1531,11 @@ char *qmp_backup(const char *backup_file, bool has_format, BackupFormat format,
  85      /* Todo: try to auto-detect format based on file name */
  86      format = has_format ? format : BACKUP_FORMAT_VMA;
  87
  88 -    /* fixme: find driver for specifued format */
  89      const BackupDriver *driver = NULL;
  90
  91 -    if (!driver) {
  92 +    if (format == BACKUP_FORMAT_VMA) {
  93 +        driver = &backup_vma_driver;
  94 +    } else {
  95          error_set(errp, ERROR_CLASS_GENERIC_ERROR, "unknown backup format");
  96          return NULL;
  97      }
  98 diff --git a/docs/specs/vma_spec.txt b/docs/specs/vma_spec.txt
  99 new file mode 100644
 100 index 0000000..052c629
 101 --- /dev/null
 102 +++ b/docs/specs/vma_spec.txt
 103 @@ -0,0 +1,24 @@
 104 +=Virtual Machine Archive format (VMA)=
 105 +
 106 +This format contains a header which includes the VM configuration as
 107 +binary blobs, and a list of devices (dev_id, name).
 108 +
 109 +The actual VM image data is stored inside extents. An extent contains
 110 +up to 64 clusters, and start with a 512 byte header containing
 111 +additional information for those clusters.
 112 +
 113 +We use a cluster size of 65536, and use 8 bytes for each
 114 +cluster in the header to store the following information:
 115 +
 116 +* 1 byte dev_id (to identity the drive)
 117 +* 2 bytes zero indicator (mark zero regions (16x4096))
 118 +* 4 bytes cluster number
 119 +* 1 byte not used (reserved)
 120 +
 121 +We only store non-zero blocks (such block is 4096 bytes).
 122 +
 123 +Each archive is marked with a uuid. The archive header and all
 124 +extent headers includes that uuid and a MD5 checksum (over header
 125 +data).
 126 +
 127 +
 128 diff --git a/vma-reader.c b/vma-reader.c
 129 new file mode 100644
 130 index 0000000..7e81847
 131 --- /dev/null
 132 +++ b/vma-reader.c
 133 @@ -0,0 +1,799 @@
 134 +/*
 135 + * VMA: Virtual Machine Archive
 136 + *
 137 + * Copyright (C) 2012 Proxmox Server Solutions
 138 + *
 139 + * Authors:
 140 + *  Dietmar Maurer (dietmar@proxmox.com)
 141 + *
 142 + * This work is licensed under the terms of the GNU GPL, version 2 or later.
 143 + * See the COPYING file in the top-level directory.
 144 + *
 145 + */
 146 +
 147 +#include <stdio.h>
 148 +#include <errno.h>
 149 +#include <unistd.h>
 150 +#include <stdio.h>
 151 +#include <string.h>
 152 +#include <sys/types.h>
 153 +#include <sys/stat.h>
 154 +#include <fcntl.h>
 155 +#include <glib.h>
 156 +#include <uuid/uuid.h>
 157 +
 158 +#include "qemu-common.h"
 159 +#include "qemu/timer.h"
 160 +#include "qemu/ratelimit.h"
 161 +#include "vma.h"
 162 +#include "block/block.h"
 163 +
 164 +#define BITS_PER_LONG  (sizeof(unsigned long) * 8)
 165 +
 166 +static unsigned char zero_vma_block[VMA_BLOCK_SIZE];
 167 +
 168 +typedef struct VmaRestoreState {
 169 +    BlockDriverState *bs;
 170 +    bool write_zeroes;
 171 +    unsigned long *bitmap;
 172 +    int bitmap_size;
 173 +}  VmaRestoreState;
 174 +
 175 +struct VmaReader {
 176 +    int fd;
 177 +    GChecksum *md5csum;
 178 +    GHashTable *blob_hash;
 179 +    unsigned char *head_data;
 180 +    VmaDeviceInfo devinfo[256];
 181 +    VmaRestoreState rstate[256];
 182 +    GList *cdata_list;
 183 +    guint8 vmstate_stream;
 184 +    uint32_t vmstate_clusters;
 185 +    /* to show restore percentage if run with -v */
 186 +    time_t start_time;
 187 +    int64_t cluster_count;
 188 +    int64_t clusters_read;
 189 +    int clusters_read_per;
 190 +};
 191 +
 192 +static guint
 193 +g_int32_hash(gconstpointer v)
 194 +{
 195 +    return *(const uint32_t *)v;
 196 +}
 197 +
 198 +static gboolean
 199 +g_int32_equal(gconstpointer v1, gconstpointer v2)
 200 +{
 201 +    return *((const uint32_t *)v1) == *((const uint32_t *)v2);
 202 +}
 203 +
 204 +static int vma_reader_get_bitmap(VmaRestoreState *rstate, int64_t cluster_num)
 205 +{
 206 +    assert(rstate);
 207 +    assert(rstate->bitmap);
 208 +
 209 +    unsigned long val, idx, bit;
 210 +
 211 +    idx = cluster_num / BITS_PER_LONG;
 212 +
 213 +    assert(rstate->bitmap_size > idx);
 214 +
 215 +    bit = cluster_num % BITS_PER_LONG;
 216 +    val = rstate->bitmap[idx];
 217 +
 218 +    return !!(val & (1UL << bit));
 219 +}
 220 +
 221 +static void vma_reader_set_bitmap(VmaRestoreState *rstate, int64_t cluster_num,
 222 +                                  int dirty)
 223 +{
 224 +    assert(rstate);
 225 +    assert(rstate->bitmap);
 226 +
 227 +    unsigned long val, idx, bit;
 228 +
 229 +    idx = cluster_num / BITS_PER_LONG;
 230 +
 231 +    assert(rstate->bitmap_size > idx);
 232 +
 233 +    bit = cluster_num % BITS_PER_LONG;
 234 +    val = rstate->bitmap[idx];
 235 +    if (dirty) {
 236 +        if (!(val & (1UL << bit))) {
 237 +            val |= 1UL << bit;
 238 +        }
 239 +    } else {
 240 +        if (val & (1UL << bit)) {
 241 +            val &= ~(1UL << bit);
 242 +        }
 243 +    }
 244 +    rstate->bitmap[idx] = val;
 245 +}
 246 +
 247 +typedef struct VmaBlob {
 248 +    uint32_t start;
 249 +    uint32_t len;
 250 +    void *data;
 251 +} VmaBlob;
 252 +
 253 +static const VmaBlob *get_header_blob(VmaReader *vmar, uint32_t pos)
 254 +{
 255 +    assert(vmar);
 256 +    assert(vmar->blob_hash);
 257 +
 258 +    return g_hash_table_lookup(vmar->blob_hash, &pos);
 259 +}
 260 +
 261 +static const char *get_header_str(VmaReader *vmar, uint32_t pos)
 262 +{
 263 +    const VmaBlob *blob = get_header_blob(vmar, pos);
 264 +    if (!blob) {
 265 +        return NULL;
 266 +    }
 267 +    const char *res = (char *)blob->data;
 268 +    if (res[blob->len-1] != '\0') {
 269 +        return NULL;
 270 +    }
 271 +    return res;
 272 +}
 273 +
 274 +static ssize_t
 275 +safe_read(int fd, unsigned char *buf, size_t count)
 276 +{
 277 +    ssize_t n;
 278 +
 279 +    do {
 280 +        n = read(fd, buf, count);
 281 +    } while (n < 0 && errno == EINTR);
 282 +
 283 +    return n;
 284 +}
 285 +
 286 +static ssize_t
 287 +full_read(int fd, unsigned char *buf, size_t len)
 288 +{
 289 +    ssize_t n;
 290 +    size_t total;
 291 +
 292 +    total = 0;
 293 +
 294 +    while (len > 0) {
 295 +        n = safe_read(fd, buf, len);
 296 +
 297 +        if (n == 0) {
 298 +            return total;
 299 +        }
 300 +
 301 +        if (n <= 0) {
 302 +            break;
 303 +        }
 304 +
 305 +        buf += n;
 306 +        total += n;
 307 +        len -= n;
 308 +    }
 309 +
 310 +    if (len) {
 311 +        return -1;
 312 +    }
 313 +
 314 +    return total;
 315 +}
 316 +
 317 +void vma_reader_destroy(VmaReader *vmar)
 318 +{
 319 +    assert(vmar);
 320 +
 321 +    if (vmar->fd >= 0) {
 322 +        close(vmar->fd);
 323 +    }
 324 +
 325 +    if (vmar->cdata_list) {
 326 +        g_list_free(vmar->cdata_list);
 327 +    }
 328 +
 329 +    int i;
 330 +    for (i = 1; i < 256; i++) {
 331 +        if (vmar->rstate[i].bitmap) {
 332 +            g_free(vmar->rstate[i].bitmap);
 333 +        }
 334 +    }
 335 +
 336 +    if (vmar->md5csum) {
 337 +        g_checksum_free(vmar->md5csum);
 338 +    }
 339 +
 340 +    if (vmar->blob_hash) {
 341 +        g_hash_table_destroy(vmar->blob_hash);
 342 +    }
 343 +
 344 +    if (vmar->head_data) {
 345 +        g_free(vmar->head_data);
 346 +    }
 347 +
 348 +    g_free(vmar);
 349 +
 350 +};
 351 +
 352 +static int vma_reader_read_head(VmaReader *vmar, Error **errp)
 353 +{
 354 +    assert(vmar);
 355 +    assert(errp);
 356 +    assert(*errp == NULL);
 357 +
 358 +    unsigned char md5sum[16];
 359 +    int i;
 360 +    int ret = 0;
 361 +
 362 +    vmar->head_data = g_malloc(sizeof(VmaHeader));
 363 +
 364 +    if (full_read(vmar->fd, vmar->head_data, sizeof(VmaHeader)) !=
 365 +        sizeof(VmaHeader)) {
 366 +        error_setg(errp, "can't read vma header - %s",
 367 +                   errno ? strerror(errno) : "got EOF");
 368 +        return -1;
 369 +    }
 370 +
 371 +    VmaHeader *h = (VmaHeader *)vmar->head_data;
 372 +
 373 +    if (h->magic != VMA_MAGIC) {
 374 +        error_setg(errp, "not a vma file - wrong magic number");
 375 +        return -1;
 376 +    }
 377 +
 378 +    uint32_t header_size = GUINT32_FROM_BE(h->header_size);
 379 +    int need = header_size - sizeof(VmaHeader);
 380 +    if (need <= 0) {
 381 +        error_setg(errp, "wrong vma header size %d", header_size);
 382 +        return -1;
 383 +    }
 384 +
 385 +    vmar->head_data = g_realloc(vmar->head_data, header_size);
 386 +    h = (VmaHeader *)vmar->head_data;
 387 +
 388 +    if (full_read(vmar->fd, vmar->head_data + sizeof(VmaHeader), need) !=
 389 +        need) {
 390 +        error_setg(errp, "can't read vma header data - %s",
 391 +                   errno ? strerror(errno) : "got EOF");
 392 +        return -1;
 393 +    }
 394 +
 395 +    memcpy(md5sum, h->md5sum, 16);
 396 +    memset(h->md5sum, 0, 16);
 397 +
 398 +    g_checksum_reset(vmar->md5csum);
 399 +    g_checksum_update(vmar->md5csum, vmar->head_data, header_size);
 400 +    gsize csize = 16;
 401 +    g_checksum_get_digest(vmar->md5csum, (guint8 *)(h->md5sum), &csize);
 402 +
 403 +    if (memcmp(md5sum, h->md5sum, 16) != 0) {
 404 +        error_setg(errp, "wrong vma header chechsum");
 405 +        return -1;
 406 +    }
 407 +
 408 +    /* we can modify header data after checksum verify */
 409 +    h->header_size = header_size;
 410 +
 411 +    h->version = GUINT32_FROM_BE(h->version);
 412 +    if (h->version != 1) {
 413 +        error_setg(errp, "wrong vma version %d", h->version);
 414 +        return -1;
 415 +    }
 416 +
 417 +    h->ctime = GUINT64_FROM_BE(h->ctime);
 418 +    h->blob_buffer_offset = GUINT32_FROM_BE(h->blob_buffer_offset);
 419 +    h->blob_buffer_size = GUINT32_FROM_BE(h->blob_buffer_size);
 420 +
 421 +    uint32_t bstart = h->blob_buffer_offset + 1;
 422 +    uint32_t bend = h->blob_buffer_offset + h->blob_buffer_size;
 423 +
 424 +    if (bstart <= sizeof(VmaHeader)) {
 425 +        error_setg(errp, "wrong vma blob buffer offset %d",
 426 +                   h->blob_buffer_offset);
 427 +        return -1;
 428 +    }
 429 +
 430 +    if (bend > header_size) {
 431 +        error_setg(errp, "wrong vma blob buffer size %d/%d",
 432 +                   h->blob_buffer_offset, h->blob_buffer_size);
 433 +        return -1;
 434 +    }
 435 +
 436 +    while ((bstart + 2) <= bend) {
 437 +        uint32_t size = vmar->head_data[bstart] +
 438 +            (vmar->head_data[bstart+1] << 8);
 439 +        if ((bstart + size + 2) <= bend) {
 440 +            VmaBlob *blob = g_new0(VmaBlob, 1);
 441 +            blob->start = bstart - h->blob_buffer_offset;
 442 +            blob->len = size;
 443 +            blob->data = vmar->head_data + bstart + 2;
 444 +            g_hash_table_insert(vmar->blob_hash, &blob->start, blob);
 445 +        }
 446 +        bstart += size + 2;
 447 +    }
 448 +
 449 +
 450 +    int count = 0;
 451 +    for (i = 1; i < 256; i++) {
 452 +        VmaDeviceInfoHeader *dih = &h->dev_info[i];
 453 +        uint32_t devname_ptr = GUINT32_FROM_BE(dih->devname_ptr);
 454 +        uint64_t size = GUINT64_FROM_BE(dih->size);
 455 +        const char *devname =  get_header_str(vmar, devname_ptr);
 456 +
 457 +        if (size && devname) {
 458 +            count++;
 459 +            vmar->devinfo[i].size = size;
 460 +            vmar->devinfo[i].devname = devname;
 461 +
 462 +            if (strcmp(devname, "vmstate") == 0) {
 463 +                vmar->vmstate_stream = i;
 464 +            }
 465 +        }
 466 +    }
 467 +
 468 +    if (!count) {
 469 +        error_setg(errp, "vma does not contain data");
 470 +        return -1;
 471 +    }
 472 +
 473 +    for (i = 0; i < VMA_MAX_CONFIGS; i++) {
 474 +        uint32_t name_ptr = GUINT32_FROM_BE(h->config_names[i]);
 475 +        uint32_t data_ptr = GUINT32_FROM_BE(h->config_data[i]);
 476 +
 477 +        if (!(name_ptr && data_ptr)) {
 478 +            continue;
 479 +        }
 480 +        const char *name =  get_header_str(vmar, name_ptr);
 481 +        const VmaBlob *blob = get_header_blob(vmar, data_ptr);
 482 +
 483 +        if (!(name && blob)) {
 484 +            error_setg(errp, "vma contains invalid data pointers");
 485 +            return -1;
 486 +        }
 487 +
 488 +        VmaConfigData *cdata = g_new0(VmaConfigData, 1);
 489 +        cdata->name = name;
 490 +        cdata->data = blob->data;
 491 +        cdata->len = blob->len;
 492 +
 493 +        vmar->cdata_list = g_list_append(vmar->cdata_list, cdata);
 494 +    }
 495 +
 496 +    return ret;
 497 +};
 498 +
 499 +VmaReader *vma_reader_create(const char *filename, Error **errp)
 500 +{
 501 +    assert(filename);
 502 +    assert(errp);
 503 +
 504 +    VmaReader *vmar = g_new0(VmaReader, 1);
 505 +
 506 +    if (strcmp(filename, "-") == 0) {
 507 +        vmar->fd = dup(0);
 508 +    } else {
 509 +        vmar->fd = open(filename, O_RDONLY);
 510 +    }
 511 +
 512 +    if (vmar->fd < 0) {
 513 +        error_setg(errp, "can't open file %s - %s\n", filename,
 514 +                   strerror(errno));
 515 +        goto err;
 516 +    }
 517 +
 518 +    vmar->md5csum = g_checksum_new(G_CHECKSUM_MD5);
 519 +    if (!vmar->md5csum) {
 520 +        error_setg(errp, "can't allocate cmsum\n");
 521 +        goto err;
 522 +    }
 523 +
 524 +    vmar->blob_hash = g_hash_table_new_full(g_int32_hash, g_int32_equal,
 525 +                                            NULL, g_free);
 526 +
 527 +    if (vma_reader_read_head(vmar, errp) < 0) {
 528 +        goto err;
 529 +    }
 530 +
 531 +    return vmar;
 532 +
 533 +err:
 534 +    if (vmar) {
 535 +        vma_reader_destroy(vmar);
 536 +    }
 537 +
 538 +    return NULL;
 539 +}
 540 +
 541 +VmaHeader *vma_reader_get_header(VmaReader *vmar)
 542 +{
 543 +    assert(vmar);
 544 +    assert(vmar->head_data);
 545 +
 546 +    return (VmaHeader *)(vmar->head_data);
 547 +}
 548 +
 549 +GList *vma_reader_get_config_data(VmaReader *vmar)
 550 +{
 551 +    assert(vmar);
 552 +    assert(vmar->head_data);
 553 +
 554 +    return vmar->cdata_list;
 555 +}
 556 +
 557 +VmaDeviceInfo *vma_reader_get_device_info(VmaReader *vmar, guint8 dev_id)
 558 +{
 559 +    assert(vmar);
 560 +    assert(dev_id);
 561 +
 562 +    if (vmar->devinfo[dev_id].size && vmar->devinfo[dev_id].devname) {
 563 +        return &vmar->devinfo[dev_id];
 564 +    }
 565 +
 566 +    return NULL;
 567 +}
 568 +
 569 +int vma_reader_register_bs(VmaReader *vmar, guint8 dev_id, BlockDriverState *bs,
 570 +                           bool write_zeroes, Error **errp)
 571 +{
 572 +    assert(vmar);
 573 +    assert(bs != NULL);
 574 +    assert(dev_id);
 575 +    assert(vmar->rstate[dev_id].bs == NULL);
 576 +
 577 +    int64_t size = bdrv_getlength(bs);
 578 +    if (size != vmar->devinfo[dev_id].size) {
 579 +        error_setg(errp, "vma_reader_register_bs for stream %s failed - "
 580 +                   "unexpected size %zd != %zd", vmar->devinfo[dev_id].devname,
 581 +                   size, vmar->devinfo[dev_id].size);
 582 +        return -1;
 583 +    }
 584 +
 585 +    vmar->rstate[dev_id].bs = bs;
 586 +    vmar->rstate[dev_id].write_zeroes = write_zeroes;
 587 +
 588 +    int64_t bitmap_size = (size/BDRV_SECTOR_SIZE) +
 589 +        (VMA_CLUSTER_SIZE/BDRV_SECTOR_SIZE) * BITS_PER_LONG - 1;
 590 +    bitmap_size /= (VMA_CLUSTER_SIZE/BDRV_SECTOR_SIZE) * BITS_PER_LONG;
 591 +
 592 +    vmar->rstate[dev_id].bitmap_size = bitmap_size;
 593 +    vmar->rstate[dev_id].bitmap = g_new0(unsigned long, bitmap_size);
 594 +
 595 +    vmar->cluster_count += size/VMA_CLUSTER_SIZE;
 596 +
 597 +    return 0;
 598 +}
 599 +
 600 +static ssize_t safe_write(int fd, void *buf, size_t count)
 601 +{
 602 +    ssize_t n;
 603 +
 604 +    do {
 605 +        n = write(fd, buf, count);
 606 +    } while (n < 0 && errno == EINTR);
 607 +
 608 +    return n;
 609 +}
 610 +
 611 +static size_t full_write(int fd, void *buf, size_t len)
 612 +{
 613 +    ssize_t n;
 614 +    size_t total;
 615 +
 616 +    total = 0;
 617 +
 618 +    while (len > 0) {
 619 +        n = safe_write(fd, buf, len);
 620 +        if (n < 0) {
 621 +            return n;
 622 +        }
 623 +        buf += n;
 624 +        total += n;
 625 +        len -= n;
 626 +    }
 627 +
 628 +    if (len) {
 629 +        /* incomplete write ? */
 630 +        return -1;
 631 +    }
 632 +
 633 +    return total;
 634 +}
 635 +
 636 +static int restore_write_data(VmaReader *vmar, guint8 dev_id,
 637 +                              BlockDriverState *bs, int vmstate_fd,
 638 +                              unsigned char *buf, int64_t sector_num,
 639 +                              int nb_sectors, Error **errp)
 640 +{
 641 +    assert(vmar);
 642 +
 643 +    if (dev_id == vmar->vmstate_stream) {
 644 +        if (vmstate_fd >= 0) {
 645 +            int len = nb_sectors * BDRV_SECTOR_SIZE;
 646 +            int res = full_write(vmstate_fd, buf, len);
 647 +            if (res < 0) {
 648 +                error_setg(errp, "write vmstate failed %d", res);
 649 +                return -1;
 650 +            }
 651 +        }
 652 +    } else {
 653 +        int res = bdrv_write(bs, sector_num, buf, nb_sectors);
 654 +        if (res < 0) {
 655 +            error_setg(errp, "bdrv_write to %s failed (%d)",
 656 +                       bdrv_get_device_name(bs), res);
 657 +            return -1;
 658 +        }
 659 +    }
 660 +    return 0;
 661 +}
 662 +static int restore_extent(VmaReader *vmar, unsigned char *buf,
 663 +                          int extent_size, int vmstate_fd,
 664 +                          bool verbose, Error **errp)
 665 +{
 666 +    assert(vmar);
 667 +    assert(buf);
 668 +
 669 +    VmaExtentHeader *ehead = (VmaExtentHeader *)buf;
 670 +    int start = VMA_EXTENT_HEADER_SIZE;
 671 +    int i;
 672 +
 673 +    for (i = 0; i < VMA_BLOCKS_PER_EXTENT; i++) {
 674 +        uint64_t block_info = GUINT64_FROM_BE(ehead->blockinfo[i]);
 675 +        uint64_t cluster_num = block_info & 0xffffffff;
 676 +        uint8_t dev_id = (block_info >> 32) & 0xff;
 677 +        uint16_t mask = block_info >> (32+16);
 678 +        int64_t max_sector;
 679 +
 680 +        if (!dev_id) {
 681 +            continue;
 682 +        }
 683 +
 684 +        VmaRestoreState *rstate = &vmar->rstate[dev_id];
 685 +        BlockDriverState *bs = NULL;
 686 +
 687 +        if (dev_id != vmar->vmstate_stream) {
 688 +            bs = rstate->bs;
 689 +            if (!bs) {
 690 +                error_setg(errp, "got wrong dev id %d", dev_id);
 691 +                return -1;
 692 +            }
 693 +
 694 +            if (vma_reader_get_bitmap(rstate, cluster_num)) {
 695 +                error_setg(errp, "found duplicated cluster %zd for stream %s",
 696 +                          cluster_num, vmar->devinfo[dev_id].devname);
 697 +                return -1;
 698 +            }
 699 +            vma_reader_set_bitmap(rstate, cluster_num, 1);
 700 +
 701 +            max_sector = vmar->devinfo[dev_id].size/BDRV_SECTOR_SIZE;
 702 +        } else {
 703 +            max_sector = G_MAXINT64;
 704 +            if (cluster_num != vmar->vmstate_clusters) {
 705 +                error_setg(errp, "found out of order vmstate data");
 706 +                return -1;
 707 +            }
 708 +            vmar->vmstate_clusters++;
 709 +        }
 710 +
 711 +        vmar->clusters_read++;
 712 +
 713 +        if (verbose) {
 714 +            time_t duration = time(NULL) - vmar->start_time;
 715 +            int percent = (vmar->clusters_read*100)/vmar->cluster_count;
 716 +            if (percent != vmar->clusters_read_per) {
 717 +                printf("progress %d%% (read %zd bytes, duration %zd sec)\n",
 718 +                       percent, vmar->clusters_read*VMA_CLUSTER_SIZE,
 719 +                       duration);
 720 +                fflush(stdout);
 721 +                vmar->clusters_read_per = percent;
 722 +            }
 723 +        }
 724 +
 725 +        /* try to write whole clusters to speedup restore */
 726 +        if (mask == 0xffff) {
 727 +            if ((start + VMA_CLUSTER_SIZE) > extent_size) {
 728 +                error_setg(errp, "short vma extent - too many blocks");
 729 +                return -1;
 730 +            }
 731 +            int64_t sector_num = (cluster_num * VMA_CLUSTER_SIZE) /
 732 +                BDRV_SECTOR_SIZE;
 733 +            int64_t end_sector = sector_num +
 734 +                VMA_CLUSTER_SIZE/BDRV_SECTOR_SIZE;
 735 +
 736 +            if (end_sector > max_sector) {
 737 +                end_sector = max_sector;
 738 +            }
 739 +
 740 +            if (end_sector <= sector_num) {
 741 +                error_setg(errp, "got wrong block address - 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, buf + start,
 747 +                                   sector_num, nb_sectors, errp) < 0) {
 748 +                return -1;
 749 +            }
 750 +
 751 +            start += VMA_CLUSTER_SIZE;
 752 +        } else {
 753 +            int j;
 754 +            int bit = 1;
 755 +
 756 +            for (j = 0; j < 16; j++) {
 757 +                int64_t sector_num = (cluster_num*VMA_CLUSTER_SIZE +
 758 +                                      j*VMA_BLOCK_SIZE)/BDRV_SECTOR_SIZE;
 759 +
 760 +                int64_t end_sector = sector_num +
 761 +                    VMA_BLOCK_SIZE/BDRV_SECTOR_SIZE;
 762 +                if (end_sector > max_sector) {
 763 +                    end_sector = max_sector;
 764 +                }
 765 +
 766 +                if (mask & bit) {
 767 +                    if ((start + VMA_BLOCK_SIZE) > extent_size) {
 768 +                        error_setg(errp, "short vma extent - too many blocks");
 769 +                        return -1;
 770 +                    }
 771 +
 772 +                    if (end_sector <= sector_num) {
 773 +                        error_setg(errp, "got wrong block address - "
 774 +                                   "write bejond end");
 775 +                        return -1;
 776 +                    }
 777 +
 778 +                    int nb_sectors = end_sector - sector_num;
 779 +                    if (restore_write_data(vmar, dev_id, bs, vmstate_fd,
 780 +                                           buf + start, sector_num,
 781 +                                           nb_sectors, errp) < 0) {
 782 +                        return -1;
 783 +                    }
 784 +
 785 +                    start += VMA_BLOCK_SIZE;
 786 +
 787 +                } else {
 788 +
 789 +                    if (rstate->write_zeroes && (end_sector > sector_num)) {
 790 +                        /* Todo: use bdrv_co_write_zeroes (but that need to
 791 +                         * be run inside coroutine?)
 792 +                         */
 793 +                        int nb_sectors = end_sector - sector_num;
 794 +                        if (restore_write_data(vmar, dev_id, bs, vmstate_fd,
 795 +                                              zero_vma_block, sector_num,
 796 +                                               nb_sectors, errp) < 0) {
 797 +                            return -1;
 798 +                        }
 799 +                    }
 800 +                }
 801 +
 802 +                bit = bit << 1;
 803 +            }
 804 +        }
 805 +    }
 806 +
 807 +    if (start != extent_size) {
 808 +        error_setg(errp, "vma extent error - missing blocks");
 809 +        return -1;
 810 +    }
 811 +
 812 +    return 0;
 813 +}
 814 +
 815 +int vma_reader_restore(VmaReader *vmar, int vmstate_fd, bool verbose,
 816 +                       Error **errp)
 817 +{
 818 +    assert(vmar);
 819 +    assert(vmar->head_data);
 820 +
 821 +    int ret = 0;
 822 +    unsigned char buf[VMA_MAX_EXTENT_SIZE];
 823 +    int buf_pos = 0;
 824 +    unsigned char md5sum[16];
 825 +    VmaHeader *h = (VmaHeader *)vmar->head_data;
 826 +
 827 +    vmar->start_time = time(NULL);
 828 +
 829 +    while (1) {
 830 +        int bytes = full_read(vmar->fd, buf + buf_pos, sizeof(buf) - buf_pos);
 831 +        if (bytes < 0) {
 832 +            error_setg(errp, "read failed - %s", strerror(errno));
 833 +            return -1;
 834 +        }
 835 +
 836 +        buf_pos += bytes;
 837 +
 838 +        if (!buf_pos) {
 839 +            break; /* EOF */
 840 +        }
 841 +
 842 +        if (buf_pos < VMA_EXTENT_HEADER_SIZE) {
 843 +            error_setg(errp, "read short extent (%d bytes)", buf_pos);
 844 +            return -1;
 845 +        }
 846 +
 847 +        VmaExtentHeader *ehead = (VmaExtentHeader *)buf;
 848 +
 849 +        /* extract md5sum */
 850 +        memcpy(md5sum, ehead->md5sum, sizeof(ehead->md5sum));
 851 +        memset(ehead->md5sum, 0, sizeof(ehead->md5sum));
 852 +
 853 +        g_checksum_reset(vmar->md5csum);
 854 +        g_checksum_update(vmar->md5csum, buf, VMA_EXTENT_HEADER_SIZE);
 855 +        gsize csize = 16;
 856 +        g_checksum_get_digest(vmar->md5csum, ehead->md5sum, &csize);
 857 +
 858 +        if (memcmp(md5sum, ehead->md5sum, 16) != 0) {
 859 +            error_setg(errp, "wrong vma extent header chechsum");
 860 +            return -1;
 861 +        }
 862 +
 863 +        if (memcmp(h->uuid, ehead->uuid, sizeof(ehead->uuid)) != 0) {
 864 +            error_setg(errp, "wrong vma extent uuid");
 865 +            return -1;
 866 +        }
 867 +
 868 +        if (ehead->magic != VMA_EXTENT_MAGIC || ehead->reserved1 != 0) {
 869 +            error_setg(errp, "wrong vma extent header magic");
 870 +            return -1;
 871 +        }
 872 +
 873 +        int block_count = GUINT16_FROM_BE(ehead->block_count);
 874 +        int extent_size = VMA_EXTENT_HEADER_SIZE + block_count*VMA_BLOCK_SIZE;
 875 +
 876 +        if (buf_pos < extent_size) {
 877 +            error_setg(errp, "short vma extent (%d < %d)", buf_pos,
 878 +                       extent_size);
 879 +            return -1;
 880 +        }
 881 +
 882 +        if (restore_extent(vmar, buf, extent_size, vmstate_fd, verbose,
 883 +                           errp) < 0) {
 884 +            return -1;
 885 +        }
 886 +
 887 +        if (buf_pos > extent_size) {
 888 +            memmove(buf, buf + extent_size, buf_pos - extent_size);
 889 +            buf_pos = buf_pos - extent_size;
 890 +        } else {
 891 +            buf_pos = 0;
 892 +        }
 893 +    }
 894 +
 895 +    bdrv_drain_all();
 896 +
 897 +    int i;
 898 +    for (i = 1; i < 256; i++) {
 899 +        VmaRestoreState *rstate = &vmar->rstate[i];
 900 +        if (!rstate->bs) {
 901 +            continue;
 902 +        }
 903 +
 904 +        if (bdrv_flush(rstate->bs) < 0) {
 905 +            error_setg(errp, "vma bdrv_flush %s failed",
 906 +                       vmar->devinfo[i].devname);
 907 +            return -1;
 908 +        }
 909 +
 910 +        if (vmar->devinfo[i].size &&
 911 +            (strcmp(vmar->devinfo[i].devname, "vmstate") != 0)) {
 912 +            assert(rstate->bitmap);
 913 +
 914 +            int64_t cluster_num, end;
 915 +
 916 +            end = (vmar->devinfo[i].size + VMA_CLUSTER_SIZE - 1) /
 917 +                VMA_CLUSTER_SIZE;
 918 +
 919 +            for (cluster_num = 0; cluster_num < end; cluster_num++) {
 920 +                if (!vma_reader_get_bitmap(rstate, cluster_num)) {
 921 +                    error_setg(errp, "detected missing cluster %zd "
 922 +                               "for stream %s", cluster_num,
 923 +                               vmar->devinfo[i].devname);
 924 +                    return -1;
 925 +                }
 926 +            }
 927 +        }
 928 +    }
 929 +
 930 +    return ret;
 931 +}
 932 +
 933 diff --git a/vma-writer.c b/vma-writer.c
 934 new file mode 100644
 935 index 0000000..761d7ca
 936 --- /dev/null
 937 +++ b/vma-writer.c
 938 @@ -0,0 +1,932 @@
 939 +/*
 940 + * VMA: Virtual Machine Archive
 941 + *
 942 + * Copyright (C) 2012 Proxmox Server Solutions
 943 + *
 944 + * Authors:
 945 + *  Dietmar Maurer (dietmar@proxmox.com)
 946 + *
 947 + * This work is licensed under the terms of the GNU GPL, version 2 or later.
 948 + * See the COPYING file in the top-level directory.
 949 + *
 950 + */
 951 +
 952 +#include <stdio.h>
 953 +#include <errno.h>
 954 +#include <unistd.h>
 955 +#include <stdio.h>
 956 +#include <string.h>
 957 +#include <sys/types.h>
 958 +#include <sys/stat.h>
 959 +#include <fcntl.h>
 960 +#include <glib.h>
 961 +#include <uuid/uuid.h>
 962 +
 963 +#include "qemu-common.h"
 964 +#include "vma.h"
 965 +#include "block/block.h"
 966 +#include "monitor/monitor.h"
 967 +
 968 +#define DEBUG_VMA 0
 969 +
 970 +#define DPRINTF(fmt, ...)\
 971 +    do { if (DEBUG_VMA) { printf("vma: " fmt, ## __VA_ARGS__); } } while (0)
 972 +
 973 +#define WRITE_BUFFERS 5
 974 +
 975 +typedef struct VmaAIOCB VmaAIOCB;
 976 +struct VmaAIOCB {
 977 +    unsigned char buffer[VMA_MAX_EXTENT_SIZE];
 978 +    VmaWriter *vmaw;
 979 +    size_t bytes;
 980 +    Coroutine *co;
 981 +};
 982 +
 983 +struct VmaWriter {
 984 +    int fd;
 985 +    FILE *cmd;
 986 +    int status;
 987 +    char errmsg[8192];
 988 +    uuid_t uuid;
 989 +    bool header_written;
 990 +    bool closed;
 991 +
 992 +    /* we always write extents */
 993 +    unsigned char outbuf[VMA_MAX_EXTENT_SIZE];
 994 +    int outbuf_pos; /* in bytes */
 995 +    int outbuf_count; /* in VMA_BLOCKS */
 996 +    uint64_t outbuf_block_info[VMA_BLOCKS_PER_EXTENT];
 997 +
 998 +    VmaAIOCB *aiocbs[WRITE_BUFFERS];
 999 +    CoQueue wqueue;
1000 +
1001 +    GChecksum *md5csum;
1002 +    CoMutex writer_lock;
1003 +    CoMutex flush_lock;
1004 +    Coroutine *co_writer;
1005 +
1006 +    /* drive informations */
1007 +    VmaStreamInfo stream_info[256];
1008 +    guint stream_count;
1009 +
1010 +    guint8 vmstate_stream;
1011 +    uint32_t vmstate_clusters;
1012 +
1013 +    /* header blob table */
1014 +    char *header_blob_table;
1015 +    uint32_t header_blob_table_size;
1016 +    uint32_t header_blob_table_pos;
1017 +
1018 +    /* store for config blobs */
1019 +    uint32_t config_names[VMA_MAX_CONFIGS]; /* offset into blob_buffer table */
1020 +    uint32_t config_data[VMA_MAX_CONFIGS];  /* offset into blob_buffer table */
1021 +    uint32_t config_count;
1022 +};
1023 +
1024 +void vma_writer_set_error(VmaWriter *vmaw, const char *fmt, ...)
1025 +{
1026 +    va_list ap;
1027 +
1028 +    if (vmaw->status < 0) {
1029 +        return;
1030 +    }
1031 +
1032 +    vmaw->status = -1;
1033 +
1034 +    va_start(ap, fmt);
1035 +    g_vsnprintf(vmaw->errmsg, sizeof(vmaw->errmsg), fmt, ap);
1036 +    va_end(ap);
1037 +
1038 +    DPRINTF("vma_writer_set_error: %s\n", vmaw->errmsg);
1039 +}
1040 +
1041 +static uint32_t allocate_header_blob(VmaWriter *vmaw, const char *data,
1042 +                                     size_t len)
1043 +{
1044 +    if (len > 65535) {
1045 +        return 0;
1046 +    }
1047 +
1048 +    if (!vmaw->header_blob_table ||
1049 +        (vmaw->header_blob_table_size <
1050 +         (vmaw->header_blob_table_pos + len + 2))) {
1051 +        int newsize = vmaw->header_blob_table_size + ((len + 2 + 511)/512)*512;
1052 +
1053 +        vmaw->header_blob_table = g_realloc(vmaw->header_blob_table, newsize);
1054 +        memset(vmaw->header_blob_table + vmaw->header_blob_table_size,
1055 +               0, newsize - vmaw->header_blob_table_size);
1056 +        vmaw->header_blob_table_size = newsize;
1057 +    }
1058 +
1059 +    uint32_t cpos = vmaw->header_blob_table_pos;
1060 +    vmaw->header_blob_table[cpos] = len & 255;
1061 +    vmaw->header_blob_table[cpos+1] = (len >> 8) & 255;
1062 +    memcpy(vmaw->header_blob_table + cpos + 2, data, len);
1063 +    vmaw->header_blob_table_pos += len + 2;
1064 +    return cpos;
1065 +}
1066 +
1067 +static uint32_t allocate_header_string(VmaWriter *vmaw, const char *str)
1068 +{
1069 +    assert(vmaw);
1070 +
1071 +    size_t len = strlen(str) + 1;
1072 +
1073 +    return allocate_header_blob(vmaw, str, len);
1074 +}
1075 +
1076 +int vma_writer_add_config(VmaWriter *vmaw, const char *name, gpointer data,
1077 +                          gsize len)
1078 +{
1079 +    assert(vmaw);
1080 +    assert(!vmaw->header_written);
1081 +    assert(vmaw->config_count < VMA_MAX_CONFIGS);
1082 +    assert(name);
1083 +    assert(data);
1084 +    assert(len);
1085 +
1086 +    uint32_t name_ptr = allocate_header_string(vmaw, name);
1087 +    if (!name_ptr) {
1088 +        return -1;
1089 +    }
1090 +
1091 +    uint32_t data_ptr = allocate_header_blob(vmaw, data, len);
1092 +    if (!data_ptr) {
1093 +        return -1;
1094 +    }
1095 +
1096 +    vmaw->config_names[vmaw->config_count] = name_ptr;
1097 +    vmaw->config_data[vmaw->config_count] = data_ptr;
1098 +
1099 +    vmaw->config_count++;
1100 +
1101 +    return 0;
1102 +}
1103 +
1104 +int vma_writer_register_stream(VmaWriter *vmaw, const char *devname,
1105 +                               size_t size)
1106 +{
1107 +    assert(vmaw);
1108 +    assert(devname);
1109 +    assert(!vmaw->status);
1110 +
1111 +    if (vmaw->header_written) {
1112 +        vma_writer_set_error(vmaw, "vma_writer_register_stream: header "
1113 +                             "already written");
1114 +        return -1;
1115 +    }
1116 +
1117 +    guint n = vmaw->stream_count + 1;
1118 +
1119 +    /* we can have dev_ids form 1 to 255 (0 reserved)
1120 +     * 255(-1) reseverd for safety
1121 +     */
1122 +    if (n > 254) {
1123 +        vma_writer_set_error(vmaw, "vma_writer_register_stream: "
1124 +                             "too many drives");
1125 +        return -1;
1126 +    }
1127 +
1128 +    if (size <= 0) {
1129 +        vma_writer_set_error(vmaw, "vma_writer_register_stream: "
1130 +                             "got strange size %zd", size);
1131 +        return -1;
1132 +    }
1133 +
1134 +    DPRINTF("vma_writer_register_stream %s %zu %d\n", devname, size, n);
1135 +
1136 +    vmaw->stream_info[n].devname = g_strdup(devname);
1137 +    vmaw->stream_info[n].size = size;
1138 +
1139 +    vmaw->stream_info[n].cluster_count = (size + VMA_CLUSTER_SIZE - 1) /
1140 +        VMA_CLUSTER_SIZE;
1141 +
1142 +    vmaw->stream_count = n;
1143 +
1144 +    if (strcmp(devname, "vmstate") == 0) {
1145 +        vmaw->vmstate_stream = n;
1146 +    }
1147 +
1148 +    return n;
1149 +}
1150 +
1151 +static void vma_co_continue_write(void *opaque)
1152 +{
1153 +    VmaWriter *vmaw = opaque;
1154 +
1155 +    DPRINTF("vma_co_continue_write\n");
1156 +    qemu_coroutine_enter(vmaw->co_writer, NULL);
1157 +}
1158 +
1159 +static int vma_co_write_finished(void *opaque)
1160 +{
1161 +    VmaWriter *vmaw = opaque;
1162 +
1163 +    return (vmaw->co_writer != 0);
1164 +}
1165 +
1166 +static ssize_t coroutine_fn
1167 +vma_co_write(VmaWriter *vmaw, const void *buf, size_t bytes)
1168 +{
1169 +    size_t done = 0;
1170 +    ssize_t ret;
1171 +
1172 +    /* atomic writes (we cannot interleave writes) */
1173 +    qemu_co_mutex_lock(&vmaw->writer_lock);
1174 +
1175 +    DPRINTF("vma_co_write enter %zd\n", bytes);
1176 +
1177 +    assert(vmaw->co_writer == NULL);
1178 +
1179 +    vmaw->co_writer = qemu_coroutine_self();
1180 +
1181 +    qemu_aio_set_fd_handler(vmaw->fd, NULL, vma_co_continue_write,
1182 +                            vma_co_write_finished, vmaw);
1183 +
1184 +    DPRINTF("vma_co_write wait until writable\n");
1185 +    qemu_coroutine_yield();
1186 +    DPRINTF("vma_co_write starting %zd\n", bytes);
1187 +
1188 +    while (done < bytes) {
1189 +        ret = write(vmaw->fd, buf + done, bytes - done);
1190 +        if (ret > 0) {
1191 +            done += ret;
1192 +            DPRINTF("vma_co_write written %zd %zd\n", done, ret);
1193 +        } else if (ret < 0) {
1194 +            if (errno == EAGAIN || errno == EWOULDBLOCK) {
1195 +                DPRINTF("vma_co_write yield %zd\n", done);
1196 +                qemu_coroutine_yield();
1197 +                DPRINTF("vma_co_write restart %zd\n", done);
1198 +            } else {
1199 +                vma_writer_set_error(vmaw, "vma_co_write write error - %s",
1200 +                                     strerror(errno));
1201 +                done = -1; /* always return failure for partial writes */
1202 +                break;
1203 +            }
1204 +        } else if (ret == 0) {
1205 +            /* should not happen - simply try again */
1206 +        }
1207 +    }
1208 +
1209 +    qemu_aio_set_fd_handler(vmaw->fd, NULL, NULL, NULL, NULL);
1210 +
1211 +    vmaw->co_writer = NULL;
1212 +
1213 +    qemu_co_mutex_unlock(&vmaw->writer_lock);
1214 +
1215 +    DPRINTF("vma_co_write leave %zd\n", done);
1216 +    return done;
1217 +}
1218 +
1219 +static void coroutine_fn vma_co_writer_task(void *opaque)
1220 +{
1221 +    VmaAIOCB *cb = opaque;
1222 +
1223 +    DPRINTF("vma_co_writer_task start\n");
1224 +
1225 +    int64_t done = vma_co_write(cb->vmaw, cb->buffer, cb->bytes);
1226 +    DPRINTF("vma_co_writer_task write done %zd\n", done);
1227 +
1228 +    if (done != cb->bytes) {
1229 +        DPRINTF("vma_co_writer_task failed write %zd %zd", cb->bytes, done);
1230 +        vma_writer_set_error(cb->vmaw, "vma_co_writer_task failed write %zd",
1231 +                             done);
1232 +    }
1233 +
1234 +    cb->bytes = 0;
1235 +
1236 +    qemu_co_queue_next(&cb->vmaw->wqueue);
1237 +
1238 +    DPRINTF("vma_co_writer_task end\n");
1239 +}
1240 +
1241 +static void coroutine_fn vma_queue_flush(VmaWriter *vmaw)
1242 +{
1243 +    DPRINTF("vma_queue_flush enter\n");
1244 +
1245 +    assert(vmaw);
1246 +
1247 +    while (1) {
1248 +        int i;
1249 +        VmaAIOCB *cb = NULL;
1250 +        for (i = 0; i < WRITE_BUFFERS; i++) {
1251 +            if (vmaw->aiocbs[i]->bytes) {
1252 +                cb = vmaw->aiocbs[i];
1253 +                DPRINTF("FOUND USED AIO BUFFER %d %zd\n", i,
1254 +                        vmaw->aiocbs[i]->bytes);
1255 +                break;
1256 +            }
1257 +        }
1258 +        if (!cb) {
1259 +            break;
1260 +        }
1261 +        qemu_co_queue_wait(&vmaw->wqueue);
1262 +    }
1263 +
1264 +    DPRINTF("vma_queue_flush leave\n");
1265 +}
1266 +
1267 +/**
1268 + * NOTE: pipe buffer size in only 4096 bytes on linux (see 'ulimit -a')
1269 + * So we need to create a coroutione to allow 'parallel' execution.
1270 + */
1271 +static ssize_t coroutine_fn
1272 +vma_queue_write(VmaWriter *vmaw, const void *buf, size_t bytes)
1273 +{
1274 +    DPRINTF("vma_queue_write enter %zd\n", bytes);
1275 +
1276 +    assert(vmaw);
1277 +    assert(buf);
1278 +    assert(bytes <= VMA_MAX_EXTENT_SIZE);
1279 +
1280 +    VmaAIOCB *cb = NULL;
1281 +    while (!cb) {
1282 +        int i;
1283 +        for (i = 0; i < WRITE_BUFFERS; i++) {
1284 +            if (!vmaw->aiocbs[i]->bytes) {
1285 +                cb = vmaw->aiocbs[i];
1286 +                break;
1287 +            }
1288 +        }
1289 +        if (!cb) {
1290 +            qemu_co_queue_wait(&vmaw->wqueue);
1291 +        }
1292 +    }
1293 +
1294 +    memcpy(cb->buffer, buf, bytes);
1295 +    cb->bytes = bytes;
1296 +    cb->vmaw = vmaw;
1297 +
1298 +    DPRINTF("vma_queue_write start %zd\n", bytes);
1299 +    cb->co = qemu_coroutine_create(vma_co_writer_task);
1300 +    qemu_coroutine_enter(cb->co, cb);
1301 +
1302 +    DPRINTF("vma_queue_write leave\n");
1303 +
1304 +    return bytes;
1305 +}
1306 +
1307 +VmaWriter *vma_writer_create(const char *filename, uuid_t uuid, Error **errp)
1308 +{
1309 +    const char *p;
1310 +
1311 +    assert(sizeof(VmaHeader) == (4096 + 8192));
1312 +    assert(sizeof(VmaExtentHeader) == 512);
1313 +
1314 +    VmaWriter *vmaw = g_new0(VmaWriter, 1);
1315 +    vmaw->fd = -1;
1316 +
1317 +    vmaw->md5csum = g_checksum_new(G_CHECKSUM_MD5);
1318 +    if (!vmaw->md5csum) {
1319 +        error_setg(errp, "can't allocate cmsum\n");
1320 +        goto err;
1321 +    }
1322 +
1323 +    if (strstart(filename, "exec:", &p)) {
1324 +        vmaw->cmd = popen(p, "w");
1325 +        if (vmaw->cmd == NULL) {
1326 +            error_setg(errp, "can't popen command '%s' - %s\n", p,
1327 +                       strerror(errno));
1328 +            goto err;
1329 +        }
1330 +        vmaw->fd = fileno(vmaw->cmd);
1331 +
1332 +        /* try to use O_NONBLOCK and O_DIRECT */
1333 +        fcntl(vmaw->fd, F_SETFL, fcntl(vmaw->fd, F_GETFL)|O_NONBLOCK);
1334 +        fcntl(vmaw->fd, F_SETFL, fcntl(vmaw->fd, F_GETFL)|O_DIRECT);
1335 +
1336 +    } else {
1337 +        struct stat st;
1338 +        int oflags;
1339 +        const char *tmp_id_str;
1340 +
1341 +        if ((stat(filename, &st) == 0) && S_ISFIFO(st.st_mode)) {
1342 +            oflags = O_NONBLOCK|O_DIRECT|O_WRONLY;
1343 +            vmaw->fd = qemu_open(filename, oflags, 0644);
1344 +        } else if (strstart(filename, "/dev/fdset/", &tmp_id_str)) {
1345 +            oflags = O_NONBLOCK|O_DIRECT|O_WRONLY;
1346 +            vmaw->fd = qemu_open(filename, oflags, 0644);
1347 +        } else if (strstart(filename, "/dev/fdname/", &tmp_id_str)) {
1348 +            vmaw->fd = monitor_get_fd(cur_mon, tmp_id_str, errp);
1349 +            if (vmaw->fd < 0) {
1350 +                goto err;
1351 +            }
1352 +            /* try to use O_NONBLOCK and O_DIRECT */
1353 +            fcntl(vmaw->fd, F_SETFL, fcntl(vmaw->fd, F_GETFL)|O_NONBLOCK);
1354 +            fcntl(vmaw->fd, F_SETFL, fcntl(vmaw->fd, F_GETFL)|O_DIRECT);
1355 +        } else  {
1356 +            oflags = O_NONBLOCK|O_DIRECT|O_WRONLY|O_CREAT|O_EXCL;
1357 +            vmaw->fd = qemu_open(filename, oflags, 0644);
1358 +        }
1359 +
1360 +        if (vmaw->fd < 0) {
1361 +            error_setg(errp, "can't open file %s - %s\n", filename,
1362 +                       strerror(errno));
1363 +            goto err;
1364 +        }
1365 +    }
1366 +
1367 +    /* we use O_DIRECT, so we need to align IO buffers */
1368 +    int i;
1369 +    for (i = 0; i < WRITE_BUFFERS; i++) {
1370 +        vmaw->aiocbs[i] = qemu_memalign(512, sizeof(VmaAIOCB));
1371 +        memset(vmaw->aiocbs[i], 0, sizeof(VmaAIOCB));
1372 +    }
1373 +
1374 +    vmaw->outbuf_count = 0;
1375 +    vmaw->outbuf_pos = VMA_EXTENT_HEADER_SIZE;
1376 +
1377 +    vmaw->header_blob_table_pos = 1; /* start at pos 1 */
1378 +
1379 +    qemu_co_mutex_init(&vmaw->writer_lock);
1380 +    qemu_co_mutex_init(&vmaw->flush_lock);
1381 +    qemu_co_queue_init(&vmaw->wqueue);
1382 +
1383 +    uuid_copy(vmaw->uuid, uuid);
1384 +
1385 +    return vmaw;
1386 +
1387 +err:
1388 +    if (vmaw) {
1389 +        if (vmaw->cmd) {
1390 +            pclose(vmaw->cmd);
1391 +        } else if (vmaw->fd >= 0) {
1392 +            close(vmaw->fd);
1393 +        }
1394 +
1395 +        if (vmaw->md5csum) {
1396 +            g_checksum_free(vmaw->md5csum);
1397 +        }
1398 +
1399 +        g_free(vmaw);
1400 +    }
1401 +
1402 +    return NULL;
1403 +}
1404 +
1405 +static int coroutine_fn vma_write_header(VmaWriter *vmaw)
1406 +{
1407 +    assert(vmaw);
1408 +    int header_clusters = 8;
1409 +    char buf[65536*header_clusters];
1410 +    VmaHeader *head = (VmaHeader *)buf;
1411 +
1412 +    int i;
1413 +
1414 +    DPRINTF("VMA WRITE HEADER\n");
1415 +
1416 +    if (vmaw->status < 0) {
1417 +        return vmaw->status;
1418 +    }
1419 +
1420 +    memset(buf, 0, sizeof(buf));
1421 +
1422 +    head->magic = VMA_MAGIC;
1423 +    head->version = GUINT32_TO_BE(1); /* v1 */
1424 +    memcpy(head->uuid, vmaw->uuid, 16);
1425 +
1426 +    time_t ctime = time(NULL);
1427 +    head->ctime = GUINT64_TO_BE(ctime);
1428 +
1429 +    if (!vmaw->stream_count) {
1430 +        return -1;
1431 +    }
1432 +
1433 +    for (i = 0; i < VMA_MAX_CONFIGS; i++) {
1434 +        head->config_names[i] = GUINT32_TO_BE(vmaw->config_names[i]);
1435 +        head->config_data[i] = GUINT32_TO_BE(vmaw->config_data[i]);
1436 +    }
1437 +
1438 +    /* 32 bytes per device (12 used currently) = 8192 bytes max */
1439 +    for (i = 1; i <= 254; i++) {
1440 +        VmaStreamInfo *si = &vmaw->stream_info[i];
1441 +        if (si->size) {
1442 +            assert(si->devname);
1443 +            uint32_t devname_ptr = allocate_header_string(vmaw, si->devname);
1444 +            if (!devname_ptr) {
1445 +                return -1;
1446 +            }
1447 +            head->dev_info[i].devname_ptr = GUINT32_TO_BE(devname_ptr);
1448 +            head->dev_info[i].size = GUINT64_TO_BE(si->size);
1449 +        }
1450 +    }
1451 +
1452 +    uint32_t header_size = sizeof(VmaHeader) + vmaw->header_blob_table_size;
1453 +    head->header_size = GUINT32_TO_BE(header_size);
1454 +
1455 +    if (header_size > sizeof(buf)) {
1456 +        return -1; /* just to be sure */
1457 +    }
1458 +
1459 +    uint32_t blob_buffer_offset = sizeof(VmaHeader);
1460 +    memcpy(buf + blob_buffer_offset, vmaw->header_blob_table,
1461 +           vmaw->header_blob_table_size);
1462 +    head->blob_buffer_offset = GUINT32_TO_BE(blob_buffer_offset);
1463 +    head->blob_buffer_size = GUINT32_TO_BE(vmaw->header_blob_table_pos);
1464 +
1465 +    g_checksum_reset(vmaw->md5csum);
1466 +    g_checksum_update(vmaw->md5csum, (const guchar *)buf, header_size);
1467 +    gsize csize = 16;
1468 +    g_checksum_get_digest(vmaw->md5csum, (guint8 *)(head->md5sum), &csize);
1469 +
1470 +    return vma_queue_write(vmaw, buf, header_size);
1471 +}
1472 +
1473 +static int coroutine_fn vma_writer_flush(VmaWriter *vmaw)
1474 +{
1475 +    assert(vmaw);
1476 +
1477 +    int ret;
1478 +    int i;
1479 +
1480 +    if (vmaw->status < 0) {
1481 +        return vmaw->status;
1482 +    }
1483 +
1484 +    if (!vmaw->header_written) {
1485 +        vmaw->header_written = true;
1486 +        ret = vma_write_header(vmaw);
1487 +        if (ret < 0) {
1488 +            vma_writer_set_error(vmaw, "vma_writer_flush: write header failed");
1489 +            return ret;
1490 +        }
1491 +    }
1492 +
1493 +    DPRINTF("VMA WRITE FLUSH %d %d\n", vmaw->outbuf_count, vmaw->outbuf_pos);
1494 +
1495 +
1496 +    VmaExtentHeader *ehead = (VmaExtentHeader *)vmaw->outbuf;
1497 +
1498 +    ehead->magic = VMA_EXTENT_MAGIC;
1499 +    ehead->reserved1 = 0;
1500 +
1501 +    for (i = 0; i < VMA_BLOCKS_PER_EXTENT; i++) {
1502 +        ehead->blockinfo[i] = GUINT64_TO_BE(vmaw->outbuf_block_info[i]);
1503 +    }
1504 +
1505 +    guint16 block_count = (vmaw->outbuf_pos - VMA_EXTENT_HEADER_SIZE) /
1506 +        VMA_BLOCK_SIZE;
1507 +
1508 +    ehead->block_count = GUINT16_TO_BE(block_count);
1509 +
1510 +    memcpy(ehead->uuid, vmaw->uuid, sizeof(ehead->uuid));
1511 +    memset(ehead->md5sum, 0, sizeof(ehead->md5sum));
1512 +
1513 +    g_checksum_reset(vmaw->md5csum);
1514 +    g_checksum_update(vmaw->md5csum, vmaw->outbuf, VMA_EXTENT_HEADER_SIZE);
1515 +    gsize csize = 16;
1516 +    g_checksum_get_digest(vmaw->md5csum, ehead->md5sum, &csize);
1517 +
1518 +    int bytes = vmaw->outbuf_pos;
1519 +    ret = vma_queue_write(vmaw, vmaw->outbuf, bytes);
1520 +    if (ret != bytes) {
1521 +        vma_writer_set_error(vmaw, "vma_writer_flush: failed write");
1522 +    }
1523 +
1524 +    vmaw->outbuf_count = 0;
1525 +    vmaw->outbuf_pos = VMA_EXTENT_HEADER_SIZE;
1526 +
1527 +    for (i = 0; i < VMA_BLOCKS_PER_EXTENT; i++) {
1528 +        vmaw->outbuf_block_info[i] = 0;
1529 +    }
1530 +
1531 +    return vmaw->status;
1532 +}
1533 +
1534 +static int vma_count_open_streams(VmaWriter *vmaw)
1535 +{
1536 +    g_assert(vmaw != NULL);
1537 +
1538 +    int i;
1539 +    int open_drives = 0;
1540 +    for (i = 0; i <= 255; i++) {
1541 +        if (vmaw->stream_info[i].size && !vmaw->stream_info[i].finished) {
1542 +            open_drives++;
1543 +        }
1544 +    }
1545 +
1546 +    return open_drives;
1547 +}
1548 +
1549 +/**
1550 + * all jobs should call this when there is no more data
1551 + * Returns: number of remaining stream (0 ==> finished)
1552 + */
1553 +int coroutine_fn
1554 +vma_writer_close_stream(VmaWriter *vmaw, uint8_t dev_id)
1555 +{
1556 +    g_assert(vmaw != NULL);
1557 +
1558 +    DPRINTF("vma_writer_set_status %d\n", dev_id);
1559 +    if (!vmaw->stream_info[dev_id].size) {
1560 +        vma_writer_set_error(vmaw, "vma_writer_close_stream: "
1561 +                             "no such stream %d", dev_id);
1562 +        return -1;
1563 +    }
1564 +    if (vmaw->stream_info[dev_id].finished) {
1565 +        vma_writer_set_error(vmaw, "vma_writer_close_stream: "
1566 +                             "stream already closed %d", dev_id);
1567 +        return -1;
1568 +    }
1569 +
1570 +    vmaw->stream_info[dev_id].finished = true;
1571 +
1572 +    int open_drives = vma_count_open_streams(vmaw);
1573 +
1574 +    if (open_drives <= 0) {
1575 +        DPRINTF("vma_writer_set_status all drives completed\n");
1576 +        qemu_co_mutex_lock(&vmaw->flush_lock);
1577 +        int ret = vma_writer_flush(vmaw);
1578 +        qemu_co_mutex_unlock(&vmaw->flush_lock);
1579 +        if (ret < 0) {
1580 +            vma_writer_set_error(vmaw, "vma_writer_close_stream: flush failed");
1581 +        }
1582 +    }
1583 +
1584 +    return open_drives;
1585 +}
1586 +
1587 +int vma_writer_get_status(VmaWriter *vmaw, VmaStatus *status)
1588 +{
1589 +    int i;
1590 +
1591 +    g_assert(vmaw != NULL);
1592 +
1593 +    if (status) {
1594 +        status->status = vmaw->status;
1595 +        g_strlcpy(status->errmsg, vmaw->errmsg, sizeof(status->errmsg));
1596 +        for (i = 0; i <= 255; i++) {
1597 +            status->stream_info[i] = vmaw->stream_info[i];
1598 +        }
1599 +
1600 +        uuid_unparse_lower(vmaw->uuid, status->uuid_str);
1601 +    }
1602 +
1603 +    status->closed = vmaw->closed;
1604 +
1605 +    return vmaw->status;
1606 +}
1607 +
1608 +static int vma_writer_get_buffer(VmaWriter *vmaw)
1609 +{
1610 +    int ret = 0;
1611 +
1612 +    qemu_co_mutex_lock(&vmaw->flush_lock);
1613 +
1614 +    /* wait until buffer is available */
1615 +    while (vmaw->outbuf_count >= (VMA_BLOCKS_PER_EXTENT - 1)) {
1616 +        ret = vma_writer_flush(vmaw);
1617 +        if (ret < 0) {
1618 +            vma_writer_set_error(vmaw, "vma_writer_get_buffer: flush failed");
1619 +            break;
1620 +        }
1621 +    }
1622 +
1623 +    qemu_co_mutex_unlock(&vmaw->flush_lock);
1624 +
1625 +    return ret;
1626 +}
1627 +
1628 +
1629 +int64_t coroutine_fn
1630 +vma_writer_write(VmaWriter *vmaw, uint8_t dev_id, int64_t cluster_num,
1631 +                 unsigned char *buf, size_t *zero_bytes)
1632 +{
1633 +    g_assert(vmaw != NULL);
1634 +    g_assert(zero_bytes != NULL);
1635 +
1636 +    *zero_bytes = 0;
1637 +
1638 +    if (vmaw->status < 0) {
1639 +        return vmaw->status;
1640 +    }
1641 +
1642 +    if (!dev_id || !vmaw->stream_info[dev_id].size) {
1643 +        vma_writer_set_error(vmaw, "vma_writer_write: "
1644 +                             "no such stream %d", dev_id);
1645 +        return -1;
1646 +    }
1647 +
1648 +    if (vmaw->stream_info[dev_id].finished) {
1649 +        vma_writer_set_error(vmaw, "vma_writer_write: "
1650 +                             "stream already closed %d", dev_id);
1651 +        return -1;
1652 +    }
1653 +
1654 +
1655 +    if (cluster_num >= (((uint64_t)1)<<32)) {
1656 +        vma_writer_set_error(vmaw, "vma_writer_write: "
1657 +                             "cluster number out of range");
1658 +        return -1;
1659 +    }
1660 +
1661 +    if (dev_id == vmaw->vmstate_stream) {
1662 +        if (cluster_num != vmaw->vmstate_clusters) {
1663 +            vma_writer_set_error(vmaw, "vma_writer_write: "
1664 +                                 "non sequential vmstate write");
1665 +        }
1666 +        vmaw->vmstate_clusters++;
1667 +    } else if (cluster_num >= vmaw->stream_info[dev_id].cluster_count) {
1668 +        vma_writer_set_error(vmaw, "vma_writer_write: cluster number too big");
1669 +        return -1;
1670 +    }
1671 +
1672 +    /* wait until buffer is available */
1673 +    if (vma_writer_get_buffer(vmaw) < 0) {
1674 +        vma_writer_set_error(vmaw, "vma_writer_write: "
1675 +                             "vma_writer_get_buffer failed");
1676 +        return -1;
1677 +    }
1678 +
1679 +    DPRINTF("VMA WRITE %d %zd\n", dev_id, cluster_num);
1680 +
1681 +    int i;
1682 +    int bit = 1;
1683 +    uint16_t mask = 0;
1684 +    for (i = 0; i < 16; i++) {
1685 +        unsigned char *vmablock = buf + (i*VMA_BLOCK_SIZE);
1686 +        if (!buffer_is_zero(vmablock, VMA_BLOCK_SIZE)) {
1687 +            mask |= bit;
1688 +            memcpy(vmaw->outbuf + vmaw->outbuf_pos, vmablock, VMA_BLOCK_SIZE);
1689 +            vmaw->outbuf_pos += VMA_BLOCK_SIZE;
1690 +        } else {
1691 +            DPRINTF("VMA WRITE %zd ZERO BLOCK %d\n", cluster_num, i);
1692 +            vmaw->stream_info[dev_id].zero_bytes += VMA_BLOCK_SIZE;
1693 +            *zero_bytes += VMA_BLOCK_SIZE;
1694 +        }
1695 +
1696 +        bit = bit << 1;
1697 +    }
1698 +
1699 +    uint64_t block_info = ((uint64_t)mask) << (32+16);
1700 +    block_info |= ((uint64_t)dev_id) << 32;
1701 +    block_info |= (cluster_num & 0xffffffff);
1702 +    vmaw->outbuf_block_info[vmaw->outbuf_count] = block_info;
1703 +
1704 +    DPRINTF("VMA WRITE MASK %zd %zx\n", cluster_num, block_info);
1705 +
1706 +    vmaw->outbuf_count++;
1707 +
1708 +    /** NOTE: We allways write whole clusters, but we correctly set
1709 +     * transferred bytes. So transferred == size when when everything
1710 +     * went OK.
1711 +     */
1712 +    size_t transferred = VMA_CLUSTER_SIZE;
1713 +
1714 +    if (dev_id != vmaw->vmstate_stream) {
1715 +        uint64_t last = (cluster_num + 1) * VMA_CLUSTER_SIZE;
1716 +        if (last > vmaw->stream_info[dev_id].size) {
1717 +            uint64_t diff = last - vmaw->stream_info[dev_id].size;
1718 +            if (diff >= VMA_CLUSTER_SIZE) {
1719 +                vma_writer_set_error(vmaw, "vma_writer_write: "
1720 +                                     "read after last cluster");
1721 +                return -1;
1722 +            }
1723 +            transferred -= diff;
1724 +        }
1725 +    }
1726 +
1727 +    vmaw->stream_info[dev_id].transferred += transferred;
1728 +
1729 +    return transferred;
1730 +}
1731 +
1732 +int vma_writer_close(VmaWriter *vmaw, Error **errp)
1733 +{
1734 +    g_assert(vmaw != NULL);
1735 +
1736 +    int i;
1737 +
1738 +    vma_queue_flush(vmaw);
1739 +
1740 +    /* this should not happen - just to be sure */
1741 +    while (!qemu_co_queue_empty(&vmaw->wqueue)) {
1742 +        DPRINTF("vma_writer_close wait\n");
1743 +        co_sleep_ns(rt_clock, 1000000);
1744 +    }
1745 +
1746 +    if (vmaw->cmd) {
1747 +        if (pclose(vmaw->cmd) < 0) {
1748 +            vma_writer_set_error(vmaw, "vma_writer_close: "
1749 +                                 "pclose failed - %s", strerror(errno));
1750 +        }
1751 +    } else {
1752 +        if (close(vmaw->fd) < 0) {
1753 +            vma_writer_set_error(vmaw, "vma_writer_close: "
1754 +                                 "close failed - %s", strerror(errno));
1755 +        }
1756 +    }
1757 +
1758 +    for (i = 0; i <= 255; i++) {
1759 +        VmaStreamInfo *si = &vmaw->stream_info[i];
1760 +        if (si->size) {
1761 +            if (!si->finished) {
1762 +                vma_writer_set_error(vmaw, "vma_writer_close: "
1763 +                                     "detected open stream '%s'", si->devname);
1764 +            } else if ((si->transferred != si->size) &&
1765 +                       (i != vmaw->vmstate_stream)) {
1766 +                vma_writer_set_error(vmaw, "vma_writer_close: "
1767 +                                     "incomplete stream '%s' (%zd != %zd)",
1768 +                                     si->devname, si->transferred, si->size);
1769 +            }
1770 +        }
1771 +    }
1772 +
1773 +    for (i = 0; i <= 255; i++) {
1774 +        vmaw->stream_info[i].finished = 1; /* mark as closed */
1775 +    }
1776 +
1777 +    vmaw->closed = 1;
1778 +
1779 +    if (vmaw->status < 0 && *errp == NULL) {
1780 +        error_setg(errp, "%s", vmaw->errmsg);
1781 +    }
1782 +
1783 +    return vmaw->status;
1784 +}
1785 +
1786 +void vma_writer_destroy(VmaWriter *vmaw)
1787 +{
1788 +    assert(vmaw);
1789 +
1790 +    int i;
1791 +
1792 +    for (i = 0; i <= 255; i++) {
1793 +        if (vmaw->stream_info[i].devname) {
1794 +            g_free(vmaw->stream_info[i].devname);
1795 +        }
1796 +    }
1797 +
1798 +    if (vmaw->md5csum) {
1799 +        g_checksum_free(vmaw->md5csum);
1800 +    }
1801 +
1802 +    for (i = 0; i < WRITE_BUFFERS; i++) {
1803 +        free(vmaw->aiocbs[i]);
1804 +    }
1805 +
1806 +    g_free(vmaw);
1807 +}
1808 +
1809 +/* backup driver plugin */
1810 +
1811 +static int vma_dump_cb(void *opaque, uint8_t dev_id, int64_t cluster_num,
1812 +                       unsigned char *buf, size_t *zero_bytes)
1813 +{
1814 +    VmaWriter *vmaw = opaque;
1815 +
1816 +    return vma_writer_write(vmaw, dev_id, cluster_num, buf, zero_bytes);
1817 +}
1818 +
1819 +static int vma_close_cb(void *opaque, Error **errp)
1820 +{
1821 +    VmaWriter *vmaw = opaque;
1822 +
1823 +    int res = vma_writer_close(vmaw, errp);
1824 +    vma_writer_destroy(vmaw);
1825 +
1826 +    return res;
1827 +}
1828 +
1829 +static int vma_complete_cb(void *opaque, uint8_t dev_id, int ret)
1830 +{
1831 +    VmaWriter *vmaw = opaque;
1832 +
1833 +    if (ret < 0) {
1834 +        vma_writer_set_error(vmaw, "backup_complete_cb %d", ret);
1835 +    }
1836 +
1837 +    return vma_writer_close_stream(vmaw, dev_id);
1838 +}
1839 +
1840 +static int vma_register_stream_cb(void *opaque, const char *devname,
1841 +                                  size_t size)
1842 +{
1843 +    VmaWriter *vmaw = opaque;
1844 +
1845 +    return vma_writer_register_stream(vmaw, devname, size);
1846 +}
1847 +
1848 +static int vma_register_config_cb(void *opaque, const char *name,
1849 +                                  gpointer data, size_t data_len)
1850 +{
1851 +    VmaWriter *vmaw = opaque;
1852 +
1853 +    return vma_writer_add_config(vmaw, name, data, data_len);
1854 +}
1855 +
1856 +static void *vma_open_cb(const char *filename, uuid_t uuid, Error **errp)
1857 +{
1858 +    return vma_writer_create(filename, uuid, errp);
1859 +}
1860 +
1861 +const BackupDriver backup_vma_driver = {
1862 +    .format = "vma",
1863 +    .open_cb = vma_open_cb,
1864 +    .close_cb = vma_close_cb,
1865 +    .register_config_cb = vma_register_config_cb,
1866 +    .register_stream_cb = vma_register_stream_cb,
1867 +    .dump_cb = vma_dump_cb,
1868 +    .complete_cb = vma_complete_cb,
1869 +};
1870 +
1871 diff --git a/vma.c b/vma.c
1872 new file mode 100644
1873 index 0000000..b2e276c
1874 --- /dev/null
1875 +++ b/vma.c
1876 @@ -0,0 +1,559 @@
1877 +/*
1878 + * VMA: Virtual Machine Archive
1879 + *
1880 + * Copyright (C) 2012 Proxmox Server Solutions
1881 + *
1882 + * Authors:
1883 + *  Dietmar Maurer (dietmar@proxmox.com)
1884 + *
1885 + * This work is licensed under the terms of the GNU GPL, version 2 or later.
1886 + * See the COPYING file in the top-level directory.
1887 + *
1888 + */
1889 +
1890 +#include <stdio.h>
1891 +#include <errno.h>
1892 +#include <unistd.h>
1893 +#include <stdio.h>
1894 +#include <string.h>
1895 +#include <sys/types.h>
1896 +#include <sys/stat.h>
1897 +#include <fcntl.h>
1898 +#include <glib.h>
1899 +
1900 +#include "qemu-common.h"
1901 +#include "qemu/error-report.h"
1902 +#include "vma.h"
1903 +#include "block/block.h"
1904 +
1905 +static void help(void)
1906 +{
1907 +    const char *help_msg =
1908 +        "usage: vma command [command options]\n"
1909 +        "\n"
1910 +        "vma list <filename>\n"
1911 +        "vma create <filename> [-c config] <archive> pathname ...\n"
1912 +        "vma extract <filename> [-r] <targetdir>\n"
1913 +        ;
1914 +
1915 +    printf("%s", help_msg);
1916 +    exit(1);
1917 +}
1918 +
1919 +static const char *extract_devname(const char *path, char **devname, int index)
1920 +{
1921 +    assert(path);
1922 +
1923 +    const char *sep = strchr(path, '=');
1924 +
1925 +    if (sep) {
1926 +        *devname = g_strndup(path, sep - path);
1927 +        path = sep + 1;
1928 +    } else {
1929 +        if (index >= 0) {
1930 +            *devname = g_strdup_printf("disk%d", index);
1931 +        } else {
1932 +            *devname = NULL;
1933 +        }
1934 +    }
1935 +
1936 +    return path;
1937 +}
1938 +
1939 +static void print_content(VmaReader *vmar)
1940 +{
1941 +    assert(vmar);
1942 +
1943 +    VmaHeader *head = vma_reader_get_header(vmar);
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 +        printf("CFG: size: %d name: %s\n", cdata->len, cdata->name);
1950 +    }
1951 +
1952 +    int i;
1953 +    VmaDeviceInfo *di;
1954 +    for (i = 1; i < 255; i++) {
1955 +        di = vma_reader_get_device_info(vmar, i);
1956 +        if (di) {
1957 +            if (strcmp(di->devname, "vmstate") == 0) {
1958 +                printf("VMSTATE: dev_id=%d memory: %zd\n", i, di->size);
1959 +            } else {
1960 +                printf("DEV: dev_id=%d size: %zd devname: %s\n",
1961 +                       i, di->size, di->devname);
1962 +            }
1963 +        }
1964 +    }
1965 +    /* ctime is the last entry we print */
1966 +    printf("CTIME: %s", ctime(&head->ctime));
1967 +    fflush(stdout);
1968 +}
1969 +
1970 +static int list_content(int argc, char **argv)
1971 +{
1972 +    int c, ret = 0;
1973 +    const char *filename;
1974 +
1975 +    for (;;) {
1976 +        c = getopt(argc, argv, "h");
1977 +        if (c == -1) {
1978 +            break;
1979 +        }
1980 +        switch (c) {
1981 +        case '?':
1982 +        case 'h':
1983 +            help();
1984 +            break;
1985 +        default:
1986 +            g_assert_not_reached();
1987 +        }
1988 +    }
1989 +
1990 +    /* Get the filename */
1991 +    if ((optind + 1) != argc) {
1992 +        help();
1993 +    }
1994 +    filename = argv[optind++];
1995 +
1996 +    Error *errp = NULL;
1997 +    VmaReader *vmar = vma_reader_create(filename, &errp);
1998 +
1999 +    if (!vmar) {
2000 +        g_error("%s", error_get_pretty(errp));
2001 +    }
2002 +
2003 +    print_content(vmar);
2004 +
2005 +    vma_reader_destroy(vmar);
2006 +
2007 +    return ret;
2008 +}
2009 +
2010 +typedef struct RestoreMap {
2011 +    char *devname;
2012 +    char *path;
2013 +    bool write_zero;
2014 +} RestoreMap;
2015 +
2016 +static int extract_content(int argc, char **argv)
2017 +{
2018 +    int c, ret = 0;
2019 +    int verbose = 0;
2020 +    const char *filename;
2021 +    const char *dirname;
2022 +    const char *readmap = NULL;
2023 +
2024 +    for (;;) {
2025 +        c = getopt(argc, argv, "hvr:");
2026 +        if (c == -1) {
2027 +            break;
2028 +        }
2029 +        switch (c) {
2030 +        case '?':
2031 +        case 'h':
2032 +            help();
2033 +            break;
2034 +        case 'r':
2035 +            readmap = optarg;
2036 +            break;
2037 +        case 'v':
2038 +            verbose = 1;
2039 +            break;
2040 +        default:
2041 +            help();
2042 +        }
2043 +    }
2044 +
2045 +    /* Get the filename */
2046 +    if ((optind + 2) != argc) {
2047 +        help();
2048 +    }
2049 +    filename = argv[optind++];
2050 +    dirname = argv[optind++];
2051 +
2052 +    Error *errp = NULL;
2053 +    VmaReader *vmar = vma_reader_create(filename, &errp);
2054 +
2055 +    if (!vmar) {
2056 +        g_error("%s", error_get_pretty(errp));
2057 +    }
2058 +
2059 +    if (mkdir(dirname, 0777) < 0) {
2060 +        g_error("unable to create target directory %s - %s",
2061 +                dirname, strerror(errno));
2062 +    }
2063 +
2064 +    GList *l = vma_reader_get_config_data(vmar);
2065 +    while (l && l->data) {
2066 +        VmaConfigData *cdata = (VmaConfigData *)l->data;
2067 +        l = g_list_next(l);
2068 +        char *cfgfn = g_strdup_printf("%s/%s", dirname, cdata->name);
2069 +        GError *err = NULL;
2070 +        if (!g_file_set_contents(cfgfn, (gchar *)cdata->data, cdata->len,
2071 +                                 &err)) {
2072 +            g_error("unable to write file: %s", err->message);
2073 +        }
2074 +    }
2075 +
2076 +    GHashTable *devmap = g_hash_table_new(g_str_hash, g_str_equal);
2077 +
2078 +    if (readmap) {
2079 +        print_content(vmar);
2080 +
2081 +        FILE *map = fopen(readmap, "r");
2082 +        if (!map) {
2083 +            g_error("unable to open fifo %s - %s", readmap, strerror(errno));
2084 +        }
2085 +
2086 +        while (1) {
2087 +            char inbuf[8192];
2088 +            char *line = fgets(inbuf, sizeof(inbuf), map);
2089 +            if (!line || line[0] == '\0' || !strcmp(line, "done\n")) {
2090 +                break;
2091 +            }
2092 +            int len = strlen(line);
2093 +            if (line[len - 1] == '\n') {
2094 +                line[len - 1] = '\0';
2095 +                if (len == 1) {
2096 +                    break;
2097 +                }
2098 +            }
2099 +
2100 +            const char *path;
2101 +            bool write_zero;
2102 +            if (line[0] == '0' && line[1] == ':') {
2103 +                path = inbuf + 2;
2104 +                write_zero = false;
2105 +            } else if (line[0] == '1' && line[1] == ':') {
2106 +                path = inbuf + 2;
2107 +                write_zero = true;
2108 +            } else {
2109 +                g_error("read map failed - parse error ('%s')", inbuf);
2110 +            }
2111 +
2112 +            char *devname = NULL;
2113 +            path = extract_devname(path, &devname, -1);
2114 +            if (!devname) {
2115 +                g_error("read map failed - no dev name specified ('%s')",
2116 +                        inbuf);
2117 +            }
2118 +
2119 +            RestoreMap *map = g_new0(RestoreMap, 1);
2120 +            map->devname = g_strdup(devname);
2121 +            map->path = g_strdup(path);
2122 +            map->write_zero = write_zero;
2123 +
2124 +            g_hash_table_insert(devmap, map->devname, map);
2125 +
2126 +        };
2127 +    }
2128 +
2129 +    int i;
2130 +    int vmstate_fd = -1;
2131 +    guint8 vmstate_stream = 0;
2132 +
2133 +    for (i = 1; i < 255; i++) {
2134 +        VmaDeviceInfo *di = vma_reader_get_device_info(vmar, i);
2135 +        if (di && (strcmp(di->devname, "vmstate") == 0)) {
2136 +            vmstate_stream = i;
2137 +            char *statefn = g_strdup_printf("%s/vmstate.bin", dirname);
2138 +            vmstate_fd = open(statefn, O_WRONLY|O_CREAT|O_EXCL, 0644);
2139 +            if (vmstate_fd < 0) {
2140 +                g_error("create vmstate file '%s' failed - %s", statefn,
2141 +                        strerror(errno));
2142 +            }
2143 +            g_free(statefn);
2144 +        } else if (di) {
2145 +            char *devfn = NULL;
2146 +            int flags = BDRV_O_RDWR|BDRV_O_CACHE_WB;
2147 +            bool write_zero = true;
2148 +
2149 +            if (readmap) {
2150 +                RestoreMap *map;
2151 +                map = (RestoreMap *)g_hash_table_lookup(devmap, di->devname);
2152 +                if (map == NULL) {
2153 +                    g_error("no device name mapping for %s", di->devname);
2154 +                }
2155 +                devfn = map->path;
2156 +                write_zero = map->write_zero;
2157 +            } else {
2158 +                devfn = g_strdup_printf("%s/tmp-disk-%s.raw",
2159 +                                        dirname, di->devname);
2160 +                printf("DEVINFO %s %zd\n", devfn, di->size);
2161 +
2162 +                bdrv_img_create(devfn, "raw", NULL, NULL, NULL, di->size,
2163 +                                flags, &errp);
2164 +                if (error_is_set(&errp)) {
2165 +                    g_error("can't create file %s: %s", devfn,
2166 +                            error_get_pretty(errp));
2167 +                }
2168 +
2169 +                /* Note: we created an empty file above, so there is no
2170 +                 * need to write zeroes (so we generate a sparse file)
2171 +                 */
2172 +                write_zero = false;
2173 +            }
2174 +
2175 +            BlockDriverState *bs = bdrv_new(di->devname);
2176 +            if (bdrv_open(bs, devfn, flags, NULL)) {
2177 +                g_error("can't open file %s", devfn);
2178 +            }
2179 +            if (vma_reader_register_bs(vmar, i, bs, write_zero, &errp) < 0) {
2180 +                g_error("%s", error_get_pretty(errp));
2181 +            }
2182 +
2183 +            if (!readmap) {
2184 +                g_free(devfn);
2185 +            }
2186 +        }
2187 +    }
2188 +
2189 +    if (vma_reader_restore(vmar, vmstate_fd, verbose, &errp) < 0) {
2190 +        g_error("restore failed - %s", error_get_pretty(errp));
2191 +    }
2192 +
2193 +    if (!readmap) {
2194 +        for (i = 1; i < 255; i++) {
2195 +            VmaDeviceInfo *di = vma_reader_get_device_info(vmar, i);
2196 +            if (di && (i != vmstate_stream)) {
2197 +                char *tmpfn = g_strdup_printf("%s/tmp-disk-%s.raw",
2198 +                                              dirname, di->devname);
2199 +                char *fn = g_strdup_printf("%s/disk-%s.raw",
2200 +                                           dirname, di->devname);
2201 +                if (rename(tmpfn, fn) != 0) {
2202 +                    g_error("rename %s to %s failed - %s",
2203 +                            tmpfn, fn, strerror(errno));
2204 +                }
2205 +            }
2206 +        }
2207 +    }
2208 +
2209 +    vma_reader_destroy(vmar);
2210 +
2211 +    bdrv_close_all();
2212 +
2213 +    return ret;
2214 +}
2215 +
2216 +typedef struct BackupCB {
2217 +    VmaWriter *vmaw;
2218 +    uint8_t dev_id;
2219 +} BackupCB;
2220 +
2221 +static int backup_dump_cb(void *opaque, BlockDriverState *bs,
2222 +                          int64_t cluster_num, unsigned char *buf)
2223 +{
2224 +    BackupCB *bcb = opaque;
2225 +    size_t zb = 0;
2226 +    if (vma_writer_write(bcb->vmaw, bcb->dev_id, cluster_num, buf, &zb) < 0) {
2227 +        g_warning("backup_dump_cb vma_writer_write failed");
2228 +        return -1;
2229 +    }
2230 +
2231 +    return 0;
2232 +}
2233 +
2234 +static void backup_complete_cb(void *opaque, int ret)
2235 +{
2236 +    BackupCB *bcb = opaque;
2237 +
2238 +    if (ret < 0) {
2239 +        vma_writer_set_error(bcb->vmaw, "backup_complete_cb %d", ret);
2240 +    }
2241 +
2242 +    if (vma_writer_close_stream(bcb->vmaw, bcb->dev_id) <= 0) {
2243 +        Error *err = NULL;
2244 +        if (vma_writer_close(bcb->vmaw, &err) != 0) {
2245 +            g_warning("vma_writer_close failed %s", error_get_pretty(err));
2246 +        }
2247 +    }
2248 +}
2249 +
2250 +static int create_archive(int argc, char **argv)
2251 +{
2252 +    int i, c, res;
2253 +    int verbose = 0;
2254 +    const char *archivename;
2255 +    GList *config_files = NULL;
2256 +
2257 +    for (;;) {
2258 +        c = getopt(argc, argv, "hvc:");
2259 +        if (c == -1) {
2260 +            break;
2261 +        }
2262 +        switch (c) {
2263 +        case '?':
2264 +        case 'h':
2265 +            help();
2266 +            break;
2267 +        case 'c':
2268 +            config_files = g_list_append(config_files, optarg);
2269 +            break;
2270 +        case 'v':
2271 +            verbose = 1;
2272 +            break;
2273 +        default:
2274 +            g_assert_not_reached();
2275 +        }
2276 +    }
2277 +
2278 +
2279 +    /* make sure we have archive name and at least one path */
2280 +    if ((optind + 2) > argc) {
2281 +        help();
2282 +    }
2283 +
2284 +    archivename = argv[optind++];
2285 +
2286 +    uuid_t uuid;
2287 +    uuid_generate(uuid);
2288 +
2289 +    Error *local_err = NULL;
2290 +    VmaWriter *vmaw = vma_writer_create(archivename, uuid, &local_err);
2291 +
2292 +    if (vmaw == NULL) {
2293 +        g_error("%s", error_get_pretty(local_err));
2294 +    }
2295 +
2296 +    GList *l = config_files;
2297 +    while (l && l->data) {
2298 +        char *name = l->data;
2299 +        char *cdata = NULL;
2300 +        gsize clen = 0;
2301 +        GError *err = NULL;
2302 +        if (!g_file_get_contents(name, &cdata, &clen, &err)) {
2303 +            unlink(archivename);
2304 +            g_error("Unable to read file: %s", err->message);
2305 +        }
2306 +
2307 +        if (vma_writer_add_config(vmaw, name, cdata, clen) != 0) {
2308 +            unlink(archivename);
2309 +            g_error("Unable to append config data %s (len = %zd)",
2310 +                    name, clen);
2311 +        }
2312 +        l = g_list_next(l);
2313 +    }
2314 +
2315 +    int ind = 0;
2316 +    while (optind < argc) {
2317 +        const char *path = argv[optind++];
2318 +        char *devname = NULL;
2319 +        path = extract_devname(path, &devname, ind++);
2320 +
2321 +        BlockDriver *drv = NULL;
2322 +        BlockDriverState *bs = bdrv_new(devname);
2323 +
2324 +        res = bdrv_open(bs, path, BDRV_O_CACHE_WB , drv);
2325 +        if (res < 0) {
2326 +            unlink(archivename);
2327 +            g_error("bdrv_open '%s' failed", path);
2328 +        }
2329 +        int64_t size = bdrv_getlength(bs);
2330 +        int dev_id = vma_writer_register_stream(vmaw, devname, size);
2331 +        if (dev_id <= 0) {
2332 +            unlink(archivename);
2333 +            g_error("vma_writer_register_stream '%s' failed", devname);
2334 +        }
2335 +
2336 +        BackupCB *bcb = g_new0(BackupCB, 1);
2337 +        bcb->vmaw = vmaw;
2338 +        bcb->dev_id = dev_id;
2339 +
2340 +        if (backup_job_create(bs, backup_dump_cb, backup_complete_cb,
2341 +                              bcb, 0) < 0) {
2342 +            unlink(archivename);
2343 +            g_error("backup_job_start failed");
2344 +        } else {
2345 +            backup_job_start(bs, false);
2346 +        }
2347 +    }
2348 +
2349 +    VmaStatus vmastat;
2350 +    int percent = 0;
2351 +    int last_percent = -1;
2352 +
2353 +    while (1) {
2354 +        main_loop_wait(false);
2355 +        vma_writer_get_status(vmaw, &vmastat);
2356 +
2357 +        if (verbose) {
2358 +
2359 +            uint64_t total = 0;
2360 +            uint64_t transferred = 0;
2361 +            uint64_t zero_bytes = 0;
2362 +
2363 +            int i;
2364 +            for (i = 0; i < 256; i++) {
2365 +                if (vmastat.stream_info[i].size) {
2366 +                    total += vmastat.stream_info[i].size;
2367 +                    transferred += vmastat.stream_info[i].transferred;
2368 +                    zero_bytes += vmastat.stream_info[i].zero_bytes;
2369 +                }
2370 +            }
2371 +            percent = (transferred*100)/total;
2372 +            if (percent != last_percent) {
2373 +                printf("progress %d%% %zd/%zd %zd\n", percent,
2374 +                       transferred, total, zero_bytes);
2375 +
2376 +                last_percent = percent;
2377 +            }
2378 +        }
2379 +
2380 +        if (vmastat.closed) {
2381 +            break;
2382 +        }
2383 +    }
2384 +
2385 +    bdrv_drain_all();
2386 +
2387 +    vma_writer_get_status(vmaw, &vmastat);
2388 +
2389 +    if (verbose) {
2390 +        for (i = 0; i < 256; i++) {
2391 +            VmaStreamInfo *si = &vmastat.stream_info[i];
2392 +            if (si->size) {
2393 +                printf("image %s: size=%zd zeros=%zd saved=%zd\n", si->devname,
2394 +                       si->size, si->zero_bytes, si->size - si->zero_bytes);
2395 +            }
2396 +        }
2397 +    }
2398 +
2399 +    if (vmastat.status < 0) {
2400 +        unlink(archivename);
2401 +        g_error("creating vma archive failed");
2402 +    }
2403 +
2404 +    return 0;
2405 +}
2406 +
2407 +int main(int argc, char **argv)
2408 +{
2409 +    const char *cmdname;
2410 +
2411 +    error_set_progname(argv[0]);
2412 +
2413 +    qemu_init_main_loop();
2414 +
2415 +    bdrv_init();
2416 +
2417 +    if (argc < 2) {
2418 +        help();
2419 +    }
2420 +
2421 +    cmdname = argv[1];
2422 +    argc--; argv++;
2423 +
2424 +
2425 +    if (!strcmp(cmdname, "list")) {
2426 +        return list_content(argc, argv);
2427 +    } else if (!strcmp(cmdname, "create")) {
2428 +        return create_archive(argc, argv);
2429 +    } else if (!strcmp(cmdname, "extract")) {
2430 +        return extract_content(argc, argv);
2431 +    }
2432 +
2433 +    help();
2434 +    return 0;
2435 +}
2436 diff --git a/vma.h b/vma.h
2437 new file mode 100644
2438 index 0000000..76d0dc8
2439 --- /dev/null
2440 +++ b/vma.h
2441 @@ -0,0 +1,145 @@
2442 +/*
2443 + * VMA: Virtual Machine Archive
2444 + *
2445 + * Copyright (C) Proxmox Server Solutions
2446 + *
2447 + * Authors:
2448 + *  Dietmar Maurer (dietmar@proxmox.com)
2449 + *
2450 + * This work is licensed under the terms of the GNU GPL, version 2 or later.
2451 + * See the COPYING file in the top-level directory.
2452 + *
2453 + */
2454 +
2455 +#ifndef BACKUP_VMA_H
2456 +#define BACKUP_VMA_H
2457 +
2458 +#include "backup.h"
2459 +#include "error.h"
2460 +
2461 +#define VMA_BLOCK_BITS 12
2462 +#define VMA_BLOCK_SIZE (1<<VMA_BLOCK_BITS)
2463 +#define VMA_CLUSTER_BITS (VMA_BLOCK_BITS+4)
2464 +#define VMA_CLUSTER_SIZE (1<<VMA_CLUSTER_BITS)
2465 +
2466 +#if VMA_CLUSTER_SIZE != 65536
2467 +#error unexpected cluster size
2468 +#endif
2469 +
2470 +#define VMA_EXTENT_HEADER_SIZE 512
2471 +#define VMA_BLOCKS_PER_EXTENT 59
2472 +#define VMA_MAX_CONFIGS 256
2473 +
2474 +#define VMA_MAX_EXTENT_SIZE \
2475 +    (VMA_EXTENT_HEADER_SIZE+VMA_CLUSTER_SIZE*VMA_BLOCKS_PER_EXTENT)
2476 +#if VMA_MAX_EXTENT_SIZE != 3867136
2477 +#error unexpected VMA_EXTENT_SIZE
2478 +#endif
2479 +
2480 +/* File Format Definitions */
2481 +
2482 +#define VMA_MAGIC (GUINT32_TO_BE(('V'<<24)|('M'<<16)|('A'<<8)|0x00))
2483 +#define VMA_EXTENT_MAGIC (GUINT32_TO_BE(('V'<<24)|('M'<<16)|('A'<<8)|'E'))
2484 +
2485 +typedef struct VmaDeviceInfoHeader {
2486 +    uint32_t devname_ptr; /* offset into blob_buffer table */
2487 +    uint32_t reserved0;
2488 +    uint64_t size; /* device size in bytes */
2489 +    uint64_t reserved1;
2490 +    uint64_t reserved2;
2491 +} VmaDeviceInfoHeader;
2492 +
2493 +typedef struct VmaHeader {
2494 +    uint32_t magic;
2495 +    uint32_t version;
2496 +    unsigned char uuid[16];
2497 +    int64_t ctime;
2498 +    unsigned char md5sum[16];
2499 +
2500 +    uint32_t blob_buffer_offset;
2501 +    uint32_t blob_buffer_size;
2502 +    uint32_t header_size;
2503 +
2504 +    unsigned char reserved[1984];
2505 +
2506 +    uint32_t config_names[VMA_MAX_CONFIGS]; /* offset into blob_buffer table */
2507 +    uint32_t config_data[VMA_MAX_CONFIGS];  /* offset into blob_buffer table */
2508 +
2509 +    VmaDeviceInfoHeader dev_info[256];
2510 +} VmaHeader;
2511 +
2512 +typedef struct VmaExtentHeader {
2513 +    uint32_t magic;
2514 +    uint16_t reserved1;
2515 +    uint16_t block_count;
2516 +    unsigned char uuid[16];
2517 +    unsigned char md5sum[16];
2518 +    uint64_t blockinfo[VMA_BLOCKS_PER_EXTENT];
2519 +} VmaExtentHeader;
2520 +
2521 +/* functions/definitions to read/write vma files */
2522 +
2523 +typedef struct VmaReader VmaReader;
2524 +
2525 +typedef struct VmaWriter VmaWriter;
2526 +
2527 +typedef struct VmaConfigData {
2528 +    const char *name;
2529 +    const void *data;
2530 +    uint32_t len;
2531 +} VmaConfigData;
2532 +
2533 +typedef struct VmaStreamInfo {
2534 +    uint64_t size;
2535 +    uint64_t cluster_count;
2536 +    uint64_t transferred;
2537 +    uint64_t zero_bytes;
2538 +    int finished;
2539 +    char *devname;
2540 +} VmaStreamInfo;
2541 +
2542 +typedef struct VmaStatus {
2543 +    int status;
2544 +    bool closed;
2545 +    char errmsg[8192];
2546 +    char uuid_str[37];
2547 +    VmaStreamInfo stream_info[256];
2548 +} VmaStatus;
2549 +
2550 +typedef struct VmaDeviceInfo {
2551 +    uint64_t size; /* device size in bytes */
2552 +    const char *devname;
2553 +} VmaDeviceInfo;
2554 +
2555 +extern const BackupDriver backup_vma_driver;
2556 +
2557 +VmaWriter *vma_writer_create(const char *filename, uuid_t uuid, Error **errp);
2558 +int vma_writer_close(VmaWriter *vmaw, Error **errp);
2559 +void vma_writer_destroy(VmaWriter *vmaw);
2560 +int vma_writer_add_config(VmaWriter *vmaw, const char *name, gpointer data,
2561 +                          size_t len);
2562 +int vma_writer_register_stream(VmaWriter *vmaw, const char *devname,
2563 +                               size_t size);
2564 +
2565 +int64_t coroutine_fn vma_writer_write(VmaWriter *vmaw, uint8_t dev_id,
2566 +                                      int64_t cluster_num, unsigned char *buf,
2567 +                                      size_t *zero_bytes);
2568 +
2569 +int coroutine_fn vma_writer_close_stream(VmaWriter *vmaw, uint8_t dev_id);
2570 +
2571 +int vma_writer_get_status(VmaWriter *vmaw, VmaStatus *status);
2572 +void vma_writer_set_error(VmaWriter *vmaw, const char *fmt, ...);
2573 +
2574 +
2575 +VmaReader *vma_reader_create(const char *filename, Error **errp);
2576 +void vma_reader_destroy(VmaReader *vmar);
2577 +VmaHeader *vma_reader_get_header(VmaReader *vmar);
2578 +GList *vma_reader_get_config_data(VmaReader *vmar);
2579 +VmaDeviceInfo *vma_reader_get_device_info(VmaReader *vmar, guint8 dev_id);
2580 +int vma_reader_register_bs(VmaReader *vmar, guint8 dev_id,
2581 +                           BlockDriverState *bs, bool write_zeroes,
2582 +                           Error **errp);
2583 +int vma_reader_restore(VmaReader *vmar, int vmstate_fd, bool verbose,
2584 +                       Error **errp);
2585 +
2586 +#endif /* BACKUP_VMA_H */
2587 --
2588 1.7.2.5
2589