1 From 2b6ca0c6087ed51bc8318731713bb3aa83f606db 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 | 933 +++++++++++++++++++++++++++++++++++++++++++++++
16 vma.c | 561 ++++++++++++++++++++++++++++
18 8 files changed, 2471 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 ba28654..a030a13 100644
65 #include "arch_init.h"
69 static QTAILQ_HEAD(drivelist, DriveInfo) drives = QTAILQ_HEAD_INITIALIZER(drives);
71 @@ -1515,10 +1516,11 @@ char *qmp_backup(const char *backupfile, bool has_format, BackupFormat format,
72 /* Todo: try to auto-detect format based on file name */
73 format = has_format ? format : BACKUP_FORMAT_VMA;
75 - /* fixme: find driver for specifued format */
76 const BackupDriver *driver = NULL;
79 + if (format == BACKUP_FORMAT_VMA) {
80 + driver = &backup_vma_driver;
82 error_set(errp, ERROR_CLASS_GENERIC_ERROR, "unknown backup 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..2217a94
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..08c4ee3
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"
958 +#include "monitor.h"
962 +#define DPRINTF(fmt, ...)\
963 + do { if (DEBUG_VMA) { printf("vma: " fmt, ## __VA_ARGS__); } } while (0)
965 +#define WRITE_BUFFERS 5
967 +typedef struct VmaAIOCB VmaAIOCB;
969 + unsigned char buffer[VMA_MAX_EXTENT_SIZE];
981 + bool header_written;
984 + /* we always write extents */
985 + unsigned char outbuf[VMA_MAX_EXTENT_SIZE];
986 + int outbuf_pos; /* in bytes */
987 + int outbuf_count; /* in VMA_BLOCKS */
988 + uint64_t outbuf_block_info[VMA_BLOCKS_PER_EXTENT];
990 + VmaAIOCB *aiocbs[WRITE_BUFFERS];
993 + GChecksum *md5csum;
994 + CoMutex writer_lock;
995 + CoMutex flush_lock;
996 + Coroutine *co_writer;
998 + /* drive informations */
999 + VmaStreamInfo stream_info[256];
1000 + guint stream_count;
1002 + guint8 vmstate_stream;
1003 + uint32_t vmstate_clusters;
1005 + /* header blob table */
1006 + char *header_blob_table;
1007 + uint32_t header_blob_table_size;
1008 + uint32_t header_blob_table_pos;
1010 + /* store for config blobs */
1011 + uint32_t config_names[VMA_MAX_CONFIGS]; /* offset into blob_buffer table */
1012 + uint32_t config_data[VMA_MAX_CONFIGS]; /* offset into blob_buffer table */
1013 + uint32_t config_count;
1016 +void vma_writer_set_error(VmaWriter *vmaw, const char *fmt, ...)
1020 + if (vmaw->status < 0) {
1024 + vmaw->status = -1;
1026 + va_start(ap, fmt);
1027 + g_vsnprintf(vmaw->errmsg, sizeof(vmaw->errmsg), fmt, ap);
1030 + DPRINTF("vma_writer_set_error: %s\n", vmaw->errmsg);
1033 +static uint32_t allocate_header_blob(VmaWriter *vmaw, const char *data,
1036 + if (len > 65535) {
1040 + if (!vmaw->header_blob_table ||
1041 + (vmaw->header_blob_table_size <
1042 + (vmaw->header_blob_table_pos + len + 2))) {
1043 + int newsize = vmaw->header_blob_table_size + ((len + 2 + 511)/512)*512;
1045 + vmaw->header_blob_table = g_realloc(vmaw->header_blob_table, newsize);
1046 + memset(vmaw->header_blob_table + vmaw->header_blob_table_size,
1047 + 0, newsize - vmaw->header_blob_table_size);
1048 + vmaw->header_blob_table_size = newsize;
1051 + uint32_t cpos = vmaw->header_blob_table_pos;
1052 + vmaw->header_blob_table[cpos] = len & 255;
1053 + vmaw->header_blob_table[cpos+1] = (len >> 8) & 255;
1054 + memcpy(vmaw->header_blob_table + cpos + 2, data, len);
1055 + vmaw->header_blob_table_pos += len + 2;
1059 +static uint32_t allocate_header_string(VmaWriter *vmaw, const char *str)
1063 + size_t len = strlen(str) + 1;
1065 + return allocate_header_blob(vmaw, str, len);
1068 +int vma_writer_add_config(VmaWriter *vmaw, const char *name, gpointer data,
1072 + assert(!vmaw->header_written);
1073 + assert(vmaw->config_count < VMA_MAX_CONFIGS);
1078 + uint32_t name_ptr = allocate_header_string(vmaw, name);
1083 + uint32_t data_ptr = allocate_header_blob(vmaw, data, len);
1088 + vmaw->config_names[vmaw->config_count] = name_ptr;
1089 + vmaw->config_data[vmaw->config_count] = data_ptr;
1091 + vmaw->config_count++;
1096 +int vma_writer_register_stream(VmaWriter *vmaw, const char *devname,
1101 + assert(!vmaw->status);
1103 + if (vmaw->header_written) {
1104 + vma_writer_set_error(vmaw, "vma_writer_register_stream: header "
1105 + "already written");
1109 + guint n = vmaw->stream_count + 1;
1111 + /* we can have dev_ids form 1 to 255 (0 reserved)
1112 + * 255(-1) reseverd for safety
1115 + vma_writer_set_error(vmaw, "vma_writer_register_stream: "
1116 + "too many drives");
1121 + vma_writer_set_error(vmaw, "vma_writer_register_stream: "
1122 + "got strange size %zd", size);
1126 + DPRINTF("vma_writer_register_stream %s %zu %d\n", devname, size, n);
1128 + vmaw->stream_info[n].devname = g_strdup(devname);
1129 + vmaw->stream_info[n].size = size;
1131 + vmaw->stream_info[n].cluster_count = (size + VMA_CLUSTER_SIZE - 1) /
1134 + vmaw->stream_count = n;
1136 + if (strcmp(devname, "vmstate") == 0) {
1137 + vmaw->vmstate_stream = n;
1143 +static void vma_co_continue_write(void *opaque)
1145 + VmaWriter *vmaw = opaque;
1147 + qemu_aio_set_fd_handler(vmaw->fd, NULL, NULL, NULL, NULL);
1149 + DPRINTF("vma_co_continue_write\n");
1150 + qemu_coroutine_enter(vmaw->co_writer, NULL);
1153 +static int vma_co_write_finished(void *opaque)
1155 + VmaWriter *vmaw = opaque;
1157 + return (vmaw->co_writer != 0);
1160 +static ssize_t coroutine_fn
1161 +vma_co_write(VmaWriter *vmaw, const void *buf, size_t bytes)
1166 + /* atomic writes (we cannot interleave writes) */
1167 + qemu_co_mutex_lock(&vmaw->writer_lock);
1169 + DPRINTF("vma_co_write enter %zd\n", bytes);
1171 + assert(vmaw->co_writer == NULL);
1173 + while (done < bytes) {
1174 + /* Note: we limit maximal write size - else VM gets slow */
1175 + ret = write(vmaw->fd, buf + done,
1176 + (bytes - done) > 4096 ? 4096 : bytes - done);
1179 + DPRINTF("vma_co_write written %zd %zd\n", done, ret);
1180 + } else if (ret < 0) {
1181 + if (errno == EAGAIN || errno == EWOULDBLOCK) {
1182 + DPRINTF("vma_co_write yield %zd\n", done);
1184 + vmaw->co_writer = qemu_coroutine_self();
1185 + qemu_aio_set_fd_handler(vmaw->fd, NULL, vma_co_continue_write,
1186 + vma_co_write_finished, vmaw);
1188 + qemu_coroutine_yield();
1189 + DPRINTF("vma_co_write restart %zd\n", done);
1191 + vma_writer_set_error(vmaw, "vma_co_write write error - %s",
1193 + done = -1; /* always return failure for partial writes */
1196 + } else if (ret == 0) {
1197 + /* should not happen - simply try again */
1201 + vmaw->co_writer = NULL;
1203 + qemu_co_mutex_unlock(&vmaw->writer_lock);
1205 + DPRINTF("vma_co_write leave %zd\n", done);
1209 +static void coroutine_fn vma_co_writer_task(void *opaque)
1211 + VmaAIOCB *cb = opaque;
1213 + DPRINTF("vma_co_writer_task start\n");
1215 + int64_t done = vma_co_write(cb->vmaw, cb->buffer, cb->bytes);
1216 + DPRINTF("vma_co_writer_task write done %zd\n", done);
1218 + if (done != cb->bytes) {
1219 + DPRINTF("vma_co_writer_task failed write %zd %zd", cb->bytes, done);
1220 + vma_writer_set_error(cb->vmaw, "vma_co_writer_task failed write %zd",
1226 + qemu_co_queue_next(&cb->vmaw->wqueue);
1228 + DPRINTF("vma_co_writer_task end\n");
1231 +static void coroutine_fn vma_queue_flush(VmaWriter *vmaw)
1233 + DPRINTF("vma_queue_flush enter\n");
1239 + VmaAIOCB *cb = NULL;
1240 + for (i = 0; i < WRITE_BUFFERS; i++) {
1241 + if (vmaw->aiocbs[i]->bytes) {
1242 + cb = vmaw->aiocbs[i];
1243 + DPRINTF("FOUND USED AIO BUFFER %d %zd\n", i,
1244 + vmaw->aiocbs[i]->bytes);
1251 + qemu_co_queue_wait(&vmaw->wqueue);
1254 + DPRINTF("vma_queue_flush leave\n");
1258 + * NOTE: pipe buffer size in only 4096 bytes on linux (see 'ulimit -a')
1259 + * So we need to create a coroutione to allow 'parallel' execution.
1261 +static ssize_t coroutine_fn
1262 +vma_queue_write(VmaWriter *vmaw, const void *buf, size_t bytes)
1264 + DPRINTF("vma_queue_write enter %zd\n", bytes);
1268 + assert(bytes <= VMA_MAX_EXTENT_SIZE);
1270 + VmaAIOCB *cb = NULL;
1273 + for (i = 0; i < WRITE_BUFFERS; i++) {
1274 + if (!vmaw->aiocbs[i]->bytes) {
1275 + cb = vmaw->aiocbs[i];
1280 + qemu_co_queue_wait(&vmaw->wqueue);
1284 + memcpy(cb->buffer, buf, bytes);
1285 + cb->bytes = bytes;
1288 + DPRINTF("vma_queue_write start %zd\n", bytes);
1289 + cb->co = qemu_coroutine_create(vma_co_writer_task);
1290 + qemu_coroutine_enter(cb->co, cb);
1292 + DPRINTF("vma_queue_write leave\n");
1297 +VmaWriter *vma_writer_create(const char *filename, uuid_t uuid, Error **errp)
1301 + assert(sizeof(VmaHeader) == (4096 + 8192));
1302 + assert(sizeof(VmaExtentHeader) == 512);
1304 + VmaWriter *vmaw = g_new0(VmaWriter, 1);
1307 + vmaw->md5csum = g_checksum_new(G_CHECKSUM_MD5);
1308 + if (!vmaw->md5csum) {
1309 + error_setg(errp, "can't allocate cmsum\n");
1313 + if (strstart(filename, "exec:", &p)) {
1314 + vmaw->cmd = popen(p, "w");
1315 + if (vmaw->cmd == NULL) {
1316 + error_setg(errp, "can't popen command '%s' - %s\n", p,
1320 + vmaw->fd = fileno(vmaw->cmd);
1322 + /* try to use O_NONBLOCK and O_DIRECT */
1323 + fcntl(vmaw->fd, F_SETFL, fcntl(vmaw->fd, F_GETFL)|O_NONBLOCK);
1324 + fcntl(vmaw->fd, F_SETFL, fcntl(vmaw->fd, F_GETFL)|O_DIRECT);
1329 + const char *tmp_id_str;
1331 + if ((stat(filename, &st) == 0) && S_ISFIFO(st.st_mode)) {
1332 + oflags = O_NONBLOCK|O_DIRECT|O_WRONLY;
1333 + vmaw->fd = qemu_open(filename, oflags, 0644);
1334 + } else if (strstart(filename, "/dev/fdset/", &tmp_id_str)) {
1335 + oflags = O_NONBLOCK|O_DIRECT|O_WRONLY;
1336 + vmaw->fd = qemu_open(filename, oflags, 0644);
1337 + } else if (strstart(filename, "/dev/fdname/", &tmp_id_str)) {
1338 + vmaw->fd = monitor_get_fd(cur_mon, tmp_id_str, errp);
1339 + if (vmaw->fd < 0) {
1342 + /* try to use O_NONBLOCK and O_DIRECT */
1343 + fcntl(vmaw->fd, F_SETFL, fcntl(vmaw->fd, F_GETFL)|O_NONBLOCK);
1344 + fcntl(vmaw->fd, F_SETFL, fcntl(vmaw->fd, F_GETFL)|O_DIRECT);
1346 + oflags = O_NONBLOCK|O_DIRECT|O_WRONLY|O_CREAT|O_EXCL;
1347 + vmaw->fd = qemu_open(filename, oflags, 0644);
1350 + if (vmaw->fd < 0) {
1351 + error_setg(errp, "can't open file %s - %s\n", filename,
1357 + /* we use O_DIRECT, so we need to align IO buffers */
1359 + for (i = 0; i < WRITE_BUFFERS; i++) {
1360 + vmaw->aiocbs[i] = qemu_memalign(512, sizeof(VmaAIOCB));
1361 + memset(vmaw->aiocbs[i], 0, sizeof(VmaAIOCB));
1364 + vmaw->outbuf_count = 0;
1365 + vmaw->outbuf_pos = VMA_EXTENT_HEADER_SIZE;
1367 + vmaw->header_blob_table_pos = 1; /* start at pos 1 */
1369 + qemu_co_mutex_init(&vmaw->writer_lock);
1370 + qemu_co_mutex_init(&vmaw->flush_lock);
1371 + qemu_co_queue_init(&vmaw->wqueue);
1373 + uuid_copy(vmaw->uuid, uuid);
1380 + pclose(vmaw->cmd);
1381 + } else if (vmaw->fd >= 0) {
1385 + if (vmaw->md5csum) {
1386 + g_checksum_free(vmaw->md5csum);
1395 +static int coroutine_fn vma_write_header(VmaWriter *vmaw)
1398 + int header_clusters = 8;
1399 + char buf[65536*header_clusters];
1400 + VmaHeader *head = (VmaHeader *)buf;
1404 + DPRINTF("VMA WRITE HEADER\n");
1406 + if (vmaw->status < 0) {
1407 + return vmaw->status;
1410 + memset(buf, 0, sizeof(buf));
1412 + head->magic = VMA_MAGIC;
1413 + head->version = GUINT32_TO_BE(1); /* v1 */
1414 + memcpy(head->uuid, vmaw->uuid, 16);
1416 + time_t ctime = time(NULL);
1417 + head->ctime = GUINT64_TO_BE(ctime);
1419 + if (!vmaw->stream_count) {
1423 + for (i = 0; i < VMA_MAX_CONFIGS; i++) {
1424 + head->config_names[i] = GUINT32_TO_BE(vmaw->config_names[i]);
1425 + head->config_data[i] = GUINT32_TO_BE(vmaw->config_data[i]);
1428 + /* 32 bytes per device (12 used currently) = 8192 bytes max */
1429 + for (i = 1; i <= 254; i++) {
1430 + VmaStreamInfo *si = &vmaw->stream_info[i];
1432 + assert(si->devname);
1433 + uint32_t devname_ptr = allocate_header_string(vmaw, si->devname);
1434 + if (!devname_ptr) {
1437 + head->dev_info[i].devname_ptr = GUINT32_TO_BE(devname_ptr);
1438 + head->dev_info[i].size = GUINT64_TO_BE(si->size);
1442 + uint32_t header_size = sizeof(VmaHeader) + vmaw->header_blob_table_size;
1443 + head->header_size = GUINT32_TO_BE(header_size);
1445 + if (header_size > sizeof(buf)) {
1446 + return -1; /* just to be sure */
1449 + uint32_t blob_buffer_offset = sizeof(VmaHeader);
1450 + memcpy(buf + blob_buffer_offset, vmaw->header_blob_table,
1451 + vmaw->header_blob_table_size);
1452 + head->blob_buffer_offset = GUINT32_TO_BE(blob_buffer_offset);
1453 + head->blob_buffer_size = GUINT32_TO_BE(vmaw->header_blob_table_pos);
1455 + g_checksum_reset(vmaw->md5csum);
1456 + g_checksum_update(vmaw->md5csum, (const guchar *)buf, header_size);
1458 + g_checksum_get_digest(vmaw->md5csum, (guint8 *)(head->md5sum), &csize);
1460 + return vma_queue_write(vmaw, buf, header_size);
1463 +static int coroutine_fn vma_writer_flush(VmaWriter *vmaw)
1470 + if (vmaw->status < 0) {
1471 + return vmaw->status;
1474 + if (!vmaw->header_written) {
1475 + vmaw->header_written = true;
1476 + ret = vma_write_header(vmaw);
1478 + vma_writer_set_error(vmaw, "vma_writer_flush: write header failed");
1483 + DPRINTF("VMA WRITE FLUSH %d %d\n", vmaw->outbuf_count, vmaw->outbuf_pos);
1486 + VmaExtentHeader *ehead = (VmaExtentHeader *)vmaw->outbuf;
1488 + ehead->magic = VMA_EXTENT_MAGIC;
1489 + ehead->reserved1 = 0;
1491 + for (i = 0; i < VMA_BLOCKS_PER_EXTENT; i++) {
1492 + ehead->blockinfo[i] = GUINT64_TO_BE(vmaw->outbuf_block_info[i]);
1495 + guint16 block_count = (vmaw->outbuf_pos - VMA_EXTENT_HEADER_SIZE) /
1498 + ehead->block_count = GUINT16_TO_BE(block_count);
1500 + memcpy(ehead->uuid, vmaw->uuid, sizeof(ehead->uuid));
1501 + memset(ehead->md5sum, 0, sizeof(ehead->md5sum));
1503 + g_checksum_reset(vmaw->md5csum);
1504 + g_checksum_update(vmaw->md5csum, vmaw->outbuf, VMA_EXTENT_HEADER_SIZE);
1506 + g_checksum_get_digest(vmaw->md5csum, ehead->md5sum, &csize);
1508 + int bytes = vmaw->outbuf_pos;
1509 + ret = vma_queue_write(vmaw, vmaw->outbuf, bytes);
1510 + if (ret != bytes) {
1511 + vma_writer_set_error(vmaw, "vma_writer_flush: failed write");
1514 + vmaw->outbuf_count = 0;
1515 + vmaw->outbuf_pos = VMA_EXTENT_HEADER_SIZE;
1517 + for (i = 0; i < VMA_BLOCKS_PER_EXTENT; i++) {
1518 + vmaw->outbuf_block_info[i] = 0;
1521 + return vmaw->status;
1524 +static int vma_count_open_streams(VmaWriter *vmaw)
1526 + g_assert(vmaw != NULL);
1529 + int open_drives = 0;
1530 + for (i = 0; i <= 255; i++) {
1531 + if (vmaw->stream_info[i].size && !vmaw->stream_info[i].finished) {
1536 + return open_drives;
1540 + * all jobs should call this when there is no more data
1541 + * Returns: number of remaining stream (0 ==> finished)
1544 +vma_writer_close_stream(VmaWriter *vmaw, uint8_t dev_id)
1546 + g_assert(vmaw != NULL);
1548 + DPRINTF("vma_writer_set_status %d\n", dev_id);
1549 + if (!vmaw->stream_info[dev_id].size) {
1550 + vma_writer_set_error(vmaw, "vma_writer_close_stream: "
1551 + "no such stream %d", dev_id);
1554 + if (vmaw->stream_info[dev_id].finished) {
1555 + vma_writer_set_error(vmaw, "vma_writer_close_stream: "
1556 + "stream already closed %d", dev_id);
1560 + vmaw->stream_info[dev_id].finished = true;
1562 + int open_drives = vma_count_open_streams(vmaw);
1564 + if (open_drives <= 0) {
1565 + DPRINTF("vma_writer_set_status all drives completed\n");
1566 + qemu_co_mutex_lock(&vmaw->flush_lock);
1567 + int ret = vma_writer_flush(vmaw);
1568 + qemu_co_mutex_unlock(&vmaw->flush_lock);
1570 + vma_writer_set_error(vmaw, "vma_writer_close_stream: flush failed");
1574 + return open_drives;
1577 +int vma_writer_get_status(VmaWriter *vmaw, VmaStatus *status)
1581 + g_assert(vmaw != NULL);
1584 + status->status = vmaw->status;
1585 + g_strlcpy(status->errmsg, vmaw->errmsg, sizeof(status->errmsg));
1586 + for (i = 0; i <= 255; i++) {
1587 + status->stream_info[i] = vmaw->stream_info[i];
1590 + uuid_unparse_lower(vmaw->uuid, status->uuid_str);
1593 + status->closed = vmaw->closed;
1595 + return vmaw->status;
1598 +static int vma_writer_get_buffer(VmaWriter *vmaw)
1602 + qemu_co_mutex_lock(&vmaw->flush_lock);
1604 + /* wait until buffer is available */
1605 + while (vmaw->outbuf_count >= (VMA_BLOCKS_PER_EXTENT - 1)) {
1606 + ret = vma_writer_flush(vmaw);
1608 + vma_writer_set_error(vmaw, "vma_writer_get_buffer: flush failed");
1613 + qemu_co_mutex_unlock(&vmaw->flush_lock);
1619 +int64_t coroutine_fn
1620 +vma_writer_write(VmaWriter *vmaw, uint8_t dev_id, int64_t cluster_num,
1621 + unsigned char *buf, size_t *zero_bytes)
1623 + g_assert(vmaw != NULL);
1624 + g_assert(zero_bytes != NULL);
1628 + if (vmaw->status < 0) {
1629 + return vmaw->status;
1632 + if (!dev_id || !vmaw->stream_info[dev_id].size) {
1633 + vma_writer_set_error(vmaw, "vma_writer_write: "
1634 + "no such stream %d", dev_id);
1638 + if (vmaw->stream_info[dev_id].finished) {
1639 + vma_writer_set_error(vmaw, "vma_writer_write: "
1640 + "stream already closed %d", dev_id);
1645 + if (cluster_num >= (((uint64_t)1)<<32)) {
1646 + vma_writer_set_error(vmaw, "vma_writer_write: "
1647 + "cluster number out of range");
1651 + if (dev_id == vmaw->vmstate_stream) {
1652 + if (cluster_num != vmaw->vmstate_clusters) {
1653 + vma_writer_set_error(vmaw, "vma_writer_write: "
1654 + "non sequential vmstate write");
1656 + vmaw->vmstate_clusters++;
1657 + } else if (cluster_num >= vmaw->stream_info[dev_id].cluster_count) {
1658 + vma_writer_set_error(vmaw, "vma_writer_write: cluster number too big");
1662 + /* wait until buffer is available */
1663 + if (vma_writer_get_buffer(vmaw) < 0) {
1664 + vma_writer_set_error(vmaw, "vma_writer_write: "
1665 + "vma_writer_get_buffer failed");
1669 + DPRINTF("VMA WRITE %d %zd\n", dev_id, cluster_num);
1673 + uint16_t mask = 0;
1674 + for (i = 0; i < 16; i++) {
1675 + unsigned char *vmablock = buf + (i*VMA_BLOCK_SIZE);
1676 + if (!buffer_is_zero(vmablock, VMA_BLOCK_SIZE)) {
1678 + memcpy(vmaw->outbuf + vmaw->outbuf_pos, vmablock, VMA_BLOCK_SIZE);
1679 + vmaw->outbuf_pos += VMA_BLOCK_SIZE;
1681 + DPRINTF("VMA WRITE %zd ZERO BLOCK %d\n", cluster_num, i);
1682 + vmaw->stream_info[dev_id].zero_bytes += VMA_BLOCK_SIZE;
1683 + *zero_bytes += VMA_BLOCK_SIZE;
1689 + uint64_t block_info = ((uint64_t)mask) << (32+16);
1690 + block_info |= ((uint64_t)dev_id) << 32;
1691 + block_info |= (cluster_num & 0xffffffff);
1692 + vmaw->outbuf_block_info[vmaw->outbuf_count] = block_info;
1694 + DPRINTF("VMA WRITE MASK %zd %zx\n", cluster_num, block_info);
1696 + vmaw->outbuf_count++;
1698 + /** NOTE: We allways write whole clusters, but we correctly set
1699 + * transferred bytes. So transferred == size when when everything
1702 + size_t transferred = VMA_CLUSTER_SIZE;
1704 + if (dev_id != vmaw->vmstate_stream) {
1705 + uint64_t last = (cluster_num + 1) * VMA_CLUSTER_SIZE;
1706 + if (last > vmaw->stream_info[dev_id].size) {
1707 + uint64_t diff = last - vmaw->stream_info[dev_id].size;
1708 + if (diff >= VMA_CLUSTER_SIZE) {
1709 + vma_writer_set_error(vmaw, "vma_writer_write: "
1710 + "read after last cluster");
1713 + transferred -= diff;
1717 + vmaw->stream_info[dev_id].transferred += transferred;
1719 + return transferred;
1722 +int vma_writer_close(VmaWriter *vmaw, Error **errp)
1724 + g_assert(vmaw != NULL);
1728 + vma_queue_flush(vmaw);
1730 + /* this should not happen - just to be sure */
1731 + while (!qemu_co_queue_empty(&vmaw->wqueue)) {
1732 + DPRINTF("vma_writer_close wait\n");
1733 + co_sleep_ns(rt_clock, 1000000);
1737 + if (pclose(vmaw->cmd) < 0) {
1738 + vma_writer_set_error(vmaw, "vma_writer_close: "
1739 + "pclose failed - %s", strerror(errno));
1742 + if (close(vmaw->fd) < 0) {
1743 + vma_writer_set_error(vmaw, "vma_writer_close: "
1744 + "close failed - %s", strerror(errno));
1748 + for (i = 0; i <= 255; i++) {
1749 + VmaStreamInfo *si = &vmaw->stream_info[i];
1751 + if (!si->finished) {
1752 + vma_writer_set_error(vmaw, "vma_writer_close: "
1753 + "detected open stream '%s'", si->devname);
1754 + } else if ((si->transferred != si->size) &&
1755 + (i != vmaw->vmstate_stream)) {
1756 + vma_writer_set_error(vmaw, "vma_writer_close: "
1757 + "incomplete stream '%s' (%zd != %zd)",
1758 + si->devname, si->transferred, si->size);
1763 + for (i = 0; i <= 255; i++) {
1764 + vmaw->stream_info[i].finished = 1; /* mark as closed */
1769 + if (vmaw->status < 0 && *errp == NULL) {
1770 + error_setg(errp, "%s", vmaw->errmsg);
1773 + return vmaw->status;
1776 +void vma_writer_destroy(VmaWriter *vmaw)
1782 + for (i = 0; i <= 255; i++) {
1783 + if (vmaw->stream_info[i].devname) {
1784 + g_free(vmaw->stream_info[i].devname);
1788 + if (vmaw->md5csum) {
1789 + g_checksum_free(vmaw->md5csum);
1792 + for (i = 0; i < WRITE_BUFFERS; i++) {
1793 + free(vmaw->aiocbs[i]);
1799 +/* backup driver plugin */
1801 +static int vma_dump_cb(void *opaque, uint8_t dev_id, int64_t cluster_num,
1802 + unsigned char *buf, size_t *zero_bytes)
1804 + VmaWriter *vmaw = opaque;
1806 + return vma_writer_write(vmaw, dev_id, cluster_num, buf, zero_bytes);
1809 +static int vma_close_cb(void *opaque, Error **errp)
1811 + VmaWriter *vmaw = opaque;
1813 + int res = vma_writer_close(vmaw, errp);
1814 + vma_writer_destroy(vmaw);
1819 +static int vma_complete_cb(void *opaque, uint8_t dev_id, int ret)
1821 + VmaWriter *vmaw = opaque;
1824 + vma_writer_set_error(vmaw, "backup_complete_cb %d", ret);
1827 + return vma_writer_close_stream(vmaw, dev_id);
1830 +static int vma_register_stream_cb(void *opaque, const char *devname,
1833 + VmaWriter *vmaw = opaque;
1835 + return vma_writer_register_stream(vmaw, devname, size);
1838 +static int vma_register_config_cb(void *opaque, const char *name,
1839 + gpointer data, size_t data_len)
1841 + VmaWriter *vmaw = opaque;
1843 + return vma_writer_add_config(vmaw, name, data, data_len);
1846 +static void *vma_open_cb(const char *filename, uuid_t uuid, Error **errp)
1848 + return vma_writer_create(filename, uuid, errp);
1851 +const BackupDriver backup_vma_driver = {
1853 + .open_cb = vma_open_cb,
1854 + .close_cb = vma_close_cb,
1855 + .register_config_cb = vma_register_config_cb,
1856 + .register_stream_cb = vma_register_stream_cb,
1857 + .dump_cb = vma_dump_cb,
1858 + .complete_cb = vma_complete_cb,
1861 diff --git a/vma.c b/vma.c
1862 new file mode 100644
1863 index 0000000..5d13906
1868 + * VMA: Virtual Machine Archive
1870 + * Copyright (C) 2012 Proxmox Server Solutions
1873 + * Dietmar Maurer (dietmar@proxmox.com)
1875 + * This work is licensed under the terms of the GNU GPL, version 2 or later.
1876 + * See the COPYING file in the top-level directory.
1882 +#include <unistd.h>
1884 +#include <string.h>
1885 +#include <sys/types.h>
1886 +#include <sys/stat.h>
1890 +#include "qemu-common.h"
1891 +#include "qemu-option.h"
1892 +#include "qemu-error.h"
1894 +#include "sysemu.h"
1895 +#include "block_int.h"
1899 +static void help(void)
1901 + const char *help_msg =
1902 + "usage: vma command [command options]\n"
1904 + "vma list <filename>\n"
1905 + "vma create <filename> [-c config] <archive> pathname ...\n"
1906 + "vma extract <filename> [-r] <targetdir>\n"
1909 + printf("%s", help_msg);
1913 +static const char *extract_devname(const char *path, char **devname, int index)
1917 + const char *sep = strchr(path, '=');
1920 + *devname = g_strndup(path, sep - path);
1924 + *devname = g_strdup_printf("disk%d", index);
1933 +static void print_content(VmaReader *vmar)
1937 + VmaHeader *head = vma_reader_get_header(vmar);
1939 + GList *l = vma_reader_get_config_data(vmar);
1940 + while (l && l->data) {
1941 + VmaConfigData *cdata = (VmaConfigData *)l->data;
1942 + l = g_list_next(l);
1943 + printf("CFG: size: %d name: %s\n", cdata->len, cdata->name);
1947 + VmaDeviceInfo *di;
1948 + for (i = 1; i < 255; i++) {
1949 + di = vma_reader_get_device_info(vmar, i);
1951 + if (strcmp(di->devname, "vmstate") == 0) {
1952 + printf("VMSTATE: dev_id=%d memory: %zd\n", i, di->size);
1954 + printf("DEV: dev_id=%d size: %zd devname: %s\n",
1955 + i, di->size, di->devname);
1959 + /* ctime is the last entry we print */
1960 + printf("CTIME: %s", ctime(&head->ctime));
1964 +static int list_content(int argc, char **argv)
1967 + const char *filename;
1970 + c = getopt(argc, argv, "h");
1980 + g_assert_not_reached();
1984 + /* Get the filename */
1985 + if ((optind + 1) != argc) {
1988 + filename = argv[optind++];
1990 + Error *errp = NULL;
1991 + VmaReader *vmar = vma_reader_create(filename, &errp);
1994 + g_error("%s", error_get_pretty(errp));
1997 + print_content(vmar);
1999 + vma_reader_destroy(vmar);
2004 +typedef struct RestoreMap {
2010 +static int extract_content(int argc, char **argv)
2014 + const char *filename;
2015 + const char *dirname;
2016 + const char *readmap = NULL;
2019 + c = getopt(argc, argv, "hvr:");
2039 + /* Get the filename */
2040 + if ((optind + 2) != argc) {
2043 + filename = argv[optind++];
2044 + dirname = argv[optind++];
2046 + Error *errp = NULL;
2047 + VmaReader *vmar = vma_reader_create(filename, &errp);
2050 + g_error("%s", error_get_pretty(errp));
2053 + if (mkdir(dirname, 0777) < 0) {
2054 + g_error("unable to create target directory %s - %s",
2055 + dirname, strerror(errno));
2058 + GList *l = vma_reader_get_config_data(vmar);
2059 + while (l && l->data) {
2060 + VmaConfigData *cdata = (VmaConfigData *)l->data;
2061 + l = g_list_next(l);
2062 + char *cfgfn = g_strdup_printf("%s/%s", dirname, cdata->name);
2063 + GError *err = NULL;
2064 + if (!g_file_set_contents(cfgfn, (gchar *)cdata->data, cdata->len,
2066 + g_error("unable to write file: %s", err->message);
2070 + GHashTable *devmap = g_hash_table_new(g_str_hash, g_str_equal);
2073 + print_content(vmar);
2075 + FILE *map = fopen(readmap, "r");
2077 + g_error("unable to open fifo %s - %s", readmap, strerror(errno));
2082 + char *line = fgets(inbuf, sizeof(inbuf), map);
2083 + if (!line || line[0] == '\0' || !strcmp(line, "done\n")) {
2086 + int len = strlen(line);
2087 + if (line[len - 1] == '\n') {
2088 + line[len - 1] = '\0';
2096 + if (line[0] == '0' && line[1] == ':') {
2098 + write_zero = false;
2099 + } else if (line[0] == '1' && line[1] == ':') {
2101 + write_zero = true;
2103 + g_error("read map failed - parse error ('%s')", inbuf);
2106 + char *devname = NULL;
2107 + path = extract_devname(path, &devname, -1);
2109 + g_error("read map failed - no dev name specified ('%s')",
2113 + RestoreMap *map = g_new0(RestoreMap, 1);
2114 + map->devname = g_strdup(devname);
2115 + map->path = g_strdup(path);
2116 + map->write_zero = write_zero;
2118 + g_hash_table_insert(devmap, map->devname, map);
2124 + int vmstate_fd = -1;
2125 + guint8 vmstate_stream = 0;
2127 + for (i = 1; i < 255; i++) {
2128 + VmaDeviceInfo *di = vma_reader_get_device_info(vmar, i);
2129 + if (di && (strcmp(di->devname, "vmstate") == 0)) {
2130 + vmstate_stream = i;
2131 + char *statefn = g_strdup_printf("%s/vmstate.bin", dirname);
2132 + vmstate_fd = open(statefn, O_WRONLY|O_CREAT|O_EXCL, 0644);
2133 + if (vmstate_fd < 0) {
2134 + g_error("create vmstate file '%s' failed - %s", statefn,
2139 + char *devfn = NULL;
2140 + int flags = BDRV_O_RDWR|BDRV_O_CACHE_WB;
2141 + bool write_zero = true;
2145 + map = (RestoreMap *)g_hash_table_lookup(devmap, di->devname);
2146 + if (map == NULL) {
2147 + g_error("no device name mapping for %s", di->devname);
2149 + devfn = map->path;
2150 + write_zero = map->write_zero;
2152 + devfn = g_strdup_printf("%s/tmp-disk-%s.raw",
2153 + dirname, di->devname);
2154 + printf("DEVINFO %s %zd\n", devfn, di->size);
2156 + if (bdrv_img_create(devfn, "raw", NULL, NULL, NULL,
2157 + di->size, flags)) {
2158 + g_error("can't create file %s", devfn);
2161 + /* Note: we created an empty file above, so there is no
2162 + * need to write zeroes (so we generate a sparse file)
2164 + write_zero = false;
2167 + BlockDriverState *bs = bdrv_new(di->devname);
2168 + if (bdrv_open(bs, devfn, flags, NULL)) {
2169 + g_error("can't open file %s", devfn);
2171 + if (vma_reader_register_bs(vmar, i, bs, write_zero, &errp) < 0) {
2172 + g_error("%s", error_get_pretty(errp));
2181 + if (vma_reader_restore(vmar, vmstate_fd, verbose, &errp) < 0) {
2182 + g_error("restore failed - %s", error_get_pretty(errp));
2186 + for (i = 1; i < 255; i++) {
2187 + VmaDeviceInfo *di = vma_reader_get_device_info(vmar, i);
2188 + if (di && (i != vmstate_stream)) {
2189 + char *tmpfn = g_strdup_printf("%s/tmp-disk-%s.raw",
2190 + dirname, di->devname);
2191 + char *fn = g_strdup_printf("%s/disk-%s.raw",
2192 + dirname, di->devname);
2193 + if (rename(tmpfn, fn) != 0) {
2194 + g_error("rename %s to %s failed - %s",
2195 + tmpfn, fn, strerror(errno));
2201 + vma_reader_destroy(vmar);
2208 +typedef struct BackupCB {
2213 +static int backup_dump_cb(void *opaque, BlockDriverState *bs,
2214 + int64_t cluster_num, unsigned char *buf)
2216 + BackupCB *bcb = opaque;
2218 + if (vma_writer_write(bcb->vmaw, bcb->dev_id, cluster_num, buf, &zb) < 0) {
2219 + g_warning("backup_dump_cb vma_writer_write failed");
2226 +static void backup_complete_cb(void *opaque, int ret)
2228 + BackupCB *bcb = opaque;
2231 + vma_writer_set_error(bcb->vmaw, "backup_complete_cb %d", ret);
2234 + if (vma_writer_close_stream(bcb->vmaw, bcb->dev_id) <= 0) {
2235 + Error *err = NULL;
2236 + if (vma_writer_close(bcb->vmaw, &err) != 0) {
2237 + g_warning("vma_writer_close failed %s", error_get_pretty(err));
2242 +static int create_archive(int argc, char **argv)
2246 + const char *archivename;
2247 + GList *config_files = NULL;
2250 + c = getopt(argc, argv, "hvc:");
2260 + config_files = g_list_append(config_files, optarg);
2266 + g_assert_not_reached();
2271 + /* make sure we have archive name and at least one path */
2272 + if ((optind + 2) > argc) {
2276 + archivename = argv[optind++];
2279 + uuid_generate(uuid);
2281 + Error *local_err = NULL;
2282 + VmaWriter *vmaw = vma_writer_create(archivename, uuid, &local_err);
2284 + if (vmaw == NULL) {
2285 + g_error("%s", error_get_pretty(local_err));
2288 + GList *l = config_files;
2289 + while (l && l->data) {
2290 + char *name = l->data;
2291 + char *cdata = NULL;
2293 + GError *err = NULL;
2294 + if (!g_file_get_contents(name, &cdata, &clen, &err)) {
2295 + unlink(archivename);
2296 + g_error("Unable to read file: %s", err->message);
2299 + if (vma_writer_add_config(vmaw, name, cdata, clen) != 0) {
2300 + unlink(archivename);
2301 + g_error("Unable to append config data %s (len = %zd)",
2304 + l = g_list_next(l);
2308 + while (optind < argc) {
2309 + const char *path = argv[optind++];
2310 + char *devname = NULL;
2311 + path = extract_devname(path, &devname, ind++);
2313 + BlockDriver *drv = NULL;
2314 + BlockDriverState *bs = bdrv_new(devname);
2316 + res = bdrv_open(bs, path, BDRV_O_CACHE_WB , drv);
2318 + unlink(archivename);
2319 + g_error("bdrv_open '%s' failed", path);
2321 + int64_t size = bdrv_getlength(bs);
2322 + int dev_id = vma_writer_register_stream(vmaw, devname, size);
2323 + if (dev_id <= 0) {
2324 + unlink(archivename);
2325 + g_error("vma_writer_register_stream '%s' failed", devname);
2328 + BackupCB *bcb = g_new0(BackupCB, 1);
2330 + bcb->dev_id = dev_id;
2332 + if (backup_job_create(bs, backup_dump_cb, backup_complete_cb,
2334 + unlink(archivename);
2335 + g_error("backup_job_start failed");
2337 + backup_job_start(bs);
2341 + VmaStatus vmastat;
2343 + int last_percent = -1;
2346 + main_loop_wait(false);
2347 + vma_writer_get_status(vmaw, &vmastat);
2351 + uint64_t total = 0;
2352 + uint64_t transferred = 0;
2353 + uint64_t zero_bytes = 0;
2356 + for (i = 0; i < 256; i++) {
2357 + if (vmastat.stream_info[i].size) {
2358 + total += vmastat.stream_info[i].size;
2359 + transferred += vmastat.stream_info[i].transferred;
2360 + zero_bytes += vmastat.stream_info[i].zero_bytes;
2363 + percent = (transferred*100)/total;
2364 + if (percent != last_percent) {
2365 + printf("progress %d%% %zd/%zd %zd\n", percent,
2366 + transferred, total, zero_bytes);
2368 + last_percent = percent;
2372 + if (vmastat.closed) {
2379 + vma_writer_get_status(vmaw, &vmastat);
2382 + for (i = 0; i < 256; i++) {
2383 + VmaStreamInfo *si = &vmastat.stream_info[i];
2385 + printf("image %s: size=%zd zeros=%zd saved=%zd\n", si->devname,
2386 + si->size, si->zero_bytes, si->size - si->zero_bytes);
2391 + if (vmastat.status < 0) {
2392 + unlink(archivename);
2393 + g_error("creating vma archive failed");
2399 +int main(int argc, char **argv)
2401 + const char *cmdname;
2403 + error_set_progname(argv[0]);
2405 + qemu_init_main_loop();
2413 + cmdname = argv[1];
2417 + if (!strcmp(cmdname, "list")) {
2418 + return list_content(argc, argv);
2419 + } else if (!strcmp(cmdname, "create")) {
2420 + return create_archive(argc, argv);
2421 + } else if (!strcmp(cmdname, "extract")) {
2422 + return extract_content(argc, argv);
2428 diff --git a/vma.h b/vma.h
2429 new file mode 100644
2430 index 0000000..76d0dc8
2435 + * VMA: Virtual Machine Archive
2437 + * Copyright (C) Proxmox Server Solutions
2440 + * Dietmar Maurer (dietmar@proxmox.com)
2442 + * This work is licensed under the terms of the GNU GPL, version 2 or later.
2443 + * See the COPYING file in the top-level directory.
2447 +#ifndef BACKUP_VMA_H
2448 +#define BACKUP_VMA_H
2450 +#include "backup.h"
2453 +#define VMA_BLOCK_BITS 12
2454 +#define VMA_BLOCK_SIZE (1<<VMA_BLOCK_BITS)
2455 +#define VMA_CLUSTER_BITS (VMA_BLOCK_BITS+4)
2456 +#define VMA_CLUSTER_SIZE (1<<VMA_CLUSTER_BITS)
2458 +#if VMA_CLUSTER_SIZE != 65536
2459 +#error unexpected cluster size
2462 +#define VMA_EXTENT_HEADER_SIZE 512
2463 +#define VMA_BLOCKS_PER_EXTENT 59
2464 +#define VMA_MAX_CONFIGS 256
2466 +#define VMA_MAX_EXTENT_SIZE \
2467 + (VMA_EXTENT_HEADER_SIZE+VMA_CLUSTER_SIZE*VMA_BLOCKS_PER_EXTENT)
2468 +#if VMA_MAX_EXTENT_SIZE != 3867136
2469 +#error unexpected VMA_EXTENT_SIZE
2472 +/* File Format Definitions */
2474 +#define VMA_MAGIC (GUINT32_TO_BE(('V'<<24)|('M'<<16)|('A'<<8)|0x00))
2475 +#define VMA_EXTENT_MAGIC (GUINT32_TO_BE(('V'<<24)|('M'<<16)|('A'<<8)|'E'))
2477 +typedef struct VmaDeviceInfoHeader {
2478 + uint32_t devname_ptr; /* offset into blob_buffer table */
2479 + uint32_t reserved0;
2480 + uint64_t size; /* device size in bytes */
2481 + uint64_t reserved1;
2482 + uint64_t reserved2;
2483 +} VmaDeviceInfoHeader;
2485 +typedef struct VmaHeader {
2488 + unsigned char uuid[16];
2490 + unsigned char md5sum[16];
2492 + uint32_t blob_buffer_offset;
2493 + uint32_t blob_buffer_size;
2494 + uint32_t header_size;
2496 + unsigned char reserved[1984];
2498 + uint32_t config_names[VMA_MAX_CONFIGS]; /* offset into blob_buffer table */
2499 + uint32_t config_data[VMA_MAX_CONFIGS]; /* offset into blob_buffer table */
2501 + VmaDeviceInfoHeader dev_info[256];
2504 +typedef struct VmaExtentHeader {
2506 + uint16_t reserved1;
2507 + uint16_t block_count;
2508 + unsigned char uuid[16];
2509 + unsigned char md5sum[16];
2510 + uint64_t blockinfo[VMA_BLOCKS_PER_EXTENT];
2513 +/* functions/definitions to read/write vma files */
2515 +typedef struct VmaReader VmaReader;
2517 +typedef struct VmaWriter VmaWriter;
2519 +typedef struct VmaConfigData {
2525 +typedef struct VmaStreamInfo {
2527 + uint64_t cluster_count;
2528 + uint64_t transferred;
2529 + uint64_t zero_bytes;
2534 +typedef struct VmaStatus {
2537 + char errmsg[8192];
2538 + char uuid_str[37];
2539 + VmaStreamInfo stream_info[256];
2542 +typedef struct VmaDeviceInfo {
2543 + uint64_t size; /* device size in bytes */
2544 + const char *devname;
2547 +extern const BackupDriver backup_vma_driver;
2549 +VmaWriter *vma_writer_create(const char *filename, uuid_t uuid, Error **errp);
2550 +int vma_writer_close(VmaWriter *vmaw, Error **errp);
2551 +void vma_writer_destroy(VmaWriter *vmaw);
2552 +int vma_writer_add_config(VmaWriter *vmaw, const char *name, gpointer data,
2554 +int vma_writer_register_stream(VmaWriter *vmaw, const char *devname,
2557 +int64_t coroutine_fn vma_writer_write(VmaWriter *vmaw, uint8_t dev_id,
2558 + int64_t cluster_num, unsigned char *buf,
2559 + size_t *zero_bytes);
2561 +int coroutine_fn vma_writer_close_stream(VmaWriter *vmaw, uint8_t dev_id);
2563 +int vma_writer_get_status(VmaWriter *vmaw, VmaStatus *status);
2564 +void vma_writer_set_error(VmaWriter *vmaw, const char *fmt, ...);
2567 +VmaReader *vma_reader_create(const char *filename, Error **errp);
2568 +void vma_reader_destroy(VmaReader *vmar);
2569 +VmaHeader *vma_reader_get_header(VmaReader *vmar);
2570 +GList *vma_reader_get_config_data(VmaReader *vmar);
2571 +VmaDeviceInfo *vma_reader_get_device_info(VmaReader *vmar, guint8 dev_id);
2572 +int vma_reader_register_bs(VmaReader *vmar, guint8 dev_id,
2573 + BlockDriverState *bs, bool write_zeroes,
2575 +int vma_reader_restore(VmaReader *vmar, int vmstate_fd, bool verbose,
2578 +#endif /* BACKUP_VMA_H */