1 From e85ba881045bcd48a285d17234ebe32d10735492 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 | 801 +++++++++++++++++++++++++++++++++++++++++
15 vma-writer.c | 920 +++++++++++++++++++++++++++++++++++++++++++++++
16 vma.c | 558 ++++++++++++++++++++++++++++
18 8 files changed, 2456 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 e89aa14..6764e73 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..b6a550b
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;
174 + /* to show restore percentage if run with -v */
176 + int64_t cluster_count;
177 + int64_t clusters_read;
178 + int clusters_read_per;
182 +g_int32_hash(gconstpointer v)
184 + return *(const uint32_t *)v;
188 +g_int32_equal(gconstpointer v1, gconstpointer v2)
190 + return *((const uint32_t *)v1) == *((const uint32_t *)v2);
193 +static int vma_reader_get_bitmap(VmaRestoreState *rstate, int64_t cluster_num)
196 + assert(rstate->bitmap);
198 + unsigned long val, idx, bit;
200 + idx = cluster_num / BITS_PER_LONG;
202 + assert(rstate->bitmap_size > idx);
204 + bit = cluster_num % BITS_PER_LONG;
205 + val = rstate->bitmap[idx];
207 + return !!(val & (1UL << bit));
210 +static void vma_reader_set_bitmap(VmaRestoreState *rstate, int64_t cluster_num,
214 + assert(rstate->bitmap);
216 + unsigned long val, idx, bit;
218 + idx = cluster_num / BITS_PER_LONG;
220 + assert(rstate->bitmap_size > idx);
222 + bit = cluster_num % BITS_PER_LONG;
223 + val = rstate->bitmap[idx];
225 + if (!(val & (1UL << bit))) {
229 + if (val & (1UL << bit)) {
230 + val &= ~(1UL << bit);
233 + rstate->bitmap[idx] = val;
236 +typedef struct VmaBlob {
242 +static const VmaBlob *get_header_blob(VmaReader *vmar, uint32_t pos)
245 + assert(vmar->blob_hash);
247 + return g_hash_table_lookup(vmar->blob_hash, &pos);
250 +static const char *get_header_str(VmaReader *vmar, uint32_t pos)
252 + const VmaBlob *blob = get_header_blob(vmar, pos);
256 + const char *res = (char *)blob->data;
257 + if (res[blob->len-1] != '\0') {
264 +safe_read(int fd, unsigned char *buf, size_t count)
269 + n = read(fd, buf, count);
270 + } while (n < 0 && errno == EINTR);
276 +full_read(int fd, unsigned char *buf, size_t len)
284 + n = safe_read(fd, buf, len);
306 +void vma_reader_destroy(VmaReader *vmar)
310 + if (vmar->fd >= 0) {
314 + if (vmar->cdata_list) {
315 + g_list_free(vmar->cdata_list);
319 + for (i = 1; i < 256; i++) {
320 + if (vmar->rstate[i].bitmap) {
321 + g_free(vmar->rstate[i].bitmap);
325 + if (vmar->md5csum) {
326 + g_checksum_free(vmar->md5csum);
329 + if (vmar->blob_hash) {
330 + g_hash_table_destroy(vmar->blob_hash);
333 + if (vmar->head_data) {
334 + g_free(vmar->head_data);
341 +static int vma_reader_read_head(VmaReader *vmar, Error **errp)
345 + assert(*errp == NULL);
347 + unsigned char md5sum[16];
351 + vmar->head_data = g_malloc(sizeof(VmaHeader));
353 + if (full_read(vmar->fd, vmar->head_data, sizeof(VmaHeader)) !=
354 + sizeof(VmaHeader)) {
355 + error_setg(errp, "can't read vma header - %s",
356 + errno ? strerror(errno) : "got EOF");
360 + VmaHeader *h = (VmaHeader *)vmar->head_data;
362 + if (h->magic != VMA_MAGIC) {
363 + error_setg(errp, "not a vma file - wrong magic number");
367 + uint32_t header_size = GUINT32_FROM_BE(h->header_size);
368 + int need = header_size - sizeof(VmaHeader);
370 + error_setg(errp, "wrong vma header size %d", header_size);
374 + vmar->head_data = g_realloc(vmar->head_data, header_size);
375 + h = (VmaHeader *)vmar->head_data;
377 + if (full_read(vmar->fd, vmar->head_data + sizeof(VmaHeader), need) !=
379 + error_setg(errp, "can't read vma header data - %s",
380 + errno ? strerror(errno) : "got EOF");
384 + memcpy(md5sum, h->md5sum, 16);
385 + memset(h->md5sum, 0, 16);
387 + g_checksum_reset(vmar->md5csum);
388 + g_checksum_update(vmar->md5csum, vmar->head_data, header_size);
390 + g_checksum_get_digest(vmar->md5csum, (guint8 *)(h->md5sum), &csize);
392 + if (memcmp(md5sum, h->md5sum, 16) != 0) {
393 + error_setg(errp, "wrong vma header chechsum");
397 + /* we can modify header data after checksum verify */
398 + h->header_size = header_size;
400 + h->version = GUINT32_FROM_BE(h->version);
401 + if (h->version != 1) {
402 + error_setg(errp, "wrong vma version %d", h->version);
406 + h->ctime = GUINT64_FROM_BE(h->ctime);
407 + h->blob_buffer_offset = GUINT32_FROM_BE(h->blob_buffer_offset);
408 + h->blob_buffer_size = GUINT32_FROM_BE(h->blob_buffer_size);
410 + uint32_t bstart = h->blob_buffer_offset + 1;
411 + uint32_t bend = h->blob_buffer_offset + h->blob_buffer_size;
413 + if (bstart <= sizeof(VmaHeader)) {
414 + error_setg(errp, "wrong vma blob buffer offset %d",
415 + h->blob_buffer_offset);
419 + if (bend > header_size) {
420 + error_setg(errp, "wrong vma blob buffer size %d/%d",
421 + h->blob_buffer_offset, h->blob_buffer_size);
425 + while ((bstart + 2) <= bend) {
426 + uint32_t size = vmar->head_data[bstart] +
427 + (vmar->head_data[bstart+1] << 8);
428 + if ((bstart + size + 2) <= bend) {
429 + VmaBlob *blob = g_new0(VmaBlob, 1);
430 + blob->start = bstart - h->blob_buffer_offset;
432 + blob->data = vmar->head_data + bstart + 2;
433 + g_hash_table_insert(vmar->blob_hash, &blob->start, blob);
435 + bstart += size + 2;
440 + for (i = 1; i < 256; i++) {
441 + VmaDeviceInfoHeader *dih = &h->dev_info[i];
442 + uint32_t devname_ptr = GUINT32_FROM_BE(dih->devname_ptr);
443 + uint64_t size = GUINT64_FROM_BE(dih->size);
444 + const char *devname = get_header_str(vmar, devname_ptr);
446 + if (size && devname) {
448 + vmar->devinfo[i].size = size;
449 + vmar->devinfo[i].devname = devname;
451 + if (strcmp(devname, "vmstate") == 0) {
452 + vmar->vmstate_stream = i;
458 + error_setg(errp, "vma does not contain data");
462 + for (i = 0; i < VMA_MAX_CONFIGS; i++) {
463 + uint32_t name_ptr = GUINT32_FROM_BE(h->config_names[i]);
464 + uint32_t data_ptr = GUINT32_FROM_BE(h->config_data[i]);
466 + if (!(name_ptr && data_ptr)) {
469 + const char *name = get_header_str(vmar, name_ptr);
470 + const VmaBlob *blob = get_header_blob(vmar, data_ptr);
472 + if (!(name && blob)) {
473 + error_setg(errp, "vma contains invalid data pointers");
477 + VmaConfigData *cdata = g_new0(VmaConfigData, 1);
478 + cdata->name = name;
479 + cdata->data = blob->data;
480 + cdata->len = blob->len;
482 + vmar->cdata_list = g_list_append(vmar->cdata_list, cdata);
488 +VmaReader *vma_reader_create(const char *filename, Error **errp)
493 + VmaReader *vmar = g_new0(VmaReader, 1);
495 + if (strcmp(filename, "-") == 0) {
498 + vmar->fd = open(filename, O_RDONLY);
501 + if (vmar->fd < 0) {
502 + error_setg(errp, "can't open file %s - %s\n", filename,
507 + vmar->md5csum = g_checksum_new(G_CHECKSUM_MD5);
508 + if (!vmar->md5csum) {
509 + error_setg(errp, "can't allocate cmsum\n");
513 + vmar->blob_hash = g_hash_table_new_full(g_int32_hash, g_int32_equal,
516 + if (vma_reader_read_head(vmar, errp) < 0) {
524 + vma_reader_destroy(vmar);
530 +VmaHeader *vma_reader_get_header(VmaReader *vmar)
533 + assert(vmar->head_data);
535 + return (VmaHeader *)(vmar->head_data);
538 +GList *vma_reader_get_config_data(VmaReader *vmar)
541 + assert(vmar->head_data);
543 + return vmar->cdata_list;
546 +VmaDeviceInfo *vma_reader_get_device_info(VmaReader *vmar, guint8 dev_id)
551 + if (vmar->devinfo[dev_id].size && vmar->devinfo[dev_id].devname) {
552 + return &vmar->devinfo[dev_id];
558 +int vma_reader_register_bs(VmaReader *vmar, guint8 dev_id, BlockDriverState *bs,
559 + bool write_zeroes, Error **errp)
562 + assert(bs != NULL);
564 + assert(vmar->rstate[dev_id].bs == NULL);
566 + int64_t size = bdrv_getlength(bs);
567 + if (size != vmar->devinfo[dev_id].size) {
568 + error_setg(errp, "vma_reader_register_bs for stream %s failed - "
569 + "unexpected size %zd != %zd", vmar->devinfo[dev_id].devname,
570 + size, vmar->devinfo[dev_id].size);
574 + vmar->rstate[dev_id].bs = bs;
575 + vmar->rstate[dev_id].write_zeroes = write_zeroes;
577 + int64_t bitmap_size = (size/BDRV_SECTOR_SIZE) +
578 + (VMA_CLUSTER_SIZE/BDRV_SECTOR_SIZE) * BITS_PER_LONG - 1;
579 + bitmap_size /= (VMA_CLUSTER_SIZE/BDRV_SECTOR_SIZE) * BITS_PER_LONG;
581 + vmar->rstate[dev_id].bitmap_size = bitmap_size;
582 + vmar->rstate[dev_id].bitmap = g_new0(unsigned long, bitmap_size);
584 + vmar->cluster_count += size/VMA_CLUSTER_SIZE;
589 +static ssize_t safe_write(int fd, void *buf, size_t count)
594 + n = write(fd, buf, count);
595 + } while (n < 0 && errno == EINTR);
600 +static size_t full_write(int fd, void *buf, size_t len)
608 + n = safe_write(fd, buf, len);
618 + /* incomplete write ? */
625 +static int restore_write_data(VmaReader *vmar, guint8 dev_id,
626 + BlockDriverState *bs, int vmstate_fd,
627 + unsigned char *buf, int64_t sector_num,
628 + int nb_sectors, Error **errp)
632 + if (dev_id == vmar->vmstate_stream) {
633 + if (vmstate_fd >= 0) {
634 + int len = nb_sectors * BDRV_SECTOR_SIZE;
635 + int res = full_write(vmstate_fd, buf, len);
637 + error_setg(errp, "write vmstate failed %d", res);
642 + int res = bdrv_write(bs, sector_num, buf, nb_sectors);
644 + error_setg(errp, "bdrv_write to %s failed (%d)",
645 + bdrv_get_device_name(bs), res);
651 +static int restore_extent(VmaReader *vmar, unsigned char *buf,
652 + int extent_size, int vmstate_fd,
653 + bool verbose, Error **errp)
658 + VmaExtentHeader *ehead = (VmaExtentHeader *)buf;
659 + int start = VMA_EXTENT_HEADER_SIZE;
662 + for (i = 0; i < VMA_BLOCKS_PER_EXTENT; i++) {
663 + uint64_t block_info = GUINT64_FROM_BE(ehead->blockinfo[i]);
664 + uint64_t cluster_num = block_info & 0xffffffff;
665 + uint8_t dev_id = (block_info >> 32) & 0xff;
666 + uint16_t mask = block_info >> (32+16);
667 + int64_t max_sector;
673 + VmaRestoreState *rstate = &vmar->rstate[dev_id];
674 + BlockDriverState *bs = NULL;
676 + if (dev_id != vmar->vmstate_stream) {
679 + error_setg(errp, "got wrong dev id %d", dev_id);
683 + if (vma_reader_get_bitmap(rstate, cluster_num)) {
684 + error_setg(errp, "found duplicated cluster %zd for stream %s",
685 + cluster_num, vmar->devinfo[dev_id].devname);
688 + vma_reader_set_bitmap(rstate, cluster_num, 1);
690 + max_sector = vmar->devinfo[dev_id].size/BDRV_SECTOR_SIZE;
692 + max_sector = G_MAXINT64;
693 + if (cluster_num != vmar->vmstate_clusters) {
694 + error_setg(errp, "found out of order vmstate data");
697 + vmar->vmstate_clusters++;
700 + vmar->clusters_read++;
703 + time_t duration = time(NULL) - vmar->start_time;
704 + int percent = (vmar->clusters_read*100)/vmar->cluster_count;
705 + if (percent != vmar->clusters_read_per) {
706 + printf("progress %d%% (read %zd bytes, duration %zd sec)\n",
707 + percent, vmar->clusters_read*VMA_CLUSTER_SIZE,
710 + vmar->clusters_read_per = percent;
714 + /* try to write whole clusters to speedup restore */
715 + if (mask == 0xffff) {
716 + if ((start + VMA_CLUSTER_SIZE) > extent_size) {
717 + error_setg(errp, "short vma extent - too many blocks");
720 + int64_t sector_num = (cluster_num * VMA_CLUSTER_SIZE) /
722 + int64_t end_sector = sector_num +
723 + VMA_CLUSTER_SIZE/BDRV_SECTOR_SIZE;
725 + if (end_sector > max_sector) {
726 + end_sector = max_sector;
729 + if (end_sector <= sector_num) {
730 + error_setg(errp, "got wrong block address - write bejond end");
734 + int nb_sectors = end_sector - sector_num;
735 + if (restore_write_data(vmar, dev_id, bs, vmstate_fd, buf + start,
736 + sector_num, nb_sectors, errp) < 0) {
740 + start += VMA_CLUSTER_SIZE;
745 + for (j = 0; j < 16; j++) {
746 + int64_t sector_num = (cluster_num*VMA_CLUSTER_SIZE +
747 + j*VMA_BLOCK_SIZE)/BDRV_SECTOR_SIZE;
749 + int64_t end_sector = sector_num +
750 + VMA_BLOCK_SIZE/BDRV_SECTOR_SIZE;
751 + if (end_sector > max_sector) {
752 + end_sector = max_sector;
756 + if ((start + VMA_BLOCK_SIZE) > extent_size) {
757 + error_setg(errp, "short vma extent - too many blocks");
761 + if (end_sector <= sector_num) {
762 + error_setg(errp, "got wrong block address - "
763 + "write bejond end");
767 + int nb_sectors = end_sector - sector_num;
768 + if (restore_write_data(vmar, dev_id, bs, vmstate_fd,
769 + buf + start, sector_num,
770 + nb_sectors, errp) < 0) {
774 + start += VMA_BLOCK_SIZE;
778 + if (rstate->write_zeroes && (end_sector > sector_num)) {
779 + /* Todo: use bdrv_co_write_zeroes (but that need to
780 + * be run inside coroutine?)
782 + int nb_sectors = end_sector - sector_num;
783 + if (restore_write_data(vmar, dev_id, bs, vmstate_fd,
784 + zero_vma_block, sector_num,
785 + nb_sectors, errp) < 0) {
796 + if (start != extent_size) {
797 + error_setg(errp, "vma extent error - missing blocks");
804 +int vma_reader_restore(VmaReader *vmar, int vmstate_fd, bool verbose,
808 + assert(vmar->head_data);
811 + unsigned char buf[VMA_MAX_EXTENT_SIZE];
813 + unsigned char md5sum[16];
814 + VmaHeader *h = (VmaHeader *)vmar->head_data;
816 + vmar->start_time = time(NULL);
819 + int bytes = full_read(vmar->fd, buf + buf_pos, sizeof(buf) - buf_pos);
821 + error_setg(errp, "read failed - %s", strerror(errno));
831 + if (buf_pos < VMA_EXTENT_HEADER_SIZE) {
832 + error_setg(errp, "read short extent (%d bytes)", buf_pos);
836 + VmaExtentHeader *ehead = (VmaExtentHeader *)buf;
838 + /* extract md5sum */
839 + memcpy(md5sum, ehead->md5sum, sizeof(ehead->md5sum));
840 + memset(ehead->md5sum, 0, sizeof(ehead->md5sum));
842 + g_checksum_reset(vmar->md5csum);
843 + g_checksum_update(vmar->md5csum, buf, VMA_EXTENT_HEADER_SIZE);
845 + g_checksum_get_digest(vmar->md5csum, ehead->md5sum, &csize);
847 + if (memcmp(md5sum, ehead->md5sum, 16) != 0) {
848 + error_setg(errp, "wrong vma extent header chechsum");
852 + if (memcmp(h->uuid, ehead->uuid, sizeof(ehead->uuid)) != 0) {
853 + error_setg(errp, "wrong vma extent uuid");
857 + if (ehead->magic != VMA_EXTENT_MAGIC || ehead->reserved1 != 0) {
858 + error_setg(errp, "wrong vma extent header magic");
862 + int block_count = GUINT16_FROM_BE(ehead->block_count);
863 + int extent_size = VMA_EXTENT_HEADER_SIZE + block_count*VMA_BLOCK_SIZE;
865 + if (buf_pos < extent_size) {
866 + error_setg(errp, "short vma extent (%d < %d)", buf_pos,
871 + if (restore_extent(vmar, buf, extent_size, vmstate_fd, verbose,
876 + if (buf_pos > extent_size) {
877 + memmove(buf, buf + extent_size, buf_pos - extent_size);
878 + buf_pos = buf_pos - extent_size;
887 + for (i = 1; i < 256; i++) {
888 + VmaRestoreState *rstate = &vmar->rstate[i];
893 + if (bdrv_flush(rstate->bs) < 0) {
894 + error_setg(errp, "vma bdrv_flush %s failed",
895 + vmar->devinfo[i].devname);
899 + if (vmar->devinfo[i].size &&
900 + (strcmp(vmar->devinfo[i].devname, "vmstate") != 0)) {
901 + assert(rstate->bitmap);
903 + int64_t cluster_num, end;
905 + end = (vmar->devinfo[i].size + VMA_CLUSTER_SIZE - 1) /
908 + for (cluster_num = 0; cluster_num < end; cluster_num++) {
909 + if (!vma_reader_get_bitmap(rstate, cluster_num)) {
910 + error_setg(errp, "detected missing cluster %zd "
911 + "for stream %s", cluster_num,
912 + vmar->devinfo[i].devname);
922 diff --git a/vma-writer.c b/vma-writer.c
924 index 0000000..c1e1afe
929 + * VMA: Virtual Machine Archive
931 + * Copyright (C) 2012 Proxmox Server Solutions
934 + * Dietmar Maurer (dietmar@proxmox.com)
936 + * This work is licensed under the terms of the GNU GPL, version 2 or later.
937 + * See the COPYING file in the top-level directory.
946 +#include <sys/types.h>
947 +#include <sys/stat.h>
950 +#include <uuid/uuid.h>
952 +#include "qemu-common.h"
953 +#include "qemu_socket.h"
954 +#include "qemu-coroutine.h"
955 +#include "qemu-aio.h"
956 +#include "qemu/ratelimit.h"
959 +#include "monitor.h"
963 +#define DPRINTF(fmt, ...)\
964 + do { if (DEBUG_VMA) { printf("vma: " fmt, ## __VA_ARGS__); } } while (0)
966 +#define WRITE_BUFFERS 5
968 +typedef struct VmaAIOCB VmaAIOCB;
971 + unsigned char buffer[VMA_MAX_EXTENT_SIZE];
982 + bool header_written;
985 + /* we always write extents */
986 + unsigned char outbuf[VMA_MAX_EXTENT_SIZE];
987 + int outbuf_pos; /* in bytes */
988 + int outbuf_count; /* in VMA_BLOCKS */
989 + uint64_t outbuf_block_info[VMA_BLOCKS_PER_EXTENT];
991 + VmaAIOCB aiocbs[WRITE_BUFFERS];
994 + GChecksum *md5csum;
995 + CoMutex writer_lock;
996 + CoMutex flush_lock;
997 + Coroutine *co_writer;
1001 + /* drive informations */
1002 + VmaStreamInfo stream_info[256];
1003 + guint stream_count;
1005 + guint8 vmstate_stream;
1006 + uint32_t vmstate_clusters;
1008 + /* header blob table */
1009 + char *header_blob_table;
1010 + uint32_t header_blob_table_size;
1011 + uint32_t header_blob_table_pos;
1013 + /* store for config blobs */
1014 + uint32_t config_names[VMA_MAX_CONFIGS]; /* offset into blob_buffer table */
1015 + uint32_t config_data[VMA_MAX_CONFIGS]; /* offset into blob_buffer table */
1016 + uint32_t config_count;
1019 +void vma_writer_set_error(VmaWriter *vmaw, const char *fmt, ...)
1023 + if (vmaw->status < 0) {
1027 + vmaw->status = -1;
1029 + va_start(ap, fmt);
1030 + g_vsnprintf(vmaw->errmsg, sizeof(vmaw->errmsg), fmt, ap);
1033 + DPRINTF("vma_writer_set_error: %s\n", vmaw->errmsg);
1036 +static uint32_t allocate_header_blob(VmaWriter *vmaw, const char *data,
1039 + if (len > 65535) {
1043 + if (!vmaw->header_blob_table ||
1044 + (vmaw->header_blob_table_size <
1045 + (vmaw->header_blob_table_pos + len + 2))) {
1046 + int newsize = vmaw->header_blob_table_size + ((len + 2 + 511)/512)*512;
1048 + vmaw->header_blob_table = g_realloc(vmaw->header_blob_table, newsize);
1049 + memset(vmaw->header_blob_table + vmaw->header_blob_table_size,
1050 + 0, newsize - vmaw->header_blob_table_size);
1051 + vmaw->header_blob_table_size = newsize;
1054 + uint32_t cpos = vmaw->header_blob_table_pos;
1055 + vmaw->header_blob_table[cpos] = len & 255;
1056 + vmaw->header_blob_table[cpos+1] = (len >> 8) & 255;
1057 + memcpy(vmaw->header_blob_table + cpos + 2, data, len);
1058 + vmaw->header_blob_table_pos += len + 2;
1062 +static uint32_t allocate_header_string(VmaWriter *vmaw, const char *str)
1066 + size_t len = strlen(str) + 1;
1068 + return allocate_header_blob(vmaw, str, len);
1071 +int vma_writer_add_config(VmaWriter *vmaw, const char *name, gpointer data,
1075 + assert(!vmaw->header_written);
1076 + assert(vmaw->config_count < VMA_MAX_CONFIGS);
1081 + uint32_t name_ptr = allocate_header_string(vmaw, name);
1086 + uint32_t data_ptr = allocate_header_blob(vmaw, data, len);
1091 + vmaw->config_names[vmaw->config_count] = name_ptr;
1092 + vmaw->config_data[vmaw->config_count] = data_ptr;
1094 + vmaw->config_count++;
1099 +int vma_writer_register_stream(VmaWriter *vmaw, const char *devname,
1104 + assert(!vmaw->status);
1106 + if (vmaw->header_written) {
1107 + vma_writer_set_error(vmaw, "vma_writer_register_stream: header "
1108 + "already written");
1112 + guint n = vmaw->stream_count + 1;
1114 + /* we can have dev_ids form 1 to 255 (0 reserved)
1115 + * 255(-1) reseverd for safety
1118 + vma_writer_set_error(vmaw, "vma_writer_register_stream: "
1119 + "too many drives");
1124 + vma_writer_set_error(vmaw, "vma_writer_register_stream: "
1125 + "got strange size %zd", size);
1129 + DPRINTF("vma_writer_register_stream %s %zu %d\n", devname, size, n);
1131 + vmaw->stream_info[n].devname = g_strdup(devname);
1132 + vmaw->stream_info[n].size = size;
1134 + vmaw->stream_info[n].cluster_count = (size + VMA_CLUSTER_SIZE - 1) /
1137 + vmaw->stream_count = n;
1139 + if (strcmp(devname, "vmstate") == 0) {
1140 + vmaw->vmstate_stream = n;
1146 +static void vma_co_continue_write(void *opaque)
1148 + VmaWriter *vmaw = opaque;
1150 + qemu_aio_set_fd_handler(vmaw->fd, NULL, NULL, NULL, NULL);
1152 + DPRINTF("vma_co_continue_write\n");
1153 + qemu_coroutine_enter(vmaw->co_writer, NULL);
1156 +static ssize_t coroutine_fn
1157 +vma_co_write(VmaWriter *vmaw, const void *buf, size_t bytes)
1162 + /* atomic writes (we cannot interleave writes) */
1163 + qemu_co_mutex_lock(&vmaw->writer_lock);
1165 + DPRINTF("vma_co_write enter %zd\n", bytes);
1167 + while (done < bytes) {
1168 + ret = write(vmaw->fd, buf + done, bytes - done);
1171 + DPRINTF("vma_co_write written %zd %zd\n", done, ret);
1172 + } else if (ret < 0) {
1173 + if (errno == EAGAIN || errno == EWOULDBLOCK) {
1174 + DPRINTF("vma_co_write yield %zd\n", done);
1176 + vmaw->co_writer = qemu_coroutine_self();
1177 + qemu_aio_set_fd_handler(vmaw->fd, NULL, vma_co_continue_write,
1180 + qemu_coroutine_yield();
1181 + DPRINTF("vma_co_write restart %zd\n", done);
1183 + vma_writer_set_error(vmaw, "vma_co_write write error - %s",
1185 + done = -1; /* always return failure for partial writes */
1188 + } else if (ret == 0) {
1189 + /* should not happen - simply try again */
1193 + qemu_co_mutex_unlock(&vmaw->writer_lock);
1195 + DPRINTF("vma_co_write leave %zd\n", done);
1199 +static void coroutine_fn vma_co_writer_task(void *opaque)
1201 + VmaAIOCB *cb = opaque;
1203 + DPRINTF("vma_co_writer_task start\n");
1205 + int64_t done = vma_co_write(cb->vmaw, cb->buffer, cb->bytes);
1206 + DPRINTF("vma_co_writer_task write done %zd\n", done);
1208 + if (done != cb->bytes) {
1209 + DPRINTF("vma_co_writer_task failed write %zd %zd", cb->bytes, done);
1210 + vma_writer_set_error(cb->vmaw, "vma_co_writer_task failed write %zd",
1216 + qemu_co_queue_next(&cb->vmaw->wqueue);
1218 + DPRINTF("vma_co_writer_task end\n");
1221 +static void coroutine_fn vma_queue_flush(VmaWriter *vmaw)
1223 + DPRINTF("vma_queue_flush enter\n");
1229 + VmaAIOCB *cb = NULL;
1230 + for (i = 0; i < WRITE_BUFFERS; i++) {
1231 + if (vmaw->aiocbs[i].bytes) {
1232 + cb = &vmaw->aiocbs[i];
1233 + DPRINTF("FOUND USED AIO BUFFER %d %zd\n", i,
1234 + vmaw->aiocbs[i].bytes);
1241 + qemu_co_queue_wait(&vmaw->wqueue);
1244 + DPRINTF("vma_queue_flush leave\n");
1248 + * NOTE: pipe buffer size in only 4096 bytes on linux (see 'ulimit -a')
1249 + * So we need to create a coroutione to allow 'parallel' execution.
1251 +static ssize_t coroutine_fn
1252 +vma_queue_write(VmaWriter *vmaw, const void *buf, size_t bytes)
1254 + DPRINTF("vma_queue_write enter %zd\n", bytes);
1258 + assert(bytes <= VMA_MAX_EXTENT_SIZE);
1260 + VmaAIOCB *cb = NULL;
1263 + for (i = 0; i < WRITE_BUFFERS; i++) {
1264 + if (!vmaw->aiocbs[i].bytes) {
1265 + cb = &vmaw->aiocbs[i];
1270 + qemu_co_queue_wait(&vmaw->wqueue);
1274 + memcpy(cb->buffer, buf, bytes);
1275 + cb->bytes = bytes;
1278 + DPRINTF("vma_queue_write start %zd\n", bytes);
1279 + cb->co = qemu_coroutine_create(vma_co_writer_task);
1280 + qemu_coroutine_enter(cb->co, cb);
1282 + DPRINTF("vma_queue_write leave\n");
1287 +VmaWriter *vma_writer_create(const char *filename, uuid_t uuid, int64_t speed,
1292 + assert(sizeof(VmaHeader) == (4096 + 8192));
1293 + assert(sizeof(VmaExtentHeader) == 512);
1295 + VmaWriter *vmaw = g_new0(VmaWriter, 1);
1298 + vmaw->md5csum = g_checksum_new(G_CHECKSUM_MD5);
1299 + if (!vmaw->md5csum) {
1300 + error_setg(errp, "can't allocate cmsum\n");
1304 + if (strstart(filename, "exec:", &p)) {
1305 + vmaw->cmd = popen(p, "w");
1306 + if (vmaw->cmd == NULL) {
1307 + error_setg(errp, "can't popen command '%s' - %s\n", p,
1311 + vmaw->fd = fileno(vmaw->cmd);
1312 + socket_set_nonblock(vmaw->fd);
1317 + const char *tmp_id_str;
1319 + if ((stat(filename, &st) == 0) && S_ISFIFO(st.st_mode)) {
1320 + oflags = O_NONBLOCK|O_WRONLY;
1321 + vmaw->fd = qemu_open(filename, oflags, 0644);
1322 + } else if (strstart(filename, "/dev/fdset/", &tmp_id_str)) {
1323 + oflags = O_NONBLOCK|O_WRONLY;
1324 + vmaw->fd = qemu_open(filename, oflags, 0644);
1325 + } else if (strstart(filename, "/dev/fdname/", &tmp_id_str)) {
1326 + vmaw->fd = monitor_get_fd(cur_mon, tmp_id_str, errp);
1327 + if (vmaw->fd < 0) {
1331 + oflags = O_NONBLOCK|O_WRONLY|O_CREAT|O_EXCL;
1332 + vmaw->fd = qemu_open(filename, oflags, 0644);
1335 + if (vmaw->fd < 0) {
1336 + error_setg(errp, "can't open file %s - %s\n", filename,
1342 + vmaw->outbuf_count = 0;
1343 + vmaw->outbuf_pos = VMA_EXTENT_HEADER_SIZE;
1345 + vmaw->header_blob_table_pos = 1; /* start at pos 1 */
1347 + qemu_co_mutex_init(&vmaw->writer_lock);
1348 + qemu_co_mutex_init(&vmaw->flush_lock);
1349 + qemu_co_queue_init(&vmaw->wqueue);
1351 + uuid_copy(vmaw->uuid, uuid);
1354 + speed = 10*1024*1024*1024LLU; /* default 10GB/s */
1357 + ratelimit_set_speed(&vmaw->limit, speed, 100000000ULL /* 0.1 sec */);
1364 + pclose(vmaw->cmd);
1365 + } else if (vmaw->fd >= 0) {
1369 + if (vmaw->md5csum) {
1370 + g_checksum_free(vmaw->md5csum);
1379 +static int coroutine_fn vma_write_header(VmaWriter *vmaw)
1382 + int header_clusters = 8;
1383 + char buf[65536*header_clusters];
1384 + VmaHeader *head = (VmaHeader *)buf;
1388 + DPRINTF("VMA WRITE HEADER\n");
1390 + if (vmaw->status < 0) {
1391 + return vmaw->status;
1394 + memset(buf, 0, sizeof(buf));
1396 + head->magic = VMA_MAGIC;
1397 + head->version = GUINT32_TO_BE(1); /* v1 */
1398 + memcpy(head->uuid, vmaw->uuid, 16);
1400 + time_t ctime = time(NULL);
1401 + head->ctime = GUINT64_TO_BE(ctime);
1403 + if (!vmaw->stream_count) {
1407 + for (i = 0; i < VMA_MAX_CONFIGS; i++) {
1408 + head->config_names[i] = GUINT32_TO_BE(vmaw->config_names[i]);
1409 + head->config_data[i] = GUINT32_TO_BE(vmaw->config_data[i]);
1412 + /* 32 bytes per device (12 used currently) = 8192 bytes max */
1413 + for (i = 1; i <= 254; i++) {
1414 + VmaStreamInfo *si = &vmaw->stream_info[i];
1416 + assert(si->devname);
1417 + uint32_t devname_ptr = allocate_header_string(vmaw, si->devname);
1418 + if (!devname_ptr) {
1421 + head->dev_info[i].devname_ptr = GUINT32_TO_BE(devname_ptr);
1422 + head->dev_info[i].size = GUINT64_TO_BE(si->size);
1426 + uint32_t header_size = sizeof(VmaHeader) + vmaw->header_blob_table_size;
1427 + head->header_size = GUINT32_TO_BE(header_size);
1429 + if (header_size > sizeof(buf)) {
1430 + return -1; /* just to be sure */
1433 + uint32_t blob_buffer_offset = sizeof(VmaHeader);
1434 + memcpy(buf + blob_buffer_offset, vmaw->header_blob_table,
1435 + vmaw->header_blob_table_size);
1436 + head->blob_buffer_offset = GUINT32_TO_BE(blob_buffer_offset);
1437 + head->blob_buffer_size = GUINT32_TO_BE(vmaw->header_blob_table_pos);
1439 + g_checksum_reset(vmaw->md5csum);
1440 + g_checksum_update(vmaw->md5csum, (const guchar *)buf, header_size);
1442 + g_checksum_get_digest(vmaw->md5csum, (guint8 *)(head->md5sum), &csize);
1444 + return vma_queue_write(vmaw, buf, header_size);
1447 +static int coroutine_fn vma_writer_flush(VmaWriter *vmaw)
1454 + if (vmaw->status < 0) {
1455 + return vmaw->status;
1458 + if (!vmaw->header_written) {
1459 + vmaw->header_written = true;
1460 + ret = vma_write_header(vmaw);
1462 + vma_writer_set_error(vmaw, "vma_writer_flush: write header failed");
1467 + DPRINTF("VMA WRITE FLUSH %d %d\n", vmaw->outbuf_count, vmaw->outbuf_pos);
1470 + VmaExtentHeader *ehead = (VmaExtentHeader *)vmaw->outbuf;
1472 + ehead->magic = VMA_EXTENT_MAGIC;
1473 + ehead->reserved1 = 0;
1475 + for (i = 0; i < VMA_BLOCKS_PER_EXTENT; i++) {
1476 + ehead->blockinfo[i] = GUINT64_TO_BE(vmaw->outbuf_block_info[i]);
1479 + guint16 block_count = (vmaw->outbuf_pos - VMA_EXTENT_HEADER_SIZE) /
1482 + ehead->block_count = GUINT16_TO_BE(block_count);
1484 + memcpy(ehead->uuid, vmaw->uuid, sizeof(ehead->uuid));
1485 + memset(ehead->md5sum, 0, sizeof(ehead->md5sum));
1487 + g_checksum_reset(vmaw->md5csum);
1488 + g_checksum_update(vmaw->md5csum, vmaw->outbuf, VMA_EXTENT_HEADER_SIZE);
1490 + g_checksum_get_digest(vmaw->md5csum, ehead->md5sum, &csize);
1492 + int bytes = vmaw->outbuf_pos;
1493 + ret = vma_queue_write(vmaw, vmaw->outbuf, bytes);
1494 + if (ret != bytes) {
1495 + vma_writer_set_error(vmaw, "vma_writer_flush: failed write");
1498 + vmaw->outbuf_count = 0;
1499 + vmaw->outbuf_pos = VMA_EXTENT_HEADER_SIZE;
1501 + for (i = 0; i < VMA_BLOCKS_PER_EXTENT; i++) {
1502 + vmaw->outbuf_block_info[i] = 0;
1505 + return vmaw->status;
1508 +static int vma_count_open_streams(VmaWriter *vmaw)
1510 + g_assert(vmaw != NULL);
1513 + int open_drives = 0;
1514 + for (i = 0; i <= 255; i++) {
1515 + if (vmaw->stream_info[i].size && !vmaw->stream_info[i].finished) {
1520 + return open_drives;
1524 + * all jobs should call this when there is no more data
1525 + * Returns: number of remaining stream (0 ==> finished)
1528 +vma_writer_close_stream(VmaWriter *vmaw, uint8_t dev_id)
1530 + g_assert(vmaw != NULL);
1532 + DPRINTF("vma_writer_set_status %d\n", dev_id);
1533 + if (!vmaw->stream_info[dev_id].size) {
1534 + vma_writer_set_error(vmaw, "vma_writer_close_stream: "
1535 + "no such stream %d", dev_id);
1538 + if (vmaw->stream_info[dev_id].finished) {
1539 + vma_writer_set_error(vmaw, "vma_writer_close_stream: "
1540 + "stream already closed %d", dev_id);
1544 + vmaw->stream_info[dev_id].finished = true;
1546 + int open_drives = vma_count_open_streams(vmaw);
1548 + if (open_drives <= 0) {
1549 + DPRINTF("vma_writer_set_status all drives completed\n");
1550 + qemu_co_mutex_lock(&vmaw->flush_lock);
1551 + int ret = vma_writer_flush(vmaw);
1552 + qemu_co_mutex_unlock(&vmaw->flush_lock);
1554 + vma_writer_set_error(vmaw, "vma_writer_close_stream: flush failed");
1558 + return open_drives;
1561 +int vma_writer_get_status(VmaWriter *vmaw, VmaStatus *status)
1565 + g_assert(vmaw != NULL);
1568 + status->status = vmaw->status;
1569 + g_strlcpy(status->errmsg, vmaw->errmsg, sizeof(status->errmsg));
1570 + for (i = 0; i <= 255; i++) {
1571 + status->stream_info[i] = vmaw->stream_info[i];
1574 + uuid_unparse_lower(vmaw->uuid, status->uuid_str);
1577 + status->closed = vmaw->closed;
1579 + return vmaw->status;
1582 +static int vma_writer_get_buffer(VmaWriter *vmaw)
1585 + /* wait until buffer is available */
1586 + while (vmaw->outbuf_count >= (VMA_BLOCKS_PER_EXTENT - 1)) {
1589 + qemu_co_mutex_lock(&vmaw->flush_lock);
1590 + res = vma_writer_flush(vmaw);
1591 + qemu_co_mutex_unlock(&vmaw->flush_lock);
1594 + vma_writer_set_error(vmaw, "vma_writer_get_buffer: flush failed");
1603 +int64_t coroutine_fn
1604 +vma_writer_write(VmaWriter *vmaw, uint8_t dev_id, int64_t cluster_num,
1605 + unsigned char *buf, size_t *zero_bytes)
1607 + g_assert(vmaw != NULL);
1608 + g_assert(zero_bytes != NULL);
1612 + if (vmaw->status < 0) {
1613 + return vmaw->status;
1616 + if (!dev_id || !vmaw->stream_info[dev_id].size) {
1617 + vma_writer_set_error(vmaw, "vma_writer_write: "
1618 + "no such stream %d", dev_id);
1622 + if (vmaw->stream_info[dev_id].finished) {
1623 + vma_writer_set_error(vmaw, "vma_writer_write: "
1624 + "stream already closed %d", dev_id);
1629 + if (cluster_num >= (((uint64_t)1)<<32)) {
1630 + vma_writer_set_error(vmaw, "vma_writer_write: "
1631 + "cluster number out of range");
1635 + if (dev_id == vmaw->vmstate_stream) {
1636 + if (cluster_num != vmaw->vmstate_clusters) {
1637 + vma_writer_set_error(vmaw, "vma_writer_write: "
1638 + "non sequential vmstate write");
1640 + vmaw->vmstate_clusters++;
1641 + } else if (cluster_num >= vmaw->stream_info[dev_id].cluster_count) {
1642 + vma_writer_set_error(vmaw, "vma_writer_write: cluster number too big");
1646 + /* wait until buffer is available */
1647 + if (vma_writer_get_buffer(vmaw) < 0) {
1648 + vma_writer_set_error(vmaw, "vma_writer_write: "
1649 + "vma_writer_get_buffer failed");
1653 + DPRINTF("VMA WRITE %zd\n", cluster_num);
1657 + uint16_t mask = 0;
1658 + for (i = 0; i < 16; i++) {
1659 + unsigned char *vmablock = buf + (i*VMA_BLOCK_SIZE);
1660 + if (buffer_is_zero(vmablock, VMA_BLOCK_SIZE)) {
1661 + DPRINTF("VMA WRITE %zd ZERO BLOCK %d\n", cluster_num, i);
1662 + vmaw->stream_info[dev_id].zero_bytes += VMA_BLOCK_SIZE;
1663 + *zero_bytes += VMA_BLOCK_SIZE;
1666 + memcpy(vmaw->outbuf + vmaw->outbuf_pos, vmablock, VMA_BLOCK_SIZE);
1667 + vmaw->outbuf_pos += VMA_BLOCK_SIZE;
1669 + vmaw->delay_ns = ratelimit_calculate_delay(&vmaw->limit,
1671 + if (vmaw->delay_ns) {
1672 + co_sleep_ns(rt_clock, vmaw->delay_ns);
1679 + uint64_t block_info = ((uint64_t)mask) << (32+16);
1680 + block_info |= ((uint64_t)dev_id) << 32;
1681 + block_info |= (cluster_num & 0xffffffff);
1682 + vmaw->outbuf_block_info[vmaw->outbuf_count] = block_info;
1684 + DPRINTF("VMA WRITE MASK %zd %zx\n", cluster_num, block_info);
1686 + vmaw->outbuf_count++;
1688 + /** NOTE: We allways write whole clusters, but we correctly set
1689 + * transferred bytes. So transferred == size when when everything
1692 + size_t transferred = VMA_CLUSTER_SIZE;
1694 + if (dev_id != vmaw->vmstate_stream) {
1695 + uint64_t last = (cluster_num + 1) * VMA_CLUSTER_SIZE;
1696 + if (last > vmaw->stream_info[dev_id].size) {
1697 + uint64_t diff = last - vmaw->stream_info[dev_id].size;
1698 + if (diff >= VMA_CLUSTER_SIZE) {
1699 + vma_writer_set_error(vmaw, "vma_writer_write: "
1700 + "read after last cluster");
1703 + transferred -= diff;
1707 + vmaw->stream_info[dev_id].transferred += transferred;
1709 + return transferred;
1712 +int vma_writer_close(VmaWriter *vmaw, Error **errp)
1714 + g_assert(vmaw != NULL);
1718 + vma_queue_flush(vmaw);
1720 + /* this should not happen - just to be sure */
1721 + while (!qemu_co_queue_empty(&vmaw->wqueue)) {
1722 + DPRINTF("vma_writer_close wait\n");
1723 + co_sleep_ns(rt_clock, 1000000);
1727 + if (pclose(vmaw->cmd) < 0) {
1728 + vma_writer_set_error(vmaw, "vma_writer_close: "
1729 + "pclose failed - %s", strerror(errno));
1732 + if (close(vmaw->fd) < 0) {
1733 + vma_writer_set_error(vmaw, "vma_writer_close: "
1734 + "close failed - %s", strerror(errno));
1738 + for (i = 0; i <= 255; i++) {
1739 + VmaStreamInfo *si = &vmaw->stream_info[i];
1741 + if (!si->finished) {
1742 + vma_writer_set_error(vmaw, "vma_writer_close: "
1743 + "detected open stream '%s'", si->devname);
1744 + } else if ((si->transferred != si->size) &&
1745 + (i != vmaw->vmstate_stream)) {
1746 + vma_writer_set_error(vmaw, "vma_writer_close: "
1747 + "incomplete stream '%s' (%zd != %zd)",
1748 + si->devname, si->transferred, si->size);
1753 + for (i = 0; i <= 255; i++) {
1754 + vmaw->stream_info[i].finished = 1; /* mark as closed */
1759 + if (vmaw->status < 0 && *errp == NULL) {
1760 + error_setg(errp, "%s", vmaw->errmsg);
1763 + return vmaw->status;
1766 +void vma_writer_destroy(VmaWriter *vmaw)
1772 + for (i = 0; i <= 255; i++) {
1773 + if (vmaw->stream_info[i].devname) {
1774 + g_free(vmaw->stream_info[i].devname);
1778 + if (vmaw->md5csum) {
1779 + g_checksum_free(vmaw->md5csum);
1785 +/* backup driver plugin */
1787 +static int vma_dump_cb(void *opaque, uint8_t dev_id, int64_t cluster_num,
1788 + unsigned char *buf, size_t *zero_bytes)
1790 + VmaWriter *vmaw = opaque;
1792 + return vma_writer_write(vmaw, dev_id, cluster_num, buf, zero_bytes);
1795 +static int vma_close_cb(void *opaque, Error **errp)
1797 + VmaWriter *vmaw = opaque;
1799 + int res = vma_writer_close(vmaw, errp);
1800 + vma_writer_destroy(vmaw);
1805 +static int vma_complete_cb(void *opaque, uint8_t dev_id, int ret)
1807 + VmaWriter *vmaw = opaque;
1810 + vma_writer_set_error(vmaw, "backup_complete_cb %d", ret);
1813 + return vma_writer_close_stream(vmaw, dev_id);
1816 +static int vma_register_stream_cb(void *opaque, const char *devname,
1819 + VmaWriter *vmaw = opaque;
1821 + return vma_writer_register_stream(vmaw, devname, size);
1824 +static int vma_register_config_cb(void *opaque, const char *name,
1825 + gpointer data, size_t data_len)
1827 + VmaWriter *vmaw = opaque;
1829 + return vma_writer_add_config(vmaw, name, data, data_len);
1832 +static void *vma_open_cb(const char *filename, uuid_t uuid, int64_t speed,
1835 + return vma_writer_create(filename, uuid, speed, errp);
1838 +const BackupDriver backup_vma_driver = {
1840 + .open_cb = vma_open_cb,
1841 + .close_cb = vma_close_cb,
1842 + .register_config_cb = vma_register_config_cb,
1843 + .register_stream_cb = vma_register_stream_cb,
1844 + .dump_cb = vma_dump_cb,
1845 + .complete_cb = vma_complete_cb,
1848 diff --git a/vma.c b/vma.c
1849 new file mode 100644
1850 index 0000000..8d6f32d
1855 + * VMA: Virtual Machine Archive
1857 + * Copyright (C) 2012 Proxmox Server Solutions
1860 + * Dietmar Maurer (dietmar@proxmox.com)
1862 + * This work is licensed under the terms of the GNU GPL, version 2 or later.
1863 + * See the COPYING file in the top-level directory.
1869 +#include <unistd.h>
1871 +#include <string.h>
1872 +#include <sys/types.h>
1873 +#include <sys/stat.h>
1877 +#include "qemu-common.h"
1878 +#include "qemu-option.h"
1879 +#include "qemu-error.h"
1881 +#include "sysemu.h"
1882 +#include "block_int.h"
1886 +static void help(void)
1888 + const char *help_msg =
1889 + "usage: vma command [command options]\n"
1891 + "vma list <filename>\n"
1892 + "vma create <filename> [-c config] <archive> pathname ...\n"
1893 + "vma extract <filename> [-r] <targetdir>\n"
1896 + printf("%s", help_msg);
1900 +static const char *extract_devname(const char *path, char **devname, int index)
1904 + const char *sep = strchr(path, '=');
1907 + *devname = g_strndup(path, sep - path);
1911 + *devname = g_strdup_printf("disk%d", index);
1920 +static void print_content(VmaReader *vmar)
1924 + VmaHeader *head = vma_reader_get_header(vmar);
1926 + GList *l = vma_reader_get_config_data(vmar);
1927 + while (l && l->data) {
1928 + VmaConfigData *cdata = (VmaConfigData *)l->data;
1929 + l = g_list_next(l);
1930 + printf("CFG: size: %d name: %s\n", cdata->len, cdata->name);
1934 + VmaDeviceInfo *di;
1935 + for (i = 1; i < 255; i++) {
1936 + di = vma_reader_get_device_info(vmar, i);
1938 + if (strcmp(di->devname, "vmstate") == 0) {
1939 + printf("VMSTATE: dev_id=%d memory: %zd\n", i, di->size);
1941 + printf("DEV: dev_id=%d size: %zd devname: %s\n",
1942 + i, di->size, di->devname);
1946 + /* ctime is the last entry we print */
1947 + printf("CTIME: %s", ctime(&head->ctime));
1951 +static int list_content(int argc, char **argv)
1954 + const char *filename;
1957 + c = getopt(argc, argv, "h");
1967 + g_assert_not_reached();
1971 + /* Get the filename */
1972 + if ((optind + 1) != argc) {
1975 + filename = argv[optind++];
1977 + Error *errp = NULL;
1978 + VmaReader *vmar = vma_reader_create(filename, &errp);
1981 + g_error("%s", error_get_pretty(errp));
1984 + print_content(vmar);
1986 + vma_reader_destroy(vmar);
1991 +typedef struct RestoreMap {
1997 +static int extract_content(int argc, char **argv)
2001 + const char *filename;
2002 + const char *dirname;
2003 + const char *readmap = NULL;
2006 + c = getopt(argc, argv, "hvr:");
2026 + /* Get the filename */
2027 + if ((optind + 2) != argc) {
2030 + filename = argv[optind++];
2031 + dirname = argv[optind++];
2033 + Error *errp = NULL;
2034 + VmaReader *vmar = vma_reader_create(filename, &errp);
2037 + g_error("%s", error_get_pretty(errp));
2040 + if (mkdir(dirname, 0777) < 0) {
2041 + g_error("unable to create target directory %s - %s",
2042 + dirname, strerror(errno));
2045 + GList *l = vma_reader_get_config_data(vmar);
2046 + while (l && l->data) {
2047 + VmaConfigData *cdata = (VmaConfigData *)l->data;
2048 + l = g_list_next(l);
2049 + char *cfgfn = g_strdup_printf("%s/%s", dirname, cdata->name);
2050 + GError *err = NULL;
2051 + if (!g_file_set_contents(cfgfn, (gchar *)cdata->data, cdata->len,
2053 + g_error("unable to write file: %s", err->message);
2057 + GHashTable *devmap = g_hash_table_new(g_str_hash, g_str_equal);
2060 + print_content(vmar);
2062 + FILE *map = fopen(readmap, "r");
2064 + g_error("unable to open fifo %s - %s", readmap, strerror(errno));
2069 + char *line = fgets(inbuf, sizeof(inbuf), map);
2070 + if (!line || line[0] == '\0' || !strcmp(line, "done\n")) {
2073 + int len = strlen(line);
2074 + if (line[len - 1] == '\n') {
2075 + line[len - 1] = '\0';
2083 + if (line[0] == '0' && line[1] == ':') {
2085 + write_zero = false;
2086 + } else if (line[0] == '1' && line[1] == ':') {
2088 + write_zero = true;
2090 + g_error("read map failed - parse error ('%s')", inbuf);
2093 + char *devname = NULL;
2094 + path = extract_devname(path, &devname, -1);
2096 + g_error("read map failed - no dev name specified ('%s')",
2100 + RestoreMap *map = g_new0(RestoreMap, 1);
2101 + map->devname = g_strdup(devname);
2102 + map->path = g_strdup(path);
2103 + map->write_zero = write_zero;
2105 + g_hash_table_insert(devmap, map->devname, map);
2111 + int vmstate_fd = -1;
2112 + guint8 vmstate_stream = 0;
2114 + for (i = 1; i < 255; i++) {
2115 + VmaDeviceInfo *di = vma_reader_get_device_info(vmar, i);
2116 + if (di && (strcmp(di->devname, "vmstate") == 0)) {
2117 + vmstate_stream = i;
2118 + char *statefn = g_strdup_printf("%s/vmstate.bin", dirname);
2119 + vmstate_fd = open(statefn, O_WRONLY|O_CREAT|O_EXCL, 0644);
2120 + if (vmstate_fd < 0) {
2121 + g_error("create vmstate file '%s' failed - %s", statefn,
2126 + char *devfn = NULL;
2127 + int flags = BDRV_O_RDWR|BDRV_O_CACHE_WB;
2128 + bool write_zero = true;
2132 + map = (RestoreMap *)g_hash_table_lookup(devmap, di->devname);
2133 + if (map == NULL) {
2134 + g_error("no device name mapping for %s", di->devname);
2136 + devfn = map->path;
2137 + write_zero = map->write_zero;
2139 + devfn = g_strdup_printf("%s/tmp-disk-%s.raw",
2140 + dirname, di->devname);
2141 + printf("DEVINFO %s %zd\n", devfn, di->size);
2143 + if (bdrv_img_create(devfn, "raw", NULL, NULL, NULL,
2144 + di->size, flags)) {
2145 + g_error("can't create file %s", devfn);
2148 + /* Note: we created an empty file above, so there is no
2149 + * need to write zeroes (so we generate a sparse file)
2151 + write_zero = false;
2154 + BlockDriverState *bs = NULL;
2155 + if (bdrv_file_open(&bs, devfn, flags)) {
2156 + g_error("can't open file %s", devfn);
2158 + if (vma_reader_register_bs(vmar, i, bs, write_zero, &errp) < 0) {
2159 + g_error("%s", error_get_pretty(errp));
2168 + if (vma_reader_restore(vmar, vmstate_fd, verbose, &errp) < 0) {
2169 + g_error("restore failed - %s", error_get_pretty(errp));
2173 + for (i = 1; i < 255; i++) {
2174 + VmaDeviceInfo *di = vma_reader_get_device_info(vmar, i);
2175 + if (di && (i != vmstate_stream)) {
2176 + char *tmpfn = g_strdup_printf("%s/tmp-disk-%s.raw",
2177 + dirname, di->devname);
2178 + char *fn = g_strdup_printf("%s/disk-%s.raw",
2179 + dirname, di->devname);
2180 + if (rename(tmpfn, fn) != 0) {
2181 + g_error("rename %s to %s failed - %s",
2182 + tmpfn, fn, strerror(errno));
2188 + vma_reader_destroy(vmar);
2195 +typedef struct BackupCB {
2200 +static int backup_dump_cb(void *opaque, BlockDriverState *bs,
2201 + int64_t cluster_num, unsigned char *buf)
2203 + BackupCB *bcb = opaque;
2205 + if (vma_writer_write(bcb->vmaw, bcb->dev_id, cluster_num, buf, &zb) < 0) {
2206 + g_warning("backup_dump_cb vma_writer_write failed");
2213 +static void backup_complete_cb(void *opaque, int ret)
2215 + BackupCB *bcb = opaque;
2218 + vma_writer_set_error(bcb->vmaw, "backup_complete_cb %d", ret);
2221 + if (vma_writer_close_stream(bcb->vmaw, bcb->dev_id) <= 0) {
2222 + Error *err = NULL;
2223 + if (vma_writer_close(bcb->vmaw, &err) != 0) {
2224 + g_warning("vma_writer_close failed %s", error_get_pretty(err));
2229 +static int create_archive(int argc, char **argv)
2233 + const char *archivename;
2234 + GList *config_files = NULL;
2237 + c = getopt(argc, argv, "hvc:");
2247 + config_files = g_list_append(config_files, optarg);
2253 + g_assert_not_reached();
2258 + /* make sure we have archive name and at least one path */
2259 + if ((optind + 2) > argc) {
2263 + archivename = argv[optind++];
2266 + uuid_generate(uuid);
2268 + Error *local_err = NULL;
2269 + VmaWriter *vmaw = vma_writer_create(archivename, uuid, 0, &local_err);
2271 + if (vmaw == NULL) {
2272 + g_error("%s", error_get_pretty(local_err));
2275 + GList *l = config_files;
2276 + while (l && l->data) {
2277 + char *name = l->data;
2278 + char *cdata = NULL;
2280 + GError *err = NULL;
2281 + if (!g_file_get_contents(name, &cdata, &clen, &err)) {
2282 + unlink(archivename);
2283 + g_error("Unable to read file: %s", err->message);
2286 + if (vma_writer_add_config(vmaw, name, cdata, clen) != 0) {
2287 + unlink(archivename);
2288 + g_error("Unable to append config data %s (len = %zd)",
2291 + l = g_list_next(l);
2295 + while (optind < argc) {
2296 + const char *path = argv[optind++];
2297 + char *devname = NULL;
2298 + path = extract_devname(path, &devname, ind++);
2300 + BlockDriver *drv = NULL;
2301 + BlockDriverState *bs = bdrv_new(devname);
2303 + res = bdrv_open(bs, path, BDRV_O_CACHE_WB , drv);
2305 + unlink(archivename);
2306 + g_error("bdrv_open '%s' failed", path);
2308 + int64_t size = bdrv_getlength(bs);
2309 + int dev_id = vma_writer_register_stream(vmaw, devname, size);
2310 + if (dev_id <= 0) {
2311 + unlink(archivename);
2312 + g_error("vma_writer_register_stream '%s' failed", devname);
2315 + BackupCB *bcb = g_new0(BackupCB, 1);
2317 + bcb->dev_id = dev_id;
2319 + if (backup_job_start(bs, backup_dump_cb, backup_complete_cb, bcb) < 0) {
2320 + unlink(archivename);
2321 + g_error("backup_job_start failed");
2325 + VmaStatus vmastat;
2327 + int last_percent = -1;
2330 + main_loop_wait(false);
2331 + vma_writer_get_status(vmaw, &vmastat);
2335 + uint64_t total = 0;
2336 + uint64_t transferred = 0;
2337 + uint64_t zero_bytes = 0;
2340 + for (i = 0; i < 256; i++) {
2341 + if (vmastat.stream_info[i].size) {
2342 + total += vmastat.stream_info[i].size;
2343 + transferred += vmastat.stream_info[i].transferred;
2344 + zero_bytes += vmastat.stream_info[i].zero_bytes;
2347 + percent = (transferred*100)/total;
2348 + if (percent != last_percent) {
2349 + printf("progress %d%% %zd/%zd %zd\n", percent,
2350 + transferred, total, zero_bytes);
2352 + last_percent = percent;
2356 + if (vmastat.closed) {
2363 + vma_writer_get_status(vmaw, &vmastat);
2366 + for (i = 0; i < 256; i++) {
2367 + VmaStreamInfo *si = &vmastat.stream_info[i];
2369 + printf("image %s: size=%zd zeros=%zd saved=%zd\n", si->devname,
2370 + si->size, si->zero_bytes, si->size - si->zero_bytes);
2375 + if (vmastat.status < 0) {
2376 + unlink(archivename);
2377 + g_error("creating vma archive failed");
2383 +int main(int argc, char **argv)
2385 + const char *cmdname;
2387 + error_set_progname(argv[0]);
2389 + qemu_init_main_loop();
2397 + cmdname = argv[1];
2401 + if (!strcmp(cmdname, "list")) {
2402 + return list_content(argc, argv);
2403 + } else if (!strcmp(cmdname, "create")) {
2404 + return create_archive(argc, argv);
2405 + } else if (!strcmp(cmdname, "extract")) {
2406 + return extract_content(argc, argv);
2412 diff --git a/vma.h b/vma.h
2413 new file mode 100644
2414 index 0000000..f30a2bc
2419 + * VMA: Virtual Machine Archive
2421 + * Copyright (C) Proxmox Server Solutions
2424 + * Dietmar Maurer (dietmar@proxmox.com)
2426 + * This work is licensed under the terms of the GNU GPL, version 2 or later.
2427 + * See the COPYING file in the top-level directory.
2431 +#ifndef BACKUP_VMA_H
2432 +#define BACKUP_VMA_H
2434 +#include "backup.h"
2437 +#define VMA_BLOCK_BITS 12
2438 +#define VMA_BLOCK_SIZE (1<<VMA_BLOCK_BITS)
2439 +#define VMA_CLUSTER_BITS (VMA_BLOCK_BITS+4)
2440 +#define VMA_CLUSTER_SIZE (1<<VMA_CLUSTER_BITS)
2442 +#if VMA_CLUSTER_SIZE != 65536
2443 +#error unexpected cluster size
2446 +#define VMA_EXTENT_HEADER_SIZE 512
2447 +#define VMA_BLOCKS_PER_EXTENT 59
2448 +#define VMA_MAX_CONFIGS 256
2450 +#define VMA_MAX_EXTENT_SIZE \
2451 + (VMA_EXTENT_HEADER_SIZE+VMA_CLUSTER_SIZE*VMA_BLOCKS_PER_EXTENT)
2452 +#if VMA_MAX_EXTENT_SIZE != 3867136
2453 +#error unexpected VMA_EXTENT_SIZE
2456 +/* File Format Definitions */
2458 +#define VMA_MAGIC (GUINT32_TO_BE(('V'<<24)|('M'<<16)|('A'<<8)|0x00))
2459 +#define VMA_EXTENT_MAGIC (GUINT32_TO_BE(('V'<<24)|('M'<<16)|('A'<<8)|'E'))
2461 +typedef struct VmaDeviceInfoHeader {
2462 + uint32_t devname_ptr; /* offset into blob_buffer table */
2463 + uint32_t reserved0;
2464 + uint64_t size; /* device size in bytes */
2465 + uint64_t reserved1;
2466 + uint64_t reserved2;
2467 +} VmaDeviceInfoHeader;
2469 +typedef struct VmaHeader {
2472 + unsigned char uuid[16];
2474 + unsigned char md5sum[16];
2476 + uint32_t blob_buffer_offset;
2477 + uint32_t blob_buffer_size;
2478 + uint32_t header_size;
2480 + unsigned char reserved[1984];
2482 + uint32_t config_names[VMA_MAX_CONFIGS]; /* offset into blob_buffer table */
2483 + uint32_t config_data[VMA_MAX_CONFIGS]; /* offset into blob_buffer table */
2485 + VmaDeviceInfoHeader dev_info[256];
2488 +typedef struct VmaExtentHeader {
2490 + uint16_t reserved1;
2491 + uint16_t block_count;
2492 + unsigned char uuid[16];
2493 + unsigned char md5sum[16];
2494 + uint64_t blockinfo[VMA_BLOCKS_PER_EXTENT];
2497 +/* functions/definitions to read/write vma files */
2499 +typedef struct VmaReader VmaReader;
2501 +typedef struct VmaWriter VmaWriter;
2503 +typedef struct VmaConfigData {
2509 +typedef struct VmaStreamInfo {
2511 + uint64_t cluster_count;
2512 + uint64_t transferred;
2513 + uint64_t zero_bytes;
2518 +typedef struct VmaStatus {
2521 + char errmsg[8192];
2522 + char uuid_str[37];
2523 + VmaStreamInfo stream_info[256];
2526 +typedef struct VmaDeviceInfo {
2527 + uint64_t size; /* device size in bytes */
2528 + const char *devname;
2531 +extern const BackupDriver backup_vma_driver;
2533 +VmaWriter *vma_writer_create(const char *filename, uuid_t uuid, int64_t speed,
2535 +int vma_writer_close(VmaWriter *vmaw, Error **errp);
2536 +void vma_writer_destroy(VmaWriter *vmaw);
2537 +int vma_writer_add_config(VmaWriter *vmaw, const char *name, gpointer data,
2539 +int vma_writer_register_stream(VmaWriter *vmaw, const char *devname,
2542 +int64_t coroutine_fn vma_writer_write(VmaWriter *vmaw, uint8_t dev_id,
2543 + int64_t cluster_num, unsigned char *buf,
2544 + size_t *zero_bytes);
2546 +int coroutine_fn vma_writer_close_stream(VmaWriter *vmaw, uint8_t dev_id);
2548 +int vma_writer_get_status(VmaWriter *vmaw, VmaStatus *status);
2549 +void vma_writer_set_error(VmaWriter *vmaw, const char *fmt, ...);
2552 +VmaReader *vma_reader_create(const char *filename, Error **errp);
2553 +void vma_reader_destroy(VmaReader *vmar);
2554 +VmaHeader *vma_reader_get_header(VmaReader *vmar);
2555 +GList *vma_reader_get_config_data(VmaReader *vmar);
2556 +VmaDeviceInfo *vma_reader_get_device_info(VmaReader *vmar, guint8 dev_id);
2557 +int vma_reader_register_bs(VmaReader *vmar, guint8 dev_id,
2558 + BlockDriverState *bs, bool write_zeroes,
2560 +int vma_reader_restore(VmaReader *vmar, int vmstate_fd, bool verbose,
2563 +#endif /* BACKUP_VMA_H */