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

   1 From 6289a43696ca6f713a5d3bb9f95a5adb608a5e13 Mon Sep 17 00:00:00 2001
   2 From: Dietmar Maurer <dietmar@proxmox.com>
   3 Date: Tue, 13 Nov 2012 11:11:38 +0100
   4 Subject: [PATCH v5 4/6] introduce new vma archive format
   5
   6 This is a very simple archive format, see docs/specs/vma_spec.txt
   7
   8 Signed-off-by: Dietmar Maurer <dietmar@proxmox.com>
   9 ---
  10  Makefile                |    3 +-
  11  Makefile.objs           |    2 +-
  12  blockdev.c              |    6 +-
  13  docs/specs/vma_spec.txt |   24 ++
  14  vma-reader.c            |  799 ++++++++++++++++++++++++++++++++++++++++
  15  vma-writer.c            |  940 +++++++++++++++++++++++++++++++++++++++++++++++
  16  vma.c                   |  561 ++++++++++++++++++++++++++++
  17  vma.h                   |  145 ++++++++
  18  8 files changed, 2476 insertions(+), 4 deletions(-)
  19  create mode 100644 docs/specs/vma_spec.txt
  20  create mode 100644 vma-reader.c
  21  create mode 100644 vma-writer.c
  22  create mode 100644 vma.c
  23  create mode 100644 vma.h
  24
  25 Index: new/Makefile
  26 ===================================================================
  27 --- new.orig/Makefile   2014-06-17 06:14:27.000000000 +0200
  28 +++ new/Makefile        2014-06-17 06:15:02.000000000 +0200
  29 @@ -146,7 +146,7 @@
  30  include $(SRC_PATH)/libcacard/Makefile
  31  endif
  32
  33 -all: $(DOCS) $(TOOLS) $(HELPERS-y) recurse-all modules
  34 +all: $(DOCS) $(TOOLS) vma$(EXESUF) $(HELPERS-y) recurse-all modules
  35
  36  config-host.h: config-host.h-timestamp
  37  config-host.h-timestamp: config-host.mak
  38 @@ -214,6 +214,7 @@
  39  qemu-img$(EXESUF): qemu-img.o $(block-obj-y) libqemuutil.a libqemustub.a
  40  qemu-nbd$(EXESUF): qemu-nbd.o $(block-obj-y) libqemuutil.a libqemustub.a
  41  qemu-io$(EXESUF): qemu-io.o $(block-obj-y) libqemuutil.a libqemustub.a
  42 +vma$(EXESUF): vma.o vma-reader.o $(block-obj-y)  libqemuutil.a libqemustub.a
  43
  44  qemu-bridge-helper$(EXESUF): qemu-bridge-helper.o
  45
  46 Index: new/vma-reader.c
  47 ===================================================================
  48 --- /dev/null   1970-01-01 00:00:00.000000000 +0000
  49 +++ new/vma-reader.c    2014-06-17 06:15:02.000000000 +0200
  50 @@ -0,0 +1,799 @@
  51 +/*
  52 + * VMA: Virtual Machine Archive
  53 + *
  54 + * Copyright (C) 2012 Proxmox Server Solutions
  55 + *
  56 + * Authors:
  57 + *  Dietmar Maurer (dietmar@proxmox.com)
  58 + *
  59 + * This work is licensed under the terms of the GNU GPL, version 2 or later.
  60 + * See the COPYING file in the top-level directory.
  61 + *
  62 + */
  63 +
  64 +#include <stdio.h>
  65 +#include <errno.h>
  66 +#include <unistd.h>
  67 +#include <stdio.h>
  68 +#include <string.h>
  69 +#include <sys/types.h>
  70 +#include <sys/stat.h>
  71 +#include <fcntl.h>
  72 +#include <glib.h>
  73 +#include <uuid/uuid.h>
  74 +
  75 +#include "qemu-common.h"
  76 +#include "qemu/timer.h"
  77 +#include "qemu/ratelimit.h"
  78 +#include "vma.h"
  79 +#include "block/block.h"
  80 +
  81 +#define BITS_PER_LONG  (sizeof(unsigned long) * CHAR_BIT)
  82 +
  83 +static unsigned char zero_vma_block[VMA_BLOCK_SIZE];
  84 +
  85 +typedef struct VmaRestoreState {
  86 +    BlockDriverState *bs;
  87 +    bool write_zeroes;
  88 +    unsigned long *bitmap;
  89 +    int bitmap_size;
  90 +}  VmaRestoreState;
  91 +
  92 +struct VmaReader {
  93 +    int fd;
  94 +    GChecksum *md5csum;
  95 +    GHashTable *blob_hash;
  96 +    unsigned char *head_data;
  97 +    VmaDeviceInfo devinfo[256];
  98 +    VmaRestoreState rstate[256];
  99 +    GList *cdata_list;
 100 +    guint8 vmstate_stream;
 101 +    uint32_t vmstate_clusters;
 102 +    /* to show restore percentage if run with -v */
 103 +    time_t start_time;
 104 +    int64_t cluster_count;
 105 +    int64_t clusters_read;
 106 +    int clusters_read_per;
 107 +};
 108 +
 109 +static guint
 110 +g_int32_hash(gconstpointer v)
 111 +{
 112 +    return *(const uint32_t *)v;
 113 +}
 114 +
 115 +static gboolean
 116 +g_int32_equal(gconstpointer v1, gconstpointer v2)
 117 +{
 118 +    return *((const uint32_t *)v1) == *((const uint32_t *)v2);
 119 +}
 120 +
 121 +static int vma_reader_get_bitmap(VmaRestoreState *rstate, int64_t cluster_num)
 122 +{
 123 +    assert(rstate);
 124 +    assert(rstate->bitmap);
 125 +
 126 +    unsigned long val, idx, bit;
 127 +
 128 +    idx = cluster_num / BITS_PER_LONG;
 129 +
 130 +    assert(rstate->bitmap_size > idx);
 131 +
 132 +    bit = cluster_num % BITS_PER_LONG;
 133 +    val = rstate->bitmap[idx];
 134 +
 135 +    return !!(val & (1UL << bit));
 136 +}
 137 +
 138 +static void vma_reader_set_bitmap(VmaRestoreState *rstate, int64_t cluster_num,
 139 +                                  int dirty)
 140 +{
 141 +    assert(rstate);
 142 +    assert(rstate->bitmap);
 143 +
 144 +    unsigned long val, idx, bit;
 145 +
 146 +    idx = cluster_num / BITS_PER_LONG;
 147 +
 148 +    assert(rstate->bitmap_size > idx);
 149 +
 150 +    bit = cluster_num % BITS_PER_LONG;
 151 +    val = rstate->bitmap[idx];
 152 +    if (dirty) {
 153 +        if (!(val & (1UL << bit))) {
 154 +            val |= 1UL << bit;
 155 +        }
 156 +    } else {
 157 +        if (val & (1UL << bit)) {
 158 +            val &= ~(1UL << bit);
 159 +        }
 160 +    }
 161 +    rstate->bitmap[idx] = val;
 162 +}
 163 +
 164 +typedef struct VmaBlob {
 165 +    uint32_t start;
 166 +    uint32_t len;
 167 +    void *data;
 168 +} VmaBlob;
 169 +
 170 +static const VmaBlob *get_header_blob(VmaReader *vmar, uint32_t pos)
 171 +{
 172 +    assert(vmar);
 173 +    assert(vmar->blob_hash);
 174 +
 175 +    return g_hash_table_lookup(vmar->blob_hash, &pos);
 176 +}
 177 +
 178 +static const char *get_header_str(VmaReader *vmar, uint32_t pos)
 179 +{
 180 +    const VmaBlob *blob = get_header_blob(vmar, pos);
 181 +    if (!blob) {
 182 +        return NULL;
 183 +    }
 184 +    const char *res = (char *)blob->data;
 185 +    if (res[blob->len-1] != '\0') {
 186 +        return NULL;
 187 +    }
 188 +    return res;
 189 +}
 190 +
 191 +static ssize_t
 192 +safe_read(int fd, unsigned char *buf, size_t count)
 193 +{
 194 +    ssize_t n;
 195 +
 196 +    do {
 197 +        n = read(fd, buf, count);
 198 +    } while (n < 0 && errno == EINTR);
 199 +
 200 +    return n;
 201 +}
 202 +
 203 +static ssize_t
 204 +full_read(int fd, unsigned char *buf, size_t len)
 205 +{
 206 +    ssize_t n;
 207 +    size_t total;
 208 +
 209 +    total = 0;
 210 +
 211 +    while (len > 0) {
 212 +        n = safe_read(fd, buf, len);
 213 +
 214 +        if (n == 0) {
 215 +            return total;
 216 +        }
 217 +
 218 +        if (n <= 0) {
 219 +            break;
 220 +        }
 221 +
 222 +        buf += n;
 223 +        total += n;
 224 +        len -= n;
 225 +    }
 226 +
 227 +    if (len) {
 228 +        return -1;
 229 +    }
 230 +
 231 +    return total;
 232 +}
 233 +
 234 +void vma_reader_destroy(VmaReader *vmar)
 235 +{
 236 +    assert(vmar);
 237 +
 238 +    if (vmar->fd >= 0) {
 239 +        close(vmar->fd);
 240 +    }
 241 +
 242 +    if (vmar->cdata_list) {
 243 +        g_list_free(vmar->cdata_list);
 244 +    }
 245 +
 246 +    int i;
 247 +    for (i = 1; i < 256; i++) {
 248 +        if (vmar->rstate[i].bitmap) {
 249 +            g_free(vmar->rstate[i].bitmap);
 250 +        }
 251 +    }
 252 +
 253 +    if (vmar->md5csum) {
 254 +        g_checksum_free(vmar->md5csum);
 255 +    }
 256 +
 257 +    if (vmar->blob_hash) {
 258 +        g_hash_table_destroy(vmar->blob_hash);
 259 +    }
 260 +
 261 +    if (vmar->head_data) {
 262 +        g_free(vmar->head_data);
 263 +    }
 264 +
 265 +    g_free(vmar);
 266 +
 267 +};
 268 +
 269 +static int vma_reader_read_head(VmaReader *vmar, Error **errp)
 270 +{
 271 +    assert(vmar);
 272 +    assert(errp);
 273 +    assert(*errp == NULL);
 274 +
 275 +    unsigned char md5sum[16];
 276 +    int i;
 277 +    int ret = 0;
 278 +
 279 +    vmar->head_data = g_malloc(sizeof(VmaHeader));
 280 +
 281 +    if (full_read(vmar->fd, vmar->head_data, sizeof(VmaHeader)) !=
 282 +        sizeof(VmaHeader)) {
 283 +        error_setg(errp, "can't read vma header - %s",
 284 +                   errno ? g_strerror(errno) : "got EOF");
 285 +        return -1;
 286 +    }
 287 +
 288 +    VmaHeader *h = (VmaHeader *)vmar->head_data;
 289 +
 290 +    if (h->magic != VMA_MAGIC) {
 291 +        error_setg(errp, "not a vma file - wrong magic number");
 292 +        return -1;
 293 +    }
 294 +
 295 +    uint32_t header_size = GUINT32_FROM_BE(h->header_size);
 296 +    int need = header_size - sizeof(VmaHeader);
 297 +    if (need <= 0) {
 298 +        error_setg(errp, "wrong vma header size %d", header_size);
 299 +        return -1;
 300 +    }
 301 +
 302 +    vmar->head_data = g_realloc(vmar->head_data, header_size);
 303 +    h = (VmaHeader *)vmar->head_data;
 304 +
 305 +    if (full_read(vmar->fd, vmar->head_data + sizeof(VmaHeader), need) !=
 306 +        need) {
 307 +        error_setg(errp, "can't read vma header data - %s",
 308 +                   errno ? g_strerror(errno) : "got EOF");
 309 +        return -1;
 310 +    }
 311 +
 312 +    memcpy(md5sum, h->md5sum, 16);
 313 +    memset(h->md5sum, 0, 16);
 314 +
 315 +    g_checksum_reset(vmar->md5csum);
 316 +    g_checksum_update(vmar->md5csum, vmar->head_data, header_size);
 317 +    gsize csize = 16;
 318 +    g_checksum_get_digest(vmar->md5csum, (guint8 *)(h->md5sum), &csize);
 319 +
 320 +    if (memcmp(md5sum, h->md5sum, 16) != 0) {
 321 +        error_setg(errp, "wrong vma header chechsum");
 322 +        return -1;
 323 +    }
 324 +
 325 +    /* we can modify header data after checksum verify */
 326 +    h->header_size = header_size;
 327 +
 328 +    h->version = GUINT32_FROM_BE(h->version);
 329 +    if (h->version != 1) {
 330 +        error_setg(errp, "wrong vma version %d", h->version);
 331 +        return -1;
 332 +    }
 333 +
 334 +    h->ctime = GUINT64_FROM_BE(h->ctime);
 335 +    h->blob_buffer_offset = GUINT32_FROM_BE(h->blob_buffer_offset);
 336 +    h->blob_buffer_size = GUINT32_FROM_BE(h->blob_buffer_size);
 337 +
 338 +    uint32_t bstart = h->blob_buffer_offset + 1;
 339 +    uint32_t bend = h->blob_buffer_offset + h->blob_buffer_size;
 340 +
 341 +    if (bstart <= sizeof(VmaHeader)) {
 342 +        error_setg(errp, "wrong vma blob buffer offset %d",
 343 +                   h->blob_buffer_offset);
 344 +        return -1;
 345 +    }
 346 +
 347 +    if (bend > header_size) {
 348 +        error_setg(errp, "wrong vma blob buffer size %d/%d",
 349 +                   h->blob_buffer_offset, h->blob_buffer_size);
 350 +        return -1;
 351 +    }
 352 +
 353 +    while ((bstart + 2) <= bend) {
 354 +        uint32_t size = vmar->head_data[bstart] +
 355 +            (vmar->head_data[bstart+1] << 8);
 356 +        if ((bstart + size + 2) <= bend) {
 357 +            VmaBlob *blob = g_new0(VmaBlob, 1);
 358 +            blob->start = bstart - h->blob_buffer_offset;
 359 +            blob->len = size;
 360 +            blob->data = vmar->head_data + bstart + 2;
 361 +            g_hash_table_insert(vmar->blob_hash, &blob->start, blob);
 362 +        }
 363 +        bstart += size + 2;
 364 +    }
 365 +
 366 +
 367 +    int count = 0;
 368 +    for (i = 1; i < 256; i++) {
 369 +        VmaDeviceInfoHeader *dih = &h->dev_info[i];
 370 +        uint32_t devname_ptr = GUINT32_FROM_BE(dih->devname_ptr);
 371 +        uint64_t size = GUINT64_FROM_BE(dih->size);
 372 +        const char *devname =  get_header_str(vmar, devname_ptr);
 373 +
 374 +        if (size && devname) {
 375 +            count++;
 376 +            vmar->devinfo[i].size = size;
 377 +            vmar->devinfo[i].devname = devname;
 378 +
 379 +            if (strcmp(devname, "vmstate") == 0) {
 380 +                vmar->vmstate_stream = i;
 381 +            }
 382 +        }
 383 +    }
 384 +
 385 +    if (!count) {
 386 +        error_setg(errp, "vma does not contain data");
 387 +        return -1;
 388 +    }
 389 +
 390 +    for (i = 0; i < VMA_MAX_CONFIGS; i++) {
 391 +        uint32_t name_ptr = GUINT32_FROM_BE(h->config_names[i]);
 392 +        uint32_t data_ptr = GUINT32_FROM_BE(h->config_data[i]);
 393 +
 394 +        if (!(name_ptr && data_ptr)) {
 395 +            continue;
 396 +        }
 397 +        const char *name =  get_header_str(vmar, name_ptr);
 398 +        const VmaBlob *blob = get_header_blob(vmar, data_ptr);
 399 +
 400 +        if (!(name && blob)) {
 401 +            error_setg(errp, "vma contains invalid data pointers");
 402 +            return -1;
 403 +        }
 404 +
 405 +        VmaConfigData *cdata = g_new0(VmaConfigData, 1);
 406 +        cdata->name = name;
 407 +        cdata->data = blob->data;
 408 +        cdata->len = blob->len;
 409 +
 410 +        vmar->cdata_list = g_list_append(vmar->cdata_list, cdata);
 411 +    }
 412 +
 413 +    return ret;
 414 +};
 415 +
 416 +VmaReader *vma_reader_create(const char *filename, Error **errp)
 417 +{
 418 +    assert(filename);
 419 +    assert(errp);
 420 +
 421 +    VmaReader *vmar = g_new0(VmaReader, 1);
 422 +
 423 +    if (strcmp(filename, "-") == 0) {
 424 +        vmar->fd = dup(0);
 425 +    } else {
 426 +        vmar->fd = open(filename, O_RDONLY);
 427 +    }
 428 +
 429 +    if (vmar->fd < 0) {
 430 +        error_setg(errp, "can't open file %s - %s\n", filename,
 431 +                   g_strerror(errno));
 432 +        goto err;
 433 +    }
 434 +
 435 +    vmar->md5csum = g_checksum_new(G_CHECKSUM_MD5);
 436 +    if (!vmar->md5csum) {
 437 +        error_setg(errp, "can't allocate cmsum\n");
 438 +        goto err;
 439 +    }
 440 +
 441 +    vmar->blob_hash = g_hash_table_new_full(g_int32_hash, g_int32_equal,
 442 +                                            NULL, g_free);
 443 +
 444 +    if (vma_reader_read_head(vmar, errp) < 0) {
 445 +        goto err;
 446 +    }
 447 +
 448 +    return vmar;
 449 +
 450 +err:
 451 +    if (vmar) {
 452 +        vma_reader_destroy(vmar);
 453 +    }
 454 +
 455 +    return NULL;
 456 +}
 457 +
 458 +VmaHeader *vma_reader_get_header(VmaReader *vmar)
 459 +{
 460 +    assert(vmar);
 461 +    assert(vmar->head_data);
 462 +
 463 +    return (VmaHeader *)(vmar->head_data);
 464 +}
 465 +
 466 +GList *vma_reader_get_config_data(VmaReader *vmar)
 467 +{
 468 +    assert(vmar);
 469 +    assert(vmar->head_data);
 470 +
 471 +    return vmar->cdata_list;
 472 +}
 473 +
 474 +VmaDeviceInfo *vma_reader_get_device_info(VmaReader *vmar, guint8 dev_id)
 475 +{
 476 +    assert(vmar);
 477 +    assert(dev_id);
 478 +
 479 +    if (vmar->devinfo[dev_id].size && vmar->devinfo[dev_id].devname) {
 480 +        return &vmar->devinfo[dev_id];
 481 +    }
 482 +
 483 +    return NULL;
 484 +}
 485 +
 486 +int vma_reader_register_bs(VmaReader *vmar, guint8 dev_id, BlockDriverState *bs,
 487 +                           bool write_zeroes, Error **errp)
 488 +{
 489 +    assert(vmar);
 490 +    assert(bs != NULL);
 491 +    assert(dev_id);
 492 +    assert(vmar->rstate[dev_id].bs == NULL);
 493 +
 494 +    int64_t size = bdrv_getlength(bs);
 495 +    if (size != vmar->devinfo[dev_id].size) {
 496 +        error_setg(errp, "vma_reader_register_bs for stream %s failed - "
 497 +                   "unexpected size %zd != %zd", vmar->devinfo[dev_id].devname,
 498 +                   size, vmar->devinfo[dev_id].size);
 499 +        return -1;
 500 +    }
 501 +
 502 +    vmar->rstate[dev_id].bs = bs;
 503 +    vmar->rstate[dev_id].write_zeroes = write_zeroes;
 504 +
 505 +    int64_t bitmap_size = (size/BDRV_SECTOR_SIZE) +
 506 +        (VMA_CLUSTER_SIZE/BDRV_SECTOR_SIZE) * BITS_PER_LONG - 1;
 507 +    bitmap_size /= (VMA_CLUSTER_SIZE/BDRV_SECTOR_SIZE) * BITS_PER_LONG;
 508 +
 509 +    vmar->rstate[dev_id].bitmap_size = bitmap_size;
 510 +    vmar->rstate[dev_id].bitmap = g_new0(unsigned long, bitmap_size);
 511 +
 512 +    vmar->cluster_count += size/VMA_CLUSTER_SIZE;
 513 +
 514 +    return 0;
 515 +}
 516 +
 517 +static ssize_t safe_write(int fd, void *buf, size_t count)
 518 +{
 519 +    ssize_t n;
 520 +
 521 +    do {
 522 +        n = write(fd, buf, count);
 523 +    } while (n < 0 && errno == EINTR);
 524 +
 525 +    return n;
 526 +}
 527 +
 528 +static size_t full_write(int fd, void *buf, size_t len)
 529 +{
 530 +    ssize_t n;
 531 +    size_t total;
 532 +
 533 +    total = 0;
 534 +
 535 +    while (len > 0) {
 536 +        n = safe_write(fd, buf, len);
 537 +        if (n < 0) {
 538 +            return n;
 539 +        }
 540 +        buf += n;
 541 +        total += n;
 542 +        len -= n;
 543 +    }
 544 +
 545 +    if (len) {
 546 +        /* incomplete write ? */
 547 +        return -1;
 548 +    }
 549 +
 550 +    return total;
 551 +}
 552 +
 553 +static int restore_write_data(VmaReader *vmar, guint8 dev_id,
 554 +                              BlockDriverState *bs, int vmstate_fd,
 555 +                              unsigned char *buf, int64_t sector_num,
 556 +                              int nb_sectors, Error **errp)
 557 +{
 558 +    assert(vmar);
 559 +
 560 +    if (dev_id == vmar->vmstate_stream) {
 561 +        if (vmstate_fd >= 0) {
 562 +            int len = nb_sectors * BDRV_SECTOR_SIZE;
 563 +            int res = full_write(vmstate_fd, buf, len);
 564 +            if (res < 0) {
 565 +                error_setg(errp, "write vmstate failed %d", res);
 566 +                return -1;
 567 +            }
 568 +        }
 569 +    } else {
 570 +        int res = bdrv_write(bs, sector_num, buf, nb_sectors);
 571 +        if (res < 0) {
 572 +            error_setg(errp, "bdrv_write to %s failed (%d)",
 573 +                       bdrv_get_device_name(bs), res);
 574 +            return -1;
 575 +        }
 576 +    }
 577 +    return 0;
 578 +}
 579 +static int restore_extent(VmaReader *vmar, unsigned char *buf,
 580 +                          int extent_size, int vmstate_fd,
 581 +                          bool verbose, Error **errp)
 582 +{
 583 +    assert(vmar);
 584 +    assert(buf);
 585 +
 586 +    VmaExtentHeader *ehead = (VmaExtentHeader *)buf;
 587 +    int start = VMA_EXTENT_HEADER_SIZE;
 588 +    int i;
 589 +
 590 +    for (i = 0; i < VMA_BLOCKS_PER_EXTENT; i++) {
 591 +        uint64_t block_info = GUINT64_FROM_BE(ehead->blockinfo[i]);
 592 +        uint64_t cluster_num = block_info & 0xffffffff;
 593 +        uint8_t dev_id = (block_info >> 32) & 0xff;
 594 +        uint16_t mask = block_info >> (32+16);
 595 +        int64_t max_sector;
 596 +
 597 +        if (!dev_id) {
 598 +            continue;
 599 +        }
 600 +
 601 +        VmaRestoreState *rstate = &vmar->rstate[dev_id];
 602 +        BlockDriverState *bs = NULL;
 603 +
 604 +        if (dev_id != vmar->vmstate_stream) {
 605 +            bs = rstate->bs;
 606 +            if (!bs) {
 607 +                error_setg(errp, "got wrong dev id %d", dev_id);
 608 +                return -1;
 609 +            }
 610 +
 611 +            if (vma_reader_get_bitmap(rstate, cluster_num)) {
 612 +                error_setg(errp, "found duplicated cluster %zd for stream %s",
 613 +                          cluster_num, vmar->devinfo[dev_id].devname);
 614 +                return -1;
 615 +            }
 616 +            vma_reader_set_bitmap(rstate, cluster_num, 1);
 617 +
 618 +            max_sector = vmar->devinfo[dev_id].size/BDRV_SECTOR_SIZE;
 619 +        } else {
 620 +            max_sector = G_MAXINT64;
 621 +            if (cluster_num != vmar->vmstate_clusters) {
 622 +                error_setg(errp, "found out of order vmstate data");
 623 +                return -1;
 624 +            }
 625 +            vmar->vmstate_clusters++;
 626 +        }
 627 +
 628 +        vmar->clusters_read++;
 629 +
 630 +        if (verbose) {
 631 +            time_t duration = time(NULL) - vmar->start_time;
 632 +            int percent = (vmar->clusters_read*100)/vmar->cluster_count;
 633 +            if (percent != vmar->clusters_read_per) {
 634 +                printf("progress %d%% (read %zd bytes, duration %zd sec)\n",
 635 +                       percent, vmar->clusters_read*VMA_CLUSTER_SIZE,
 636 +                       duration);
 637 +                fflush(stdout);
 638 +                vmar->clusters_read_per = percent;
 639 +            }
 640 +        }
 641 +
 642 +        /* try to write whole clusters to speedup restore */
 643 +        if (mask == 0xffff) {
 644 +            if ((start + VMA_CLUSTER_SIZE) > extent_size) {
 645 +                error_setg(errp, "short vma extent - too many blocks");
 646 +                return -1;
 647 +            }
 648 +            int64_t sector_num = (cluster_num * VMA_CLUSTER_SIZE) /
 649 +                BDRV_SECTOR_SIZE;
 650 +            int64_t end_sector = sector_num +
 651 +                VMA_CLUSTER_SIZE/BDRV_SECTOR_SIZE;
 652 +
 653 +            if (end_sector > max_sector) {
 654 +                end_sector = max_sector;
 655 +            }
 656 +
 657 +            if (end_sector <= sector_num) {
 658 +                error_setg(errp, "got wrong block address - write bejond end");
 659 +                return -1;
 660 +            }
 661 +
 662 +            int nb_sectors = end_sector - sector_num;
 663 +            if (restore_write_data(vmar, dev_id, bs, vmstate_fd, buf + start,
 664 +                                   sector_num, nb_sectors, errp) < 0) {
 665 +                return -1;
 666 +            }
 667 +
 668 +            start += VMA_CLUSTER_SIZE;
 669 +        } else {
 670 +            int j;
 671 +            int bit = 1;
 672 +
 673 +            for (j = 0; j < 16; j++) {
 674 +                int64_t sector_num = (cluster_num*VMA_CLUSTER_SIZE +
 675 +                                      j*VMA_BLOCK_SIZE)/BDRV_SECTOR_SIZE;
 676 +
 677 +                int64_t end_sector = sector_num +
 678 +                    VMA_BLOCK_SIZE/BDRV_SECTOR_SIZE;
 679 +                if (end_sector > max_sector) {
 680 +                    end_sector = max_sector;
 681 +                }
 682 +
 683 +                if (mask & bit) {
 684 +                    if ((start + VMA_BLOCK_SIZE) > extent_size) {
 685 +                        error_setg(errp, "short vma extent - too many blocks");
 686 +                        return -1;
 687 +                    }
 688 +
 689 +                    if (end_sector <= sector_num) {
 690 +                        error_setg(errp, "got wrong block address - "
 691 +                                   "write bejond end");
 692 +                        return -1;
 693 +                    }
 694 +
 695 +                    int nb_sectors = end_sector - sector_num;
 696 +                    if (restore_write_data(vmar, dev_id, bs, vmstate_fd,
 697 +                                           buf + start, sector_num,
 698 +                                           nb_sectors, errp) < 0) {
 699 +                        return -1;
 700 +                    }
 701 +
 702 +                    start += VMA_BLOCK_SIZE;
 703 +
 704 +                } else {
 705 +
 706 +                    if (rstate->write_zeroes && (end_sector > sector_num)) {
 707 +                        /* Todo: use bdrv_co_write_zeroes (but that need to
 708 +                         * be run inside coroutine?)
 709 +                         */
 710 +                        int nb_sectors = end_sector - sector_num;
 711 +                        if (restore_write_data(vmar, dev_id, bs, vmstate_fd,
 712 +                                              zero_vma_block, sector_num,
 713 +                                               nb_sectors, errp) < 0) {
 714 +                            return -1;
 715 +                        }
 716 +                    }
 717 +                }
 718 +
 719 +                bit = bit << 1;
 720 +            }
 721 +        }
 722 +    }
 723 +
 724 +    if (start != extent_size) {
 725 +        error_setg(errp, "vma extent error - missing blocks");
 726 +        return -1;
 727 +    }
 728 +
 729 +    return 0;
 730 +}
 731 +
 732 +int vma_reader_restore(VmaReader *vmar, int vmstate_fd, bool verbose,
 733 +                       Error **errp)
 734 +{
 735 +    assert(vmar);
 736 +    assert(vmar->head_data);
 737 +
 738 +    int ret = 0;
 739 +    unsigned char buf[VMA_MAX_EXTENT_SIZE];
 740 +    int buf_pos = 0;
 741 +    unsigned char md5sum[16];
 742 +    VmaHeader *h = (VmaHeader *)vmar->head_data;
 743 +
 744 +    vmar->start_time = time(NULL);
 745 +
 746 +    while (1) {
 747 +        int bytes = full_read(vmar->fd, buf + buf_pos, sizeof(buf) - buf_pos);
 748 +        if (bytes < 0) {
 749 +            error_setg(errp, "read failed - %s", g_strerror(errno));
 750 +            return -1;
 751 +        }
 752 +
 753 +        buf_pos += bytes;
 754 +
 755 +        if (!buf_pos) {
 756 +            break; /* EOF */
 757 +        }
 758 +
 759 +        if (buf_pos < VMA_EXTENT_HEADER_SIZE) {
 760 +            error_setg(errp, "read short extent (%d bytes)", buf_pos);
 761 +            return -1;
 762 +        }
 763 +
 764 +        VmaExtentHeader *ehead = (VmaExtentHeader *)buf;
 765 +
 766 +        /* extract md5sum */
 767 +        memcpy(md5sum, ehead->md5sum, sizeof(ehead->md5sum));
 768 +        memset(ehead->md5sum, 0, sizeof(ehead->md5sum));
 769 +
 770 +        g_checksum_reset(vmar->md5csum);
 771 +        g_checksum_update(vmar->md5csum, buf, VMA_EXTENT_HEADER_SIZE);
 772 +        gsize csize = 16;
 773 +        g_checksum_get_digest(vmar->md5csum, ehead->md5sum, &csize);
 774 +
 775 +        if (memcmp(md5sum, ehead->md5sum, 16) != 0) {
 776 +            error_setg(errp, "wrong vma extent header chechsum");
 777 +            return -1;
 778 +        }
 779 +
 780 +        if (memcmp(h->uuid, ehead->uuid, sizeof(ehead->uuid)) != 0) {
 781 +            error_setg(errp, "wrong vma extent uuid");
 782 +            return -1;
 783 +        }
 784 +
 785 +        if (ehead->magic != VMA_EXTENT_MAGIC || ehead->reserved1 != 0) {
 786 +            error_setg(errp, "wrong vma extent header magic");
 787 +            return -1;
 788 +        }
 789 +
 790 +        int block_count = GUINT16_FROM_BE(ehead->block_count);
 791 +        int extent_size = VMA_EXTENT_HEADER_SIZE + block_count*VMA_BLOCK_SIZE;
 792 +
 793 +        if (buf_pos < extent_size) {
 794 +            error_setg(errp, "short vma extent (%d < %d)", buf_pos,
 795 +                       extent_size);
 796 +            return -1;
 797 +        }
 798 +
 799 +        if (restore_extent(vmar, buf, extent_size, vmstate_fd, verbose,
 800 +                           errp) < 0) {
 801 +            return -1;
 802 +        }
 803 +
 804 +        if (buf_pos > extent_size) {
 805 +            memmove(buf, buf + extent_size, buf_pos - extent_size);
 806 +            buf_pos = buf_pos - extent_size;
 807 +        } else {
 808 +            buf_pos = 0;
 809 +        }
 810 +    }
 811 +
 812 +    bdrv_drain_all();
 813 +
 814 +    int i;
 815 +    for (i = 1; i < 256; i++) {
 816 +        VmaRestoreState *rstate = &vmar->rstate[i];
 817 +        if (!rstate->bs) {
 818 +            continue;
 819 +        }
 820 +
 821 +        if (bdrv_flush(rstate->bs) < 0) {
 822 +            error_setg(errp, "vma bdrv_flush %s failed",
 823 +                       vmar->devinfo[i].devname);
 824 +            return -1;
 825 +        }
 826 +
 827 +        if (vmar->devinfo[i].size &&
 828 +            (strcmp(vmar->devinfo[i].devname, "vmstate") != 0)) {
 829 +            assert(rstate->bitmap);
 830 +
 831 +            int64_t cluster_num, end;
 832 +
 833 +            end = (vmar->devinfo[i].size + VMA_CLUSTER_SIZE - 1) /
 834 +                VMA_CLUSTER_SIZE;
 835 +
 836 +            for (cluster_num = 0; cluster_num < end; cluster_num++) {
 837 +                if (!vma_reader_get_bitmap(rstate, cluster_num)) {
 838 +                    error_setg(errp, "detected missing cluster %zd "
 839 +                               "for stream %s", cluster_num,
 840 +                               vmar->devinfo[i].devname);
 841 +                    return -1;
 842 +                }
 843 +            }
 844 +        }
 845 +    }
 846 +
 847 +    return ret;
 848 +}
 849 +
 850 Index: new/vma-writer.c
 851 ===================================================================
 852 --- /dev/null   1970-01-01 00:00:00.000000000 +0000
 853 +++ new/vma-writer.c    2014-06-17 06:15:02.000000000 +0200
 854 @@ -0,0 +1,875 @@
 855 +/*
 856 + * VMA: Virtual Machine Archive
 857 + *
 858 + * Copyright (C) 2012 Proxmox Server Solutions
 859 + *
 860 + * Authors:
 861 + *  Dietmar Maurer (dietmar@proxmox.com)
 862 + *
 863 + * This work is licensed under the terms of the GNU GPL, version 2 or later.
 864 + * See the COPYING file in the top-level directory.
 865 + *
 866 + */
 867 +
 868 +#include <stdio.h>
 869 +#include <errno.h>
 870 +#include <unistd.h>
 871 +#include <stdio.h>
 872 +#include <string.h>
 873 +#include <sys/types.h>
 874 +#include <sys/stat.h>
 875 +#include <fcntl.h>
 876 +#include <glib.h>
 877 +#include <uuid/uuid.h>
 878 +
 879 +#include "qemu-common.h"
 880 +#include "vma.h"
 881 +#include "block/block.h"
 882 +#include "monitor/monitor.h"
 883 +
 884 +#define DEBUG_VMA 0
 885 +
 886 +#define DPRINTF(fmt, ...)\
 887 +    do { if (DEBUG_VMA) { printf("vma: " fmt, ## __VA_ARGS__); } } while (0)
 888 +
 889 +#define WRITE_BUFFERS 5
 890 +
 891 +typedef struct VmaAIOCB VmaAIOCB;
 892 +struct VmaAIOCB {
 893 +    unsigned char buffer[VMA_MAX_EXTENT_SIZE];
 894 +    VmaWriter *vmaw;
 895 +    size_t bytes;
 896 +    Coroutine *co;
 897 +};
 898 +
 899 +struct VmaWriter {
 900 +    int fd;
 901 +    FILE *cmd;
 902 +    int status;
 903 +    char errmsg[8192];
 904 +    uuid_t uuid;
 905 +    bool header_written;
 906 +    bool closed;
 907 +
 908 +    /* we always write extents */
 909 +    unsigned char outbuf[VMA_MAX_EXTENT_SIZE];
 910 +    int outbuf_pos; /* in bytes */
 911 +    int outbuf_count; /* in VMA_BLOCKS */
 912 +    uint64_t outbuf_block_info[VMA_BLOCKS_PER_EXTENT];
 913 +
 914 +    VmaAIOCB *aiocbs[WRITE_BUFFERS];
 915 +    CoQueue wqueue;
 916 +
 917 +    GChecksum *md5csum;
 918 +    CoMutex writer_lock;
 919 +    CoMutex flush_lock;
 920 +    Coroutine *co_writer;
 921 +
 922 +    /* drive informations */
 923 +    VmaStreamInfo stream_info[256];
 924 +    guint stream_count;
 925 +
 926 +    guint8 vmstate_stream;
 927 +    uint32_t vmstate_clusters;
 928 +
 929 +    /* header blob table */
 930 +    char *header_blob_table;
 931 +    uint32_t header_blob_table_size;
 932 +    uint32_t header_blob_table_pos;
 933 +
 934 +    /* store for config blobs */
 935 +    uint32_t config_names[VMA_MAX_CONFIGS]; /* offset into blob_buffer table */
 936 +    uint32_t config_data[VMA_MAX_CONFIGS];  /* offset into blob_buffer table */
 937 +    uint32_t config_count;
 938 +};
 939 +
 940 +void vma_writer_set_error(VmaWriter *vmaw, const char *fmt, ...)
 941 +{
 942 +    va_list ap;
 943 +
 944 +    if (vmaw->status < 0) {
 945 +        return;
 946 +    }
 947 +
 948 +    vmaw->status = -1;
 949 +
 950 +    va_start(ap, fmt);
 951 +    g_vsnprintf(vmaw->errmsg, sizeof(vmaw->errmsg), fmt, ap);
 952 +    va_end(ap);
 953 +
 954 +    DPRINTF("vma_writer_set_error: %s\n", vmaw->errmsg);
 955 +}
 956 +
 957 +static uint32_t allocate_header_blob(VmaWriter *vmaw, const char *data,
 958 +                                     size_t len)
 959 +{
 960 +    if (len > 65535) {
 961 +        return 0;
 962 +    }
 963 +
 964 +    if (!vmaw->header_blob_table ||
 965 +        (vmaw->header_blob_table_size <
 966 +         (vmaw->header_blob_table_pos + len + 2))) {
 967 +        int newsize = vmaw->header_blob_table_size + ((len + 2 + 511)/512)*512;
 968 +
 969 +        vmaw->header_blob_table = g_realloc(vmaw->header_blob_table, newsize);
 970 +        memset(vmaw->header_blob_table + vmaw->header_blob_table_size,
 971 +               0, newsize - vmaw->header_blob_table_size);
 972 +        vmaw->header_blob_table_size = newsize;
 973 +    }
 974 +
 975 +    uint32_t cpos = vmaw->header_blob_table_pos;
 976 +    vmaw->header_blob_table[cpos] = len & 255;
 977 +    vmaw->header_blob_table[cpos+1] = (len >> 8) & 255;
 978 +    memcpy(vmaw->header_blob_table + cpos + 2, data, len);
 979 +    vmaw->header_blob_table_pos += len + 2;
 980 +    return cpos;
 981 +}
 982 +
 983 +static uint32_t allocate_header_string(VmaWriter *vmaw, const char *str)
 984 +{
 985 +    assert(vmaw);
 986 +
 987 +    size_t len = strlen(str) + 1;
 988 +
 989 +    return allocate_header_blob(vmaw, str, len);
 990 +}
 991 +
 992 +int vma_writer_add_config(VmaWriter *vmaw, const char *name, gpointer data,
 993 +                          gsize len)
 994 +{
 995 +    assert(vmaw);
 996 +    assert(!vmaw->header_written);
 997 +    assert(vmaw->config_count < VMA_MAX_CONFIGS);
 998 +    assert(name);
 999 +    assert(data);
1000 +    assert(len);
1001 +
1002 +    gchar *basename = g_path_get_basename(name);
1003 +    uint32_t name_ptr = allocate_header_string(vmaw, basename);
1004 +    g_free(basename);
1005 +
1006 +    if (!name_ptr) {
1007 +        return -1;
1008 +    }
1009 +
1010 +    uint32_t data_ptr = allocate_header_blob(vmaw, data, len);
1011 +    if (!data_ptr) {
1012 +        return -1;
1013 +    }
1014 +
1015 +    vmaw->config_names[vmaw->config_count] = name_ptr;
1016 +    vmaw->config_data[vmaw->config_count] = data_ptr;
1017 +
1018 +    vmaw->config_count++;
1019 +
1020 +    return 0;
1021 +}
1022 +
1023 +int vma_writer_register_stream(VmaWriter *vmaw, const char *devname,
1024 +                               size_t size)
1025 +{
1026 +    assert(vmaw);
1027 +    assert(devname);
1028 +    assert(!vmaw->status);
1029 +
1030 +    if (vmaw->header_written) {
1031 +        vma_writer_set_error(vmaw, "vma_writer_register_stream: header "
1032 +                             "already written");
1033 +        return -1;
1034 +    }
1035 +
1036 +    guint n = vmaw->stream_count + 1;
1037 +
1038 +    /* we can have dev_ids form 1 to 255 (0 reserved)
1039 +     * 255(-1) reseverd for safety
1040 +     */
1041 +    if (n > 254) {
1042 +        vma_writer_set_error(vmaw, "vma_writer_register_stream: "
1043 +                             "too many drives");
1044 +        return -1;
1045 +    }
1046 +
1047 +    if (size <= 0) {
1048 +        vma_writer_set_error(vmaw, "vma_writer_register_stream: "
1049 +                             "got strange size %zd", size);
1050 +        return -1;
1051 +    }
1052 +
1053 +    DPRINTF("vma_writer_register_stream %s %zu %d\n", devname, size, n);
1054 +
1055 +    vmaw->stream_info[n].devname = g_strdup(devname);
1056 +    vmaw->stream_info[n].size = size;
1057 +
1058 +    vmaw->stream_info[n].cluster_count = (size + VMA_CLUSTER_SIZE - 1) /
1059 +        VMA_CLUSTER_SIZE;
1060 +
1061 +    vmaw->stream_count = n;
1062 +
1063 +    if (strcmp(devname, "vmstate") == 0) {
1064 +        vmaw->vmstate_stream = n;
1065 +    }
1066 +
1067 +    return n;
1068 +}
1069 +
1070 +static void vma_co_continue_write(void *opaque)
1071 +{
1072 +    VmaWriter *vmaw = opaque;
1073 +
1074 +    DPRINTF("vma_co_continue_write\n");
1075 +    qemu_coroutine_enter(vmaw->co_writer, NULL);
1076 +}
1077 +
1078 +static ssize_t coroutine_fn
1079 +vma_co_write(VmaWriter *vmaw, const void *buf, size_t bytes)
1080 +{
1081 +    size_t done = 0;
1082 +    ssize_t ret;
1083 +
1084 +    /* atomic writes (we cannot interleave writes) */
1085 +    qemu_co_mutex_lock(&vmaw->writer_lock);
1086 +
1087 +    DPRINTF("vma_co_write enter %zd\n", bytes);
1088 +
1089 +    assert(vmaw->co_writer == NULL);
1090 +
1091 +    vmaw->co_writer = qemu_coroutine_self();
1092 +
1093 +    qemu_aio_set_fd_handler(vmaw->fd, NULL, vma_co_continue_write, vmaw);
1094 +
1095 +    DPRINTF("vma_co_write wait until writable\n");
1096 +    qemu_coroutine_yield();
1097 +    DPRINTF("vma_co_write starting %zd\n", bytes);
1098 +
1099 +    while (done < bytes) {
1100 +        ret = write(vmaw->fd, buf + done, bytes - done);
1101 +        if (ret > 0) {
1102 +            done += ret;
1103 +            DPRINTF("vma_co_write written %zd %zd\n", done, ret);
1104 +        } else if (ret < 0) {
1105 +            if (errno == EAGAIN || errno == EWOULDBLOCK) {
1106 +                DPRINTF("vma_co_write yield %zd\n", done);
1107 +                qemu_coroutine_yield();
1108 +                DPRINTF("vma_co_write restart %zd\n", done);
1109 +            } else {
1110 +                vma_writer_set_error(vmaw, "vma_co_write write error - %s",
1111 +                                     g_strerror(errno));
1112 +                done = -1; /* always return failure for partial writes */
1113 +                break;
1114 +            }
1115 +        } else if (ret == 0) {
1116 +            /* should not happen - simply try again */
1117 +        }
1118 +    }
1119 +
1120 +    qemu_aio_set_fd_handler(vmaw->fd, NULL, NULL, NULL);
1121 +
1122 +    vmaw->co_writer = NULL;
1123 +
1124 +    qemu_co_mutex_unlock(&vmaw->writer_lock);
1125 +
1126 +    DPRINTF("vma_co_write leave %zd\n", done);
1127 +    return done;
1128 +}
1129 +
1130 +static void coroutine_fn vma_co_writer_task(void *opaque)
1131 +{
1132 +    VmaAIOCB *cb = opaque;
1133 +
1134 +    DPRINTF("vma_co_writer_task start\n");
1135 +
1136 +    int64_t done = vma_co_write(cb->vmaw, cb->buffer, cb->bytes);
1137 +    DPRINTF("vma_co_writer_task write done %zd\n", done);
1138 +
1139 +    if (done != cb->bytes) {
1140 +        DPRINTF("vma_co_writer_task failed write %zd %zd", cb->bytes, done);
1141 +        vma_writer_set_error(cb->vmaw, "vma_co_writer_task failed write %zd",
1142 +                             done);
1143 +    }
1144 +
1145 +    cb->bytes = 0;
1146 +
1147 +    qemu_co_queue_next(&cb->vmaw->wqueue);
1148 +
1149 +    DPRINTF("vma_co_writer_task end\n");
1150 +}
1151 +
1152 +static void coroutine_fn vma_queue_flush(VmaWriter *vmaw)
1153 +{
1154 +    DPRINTF("vma_queue_flush enter\n");
1155 +
1156 +    assert(vmaw);
1157 +
1158 +    while (1) {
1159 +        int i;
1160 +        VmaAIOCB *cb = NULL;
1161 +        for (i = 0; i < WRITE_BUFFERS; i++) {
1162 +            if (vmaw->aiocbs[i]->bytes) {
1163 +                cb = vmaw->aiocbs[i];
1164 +                DPRINTF("FOUND USED AIO BUFFER %d %zd\n", i,
1165 +                        vmaw->aiocbs[i]->bytes);
1166 +                break;
1167 +            }
1168 +        }
1169 +        if (!cb) {
1170 +            break;
1171 +        }
1172 +        qemu_co_queue_wait(&vmaw->wqueue);
1173 +    }
1174 +
1175 +    DPRINTF("vma_queue_flush leave\n");
1176 +}
1177 +
1178 +/**
1179 + * NOTE: pipe buffer size in only 4096 bytes on linux (see 'ulimit -a')
1180 + * So we need to create a coroutione to allow 'parallel' execution.
1181 + */
1182 +static ssize_t coroutine_fn
1183 +vma_queue_write(VmaWriter *vmaw, const void *buf, size_t bytes)
1184 +{
1185 +    DPRINTF("vma_queue_write enter %zd\n", bytes);
1186 +
1187 +    assert(vmaw);
1188 +    assert(buf);
1189 +    assert(bytes <= VMA_MAX_EXTENT_SIZE);
1190 +
1191 +    VmaAIOCB *cb = NULL;
1192 +    while (!cb) {
1193 +        int i;
1194 +        for (i = 0; i < WRITE_BUFFERS; i++) {
1195 +            if (!vmaw->aiocbs[i]->bytes) {
1196 +                cb = vmaw->aiocbs[i];
1197 +                break;
1198 +            }
1199 +        }
1200 +        if (!cb) {
1201 +            qemu_co_queue_wait(&vmaw->wqueue);
1202 +        }
1203 +    }
1204 +
1205 +    memcpy(cb->buffer, buf, bytes);
1206 +    cb->bytes = bytes;
1207 +    cb->vmaw = vmaw;
1208 +
1209 +    DPRINTF("vma_queue_write start %zd\n", bytes);
1210 +    cb->co = qemu_coroutine_create(vma_co_writer_task);
1211 +    qemu_coroutine_enter(cb->co, cb);
1212 +
1213 +    DPRINTF("vma_queue_write leave\n");
1214 +
1215 +    return bytes;
1216 +}
1217 +
1218 +VmaWriter *vma_writer_create(const char *filename, uuid_t uuid, Error **errp)
1219 +{
1220 +    const char *p;
1221 +
1222 +    assert(sizeof(VmaHeader) == (4096 + 8192));
1223 +    assert(G_STRUCT_OFFSET(VmaHeader, config_names) == 2044);
1224 +    assert(G_STRUCT_OFFSET(VmaHeader, config_data) == 3068);
1225 +    assert(G_STRUCT_OFFSET(VmaHeader, dev_info) == 4096);
1226 +    assert(sizeof(VmaExtentHeader) == 512);
1227 +
1228 +    VmaWriter *vmaw = g_new0(VmaWriter, 1);
1229 +    vmaw->fd = -1;
1230 +
1231 +    vmaw->md5csum = g_checksum_new(G_CHECKSUM_MD5);
1232 +    if (!vmaw->md5csum) {
1233 +        error_setg(errp, "can't allocate cmsum\n");
1234 +        goto err;
1235 +    }
1236 +
1237 +    if (strstart(filename, "exec:", &p)) {
1238 +        vmaw->cmd = popen(p, "w");
1239 +        if (vmaw->cmd == NULL) {
1240 +            error_setg(errp, "can't popen command '%s' - %s\n", p,
1241 +                       g_strerror(errno));
1242 +            goto err;
1243 +        }
1244 +        vmaw->fd = fileno(vmaw->cmd);
1245 +
1246 +        /* try to use O_NONBLOCK and O_DIRECT */
1247 +        fcntl(vmaw->fd, F_SETFL, fcntl(vmaw->fd, F_GETFL)|O_NONBLOCK);
1248 +        fcntl(vmaw->fd, F_SETFL, fcntl(vmaw->fd, F_GETFL)|O_DIRECT);
1249 +
1250 +    } else {
1251 +        struct stat st;
1252 +        int oflags;
1253 +        const char *tmp_id_str;
1254 +
1255 +        if ((stat(filename, &st) == 0) && S_ISFIFO(st.st_mode)) {
1256 +            oflags = O_NONBLOCK|O_DIRECT|O_WRONLY;
1257 +            vmaw->fd = qemu_open(filename, oflags, 0644);
1258 +        } else if (strstart(filename, "/dev/fdset/", &tmp_id_str)) {
1259 +            oflags = O_NONBLOCK|O_DIRECT|O_WRONLY;
1260 +            vmaw->fd = qemu_open(filename, oflags, 0644);
1261 +        } else if (strstart(filename, "/dev/fdname/", &tmp_id_str)) {
1262 +            vmaw->fd = monitor_get_fd(cur_mon, tmp_id_str, errp);
1263 +            if (vmaw->fd < 0) {
1264 +                goto err;
1265 +            }
1266 +            /* try to use O_NONBLOCK and O_DIRECT */
1267 +            fcntl(vmaw->fd, F_SETFL, fcntl(vmaw->fd, F_GETFL)|O_NONBLOCK);
1268 +            fcntl(vmaw->fd, F_SETFL, fcntl(vmaw->fd, F_GETFL)|O_DIRECT);
1269 +        } else  {
1270 +            oflags = O_NONBLOCK|O_DIRECT|O_WRONLY|O_CREAT|O_EXCL;
1271 +            vmaw->fd = qemu_open(filename, oflags, 0644);
1272 +        }
1273 +
1274 +        if (vmaw->fd < 0) {
1275 +            error_setg(errp, "can't open file %s - %s\n", filename,
1276 +                       g_strerror(errno));
1277 +            goto err;
1278 +        }
1279 +    }
1280 +
1281 +    /* we use O_DIRECT, so we need to align IO buffers */
1282 +    int i;
1283 +    for (i = 0; i < WRITE_BUFFERS; i++) {
1284 +        vmaw->aiocbs[i] = qemu_memalign(512, sizeof(VmaAIOCB));
1285 +        memset(vmaw->aiocbs[i], 0, sizeof(VmaAIOCB));
1286 +    }
1287 +
1288 +    vmaw->outbuf_count = 0;
1289 +    vmaw->outbuf_pos = VMA_EXTENT_HEADER_SIZE;
1290 +
1291 +    vmaw->header_blob_table_pos = 1; /* start at pos 1 */
1292 +
1293 +    qemu_co_mutex_init(&vmaw->writer_lock);
1294 +    qemu_co_mutex_init(&vmaw->flush_lock);
1295 +    qemu_co_queue_init(&vmaw->wqueue);
1296 +
1297 +    uuid_copy(vmaw->uuid, uuid);
1298 +
1299 +    return vmaw;
1300 +
1301 +err:
1302 +    if (vmaw) {
1303 +        if (vmaw->cmd) {
1304 +            pclose(vmaw->cmd);
1305 +        } else if (vmaw->fd >= 0) {
1306 +            close(vmaw->fd);
1307 +        }
1308 +
1309 +        if (vmaw->md5csum) {
1310 +            g_checksum_free(vmaw->md5csum);
1311 +        }
1312 +
1313 +        g_free(vmaw);
1314 +    }
1315 +
1316 +    return NULL;
1317 +}
1318 +
1319 +static int coroutine_fn vma_write_header(VmaWriter *vmaw)
1320 +{
1321 +    assert(vmaw);
1322 +    int header_clusters = 8;
1323 +    char buf[65536*header_clusters];
1324 +    VmaHeader *head = (VmaHeader *)buf;
1325 +
1326 +    int i;
1327 +
1328 +    DPRINTF("VMA WRITE HEADER\n");
1329 +
1330 +    if (vmaw->status < 0) {
1331 +        return vmaw->status;
1332 +    }
1333 +
1334 +    memset(buf, 0, sizeof(buf));
1335 +
1336 +    head->magic = VMA_MAGIC;
1337 +    head->version = GUINT32_TO_BE(1); /* v1 */
1338 +    memcpy(head->uuid, vmaw->uuid, 16);
1339 +
1340 +    time_t ctime = time(NULL);
1341 +    head->ctime = GUINT64_TO_BE(ctime);
1342 +
1343 +    if (!vmaw->stream_count) {
1344 +        return -1;
1345 +    }
1346 +
1347 +    for (i = 0; i < VMA_MAX_CONFIGS; i++) {
1348 +        head->config_names[i] = GUINT32_TO_BE(vmaw->config_names[i]);
1349 +        head->config_data[i] = GUINT32_TO_BE(vmaw->config_data[i]);
1350 +    }
1351 +
1352 +    /* 32 bytes per device (12 used currently) = 8192 bytes max */
1353 +    for (i = 1; i <= 254; i++) {
1354 +        VmaStreamInfo *si = &vmaw->stream_info[i];
1355 +        if (si->size) {
1356 +            assert(si->devname);
1357 +            uint32_t devname_ptr = allocate_header_string(vmaw, si->devname);
1358 +            if (!devname_ptr) {
1359 +                return -1;
1360 +            }
1361 +            head->dev_info[i].devname_ptr = GUINT32_TO_BE(devname_ptr);
1362 +            head->dev_info[i].size = GUINT64_TO_BE(si->size);
1363 +        }
1364 +    }
1365 +
1366 +    uint32_t header_size = sizeof(VmaHeader) + vmaw->header_blob_table_size;
1367 +    head->header_size = GUINT32_TO_BE(header_size);
1368 +
1369 +    if (header_size > sizeof(buf)) {
1370 +        return -1; /* just to be sure */
1371 +    }
1372 +
1373 +    uint32_t blob_buffer_offset = sizeof(VmaHeader);
1374 +    memcpy(buf + blob_buffer_offset, vmaw->header_blob_table,
1375 +           vmaw->header_blob_table_size);
1376 +    head->blob_buffer_offset = GUINT32_TO_BE(blob_buffer_offset);
1377 +    head->blob_buffer_size = GUINT32_TO_BE(vmaw->header_blob_table_pos);
1378 +
1379 +    g_checksum_reset(vmaw->md5csum);
1380 +    g_checksum_update(vmaw->md5csum, (const guchar *)buf, header_size);
1381 +    gsize csize = 16;
1382 +    g_checksum_get_digest(vmaw->md5csum, (guint8 *)(head->md5sum), &csize);
1383 +
1384 +    return vma_queue_write(vmaw, buf, header_size);
1385 +}
1386 +
1387 +static int coroutine_fn vma_writer_flush(VmaWriter *vmaw)
1388 +{
1389 +    assert(vmaw);
1390 +
1391 +    int ret;
1392 +    int i;
1393 +
1394 +    if (vmaw->status < 0) {
1395 +        return vmaw->status;
1396 +    }
1397 +
1398 +    if (!vmaw->header_written) {
1399 +        vmaw->header_written = true;
1400 +        ret = vma_write_header(vmaw);
1401 +        if (ret < 0) {
1402 +            vma_writer_set_error(vmaw, "vma_writer_flush: write header failed");
1403 +            return ret;
1404 +        }
1405 +    }
1406 +
1407 +    DPRINTF("VMA WRITE FLUSH %d %d\n", vmaw->outbuf_count, vmaw->outbuf_pos);
1408 +
1409 +
1410 +    VmaExtentHeader *ehead = (VmaExtentHeader *)vmaw->outbuf;
1411 +
1412 +    ehead->magic = VMA_EXTENT_MAGIC;
1413 +    ehead->reserved1 = 0;
1414 +
1415 +    for (i = 0; i < VMA_BLOCKS_PER_EXTENT; i++) {
1416 +        ehead->blockinfo[i] = GUINT64_TO_BE(vmaw->outbuf_block_info[i]);
1417 +    }
1418 +
1419 +    guint16 block_count = (vmaw->outbuf_pos - VMA_EXTENT_HEADER_SIZE) /
1420 +        VMA_BLOCK_SIZE;
1421 +
1422 +    ehead->block_count = GUINT16_TO_BE(block_count);
1423 +
1424 +    memcpy(ehead->uuid, vmaw->uuid, sizeof(ehead->uuid));
1425 +    memset(ehead->md5sum, 0, sizeof(ehead->md5sum));
1426 +
1427 +    g_checksum_reset(vmaw->md5csum);
1428 +    g_checksum_update(vmaw->md5csum, vmaw->outbuf, VMA_EXTENT_HEADER_SIZE);
1429 +    gsize csize = 16;
1430 +    g_checksum_get_digest(vmaw->md5csum, ehead->md5sum, &csize);
1431 +
1432 +    int bytes = vmaw->outbuf_pos;
1433 +    ret = vma_queue_write(vmaw, vmaw->outbuf, bytes);
1434 +    if (ret != bytes) {
1435 +        vma_writer_set_error(vmaw, "vma_writer_flush: failed write");
1436 +    }
1437 +
1438 +    vmaw->outbuf_count = 0;
1439 +    vmaw->outbuf_pos = VMA_EXTENT_HEADER_SIZE;
1440 +
1441 +    for (i = 0; i < VMA_BLOCKS_PER_EXTENT; i++) {
1442 +        vmaw->outbuf_block_info[i] = 0;
1443 +    }
1444 +
1445 +    return vmaw->status;
1446 +}
1447 +
1448 +static int vma_count_open_streams(VmaWriter *vmaw)
1449 +{
1450 +    g_assert(vmaw != NULL);
1451 +
1452 +    int i;
1453 +    int open_drives = 0;
1454 +    for (i = 0; i <= 255; i++) {
1455 +        if (vmaw->stream_info[i].size && !vmaw->stream_info[i].finished) {
1456 +            open_drives++;
1457 +        }
1458 +    }
1459 +
1460 +    return open_drives;
1461 +}
1462 +
1463 +/**
1464 + * all jobs should call this when there is no more data
1465 + * Returns: number of remaining stream (0 ==> finished)
1466 + */
1467 +int coroutine_fn
1468 +vma_writer_close_stream(VmaWriter *vmaw, uint8_t dev_id)
1469 +{
1470 +    g_assert(vmaw != NULL);
1471 +
1472 +    DPRINTF("vma_writer_set_status %d\n", dev_id);
1473 +    if (!vmaw->stream_info[dev_id].size) {
1474 +        vma_writer_set_error(vmaw, "vma_writer_close_stream: "
1475 +                             "no such stream %d", dev_id);
1476 +        return -1;
1477 +    }
1478 +    if (vmaw->stream_info[dev_id].finished) {
1479 +        vma_writer_set_error(vmaw, "vma_writer_close_stream: "
1480 +                             "stream already closed %d", dev_id);
1481 +        return -1;
1482 +    }
1483 +
1484 +    vmaw->stream_info[dev_id].finished = true;
1485 +
1486 +    int open_drives = vma_count_open_streams(vmaw);
1487 +
1488 +    if (open_drives <= 0) {
1489 +        DPRINTF("vma_writer_set_status all drives completed\n");
1490 +        qemu_co_mutex_lock(&vmaw->flush_lock);
1491 +        int ret = vma_writer_flush(vmaw);
1492 +        qemu_co_mutex_unlock(&vmaw->flush_lock);
1493 +        if (ret < 0) {
1494 +            vma_writer_set_error(vmaw, "vma_writer_close_stream: flush failed");
1495 +        }
1496 +    }
1497 +
1498 +    return open_drives;
1499 +}
1500 +
1501 +int vma_writer_get_status(VmaWriter *vmaw, VmaStatus *status)
1502 +{
1503 +    int i;
1504 +
1505 +    g_assert(vmaw != NULL);
1506 +
1507 +    if (status) {
1508 +        status->status = vmaw->status;
1509 +        g_strlcpy(status->errmsg, vmaw->errmsg, sizeof(status->errmsg));
1510 +        for (i = 0; i <= 255; i++) {
1511 +            status->stream_info[i] = vmaw->stream_info[i];
1512 +        }
1513 +
1514 +        uuid_unparse_lower(vmaw->uuid, status->uuid_str);
1515 +    }
1516 +
1517 +    status->closed = vmaw->closed;
1518 +
1519 +    return vmaw->status;
1520 +}
1521 +
1522 +static int vma_writer_get_buffer(VmaWriter *vmaw)
1523 +{
1524 +    int ret = 0;
1525 +
1526 +    qemu_co_mutex_lock(&vmaw->flush_lock);
1527 +
1528 +    /* wait until buffer is available */
1529 +    while (vmaw->outbuf_count >= (VMA_BLOCKS_PER_EXTENT - 1)) {
1530 +        ret = vma_writer_flush(vmaw);
1531 +        if (ret < 0) {
1532 +            vma_writer_set_error(vmaw, "vma_writer_get_buffer: flush failed");
1533 +            break;
1534 +        }
1535 +    }
1536 +
1537 +    qemu_co_mutex_unlock(&vmaw->flush_lock);
1538 +
1539 +    return ret;
1540 +}
1541 +
1542 +
1543 +int64_t coroutine_fn
1544 +vma_writer_write(VmaWriter *vmaw, uint8_t dev_id, int64_t cluster_num,
1545 +                 unsigned char *buf, size_t *zero_bytes)
1546 +{
1547 +    g_assert(vmaw != NULL);
1548 +    g_assert(zero_bytes != NULL);
1549 +
1550 +    *zero_bytes = 0;
1551 +
1552 +    if (vmaw->status < 0) {
1553 +        return vmaw->status;
1554 +    }
1555 +
1556 +    if (!dev_id || !vmaw->stream_info[dev_id].size) {
1557 +        vma_writer_set_error(vmaw, "vma_writer_write: "
1558 +                             "no such stream %d", dev_id);
1559 +        return -1;
1560 +    }
1561 +
1562 +    if (vmaw->stream_info[dev_id].finished) {
1563 +        vma_writer_set_error(vmaw, "vma_writer_write: "
1564 +                             "stream already closed %d", dev_id);
1565 +        return -1;
1566 +    }
1567 +
1568 +
1569 +    if (cluster_num >= (((uint64_t)1)<<32)) {
1570 +        vma_writer_set_error(vmaw, "vma_writer_write: "
1571 +                             "cluster number out of range");
1572 +        return -1;
1573 +    }
1574 +
1575 +    if (dev_id == vmaw->vmstate_stream) {
1576 +        if (cluster_num != vmaw->vmstate_clusters) {
1577 +            vma_writer_set_error(vmaw, "vma_writer_write: "
1578 +                                 "non sequential vmstate write");
1579 +        }
1580 +        vmaw->vmstate_clusters++;
1581 +    } else if (cluster_num >= vmaw->stream_info[dev_id].cluster_count) {
1582 +        vma_writer_set_error(vmaw, "vma_writer_write: cluster number too big");
1583 +        return -1;
1584 +    }
1585 +
1586 +    /* wait until buffer is available */
1587 +    if (vma_writer_get_buffer(vmaw) < 0) {
1588 +        vma_writer_set_error(vmaw, "vma_writer_write: "
1589 +                             "vma_writer_get_buffer failed");
1590 +        return -1;
1591 +    }
1592 +
1593 +    DPRINTF("VMA WRITE %d %zd\n", dev_id, cluster_num);
1594 +
1595 +    uint16_t mask = 0;
1596 +
1597 +    if (buf) {
1598 +        int i;
1599 +        int bit = 1;
1600 +        for (i = 0; i < 16; i++) {
1601 +            unsigned char *vmablock = buf + (i*VMA_BLOCK_SIZE);
1602 +            if (!buffer_is_zero(vmablock, VMA_BLOCK_SIZE)) {
1603 +                mask |= bit;
1604 +                memcpy(vmaw->outbuf + vmaw->outbuf_pos, vmablock,
1605 +                       VMA_BLOCK_SIZE);
1606 +                vmaw->outbuf_pos += VMA_BLOCK_SIZE;
1607 +            } else {
1608 +                DPRINTF("VMA WRITE %zd ZERO BLOCK %d\n", cluster_num, i);
1609 +                vmaw->stream_info[dev_id].zero_bytes += VMA_BLOCK_SIZE;
1610 +                *zero_bytes += VMA_BLOCK_SIZE;
1611 +            }
1612 +
1613 +            bit = bit << 1;
1614 +        }
1615 +    } else {
1616 +        DPRINTF("VMA WRITE %zd ZERO CLUSTER\n", cluster_num);
1617 +        vmaw->stream_info[dev_id].zero_bytes += VMA_CLUSTER_SIZE;
1618 +        *zero_bytes += VMA_CLUSTER_SIZE;
1619 +    }
1620 +
1621 +    uint64_t block_info = ((uint64_t)mask) << (32+16);
1622 +    block_info |= ((uint64_t)dev_id) << 32;
1623 +    block_info |= (cluster_num & 0xffffffff);
1624 +    vmaw->outbuf_block_info[vmaw->outbuf_count] = block_info;
1625 +
1626 +    DPRINTF("VMA WRITE MASK %zd %zx\n", cluster_num, block_info);
1627 +
1628 +    vmaw->outbuf_count++;
1629 +
1630 +    /** NOTE: We allways write whole clusters, but we correctly set
1631 +     * transferred bytes. So transferred == size when when everything
1632 +     * went OK.
1633 +     */
1634 +    size_t transferred = VMA_CLUSTER_SIZE;
1635 +
1636 +    if (dev_id != vmaw->vmstate_stream) {
1637 +        uint64_t last = (cluster_num + 1) * VMA_CLUSTER_SIZE;
1638 +        if (last > vmaw->stream_info[dev_id].size) {
1639 +            uint64_t diff = last - vmaw->stream_info[dev_id].size;
1640 +            if (diff >= VMA_CLUSTER_SIZE) {
1641 +                vma_writer_set_error(vmaw, "vma_writer_write: "
1642 +                                     "read after last cluster");
1643 +                return -1;
1644 +            }
1645 +            transferred -= diff;
1646 +        }
1647 +    }
1648 +
1649 +    vmaw->stream_info[dev_id].transferred += transferred;
1650 +
1651 +    return transferred;
1652 +}
1653 +
1654 +int vma_writer_close(VmaWriter *vmaw, Error **errp)
1655 +{
1656 +    g_assert(vmaw != NULL);
1657 +
1658 +    int i;
1659 +
1660 +    vma_queue_flush(vmaw);
1661 +
1662 +    /* this should not happen - just to be sure */
1663 +    while (!qemu_co_queue_empty(&vmaw->wqueue)) {
1664 +        DPRINTF("vma_writer_close wait\n");
1665 +        co_sleep_ns(QEMU_CLOCK_REALTIME, 1000000);
1666 +    }
1667 +
1668 +    if (vmaw->cmd) {
1669 +        if (pclose(vmaw->cmd) < 0) {
1670 +            vma_writer_set_error(vmaw, "vma_writer_close: "
1671 +                                 "pclose failed - %s", g_strerror(errno));
1672 +        }
1673 +    } else {
1674 +        if (close(vmaw->fd) < 0) {
1675 +            vma_writer_set_error(vmaw, "vma_writer_close: "
1676 +                                 "close failed - %s", g_strerror(errno));
1677 +        }
1678 +    }
1679 +
1680 +    for (i = 0; i <= 255; i++) {
1681 +        VmaStreamInfo *si = &vmaw->stream_info[i];
1682 +        if (si->size) {
1683 +            if (!si->finished) {
1684 +                vma_writer_set_error(vmaw, "vma_writer_close: "
1685 +                                     "detected open stream '%s'", si->devname);
1686 +            } else if ((si->transferred != si->size) &&
1687 +                       (i != vmaw->vmstate_stream)) {
1688 +                vma_writer_set_error(vmaw, "vma_writer_close: "
1689 +                                     "incomplete stream '%s' (%zd != %zd)",
1690 +                                     si->devname, si->transferred, si->size);
1691 +            }
1692 +        }
1693 +    }
1694 +
1695 +    for (i = 0; i <= 255; i++) {
1696 +        vmaw->stream_info[i].finished = 1; /* mark as closed */
1697 +    }
1698 +
1699 +    vmaw->closed = 1;
1700 +
1701 +    if (vmaw->status < 0 && *errp == NULL) {
1702 +        error_setg(errp, "%s", vmaw->errmsg);
1703 +    }
1704 +
1705 +    return vmaw->status;
1706 +}
1707 +
1708 +void vma_writer_destroy(VmaWriter *vmaw)
1709 +{
1710 +    assert(vmaw);
1711 +
1712 +    int i;
1713 +
1714 +    for (i = 0; i <= 255; i++) {
1715 +        if (vmaw->stream_info[i].devname) {
1716 +            g_free(vmaw->stream_info[i].devname);
1717 +        }
1718 +    }
1719 +
1720 +    if (vmaw->md5csum) {
1721 +        g_checksum_free(vmaw->md5csum);
1722 +    }
1723 +
1724 +    for (i = 0; i < WRITE_BUFFERS; i++) {
1725 +        free(vmaw->aiocbs[i]);
1726 +    }
1727 +
1728 +    g_free(vmaw);
1729 +}
1730 Index: new/vma.c
1731 ===================================================================
1732 --- /dev/null   1970-01-01 00:00:00.000000000 +0000
1733 +++ new/vma.c   2014-06-17 06:15:02.000000000 +0200
1734 @@ -0,0 +1,583 @@
1735 +/*
1736 + * VMA: Virtual Machine Archive
1737 + *
1738 + * Copyright (C) 2012-2013 Proxmox Server Solutions
1739 + *
1740 + * Authors:
1741 + *  Dietmar Maurer (dietmar@proxmox.com)
1742 + *
1743 + * This work is licensed under the terms of the GNU GPL, version 2 or later.
1744 + * See the COPYING file in the top-level directory.
1745 + *
1746 + */
1747 +
1748 +#include <stdio.h>
1749 +#include <errno.h>
1750 +#include <unistd.h>
1751 +#include <stdio.h>
1752 +#include <string.h>
1753 +#include <sys/types.h>
1754 +#include <sys/stat.h>
1755 +#include <fcntl.h>
1756 +#include <glib.h>
1757 +
1758 +#include "vma.h"
1759 +#include "qemu-common.h"
1760 +#include "qemu/error-report.h"
1761 +#include "qemu/main-loop.h"
1762 +
1763 +static void help(void)
1764 +{
1765 +    const char *help_msg =
1766 +        "usage: vma command [command options]\n"
1767 +        "\n"
1768 +        "vma list <filename>\n"
1769 +        "vma create <filename> [-c config] <archive> pathname ...\n"
1770 +        "vma extract <filename> [-r <fifo>] <targetdir>\n"
1771 +        ;
1772 +
1773 +    printf("%s", help_msg);
1774 +    exit(1);
1775 +}
1776 +
1777 +static const char *extract_devname(const char *path, char **devname, int index)
1778 +{
1779 +    assert(path);
1780 +
1781 +    const char *sep = strchr(path, '=');
1782 +
1783 +    if (sep) {
1784 +        *devname = g_strndup(path, sep - path);
1785 +        path = sep + 1;
1786 +    } else {
1787 +        if (index >= 0) {
1788 +            *devname = g_strdup_printf("disk%d", index);
1789 +        } else {
1790 +            *devname = NULL;
1791 +        }
1792 +    }
1793 +
1794 +    return path;
1795 +}
1796 +
1797 +static void print_content(VmaReader *vmar)
1798 +{
1799 +    assert(vmar);
1800 +
1801 +    VmaHeader *head = vma_reader_get_header(vmar);
1802 +
1803 +    GList *l = vma_reader_get_config_data(vmar);
1804 +    while (l && l->data) {
1805 +        VmaConfigData *cdata = (VmaConfigData *)l->data;
1806 +        l = g_list_next(l);
1807 +        printf("CFG: size: %d name: %s\n", cdata->len, cdata->name);
1808 +    }
1809 +
1810 +    int i;
1811 +    VmaDeviceInfo *di;
1812 +    for (i = 1; i < 255; i++) {
1813 +        di = vma_reader_get_device_info(vmar, i);
1814 +        if (di) {
1815 +            if (strcmp(di->devname, "vmstate") == 0) {
1816 +                printf("VMSTATE: dev_id=%d memory: %zd\n", i, di->size);
1817 +            } else {
1818 +                printf("DEV: dev_id=%d size: %zd devname: %s\n",
1819 +                       i, di->size, di->devname);
1820 +            }
1821 +        }
1822 +    }
1823 +    /* ctime is the last entry we print */
1824 +    printf("CTIME: %s", ctime(&head->ctime));
1825 +    fflush(stdout);
1826 +}
1827 +
1828 +static int list_content(int argc, char **argv)
1829 +{
1830 +    int c, ret = 0;
1831 +    const char *filename;
1832 +
1833 +    for (;;) {
1834 +        c = getopt(argc, argv, "h");
1835 +        if (c == -1) {
1836 +            break;
1837 +        }
1838 +        switch (c) {
1839 +        case '?':
1840 +        case 'h':
1841 +            help();
1842 +            break;
1843 +        default:
1844 +            g_assert_not_reached();
1845 +        }
1846 +    }
1847 +
1848 +    /* Get the filename */
1849 +    if ((optind + 1) != argc) {
1850 +        help();
1851 +    }
1852 +    filename = argv[optind++];
1853 +
1854 +    Error *errp = NULL;
1855 +    VmaReader *vmar = vma_reader_create(filename, &errp);
1856 +
1857 +    if (!vmar) {
1858 +        g_error("%s", error_get_pretty(errp));
1859 +    }
1860 +
1861 +    print_content(vmar);
1862 +
1863 +    vma_reader_destroy(vmar);
1864 +
1865 +    return ret;
1866 +}
1867 +
1868 +typedef struct RestoreMap {
1869 +    char *devname;
1870 +    char *path;
1871 +    bool write_zero;
1872 +} RestoreMap;
1873 +
1874 +static int extract_content(int argc, char **argv)
1875 +{
1876 +    int c, ret = 0;
1877 +    int verbose = 0;
1878 +    const char *filename;
1879 +    const char *dirname;
1880 +    const char *readmap = NULL;
1881 +
1882 +    for (;;) {
1883 +        c = getopt(argc, argv, "hvr:");
1884 +        if (c == -1) {
1885 +            break;
1886 +        }
1887 +        switch (c) {
1888 +        case '?':
1889 +        case 'h':
1890 +            help();
1891 +            break;
1892 +        case 'r':
1893 +            readmap = optarg;
1894 +            break;
1895 +        case 'v':
1896 +            verbose = 1;
1897 +            break;
1898 +        default:
1899 +            help();
1900 +        }
1901 +    }
1902 +
1903 +    /* Get the filename */
1904 +    if ((optind + 2) != argc) {
1905 +        help();
1906 +    }
1907 +    filename = argv[optind++];
1908 +    dirname = argv[optind++];
1909 +
1910 +    Error *errp = NULL;
1911 +    VmaReader *vmar = vma_reader_create(filename, &errp);
1912 +
1913 +    if (!vmar) {
1914 +        g_error("%s", error_get_pretty(errp));
1915 +    }
1916 +
1917 +    if (mkdir(dirname, 0777) < 0) {
1918 +        g_error("unable to create target directory %s - %s",
1919 +                dirname, g_strerror(errno));
1920 +    }
1921 +
1922 +    GList *l = vma_reader_get_config_data(vmar);
1923 +    while (l && l->data) {
1924 +        VmaConfigData *cdata = (VmaConfigData *)l->data;
1925 +        l = g_list_next(l);
1926 +        char *cfgfn = g_strdup_printf("%s/%s", dirname, cdata->name);
1927 +        GError *err = NULL;
1928 +        if (!g_file_set_contents(cfgfn, (gchar *)cdata->data, cdata->len,
1929 +                                 &err)) {
1930 +            g_error("unable to write file: %s", err->message);
1931 +        }
1932 +    }
1933 +
1934 +    GHashTable *devmap = g_hash_table_new(g_str_hash, g_str_equal);
1935 +
1936 +    if (readmap) {
1937 +        print_content(vmar);
1938 +
1939 +        FILE *map = fopen(readmap, "r");
1940 +        if (!map) {
1941 +            g_error("unable to open fifo %s - %s", readmap, g_strerror(errno));
1942 +        }
1943 +
1944 +        while (1) {
1945 +            char inbuf[8192];
1946 +            char *line = fgets(inbuf, sizeof(inbuf), map);
1947 +            if (!line || line[0] == '\0' || !strcmp(line, "done\n")) {
1948 +                break;
1949 +            }
1950 +            int len = strlen(line);
1951 +            if (line[len - 1] == '\n') {
1952 +                line[len - 1] = '\0';
1953 +                if (len == 1) {
1954 +                    break;
1955 +                }
1956 +            }
1957 +
1958 +            const char *path;
1959 +            bool write_zero;
1960 +            if (line[0] == '0' && line[1] == ':') {
1961 +                path = inbuf + 2;
1962 +                write_zero = false;
1963 +            } else if (line[0] == '1' && line[1] == ':') {
1964 +                path = inbuf + 2;
1965 +                write_zero = true;
1966 +            } else {
1967 +                g_error("read map failed - parse error ('%s')", inbuf);
1968 +            }
1969 +
1970 +            char *devname = NULL;
1971 +            path = extract_devname(path, &devname, -1);
1972 +            if (!devname) {
1973 +                g_error("read map failed - no dev name specified ('%s')",
1974 +                        inbuf);
1975 +            }
1976 +
1977 +            RestoreMap *map = g_new0(RestoreMap, 1);
1978 +            map->devname = g_strdup(devname);
1979 +            map->path = g_strdup(path);
1980 +            map->write_zero = write_zero;
1981 +
1982 +            g_hash_table_insert(devmap, map->devname, map);
1983 +
1984 +        };
1985 +    }
1986 +
1987 +    int i;
1988 +    int vmstate_fd = -1;
1989 +    guint8 vmstate_stream = 0;
1990 +
1991 +    for (i = 1; i < 255; i++) {
1992 +        VmaDeviceInfo *di = vma_reader_get_device_info(vmar, i);
1993 +        if (di && (strcmp(di->devname, "vmstate") == 0)) {
1994 +            vmstate_stream = i;
1995 +            char *statefn = g_strdup_printf("%s/vmstate.bin", dirname);
1996 +            vmstate_fd = open(statefn, O_WRONLY|O_CREAT|O_EXCL, 0644);
1997 +            if (vmstate_fd < 0) {
1998 +                g_error("create vmstate file '%s' failed - %s", statefn,
1999 +                        g_strerror(errno));
2000 +            }
2001 +            g_free(statefn);
2002 +        } else if (di) {
2003 +            char *devfn = NULL;
2004 +            int flags = BDRV_O_RDWR|BDRV_O_CACHE_WB;
2005 +            bool write_zero = true;
2006 +
2007 +            if (readmap) {
2008 +                RestoreMap *map;
2009 +                map = (RestoreMap *)g_hash_table_lookup(devmap, di->devname);
2010 +                if (map == NULL) {
2011 +                    g_error("no device name mapping for %s", di->devname);
2012 +                }
2013 +                devfn = map->path;
2014 +                write_zero = map->write_zero;
2015 +            } else {
2016 +                devfn = g_strdup_printf("%s/tmp-disk-%s.raw",
2017 +                                        dirname, di->devname);
2018 +                printf("DEVINFO %s %zd\n", devfn, di->size);
2019 +
2020 +                bdrv_img_create(devfn, "raw", NULL, NULL, NULL, di->size,
2021 +                                flags, &errp, 0);
2022 +                if (errp) {
2023 +                    g_error("can't create file %s: %s", devfn,
2024 +                            error_get_pretty(errp));
2025 +                }
2026 +
2027 +                /* Note: we created an empty file above, so there is no
2028 +                 * need to write zeroes (so we generate a sparse file)
2029 +                 */
2030 +                write_zero = false;
2031 +            }
2032 +
2033 +            BlockDriverState *bs = bdrv_new(di->devname, &errp);
2034 +            if (errp || bdrv_open(&bs, devfn, NULL, NULL, flags, NULL, &errp)) {
2035 +                g_error("can't open file %s - %s", devfn,
2036 +                        error_get_pretty(errp));
2037 +            }
2038 +            if (vma_reader_register_bs(vmar, i, bs, write_zero, &errp) < 0) {
2039 +                g_error("%s", error_get_pretty(errp));
2040 +            }
2041 +
2042 +            if (!readmap) {
2043 +                g_free(devfn);
2044 +            }
2045 +        }
2046 +    }
2047 +
2048 +    if (vma_reader_restore(vmar, vmstate_fd, verbose, &errp) < 0) {
2049 +        g_error("restore failed - %s", error_get_pretty(errp));
2050 +    }
2051 +
2052 +    if (!readmap) {
2053 +        for (i = 1; i < 255; i++) {
2054 +            VmaDeviceInfo *di = vma_reader_get_device_info(vmar, i);
2055 +            if (di && (i != vmstate_stream)) {
2056 +                char *tmpfn = g_strdup_printf("%s/tmp-disk-%s.raw",
2057 +                                              dirname, di->devname);
2058 +                char *fn = g_strdup_printf("%s/disk-%s.raw",
2059 +                                           dirname, di->devname);
2060 +                if (rename(tmpfn, fn) != 0) {
2061 +                    g_error("rename %s to %s failed - %s",
2062 +                            tmpfn, fn, g_strerror(errno));
2063 +                }
2064 +            }
2065 +        }
2066 +    }
2067 +
2068 +    vma_reader_destroy(vmar);
2069 +
2070 +    bdrv_close_all();
2071 +
2072 +    return ret;
2073 +}
2074 +
2075 +typedef struct BackupJob {
2076 +    BlockDriverState *bs;
2077 +    int64_t len;
2078 +    VmaWriter *vmaw;
2079 +    uint8_t dev_id;
2080 +} BackupJob;
2081 +
2082 +#define BACKUP_SECTORS_PER_CLUSTER (VMA_CLUSTER_SIZE / BDRV_SECTOR_SIZE)
2083 +
2084 +static void coroutine_fn backup_run(void *opaque)
2085 +{
2086 +    BackupJob *job = (BackupJob *)opaque;
2087 +    struct iovec iov;
2088 +    QEMUIOVector qiov;
2089 +
2090 +    int64_t start, end;
2091 +    int ret = 0;
2092 +
2093 +    unsigned char *buf = qemu_blockalign(job->bs, VMA_CLUSTER_SIZE);
2094 +
2095 +    start = 0;
2096 +    end = DIV_ROUND_UP(job->len / BDRV_SECTOR_SIZE,
2097 +                       BACKUP_SECTORS_PER_CLUSTER);
2098 +
2099 +    for (; start < end; start++) {
2100 +        iov.iov_base = buf;
2101 +        iov.iov_len = VMA_CLUSTER_SIZE;
2102 +        qemu_iovec_init_external(&qiov, &iov, 1);
2103 +
2104 +        ret = bdrv_co_readv(job->bs, start * BACKUP_SECTORS_PER_CLUSTER,
2105 +                            BACKUP_SECTORS_PER_CLUSTER, &qiov);
2106 +        if (ret < 0) {
2107 +            vma_writer_set_error(job->vmaw, "read error", -1);
2108 +            goto out;
2109 +        }
2110 +
2111 +        size_t zb = 0;
2112 +        if (vma_writer_write(job->vmaw, job->dev_id, start, buf, &zb) < 0) {
2113 +            vma_writer_set_error(job->vmaw, "backup_dump_cb vma_writer_write failed", -1);
2114 +            goto out;
2115 +        }
2116 +    }
2117 +
2118 +
2119 +out:
2120 +    if (vma_writer_close_stream(job->vmaw, job->dev_id) <= 0) {
2121 +        Error *err = NULL;
2122 +        if (vma_writer_close(job->vmaw, &err) != 0) {
2123 +            g_warning("vma_writer_close failed %s", error_get_pretty(err));
2124 +        }
2125 +    }
2126 +}
2127 +
2128 +static int create_archive(int argc, char **argv)
2129 +{
2130 +    int i, c, res;
2131 +    int verbose = 0;
2132 +    const char *archivename;
2133 +    GList *config_files = NULL;
2134 +
2135 +    for (;;) {
2136 +        c = getopt(argc, argv, "hvc:");
2137 +        if (c == -1) {
2138 +            break;
2139 +        }
2140 +        switch (c) {
2141 +        case '?':
2142 +        case 'h':
2143 +            help();
2144 +            break;
2145 +        case 'c':
2146 +            config_files = g_list_append(config_files, optarg);
2147 +            break;
2148 +        case 'v':
2149 +            verbose = 1;
2150 +            break;
2151 +        default:
2152 +            g_assert_not_reached();
2153 +        }
2154 +    }
2155 +
2156 +
2157 +    /* make sure we have archive name and at least one path */
2158 +    if ((optind + 2) > argc) {
2159 +        help();
2160 +    }
2161 +
2162 +    archivename = argv[optind++];
2163 +
2164 +    uuid_t uuid;
2165 +    uuid_generate(uuid);
2166 +
2167 +    Error *local_err = NULL;
2168 +    VmaWriter *vmaw = vma_writer_create(archivename, uuid, &local_err);
2169 +
2170 +    if (vmaw == NULL) {
2171 +        g_error("%s", error_get_pretty(local_err));
2172 +    }
2173 +
2174 +    GList *l = config_files;
2175 +    while (l && l->data) {
2176 +        char *name = l->data;
2177 +        char *cdata = NULL;
2178 +        gsize clen = 0;
2179 +        GError *err = NULL;
2180 +        if (!g_file_get_contents(name, &cdata, &clen, &err)) {
2181 +            unlink(archivename);
2182 +            g_error("Unable to read file: %s", err->message);
2183 +        }
2184 +
2185 +        if (vma_writer_add_config(vmaw, name, cdata, clen) != 0) {
2186 +            unlink(archivename);
2187 +            g_error("Unable to append config data %s (len = %zd)",
2188 +                    name, clen);
2189 +        }
2190 +        l = g_list_next(l);
2191 +    }
2192 +
2193 +    int ind = 0;
2194 +    while (optind < argc) {
2195 +        const char *path = argv[optind++];
2196 +        char *devname = NULL;
2197 +        path = extract_devname(path, &devname, ind++);
2198 +
2199 +        BlockDriver *drv = NULL;
2200 +        Error *errp = NULL;
2201 +        BlockDriverState *bs = bdrv_new(devname, &errp);
2202 +
2203 +        if (errp) {
2204 +            g_error("bdrv_new '%s' failed - %s", devname, error_get_pretty(errp));
2205 +        }
2206 +
2207 +        res = bdrv_open(&bs, path, NULL, NULL, BDRV_O_CACHE_WB , drv, &errp);
2208 +        if (res < 0) {
2209 +            unlink(archivename);
2210 +            g_error("bdrv_open '%s' failed - %s", path, error_get_pretty(errp));
2211 +        }
2212 +        int64_t size = bdrv_getlength(bs);
2213 +        int dev_id = vma_writer_register_stream(vmaw, devname, size);
2214 +        if (dev_id <= 0) {
2215 +            unlink(archivename);
2216 +            g_error("vma_writer_register_stream '%s' failed", devname);
2217 +        }
2218 +
2219 +        BackupJob *job = g_new0(BackupJob, 1);
2220 +        job->len = size;
2221 +        job->bs = bs;
2222 +        job->vmaw = vmaw;
2223 +        job->dev_id = dev_id;
2224 +
2225 +        Coroutine *co = qemu_coroutine_create(backup_run);
2226 +        qemu_coroutine_enter(co, job);
2227 +    }
2228 +
2229 +    VmaStatus vmastat;
2230 +    int percent = 0;
2231 +    int last_percent = -1;
2232 +
2233 +    while (1) {
2234 +        main_loop_wait(false);
2235 +        vma_writer_get_status(vmaw, &vmastat);
2236 +
2237 +        if (verbose) {
2238 +
2239 +            uint64_t total = 0;
2240 +            uint64_t transferred = 0;
2241 +            uint64_t zero_bytes = 0;
2242 +
2243 +            int i;
2244 +            for (i = 0; i < 256; i++) {
2245 +                if (vmastat.stream_info[i].size) {
2246 +                    total += vmastat.stream_info[i].size;
2247 +                    transferred += vmastat.stream_info[i].transferred;
2248 +                    zero_bytes += vmastat.stream_info[i].zero_bytes;
2249 +                }
2250 +            }
2251 +            percent = (transferred*100)/total;
2252 +            if (percent != last_percent) {
2253 +                fprintf(stderr, "progress %d%% %zd/%zd %zd\n", percent,
2254 +                        transferred, total, zero_bytes);
2255 +                fflush(stderr);
2256 +
2257 +                last_percent = percent;
2258 +            }
2259 +        }
2260 +
2261 +        if (vmastat.closed) {
2262 +            break;
2263 +        }
2264 +    }
2265 +
2266 +    bdrv_drain_all();
2267 +
2268 +    vma_writer_get_status(vmaw, &vmastat);
2269 +
2270 +    if (verbose) {
2271 +        for (i = 0; i < 256; i++) {
2272 +            VmaStreamInfo *si = &vmastat.stream_info[i];
2273 +            if (si->size) {
2274 +                fprintf(stderr, "image %s: size=%zd zeros=%zd saved=%zd\n",
2275 +                        si->devname, si->size, si->zero_bytes,
2276 +                        si->size - si->zero_bytes);
2277 +            }
2278 +        }
2279 +    }
2280 +
2281 +    if (vmastat.status < 0) {
2282 +        unlink(archivename);
2283 +        g_error("creating vma archive failed");
2284 +    }
2285 +
2286 +    return 0;
2287 +}
2288 +
2289 +int main(int argc, char **argv)
2290 +{
2291 +    const char *cmdname;
2292 +
2293 +    error_set_progname(argv[0]);
2294 +
2295 +    qemu_init_main_loop();
2296 +
2297 +    bdrv_init();
2298 +
2299 +    if (argc < 2) {
2300 +        help();
2301 +    }
2302 +
2303 +    cmdname = argv[1];
2304 +    argc--; argv++;
2305 +
2306 +
2307 +    if (!strcmp(cmdname, "list")) {
2308 +        return list_content(argc, argv);
2309 +    } else if (!strcmp(cmdname, "create")) {
2310 +        return create_archive(argc, argv);
2311 +    } else if (!strcmp(cmdname, "extract")) {
2312 +        return extract_content(argc, argv);
2313 +    }
2314 +
2315 +    help();
2316 +    return 0;
2317 +}
2318 Index: new/vma.h
2319 ===================================================================
2320 --- /dev/null   1970-01-01 00:00:00.000000000 +0000
2321 +++ new/vma.h   2014-06-17 06:15:02.000000000 +0200
2322 @@ -0,0 +1,146 @@
2323 +/*
2324 + * VMA: Virtual Machine Archive
2325 + *
2326 + * Copyright (C) Proxmox Server Solutions
2327 + *
2328 + * Authors:
2329 + *  Dietmar Maurer (dietmar@proxmox.com)
2330 + *
2331 + * This work is licensed under the terms of the GNU GPL, version 2 or later.
2332 + * See the COPYING file in the top-level directory.
2333 + *
2334 + */
2335 +
2336 +#ifndef BACKUP_VMA_H
2337 +#define BACKUP_VMA_H
2338 +
2339 +#include <uuid/uuid.h>
2340 +#include "qapi/error.h"
2341 +#include "block/block.h"
2342 +
2343 +#define VMA_BLOCK_BITS 12
2344 +#define VMA_BLOCK_SIZE (1<<VMA_BLOCK_BITS)
2345 +#define VMA_CLUSTER_BITS (VMA_BLOCK_BITS+4)
2346 +#define VMA_CLUSTER_SIZE (1<<VMA_CLUSTER_BITS)
2347 +
2348 +#if VMA_CLUSTER_SIZE != 65536
2349 +#error unexpected cluster size
2350 +#endif
2351 +
2352 +#define VMA_EXTENT_HEADER_SIZE 512
2353 +#define VMA_BLOCKS_PER_EXTENT 59
2354 +#define VMA_MAX_CONFIGS 256
2355 +
2356 +#define VMA_MAX_EXTENT_SIZE \
2357 +    (VMA_EXTENT_HEADER_SIZE+VMA_CLUSTER_SIZE*VMA_BLOCKS_PER_EXTENT)
2358 +#if VMA_MAX_EXTENT_SIZE != 3867136
2359 +#error unexpected VMA_EXTENT_SIZE
2360 +#endif
2361 +
2362 +/* File Format Definitions */
2363 +
2364 +#define VMA_MAGIC (GUINT32_TO_BE(('V'<<24)|('M'<<16)|('A'<<8)|0x00))
2365 +#define VMA_EXTENT_MAGIC (GUINT32_TO_BE(('V'<<24)|('M'<<16)|('A'<<8)|'E'))
2366 +
2367 +typedef struct VmaDeviceInfoHeader {
2368 +    uint32_t devname_ptr; /* offset into blob_buffer table */
2369 +    uint32_t reserved0;
2370 +    uint64_t size; /* device size in bytes */
2371 +    uint64_t reserved1;
2372 +    uint64_t reserved2;
2373 +} VmaDeviceInfoHeader;
2374 +
2375 +typedef struct VmaHeader {
2376 +    uint32_t magic;
2377 +    uint32_t version;
2378 +    unsigned char uuid[16];
2379 +    int64_t ctime;
2380 +    unsigned char md5sum[16];
2381 +
2382 +    uint32_t blob_buffer_offset;
2383 +    uint32_t blob_buffer_size;
2384 +    uint32_t header_size;
2385 +
2386 +    unsigned char reserved[1984];
2387 +
2388 +    uint32_t config_names[VMA_MAX_CONFIGS]; /* offset into blob_buffer table */
2389 +    uint32_t config_data[VMA_MAX_CONFIGS];  /* offset into blob_buffer table */
2390 +
2391 +    uint32_t reserved1;
2392 +
2393 +    VmaDeviceInfoHeader dev_info[256];
2394 +} VmaHeader;
2395 +
2396 +typedef struct VmaExtentHeader {
2397 +    uint32_t magic;
2398 +    uint16_t reserved1;
2399 +    uint16_t block_count;
2400 +    unsigned char uuid[16];
2401 +    unsigned char md5sum[16];
2402 +    uint64_t blockinfo[VMA_BLOCKS_PER_EXTENT];
2403 +} VmaExtentHeader;
2404 +
2405 +/* functions/definitions to read/write vma files */
2406 +
2407 +typedef struct VmaReader VmaReader;
2408 +
2409 +typedef struct VmaWriter VmaWriter;
2410 +
2411 +typedef struct VmaConfigData {
2412 +    const char *name;
2413 +    const void *data;
2414 +    uint32_t len;
2415 +} VmaConfigData;
2416 +
2417 +typedef struct VmaStreamInfo {
2418 +    uint64_t size;
2419 +    uint64_t cluster_count;
2420 +    uint64_t transferred;
2421 +    uint64_t zero_bytes;
2422 +    int finished;
2423 +    char *devname;
2424 +} VmaStreamInfo;
2425 +
2426 +typedef struct VmaStatus {
2427 +    int status;
2428 +    bool closed;
2429 +    char errmsg[8192];
2430 +    char uuid_str[37];
2431 +    VmaStreamInfo stream_info[256];
2432 +} VmaStatus;
2433 +
2434 +typedef struct VmaDeviceInfo {
2435 +    uint64_t size; /* device size in bytes */
2436 +    const char *devname;
2437 +} VmaDeviceInfo;
2438 +
2439 +VmaWriter *vma_writer_create(const char *filename, uuid_t uuid, Error **errp);
2440 +int vma_writer_close(VmaWriter *vmaw, Error **errp);
2441 +void vma_writer_destroy(VmaWriter *vmaw);
2442 +int vma_writer_add_config(VmaWriter *vmaw, const char *name, gpointer data,
2443 +                          size_t len);
2444 +int vma_writer_register_stream(VmaWriter *vmaw, const char *devname,
2445 +                               size_t size);
2446 +
2447 +int64_t coroutine_fn vma_writer_write(VmaWriter *vmaw, uint8_t dev_id,
2448 +                                      int64_t cluster_num, unsigned char *buf,
2449 +                                      size_t *zero_bytes);
2450 +
2451 +int coroutine_fn vma_writer_close_stream(VmaWriter *vmaw, uint8_t dev_id);
2452 +
2453 +int vma_writer_get_status(VmaWriter *vmaw, VmaStatus *status);
2454 +void vma_writer_set_error(VmaWriter *vmaw, const char *fmt, ...);
2455 +
2456 +
2457 +VmaReader *vma_reader_create(const char *filename, Error **errp);
2458 +void vma_reader_destroy(VmaReader *vmar);
2459 +VmaHeader *vma_reader_get_header(VmaReader *vmar);
2460 +GList *vma_reader_get_config_data(VmaReader *vmar);
2461 +VmaDeviceInfo *vma_reader_get_device_info(VmaReader *vmar, guint8 dev_id);
2462 +int vma_reader_register_bs(VmaReader *vmar, guint8 dev_id,
2463 +                           BlockDriverState *bs, bool write_zeroes,
2464 +                           Error **errp);
2465 +int vma_reader_restore(VmaReader *vmar, int vmstate_fd, bool verbose,
2466 +                       Error **errp);
2467 +
2468 +#endif /* BACKUP_VMA_H */
2469 Index: new/Makefile.objs
2470 ===================================================================
2471 --- new.orig/Makefile.objs      2014-06-17 06:14:27.000000000 +0200
2472 +++ new/Makefile.objs   2014-06-17 06:15:02.000000000 +0200
2473 @@ -14,6 +14,7 @@
2474  block-obj-y += block/
2475  block-obj-y += qapi-types.o qapi-visit.o
2476  block-obj-y += qemu-io-cmds.o
2477 +block-obj-y += vma-writer.o
2478
2479  block-obj-y += qemu-coroutine.o qemu-coroutine-lock.o qemu-coroutine-io.o
2480  block-obj-y += qemu-coroutine-sleep.o