1 From ada8961d41298b340dc624da571cba760ab39158 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>
13 docs/specs/vma_spec.txt | 24 ++
14 vma-reader.c | 776 +++++++++++++++++++++++++++++++++++++++
15 vma-writer.c | 920 +++++++++++++++++++++++++++++++++++++++++++++++
16 vma.c | 555 ++++++++++++++++++++++++++++
18 8 files changed, 2427 insertions(+), 4 deletions(-)
19 create mode 100644 docs/specs/vma_spec.txt
20 create mode 100644 vma-reader.c
21 create mode 100644 vma-writer.c
22 create mode 100644 vma.c
23 create mode 100644 vma.h
25 diff --git a/Makefile b/Makefile
26 index 9ecbcbb..30a9268 100644
29 @@ -100,7 +100,7 @@ defconfig:
31 -include config-all-devices.mak
33 -all: $(DOCS) $(TOOLS) $(HELPERS-y) recurse-all
34 +all: $(DOCS) $(TOOLS) vma$(EXESUF) $(HELPERS-y) recurse-all
36 config-host.h: config-host.h-timestamp
37 config-host.h-timestamp: config-host.mak
38 @@ -194,6 +194,7 @@ tools-obj-$(CONFIG_POSIX) += compatfd.o
39 qemu-img$(EXESUF): qemu-img.o $(tools-obj-y) $(block-obj-y) libqemustub.a
40 qemu-nbd$(EXESUF): qemu-nbd.o $(tools-obj-y) $(block-obj-y) libqemustub.a
41 qemu-io$(EXESUF): qemu-io.o cmd.o $(tools-obj-y) $(block-obj-y) libqemustub.a
42 +vma$(EXESUF): vma.o vma-writer.o vma-reader.o $(tools-obj-y) $(block-obj-y) libqemustub.a
44 qemu-bridge-helper$(EXESUF): qemu-bridge-helper.o
46 diff --git a/Makefile.objs b/Makefile.objs
47 index cb46be5..b5732e2 100644
50 @@ -48,7 +48,7 @@ coroutine-obj-$(CONFIG_WIN32) += coroutine-win32.o
51 block-obj-y = iov.o cache-utils.o qemu-option.o module.o async.o
52 block-obj-y += nbd.o block.o blockjob.o aes.o qemu-config.o
53 block-obj-y += thread-pool.o qemu-progress.o qemu-sockets.o uri.o notify.o
54 -block-obj-y += backup.o
55 +block-obj-y += vma-writer.o backup.o
56 block-obj-y += $(coroutine-obj-y) $(qobject-obj-y) $(version-obj-y)
57 block-obj-$(CONFIG_POSIX) += event_notifier-posix.o aio-posix.o
58 block-obj-$(CONFIG_WIN32) += event_notifier-win32.o aio-win32.o
59 diff --git a/blockdev.c b/blockdev.c
60 index 9e85ffe..139e350 100644
65 #include "arch_init.h"
69 static QTAILQ_HEAD(drivelist, DriveInfo) drives = QTAILQ_HEAD_INITIALIZER(drives);
71 @@ -1483,10 +1484,11 @@ char *qmp_backup(const char *backupfile, bool has_format, const char *format,
72 /* Todo: try to auto-detect format based on file name */
73 format = has_format ? format : "vma";
75 - /* fixme: find driver for specifued format */
76 const BackupDriver *driver = NULL;
79 + if (strcmp(format, "vma") == 0) {
80 + driver = &backup_vma_driver;
82 error_set(errp, ERROR_CLASS_GENERIC_ERROR,
83 "no backup driver for format '%s'", format);
85 diff --git a/docs/specs/vma_spec.txt b/docs/specs/vma_spec.txt
87 index 0000000..052c629
89 +++ b/docs/specs/vma_spec.txt
91 +=Virtual Machine Archive format (VMA)=
93 +This format contains a header which includes the VM configuration as
94 +binary blobs, and a list of devices (dev_id, name).
96 +The actual VM image data is stored inside extents. An extent contains
97 +up to 64 clusters, and start with a 512 byte header containing
98 +additional information for those clusters.
100 +We use a cluster size of 65536, and use 8 bytes for each
101 +cluster in the header to store the following information:
103 +* 1 byte dev_id (to identity the drive)
104 +* 2 bytes zero indicator (mark zero regions (16x4096))
105 +* 4 bytes cluster number
106 +* 1 byte not used (reserved)
108 +We only store non-zero blocks (such block is 4096 bytes).
110 +Each archive is marked with a uuid. The archive header and all
111 +extent headers includes that uuid and a MD5 checksum (over header
115 diff --git a/vma-reader.c b/vma-reader.c
117 index 0000000..b05d7da
122 + * VMA: Virtual Machine Archive
124 + * Copyright (C) 2012 Proxmox Server Solutions
127 + * Dietmar Maurer (dietmar@proxmox.com)
129 + * This work is licensed under the terms of the GNU GPL, version 2 or later.
130 + * See the COPYING file in the top-level directory.
139 +#include <sys/types.h>
140 +#include <sys/stat.h>
143 +#include <uuid/uuid.h>
145 +#include "qemu-common.h"
146 +#include "qemu_socket.h"
147 +#include "qemu-coroutine.h"
148 +#include "qemu-aio.h"
149 +#include "qemu/ratelimit.h"
153 +#define BITS_PER_LONG (sizeof(unsigned long) * 8)
155 +static unsigned char zero_vma_block[VMA_BLOCK_SIZE];
157 +typedef struct VmaRestoreState {
158 + BlockDriverState *bs;
160 + unsigned long *bitmap;
166 + GChecksum *md5csum;
167 + GHashTable *blob_hash;
168 + unsigned char *head_data;
169 + VmaDeviceInfo devinfo[256];
170 + VmaRestoreState rstate[256];
172 + guint8 vmstate_stream;
173 + uint32_t vmstate_clusters;
177 +g_int32_hash(gconstpointer v)
179 + return *(const uint32_t *)v;
183 +g_int32_equal(gconstpointer v1, gconstpointer v2)
185 + return *((const uint32_t *)v1) == *((const uint32_t *)v2);
188 +static int vma_reader_get_bitmap(VmaRestoreState *rstate, int64_t cluster_num)
191 + assert(rstate->bitmap);
193 + unsigned long val, idx, bit;
195 + idx = cluster_num / BITS_PER_LONG;
197 + assert(rstate->bitmap_size > idx);
199 + bit = cluster_num % BITS_PER_LONG;
200 + val = rstate->bitmap[idx];
202 + return !!(val & (1UL << bit));
205 +static void vma_reader_set_bitmap(VmaRestoreState *rstate, int64_t cluster_num,
209 + assert(rstate->bitmap);
211 + unsigned long val, idx, bit;
213 + idx = cluster_num / BITS_PER_LONG;
215 + assert(rstate->bitmap_size > idx);
217 + bit = cluster_num % BITS_PER_LONG;
218 + val = rstate->bitmap[idx];
220 + if (!(val & (1UL << bit))) {
224 + if (val & (1UL << bit)) {
225 + val &= ~(1UL << bit);
228 + rstate->bitmap[idx] = val;
231 +typedef struct VmaBlob {
237 +static const VmaBlob *get_header_blob(VmaReader *vmar, uint32_t pos)
240 + assert(vmar->blob_hash);
242 + return g_hash_table_lookup(vmar->blob_hash, &pos);
245 +static const char *get_header_str(VmaReader *vmar, uint32_t pos)
247 + const VmaBlob *blob = get_header_blob(vmar, pos);
251 + const char *res = (char *)blob->data;
252 + if (res[blob->len-1] != '\0') {
259 +safe_read(int fd, unsigned char *buf, size_t count)
264 + n = read(fd, buf, count);
265 + } while (n < 0 && errno == EINTR);
271 +full_read(int fd, unsigned char *buf, size_t len)
279 + n = safe_read(fd, buf, len);
301 +void vma_reader_destroy(VmaReader *vmar)
305 + if (vmar->fd >= 0) {
309 + if (vmar->cdata_list) {
310 + g_list_free(vmar->cdata_list);
314 + for (i = 1; i < 256; i++) {
315 + if (vmar->rstate[i].bitmap) {
316 + g_free(vmar->rstate[i].bitmap);
320 + if (vmar->md5csum) {
321 + g_checksum_free(vmar->md5csum);
324 + if (vmar->blob_hash) {
325 + g_hash_table_destroy(vmar->blob_hash);
328 + if (vmar->head_data) {
329 + g_free(vmar->head_data);
336 +static int vma_reader_read_head(VmaReader *vmar, Error **errp)
340 + assert(*errp == NULL);
342 + unsigned char md5sum[16];
346 + vmar->head_data = g_malloc(sizeof(VmaHeader));
348 + if (full_read(vmar->fd, vmar->head_data, sizeof(VmaHeader)) !=
349 + sizeof(VmaHeader)) {
350 + error_setg(errp, "can't read vma header - %s",
351 + errno ? strerror(errno) : "got EOF");
355 + VmaHeader *h = (VmaHeader *)vmar->head_data;
357 + if (h->magic != VMA_MAGIC) {
358 + error_setg(errp, "not a vma file - wrong magic number");
362 + uint32_t header_size = GUINT32_FROM_BE(h->header_size);
363 + int need = header_size - sizeof(VmaHeader);
365 + error_setg(errp, "wrong vma header size %d", header_size);
369 + vmar->head_data = g_realloc(vmar->head_data, header_size);
370 + h = (VmaHeader *)vmar->head_data;
372 + if (full_read(vmar->fd, vmar->head_data + sizeof(VmaHeader), need) !=
374 + error_setg(errp, "can't read vma header data - %s",
375 + errno ? strerror(errno) : "got EOF");
379 + memcpy(md5sum, h->md5sum, 16);
380 + memset(h->md5sum, 0, 16);
382 + g_checksum_reset(vmar->md5csum);
383 + g_checksum_update(vmar->md5csum, vmar->head_data, header_size);
385 + g_checksum_get_digest(vmar->md5csum, (guint8 *)(h->md5sum), &csize);
387 + if (memcmp(md5sum, h->md5sum, 16) != 0) {
388 + error_setg(errp, "wrong vma header chechsum");
392 + /* we can modify header data after checksum verify */
393 + h->header_size = header_size;
395 + h->version = GUINT32_FROM_BE(h->version);
396 + if (h->version != 1) {
397 + error_setg(errp, "wrong vma version %d", h->version);
401 + h->ctime = GUINT64_FROM_BE(h->ctime);
402 + h->blob_buffer_offset = GUINT32_FROM_BE(h->blob_buffer_offset);
403 + h->blob_buffer_size = GUINT32_FROM_BE(h->blob_buffer_size);
405 + uint32_t bstart = h->blob_buffer_offset + 1;
406 + uint32_t bend = h->blob_buffer_offset + h->blob_buffer_size;
408 + if (bstart <= sizeof(VmaHeader)) {
409 + error_setg(errp, "wrong vma blob buffer offset %d",
410 + h->blob_buffer_offset);
414 + if (bend > header_size) {
415 + error_setg(errp, "wrong vma blob buffer size %d/%d",
416 + h->blob_buffer_offset, h->blob_buffer_size);
420 + while ((bstart + 2) <= bend) {
421 + uint32_t size = vmar->head_data[bstart] +
422 + (vmar->head_data[bstart+1] << 8);
423 + if ((bstart + size + 2) <= bend) {
424 + VmaBlob *blob = g_new0(VmaBlob, 1);
425 + blob->start = bstart - h->blob_buffer_offset;
427 + blob->data = vmar->head_data + bstart + 2;
428 + g_hash_table_insert(vmar->blob_hash, &blob->start, blob);
430 + bstart += size + 2;
435 + for (i = 1; i < 256; i++) {
436 + VmaDeviceInfoHeader *dih = &h->dev_info[i];
437 + uint32_t devname_ptr = GUINT32_FROM_BE(dih->devname_ptr);
438 + uint64_t size = GUINT64_FROM_BE(dih->size);
439 + const char *devname = get_header_str(vmar, devname_ptr);
441 + if (size && devname) {
443 + vmar->devinfo[i].size = size;
444 + vmar->devinfo[i].devname = devname;
446 + if (strcmp(devname, "vmstate") == 0) {
447 + vmar->vmstate_stream = i;
453 + error_setg(errp, "vma does not contain data");
457 + for (i = 0; i < VMA_MAX_CONFIGS; i++) {
458 + uint32_t name_ptr = GUINT32_FROM_BE(h->config_names[i]);
459 + uint32_t data_ptr = GUINT32_FROM_BE(h->config_data[i]);
461 + if (!(name_ptr && data_ptr)) {
464 + const char *name = get_header_str(vmar, name_ptr);
465 + const VmaBlob *blob = get_header_blob(vmar, data_ptr);
467 + if (!(name && blob)) {
468 + error_setg(errp, "vma contains invalid data pointers");
472 + VmaConfigData *cdata = g_new0(VmaConfigData, 1);
473 + cdata->name = name;
474 + cdata->data = blob->data;
475 + cdata->len = blob->len;
477 + vmar->cdata_list = g_list_append(vmar->cdata_list, cdata);
483 +VmaReader *vma_reader_create(const char *filename, Error **errp)
488 + VmaReader *vmar = g_new0(VmaReader, 1);
490 + if (strcmp(filename, "-") == 0) {
493 + vmar->fd = open(filename, O_RDONLY);
496 + if (vmar->fd < 0) {
497 + error_setg(errp, "can't open file %s - %s\n", filename,
502 + vmar->md5csum = g_checksum_new(G_CHECKSUM_MD5);
503 + if (!vmar->md5csum) {
504 + error_setg(errp, "can't allocate cmsum\n");
508 + vmar->blob_hash = g_hash_table_new_full(g_int32_hash, g_int32_equal,
511 + if (vma_reader_read_head(vmar, errp) < 0) {
519 + vma_reader_destroy(vmar);
525 +VmaHeader *vma_reader_get_header(VmaReader *vmar)
528 + assert(vmar->head_data);
530 + return (VmaHeader *)(vmar->head_data);
533 +GList *vma_reader_get_config_data(VmaReader *vmar)
536 + assert(vmar->head_data);
538 + return vmar->cdata_list;
541 +VmaDeviceInfo *vma_reader_get_device_info(VmaReader *vmar, guint8 dev_id)
546 + if (vmar->devinfo[dev_id].size && vmar->devinfo[dev_id].devname) {
547 + return &vmar->devinfo[dev_id];
553 +int vma_reader_register_bs(VmaReader *vmar, guint8 dev_id, BlockDriverState *bs,
554 + bool write_zeroes, Error **errp)
557 + assert(bs != NULL);
559 + assert(vmar->rstate[dev_id].bs == NULL);
561 + int64_t size = bdrv_getlength(bs);
562 + if (size != vmar->devinfo[dev_id].size) {
563 + error_setg(errp, "vma_reader_register_bs for stream %s failed - "
564 + "unexpected size %zd != %zd", vmar->devinfo[dev_id].devname,
565 + size, vmar->devinfo[dev_id].size);
569 + vmar->rstate[dev_id].bs = bs;
570 + vmar->rstate[dev_id].write_zeroes = write_zeroes;
572 + int64_t bitmap_size = (size/BDRV_SECTOR_SIZE) +
573 + (VMA_CLUSTER_SIZE/BDRV_SECTOR_SIZE) * BITS_PER_LONG - 1;
574 + bitmap_size /= (VMA_CLUSTER_SIZE/BDRV_SECTOR_SIZE) * BITS_PER_LONG;
576 + vmar->rstate[dev_id].bitmap_size = bitmap_size;
577 + vmar->rstate[dev_id].bitmap = g_new0(unsigned long, bitmap_size);
582 +static ssize_t safe_write(int fd, void *buf, size_t count)
587 + n = write(fd, buf, count);
588 + } while (n < 0 && errno == EINTR);
593 +static size_t full_write(int fd, void *buf, size_t len)
601 + n = safe_write(fd, buf, len);
611 + /* incomplete write ? */
618 +static int restore_write_data(VmaReader *vmar, guint8 dev_id,
619 + BlockDriverState *bs, int vmstate_fd,
620 + unsigned char *buf, int64_t sector_num,
621 + int nb_sectors, Error **errp)
625 + if (dev_id == vmar->vmstate_stream) {
626 + if (vmstate_fd >= 0) {
627 + int len = nb_sectors * BDRV_SECTOR_SIZE;
628 + int res = full_write(vmstate_fd, buf, len);
630 + error_setg(errp, "write vmstate failed %d", res);
635 + int res = bdrv_write(bs, sector_num, buf, nb_sectors);
637 + error_setg(errp, "bdrv_write to %s failed (%d)",
638 + bdrv_get_device_name(bs), res);
644 +static int restore_extent(VmaReader *vmar, unsigned char *buf,
645 + int extent_size, int vmstate_fd, Error **errp)
650 + VmaExtentHeader *ehead = (VmaExtentHeader *)buf;
651 + int start = VMA_EXTENT_HEADER_SIZE;
654 + for (i = 0; i < VMA_BLOCKS_PER_EXTENT; i++) {
655 + uint64_t block_info = GUINT64_FROM_BE(ehead->blockinfo[i]);
656 + uint32_t cluster_num = block_info & 0xffffffff;
657 + uint8_t dev_id = (block_info >> 32) & 0xff;
658 + uint16_t mask = block_info >> (32+16);
659 + int64_t max_sector;
665 + VmaRestoreState *rstate = &vmar->rstate[dev_id];
666 + BlockDriverState *bs = NULL;
668 + if (dev_id != vmar->vmstate_stream) {
671 + error_setg(errp, "got wrong dev id %d", dev_id);
675 + if (vma_reader_get_bitmap(rstate, cluster_num)) {
676 + error_setg(errp, "found duplicated cluster %d for stream %s",
677 + cluster_num, vmar->devinfo[dev_id].devname);
680 + vma_reader_set_bitmap(rstate, cluster_num, 1);
682 + max_sector = vmar->devinfo[dev_id].size/BDRV_SECTOR_SIZE;
684 + max_sector = G_MAXINT64;
685 + if (cluster_num != vmar->vmstate_clusters) {
686 + error_setg(errp, "found out of order vmstate data");
689 + vmar->vmstate_clusters++;
692 + /* try to write whole clusters to speedup restore */
693 + if (mask == 0xffff) {
694 + if ((start + VMA_CLUSTER_SIZE) > extent_size) {
695 + error_setg(errp, "short vma extent - too many blocks");
698 + int64_t sector_num = (cluster_num * VMA_CLUSTER_SIZE) /
700 + int64_t end_sector = sector_num +
701 + VMA_CLUSTER_SIZE/BDRV_SECTOR_SIZE;
703 + if (end_sector > max_sector) {
704 + end_sector = max_sector;
707 + if (end_sector <= sector_num) {
708 + error_setg(errp, "got wrong block address - write bejond end");
712 + int nb_sectors = end_sector - sector_num;
713 + if (restore_write_data(vmar, dev_id, bs, vmstate_fd, buf + start,
714 + sector_num, nb_sectors, errp) < 0) {
718 + start += VMA_CLUSTER_SIZE;
723 + for (j = 0; j < 16; j++) {
724 + int64_t sector_num = (cluster_num*VMA_CLUSTER_SIZE +
725 + j*VMA_BLOCK_SIZE)/BDRV_SECTOR_SIZE;
727 + int64_t end_sector = sector_num +
728 + VMA_BLOCK_SIZE/BDRV_SECTOR_SIZE;
729 + if (end_sector > max_sector) {
730 + end_sector = max_sector;
734 + if ((start + VMA_BLOCK_SIZE) > extent_size) {
735 + error_setg(errp, "short vma extent - too many blocks");
739 + if (end_sector <= sector_num) {
740 + error_setg(errp, "got wrong block address - "
741 + "write bejond end");
745 + int nb_sectors = end_sector - sector_num;
746 + if (restore_write_data(vmar, dev_id, bs, vmstate_fd,
747 + buf + start, sector_num,
748 + nb_sectors, errp) < 0) {
752 + start += VMA_BLOCK_SIZE;
756 + if (rstate->write_zeroes & (end_sector > sector_num)) {
757 + /* Todo: use bdrv_co_write_zeroes (but that need to
758 + * be run inside coroutine?)
760 + int nb_sectors = end_sector - sector_num;
761 + if (restore_write_data(vmar, dev_id, bs, vmstate_fd,
762 + zero_vma_block, sector_num,
763 + nb_sectors, errp) < 0) {
774 + if (start != extent_size) {
775 + error_setg(errp, "vma extent error - missing blocks");
782 +int vma_reader_restore(VmaReader *vmar, int vmstate_fd, Error **errp)
785 + assert(vmar->head_data);
788 + unsigned char buf[VMA_MAX_EXTENT_SIZE];
790 + unsigned char md5sum[16];
791 + VmaHeader *h = (VmaHeader *)vmar->head_data;
795 + int bytes = full_read(vmar->fd, buf + buf_pos, sizeof(buf) - buf_pos);
797 + error_setg(errp, "read failed - %s", strerror(errno));
807 + if (buf_pos < VMA_EXTENT_HEADER_SIZE) {
808 + error_setg(errp, "read short extent (%d bytes)", buf_pos);
812 + VmaExtentHeader *ehead = (VmaExtentHeader *)buf;
814 + /* extract md5sum */
815 + memcpy(md5sum, ehead->md5sum, sizeof(ehead->md5sum));
816 + memset(ehead->md5sum, 0, sizeof(ehead->md5sum));
818 + g_checksum_reset(vmar->md5csum);
819 + g_checksum_update(vmar->md5csum, buf, VMA_EXTENT_HEADER_SIZE);
821 + g_checksum_get_digest(vmar->md5csum, ehead->md5sum, &csize);
823 + if (memcmp(md5sum, ehead->md5sum, 16) != 0) {
824 + error_setg(errp, "wrong vma extent header chechsum");
828 + if (memcmp(h->uuid, ehead->uuid, sizeof(ehead->uuid)) != 0) {
829 + error_setg(errp, "wrong vma extent uuid");
833 + if (ehead->magic != VMA_EXTENT_MAGIC || ehead->reserved1 != 0) {
834 + error_setg(errp, "wrong vma extent header magic");
838 + int block_count = GUINT16_FROM_BE(ehead->block_count);
839 + int extent_size = VMA_EXTENT_HEADER_SIZE + block_count*VMA_BLOCK_SIZE;
841 + if (buf_pos < extent_size) {
842 + error_setg(errp, "short vma extent (%d < %d)", buf_pos,
847 + if (restore_extent(vmar, buf, extent_size, vmstate_fd, errp) < 0) {
851 + if (buf_pos > extent_size) {
852 + memmove(buf, buf + extent_size, buf_pos - extent_size);
853 + buf_pos = buf_pos - extent_size;
862 + for (i = 1; i < 256; i++) {
863 + VmaRestoreState *rstate = &vmar->rstate[i];
868 + if (bdrv_flush(rstate->bs) < 0) {
869 + error_setg(errp, "vma bdrv_flush %s failed",
870 + vmar->devinfo[i].devname);
874 + if (vmar->devinfo[i].size &&
875 + (strcmp(vmar->devinfo[i].devname, "vmstate") != 0)) {
876 + assert(rstate->bitmap);
878 + int64_t cluster_num, end;
880 + end = (vmar->devinfo[i].size + VMA_CLUSTER_SIZE - 1) /
883 + for (cluster_num = 0; cluster_num < end; cluster_num++) {
884 + if (!vma_reader_get_bitmap(rstate, cluster_num)) {
885 + error_setg(errp, "detected missing cluster %zd "
886 + "for stream %s", cluster_num,
887 + vmar->devinfo[i].devname);
897 diff --git a/vma-writer.c b/vma-writer.c
899 index 0000000..c1e1afe
904 + * VMA: Virtual Machine Archive
906 + * Copyright (C) 2012 Proxmox Server Solutions
909 + * Dietmar Maurer (dietmar@proxmox.com)
911 + * This work is licensed under the terms of the GNU GPL, version 2 or later.
912 + * See the COPYING file in the top-level directory.
921 +#include <sys/types.h>
922 +#include <sys/stat.h>
925 +#include <uuid/uuid.h>
927 +#include "qemu-common.h"
928 +#include "qemu_socket.h"
929 +#include "qemu-coroutine.h"
930 +#include "qemu-aio.h"
931 +#include "qemu/ratelimit.h"
934 +#include "monitor.h"
938 +#define DPRINTF(fmt, ...)\
939 + do { if (DEBUG_VMA) { printf("vma: " fmt, ## __VA_ARGS__); } } while (0)
941 +#define WRITE_BUFFERS 5
943 +typedef struct VmaAIOCB VmaAIOCB;
946 + unsigned char buffer[VMA_MAX_EXTENT_SIZE];
957 + bool header_written;
960 + /* we always write extents */
961 + unsigned char outbuf[VMA_MAX_EXTENT_SIZE];
962 + int outbuf_pos; /* in bytes */
963 + int outbuf_count; /* in VMA_BLOCKS */
964 + uint64_t outbuf_block_info[VMA_BLOCKS_PER_EXTENT];
966 + VmaAIOCB aiocbs[WRITE_BUFFERS];
969 + GChecksum *md5csum;
970 + CoMutex writer_lock;
971 + CoMutex flush_lock;
972 + Coroutine *co_writer;
976 + /* drive informations */
977 + VmaStreamInfo stream_info[256];
978 + guint stream_count;
980 + guint8 vmstate_stream;
981 + uint32_t vmstate_clusters;
983 + /* header blob table */
984 + char *header_blob_table;
985 + uint32_t header_blob_table_size;
986 + uint32_t header_blob_table_pos;
988 + /* store for config blobs */
989 + uint32_t config_names[VMA_MAX_CONFIGS]; /* offset into blob_buffer table */
990 + uint32_t config_data[VMA_MAX_CONFIGS]; /* offset into blob_buffer table */
991 + uint32_t config_count;
994 +void vma_writer_set_error(VmaWriter *vmaw, const char *fmt, ...)
998 + if (vmaw->status < 0) {
1002 + vmaw->status = -1;
1004 + va_start(ap, fmt);
1005 + g_vsnprintf(vmaw->errmsg, sizeof(vmaw->errmsg), fmt, ap);
1008 + DPRINTF("vma_writer_set_error: %s\n", vmaw->errmsg);
1011 +static uint32_t allocate_header_blob(VmaWriter *vmaw, const char *data,
1014 + if (len > 65535) {
1018 + if (!vmaw->header_blob_table ||
1019 + (vmaw->header_blob_table_size <
1020 + (vmaw->header_blob_table_pos + len + 2))) {
1021 + int newsize = vmaw->header_blob_table_size + ((len + 2 + 511)/512)*512;
1023 + vmaw->header_blob_table = g_realloc(vmaw->header_blob_table, newsize);
1024 + memset(vmaw->header_blob_table + vmaw->header_blob_table_size,
1025 + 0, newsize - vmaw->header_blob_table_size);
1026 + vmaw->header_blob_table_size = newsize;
1029 + uint32_t cpos = vmaw->header_blob_table_pos;
1030 + vmaw->header_blob_table[cpos] = len & 255;
1031 + vmaw->header_blob_table[cpos+1] = (len >> 8) & 255;
1032 + memcpy(vmaw->header_blob_table + cpos + 2, data, len);
1033 + vmaw->header_blob_table_pos += len + 2;
1037 +static uint32_t allocate_header_string(VmaWriter *vmaw, const char *str)
1041 + size_t len = strlen(str) + 1;
1043 + return allocate_header_blob(vmaw, str, len);
1046 +int vma_writer_add_config(VmaWriter *vmaw, const char *name, gpointer data,
1050 + assert(!vmaw->header_written);
1051 + assert(vmaw->config_count < VMA_MAX_CONFIGS);
1056 + uint32_t name_ptr = allocate_header_string(vmaw, name);
1061 + uint32_t data_ptr = allocate_header_blob(vmaw, data, len);
1066 + vmaw->config_names[vmaw->config_count] = name_ptr;
1067 + vmaw->config_data[vmaw->config_count] = data_ptr;
1069 + vmaw->config_count++;
1074 +int vma_writer_register_stream(VmaWriter *vmaw, const char *devname,
1079 + assert(!vmaw->status);
1081 + if (vmaw->header_written) {
1082 + vma_writer_set_error(vmaw, "vma_writer_register_stream: header "
1083 + "already written");
1087 + guint n = vmaw->stream_count + 1;
1089 + /* we can have dev_ids form 1 to 255 (0 reserved)
1090 + * 255(-1) reseverd for safety
1093 + vma_writer_set_error(vmaw, "vma_writer_register_stream: "
1094 + "too many drives");
1099 + vma_writer_set_error(vmaw, "vma_writer_register_stream: "
1100 + "got strange size %zd", size);
1104 + DPRINTF("vma_writer_register_stream %s %zu %d\n", devname, size, n);
1106 + vmaw->stream_info[n].devname = g_strdup(devname);
1107 + vmaw->stream_info[n].size = size;
1109 + vmaw->stream_info[n].cluster_count = (size + VMA_CLUSTER_SIZE - 1) /
1112 + vmaw->stream_count = n;
1114 + if (strcmp(devname, "vmstate") == 0) {
1115 + vmaw->vmstate_stream = n;
1121 +static void vma_co_continue_write(void *opaque)
1123 + VmaWriter *vmaw = opaque;
1125 + qemu_aio_set_fd_handler(vmaw->fd, NULL, NULL, NULL, NULL);
1127 + DPRINTF("vma_co_continue_write\n");
1128 + qemu_coroutine_enter(vmaw->co_writer, NULL);
1131 +static ssize_t coroutine_fn
1132 +vma_co_write(VmaWriter *vmaw, const void *buf, size_t bytes)
1137 + /* atomic writes (we cannot interleave writes) */
1138 + qemu_co_mutex_lock(&vmaw->writer_lock);
1140 + DPRINTF("vma_co_write enter %zd\n", bytes);
1142 + while (done < bytes) {
1143 + ret = write(vmaw->fd, buf + done, bytes - done);
1146 + DPRINTF("vma_co_write written %zd %zd\n", done, ret);
1147 + } else if (ret < 0) {
1148 + if (errno == EAGAIN || errno == EWOULDBLOCK) {
1149 + DPRINTF("vma_co_write yield %zd\n", done);
1151 + vmaw->co_writer = qemu_coroutine_self();
1152 + qemu_aio_set_fd_handler(vmaw->fd, NULL, vma_co_continue_write,
1155 + qemu_coroutine_yield();
1156 + DPRINTF("vma_co_write restart %zd\n", done);
1158 + vma_writer_set_error(vmaw, "vma_co_write write error - %s",
1160 + done = -1; /* always return failure for partial writes */
1163 + } else if (ret == 0) {
1164 + /* should not happen - simply try again */
1168 + qemu_co_mutex_unlock(&vmaw->writer_lock);
1170 + DPRINTF("vma_co_write leave %zd\n", done);
1174 +static void coroutine_fn vma_co_writer_task(void *opaque)
1176 + VmaAIOCB *cb = opaque;
1178 + DPRINTF("vma_co_writer_task start\n");
1180 + int64_t done = vma_co_write(cb->vmaw, cb->buffer, cb->bytes);
1181 + DPRINTF("vma_co_writer_task write done %zd\n", done);
1183 + if (done != cb->bytes) {
1184 + DPRINTF("vma_co_writer_task failed write %zd %zd", cb->bytes, done);
1185 + vma_writer_set_error(cb->vmaw, "vma_co_writer_task failed write %zd",
1191 + qemu_co_queue_next(&cb->vmaw->wqueue);
1193 + DPRINTF("vma_co_writer_task end\n");
1196 +static void coroutine_fn vma_queue_flush(VmaWriter *vmaw)
1198 + DPRINTF("vma_queue_flush enter\n");
1204 + VmaAIOCB *cb = NULL;
1205 + for (i = 0; i < WRITE_BUFFERS; i++) {
1206 + if (vmaw->aiocbs[i].bytes) {
1207 + cb = &vmaw->aiocbs[i];
1208 + DPRINTF("FOUND USED AIO BUFFER %d %zd\n", i,
1209 + vmaw->aiocbs[i].bytes);
1216 + qemu_co_queue_wait(&vmaw->wqueue);
1219 + DPRINTF("vma_queue_flush leave\n");
1223 + * NOTE: pipe buffer size in only 4096 bytes on linux (see 'ulimit -a')
1224 + * So we need to create a coroutione to allow 'parallel' execution.
1226 +static ssize_t coroutine_fn
1227 +vma_queue_write(VmaWriter *vmaw, const void *buf, size_t bytes)
1229 + DPRINTF("vma_queue_write enter %zd\n", bytes);
1233 + assert(bytes <= VMA_MAX_EXTENT_SIZE);
1235 + VmaAIOCB *cb = NULL;
1238 + for (i = 0; i < WRITE_BUFFERS; i++) {
1239 + if (!vmaw->aiocbs[i].bytes) {
1240 + cb = &vmaw->aiocbs[i];
1245 + qemu_co_queue_wait(&vmaw->wqueue);
1249 + memcpy(cb->buffer, buf, bytes);
1250 + cb->bytes = bytes;
1253 + DPRINTF("vma_queue_write start %zd\n", bytes);
1254 + cb->co = qemu_coroutine_create(vma_co_writer_task);
1255 + qemu_coroutine_enter(cb->co, cb);
1257 + DPRINTF("vma_queue_write leave\n");
1262 +VmaWriter *vma_writer_create(const char *filename, uuid_t uuid, int64_t speed,
1267 + assert(sizeof(VmaHeader) == (4096 + 8192));
1268 + assert(sizeof(VmaExtentHeader) == 512);
1270 + VmaWriter *vmaw = g_new0(VmaWriter, 1);
1273 + vmaw->md5csum = g_checksum_new(G_CHECKSUM_MD5);
1274 + if (!vmaw->md5csum) {
1275 + error_setg(errp, "can't allocate cmsum\n");
1279 + if (strstart(filename, "exec:", &p)) {
1280 + vmaw->cmd = popen(p, "w");
1281 + if (vmaw->cmd == NULL) {
1282 + error_setg(errp, "can't popen command '%s' - %s\n", p,
1286 + vmaw->fd = fileno(vmaw->cmd);
1287 + socket_set_nonblock(vmaw->fd);
1292 + const char *tmp_id_str;
1294 + if ((stat(filename, &st) == 0) && S_ISFIFO(st.st_mode)) {
1295 + oflags = O_NONBLOCK|O_WRONLY;
1296 + vmaw->fd = qemu_open(filename, oflags, 0644);
1297 + } else if (strstart(filename, "/dev/fdset/", &tmp_id_str)) {
1298 + oflags = O_NONBLOCK|O_WRONLY;
1299 + vmaw->fd = qemu_open(filename, oflags, 0644);
1300 + } else if (strstart(filename, "/dev/fdname/", &tmp_id_str)) {
1301 + vmaw->fd = monitor_get_fd(cur_mon, tmp_id_str, errp);
1302 + if (vmaw->fd < 0) {
1306 + oflags = O_NONBLOCK|O_WRONLY|O_CREAT|O_EXCL;
1307 + vmaw->fd = qemu_open(filename, oflags, 0644);
1310 + if (vmaw->fd < 0) {
1311 + error_setg(errp, "can't open file %s - %s\n", filename,
1317 + vmaw->outbuf_count = 0;
1318 + vmaw->outbuf_pos = VMA_EXTENT_HEADER_SIZE;
1320 + vmaw->header_blob_table_pos = 1; /* start at pos 1 */
1322 + qemu_co_mutex_init(&vmaw->writer_lock);
1323 + qemu_co_mutex_init(&vmaw->flush_lock);
1324 + qemu_co_queue_init(&vmaw->wqueue);
1326 + uuid_copy(vmaw->uuid, uuid);
1329 + speed = 10*1024*1024*1024LLU; /* default 10GB/s */
1332 + ratelimit_set_speed(&vmaw->limit, speed, 100000000ULL /* 0.1 sec */);
1339 + pclose(vmaw->cmd);
1340 + } else if (vmaw->fd >= 0) {
1344 + if (vmaw->md5csum) {
1345 + g_checksum_free(vmaw->md5csum);
1354 +static int coroutine_fn vma_write_header(VmaWriter *vmaw)
1357 + int header_clusters = 8;
1358 + char buf[65536*header_clusters];
1359 + VmaHeader *head = (VmaHeader *)buf;
1363 + DPRINTF("VMA WRITE HEADER\n");
1365 + if (vmaw->status < 0) {
1366 + return vmaw->status;
1369 + memset(buf, 0, sizeof(buf));
1371 + head->magic = VMA_MAGIC;
1372 + head->version = GUINT32_TO_BE(1); /* v1 */
1373 + memcpy(head->uuid, vmaw->uuid, 16);
1375 + time_t ctime = time(NULL);
1376 + head->ctime = GUINT64_TO_BE(ctime);
1378 + if (!vmaw->stream_count) {
1382 + for (i = 0; i < VMA_MAX_CONFIGS; i++) {
1383 + head->config_names[i] = GUINT32_TO_BE(vmaw->config_names[i]);
1384 + head->config_data[i] = GUINT32_TO_BE(vmaw->config_data[i]);
1387 + /* 32 bytes per device (12 used currently) = 8192 bytes max */
1388 + for (i = 1; i <= 254; i++) {
1389 + VmaStreamInfo *si = &vmaw->stream_info[i];
1391 + assert(si->devname);
1392 + uint32_t devname_ptr = allocate_header_string(vmaw, si->devname);
1393 + if (!devname_ptr) {
1396 + head->dev_info[i].devname_ptr = GUINT32_TO_BE(devname_ptr);
1397 + head->dev_info[i].size = GUINT64_TO_BE(si->size);
1401 + uint32_t header_size = sizeof(VmaHeader) + vmaw->header_blob_table_size;
1402 + head->header_size = GUINT32_TO_BE(header_size);
1404 + if (header_size > sizeof(buf)) {
1405 + return -1; /* just to be sure */
1408 + uint32_t blob_buffer_offset = sizeof(VmaHeader);
1409 + memcpy(buf + blob_buffer_offset, vmaw->header_blob_table,
1410 + vmaw->header_blob_table_size);
1411 + head->blob_buffer_offset = GUINT32_TO_BE(blob_buffer_offset);
1412 + head->blob_buffer_size = GUINT32_TO_BE(vmaw->header_blob_table_pos);
1414 + g_checksum_reset(vmaw->md5csum);
1415 + g_checksum_update(vmaw->md5csum, (const guchar *)buf, header_size);
1417 + g_checksum_get_digest(vmaw->md5csum, (guint8 *)(head->md5sum), &csize);
1419 + return vma_queue_write(vmaw, buf, header_size);
1422 +static int coroutine_fn vma_writer_flush(VmaWriter *vmaw)
1429 + if (vmaw->status < 0) {
1430 + return vmaw->status;
1433 + if (!vmaw->header_written) {
1434 + vmaw->header_written = true;
1435 + ret = vma_write_header(vmaw);
1437 + vma_writer_set_error(vmaw, "vma_writer_flush: write header failed");
1442 + DPRINTF("VMA WRITE FLUSH %d %d\n", vmaw->outbuf_count, vmaw->outbuf_pos);
1445 + VmaExtentHeader *ehead = (VmaExtentHeader *)vmaw->outbuf;
1447 + ehead->magic = VMA_EXTENT_MAGIC;
1448 + ehead->reserved1 = 0;
1450 + for (i = 0; i < VMA_BLOCKS_PER_EXTENT; i++) {
1451 + ehead->blockinfo[i] = GUINT64_TO_BE(vmaw->outbuf_block_info[i]);
1454 + guint16 block_count = (vmaw->outbuf_pos - VMA_EXTENT_HEADER_SIZE) /
1457 + ehead->block_count = GUINT16_TO_BE(block_count);
1459 + memcpy(ehead->uuid, vmaw->uuid, sizeof(ehead->uuid));
1460 + memset(ehead->md5sum, 0, sizeof(ehead->md5sum));
1462 + g_checksum_reset(vmaw->md5csum);
1463 + g_checksum_update(vmaw->md5csum, vmaw->outbuf, VMA_EXTENT_HEADER_SIZE);
1465 + g_checksum_get_digest(vmaw->md5csum, ehead->md5sum, &csize);
1467 + int bytes = vmaw->outbuf_pos;
1468 + ret = vma_queue_write(vmaw, vmaw->outbuf, bytes);
1469 + if (ret != bytes) {
1470 + vma_writer_set_error(vmaw, "vma_writer_flush: failed write");
1473 + vmaw->outbuf_count = 0;
1474 + vmaw->outbuf_pos = VMA_EXTENT_HEADER_SIZE;
1476 + for (i = 0; i < VMA_BLOCKS_PER_EXTENT; i++) {
1477 + vmaw->outbuf_block_info[i] = 0;
1480 + return vmaw->status;
1483 +static int vma_count_open_streams(VmaWriter *vmaw)
1485 + g_assert(vmaw != NULL);
1488 + int open_drives = 0;
1489 + for (i = 0; i <= 255; i++) {
1490 + if (vmaw->stream_info[i].size && !vmaw->stream_info[i].finished) {
1495 + return open_drives;
1499 + * all jobs should call this when there is no more data
1500 + * Returns: number of remaining stream (0 ==> finished)
1503 +vma_writer_close_stream(VmaWriter *vmaw, uint8_t dev_id)
1505 + g_assert(vmaw != NULL);
1507 + DPRINTF("vma_writer_set_status %d\n", dev_id);
1508 + if (!vmaw->stream_info[dev_id].size) {
1509 + vma_writer_set_error(vmaw, "vma_writer_close_stream: "
1510 + "no such stream %d", dev_id);
1513 + if (vmaw->stream_info[dev_id].finished) {
1514 + vma_writer_set_error(vmaw, "vma_writer_close_stream: "
1515 + "stream already closed %d", dev_id);
1519 + vmaw->stream_info[dev_id].finished = true;
1521 + int open_drives = vma_count_open_streams(vmaw);
1523 + if (open_drives <= 0) {
1524 + DPRINTF("vma_writer_set_status all drives completed\n");
1525 + qemu_co_mutex_lock(&vmaw->flush_lock);
1526 + int ret = vma_writer_flush(vmaw);
1527 + qemu_co_mutex_unlock(&vmaw->flush_lock);
1529 + vma_writer_set_error(vmaw, "vma_writer_close_stream: flush failed");
1533 + return open_drives;
1536 +int vma_writer_get_status(VmaWriter *vmaw, VmaStatus *status)
1540 + g_assert(vmaw != NULL);
1543 + status->status = vmaw->status;
1544 + g_strlcpy(status->errmsg, vmaw->errmsg, sizeof(status->errmsg));
1545 + for (i = 0; i <= 255; i++) {
1546 + status->stream_info[i] = vmaw->stream_info[i];
1549 + uuid_unparse_lower(vmaw->uuid, status->uuid_str);
1552 + status->closed = vmaw->closed;
1554 + return vmaw->status;
1557 +static int vma_writer_get_buffer(VmaWriter *vmaw)
1560 + /* wait until buffer is available */
1561 + while (vmaw->outbuf_count >= (VMA_BLOCKS_PER_EXTENT - 1)) {
1564 + qemu_co_mutex_lock(&vmaw->flush_lock);
1565 + res = vma_writer_flush(vmaw);
1566 + qemu_co_mutex_unlock(&vmaw->flush_lock);
1569 + vma_writer_set_error(vmaw, "vma_writer_get_buffer: flush failed");
1578 +int64_t coroutine_fn
1579 +vma_writer_write(VmaWriter *vmaw, uint8_t dev_id, int64_t cluster_num,
1580 + unsigned char *buf, size_t *zero_bytes)
1582 + g_assert(vmaw != NULL);
1583 + g_assert(zero_bytes != NULL);
1587 + if (vmaw->status < 0) {
1588 + return vmaw->status;
1591 + if (!dev_id || !vmaw->stream_info[dev_id].size) {
1592 + vma_writer_set_error(vmaw, "vma_writer_write: "
1593 + "no such stream %d", dev_id);
1597 + if (vmaw->stream_info[dev_id].finished) {
1598 + vma_writer_set_error(vmaw, "vma_writer_write: "
1599 + "stream already closed %d", dev_id);
1604 + if (cluster_num >= (((uint64_t)1)<<32)) {
1605 + vma_writer_set_error(vmaw, "vma_writer_write: "
1606 + "cluster number out of range");
1610 + if (dev_id == vmaw->vmstate_stream) {
1611 + if (cluster_num != vmaw->vmstate_clusters) {
1612 + vma_writer_set_error(vmaw, "vma_writer_write: "
1613 + "non sequential vmstate write");
1615 + vmaw->vmstate_clusters++;
1616 + } else if (cluster_num >= vmaw->stream_info[dev_id].cluster_count) {
1617 + vma_writer_set_error(vmaw, "vma_writer_write: cluster number too big");
1621 + /* wait until buffer is available */
1622 + if (vma_writer_get_buffer(vmaw) < 0) {
1623 + vma_writer_set_error(vmaw, "vma_writer_write: "
1624 + "vma_writer_get_buffer failed");
1628 + DPRINTF("VMA WRITE %zd\n", cluster_num);
1632 + uint16_t mask = 0;
1633 + for (i = 0; i < 16; i++) {
1634 + unsigned char *vmablock = buf + (i*VMA_BLOCK_SIZE);
1635 + if (buffer_is_zero(vmablock, VMA_BLOCK_SIZE)) {
1636 + DPRINTF("VMA WRITE %zd ZERO BLOCK %d\n", cluster_num, i);
1637 + vmaw->stream_info[dev_id].zero_bytes += VMA_BLOCK_SIZE;
1638 + *zero_bytes += VMA_BLOCK_SIZE;
1641 + memcpy(vmaw->outbuf + vmaw->outbuf_pos, vmablock, VMA_BLOCK_SIZE);
1642 + vmaw->outbuf_pos += VMA_BLOCK_SIZE;
1644 + vmaw->delay_ns = ratelimit_calculate_delay(&vmaw->limit,
1646 + if (vmaw->delay_ns) {
1647 + co_sleep_ns(rt_clock, vmaw->delay_ns);
1654 + uint64_t block_info = ((uint64_t)mask) << (32+16);
1655 + block_info |= ((uint64_t)dev_id) << 32;
1656 + block_info |= (cluster_num & 0xffffffff);
1657 + vmaw->outbuf_block_info[vmaw->outbuf_count] = block_info;
1659 + DPRINTF("VMA WRITE MASK %zd %zx\n", cluster_num, block_info);
1661 + vmaw->outbuf_count++;
1663 + /** NOTE: We allways write whole clusters, but we correctly set
1664 + * transferred bytes. So transferred == size when when everything
1667 + size_t transferred = VMA_CLUSTER_SIZE;
1669 + if (dev_id != vmaw->vmstate_stream) {
1670 + uint64_t last = (cluster_num + 1) * VMA_CLUSTER_SIZE;
1671 + if (last > vmaw->stream_info[dev_id].size) {
1672 + uint64_t diff = last - vmaw->stream_info[dev_id].size;
1673 + if (diff >= VMA_CLUSTER_SIZE) {
1674 + vma_writer_set_error(vmaw, "vma_writer_write: "
1675 + "read after last cluster");
1678 + transferred -= diff;
1682 + vmaw->stream_info[dev_id].transferred += transferred;
1684 + return transferred;
1687 +int vma_writer_close(VmaWriter *vmaw, Error **errp)
1689 + g_assert(vmaw != NULL);
1693 + vma_queue_flush(vmaw);
1695 + /* this should not happen - just to be sure */
1696 + while (!qemu_co_queue_empty(&vmaw->wqueue)) {
1697 + DPRINTF("vma_writer_close wait\n");
1698 + co_sleep_ns(rt_clock, 1000000);
1702 + if (pclose(vmaw->cmd) < 0) {
1703 + vma_writer_set_error(vmaw, "vma_writer_close: "
1704 + "pclose failed - %s", strerror(errno));
1707 + if (close(vmaw->fd) < 0) {
1708 + vma_writer_set_error(vmaw, "vma_writer_close: "
1709 + "close failed - %s", strerror(errno));
1713 + for (i = 0; i <= 255; i++) {
1714 + VmaStreamInfo *si = &vmaw->stream_info[i];
1716 + if (!si->finished) {
1717 + vma_writer_set_error(vmaw, "vma_writer_close: "
1718 + "detected open stream '%s'", si->devname);
1719 + } else if ((si->transferred != si->size) &&
1720 + (i != vmaw->vmstate_stream)) {
1721 + vma_writer_set_error(vmaw, "vma_writer_close: "
1722 + "incomplete stream '%s' (%zd != %zd)",
1723 + si->devname, si->transferred, si->size);
1728 + for (i = 0; i <= 255; i++) {
1729 + vmaw->stream_info[i].finished = 1; /* mark as closed */
1734 + if (vmaw->status < 0 && *errp == NULL) {
1735 + error_setg(errp, "%s", vmaw->errmsg);
1738 + return vmaw->status;
1741 +void vma_writer_destroy(VmaWriter *vmaw)
1747 + for (i = 0; i <= 255; i++) {
1748 + if (vmaw->stream_info[i].devname) {
1749 + g_free(vmaw->stream_info[i].devname);
1753 + if (vmaw->md5csum) {
1754 + g_checksum_free(vmaw->md5csum);
1760 +/* backup driver plugin */
1762 +static int vma_dump_cb(void *opaque, uint8_t dev_id, int64_t cluster_num,
1763 + unsigned char *buf, size_t *zero_bytes)
1765 + VmaWriter *vmaw = opaque;
1767 + return vma_writer_write(vmaw, dev_id, cluster_num, buf, zero_bytes);
1770 +static int vma_close_cb(void *opaque, Error **errp)
1772 + VmaWriter *vmaw = opaque;
1774 + int res = vma_writer_close(vmaw, errp);
1775 + vma_writer_destroy(vmaw);
1780 +static int vma_complete_cb(void *opaque, uint8_t dev_id, int ret)
1782 + VmaWriter *vmaw = opaque;
1785 + vma_writer_set_error(vmaw, "backup_complete_cb %d", ret);
1788 + return vma_writer_close_stream(vmaw, dev_id);
1791 +static int vma_register_stream_cb(void *opaque, const char *devname,
1794 + VmaWriter *vmaw = opaque;
1796 + return vma_writer_register_stream(vmaw, devname, size);
1799 +static int vma_register_config_cb(void *opaque, const char *name,
1800 + gpointer data, size_t data_len)
1802 + VmaWriter *vmaw = opaque;
1804 + return vma_writer_add_config(vmaw, name, data, data_len);
1807 +static void *vma_open_cb(const char *filename, uuid_t uuid, int64_t speed,
1810 + return vma_writer_create(filename, uuid, speed, errp);
1813 +const BackupDriver backup_vma_driver = {
1815 + .open_cb = vma_open_cb,
1816 + .close_cb = vma_close_cb,
1817 + .register_config_cb = vma_register_config_cb,
1818 + .register_stream_cb = vma_register_stream_cb,
1819 + .dump_cb = vma_dump_cb,
1820 + .complete_cb = vma_complete_cb,
1823 diff --git a/vma.c b/vma.c
1824 new file mode 100644
1825 index 0000000..7a88d3f
1830 + * VMA: Virtual Machine Archive
1832 + * Copyright (C) 2012 Proxmox Server Solutions
1835 + * Dietmar Maurer (dietmar@proxmox.com)
1837 + * This work is licensed under the terms of the GNU GPL, version 2 or later.
1838 + * See the COPYING file in the top-level directory.
1844 +#include <unistd.h>
1846 +#include <string.h>
1847 +#include <sys/types.h>
1848 +#include <sys/stat.h>
1852 +#include "qemu-common.h"
1853 +#include "qemu-option.h"
1854 +#include "qemu-error.h"
1856 +#include "sysemu.h"
1857 +#include "block_int.h"
1861 +static void help(void)
1863 + const char *help_msg =
1864 + "usage: vma command [command options]\n"
1866 + "vma list <filename>\n"
1867 + "vma create <filename> [-c config] <archive> pathname ...\n"
1868 + "vma extract <filename> [-r] <targetdir>\n"
1871 + printf("%s", help_msg);
1875 +static const char *extract_devname(const char *path, char **devname, int index)
1879 + const char *sep = strchr(path, '=');
1882 + *devname = g_strndup(path, sep - path);
1886 + *devname = g_strdup_printf("disk%d", index);
1895 +static void print_content(VmaReader *vmar)
1899 + VmaHeader *head = vma_reader_get_header(vmar);
1901 + printf("CTIME: %s", ctime(&head->ctime));
1903 + GList *l = vma_reader_get_config_data(vmar);
1904 + while (l && l->data) {
1905 + VmaConfigData *cdata = (VmaConfigData *)l->data;
1906 + l = g_list_next(l);
1907 + printf("CFG: size: %d name: %s\n", cdata->len, cdata->name);
1911 + VmaDeviceInfo *di;
1912 + for (i = 1; i < 255; i++) {
1913 + di = vma_reader_get_device_info(vmar, i);
1915 + if (strcmp(di->devname, "vmstate") == 0) {
1916 + printf("VMSTATE: dev_id=%d memory: %zd\n", i, di->size);
1918 + printf("DEV: dev_id=%d size: %zd devname: %s\n",
1919 + i, di->size, di->devname);
1925 +static int list_content(int argc, char **argv)
1928 + const char *filename;
1931 + c = getopt(argc, argv, "h");
1941 + g_assert_not_reached();
1945 + /* Get the filename */
1946 + if ((optind + 1) != argc) {
1949 + filename = argv[optind++];
1951 + Error *errp = NULL;
1952 + VmaReader *vmar = vma_reader_create(filename, &errp);
1955 + g_error("%s", error_get_pretty(errp));
1958 + print_content(vmar);
1960 + vma_reader_destroy(vmar);
1965 +typedef struct RestoreMap {
1971 +static int extract_content(int argc, char **argv)
1974 + const char *filename;
1975 + const char *dirname;
1976 + const char *readmap = NULL;
1979 + c = getopt(argc, argv, "hr:");
1996 + /* Get the filename */
1997 + if ((optind + 2) != argc) {
2000 + filename = argv[optind++];
2001 + dirname = argv[optind++];
2003 + Error *errp = NULL;
2004 + VmaReader *vmar = vma_reader_create(filename, &errp);
2007 + g_error("%s", error_get_pretty(errp));
2010 + if (mkdir(dirname, 0777) < 0) {
2011 + g_error("unable to create target directory %s - %s",
2012 + dirname, strerror(errno));
2015 + GList *l = vma_reader_get_config_data(vmar);
2016 + while (l && l->data) {
2017 + VmaConfigData *cdata = (VmaConfigData *)l->data;
2018 + l = g_list_next(l);
2019 + char *cfgfn = g_strdup_printf("%s/%s", dirname, cdata->name);
2020 + GError *err = NULL;
2021 + if (!g_file_set_contents(cfgfn, (gchar *)cdata->data, cdata->len,
2023 + g_error("unable to write file: %s", err->message);
2027 + GHashTable *devmap = g_hash_table_new(g_str_hash, g_str_equal);
2030 + print_content(vmar);
2032 + FILE *map = fopen(readmap, "r");
2034 + g_error("unable to open fifo %s - %s", readmap, strerror(errno));
2039 + char *line = fgets(inbuf, sizeof(inbuf), map);
2040 + if (!line || line[0] == '\0' || !strcmp(line, "done\n")) {
2043 + int len = strlen(line);
2044 + if (line[len - 1] == '\n') {
2045 + line[len - 1] = '\0';
2053 + if (line[0] == '0' && line[1] == ':') {
2055 + write_zero = false;
2056 + } else if (line[0] == '1' && line[1] == ':') {
2058 + write_zero = true;
2060 + g_error("read map failed - parse error ('%s')", inbuf);
2063 + char *devname = NULL;
2064 + path = extract_devname(path, &devname, -1);
2066 + g_error("read map failed - no dev name specified ('%s')",
2070 + printf("TEST %s %s\n", path, devname);
2072 + RestoreMap *map = g_new0(RestoreMap, 1);
2073 + map->devname = g_strdup(devname);
2074 + map->path = g_strdup(path);
2075 + map->write_zero = write_zero;
2077 + g_hash_table_insert(devmap, map->devname, map);
2083 + int vmstate_fd = -1;
2084 + guint8 vmstate_stream = 0;
2086 + for (i = 1; i < 255; i++) {
2087 + VmaDeviceInfo *di = vma_reader_get_device_info(vmar, i);
2088 + if (di && (strcmp(di->devname, "vmstate") == 0)) {
2089 + vmstate_stream = i;
2090 + char *statefn = g_strdup_printf("%s/vmstate.bin", dirname);
2091 + vmstate_fd = open(statefn, O_WRONLY|O_CREAT|O_EXCL, 0644);
2092 + if (vmstate_fd < 0) {
2093 + g_error("create vmstate file '%s' failed - %s", statefn,
2098 + char *devfn = NULL;
2099 + int flags = BDRV_O_RDWR|BDRV_O_CACHE_WB;
2100 + bool write_zero = true;
2104 + map = (RestoreMap *)g_hash_table_lookup(devmap, di->devname);
2105 + if (map == NULL) {
2106 + g_error("no device name mapping for %s", di->devname);
2108 + devfn = map->path;
2109 + write_zero = map->write_zero;
2111 + devfn = g_strdup_printf("%s/tmp-disk-%s.raw",
2112 + dirname, di->devname);
2113 + printf("DEVINFO %s %zd\n", devfn, di->size);
2115 + if (bdrv_img_create(devfn, "raw", NULL, NULL, NULL,
2116 + di->size, flags)) {
2117 + g_error("can't create file %s", devfn);
2120 + /* Note: we created an empty file above, so there is no
2121 + * need to write zeroes (so we generate a sparse file)
2123 + write_zero = false;
2126 + BlockDriverState *bs = NULL;
2127 + if (bdrv_file_open(&bs, devfn, flags)) {
2128 + g_error("can't open file %s", devfn);
2130 + if (vma_reader_register_bs(vmar, i, bs, write_zero, &errp) < 0) {
2131 + g_error("%s", error_get_pretty(errp));
2140 + if (vma_reader_restore(vmar, vmstate_fd, &errp) < 0) {
2141 + g_error("restore failed - %s", error_get_pretty(errp));
2145 + for (i = 1; i < 255; i++) {
2146 + VmaDeviceInfo *di = vma_reader_get_device_info(vmar, i);
2147 + if (di && (i != vmstate_stream)) {
2148 + char *tmpfn = g_strdup_printf("%s/tmp-disk-%s.raw",
2149 + dirname, di->devname);
2150 + char *fn = g_strdup_printf("%s/disk-%s.raw",
2151 + dirname, di->devname);
2152 + if (rename(tmpfn, fn) != 0) {
2153 + g_error("rename %s to %s failed - %s",
2154 + tmpfn, fn, strerror(errno));
2160 + vma_reader_destroy(vmar);
2167 +typedef struct BackupCB {
2172 +static int backup_dump_cb(void *opaque, BlockDriverState *bs,
2173 + int64_t cluster_num, unsigned char *buf)
2175 + BackupCB *bcb = opaque;
2177 + if (vma_writer_write(bcb->vmaw, bcb->dev_id, cluster_num, buf, &zb) < 0) {
2178 + g_warning("backup_dump_cb vma_writer_write failed");
2185 +static void backup_complete_cb(void *opaque, int ret)
2187 + BackupCB *bcb = opaque;
2190 + vma_writer_set_error(bcb->vmaw, "backup_complete_cb %d", ret);
2193 + if (vma_writer_close_stream(bcb->vmaw, bcb->dev_id) <= 0) {
2194 + Error *err = NULL;
2195 + if (vma_writer_close(bcb->vmaw, &err) != 0) {
2196 + g_warning("vma_writer_close failed %s", error_get_pretty(err));
2201 +static int create_archive(int argc, char **argv)
2205 + const char *archivename;
2206 + GList *config_files = NULL;
2209 + c = getopt(argc, argv, "hvc:");
2219 + config_files = g_list_append(config_files, optarg);
2225 + g_assert_not_reached();
2230 + /* make sure we have archive name and at least one path */
2231 + if ((optind + 2) > argc) {
2235 + archivename = argv[optind++];
2238 + uuid_generate(uuid);
2240 + Error *local_err = NULL;
2241 + VmaWriter *vmaw = vma_writer_create(archivename, uuid, 0, &local_err);
2243 + if (vmaw == NULL) {
2244 + g_error("%s", error_get_pretty(local_err));
2247 + GList *l = config_files;
2248 + while (l && l->data) {
2249 + char *name = l->data;
2250 + char *cdata = NULL;
2252 + GError *err = NULL;
2253 + if (!g_file_get_contents(name, &cdata, &clen, &err)) {
2254 + unlink(archivename);
2255 + g_error("Unable to read file: %s", err->message);
2258 + if (vma_writer_add_config(vmaw, name, cdata, clen) != 0) {
2259 + unlink(archivename);
2260 + g_error("Unable to append config data %s (len = %zd)",
2263 + l = g_list_next(l);
2267 + while (optind < argc) {
2268 + const char *path = argv[optind++];
2269 + char *devname = NULL;
2270 + path = extract_devname(path, &devname, ind++);
2272 + BlockDriver *drv = NULL;
2273 + BlockDriverState *bs = bdrv_new(devname);
2275 + res = bdrv_open(bs, path, BDRV_O_CACHE_WB , drv);
2277 + unlink(archivename);
2278 + g_error("bdrv_open '%s' failed", path);
2280 + int64_t size = bdrv_getlength(bs);
2281 + int dev_id = vma_writer_register_stream(vmaw, devname, size);
2282 + if (dev_id <= 0) {
2283 + unlink(archivename);
2284 + g_error("vma_writer_register_stream '%s' failed", devname);
2287 + BackupCB *bcb = g_new0(BackupCB, 1);
2289 + bcb->dev_id = dev_id;
2291 + if (backup_job_start(bs, backup_dump_cb, backup_complete_cb, bcb) < 0) {
2292 + unlink(archivename);
2293 + g_error("backup_job_start failed");
2297 + VmaStatus vmastat;
2299 + int last_percent = -1;
2302 + main_loop_wait(false);
2303 + vma_writer_get_status(vmaw, &vmastat);
2307 + uint64_t total = 0;
2308 + uint64_t transferred = 0;
2309 + uint64_t zero_bytes = 0;
2312 + for (i = 0; i < 256; i++) {
2313 + if (vmastat.stream_info[i].size) {
2314 + total += vmastat.stream_info[i].size;
2315 + transferred += vmastat.stream_info[i].transferred;
2316 + zero_bytes += vmastat.stream_info[i].zero_bytes;
2319 + percent = (transferred*100)/total;
2320 + if (percent != last_percent) {
2321 + printf("progress %d%% %zd/%zd %zd\n", percent,
2322 + transferred, total, zero_bytes);
2324 + last_percent = percent;
2328 + if (vmastat.closed) {
2335 + vma_writer_get_status(vmaw, &vmastat);
2338 + for (i = 0; i < 256; i++) {
2339 + VmaStreamInfo *si = &vmastat.stream_info[i];
2341 + printf("image %s: size=%zd zeros=%zd saved=%zd\n", si->devname,
2342 + si->size, si->zero_bytes, si->size - si->zero_bytes);
2347 + if (vmastat.status < 0) {
2348 + unlink(archivename);
2349 + g_error("creating vma archive failed");
2355 +int main(int argc, char **argv)
2357 + const char *cmdname;
2359 + error_set_progname(argv[0]);
2361 + qemu_init_main_loop();
2369 + cmdname = argv[1];
2373 + if (!strcmp(cmdname, "list")) {
2374 + return list_content(argc, argv);
2375 + } else if (!strcmp(cmdname, "create")) {
2376 + return create_archive(argc, argv);
2377 + } else if (!strcmp(cmdname, "extract")) {
2378 + return extract_content(argc, argv);
2384 diff --git a/vma.h b/vma.h
2385 new file mode 100644
2386 index 0000000..10800a1
2391 + * VMA: Virtual Machine Archive
2393 + * Copyright (C) Proxmox Server Solutions
2396 + * Dietmar Maurer (dietmar@proxmox.com)
2398 + * This work is licensed under the terms of the GNU GPL, version 2 or later.
2399 + * See the COPYING file in the top-level directory.
2403 +#ifndef BACKUP_VMA_H
2404 +#define BACKUP_VMA_H
2406 +#include "backup.h"
2409 +#define VMA_BLOCK_BITS 12
2410 +#define VMA_BLOCK_SIZE (1<<VMA_BLOCK_BITS)
2411 +#define VMA_CLUSTER_BITS (VMA_BLOCK_BITS+4)
2412 +#define VMA_CLUSTER_SIZE (1<<VMA_CLUSTER_BITS)
2414 +#if VMA_CLUSTER_SIZE != 65536
2415 +#error unexpected cluster size
2418 +#define VMA_EXTENT_HEADER_SIZE 512
2419 +#define VMA_BLOCKS_PER_EXTENT 59
2420 +#define VMA_MAX_CONFIGS 256
2422 +#define VMA_MAX_EXTENT_SIZE \
2423 + (VMA_EXTENT_HEADER_SIZE+VMA_CLUSTER_SIZE*VMA_BLOCKS_PER_EXTENT)
2424 +#if VMA_MAX_EXTENT_SIZE != 3867136
2425 +#error unexpected VMA_EXTENT_SIZE
2428 +/* File Format Definitions */
2430 +#define VMA_MAGIC (GUINT32_TO_BE(('V'<<24)|('M'<<16)|('A'<<8)|0x00))
2431 +#define VMA_EXTENT_MAGIC (GUINT32_TO_BE(('V'<<24)|('M'<<16)|('A'<<8)|'E'))
2433 +typedef struct VmaDeviceInfoHeader {
2434 + uint32_t devname_ptr; /* offset into blob_buffer table */
2435 + uint32_t reserved0;
2436 + uint64_t size; /* device size in bytes */
2437 + uint64_t reserved1;
2438 + uint64_t reserved2;
2439 +} VmaDeviceInfoHeader;
2441 +typedef struct VmaHeader {
2444 + unsigned char uuid[16];
2446 + unsigned char md5sum[16];
2448 + uint32_t blob_buffer_offset;
2449 + uint32_t blob_buffer_size;
2450 + uint32_t header_size;
2452 + unsigned char reserved[1984];
2454 + uint32_t config_names[VMA_MAX_CONFIGS]; /* offset into blob_buffer table */
2455 + uint32_t config_data[VMA_MAX_CONFIGS]; /* offset into blob_buffer table */
2457 + VmaDeviceInfoHeader dev_info[256];
2460 +typedef struct VmaExtentHeader {
2462 + uint16_t reserved1;
2463 + uint16_t block_count;
2464 + unsigned char uuid[16];
2465 + unsigned char md5sum[16];
2466 + uint64_t blockinfo[VMA_BLOCKS_PER_EXTENT];
2469 +/* functions/definitions to read/write vma files */
2471 +typedef struct VmaReader VmaReader;
2473 +typedef struct VmaWriter VmaWriter;
2475 +typedef struct VmaConfigData {
2481 +typedef struct VmaStreamInfo {
2483 + uint64_t cluster_count;
2484 + uint64_t transferred;
2485 + uint64_t zero_bytes;
2490 +typedef struct VmaStatus {
2493 + char errmsg[8192];
2494 + char uuid_str[37];
2495 + VmaStreamInfo stream_info[256];
2498 +typedef struct VmaDeviceInfo {
2499 + uint64_t size; /* device size in bytes */
2500 + const char *devname;
2503 +extern const BackupDriver backup_vma_driver;
2505 +VmaWriter *vma_writer_create(const char *filename, uuid_t uuid, int64_t speed,
2507 +int vma_writer_close(VmaWriter *vmaw, Error **errp);
2508 +void vma_writer_destroy(VmaWriter *vmaw);
2509 +int vma_writer_add_config(VmaWriter *vmaw, const char *name, gpointer data,
2511 +int vma_writer_register_stream(VmaWriter *vmaw, const char *devname,
2514 +int64_t coroutine_fn vma_writer_write(VmaWriter *vmaw, uint8_t dev_id,
2515 + int64_t cluster_num, unsigned char *buf,
2516 + size_t *zero_bytes);
2518 +int coroutine_fn vma_writer_close_stream(VmaWriter *vmaw, uint8_t dev_id);
2520 +int vma_writer_get_status(VmaWriter *vmaw, VmaStatus *status);
2521 +void vma_writer_set_error(VmaWriter *vmaw, const char *fmt, ...);
2524 +VmaReader *vma_reader_create(const char *filename, Error **errp);
2525 +void vma_reader_destroy(VmaReader *vmar);
2526 +VmaHeader *vma_reader_get_header(VmaReader *vmar);
2527 +GList *vma_reader_get_config_data(VmaReader *vmar);
2528 +VmaDeviceInfo *vma_reader_get_device_info(VmaReader *vmar, guint8 dev_id);
2529 +int vma_reader_register_bs(VmaReader *vmar, guint8 dev_id,
2530 + BlockDriverState *bs, bool write_zeroes,
2532 +int vma_reader_restore(VmaReader *vmar, int vmstate_fd, Error **errp);
2534 +#endif /* BACKUP_VMA_H */