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

   1 From ff878b16c010752dd46a474329d523cded78a004 Mon Sep 17 00:00:00 2001
   2 From: Dietmar Maurer <dietmar@proxmox.com>
   3 Date: Tue, 13 Nov 2012 11:11:38 +0100
   4 Subject: [PATCH v3 4/6] introduce new vma archive format
   5
   6 This is a very simple archive format, see docs/specs/vma_spec.txt
   7
   8 Signed-off-by: Dietmar Maurer <dietmar@proxmox.com>
   9 ---
  10  Makefile                |    3 +-
  11  Makefile.objs           |    2 +-
  12  blockdev.c              |    6 +-
  13  docs/specs/vma_spec.txt |   24 ++
  14  vma-reader.c            |  772 ++++++++++++++++++++++++++++++++++++++++
  15  vma-writer.c            |  900 +++++++++++++++++++++++++++++++++++++++++++++++
  16  vma.c                   |  550 +++++++++++++++++++++++++++++
  17  vma.h                   |  145 ++++++++
  18  8 files changed, 2398 insertions(+), 4 deletions(-)
  19  create mode 100644 docs/specs/vma_spec.txt
  20  create mode 100644 vma-reader.c
  21  create mode 100644 vma-writer.c
  22  create mode 100644 vma.c
  23  create mode 100644 vma.h
  24
  25 diff --git a/Makefile b/Makefile
  26 index 9ecbcbb..30a9268 100644
  27 --- a/Makefile
  28 +++ b/Makefile
  29 @@ -100,7 +100,7 @@ defconfig:
  30
  31  -include config-all-devices.mak
  32
  33 -all: $(DOCS) $(TOOLS) $(HELPERS-y) recurse-all
  34 +all: $(DOCS) $(TOOLS) vma$(EXESUF) $(HELPERS-y) recurse-all
  35
  36  config-host.h: config-host.h-timestamp
  37  config-host.h-timestamp: config-host.mak
  38 @@ -194,6 +194,7 @@ tools-obj-$(CONFIG_POSIX) += compatfd.o
  39  qemu-img$(EXESUF): qemu-img.o $(tools-obj-y) $(block-obj-y) libqemustub.a
  40  qemu-nbd$(EXESUF): qemu-nbd.o $(tools-obj-y) $(block-obj-y) libqemustub.a
  41  qemu-io$(EXESUF): qemu-io.o cmd.o $(tools-obj-y) $(block-obj-y) libqemustub.a
  42 +vma$(EXESUF): vma.o vma-writer.o vma-reader.o $(tools-obj-y) $(block-obj-y) libqemustub.a
  43
  44  qemu-bridge-helper$(EXESUF): qemu-bridge-helper.o
  45
  46 diff --git a/Makefile.objs b/Makefile.objs
  47 index cb46be5..b5732e2 100644
  48 --- a/Makefile.objs
  49 +++ b/Makefile.objs
  50 @@ -48,7 +48,7 @@ coroutine-obj-$(CONFIG_WIN32) += coroutine-win32.o
  51  block-obj-y = iov.o cache-utils.o qemu-option.o module.o async.o
  52  block-obj-y += nbd.o block.o blockjob.o aes.o qemu-config.o
  53  block-obj-y += thread-pool.o qemu-progress.o qemu-sockets.o uri.o notify.o
  54 -block-obj-y += backup.o
  55 +block-obj-y += vma-writer.o backup.o
  56  block-obj-y += $(coroutine-obj-y) $(qobject-obj-y) $(version-obj-y)
  57  block-obj-$(CONFIG_POSIX) += event_notifier-posix.o aio-posix.o
  58  block-obj-$(CONFIG_WIN32) += event_notifier-win32.o aio-win32.o
  59 diff --git a/blockdev.c b/blockdev.c
  60 index c635d21..f424933 100644
  61 --- a/blockdev.c
  62 +++ b/blockdev.c
  63 @@ -21,6 +21,7 @@
  64  #include "trace.h"
  65  #include "arch_init.h"
  66  #include "backup.h"
  67 +#include "vma.h"
  68
  69  static QTAILQ_HEAD(drivelist, DriveInfo) drives = QTAILQ_HEAD_INITIALIZER(drives);
  70
  71 @@ -1483,10 +1484,11 @@ char *qmp_backup(const char *backupfile, bool has_format, const char *format,
  72      /* Todo: try to auto-detect format based on file name */
  73      format = has_format ? format : "vma";
  74
  75 -    /* fixme: find driver for specifued format */
  76      const BackupDriver *driver = NULL;
  77
  78 -    if (!driver) {
  79 +    if (strcmp(format, "vma") == 0) {
  80 +        driver = &backup_vma_driver;
  81 +    } else {
  82          error_set(errp, ERROR_CLASS_GENERIC_ERROR,
  83                    "no backup driver for format '%s'", format);
  84          return NULL;
  85 diff --git a/docs/specs/vma_spec.txt b/docs/specs/vma_spec.txt
  86 new file mode 100644
  87 index 0000000..052c629
  88 --- /dev/null
  89 +++ b/docs/specs/vma_spec.txt
  90 @@ -0,0 +1,24 @@
  91 +=Virtual Machine Archive format (VMA)=
  92 +
  93 +This format contains a header which includes the VM configuration as
  94 +binary blobs, and a list of devices (dev_id, name).
  95 +
  96 +The actual VM image data is stored inside extents. An extent contains
  97 +up to 64 clusters, and start with a 512 byte header containing
  98 +additional information for those clusters.
  99 +
 100 +We use a cluster size of 65536, and use 8 bytes for each
 101 +cluster in the header to store the following information:
 102 +
 103 +* 1 byte dev_id (to identity the drive)
 104 +* 2 bytes zero indicator (mark zero regions (16x4096))
 105 +* 4 bytes cluster number
 106 +* 1 byte not used (reserved)
 107 +
 108 +We only store non-zero blocks (such block is 4096 bytes).
 109 +
 110 +Each archive is marked with a uuid. The archive header and all
 111 +extent headers includes that uuid and a MD5 checksum (over header
 112 +data).
 113 +
 114 +
 115 diff --git a/vma-reader.c b/vma-reader.c
 116 new file mode 100644
 117 index 0000000..154c96b
 118 --- /dev/null
 119 +++ b/vma-reader.c
 120 @@ -0,0 +1,772 @@
 121 +/*
 122 + * VMA: Virtual Machine Archive
 123 + *
 124 + * Copyright (C) 2012 Proxmox Server Solutions
 125 + *
 126 + * Authors:
 127 + *  Dietmar Maurer (dietmar@proxmox.com)
 128 + *
 129 + * This work is licensed under the terms of the GNU GPL, version 2 or later.
 130 + * See the COPYING file in the top-level directory.
 131 + *
 132 + */
 133 +
 134 +#include <stdio.h>
 135 +#include <errno.h>
 136 +#include <unistd.h>
 137 +#include <stdio.h>
 138 +#include <string.h>
 139 +#include <sys/types.h>
 140 +#include <sys/stat.h>
 141 +#include <fcntl.h>
 142 +#include <glib.h>
 143 +#include <uuid/uuid.h>
 144 +
 145 +#include "qemu-common.h"
 146 +#include "qemu_socket.h"
 147 +#include "qemu-coroutine.h"
 148 +#include "qemu-aio.h"
 149 +#include "qemu/ratelimit.h"
 150 +#include "vma.h"
 151 +#include "block.h"
 152 +
 153 +#define BITS_PER_LONG  (sizeof(unsigned long) * 8)
 154 +
 155 +static unsigned char zero_vma_block[VMA_BLOCK_SIZE];
 156 +
 157 +typedef struct VmaRestoreState {
 158 +    BlockDriverState *bs;
 159 +    bool write_zeroes;
 160 +    unsigned long *bitmap;
 161 +    int bitmap_size;
 162 +}  VmaRestoreState;
 163 +
 164 +struct VmaReader {
 165 +    int fd;
 166 +    GChecksum *md5csum;
 167 +    GHashTable *blob_hash;
 168 +    unsigned char *head_data;
 169 +    VmaDeviceInfo devinfo[256];
 170 +    VmaRestoreState rstate[256];
 171 +    GList *cdata_list;
 172 +    guint8 vmstate_stream;
 173 +    uint32_t vmstate_clusters;
 174 +};
 175 +
 176 +static guint
 177 +g_int32_hash(gconstpointer v)
 178 +{
 179 +    return *(const uint32_t *)v;
 180 +}
 181 +
 182 +static gboolean
 183 +g_int32_equal(gconstpointer v1, gconstpointer v2)
 184 +{
 185 +    return *((const uint32_t *)v1) == *((const uint32_t *)v2);
 186 +}
 187 +
 188 +static int vma_reader_get_bitmap(VmaRestoreState *rstate, int64_t cluster_num)
 189 +{
 190 +    assert(rstate);
 191 +    assert(rstate->bitmap);
 192 +
 193 +    unsigned long val, idx, bit;
 194 +
 195 +    idx = cluster_num / BITS_PER_LONG;
 196 +
 197 +    assert(rstate->bitmap_size > idx);
 198 +
 199 +    bit = cluster_num % BITS_PER_LONG;
 200 +    val = rstate->bitmap[idx];
 201 +
 202 +    return !!(val & (1UL << bit));
 203 +}
 204 +
 205 +static void vma_reader_set_bitmap(VmaRestoreState *rstate, int64_t cluster_num,
 206 +                                  int dirty)
 207 +{
 208 +    assert(rstate);
 209 +    assert(rstate->bitmap);
 210 +
 211 +    unsigned long val, idx, bit;
 212 +
 213 +    idx = cluster_num / BITS_PER_LONG;
 214 +
 215 +    assert(rstate->bitmap_size > idx);
 216 +
 217 +    bit = cluster_num % BITS_PER_LONG;
 218 +    val = rstate->bitmap[idx];
 219 +    if (dirty) {
 220 +        if (!(val & (1UL << bit))) {
 221 +            val |= 1UL << bit;
 222 +        }
 223 +    } else {
 224 +        if (val & (1UL << bit)) {
 225 +            val &= ~(1UL << bit);
 226 +        }
 227 +    }
 228 +    rstate->bitmap[idx] = val;
 229 +}
 230 +
 231 +typedef struct VmaBlob {
 232 +    uint32_t start;
 233 +    uint32_t len;
 234 +    void *data;
 235 +} VmaBlob;
 236 +
 237 +static const VmaBlob *get_header_blob(VmaReader *vmar, uint32_t pos)
 238 +{
 239 +    assert(vmar);
 240 +    assert(vmar->blob_hash);
 241 +
 242 +    return g_hash_table_lookup(vmar->blob_hash, &pos);
 243 +}
 244 +
 245 +static const char *get_header_str(VmaReader *vmar, uint32_t pos)
 246 +{
 247 +    const VmaBlob *blob = get_header_blob(vmar, pos);
 248 +    if (!blob) {
 249 +        return NULL;
 250 +    }
 251 +    const char *res = (char *)blob->data;
 252 +    if (res[blob->len-1] != '\0') {
 253 +        return NULL;
 254 +    }
 255 +    return res;
 256 +}
 257 +
 258 +static ssize_t
 259 +safe_read(int fd, unsigned char *buf, size_t count)
 260 +{
 261 +    ssize_t n;
 262 +
 263 +    do {
 264 +        n = read(fd, buf, count);
 265 +    } while (n < 0 && errno == EINTR);
 266 +
 267 +    return n;
 268 +}
 269 +
 270 +static ssize_t
 271 +full_read(int fd, unsigned char *buf, size_t len)
 272 +{
 273 +    ssize_t n;
 274 +    size_t total;
 275 +
 276 +    total = 0;
 277 +
 278 +    while (len > 0) {
 279 +        n = safe_read(fd, buf, len);
 280 +
 281 +        if (n == 0) {
 282 +            return total;
 283 +        }
 284 +
 285 +        if (n <= 0) {
 286 +            break;
 287 +        }
 288 +
 289 +        buf += n;
 290 +        total += n;
 291 +        len -= n;
 292 +    }
 293 +
 294 +    if (len) {
 295 +        return -1;
 296 +    }
 297 +
 298 +    return total;
 299 +}
 300 +
 301 +void vma_reader_destroy(VmaReader *vmar)
 302 +{
 303 +    assert(vmar);
 304 +
 305 +    if (vmar->fd >= 0) {
 306 +        close(vmar->fd);
 307 +    }
 308 +
 309 +    if (vmar->cdata_list) {
 310 +        g_list_free(vmar->cdata_list);
 311 +    }
 312 +
 313 +    int i;
 314 +    for (i = 1; i < 256; i++) {
 315 +        if (vmar->rstate[i].bitmap) {
 316 +            g_free(vmar->rstate[i].bitmap);
 317 +        }
 318 +    }
 319 +
 320 +    if (vmar->md5csum) {
 321 +        g_checksum_free(vmar->md5csum);
 322 +    }
 323 +
 324 +    if (vmar->blob_hash) {
 325 +        g_hash_table_destroy(vmar->blob_hash);
 326 +    }
 327 +
 328 +    if (vmar->head_data) {
 329 +        g_free(vmar->head_data);
 330 +    }
 331 +
 332 +    g_free(vmar);
 333 +
 334 +};
 335 +
 336 +static int vma_reader_read_head(VmaReader *vmar, Error **errp)
 337 +{
 338 +    assert(vmar);
 339 +    assert(errp);
 340 +    assert(*errp == NULL);
 341 +
 342 +    unsigned char md5sum[16];
 343 +    int i;
 344 +    int ret = 0;
 345 +
 346 +    vmar->head_data = g_malloc(sizeof(VmaHeader));
 347 +
 348 +    if (full_read(vmar->fd, vmar->head_data, sizeof(VmaHeader)) !=
 349 +        sizeof(VmaHeader)) {
 350 +        error_setg(errp, "can't read vma header - %s",
 351 +                   errno ? strerror(errno) : "got EOF");
 352 +        return -1;
 353 +    }
 354 +
 355 +    VmaHeader *h = (VmaHeader *)vmar->head_data;
 356 +
 357 +    if (h->magic != VMA_MAGIC) {
 358 +        error_setg(errp, "not a vma file - wrong magic number");
 359 +        return -1;
 360 +    }
 361 +
 362 +    uint32_t header_size = GUINT32_FROM_BE(h->header_size);
 363 +    int need = header_size - sizeof(VmaHeader);
 364 +    if (need <= 0) {
 365 +        error_setg(errp, "wrong vma header size %d", header_size);
 366 +        return -1;
 367 +    }
 368 +
 369 +    vmar->head_data = g_realloc(vmar->head_data, header_size);
 370 +    h = (VmaHeader *)vmar->head_data;
 371 +
 372 +    if (full_read(vmar->fd, vmar->head_data + sizeof(VmaHeader), need) !=
 373 +        need) {
 374 +        error_setg(errp, "can't read vma header data - %s",
 375 +                   errno ? strerror(errno) : "got EOF");
 376 +        return -1;
 377 +    }
 378 +
 379 +    memcpy(md5sum, h->md5sum, 16);
 380 +    memset(h->md5sum, 0, 16);
 381 +
 382 +    g_checksum_reset(vmar->md5csum);
 383 +    g_checksum_update(vmar->md5csum, vmar->head_data, header_size);
 384 +    gsize csize = 16;
 385 +    g_checksum_get_digest(vmar->md5csum, (guint8 *)(h->md5sum), &csize);
 386 +
 387 +    if (memcmp(md5sum, h->md5sum, 16) != 0) {
 388 +        error_setg(errp, "wrong vma header chechsum");
 389 +        return -1;
 390 +    }
 391 +
 392 +    /* we can modify header data after checksum verify */
 393 +    h->header_size = header_size;
 394 +
 395 +    h->version = GUINT32_FROM_BE(h->version);
 396 +    if (h->version != 1) {
 397 +        error_setg(errp, "wrong vma version %d", h->version);
 398 +        return -1;
 399 +    }
 400 +
 401 +    h->ctime = GUINT64_FROM_BE(h->ctime);
 402 +    h->blob_buffer_offset = GUINT32_FROM_BE(h->blob_buffer_offset);
 403 +    h->blob_buffer_size = GUINT32_FROM_BE(h->blob_buffer_size);
 404 +
 405 +    uint32_t bstart = h->blob_buffer_offset + 1;
 406 +    uint32_t bend = h->blob_buffer_offset + h->blob_buffer_size;
 407 +
 408 +    if (bstart <= sizeof(VmaHeader)) {
 409 +        error_setg(errp, "wrong vma blob buffer offset %d",
 410 +                   h->blob_buffer_offset);
 411 +        return -1;
 412 +    }
 413 +
 414 +    if (bend > header_size) {
 415 +        error_setg(errp, "wrong vma blob buffer size %d/%d",
 416 +                   h->blob_buffer_offset, h->blob_buffer_size);
 417 +        return -1;
 418 +    }
 419 +
 420 +    while ((bstart + 2) <= bend) {
 421 +        uint32_t size = vmar->head_data[bstart] +
 422 +            (vmar->head_data[bstart+1] << 8);
 423 +        if ((bstart + size + 2) <= bend) {
 424 +            VmaBlob *blob = g_new0(VmaBlob, 1);
 425 +            blob->start = bstart - h->blob_buffer_offset;
 426 +            blob->len = size;
 427 +            blob->data = vmar->head_data + bstart + 2;
 428 +            g_hash_table_insert(vmar->blob_hash, &blob->start, blob);
 429 +        }
 430 +        bstart += size + 2;
 431 +    }
 432 +
 433 +
 434 +    int count = 0;
 435 +    for (i = 1; i < 256; i++) {
 436 +        VmaDeviceInfoHeader *dih = &h->dev_info[i];
 437 +        uint32_t devname_ptr = GUINT32_FROM_BE(dih->devname_ptr);
 438 +        uint64_t size = GUINT64_FROM_BE(dih->size);
 439 +        const char *devname =  get_header_str(vmar, devname_ptr);
 440 +
 441 +        if (size && devname) {
 442 +            count++;
 443 +            vmar->devinfo[i].size = size;
 444 +            vmar->devinfo[i].devname = devname;
 445 +
 446 +            if (strcmp(devname, "vmstate") == 0) {
 447 +                vmar->vmstate_stream = i;
 448 +            }
 449 +        }
 450 +    }
 451 +
 452 +    if (!count) {
 453 +        error_setg(errp, "vma does not contain data");
 454 +        return -1;
 455 +    }
 456 +
 457 +    for (i = 0; i < VMA_MAX_CONFIGS; i++) {
 458 +        uint32_t name_ptr = GUINT32_FROM_BE(h->config_names[i]);
 459 +        uint32_t data_ptr = GUINT32_FROM_BE(h->config_data[i]);
 460 +
 461 +        if (!(name_ptr && data_ptr)) {
 462 +            continue;
 463 +        }
 464 +        const char *name =  get_header_str(vmar, name_ptr);
 465 +        const VmaBlob *blob = get_header_blob(vmar, data_ptr);
 466 +
 467 +        if (!(name && blob)) {
 468 +            error_setg(errp, "vma contains invalid data pointers");
 469 +            return -1;
 470 +        }
 471 +
 472 +        VmaConfigData *cdata = g_new0(VmaConfigData, 1);
 473 +        cdata->name = name;
 474 +        cdata->data = blob->data;
 475 +        cdata->len = blob->len;
 476 +
 477 +        vmar->cdata_list = g_list_append(vmar->cdata_list, cdata);
 478 +    }
 479 +
 480 +    return ret;
 481 +};
 482 +
 483 +VmaReader *vma_reader_create(const char *filename, Error **errp)
 484 +{
 485 +    assert(filename);
 486 +    assert(errp);
 487 +
 488 +    VmaReader *vmar = g_new0(VmaReader, 1);
 489 +
 490 +    vmar->fd = open(filename, O_RDONLY);
 491 +
 492 +    if (vmar->fd < 0) {
 493 +        error_setg(errp, "can't open file %s - %s\n", filename,
 494 +                   strerror(errno));
 495 +        goto err;
 496 +    }
 497 +
 498 +    vmar->md5csum = g_checksum_new(G_CHECKSUM_MD5);
 499 +    if (!vmar->md5csum) {
 500 +        error_setg(errp, "can't allocate cmsum\n");
 501 +        goto err;
 502 +    }
 503 +
 504 +    vmar->blob_hash = g_hash_table_new_full(g_int32_hash, g_int32_equal,
 505 +                                            NULL, g_free);
 506 +
 507 +    if (vma_reader_read_head(vmar, errp) < 0) {
 508 +        goto err;
 509 +    }
 510 +
 511 +    return vmar;
 512 +
 513 +err:
 514 +    if (vmar) {
 515 +        vma_reader_destroy(vmar);
 516 +    }
 517 +
 518 +    return NULL;
 519 +}
 520 +
 521 +VmaHeader *vma_reader_get_header(VmaReader *vmar)
 522 +{
 523 +    assert(vmar);
 524 +    assert(vmar->head_data);
 525 +
 526 +    return (VmaHeader *)(vmar->head_data);
 527 +}
 528 +
 529 +GList *vma_reader_get_config_data(VmaReader *vmar)
 530 +{
 531 +    assert(vmar);
 532 +    assert(vmar->head_data);
 533 +
 534 +    return vmar->cdata_list;
 535 +}
 536 +
 537 +VmaDeviceInfo *vma_reader_get_device_info(VmaReader *vmar, guint8 dev_id)
 538 +{
 539 +    assert(vmar);
 540 +    assert(dev_id);
 541 +
 542 +    if (vmar->devinfo[dev_id].size && vmar->devinfo[dev_id].devname) {
 543 +        return &vmar->devinfo[dev_id];
 544 +    }
 545 +
 546 +    return NULL;
 547 +}
 548 +
 549 +int vma_reader_register_bs(VmaReader *vmar, guint8 dev_id, BlockDriverState *bs,
 550 +                           bool write_zeroes, Error **errp)
 551 +{
 552 +    assert(vmar);
 553 +    assert(bs != NULL);
 554 +    assert(dev_id);
 555 +    assert(vmar->rstate[dev_id].bs == NULL);
 556 +
 557 +    int64_t size = bdrv_getlength(bs);
 558 +    if (size != vmar->devinfo[dev_id].size) {
 559 +        error_setg(errp, "vma_reader_register_bs for stream %s failed - "
 560 +                   "unexpected size %zd != %zd", vmar->devinfo[dev_id].devname,
 561 +                   size, vmar->devinfo[dev_id].size);
 562 +        return -1;
 563 +    }
 564 +
 565 +    vmar->rstate[dev_id].bs = bs;
 566 +    vmar->rstate[dev_id].write_zeroes = write_zeroes;
 567 +
 568 +    int64_t bitmap_size = (size/BDRV_SECTOR_SIZE) +
 569 +        (VMA_CLUSTER_SIZE/BDRV_SECTOR_SIZE) * BITS_PER_LONG - 1;
 570 +    bitmap_size /= (VMA_CLUSTER_SIZE/BDRV_SECTOR_SIZE) * BITS_PER_LONG;
 571 +
 572 +    vmar->rstate[dev_id].bitmap_size = bitmap_size;
 573 +    vmar->rstate[dev_id].bitmap = g_new0(unsigned long, bitmap_size);
 574 +
 575 +    return 0;
 576 +}
 577 +
 578 +static ssize_t safe_write(int fd, void *buf, size_t count)
 579 +{
 580 +    ssize_t n;
 581 +
 582 +    do {
 583 +        n = write(fd, buf, count);
 584 +    } while (n < 0 && errno == EINTR);
 585 +
 586 +    return n;
 587 +}
 588 +
 589 +static size_t full_write(int fd, void *buf, size_t len)
 590 +{
 591 +    ssize_t n;
 592 +    size_t total;
 593 +
 594 +    total = 0;
 595 +
 596 +    while (len > 0) {
 597 +        n = safe_write(fd, buf, len);
 598 +        if (n < 0) {
 599 +            return n;
 600 +        }
 601 +        buf += n;
 602 +        total += n;
 603 +        len -= n;
 604 +    }
 605 +
 606 +    if (len) {
 607 +        /* incomplete write ? */
 608 +        return -1;
 609 +    }
 610 +
 611 +    return total;
 612 +}
 613 +
 614 +static int restore_write_data(VmaReader *vmar, guint8 dev_id,
 615 +                              BlockDriverState *bs, int vmstate_fd,
 616 +                              unsigned char *buf, int64_t sector_num,
 617 +                              int nb_sectors, Error **errp)
 618 +{
 619 +    assert(vmar);
 620 +
 621 +    if (dev_id == vmar->vmstate_stream) {
 622 +        if (vmstate_fd >= 0) {
 623 +            int len = nb_sectors * BDRV_SECTOR_SIZE;
 624 +            int res = full_write(vmstate_fd, buf, len);
 625 +            if (res < 0) {
 626 +                error_setg(errp, "write vmstate failed %d", res);
 627 +                return -1;
 628 +            }
 629 +        }
 630 +    } else {
 631 +        int res = bdrv_write(bs, sector_num, buf, nb_sectors);
 632 +        if (res < 0) {
 633 +            error_setg(errp, "bdrv_write to %s failed (%d)",
 634 +                       bdrv_get_device_name(bs), res);
 635 +            return -1;
 636 +        }
 637 +    }
 638 +    return 0;
 639 +}
 640 +static int restore_extent(VmaReader *vmar, unsigned char *buf,
 641 +                          int extent_size, int vmstate_fd, Error **errp)
 642 +{
 643 +    assert(vmar);
 644 +    assert(buf);
 645 +
 646 +    VmaExtentHeader *ehead = (VmaExtentHeader *)buf;
 647 +    int start = VMA_EXTENT_HEADER_SIZE;
 648 +    int i;
 649 +
 650 +    for (i = 0; i < VMA_BLOCKS_PER_EXTENT; i++) {
 651 +        uint64_t block_info = GUINT64_FROM_BE(ehead->blockinfo[i]);
 652 +        uint32_t cluster_num = block_info &  0xffffffff;
 653 +        uint8_t dev_id = (block_info >> 32) &  0xff;
 654 +        uint16_t mask = block_info >> (32+16);
 655 +        int64_t max_sector;
 656 +
 657 +        if (!dev_id) {
 658 +            continue;
 659 +        }
 660 +
 661 +        VmaRestoreState *rstate = &vmar->rstate[dev_id];
 662 +        BlockDriverState *bs = NULL;
 663 +
 664 +        if (dev_id != vmar->vmstate_stream) {
 665 +            bs = rstate->bs;
 666 +            if (!bs) {
 667 +                error_setg(errp, "got wrong dev id %d", dev_id);
 668 +                return -1;
 669 +            }
 670 +
 671 +            if (vma_reader_get_bitmap(rstate, cluster_num)) {
 672 +                error_setg(errp, "found duplicated cluster %d for stream %s",
 673 +                          cluster_num, vmar->devinfo[dev_id].devname);
 674 +                return -1;
 675 +            }
 676 +            vma_reader_set_bitmap(rstate, cluster_num, 1);
 677 +
 678 +            max_sector = vmar->devinfo[dev_id].size/BDRV_SECTOR_SIZE;
 679 +        } else {
 680 +            max_sector = G_MAXINT64;
 681 +            if (cluster_num != vmar->vmstate_clusters) {
 682 +                error_setg(errp, "found out of order vmstate data");
 683 +                return -1;
 684 +            }
 685 +            vmar->vmstate_clusters++;
 686 +        }
 687 +
 688 +        /* try to write whole clusters to speedup restore */
 689 +        if (mask == 0xffff) {
 690 +            if ((start + VMA_CLUSTER_SIZE) > extent_size) {
 691 +                error_setg(errp, "short vma extent - too many blocks");
 692 +                return -1;
 693 +            }
 694 +            int64_t sector_num = (cluster_num * VMA_CLUSTER_SIZE) /
 695 +                BDRV_SECTOR_SIZE;
 696 +            int64_t end_sector = sector_num +
 697 +                VMA_CLUSTER_SIZE/BDRV_SECTOR_SIZE;
 698 +
 699 +            if (end_sector > max_sector) {
 700 +                end_sector = max_sector;
 701 +            }
 702 +
 703 +            if (end_sector <= sector_num) {
 704 +                error_setg(errp, "got wrong block address - write bejond end");
 705 +                return -1;
 706 +            }
 707 +
 708 +            int nb_sectors = end_sector - sector_num;
 709 +            if (restore_write_data(vmar, dev_id, bs, vmstate_fd, buf + start,
 710 +                                   sector_num, nb_sectors, errp) < 0) {
 711 +                return -1;
 712 +            }
 713 +
 714 +            start += VMA_CLUSTER_SIZE;
 715 +        } else {
 716 +            int j;
 717 +            int bit = 1;
 718 +
 719 +            for (j = 0; j < 16; j++) {
 720 +                int64_t sector_num = (cluster_num*VMA_CLUSTER_SIZE +
 721 +                                      j*VMA_BLOCK_SIZE)/BDRV_SECTOR_SIZE;
 722 +
 723 +                int64_t end_sector = sector_num +
 724 +                    VMA_BLOCK_SIZE/BDRV_SECTOR_SIZE;
 725 +                if (end_sector > max_sector) {
 726 +                    end_sector = max_sector;
 727 +                }
 728 +
 729 +                if (mask & bit) {
 730 +                    if ((start + VMA_BLOCK_SIZE) > extent_size) {
 731 +                        error_setg(errp, "short vma extent - too many blocks");
 732 +                        return -1;
 733 +                    }
 734 +
 735 +                    if (end_sector <= sector_num) {
 736 +                        error_setg(errp, "got wrong block address - "
 737 +                                   "write bejond end");
 738 +                        return -1;
 739 +                    }
 740 +
 741 +                    int nb_sectors = end_sector - sector_num;
 742 +                    if (restore_write_data(vmar, dev_id, bs, vmstate_fd,
 743 +                                           buf + start, sector_num,
 744 +                                           nb_sectors, errp) < 0) {
 745 +                        return -1;
 746 +                    }
 747 +
 748 +                    start += VMA_BLOCK_SIZE;
 749 +
 750 +                } else {
 751 +
 752 +                    if (rstate->write_zeroes & (end_sector > sector_num)) {
 753 +                        /* Todo: use bdrv_co_write_zeroes (but that need to
 754 +                         * be run inside coroutine?)
 755 +                         */
 756 +                        int nb_sectors = end_sector - sector_num;
 757 +                        if (restore_write_data(vmar, dev_id, bs, vmstate_fd,
 758 +                                              zero_vma_block, sector_num,
 759 +                                               nb_sectors, errp) < 0) {
 760 +                            return -1;
 761 +                        }
 762 +                    }
 763 +                }
 764 +
 765 +                bit = bit << 1;
 766 +            }
 767 +        }
 768 +    }
 769 +
 770 +    if (start != extent_size) {
 771 +        error_setg(errp, "vma extent error - missing blocks");
 772 +        return -1;
 773 +    }
 774 +
 775 +    return 0;
 776 +}
 777 +
 778 +int vma_reader_restore(VmaReader *vmar, int vmstate_fd, Error **errp)
 779 +{
 780 +    assert(vmar);
 781 +    assert(vmar->head_data);
 782 +
 783 +    int ret = 0;
 784 +    unsigned char buf[VMA_MAX_EXTENT_SIZE];
 785 +    int buf_pos = 0;
 786 +    unsigned char md5sum[16];
 787 +    VmaHeader *h = (VmaHeader *)vmar->head_data;
 788 +
 789 +
 790 +    while (1) {
 791 +        int bytes = full_read(vmar->fd, buf + buf_pos, sizeof(buf) - buf_pos);
 792 +        if (bytes < 0) {
 793 +            error_setg(errp, "read failed - %s", strerror(errno));
 794 +            return -1;
 795 +        }
 796 +
 797 +        buf_pos += bytes;
 798 +
 799 +        if (!buf_pos) {
 800 +            break; /* EOF */
 801 +        }
 802 +
 803 +        if (buf_pos < VMA_EXTENT_HEADER_SIZE) {
 804 +            error_setg(errp, "read short extent (%d bytes)", buf_pos);
 805 +            return -1;
 806 +        }
 807 +
 808 +        VmaExtentHeader *ehead = (VmaExtentHeader *)buf;
 809 +
 810 +        /* extract md5sum */
 811 +        memcpy(md5sum, ehead->md5sum, sizeof(ehead->md5sum));
 812 +        memset(ehead->md5sum, 0, sizeof(ehead->md5sum));
 813 +
 814 +        g_checksum_reset(vmar->md5csum);
 815 +        g_checksum_update(vmar->md5csum, buf, VMA_EXTENT_HEADER_SIZE);
 816 +        gsize csize = 16;
 817 +        g_checksum_get_digest(vmar->md5csum, ehead->md5sum, &csize);
 818 +
 819 +        if (memcmp(md5sum, ehead->md5sum, 16) != 0) {
 820 +            error_setg(errp, "wrong vma extent header chechsum");
 821 +            return -1;
 822 +        }
 823 +
 824 +        if (memcmp(h->uuid, ehead->uuid, sizeof(ehead->uuid)) != 0) {
 825 +            error_setg(errp, "wrong vma extent uuid");
 826 +            return -1;
 827 +        }
 828 +
 829 +        if (ehead->magic != VMA_EXTENT_MAGIC || ehead->reserved1 != 0) {
 830 +            error_setg(errp, "wrong vma extent header magic");
 831 +            return -1;
 832 +        }
 833 +
 834 +        int block_count = GUINT16_FROM_BE(ehead->block_count);
 835 +        int extent_size = VMA_EXTENT_HEADER_SIZE + block_count*VMA_BLOCK_SIZE;
 836 +
 837 +        if (buf_pos < extent_size) {
 838 +            error_setg(errp, "short vma extent (%d < %d)", buf_pos,
 839 +                       extent_size);
 840 +            return -1;
 841 +        }
 842 +
 843 +        if (restore_extent(vmar, buf, extent_size, vmstate_fd, errp) < 0) {
 844 +            return -1;
 845 +        }
 846 +
 847 +        if (buf_pos > extent_size) {
 848 +            memmove(buf, buf + extent_size, buf_pos - extent_size);
 849 +            buf_pos = buf_pos - extent_size;
 850 +        } else {
 851 +            buf_pos = 0;
 852 +        }
 853 +    }
 854 +
 855 +    bdrv_drain_all();
 856 +
 857 +    int i;
 858 +    for (i = 1; i < 256; i++) {
 859 +        VmaRestoreState *rstate = &vmar->rstate[i];
 860 +        if (!rstate->bs) {
 861 +            continue;
 862 +        }
 863 +
 864 +        if (bdrv_flush(rstate->bs) < 0) {
 865 +            error_setg(errp, "vma bdrv_flush %s failed",
 866 +                       vmar->devinfo[i].devname);
 867 +            return -1;
 868 +        }
 869 +
 870 +        if (vmar->devinfo[i].size &&
 871 +            (strcmp(vmar->devinfo[i].devname, "vmstate") != 0)) {
 872 +            assert(rstate->bitmap);
 873 +
 874 +            int64_t cluster_num, end;
 875 +
 876 +            end = (vmar->devinfo[i].size + VMA_CLUSTER_SIZE - 1) /
 877 +                VMA_CLUSTER_SIZE;
 878 +
 879 +            for (cluster_num = 0; cluster_num < end; cluster_num++) {
 880 +                if (!vma_reader_get_bitmap(rstate, cluster_num)) {
 881 +                    error_setg(errp, "detected missing cluster %zd "
 882 +                               "for stream %s", cluster_num,
 883 +                               vmar->devinfo[i].devname);
 884 +                    return -1;
 885 +                }
 886 +            }
 887 +        }
 888 +    }
 889 +
 890 +    return ret;
 891 +}
 892 +
 893 diff --git a/vma-writer.c b/vma-writer.c
 894 new file mode 100644
 895 index 0000000..02d4447
 896 --- /dev/null
 897 +++ b/vma-writer.c
 898 @@ -0,0 +1,900 @@
 899 +/*
 900 + * VMA: Virtual Machine Archive
 901 + *
 902 + * Copyright (C) 2012 Proxmox Server Solutions
 903 + *
 904 + * Authors:
 905 + *  Dietmar Maurer (dietmar@proxmox.com)
 906 + *
 907 + * This work is licensed under the terms of the GNU GPL, version 2 or later.
 908 + * See the COPYING file in the top-level directory.
 909 + *
 910 + */
 911 +
 912 +#include <stdio.h>
 913 +#include <errno.h>
 914 +#include <unistd.h>
 915 +#include <stdio.h>
 916 +#include <string.h>
 917 +#include <sys/types.h>
 918 +#include <sys/stat.h>
 919 +#include <fcntl.h>
 920 +#include <glib.h>
 921 +#include <uuid/uuid.h>
 922 +
 923 +#include "qemu-common.h"
 924 +#include "qemu_socket.h"
 925 +#include "qemu-coroutine.h"
 926 +#include "qemu-aio.h"
 927 +#include "qemu/ratelimit.h"
 928 +#include "vma.h"
 929 +#include "block.h"
 930 +
 931 +#define DEBUG_VMA 0
 932 +
 933 +#define DPRINTF(fmt, ...)\
 934 +    do { if (DEBUG_VMA) { printf("vma: " fmt, ## __VA_ARGS__); } } while (0)
 935 +
 936 +#define WRITE_BUFFERS 5
 937 +
 938 +typedef struct VmaAIOCB VmaAIOCB;
 939 +struct VmaAIOCB {
 940 +    VmaWriter *vmaw;
 941 +    unsigned char buffer[VMA_MAX_EXTENT_SIZE];
 942 +    size_t bytes;
 943 +    Coroutine *co;
 944 +};
 945 +
 946 +struct VmaWriter {
 947 +    int fd;
 948 +    FILE *cmd;
 949 +    int status;
 950 +    char errmsg[8192];
 951 +    uuid_t uuid;
 952 +    bool header_written;
 953 +    bool closed;
 954 +
 955 +    /* we always write extents */
 956 +    unsigned char outbuf[VMA_MAX_EXTENT_SIZE];
 957 +    int outbuf_pos; /* in bytes */
 958 +    int outbuf_count; /* in VMA_BLOCKS */
 959 +    uint64_t outbuf_block_info[VMA_BLOCKS_PER_EXTENT];
 960 +
 961 +    VmaAIOCB aiocbs[WRITE_BUFFERS];
 962 +    CoQueue wqueue;
 963 +
 964 +    GChecksum *md5csum;
 965 +    CoMutex writer_lock;
 966 +    CoMutex flush_lock;
 967 +    Coroutine *co_writer;
 968 +    RateLimit limit;
 969 +    uint64_t delay_ns;
 970 +
 971 +    /* drive informations */
 972 +    VmaStreamInfo stream_info[256];
 973 +    guint stream_count;
 974 +
 975 +    guint8 vmstate_stream;
 976 +    uint32_t vmstate_clusters;
 977 +
 978 +    /* header blob table */
 979 +    char *header_blob_table;
 980 +    uint32_t header_blob_table_size;
 981 +    uint32_t header_blob_table_pos;
 982 +
 983 +    /* store for config blobs */
 984 +    uint32_t config_names[VMA_MAX_CONFIGS]; /* offset into blob_buffer table */
 985 +    uint32_t config_data[VMA_MAX_CONFIGS];  /* offset into blob_buffer table */
 986 +    uint32_t config_count;
 987 +};
 988 +
 989 +void vma_writer_set_error(VmaWriter *vmaw, const char *fmt, ...)
 990 +{
 991 +    va_list ap;
 992 +
 993 +    if (vmaw->status < 0) {
 994 +        return;
 995 +    }
 996 +
 997 +    vmaw->status = -1;
 998 +
 999 +    va_start(ap, fmt);
1000 +    g_vsnprintf(vmaw->errmsg, sizeof(vmaw->errmsg), fmt, ap);
1001 +    va_end(ap);
1002 +
1003 +    DPRINTF("vma_writer_set_error: %s\n", vmaw->errmsg);
1004 +}
1005 +
1006 +static uint32_t allocate_header_blob(VmaWriter *vmaw, const char *data,
1007 +                                     size_t len)
1008 +{
1009 +    if (len > 65535) {
1010 +        return 0;
1011 +    }
1012 +
1013 +    if (!vmaw->header_blob_table ||
1014 +        (vmaw->header_blob_table_size <
1015 +         (vmaw->header_blob_table_pos + len + 2))) {
1016 +        int newsize = vmaw->header_blob_table_size + ((len + 2 + 511)/512)*512;
1017 +
1018 +        vmaw->header_blob_table = g_realloc(vmaw->header_blob_table, newsize);
1019 +        memset(vmaw->header_blob_table + vmaw->header_blob_table_size,
1020 +               0, newsize - vmaw->header_blob_table_size);
1021 +        vmaw->header_blob_table_size = newsize;
1022 +    }
1023 +
1024 +    uint32_t cpos = vmaw->header_blob_table_pos;
1025 +    vmaw->header_blob_table[cpos] = len & 255;
1026 +    vmaw->header_blob_table[cpos+1] = (len >> 8) & 255;
1027 +    memcpy(vmaw->header_blob_table + cpos + 2, data, len);
1028 +    vmaw->header_blob_table_pos += len + 2;
1029 +    return cpos;
1030 +}
1031 +
1032 +static uint32_t allocate_header_string(VmaWriter *vmaw, const char *str)
1033 +{
1034 +    assert(vmaw);
1035 +
1036 +    size_t len = strlen(str) + 1;
1037 +
1038 +    return allocate_header_blob(vmaw, str, len);
1039 +}
1040 +
1041 +int vma_writer_add_config(VmaWriter *vmaw, const char *name, gpointer data,
1042 +                          gsize len)
1043 +{
1044 +    assert(vmaw);
1045 +    assert(!vmaw->header_written);
1046 +    assert(vmaw->config_count < VMA_MAX_CONFIGS);
1047 +    assert(name);
1048 +    assert(data);
1049 +    assert(len);
1050 +
1051 +    uint32_t name_ptr = allocate_header_string(vmaw, name);
1052 +    if (!name_ptr) {
1053 +        return -1;
1054 +    }
1055 +
1056 +    uint32_t data_ptr = allocate_header_blob(vmaw, data, len);
1057 +    if (!data_ptr) {
1058 +        return -1;
1059 +    }
1060 +
1061 +    vmaw->config_names[vmaw->config_count] = name_ptr;
1062 +    vmaw->config_data[vmaw->config_count] = data_ptr;
1063 +
1064 +    vmaw->config_count++;
1065 +
1066 +    return 0;
1067 +}
1068 +
1069 +int vma_writer_register_stream(VmaWriter *vmaw, const char *devname,
1070 +                               size_t size)
1071 +{
1072 +    assert(vmaw);
1073 +    assert(devname);
1074 +    assert(!vmaw->status);
1075 +
1076 +    if (vmaw->header_written) {
1077 +        vma_writer_set_error(vmaw, "vma_writer_register_stream: header "
1078 +                             "already written");
1079 +        return -1;
1080 +    }
1081 +
1082 +    guint n = vmaw->stream_count + 1;
1083 +
1084 +    /* we can have dev_ids form 1 to 255 (0 reserved)
1085 +     * 255(-1) reseverd for safety
1086 +     */
1087 +    if (n > 254) {
1088 +        vma_writer_set_error(vmaw, "vma_writer_register_stream: "
1089 +                             "too many drives");
1090 +        return -1;
1091 +    }
1092 +
1093 +    if (size <= 0) {
1094 +        vma_writer_set_error(vmaw, "vma_writer_register_stream: "
1095 +                             "got strange size %zd", size);
1096 +        return -1;
1097 +    }
1098 +
1099 +    DPRINTF("vma_writer_register_stream %s %zu %d\n", devname, size, n);
1100 +
1101 +    vmaw->stream_info[n].devname = g_strdup(devname);
1102 +    vmaw->stream_info[n].size = size;
1103 +
1104 +    vmaw->stream_info[n].cluster_count = (size + VMA_CLUSTER_SIZE - 1) /
1105 +        VMA_CLUSTER_SIZE;
1106 +
1107 +    vmaw->stream_count = n;
1108 +
1109 +    if (strcmp(devname, "vmstate") == 0) {
1110 +        vmaw->vmstate_stream = n;
1111 +    }
1112 +
1113 +    return n;
1114 +}
1115 +
1116 +static void vma_co_continue_write(void *opaque)
1117 +{
1118 +    VmaWriter *vmaw = opaque;
1119 +
1120 +    qemu_aio_set_fd_handler(vmaw->fd, NULL, NULL, NULL, NULL);
1121 +
1122 +    DPRINTF("vma_co_continue_write\n");
1123 +    qemu_coroutine_enter(vmaw->co_writer, NULL);
1124 +}
1125 +
1126 +static ssize_t coroutine_fn
1127 +vma_co_write(VmaWriter *vmaw, const void *buf, size_t bytes)
1128 +{
1129 +    size_t done = 0;
1130 +    ssize_t ret;
1131 +
1132 +    /* atomic writes (we cannot interleave writes) */
1133 +    qemu_co_mutex_lock(&vmaw->writer_lock);
1134 +
1135 +    DPRINTF("vma_co_write enter %zd\n", bytes);
1136 +
1137 +    while (done < bytes) {
1138 +        ret = write(vmaw->fd, buf + done, bytes - done);
1139 +        if (ret > 0) {
1140 +            done += ret;
1141 +            DPRINTF("vma_co_write written %zd %zd\n", done, ret);
1142 +        } else if (ret < 0) {
1143 +            if (errno == EAGAIN || errno == EWOULDBLOCK) {
1144 +                DPRINTF("vma_co_write yield %zd\n", done);
1145 +
1146 +                vmaw->co_writer = qemu_coroutine_self();
1147 +                qemu_aio_set_fd_handler(vmaw->fd, NULL, vma_co_continue_write,
1148 +                                        NULL, vmaw);
1149 +
1150 +                qemu_coroutine_yield();
1151 +                DPRINTF("vma_co_write restart %zd\n", done);
1152 +            } else {
1153 +                vma_writer_set_error(vmaw, "vma_co_write write error - %s",
1154 +                                     strerror(errno));
1155 +                done = -1; /* always return failure for partial writes */
1156 +                break;
1157 +            }
1158 +        } else if (ret == 0) {
1159 +            /* should not happen - simply try again */
1160 +        }
1161 +    }
1162 +
1163 +    qemu_co_mutex_unlock(&vmaw->writer_lock);
1164 +
1165 +    DPRINTF("vma_co_write leave %zd\n", done);
1166 +    return done;
1167 +}
1168 +
1169 +static void coroutine_fn vma_co_writer_task(void *opaque)
1170 +{
1171 +    VmaAIOCB *cb = opaque;
1172 +
1173 +    DPRINTF("vma_co_writer_task start\n");
1174 +
1175 +    int64_t done = vma_co_write(cb->vmaw, cb->buffer, cb->bytes);
1176 +    DPRINTF("vma_co_writer_task write done %zd\n", done);
1177 +
1178 +    if (done != cb->bytes) {
1179 +        DPRINTF("vma_co_writer_task failed write %zd %zd", cb->bytes, done);
1180 +        vma_writer_set_error(cb->vmaw, "vma_co_writer_task failed write %zd",
1181 +                             done);
1182 +    }
1183 +
1184 +    cb->bytes = 0;
1185 +
1186 +    qemu_co_queue_next(&cb->vmaw->wqueue);
1187 +
1188 +    DPRINTF("vma_co_writer_task end\n");
1189 +}
1190 +
1191 +static void coroutine_fn vma_queue_flush(VmaWriter *vmaw)
1192 +{
1193 +    DPRINTF("vma_queue_flush enter\n");
1194 +
1195 +    assert(vmaw);
1196 +
1197 +    while (1) {
1198 +        int i;
1199 +        VmaAIOCB *cb = NULL;
1200 +        for (i = 0; i < WRITE_BUFFERS; i++) {
1201 +            if (vmaw->aiocbs[i].bytes) {
1202 +                cb = &vmaw->aiocbs[i];
1203 +                DPRINTF("FOUND USED AIO BUFFER %d %zd\n", i,
1204 +                        vmaw->aiocbs[i].bytes);
1205 +                break;
1206 +            }
1207 +        }
1208 +        if (!cb) {
1209 +            break;
1210 +        }
1211 +        qemu_co_queue_wait(&vmaw->wqueue);
1212 +    }
1213 +
1214 +    DPRINTF("vma_queue_flush leave\n");
1215 +}
1216 +
1217 +/**
1218 + * NOTE: pipe buffer size in only 4096 bytes on linux (see 'ulimit -a')
1219 + * So we need to create a coroutione to allow 'parallel' execution.
1220 + */
1221 +static ssize_t coroutine_fn
1222 +vma_queue_write(VmaWriter *vmaw, const void *buf, size_t bytes)
1223 +{
1224 +    DPRINTF("vma_queue_write enter %zd\n", bytes);
1225 +
1226 +    assert(vmaw);
1227 +    assert(buf);
1228 +    assert(bytes <= VMA_MAX_EXTENT_SIZE);
1229 +
1230 +    VmaAIOCB *cb = NULL;
1231 +    while (!cb) {
1232 +        int i;
1233 +        for (i = 0; i < WRITE_BUFFERS; i++) {
1234 +            if (!vmaw->aiocbs[i].bytes) {
1235 +                cb = &vmaw->aiocbs[i];
1236 +                break;
1237 +            }
1238 +        }
1239 +        if (!cb) {
1240 +            qemu_co_queue_wait(&vmaw->wqueue);
1241 +        }
1242 +    }
1243 +
1244 +    memcpy(cb->buffer, buf, bytes);
1245 +    cb->bytes = bytes;
1246 +    cb->vmaw = vmaw;
1247 +
1248 +    DPRINTF("vma_queue_write start %zd\n", bytes);
1249 +    cb->co = qemu_coroutine_create(vma_co_writer_task);
1250 +    qemu_coroutine_enter(cb->co, cb);
1251 +
1252 +    DPRINTF("vma_queue_write leave\n");
1253 +
1254 +    return bytes;
1255 +}
1256 +
1257 +VmaWriter *vma_writer_create(const char *filename, uuid_t uuid, int64_t speed,
1258 +                             Error **errp)
1259 +{
1260 +    const char *p;
1261 +
1262 +    assert(sizeof(VmaHeader) == (4096 + 8192));
1263 +    assert(sizeof(VmaExtentHeader) == 512);
1264 +
1265 +    VmaWriter *vmaw = g_new0(VmaWriter, 1);
1266 +    vmaw->fd = -1;
1267 +
1268 +    vmaw->md5csum = g_checksum_new(G_CHECKSUM_MD5);
1269 +    if (!vmaw->md5csum) {
1270 +        error_setg(errp, "can't allocate cmsum\n");
1271 +        goto err;
1272 +    }
1273 +
1274 +    if (strstart(filename, "exec:", &p)) {
1275 +        vmaw->cmd = popen(p, "w");
1276 +        if (vmaw->cmd == NULL) {
1277 +            error_setg(errp, "can't popen command '%s' - %s\n", p,
1278 +                       strerror(errno));
1279 +            goto err;
1280 +        }
1281 +        vmaw->fd = fileno(vmaw->cmd);
1282 +        socket_set_nonblock(vmaw->fd);
1283 +
1284 +    } else {
1285 +        vmaw->fd = open(filename, O_NONBLOCK|O_WRONLY|O_CREAT|O_EXCL, 0644);
1286 +        if (vmaw->fd < 0) {
1287 +            error_setg(errp, "can't open file %s - %s\n", filename,
1288 +                       strerror(errno));
1289 +            goto err;
1290 +        }
1291 +    }
1292 +
1293 +    vmaw->outbuf_count = 0;
1294 +    vmaw->outbuf_pos = VMA_EXTENT_HEADER_SIZE;
1295 +
1296 +    vmaw->header_blob_table_pos = 1; /* start at pos 1 */
1297 +
1298 +    qemu_co_mutex_init(&vmaw->writer_lock);
1299 +    qemu_co_mutex_init(&vmaw->flush_lock);
1300 +    qemu_co_queue_init(&vmaw->wqueue);
1301 +
1302 +    uuid_copy(vmaw->uuid, uuid);
1303 +
1304 +    if (speed <= 0) {
1305 +        speed = 10*1024*1024*1024LLU; /* default 10GB/s */
1306 +    }
1307 +
1308 +    ratelimit_set_speed(&vmaw->limit, speed, 100000000ULL /* 0.1 sec */);
1309 +
1310 +    return vmaw;
1311 +
1312 +err:
1313 +    if (vmaw) {
1314 +        if (vmaw->cmd) {
1315 +            pclose(vmaw->cmd);
1316 +        } else if (vmaw->fd >= 0) {
1317 +            close(vmaw->fd);
1318 +        }
1319 +
1320 +        if (vmaw->md5csum) {
1321 +            g_checksum_free(vmaw->md5csum);
1322 +        }
1323 +
1324 +        g_free(vmaw);
1325 +    }
1326 +
1327 +    return NULL;
1328 +}
1329 +
1330 +static int coroutine_fn vma_write_header(VmaWriter *vmaw)
1331 +{
1332 +    assert(vmaw);
1333 +    int header_clusters = 8;
1334 +    char buf[65536*header_clusters];
1335 +    VmaHeader *head = (VmaHeader *)buf;
1336 +
1337 +    int i;
1338 +
1339 +    DPRINTF("VMA WRITE HEADER\n");
1340 +
1341 +    if (vmaw->status < 0) {
1342 +        return vmaw->status;
1343 +    }
1344 +
1345 +    memset(buf, 0, sizeof(buf));
1346 +
1347 +    head->magic = VMA_MAGIC;
1348 +    head->version = GUINT32_TO_BE(1); /* v1 */
1349 +    memcpy(head->uuid, vmaw->uuid, 16);
1350 +
1351 +    time_t ctime = time(NULL);
1352 +    head->ctime = GUINT64_TO_BE(ctime);
1353 +
1354 +    if (!vmaw->stream_count) {
1355 +        return -1;
1356 +    }
1357 +
1358 +    for (i = 0; i < VMA_MAX_CONFIGS; i++) {
1359 +        head->config_names[i] = GUINT32_TO_BE(vmaw->config_names[i]);
1360 +        head->config_data[i] = GUINT32_TO_BE(vmaw->config_data[i]);
1361 +    }
1362 +
1363 +    /* 32 bytes per device (12 used currently) = 8192 bytes max */
1364 +    for (i = 1; i <= 254; i++) {
1365 +        VmaStreamInfo *si = &vmaw->stream_info[i];
1366 +        if (si->size) {
1367 +            assert(si->devname);
1368 +            uint32_t devname_ptr = allocate_header_string(vmaw, si->devname);
1369 +            if (!devname_ptr) {
1370 +                return -1;
1371 +            }
1372 +            head->dev_info[i].devname_ptr = GUINT32_TO_BE(devname_ptr);
1373 +            head->dev_info[i].size = GUINT64_TO_BE(si->size);
1374 +        }
1375 +    }
1376 +
1377 +    uint32_t header_size = sizeof(VmaHeader) + vmaw->header_blob_table_size;
1378 +    head->header_size = GUINT32_TO_BE(header_size);
1379 +
1380 +    if (header_size > sizeof(buf)) {
1381 +        return -1; /* just to be sure */
1382 +    }
1383 +
1384 +    uint32_t blob_buffer_offset = sizeof(VmaHeader);
1385 +    memcpy(buf + blob_buffer_offset, vmaw->header_blob_table,
1386 +           vmaw->header_blob_table_size);
1387 +    head->blob_buffer_offset = GUINT32_TO_BE(blob_buffer_offset);
1388 +    head->blob_buffer_size = GUINT32_TO_BE(vmaw->header_blob_table_pos);
1389 +
1390 +    g_checksum_reset(vmaw->md5csum);
1391 +    g_checksum_update(vmaw->md5csum, (const guchar *)buf, header_size);
1392 +    gsize csize = 16;
1393 +    g_checksum_get_digest(vmaw->md5csum, (guint8 *)(head->md5sum), &csize);
1394 +
1395 +    return vma_queue_write(vmaw, buf, header_size);
1396 +}
1397 +
1398 +static int coroutine_fn vma_writer_flush(VmaWriter *vmaw)
1399 +{
1400 +    assert(vmaw);
1401 +
1402 +    int ret;
1403 +    int i;
1404 +
1405 +    if (vmaw->status < 0) {
1406 +        return vmaw->status;
1407 +    }
1408 +
1409 +    if (!vmaw->header_written) {
1410 +        vmaw->header_written = true;
1411 +        ret = vma_write_header(vmaw);
1412 +        if (ret < 0) {
1413 +            vma_writer_set_error(vmaw, "vma_writer_flush: write header failed");
1414 +            return ret;
1415 +        }
1416 +    }
1417 +
1418 +    DPRINTF("VMA WRITE FLUSH %d %d\n", vmaw->outbuf_count, vmaw->outbuf_pos);
1419 +
1420 +
1421 +    VmaExtentHeader *ehead = (VmaExtentHeader *)vmaw->outbuf;
1422 +
1423 +    ehead->magic = VMA_EXTENT_MAGIC;
1424 +    ehead->reserved1 = 0;
1425 +
1426 +    for (i = 0; i < VMA_BLOCKS_PER_EXTENT; i++) {
1427 +        ehead->blockinfo[i] = GUINT64_TO_BE(vmaw->outbuf_block_info[i]);
1428 +    }
1429 +
1430 +    guint16 block_count = (vmaw->outbuf_pos - VMA_EXTENT_HEADER_SIZE) /
1431 +        VMA_BLOCK_SIZE;
1432 +
1433 +    ehead->block_count = GUINT16_TO_BE(block_count);
1434 +
1435 +    memcpy(ehead->uuid, vmaw->uuid, sizeof(ehead->uuid));
1436 +    memset(ehead->md5sum, 0, sizeof(ehead->md5sum));
1437 +
1438 +    g_checksum_reset(vmaw->md5csum);
1439 +    g_checksum_update(vmaw->md5csum, vmaw->outbuf, VMA_EXTENT_HEADER_SIZE);
1440 +    gsize csize = 16;
1441 +    g_checksum_get_digest(vmaw->md5csum, ehead->md5sum, &csize);
1442 +
1443 +    int bytes = vmaw->outbuf_pos;
1444 +    ret = vma_queue_write(vmaw, vmaw->outbuf, bytes);
1445 +    if (ret != bytes) {
1446 +        vma_writer_set_error(vmaw, "vma_writer_flush: failed write");
1447 +    }
1448 +
1449 +    vmaw->outbuf_count = 0;
1450 +    vmaw->outbuf_pos = VMA_EXTENT_HEADER_SIZE;
1451 +
1452 +    for (i = 0; i < VMA_BLOCKS_PER_EXTENT; i++) {
1453 +        vmaw->outbuf_block_info[i] = 0;
1454 +    }
1455 +
1456 +    return vmaw->status;
1457 +}
1458 +
1459 +static int vma_count_open_streams(VmaWriter *vmaw)
1460 +{
1461 +    g_assert(vmaw != NULL);
1462 +
1463 +    int i;
1464 +    int open_drives = 0;
1465 +    for (i = 0; i <= 255; i++) {
1466 +        if (vmaw->stream_info[i].size && !vmaw->stream_info[i].finished) {
1467 +            open_drives++;
1468 +        }
1469 +    }
1470 +
1471 +    return open_drives;
1472 +}
1473 +
1474 +/**
1475 + * all jobs should call this when there is no more data
1476 + * Returns: number of remaining stream (0 ==> finished)
1477 + */
1478 +int coroutine_fn
1479 +vma_writer_close_stream(VmaWriter *vmaw, uint8_t dev_id)
1480 +{
1481 +    g_assert(vmaw != NULL);
1482 +
1483 +    DPRINTF("vma_writer_set_status %d\n", dev_id);
1484 +    if (!vmaw->stream_info[dev_id].size) {
1485 +        vma_writer_set_error(vmaw, "vma_writer_close_stream: "
1486 +                             "no such stream %d", dev_id);
1487 +        return -1;
1488 +    }
1489 +    if (vmaw->stream_info[dev_id].finished) {
1490 +        vma_writer_set_error(vmaw, "vma_writer_close_stream: "
1491 +                             "stream already closed %d", dev_id);
1492 +        return -1;
1493 +    }
1494 +
1495 +    vmaw->stream_info[dev_id].finished = true;
1496 +
1497 +    int open_drives = vma_count_open_streams(vmaw);
1498 +
1499 +    if (open_drives <= 0) {
1500 +        DPRINTF("vma_writer_set_status all drives completed\n");
1501 +        qemu_co_mutex_lock(&vmaw->flush_lock);
1502 +        int ret = vma_writer_flush(vmaw);
1503 +        qemu_co_mutex_unlock(&vmaw->flush_lock);
1504 +        if (ret < 0) {
1505 +            vma_writer_set_error(vmaw, "vma_writer_close_stream: flush failed");
1506 +        }
1507 +    }
1508 +
1509 +    return open_drives;
1510 +}
1511 +
1512 +int vma_writer_get_status(VmaWriter *vmaw, VmaStatus *status)
1513 +{
1514 +    int i;
1515 +
1516 +    g_assert(vmaw != NULL);
1517 +
1518 +    if (status) {
1519 +        status->status = vmaw->status;
1520 +        g_strlcpy(status->errmsg, vmaw->errmsg, sizeof(status->errmsg));
1521 +        for (i = 0; i <= 255; i++) {
1522 +            status->stream_info[i] = vmaw->stream_info[i];
1523 +        }
1524 +
1525 +        uuid_unparse_lower(vmaw->uuid, status->uuid_str);
1526 +    }
1527 +
1528 +    status->closed = vmaw->closed;
1529 +
1530 +    return vmaw->status;
1531 +}
1532 +
1533 +static int vma_writer_get_buffer(VmaWriter *vmaw)
1534 +{
1535 +
1536 +    /* wait until buffer is available */
1537 +    while (vmaw->outbuf_count >= (VMA_BLOCKS_PER_EXTENT - 1)) {
1538 +        int res = 0;
1539 +
1540 +        qemu_co_mutex_lock(&vmaw->flush_lock);
1541 +        res = vma_writer_flush(vmaw);
1542 +        qemu_co_mutex_unlock(&vmaw->flush_lock);
1543 +
1544 +        if (res < 0) {
1545 +            vma_writer_set_error(vmaw, "vma_writer_get_buffer: flush failed");
1546 +            return -1;
1547 +        }
1548 +    }
1549 +
1550 +    return 0;
1551 +}
1552 +
1553 +
1554 +int64_t coroutine_fn
1555 +vma_writer_write(VmaWriter *vmaw, uint8_t dev_id, int64_t cluster_num,
1556 +                 unsigned char *buf, size_t *zero_bytes)
1557 +{
1558 +    g_assert(vmaw != NULL);
1559 +    g_assert(zero_bytes != NULL);
1560 +
1561 +    *zero_bytes = 0;
1562 +
1563 +    if (vmaw->status < 0) {
1564 +        return vmaw->status;
1565 +    }
1566 +
1567 +    if (!dev_id || !vmaw->stream_info[dev_id].size) {
1568 +        vma_writer_set_error(vmaw, "vma_writer_write: "
1569 +                             "no such stream %d", dev_id);
1570 +        return -1;
1571 +    }
1572 +
1573 +    if (vmaw->stream_info[dev_id].finished) {
1574 +        vma_writer_set_error(vmaw, "vma_writer_write: "
1575 +                             "stream already closed %d", dev_id);
1576 +        return -1;
1577 +    }
1578 +
1579 +
1580 +    if (cluster_num >= (((uint64_t)1)<<32)) {
1581 +        vma_writer_set_error(vmaw, "vma_writer_write: "
1582 +                             "cluster number out of range");
1583 +        return -1;
1584 +    }
1585 +
1586 +    if (dev_id == vmaw->vmstate_stream) {
1587 +        if (cluster_num != vmaw->vmstate_clusters) {
1588 +            vma_writer_set_error(vmaw, "vma_writer_write: "
1589 +                                 "non sequential vmstate write");
1590 +        }
1591 +        vmaw->vmstate_clusters++;
1592 +    } else if (cluster_num >= vmaw->stream_info[dev_id].cluster_count) {
1593 +        vma_writer_set_error(vmaw, "vma_writer_write: cluster number too big");
1594 +        return -1;
1595 +    }
1596 +
1597 +    /* wait until buffer is available */
1598 +    if (vma_writer_get_buffer(vmaw) < 0) {
1599 +        vma_writer_set_error(vmaw, "vma_writer_write: "
1600 +                             "vma_writer_get_buffer failed");
1601 +        return -1;
1602 +    }
1603 +
1604 +    DPRINTF("VMA WRITE %zd\n", cluster_num);
1605 +
1606 +    int i;
1607 +    int bit = 1;
1608 +    uint16_t mask = 0;
1609 +    for (i = 0; i < 16; i++) {
1610 +        unsigned char *vmablock = buf + (i*VMA_BLOCK_SIZE);
1611 +        if (buffer_is_zero(vmablock, VMA_BLOCK_SIZE)) {
1612 +            DPRINTF("VMA WRITE %zd ZERO BLOCK %d\n", cluster_num, i);
1613 +            vmaw->stream_info[dev_id].zero_bytes += VMA_BLOCK_SIZE;
1614 +            *zero_bytes += VMA_BLOCK_SIZE;
1615 +        } else {
1616 +            mask |= bit;
1617 +            memcpy(vmaw->outbuf + vmaw->outbuf_pos, vmablock, VMA_BLOCK_SIZE);
1618 +            vmaw->outbuf_pos += VMA_BLOCK_SIZE;
1619 +
1620 +            vmaw->delay_ns = ratelimit_calculate_delay(&vmaw->limit,
1621 +                                                       VMA_BLOCK_SIZE);
1622 +            if (vmaw->delay_ns) {
1623 +                co_sleep_ns(rt_clock, vmaw->delay_ns);
1624 +            }
1625 +        }
1626 +
1627 +        bit = bit << 1;
1628 +    }
1629 +
1630 +    uint64_t block_info = ((uint64_t)mask) << (32+16);
1631 +    block_info |= ((uint64_t)dev_id) << 32;
1632 +    block_info |= (cluster_num & 0xffffffff);
1633 +    vmaw->outbuf_block_info[vmaw->outbuf_count] =  block_info;
1634 +
1635 +    DPRINTF("VMA WRITE MASK %zd %zx\n", cluster_num, block_info);
1636 +
1637 +    vmaw->outbuf_count++;
1638 +
1639 +    /** NOTE: We allways write whole clusters, but we correctly set
1640 +     * transferred bytes. So transferred == size when when everything
1641 +     * went OK.
1642 +     */
1643 +    size_t transferred = VMA_CLUSTER_SIZE;
1644 +
1645 +    if (dev_id != vmaw->vmstate_stream) {
1646 +        uint64_t last = (cluster_num + 1) * VMA_CLUSTER_SIZE;
1647 +        if (last > vmaw->stream_info[dev_id].size) {
1648 +            uint64_t diff = last - vmaw->stream_info[dev_id].size;
1649 +            if (diff >= VMA_CLUSTER_SIZE) {
1650 +                vma_writer_set_error(vmaw, "vma_writer_write: "
1651 +                                     "read after last cluster");
1652 +                return -1;
1653 +            }
1654 +            transferred -= diff;
1655 +        }
1656 +    }
1657 +
1658 +    vmaw->stream_info[dev_id].transferred += transferred;
1659 +
1660 +    return transferred;
1661 +}
1662 +
1663 +int vma_writer_close(VmaWriter *vmaw, Error **errp)
1664 +{
1665 +    g_assert(vmaw != NULL);
1666 +
1667 +    int i;
1668 +
1669 +    vma_queue_flush(vmaw);
1670 +
1671 +    /* this should not happen - just to be sure */
1672 +    while (!qemu_co_queue_empty(&vmaw->wqueue)) {
1673 +        DPRINTF("vma_writer_close wait\n");
1674 +        co_sleep_ns(rt_clock, 1000000);
1675 +    }
1676 +
1677 +    if (vmaw->cmd) {
1678 +        if (pclose(vmaw->cmd) < 0) {
1679 +            vma_writer_set_error(vmaw, "vma_writer_close: "
1680 +                                 "pclose failed - %s", strerror(errno));
1681 +        }
1682 +    } else {
1683 +        if (close(vmaw->fd) < 0) {
1684 +            vma_writer_set_error(vmaw, "vma_writer_close: "
1685 +                                 "close failed - %s", strerror(errno));
1686 +        }
1687 +    }
1688 +
1689 +    for (i = 0; i <= 255; i++) {
1690 +        VmaStreamInfo *si = &vmaw->stream_info[i];
1691 +        if (si->size) {
1692 +            if (!si->finished) {
1693 +                vma_writer_set_error(vmaw, "vma_writer_close: "
1694 +                                     "detected open stream '%s'", si->devname);
1695 +            } else if ((si->transferred != si->size) &&
1696 +                       (i != vmaw->vmstate_stream)) {
1697 +                vma_writer_set_error(vmaw, "vma_writer_close: "
1698 +                                     "incomplete stream '%s' (%zd != %zd)",
1699 +                                     si->devname, si->transferred, si->size);
1700 +            }
1701 +        }
1702 +    }
1703 +
1704 +    for (i = 0; i <= 255; i++) {
1705 +        vmaw->stream_info[i].finished = 1; /* mark as closed */
1706 +    }
1707 +
1708 +    vmaw->closed = 1;
1709 +
1710 +    if (vmaw->status < 0 && *errp == NULL) {
1711 +        error_setg(errp, "%s", vmaw->errmsg);
1712 +    }
1713 +
1714 +    return vmaw->status;
1715 +}
1716 +
1717 +void vma_writer_destroy(VmaWriter *vmaw)
1718 +{
1719 +    assert(vmaw);
1720 +
1721 +    int i;
1722 +
1723 +    for (i = 0; i <= 255; i++) {
1724 +        if (vmaw->stream_info[i].devname) {
1725 +            g_free(vmaw->stream_info[i].devname);
1726 +        }
1727 +    }
1728 +
1729 +    if (vmaw->md5csum) {
1730 +        g_checksum_free(vmaw->md5csum);
1731 +    }
1732 +
1733 +    g_free(vmaw);
1734 +}
1735 +
1736 +/* backup driver plugin */
1737 +
1738 +static int vma_dump_cb(void *opaque, uint8_t dev_id, int64_t cluster_num,
1739 +                       unsigned char *buf, size_t *zero_bytes)
1740 +{
1741 +    VmaWriter *vmaw = opaque;
1742 +
1743 +    return vma_writer_write(vmaw, dev_id, cluster_num, buf, zero_bytes);
1744 +}
1745 +
1746 +static int vma_close_cb(void *opaque, Error **errp)
1747 +{
1748 +    VmaWriter *vmaw = opaque;
1749 +
1750 +    int res = vma_writer_close(vmaw, errp);
1751 +    vma_writer_destroy(vmaw);
1752 +
1753 +    return res;
1754 +}
1755 +
1756 +static int vma_complete_cb(void *opaque, uint8_t dev_id, int ret)
1757 +{
1758 +    VmaWriter *vmaw = opaque;
1759 +
1760 +    if (ret < 0) {
1761 +        vma_writer_set_error(vmaw, "backup_complete_cb %d", ret);
1762 +    }
1763 +
1764 +    return vma_writer_close_stream(vmaw, dev_id);
1765 +}
1766 +
1767 +static int vma_register_stream_cb(void *opaque, const char *devname,
1768 +                                  size_t size)
1769 +{
1770 +    VmaWriter *vmaw = opaque;
1771 +
1772 +    return vma_writer_register_stream(vmaw, devname, size);
1773 +}
1774 +
1775 +static int vma_register_config_cb(void *opaque, const char *name,
1776 +                                  gpointer data, size_t data_len)
1777 +{
1778 +    VmaWriter *vmaw = opaque;
1779 +
1780 +    return vma_writer_add_config(vmaw, name, data, data_len);
1781 +}
1782 +
1783 +static void *vma_open_cb(const char *filename, uuid_t uuid, int64_t speed,
1784 +                         Error **errp)
1785 +{
1786 +    return vma_writer_create(filename, uuid, speed, errp);
1787 +}
1788 +
1789 +const BackupDriver backup_vma_driver = {
1790 +    .format = "vma",
1791 +    .open_cb = vma_open_cb,
1792 +    .close_cb = vma_close_cb,
1793 +    .register_config_cb = vma_register_config_cb,
1794 +    .register_stream_cb = vma_register_stream_cb,
1795 +    .dump_cb = vma_dump_cb,
1796 +    .complete_cb = vma_complete_cb,
1797 +};
1798 +
1799 diff --git a/vma.c b/vma.c
1800 new file mode 100644
1801 index 0000000..69af80c
1802 --- /dev/null
1803 +++ b/vma.c
1804 @@ -0,0 +1,550 @@
1805 +/*
1806 + * VMA: Virtual Machine Archive
1807 + *
1808 + * Copyright (C) 2012 Proxmox Server Solutions
1809 + *
1810 + * Authors:
1811 + *  Dietmar Maurer (dietmar@proxmox.com)
1812 + *
1813 + * This work is licensed under the terms of the GNU GPL, version 2 or later.
1814 + * See the COPYING file in the top-level directory.
1815 + *
1816 + */
1817 +
1818 +#include <stdio.h>
1819 +#include <errno.h>
1820 +#include <unistd.h>
1821 +#include <stdio.h>
1822 +#include <string.h>
1823 +#include <sys/types.h>
1824 +#include <sys/stat.h>
1825 +#include <fcntl.h>
1826 +#include <glib.h>
1827 +
1828 +#include "qemu-common.h"
1829 +#include "qemu-option.h"
1830 +#include "qemu-error.h"
1831 +#include "osdep.h"
1832 +#include "sysemu.h"
1833 +#include "block_int.h"
1834 +#include <stdio.h>
1835 +#include "vma.h"
1836 +
1837 +static void help(void)
1838 +{
1839 +    const char *help_msg =
1840 +        "usage: vma command [command options]\n"
1841 +        "\n"
1842 +        "vma list <filename>\n"
1843 +        "vma create <filename> [-c config] <archive> pathname ...\n"
1844 +        "vma extract <filename> [-r] <targetdir>\n"
1845 +        ;
1846 +
1847 +    printf("%s", help_msg);
1848 +    exit(1);
1849 +}
1850 +
1851 +static const char *extract_devname(const char *path, char **devname, int index)
1852 +{
1853 +    assert(path);
1854 +
1855 +    const char *sep = strchr(path, '=');
1856 +
1857 +    if (sep) {
1858 +        *devname = g_strndup(path, sep - path);
1859 +        path = sep + 1;
1860 +    } else {
1861 +        if (index >= 0) {
1862 +            *devname = g_strdup_printf("disk%d", index);
1863 +        } else {
1864 +            *devname = NULL;
1865 +        }
1866 +    }
1867 +
1868 +    return path;
1869 +}
1870 +
1871 +static void print_content(VmaReader *vmar)
1872 +{
1873 +    assert(vmar);
1874 +
1875 +    VmaHeader *head = vma_reader_get_header(vmar);
1876 +
1877 +    printf("CTIME: %s", ctime(&head->ctime));
1878 +
1879 +    GList *l = vma_reader_get_config_data(vmar);
1880 +    while (l && l->data) {
1881 +        VmaConfigData *cdata = (VmaConfigData *)l->data;
1882 +        l = g_list_next(l);
1883 +        printf("CFG: size: %d name: %s\n", cdata->len, cdata->name);
1884 +    }
1885 +
1886 +    int i;
1887 +    VmaDeviceInfo *di;
1888 +    for (i = 1; i < 255; i++) {
1889 +        di = vma_reader_get_device_info(vmar, i);
1890 +        if (di) {
1891 +            if (strcmp(di->devname, "vmstate") == 0) {
1892 +                printf("VMSTATE: dev_id=%d memory: %zd\n", i, di->size);
1893 +            } else {
1894 +                printf("DEV: dev_id=%d size: %zd devname: %s\n",
1895 +                       i, di->size, di->devname);
1896 +            }
1897 +        }
1898 +    }
1899 +}
1900 +
1901 +static int list_content(int argc, char **argv)
1902 +{
1903 +    int c, ret = 0;
1904 +    const char *filename;
1905 +
1906 +    for (;;) {
1907 +        c = getopt(argc, argv, "h");
1908 +        if (c == -1) {
1909 +            break;
1910 +        }
1911 +        switch (c) {
1912 +        case '?':
1913 +        case 'h':
1914 +            help();
1915 +            break;
1916 +        default:
1917 +            g_assert_not_reached();
1918 +        }
1919 +    }
1920 +
1921 +    /* Get the filename */
1922 +    if ((optind + 1) != argc) {
1923 +        help();
1924 +    }
1925 +    filename = argv[optind++];
1926 +
1927 +    Error *errp = NULL;
1928 +    VmaReader *vmar = vma_reader_create(filename, &errp);
1929 +
1930 +    if (!vmar) {
1931 +        g_error("%s", error_get_pretty(errp));
1932 +    }
1933 +
1934 +    print_content(vmar);
1935 +
1936 +    vma_reader_destroy(vmar);
1937 +
1938 +    return ret;
1939 +}
1940 +
1941 +typedef struct RestoreMap {
1942 +    char *devname;
1943 +    char *path;
1944 +    bool write_zero;
1945 +} RestoreMap;
1946 +
1947 +static int extract_content(int argc, char **argv)
1948 +{
1949 +    int c, ret = 0;
1950 +    const char *filename;
1951 +    const char *dirname;
1952 +    int readmap = 0;
1953 +
1954 +    for (;;) {
1955 +        c = getopt(argc, argv, "hr");
1956 +        if (c == -1) {
1957 +            break;
1958 +        }
1959 +        switch (c) {
1960 +        case '?':
1961 +        case 'h':
1962 +            help();
1963 +            break;
1964 +        case 'r':
1965 +            readmap = 1;
1966 +            break;
1967 +        default:
1968 +            help();
1969 +        }
1970 +    }
1971 +
1972 +    /* Get the filename */
1973 +    if ((optind + 2) != argc) {
1974 +        help();
1975 +    }
1976 +    filename = argv[optind++];
1977 +    dirname = argv[optind++];
1978 +
1979 +    Error *errp = NULL;
1980 +    VmaReader *vmar = vma_reader_create(filename, &errp);
1981 +
1982 +    if (!vmar) {
1983 +        g_error("%s", error_get_pretty(errp));
1984 +    }
1985 +
1986 +    if (mkdir(dirname, 0777) < 0) {
1987 +        g_error("unable to create target directory %s - %s",
1988 +                dirname, strerror(errno));
1989 +    }
1990 +
1991 +    GList *l = vma_reader_get_config_data(vmar);
1992 +    while (l && l->data) {
1993 +        VmaConfigData *cdata = (VmaConfigData *)l->data;
1994 +        l = g_list_next(l);
1995 +        char *cfgfn = g_strdup_printf("%s/%s", dirname, cdata->name);
1996 +        GError *err = NULL;
1997 +        if (!g_file_set_contents(cfgfn, (gchar *)cdata->data, cdata->len,
1998 +                                 &err)) {
1999 +            g_error("Unable to write file: %s", err->message);
2000 +        }
2001 +    }
2002 +
2003 +    GHashTable *devmap = g_hash_table_new(g_str_hash, g_str_equal);
2004 +
2005 +    if (readmap) {
2006 +        print_content(vmar);
2007 +
2008 +        while (1) {
2009 +            char inbuf[8192];
2010 +            char *line = fgets(inbuf, sizeof(inbuf), stdin);
2011 +            if (!line || line[0] == '\0' || !strcmp(line, "done\n")) {
2012 +                break;
2013 +            }
2014 +            int len = strlen(line);
2015 +            if (line[len - 1] == '\n') {
2016 +                line[len - 1] = '\0';
2017 +                if (len == 1) {
2018 +                    break;
2019 +                }
2020 +            }
2021 +
2022 +            const char *path;
2023 +            bool write_zero;
2024 +            if (line[0] == '0' && line[1] == ':') {
2025 +                path = inbuf + 2;
2026 +                write_zero = false;
2027 +            } else if (line[0] == '1' && line[1] == ':') {
2028 +                path = inbuf + 2;
2029 +                write_zero = true;
2030 +           } else {
2031 +                g_error("read map failed - parse error ('%s')", inbuf);
2032 +            }
2033 +
2034 +            char *devname = NULL;
2035 +            path = extract_devname(path, &devname, -1);
2036 +            if (!devname) {
2037 +                g_error("read map failed - no dev name specified ('%s')",
2038 +                        inbuf);
2039 +            }
2040 +
2041 +            printf("TEST %s %s\n", path, devname);
2042 +
2043 +            RestoreMap *map = g_new0(RestoreMap, 1);
2044 +            map->devname = g_strdup(devname);
2045 +            map->path = g_strdup(path);
2046 +            map->write_zero = write_zero;
2047 +
2048 +            g_hash_table_insert(devmap, map->devname, map);
2049 +
2050 +        };
2051 +    }
2052 +
2053 +    int i;
2054 +    int vmstate_fd = -1;
2055 +    guint8 vmstate_stream = 0;
2056 +
2057 +    for (i = 1; i < 255; i++) {
2058 +        VmaDeviceInfo *di = vma_reader_get_device_info(vmar, i);
2059 +        if (di && (strcmp(di->devname, "vmstate") == 0)) {
2060 +            vmstate_stream = i;
2061 +            char *statefn = g_strdup_printf("%s/vmstate.bin", dirname);
2062 +            vmstate_fd = open(statefn, O_WRONLY|O_CREAT|O_EXCL, 0644);
2063 +            if (vmstate_fd < 0) {
2064 +                g_error("create vmstate file '%s' failed - %s", statefn,
2065 +                        strerror(errno));
2066 +            }
2067 +            g_free(statefn);
2068 +        } else if (di) {
2069 +            char *devfn = NULL;
2070 +            int flags = BDRV_O_RDWR|BDRV_O_CACHE_WB;
2071 +            bool write_zero = true;
2072 +
2073 +            if (readmap) {
2074 +                RestoreMap *map;
2075 +                map = (RestoreMap *)g_hash_table_lookup(devmap, di->devname);
2076 +                if (map == NULL) {
2077 +                    g_error("no device name mapping for %s", di->devname);
2078 +                }
2079 +                devfn = map->path;
2080 +                write_zero = map->write_zero;
2081 +            } else {
2082 +                devfn = g_strdup_printf("%s/tmp-disk-%s.raw",
2083 +                                        dirname, di->devname);
2084 +                printf("DEVINFO %s %zd\n", devfn, di->size);
2085 +
2086 +                if (bdrv_img_create(devfn, "raw", NULL, NULL, NULL,
2087 +                                    di->size, flags)) {
2088 +                    g_error("can't create file %s", devfn);
2089 +                }
2090 +
2091 +                /* Note: we created an empty file above, so there is no
2092 +                 * need to write zeroes (so we generate a sparse file)
2093 +                 */
2094 +                write_zero = false;
2095 +            }
2096 +
2097 +            BlockDriverState *bs = NULL;
2098 +            if (bdrv_file_open(&bs, devfn, flags)) {
2099 +                g_error("can't open file %s", devfn);
2100 +            }
2101 +            if (vma_reader_register_bs(vmar, i, bs, write_zero, &errp) < 0) {
2102 +                g_error("%s", error_get_pretty(errp));
2103 +            }
2104 +
2105 +            if (!readmap) {
2106 +                g_free(devfn);
2107 +            }
2108 +        }
2109 +    }
2110 +
2111 +    if (vma_reader_restore(vmar, vmstate_fd, &errp) < 0) {
2112 +        g_error("restore failed - %s", error_get_pretty(errp));
2113 +    }
2114 +
2115 +    if (!readmap) {
2116 +        for (i = 1; i < 255; i++) {
2117 +            VmaDeviceInfo *di = vma_reader_get_device_info(vmar, i);
2118 +            if (di && (i != vmstate_stream)) {
2119 +                char *tmpfn = g_strdup_printf("%s/tmp-disk-%s.raw",
2120 +                                              dirname, di->devname);
2121 +                char *fn = g_strdup_printf("%s/disk-%s.raw",
2122 +                                           dirname, di->devname);
2123 +                if (rename(tmpfn, fn) != 0) {
2124 +                    g_error("rename %s to %s failed - %s",
2125 +                            tmpfn, fn, strerror(errno));
2126 +                }
2127 +            }
2128 +        }
2129 +    }
2130 +
2131 +    vma_reader_destroy(vmar);
2132 +
2133 +    bdrv_close_all();
2134 +
2135 +    return ret;
2136 +}
2137 +
2138 +typedef struct BackupCB {
2139 +    VmaWriter *vmaw;
2140 +    uint8_t dev_id;
2141 +} BackupCB;
2142 +
2143 +static int backup_dump_cb(void *opaque, BlockDriverState *bs,
2144 +                          int64_t cluster_num, unsigned char *buf)
2145 +{
2146 +    BackupCB *bcb = opaque;
2147 +    size_t zb = 0;
2148 +    if (vma_writer_write(bcb->vmaw, bcb->dev_id, cluster_num, buf, &zb) < 0) {
2149 +        g_warning("backup_dump_cb vma_writer_write failed");
2150 +        return -1;
2151 +    }
2152 +
2153 +    return 0;
2154 +}
2155 +
2156 +static void backup_complete_cb(void *opaque, int ret)
2157 +{
2158 +    BackupCB *bcb = opaque;
2159 +
2160 +    if (ret < 0) {
2161 +        vma_writer_set_error(bcb->vmaw, "backup_complete_cb %d", ret);
2162 +    }
2163 +
2164 +    if (vma_writer_close_stream(bcb->vmaw, bcb->dev_id) <= 0) {
2165 +        Error *err = NULL;
2166 +        if (vma_writer_close(bcb->vmaw, &err) != 0) {
2167 +            g_warning("vma_writer_close failed %s", error_get_pretty(err));
2168 +        }
2169 +    }
2170 +}
2171 +
2172 +static int create_archive(int argc, char **argv)
2173 +{
2174 +    int i, c, res;
2175 +    int verbose = 0;
2176 +    const char *archivename;
2177 +    GList *config_files = NULL;
2178 +
2179 +    for (;;) {
2180 +        c = getopt(argc, argv, "hvc:");
2181 +        if (c == -1) {
2182 +            break;
2183 +        }
2184 +        switch (c) {
2185 +        case '?':
2186 +        case 'h':
2187 +            help();
2188 +            break;
2189 +        case 'c':
2190 +            config_files = g_list_append(config_files, optarg);
2191 +            break;
2192 +        case 'v':
2193 +            verbose = 1;
2194 +            break;
2195 +        default:
2196 +            g_assert_not_reached();
2197 +        }
2198 +    }
2199 +
2200 +
2201 +    /* make sure we have archive name and at least one path */
2202 +    if ((optind + 2) > argc) {
2203 +        help();
2204 +    }
2205 +
2206 +    archivename = argv[optind++];
2207 +
2208 +    uuid_t uuid;
2209 +    uuid_generate(uuid);
2210 +
2211 +    Error *local_err = NULL;
2212 +    VmaWriter *vmaw = vma_writer_create(archivename, uuid, 0, &local_err);
2213 +
2214 +    if (vmaw == NULL) {
2215 +        g_error("%s", error_get_pretty(local_err));
2216 +    }
2217 +
2218 +    GList *l = config_files;
2219 +    while (l && l->data) {
2220 +        char *name = l->data;
2221 +        char *cdata = NULL;
2222 +        gsize clen = 0;
2223 +        GError *err = NULL;
2224 +        if (!g_file_get_contents(name, &cdata, &clen, &err)) {
2225 +            unlink(archivename);
2226 +            g_error("Unable to read file: %s", err->message);
2227 +        }
2228 +
2229 +        if (vma_writer_add_config(vmaw, name, cdata, clen) != 0) {
2230 +            unlink(archivename);
2231 +            g_error("Unable to append config data %s (len = %zd)",
2232 +                    name, clen);
2233 +        }
2234 +        l = g_list_next(l);
2235 +    }
2236 +
2237 +    int ind = 0;
2238 +    while (optind < argc) {
2239 +        const char *path = argv[optind++];
2240 +        char *devname = NULL;
2241 +        path = extract_devname(path, &devname, ind++);
2242 +
2243 +        BlockDriver *drv = NULL;
2244 +        BlockDriverState *bs = bdrv_new(devname);
2245 +
2246 +        res = bdrv_open(bs, path, BDRV_O_CACHE_WB , drv);
2247 +        if (res < 0) {
2248 +            unlink(archivename);
2249 +            g_error("bdrv_open '%s' failed", path);
2250 +        }
2251 +        int64_t size = bdrv_getlength(bs);
2252 +        int dev_id = vma_writer_register_stream(vmaw, devname, size);
2253 +        if (dev_id <= 0) {
2254 +            unlink(archivename);
2255 +            g_error("vma_writer_register_stream '%s' failed", devname);
2256 +        }
2257 +
2258 +        BackupCB *bcb = g_new0(BackupCB, 1);
2259 +        bcb->vmaw = vmaw;
2260 +        bcb->dev_id = dev_id;
2261 +
2262 +        if (backup_job_start(bs, backup_dump_cb, backup_complete_cb, bcb) < 0) {
2263 +            unlink(archivename);
2264 +            g_error("backup_job_start failed");
2265 +        }
2266 +    }
2267 +
2268 +    VmaStatus vmastat;
2269 +    int percent = 0;
2270 +    int last_percent = -1;
2271 +
2272 +    while (1) {
2273 +        main_loop_wait(false);
2274 +        vma_writer_get_status(vmaw, &vmastat);
2275 +
2276 +        if (verbose) {
2277 +
2278 +            uint64_t total = 0;
2279 +            uint64_t transferred = 0;
2280 +            uint64_t zero_bytes = 0;
2281 +
2282 +            int i;
2283 +            for (i = 0; i < 256; i++) {
2284 +                if (vmastat.stream_info[i].size) {
2285 +                    total += vmastat.stream_info[i].size;
2286 +                    transferred += vmastat.stream_info[i].transferred;
2287 +                    zero_bytes += vmastat.stream_info[i].zero_bytes;
2288 +                }
2289 +            }
2290 +            percent = (transferred*100)/total;
2291 +            if (percent != last_percent) {
2292 +                printf("progress %d%% %zd/%zd %zd\n", percent,
2293 +                       transferred, total, zero_bytes);
2294 +
2295 +                last_percent = percent;
2296 +            }
2297 +        }
2298 +
2299 +        if (vmastat.closed) {
2300 +            break;
2301 +        }
2302 +    }
2303 +
2304 +    bdrv_drain_all();
2305 +
2306 +    vma_writer_get_status(vmaw, &vmastat);
2307 +
2308 +    if (verbose) {
2309 +        for (i = 0; i < 256; i++) {
2310 +            VmaStreamInfo *si = &vmastat.stream_info[i];
2311 +            if (si->size) {
2312 +                printf("image %s: size=%zd zeros=%zd saved=%zd\n", si->devname,
2313 +                       si->size, si->zero_bytes, si->size - si->zero_bytes);
2314 +            }
2315 +        }
2316 +    }
2317 +
2318 +    if (vmastat.status < 0) {
2319 +        unlink(archivename);
2320 +        g_error("creating vma archive failed");
2321 +    }
2322 +
2323 +    return 0;
2324 +}
2325 +
2326 +int main(int argc, char **argv)
2327 +{
2328 +    const char *cmdname;
2329 +
2330 +    error_set_progname(argv[0]);
2331 +
2332 +    qemu_init_main_loop();
2333 +
2334 +    bdrv_init();
2335 +
2336 +    if (argc < 2) {
2337 +        help();
2338 +    }
2339 +
2340 +    cmdname = argv[1];
2341 +    argc--; argv++;
2342 +
2343 +
2344 +    if (!strcmp(cmdname, "list")) {
2345 +        return list_content(argc, argv);
2346 +    } else if (!strcmp(cmdname, "create")) {
2347 +        return create_archive(argc, argv);
2348 +    } else if (!strcmp(cmdname, "extract")) {
2349 +        return extract_content(argc, argv);
2350 +    }
2351 +
2352 +    help();
2353 +    return 0;
2354 +}
2355 diff --git a/vma.h b/vma.h
2356 new file mode 100644
2357 index 0000000..10800a1
2358 --- /dev/null
2359 +++ b/vma.h
2360 @@ -0,0 +1,145 @@
2361 +/*
2362 + * VMA: Virtual Machine Archive
2363 + *
2364 + * Copyright (C) Proxmox Server Solutions
2365 + *
2366 + * Authors:
2367 + *  Dietmar Maurer (dietmar@proxmox.com)
2368 + *
2369 + * This work is licensed under the terms of the GNU GPL, version 2 or later.
2370 + * See the COPYING file in the top-level directory.
2371 + *
2372 + */
2373 +
2374 +#ifndef BACKUP_VMA_H
2375 +#define BACKUP_VMA_H
2376 +
2377 +#include "backup.h"
2378 +#include "error.h"
2379 +
2380 +#define VMA_BLOCK_BITS 12
2381 +#define VMA_BLOCK_SIZE (1<<VMA_BLOCK_BITS)
2382 +#define VMA_CLUSTER_BITS (VMA_BLOCK_BITS+4)
2383 +#define VMA_CLUSTER_SIZE (1<<VMA_CLUSTER_BITS)
2384 +
2385 +#if VMA_CLUSTER_SIZE != 65536
2386 +#error unexpected cluster size
2387 +#endif
2388 +
2389 +#define VMA_EXTENT_HEADER_SIZE 512
2390 +#define VMA_BLOCKS_PER_EXTENT 59
2391 +#define VMA_MAX_CONFIGS 256
2392 +
2393 +#define VMA_MAX_EXTENT_SIZE \
2394 +    (VMA_EXTENT_HEADER_SIZE+VMA_CLUSTER_SIZE*VMA_BLOCKS_PER_EXTENT)
2395 +#if VMA_MAX_EXTENT_SIZE != 3867136
2396 +#error unexpected VMA_EXTENT_SIZE
2397 +#endif
2398 +
2399 +/* File Format Definitions */
2400 +
2401 +#define VMA_MAGIC (GUINT32_TO_BE(('V'<<24)|('M'<<16)|('A'<<8)|0x00))
2402 +#define VMA_EXTENT_MAGIC (GUINT32_TO_BE(('V'<<24)|('M'<<16)|('A'<<8)|'E'))
2403 +
2404 +typedef struct VmaDeviceInfoHeader {
2405 +    uint32_t devname_ptr; /* offset into blob_buffer table */
2406 +    uint32_t reserved0;
2407 +    uint64_t size; /* device size in bytes */
2408 +    uint64_t reserved1;
2409 +    uint64_t reserved2;
2410 +} VmaDeviceInfoHeader;
2411 +
2412 +typedef struct VmaHeader {
2413 +    uint32_t magic;
2414 +    uint32_t version;
2415 +    unsigned char uuid[16];
2416 +    int64_t ctime;
2417 +    unsigned char md5sum[16];
2418 +
2419 +    uint32_t blob_buffer_offset;
2420 +    uint32_t blob_buffer_size;
2421 +    uint32_t header_size;
2422 +
2423 +    unsigned char reserved[1984];
2424 +
2425 +    uint32_t config_names[VMA_MAX_CONFIGS]; /* offset into blob_buffer table */
2426 +    uint32_t config_data[VMA_MAX_CONFIGS];  /* offset into blob_buffer table */
2427 +
2428 +    VmaDeviceInfoHeader dev_info[256];
2429 +} VmaHeader;
2430 +
2431 +typedef struct VmaExtentHeader {
2432 +    uint32_t magic;
2433 +    uint16_t reserved1;
2434 +    uint16_t block_count;
2435 +    unsigned char uuid[16];
2436 +    unsigned char md5sum[16];
2437 +    uint64_t blockinfo[VMA_BLOCKS_PER_EXTENT];
2438 +} VmaExtentHeader;
2439 +
2440 +/* functions/definitions to read/write vma files */
2441 +
2442 +typedef struct VmaReader VmaReader;
2443 +
2444 +typedef struct VmaWriter VmaWriter;
2445 +
2446 +typedef struct VmaConfigData {
2447 +    const char *name;
2448 +    const void *data;
2449 +    uint32_t len;
2450 +} VmaConfigData;
2451 +
2452 +typedef struct VmaStreamInfo {
2453 +    uint64_t size;
2454 +    uint64_t cluster_count;
2455 +    uint64_t transferred;
2456 +    uint64_t zero_bytes;
2457 +    int finished;
2458 +    char *devname;
2459 +} VmaStreamInfo;
2460 +
2461 +typedef struct VmaStatus {
2462 +    int status;
2463 +    bool closed;
2464 +    char errmsg[8192];
2465 +    char uuid_str[37];
2466 +    VmaStreamInfo stream_info[256];
2467 +} VmaStatus;
2468 +
2469 +typedef struct VmaDeviceInfo {
2470 +    uint64_t size; /* device size in bytes */
2471 +    const char *devname;
2472 +} VmaDeviceInfo;
2473 +
2474 +extern const BackupDriver backup_vma_driver;
2475 +
2476 +VmaWriter *vma_writer_create(const char *filename, uuid_t uuid, int64_t speed,
2477 +                             Error **errp);
2478 +int vma_writer_close(VmaWriter *vmaw, Error **errp);
2479 +void vma_writer_destroy(VmaWriter *vmaw);
2480 +int vma_writer_add_config(VmaWriter *vmaw, const char *name, gpointer data,
2481 +                          size_t len);
2482 +int vma_writer_register_stream(VmaWriter *vmaw, const char *devname,
2483 +                               size_t size);
2484 +
2485 +int64_t coroutine_fn vma_writer_write(VmaWriter *vmaw, uint8_t dev_id,
2486 +                                      int64_t cluster_num, unsigned char *buf,
2487 +                                      size_t *zero_bytes);
2488 +
2489 +int coroutine_fn vma_writer_close_stream(VmaWriter *vmaw, uint8_t dev_id);
2490 +
2491 +int vma_writer_get_status(VmaWriter *vmaw, VmaStatus *status);
2492 +void vma_writer_set_error(VmaWriter *vmaw, const char *fmt, ...);
2493 +
2494 +
2495 +VmaReader *vma_reader_create(const char *filename, Error **errp);
2496 +void vma_reader_destroy(VmaReader *vmar);
2497 +VmaHeader *vma_reader_get_header(VmaReader *vmar);
2498 +GList *vma_reader_get_config_data(VmaReader *vmar);
2499 +VmaDeviceInfo *vma_reader_get_device_info(VmaReader *vmar, guint8 dev_id);
2500 +int vma_reader_register_bs(VmaReader *vmar, guint8 dev_id,
2501 +                           BlockDriverState *bs, bool write_zeroes,
2502 +                           Error **errp);
2503 +int vma_reader_restore(VmaReader *vmar, int vmstate_fd, Error **errp);
2504 +
2505 +#endif /* BACKUP_VMA_H */
2506 --
2507 1.7.2.5
2508