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

   1 From 6289a43696ca6f713a5d3bb9f95a5adb608a5e13 Mon Sep 17 00:00:00 2001
   2 From: Dietmar Maurer <dietmar@proxmox.com>
   3 Date: Tue, 13 Nov 2012 11:11:38 +0100
   4 Subject: [PATCH v5 4/6] introduce new vma archive format
   5
   6 This is a very simple archive format, see docs/specs/vma_spec.txt
   7
   8 Signed-off-by: Dietmar Maurer <dietmar@proxmox.com>
   9 ---
  10  Makefile                |    3 +-
  11  Makefile.objs           |    2 +-
  12  blockdev.c              |    6 +-
  13  docs/specs/vma_spec.txt |   24 ++
  14  vma-reader.c            |  799 ++++++++++++++++++++++++++++++++++++++++
  15  vma-writer.c            |  940 +++++++++++++++++++++++++++++++++++++++++++++++
  16  vma.c                   |  561 ++++++++++++++++++++++++++++
  17  vma.h                   |  145 ++++++++
  18  8 files changed, 2476 insertions(+), 4 deletions(-)
  19  create mode 100644 docs/specs/vma_spec.txt
  20  create mode 100644 vma-reader.c
  21  create mode 100644 vma-writer.c
  22  create mode 100644 vma.c
  23  create mode 100644 vma.h
  24
  25 Index: new/Makefile
  26 ===================================================================
  27 --- new.orig/Makefile   2013-12-04 12:07:18.000000000 +0100
  28 +++ new/Makefile        2013-12-04 12:07:25.000000000 +0100
  29 @@ -128,7 +128,7 @@
  30  include $(SRC_PATH)/libcacard/Makefile
  31  endif
  32
  33 -all: $(DOCS) $(TOOLS) $(HELPERS-y) recurse-all
  34 +all: $(DOCS) $(TOOLS) vma$(EXESUF) $(HELPERS-y) recurse-all
  35
  36  config-host.h: config-host.h-timestamp
  37  config-host.h-timestamp: config-host.mak
  38 @@ -194,6 +194,7 @@
  39  qemu-img$(EXESUF): qemu-img.o $(block-obj-y) libqemuutil.a libqemustub.a
  40  qemu-nbd$(EXESUF): qemu-nbd.o $(block-obj-y) libqemuutil.a libqemustub.a
  41  qemu-io$(EXESUF): qemu-io.o $(block-obj-y) libqemuutil.a libqemustub.a
  42 +vma$(EXESUF): vma.o vma-reader.o $(block-obj-y)  libqemuutil.a libqemustub.a
  43
  44  qemu-bridge-helper$(EXESUF): qemu-bridge-helper.o
  45
  46 Index: new/docs/specs/vma_spec.txt
  47 ===================================================================
  48 --- /dev/null   1970-01-01 00:00:00.000000000 +0000
  49 +++ new/docs/specs/vma_spec.txt 2013-12-05 08:14:11.000000000 +0100
  50 @@ -0,0 +1,132 @@
  51 += Virtual Machine Archive format (VMA) =
  52 +
  53 +This format contains a header which includes the VM configuration as
  54 +binary blobs, and a list of devices (dev_id, name).
  55 +
  56 +The actual VM image data is stored inside extents. An extent contains
  57 +up to 64 clusters, and start with a 512 byte header containing
  58 +additional information for those clusters.
  59 +
  60 +We use a cluster size of 65536, and use 8 bytes for each
  61 +cluster in the header to store the following information:
  62 +
  63 +* 1 byte dev_id (to identity the drive)
  64 +* 1 byte not used (reserved)
  65 +* 2 bytes zero indicator (mark zero regions (16x4096))
  66 +* 4 bytes cluster number
  67 +
  68 +We only store non-zero blocks (such block is 4096 bytes).
  69 +
  70 +Each archive is marked with a uuid. The archive header and all
  71 +extent headers includes that uuid and a MD5 checksum (over header
  72 +data).
  73 +
  74 +All numbers in VMA archive are stored in Big Endian byte order.
  75 +
  76 +== VMA Header ==
  77 +
  78 +    Byte  0 -  3:   magic
  79 +                    VMA magic string ("VMA\x00")
  80 +
  81 +         4 -  7:   version
  82 +                    Version number (valid value is 1)
  83 +
  84 +          8 - 23:   uuid
  85 +                   Unique ID, Same uuid is used to mark extents.
  86 +
  87 +         24 - 31:   ctime
  88 +                   Backup time stamp (seconds since epoch)
  89 +
  90 +         32 - 47:   md5sum
  91 +                   Header checksum (from byte 0 to header_size). This field
  92 +                   is filled with zero to generate the checksum.
  93 +
  94 +         48 - 51:   blob_buffer_offset
  95 +                   Start of blob buffer (multiple of 512)
  96 +
  97 +         52 - 55:   blob_buffer_size
  98 +                   Size of blob buffer (multiple of 512)
  99 +
 100 +         56 - 59:   header_size
 101 +                   Overall size of this header (multiple of 512)
 102 +
 103 +        60 - 2043: reserved
 104 +
 105 +       2044 - 3067: uint32_t config_names[256]
 106 +                           Offsets into blob_buffer table
 107 +
 108 +       3068 - 4091: uint32_t config_data[256]
 109 +                           Offsets into blob_buffer table
 110 +
 111 +       4092 - 4095: reserved
 112 +
 113 +      4096 - 12287: VmaDeviceInfoHeader dev_info[256]
 114 +                    The offset in this table is used as 'dev_id' inside
 115 +                   the data streams.
 116 +
 117 +      12288 - header_size: Blob buffer
 118 +
 119 +
 120 +=== Devive Info Header (VmaDeviceInfoHeader) ===
 121 +
 122 +This is use to store details about the contained disk images.
 123 +
 124 +    Byte  0 -  3:   devive name (offsets into blob_buffer table)
 125 +
 126 +         4 -  7:   reserved
 127 +
 128 +         8 - 15:   device size in bytes
 129 +
 130 +        16 - 31:   reserved
 131 +
 132 +Note: Devive name 'vmstate' is reserved to store VM RAM state.
 133 +
 134 +=== Blob buffer ===
 135 +
 136 +The blob buffer is used to store both configuration file names and
 137 +configuration data.
 138 +
 139 +This region contain a list of binary data blobs. Each blob starts with
 140 +a 2 byte size field, followed by the actual data.
 141 +
 142 +== Image Data Streams ==
 143 +
 144 +The VMA header is followed by the image data stream. Image data is grouped
 145 +with extents, which contains up to 59 clusters from different images.
 146 +
 147 +=== VMA Extent Header ===
 148 +
 149 +    Byte  0 -  3:   magic
 150 +                    VMA extent magic string ("VMAE")
 151 +
 152 +         4 -  5:   reserved
 153 +
 154 +         6 -  7:   block_count
 155 +                   Overall number of contained 4K block
 156 +
 157 +          8 - 23:   uuid
 158 +                   Unique ID, Same uuid as used in the VMA header.
 159 +
 160 +         24 - 39:   md5sum
 161 +                   Header checksum (from byte 0 to header_size). This field
 162 +                   is filled with zero to generate the checksum.
 163 +
 164 +        40 - 511:   blockinfo[59]
 165 +
 166 +
 167 +Each 'blockinfo' (8 bytes) give further details about contained clusters:
 168 +
 169 +    Byte  0 - 1:   mask
 170 +                  Bitmap used to indicate non-zero 4K blocks inside the
 171 +                  cluster.
 172 +
 173 +              2:   reserved
 174 +
 175 +              3:   dev_id
 176 +                  Device ID (offset into dev_info table)
 177 +
 178 +          4 - 7:   cluster_num
 179 +
 180 +The extend header if followed by the actual cluster data, where we only
 181 +store non-zero 4K blocks.
 182 +
 183 Index: new/vma-reader.c
 184 ===================================================================
 185 --- /dev/null   1970-01-01 00:00:00.000000000 +0000
 186 +++ new/vma-reader.c    2013-12-04 12:07:25.000000000 +0100
 187 @@ -0,0 +1,799 @@
 188 +/*
 189 + * VMA: Virtual Machine Archive
 190 + *
 191 + * Copyright (C) 2012 Proxmox Server Solutions
 192 + *
 193 + * Authors:
 194 + *  Dietmar Maurer (dietmar@proxmox.com)
 195 + *
 196 + * This work is licensed under the terms of the GNU GPL, version 2 or later.
 197 + * See the COPYING file in the top-level directory.
 198 + *
 199 + */
 200 +
 201 +#include <stdio.h>
 202 +#include <errno.h>
 203 +#include <unistd.h>
 204 +#include <stdio.h>
 205 +#include <string.h>
 206 +#include <sys/types.h>
 207 +#include <sys/stat.h>
 208 +#include <fcntl.h>
 209 +#include <glib.h>
 210 +#include <uuid/uuid.h>
 211 +
 212 +#include "qemu-common.h"
 213 +#include "qemu/timer.h"
 214 +#include "qemu/ratelimit.h"
 215 +#include "vma.h"
 216 +#include "block/block.h"
 217 +
 218 +#define BITS_PER_LONG  (sizeof(unsigned long) * CHAR_BIT)
 219 +
 220 +static unsigned char zero_vma_block[VMA_BLOCK_SIZE];
 221 +
 222 +typedef struct VmaRestoreState {
 223 +    BlockDriverState *bs;
 224 +    bool write_zeroes;
 225 +    unsigned long *bitmap;
 226 +    int bitmap_size;
 227 +}  VmaRestoreState;
 228 +
 229 +struct VmaReader {
 230 +    int fd;
 231 +    GChecksum *md5csum;
 232 +    GHashTable *blob_hash;
 233 +    unsigned char *head_data;
 234 +    VmaDeviceInfo devinfo[256];
 235 +    VmaRestoreState rstate[256];
 236 +    GList *cdata_list;
 237 +    guint8 vmstate_stream;
 238 +    uint32_t vmstate_clusters;
 239 +    /* to show restore percentage if run with -v */
 240 +    time_t start_time;
 241 +    int64_t cluster_count;
 242 +    int64_t clusters_read;
 243 +    int clusters_read_per;
 244 +};
 245 +
 246 +static guint
 247 +g_int32_hash(gconstpointer v)
 248 +{
 249 +    return *(const uint32_t *)v;
 250 +}
 251 +
 252 +static gboolean
 253 +g_int32_equal(gconstpointer v1, gconstpointer v2)
 254 +{
 255 +    return *((const uint32_t *)v1) == *((const uint32_t *)v2);
 256 +}
 257 +
 258 +static int vma_reader_get_bitmap(VmaRestoreState *rstate, int64_t cluster_num)
 259 +{
 260 +    assert(rstate);
 261 +    assert(rstate->bitmap);
 262 +
 263 +    unsigned long val, idx, bit;
 264 +
 265 +    idx = cluster_num / BITS_PER_LONG;
 266 +
 267 +    assert(rstate->bitmap_size > idx);
 268 +
 269 +    bit = cluster_num % BITS_PER_LONG;
 270 +    val = rstate->bitmap[idx];
 271 +
 272 +    return !!(val & (1UL << bit));
 273 +}
 274 +
 275 +static void vma_reader_set_bitmap(VmaRestoreState *rstate, int64_t cluster_num,
 276 +                                  int dirty)
 277 +{
 278 +    assert(rstate);
 279 +    assert(rstate->bitmap);
 280 +
 281 +    unsigned long val, idx, bit;
 282 +
 283 +    idx = cluster_num / BITS_PER_LONG;
 284 +
 285 +    assert(rstate->bitmap_size > idx);
 286 +
 287 +    bit = cluster_num % BITS_PER_LONG;
 288 +    val = rstate->bitmap[idx];
 289 +    if (dirty) {
 290 +        if (!(val & (1UL << bit))) {
 291 +            val |= 1UL << bit;
 292 +        }
 293 +    } else {
 294 +        if (val & (1UL << bit)) {
 295 +            val &= ~(1UL << bit);
 296 +        }
 297 +    }
 298 +    rstate->bitmap[idx] = val;
 299 +}
 300 +
 301 +typedef struct VmaBlob {
 302 +    uint32_t start;
 303 +    uint32_t len;
 304 +    void *data;
 305 +} VmaBlob;
 306 +
 307 +static const VmaBlob *get_header_blob(VmaReader *vmar, uint32_t pos)
 308 +{
 309 +    assert(vmar);
 310 +    assert(vmar->blob_hash);
 311 +
 312 +    return g_hash_table_lookup(vmar->blob_hash, &pos);
 313 +}
 314 +
 315 +static const char *get_header_str(VmaReader *vmar, uint32_t pos)
 316 +{
 317 +    const VmaBlob *blob = get_header_blob(vmar, pos);
 318 +    if (!blob) {
 319 +        return NULL;
 320 +    }
 321 +    const char *res = (char *)blob->data;
 322 +    if (res[blob->len-1] != '\0') {
 323 +        return NULL;
 324 +    }
 325 +    return res;
 326 +}
 327 +
 328 +static ssize_t
 329 +safe_read(int fd, unsigned char *buf, size_t count)
 330 +{
 331 +    ssize_t n;
 332 +
 333 +    do {
 334 +        n = read(fd, buf, count);
 335 +    } while (n < 0 && errno == EINTR);
 336 +
 337 +    return n;
 338 +}
 339 +
 340 +static ssize_t
 341 +full_read(int fd, unsigned char *buf, size_t len)
 342 +{
 343 +    ssize_t n;
 344 +    size_t total;
 345 +
 346 +    total = 0;
 347 +
 348 +    while (len > 0) {
 349 +        n = safe_read(fd, buf, len);
 350 +
 351 +        if (n == 0) {
 352 +            return total;
 353 +        }
 354 +
 355 +        if (n <= 0) {
 356 +            break;
 357 +        }
 358 +
 359 +        buf += n;
 360 +        total += n;
 361 +        len -= n;
 362 +    }
 363 +
 364 +    if (len) {
 365 +        return -1;
 366 +    }
 367 +
 368 +    return total;
 369 +}
 370 +
 371 +void vma_reader_destroy(VmaReader *vmar)
 372 +{
 373 +    assert(vmar);
 374 +
 375 +    if (vmar->fd >= 0) {
 376 +        close(vmar->fd);
 377 +    }
 378 +
 379 +    if (vmar->cdata_list) {
 380 +        g_list_free(vmar->cdata_list);
 381 +    }
 382 +
 383 +    int i;
 384 +    for (i = 1; i < 256; i++) {
 385 +        if (vmar->rstate[i].bitmap) {
 386 +            g_free(vmar->rstate[i].bitmap);
 387 +        }
 388 +    }
 389 +
 390 +    if (vmar->md5csum) {
 391 +        g_checksum_free(vmar->md5csum);
 392 +    }
 393 +
 394 +    if (vmar->blob_hash) {
 395 +        g_hash_table_destroy(vmar->blob_hash);
 396 +    }
 397 +
 398 +    if (vmar->head_data) {
 399 +        g_free(vmar->head_data);
 400 +    }
 401 +
 402 +    g_free(vmar);
 403 +
 404 +};
 405 +
 406 +static int vma_reader_read_head(VmaReader *vmar, Error **errp)
 407 +{
 408 +    assert(vmar);
 409 +    assert(errp);
 410 +    assert(*errp == NULL);
 411 +
 412 +    unsigned char md5sum[16];
 413 +    int i;
 414 +    int ret = 0;
 415 +
 416 +    vmar->head_data = g_malloc(sizeof(VmaHeader));
 417 +
 418 +    if (full_read(vmar->fd, vmar->head_data, sizeof(VmaHeader)) !=
 419 +        sizeof(VmaHeader)) {
 420 +        error_setg(errp, "can't read vma header - %s",
 421 +                   errno ? g_strerror(errno) : "got EOF");
 422 +        return -1;
 423 +    }
 424 +
 425 +    VmaHeader *h = (VmaHeader *)vmar->head_data;
 426 +
 427 +    if (h->magic != VMA_MAGIC) {
 428 +        error_setg(errp, "not a vma file - wrong magic number");
 429 +        return -1;
 430 +    }
 431 +
 432 +    uint32_t header_size = GUINT32_FROM_BE(h->header_size);
 433 +    int need = header_size - sizeof(VmaHeader);
 434 +    if (need <= 0) {
 435 +        error_setg(errp, "wrong vma header size %d", header_size);
 436 +        return -1;
 437 +    }
 438 +
 439 +    vmar->head_data = g_realloc(vmar->head_data, header_size);
 440 +    h = (VmaHeader *)vmar->head_data;
 441 +
 442 +    if (full_read(vmar->fd, vmar->head_data + sizeof(VmaHeader), need) !=
 443 +        need) {
 444 +        error_setg(errp, "can't read vma header data - %s",
 445 +                   errno ? g_strerror(errno) : "got EOF");
 446 +        return -1;
 447 +    }
 448 +
 449 +    memcpy(md5sum, h->md5sum, 16);
 450 +    memset(h->md5sum, 0, 16);
 451 +
 452 +    g_checksum_reset(vmar->md5csum);
 453 +    g_checksum_update(vmar->md5csum, vmar->head_data, header_size);
 454 +    gsize csize = 16;
 455 +    g_checksum_get_digest(vmar->md5csum, (guint8 *)(h->md5sum), &csize);
 456 +
 457 +    if (memcmp(md5sum, h->md5sum, 16) != 0) {
 458 +        error_setg(errp, "wrong vma header chechsum");
 459 +        return -1;
 460 +    }
 461 +
 462 +    /* we can modify header data after checksum verify */
 463 +    h->header_size = header_size;
 464 +
 465 +    h->version = GUINT32_FROM_BE(h->version);
 466 +    if (h->version != 1) {
 467 +        error_setg(errp, "wrong vma version %d", h->version);
 468 +        return -1;
 469 +    }
 470 +
 471 +    h->ctime = GUINT64_FROM_BE(h->ctime);
 472 +    h->blob_buffer_offset = GUINT32_FROM_BE(h->blob_buffer_offset);
 473 +    h->blob_buffer_size = GUINT32_FROM_BE(h->blob_buffer_size);
 474 +
 475 +    uint32_t bstart = h->blob_buffer_offset + 1;
 476 +    uint32_t bend = h->blob_buffer_offset + h->blob_buffer_size;
 477 +
 478 +    if (bstart <= sizeof(VmaHeader)) {
 479 +        error_setg(errp, "wrong vma blob buffer offset %d",
 480 +                   h->blob_buffer_offset);
 481 +        return -1;
 482 +    }
 483 +
 484 +    if (bend > header_size) {
 485 +        error_setg(errp, "wrong vma blob buffer size %d/%d",
 486 +                   h->blob_buffer_offset, h->blob_buffer_size);
 487 +        return -1;
 488 +    }
 489 +
 490 +    while ((bstart + 2) <= bend) {
 491 +        uint32_t size = vmar->head_data[bstart] +
 492 +            (vmar->head_data[bstart+1] << 8);
 493 +        if ((bstart + size + 2) <= bend) {
 494 +            VmaBlob *blob = g_new0(VmaBlob, 1);
 495 +            blob->start = bstart - h->blob_buffer_offset;
 496 +            blob->len = size;
 497 +            blob->data = vmar->head_data + bstart + 2;
 498 +            g_hash_table_insert(vmar->blob_hash, &blob->start, blob);
 499 +        }
 500 +        bstart += size + 2;
 501 +    }
 502 +
 503 +
 504 +    int count = 0;
 505 +    for (i = 1; i < 256; i++) {
 506 +        VmaDeviceInfoHeader *dih = &h->dev_info[i];
 507 +        uint32_t devname_ptr = GUINT32_FROM_BE(dih->devname_ptr);
 508 +        uint64_t size = GUINT64_FROM_BE(dih->size);
 509 +        const char *devname =  get_header_str(vmar, devname_ptr);
 510 +
 511 +        if (size && devname) {
 512 +            count++;
 513 +            vmar->devinfo[i].size = size;
 514 +            vmar->devinfo[i].devname = devname;
 515 +
 516 +            if (strcmp(devname, "vmstate") == 0) {
 517 +                vmar->vmstate_stream = i;
 518 +            }
 519 +        }
 520 +    }
 521 +
 522 +    if (!count) {
 523 +        error_setg(errp, "vma does not contain data");
 524 +        return -1;
 525 +    }
 526 +
 527 +    for (i = 0; i < VMA_MAX_CONFIGS; i++) {
 528 +        uint32_t name_ptr = GUINT32_FROM_BE(h->config_names[i]);
 529 +        uint32_t data_ptr = GUINT32_FROM_BE(h->config_data[i]);
 530 +
 531 +        if (!(name_ptr && data_ptr)) {
 532 +            continue;
 533 +        }
 534 +        const char *name =  get_header_str(vmar, name_ptr);
 535 +        const VmaBlob *blob = get_header_blob(vmar, data_ptr);
 536 +
 537 +        if (!(name && blob)) {
 538 +            error_setg(errp, "vma contains invalid data pointers");
 539 +            return -1;
 540 +        }
 541 +
 542 +        VmaConfigData *cdata = g_new0(VmaConfigData, 1);
 543 +        cdata->name = name;
 544 +        cdata->data = blob->data;
 545 +        cdata->len = blob->len;
 546 +
 547 +        vmar->cdata_list = g_list_append(vmar->cdata_list, cdata);
 548 +    }
 549 +
 550 +    return ret;
 551 +};
 552 +
 553 +VmaReader *vma_reader_create(const char *filename, Error **errp)
 554 +{
 555 +    assert(filename);
 556 +    assert(errp);
 557 +
 558 +    VmaReader *vmar = g_new0(VmaReader, 1);
 559 +
 560 +    if (strcmp(filename, "-") == 0) {
 561 +        vmar->fd = dup(0);
 562 +    } else {
 563 +        vmar->fd = open(filename, O_RDONLY);
 564 +    }
 565 +
 566 +    if (vmar->fd < 0) {
 567 +        error_setg(errp, "can't open file %s - %s\n", filename,
 568 +                   g_strerror(errno));
 569 +        goto err;
 570 +    }
 571 +
 572 +    vmar->md5csum = g_checksum_new(G_CHECKSUM_MD5);
 573 +    if (!vmar->md5csum) {
 574 +        error_setg(errp, "can't allocate cmsum\n");
 575 +        goto err;
 576 +    }
 577 +
 578 +    vmar->blob_hash = g_hash_table_new_full(g_int32_hash, g_int32_equal,
 579 +                                            NULL, g_free);
 580 +
 581 +    if (vma_reader_read_head(vmar, errp) < 0) {
 582 +        goto err;
 583 +    }
 584 +
 585 +    return vmar;
 586 +
 587 +err:
 588 +    if (vmar) {
 589 +        vma_reader_destroy(vmar);
 590 +    }
 591 +
 592 +    return NULL;
 593 +}
 594 +
 595 +VmaHeader *vma_reader_get_header(VmaReader *vmar)
 596 +{
 597 +    assert(vmar);
 598 +    assert(vmar->head_data);
 599 +
 600 +    return (VmaHeader *)(vmar->head_data);
 601 +}
 602 +
 603 +GList *vma_reader_get_config_data(VmaReader *vmar)
 604 +{
 605 +    assert(vmar);
 606 +    assert(vmar->head_data);
 607 +
 608 +    return vmar->cdata_list;
 609 +}
 610 +
 611 +VmaDeviceInfo *vma_reader_get_device_info(VmaReader *vmar, guint8 dev_id)
 612 +{
 613 +    assert(vmar);
 614 +    assert(dev_id);
 615 +
 616 +    if (vmar->devinfo[dev_id].size && vmar->devinfo[dev_id].devname) {
 617 +        return &vmar->devinfo[dev_id];
 618 +    }
 619 +
 620 +    return NULL;
 621 +}
 622 +
 623 +int vma_reader_register_bs(VmaReader *vmar, guint8 dev_id, BlockDriverState *bs,
 624 +                           bool write_zeroes, Error **errp)
 625 +{
 626 +    assert(vmar);
 627 +    assert(bs != NULL);
 628 +    assert(dev_id);
 629 +    assert(vmar->rstate[dev_id].bs == NULL);
 630 +
 631 +    int64_t size = bdrv_getlength(bs);
 632 +    if (size != vmar->devinfo[dev_id].size) {
 633 +        error_setg(errp, "vma_reader_register_bs for stream %s failed - "
 634 +                   "unexpected size %zd != %zd", vmar->devinfo[dev_id].devname,
 635 +                   size, vmar->devinfo[dev_id].size);
 636 +        return -1;
 637 +    }
 638 +
 639 +    vmar->rstate[dev_id].bs = bs;
 640 +    vmar->rstate[dev_id].write_zeroes = write_zeroes;
 641 +
 642 +    int64_t bitmap_size = (size/BDRV_SECTOR_SIZE) +
 643 +        (VMA_CLUSTER_SIZE/BDRV_SECTOR_SIZE) * BITS_PER_LONG - 1;
 644 +    bitmap_size /= (VMA_CLUSTER_SIZE/BDRV_SECTOR_SIZE) * BITS_PER_LONG;
 645 +
 646 +    vmar->rstate[dev_id].bitmap_size = bitmap_size;
 647 +    vmar->rstate[dev_id].bitmap = g_new0(unsigned long, bitmap_size);
 648 +
 649 +    vmar->cluster_count += size/VMA_CLUSTER_SIZE;
 650 +
 651 +    return 0;
 652 +}
 653 +
 654 +static ssize_t safe_write(int fd, void *buf, size_t count)
 655 +{
 656 +    ssize_t n;
 657 +
 658 +    do {
 659 +        n = write(fd, buf, count);
 660 +    } while (n < 0 && errno == EINTR);
 661 +
 662 +    return n;
 663 +}
 664 +
 665 +static size_t full_write(int fd, void *buf, size_t len)
 666 +{
 667 +    ssize_t n;
 668 +    size_t total;
 669 +
 670 +    total = 0;
 671 +
 672 +    while (len > 0) {
 673 +        n = safe_write(fd, buf, len);
 674 +        if (n < 0) {
 675 +            return n;
 676 +        }
 677 +        buf += n;
 678 +        total += n;
 679 +        len -= n;
 680 +    }
 681 +
 682 +    if (len) {
 683 +        /* incomplete write ? */
 684 +        return -1;
 685 +    }
 686 +
 687 +    return total;
 688 +}
 689 +
 690 +static int restore_write_data(VmaReader *vmar, guint8 dev_id,
 691 +                              BlockDriverState *bs, int vmstate_fd,
 692 +                              unsigned char *buf, int64_t sector_num,
 693 +                              int nb_sectors, Error **errp)
 694 +{
 695 +    assert(vmar);
 696 +
 697 +    if (dev_id == vmar->vmstate_stream) {
 698 +        if (vmstate_fd >= 0) {
 699 +            int len = nb_sectors * BDRV_SECTOR_SIZE;
 700 +            int res = full_write(vmstate_fd, buf, len);
 701 +            if (res < 0) {
 702 +                error_setg(errp, "write vmstate failed %d", res);
 703 +                return -1;
 704 +            }
 705 +        }
 706 +    } else {
 707 +        int res = bdrv_write(bs, sector_num, buf, nb_sectors);
 708 +        if (res < 0) {
 709 +            error_setg(errp, "bdrv_write to %s failed (%d)",
 710 +                       bdrv_get_device_name(bs), res);
 711 +            return -1;
 712 +        }
 713 +    }
 714 +    return 0;
 715 +}
 716 +static int restore_extent(VmaReader *vmar, unsigned char *buf,
 717 +                          int extent_size, int vmstate_fd,
 718 +                          bool verbose, Error **errp)
 719 +{
 720 +    assert(vmar);
 721 +    assert(buf);
 722 +
 723 +    VmaExtentHeader *ehead = (VmaExtentHeader *)buf;
 724 +    int start = VMA_EXTENT_HEADER_SIZE;
 725 +    int i;
 726 +
 727 +    for (i = 0; i < VMA_BLOCKS_PER_EXTENT; i++) {
 728 +        uint64_t block_info = GUINT64_FROM_BE(ehead->blockinfo[i]);
 729 +        uint64_t cluster_num = block_info & 0xffffffff;
 730 +        uint8_t dev_id = (block_info >> 32) & 0xff;
 731 +        uint16_t mask = block_info >> (32+16);
 732 +        int64_t max_sector;
 733 +
 734 +        if (!dev_id) {
 735 +            continue;
 736 +        }
 737 +
 738 +        VmaRestoreState *rstate = &vmar->rstate[dev_id];
 739 +        BlockDriverState *bs = NULL;
 740 +
 741 +        if (dev_id != vmar->vmstate_stream) {
 742 +            bs = rstate->bs;
 743 +            if (!bs) {
 744 +                error_setg(errp, "got wrong dev id %d", dev_id);
 745 +                return -1;
 746 +            }
 747 +
 748 +            if (vma_reader_get_bitmap(rstate, cluster_num)) {
 749 +                error_setg(errp, "found duplicated cluster %zd for stream %s",
 750 +                          cluster_num, vmar->devinfo[dev_id].devname);
 751 +                return -1;
 752 +            }
 753 +            vma_reader_set_bitmap(rstate, cluster_num, 1);
 754 +
 755 +            max_sector = vmar->devinfo[dev_id].size/BDRV_SECTOR_SIZE;
 756 +        } else {
 757 +            max_sector = G_MAXINT64;
 758 +            if (cluster_num != vmar->vmstate_clusters) {
 759 +                error_setg(errp, "found out of order vmstate data");
 760 +                return -1;
 761 +            }
 762 +            vmar->vmstate_clusters++;
 763 +        }
 764 +
 765 +        vmar->clusters_read++;
 766 +
 767 +        if (verbose) {
 768 +            time_t duration = time(NULL) - vmar->start_time;
 769 +            int percent = (vmar->clusters_read*100)/vmar->cluster_count;
 770 +            if (percent != vmar->clusters_read_per) {
 771 +                printf("progress %d%% (read %zd bytes, duration %zd sec)\n",
 772 +                       percent, vmar->clusters_read*VMA_CLUSTER_SIZE,
 773 +                       duration);
 774 +                fflush(stdout);
 775 +                vmar->clusters_read_per = percent;
 776 +            }
 777 +        }
 778 +
 779 +        /* try to write whole clusters to speedup restore */
 780 +        if (mask == 0xffff) {
 781 +            if ((start + VMA_CLUSTER_SIZE) > extent_size) {
 782 +                error_setg(errp, "short vma extent - too many blocks");
 783 +                return -1;
 784 +            }
 785 +            int64_t sector_num = (cluster_num * VMA_CLUSTER_SIZE) /
 786 +                BDRV_SECTOR_SIZE;
 787 +            int64_t end_sector = sector_num +
 788 +                VMA_CLUSTER_SIZE/BDRV_SECTOR_SIZE;
 789 +
 790 +            if (end_sector > max_sector) {
 791 +                end_sector = max_sector;
 792 +            }
 793 +
 794 +            if (end_sector <= sector_num) {
 795 +                error_setg(errp, "got wrong block address - write bejond end");
 796 +                return -1;
 797 +            }
 798 +
 799 +            int nb_sectors = end_sector - sector_num;
 800 +            if (restore_write_data(vmar, dev_id, bs, vmstate_fd, buf + start,
 801 +                                   sector_num, nb_sectors, errp) < 0) {
 802 +                return -1;
 803 +            }
 804 +
 805 +            start += VMA_CLUSTER_SIZE;
 806 +        } else {
 807 +            int j;
 808 +            int bit = 1;
 809 +
 810 +            for (j = 0; j < 16; j++) {
 811 +                int64_t sector_num = (cluster_num*VMA_CLUSTER_SIZE +
 812 +                                      j*VMA_BLOCK_SIZE)/BDRV_SECTOR_SIZE;
 813 +
 814 +                int64_t end_sector = sector_num +
 815 +                    VMA_BLOCK_SIZE/BDRV_SECTOR_SIZE;
 816 +                if (end_sector > max_sector) {
 817 +                    end_sector = max_sector;
 818 +                }
 819 +
 820 +                if (mask & bit) {
 821 +                    if ((start + VMA_BLOCK_SIZE) > extent_size) {
 822 +                        error_setg(errp, "short vma extent - too many blocks");
 823 +                        return -1;
 824 +                    }
 825 +
 826 +                    if (end_sector <= sector_num) {
 827 +                        error_setg(errp, "got wrong block address - "
 828 +                                   "write bejond end");
 829 +                        return -1;
 830 +                    }
 831 +
 832 +                    int nb_sectors = end_sector - sector_num;
 833 +                    if (restore_write_data(vmar, dev_id, bs, vmstate_fd,
 834 +                                           buf + start, sector_num,
 835 +                                           nb_sectors, errp) < 0) {
 836 +                        return -1;
 837 +                    }
 838 +
 839 +                    start += VMA_BLOCK_SIZE;
 840 +
 841 +                } else {
 842 +
 843 +                    if (rstate->write_zeroes && (end_sector > sector_num)) {
 844 +                        /* Todo: use bdrv_co_write_zeroes (but that need to
 845 +                         * be run inside coroutine?)
 846 +                         */
 847 +                        int nb_sectors = end_sector - sector_num;
 848 +                        if (restore_write_data(vmar, dev_id, bs, vmstate_fd,
 849 +                                              zero_vma_block, sector_num,
 850 +                                               nb_sectors, errp) < 0) {
 851 +                            return -1;
 852 +                        }
 853 +                    }
 854 +                }
 855 +
 856 +                bit = bit << 1;
 857 +            }
 858 +        }
 859 +    }
 860 +
 861 +    if (start != extent_size) {
 862 +        error_setg(errp, "vma extent error - missing blocks");
 863 +        return -1;
 864 +    }
 865 +
 866 +    return 0;
 867 +}
 868 +
 869 +int vma_reader_restore(VmaReader *vmar, int vmstate_fd, bool verbose,
 870 +                       Error **errp)
 871 +{
 872 +    assert(vmar);
 873 +    assert(vmar->head_data);
 874 +
 875 +    int ret = 0;
 876 +    unsigned char buf[VMA_MAX_EXTENT_SIZE];
 877 +    int buf_pos = 0;
 878 +    unsigned char md5sum[16];
 879 +    VmaHeader *h = (VmaHeader *)vmar->head_data;
 880 +
 881 +    vmar->start_time = time(NULL);
 882 +
 883 +    while (1) {
 884 +        int bytes = full_read(vmar->fd, buf + buf_pos, sizeof(buf) - buf_pos);
 885 +        if (bytes < 0) {
 886 +            error_setg(errp, "read failed - %s", g_strerror(errno));
 887 +            return -1;
 888 +        }
 889 +
 890 +        buf_pos += bytes;
 891 +
 892 +        if (!buf_pos) {
 893 +            break; /* EOF */
 894 +        }
 895 +
 896 +        if (buf_pos < VMA_EXTENT_HEADER_SIZE) {
 897 +            error_setg(errp, "read short extent (%d bytes)", buf_pos);
 898 +            return -1;
 899 +        }
 900 +
 901 +        VmaExtentHeader *ehead = (VmaExtentHeader *)buf;
 902 +
 903 +        /* extract md5sum */
 904 +        memcpy(md5sum, ehead->md5sum, sizeof(ehead->md5sum));
 905 +        memset(ehead->md5sum, 0, sizeof(ehead->md5sum));
 906 +
 907 +        g_checksum_reset(vmar->md5csum);
 908 +        g_checksum_update(vmar->md5csum, buf, VMA_EXTENT_HEADER_SIZE);
 909 +        gsize csize = 16;
 910 +        g_checksum_get_digest(vmar->md5csum, ehead->md5sum, &csize);
 911 +
 912 +        if (memcmp(md5sum, ehead->md5sum, 16) != 0) {
 913 +            error_setg(errp, "wrong vma extent header chechsum");
 914 +            return -1;
 915 +        }
 916 +
 917 +        if (memcmp(h->uuid, ehead->uuid, sizeof(ehead->uuid)) != 0) {
 918 +            error_setg(errp, "wrong vma extent uuid");
 919 +            return -1;
 920 +        }
 921 +
 922 +        if (ehead->magic != VMA_EXTENT_MAGIC || ehead->reserved1 != 0) {
 923 +            error_setg(errp, "wrong vma extent header magic");
 924 +            return -1;
 925 +        }
 926 +
 927 +        int block_count = GUINT16_FROM_BE(ehead->block_count);
 928 +        int extent_size = VMA_EXTENT_HEADER_SIZE + block_count*VMA_BLOCK_SIZE;
 929 +
 930 +        if (buf_pos < extent_size) {
 931 +            error_setg(errp, "short vma extent (%d < %d)", buf_pos,
 932 +                       extent_size);
 933 +            return -1;
 934 +        }
 935 +
 936 +        if (restore_extent(vmar, buf, extent_size, vmstate_fd, verbose,
 937 +                           errp) < 0) {
 938 +            return -1;
 939 +        }
 940 +
 941 +        if (buf_pos > extent_size) {
 942 +            memmove(buf, buf + extent_size, buf_pos - extent_size);
 943 +            buf_pos = buf_pos - extent_size;
 944 +        } else {
 945 +            buf_pos = 0;
 946 +        }
 947 +    }
 948 +
 949 +    bdrv_drain_all();
 950 +
 951 +    int i;
 952 +    for (i = 1; i < 256; i++) {
 953 +        VmaRestoreState *rstate = &vmar->rstate[i];
 954 +        if (!rstate->bs) {
 955 +            continue;
 956 +        }
 957 +
 958 +        if (bdrv_flush(rstate->bs) < 0) {
 959 +            error_setg(errp, "vma bdrv_flush %s failed",
 960 +                       vmar->devinfo[i].devname);
 961 +            return -1;
 962 +        }
 963 +
 964 +        if (vmar->devinfo[i].size &&
 965 +            (strcmp(vmar->devinfo[i].devname, "vmstate") != 0)) {
 966 +            assert(rstate->bitmap);
 967 +
 968 +            int64_t cluster_num, end;
 969 +
 970 +            end = (vmar->devinfo[i].size + VMA_CLUSTER_SIZE - 1) /
 971 +                VMA_CLUSTER_SIZE;
 972 +
 973 +            for (cluster_num = 0; cluster_num < end; cluster_num++) {
 974 +                if (!vma_reader_get_bitmap(rstate, cluster_num)) {
 975 +                    error_setg(errp, "detected missing cluster %zd "
 976 +                               "for stream %s", cluster_num,
 977 +                               vmar->devinfo[i].devname);
 978 +                    return -1;
 979 +                }
 980 +            }
 981 +        }
 982 +    }
 983 +
 984 +    return ret;
 985 +}
 986 +
 987 Index: new/vma-writer.c
 988 ===================================================================
 989 --- /dev/null   1970-01-01 00:00:00.000000000 +0000
 990 +++ new/vma-writer.c    2013-12-04 14:26:06.000000000 +0100
 991 @@ -0,0 +1,875 @@
 992 +/*
 993 + * VMA: Virtual Machine Archive
 994 + *
 995 + * Copyright (C) 2012 Proxmox Server Solutions
 996 + *
 997 + * Authors:
 998 + *  Dietmar Maurer (dietmar@proxmox.com)
 999 + *
1000 + * This work is licensed under the terms of the GNU GPL, version 2 or later.
1001 + * See the COPYING file in the top-level directory.
1002 + *
1003 + */
1004 +
1005 +#include <stdio.h>
1006 +#include <errno.h>
1007 +#include <unistd.h>
1008 +#include <stdio.h>
1009 +#include <string.h>
1010 +#include <sys/types.h>
1011 +#include <sys/stat.h>
1012 +#include <fcntl.h>
1013 +#include <glib.h>
1014 +#include <uuid/uuid.h>
1015 +
1016 +#include "qemu-common.h"
1017 +#include "vma.h"
1018 +#include "block/block.h"
1019 +#include "monitor/monitor.h"
1020 +
1021 +#define DEBUG_VMA 0
1022 +
1023 +#define DPRINTF(fmt, ...)\
1024 +    do { if (DEBUG_VMA) { printf("vma: " fmt, ## __VA_ARGS__); } } while (0)
1025 +
1026 +#define WRITE_BUFFERS 5
1027 +
1028 +typedef struct VmaAIOCB VmaAIOCB;
1029 +struct VmaAIOCB {
1030 +    unsigned char buffer[VMA_MAX_EXTENT_SIZE];
1031 +    VmaWriter *vmaw;
1032 +    size_t bytes;
1033 +    Coroutine *co;
1034 +};
1035 +
1036 +struct VmaWriter {
1037 +    int fd;
1038 +    FILE *cmd;
1039 +    int status;
1040 +    char errmsg[8192];
1041 +    uuid_t uuid;
1042 +    bool header_written;
1043 +    bool closed;
1044 +
1045 +    /* we always write extents */
1046 +    unsigned char outbuf[VMA_MAX_EXTENT_SIZE];
1047 +    int outbuf_pos; /* in bytes */
1048 +    int outbuf_count; /* in VMA_BLOCKS */
1049 +    uint64_t outbuf_block_info[VMA_BLOCKS_PER_EXTENT];
1050 +
1051 +    VmaAIOCB *aiocbs[WRITE_BUFFERS];
1052 +    CoQueue wqueue;
1053 +
1054 +    GChecksum *md5csum;
1055 +    CoMutex writer_lock;
1056 +    CoMutex flush_lock;
1057 +    Coroutine *co_writer;
1058 +
1059 +    /* drive informations */
1060 +    VmaStreamInfo stream_info[256];
1061 +    guint stream_count;
1062 +
1063 +    guint8 vmstate_stream;
1064 +    uint32_t vmstate_clusters;
1065 +
1066 +    /* header blob table */
1067 +    char *header_blob_table;
1068 +    uint32_t header_blob_table_size;
1069 +    uint32_t header_blob_table_pos;
1070 +
1071 +    /* store for config blobs */
1072 +    uint32_t config_names[VMA_MAX_CONFIGS]; /* offset into blob_buffer table */
1073 +    uint32_t config_data[VMA_MAX_CONFIGS];  /* offset into blob_buffer table */
1074 +    uint32_t config_count;
1075 +};
1076 +
1077 +void vma_writer_set_error(VmaWriter *vmaw, const char *fmt, ...)
1078 +{
1079 +    va_list ap;
1080 +
1081 +    if (vmaw->status < 0) {
1082 +        return;
1083 +    }
1084 +
1085 +    vmaw->status = -1;
1086 +
1087 +    va_start(ap, fmt);
1088 +    g_vsnprintf(vmaw->errmsg, sizeof(vmaw->errmsg), fmt, ap);
1089 +    va_end(ap);
1090 +
1091 +    DPRINTF("vma_writer_set_error: %s\n", vmaw->errmsg);
1092 +}
1093 +
1094 +static uint32_t allocate_header_blob(VmaWriter *vmaw, const char *data,
1095 +                                     size_t len)
1096 +{
1097 +    if (len > 65535) {
1098 +        return 0;
1099 +    }
1100 +
1101 +    if (!vmaw->header_blob_table ||
1102 +        (vmaw->header_blob_table_size <
1103 +         (vmaw->header_blob_table_pos + len + 2))) {
1104 +        int newsize = vmaw->header_blob_table_size + ((len + 2 + 511)/512)*512;
1105 +
1106 +        vmaw->header_blob_table = g_realloc(vmaw->header_blob_table, newsize);
1107 +        memset(vmaw->header_blob_table + vmaw->header_blob_table_size,
1108 +               0, newsize - vmaw->header_blob_table_size);
1109 +        vmaw->header_blob_table_size = newsize;
1110 +    }
1111 +
1112 +    uint32_t cpos = vmaw->header_blob_table_pos;
1113 +    vmaw->header_blob_table[cpos] = len & 255;
1114 +    vmaw->header_blob_table[cpos+1] = (len >> 8) & 255;
1115 +    memcpy(vmaw->header_blob_table + cpos + 2, data, len);
1116 +    vmaw->header_blob_table_pos += len + 2;
1117 +    return cpos;
1118 +}
1119 +
1120 +static uint32_t allocate_header_string(VmaWriter *vmaw, const char *str)
1121 +{
1122 +    assert(vmaw);
1123 +
1124 +    size_t len = strlen(str) + 1;
1125 +
1126 +    return allocate_header_blob(vmaw, str, len);
1127 +}
1128 +
1129 +int vma_writer_add_config(VmaWriter *vmaw, const char *name, gpointer data,
1130 +                          gsize len)
1131 +{
1132 +    assert(vmaw);
1133 +    assert(!vmaw->header_written);
1134 +    assert(vmaw->config_count < VMA_MAX_CONFIGS);
1135 +    assert(name);
1136 +    assert(data);
1137 +    assert(len);
1138 +
1139 +    gchar *basename = g_path_get_basename(name);
1140 +    uint32_t name_ptr = allocate_header_string(vmaw, basename);
1141 +    g_free(basename);
1142 +
1143 +    if (!name_ptr) {
1144 +        return -1;
1145 +    }
1146 +
1147 +    uint32_t data_ptr = allocate_header_blob(vmaw, data, len);
1148 +    if (!data_ptr) {
1149 +        return -1;
1150 +    }
1151 +
1152 +    vmaw->config_names[vmaw->config_count] = name_ptr;
1153 +    vmaw->config_data[vmaw->config_count] = data_ptr;
1154 +
1155 +    vmaw->config_count++;
1156 +
1157 +    return 0;
1158 +}
1159 +
1160 +int vma_writer_register_stream(VmaWriter *vmaw, const char *devname,
1161 +                               size_t size)
1162 +{
1163 +    assert(vmaw);
1164 +    assert(devname);
1165 +    assert(!vmaw->status);
1166 +
1167 +    if (vmaw->header_written) {
1168 +        vma_writer_set_error(vmaw, "vma_writer_register_stream: header "
1169 +                             "already written");
1170 +        return -1;
1171 +    }
1172 +
1173 +    guint n = vmaw->stream_count + 1;
1174 +
1175 +    /* we can have dev_ids form 1 to 255 (0 reserved)
1176 +     * 255(-1) reseverd for safety
1177 +     */
1178 +    if (n > 254) {
1179 +        vma_writer_set_error(vmaw, "vma_writer_register_stream: "
1180 +                             "too many drives");
1181 +        return -1;
1182 +    }
1183 +
1184 +    if (size <= 0) {
1185 +        vma_writer_set_error(vmaw, "vma_writer_register_stream: "
1186 +                             "got strange size %zd", size);
1187 +        return -1;
1188 +    }
1189 +
1190 +    DPRINTF("vma_writer_register_stream %s %zu %d\n", devname, size, n);
1191 +
1192 +    vmaw->stream_info[n].devname = g_strdup(devname);
1193 +    vmaw->stream_info[n].size = size;
1194 +
1195 +    vmaw->stream_info[n].cluster_count = (size + VMA_CLUSTER_SIZE - 1) /
1196 +        VMA_CLUSTER_SIZE;
1197 +
1198 +    vmaw->stream_count = n;
1199 +
1200 +    if (strcmp(devname, "vmstate") == 0) {
1201 +        vmaw->vmstate_stream = n;
1202 +    }
1203 +
1204 +    return n;
1205 +}
1206 +
1207 +static void vma_co_continue_write(void *opaque)
1208 +{
1209 +    VmaWriter *vmaw = opaque;
1210 +
1211 +    DPRINTF("vma_co_continue_write\n");
1212 +    qemu_coroutine_enter(vmaw->co_writer, NULL);
1213 +}
1214 +
1215 +static ssize_t coroutine_fn
1216 +vma_co_write(VmaWriter *vmaw, const void *buf, size_t bytes)
1217 +{
1218 +    size_t done = 0;
1219 +    ssize_t ret;
1220 +
1221 +    /* atomic writes (we cannot interleave writes) */
1222 +    qemu_co_mutex_lock(&vmaw->writer_lock);
1223 +
1224 +    DPRINTF("vma_co_write enter %zd\n", bytes);
1225 +
1226 +    assert(vmaw->co_writer == NULL);
1227 +
1228 +    vmaw->co_writer = qemu_coroutine_self();
1229 +
1230 +    qemu_aio_set_fd_handler(vmaw->fd, NULL, vma_co_continue_write, vmaw);
1231 +
1232 +    DPRINTF("vma_co_write wait until writable\n");
1233 +    qemu_coroutine_yield();
1234 +    DPRINTF("vma_co_write starting %zd\n", bytes);
1235 +
1236 +    while (done < bytes) {
1237 +        ret = write(vmaw->fd, buf + done, bytes - done);
1238 +        if (ret > 0) {
1239 +            done += ret;
1240 +            DPRINTF("vma_co_write written %zd %zd\n", done, ret);
1241 +        } else if (ret < 0) {
1242 +            if (errno == EAGAIN || errno == EWOULDBLOCK) {
1243 +                DPRINTF("vma_co_write yield %zd\n", done);
1244 +                qemu_coroutine_yield();
1245 +                DPRINTF("vma_co_write restart %zd\n", done);
1246 +            } else {
1247 +                vma_writer_set_error(vmaw, "vma_co_write write error - %s",
1248 +                                     g_strerror(errno));
1249 +                done = -1; /* always return failure for partial writes */
1250 +                break;
1251 +            }
1252 +        } else if (ret == 0) {
1253 +            /* should not happen - simply try again */
1254 +        }
1255 +    }
1256 +
1257 +    qemu_aio_set_fd_handler(vmaw->fd, NULL, NULL, NULL);
1258 +
1259 +    vmaw->co_writer = NULL;
1260 +
1261 +    qemu_co_mutex_unlock(&vmaw->writer_lock);
1262 +
1263 +    DPRINTF("vma_co_write leave %zd\n", done);
1264 +    return done;
1265 +}
1266 +
1267 +static void coroutine_fn vma_co_writer_task(void *opaque)
1268 +{
1269 +    VmaAIOCB *cb = opaque;
1270 +
1271 +    DPRINTF("vma_co_writer_task start\n");
1272 +
1273 +    int64_t done = vma_co_write(cb->vmaw, cb->buffer, cb->bytes);
1274 +    DPRINTF("vma_co_writer_task write done %zd\n", done);
1275 +
1276 +    if (done != cb->bytes) {
1277 +        DPRINTF("vma_co_writer_task failed write %zd %zd", cb->bytes, done);
1278 +        vma_writer_set_error(cb->vmaw, "vma_co_writer_task failed write %zd",
1279 +                             done);
1280 +    }
1281 +
1282 +    cb->bytes = 0;
1283 +
1284 +    qemu_co_queue_next(&cb->vmaw->wqueue);
1285 +
1286 +    DPRINTF("vma_co_writer_task end\n");
1287 +}
1288 +
1289 +static void coroutine_fn vma_queue_flush(VmaWriter *vmaw)
1290 +{
1291 +    DPRINTF("vma_queue_flush enter\n");
1292 +
1293 +    assert(vmaw);
1294 +
1295 +    while (1) {
1296 +        int i;
1297 +        VmaAIOCB *cb = NULL;
1298 +        for (i = 0; i < WRITE_BUFFERS; i++) {
1299 +            if (vmaw->aiocbs[i]->bytes) {
1300 +                cb = vmaw->aiocbs[i];
1301 +                DPRINTF("FOUND USED AIO BUFFER %d %zd\n", i,
1302 +                        vmaw->aiocbs[i]->bytes);
1303 +                break;
1304 +            }
1305 +        }
1306 +        if (!cb) {
1307 +            break;
1308 +        }
1309 +        qemu_co_queue_wait(&vmaw->wqueue);
1310 +    }
1311 +
1312 +    DPRINTF("vma_queue_flush leave\n");
1313 +}
1314 +
1315 +/**
1316 + * NOTE: pipe buffer size in only 4096 bytes on linux (see 'ulimit -a')
1317 + * So we need to create a coroutione to allow 'parallel' execution.
1318 + */
1319 +static ssize_t coroutine_fn
1320 +vma_queue_write(VmaWriter *vmaw, const void *buf, size_t bytes)
1321 +{
1322 +    DPRINTF("vma_queue_write enter %zd\n", bytes);
1323 +
1324 +    assert(vmaw);
1325 +    assert(buf);
1326 +    assert(bytes <= VMA_MAX_EXTENT_SIZE);
1327 +
1328 +    VmaAIOCB *cb = NULL;
1329 +    while (!cb) {
1330 +        int i;
1331 +        for (i = 0; i < WRITE_BUFFERS; i++) {
1332 +            if (!vmaw->aiocbs[i]->bytes) {
1333 +                cb = vmaw->aiocbs[i];
1334 +                break;
1335 +            }
1336 +        }
1337 +        if (!cb) {
1338 +            qemu_co_queue_wait(&vmaw->wqueue);
1339 +        }
1340 +    }
1341 +
1342 +    memcpy(cb->buffer, buf, bytes);
1343 +    cb->bytes = bytes;
1344 +    cb->vmaw = vmaw;
1345 +
1346 +    DPRINTF("vma_queue_write start %zd\n", bytes);
1347 +    cb->co = qemu_coroutine_create(vma_co_writer_task);
1348 +    qemu_coroutine_enter(cb->co, cb);
1349 +
1350 +    DPRINTF("vma_queue_write leave\n");
1351 +
1352 +    return bytes;
1353 +}
1354 +
1355 +VmaWriter *vma_writer_create(const char *filename, uuid_t uuid, Error **errp)
1356 +{
1357 +    const char *p;
1358 +
1359 +    assert(sizeof(VmaHeader) == (4096 + 8192));
1360 +    assert(G_STRUCT_OFFSET(VmaHeader, config_names) == 2044);
1361 +    assert(G_STRUCT_OFFSET(VmaHeader, config_data) == 3068);
1362 +    assert(G_STRUCT_OFFSET(VmaHeader, dev_info) == 4096);
1363 +    assert(sizeof(VmaExtentHeader) == 512);
1364 +
1365 +    VmaWriter *vmaw = g_new0(VmaWriter, 1);
1366 +    vmaw->fd = -1;
1367 +
1368 +    vmaw->md5csum = g_checksum_new(G_CHECKSUM_MD5);
1369 +    if (!vmaw->md5csum) {
1370 +        error_setg(errp, "can't allocate cmsum\n");
1371 +        goto err;
1372 +    }
1373 +
1374 +    if (strstart(filename, "exec:", &p)) {
1375 +        vmaw->cmd = popen(p, "w");
1376 +        if (vmaw->cmd == NULL) {
1377 +            error_setg(errp, "can't popen command '%s' - %s\n", p,
1378 +                       g_strerror(errno));
1379 +            goto err;
1380 +        }
1381 +        vmaw->fd = fileno(vmaw->cmd);
1382 +
1383 +        /* try to use O_NONBLOCK and O_DIRECT */
1384 +        fcntl(vmaw->fd, F_SETFL, fcntl(vmaw->fd, F_GETFL)|O_NONBLOCK);
1385 +        fcntl(vmaw->fd, F_SETFL, fcntl(vmaw->fd, F_GETFL)|O_DIRECT);
1386 +
1387 +    } else {
1388 +        struct stat st;
1389 +        int oflags;
1390 +        const char *tmp_id_str;
1391 +
1392 +        if ((stat(filename, &st) == 0) && S_ISFIFO(st.st_mode)) {
1393 +            oflags = O_NONBLOCK|O_DIRECT|O_WRONLY;
1394 +            vmaw->fd = qemu_open(filename, oflags, 0644);
1395 +        } else if (strstart(filename, "/dev/fdset/", &tmp_id_str)) {
1396 +            oflags = O_NONBLOCK|O_DIRECT|O_WRONLY;
1397 +            vmaw->fd = qemu_open(filename, oflags, 0644);
1398 +        } else if (strstart(filename, "/dev/fdname/", &tmp_id_str)) {
1399 +            vmaw->fd = monitor_get_fd(cur_mon, tmp_id_str, errp);
1400 +            if (vmaw->fd < 0) {
1401 +                goto err;
1402 +            }
1403 +            /* try to use O_NONBLOCK and O_DIRECT */
1404 +            fcntl(vmaw->fd, F_SETFL, fcntl(vmaw->fd, F_GETFL)|O_NONBLOCK);
1405 +            fcntl(vmaw->fd, F_SETFL, fcntl(vmaw->fd, F_GETFL)|O_DIRECT);
1406 +        } else  {
1407 +            oflags = O_NONBLOCK|O_DIRECT|O_WRONLY|O_CREAT|O_EXCL;
1408 +            vmaw->fd = qemu_open(filename, oflags, 0644);
1409 +        }
1410 +
1411 +        if (vmaw->fd < 0) {
1412 +            error_setg(errp, "can't open file %s - %s\n", filename,
1413 +                       g_strerror(errno));
1414 +            goto err;
1415 +        }
1416 +    }
1417 +
1418 +    /* we use O_DIRECT, so we need to align IO buffers */
1419 +    int i;
1420 +    for (i = 0; i < WRITE_BUFFERS; i++) {
1421 +        vmaw->aiocbs[i] = qemu_memalign(512, sizeof(VmaAIOCB));
1422 +        memset(vmaw->aiocbs[i], 0, sizeof(VmaAIOCB));
1423 +    }
1424 +
1425 +    vmaw->outbuf_count = 0;
1426 +    vmaw->outbuf_pos = VMA_EXTENT_HEADER_SIZE;
1427 +
1428 +    vmaw->header_blob_table_pos = 1; /* start at pos 1 */
1429 +
1430 +    qemu_co_mutex_init(&vmaw->writer_lock);
1431 +    qemu_co_mutex_init(&vmaw->flush_lock);
1432 +    qemu_co_queue_init(&vmaw->wqueue);
1433 +
1434 +    uuid_copy(vmaw->uuid, uuid);
1435 +
1436 +    return vmaw;
1437 +
1438 +err:
1439 +    if (vmaw) {
1440 +        if (vmaw->cmd) {
1441 +            pclose(vmaw->cmd);
1442 +        } else if (vmaw->fd >= 0) {
1443 +            close(vmaw->fd);
1444 +        }
1445 +
1446 +        if (vmaw->md5csum) {
1447 +            g_checksum_free(vmaw->md5csum);
1448 +        }
1449 +
1450 +        g_free(vmaw);
1451 +    }
1452 +
1453 +    return NULL;
1454 +}
1455 +
1456 +static int coroutine_fn vma_write_header(VmaWriter *vmaw)
1457 +{
1458 +    assert(vmaw);
1459 +    int header_clusters = 8;
1460 +    char buf[65536*header_clusters];
1461 +    VmaHeader *head = (VmaHeader *)buf;
1462 +
1463 +    int i;
1464 +
1465 +    DPRINTF("VMA WRITE HEADER\n");
1466 +
1467 +    if (vmaw->status < 0) {
1468 +        return vmaw->status;
1469 +    }
1470 +
1471 +    memset(buf, 0, sizeof(buf));
1472 +
1473 +    head->magic = VMA_MAGIC;
1474 +    head->version = GUINT32_TO_BE(1); /* v1 */
1475 +    memcpy(head->uuid, vmaw->uuid, 16);
1476 +
1477 +    time_t ctime = time(NULL);
1478 +    head->ctime = GUINT64_TO_BE(ctime);
1479 +
1480 +    if (!vmaw->stream_count) {
1481 +        return -1;
1482 +    }
1483 +
1484 +    for (i = 0; i < VMA_MAX_CONFIGS; i++) {
1485 +        head->config_names[i] = GUINT32_TO_BE(vmaw->config_names[i]);
1486 +        head->config_data[i] = GUINT32_TO_BE(vmaw->config_data[i]);
1487 +    }
1488 +
1489 +    /* 32 bytes per device (12 used currently) = 8192 bytes max */
1490 +    for (i = 1; i <= 254; i++) {
1491 +        VmaStreamInfo *si = &vmaw->stream_info[i];
1492 +        if (si->size) {
1493 +            assert(si->devname);
1494 +            uint32_t devname_ptr = allocate_header_string(vmaw, si->devname);
1495 +            if (!devname_ptr) {
1496 +                return -1;
1497 +            }
1498 +            head->dev_info[i].devname_ptr = GUINT32_TO_BE(devname_ptr);
1499 +            head->dev_info[i].size = GUINT64_TO_BE(si->size);
1500 +        }
1501 +    }
1502 +
1503 +    uint32_t header_size = sizeof(VmaHeader) + vmaw->header_blob_table_size;
1504 +    head->header_size = GUINT32_TO_BE(header_size);
1505 +
1506 +    if (header_size > sizeof(buf)) {
1507 +        return -1; /* just to be sure */
1508 +    }
1509 +
1510 +    uint32_t blob_buffer_offset = sizeof(VmaHeader);
1511 +    memcpy(buf + blob_buffer_offset, vmaw->header_blob_table,
1512 +           vmaw->header_blob_table_size);
1513 +    head->blob_buffer_offset = GUINT32_TO_BE(blob_buffer_offset);
1514 +    head->blob_buffer_size = GUINT32_TO_BE(vmaw->header_blob_table_pos);
1515 +
1516 +    g_checksum_reset(vmaw->md5csum);
1517 +    g_checksum_update(vmaw->md5csum, (const guchar *)buf, header_size);
1518 +    gsize csize = 16;
1519 +    g_checksum_get_digest(vmaw->md5csum, (guint8 *)(head->md5sum), &csize);
1520 +
1521 +    return vma_queue_write(vmaw, buf, header_size);
1522 +}
1523 +
1524 +static int coroutine_fn vma_writer_flush(VmaWriter *vmaw)
1525 +{
1526 +    assert(vmaw);
1527 +
1528 +    int ret;
1529 +    int i;
1530 +
1531 +    if (vmaw->status < 0) {
1532 +        return vmaw->status;
1533 +    }
1534 +
1535 +    if (!vmaw->header_written) {
1536 +        vmaw->header_written = true;
1537 +        ret = vma_write_header(vmaw);
1538 +        if (ret < 0) {
1539 +            vma_writer_set_error(vmaw, "vma_writer_flush: write header failed");
1540 +            return ret;
1541 +        }
1542 +    }
1543 +
1544 +    DPRINTF("VMA WRITE FLUSH %d %d\n", vmaw->outbuf_count, vmaw->outbuf_pos);
1545 +
1546 +
1547 +    VmaExtentHeader *ehead = (VmaExtentHeader *)vmaw->outbuf;
1548 +
1549 +    ehead->magic = VMA_EXTENT_MAGIC;
1550 +    ehead->reserved1 = 0;
1551 +
1552 +    for (i = 0; i < VMA_BLOCKS_PER_EXTENT; i++) {
1553 +        ehead->blockinfo[i] = GUINT64_TO_BE(vmaw->outbuf_block_info[i]);
1554 +    }
1555 +
1556 +    guint16 block_count = (vmaw->outbuf_pos - VMA_EXTENT_HEADER_SIZE) /
1557 +        VMA_BLOCK_SIZE;
1558 +
1559 +    ehead->block_count = GUINT16_TO_BE(block_count);
1560 +
1561 +    memcpy(ehead->uuid, vmaw->uuid, sizeof(ehead->uuid));
1562 +    memset(ehead->md5sum, 0, sizeof(ehead->md5sum));
1563 +
1564 +    g_checksum_reset(vmaw->md5csum);
1565 +    g_checksum_update(vmaw->md5csum, vmaw->outbuf, VMA_EXTENT_HEADER_SIZE);
1566 +    gsize csize = 16;
1567 +    g_checksum_get_digest(vmaw->md5csum, ehead->md5sum, &csize);
1568 +
1569 +    int bytes = vmaw->outbuf_pos;
1570 +    ret = vma_queue_write(vmaw, vmaw->outbuf, bytes);
1571 +    if (ret != bytes) {
1572 +        vma_writer_set_error(vmaw, "vma_writer_flush: failed write");
1573 +    }
1574 +
1575 +    vmaw->outbuf_count = 0;
1576 +    vmaw->outbuf_pos = VMA_EXTENT_HEADER_SIZE;
1577 +
1578 +    for (i = 0; i < VMA_BLOCKS_PER_EXTENT; i++) {
1579 +        vmaw->outbuf_block_info[i] = 0;
1580 +    }
1581 +
1582 +    return vmaw->status;
1583 +}
1584 +
1585 +static int vma_count_open_streams(VmaWriter *vmaw)
1586 +{
1587 +    g_assert(vmaw != NULL);
1588 +
1589 +    int i;
1590 +    int open_drives = 0;
1591 +    for (i = 0; i <= 255; i++) {
1592 +        if (vmaw->stream_info[i].size && !vmaw->stream_info[i].finished) {
1593 +            open_drives++;
1594 +        }
1595 +    }
1596 +
1597 +    return open_drives;
1598 +}
1599 +
1600 +/**
1601 + * all jobs should call this when there is no more data
1602 + * Returns: number of remaining stream (0 ==> finished)
1603 + */
1604 +int coroutine_fn
1605 +vma_writer_close_stream(VmaWriter *vmaw, uint8_t dev_id)
1606 +{
1607 +    g_assert(vmaw != NULL);
1608 +
1609 +    DPRINTF("vma_writer_set_status %d\n", dev_id);
1610 +    if (!vmaw->stream_info[dev_id].size) {
1611 +        vma_writer_set_error(vmaw, "vma_writer_close_stream: "
1612 +                             "no such stream %d", dev_id);
1613 +        return -1;
1614 +    }
1615 +    if (vmaw->stream_info[dev_id].finished) {
1616 +        vma_writer_set_error(vmaw, "vma_writer_close_stream: "
1617 +                             "stream already closed %d", dev_id);
1618 +        return -1;
1619 +    }
1620 +
1621 +    vmaw->stream_info[dev_id].finished = true;
1622 +
1623 +    int open_drives = vma_count_open_streams(vmaw);
1624 +
1625 +    if (open_drives <= 0) {
1626 +        DPRINTF("vma_writer_set_status all drives completed\n");
1627 +        qemu_co_mutex_lock(&vmaw->flush_lock);
1628 +        int ret = vma_writer_flush(vmaw);
1629 +        qemu_co_mutex_unlock(&vmaw->flush_lock);
1630 +        if (ret < 0) {
1631 +            vma_writer_set_error(vmaw, "vma_writer_close_stream: flush failed");
1632 +        }
1633 +    }
1634 +
1635 +    return open_drives;
1636 +}
1637 +
1638 +int vma_writer_get_status(VmaWriter *vmaw, VmaStatus *status)
1639 +{
1640 +    int i;
1641 +
1642 +    g_assert(vmaw != NULL);
1643 +
1644 +    if (status) {
1645 +        status->status = vmaw->status;
1646 +        g_strlcpy(status->errmsg, vmaw->errmsg, sizeof(status->errmsg));
1647 +        for (i = 0; i <= 255; i++) {
1648 +            status->stream_info[i] = vmaw->stream_info[i];
1649 +        }
1650 +
1651 +        uuid_unparse_lower(vmaw->uuid, status->uuid_str);
1652 +    }
1653 +
1654 +    status->closed = vmaw->closed;
1655 +
1656 +    return vmaw->status;
1657 +}
1658 +
1659 +static int vma_writer_get_buffer(VmaWriter *vmaw)
1660 +{
1661 +    int ret = 0;
1662 +
1663 +    qemu_co_mutex_lock(&vmaw->flush_lock);
1664 +
1665 +    /* wait until buffer is available */
1666 +    while (vmaw->outbuf_count >= (VMA_BLOCKS_PER_EXTENT - 1)) {
1667 +        ret = vma_writer_flush(vmaw);
1668 +        if (ret < 0) {
1669 +            vma_writer_set_error(vmaw, "vma_writer_get_buffer: flush failed");
1670 +            break;
1671 +        }
1672 +    }
1673 +
1674 +    qemu_co_mutex_unlock(&vmaw->flush_lock);
1675 +
1676 +    return ret;
1677 +}
1678 +
1679 +
1680 +int64_t coroutine_fn
1681 +vma_writer_write(VmaWriter *vmaw, uint8_t dev_id, int64_t cluster_num,
1682 +                 unsigned char *buf, size_t *zero_bytes)
1683 +{
1684 +    g_assert(vmaw != NULL);
1685 +    g_assert(zero_bytes != NULL);
1686 +
1687 +    *zero_bytes = 0;
1688 +
1689 +    if (vmaw->status < 0) {
1690 +        return vmaw->status;
1691 +    }
1692 +
1693 +    if (!dev_id || !vmaw->stream_info[dev_id].size) {
1694 +        vma_writer_set_error(vmaw, "vma_writer_write: "
1695 +                             "no such stream %d", dev_id);
1696 +        return -1;
1697 +    }
1698 +
1699 +    if (vmaw->stream_info[dev_id].finished) {
1700 +        vma_writer_set_error(vmaw, "vma_writer_write: "
1701 +                             "stream already closed %d", dev_id);
1702 +        return -1;
1703 +    }
1704 +
1705 +
1706 +    if (cluster_num >= (((uint64_t)1)<<32)) {
1707 +        vma_writer_set_error(vmaw, "vma_writer_write: "
1708 +                             "cluster number out of range");
1709 +        return -1;
1710 +    }
1711 +
1712 +    if (dev_id == vmaw->vmstate_stream) {
1713 +        if (cluster_num != vmaw->vmstate_clusters) {
1714 +            vma_writer_set_error(vmaw, "vma_writer_write: "
1715 +                                 "non sequential vmstate write");
1716 +        }
1717 +        vmaw->vmstate_clusters++;
1718 +    } else if (cluster_num >= vmaw->stream_info[dev_id].cluster_count) {
1719 +        vma_writer_set_error(vmaw, "vma_writer_write: cluster number too big");
1720 +        return -1;
1721 +    }
1722 +
1723 +    /* wait until buffer is available */
1724 +    if (vma_writer_get_buffer(vmaw) < 0) {
1725 +        vma_writer_set_error(vmaw, "vma_writer_write: "
1726 +                             "vma_writer_get_buffer failed");
1727 +        return -1;
1728 +    }
1729 +
1730 +    DPRINTF("VMA WRITE %d %zd\n", dev_id, cluster_num);
1731 +
1732 +    uint16_t mask = 0;
1733 +
1734 +    if (buf) {
1735 +        int i;
1736 +        int bit = 1;
1737 +        for (i = 0; i < 16; i++) {
1738 +            unsigned char *vmablock = buf + (i*VMA_BLOCK_SIZE);
1739 +            if (!buffer_is_zero(vmablock, VMA_BLOCK_SIZE)) {
1740 +                mask |= bit;
1741 +                memcpy(vmaw->outbuf + vmaw->outbuf_pos, vmablock,
1742 +                       VMA_BLOCK_SIZE);
1743 +                vmaw->outbuf_pos += VMA_BLOCK_SIZE;
1744 +            } else {
1745 +                DPRINTF("VMA WRITE %zd ZERO BLOCK %d\n", cluster_num, i);
1746 +                vmaw->stream_info[dev_id].zero_bytes += VMA_BLOCK_SIZE;
1747 +                *zero_bytes += VMA_BLOCK_SIZE;
1748 +            }
1749 +
1750 +            bit = bit << 1;
1751 +        }
1752 +    } else {
1753 +        DPRINTF("VMA WRITE %zd ZERO CLUSTER\n", cluster_num);
1754 +        vmaw->stream_info[dev_id].zero_bytes += VMA_CLUSTER_SIZE;
1755 +        *zero_bytes += VMA_CLUSTER_SIZE;
1756 +    }
1757 +
1758 +    uint64_t block_info = ((uint64_t)mask) << (32+16);
1759 +    block_info |= ((uint64_t)dev_id) << 32;
1760 +    block_info |= (cluster_num & 0xffffffff);
1761 +    vmaw->outbuf_block_info[vmaw->outbuf_count] = block_info;
1762 +
1763 +    DPRINTF("VMA WRITE MASK %zd %zx\n", cluster_num, block_info);
1764 +
1765 +    vmaw->outbuf_count++;
1766 +
1767 +    /** NOTE: We allways write whole clusters, but we correctly set
1768 +     * transferred bytes. So transferred == size when when everything
1769 +     * went OK.
1770 +     */
1771 +    size_t transferred = VMA_CLUSTER_SIZE;
1772 +
1773 +    if (dev_id != vmaw->vmstate_stream) {
1774 +        uint64_t last = (cluster_num + 1) * VMA_CLUSTER_SIZE;
1775 +        if (last > vmaw->stream_info[dev_id].size) {
1776 +            uint64_t diff = last - vmaw->stream_info[dev_id].size;
1777 +            if (diff >= VMA_CLUSTER_SIZE) {
1778 +                vma_writer_set_error(vmaw, "vma_writer_write: "
1779 +                                     "read after last cluster");
1780 +                return -1;
1781 +            }
1782 +            transferred -= diff;
1783 +        }
1784 +    }
1785 +
1786 +    vmaw->stream_info[dev_id].transferred += transferred;
1787 +
1788 +    return transferred;
1789 +}
1790 +
1791 +int vma_writer_close(VmaWriter *vmaw, Error **errp)
1792 +{
1793 +    g_assert(vmaw != NULL);
1794 +
1795 +    int i;
1796 +
1797 +    vma_queue_flush(vmaw);
1798 +
1799 +    /* this should not happen - just to be sure */
1800 +    while (!qemu_co_queue_empty(&vmaw->wqueue)) {
1801 +        DPRINTF("vma_writer_close wait\n");
1802 +        co_sleep_ns(QEMU_CLOCK_REALTIME, 1000000);
1803 +    }
1804 +
1805 +    if (vmaw->cmd) {
1806 +        if (pclose(vmaw->cmd) < 0) {
1807 +            vma_writer_set_error(vmaw, "vma_writer_close: "
1808 +                                 "pclose failed - %s", g_strerror(errno));
1809 +        }
1810 +    } else {
1811 +        if (close(vmaw->fd) < 0) {
1812 +            vma_writer_set_error(vmaw, "vma_writer_close: "
1813 +                                 "close failed - %s", g_strerror(errno));
1814 +        }
1815 +    }
1816 +
1817 +    for (i = 0; i <= 255; i++) {
1818 +        VmaStreamInfo *si = &vmaw->stream_info[i];
1819 +        if (si->size) {
1820 +            if (!si->finished) {
1821 +                vma_writer_set_error(vmaw, "vma_writer_close: "
1822 +                                     "detected open stream '%s'", si->devname);
1823 +            } else if ((si->transferred != si->size) &&
1824 +                       (i != vmaw->vmstate_stream)) {
1825 +                vma_writer_set_error(vmaw, "vma_writer_close: "
1826 +                                     "incomplete stream '%s' (%zd != %zd)",
1827 +                                     si->devname, si->transferred, si->size);
1828 +            }
1829 +        }
1830 +    }
1831 +
1832 +    for (i = 0; i <= 255; i++) {
1833 +        vmaw->stream_info[i].finished = 1; /* mark as closed */
1834 +    }
1835 +
1836 +    vmaw->closed = 1;
1837 +
1838 +    if (vmaw->status < 0 && *errp == NULL) {
1839 +        error_setg(errp, "%s", vmaw->errmsg);
1840 +    }
1841 +
1842 +    return vmaw->status;
1843 +}
1844 +
1845 +void vma_writer_destroy(VmaWriter *vmaw)
1846 +{
1847 +    assert(vmaw);
1848 +
1849 +    int i;
1850 +
1851 +    for (i = 0; i <= 255; i++) {
1852 +        if (vmaw->stream_info[i].devname) {
1853 +            g_free(vmaw->stream_info[i].devname);
1854 +        }
1855 +    }
1856 +
1857 +    if (vmaw->md5csum) {
1858 +        g_checksum_free(vmaw->md5csum);
1859 +    }
1860 +
1861 +    for (i = 0; i < WRITE_BUFFERS; i++) {
1862 +        free(vmaw->aiocbs[i]);
1863 +    }
1864 +
1865 +    g_free(vmaw);
1866 +}
1867 Index: new/vma.c
1868 ===================================================================
1869 --- /dev/null   1970-01-01 00:00:00.000000000 +0000
1870 +++ new/vma.c   2013-12-04 12:07:25.000000000 +0100
1871 @@ -0,0 +1,579 @@
1872 +/*
1873 + * VMA: Virtual Machine Archive
1874 + *
1875 + * Copyright (C) 2012-2013 Proxmox Server Solutions
1876 + *
1877 + * Authors:
1878 + *  Dietmar Maurer (dietmar@proxmox.com)
1879 + *
1880 + * This work is licensed under the terms of the GNU GPL, version 2 or later.
1881 + * See the COPYING file in the top-level directory.
1882 + *
1883 + */
1884 +
1885 +#include <stdio.h>
1886 +#include <errno.h>
1887 +#include <unistd.h>
1888 +#include <stdio.h>
1889 +#include <string.h>
1890 +#include <sys/types.h>
1891 +#include <sys/stat.h>
1892 +#include <fcntl.h>
1893 +#include <glib.h>
1894 +
1895 +#include "vma.h"
1896 +#include "qemu-common.h"
1897 +#include "qemu/error-report.h"
1898 +#include "qemu/main-loop.h"
1899 +
1900 +static void help(void)
1901 +{
1902 +    const char *help_msg =
1903 +        "usage: vma command [command options]\n"
1904 +        "\n"
1905 +        "vma list <filename>\n"
1906 +        "vma create <filename> [-c config] <archive> pathname ...\n"
1907 +        "vma extract <filename> [-r <fifo>] <targetdir>\n"
1908 +        ;
1909 +
1910 +    printf("%s", help_msg);
1911 +    exit(1);
1912 +}
1913 +
1914 +static const char *extract_devname(const char *path, char **devname, int index)
1915 +{
1916 +    assert(path);
1917 +
1918 +    const char *sep = strchr(path, '=');
1919 +
1920 +    if (sep) {
1921 +        *devname = g_strndup(path, sep - path);
1922 +        path = sep + 1;
1923 +    } else {
1924 +        if (index >= 0) {
1925 +            *devname = g_strdup_printf("disk%d", index);
1926 +        } else {
1927 +            *devname = NULL;
1928 +        }
1929 +    }
1930 +
1931 +    return path;
1932 +}
1933 +
1934 +static void print_content(VmaReader *vmar)
1935 +{
1936 +    assert(vmar);
1937 +
1938 +    VmaHeader *head = vma_reader_get_header(vmar);
1939 +
1940 +    GList *l = vma_reader_get_config_data(vmar);
1941 +    while (l && l->data) {
1942 +        VmaConfigData *cdata = (VmaConfigData *)l->data;
1943 +        l = g_list_next(l);
1944 +        printf("CFG: size: %d name: %s\n", cdata->len, cdata->name);
1945 +    }
1946 +
1947 +    int i;
1948 +    VmaDeviceInfo *di;
1949 +    for (i = 1; i < 255; i++) {
1950 +        di = vma_reader_get_device_info(vmar, i);
1951 +        if (di) {
1952 +            if (strcmp(di->devname, "vmstate") == 0) {
1953 +                printf("VMSTATE: dev_id=%d memory: %zd\n", i, di->size);
1954 +            } else {
1955 +                printf("DEV: dev_id=%d size: %zd devname: %s\n",
1956 +                       i, di->size, di->devname);
1957 +            }
1958 +        }
1959 +    }
1960 +    /* ctime is the last entry we print */
1961 +    printf("CTIME: %s", ctime(&head->ctime));
1962 +    fflush(stdout);
1963 +}
1964 +
1965 +static int list_content(int argc, char **argv)
1966 +{
1967 +    int c, ret = 0;
1968 +    const char *filename;
1969 +
1970 +    for (;;) {
1971 +        c = getopt(argc, argv, "h");
1972 +        if (c == -1) {
1973 +            break;
1974 +        }
1975 +        switch (c) {
1976 +        case '?':
1977 +        case 'h':
1978 +            help();
1979 +            break;
1980 +        default:
1981 +            g_assert_not_reached();
1982 +        }
1983 +    }
1984 +
1985 +    /* Get the filename */
1986 +    if ((optind + 1) != argc) {
1987 +        help();
1988 +    }
1989 +    filename = argv[optind++];
1990 +
1991 +    Error *errp = NULL;
1992 +    VmaReader *vmar = vma_reader_create(filename, &errp);
1993 +
1994 +    if (!vmar) {
1995 +        g_error("%s", error_get_pretty(errp));
1996 +    }
1997 +
1998 +    print_content(vmar);
1999 +
2000 +    vma_reader_destroy(vmar);
2001 +
2002 +    return ret;
2003 +}
2004 +
2005 +typedef struct RestoreMap {
2006 +    char *devname;
2007 +    char *path;
2008 +    bool write_zero;
2009 +} RestoreMap;
2010 +
2011 +static int extract_content(int argc, char **argv)
2012 +{
2013 +    int c, ret = 0;
2014 +    int verbose = 0;
2015 +    const char *filename;
2016 +    const char *dirname;
2017 +    const char *readmap = NULL;
2018 +
2019 +    for (;;) {
2020 +        c = getopt(argc, argv, "hvr:");
2021 +        if (c == -1) {
2022 +            break;
2023 +        }
2024 +        switch (c) {
2025 +        case '?':
2026 +        case 'h':
2027 +            help();
2028 +            break;
2029 +        case 'r':
2030 +            readmap = optarg;
2031 +            break;
2032 +        case 'v':
2033 +            verbose = 1;
2034 +            break;
2035 +        default:
2036 +            help();
2037 +        }
2038 +    }
2039 +
2040 +    /* Get the filename */
2041 +    if ((optind + 2) != argc) {
2042 +        help();
2043 +    }
2044 +    filename = argv[optind++];
2045 +    dirname = argv[optind++];
2046 +
2047 +    Error *errp = NULL;
2048 +    VmaReader *vmar = vma_reader_create(filename, &errp);
2049 +
2050 +    if (!vmar) {
2051 +        g_error("%s", error_get_pretty(errp));
2052 +    }
2053 +
2054 +    if (mkdir(dirname, 0777) < 0) {
2055 +        g_error("unable to create target directory %s - %s",
2056 +                dirname, g_strerror(errno));
2057 +    }
2058 +
2059 +    GList *l = vma_reader_get_config_data(vmar);
2060 +    while (l && l->data) {
2061 +        VmaConfigData *cdata = (VmaConfigData *)l->data;
2062 +        l = g_list_next(l);
2063 +        char *cfgfn = g_strdup_printf("%s/%s", dirname, cdata->name);
2064 +        GError *err = NULL;
2065 +        if (!g_file_set_contents(cfgfn, (gchar *)cdata->data, cdata->len,
2066 +                                 &err)) {
2067 +            g_error("unable to write file: %s", err->message);
2068 +        }
2069 +    }
2070 +
2071 +    GHashTable *devmap = g_hash_table_new(g_str_hash, g_str_equal);
2072 +
2073 +    if (readmap) {
2074 +        print_content(vmar);
2075 +
2076 +        FILE *map = fopen(readmap, "r");
2077 +        if (!map) {
2078 +            g_error("unable to open fifo %s - %s", readmap, g_strerror(errno));
2079 +        }
2080 +
2081 +        while (1) {
2082 +            char inbuf[8192];
2083 +            char *line = fgets(inbuf, sizeof(inbuf), map);
2084 +            if (!line || line[0] == '\0' || !strcmp(line, "done\n")) {
2085 +                break;
2086 +            }
2087 +            int len = strlen(line);
2088 +            if (line[len - 1] == '\n') {
2089 +                line[len - 1] = '\0';
2090 +                if (len == 1) {
2091 +                    break;
2092 +                }
2093 +            }
2094 +
2095 +            const char *path;
2096 +            bool write_zero;
2097 +            if (line[0] == '0' && line[1] == ':') {
2098 +                path = inbuf + 2;
2099 +                write_zero = false;
2100 +            } else if (line[0] == '1' && line[1] == ':') {
2101 +                path = inbuf + 2;
2102 +                write_zero = true;
2103 +            } else {
2104 +                g_error("read map failed - parse error ('%s')", inbuf);
2105 +            }
2106 +
2107 +            char *devname = NULL;
2108 +            path = extract_devname(path, &devname, -1);
2109 +            if (!devname) {
2110 +                g_error("read map failed - no dev name specified ('%s')",
2111 +                        inbuf);
2112 +            }
2113 +
2114 +            RestoreMap *map = g_new0(RestoreMap, 1);
2115 +            map->devname = g_strdup(devname);
2116 +            map->path = g_strdup(path);
2117 +            map->write_zero = write_zero;
2118 +
2119 +            g_hash_table_insert(devmap, map->devname, map);
2120 +
2121 +        };
2122 +    }
2123 +
2124 +    int i;
2125 +    int vmstate_fd = -1;
2126 +    guint8 vmstate_stream = 0;
2127 +
2128 +    for (i = 1; i < 255; i++) {
2129 +        VmaDeviceInfo *di = vma_reader_get_device_info(vmar, i);
2130 +        if (di && (strcmp(di->devname, "vmstate") == 0)) {
2131 +            vmstate_stream = i;
2132 +            char *statefn = g_strdup_printf("%s/vmstate.bin", dirname);
2133 +            vmstate_fd = open(statefn, O_WRONLY|O_CREAT|O_EXCL, 0644);
2134 +            if (vmstate_fd < 0) {
2135 +                g_error("create vmstate file '%s' failed - %s", statefn,
2136 +                        g_strerror(errno));
2137 +            }
2138 +            g_free(statefn);
2139 +        } else if (di) {
2140 +            char *devfn = NULL;
2141 +            int flags = BDRV_O_RDWR|BDRV_O_CACHE_WB;
2142 +            bool write_zero = true;
2143 +
2144 +            if (readmap) {
2145 +                RestoreMap *map;
2146 +                map = (RestoreMap *)g_hash_table_lookup(devmap, di->devname);
2147 +                if (map == NULL) {
2148 +                    g_error("no device name mapping for %s", di->devname);
2149 +                }
2150 +                devfn = map->path;
2151 +                write_zero = map->write_zero;
2152 +            } else {
2153 +                devfn = g_strdup_printf("%s/tmp-disk-%s.raw",
2154 +                                        dirname, di->devname);
2155 +                printf("DEVINFO %s %zd\n", devfn, di->size);
2156 +
2157 +                bdrv_img_create(devfn, "raw", NULL, NULL, NULL, di->size,
2158 +                                flags, &errp, 0);
2159 +                if (error_is_set(&errp)) {
2160 +                    g_error("can't create file %s: %s", devfn,
2161 +                            error_get_pretty(errp));
2162 +                }
2163 +
2164 +                /* Note: we created an empty file above, so there is no
2165 +                 * need to write zeroes (so we generate a sparse file)
2166 +                 */
2167 +                write_zero = false;
2168 +            }
2169 +
2170 +            BlockDriverState *bs = bdrv_new(di->devname);
2171 +            if (bdrv_open(bs, devfn, NULL, flags, NULL, &errp)) {
2172 +                g_error("can't open file %s - %s", devfn,
2173 +                        error_get_pretty(errp));
2174 +            }
2175 +            if (vma_reader_register_bs(vmar, i, bs, write_zero, &errp) < 0) {
2176 +                g_error("%s", error_get_pretty(errp));
2177 +            }
2178 +
2179 +            if (!readmap) {
2180 +                g_free(devfn);
2181 +            }
2182 +        }
2183 +    }
2184 +
2185 +    if (vma_reader_restore(vmar, vmstate_fd, verbose, &errp) < 0) {
2186 +        g_error("restore failed - %s", error_get_pretty(errp));
2187 +    }
2188 +
2189 +    if (!readmap) {
2190 +        for (i = 1; i < 255; i++) {
2191 +            VmaDeviceInfo *di = vma_reader_get_device_info(vmar, i);
2192 +            if (di && (i != vmstate_stream)) {
2193 +                char *tmpfn = g_strdup_printf("%s/tmp-disk-%s.raw",
2194 +                                              dirname, di->devname);
2195 +                char *fn = g_strdup_printf("%s/disk-%s.raw",
2196 +                                           dirname, di->devname);
2197 +                if (rename(tmpfn, fn) != 0) {
2198 +                    g_error("rename %s to %s failed - %s",
2199 +                            tmpfn, fn, g_strerror(errno));
2200 +                }
2201 +            }
2202 +        }
2203 +    }
2204 +
2205 +    vma_reader_destroy(vmar);
2206 +
2207 +    bdrv_close_all();
2208 +
2209 +    return ret;
2210 +}
2211 +
2212 +typedef struct BackupJob {
2213 +    BlockDriverState *bs;
2214 +    int64_t len;
2215 +    VmaWriter *vmaw;
2216 +    uint8_t dev_id;
2217 +} BackupJob;
2218 +
2219 +#define BACKUP_SECTORS_PER_CLUSTER (VMA_CLUSTER_SIZE / BDRV_SECTOR_SIZE)
2220 +
2221 +static void coroutine_fn backup_run(void *opaque)
2222 +{
2223 +    BackupJob *job = (BackupJob *)opaque;
2224 +    struct iovec iov;
2225 +    QEMUIOVector qiov;
2226 +
2227 +    int64_t start, end;
2228 +    int ret = 0;
2229 +
2230 +    unsigned char *buf = qemu_blockalign(job->bs, VMA_CLUSTER_SIZE);
2231 +
2232 +    start = 0;
2233 +    end = DIV_ROUND_UP(job->len / BDRV_SECTOR_SIZE,
2234 +                       BACKUP_SECTORS_PER_CLUSTER);
2235 +
2236 +    for (; start < end; start++) {
2237 +        iov.iov_base = buf;
2238 +        iov.iov_len = VMA_CLUSTER_SIZE;
2239 +        qemu_iovec_init_external(&qiov, &iov, 1);
2240 +
2241 +        ret = bdrv_co_readv(job->bs, start * BACKUP_SECTORS_PER_CLUSTER,
2242 +                            BACKUP_SECTORS_PER_CLUSTER, &qiov);
2243 +        if (ret < 0) {
2244 +            vma_writer_set_error(job->vmaw, "read error", -1);
2245 +            goto out;
2246 +        }
2247 +
2248 +        size_t zb = 0;
2249 +        if (vma_writer_write(job->vmaw, job->dev_id, start, buf, &zb) < 0) {
2250 +            vma_writer_set_error(job->vmaw, "backup_dump_cb vma_writer_write failed", -1);
2251 +            goto out;
2252 +        }
2253 +    }
2254 +
2255 +
2256 +out:
2257 +    if (vma_writer_close_stream(job->vmaw, job->dev_id) <= 0) {
2258 +        Error *err = NULL;
2259 +        if (vma_writer_close(job->vmaw, &err) != 0) {
2260 +            g_warning("vma_writer_close failed %s", error_get_pretty(err));
2261 +        }
2262 +    }
2263 +}
2264 +
2265 +static int create_archive(int argc, char **argv)
2266 +{
2267 +    int i, c, res;
2268 +    int verbose = 0;
2269 +    const char *archivename;
2270 +    GList *config_files = NULL;
2271 +
2272 +    for (;;) {
2273 +        c = getopt(argc, argv, "hvc:");
2274 +        if (c == -1) {
2275 +            break;
2276 +        }
2277 +        switch (c) {
2278 +        case '?':
2279 +        case 'h':
2280 +            help();
2281 +            break;
2282 +        case 'c':
2283 +            config_files = g_list_append(config_files, optarg);
2284 +            break;
2285 +        case 'v':
2286 +            verbose = 1;
2287 +            break;
2288 +        default:
2289 +            g_assert_not_reached();
2290 +        }
2291 +    }
2292 +
2293 +
2294 +    /* make sure we have archive name and at least one path */
2295 +    if ((optind + 2) > argc) {
2296 +        help();
2297 +    }
2298 +
2299 +    archivename = argv[optind++];
2300 +
2301 +    uuid_t uuid;
2302 +    uuid_generate(uuid);
2303 +
2304 +    Error *local_err = NULL;
2305 +    VmaWriter *vmaw = vma_writer_create(archivename, uuid, &local_err);
2306 +
2307 +    if (vmaw == NULL) {
2308 +        g_error("%s", error_get_pretty(local_err));
2309 +    }
2310 +
2311 +    GList *l = config_files;
2312 +    while (l && l->data) {
2313 +        char *name = l->data;
2314 +        char *cdata = NULL;
2315 +        gsize clen = 0;
2316 +        GError *err = NULL;
2317 +        if (!g_file_get_contents(name, &cdata, &clen, &err)) {
2318 +            unlink(archivename);
2319 +            g_error("Unable to read file: %s", err->message);
2320 +        }
2321 +
2322 +        if (vma_writer_add_config(vmaw, name, cdata, clen) != 0) {
2323 +            unlink(archivename);
2324 +            g_error("Unable to append config data %s (len = %zd)",
2325 +                    name, clen);
2326 +        }
2327 +        l = g_list_next(l);
2328 +    }
2329 +
2330 +    int ind = 0;
2331 +    while (optind < argc) {
2332 +        const char *path = argv[optind++];
2333 +        char *devname = NULL;
2334 +        path = extract_devname(path, &devname, ind++);
2335 +
2336 +        BlockDriver *drv = NULL;
2337 +        BlockDriverState *bs = bdrv_new(devname);
2338 +        Error *errp = NULL;
2339 +
2340 +        res = bdrv_open(bs, path, NULL, BDRV_O_CACHE_WB , drv, &errp);
2341 +        if (res < 0) {
2342 +            unlink(archivename);
2343 +            g_error("bdrv_open '%s' failed - %s", path, error_get_pretty(errp));
2344 +        }
2345 +        int64_t size = bdrv_getlength(bs);
2346 +        int dev_id = vma_writer_register_stream(vmaw, devname, size);
2347 +        if (dev_id <= 0) {
2348 +            unlink(archivename);
2349 +            g_error("vma_writer_register_stream '%s' failed", devname);
2350 +        }
2351 +
2352 +        BackupJob *job = g_new0(BackupJob, 1);
2353 +        job->len = size;
2354 +        job->bs = bs;
2355 +        job->vmaw = vmaw;
2356 +        job->dev_id = dev_id;
2357 +
2358 +        Coroutine *co = qemu_coroutine_create(backup_run);
2359 +        qemu_coroutine_enter(co, job);
2360 +    }
2361 +
2362 +    VmaStatus vmastat;
2363 +    int percent = 0;
2364 +    int last_percent = -1;
2365 +
2366 +    while (1) {
2367 +        main_loop_wait(false);
2368 +        vma_writer_get_status(vmaw, &vmastat);
2369 +
2370 +        if (verbose) {
2371 +
2372 +            uint64_t total = 0;
2373 +            uint64_t transferred = 0;
2374 +            uint64_t zero_bytes = 0;
2375 +
2376 +            int i;
2377 +            for (i = 0; i < 256; i++) {
2378 +                if (vmastat.stream_info[i].size) {
2379 +                    total += vmastat.stream_info[i].size;
2380 +                    transferred += vmastat.stream_info[i].transferred;
2381 +                    zero_bytes += vmastat.stream_info[i].zero_bytes;
2382 +                }
2383 +            }
2384 +            percent = (transferred*100)/total;
2385 +            if (percent != last_percent) {
2386 +                fprintf(stderr, "progress %d%% %zd/%zd %zd\n", percent,
2387 +                        transferred, total, zero_bytes);
2388 +                fflush(stderr);
2389 +
2390 +                last_percent = percent;
2391 +            }
2392 +        }
2393 +
2394 +        if (vmastat.closed) {
2395 +            break;
2396 +        }
2397 +    }
2398 +
2399 +    bdrv_drain_all();
2400 +
2401 +    vma_writer_get_status(vmaw, &vmastat);
2402 +
2403 +    if (verbose) {
2404 +        for (i = 0; i < 256; i++) {
2405 +            VmaStreamInfo *si = &vmastat.stream_info[i];
2406 +            if (si->size) {
2407 +                fprintf(stderr, "image %s: size=%zd zeros=%zd saved=%zd\n",
2408 +                        si->devname, si->size, si->zero_bytes,
2409 +                        si->size - si->zero_bytes);
2410 +            }
2411 +        }
2412 +    }
2413 +
2414 +    if (vmastat.status < 0) {
2415 +        unlink(archivename);
2416 +        g_error("creating vma archive failed");
2417 +    }
2418 +
2419 +    return 0;
2420 +}
2421 +
2422 +int main(int argc, char **argv)
2423 +{
2424 +    const char *cmdname;
2425 +
2426 +    error_set_progname(argv[0]);
2427 +
2428 +    qemu_init_main_loop();
2429 +
2430 +    bdrv_init();
2431 +
2432 +    if (argc < 2) {
2433 +        help();
2434 +    }
2435 +
2436 +    cmdname = argv[1];
2437 +    argc--; argv++;
2438 +
2439 +
2440 +    if (!strcmp(cmdname, "list")) {
2441 +        return list_content(argc, argv);
2442 +    } else if (!strcmp(cmdname, "create")) {
2443 +        return create_archive(argc, argv);
2444 +    } else if (!strcmp(cmdname, "extract")) {
2445 +        return extract_content(argc, argv);
2446 +    }
2447 +
2448 +    help();
2449 +    return 0;
2450 +}
2451 Index: new/vma.h
2452 ===================================================================
2453 --- /dev/null   1970-01-01 00:00:00.000000000 +0000
2454 +++ new/vma.h   2013-12-04 14:27:56.000000000 +0100
2455 @@ -0,0 +1,146 @@
2456 +/*
2457 + * VMA: Virtual Machine Archive
2458 + *
2459 + * Copyright (C) Proxmox Server Solutions
2460 + *
2461 + * Authors:
2462 + *  Dietmar Maurer (dietmar@proxmox.com)
2463 + *
2464 + * This work is licensed under the terms of the GNU GPL, version 2 or later.
2465 + * See the COPYING file in the top-level directory.
2466 + *
2467 + */
2468 +
2469 +#ifndef BACKUP_VMA_H
2470 +#define BACKUP_VMA_H
2471 +
2472 +#include <uuid/uuid.h>
2473 +#include "qapi/error.h"
2474 +#include "block/block.h"
2475 +
2476 +#define VMA_BLOCK_BITS 12
2477 +#define VMA_BLOCK_SIZE (1<<VMA_BLOCK_BITS)
2478 +#define VMA_CLUSTER_BITS (VMA_BLOCK_BITS+4)
2479 +#define VMA_CLUSTER_SIZE (1<<VMA_CLUSTER_BITS)
2480 +
2481 +#if VMA_CLUSTER_SIZE != 65536
2482 +#error unexpected cluster size
2483 +#endif
2484 +
2485 +#define VMA_EXTENT_HEADER_SIZE 512
2486 +#define VMA_BLOCKS_PER_EXTENT 59
2487 +#define VMA_MAX_CONFIGS 256
2488 +
2489 +#define VMA_MAX_EXTENT_SIZE \
2490 +    (VMA_EXTENT_HEADER_SIZE+VMA_CLUSTER_SIZE*VMA_BLOCKS_PER_EXTENT)
2491 +#if VMA_MAX_EXTENT_SIZE != 3867136
2492 +#error unexpected VMA_EXTENT_SIZE
2493 +#endif
2494 +
2495 +/* File Format Definitions */
2496 +
2497 +#define VMA_MAGIC (GUINT32_TO_BE(('V'<<24)|('M'<<16)|('A'<<8)|0x00))
2498 +#define VMA_EXTENT_MAGIC (GUINT32_TO_BE(('V'<<24)|('M'<<16)|('A'<<8)|'E'))
2499 +
2500 +typedef struct VmaDeviceInfoHeader {
2501 +    uint32_t devname_ptr; /* offset into blob_buffer table */
2502 +    uint32_t reserved0;
2503 +    uint64_t size; /* device size in bytes */
2504 +    uint64_t reserved1;
2505 +    uint64_t reserved2;
2506 +} VmaDeviceInfoHeader;
2507 +
2508 +typedef struct VmaHeader {
2509 +    uint32_t magic;
2510 +    uint32_t version;
2511 +    unsigned char uuid[16];
2512 +    int64_t ctime;
2513 +    unsigned char md5sum[16];
2514 +
2515 +    uint32_t blob_buffer_offset;
2516 +    uint32_t blob_buffer_size;
2517 +    uint32_t header_size;
2518 +
2519 +    unsigned char reserved[1984];
2520 +
2521 +    uint32_t config_names[VMA_MAX_CONFIGS]; /* offset into blob_buffer table */
2522 +    uint32_t config_data[VMA_MAX_CONFIGS];  /* offset into blob_buffer table */
2523 +
2524 +    uint32_t reserved1;
2525 +
2526 +    VmaDeviceInfoHeader dev_info[256];
2527 +} VmaHeader;
2528 +
2529 +typedef struct VmaExtentHeader {
2530 +    uint32_t magic;
2531 +    uint16_t reserved1;
2532 +    uint16_t block_count;
2533 +    unsigned char uuid[16];
2534 +    unsigned char md5sum[16];
2535 +    uint64_t blockinfo[VMA_BLOCKS_PER_EXTENT];
2536 +} VmaExtentHeader;
2537 +
2538 +/* functions/definitions to read/write vma files */
2539 +
2540 +typedef struct VmaReader VmaReader;
2541 +
2542 +typedef struct VmaWriter VmaWriter;
2543 +
2544 +typedef struct VmaConfigData {
2545 +    const char *name;
2546 +    const void *data;
2547 +    uint32_t len;
2548 +} VmaConfigData;
2549 +
2550 +typedef struct VmaStreamInfo {
2551 +    uint64_t size;
2552 +    uint64_t cluster_count;
2553 +    uint64_t transferred;
2554 +    uint64_t zero_bytes;
2555 +    int finished;
2556 +    char *devname;
2557 +} VmaStreamInfo;
2558 +
2559 +typedef struct VmaStatus {
2560 +    int status;
2561 +    bool closed;
2562 +    char errmsg[8192];
2563 +    char uuid_str[37];
2564 +    VmaStreamInfo stream_info[256];
2565 +} VmaStatus;
2566 +
2567 +typedef struct VmaDeviceInfo {
2568 +    uint64_t size; /* device size in bytes */
2569 +    const char *devname;
2570 +} VmaDeviceInfo;
2571 +
2572 +VmaWriter *vma_writer_create(const char *filename, uuid_t uuid, Error **errp);
2573 +int vma_writer_close(VmaWriter *vmaw, Error **errp);
2574 +void vma_writer_destroy(VmaWriter *vmaw);
2575 +int vma_writer_add_config(VmaWriter *vmaw, const char *name, gpointer data,
2576 +                          size_t len);
2577 +int vma_writer_register_stream(VmaWriter *vmaw, const char *devname,
2578 +                               size_t size);
2579 +
2580 +int64_t coroutine_fn vma_writer_write(VmaWriter *vmaw, uint8_t dev_id,
2581 +                                      int64_t cluster_num, unsigned char *buf,
2582 +                                      size_t *zero_bytes);
2583 +
2584 +int coroutine_fn vma_writer_close_stream(VmaWriter *vmaw, uint8_t dev_id);
2585 +
2586 +int vma_writer_get_status(VmaWriter *vmaw, VmaStatus *status);
2587 +void vma_writer_set_error(VmaWriter *vmaw, const char *fmt, ...);
2588 +
2589 +
2590 +VmaReader *vma_reader_create(const char *filename, Error **errp);
2591 +void vma_reader_destroy(VmaReader *vmar);
2592 +VmaHeader *vma_reader_get_header(VmaReader *vmar);
2593 +GList *vma_reader_get_config_data(VmaReader *vmar);
2594 +VmaDeviceInfo *vma_reader_get_device_info(VmaReader *vmar, guint8 dev_id);
2595 +int vma_reader_register_bs(VmaReader *vmar, guint8 dev_id,
2596 +                           BlockDriverState *bs, bool write_zeroes,
2597 +                           Error **errp);
2598 +int vma_reader_restore(VmaReader *vmar, int vmstate_fd, bool verbose,
2599 +                       Error **errp);
2600 +
2601 +#endif /* BACKUP_VMA_H */
2602 Index: new/Makefile.objs
2603 ===================================================================
2604 --- new.orig/Makefile.objs      2013-12-04 12:07:18.000000000 +0100
2605 +++ new/Makefile.objs   2013-12-04 12:07:25.000000000 +0100
2606 @@ -14,6 +14,7 @@
2607  block-obj-y += block/
2608  block-obj-y += qapi-types.o qapi-visit.o
2609  block-obj-y += qemu-io-cmds.o
2610 +block-obj-y += vma-writer.o
2611
2612  block-obj-y += qemu-coroutine.o qemu-coroutine-lock.o qemu-coroutine-io.o
2613  block-obj-y += qemu-coroutine-sleep.o