1 From c04d17631e00ff368e42dc5d24e94152b6526014 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
6 This is a very simple archive format, see docs/specs/vma_spec.txt
8 Signed-off-by: Dietmar Maurer <dietmar@proxmox.com>
14 docs/specs/vma_spec.txt | 24 ++
15 vma-reader.c | 799 ++++++++++++++++++++++++++++++++++++++++
16 vma-writer.c | 932 +++++++++++++++++++++++++++++++++++++++++++++++
17 vma.c | 559 ++++++++++++++++++++++++++++
19 9 files changed, 2467 insertions(+), 4 deletions(-)
20 create mode 100644 docs/specs/vma_spec.txt
21 create mode 100644 vma-reader.c
22 create mode 100644 vma-writer.c
23 create mode 100644 vma.c
24 create mode 100644 vma.h
26 diff --git a/Makefile b/Makefile
27 index 0d9099a..16f1c25 100644
30 @@ -115,7 +115,7 @@ ifeq ($(CONFIG_SMARTCARD_NSS),y)
31 include $(SRC_PATH)/libcacard/Makefile
34 -all: $(DOCS) $(TOOLS) $(HELPERS-y) recurse-all
35 +all: $(DOCS) $(TOOLS) vma$(EXESUF) $(HELPERS-y) recurse-all
37 config-host.h: config-host.h-timestamp
38 config-host.h-timestamp: config-host.mak
39 @@ -167,6 +167,7 @@ qemu-img.o: qemu-img-cmds.h
40 qemu-img$(EXESUF): qemu-img.o $(block-obj-y) libqemuutil.a libqemustub.a
41 qemu-nbd$(EXESUF): qemu-nbd.o $(block-obj-y) libqemuutil.a libqemustub.a
42 qemu-io$(EXESUF): qemu-io.o cmd.o $(block-obj-y) libqemuutil.a libqemustub.a
43 +vma$(EXESUF): vma.o vma-writer.o vma-reader.o $(block-obj-y) libqemuutil.a libqemustub.a
45 qemu-bridge-helper$(EXESUF): qemu-bridge-helper.o
47 diff --git a/Makefile.objs b/Makefile.objs
48 index d72822b..111b9f8 100644
51 @@ -13,7 +13,7 @@ block-obj-$(CONFIG_POSIX) += aio-posix.o
52 block-obj-$(CONFIG_WIN32) += aio-win32.o
54 block-obj-y += qapi-types.o qapi-visit.o
55 -block-obj-y += backup.o
56 +block-obj-y += vma-writer.o backup.o
58 block-obj-y += qemu-coroutine.o qemu-coroutine-lock.o qemu-coroutine-io.o
59 block-obj-y += qemu-coroutine-sleep.o
60 diff --git a/backup.h b/backup.h
61 index be52ea4..3a81e52 100644
67 #include <uuid/uuid.h>
68 +#include "block/block.h"
70 #define BACKUP_CLUSTER_BITS 16
71 #define BACKUP_CLUSTER_SIZE (1<<BACKUP_CLUSTER_BITS)
72 diff --git a/blockdev.c b/blockdev.c
73 index 92a7ca3..aad6e0c 100644
78 #include "sysemu/arch_init.h"
82 static QTAILQ_HEAD(drivelist, DriveInfo) drives = QTAILQ_HEAD_INITIALIZER(drives);
84 @@ -1530,10 +1531,11 @@ char *qmp_backup(const char *backupfile, bool has_format, BackupFormat format,
85 /* Todo: try to auto-detect format based on file name */
86 format = has_format ? format : BACKUP_FORMAT_VMA;
88 - /* fixme: find driver for specifued format */
89 const BackupDriver *driver = NULL;
92 + if (format == BACKUP_FORMAT_VMA) {
93 + driver = &backup_vma_driver;
95 error_set(errp, ERROR_CLASS_GENERIC_ERROR, "unknown backup format");
98 diff --git a/docs/specs/vma_spec.txt b/docs/specs/vma_spec.txt
100 index 0000000..052c629
102 +++ b/docs/specs/vma_spec.txt
104 +=Virtual Machine Archive format (VMA)=
106 +This format contains a header which includes the VM configuration as
107 +binary blobs, and a list of devices (dev_id, name).
109 +The actual VM image data is stored inside extents. An extent contains
110 +up to 64 clusters, and start with a 512 byte header containing
111 +additional information for those clusters.
113 +We use a cluster size of 65536, and use 8 bytes for each
114 +cluster in the header to store the following information:
116 +* 1 byte dev_id (to identity the drive)
117 +* 2 bytes zero indicator (mark zero regions (16x4096))
118 +* 4 bytes cluster number
119 +* 1 byte not used (reserved)
121 +We only store non-zero blocks (such block is 4096 bytes).
123 +Each archive is marked with a uuid. The archive header and all
124 +extent headers includes that uuid and a MD5 checksum (over header
128 diff --git a/vma-reader.c b/vma-reader.c
130 index 0000000..7e81847
135 + * VMA: Virtual Machine Archive
137 + * Copyright (C) 2012 Proxmox Server Solutions
140 + * Dietmar Maurer (dietmar@proxmox.com)
142 + * This work is licensed under the terms of the GNU GPL, version 2 or later.
143 + * See the COPYING file in the top-level directory.
152 +#include <sys/types.h>
153 +#include <sys/stat.h>
156 +#include <uuid/uuid.h>
158 +#include "qemu-common.h"
159 +#include "qemu/timer.h"
160 +#include "qemu/ratelimit.h"
162 +#include "block/block.h"
164 +#define BITS_PER_LONG (sizeof(unsigned long) * 8)
166 +static unsigned char zero_vma_block[VMA_BLOCK_SIZE];
168 +typedef struct VmaRestoreState {
169 + BlockDriverState *bs;
171 + unsigned long *bitmap;
177 + GChecksum *md5csum;
178 + GHashTable *blob_hash;
179 + unsigned char *head_data;
180 + VmaDeviceInfo devinfo[256];
181 + VmaRestoreState rstate[256];
183 + guint8 vmstate_stream;
184 + uint32_t vmstate_clusters;
185 + /* to show restore percentage if run with -v */
187 + int64_t cluster_count;
188 + int64_t clusters_read;
189 + int clusters_read_per;
193 +g_int32_hash(gconstpointer v)
195 + return *(const uint32_t *)v;
199 +g_int32_equal(gconstpointer v1, gconstpointer v2)
201 + return *((const uint32_t *)v1) == *((const uint32_t *)v2);
204 +static int vma_reader_get_bitmap(VmaRestoreState *rstate, int64_t cluster_num)
207 + assert(rstate->bitmap);
209 + unsigned long val, idx, bit;
211 + idx = cluster_num / BITS_PER_LONG;
213 + assert(rstate->bitmap_size > idx);
215 + bit = cluster_num % BITS_PER_LONG;
216 + val = rstate->bitmap[idx];
218 + return !!(val & (1UL << bit));
221 +static void vma_reader_set_bitmap(VmaRestoreState *rstate, int64_t cluster_num,
225 + assert(rstate->bitmap);
227 + unsigned long val, idx, bit;
229 + idx = cluster_num / BITS_PER_LONG;
231 + assert(rstate->bitmap_size > idx);
233 + bit = cluster_num % BITS_PER_LONG;
234 + val = rstate->bitmap[idx];
236 + if (!(val & (1UL << bit))) {
240 + if (val & (1UL << bit)) {
241 + val &= ~(1UL << bit);
244 + rstate->bitmap[idx] = val;
247 +typedef struct VmaBlob {
253 +static const VmaBlob *get_header_blob(VmaReader *vmar, uint32_t pos)
256 + assert(vmar->blob_hash);
258 + return g_hash_table_lookup(vmar->blob_hash, &pos);
261 +static const char *get_header_str(VmaReader *vmar, uint32_t pos)
263 + const VmaBlob *blob = get_header_blob(vmar, pos);
267 + const char *res = (char *)blob->data;
268 + if (res[blob->len-1] != '\0') {
275 +safe_read(int fd, unsigned char *buf, size_t count)
280 + n = read(fd, buf, count);
281 + } while (n < 0 && errno == EINTR);
287 +full_read(int fd, unsigned char *buf, size_t len)
295 + n = safe_read(fd, buf, len);
317 +void vma_reader_destroy(VmaReader *vmar)
321 + if (vmar->fd >= 0) {
325 + if (vmar->cdata_list) {
326 + g_list_free(vmar->cdata_list);
330 + for (i = 1; i < 256; i++) {
331 + if (vmar->rstate[i].bitmap) {
332 + g_free(vmar->rstate[i].bitmap);
336 + if (vmar->md5csum) {
337 + g_checksum_free(vmar->md5csum);
340 + if (vmar->blob_hash) {
341 + g_hash_table_destroy(vmar->blob_hash);
344 + if (vmar->head_data) {
345 + g_free(vmar->head_data);
352 +static int vma_reader_read_head(VmaReader *vmar, Error **errp)
356 + assert(*errp == NULL);
358 + unsigned char md5sum[16];
362 + vmar->head_data = g_malloc(sizeof(VmaHeader));
364 + if (full_read(vmar->fd, vmar->head_data, sizeof(VmaHeader)) !=
365 + sizeof(VmaHeader)) {
366 + error_setg(errp, "can't read vma header - %s",
367 + errno ? strerror(errno) : "got EOF");
371 + VmaHeader *h = (VmaHeader *)vmar->head_data;
373 + if (h->magic != VMA_MAGIC) {
374 + error_setg(errp, "not a vma file - wrong magic number");
378 + uint32_t header_size = GUINT32_FROM_BE(h->header_size);
379 + int need = header_size - sizeof(VmaHeader);
381 + error_setg(errp, "wrong vma header size %d", header_size);
385 + vmar->head_data = g_realloc(vmar->head_data, header_size);
386 + h = (VmaHeader *)vmar->head_data;
388 + if (full_read(vmar->fd, vmar->head_data + sizeof(VmaHeader), need) !=
390 + error_setg(errp, "can't read vma header data - %s",
391 + errno ? strerror(errno) : "got EOF");
395 + memcpy(md5sum, h->md5sum, 16);
396 + memset(h->md5sum, 0, 16);
398 + g_checksum_reset(vmar->md5csum);
399 + g_checksum_update(vmar->md5csum, vmar->head_data, header_size);
401 + g_checksum_get_digest(vmar->md5csum, (guint8 *)(h->md5sum), &csize);
403 + if (memcmp(md5sum, h->md5sum, 16) != 0) {
404 + error_setg(errp, "wrong vma header chechsum");
408 + /* we can modify header data after checksum verify */
409 + h->header_size = header_size;
411 + h->version = GUINT32_FROM_BE(h->version);
412 + if (h->version != 1) {
413 + error_setg(errp, "wrong vma version %d", h->version);
417 + h->ctime = GUINT64_FROM_BE(h->ctime);
418 + h->blob_buffer_offset = GUINT32_FROM_BE(h->blob_buffer_offset);
419 + h->blob_buffer_size = GUINT32_FROM_BE(h->blob_buffer_size);
421 + uint32_t bstart = h->blob_buffer_offset + 1;
422 + uint32_t bend = h->blob_buffer_offset + h->blob_buffer_size;
424 + if (bstart <= sizeof(VmaHeader)) {
425 + error_setg(errp, "wrong vma blob buffer offset %d",
426 + h->blob_buffer_offset);
430 + if (bend > header_size) {
431 + error_setg(errp, "wrong vma blob buffer size %d/%d",
432 + h->blob_buffer_offset, h->blob_buffer_size);
436 + while ((bstart + 2) <= bend) {
437 + uint32_t size = vmar->head_data[bstart] +
438 + (vmar->head_data[bstart+1] << 8);
439 + if ((bstart + size + 2) <= bend) {
440 + VmaBlob *blob = g_new0(VmaBlob, 1);
441 + blob->start = bstart - h->blob_buffer_offset;
443 + blob->data = vmar->head_data + bstart + 2;
444 + g_hash_table_insert(vmar->blob_hash, &blob->start, blob);
446 + bstart += size + 2;
451 + for (i = 1; i < 256; i++) {
452 + VmaDeviceInfoHeader *dih = &h->dev_info[i];
453 + uint32_t devname_ptr = GUINT32_FROM_BE(dih->devname_ptr);
454 + uint64_t size = GUINT64_FROM_BE(dih->size);
455 + const char *devname = get_header_str(vmar, devname_ptr);
457 + if (size && devname) {
459 + vmar->devinfo[i].size = size;
460 + vmar->devinfo[i].devname = devname;
462 + if (strcmp(devname, "vmstate") == 0) {
463 + vmar->vmstate_stream = i;
469 + error_setg(errp, "vma does not contain data");
473 + for (i = 0; i < VMA_MAX_CONFIGS; i++) {
474 + uint32_t name_ptr = GUINT32_FROM_BE(h->config_names[i]);
475 + uint32_t data_ptr = GUINT32_FROM_BE(h->config_data[i]);
477 + if (!(name_ptr && data_ptr)) {
480 + const char *name = get_header_str(vmar, name_ptr);
481 + const VmaBlob *blob = get_header_blob(vmar, data_ptr);
483 + if (!(name && blob)) {
484 + error_setg(errp, "vma contains invalid data pointers");
488 + VmaConfigData *cdata = g_new0(VmaConfigData, 1);
489 + cdata->name = name;
490 + cdata->data = blob->data;
491 + cdata->len = blob->len;
493 + vmar->cdata_list = g_list_append(vmar->cdata_list, cdata);
499 +VmaReader *vma_reader_create(const char *filename, Error **errp)
504 + VmaReader *vmar = g_new0(VmaReader, 1);
506 + if (strcmp(filename, "-") == 0) {
509 + vmar->fd = open(filename, O_RDONLY);
512 + if (vmar->fd < 0) {
513 + error_setg(errp, "can't open file %s - %s\n", filename,
518 + vmar->md5csum = g_checksum_new(G_CHECKSUM_MD5);
519 + if (!vmar->md5csum) {
520 + error_setg(errp, "can't allocate cmsum\n");
524 + vmar->blob_hash = g_hash_table_new_full(g_int32_hash, g_int32_equal,
527 + if (vma_reader_read_head(vmar, errp) < 0) {
535 + vma_reader_destroy(vmar);
541 +VmaHeader *vma_reader_get_header(VmaReader *vmar)
544 + assert(vmar->head_data);
546 + return (VmaHeader *)(vmar->head_data);
549 +GList *vma_reader_get_config_data(VmaReader *vmar)
552 + assert(vmar->head_data);
554 + return vmar->cdata_list;
557 +VmaDeviceInfo *vma_reader_get_device_info(VmaReader *vmar, guint8 dev_id)
562 + if (vmar->devinfo[dev_id].size && vmar->devinfo[dev_id].devname) {
563 + return &vmar->devinfo[dev_id];
569 +int vma_reader_register_bs(VmaReader *vmar, guint8 dev_id, BlockDriverState *bs,
570 + bool write_zeroes, Error **errp)
573 + assert(bs != NULL);
575 + assert(vmar->rstate[dev_id].bs == NULL);
577 + int64_t size = bdrv_getlength(bs);
578 + if (size != vmar->devinfo[dev_id].size) {
579 + error_setg(errp, "vma_reader_register_bs for stream %s failed - "
580 + "unexpected size %zd != %zd", vmar->devinfo[dev_id].devname,
581 + size, vmar->devinfo[dev_id].size);
585 + vmar->rstate[dev_id].bs = bs;
586 + vmar->rstate[dev_id].write_zeroes = write_zeroes;
588 + int64_t bitmap_size = (size/BDRV_SECTOR_SIZE) +
589 + (VMA_CLUSTER_SIZE/BDRV_SECTOR_SIZE) * BITS_PER_LONG - 1;
590 + bitmap_size /= (VMA_CLUSTER_SIZE/BDRV_SECTOR_SIZE) * BITS_PER_LONG;
592 + vmar->rstate[dev_id].bitmap_size = bitmap_size;
593 + vmar->rstate[dev_id].bitmap = g_new0(unsigned long, bitmap_size);
595 + vmar->cluster_count += size/VMA_CLUSTER_SIZE;
600 +static ssize_t safe_write(int fd, void *buf, size_t count)
605 + n = write(fd, buf, count);
606 + } while (n < 0 && errno == EINTR);
611 +static size_t full_write(int fd, void *buf, size_t len)
619 + n = safe_write(fd, buf, len);
629 + /* incomplete write ? */
636 +static int restore_write_data(VmaReader *vmar, guint8 dev_id,
637 + BlockDriverState *bs, int vmstate_fd,
638 + unsigned char *buf, int64_t sector_num,
639 + int nb_sectors, Error **errp)
643 + if (dev_id == vmar->vmstate_stream) {
644 + if (vmstate_fd >= 0) {
645 + int len = nb_sectors * BDRV_SECTOR_SIZE;
646 + int res = full_write(vmstate_fd, buf, len);
648 + error_setg(errp, "write vmstate failed %d", res);
653 + int res = bdrv_write(bs, sector_num, buf, nb_sectors);
655 + error_setg(errp, "bdrv_write to %s failed (%d)",
656 + bdrv_get_device_name(bs), res);
662 +static int restore_extent(VmaReader *vmar, unsigned char *buf,
663 + int extent_size, int vmstate_fd,
664 + bool verbose, Error **errp)
669 + VmaExtentHeader *ehead = (VmaExtentHeader *)buf;
670 + int start = VMA_EXTENT_HEADER_SIZE;
673 + for (i = 0; i < VMA_BLOCKS_PER_EXTENT; i++) {
674 + uint64_t block_info = GUINT64_FROM_BE(ehead->blockinfo[i]);
675 + uint64_t cluster_num = block_info & 0xffffffff;
676 + uint8_t dev_id = (block_info >> 32) & 0xff;
677 + uint16_t mask = block_info >> (32+16);
678 + int64_t max_sector;
684 + VmaRestoreState *rstate = &vmar->rstate[dev_id];
685 + BlockDriverState *bs = NULL;
687 + if (dev_id != vmar->vmstate_stream) {
690 + error_setg(errp, "got wrong dev id %d", dev_id);
694 + if (vma_reader_get_bitmap(rstate, cluster_num)) {
695 + error_setg(errp, "found duplicated cluster %zd for stream %s",
696 + cluster_num, vmar->devinfo[dev_id].devname);
699 + vma_reader_set_bitmap(rstate, cluster_num, 1);
701 + max_sector = vmar->devinfo[dev_id].size/BDRV_SECTOR_SIZE;
703 + max_sector = G_MAXINT64;
704 + if (cluster_num != vmar->vmstate_clusters) {
705 + error_setg(errp, "found out of order vmstate data");
708 + vmar->vmstate_clusters++;
711 + vmar->clusters_read++;
714 + time_t duration = time(NULL) - vmar->start_time;
715 + int percent = (vmar->clusters_read*100)/vmar->cluster_count;
716 + if (percent != vmar->clusters_read_per) {
717 + printf("progress %d%% (read %zd bytes, duration %zd sec)\n",
718 + percent, vmar->clusters_read*VMA_CLUSTER_SIZE,
721 + vmar->clusters_read_per = percent;
725 + /* try to write whole clusters to speedup restore */
726 + if (mask == 0xffff) {
727 + if ((start + VMA_CLUSTER_SIZE) > extent_size) {
728 + error_setg(errp, "short vma extent - too many blocks");
731 + int64_t sector_num = (cluster_num * VMA_CLUSTER_SIZE) /
733 + int64_t end_sector = sector_num +
734 + VMA_CLUSTER_SIZE/BDRV_SECTOR_SIZE;
736 + if (end_sector > max_sector) {
737 + end_sector = max_sector;
740 + if (end_sector <= sector_num) {
741 + error_setg(errp, "got wrong block address - write bejond end");
745 + int nb_sectors = end_sector - sector_num;
746 + if (restore_write_data(vmar, dev_id, bs, vmstate_fd, buf + start,
747 + sector_num, nb_sectors, errp) < 0) {
751 + start += VMA_CLUSTER_SIZE;
756 + for (j = 0; j < 16; j++) {
757 + int64_t sector_num = (cluster_num*VMA_CLUSTER_SIZE +
758 + j*VMA_BLOCK_SIZE)/BDRV_SECTOR_SIZE;
760 + int64_t end_sector = sector_num +
761 + VMA_BLOCK_SIZE/BDRV_SECTOR_SIZE;
762 + if (end_sector > max_sector) {
763 + end_sector = max_sector;
767 + if ((start + VMA_BLOCK_SIZE) > extent_size) {
768 + error_setg(errp, "short vma extent - too many blocks");
772 + if (end_sector <= sector_num) {
773 + error_setg(errp, "got wrong block address - "
774 + "write bejond end");
778 + int nb_sectors = end_sector - sector_num;
779 + if (restore_write_data(vmar, dev_id, bs, vmstate_fd,
780 + buf + start, sector_num,
781 + nb_sectors, errp) < 0) {
785 + start += VMA_BLOCK_SIZE;
789 + if (rstate->write_zeroes && (end_sector > sector_num)) {
790 + /* Todo: use bdrv_co_write_zeroes (but that need to
791 + * be run inside coroutine?)
793 + int nb_sectors = end_sector - sector_num;
794 + if (restore_write_data(vmar, dev_id, bs, vmstate_fd,
795 + zero_vma_block, sector_num,
796 + nb_sectors, errp) < 0) {
807 + if (start != extent_size) {
808 + error_setg(errp, "vma extent error - missing blocks");
815 +int vma_reader_restore(VmaReader *vmar, int vmstate_fd, bool verbose,
819 + assert(vmar->head_data);
822 + unsigned char buf[VMA_MAX_EXTENT_SIZE];
824 + unsigned char md5sum[16];
825 + VmaHeader *h = (VmaHeader *)vmar->head_data;
827 + vmar->start_time = time(NULL);
830 + int bytes = full_read(vmar->fd, buf + buf_pos, sizeof(buf) - buf_pos);
832 + error_setg(errp, "read failed - %s", strerror(errno));
842 + if (buf_pos < VMA_EXTENT_HEADER_SIZE) {
843 + error_setg(errp, "read short extent (%d bytes)", buf_pos);
847 + VmaExtentHeader *ehead = (VmaExtentHeader *)buf;
849 + /* extract md5sum */
850 + memcpy(md5sum, ehead->md5sum, sizeof(ehead->md5sum));
851 + memset(ehead->md5sum, 0, sizeof(ehead->md5sum));
853 + g_checksum_reset(vmar->md5csum);
854 + g_checksum_update(vmar->md5csum, buf, VMA_EXTENT_HEADER_SIZE);
856 + g_checksum_get_digest(vmar->md5csum, ehead->md5sum, &csize);
858 + if (memcmp(md5sum, ehead->md5sum, 16) != 0) {
859 + error_setg(errp, "wrong vma extent header chechsum");
863 + if (memcmp(h->uuid, ehead->uuid, sizeof(ehead->uuid)) != 0) {
864 + error_setg(errp, "wrong vma extent uuid");
868 + if (ehead->magic != VMA_EXTENT_MAGIC || ehead->reserved1 != 0) {
869 + error_setg(errp, "wrong vma extent header magic");
873 + int block_count = GUINT16_FROM_BE(ehead->block_count);
874 + int extent_size = VMA_EXTENT_HEADER_SIZE + block_count*VMA_BLOCK_SIZE;
876 + if (buf_pos < extent_size) {
877 + error_setg(errp, "short vma extent (%d < %d)", buf_pos,
882 + if (restore_extent(vmar, buf, extent_size, vmstate_fd, verbose,
887 + if (buf_pos > extent_size) {
888 + memmove(buf, buf + extent_size, buf_pos - extent_size);
889 + buf_pos = buf_pos - extent_size;
898 + for (i = 1; i < 256; i++) {
899 + VmaRestoreState *rstate = &vmar->rstate[i];
904 + if (bdrv_flush(rstate->bs) < 0) {
905 + error_setg(errp, "vma bdrv_flush %s failed",
906 + vmar->devinfo[i].devname);
910 + if (vmar->devinfo[i].size &&
911 + (strcmp(vmar->devinfo[i].devname, "vmstate") != 0)) {
912 + assert(rstate->bitmap);
914 + int64_t cluster_num, end;
916 + end = (vmar->devinfo[i].size + VMA_CLUSTER_SIZE - 1) /
919 + for (cluster_num = 0; cluster_num < end; cluster_num++) {
920 + if (!vma_reader_get_bitmap(rstate, cluster_num)) {
921 + error_setg(errp, "detected missing cluster %zd "
922 + "for stream %s", cluster_num,
923 + vmar->devinfo[i].devname);
933 diff --git a/vma-writer.c b/vma-writer.c
935 index 0000000..04a346e
940 + * VMA: Virtual Machine Archive
942 + * Copyright (C) 2012 Proxmox Server Solutions
945 + * Dietmar Maurer (dietmar@proxmox.com)
947 + * This work is licensed under the terms of the GNU GPL, version 2 or later.
948 + * See the COPYING file in the top-level directory.
957 +#include <sys/types.h>
958 +#include <sys/stat.h>
961 +#include <uuid/uuid.h>
963 +#include "qemu-common.h"
965 +#include "block/block.h"
966 +#include "monitor/monitor.h"
970 +#define DPRINTF(fmt, ...)\
971 + do { if (DEBUG_VMA) { printf("vma: " fmt, ## __VA_ARGS__); } } while (0)
973 +#define WRITE_BUFFERS 5
975 +typedef struct VmaAIOCB VmaAIOCB;
977 + unsigned char buffer[VMA_MAX_EXTENT_SIZE];
989 + bool header_written;
992 + /* we always write extents */
993 + unsigned char outbuf[VMA_MAX_EXTENT_SIZE];
994 + int outbuf_pos; /* in bytes */
995 + int outbuf_count; /* in VMA_BLOCKS */
996 + uint64_t outbuf_block_info[VMA_BLOCKS_PER_EXTENT];
998 + VmaAIOCB *aiocbs[WRITE_BUFFERS];
1001 + GChecksum *md5csum;
1002 + CoMutex writer_lock;
1003 + CoMutex flush_lock;
1004 + Coroutine *co_writer;
1006 + /* drive informations */
1007 + VmaStreamInfo stream_info[256];
1008 + guint stream_count;
1010 + guint8 vmstate_stream;
1011 + uint32_t vmstate_clusters;
1013 + /* header blob table */
1014 + char *header_blob_table;
1015 + uint32_t header_blob_table_size;
1016 + uint32_t header_blob_table_pos;
1018 + /* store for config blobs */
1019 + uint32_t config_names[VMA_MAX_CONFIGS]; /* offset into blob_buffer table */
1020 + uint32_t config_data[VMA_MAX_CONFIGS]; /* offset into blob_buffer table */
1021 + uint32_t config_count;
1024 +void vma_writer_set_error(VmaWriter *vmaw, const char *fmt, ...)
1028 + if (vmaw->status < 0) {
1032 + vmaw->status = -1;
1034 + va_start(ap, fmt);
1035 + g_vsnprintf(vmaw->errmsg, sizeof(vmaw->errmsg), fmt, ap);
1038 + DPRINTF("vma_writer_set_error: %s\n", vmaw->errmsg);
1041 +static uint32_t allocate_header_blob(VmaWriter *vmaw, const char *data,
1044 + if (len > 65535) {
1048 + if (!vmaw->header_blob_table ||
1049 + (vmaw->header_blob_table_size <
1050 + (vmaw->header_blob_table_pos + len + 2))) {
1051 + int newsize = vmaw->header_blob_table_size + ((len + 2 + 511)/512)*512;
1053 + vmaw->header_blob_table = g_realloc(vmaw->header_blob_table, newsize);
1054 + memset(vmaw->header_blob_table + vmaw->header_blob_table_size,
1055 + 0, newsize - vmaw->header_blob_table_size);
1056 + vmaw->header_blob_table_size = newsize;
1059 + uint32_t cpos = vmaw->header_blob_table_pos;
1060 + vmaw->header_blob_table[cpos] = len & 255;
1061 + vmaw->header_blob_table[cpos+1] = (len >> 8) & 255;
1062 + memcpy(vmaw->header_blob_table + cpos + 2, data, len);
1063 + vmaw->header_blob_table_pos += len + 2;
1067 +static uint32_t allocate_header_string(VmaWriter *vmaw, const char *str)
1071 + size_t len = strlen(str) + 1;
1073 + return allocate_header_blob(vmaw, str, len);
1076 +int vma_writer_add_config(VmaWriter *vmaw, const char *name, gpointer data,
1080 + assert(!vmaw->header_written);
1081 + assert(vmaw->config_count < VMA_MAX_CONFIGS);
1086 + uint32_t name_ptr = allocate_header_string(vmaw, name);
1091 + uint32_t data_ptr = allocate_header_blob(vmaw, data, len);
1096 + vmaw->config_names[vmaw->config_count] = name_ptr;
1097 + vmaw->config_data[vmaw->config_count] = data_ptr;
1099 + vmaw->config_count++;
1104 +int vma_writer_register_stream(VmaWriter *vmaw, const char *devname,
1109 + assert(!vmaw->status);
1111 + if (vmaw->header_written) {
1112 + vma_writer_set_error(vmaw, "vma_writer_register_stream: header "
1113 + "already written");
1117 + guint n = vmaw->stream_count + 1;
1119 + /* we can have dev_ids form 1 to 255 (0 reserved)
1120 + * 255(-1) reseverd for safety
1123 + vma_writer_set_error(vmaw, "vma_writer_register_stream: "
1124 + "too many drives");
1129 + vma_writer_set_error(vmaw, "vma_writer_register_stream: "
1130 + "got strange size %zd", size);
1134 + DPRINTF("vma_writer_register_stream %s %zu %d\n", devname, size, n);
1136 + vmaw->stream_info[n].devname = g_strdup(devname);
1137 + vmaw->stream_info[n].size = size;
1139 + vmaw->stream_info[n].cluster_count = (size + VMA_CLUSTER_SIZE - 1) /
1142 + vmaw->stream_count = n;
1144 + if (strcmp(devname, "vmstate") == 0) {
1145 + vmaw->vmstate_stream = n;
1151 +static void vma_co_continue_write(void *opaque)
1153 + VmaWriter *vmaw = opaque;
1155 + DPRINTF("vma_co_continue_write\n");
1156 + qemu_coroutine_enter(vmaw->co_writer, NULL);
1159 +static int vma_co_write_finished(void *opaque)
1161 + VmaWriter *vmaw = opaque;
1163 + return (vmaw->co_writer != 0);
1166 +static ssize_t coroutine_fn
1167 +vma_co_write(VmaWriter *vmaw, const void *buf, size_t bytes)
1172 + /* atomic writes (we cannot interleave writes) */
1173 + qemu_co_mutex_lock(&vmaw->writer_lock);
1175 + DPRINTF("vma_co_write enter %zd\n", bytes);
1177 + assert(vmaw->co_writer == NULL);
1179 + vmaw->co_writer = qemu_coroutine_self();
1181 + qemu_aio_set_fd_handler(vmaw->fd, NULL, vma_co_continue_write,
1182 + vma_co_write_finished, vmaw);
1184 + DPRINTF("vma_co_write wait until writable\n");
1185 + qemu_coroutine_yield();
1186 + DPRINTF("vma_co_write starting %zd\n", bytes);
1188 + while (done < bytes) {
1189 + ret = write(vmaw->fd, buf + done, bytes - done);
1192 + DPRINTF("vma_co_write written %zd %zd\n", done, ret);
1193 + } else if (ret < 0) {
1194 + if (errno == EAGAIN || errno == EWOULDBLOCK) {
1195 + DPRINTF("vma_co_write yield %zd\n", done);
1196 + qemu_coroutine_yield();
1197 + DPRINTF("vma_co_write restart %zd\n", done);
1199 + vma_writer_set_error(vmaw, "vma_co_write write error - %s",
1201 + done = -1; /* always return failure for partial writes */
1204 + } else if (ret == 0) {
1205 + /* should not happen - simply try again */
1209 + qemu_aio_set_fd_handler(vmaw->fd, NULL, NULL, NULL, NULL);
1211 + vmaw->co_writer = NULL;
1213 + qemu_co_mutex_unlock(&vmaw->writer_lock);
1215 + DPRINTF("vma_co_write leave %zd\n", done);
1219 +static void coroutine_fn vma_co_writer_task(void *opaque)
1221 + VmaAIOCB *cb = opaque;
1223 + DPRINTF("vma_co_writer_task start\n");
1225 + int64_t done = vma_co_write(cb->vmaw, cb->buffer, cb->bytes);
1226 + DPRINTF("vma_co_writer_task write done %zd\n", done);
1228 + if (done != cb->bytes) {
1229 + DPRINTF("vma_co_writer_task failed write %zd %zd", cb->bytes, done);
1230 + vma_writer_set_error(cb->vmaw, "vma_co_writer_task failed write %zd",
1236 + qemu_co_queue_next(&cb->vmaw->wqueue);
1238 + DPRINTF("vma_co_writer_task end\n");
1241 +static void coroutine_fn vma_queue_flush(VmaWriter *vmaw)
1243 + DPRINTF("vma_queue_flush enter\n");
1249 + VmaAIOCB *cb = NULL;
1250 + for (i = 0; i < WRITE_BUFFERS; i++) {
1251 + if (vmaw->aiocbs[i]->bytes) {
1252 + cb = vmaw->aiocbs[i];
1253 + DPRINTF("FOUND USED AIO BUFFER %d %zd\n", i,
1254 + vmaw->aiocbs[i]->bytes);
1261 + qemu_co_queue_wait(&vmaw->wqueue);
1264 + DPRINTF("vma_queue_flush leave\n");
1268 + * NOTE: pipe buffer size in only 4096 bytes on linux (see 'ulimit -a')
1269 + * So we need to create a coroutione to allow 'parallel' execution.
1271 +static ssize_t coroutine_fn
1272 +vma_queue_write(VmaWriter *vmaw, const void *buf, size_t bytes)
1274 + DPRINTF("vma_queue_write enter %zd\n", bytes);
1278 + assert(bytes <= VMA_MAX_EXTENT_SIZE);
1280 + VmaAIOCB *cb = NULL;
1283 + for (i = 0; i < WRITE_BUFFERS; i++) {
1284 + if (!vmaw->aiocbs[i]->bytes) {
1285 + cb = vmaw->aiocbs[i];
1290 + qemu_co_queue_wait(&vmaw->wqueue);
1294 + memcpy(cb->buffer, buf, bytes);
1295 + cb->bytes = bytes;
1298 + DPRINTF("vma_queue_write start %zd\n", bytes);
1299 + cb->co = qemu_coroutine_create(vma_co_writer_task);
1300 + qemu_coroutine_enter(cb->co, cb);
1302 + DPRINTF("vma_queue_write leave\n");
1307 +VmaWriter *vma_writer_create(const char *filename, uuid_t uuid, Error **errp)
1311 + assert(sizeof(VmaHeader) == (4096 + 8192));
1312 + assert(sizeof(VmaExtentHeader) == 512);
1314 + VmaWriter *vmaw = g_new0(VmaWriter, 1);
1317 + vmaw->md5csum = g_checksum_new(G_CHECKSUM_MD5);
1318 + if (!vmaw->md5csum) {
1319 + error_setg(errp, "can't allocate cmsum\n");
1323 + if (strstart(filename, "exec:", &p)) {
1324 + vmaw->cmd = popen(p, "w");
1325 + if (vmaw->cmd == NULL) {
1326 + error_setg(errp, "can't popen command '%s' - %s\n", p,
1330 + vmaw->fd = fileno(vmaw->cmd);
1332 + /* try to use O_NONBLOCK and O_DIRECT */
1333 + fcntl(vmaw->fd, F_SETFL, fcntl(vmaw->fd, F_GETFL)|O_NONBLOCK);
1334 + fcntl(vmaw->fd, F_SETFL, fcntl(vmaw->fd, F_GETFL)|O_DIRECT);
1339 + const char *tmp_id_str;
1341 + if ((stat(filename, &st) == 0) && S_ISFIFO(st.st_mode)) {
1342 + oflags = O_NONBLOCK|O_DIRECT|O_WRONLY;
1343 + vmaw->fd = qemu_open(filename, oflags, 0644);
1344 + } else if (strstart(filename, "/dev/fdset/", &tmp_id_str)) {
1345 + oflags = O_NONBLOCK|O_DIRECT|O_WRONLY;
1346 + vmaw->fd = qemu_open(filename, oflags, 0644);
1347 + } else if (strstart(filename, "/dev/fdname/", &tmp_id_str)) {
1348 + vmaw->fd = monitor_get_fd(cur_mon, tmp_id_str, errp);
1349 + if (vmaw->fd < 0) {
1352 + /* try to use O_NONBLOCK and O_DIRECT */
1353 + fcntl(vmaw->fd, F_SETFL, fcntl(vmaw->fd, F_GETFL)|O_NONBLOCK);
1354 + fcntl(vmaw->fd, F_SETFL, fcntl(vmaw->fd, F_GETFL)|O_DIRECT);
1356 + oflags = O_NONBLOCK|O_DIRECT|O_WRONLY|O_CREAT|O_EXCL;
1357 + vmaw->fd = qemu_open(filename, oflags, 0644);
1360 + if (vmaw->fd < 0) {
1361 + error_setg(errp, "can't open file %s - %s\n", filename,
1367 + /* we use O_DIRECT, so we need to align IO buffers */
1369 + for (i = 0; i < WRITE_BUFFERS; i++) {
1370 + vmaw->aiocbs[i] = qemu_memalign(512, sizeof(VmaAIOCB));
1371 + memset(vmaw->aiocbs[i], 0, sizeof(VmaAIOCB));
1374 + vmaw->outbuf_count = 0;
1375 + vmaw->outbuf_pos = VMA_EXTENT_HEADER_SIZE;
1377 + vmaw->header_blob_table_pos = 1; /* start at pos 1 */
1379 + qemu_co_mutex_init(&vmaw->writer_lock);
1380 + qemu_co_mutex_init(&vmaw->flush_lock);
1381 + qemu_co_queue_init(&vmaw->wqueue);
1383 + uuid_copy(vmaw->uuid, uuid);
1390 + pclose(vmaw->cmd);
1391 + } else if (vmaw->fd >= 0) {
1395 + if (vmaw->md5csum) {
1396 + g_checksum_free(vmaw->md5csum);
1405 +static int coroutine_fn vma_write_header(VmaWriter *vmaw)
1408 + int header_clusters = 8;
1409 + char buf[65536*header_clusters];
1410 + VmaHeader *head = (VmaHeader *)buf;
1414 + DPRINTF("VMA WRITE HEADER\n");
1416 + if (vmaw->status < 0) {
1417 + return vmaw->status;
1420 + memset(buf, 0, sizeof(buf));
1422 + head->magic = VMA_MAGIC;
1423 + head->version = GUINT32_TO_BE(1); /* v1 */
1424 + memcpy(head->uuid, vmaw->uuid, 16);
1426 + time_t ctime = time(NULL);
1427 + head->ctime = GUINT64_TO_BE(ctime);
1429 + if (!vmaw->stream_count) {
1433 + for (i = 0; i < VMA_MAX_CONFIGS; i++) {
1434 + head->config_names[i] = GUINT32_TO_BE(vmaw->config_names[i]);
1435 + head->config_data[i] = GUINT32_TO_BE(vmaw->config_data[i]);
1438 + /* 32 bytes per device (12 used currently) = 8192 bytes max */
1439 + for (i = 1; i <= 254; i++) {
1440 + VmaStreamInfo *si = &vmaw->stream_info[i];
1442 + assert(si->devname);
1443 + uint32_t devname_ptr = allocate_header_string(vmaw, si->devname);
1444 + if (!devname_ptr) {
1447 + head->dev_info[i].devname_ptr = GUINT32_TO_BE(devname_ptr);
1448 + head->dev_info[i].size = GUINT64_TO_BE(si->size);
1452 + uint32_t header_size = sizeof(VmaHeader) + vmaw->header_blob_table_size;
1453 + head->header_size = GUINT32_TO_BE(header_size);
1455 + if (header_size > sizeof(buf)) {
1456 + return -1; /* just to be sure */
1459 + uint32_t blob_buffer_offset = sizeof(VmaHeader);
1460 + memcpy(buf + blob_buffer_offset, vmaw->header_blob_table,
1461 + vmaw->header_blob_table_size);
1462 + head->blob_buffer_offset = GUINT32_TO_BE(blob_buffer_offset);
1463 + head->blob_buffer_size = GUINT32_TO_BE(vmaw->header_blob_table_pos);
1465 + g_checksum_reset(vmaw->md5csum);
1466 + g_checksum_update(vmaw->md5csum, (const guchar *)buf, header_size);
1468 + g_checksum_get_digest(vmaw->md5csum, (guint8 *)(head->md5sum), &csize);
1470 + return vma_queue_write(vmaw, buf, header_size);
1473 +static int coroutine_fn vma_writer_flush(VmaWriter *vmaw)
1480 + if (vmaw->status < 0) {
1481 + return vmaw->status;
1484 + if (!vmaw->header_written) {
1485 + vmaw->header_written = true;
1486 + ret = vma_write_header(vmaw);
1488 + vma_writer_set_error(vmaw, "vma_writer_flush: write header failed");
1493 + DPRINTF("VMA WRITE FLUSH %d %d\n", vmaw->outbuf_count, vmaw->outbuf_pos);
1496 + VmaExtentHeader *ehead = (VmaExtentHeader *)vmaw->outbuf;
1498 + ehead->magic = VMA_EXTENT_MAGIC;
1499 + ehead->reserved1 = 0;
1501 + for (i = 0; i < VMA_BLOCKS_PER_EXTENT; i++) {
1502 + ehead->blockinfo[i] = GUINT64_TO_BE(vmaw->outbuf_block_info[i]);
1505 + guint16 block_count = (vmaw->outbuf_pos - VMA_EXTENT_HEADER_SIZE) /
1508 + ehead->block_count = GUINT16_TO_BE(block_count);
1510 + memcpy(ehead->uuid, vmaw->uuid, sizeof(ehead->uuid));
1511 + memset(ehead->md5sum, 0, sizeof(ehead->md5sum));
1513 + g_checksum_reset(vmaw->md5csum);
1514 + g_checksum_update(vmaw->md5csum, vmaw->outbuf, VMA_EXTENT_HEADER_SIZE);
1516 + g_checksum_get_digest(vmaw->md5csum, ehead->md5sum, &csize);
1518 + int bytes = vmaw->outbuf_pos;
1519 + ret = vma_queue_write(vmaw, vmaw->outbuf, bytes);
1520 + if (ret != bytes) {
1521 + vma_writer_set_error(vmaw, "vma_writer_flush: failed write");
1524 + vmaw->outbuf_count = 0;
1525 + vmaw->outbuf_pos = VMA_EXTENT_HEADER_SIZE;
1527 + for (i = 0; i < VMA_BLOCKS_PER_EXTENT; i++) {
1528 + vmaw->outbuf_block_info[i] = 0;
1531 + return vmaw->status;
1534 +static int vma_count_open_streams(VmaWriter *vmaw)
1536 + g_assert(vmaw != NULL);
1539 + int open_drives = 0;
1540 + for (i = 0; i <= 255; i++) {
1541 + if (vmaw->stream_info[i].size && !vmaw->stream_info[i].finished) {
1546 + return open_drives;
1550 + * all jobs should call this when there is no more data
1551 + * Returns: number of remaining stream (0 ==> finished)
1554 +vma_writer_close_stream(VmaWriter *vmaw, uint8_t dev_id)
1556 + g_assert(vmaw != NULL);
1558 + DPRINTF("vma_writer_set_status %d\n", dev_id);
1559 + if (!vmaw->stream_info[dev_id].size) {
1560 + vma_writer_set_error(vmaw, "vma_writer_close_stream: "
1561 + "no such stream %d", dev_id);
1564 + if (vmaw->stream_info[dev_id].finished) {
1565 + vma_writer_set_error(vmaw, "vma_writer_close_stream: "
1566 + "stream already closed %d", dev_id);
1570 + vmaw->stream_info[dev_id].finished = true;
1572 + int open_drives = vma_count_open_streams(vmaw);
1574 + if (open_drives <= 0) {
1575 + DPRINTF("vma_writer_set_status all drives completed\n");
1576 + qemu_co_mutex_lock(&vmaw->flush_lock);
1577 + int ret = vma_writer_flush(vmaw);
1578 + qemu_co_mutex_unlock(&vmaw->flush_lock);
1580 + vma_writer_set_error(vmaw, "vma_writer_close_stream: flush failed");
1584 + return open_drives;
1587 +int vma_writer_get_status(VmaWriter *vmaw, VmaStatus *status)
1591 + g_assert(vmaw != NULL);
1594 + status->status = vmaw->status;
1595 + g_strlcpy(status->errmsg, vmaw->errmsg, sizeof(status->errmsg));
1596 + for (i = 0; i <= 255; i++) {
1597 + status->stream_info[i] = vmaw->stream_info[i];
1600 + uuid_unparse_lower(vmaw->uuid, status->uuid_str);
1603 + status->closed = vmaw->closed;
1605 + return vmaw->status;
1608 +static int vma_writer_get_buffer(VmaWriter *vmaw)
1612 + qemu_co_mutex_lock(&vmaw->flush_lock);
1614 + /* wait until buffer is available */
1615 + while (vmaw->outbuf_count >= (VMA_BLOCKS_PER_EXTENT - 1)) {
1616 + ret = vma_writer_flush(vmaw);
1618 + vma_writer_set_error(vmaw, "vma_writer_get_buffer: flush failed");
1623 + qemu_co_mutex_unlock(&vmaw->flush_lock);
1629 +int64_t coroutine_fn
1630 +vma_writer_write(VmaWriter *vmaw, uint8_t dev_id, int64_t cluster_num,
1631 + unsigned char *buf, size_t *zero_bytes)
1633 + g_assert(vmaw != NULL);
1634 + g_assert(zero_bytes != NULL);
1638 + if (vmaw->status < 0) {
1639 + return vmaw->status;
1642 + if (!dev_id || !vmaw->stream_info[dev_id].size) {
1643 + vma_writer_set_error(vmaw, "vma_writer_write: "
1644 + "no such stream %d", dev_id);
1648 + if (vmaw->stream_info[dev_id].finished) {
1649 + vma_writer_set_error(vmaw, "vma_writer_write: "
1650 + "stream already closed %d", dev_id);
1655 + if (cluster_num >= (((uint64_t)1)<<32)) {
1656 + vma_writer_set_error(vmaw, "vma_writer_write: "
1657 + "cluster number out of range");
1661 + if (dev_id == vmaw->vmstate_stream) {
1662 + if (cluster_num != vmaw->vmstate_clusters) {
1663 + vma_writer_set_error(vmaw, "vma_writer_write: "
1664 + "non sequential vmstate write");
1666 + vmaw->vmstate_clusters++;
1667 + } else if (cluster_num >= vmaw->stream_info[dev_id].cluster_count) {
1668 + vma_writer_set_error(vmaw, "vma_writer_write: cluster number too big");
1672 + /* wait until buffer is available */
1673 + if (vma_writer_get_buffer(vmaw) < 0) {
1674 + vma_writer_set_error(vmaw, "vma_writer_write: "
1675 + "vma_writer_get_buffer failed");
1679 + DPRINTF("VMA WRITE %d %zd\n", dev_id, cluster_num);
1683 + uint16_t mask = 0;
1684 + for (i = 0; i < 16; i++) {
1685 + unsigned char *vmablock = buf + (i*VMA_BLOCK_SIZE);
1686 + if (!buffer_is_zero(vmablock, VMA_BLOCK_SIZE)) {
1688 + memcpy(vmaw->outbuf + vmaw->outbuf_pos, vmablock, VMA_BLOCK_SIZE);
1689 + vmaw->outbuf_pos += VMA_BLOCK_SIZE;
1691 + DPRINTF("VMA WRITE %zd ZERO BLOCK %d\n", cluster_num, i);
1692 + vmaw->stream_info[dev_id].zero_bytes += VMA_BLOCK_SIZE;
1693 + *zero_bytes += VMA_BLOCK_SIZE;
1699 + uint64_t block_info = ((uint64_t)mask) << (32+16);
1700 + block_info |= ((uint64_t)dev_id) << 32;
1701 + block_info |= (cluster_num & 0xffffffff);
1702 + vmaw->outbuf_block_info[vmaw->outbuf_count] = block_info;
1704 + DPRINTF("VMA WRITE MASK %zd %zx\n", cluster_num, block_info);
1706 + vmaw->outbuf_count++;
1708 + /** NOTE: We allways write whole clusters, but we correctly set
1709 + * transferred bytes. So transferred == size when when everything
1712 + size_t transferred = VMA_CLUSTER_SIZE;
1714 + if (dev_id != vmaw->vmstate_stream) {
1715 + uint64_t last = (cluster_num + 1) * VMA_CLUSTER_SIZE;
1716 + if (last > vmaw->stream_info[dev_id].size) {
1717 + uint64_t diff = last - vmaw->stream_info[dev_id].size;
1718 + if (diff >= VMA_CLUSTER_SIZE) {
1719 + vma_writer_set_error(vmaw, "vma_writer_write: "
1720 + "read after last cluster");
1723 + transferred -= diff;
1727 + vmaw->stream_info[dev_id].transferred += transferred;
1729 + return transferred;
1732 +int vma_writer_close(VmaWriter *vmaw, Error **errp)
1734 + g_assert(vmaw != NULL);
1738 + vma_queue_flush(vmaw);
1740 + /* this should not happen - just to be sure */
1741 + while (!qemu_co_queue_empty(&vmaw->wqueue)) {
1742 + DPRINTF("vma_writer_close wait\n");
1743 + co_sleep_ns(rt_clock, 1000000);
1747 + if (pclose(vmaw->cmd) < 0) {
1748 + vma_writer_set_error(vmaw, "vma_writer_close: "
1749 + "pclose failed - %s", strerror(errno));
1752 + if (close(vmaw->fd) < 0) {
1753 + vma_writer_set_error(vmaw, "vma_writer_close: "
1754 + "close failed - %s", strerror(errno));
1758 + for (i = 0; i <= 255; i++) {
1759 + VmaStreamInfo *si = &vmaw->stream_info[i];
1761 + if (!si->finished) {
1762 + vma_writer_set_error(vmaw, "vma_writer_close: "
1763 + "detected open stream '%s'", si->devname);
1764 + } else if ((si->transferred != si->size) &&
1765 + (i != vmaw->vmstate_stream)) {
1766 + vma_writer_set_error(vmaw, "vma_writer_close: "
1767 + "incomplete stream '%s' (%zd != %zd)",
1768 + si->devname, si->transferred, si->size);
1773 + for (i = 0; i <= 255; i++) {
1774 + vmaw->stream_info[i].finished = 1; /* mark as closed */
1779 + if (vmaw->status < 0 && *errp == NULL) {
1780 + error_setg(errp, "%s", vmaw->errmsg);
1783 + return vmaw->status;
1786 +void vma_writer_destroy(VmaWriter *vmaw)
1792 + for (i = 0; i <= 255; i++) {
1793 + if (vmaw->stream_info[i].devname) {
1794 + g_free(vmaw->stream_info[i].devname);
1798 + if (vmaw->md5csum) {
1799 + g_checksum_free(vmaw->md5csum);
1802 + for (i = 0; i < WRITE_BUFFERS; i++) {
1803 + free(vmaw->aiocbs[i]);
1809 +/* backup driver plugin */
1811 +static int vma_dump_cb(void *opaque, uint8_t dev_id, int64_t cluster_num,
1812 + unsigned char *buf, size_t *zero_bytes)
1814 + VmaWriter *vmaw = opaque;
1816 + return vma_writer_write(vmaw, dev_id, cluster_num, buf, zero_bytes);
1819 +static int vma_close_cb(void *opaque, Error **errp)
1821 + VmaWriter *vmaw = opaque;
1823 + int res = vma_writer_close(vmaw, errp);
1824 + vma_writer_destroy(vmaw);
1829 +static int vma_complete_cb(void *opaque, uint8_t dev_id, int ret)
1831 + VmaWriter *vmaw = opaque;
1834 + vma_writer_set_error(vmaw, "backup_complete_cb %d", ret);
1837 + return vma_writer_close_stream(vmaw, dev_id);
1840 +static int vma_register_stream_cb(void *opaque, const char *devname,
1843 + VmaWriter *vmaw = opaque;
1845 + return vma_writer_register_stream(vmaw, devname, size);
1848 +static int vma_register_config_cb(void *opaque, const char *name,
1849 + gpointer data, size_t data_len)
1851 + VmaWriter *vmaw = opaque;
1853 + return vma_writer_add_config(vmaw, name, data, data_len);
1856 +static void *vma_open_cb(const char *filename, uuid_t uuid, Error **errp)
1858 + return vma_writer_create(filename, uuid, errp);
1861 +const BackupDriver backup_vma_driver = {
1863 + .open_cb = vma_open_cb,
1864 + .close_cb = vma_close_cb,
1865 + .register_config_cb = vma_register_config_cb,
1866 + .register_stream_cb = vma_register_stream_cb,
1867 + .dump_cb = vma_dump_cb,
1868 + .complete_cb = vma_complete_cb,
1871 diff --git a/vma.c b/vma.c
1872 new file mode 100644
1873 index 0000000..64da334
1878 + * VMA: Virtual Machine Archive
1880 + * Copyright (C) 2012 Proxmox Server Solutions
1883 + * Dietmar Maurer (dietmar@proxmox.com)
1885 + * This work is licensed under the terms of the GNU GPL, version 2 or later.
1886 + * See the COPYING file in the top-level directory.
1892 +#include <unistd.h>
1894 +#include <string.h>
1895 +#include <sys/types.h>
1896 +#include <sys/stat.h>
1900 +#include "qemu-common.h"
1901 +#include "qemu/error-report.h"
1903 +#include "block/block.h"
1905 +static void help(void)
1907 + const char *help_msg =
1908 + "usage: vma command [command options]\n"
1910 + "vma list <filename>\n"
1911 + "vma create <filename> [-c config] <archive> pathname ...\n"
1912 + "vma extract <filename> [-r] <targetdir>\n"
1915 + printf("%s", help_msg);
1919 +static const char *extract_devname(const char *path, char **devname, int index)
1923 + const char *sep = strchr(path, '=');
1926 + *devname = g_strndup(path, sep - path);
1930 + *devname = g_strdup_printf("disk%d", index);
1939 +static void print_content(VmaReader *vmar)
1943 + VmaHeader *head = vma_reader_get_header(vmar);
1945 + GList *l = vma_reader_get_config_data(vmar);
1946 + while (l && l->data) {
1947 + VmaConfigData *cdata = (VmaConfigData *)l->data;
1948 + l = g_list_next(l);
1949 + printf("CFG: size: %d name: %s\n", cdata->len, cdata->name);
1953 + VmaDeviceInfo *di;
1954 + for (i = 1; i < 255; i++) {
1955 + di = vma_reader_get_device_info(vmar, i);
1957 + if (strcmp(di->devname, "vmstate") == 0) {
1958 + printf("VMSTATE: dev_id=%d memory: %zd\n", i, di->size);
1960 + printf("DEV: dev_id=%d size: %zd devname: %s\n",
1961 + i, di->size, di->devname);
1965 + /* ctime is the last entry we print */
1966 + printf("CTIME: %s", ctime(&head->ctime));
1970 +static int list_content(int argc, char **argv)
1973 + const char *filename;
1976 + c = getopt(argc, argv, "h");
1986 + g_assert_not_reached();
1990 + /* Get the filename */
1991 + if ((optind + 1) != argc) {
1994 + filename = argv[optind++];
1996 + Error *errp = NULL;
1997 + VmaReader *vmar = vma_reader_create(filename, &errp);
2000 + g_error("%s", error_get_pretty(errp));
2003 + print_content(vmar);
2005 + vma_reader_destroy(vmar);
2010 +typedef struct RestoreMap {
2016 +static int extract_content(int argc, char **argv)
2020 + const char *filename;
2021 + const char *dirname;
2022 + const char *readmap = NULL;
2025 + c = getopt(argc, argv, "hvr:");
2045 + /* Get the filename */
2046 + if ((optind + 2) != argc) {
2049 + filename = argv[optind++];
2050 + dirname = argv[optind++];
2052 + Error *errp = NULL;
2053 + VmaReader *vmar = vma_reader_create(filename, &errp);
2056 + g_error("%s", error_get_pretty(errp));
2059 + if (mkdir(dirname, 0777) < 0) {
2060 + g_error("unable to create target directory %s - %s",
2061 + dirname, strerror(errno));
2064 + GList *l = vma_reader_get_config_data(vmar);
2065 + while (l && l->data) {
2066 + VmaConfigData *cdata = (VmaConfigData *)l->data;
2067 + l = g_list_next(l);
2068 + char *cfgfn = g_strdup_printf("%s/%s", dirname, cdata->name);
2069 + GError *err = NULL;
2070 + if (!g_file_set_contents(cfgfn, (gchar *)cdata->data, cdata->len,
2072 + g_error("unable to write file: %s", err->message);
2076 + GHashTable *devmap = g_hash_table_new(g_str_hash, g_str_equal);
2079 + print_content(vmar);
2081 + FILE *map = fopen(readmap, "r");
2083 + g_error("unable to open fifo %s - %s", readmap, strerror(errno));
2088 + char *line = fgets(inbuf, sizeof(inbuf), map);
2089 + if (!line || line[0] == '\0' || !strcmp(line, "done\n")) {
2092 + int len = strlen(line);
2093 + if (line[len - 1] == '\n') {
2094 + line[len - 1] = '\0';
2102 + if (line[0] == '0' && line[1] == ':') {
2104 + write_zero = false;
2105 + } else if (line[0] == '1' && line[1] == ':') {
2107 + write_zero = true;
2109 + g_error("read map failed - parse error ('%s')", inbuf);
2112 + char *devname = NULL;
2113 + path = extract_devname(path, &devname, -1);
2115 + g_error("read map failed - no dev name specified ('%s')",
2119 + RestoreMap *map = g_new0(RestoreMap, 1);
2120 + map->devname = g_strdup(devname);
2121 + map->path = g_strdup(path);
2122 + map->write_zero = write_zero;
2124 + g_hash_table_insert(devmap, map->devname, map);
2130 + int vmstate_fd = -1;
2131 + guint8 vmstate_stream = 0;
2133 + for (i = 1; i < 255; i++) {
2134 + VmaDeviceInfo *di = vma_reader_get_device_info(vmar, i);
2135 + if (di && (strcmp(di->devname, "vmstate") == 0)) {
2136 + vmstate_stream = i;
2137 + char *statefn = g_strdup_printf("%s/vmstate.bin", dirname);
2138 + vmstate_fd = open(statefn, O_WRONLY|O_CREAT|O_EXCL, 0644);
2139 + if (vmstate_fd < 0) {
2140 + g_error("create vmstate file '%s' failed - %s", statefn,
2145 + char *devfn = NULL;
2146 + int flags = BDRV_O_RDWR|BDRV_O_CACHE_WB;
2147 + bool write_zero = true;
2151 + map = (RestoreMap *)g_hash_table_lookup(devmap, di->devname);
2152 + if (map == NULL) {
2153 + g_error("no device name mapping for %s", di->devname);
2155 + devfn = map->path;
2156 + write_zero = map->write_zero;
2158 + devfn = g_strdup_printf("%s/tmp-disk-%s.raw",
2159 + dirname, di->devname);
2160 + printf("DEVINFO %s %zd\n", devfn, di->size);
2162 + bdrv_img_create(devfn, "raw", NULL, NULL, NULL, di->size,
2164 + if (error_is_set(&errp)) {
2165 + g_error("can't create file %s: %s", devfn,
2166 + error_get_pretty(errp));
2169 + /* Note: we created an empty file above, so there is no
2170 + * need to write zeroes (so we generate a sparse file)
2172 + write_zero = false;
2175 + BlockDriverState *bs = bdrv_new(di->devname);
2176 + if (bdrv_open(bs, devfn, flags, NULL)) {
2177 + g_error("can't open file %s", devfn);
2179 + if (vma_reader_register_bs(vmar, i, bs, write_zero, &errp) < 0) {
2180 + g_error("%s", error_get_pretty(errp));
2189 + if (vma_reader_restore(vmar, vmstate_fd, verbose, &errp) < 0) {
2190 + g_error("restore failed - %s", error_get_pretty(errp));
2194 + for (i = 1; i < 255; i++) {
2195 + VmaDeviceInfo *di = vma_reader_get_device_info(vmar, i);
2196 + if (di && (i != vmstate_stream)) {
2197 + char *tmpfn = g_strdup_printf("%s/tmp-disk-%s.raw",
2198 + dirname, di->devname);
2199 + char *fn = g_strdup_printf("%s/disk-%s.raw",
2200 + dirname, di->devname);
2201 + if (rename(tmpfn, fn) != 0) {
2202 + g_error("rename %s to %s failed - %s",
2203 + tmpfn, fn, strerror(errno));
2209 + vma_reader_destroy(vmar);
2216 +typedef struct BackupCB {
2221 +static int backup_dump_cb(void *opaque, BlockDriverState *bs,
2222 + int64_t cluster_num, unsigned char *buf)
2224 + BackupCB *bcb = opaque;
2226 + if (vma_writer_write(bcb->vmaw, bcb->dev_id, cluster_num, buf, &zb) < 0) {
2227 + g_warning("backup_dump_cb vma_writer_write failed");
2234 +static void backup_complete_cb(void *opaque, int ret)
2236 + BackupCB *bcb = opaque;
2239 + vma_writer_set_error(bcb->vmaw, "backup_complete_cb %d", ret);
2242 + if (vma_writer_close_stream(bcb->vmaw, bcb->dev_id) <= 0) {
2243 + Error *err = NULL;
2244 + if (vma_writer_close(bcb->vmaw, &err) != 0) {
2245 + g_warning("vma_writer_close failed %s", error_get_pretty(err));
2250 +static int create_archive(int argc, char **argv)
2254 + const char *archivename;
2255 + GList *config_files = NULL;
2258 + c = getopt(argc, argv, "hvc:");
2268 + config_files = g_list_append(config_files, optarg);
2274 + g_assert_not_reached();
2279 + /* make sure we have archive name and at least one path */
2280 + if ((optind + 2) > argc) {
2284 + archivename = argv[optind++];
2287 + uuid_generate(uuid);
2289 + Error *local_err = NULL;
2290 + VmaWriter *vmaw = vma_writer_create(archivename, uuid, &local_err);
2292 + if (vmaw == NULL) {
2293 + g_error("%s", error_get_pretty(local_err));
2296 + GList *l = config_files;
2297 + while (l && l->data) {
2298 + char *name = l->data;
2299 + char *cdata = NULL;
2301 + GError *err = NULL;
2302 + if (!g_file_get_contents(name, &cdata, &clen, &err)) {
2303 + unlink(archivename);
2304 + g_error("Unable to read file: %s", err->message);
2307 + if (vma_writer_add_config(vmaw, name, cdata, clen) != 0) {
2308 + unlink(archivename);
2309 + g_error("Unable to append config data %s (len = %zd)",
2312 + l = g_list_next(l);
2316 + while (optind < argc) {
2317 + const char *path = argv[optind++];
2318 + char *devname = NULL;
2319 + path = extract_devname(path, &devname, ind++);
2321 + BlockDriver *drv = NULL;
2322 + BlockDriverState *bs = bdrv_new(devname);
2324 + res = bdrv_open(bs, path, BDRV_O_CACHE_WB , drv);
2326 + unlink(archivename);
2327 + g_error("bdrv_open '%s' failed", path);
2329 + int64_t size = bdrv_getlength(bs);
2330 + int dev_id = vma_writer_register_stream(vmaw, devname, size);
2331 + if (dev_id <= 0) {
2332 + unlink(archivename);
2333 + g_error("vma_writer_register_stream '%s' failed", devname);
2336 + BackupCB *bcb = g_new0(BackupCB, 1);
2338 + bcb->dev_id = dev_id;
2340 + if (backup_job_create(bs, backup_dump_cb, backup_complete_cb,
2342 + unlink(archivename);
2343 + g_error("backup_job_start failed");
2345 + backup_job_start(bs, false);
2349 + VmaStatus vmastat;
2351 + int last_percent = -1;
2354 + main_loop_wait(false);
2355 + vma_writer_get_status(vmaw, &vmastat);
2359 + uint64_t total = 0;
2360 + uint64_t transferred = 0;
2361 + uint64_t zero_bytes = 0;
2364 + for (i = 0; i < 256; i++) {
2365 + if (vmastat.stream_info[i].size) {
2366 + total += vmastat.stream_info[i].size;
2367 + transferred += vmastat.stream_info[i].transferred;
2368 + zero_bytes += vmastat.stream_info[i].zero_bytes;
2371 + percent = (transferred*100)/total;
2372 + if (percent != last_percent) {
2373 + printf("progress %d%% %zd/%zd %zd\n", percent,
2374 + transferred, total, zero_bytes);
2376 + last_percent = percent;
2380 + if (vmastat.closed) {
2387 + vma_writer_get_status(vmaw, &vmastat);
2390 + for (i = 0; i < 256; i++) {
2391 + VmaStreamInfo *si = &vmastat.stream_info[i];
2393 + printf("image %s: size=%zd zeros=%zd saved=%zd\n", si->devname,
2394 + si->size, si->zero_bytes, si->size - si->zero_bytes);
2399 + if (vmastat.status < 0) {
2400 + unlink(archivename);
2401 + g_error("creating vma archive failed");
2407 +int main(int argc, char **argv)
2409 + const char *cmdname;
2411 + error_set_progname(argv[0]);
2413 + qemu_init_main_loop();
2421 + cmdname = argv[1];
2425 + if (!strcmp(cmdname, "list")) {
2426 + return list_content(argc, argv);
2427 + } else if (!strcmp(cmdname, "create")) {
2428 + return create_archive(argc, argv);
2429 + } else if (!strcmp(cmdname, "extract")) {
2430 + return extract_content(argc, argv);
2436 diff --git a/vma.h b/vma.h
2437 new file mode 100644
2438 index 0000000..76d0dc8
2443 + * VMA: Virtual Machine Archive
2445 + * Copyright (C) Proxmox Server Solutions
2448 + * Dietmar Maurer (dietmar@proxmox.com)
2450 + * This work is licensed under the terms of the GNU GPL, version 2 or later.
2451 + * See the COPYING file in the top-level directory.
2455 +#ifndef BACKUP_VMA_H
2456 +#define BACKUP_VMA_H
2458 +#include "backup.h"
2461 +#define VMA_BLOCK_BITS 12
2462 +#define VMA_BLOCK_SIZE (1<<VMA_BLOCK_BITS)
2463 +#define VMA_CLUSTER_BITS (VMA_BLOCK_BITS+4)
2464 +#define VMA_CLUSTER_SIZE (1<<VMA_CLUSTER_BITS)
2466 +#if VMA_CLUSTER_SIZE != 65536
2467 +#error unexpected cluster size
2470 +#define VMA_EXTENT_HEADER_SIZE 512
2471 +#define VMA_BLOCKS_PER_EXTENT 59
2472 +#define VMA_MAX_CONFIGS 256
2474 +#define VMA_MAX_EXTENT_SIZE \
2475 + (VMA_EXTENT_HEADER_SIZE+VMA_CLUSTER_SIZE*VMA_BLOCKS_PER_EXTENT)
2476 +#if VMA_MAX_EXTENT_SIZE != 3867136
2477 +#error unexpected VMA_EXTENT_SIZE
2480 +/* File Format Definitions */
2482 +#define VMA_MAGIC (GUINT32_TO_BE(('V'<<24)|('M'<<16)|('A'<<8)|0x00))
2483 +#define VMA_EXTENT_MAGIC (GUINT32_TO_BE(('V'<<24)|('M'<<16)|('A'<<8)|'E'))
2485 +typedef struct VmaDeviceInfoHeader {
2486 + uint32_t devname_ptr; /* offset into blob_buffer table */
2487 + uint32_t reserved0;
2488 + uint64_t size; /* device size in bytes */
2489 + uint64_t reserved1;
2490 + uint64_t reserved2;
2491 +} VmaDeviceInfoHeader;
2493 +typedef struct VmaHeader {
2496 + unsigned char uuid[16];
2498 + unsigned char md5sum[16];
2500 + uint32_t blob_buffer_offset;
2501 + uint32_t blob_buffer_size;
2502 + uint32_t header_size;
2504 + unsigned char reserved[1984];
2506 + uint32_t config_names[VMA_MAX_CONFIGS]; /* offset into blob_buffer table */
2507 + uint32_t config_data[VMA_MAX_CONFIGS]; /* offset into blob_buffer table */
2509 + VmaDeviceInfoHeader dev_info[256];
2512 +typedef struct VmaExtentHeader {
2514 + uint16_t reserved1;
2515 + uint16_t block_count;
2516 + unsigned char uuid[16];
2517 + unsigned char md5sum[16];
2518 + uint64_t blockinfo[VMA_BLOCKS_PER_EXTENT];
2521 +/* functions/definitions to read/write vma files */
2523 +typedef struct VmaReader VmaReader;
2525 +typedef struct VmaWriter VmaWriter;
2527 +typedef struct VmaConfigData {
2533 +typedef struct VmaStreamInfo {
2535 + uint64_t cluster_count;
2536 + uint64_t transferred;
2537 + uint64_t zero_bytes;
2542 +typedef struct VmaStatus {
2545 + char errmsg[8192];
2546 + char uuid_str[37];
2547 + VmaStreamInfo stream_info[256];
2550 +typedef struct VmaDeviceInfo {
2551 + uint64_t size; /* device size in bytes */
2552 + const char *devname;
2555 +extern const BackupDriver backup_vma_driver;
2557 +VmaWriter *vma_writer_create(const char *filename, uuid_t uuid, Error **errp);
2558 +int vma_writer_close(VmaWriter *vmaw, Error **errp);
2559 +void vma_writer_destroy(VmaWriter *vmaw);
2560 +int vma_writer_add_config(VmaWriter *vmaw, const char *name, gpointer data,
2562 +int vma_writer_register_stream(VmaWriter *vmaw, const char *devname,
2565 +int64_t coroutine_fn vma_writer_write(VmaWriter *vmaw, uint8_t dev_id,
2566 + int64_t cluster_num, unsigned char *buf,
2567 + size_t *zero_bytes);
2569 +int coroutine_fn vma_writer_close_stream(VmaWriter *vmaw, uint8_t dev_id);
2571 +int vma_writer_get_status(VmaWriter *vmaw, VmaStatus *status);
2572 +void vma_writer_set_error(VmaWriter *vmaw, const char *fmt, ...);
2575 +VmaReader *vma_reader_create(const char *filename, Error **errp);
2576 +void vma_reader_destroy(VmaReader *vmar);
2577 +VmaHeader *vma_reader_get_header(VmaReader *vmar);
2578 +GList *vma_reader_get_config_data(VmaReader *vmar);
2579 +VmaDeviceInfo *vma_reader_get_device_info(VmaReader *vmar, guint8 dev_id);
2580 +int vma_reader_register_bs(VmaReader *vmar, guint8 dev_id,
2581 + BlockDriverState *bs, bool write_zeroes,
2583 +int vma_reader_restore(VmaReader *vmar, int vmstate_fd, bool verbose,
2586 +#endif /* BACKUP_VMA_H */