debian/patches/pve/0027-PVE-Backup-add-vma-backup-format-code.patch

   1 From 0000000000000000000000000000000000000000 Mon Sep 17 00:00:00 2001
   2 From: Dietmar Maurer <dietmar@proxmox.com>
   3 Date: Mon, 6 Apr 2020 12:16:57 +0200
   4 Subject: [PATCH] PVE-Backup: add vma backup format code
   5
   6 ---
   7  Makefile      |   3 +-
   8  Makefile.objs |   1 +
   9  vma-reader.c  | 857 ++++++++++++++++++++++++++++++++++++++++++++++++++
  10  vma-writer.c  | 790 ++++++++++++++++++++++++++++++++++++++++++++++
  11  vma.c         | 839 ++++++++++++++++++++++++++++++++++++++++++++++++
  12  vma.h         | 150 +++++++++
  13  6 files changed, 2639 insertions(+), 1 deletion(-)
  14  create mode 100644 vma-reader.c
  15  create mode 100644 vma-writer.c
  16  create mode 100644 vma.c
  17  create mode 100644 vma.h
  18
  19 diff --git a/Makefile b/Makefile
  20 index 8a9113e666..74c2039005 100644
  21 --- a/Makefile
  22 +++ b/Makefile
  23 @@ -479,7 +479,7 @@ dummy := $(call unnest-vars,, \
  24
  25  include $(SRC_PATH)/tests/Makefile.include
  26
  27 -all: $(DOCS) $(if $(BUILD_DOCS),sphinxdocs) $(TOOLS) $(HELPERS-y) recurse-all modules $(vhost-user-json-y)
  28 +all: $(DOCS) $(if $(BUILD_DOCS),sphinxdocs) $(TOOLS) vma$(EXESUF) $(HELPERS-y) recurse-all modules $(vhost-user-json-y)
  29
  30  qemu-version.h: FORCE
  31         $(call quiet-command, \
  32 @@ -608,6 +608,7 @@ qemu-img$(EXESUF): qemu-img.o $(authz-obj-y) $(block-obj-y) $(crypto-obj-y) $(io
  33  qemu-nbd$(EXESUF): qemu-nbd.o $(authz-obj-y) $(block-obj-y) $(crypto-obj-y) $(io-obj-y) $(qom-obj-y) $(COMMON_LDADDS)
  34  qemu-io$(EXESUF): qemu-io.o $(authz-obj-y) $(block-obj-y) $(crypto-obj-y) $(io-obj-y) $(qom-obj-y) $(COMMON_LDADDS)
  35  qemu-storage-daemon$(EXESUF): qemu-storage-daemon.o $(authz-obj-y) $(block-obj-y) $(crypto-obj-y) $(chardev-obj-y) $(io-obj-y) $(qom-obj-y) $(storage-daemon-obj-y) $(COMMON_LDADDS)
  36 +vma$(EXESUF): vma.o vma-reader.o $(authz-obj-y) $(block-obj-y) $(crypto-obj-y) $(io-obj-y) $(qom-obj-y) $(COMMON_LDADDS)
  37
  38  qemu-bridge-helper$(EXESUF): qemu-bridge-helper.o $(COMMON_LDADDS)
  39
  40 diff --git a/Makefile.objs b/Makefile.objs
  41 index d0b4dde836..05031a3da7 100644
  42 --- a/Makefile.objs
  43 +++ b/Makefile.objs
  44 @@ -18,6 +18,7 @@ block-obj-y += block.o blockjob.o job.o
  45  block-obj-y += block/ scsi/
  46  block-obj-y += qemu-io-cmds.o
  47  block-obj-$(CONFIG_REPLICATION) += replication.o
  48 +block-obj-y += vma-writer.o
  49
  50  block-obj-m = block/
  51
  52 diff --git a/vma-reader.c b/vma-reader.c
  53 new file mode 100644
  54 index 0000000000..2b1d1cdab3
  55 --- /dev/null
  56 +++ b/vma-reader.c
  57 @@ -0,0 +1,857 @@
  58 +/*
  59 + * VMA: Virtual Machine Archive
  60 + *
  61 + * Copyright (C) 2012 Proxmox Server Solutions
  62 + *
  63 + * Authors:
  64 + *  Dietmar Maurer (dietmar@proxmox.com)
  65 + *
  66 + * This work is licensed under the terms of the GNU GPL, version 2 or later.
  67 + * See the COPYING file in the top-level directory.
  68 + *
  69 + */
  70 +
  71 +#include "qemu/osdep.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 +#include "sysemu/block-backend.h"
  81 +
  82 +static unsigned char zero_vma_block[VMA_BLOCK_SIZE];
  83 +
  84 +typedef struct VmaRestoreState {
  85 +    BlockBackend *target;
  86 +    bool write_zeroes;
  87 +    unsigned long *bitmap;
  88 +    int bitmap_size;
  89 +}  VmaRestoreState;
  90 +
  91 +struct VmaReader {
  92 +    int fd;
  93 +    GChecksum *md5csum;
  94 +    GHashTable *blob_hash;
  95 +    unsigned char *head_data;
  96 +    VmaDeviceInfo devinfo[256];
  97 +    VmaRestoreState rstate[256];
  98 +    GList *cdata_list;
  99 +    guint8 vmstate_stream;
 100 +    uint32_t vmstate_clusters;
 101 +    /* to show restore percentage if run with -v */
 102 +    time_t start_time;
 103 +    int64_t cluster_count;
 104 +    int64_t clusters_read;
 105 +    int64_t zero_cluster_data;
 106 +    int64_t partial_zero_cluster_data;
 107 +    int clusters_read_per;
 108 +};
 109 +
 110 +static guint
 111 +g_int32_hash(gconstpointer v)
 112 +{
 113 +    return *(const uint32_t *)v;
 114 +}
 115 +
 116 +static gboolean
 117 +g_int32_equal(gconstpointer v1, gconstpointer v2)
 118 +{
 119 +    return *((const uint32_t *)v1) == *((const uint32_t *)v2);
 120 +}
 121 +
 122 +static int vma_reader_get_bitmap(VmaRestoreState *rstate, int64_t cluster_num)
 123 +{
 124 +    assert(rstate);
 125 +    assert(rstate->bitmap);
 126 +
 127 +    unsigned long val, idx, bit;
 128 +
 129 +    idx = cluster_num / BITS_PER_LONG;
 130 +
 131 +    assert(rstate->bitmap_size > idx);
 132 +
 133 +    bit = cluster_num % BITS_PER_LONG;
 134 +    val = rstate->bitmap[idx];
 135 +
 136 +    return !!(val & (1UL << bit));
 137 +}
 138 +
 139 +static void vma_reader_set_bitmap(VmaRestoreState *rstate, int64_t cluster_num,
 140 +                                  int dirty)
 141 +{
 142 +    assert(rstate);
 143 +    assert(rstate->bitmap);
 144 +
 145 +    unsigned long val, idx, bit;
 146 +
 147 +    idx = cluster_num / BITS_PER_LONG;
 148 +
 149 +    assert(rstate->bitmap_size > idx);
 150 +
 151 +    bit = cluster_num % BITS_PER_LONG;
 152 +    val = rstate->bitmap[idx];
 153 +    if (dirty) {
 154 +        if (!(val & (1UL << bit))) {
 155 +            val |= 1UL << bit;
 156 +        }
 157 +    } else {
 158 +        if (val & (1UL << bit)) {
 159 +            val &= ~(1UL << bit);
 160 +        }
 161 +    }
 162 +    rstate->bitmap[idx] = val;
 163 +}
 164 +
 165 +typedef struct VmaBlob {
 166 +    uint32_t start;
 167 +    uint32_t len;
 168 +    void *data;
 169 +} VmaBlob;
 170 +
 171 +static const VmaBlob *get_header_blob(VmaReader *vmar, uint32_t pos)
 172 +{
 173 +    assert(vmar);
 174 +    assert(vmar->blob_hash);
 175 +
 176 +    return g_hash_table_lookup(vmar->blob_hash, &pos);
 177 +}
 178 +
 179 +static const char *get_header_str(VmaReader *vmar, uint32_t pos)
 180 +{
 181 +    const VmaBlob *blob = get_header_blob(vmar, pos);
 182 +    if (!blob) {
 183 +        return NULL;
 184 +    }
 185 +    const char *res = (char *)blob->data;
 186 +    if (res[blob->len-1] != '\0') {
 187 +        return NULL;
 188 +    }
 189 +    return res;
 190 +}
 191 +
 192 +static ssize_t
 193 +safe_read(int fd, unsigned char *buf, size_t count)
 194 +{
 195 +    ssize_t n;
 196 +
 197 +    do {
 198 +        n = read(fd, buf, count);
 199 +    } while (n < 0 && errno == EINTR);
 200 +
 201 +    return n;
 202 +}
 203 +
 204 +static ssize_t
 205 +full_read(int fd, unsigned char *buf, size_t len)
 206 +{
 207 +    ssize_t n;
 208 +    size_t total;
 209 +
 210 +    total = 0;
 211 +
 212 +    while (len > 0) {
 213 +        n = safe_read(fd, buf, len);
 214 +
 215 +        if (n == 0) {
 216 +            return total;
 217 +        }
 218 +
 219 +        if (n <= 0) {
 220 +            break;
 221 +        }
 222 +
 223 +        buf += n;
 224 +        total += n;
 225 +        len -= n;
 226 +    }
 227 +
 228 +    if (len) {
 229 +        return -1;
 230 +    }
 231 +
 232 +    return total;
 233 +}
 234 +
 235 +void vma_reader_destroy(VmaReader *vmar)
 236 +{
 237 +    assert(vmar);
 238 +
 239 +    if (vmar->fd >= 0) {
 240 +        close(vmar->fd);
 241 +    }
 242 +
 243 +    if (vmar->cdata_list) {
 244 +        g_list_free(vmar->cdata_list);
 245 +    }
 246 +
 247 +    int i;
 248 +    for (i = 1; i < 256; i++) {
 249 +        if (vmar->rstate[i].bitmap) {
 250 +            g_free(vmar->rstate[i].bitmap);
 251 +        }
 252 +    }
 253 +
 254 +    if (vmar->md5csum) {
 255 +        g_checksum_free(vmar->md5csum);
 256 +    }
 257 +
 258 +    if (vmar->blob_hash) {
 259 +        g_hash_table_destroy(vmar->blob_hash);
 260 +    }
 261 +
 262 +    if (vmar->head_data) {
 263 +        g_free(vmar->head_data);
 264 +    }
 265 +
 266 +    g_free(vmar);
 267 +
 268 +};
 269 +
 270 +static int vma_reader_read_head(VmaReader *vmar, Error **errp)
 271 +{
 272 +    assert(vmar);
 273 +    assert(errp);
 274 +    assert(*errp == NULL);
 275 +
 276 +    unsigned char md5sum[16];
 277 +    int i;
 278 +    int ret = 0;
 279 +
 280 +    vmar->head_data = g_malloc(sizeof(VmaHeader));
 281 +
 282 +    if (full_read(vmar->fd, vmar->head_data, sizeof(VmaHeader)) !=
 283 +        sizeof(VmaHeader)) {
 284 +        error_setg(errp, "can't read vma header - %s",
 285 +                   errno ? g_strerror(errno) : "got EOF");
 286 +        return -1;
 287 +    }
 288 +
 289 +    VmaHeader *h = (VmaHeader *)vmar->head_data;
 290 +
 291 +    if (h->magic != VMA_MAGIC) {
 292 +        error_setg(errp, "not a vma file - wrong magic number");
 293 +        return -1;
 294 +    }
 295 +
 296 +    uint32_t header_size = GUINT32_FROM_BE(h->header_size);
 297 +    int need = header_size - sizeof(VmaHeader);
 298 +    if (need <= 0) {
 299 +        error_setg(errp, "wrong vma header size %d", header_size);
 300 +        return -1;
 301 +    }
 302 +
 303 +    vmar->head_data = g_realloc(vmar->head_data, header_size);
 304 +    h = (VmaHeader *)vmar->head_data;
 305 +
 306 +    if (full_read(vmar->fd, vmar->head_data + sizeof(VmaHeader), need) !=
 307 +        need) {
 308 +        error_setg(errp, "can't read vma header data - %s",
 309 +                   errno ? g_strerror(errno) : "got EOF");
 310 +        return -1;
 311 +    }
 312 +
 313 +    memcpy(md5sum, h->md5sum, 16);
 314 +    memset(h->md5sum, 0, 16);
 315 +
 316 +    g_checksum_reset(vmar->md5csum);
 317 +    g_checksum_update(vmar->md5csum, vmar->head_data, header_size);
 318 +    gsize csize = 16;
 319 +    g_checksum_get_digest(vmar->md5csum, (guint8 *)(h->md5sum), &csize);
 320 +
 321 +    if (memcmp(md5sum, h->md5sum, 16) != 0) {
 322 +        error_setg(errp, "wrong vma header chechsum");
 323 +        return -1;
 324 +    }
 325 +
 326 +    /* we can modify header data after checksum verify */
 327 +    h->header_size = header_size;
 328 +
 329 +    h->version = GUINT32_FROM_BE(h->version);
 330 +    if (h->version != 1) {
 331 +        error_setg(errp, "wrong vma version %d", h->version);
 332 +        return -1;
 333 +    }
 334 +
 335 +    h->ctime = GUINT64_FROM_BE(h->ctime);
 336 +    h->blob_buffer_offset = GUINT32_FROM_BE(h->blob_buffer_offset);
 337 +    h->blob_buffer_size = GUINT32_FROM_BE(h->blob_buffer_size);
 338 +
 339 +    uint32_t bstart = h->blob_buffer_offset + 1;
 340 +    uint32_t bend = h->blob_buffer_offset + h->blob_buffer_size;
 341 +
 342 +    if (bstart <= sizeof(VmaHeader)) {
 343 +        error_setg(errp, "wrong vma blob buffer offset %d",
 344 +                   h->blob_buffer_offset);
 345 +        return -1;
 346 +    }
 347 +
 348 +    if (bend > header_size) {
 349 +        error_setg(errp, "wrong vma blob buffer size %d/%d",
 350 +                   h->blob_buffer_offset, h->blob_buffer_size);
 351 +        return -1;
 352 +    }
 353 +
 354 +    while ((bstart + 2) <= bend) {
 355 +        uint32_t size = vmar->head_data[bstart] +
 356 +            (vmar->head_data[bstart+1] << 8);
 357 +        if ((bstart + size + 2) <= bend) {
 358 +            VmaBlob *blob = g_new0(VmaBlob, 1);
 359 +            blob->start = bstart - h->blob_buffer_offset;
 360 +            blob->len = size;
 361 +            blob->data = vmar->head_data + bstart + 2;
 362 +            g_hash_table_insert(vmar->blob_hash, &blob->start, blob);
 363 +        }
 364 +        bstart += size + 2;
 365 +    }
 366 +
 367 +
 368 +    int count = 0;
 369 +    for (i = 1; i < 256; i++) {
 370 +        VmaDeviceInfoHeader *dih = &h->dev_info[i];
 371 +        uint32_t devname_ptr = GUINT32_FROM_BE(dih->devname_ptr);
 372 +        uint64_t size = GUINT64_FROM_BE(dih->size);
 373 +        const char *devname =  get_header_str(vmar, devname_ptr);
 374 +
 375 +        if (size && devname) {
 376 +            count++;
 377 +            vmar->devinfo[i].size = size;
 378 +            vmar->devinfo[i].devname = devname;
 379 +
 380 +            if (strcmp(devname, "vmstate") == 0) {
 381 +                vmar->vmstate_stream = i;
 382 +            }
 383 +        }
 384 +    }
 385 +
 386 +    for (i = 0; i < VMA_MAX_CONFIGS; i++) {
 387 +        uint32_t name_ptr = GUINT32_FROM_BE(h->config_names[i]);
 388 +        uint32_t data_ptr = GUINT32_FROM_BE(h->config_data[i]);
 389 +
 390 +        if (!(name_ptr && data_ptr)) {
 391 +            continue;
 392 +        }
 393 +        const char *name =  get_header_str(vmar, name_ptr);
 394 +        const VmaBlob *blob = get_header_blob(vmar, data_ptr);
 395 +
 396 +        if (!(name && blob)) {
 397 +            error_setg(errp, "vma contains invalid data pointers");
 398 +            return -1;
 399 +        }
 400 +
 401 +        VmaConfigData *cdata = g_new0(VmaConfigData, 1);
 402 +        cdata->name = name;
 403 +        cdata->data = blob->data;
 404 +        cdata->len = blob->len;
 405 +
 406 +        vmar->cdata_list = g_list_append(vmar->cdata_list, cdata);
 407 +    }
 408 +
 409 +    return ret;
 410 +};
 411 +
 412 +VmaReader *vma_reader_create(const char *filename, Error **errp)
 413 +{
 414 +    assert(filename);
 415 +    assert(errp);
 416 +
 417 +    VmaReader *vmar = g_new0(VmaReader, 1);
 418 +
 419 +    if (strcmp(filename, "-") == 0) {
 420 +        vmar->fd = dup(0);
 421 +    } else {
 422 +        vmar->fd = open(filename, O_RDONLY);
 423 +    }
 424 +
 425 +    if (vmar->fd < 0) {
 426 +        error_setg(errp, "can't open file %s - %s\n", filename,
 427 +                   g_strerror(errno));
 428 +        goto err;
 429 +    }
 430 +
 431 +    vmar->md5csum = g_checksum_new(G_CHECKSUM_MD5);
 432 +    if (!vmar->md5csum) {
 433 +        error_setg(errp, "can't allocate cmsum\n");
 434 +        goto err;
 435 +    }
 436 +
 437 +    vmar->blob_hash = g_hash_table_new_full(g_int32_hash, g_int32_equal,
 438 +                                            NULL, g_free);
 439 +
 440 +    if (vma_reader_read_head(vmar, errp) < 0) {
 441 +        goto err;
 442 +    }
 443 +
 444 +    return vmar;
 445 +
 446 +err:
 447 +    if (vmar) {
 448 +        vma_reader_destroy(vmar);
 449 +    }
 450 +
 451 +    return NULL;
 452 +}
 453 +
 454 +VmaHeader *vma_reader_get_header(VmaReader *vmar)
 455 +{
 456 +    assert(vmar);
 457 +    assert(vmar->head_data);
 458 +
 459 +    return (VmaHeader *)(vmar->head_data);
 460 +}
 461 +
 462 +GList *vma_reader_get_config_data(VmaReader *vmar)
 463 +{
 464 +    assert(vmar);
 465 +    assert(vmar->head_data);
 466 +
 467 +    return vmar->cdata_list;
 468 +}
 469 +
 470 +VmaDeviceInfo *vma_reader_get_device_info(VmaReader *vmar, guint8 dev_id)
 471 +{
 472 +    assert(vmar);
 473 +    assert(dev_id);
 474 +
 475 +    if (vmar->devinfo[dev_id].size && vmar->devinfo[dev_id].devname) {
 476 +        return &vmar->devinfo[dev_id];
 477 +    }
 478 +
 479 +    return NULL;
 480 +}
 481 +
 482 +static void allocate_rstate(VmaReader *vmar,  guint8 dev_id,
 483 +                            BlockBackend *target, bool write_zeroes)
 484 +{
 485 +    assert(vmar);
 486 +    assert(dev_id);
 487 +
 488 +    vmar->rstate[dev_id].target = target;
 489 +    vmar->rstate[dev_id].write_zeroes = write_zeroes;
 490 +
 491 +    int64_t size = vmar->devinfo[dev_id].size;
 492 +
 493 +    int64_t bitmap_size = (size/BDRV_SECTOR_SIZE) +
 494 +        (VMA_CLUSTER_SIZE/BDRV_SECTOR_SIZE) * BITS_PER_LONG - 1;
 495 +    bitmap_size /= (VMA_CLUSTER_SIZE/BDRV_SECTOR_SIZE) * BITS_PER_LONG;
 496 +
 497 +    vmar->rstate[dev_id].bitmap_size = bitmap_size;
 498 +    vmar->rstate[dev_id].bitmap = g_new0(unsigned long, bitmap_size);
 499 +
 500 +    vmar->cluster_count += size/VMA_CLUSTER_SIZE;
 501 +}
 502 +
 503 +int vma_reader_register_bs(VmaReader *vmar, guint8 dev_id, BlockBackend *target,
 504 +                           bool write_zeroes, Error **errp)
 505 +{
 506 +    assert(vmar);
 507 +    assert(target != NULL);
 508 +    assert(dev_id);
 509 +    assert(vmar->rstate[dev_id].target == NULL);
 510 +
 511 +    int64_t size = blk_getlength(target);
 512 +    int64_t size_diff = size - vmar->devinfo[dev_id].size;
 513 +
 514 +    /* storage types can have different size restrictions, so it
 515 +     * is not always possible to create an image with exact size.
 516 +     * So we tolerate a size difference up to 4MB.
 517 +     */
 518 +    if ((size_diff < 0) || (size_diff > 4*1024*1024)) {
 519 +        error_setg(errp, "vma_reader_register_bs for stream %s failed - "
 520 +                   "unexpected size %zd != %zd", vmar->devinfo[dev_id].devname,
 521 +                   size, vmar->devinfo[dev_id].size);
 522 +        return -1;
 523 +    }
 524 +
 525 +    allocate_rstate(vmar, dev_id, target, write_zeroes);
 526 +
 527 +    return 0;
 528 +}
 529 +
 530 +static ssize_t safe_write(int fd, void *buf, size_t count)
 531 +{
 532 +    ssize_t n;
 533 +
 534 +    do {
 535 +        n = write(fd, buf, count);
 536 +    } while (n < 0 && errno == EINTR);
 537 +
 538 +    return n;
 539 +}
 540 +
 541 +static size_t full_write(int fd, void *buf, size_t len)
 542 +{
 543 +    ssize_t n;
 544 +    size_t total;
 545 +
 546 +    total = 0;
 547 +
 548 +    while (len > 0) {
 549 +        n = safe_write(fd, buf, len);
 550 +        if (n < 0) {
 551 +            return n;
 552 +        }
 553 +        buf += n;
 554 +        total += n;
 555 +        len -= n;
 556 +    }
 557 +
 558 +    if (len) {
 559 +        /* incomplete write ? */
 560 +        return -1;
 561 +    }
 562 +
 563 +    return total;
 564 +}
 565 +
 566 +static int restore_write_data(VmaReader *vmar, guint8 dev_id,
 567 +                              BlockBackend *target, int vmstate_fd,
 568 +                              unsigned char *buf, int64_t sector_num,
 569 +                              int nb_sectors, Error **errp)
 570 +{
 571 +    assert(vmar);
 572 +
 573 +    if (dev_id == vmar->vmstate_stream) {
 574 +        if (vmstate_fd >= 0) {
 575 +            int len = nb_sectors * BDRV_SECTOR_SIZE;
 576 +            int res = full_write(vmstate_fd, buf, len);
 577 +            if (res < 0) {
 578 +                error_setg(errp, "write vmstate failed %d", res);
 579 +                return -1;
 580 +            }
 581 +        }
 582 +    } else {
 583 +        int res = blk_pwrite(target, sector_num * BDRV_SECTOR_SIZE, buf, nb_sectors * BDRV_SECTOR_SIZE, 0);
 584 +        if (res < 0) {
 585 +            error_setg(errp, "blk_pwrite to %s failed (%d)",
 586 +                       bdrv_get_device_name(blk_bs(target)), res);
 587 +            return -1;
 588 +        }
 589 +    }
 590 +    return 0;
 591 +}
 592 +
 593 +static int restore_extent(VmaReader *vmar, unsigned char *buf,
 594 +                          int extent_size, int vmstate_fd,
 595 +                          bool verbose, bool verify, Error **errp)
 596 +{
 597 +    assert(vmar);
 598 +    assert(buf);
 599 +
 600 +    VmaExtentHeader *ehead = (VmaExtentHeader *)buf;
 601 +    int start = VMA_EXTENT_HEADER_SIZE;
 602 +    int i;
 603 +
 604 +    for (i = 0; i < VMA_BLOCKS_PER_EXTENT; i++) {
 605 +        uint64_t block_info = GUINT64_FROM_BE(ehead->blockinfo[i]);
 606 +        uint64_t cluster_num = block_info & 0xffffffff;
 607 +        uint8_t dev_id = (block_info >> 32) & 0xff;
 608 +        uint16_t mask = block_info >> (32+16);
 609 +        int64_t max_sector;
 610 +
 611 +        if (!dev_id) {
 612 +            continue;
 613 +        }
 614 +
 615 +        VmaRestoreState *rstate = &vmar->rstate[dev_id];
 616 +        BlockBackend *target = NULL;
 617 +
 618 +        if (dev_id != vmar->vmstate_stream) {
 619 +            target = rstate->target;
 620 +            if (!verify && !target) {
 621 +                error_setg(errp, "got wrong dev id %d", dev_id);
 622 +                return -1;
 623 +            }
 624 +
 625 +            if (vma_reader_get_bitmap(rstate, cluster_num)) {
 626 +                error_setg(errp, "found duplicated cluster %zd for stream %s",
 627 +                          cluster_num, vmar->devinfo[dev_id].devname);
 628 +                return -1;
 629 +            }
 630 +            vma_reader_set_bitmap(rstate, cluster_num, 1);
 631 +
 632 +            max_sector = vmar->devinfo[dev_id].size/BDRV_SECTOR_SIZE;
 633 +        } else {
 634 +            max_sector = G_MAXINT64;
 635 +            if (cluster_num != vmar->vmstate_clusters) {
 636 +                error_setg(errp, "found out of order vmstate data");
 637 +                return -1;
 638 +            }
 639 +            vmar->vmstate_clusters++;
 640 +        }
 641 +
 642 +        vmar->clusters_read++;
 643 +
 644 +        if (verbose) {
 645 +            time_t duration = time(NULL) - vmar->start_time;
 646 +            int percent = (vmar->clusters_read*100)/vmar->cluster_count;
 647 +            if (percent != vmar->clusters_read_per) {
 648 +                printf("progress %d%% (read %zd bytes, duration %zd sec)\n",
 649 +                       percent, vmar->clusters_read*VMA_CLUSTER_SIZE,
 650 +                       duration);
 651 +                fflush(stdout);
 652 +                vmar->clusters_read_per = percent;
 653 +            }
 654 +        }
 655 +
 656 +        /* try to write whole clusters to speedup restore */
 657 +        if (mask == 0xffff) {
 658 +            if ((start + VMA_CLUSTER_SIZE) > extent_size) {
 659 +                error_setg(errp, "short vma extent - too many blocks");
 660 +                return -1;
 661 +            }
 662 +            int64_t sector_num = (cluster_num * VMA_CLUSTER_SIZE) /
 663 +                BDRV_SECTOR_SIZE;
 664 +            int64_t end_sector = sector_num +
 665 +                VMA_CLUSTER_SIZE/BDRV_SECTOR_SIZE;
 666 +
 667 +            if (end_sector > max_sector) {
 668 +                end_sector = max_sector;
 669 +            }
 670 +
 671 +            if (end_sector <= sector_num) {
 672 +                error_setg(errp, "got wrong block address - write beyond end");
 673 +                return -1;
 674 +            }
 675 +
 676 +            if (!verify) {
 677 +                int nb_sectors = end_sector - sector_num;
 678 +                if (restore_write_data(vmar, dev_id, target, vmstate_fd,
 679 +                                       buf + start, sector_num, nb_sectors,
 680 +                                       errp) < 0) {
 681 +                    return -1;
 682 +                }
 683 +            }
 684 +
 685 +            start += VMA_CLUSTER_SIZE;
 686 +        } else {
 687 +            int j;
 688 +            int bit = 1;
 689 +
 690 +            for (j = 0; j < 16; j++) {
 691 +                int64_t sector_num = (cluster_num*VMA_CLUSTER_SIZE +
 692 +                                      j*VMA_BLOCK_SIZE)/BDRV_SECTOR_SIZE;
 693 +
 694 +                int64_t end_sector = sector_num +
 695 +                    VMA_BLOCK_SIZE/BDRV_SECTOR_SIZE;
 696 +                if (end_sector > max_sector) {
 697 +                    end_sector = max_sector;
 698 +                }
 699 +
 700 +                if (mask & bit) {
 701 +                    if ((start + VMA_BLOCK_SIZE) > extent_size) {
 702 +                        error_setg(errp, "short vma extent - too many blocks");
 703 +                        return -1;
 704 +                    }
 705 +
 706 +                    if (end_sector <= sector_num) {
 707 +                        error_setg(errp, "got wrong block address - "
 708 +                                   "write beyond end");
 709 +                        return -1;
 710 +                    }
 711 +
 712 +                    if (!verify) {
 713 +                        int nb_sectors = end_sector - sector_num;
 714 +                        if (restore_write_data(vmar, dev_id, target, vmstate_fd,
 715 +                                               buf + start, sector_num,
 716 +                                               nb_sectors, errp) < 0) {
 717 +                            return -1;
 718 +                        }
 719 +                    }
 720 +
 721 +                    start += VMA_BLOCK_SIZE;
 722 +
 723 +                } else {
 724 +
 725 +
 726 +                    if (end_sector > sector_num) {
 727 +                        /* Todo: use bdrv_co_write_zeroes (but that need to
 728 +                         * be run inside coroutine?)
 729 +                         */
 730 +                        int nb_sectors = end_sector - sector_num;
 731 +                        int zero_size = BDRV_SECTOR_SIZE*nb_sectors;
 732 +                        vmar->zero_cluster_data += zero_size;
 733 +                        if (mask != 0) {
 734 +                            vmar->partial_zero_cluster_data += zero_size;
 735 +                        }
 736 +
 737 +                        if (rstate->write_zeroes && !verify) {
 738 +                            if (restore_write_data(vmar, dev_id, target, vmstate_fd,
 739 +                                                   zero_vma_block, sector_num,
 740 +                                                   nb_sectors, errp) < 0) {
 741 +                                return -1;
 742 +                            }
 743 +                        }
 744 +                    }
 745 +                }
 746 +
 747 +                bit = bit << 1;
 748 +            }
 749 +        }
 750 +    }
 751 +
 752 +    if (start != extent_size) {
 753 +        error_setg(errp, "vma extent error - missing blocks");
 754 +        return -1;
 755 +    }
 756 +
 757 +    return 0;
 758 +}
 759 +
 760 +static int vma_reader_restore_full(VmaReader *vmar, int vmstate_fd,
 761 +                                   bool verbose, bool verify,
 762 +                                   Error **errp)
 763 +{
 764 +    assert(vmar);
 765 +    assert(vmar->head_data);
 766 +
 767 +    int ret = 0;
 768 +    unsigned char buf[VMA_MAX_EXTENT_SIZE];
 769 +    int buf_pos = 0;
 770 +    unsigned char md5sum[16];
 771 +    VmaHeader *h = (VmaHeader *)vmar->head_data;
 772 +
 773 +    vmar->start_time = time(NULL);
 774 +
 775 +    while (1) {
 776 +        int bytes = full_read(vmar->fd, buf + buf_pos, sizeof(buf) - buf_pos);
 777 +        if (bytes < 0) {
 778 +            error_setg(errp, "read failed - %s", g_strerror(errno));
 779 +            return -1;
 780 +        }
 781 +
 782 +        buf_pos += bytes;
 783 +
 784 +        if (!buf_pos) {
 785 +            break; /* EOF */
 786 +        }
 787 +
 788 +        if (buf_pos < VMA_EXTENT_HEADER_SIZE) {
 789 +            error_setg(errp, "read short extent (%d bytes)", buf_pos);
 790 +            return -1;
 791 +        }
 792 +
 793 +        VmaExtentHeader *ehead = (VmaExtentHeader *)buf;
 794 +
 795 +        /* extract md5sum */
 796 +        memcpy(md5sum, ehead->md5sum, sizeof(ehead->md5sum));
 797 +        memset(ehead->md5sum, 0, sizeof(ehead->md5sum));
 798 +
 799 +        g_checksum_reset(vmar->md5csum);
 800 +        g_checksum_update(vmar->md5csum, buf, VMA_EXTENT_HEADER_SIZE);
 801 +        gsize csize = 16;
 802 +        g_checksum_get_digest(vmar->md5csum, ehead->md5sum, &csize);
 803 +
 804 +        if (memcmp(md5sum, ehead->md5sum, 16) != 0) {
 805 +            error_setg(errp, "wrong vma extent header chechsum");
 806 +            return -1;
 807 +        }
 808 +
 809 +        if (memcmp(h->uuid, ehead->uuid, sizeof(ehead->uuid)) != 0) {
 810 +            error_setg(errp, "wrong vma extent uuid");
 811 +            return -1;
 812 +        }
 813 +
 814 +        if (ehead->magic != VMA_EXTENT_MAGIC || ehead->reserved1 != 0) {
 815 +            error_setg(errp, "wrong vma extent header magic");
 816 +            return -1;
 817 +        }
 818 +
 819 +        int block_count = GUINT16_FROM_BE(ehead->block_count);
 820 +        int extent_size = VMA_EXTENT_HEADER_SIZE + block_count*VMA_BLOCK_SIZE;
 821 +
 822 +        if (buf_pos < extent_size) {
 823 +            error_setg(errp, "short vma extent (%d < %d)", buf_pos,
 824 +                       extent_size);
 825 +            return -1;
 826 +        }
 827 +
 828 +        if (restore_extent(vmar, buf, extent_size, vmstate_fd, verbose,
 829 +                           verify, errp) < 0) {
 830 +            return -1;
 831 +        }
 832 +
 833 +        if (buf_pos > extent_size) {
 834 +            memmove(buf, buf + extent_size, buf_pos - extent_size);
 835 +            buf_pos = buf_pos - extent_size;
 836 +        } else {
 837 +            buf_pos = 0;
 838 +        }
 839 +    }
 840 +
 841 +    bdrv_drain_all();
 842 +
 843 +    int i;
 844 +    for (i = 1; i < 256; i++) {
 845 +        VmaRestoreState *rstate = &vmar->rstate[i];
 846 +        if (!rstate->target) {
 847 +            continue;
 848 +        }
 849 +
 850 +        if (blk_flush(rstate->target) < 0) {
 851 +            error_setg(errp, "vma blk_flush %s failed",
 852 +                       vmar->devinfo[i].devname);
 853 +            return -1;
 854 +        }
 855 +
 856 +        if (vmar->devinfo[i].size &&
 857 +            (strcmp(vmar->devinfo[i].devname, "vmstate") != 0)) {
 858 +            assert(rstate->bitmap);
 859 +
 860 +            int64_t cluster_num, end;
 861 +
 862 +            end = (vmar->devinfo[i].size + VMA_CLUSTER_SIZE - 1) /
 863 +                VMA_CLUSTER_SIZE;
 864 +
 865 +            for (cluster_num = 0; cluster_num < end; cluster_num++) {
 866 +                if (!vma_reader_get_bitmap(rstate, cluster_num)) {
 867 +                    error_setg(errp, "detected missing cluster %zd "
 868 +                               "for stream %s", cluster_num,
 869 +                               vmar->devinfo[i].devname);
 870 +                    return -1;
 871 +                }
 872 +            }
 873 +        }
 874 +    }
 875 +
 876 +    if (verbose) {
 877 +        if (vmar->clusters_read) {
 878 +            printf("total bytes read %zd, sparse bytes %zd (%.3g%%)\n",
 879 +                   vmar->clusters_read*VMA_CLUSTER_SIZE,
 880 +                   vmar->zero_cluster_data,
 881 +                   (double)(100.0*vmar->zero_cluster_data)/
 882 +                   (vmar->clusters_read*VMA_CLUSTER_SIZE));
 883 +
 884 +            int64_t datasize = vmar->clusters_read*VMA_CLUSTER_SIZE-vmar->zero_cluster_data;
 885 +            if (datasize) { // this does not make sense for empty files
 886 +                printf("space reduction due to 4K zero blocks %.3g%%\n",
 887 +                       (double)(100.0*vmar->partial_zero_cluster_data) / datasize);
 888 +            }
 889 +        } else {
 890 +            printf("vma archive contains no image data\n");
 891 +        }
 892 +    }
 893 +    return ret;
 894 +}
 895 +
 896 +int vma_reader_restore(VmaReader *vmar, int vmstate_fd, bool verbose,
 897 +                       Error **errp)
 898 +{
 899 +    return vma_reader_restore_full(vmar, vmstate_fd, verbose, false, errp);
 900 +}
 901 +
 902 +int vma_reader_verify(VmaReader *vmar, bool verbose, Error **errp)
 903 +{
 904 +    guint8 dev_id;
 905 +
 906 +    for (dev_id = 1; dev_id < 255; dev_id++) {
 907 +        if (vma_reader_get_device_info(vmar, dev_id)) {
 908 +            allocate_rstate(vmar, dev_id, NULL, false);
 909 +        }
 910 +    }
 911 +
 912 +    return vma_reader_restore_full(vmar, -1, verbose, true, errp);
 913 +}
 914 +
 915 diff --git a/vma-writer.c b/vma-writer.c
 916 new file mode 100644
 917 index 0000000000..f5d2c5d23c
 918 --- /dev/null
 919 +++ b/vma-writer.c
 920 @@ -0,0 +1,790 @@
 921 +/*
 922 + * VMA: Virtual Machine Archive
 923 + *
 924 + * Copyright (C) 2012 Proxmox Server Solutions
 925 + *
 926 + * Authors:
 927 + *  Dietmar Maurer (dietmar@proxmox.com)
 928 + *
 929 + * This work is licensed under the terms of the GNU GPL, version 2 or later.
 930 + * See the COPYING file in the top-level directory.
 931 + *
 932 + */
 933 +
 934 +#include "qemu/osdep.h"
 935 +#include <glib.h>
 936 +#include <uuid/uuid.h>
 937 +
 938 +#include "vma.h"
 939 +#include "block/block.h"
 940 +#include "monitor/monitor.h"
 941 +#include "qemu/main-loop.h"
 942 +#include "qemu/coroutine.h"
 943 +#include "qemu/cutils.h"
 944 +
 945 +#define DEBUG_VMA 0
 946 +
 947 +#define DPRINTF(fmt, ...)\
 948 +    do { if (DEBUG_VMA) { printf("vma: " fmt, ## __VA_ARGS__); } } while (0)
 949 +
 950 +#define WRITE_BUFFERS 5
 951 +#define HEADER_CLUSTERS 8
 952 +#define HEADERBUF_SIZE (VMA_CLUSTER_SIZE*HEADER_CLUSTERS)
 953 +
 954 +struct VmaWriter {
 955 +    int fd;
 956 +    FILE *cmd;
 957 +    int status;
 958 +    char errmsg[8192];
 959 +    uuid_t uuid;
 960 +    bool header_written;
 961 +    bool closed;
 962 +
 963 +    /* we always write extents */
 964 +    unsigned char *outbuf;
 965 +    int outbuf_pos; /* in bytes */
 966 +    int outbuf_count; /* in VMA_BLOCKS */
 967 +    uint64_t outbuf_block_info[VMA_BLOCKS_PER_EXTENT];
 968 +
 969 +    unsigned char *headerbuf;
 970 +
 971 +    GChecksum *md5csum;
 972 +    CoMutex flush_lock;
 973 +    Coroutine *co_writer;
 974 +
 975 +    /* drive informations */
 976 +    VmaStreamInfo stream_info[256];
 977 +    guint stream_count;
 978 +
 979 +    guint8 vmstate_stream;
 980 +    uint32_t vmstate_clusters;
 981 +
 982 +    /* header blob table */
 983 +    char *header_blob_table;
 984 +    uint32_t header_blob_table_size;
 985 +    uint32_t header_blob_table_pos;
 986 +
 987 +    /* store for config blobs */
 988 +    uint32_t config_names[VMA_MAX_CONFIGS]; /* offset into blob_buffer table */
 989 +    uint32_t config_data[VMA_MAX_CONFIGS];  /* offset into blob_buffer table */
 990 +    uint32_t config_count;
 991 +};
 992 +
 993 +void vma_writer_set_error(VmaWriter *vmaw, const char *fmt, ...)
 994 +{
 995 +    va_list ap;
 996 +
 997 +    if (vmaw->status < 0) {
 998 +        return;
 999 +    }
1000 +
1001 +    vmaw->status = -1;
1002 +
1003 +    va_start(ap, fmt);
1004 +    g_vsnprintf(vmaw->errmsg, sizeof(vmaw->errmsg), fmt, ap);
1005 +    va_end(ap);
1006 +
1007 +    DPRINTF("vma_writer_set_error: %s\n", vmaw->errmsg);
1008 +}
1009 +
1010 +static uint32_t allocate_header_blob(VmaWriter *vmaw, const char *data,
1011 +                                     size_t len)
1012 +{
1013 +    if (len > 65535) {
1014 +        return 0;
1015 +    }
1016 +
1017 +    if (!vmaw->header_blob_table ||
1018 +        (vmaw->header_blob_table_size <
1019 +         (vmaw->header_blob_table_pos + len + 2))) {
1020 +        int newsize = vmaw->header_blob_table_size + ((len + 2 + 511)/512)*512;
1021 +
1022 +        vmaw->header_blob_table = g_realloc(vmaw->header_blob_table, newsize);
1023 +        memset(vmaw->header_blob_table + vmaw->header_blob_table_size,
1024 +               0, newsize - vmaw->header_blob_table_size);
1025 +        vmaw->header_blob_table_size = newsize;
1026 +    }
1027 +
1028 +    uint32_t cpos = vmaw->header_blob_table_pos;
1029 +    vmaw->header_blob_table[cpos] = len & 255;
1030 +    vmaw->header_blob_table[cpos+1] = (len >> 8) & 255;
1031 +    memcpy(vmaw->header_blob_table + cpos + 2, data, len);
1032 +    vmaw->header_blob_table_pos += len + 2;
1033 +    return cpos;
1034 +}
1035 +
1036 +static uint32_t allocate_header_string(VmaWriter *vmaw, const char *str)
1037 +{
1038 +    assert(vmaw);
1039 +
1040 +    size_t len = strlen(str) + 1;
1041 +
1042 +    return allocate_header_blob(vmaw, str, len);
1043 +}
1044 +
1045 +int vma_writer_add_config(VmaWriter *vmaw, const char *name, gpointer data,
1046 +                          gsize len)
1047 +{
1048 +    assert(vmaw);
1049 +    assert(!vmaw->header_written);
1050 +    assert(vmaw->config_count < VMA_MAX_CONFIGS);
1051 +    assert(name);
1052 +    assert(data);
1053 +
1054 +    gchar *basename = g_path_get_basename(name);
1055 +    uint32_t name_ptr = allocate_header_string(vmaw, basename);
1056 +    g_free(basename);
1057 +
1058 +    if (!name_ptr) {
1059 +        return -1;
1060 +    }
1061 +
1062 +    uint32_t data_ptr = allocate_header_blob(vmaw, data, len);
1063 +    if (!data_ptr) {
1064 +        return -1;
1065 +    }
1066 +
1067 +    vmaw->config_names[vmaw->config_count] = name_ptr;
1068 +    vmaw->config_data[vmaw->config_count] = data_ptr;
1069 +
1070 +    vmaw->config_count++;
1071 +
1072 +    return 0;
1073 +}
1074 +
1075 +int vma_writer_register_stream(VmaWriter *vmaw, const char *devname,
1076 +                               size_t size)
1077 +{
1078 +    assert(vmaw);
1079 +    assert(devname);
1080 +    assert(!vmaw->status);
1081 +
1082 +    if (vmaw->header_written) {
1083 +        vma_writer_set_error(vmaw, "vma_writer_register_stream: header "
1084 +                             "already written");
1085 +        return -1;
1086 +    }
1087 +
1088 +    guint n = vmaw->stream_count + 1;
1089 +
1090 +    /* we can have dev_ids form 1 to 255 (0 reserved)
1091 +     * 255(-1) reseverd for safety
1092 +     */
1093 +    if (n > 254) {
1094 +        vma_writer_set_error(vmaw, "vma_writer_register_stream: "
1095 +                             "too many drives");
1096 +        return -1;
1097 +    }
1098 +
1099 +    if (size <= 0) {
1100 +        vma_writer_set_error(vmaw, "vma_writer_register_stream: "
1101 +                             "got strange size %zd", size);
1102 +        return -1;
1103 +    }
1104 +
1105 +    DPRINTF("vma_writer_register_stream %s %zu %d\n", devname, size, n);
1106 +
1107 +    vmaw->stream_info[n].devname = g_strdup(devname);
1108 +    vmaw->stream_info[n].size = size;
1109 +
1110 +    vmaw->stream_info[n].cluster_count = (size + VMA_CLUSTER_SIZE - 1) /
1111 +        VMA_CLUSTER_SIZE;
1112 +
1113 +    vmaw->stream_count = n;
1114 +
1115 +    if (strcmp(devname, "vmstate") == 0) {
1116 +        vmaw->vmstate_stream = n;
1117 +    }
1118 +
1119 +    return n;
1120 +}
1121 +
1122 +static void coroutine_fn yield_until_fd_writable(int fd)
1123 +{
1124 +    assert(qemu_in_coroutine());
1125 +    AioContext *ctx = qemu_get_current_aio_context();
1126 +    aio_set_fd_handler(ctx, fd, false, NULL, (IOHandler *)qemu_coroutine_enter,
1127 +                       NULL, qemu_coroutine_self());
1128 +    qemu_coroutine_yield();
1129 +    aio_set_fd_handler(ctx, fd, false, NULL, NULL, NULL, NULL);
1130 +}
1131 +
1132 +static ssize_t coroutine_fn
1133 +vma_queue_write(VmaWriter *vmaw, const void *buf, size_t bytes)
1134 +{
1135 +    DPRINTF("vma_queue_write enter %zd\n", bytes);
1136 +
1137 +    assert(vmaw);
1138 +    assert(buf);
1139 +    assert(bytes <= VMA_MAX_EXTENT_SIZE);
1140 +
1141 +    size_t done = 0;
1142 +    ssize_t ret;
1143 +
1144 +    assert(vmaw->co_writer == NULL);
1145 +
1146 +    vmaw->co_writer = qemu_coroutine_self();
1147 +
1148 +    while (done < bytes) {
1149 +        if (vmaw->status < 0) {
1150 +            DPRINTF("vma_queue_write detected canceled backup\n");
1151 +            done = -1;
1152 +            break;
1153 +        }
1154 +        yield_until_fd_writable(vmaw->fd);
1155 +        ret = write(vmaw->fd, buf + done, bytes - done);
1156 +        if (ret > 0) {
1157 +            done += ret;
1158 +            DPRINTF("vma_queue_write written %zd %zd\n", done, ret);
1159 +        } else if (ret < 0) {
1160 +            if (errno == EAGAIN || errno == EWOULDBLOCK) {
1161 +                /* try again */
1162 +           } else {
1163 +                vma_writer_set_error(vmaw, "vma_queue_write: write error - %s",
1164 +                                     g_strerror(errno));
1165 +                done = -1; /* always return failure for partial writes */
1166 +                break;
1167 +            }
1168 +        } else if (ret == 0) {
1169 +            /* should not happen - simply try again */
1170 +        }
1171 +    }
1172 +
1173 +    vmaw->co_writer = NULL;
1174 +
1175 +    return (done == bytes) ? bytes : -1;
1176 +}
1177 +
1178 +VmaWriter *vma_writer_create(const char *filename, uuid_t uuid, Error **errp)
1179 +{
1180 +    const char *p;
1181 +
1182 +    assert(sizeof(VmaHeader) == (4096 + 8192));
1183 +    assert(G_STRUCT_OFFSET(VmaHeader, config_names) == 2044);
1184 +    assert(G_STRUCT_OFFSET(VmaHeader, config_data) == 3068);
1185 +    assert(G_STRUCT_OFFSET(VmaHeader, dev_info) == 4096);
1186 +    assert(sizeof(VmaExtentHeader) == 512);
1187 +
1188 +    VmaWriter *vmaw = g_new0(VmaWriter, 1);
1189 +    vmaw->fd = -1;
1190 +
1191 +    vmaw->md5csum = g_checksum_new(G_CHECKSUM_MD5);
1192 +    if (!vmaw->md5csum) {
1193 +        error_setg(errp, "can't allocate cmsum\n");
1194 +        goto err;
1195 +    }
1196 +
1197 +    if (strstart(filename, "exec:", &p)) {
1198 +        vmaw->cmd = popen(p, "w");
1199 +        if (vmaw->cmd == NULL) {
1200 +            error_setg(errp, "can't popen command '%s' - %s\n", p,
1201 +                       g_strerror(errno));
1202 +            goto err;
1203 +        }
1204 +        vmaw->fd = fileno(vmaw->cmd);
1205 +
1206 +        /* try to use O_NONBLOCK */
1207 +        fcntl(vmaw->fd, F_SETFL, fcntl(vmaw->fd, F_GETFL)|O_NONBLOCK);
1208 +
1209 +    } else {
1210 +        struct stat st;
1211 +        int oflags;
1212 +        const char *tmp_id_str;
1213 +
1214 +        if ((stat(filename, &st) == 0) && S_ISFIFO(st.st_mode)) {
1215 +            oflags = O_NONBLOCK|O_WRONLY;
1216 +            vmaw->fd = qemu_open(filename, oflags, 0644);
1217 +        } else if (strstart(filename, "/dev/fdset/", &tmp_id_str)) {
1218 +            oflags = O_NONBLOCK|O_WRONLY;
1219 +            vmaw->fd = qemu_open(filename, oflags, 0644);
1220 +        } else if (strstart(filename, "/dev/fdname/", &tmp_id_str)) {
1221 +            vmaw->fd = monitor_get_fd(cur_mon, tmp_id_str, errp);
1222 +            if (vmaw->fd < 0) {
1223 +                goto err;
1224 +            }
1225 +            /* try to use O_NONBLOCK */
1226 +            fcntl(vmaw->fd, F_SETFL, fcntl(vmaw->fd, F_GETFL)|O_NONBLOCK);
1227 +        } else  {
1228 +            oflags = O_NONBLOCK|O_DIRECT|O_WRONLY|O_CREAT|O_EXCL;
1229 +            vmaw->fd = qemu_open(filename, oflags, 0644);
1230 +        }
1231 +
1232 +        if (vmaw->fd < 0) {
1233 +            error_setg(errp, "can't open file %s - %s\n", filename,
1234 +                       g_strerror(errno));
1235 +            goto err;
1236 +        }
1237 +    }
1238 +
1239 +    /* we use O_DIRECT, so we need to align IO buffers */
1240 +
1241 +    vmaw->outbuf = qemu_memalign(512, VMA_MAX_EXTENT_SIZE);
1242 +    vmaw->headerbuf = qemu_memalign(512, HEADERBUF_SIZE);
1243 +
1244 +    vmaw->outbuf_count = 0;
1245 +    vmaw->outbuf_pos = VMA_EXTENT_HEADER_SIZE;
1246 +
1247 +    vmaw->header_blob_table_pos = 1; /* start at pos 1 */
1248 +
1249 +    qemu_co_mutex_init(&vmaw->flush_lock);
1250 +
1251 +    uuid_copy(vmaw->uuid, uuid);
1252 +
1253 +    return vmaw;
1254 +
1255 +err:
1256 +    if (vmaw) {
1257 +        if (vmaw->cmd) {
1258 +            pclose(vmaw->cmd);
1259 +        } else if (vmaw->fd >= 0) {
1260 +            close(vmaw->fd);
1261 +        }
1262 +
1263 +        if (vmaw->md5csum) {
1264 +            g_checksum_free(vmaw->md5csum);
1265 +        }
1266 +
1267 +        g_free(vmaw);
1268 +    }
1269 +
1270 +    return NULL;
1271 +}
1272 +
1273 +static int coroutine_fn vma_write_header(VmaWriter *vmaw)
1274 +{
1275 +    assert(vmaw);
1276 +    unsigned char *buf = vmaw->headerbuf;
1277 +    VmaHeader *head = (VmaHeader *)buf;
1278 +
1279 +    int i;
1280 +
1281 +    DPRINTF("VMA WRITE HEADER\n");
1282 +
1283 +    if (vmaw->status < 0) {
1284 +        return vmaw->status;
1285 +    }
1286 +
1287 +    memset(buf, 0, HEADERBUF_SIZE);
1288 +
1289 +    head->magic = VMA_MAGIC;
1290 +    head->version = GUINT32_TO_BE(1); /* v1 */
1291 +    memcpy(head->uuid, vmaw->uuid, 16);
1292 +
1293 +    time_t ctime = time(NULL);
1294 +    head->ctime = GUINT64_TO_BE(ctime);
1295 +
1296 +    for (i = 0; i < VMA_MAX_CONFIGS; i++) {
1297 +        head->config_names[i] = GUINT32_TO_BE(vmaw->config_names[i]);
1298 +        head->config_data[i] = GUINT32_TO_BE(vmaw->config_data[i]);
1299 +    }
1300 +
1301 +    /* 32 bytes per device (12 used currently) = 8192 bytes max */
1302 +    for (i = 1; i <= 254; i++) {
1303 +        VmaStreamInfo *si = &vmaw->stream_info[i];
1304 +        if (si->size) {
1305 +            assert(si->devname);
1306 +            uint32_t devname_ptr = allocate_header_string(vmaw, si->devname);
1307 +            if (!devname_ptr) {
1308 +                return -1;
1309 +            }
1310 +            head->dev_info[i].devname_ptr = GUINT32_TO_BE(devname_ptr);
1311 +            head->dev_info[i].size = GUINT64_TO_BE(si->size);
1312 +        }
1313 +    }
1314 +
1315 +    uint32_t header_size = sizeof(VmaHeader) + vmaw->header_blob_table_size;
1316 +    head->header_size = GUINT32_TO_BE(header_size);
1317 +
1318 +    if (header_size > HEADERBUF_SIZE) {
1319 +        return -1; /* just to be sure */
1320 +    }
1321 +
1322 +    uint32_t blob_buffer_offset = sizeof(VmaHeader);
1323 +    memcpy(buf + blob_buffer_offset, vmaw->header_blob_table,
1324 +           vmaw->header_blob_table_size);
1325 +    head->blob_buffer_offset = GUINT32_TO_BE(blob_buffer_offset);
1326 +    head->blob_buffer_size = GUINT32_TO_BE(vmaw->header_blob_table_pos);
1327 +
1328 +    g_checksum_reset(vmaw->md5csum);
1329 +    g_checksum_update(vmaw->md5csum, (const guchar *)buf, header_size);
1330 +    gsize csize = 16;
1331 +    g_checksum_get_digest(vmaw->md5csum, (guint8 *)(head->md5sum), &csize);
1332 +
1333 +    return vma_queue_write(vmaw, buf, header_size);
1334 +}
1335 +
1336 +static int coroutine_fn vma_writer_flush(VmaWriter *vmaw)
1337 +{
1338 +    assert(vmaw);
1339 +
1340 +    int ret;
1341 +    int i;
1342 +
1343 +    if (vmaw->status < 0) {
1344 +        return vmaw->status;
1345 +    }
1346 +
1347 +    if (!vmaw->header_written) {
1348 +        vmaw->header_written = true;
1349 +        ret = vma_write_header(vmaw);
1350 +        if (ret < 0) {
1351 +            vma_writer_set_error(vmaw, "vma_writer_flush: write header failed");
1352 +            return ret;
1353 +        }
1354 +    }
1355 +
1356 +    DPRINTF("VMA WRITE FLUSH %d %d\n", vmaw->outbuf_count, vmaw->outbuf_pos);
1357 +
1358 +
1359 +    VmaExtentHeader *ehead = (VmaExtentHeader *)vmaw->outbuf;
1360 +
1361 +    ehead->magic = VMA_EXTENT_MAGIC;
1362 +    ehead->reserved1 = 0;
1363 +
1364 +    for (i = 0; i < VMA_BLOCKS_PER_EXTENT; i++) {
1365 +        ehead->blockinfo[i] = GUINT64_TO_BE(vmaw->outbuf_block_info[i]);
1366 +    }
1367 +
1368 +    guint16 block_count = (vmaw->outbuf_pos - VMA_EXTENT_HEADER_SIZE) /
1369 +        VMA_BLOCK_SIZE;
1370 +
1371 +    ehead->block_count = GUINT16_TO_BE(block_count);
1372 +
1373 +    memcpy(ehead->uuid, vmaw->uuid, sizeof(ehead->uuid));
1374 +    memset(ehead->md5sum, 0, sizeof(ehead->md5sum));
1375 +
1376 +    g_checksum_reset(vmaw->md5csum);
1377 +    g_checksum_update(vmaw->md5csum, vmaw->outbuf, VMA_EXTENT_HEADER_SIZE);
1378 +    gsize csize = 16;
1379 +    g_checksum_get_digest(vmaw->md5csum, ehead->md5sum, &csize);
1380 +
1381 +    int bytes = vmaw->outbuf_pos;
1382 +    ret = vma_queue_write(vmaw, vmaw->outbuf, bytes);
1383 +    if (ret != bytes) {
1384 +        vma_writer_set_error(vmaw, "vma_writer_flush: failed write");
1385 +    }
1386 +
1387 +    vmaw->outbuf_count = 0;
1388 +    vmaw->outbuf_pos = VMA_EXTENT_HEADER_SIZE;
1389 +
1390 +    for (i = 0; i < VMA_BLOCKS_PER_EXTENT; i++) {
1391 +        vmaw->outbuf_block_info[i] = 0;
1392 +    }
1393 +
1394 +    return vmaw->status;
1395 +}
1396 +
1397 +static int vma_count_open_streams(VmaWriter *vmaw)
1398 +{
1399 +    g_assert(vmaw != NULL);
1400 +
1401 +    int i;
1402 +    int open_drives = 0;
1403 +    for (i = 0; i <= 255; i++) {
1404 +        if (vmaw->stream_info[i].size && !vmaw->stream_info[i].finished) {
1405 +            open_drives++;
1406 +        }
1407 +    }
1408 +
1409 +    return open_drives;
1410 +}
1411 +
1412 +
1413 +/**
1414 + * You need to call this if the vma archive does not contain
1415 + * any data stream.
1416 + */
1417 +int coroutine_fn
1418 +vma_writer_flush_output(VmaWriter *vmaw)
1419 +{
1420 +    qemu_co_mutex_lock(&vmaw->flush_lock);
1421 +    int ret = vma_writer_flush(vmaw);
1422 +    qemu_co_mutex_unlock(&vmaw->flush_lock);
1423 +    if (ret < 0) {
1424 +        vma_writer_set_error(vmaw, "vma_writer_flush_header failed");
1425 +    }
1426 +    return ret;
1427 +}
1428 +
1429 +/**
1430 + * all jobs should call this when there is no more data
1431 + * Returns: number of remaining stream (0 ==> finished)
1432 + */
1433 +int coroutine_fn
1434 +vma_writer_close_stream(VmaWriter *vmaw, uint8_t dev_id)
1435 +{
1436 +    g_assert(vmaw != NULL);
1437 +
1438 +    DPRINTF("vma_writer_set_status %d\n", dev_id);
1439 +    if (!vmaw->stream_info[dev_id].size) {
1440 +        vma_writer_set_error(vmaw, "vma_writer_close_stream: "
1441 +                             "no such stream %d", dev_id);
1442 +        return -1;
1443 +    }
1444 +    if (vmaw->stream_info[dev_id].finished) {
1445 +        vma_writer_set_error(vmaw, "vma_writer_close_stream: "
1446 +                             "stream already closed %d", dev_id);
1447 +        return -1;
1448 +    }
1449 +
1450 +    vmaw->stream_info[dev_id].finished = true;
1451 +
1452 +    int open_drives = vma_count_open_streams(vmaw);
1453 +
1454 +    if (open_drives <= 0) {
1455 +        DPRINTF("vma_writer_set_status all drives completed\n");
1456 +        vma_writer_flush_output(vmaw);
1457 +    }
1458 +
1459 +    return open_drives;
1460 +}
1461 +
1462 +int vma_writer_get_status(VmaWriter *vmaw, VmaStatus *status)
1463 +{
1464 +    int i;
1465 +
1466 +    g_assert(vmaw != NULL);
1467 +
1468 +    if (status) {
1469 +        status->status = vmaw->status;
1470 +        g_strlcpy(status->errmsg, vmaw->errmsg, sizeof(status->errmsg));
1471 +        for (i = 0; i <= 255; i++) {
1472 +            status->stream_info[i] = vmaw->stream_info[i];
1473 +        }
1474 +
1475 +        uuid_unparse_lower(vmaw->uuid, status->uuid_str);
1476 +    }
1477 +
1478 +    status->closed = vmaw->closed;
1479 +
1480 +    return vmaw->status;
1481 +}
1482 +
1483 +static int vma_writer_get_buffer(VmaWriter *vmaw)
1484 +{
1485 +    int ret = 0;
1486 +
1487 +    qemu_co_mutex_lock(&vmaw->flush_lock);
1488 +
1489 +    /* wait until buffer is available */
1490 +    while (vmaw->outbuf_count >= (VMA_BLOCKS_PER_EXTENT - 1)) {
1491 +        ret = vma_writer_flush(vmaw);
1492 +        if (ret < 0) {
1493 +            vma_writer_set_error(vmaw, "vma_writer_get_buffer: flush failed");
1494 +            break;
1495 +        }
1496 +    }
1497 +
1498 +    qemu_co_mutex_unlock(&vmaw->flush_lock);
1499 +
1500 +    return ret;
1501 +}
1502 +
1503 +
1504 +int64_t coroutine_fn
1505 +vma_writer_write(VmaWriter *vmaw, uint8_t dev_id, int64_t cluster_num,
1506 +                 const unsigned char *buf, size_t *zero_bytes)
1507 +{
1508 +    g_assert(vmaw != NULL);
1509 +    g_assert(zero_bytes != NULL);
1510 +
1511 +    *zero_bytes = 0;
1512 +
1513 +    if (vmaw->status < 0) {
1514 +        return vmaw->status;
1515 +    }
1516 +
1517 +    if (!dev_id || !vmaw->stream_info[dev_id].size) {
1518 +        vma_writer_set_error(vmaw, "vma_writer_write: "
1519 +                             "no such stream %d", dev_id);
1520 +        return -1;
1521 +    }
1522 +
1523 +    if (vmaw->stream_info[dev_id].finished) {
1524 +        vma_writer_set_error(vmaw, "vma_writer_write: "
1525 +                             "stream already closed %d", dev_id);
1526 +        return -1;
1527 +    }
1528 +
1529 +
1530 +    if (cluster_num >= (((uint64_t)1)<<32)) {
1531 +        vma_writer_set_error(vmaw, "vma_writer_write: "
1532 +                             "cluster number out of range");
1533 +        return -1;
1534 +    }
1535 +
1536 +    if (dev_id == vmaw->vmstate_stream) {
1537 +        if (cluster_num != vmaw->vmstate_clusters) {
1538 +            vma_writer_set_error(vmaw, "vma_writer_write: "
1539 +                                 "non sequential vmstate write");
1540 +        }
1541 +        vmaw->vmstate_clusters++;
1542 +    } else if (cluster_num >= vmaw->stream_info[dev_id].cluster_count) {
1543 +        vma_writer_set_error(vmaw, "vma_writer_write: cluster number too big");
1544 +        return -1;
1545 +    }
1546 +
1547 +    /* wait until buffer is available */
1548 +    if (vma_writer_get_buffer(vmaw) < 0) {
1549 +        vma_writer_set_error(vmaw, "vma_writer_write: "
1550 +                             "vma_writer_get_buffer failed");
1551 +        return -1;
1552 +    }
1553 +
1554 +    DPRINTF("VMA WRITE %d %zd\n", dev_id, cluster_num);
1555 +
1556 +    uint64_t dev_size = vmaw->stream_info[dev_id].size;
1557 +    uint16_t mask = 0;
1558 +
1559 +    if (buf) {
1560 +        int i;
1561 +        int bit = 1;
1562 +        uint64_t byte_offset = cluster_num * VMA_CLUSTER_SIZE;
1563 +        for (i = 0; i < 16; i++) {
1564 +            const unsigned char *vmablock = buf + (i*VMA_BLOCK_SIZE);
1565 +
1566 +            // Note: If the source is not 64k-aligned, we might reach 4k blocks
1567 +            // after the end of the device. Always mark these as zero in the
1568 +            // mask, so the restore handles them correctly.
1569 +            if (byte_offset < dev_size &&
1570 +                !buffer_is_zero(vmablock, VMA_BLOCK_SIZE))
1571 +            {
1572 +                mask |= bit;
1573 +                memcpy(vmaw->outbuf + vmaw->outbuf_pos, vmablock,
1574 +                       VMA_BLOCK_SIZE);
1575 +
1576 +                // prevent memory leakage on unaligned last block
1577 +                if (byte_offset + VMA_BLOCK_SIZE > dev_size) {
1578 +                    uint64_t real_data_in_block = dev_size - byte_offset;
1579 +                    memset(vmaw->outbuf + vmaw->outbuf_pos + real_data_in_block,
1580 +                           0, VMA_BLOCK_SIZE - real_data_in_block);
1581 +                }
1582 +
1583 +                vmaw->outbuf_pos += VMA_BLOCK_SIZE;
1584 +            } else {
1585 +                DPRINTF("VMA WRITE %zd ZERO BLOCK %d\n", cluster_num, i);
1586 +                vmaw->stream_info[dev_id].zero_bytes += VMA_BLOCK_SIZE;
1587 +                *zero_bytes += VMA_BLOCK_SIZE;
1588 +            }
1589 +
1590 +            byte_offset += VMA_BLOCK_SIZE;
1591 +            bit = bit << 1;
1592 +        }
1593 +    } else {
1594 +        DPRINTF("VMA WRITE %zd ZERO CLUSTER\n", cluster_num);
1595 +        vmaw->stream_info[dev_id].zero_bytes += VMA_CLUSTER_SIZE;
1596 +        *zero_bytes += VMA_CLUSTER_SIZE;
1597 +    }
1598 +
1599 +    uint64_t block_info = ((uint64_t)mask) << (32+16);
1600 +    block_info |= ((uint64_t)dev_id) << 32;
1601 +    block_info |= (cluster_num & 0xffffffff);
1602 +    vmaw->outbuf_block_info[vmaw->outbuf_count] = block_info;
1603 +
1604 +    DPRINTF("VMA WRITE MASK %zd %zx\n", cluster_num, block_info);
1605 +
1606 +    vmaw->outbuf_count++;
1607 +
1608 +    /** NOTE: We allways write whole clusters, but we correctly set
1609 +     * transferred bytes. So transferred == size when when everything
1610 +     * went OK.
1611 +     */
1612 +    size_t transferred = VMA_CLUSTER_SIZE;
1613 +
1614 +    if (dev_id != vmaw->vmstate_stream) {
1615 +        uint64_t last = (cluster_num + 1) * VMA_CLUSTER_SIZE;
1616 +        if (last > dev_size) {
1617 +            uint64_t diff = last - dev_size;
1618 +            if (diff >= VMA_CLUSTER_SIZE) {
1619 +                vma_writer_set_error(vmaw, "vma_writer_write: "
1620 +                                     "read after last cluster");
1621 +                return -1;
1622 +            }
1623 +            transferred -= diff;
1624 +        }
1625 +    }
1626 +
1627 +    vmaw->stream_info[dev_id].transferred += transferred;
1628 +
1629 +    return transferred;
1630 +}
1631 +
1632 +void vma_writer_error_propagate(VmaWriter *vmaw, Error **errp)
1633 +{
1634 +    if (vmaw->status < 0 && *errp == NULL) {
1635 +        error_setg(errp, "%s", vmaw->errmsg);
1636 +    }
1637 +}
1638 +
1639 +int vma_writer_close(VmaWriter *vmaw, Error **errp)
1640 +{
1641 +    g_assert(vmaw != NULL);
1642 +
1643 +    int i;
1644 +
1645 +    qemu_co_mutex_lock(&vmaw->flush_lock); // wait for pending writes
1646 +
1647 +    assert(vmaw->co_writer == NULL);
1648 +
1649 +    if (vmaw->cmd) {
1650 +        if (pclose(vmaw->cmd) < 0) {
1651 +            vma_writer_set_error(vmaw, "vma_writer_close: "
1652 +                                 "pclose failed - %s", g_strerror(errno));
1653 +        }
1654 +    } else {
1655 +        if (close(vmaw->fd) < 0) {
1656 +            vma_writer_set_error(vmaw, "vma_writer_close: "
1657 +                                 "close failed - %s", g_strerror(errno));
1658 +        }
1659 +    }
1660 +
1661 +    for (i = 0; i <= 255; i++) {
1662 +        VmaStreamInfo *si = &vmaw->stream_info[i];
1663 +        if (si->size) {
1664 +            if (!si->finished) {
1665 +                vma_writer_set_error(vmaw, "vma_writer_close: "
1666 +                                     "detected open stream '%s'", si->devname);
1667 +            } else if ((si->transferred != si->size) &&
1668 +                       (i != vmaw->vmstate_stream)) {
1669 +                vma_writer_set_error(vmaw, "vma_writer_close: "
1670 +                                     "incomplete stream '%s' (%zd != %zd)",
1671 +                                     si->devname, si->transferred, si->size);
1672 +            }
1673 +        }
1674 +    }
1675 +
1676 +    for (i = 0; i <= 255; i++) {
1677 +        vmaw->stream_info[i].finished = 1; /* mark as closed */
1678 +    }
1679 +
1680 +    vmaw->closed = 1;
1681 +
1682 +    if (vmaw->status < 0 && *errp == NULL) {
1683 +        error_setg(errp, "%s", vmaw->errmsg);
1684 +    }
1685 +
1686 +    qemu_co_mutex_unlock(&vmaw->flush_lock);
1687 +
1688 +    return vmaw->status;
1689 +}
1690 +
1691 +void vma_writer_destroy(VmaWriter *vmaw)
1692 +{
1693 +    assert(vmaw);
1694 +
1695 +    int i;
1696 +
1697 +    for (i = 0; i <= 255; i++) {
1698 +        if (vmaw->stream_info[i].devname) {
1699 +            g_free(vmaw->stream_info[i].devname);
1700 +        }
1701 +    }
1702 +
1703 +    if (vmaw->md5csum) {
1704 +        g_checksum_free(vmaw->md5csum);
1705 +    }
1706 +
1707 +    qemu_vfree(vmaw->headerbuf);
1708 +    qemu_vfree(vmaw->outbuf);
1709 +    g_free(vmaw);
1710 +}
1711 diff --git a/vma.c b/vma.c
1712 new file mode 100644
1713 index 0000000000..2eea2fc281
1714 --- /dev/null
1715 +++ b/vma.c
1716 @@ -0,0 +1,839 @@
1717 +/*
1718 + * VMA: Virtual Machine Archive
1719 + *
1720 + * Copyright (C) 2012-2013 Proxmox Server Solutions
1721 + *
1722 + * Authors:
1723 + *  Dietmar Maurer (dietmar@proxmox.com)
1724 + *
1725 + * This work is licensed under the terms of the GNU GPL, version 2 or later.
1726 + * See the COPYING file in the top-level directory.
1727 + *
1728 + */
1729 +
1730 +#include "qemu/osdep.h"
1731 +#include <glib.h>
1732 +
1733 +#include "vma.h"
1734 +#include "qemu-common.h"
1735 +#include "qemu/module.h"
1736 +#include "qemu/error-report.h"
1737 +#include "qemu/main-loop.h"
1738 +#include "qemu/cutils.h"
1739 +#include "qapi/qmp/qdict.h"
1740 +#include "sysemu/block-backend.h"
1741 +
1742 +static void help(void)
1743 +{
1744 +    const char *help_msg =
1745 +        "usage: vma command [command options]\n"
1746 +        "\n"
1747 +        "vma list <filename>\n"
1748 +        "vma config <filename> [-c config]\n"
1749 +        "vma create <filename> [-c config] pathname ...\n"
1750 +        "vma extract <filename> [-r <fifo>] <targetdir>\n"
1751 +        "vma verify <filename> [-v]\n"
1752 +        ;
1753 +
1754 +    printf("%s", help_msg);
1755 +    exit(1);
1756 +}
1757 +
1758 +static const char *extract_devname(const char *path, char **devname, int index)
1759 +{
1760 +    assert(path);
1761 +
1762 +    const char *sep = strchr(path, '=');
1763 +
1764 +    if (sep) {
1765 +        *devname = g_strndup(path, sep - path);
1766 +        path = sep + 1;
1767 +    } else {
1768 +        if (index >= 0) {
1769 +            *devname = g_strdup_printf("disk%d", index);
1770 +        } else {
1771 +            *devname = NULL;
1772 +        }
1773 +    }
1774 +
1775 +    return path;
1776 +}
1777 +
1778 +static void print_content(VmaReader *vmar)
1779 +{
1780 +    assert(vmar);
1781 +
1782 +    VmaHeader *head = vma_reader_get_header(vmar);
1783 +
1784 +    GList *l = vma_reader_get_config_data(vmar);
1785 +    while (l && l->data) {
1786 +        VmaConfigData *cdata = (VmaConfigData *)l->data;
1787 +        l = g_list_next(l);
1788 +        printf("CFG: size: %d name: %s\n", cdata->len, cdata->name);
1789 +    }
1790 +
1791 +    int i;
1792 +    VmaDeviceInfo *di;
1793 +    for (i = 1; i < 255; i++) {
1794 +        di = vma_reader_get_device_info(vmar, i);
1795 +        if (di) {
1796 +            if (strcmp(di->devname, "vmstate") == 0) {
1797 +                printf("VMSTATE: dev_id=%d memory: %zd\n", i, di->size);
1798 +            } else {
1799 +                printf("DEV: dev_id=%d size: %zd devname: %s\n",
1800 +                       i, di->size, di->devname);
1801 +            }
1802 +        }
1803 +    }
1804 +    /* ctime is the last entry we print */
1805 +    printf("CTIME: %s", ctime(&head->ctime));
1806 +    fflush(stdout);
1807 +}
1808 +
1809 +static int list_content(int argc, char **argv)
1810 +{
1811 +    int c, ret = 0;
1812 +    const char *filename;
1813 +
1814 +    for (;;) {
1815 +        c = getopt(argc, argv, "h");
1816 +        if (c == -1) {
1817 +            break;
1818 +        }
1819 +        switch (c) {
1820 +        case '?':
1821 +        case 'h':
1822 +            help();
1823 +            break;
1824 +        default:
1825 +            g_assert_not_reached();
1826 +        }
1827 +    }
1828 +
1829 +    /* Get the filename */
1830 +    if ((optind + 1) != argc) {
1831 +        help();
1832 +    }
1833 +    filename = argv[optind++];
1834 +
1835 +    Error *errp = NULL;
1836 +    VmaReader *vmar = vma_reader_create(filename, &errp);
1837 +
1838 +    if (!vmar) {
1839 +        g_error("%s", error_get_pretty(errp));
1840 +    }
1841 +
1842 +    print_content(vmar);
1843 +
1844 +    vma_reader_destroy(vmar);
1845 +
1846 +    return ret;
1847 +}
1848 +
1849 +typedef struct RestoreMap {
1850 +    char *devname;
1851 +    char *path;
1852 +    char *format;
1853 +    uint64_t throttling_bps;
1854 +    char *throttling_group;
1855 +    char *cache;
1856 +    bool write_zero;
1857 +} RestoreMap;
1858 +
1859 +static bool try_parse_option(char **line, const char *optname, char **out, const char *inbuf) {
1860 +    size_t optlen = strlen(optname);
1861 +    if (strncmp(*line, optname, optlen) != 0 || (*line)[optlen] != '=') {
1862 +        return false;
1863 +    }
1864 +    if (*out) {
1865 +        g_error("read map failed - duplicate value for option '%s'", optname);
1866 +    }
1867 +    char *value = (*line) + optlen + 1; /* including a '=' */
1868 +    char *colon = strchr(value, ':');
1869 +    if (!colon) {
1870 +        g_error("read map failed - option '%s' not terminated ('%s')",
1871 +                optname, inbuf);
1872 +    }
1873 +    *line = colon+1;
1874 +    *out = g_strndup(value, colon - value);
1875 +    return true;
1876 +}
1877 +
1878 +static uint64_t verify_u64(const char *text) {
1879 +    uint64_t value;
1880 +    const char *endptr = NULL;
1881 +    if (qemu_strtou64(text, &endptr, 0, &value) != 0 || !endptr || *endptr) {
1882 +        g_error("read map failed - not a number: %s", text);
1883 +    }
1884 +    return value;
1885 +}
1886 +
1887 +static int extract_content(int argc, char **argv)
1888 +{
1889 +    int c, ret = 0;
1890 +    int verbose = 0;
1891 +    const char *filename;
1892 +    const char *dirname;
1893 +    const char *readmap = NULL;
1894 +
1895 +    for (;;) {
1896 +        c = getopt(argc, argv, "hvr:");
1897 +        if (c == -1) {
1898 +            break;
1899 +        }
1900 +        switch (c) {
1901 +        case '?':
1902 +        case 'h':
1903 +            help();
1904 +            break;
1905 +        case 'r':
1906 +            readmap = optarg;
1907 +            break;
1908 +        case 'v':
1909 +            verbose = 1;
1910 +            break;
1911 +        default:
1912 +            help();
1913 +        }
1914 +    }
1915 +
1916 +    /* Get the filename */
1917 +    if ((optind + 2) != argc) {
1918 +        help();
1919 +    }
1920 +    filename = argv[optind++];
1921 +    dirname = argv[optind++];
1922 +
1923 +    Error *errp = NULL;
1924 +    VmaReader *vmar = vma_reader_create(filename, &errp);
1925 +
1926 +    if (!vmar) {
1927 +        g_error("%s", error_get_pretty(errp));
1928 +    }
1929 +
1930 +    if (mkdir(dirname, 0777) < 0) {
1931 +        g_error("unable to create target directory %s - %s",
1932 +                dirname, g_strerror(errno));
1933 +    }
1934 +
1935 +    GList *l = vma_reader_get_config_data(vmar);
1936 +    while (l && l->data) {
1937 +        VmaConfigData *cdata = (VmaConfigData *)l->data;
1938 +        l = g_list_next(l);
1939 +        char *cfgfn = g_strdup_printf("%s/%s", dirname, cdata->name);
1940 +        GError *err = NULL;
1941 +        if (!g_file_set_contents(cfgfn, (gchar *)cdata->data, cdata->len,
1942 +                                 &err)) {
1943 +            g_error("unable to write file: %s", err->message);
1944 +        }
1945 +    }
1946 +
1947 +    GHashTable *devmap = g_hash_table_new(g_str_hash, g_str_equal);
1948 +
1949 +    if (readmap) {
1950 +        print_content(vmar);
1951 +
1952 +        FILE *map = fopen(readmap, "r");
1953 +        if (!map) {
1954 +            g_error("unable to open fifo %s - %s", readmap, g_strerror(errno));
1955 +        }
1956 +
1957 +        while (1) {
1958 +            char inbuf[8192];
1959 +            char *line = fgets(inbuf, sizeof(inbuf), map);
1960 +            char *format = NULL;
1961 +            char *bps = NULL;
1962 +            char *group = NULL;
1963 +            char *cache = NULL;
1964 +            if (!line || line[0] == '\0' || !strcmp(line, "done\n")) {
1965 +                break;
1966 +            }
1967 +            int len = strlen(line);
1968 +            if (line[len - 1] == '\n') {
1969 +                line[len - 1] = '\0';
1970 +                if (len == 1) {
1971 +                    break;
1972 +                }
1973 +            }
1974 +
1975 +            while (1) {
1976 +                if (!try_parse_option(&line, "format", &format, inbuf) &&
1977 +                    !try_parse_option(&line, "throttling.bps", &bps, inbuf) &&
1978 +                    !try_parse_option(&line, "throttling.group", &group, inbuf) &&
1979 +                    !try_parse_option(&line, "cache", &cache, inbuf))
1980 +                {
1981 +                    break;
1982 +                }
1983 +            }
1984 +
1985 +            uint64_t bps_value = 0;
1986 +            if (bps) {
1987 +                bps_value = verify_u64(bps);
1988 +                g_free(bps);
1989 +            }
1990 +
1991 +            const char *path;
1992 +            bool write_zero;
1993 +            if (line[0] == '0' && line[1] == ':') {
1994 +                path = line + 2;
1995 +                write_zero = false;
1996 +            } else if (line[0] == '1' && line[1] == ':') {
1997 +                path = line + 2;
1998 +                write_zero = true;
1999 +            } else {
2000 +                g_error("read map failed - parse error ('%s')", inbuf);
2001 +            }
2002 +
2003 +            char *devname = NULL;
2004 +            path = extract_devname(path, &devname, -1);
2005 +            if (!devname) {
2006 +                g_error("read map failed - no dev name specified ('%s')",
2007 +                        inbuf);
2008 +            }
2009 +
2010 +            RestoreMap *map = g_new0(RestoreMap, 1);
2011 +            map->devname = g_strdup(devname);
2012 +            map->path = g_strdup(path);
2013 +            map->format = format;
2014 +            map->throttling_bps = bps_value;
2015 +            map->throttling_group = group;
2016 +            map->cache = cache;
2017 +            map->write_zero = write_zero;
2018 +
2019 +            g_hash_table_insert(devmap, map->devname, map);
2020 +
2021 +        };
2022 +    }
2023 +
2024 +    int i;
2025 +    int vmstate_fd = -1;
2026 +    guint8 vmstate_stream = 0;
2027 +
2028 +    BlockBackend *blk = NULL;
2029 +
2030 +    for (i = 1; i < 255; i++) {
2031 +        VmaDeviceInfo *di = vma_reader_get_device_info(vmar, i);
2032 +        if (di && (strcmp(di->devname, "vmstate") == 0)) {
2033 +            vmstate_stream = i;
2034 +            char *statefn = g_strdup_printf("%s/vmstate.bin", dirname);
2035 +            vmstate_fd = open(statefn, O_WRONLY|O_CREAT|O_EXCL, 0644);
2036 +            if (vmstate_fd < 0) {
2037 +                g_error("create vmstate file '%s' failed - %s", statefn,
2038 +                        g_strerror(errno));
2039 +            }
2040 +            g_free(statefn);
2041 +        } else if (di) {
2042 +            char *devfn = NULL;
2043 +            const char *format = NULL;
2044 +            uint64_t throttling_bps = 0;
2045 +            const char *throttling_group = NULL;
2046 +            const char *cache = NULL;
2047 +            int flags = BDRV_O_RDWR;
2048 +            bool write_zero = true;
2049 +
2050 +            if (readmap) {
2051 +                RestoreMap *map;
2052 +                map = (RestoreMap *)g_hash_table_lookup(devmap, di->devname);
2053 +                if (map == NULL) {
2054 +                    g_error("no device name mapping for %s", di->devname);
2055 +                }
2056 +                devfn = map->path;
2057 +                format = map->format;
2058 +                throttling_bps = map->throttling_bps;
2059 +                throttling_group = map->throttling_group;
2060 +                cache = map->cache;
2061 +                write_zero = map->write_zero;
2062 +            } else {
2063 +                devfn = g_strdup_printf("%s/tmp-disk-%s.raw",
2064 +                                        dirname, di->devname);
2065 +                printf("DEVINFO %s %zd\n", devfn, di->size);
2066 +
2067 +                bdrv_img_create(devfn, "raw", NULL, NULL, NULL, di->size,
2068 +                                flags, true, &errp);
2069 +                if (errp) {
2070 +                    g_error("can't create file %s: %s", devfn,
2071 +                            error_get_pretty(errp));
2072 +                }
2073 +
2074 +                /* Note: we created an empty file above, so there is no
2075 +                 * need to write zeroes (so we generate a sparse file)
2076 +                 */
2077 +                write_zero = false;
2078 +            }
2079 +
2080 +           size_t devlen = strlen(devfn);
2081 +           QDict *options = NULL;
2082 +            bool writethrough;
2083 +            if (format) {
2084 +                /* explicit format from commandline */
2085 +                options = qdict_new();
2086 +                qdict_put_str(options, "driver", format);
2087 +            } else if ((devlen > 4 && strcmp(devfn+devlen-4, ".raw") == 0) ||
2088 +                      strncmp(devfn, "/dev/", 5) == 0)
2089 +           {
2090 +                /* This part is now deprecated for PVE as well (just as qemu
2091 +                 * deprecated not specifying an explicit raw format, too.
2092 +                 */
2093 +               /* explicit raw format */
2094 +               options = qdict_new();
2095 +               qdict_put_str(options, "driver", "raw");
2096 +           }
2097 +            if (cache && bdrv_parse_cache_mode(cache, &flags, &writethrough)) {
2098 +                g_error("invalid cache option: %s\n", cache);
2099 +            }
2100 +
2101 +           if (errp || !(blk = blk_new_open(devfn, NULL, options, flags, &errp))) {
2102 +                g_error("can't open file %s - %s", devfn,
2103 +                        error_get_pretty(errp));
2104 +            }
2105 +
2106 +            if (cache) {
2107 +                blk_set_enable_write_cache(blk, !writethrough);
2108 +            }
2109 +
2110 +            if (throttling_group) {
2111 +                blk_io_limits_enable(blk, throttling_group);
2112 +            }
2113 +
2114 +            if (throttling_bps) {
2115 +                if (!throttling_group) {
2116 +                    blk_io_limits_enable(blk, devfn);
2117 +                }
2118 +
2119 +                ThrottleConfig cfg;
2120 +                throttle_config_init(&cfg);
2121 +                cfg.buckets[THROTTLE_BPS_WRITE].avg = throttling_bps;
2122 +                Error *err = NULL;
2123 +                if (!throttle_is_valid(&cfg, &err)) {
2124 +                    error_report_err(err);
2125 +                    g_error("failed to apply throttling");
2126 +                }
2127 +                blk_set_io_limits(blk, &cfg);
2128 +            }
2129 +
2130 +            if (vma_reader_register_bs(vmar, i, blk, write_zero, &errp) < 0) {
2131 +                g_error("%s", error_get_pretty(errp));
2132 +            }
2133 +
2134 +            if (!readmap) {
2135 +                g_free(devfn);
2136 +            }
2137 +        }
2138 +    }
2139 +
2140 +    if (vma_reader_restore(vmar, vmstate_fd, verbose, &errp) < 0) {
2141 +        g_error("restore failed - %s", error_get_pretty(errp));
2142 +    }
2143 +
2144 +    if (!readmap) {
2145 +        for (i = 1; i < 255; i++) {
2146 +            VmaDeviceInfo *di = vma_reader_get_device_info(vmar, i);
2147 +            if (di && (i != vmstate_stream)) {
2148 +                char *tmpfn = g_strdup_printf("%s/tmp-disk-%s.raw",
2149 +                                              dirname, di->devname);
2150 +                char *fn = g_strdup_printf("%s/disk-%s.raw",
2151 +                                           dirname, di->devname);
2152 +                if (rename(tmpfn, fn) != 0) {
2153 +                    g_error("rename %s to %s failed - %s",
2154 +                            tmpfn, fn, g_strerror(errno));
2155 +                }
2156 +            }
2157 +        }
2158 +    }
2159 +
2160 +    vma_reader_destroy(vmar);
2161 +
2162 +    blk_unref(blk);
2163 +
2164 +    bdrv_close_all();
2165 +
2166 +    return ret;
2167 +}
2168 +
2169 +static int verify_content(int argc, char **argv)
2170 +{
2171 +    int c, ret = 0;
2172 +    int verbose = 0;
2173 +    const char *filename;
2174 +
2175 +    for (;;) {
2176 +        c = getopt(argc, argv, "hv");
2177 +        if (c == -1) {
2178 +            break;
2179 +        }
2180 +        switch (c) {
2181 +        case '?':
2182 +        case 'h':
2183 +            help();
2184 +            break;
2185 +        case 'v':
2186 +            verbose = 1;
2187 +            break;
2188 +        default:
2189 +            help();
2190 +        }
2191 +    }
2192 +
2193 +    /* Get the filename */
2194 +    if ((optind + 1) != argc) {
2195 +        help();
2196 +    }
2197 +    filename = argv[optind++];
2198 +
2199 +    Error *errp = NULL;
2200 +    VmaReader *vmar = vma_reader_create(filename, &errp);
2201 +
2202 +    if (!vmar) {
2203 +        g_error("%s", error_get_pretty(errp));
2204 +    }
2205 +
2206 +    if (verbose) {
2207 +        print_content(vmar);
2208 +    }
2209 +
2210 +    if (vma_reader_verify(vmar, verbose, &errp) < 0) {
2211 +        g_error("verify failed - %s", error_get_pretty(errp));
2212 +    }
2213 +
2214 +    vma_reader_destroy(vmar);
2215 +
2216 +    bdrv_close_all();
2217 +
2218 +    return ret;
2219 +}
2220 +
2221 +typedef struct BackupJob {
2222 +    BlockBackend *target;
2223 +    int64_t len;
2224 +    VmaWriter *vmaw;
2225 +    uint8_t dev_id;
2226 +} BackupJob;
2227 +
2228 +#define BACKUP_SECTORS_PER_CLUSTER (VMA_CLUSTER_SIZE / BDRV_SECTOR_SIZE)
2229 +
2230 +static void coroutine_fn backup_run_empty(void *opaque)
2231 +{
2232 +    VmaWriter *vmaw = (VmaWriter *)opaque;
2233 +
2234 +    vma_writer_flush_output(vmaw);
2235 +
2236 +    Error *err = NULL;
2237 +    if (vma_writer_close(vmaw, &err) != 0) {
2238 +        g_warning("vma_writer_close failed %s", error_get_pretty(err));
2239 +    }
2240 +}
2241 +
2242 +static void coroutine_fn backup_run(void *opaque)
2243 +{
2244 +    BackupJob *job = (BackupJob *)opaque;
2245 +    struct iovec iov;
2246 +    QEMUIOVector qiov;
2247 +
2248 +    int64_t start, end;
2249 +    int ret = 0;
2250 +
2251 +    unsigned char *buf = blk_blockalign(job->target, VMA_CLUSTER_SIZE);
2252 +
2253 +    start = 0;
2254 +    end = DIV_ROUND_UP(job->len / BDRV_SECTOR_SIZE,
2255 +                       BACKUP_SECTORS_PER_CLUSTER);
2256 +
2257 +    for (; start < end; start++) {
2258 +        iov.iov_base = buf;
2259 +        iov.iov_len = VMA_CLUSTER_SIZE;
2260 +        qemu_iovec_init_external(&qiov, &iov, 1);
2261 +
2262 +        ret = blk_co_preadv(job->target, start * VMA_CLUSTER_SIZE,
2263 +                            VMA_CLUSTER_SIZE, &qiov, 0);
2264 +        if (ret < 0) {
2265 +            vma_writer_set_error(job->vmaw, "read error", -1);
2266 +            goto out;
2267 +        }
2268 +
2269 +        size_t zb = 0;
2270 +        if (vma_writer_write(job->vmaw, job->dev_id, start, buf, &zb) < 0) {
2271 +            vma_writer_set_error(job->vmaw, "backup_dump_cb vma_writer_write failed", -1);
2272 +            goto out;
2273 +        }
2274 +    }
2275 +
2276 +
2277 +out:
2278 +    if (vma_writer_close_stream(job->vmaw, job->dev_id) <= 0) {
2279 +        Error *err = NULL;
2280 +        if (vma_writer_close(job->vmaw, &err) != 0) {
2281 +            g_warning("vma_writer_close failed %s", error_get_pretty(err));
2282 +        }
2283 +    }
2284 +    qemu_vfree(buf);
2285 +}
2286 +
2287 +static int create_archive(int argc, char **argv)
2288 +{
2289 +    int i, c;
2290 +    int verbose = 0;
2291 +    const char *archivename;
2292 +    GList *config_files = NULL;
2293 +
2294 +    for (;;) {
2295 +        c = getopt(argc, argv, "hvc:");
2296 +        if (c == -1) {
2297 +            break;
2298 +        }
2299 +        switch (c) {
2300 +        case '?':
2301 +        case 'h':
2302 +            help();
2303 +            break;
2304 +        case 'c':
2305 +            config_files = g_list_append(config_files, optarg);
2306 +            break;
2307 +        case 'v':
2308 +            verbose = 1;
2309 +            break;
2310 +        default:
2311 +            g_assert_not_reached();
2312 +        }
2313 +    }
2314 +
2315 +
2316 +    /* make sure we an archive name */
2317 +    if ((optind + 1) > argc) {
2318 +        help();
2319 +    }
2320 +
2321 +    archivename = argv[optind++];
2322 +
2323 +    uuid_t uuid;
2324 +    uuid_generate(uuid);
2325 +
2326 +    Error *local_err = NULL;
2327 +    VmaWriter *vmaw = vma_writer_create(archivename, uuid, &local_err);
2328 +
2329 +    if (vmaw == NULL) {
2330 +        g_error("%s", error_get_pretty(local_err));
2331 +    }
2332 +
2333 +    GList *l = config_files;
2334 +    while (l && l->data) {
2335 +        char *name = l->data;
2336 +        char *cdata = NULL;
2337 +        gsize clen = 0;
2338 +        GError *err = NULL;
2339 +        if (!g_file_get_contents(name, &cdata, &clen, &err)) {
2340 +            unlink(archivename);
2341 +            g_error("Unable to read file: %s", err->message);
2342 +        }
2343 +
2344 +        if (vma_writer_add_config(vmaw, name, cdata, clen) != 0) {
2345 +            unlink(archivename);
2346 +            g_error("Unable to append config data %s (len = %zd)",
2347 +                    name, clen);
2348 +        }
2349 +        l = g_list_next(l);
2350 +    }
2351 +
2352 +    int devcount = 0;
2353 +    while (optind < argc) {
2354 +        const char *path = argv[optind++];
2355 +        char *devname = NULL;
2356 +        path = extract_devname(path, &devname, devcount++);
2357 +
2358 +        Error *errp = NULL;
2359 +        BlockBackend *target;
2360 +
2361 +        target = blk_new_open(path, NULL, NULL, 0, &errp);
2362 +        if (!target) {
2363 +            unlink(archivename);
2364 +            g_error("bdrv_open '%s' failed - %s", path, error_get_pretty(errp));
2365 +        }
2366 +        int64_t size = blk_getlength(target);
2367 +        int dev_id = vma_writer_register_stream(vmaw, devname, size);
2368 +        if (dev_id <= 0) {
2369 +            unlink(archivename);
2370 +            g_error("vma_writer_register_stream '%s' failed", devname);
2371 +        }
2372 +
2373 +        BackupJob *job = g_new0(BackupJob, 1);
2374 +        job->len = size;
2375 +        job->target = target;
2376 +        job->vmaw = vmaw;
2377 +        job->dev_id = dev_id;
2378 +
2379 +        Coroutine *co = qemu_coroutine_create(backup_run, job);
2380 +        qemu_coroutine_enter(co);
2381 +    }
2382 +
2383 +    VmaStatus vmastat;
2384 +    int percent = 0;
2385 +    int last_percent = -1;
2386 +
2387 +    if (devcount) {
2388 +        while (1) {
2389 +            main_loop_wait(false);
2390 +            vma_writer_get_status(vmaw, &vmastat);
2391 +
2392 +            if (verbose) {
2393 +
2394 +                uint64_t total = 0;
2395 +                uint64_t transferred = 0;
2396 +                uint64_t zero_bytes = 0;
2397 +
2398 +                int i;
2399 +                for (i = 0; i < 256; i++) {
2400 +                    if (vmastat.stream_info[i].size) {
2401 +                        total += vmastat.stream_info[i].size;
2402 +                        transferred += vmastat.stream_info[i].transferred;
2403 +                        zero_bytes += vmastat.stream_info[i].zero_bytes;
2404 +                    }
2405 +                }
2406 +                percent = (transferred*100)/total;
2407 +                if (percent != last_percent) {
2408 +                    fprintf(stderr, "progress %d%% %zd/%zd %zd\n", percent,
2409 +                            transferred, total, zero_bytes);
2410 +                    fflush(stderr);
2411 +
2412 +                    last_percent = percent;
2413 +                }
2414 +            }
2415 +
2416 +            if (vmastat.closed) {
2417 +                break;
2418 +            }
2419 +        }
2420 +    } else {
2421 +        Coroutine *co = qemu_coroutine_create(backup_run_empty, vmaw);
2422 +        qemu_coroutine_enter(co);
2423 +        while (1) {
2424 +            main_loop_wait(false);
2425 +            vma_writer_get_status(vmaw, &vmastat);
2426 +            if (vmastat.closed) {
2427 +                    break;
2428 +            }
2429 +        }
2430 +    }
2431 +
2432 +    bdrv_drain_all();
2433 +
2434 +    vma_writer_get_status(vmaw, &vmastat);
2435 +
2436 +    if (verbose) {
2437 +        for (i = 0; i < 256; i++) {
2438 +            VmaStreamInfo *si = &vmastat.stream_info[i];
2439 +            if (si->size) {
2440 +                fprintf(stderr, "image %s: size=%zd zeros=%zd saved=%zd\n",
2441 +                        si->devname, si->size, si->zero_bytes,
2442 +                        si->size - si->zero_bytes);
2443 +            }
2444 +        }
2445 +    }
2446 +
2447 +    if (vmastat.status < 0) {
2448 +        unlink(archivename);
2449 +        g_error("creating vma archive failed");
2450 +    }
2451 +
2452 +    vma_writer_destroy(vmaw);
2453 +    return 0;
2454 +}
2455 +
2456 +static int dump_config(int argc, char **argv)
2457 +{
2458 +    int c, ret = 0;
2459 +    const char *filename;
2460 +    const char *config_name = "qemu-server.conf";
2461 +
2462 +    for (;;) {
2463 +        c = getopt(argc, argv, "hc:");
2464 +        if (c == -1) {
2465 +            break;
2466 +        }
2467 +        switch (c) {
2468 +        case '?':
2469 +        case 'h':
2470 +            help();
2471 +            break;
2472 +        case 'c':
2473 +            config_name = optarg;
2474 +            break;
2475 +        default:
2476 +            help();
2477 +        }
2478 +    }
2479 +
2480 +    /* Get the filename */
2481 +    if ((optind + 1) != argc) {
2482 +        help();
2483 +    }
2484 +    filename = argv[optind++];
2485 +
2486 +    Error *errp = NULL;
2487 +    VmaReader *vmar = vma_reader_create(filename, &errp);
2488 +
2489 +    if (!vmar) {
2490 +        g_error("%s", error_get_pretty(errp));
2491 +    }
2492 +
2493 +    int found = 0;
2494 +    GList *l = vma_reader_get_config_data(vmar);
2495 +    while (l && l->data) {
2496 +        VmaConfigData *cdata = (VmaConfigData *)l->data;
2497 +        l = g_list_next(l);
2498 +        if (strcmp(cdata->name, config_name) == 0) {
2499 +            found = 1;
2500 +            fwrite(cdata->data,  cdata->len, 1, stdout);
2501 +            break;
2502 +        }
2503 +    }
2504 +
2505 +    vma_reader_destroy(vmar);
2506 +
2507 +    bdrv_close_all();
2508 +
2509 +    if (!found) {
2510 +        fprintf(stderr, "unable to find configuration data '%s'\n", config_name);
2511 +        return -1;
2512 +    }
2513 +
2514 +    return ret;
2515 +}
2516 +
2517 +int main(int argc, char **argv)
2518 +{
2519 +    const char *cmdname;
2520 +    Error *main_loop_err = NULL;
2521 +
2522 +    error_init(argv[0]);
2523 +    module_call_init(MODULE_INIT_TRACE);
2524 +    qemu_init_exec_dir(argv[0]);
2525 +
2526 +    if (qemu_init_main_loop(&main_loop_err)) {
2527 +        g_error("%s", error_get_pretty(main_loop_err));
2528 +    }
2529 +
2530 +    bdrv_init();
2531 +    module_call_init(MODULE_INIT_QOM);
2532 +
2533 +    if (argc < 2) {
2534 +        help();
2535 +    }
2536 +
2537 +    cmdname = argv[1];
2538 +    argc--; argv++;
2539 +
2540 +
2541 +    if (!strcmp(cmdname, "list")) {
2542 +        return list_content(argc, argv);
2543 +    } else if (!strcmp(cmdname, "create")) {
2544 +        return create_archive(argc, argv);
2545 +    } else if (!strcmp(cmdname, "extract")) {
2546 +        return extract_content(argc, argv);
2547 +    } else if (!strcmp(cmdname, "verify")) {
2548 +        return verify_content(argc, argv);
2549 +    } else if (!strcmp(cmdname, "config")) {
2550 +        return dump_config(argc, argv);
2551 +    }
2552 +
2553 +    help();
2554 +    return 0;
2555 +}
2556 diff --git a/vma.h b/vma.h
2557 new file mode 100644
2558 index 0000000000..c895c97f6d
2559 --- /dev/null
2560 +++ b/vma.h
2561 @@ -0,0 +1,150 @@
2562 +/*
2563 + * VMA: Virtual Machine Archive
2564 + *
2565 + * Copyright (C) Proxmox Server Solutions
2566 + *
2567 + * Authors:
2568 + *  Dietmar Maurer (dietmar@proxmox.com)
2569 + *
2570 + * This work is licensed under the terms of the GNU GPL, version 2 or later.
2571 + * See the COPYING file in the top-level directory.
2572 + *
2573 + */
2574 +
2575 +#ifndef BACKUP_VMA_H
2576 +#define BACKUP_VMA_H
2577 +
2578 +#include <uuid/uuid.h>
2579 +#include "qapi/error.h"
2580 +#include "block/block.h"
2581 +
2582 +#define VMA_BLOCK_BITS 12
2583 +#define VMA_BLOCK_SIZE (1<<VMA_BLOCK_BITS)
2584 +#define VMA_CLUSTER_BITS (VMA_BLOCK_BITS+4)
2585 +#define VMA_CLUSTER_SIZE (1<<VMA_CLUSTER_BITS)
2586 +
2587 +#if VMA_CLUSTER_SIZE != 65536
2588 +#error unexpected cluster size
2589 +#endif
2590 +
2591 +#define VMA_EXTENT_HEADER_SIZE 512
2592 +#define VMA_BLOCKS_PER_EXTENT 59
2593 +#define VMA_MAX_CONFIGS 256
2594 +
2595 +#define VMA_MAX_EXTENT_SIZE \
2596 +    (VMA_EXTENT_HEADER_SIZE+VMA_CLUSTER_SIZE*VMA_BLOCKS_PER_EXTENT)
2597 +#if VMA_MAX_EXTENT_SIZE != 3867136
2598 +#error unexpected VMA_EXTENT_SIZE
2599 +#endif
2600 +
2601 +/* File Format Definitions */
2602 +
2603 +#define VMA_MAGIC (GUINT32_TO_BE(('V'<<24)|('M'<<16)|('A'<<8)|0x00))
2604 +#define VMA_EXTENT_MAGIC (GUINT32_TO_BE(('V'<<24)|('M'<<16)|('A'<<8)|'E'))
2605 +
2606 +typedef struct VmaDeviceInfoHeader {
2607 +    uint32_t devname_ptr; /* offset into blob_buffer table */
2608 +    uint32_t reserved0;
2609 +    uint64_t size; /* device size in bytes */
2610 +    uint64_t reserved1;
2611 +    uint64_t reserved2;
2612 +} VmaDeviceInfoHeader;
2613 +
2614 +typedef struct VmaHeader {
2615 +    uint32_t magic;
2616 +    uint32_t version;
2617 +    unsigned char uuid[16];
2618 +    int64_t ctime;
2619 +    unsigned char md5sum[16];
2620 +
2621 +    uint32_t blob_buffer_offset;
2622 +    uint32_t blob_buffer_size;
2623 +    uint32_t header_size;
2624 +
2625 +    unsigned char reserved[1984];
2626 +
2627 +    uint32_t config_names[VMA_MAX_CONFIGS]; /* offset into blob_buffer table */
2628 +    uint32_t config_data[VMA_MAX_CONFIGS];  /* offset into blob_buffer table */
2629 +
2630 +    uint32_t reserved1;
2631 +
2632 +    VmaDeviceInfoHeader dev_info[256];
2633 +} VmaHeader;
2634 +
2635 +typedef struct VmaExtentHeader {
2636 +    uint32_t magic;
2637 +    uint16_t reserved1;
2638 +    uint16_t block_count;
2639 +    unsigned char uuid[16];
2640 +    unsigned char md5sum[16];
2641 +    uint64_t blockinfo[VMA_BLOCKS_PER_EXTENT];
2642 +} VmaExtentHeader;
2643 +
2644 +/* functions/definitions to read/write vma files */
2645 +
2646 +typedef struct VmaReader VmaReader;
2647 +
2648 +typedef struct VmaWriter VmaWriter;
2649 +
2650 +typedef struct VmaConfigData {
2651 +    const char *name;
2652 +    const void *data;
2653 +    uint32_t len;
2654 +} VmaConfigData;
2655 +
2656 +typedef struct VmaStreamInfo {
2657 +    uint64_t size;
2658 +    uint64_t cluster_count;
2659 +    uint64_t transferred;
2660 +    uint64_t zero_bytes;
2661 +    int finished;
2662 +    char *devname;
2663 +} VmaStreamInfo;
2664 +
2665 +typedef struct VmaStatus {
2666 +    int status;
2667 +    bool closed;
2668 +    char errmsg[8192];
2669 +    char uuid_str[37];
2670 +    VmaStreamInfo stream_info[256];
2671 +} VmaStatus;
2672 +
2673 +typedef struct VmaDeviceInfo {
2674 +    uint64_t size; /* device size in bytes */
2675 +    const char *devname;
2676 +} VmaDeviceInfo;
2677 +
2678 +VmaWriter *vma_writer_create(const char *filename, uuid_t uuid, Error **errp);
2679 +int vma_writer_close(VmaWriter *vmaw, Error **errp);
2680 +void vma_writer_error_propagate(VmaWriter *vmaw, Error **errp);
2681 +void vma_writer_destroy(VmaWriter *vmaw);
2682 +int vma_writer_add_config(VmaWriter *vmaw, const char *name, gpointer data,
2683 +                          size_t len);
2684 +int vma_writer_register_stream(VmaWriter *vmaw, const char *devname,
2685 +                               size_t size);
2686 +
2687 +int64_t coroutine_fn vma_writer_write(VmaWriter *vmaw, uint8_t dev_id,
2688 +                                      int64_t cluster_num,
2689 +                                      const unsigned char *buf,
2690 +                                      size_t *zero_bytes);
2691 +
2692 +int coroutine_fn vma_writer_close_stream(VmaWriter *vmaw, uint8_t dev_id);
2693 +int coroutine_fn vma_writer_flush_output(VmaWriter *vmaw);
2694 +
2695 +int vma_writer_get_status(VmaWriter *vmaw, VmaStatus *status);
2696 +void vma_writer_set_error(VmaWriter *vmaw, const char *fmt, ...);
2697 +
2698 +
2699 +VmaReader *vma_reader_create(const char *filename, Error **errp);
2700 +void vma_reader_destroy(VmaReader *vmar);
2701 +VmaHeader *vma_reader_get_header(VmaReader *vmar);
2702 +GList *vma_reader_get_config_data(VmaReader *vmar);
2703 +VmaDeviceInfo *vma_reader_get_device_info(VmaReader *vmar, guint8 dev_id);
2704 +int vma_reader_register_bs(VmaReader *vmar, guint8 dev_id,
2705 +                           BlockBackend *target, bool write_zeroes,
2706 +                           Error **errp);
2707 +int vma_reader_restore(VmaReader *vmar, int vmstate_fd, bool verbose,
2708 +                       Error **errp);
2709 +int vma_reader_verify(VmaReader *vmar, bool verbose, Error **errp);
2710 +
2711 +#endif /* BACKUP_VMA_H */