1 From 0000000000000000000000000000000000000000 Mon Sep 17 00:00:00 2001
2 From: Dietmar Maurer <dietmar@proxmox.com>
3 Date: Mon, 6 Apr 2020 12:16:57 +0200
4 Subject: [PATCH] PVE-Backup: add vma backup format code
6 Notes about partial restoring: skipping a certain drive is done via a
7 map line of the form skip=drive-scsi0. Since in PVE, most archives are
8 compressed and piped to vma for restore, it's not easily possible to
11 Signed-off-by: Thomas Lamprecht <t.lamprecht@proxmox.com>
12 [FE: improvements during create
13 allow partial restore]
14 Signed-off-by: Fiona Ebner <f.ebner@proxmox.com>
16 block/meson.build | 2 +
18 vma-reader.c | 867 ++++++++++++++++++++++++++++++++++++++++++++
19 vma-writer.c | 793 ++++++++++++++++++++++++++++++++++++++++
20 vma.c | 900 ++++++++++++++++++++++++++++++++++++++++++++++
22 6 files changed, 2717 insertions(+)
23 create mode 100644 vma-reader.c
24 create mode 100644 vma-writer.c
25 create mode 100644 vma.c
26 create mode 100644 vma.h
28 diff --git a/block/meson.build b/block/meson.build
29 index 253fe49fa2..744b698a82 100644
30 --- a/block/meson.build
31 +++ b/block/meson.build
32 @@ -47,6 +47,8 @@ block_ss.add(files(
34 ), zstd, zlib, gnutls)
36 +block_ss.add(files('../vma-writer.c'), libuuid)
38 softmmu_ss.add(when: 'CONFIG_TCG', if_true: files('blkreplay.c'))
39 softmmu_ss.add(files('block-ram-registrar.c'))
41 diff --git a/meson.build b/meson.build
42 index 30447cfaef..38a4e2bcef 100644
45 @@ -1527,6 +1527,8 @@ keyutils = dependency('libkeyutils', required: false,
47 has_gettid = cc.has_function('gettid')
49 +libuuid = cc.find_library('uuid', required: true)
52 selinux = dependency('libselinux',
53 required: get_option('selinux'),
54 @@ -3650,6 +3652,9 @@ if have_tools
55 dependencies: [blockdev, qemuutil, gnutls, selinux],
58 + vma = executable('vma', files('vma.c', 'vma-reader.c') + genh,
59 + dependencies: [authz, block, crypto, io, qom], install: true)
61 subdir('storage-daemon')
62 subdir('contrib/rdmacm-mux')
63 subdir('contrib/elf2dmp')
64 diff --git a/vma-reader.c b/vma-reader.c
66 index 0000000000..81a891c6b1
71 + * VMA: Virtual Machine Archive
73 + * Copyright (C) 2012 Proxmox Server Solutions
76 + * Dietmar Maurer (dietmar@proxmox.com)
78 + * This work is licensed under the terms of the GNU GPL, version 2 or later.
79 + * See the COPYING file in the top-level directory.
83 +#include "qemu/osdep.h"
85 +#include <uuid/uuid.h>
87 +#include "qemu/timer.h"
88 +#include "qemu/ratelimit.h"
90 +#include "block/block.h"
91 +#include "sysemu/block-backend.h"
93 +static unsigned char zero_vma_block[VMA_BLOCK_SIZE];
95 +typedef struct VmaRestoreState {
96 + BlockBackend *target;
98 + unsigned long *bitmap;
105 + GChecksum *md5csum;
106 + GHashTable *blob_hash;
107 + unsigned char *head_data;
108 + VmaDeviceInfo devinfo[256];
109 + VmaRestoreState rstate[256];
111 + guint8 vmstate_stream;
112 + uint32_t vmstate_clusters;
113 + /* to show restore percentage if run with -v */
115 + int64_t cluster_count;
116 + int64_t clusters_read;
117 + int64_t zero_cluster_data;
118 + int64_t partial_zero_cluster_data;
119 + int clusters_read_per;
123 +g_int32_hash(gconstpointer v)
125 + return *(const uint32_t *)v;
129 +g_int32_equal(gconstpointer v1, gconstpointer v2)
131 + return *((const uint32_t *)v1) == *((const uint32_t *)v2);
134 +static int vma_reader_get_bitmap(VmaRestoreState *rstate, int64_t cluster_num)
137 + assert(rstate->bitmap);
139 + unsigned long val, idx, bit;
141 + idx = cluster_num / BITS_PER_LONG;
143 + assert(rstate->bitmap_size > idx);
145 + bit = cluster_num % BITS_PER_LONG;
146 + val = rstate->bitmap[idx];
148 + return !!(val & (1UL << bit));
151 +static void vma_reader_set_bitmap(VmaRestoreState *rstate, int64_t cluster_num,
155 + assert(rstate->bitmap);
157 + unsigned long val, idx, bit;
159 + idx = cluster_num / BITS_PER_LONG;
161 + assert(rstate->bitmap_size > idx);
163 + bit = cluster_num % BITS_PER_LONG;
164 + val = rstate->bitmap[idx];
166 + if (!(val & (1UL << bit))) {
170 + if (val & (1UL << bit)) {
171 + val &= ~(1UL << bit);
174 + rstate->bitmap[idx] = val;
177 +typedef struct VmaBlob {
183 +static const VmaBlob *get_header_blob(VmaReader *vmar, uint32_t pos)
186 + assert(vmar->blob_hash);
188 + return g_hash_table_lookup(vmar->blob_hash, &pos);
191 +static const char *get_header_str(VmaReader *vmar, uint32_t pos)
193 + const VmaBlob *blob = get_header_blob(vmar, pos);
197 + const char *res = (char *)blob->data;
198 + if (res[blob->len-1] != '\0') {
205 +safe_read(int fd, unsigned char *buf, size_t count)
210 + n = read(fd, buf, count);
211 + } while (n < 0 && errno == EINTR);
217 +full_read(int fd, unsigned char *buf, size_t len)
225 + n = safe_read(fd, buf, len);
247 +void vma_reader_destroy(VmaReader *vmar)
251 + if (vmar->fd >= 0) {
255 + if (vmar->cdata_list) {
256 + g_list_free(vmar->cdata_list);
260 + for (i = 1; i < 256; i++) {
261 + if (vmar->rstate[i].bitmap) {
262 + g_free(vmar->rstate[i].bitmap);
264 + if (vmar->rstate[i].target) {
265 + blk_unref(vmar->rstate[i].target);
269 + if (vmar->md5csum) {
270 + g_checksum_free(vmar->md5csum);
273 + if (vmar->blob_hash) {
274 + g_hash_table_destroy(vmar->blob_hash);
277 + if (vmar->head_data) {
278 + g_free(vmar->head_data);
285 +static int vma_reader_read_head(VmaReader *vmar, Error **errp)
289 + assert(*errp == NULL);
291 + unsigned char md5sum[16];
295 + vmar->head_data = g_malloc(sizeof(VmaHeader));
297 + if (full_read(vmar->fd, vmar->head_data, sizeof(VmaHeader)) !=
298 + sizeof(VmaHeader)) {
299 + error_setg(errp, "can't read vma header - %s",
300 + errno ? g_strerror(errno) : "got EOF");
304 + VmaHeader *h = (VmaHeader *)vmar->head_data;
306 + if (h->magic != VMA_MAGIC) {
307 + error_setg(errp, "not a vma file - wrong magic number");
311 + uint32_t header_size = GUINT32_FROM_BE(h->header_size);
312 + int need = header_size - sizeof(VmaHeader);
314 + error_setg(errp, "wrong vma header size %d", header_size);
318 + vmar->head_data = g_realloc(vmar->head_data, header_size);
319 + h = (VmaHeader *)vmar->head_data;
321 + if (full_read(vmar->fd, vmar->head_data + sizeof(VmaHeader), need) !=
323 + error_setg(errp, "can't read vma header data - %s",
324 + errno ? g_strerror(errno) : "got EOF");
328 + memcpy(md5sum, h->md5sum, 16);
329 + memset(h->md5sum, 0, 16);
331 + g_checksum_reset(vmar->md5csum);
332 + g_checksum_update(vmar->md5csum, vmar->head_data, header_size);
334 + g_checksum_get_digest(vmar->md5csum, (guint8 *)(h->md5sum), &csize);
336 + if (memcmp(md5sum, h->md5sum, 16) != 0) {
337 + error_setg(errp, "wrong vma header chechsum");
341 + /* we can modify header data after checksum verify */
342 + h->header_size = header_size;
344 + h->version = GUINT32_FROM_BE(h->version);
345 + if (h->version != 1) {
346 + error_setg(errp, "wrong vma version %d", h->version);
350 + h->ctime = GUINT64_FROM_BE(h->ctime);
351 + h->blob_buffer_offset = GUINT32_FROM_BE(h->blob_buffer_offset);
352 + h->blob_buffer_size = GUINT32_FROM_BE(h->blob_buffer_size);
354 + uint32_t bstart = h->blob_buffer_offset + 1;
355 + uint32_t bend = h->blob_buffer_offset + h->blob_buffer_size;
357 + if (bstart <= sizeof(VmaHeader)) {
358 + error_setg(errp, "wrong vma blob buffer offset %d",
359 + h->blob_buffer_offset);
363 + if (bend > header_size) {
364 + error_setg(errp, "wrong vma blob buffer size %d/%d",
365 + h->blob_buffer_offset, h->blob_buffer_size);
369 + while ((bstart + 2) <= bend) {
370 + uint32_t size = vmar->head_data[bstart] +
371 + (vmar->head_data[bstart+1] << 8);
372 + if ((bstart + size + 2) <= bend) {
373 + VmaBlob *blob = g_new0(VmaBlob, 1);
374 + blob->start = bstart - h->blob_buffer_offset;
376 + blob->data = vmar->head_data + bstart + 2;
377 + g_hash_table_insert(vmar->blob_hash, &blob->start, blob);
379 + bstart += size + 2;
384 + for (i = 1; i < 256; i++) {
385 + VmaDeviceInfoHeader *dih = &h->dev_info[i];
386 + uint32_t devname_ptr = GUINT32_FROM_BE(dih->devname_ptr);
387 + uint64_t size = GUINT64_FROM_BE(dih->size);
388 + const char *devname = get_header_str(vmar, devname_ptr);
390 + if (size && devname) {
392 + vmar->devinfo[i].size = size;
393 + vmar->devinfo[i].devname = devname;
395 + if (strcmp(devname, "vmstate") == 0) {
396 + vmar->vmstate_stream = i;
401 + for (i = 0; i < VMA_MAX_CONFIGS; i++) {
402 + uint32_t name_ptr = GUINT32_FROM_BE(h->config_names[i]);
403 + uint32_t data_ptr = GUINT32_FROM_BE(h->config_data[i]);
405 + if (!(name_ptr && data_ptr)) {
408 + const char *name = get_header_str(vmar, name_ptr);
409 + const VmaBlob *blob = get_header_blob(vmar, data_ptr);
411 + if (!(name && blob)) {
412 + error_setg(errp, "vma contains invalid data pointers");
416 + VmaConfigData *cdata = g_new0(VmaConfigData, 1);
417 + cdata->name = name;
418 + cdata->data = blob->data;
419 + cdata->len = blob->len;
421 + vmar->cdata_list = g_list_append(vmar->cdata_list, cdata);
427 +VmaReader *vma_reader_create(const char *filename, Error **errp)
432 + VmaReader *vmar = g_new0(VmaReader, 1);
434 + if (strcmp(filename, "-") == 0) {
437 + vmar->fd = open(filename, O_RDONLY);
440 + if (vmar->fd < 0) {
441 + error_setg(errp, "can't open file %s - %s\n", filename,
442 + g_strerror(errno));
446 + vmar->md5csum = g_checksum_new(G_CHECKSUM_MD5);
447 + if (!vmar->md5csum) {
448 + error_setg(errp, "can't allocate cmsum\n");
452 + vmar->blob_hash = g_hash_table_new_full(g_int32_hash, g_int32_equal,
455 + if (vma_reader_read_head(vmar, errp) < 0) {
463 + vma_reader_destroy(vmar);
469 +VmaHeader *vma_reader_get_header(VmaReader *vmar)
472 + assert(vmar->head_data);
474 + return (VmaHeader *)(vmar->head_data);
477 +GList *vma_reader_get_config_data(VmaReader *vmar)
480 + assert(vmar->head_data);
482 + return vmar->cdata_list;
485 +VmaDeviceInfo *vma_reader_get_device_info(VmaReader *vmar, guint8 dev_id)
490 + if (vmar->devinfo[dev_id].size && vmar->devinfo[dev_id].devname) {
491 + return &vmar->devinfo[dev_id];
497 +static void allocate_rstate(VmaReader *vmar, guint8 dev_id,
498 + BlockBackend *target, bool write_zeroes, bool skip)
503 + vmar->rstate[dev_id].target = target;
504 + vmar->rstate[dev_id].write_zeroes = write_zeroes;
505 + vmar->rstate[dev_id].skip = skip;
507 + int64_t size = vmar->devinfo[dev_id].size;
509 + int64_t bitmap_size = (size/BDRV_SECTOR_SIZE) +
510 + (VMA_CLUSTER_SIZE/BDRV_SECTOR_SIZE) * BITS_PER_LONG - 1;
511 + bitmap_size /= (VMA_CLUSTER_SIZE/BDRV_SECTOR_SIZE) * BITS_PER_LONG;
513 + vmar->rstate[dev_id].bitmap_size = bitmap_size;
514 + vmar->rstate[dev_id].bitmap = g_new0(unsigned long, bitmap_size);
516 + vmar->cluster_count += size/VMA_CLUSTER_SIZE;
519 +int vma_reader_register_bs(VmaReader *vmar, guint8 dev_id, BlockBackend *target,
520 + bool write_zeroes, bool skip, Error **errp)
523 + assert(target != NULL || skip);
525 + assert(vmar->rstate[dev_id].target == NULL && !vmar->rstate[dev_id].skip);
527 + if (target != NULL) {
528 + int64_t size = blk_getlength(target);
529 + int64_t size_diff = size - vmar->devinfo[dev_id].size;
531 + /* storage types can have different size restrictions, so it
532 + * is not always possible to create an image with exact size.
533 + * So we tolerate a size difference up to 4MB.
535 + if ((size_diff < 0) || (size_diff > 4*1024*1024)) {
536 + error_setg(errp, "vma_reader_register_bs for stream %s failed - "
537 + "unexpected size %zd != %zd", vmar->devinfo[dev_id].devname,
538 + size, vmar->devinfo[dev_id].size);
543 + allocate_rstate(vmar, dev_id, target, write_zeroes, skip);
548 +static ssize_t safe_write(int fd, void *buf, size_t count)
553 + n = write(fd, buf, count);
554 + } while (n < 0 && errno == EINTR);
559 +static size_t full_write(int fd, void *buf, size_t len)
567 + n = safe_write(fd, buf, len);
577 + /* incomplete write ? */
584 +static int restore_write_data(VmaReader *vmar, guint8 dev_id,
585 + BlockBackend *target, int vmstate_fd,
586 + unsigned char *buf, int64_t sector_num,
587 + int nb_sectors, Error **errp)
591 + if (dev_id == vmar->vmstate_stream) {
592 + if (vmstate_fd >= 0) {
593 + int len = nb_sectors * BDRV_SECTOR_SIZE;
594 + int res = full_write(vmstate_fd, buf, len);
596 + error_setg(errp, "write vmstate failed %d", res);
601 + int res = blk_pwrite(target, sector_num * BDRV_SECTOR_SIZE, nb_sectors * BDRV_SECTOR_SIZE, buf, 0);
603 + error_setg(errp, "blk_pwrite to %s failed (%d)",
604 + bdrv_get_device_name(blk_bs(target)), res);
611 +static int restore_extent(VmaReader *vmar, unsigned char *buf,
612 + int extent_size, int vmstate_fd,
613 + bool verbose, bool verify, Error **errp)
618 + VmaExtentHeader *ehead = (VmaExtentHeader *)buf;
619 + int start = VMA_EXTENT_HEADER_SIZE;
622 + for (i = 0; i < VMA_BLOCKS_PER_EXTENT; i++) {
623 + uint64_t block_info = GUINT64_FROM_BE(ehead->blockinfo[i]);
624 + uint64_t cluster_num = block_info & 0xffffffff;
625 + uint8_t dev_id = (block_info >> 32) & 0xff;
626 + uint16_t mask = block_info >> (32+16);
627 + int64_t max_sector;
633 + VmaRestoreState *rstate = &vmar->rstate[dev_id];
634 + BlockBackend *target = NULL;
636 + bool skip = rstate->skip;
638 + if (dev_id != vmar->vmstate_stream) {
639 + target = rstate->target;
640 + if (!verify && !target && !skip) {
641 + error_setg(errp, "got wrong dev id %d", dev_id);
646 + if (vma_reader_get_bitmap(rstate, cluster_num)) {
647 + error_setg(errp, "found duplicated cluster %zd for stream %s",
648 + cluster_num, vmar->devinfo[dev_id].devname);
651 + vma_reader_set_bitmap(rstate, cluster_num, 1);
654 + max_sector = vmar->devinfo[dev_id].size/BDRV_SECTOR_SIZE;
656 + max_sector = G_MAXINT64;
657 + if (cluster_num != vmar->vmstate_clusters) {
658 + error_setg(errp, "found out of order vmstate data");
661 + vmar->vmstate_clusters++;
664 + vmar->clusters_read++;
667 + time_t duration = time(NULL) - vmar->start_time;
668 + int percent = (vmar->clusters_read*100)/vmar->cluster_count;
669 + if (percent != vmar->clusters_read_per) {
670 + printf("progress %d%% (read %zd bytes, duration %zd sec)\n",
671 + percent, vmar->clusters_read*VMA_CLUSTER_SIZE,
674 + vmar->clusters_read_per = percent;
678 + /* try to write whole clusters to speedup restore */
679 + if (mask == 0xffff) {
680 + if ((start + VMA_CLUSTER_SIZE) > extent_size) {
681 + error_setg(errp, "short vma extent - too many blocks");
684 + int64_t sector_num = (cluster_num * VMA_CLUSTER_SIZE) /
686 + int64_t end_sector = sector_num +
687 + VMA_CLUSTER_SIZE/BDRV_SECTOR_SIZE;
689 + if (end_sector > max_sector) {
690 + end_sector = max_sector;
693 + if (end_sector <= sector_num) {
694 + error_setg(errp, "got wrong block address - write beyond end");
698 + if (!verify && !skip) {
699 + int nb_sectors = end_sector - sector_num;
700 + if (restore_write_data(vmar, dev_id, target, vmstate_fd,
701 + buf + start, sector_num, nb_sectors,
707 + start += VMA_CLUSTER_SIZE;
712 + for (j = 0; j < 16; j++) {
713 + int64_t sector_num = (cluster_num*VMA_CLUSTER_SIZE +
714 + j*VMA_BLOCK_SIZE)/BDRV_SECTOR_SIZE;
716 + int64_t end_sector = sector_num +
717 + VMA_BLOCK_SIZE/BDRV_SECTOR_SIZE;
718 + if (end_sector > max_sector) {
719 + end_sector = max_sector;
723 + if ((start + VMA_BLOCK_SIZE) > extent_size) {
724 + error_setg(errp, "short vma extent - too many blocks");
728 + if (end_sector <= sector_num) {
729 + error_setg(errp, "got wrong block address - "
730 + "write beyond end");
734 + if (!verify && !skip) {
735 + int nb_sectors = end_sector - sector_num;
736 + if (restore_write_data(vmar, dev_id, target, vmstate_fd,
737 + buf + start, sector_num,
738 + nb_sectors, errp) < 0) {
743 + start += VMA_BLOCK_SIZE;
748 + if (end_sector > sector_num) {
749 + /* Todo: use bdrv_co_write_zeroes (but that need to
750 + * be run inside coroutine?)
752 + int nb_sectors = end_sector - sector_num;
753 + int zero_size = BDRV_SECTOR_SIZE*nb_sectors;
754 + vmar->zero_cluster_data += zero_size;
756 + vmar->partial_zero_cluster_data += zero_size;
759 + if (rstate->write_zeroes && !verify && !skip) {
760 + if (restore_write_data(vmar, dev_id, target, vmstate_fd,
761 + zero_vma_block, sector_num,
762 + nb_sectors, errp) < 0) {
774 + if (start != extent_size) {
775 + error_setg(errp, "vma extent error - missing blocks");
782 +static int vma_reader_restore_full(VmaReader *vmar, int vmstate_fd,
783 + bool verbose, bool verify,
787 + assert(vmar->head_data);
790 + unsigned char buf[VMA_MAX_EXTENT_SIZE];
792 + unsigned char md5sum[16];
793 + VmaHeader *h = (VmaHeader *)vmar->head_data;
795 + vmar->start_time = time(NULL);
798 + int bytes = full_read(vmar->fd, buf + buf_pos, sizeof(buf) - buf_pos);
800 + error_setg(errp, "read failed - %s", g_strerror(errno));
810 + if (buf_pos < VMA_EXTENT_HEADER_SIZE) {
811 + error_setg(errp, "read short extent (%d bytes)", buf_pos);
815 + VmaExtentHeader *ehead = (VmaExtentHeader *)buf;
817 + /* extract md5sum */
818 + memcpy(md5sum, ehead->md5sum, sizeof(ehead->md5sum));
819 + memset(ehead->md5sum, 0, sizeof(ehead->md5sum));
821 + g_checksum_reset(vmar->md5csum);
822 + g_checksum_update(vmar->md5csum, buf, VMA_EXTENT_HEADER_SIZE);
824 + g_checksum_get_digest(vmar->md5csum, ehead->md5sum, &csize);
826 + if (memcmp(md5sum, ehead->md5sum, 16) != 0) {
827 + error_setg(errp, "wrong vma extent header chechsum");
831 + if (memcmp(h->uuid, ehead->uuid, sizeof(ehead->uuid)) != 0) {
832 + error_setg(errp, "wrong vma extent uuid");
836 + if (ehead->magic != VMA_EXTENT_MAGIC || ehead->reserved1 != 0) {
837 + error_setg(errp, "wrong vma extent header magic");
841 + int block_count = GUINT16_FROM_BE(ehead->block_count);
842 + int extent_size = VMA_EXTENT_HEADER_SIZE + block_count*VMA_BLOCK_SIZE;
844 + if (buf_pos < extent_size) {
845 + error_setg(errp, "short vma extent (%d < %d)", buf_pos,
850 + if (restore_extent(vmar, buf, extent_size, vmstate_fd, verbose,
851 + verify, errp) < 0) {
855 + if (buf_pos > extent_size) {
856 + memmove(buf, buf + extent_size, buf_pos - extent_size);
857 + buf_pos = buf_pos - extent_size;
866 + for (i = 1; i < 256; i++) {
867 + VmaRestoreState *rstate = &vmar->rstate[i];
868 + if (!rstate->target) {
872 + if (blk_flush(rstate->target) < 0) {
873 + error_setg(errp, "vma blk_flush %s failed",
874 + vmar->devinfo[i].devname);
878 + if (vmar->devinfo[i].size &&
879 + (strcmp(vmar->devinfo[i].devname, "vmstate") != 0)) {
880 + assert(rstate->bitmap);
882 + int64_t cluster_num, end;
884 + end = (vmar->devinfo[i].size + VMA_CLUSTER_SIZE - 1) /
887 + for (cluster_num = 0; cluster_num < end; cluster_num++) {
888 + if (!vma_reader_get_bitmap(rstate, cluster_num)) {
889 + error_setg(errp, "detected missing cluster %zd "
890 + "for stream %s", cluster_num,
891 + vmar->devinfo[i].devname);
899 + if (vmar->clusters_read) {
900 + printf("total bytes read %zd, sparse bytes %zd (%.3g%%)\n",
901 + vmar->clusters_read*VMA_CLUSTER_SIZE,
902 + vmar->zero_cluster_data,
903 + (double)(100.0*vmar->zero_cluster_data)/
904 + (vmar->clusters_read*VMA_CLUSTER_SIZE));
906 + int64_t datasize = vmar->clusters_read*VMA_CLUSTER_SIZE-vmar->zero_cluster_data;
907 + if (datasize) { // this does not make sense for empty files
908 + printf("space reduction due to 4K zero blocks %.3g%%\n",
909 + (double)(100.0*vmar->partial_zero_cluster_data) / datasize);
912 + printf("vma archive contains no image data\n");
918 +int vma_reader_restore(VmaReader *vmar, int vmstate_fd, bool verbose,
921 + return vma_reader_restore_full(vmar, vmstate_fd, verbose, false, errp);
924 +int vma_reader_verify(VmaReader *vmar, bool verbose, Error **errp)
928 + for (dev_id = 1; dev_id < 255; dev_id++) {
929 + if (vma_reader_get_device_info(vmar, dev_id)) {
930 + allocate_rstate(vmar, dev_id, NULL, false, false);
934 + return vma_reader_restore_full(vmar, -1, verbose, true, errp);
937 diff --git a/vma-writer.c b/vma-writer.c
939 index 0000000000..ac7da237d0
944 + * VMA: Virtual Machine Archive
946 + * Copyright (C) 2012 Proxmox Server Solutions
949 + * Dietmar Maurer (dietmar@proxmox.com)
951 + * This work is licensed under the terms of the GNU GPL, version 2 or later.
952 + * See the COPYING file in the top-level directory.
956 +#include "qemu/osdep.h"
958 +#include <uuid/uuid.h>
961 +#include "block/block.h"
962 +#include "monitor/monitor.h"
963 +#include "qemu/main-loop.h"
964 +#include "qemu/coroutine.h"
965 +#include "qemu/cutils.h"
966 +#include "qemu/memalign.h"
970 +#define DPRINTF(fmt, ...)\
971 + do { if (DEBUG_VMA) { printf("vma: " fmt, ## __VA_ARGS__); } } while (0)
973 +#define WRITE_BUFFERS 5
974 +#define HEADER_CLUSTERS 8
975 +#define HEADERBUF_SIZE (VMA_CLUSTER_SIZE*HEADER_CLUSTERS)
983 + bool header_written;
986 + /* we always write extents */
987 + unsigned char *outbuf;
988 + int outbuf_pos; /* in bytes */
989 + int outbuf_count; /* in VMA_BLOCKS */
990 + uint64_t outbuf_block_info[VMA_BLOCKS_PER_EXTENT];
992 + unsigned char *headerbuf;
994 + GChecksum *md5csum;
995 + CoMutex flush_lock;
996 + Coroutine *co_writer;
998 + /* drive informations */
999 + VmaStreamInfo stream_info[256];
1000 + guint stream_count;
1002 + guint8 vmstate_stream;
1003 + uint32_t vmstate_clusters;
1005 + /* header blob table */
1006 + char *header_blob_table;
1007 + uint32_t header_blob_table_size;
1008 + uint32_t header_blob_table_pos;
1010 + /* store for config blobs */
1011 + uint32_t config_names[VMA_MAX_CONFIGS]; /* offset into blob_buffer table */
1012 + uint32_t config_data[VMA_MAX_CONFIGS]; /* offset into blob_buffer table */
1013 + uint32_t config_count;
1016 +void vma_writer_set_error(VmaWriter *vmaw, const char *fmt, ...)
1020 + if (vmaw->status < 0) {
1024 + vmaw->status = -1;
1026 + va_start(ap, fmt);
1027 + g_vsnprintf(vmaw->errmsg, sizeof(vmaw->errmsg), fmt, ap);
1030 + DPRINTF("vma_writer_set_error: %s\n", vmaw->errmsg);
1033 +static uint32_t allocate_header_blob(VmaWriter *vmaw, const char *data,
1036 + if (len > 65535) {
1040 + if (!vmaw->header_blob_table ||
1041 + (vmaw->header_blob_table_size <
1042 + (vmaw->header_blob_table_pos + len + 2))) {
1043 + int newsize = vmaw->header_blob_table_size + ((len + 2 + 511)/512)*512;
1045 + vmaw->header_blob_table = g_realloc(vmaw->header_blob_table, newsize);
1046 + memset(vmaw->header_blob_table + vmaw->header_blob_table_size,
1047 + 0, newsize - vmaw->header_blob_table_size);
1048 + vmaw->header_blob_table_size = newsize;
1051 + uint32_t cpos = vmaw->header_blob_table_pos;
1052 + vmaw->header_blob_table[cpos] = len & 255;
1053 + vmaw->header_blob_table[cpos+1] = (len >> 8) & 255;
1054 + memcpy(vmaw->header_blob_table + cpos + 2, data, len);
1055 + vmaw->header_blob_table_pos += len + 2;
1059 +static uint32_t allocate_header_string(VmaWriter *vmaw, const char *str)
1063 + size_t len = strlen(str) + 1;
1065 + return allocate_header_blob(vmaw, str, len);
1068 +int vma_writer_add_config(VmaWriter *vmaw, const char *name, gpointer data,
1072 + assert(!vmaw->header_written);
1073 + assert(vmaw->config_count < VMA_MAX_CONFIGS);
1077 + gchar *basename = g_path_get_basename(name);
1078 + uint32_t name_ptr = allocate_header_string(vmaw, basename);
1085 + uint32_t data_ptr = allocate_header_blob(vmaw, data, len);
1090 + vmaw->config_names[vmaw->config_count] = name_ptr;
1091 + vmaw->config_data[vmaw->config_count] = data_ptr;
1093 + vmaw->config_count++;
1098 +int vma_writer_register_stream(VmaWriter *vmaw, const char *devname,
1103 + assert(!vmaw->status);
1105 + if (vmaw->header_written) {
1106 + vma_writer_set_error(vmaw, "vma_writer_register_stream: header "
1107 + "already written");
1111 + guint n = vmaw->stream_count + 1;
1113 + /* we can have dev_ids form 1 to 255 (0 reserved)
1114 + * 255(-1) reseverd for safety
1117 + vma_writer_set_error(vmaw, "vma_writer_register_stream: "
1118 + "too many drives");
1123 + vma_writer_set_error(vmaw, "vma_writer_register_stream: "
1124 + "got strange size %zd", size);
1128 + DPRINTF("vma_writer_register_stream %s %zu %d\n", devname, size, n);
1130 + vmaw->stream_info[n].devname = g_strdup(devname);
1131 + vmaw->stream_info[n].size = size;
1133 + vmaw->stream_info[n].cluster_count = (size + VMA_CLUSTER_SIZE - 1) /
1136 + vmaw->stream_count = n;
1138 + if (strcmp(devname, "vmstate") == 0) {
1139 + vmaw->vmstate_stream = n;
1145 +static void coroutine_fn yield_until_fd_writable(int fd)
1147 + assert(qemu_in_coroutine());
1148 + AioContext *ctx = qemu_get_current_aio_context();
1149 + aio_set_fd_handler(ctx, fd, false, NULL, (IOHandler *)qemu_coroutine_enter,
1150 + NULL, NULL, qemu_coroutine_self());
1151 + qemu_coroutine_yield();
1152 + aio_set_fd_handler(ctx, fd, false, NULL, NULL, NULL, NULL, NULL);
1155 +static ssize_t coroutine_fn
1156 +vma_queue_write(VmaWriter *vmaw, const void *buf, size_t bytes)
1158 + DPRINTF("vma_queue_write enter %zd\n", bytes);
1162 + assert(bytes <= VMA_MAX_EXTENT_SIZE);
1167 + assert(vmaw->co_writer == NULL);
1169 + vmaw->co_writer = qemu_coroutine_self();
1171 + while (done < bytes) {
1172 + if (vmaw->status < 0) {
1173 + DPRINTF("vma_queue_write detected canceled backup\n");
1177 + yield_until_fd_writable(vmaw->fd);
1178 + ret = write(vmaw->fd, buf + done, bytes - done);
1181 + DPRINTF("vma_queue_write written %zd %zd\n", done, ret);
1182 + } else if (ret < 0) {
1183 + if (errno == EAGAIN || errno == EWOULDBLOCK) {
1186 + vma_writer_set_error(vmaw, "vma_queue_write: write error - %s",
1187 + g_strerror(errno));
1188 + done = -1; /* always return failure for partial writes */
1191 + } else if (ret == 0) {
1192 + /* should not happen - simply try again */
1196 + vmaw->co_writer = NULL;
1198 + return (done == bytes) ? bytes : -1;
1201 +VmaWriter *vma_writer_create(const char *filename, uuid_t uuid, Error **errp)
1205 + assert(sizeof(VmaHeader) == (4096 + 8192));
1206 + assert(G_STRUCT_OFFSET(VmaHeader, config_names) == 2044);
1207 + assert(G_STRUCT_OFFSET(VmaHeader, config_data) == 3068);
1208 + assert(G_STRUCT_OFFSET(VmaHeader, dev_info) == 4096);
1209 + assert(sizeof(VmaExtentHeader) == 512);
1211 + VmaWriter *vmaw = g_new0(VmaWriter, 1);
1214 + vmaw->md5csum = g_checksum_new(G_CHECKSUM_MD5);
1215 + if (!vmaw->md5csum) {
1216 + error_setg(errp, "can't allocate cmsum\n");
1220 + if (strstart(filename, "exec:", &p)) {
1221 + vmaw->cmd = popen(p, "w");
1222 + if (vmaw->cmd == NULL) {
1223 + error_setg(errp, "can't popen command '%s' - %s\n", p,
1224 + g_strerror(errno));
1227 + vmaw->fd = fileno(vmaw->cmd);
1229 + /* try to use O_NONBLOCK */
1230 + fcntl(vmaw->fd, F_SETFL, fcntl(vmaw->fd, F_GETFL)|O_NONBLOCK);
1235 + const char *tmp_id_str;
1237 + if ((stat(filename, &st) == 0) && S_ISFIFO(st.st_mode)) {
1238 + oflags = O_NONBLOCK|O_WRONLY;
1239 + vmaw->fd = qemu_open(filename, oflags, errp);
1240 + } else if (strstart(filename, "/dev/fdset/", &tmp_id_str)) {
1241 + oflags = O_NONBLOCK|O_WRONLY;
1242 + vmaw->fd = qemu_open(filename, oflags, errp);
1243 + } else if (strstart(filename, "/dev/fdname/", &tmp_id_str)) {
1244 + vmaw->fd = monitor_get_fd(monitor_cur(), tmp_id_str, errp);
1245 + if (vmaw->fd < 0) {
1248 + /* try to use O_NONBLOCK */
1249 + fcntl(vmaw->fd, F_SETFL, fcntl(vmaw->fd, F_GETFL)|O_NONBLOCK);
1251 + oflags = O_NONBLOCK|O_DIRECT|O_WRONLY|O_EXCL;
1252 + vmaw->fd = qemu_create(filename, oflags, 0644, errp);
1255 + if (vmaw->fd < 0) {
1256 + error_free(*errp);
1258 + error_setg(errp, "can't open file %s - %s\n", filename,
1259 + g_strerror(errno));
1264 + /* we use O_DIRECT, so we need to align IO buffers */
1266 + vmaw->outbuf = qemu_memalign(512, VMA_MAX_EXTENT_SIZE);
1267 + vmaw->headerbuf = qemu_memalign(512, HEADERBUF_SIZE);
1269 + vmaw->outbuf_count = 0;
1270 + vmaw->outbuf_pos = VMA_EXTENT_HEADER_SIZE;
1272 + vmaw->header_blob_table_pos = 1; /* start at pos 1 */
1274 + qemu_co_mutex_init(&vmaw->flush_lock);
1276 + uuid_copy(vmaw->uuid, uuid);
1283 + pclose(vmaw->cmd);
1284 + } else if (vmaw->fd >= 0) {
1288 + if (vmaw->md5csum) {
1289 + g_checksum_free(vmaw->md5csum);
1298 +static int coroutine_fn vma_write_header(VmaWriter *vmaw)
1301 + unsigned char *buf = vmaw->headerbuf;
1302 + VmaHeader *head = (VmaHeader *)buf;
1306 + DPRINTF("VMA WRITE HEADER\n");
1308 + if (vmaw->status < 0) {
1309 + return vmaw->status;
1312 + memset(buf, 0, HEADERBUF_SIZE);
1314 + head->magic = VMA_MAGIC;
1315 + head->version = GUINT32_TO_BE(1); /* v1 */
1316 + memcpy(head->uuid, vmaw->uuid, 16);
1318 + time_t ctime = time(NULL);
1319 + head->ctime = GUINT64_TO_BE(ctime);
1321 + for (i = 0; i < VMA_MAX_CONFIGS; i++) {
1322 + head->config_names[i] = GUINT32_TO_BE(vmaw->config_names[i]);
1323 + head->config_data[i] = GUINT32_TO_BE(vmaw->config_data[i]);
1326 + /* 32 bytes per device (12 used currently) = 8192 bytes max */
1327 + for (i = 1; i <= 254; i++) {
1328 + VmaStreamInfo *si = &vmaw->stream_info[i];
1330 + assert(si->devname);
1331 + uint32_t devname_ptr = allocate_header_string(vmaw, si->devname);
1332 + if (!devname_ptr) {
1335 + head->dev_info[i].devname_ptr = GUINT32_TO_BE(devname_ptr);
1336 + head->dev_info[i].size = GUINT64_TO_BE(si->size);
1340 + uint32_t header_size = sizeof(VmaHeader) + vmaw->header_blob_table_size;
1341 + head->header_size = GUINT32_TO_BE(header_size);
1343 + if (header_size > HEADERBUF_SIZE) {
1344 + return -1; /* just to be sure */
1347 + uint32_t blob_buffer_offset = sizeof(VmaHeader);
1348 + memcpy(buf + blob_buffer_offset, vmaw->header_blob_table,
1349 + vmaw->header_blob_table_size);
1350 + head->blob_buffer_offset = GUINT32_TO_BE(blob_buffer_offset);
1351 + head->blob_buffer_size = GUINT32_TO_BE(vmaw->header_blob_table_pos);
1353 + g_checksum_reset(vmaw->md5csum);
1354 + g_checksum_update(vmaw->md5csum, (const guchar *)buf, header_size);
1356 + g_checksum_get_digest(vmaw->md5csum, (guint8 *)(head->md5sum), &csize);
1358 + return vma_queue_write(vmaw, buf, header_size);
1361 +static int coroutine_fn vma_writer_flush(VmaWriter *vmaw)
1368 + if (vmaw->status < 0) {
1369 + return vmaw->status;
1372 + if (!vmaw->header_written) {
1373 + vmaw->header_written = true;
1374 + ret = vma_write_header(vmaw);
1376 + vma_writer_set_error(vmaw, "vma_writer_flush: write header failed");
1381 + DPRINTF("VMA WRITE FLUSH %d %d\n", vmaw->outbuf_count, vmaw->outbuf_pos);
1384 + VmaExtentHeader *ehead = (VmaExtentHeader *)vmaw->outbuf;
1386 + ehead->magic = VMA_EXTENT_MAGIC;
1387 + ehead->reserved1 = 0;
1389 + for (i = 0; i < VMA_BLOCKS_PER_EXTENT; i++) {
1390 + ehead->blockinfo[i] = GUINT64_TO_BE(vmaw->outbuf_block_info[i]);
1393 + guint16 block_count = (vmaw->outbuf_pos - VMA_EXTENT_HEADER_SIZE) /
1396 + ehead->block_count = GUINT16_TO_BE(block_count);
1398 + memcpy(ehead->uuid, vmaw->uuid, sizeof(ehead->uuid));
1399 + memset(ehead->md5sum, 0, sizeof(ehead->md5sum));
1401 + g_checksum_reset(vmaw->md5csum);
1402 + g_checksum_update(vmaw->md5csum, vmaw->outbuf, VMA_EXTENT_HEADER_SIZE);
1404 + g_checksum_get_digest(vmaw->md5csum, ehead->md5sum, &csize);
1406 + int bytes = vmaw->outbuf_pos;
1407 + ret = vma_queue_write(vmaw, vmaw->outbuf, bytes);
1408 + if (ret != bytes) {
1409 + vma_writer_set_error(vmaw, "vma_writer_flush: failed write");
1412 + vmaw->outbuf_count = 0;
1413 + vmaw->outbuf_pos = VMA_EXTENT_HEADER_SIZE;
1415 + for (i = 0; i < VMA_BLOCKS_PER_EXTENT; i++) {
1416 + vmaw->outbuf_block_info[i] = 0;
1419 + return vmaw->status;
1422 +static int vma_count_open_streams(VmaWriter *vmaw)
1424 + g_assert(vmaw != NULL);
1427 + int open_drives = 0;
1428 + for (i = 0; i <= 255; i++) {
1429 + if (vmaw->stream_info[i].size && !vmaw->stream_info[i].finished) {
1434 + return open_drives;
1439 + * You need to call this if the vma archive does not contain
1440 + * any data stream.
1443 +vma_writer_flush_output(VmaWriter *vmaw)
1445 + qemu_co_mutex_lock(&vmaw->flush_lock);
1446 + int ret = vma_writer_flush(vmaw);
1447 + qemu_co_mutex_unlock(&vmaw->flush_lock);
1449 + vma_writer_set_error(vmaw, "vma_writer_flush_header failed");
1455 + * all jobs should call this when there is no more data
1456 + * Returns: number of remaining stream (0 ==> finished)
1459 +vma_writer_close_stream(VmaWriter *vmaw, uint8_t dev_id)
1461 + g_assert(vmaw != NULL);
1463 + DPRINTF("vma_writer_set_status %d\n", dev_id);
1464 + if (!vmaw->stream_info[dev_id].size) {
1465 + vma_writer_set_error(vmaw, "vma_writer_close_stream: "
1466 + "no such stream %d", dev_id);
1469 + if (vmaw->stream_info[dev_id].finished) {
1470 + vma_writer_set_error(vmaw, "vma_writer_close_stream: "
1471 + "stream already closed %d", dev_id);
1475 + vmaw->stream_info[dev_id].finished = true;
1477 + int open_drives = vma_count_open_streams(vmaw);
1479 + if (open_drives <= 0) {
1480 + DPRINTF("vma_writer_set_status all drives completed\n");
1481 + vma_writer_flush_output(vmaw);
1484 + return open_drives;
1487 +int vma_writer_get_status(VmaWriter *vmaw, VmaStatus *status)
1491 + g_assert(vmaw != NULL);
1494 + status->status = vmaw->status;
1495 + g_strlcpy(status->errmsg, vmaw->errmsg, sizeof(status->errmsg));
1496 + for (i = 0; i <= 255; i++) {
1497 + status->stream_info[i] = vmaw->stream_info[i];
1500 + uuid_unparse_lower(vmaw->uuid, status->uuid_str);
1503 + status->closed = vmaw->closed;
1505 + return vmaw->status;
1508 +static int vma_writer_get_buffer(VmaWriter *vmaw)
1512 + qemu_co_mutex_lock(&vmaw->flush_lock);
1514 + /* wait until buffer is available */
1515 + while (vmaw->outbuf_count >= (VMA_BLOCKS_PER_EXTENT - 1)) {
1516 + ret = vma_writer_flush(vmaw);
1518 + vma_writer_set_error(vmaw, "vma_writer_get_buffer: flush failed");
1523 + qemu_co_mutex_unlock(&vmaw->flush_lock);
1529 +int64_t coroutine_fn
1530 +vma_writer_write(VmaWriter *vmaw, uint8_t dev_id, int64_t cluster_num,
1531 + const unsigned char *buf, size_t *zero_bytes)
1533 + g_assert(vmaw != NULL);
1534 + g_assert(zero_bytes != NULL);
1538 + if (vmaw->status < 0) {
1539 + return vmaw->status;
1542 + if (!dev_id || !vmaw->stream_info[dev_id].size) {
1543 + vma_writer_set_error(vmaw, "vma_writer_write: "
1544 + "no such stream %d", dev_id);
1548 + if (vmaw->stream_info[dev_id].finished) {
1549 + vma_writer_set_error(vmaw, "vma_writer_write: "
1550 + "stream already closed %d", dev_id);
1555 + if (cluster_num >= (((uint64_t)1)<<32)) {
1556 + vma_writer_set_error(vmaw, "vma_writer_write: "
1557 + "cluster number out of range");
1561 + if (dev_id == vmaw->vmstate_stream) {
1562 + if (cluster_num != vmaw->vmstate_clusters) {
1563 + vma_writer_set_error(vmaw, "vma_writer_write: "
1564 + "non sequential vmstate write");
1566 + vmaw->vmstate_clusters++;
1567 + } else if (cluster_num >= vmaw->stream_info[dev_id].cluster_count) {
1568 + vma_writer_set_error(vmaw, "vma_writer_write: cluster number too big");
1572 + /* wait until buffer is available */
1573 + if (vma_writer_get_buffer(vmaw) < 0) {
1574 + vma_writer_set_error(vmaw, "vma_writer_write: "
1575 + "vma_writer_get_buffer failed");
1579 + DPRINTF("VMA WRITE %d %zd\n", dev_id, cluster_num);
1581 + uint64_t dev_size = vmaw->stream_info[dev_id].size;
1582 + uint16_t mask = 0;
1587 + uint64_t byte_offset = cluster_num * VMA_CLUSTER_SIZE;
1588 + for (i = 0; i < 16; i++) {
1589 + const unsigned char *vmablock = buf + (i*VMA_BLOCK_SIZE);
1591 + // Note: If the source is not 64k-aligned, we might reach 4k blocks
1592 + // after the end of the device. Always mark these as zero in the
1593 + // mask, so the restore handles them correctly.
1594 + if (byte_offset < dev_size &&
1595 + !buffer_is_zero(vmablock, VMA_BLOCK_SIZE))
1598 + memcpy(vmaw->outbuf + vmaw->outbuf_pos, vmablock,
1601 + // prevent memory leakage on unaligned last block
1602 + if (byte_offset + VMA_BLOCK_SIZE > dev_size) {
1603 + uint64_t real_data_in_block = dev_size - byte_offset;
1604 + memset(vmaw->outbuf + vmaw->outbuf_pos + real_data_in_block,
1605 + 0, VMA_BLOCK_SIZE - real_data_in_block);
1608 + vmaw->outbuf_pos += VMA_BLOCK_SIZE;
1610 + DPRINTF("VMA WRITE %zd ZERO BLOCK %d\n", cluster_num, i);
1611 + vmaw->stream_info[dev_id].zero_bytes += VMA_BLOCK_SIZE;
1612 + *zero_bytes += VMA_BLOCK_SIZE;
1615 + byte_offset += VMA_BLOCK_SIZE;
1619 + DPRINTF("VMA WRITE %zd ZERO CLUSTER\n", cluster_num);
1620 + vmaw->stream_info[dev_id].zero_bytes += VMA_CLUSTER_SIZE;
1621 + *zero_bytes += VMA_CLUSTER_SIZE;
1624 + uint64_t block_info = ((uint64_t)mask) << (32+16);
1625 + block_info |= ((uint64_t)dev_id) << 32;
1626 + block_info |= (cluster_num & 0xffffffff);
1627 + vmaw->outbuf_block_info[vmaw->outbuf_count] = block_info;
1629 + DPRINTF("VMA WRITE MASK %zd %zx\n", cluster_num, block_info);
1631 + vmaw->outbuf_count++;
1633 + /** NOTE: We allways write whole clusters, but we correctly set
1634 + * transferred bytes. So transferred == size when when everything
1637 + size_t transferred = VMA_CLUSTER_SIZE;
1639 + if (dev_id != vmaw->vmstate_stream) {
1640 + uint64_t last = (cluster_num + 1) * VMA_CLUSTER_SIZE;
1641 + if (last > dev_size) {
1642 + uint64_t diff = last - dev_size;
1643 + if (diff >= VMA_CLUSTER_SIZE) {
1644 + vma_writer_set_error(vmaw, "vma_writer_write: "
1645 + "read after last cluster");
1648 + transferred -= diff;
1652 + vmaw->stream_info[dev_id].transferred += transferred;
1654 + return transferred;
1657 +void vma_writer_error_propagate(VmaWriter *vmaw, Error **errp)
1659 + if (vmaw->status < 0 && *errp == NULL) {
1660 + error_setg(errp, "%s", vmaw->errmsg);
1664 +int vma_writer_close(VmaWriter *vmaw, Error **errp)
1666 + g_assert(vmaw != NULL);
1670 + qemu_co_mutex_lock(&vmaw->flush_lock); // wait for pending writes
1672 + assert(vmaw->co_writer == NULL);
1675 + if (pclose(vmaw->cmd) < 0) {
1676 + vma_writer_set_error(vmaw, "vma_writer_close: "
1677 + "pclose failed - %s", g_strerror(errno));
1680 + if (close(vmaw->fd) < 0) {
1681 + vma_writer_set_error(vmaw, "vma_writer_close: "
1682 + "close failed - %s", g_strerror(errno));
1686 + for (i = 0; i <= 255; i++) {
1687 + VmaStreamInfo *si = &vmaw->stream_info[i];
1689 + if (!si->finished) {
1690 + vma_writer_set_error(vmaw, "vma_writer_close: "
1691 + "detected open stream '%s'", si->devname);
1692 + } else if ((si->transferred != si->size) &&
1693 + (i != vmaw->vmstate_stream)) {
1694 + vma_writer_set_error(vmaw, "vma_writer_close: "
1695 + "incomplete stream '%s' (%zd != %zd)",
1696 + si->devname, si->transferred, si->size);
1701 + for (i = 0; i <= 255; i++) {
1702 + vmaw->stream_info[i].finished = 1; /* mark as closed */
1707 + if (vmaw->status < 0 && *errp == NULL) {
1708 + error_setg(errp, "%s", vmaw->errmsg);
1711 + qemu_co_mutex_unlock(&vmaw->flush_lock);
1713 + return vmaw->status;
1716 +void vma_writer_destroy(VmaWriter *vmaw)
1722 + for (i = 0; i <= 255; i++) {
1723 + if (vmaw->stream_info[i].devname) {
1724 + g_free(vmaw->stream_info[i].devname);
1728 + if (vmaw->md5csum) {
1729 + g_checksum_free(vmaw->md5csum);
1732 + qemu_vfree(vmaw->headerbuf);
1733 + qemu_vfree(vmaw->outbuf);
1736 diff --git a/vma.c b/vma.c
1737 new file mode 100644
1738 index 0000000000..347f6283ca
1743 + * VMA: Virtual Machine Archive
1745 + * Copyright (C) 2012-2013 Proxmox Server Solutions
1748 + * Dietmar Maurer (dietmar@proxmox.com)
1750 + * This work is licensed under the terms of the GNU GPL, version 2 or later.
1751 + * See the COPYING file in the top-level directory.
1755 +#include "qemu/osdep.h"
1759 +#include "qemu/module.h"
1760 +#include "qemu/error-report.h"
1761 +#include "qemu/main-loop.h"
1762 +#include "qemu/cutils.h"
1763 +#include "qemu/memalign.h"
1764 +#include "qapi/qmp/qdict.h"
1765 +#include "sysemu/block-backend.h"
1767 +static void help(void)
1769 + const char *help_msg =
1770 + "usage: vma command [command options]\n"
1772 + "vma list <filename>\n"
1773 + "vma config <filename> [-c config]\n"
1774 + "vma create <filename> [-c config] pathname ...\n"
1775 + "vma extract <filename> [-d <drive-list>] [-r <fifo>] <targetdir>\n"
1776 + "vma verify <filename> [-v]\n"
1779 + printf("%s", help_msg);
1783 +static const char *extract_devname(const char *path, char **devname, int index)
1787 + const char *sep = strchr(path, '=');
1790 + *devname = g_strndup(path, sep - path);
1794 + *devname = g_strdup_printf("disk%d", index);
1803 +static void print_content(VmaReader *vmar)
1807 + VmaHeader *head = vma_reader_get_header(vmar);
1809 + GList *l = vma_reader_get_config_data(vmar);
1810 + while (l && l->data) {
1811 + VmaConfigData *cdata = (VmaConfigData *)l->data;
1812 + l = g_list_next(l);
1813 + printf("CFG: size: %d name: %s\n", cdata->len, cdata->name);
1817 + VmaDeviceInfo *di;
1818 + for (i = 1; i < 255; i++) {
1819 + di = vma_reader_get_device_info(vmar, i);
1821 + if (strcmp(di->devname, "vmstate") == 0) {
1822 + printf("VMSTATE: dev_id=%d memory: %zd\n", i, di->size);
1824 + printf("DEV: dev_id=%d size: %zd devname: %s\n",
1825 + i, di->size, di->devname);
1829 + /* ctime is the last entry we print */
1830 + printf("CTIME: %s", ctime(&head->ctime));
1834 +static int list_content(int argc, char **argv)
1837 + const char *filename;
1840 + c = getopt(argc, argv, "h");
1850 + g_assert_not_reached();
1854 + /* Get the filename */
1855 + if ((optind + 1) != argc) {
1858 + filename = argv[optind++];
1860 + Error *errp = NULL;
1861 + VmaReader *vmar = vma_reader_create(filename, &errp);
1864 + g_error("%s", error_get_pretty(errp));
1867 + print_content(vmar);
1869 + vma_reader_destroy(vmar);
1874 +typedef struct RestoreMap {
1878 + uint64_t throttling_bps;
1879 + char *throttling_group;
1885 +static bool try_parse_option(char **line, const char *optname, char **out, const char *inbuf) {
1886 + size_t optlen = strlen(optname);
1887 + if (strncmp(*line, optname, optlen) != 0 || (*line)[optlen] != '=') {
1891 + g_error("read map failed - duplicate value for option '%s'", optname);
1893 + char *value = (*line) + optlen + 1; /* including a '=' */
1894 + char *colon = strchr(value, ':');
1896 + g_error("read map failed - option '%s' not terminated ('%s')",
1900 + *out = g_strndup(value, colon - value);
1904 +static uint64_t verify_u64(const char *text) {
1906 + const char *endptr = NULL;
1907 + if (qemu_strtou64(text, &endptr, 0, &value) != 0 || !endptr || *endptr) {
1908 + g_error("read map failed - not a number: %s", text);
1913 +static int extract_content(int argc, char **argv)
1917 + const char *filename;
1918 + const char *dirname;
1919 + const char *readmap = NULL;
1920 + gchar **drive_list = NULL;
1923 + c = getopt(argc, argv, "hvd:r:");
1933 + drive_list = g_strsplit(optarg, ",", 254);
1946 + /* Get the filename */
1947 + if ((optind + 2) != argc) {
1950 + filename = argv[optind++];
1951 + dirname = argv[optind++];
1953 + Error *errp = NULL;
1954 + VmaReader *vmar = vma_reader_create(filename, &errp);
1957 + g_error("%s", error_get_pretty(errp));
1960 + if (mkdir(dirname, 0777) < 0) {
1961 + g_error("unable to create target directory %s - %s",
1962 + dirname, g_strerror(errno));
1965 + GList *l = vma_reader_get_config_data(vmar);
1966 + while (l && l->data) {
1967 + VmaConfigData *cdata = (VmaConfigData *)l->data;
1968 + l = g_list_next(l);
1969 + char *cfgfn = g_strdup_printf("%s/%s", dirname, cdata->name);
1970 + GError *err = NULL;
1971 + if (!g_file_set_contents(cfgfn, (gchar *)cdata->data, cdata->len,
1973 + g_error("unable to write file: %s", err->message);
1977 + GHashTable *devmap = g_hash_table_new(g_str_hash, g_str_equal);
1980 + print_content(vmar);
1982 + FILE *map = fopen(readmap, "r");
1984 + g_error("unable to open fifo %s - %s", readmap, g_strerror(errno));
1989 + char *line = fgets(inbuf, sizeof(inbuf), map);
1990 + char *format = NULL;
1992 + char *group = NULL;
1993 + char *cache = NULL;
1994 + char *devname = NULL;
1995 + bool skip = false;
1996 + uint64_t bps_value = 0;
1997 + const char *path = NULL;
1998 + bool write_zero = true;
2000 + if (!line || line[0] == '\0' || !strcmp(line, "done\n")) {
2003 + int len = strlen(line);
2004 + if (line[len - 1] == '\n') {
2005 + line[len - 1] = '\0';
2012 + if (strncmp(line, "skip", 4) == 0) {
2013 + if (len < 6 || line[4] != '=') {
2014 + g_error("read map failed - option 'skip' has no value ('%s')",
2017 + devname = line + 5;
2022 + if (!try_parse_option(&line, "format", &format, inbuf) &&
2023 + !try_parse_option(&line, "throttling.bps", &bps, inbuf) &&
2024 + !try_parse_option(&line, "throttling.group", &group, inbuf) &&
2025 + !try_parse_option(&line, "cache", &cache, inbuf))
2032 + bps_value = verify_u64(bps);
2036 + if (line[0] == '0' && line[1] == ':') {
2038 + write_zero = false;
2039 + } else if (line[0] == '1' && line[1] == ':') {
2041 + write_zero = true;
2043 + g_error("read map failed - parse error ('%s')", inbuf);
2046 + path = extract_devname(path, &devname, -1);
2050 + g_error("read map failed - no dev name specified ('%s')",
2054 + RestoreMap *map = g_new0(RestoreMap, 1);
2055 + map->devname = g_strdup(devname);
2056 + map->path = g_strdup(path);
2057 + map->format = format;
2058 + map->throttling_bps = bps_value;
2059 + map->throttling_group = group;
2060 + map->cache = cache;
2061 + map->write_zero = write_zero;
2064 + g_hash_table_insert(devmap, map->devname, map);
2070 + int vmstate_fd = -1;
2071 + bool drive_rename_bitmap[255];
2072 + memset(drive_rename_bitmap, 0, sizeof(drive_rename_bitmap));
2074 + for (i = 1; i < 255; i++) {
2075 + VmaDeviceInfo *di = vma_reader_get_device_info(vmar, i);
2076 + if (di && (strcmp(di->devname, "vmstate") == 0)) {
2077 + char *statefn = g_strdup_printf("%s/vmstate.bin", dirname);
2078 + vmstate_fd = open(statefn, O_WRONLY|O_CREAT|O_EXCL, 0644);
2079 + if (vmstate_fd < 0) {
2080 + g_error("create vmstate file '%s' failed - %s", statefn,
2081 + g_strerror(errno));
2085 + char *devfn = NULL;
2086 + const char *format = NULL;
2087 + uint64_t throttling_bps = 0;
2088 + const char *throttling_group = NULL;
2089 + const char *cache = NULL;
2090 + int flags = BDRV_O_RDWR;
2091 + bool write_zero = true;
2092 + bool skip = false;
2094 + BlockBackend *blk = NULL;
2099 + for (j = 0; drive_list[j]; j++) {
2100 + if (strcmp(drive_list[j], di->devname) == 0) {
2102 + drive_rename_bitmap[i] = true;
2107 + drive_rename_bitmap[i] = true;
2110 + if (!skip && readmap) {
2112 + map = (RestoreMap *)g_hash_table_lookup(devmap, di->devname);
2113 + if (map == NULL) {
2114 + g_error("no device name mapping for %s", di->devname);
2116 + devfn = map->path;
2117 + format = map->format;
2118 + throttling_bps = map->throttling_bps;
2119 + throttling_group = map->throttling_group;
2120 + cache = map->cache;
2121 + write_zero = map->write_zero;
2123 + } else if (!skip) {
2124 + devfn = g_strdup_printf("%s/tmp-disk-%s.raw",
2125 + dirname, di->devname);
2126 + printf("DEVINFO %s %zd\n", devfn, di->size);
2128 + bdrv_img_create(devfn, "raw", NULL, NULL, NULL, di->size,
2129 + flags, true, &errp);
2131 + g_error("can't create file %s: %s", devfn,
2132 + error_get_pretty(errp));
2135 + /* Note: we created an empty file above, so there is no
2136 + * need to write zeroes (so we generate a sparse file)
2138 + write_zero = false;
2142 + size_t devlen = strlen(devfn);
2143 + QDict *options = NULL;
2144 + bool writethrough;
2146 + /* explicit format from commandline */
2147 + options = qdict_new();
2148 + qdict_put_str(options, "driver", format);
2149 + } else if ((devlen > 4 && strcmp(devfn+devlen-4, ".raw") == 0) ||
2150 + strncmp(devfn, "/dev/", 5) == 0)
2152 + /* This part is now deprecated for PVE as well (just as qemu
2153 + * deprecated not specifying an explicit raw format, too.
2155 + /* explicit raw format */
2156 + options = qdict_new();
2157 + qdict_put_str(options, "driver", "raw");
2160 + if (cache && bdrv_parse_cache_mode(cache, &flags, &writethrough)) {
2161 + g_error("invalid cache option: %s\n", cache);
2164 + if (errp || !(blk = blk_new_open(devfn, NULL, options, flags, &errp))) {
2165 + g_error("can't open file %s - %s", devfn,
2166 + error_get_pretty(errp));
2170 + blk_set_enable_write_cache(blk, !writethrough);
2173 + if (throttling_group) {
2174 + blk_io_limits_enable(blk, throttling_group);
2177 + if (throttling_bps) {
2178 + if (!throttling_group) {
2179 + blk_io_limits_enable(blk, devfn);
2182 + ThrottleConfig cfg;
2183 + throttle_config_init(&cfg);
2184 + cfg.buckets[THROTTLE_BPS_WRITE].avg = throttling_bps;
2185 + Error *err = NULL;
2186 + if (!throttle_is_valid(&cfg, &err)) {
2187 + error_report_err(err);
2188 + g_error("failed to apply throttling");
2190 + blk_set_io_limits(blk, &cfg);
2194 + if (vma_reader_register_bs(vmar, i, blk, write_zero, skip, &errp) < 0) {
2195 + g_error("%s", error_get_pretty(errp));
2205 + g_strfreev(drive_list);
2208 + if (vma_reader_restore(vmar, vmstate_fd, verbose, &errp) < 0) {
2209 + g_error("restore failed - %s", error_get_pretty(errp));
2213 + for (i = 1; i < 255; i++) {
2214 + VmaDeviceInfo *di = vma_reader_get_device_info(vmar, i);
2215 + if (di && drive_rename_bitmap[i]) {
2216 + char *tmpfn = g_strdup_printf("%s/tmp-disk-%s.raw",
2217 + dirname, di->devname);
2218 + char *fn = g_strdup_printf("%s/disk-%s.raw",
2219 + dirname, di->devname);
2220 + if (rename(tmpfn, fn) != 0) {
2221 + g_error("rename %s to %s failed - %s",
2222 + tmpfn, fn, g_strerror(errno));
2228 + vma_reader_destroy(vmar);
2235 +static int verify_content(int argc, char **argv)
2239 + const char *filename;
2242 + c = getopt(argc, argv, "hv");
2259 + /* Get the filename */
2260 + if ((optind + 1) != argc) {
2263 + filename = argv[optind++];
2265 + Error *errp = NULL;
2266 + VmaReader *vmar = vma_reader_create(filename, &errp);
2269 + g_error("%s", error_get_pretty(errp));
2273 + print_content(vmar);
2276 + if (vma_reader_verify(vmar, verbose, &errp) < 0) {
2277 + g_error("verify failed - %s", error_get_pretty(errp));
2280 + vma_reader_destroy(vmar);
2287 +typedef struct BackupJob {
2288 + BlockBackend *target;
2294 +#define BACKUP_SECTORS_PER_CLUSTER (VMA_CLUSTER_SIZE / BDRV_SECTOR_SIZE)
2296 +static void coroutine_fn backup_run_empty(void *opaque)
2298 + VmaWriter *vmaw = (VmaWriter *)opaque;
2300 + vma_writer_flush_output(vmaw);
2302 + Error *err = NULL;
2303 + if (vma_writer_close(vmaw, &err) != 0) {
2304 + g_warning("vma_writer_close failed %s", error_get_pretty(err));
2308 +static void coroutine_fn backup_run(void *opaque)
2310 + BackupJob *job = (BackupJob *)opaque;
2312 + QEMUIOVector qiov;
2314 + int64_t start, end, readlen;
2317 + unsigned char *buf = blk_blockalign(job->target, VMA_CLUSTER_SIZE);
2320 + end = DIV_ROUND_UP(job->len / BDRV_SECTOR_SIZE,
2321 + BACKUP_SECTORS_PER_CLUSTER);
2323 + for (; start < end; start++) {
2324 + iov.iov_base = buf;
2325 + iov.iov_len = VMA_CLUSTER_SIZE;
2326 + qemu_iovec_init_external(&qiov, &iov, 1);
2328 + if (start + 1 == end) {
2329 + memset(buf, 0, VMA_CLUSTER_SIZE);
2330 + readlen = job->len - start * VMA_CLUSTER_SIZE;
2331 + assert(readlen > 0 && readlen <= VMA_CLUSTER_SIZE);
2333 + readlen = VMA_CLUSTER_SIZE;
2336 + ret = blk_co_preadv(job->target, start * VMA_CLUSTER_SIZE,
2337 + readlen, &qiov, 0);
2339 + vma_writer_set_error(job->vmaw, "read error");
2344 + if (vma_writer_write(job->vmaw, job->dev_id, start, buf, &zb) < 0) {
2345 + vma_writer_set_error(job->vmaw, "backup_dump_cb vma_writer_write failed");
2352 + if (vma_writer_close_stream(job->vmaw, job->dev_id) <= 0) {
2353 + Error *err = NULL;
2354 + if (vma_writer_close(job->vmaw, &err) != 0) {
2355 + g_warning("vma_writer_close failed %s", error_get_pretty(err));
2361 +static int create_archive(int argc, char **argv)
2365 + const char *archivename;
2366 + GList *backup_coroutines = NULL;
2367 + GList *config_files = NULL;
2370 + c = getopt(argc, argv, "hvc:");
2380 + config_files = g_list_append(config_files, optarg);
2386 + g_assert_not_reached();
2391 + /* make sure we an archive name */
2392 + if ((optind + 1) > argc) {
2396 + archivename = argv[optind++];
2399 + uuid_generate(uuid);
2401 + Error *local_err = NULL;
2402 + VmaWriter *vmaw = vma_writer_create(archivename, uuid, &local_err);
2404 + if (vmaw == NULL) {
2405 + g_error("%s", error_get_pretty(local_err));
2408 + GList *l = config_files;
2409 + while (l && l->data) {
2410 + char *name = l->data;
2411 + char *cdata = NULL;
2413 + GError *err = NULL;
2414 + if (!g_file_get_contents(name, &cdata, &clen, &err)) {
2415 + unlink(archivename);
2416 + g_error("Unable to read file: %s", err->message);
2419 + if (vma_writer_add_config(vmaw, name, cdata, clen) != 0) {
2420 + unlink(archivename);
2421 + g_error("Unable to append config data %s (len = %zd)",
2424 + l = g_list_next(l);
2428 + while (optind < argc) {
2429 + const char *path = argv[optind++];
2430 + char *devname = NULL;
2431 + path = extract_devname(path, &devname, devcount++);
2433 + Error *errp = NULL;
2434 + BlockBackend *target;
2436 + target = blk_new_open(path, NULL, NULL, 0, &errp);
2438 + unlink(archivename);
2439 + g_error("bdrv_open '%s' failed - %s", path, error_get_pretty(errp));
2441 + int64_t size = blk_getlength(target);
2442 + int dev_id = vma_writer_register_stream(vmaw, devname, size);
2443 + if (dev_id <= 0) {
2444 + unlink(archivename);
2445 + g_error("vma_writer_register_stream '%s' failed", devname);
2448 + BackupJob *job = g_new0(BackupJob, 1);
2450 + job->target = target;
2452 + job->dev_id = dev_id;
2454 + Coroutine *co = qemu_coroutine_create(backup_run, job);
2455 + // Don't enter coroutine yet, because it might write the header before
2456 + // all streams can be registered.
2457 + backup_coroutines = g_list_append(backup_coroutines, co);
2460 + VmaStatus vmastat;
2462 + int last_percent = -1;
2465 + GList *entry = backup_coroutines;
2466 + while (entry && entry->data) {
2467 + Coroutine *co = entry->data;
2468 + qemu_coroutine_enter(co);
2469 + entry = g_list_next(entry);
2473 + main_loop_wait(false);
2474 + vma_writer_get_status(vmaw, &vmastat);
2478 + uint64_t total = 0;
2479 + uint64_t transferred = 0;
2480 + uint64_t zero_bytes = 0;
2483 + for (i = 0; i < 256; i++) {
2484 + if (vmastat.stream_info[i].size) {
2485 + total += vmastat.stream_info[i].size;
2486 + transferred += vmastat.stream_info[i].transferred;
2487 + zero_bytes += vmastat.stream_info[i].zero_bytes;
2490 + percent = (transferred*100)/total;
2491 + if (percent != last_percent) {
2492 + fprintf(stderr, "progress %d%% %zd/%zd %zd\n", percent,
2493 + transferred, total, zero_bytes);
2496 + last_percent = percent;
2500 + if (vmastat.closed) {
2505 + Coroutine *co = qemu_coroutine_create(backup_run_empty, vmaw);
2506 + qemu_coroutine_enter(co);
2508 + main_loop_wait(false);
2509 + vma_writer_get_status(vmaw, &vmastat);
2510 + if (vmastat.closed) {
2518 + vma_writer_get_status(vmaw, &vmastat);
2521 + for (i = 0; i < 256; i++) {
2522 + VmaStreamInfo *si = &vmastat.stream_info[i];
2524 + fprintf(stderr, "image %s: size=%zd zeros=%zd saved=%zd\n",
2525 + si->devname, si->size, si->zero_bytes,
2526 + si->size - si->zero_bytes);
2531 + if (vmastat.status < 0) {
2532 + unlink(archivename);
2533 + g_error("creating vma archive failed");
2536 + g_list_free(backup_coroutines);
2537 + g_list_free(config_files);
2538 + vma_writer_destroy(vmaw);
2542 +static int dump_config(int argc, char **argv)
2545 + const char *filename;
2546 + const char *config_name = "qemu-server.conf";
2549 + c = getopt(argc, argv, "hc:");
2559 + config_name = optarg;
2566 + /* Get the filename */
2567 + if ((optind + 1) != argc) {
2570 + filename = argv[optind++];
2572 + Error *errp = NULL;
2573 + VmaReader *vmar = vma_reader_create(filename, &errp);
2576 + g_error("%s", error_get_pretty(errp));
2580 + GList *l = vma_reader_get_config_data(vmar);
2581 + while (l && l->data) {
2582 + VmaConfigData *cdata = (VmaConfigData *)l->data;
2583 + l = g_list_next(l);
2584 + if (strcmp(cdata->name, config_name) == 0) {
2586 + fwrite(cdata->data, cdata->len, 1, stdout);
2591 + vma_reader_destroy(vmar);
2596 + fprintf(stderr, "unable to find configuration data '%s'\n", config_name);
2603 +int main(int argc, char **argv)
2605 + const char *cmdname;
2606 + Error *main_loop_err = NULL;
2608 + error_init(argv[0]);
2609 + module_call_init(MODULE_INIT_TRACE);
2610 + qemu_init_exec_dir(argv[0]);
2612 + if (qemu_init_main_loop(&main_loop_err)) {
2613 + g_error("%s", error_get_pretty(main_loop_err));
2617 + module_call_init(MODULE_INIT_QOM);
2623 + cmdname = argv[1];
2627 + if (!strcmp(cmdname, "list")) {
2628 + return list_content(argc, argv);
2629 + } else if (!strcmp(cmdname, "create")) {
2630 + return create_archive(argc, argv);
2631 + } else if (!strcmp(cmdname, "extract")) {
2632 + return extract_content(argc, argv);
2633 + } else if (!strcmp(cmdname, "verify")) {
2634 + return verify_content(argc, argv);
2635 + } else if (!strcmp(cmdname, "config")) {
2636 + return dump_config(argc, argv);
2642 diff --git a/vma.h b/vma.h
2643 new file mode 100644
2644 index 0000000000..86d2873aa5
2649 + * VMA: Virtual Machine Archive
2651 + * Copyright (C) Proxmox Server Solutions
2654 + * Dietmar Maurer (dietmar@proxmox.com)
2656 + * This work is licensed under the terms of the GNU GPL, version 2 or later.
2657 + * See the COPYING file in the top-level directory.
2661 +#ifndef BACKUP_VMA_H
2662 +#define BACKUP_VMA_H
2664 +#include <uuid/uuid.h>
2665 +#include "qapi/error.h"
2666 +#include "block/block.h"
2668 +#define VMA_BLOCK_BITS 12
2669 +#define VMA_BLOCK_SIZE (1<<VMA_BLOCK_BITS)
2670 +#define VMA_CLUSTER_BITS (VMA_BLOCK_BITS+4)
2671 +#define VMA_CLUSTER_SIZE (1<<VMA_CLUSTER_BITS)
2673 +#if VMA_CLUSTER_SIZE != 65536
2674 +#error unexpected cluster size
2677 +#define VMA_EXTENT_HEADER_SIZE 512
2678 +#define VMA_BLOCKS_PER_EXTENT 59
2679 +#define VMA_MAX_CONFIGS 256
2681 +#define VMA_MAX_EXTENT_SIZE \
2682 + (VMA_EXTENT_HEADER_SIZE+VMA_CLUSTER_SIZE*VMA_BLOCKS_PER_EXTENT)
2683 +#if VMA_MAX_EXTENT_SIZE != 3867136
2684 +#error unexpected VMA_EXTENT_SIZE
2687 +/* File Format Definitions */
2689 +#define VMA_MAGIC (GUINT32_TO_BE(('V'<<24)|('M'<<16)|('A'<<8)|0x00))
2690 +#define VMA_EXTENT_MAGIC (GUINT32_TO_BE(('V'<<24)|('M'<<16)|('A'<<8)|'E'))
2692 +typedef struct VmaDeviceInfoHeader {
2693 + uint32_t devname_ptr; /* offset into blob_buffer table */
2694 + uint32_t reserved0;
2695 + uint64_t size; /* device size in bytes */
2696 + uint64_t reserved1;
2697 + uint64_t reserved2;
2698 +} VmaDeviceInfoHeader;
2700 +typedef struct VmaHeader {
2703 + unsigned char uuid[16];
2705 + unsigned char md5sum[16];
2707 + uint32_t blob_buffer_offset;
2708 + uint32_t blob_buffer_size;
2709 + uint32_t header_size;
2711 + unsigned char reserved[1984];
2713 + uint32_t config_names[VMA_MAX_CONFIGS]; /* offset into blob_buffer table */
2714 + uint32_t config_data[VMA_MAX_CONFIGS]; /* offset into blob_buffer table */
2716 + uint32_t reserved1;
2718 + VmaDeviceInfoHeader dev_info[256];
2721 +typedef struct VmaExtentHeader {
2723 + uint16_t reserved1;
2724 + uint16_t block_count;
2725 + unsigned char uuid[16];
2726 + unsigned char md5sum[16];
2727 + uint64_t blockinfo[VMA_BLOCKS_PER_EXTENT];
2730 +/* functions/definitions to read/write vma files */
2732 +typedef struct VmaReader VmaReader;
2734 +typedef struct VmaWriter VmaWriter;
2736 +typedef struct VmaConfigData {
2742 +typedef struct VmaStreamInfo {
2744 + uint64_t cluster_count;
2745 + uint64_t transferred;
2746 + uint64_t zero_bytes;
2751 +typedef struct VmaStatus {
2754 + char errmsg[8192];
2755 + char uuid_str[37];
2756 + VmaStreamInfo stream_info[256];
2759 +typedef struct VmaDeviceInfo {
2760 + uint64_t size; /* device size in bytes */
2761 + const char *devname;
2764 +VmaWriter *vma_writer_create(const char *filename, uuid_t uuid, Error **errp);
2765 +int vma_writer_close(VmaWriter *vmaw, Error **errp);
2766 +void vma_writer_error_propagate(VmaWriter *vmaw, Error **errp);
2767 +void vma_writer_destroy(VmaWriter *vmaw);
2768 +int vma_writer_add_config(VmaWriter *vmaw, const char *name, gpointer data,
2770 +int vma_writer_register_stream(VmaWriter *vmaw, const char *devname,
2773 +int64_t coroutine_fn vma_writer_write(VmaWriter *vmaw, uint8_t dev_id,
2774 + int64_t cluster_num,
2775 + const unsigned char *buf,
2776 + size_t *zero_bytes);
2778 +int coroutine_fn vma_writer_close_stream(VmaWriter *vmaw, uint8_t dev_id);
2779 +int coroutine_fn vma_writer_flush_output(VmaWriter *vmaw);
2781 +int vma_writer_get_status(VmaWriter *vmaw, VmaStatus *status);
2782 +void vma_writer_set_error(VmaWriter *vmaw, const char *fmt, ...) G_GNUC_PRINTF(2, 3);
2785 +VmaReader *vma_reader_create(const char *filename, Error **errp);
2786 +void vma_reader_destroy(VmaReader *vmar);
2787 +VmaHeader *vma_reader_get_header(VmaReader *vmar);
2788 +GList *vma_reader_get_config_data(VmaReader *vmar);
2789 +VmaDeviceInfo *vma_reader_get_device_info(VmaReader *vmar, guint8 dev_id);
2790 +int vma_reader_register_bs(VmaReader *vmar, guint8 dev_id,
2791 + BlockBackend *target, bool write_zeroes,
2792 + bool skip, Error **errp);
2793 +int vma_reader_restore(VmaReader *vmar, int vmstate_fd, bool verbose,
2795 +int vma_reader_verify(VmaReader *vmar, bool verbose, Error **errp);
2797 +#endif /* BACKUP_VMA_H */