1 From 6289a43696ca6f713a5d3bb9f95a5adb608a5e13 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 v5 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 | 799 ++++++++++++++++++++++++++++++++++++++++
15 vma-writer.c | 940 +++++++++++++++++++++++++++++++++++++++++++++++
16 vma.c | 561 ++++++++++++++++++++++++++++
18 8 files changed, 2476 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
26 ===================================================================
27 --- new.orig/Makefile 2014-06-17 06:14:27.000000000 +0200
28 +++ new/Makefile 2014-06-17 06:15:02.000000000 +0200
30 include $(SRC_PATH)/libcacard/Makefile
33 -all: $(DOCS) $(TOOLS) $(HELPERS-y) recurse-all modules
34 +all: $(DOCS) $(TOOLS) vma$(EXESUF) $(HELPERS-y) recurse-all modules
36 config-host.h: config-host.h-timestamp
37 config-host.h-timestamp: config-host.mak
39 qemu-img$(EXESUF): qemu-img.o $(block-obj-y) libqemuutil.a libqemustub.a
40 qemu-nbd$(EXESUF): qemu-nbd.o $(block-obj-y) libqemuutil.a libqemustub.a
41 qemu-io$(EXESUF): qemu-io.o $(block-obj-y) libqemuutil.a libqemustub.a
42 +vma$(EXESUF): vma.o vma-reader.o $(block-obj-y) libqemuutil.a libqemustub.a
44 qemu-bridge-helper$(EXESUF): qemu-bridge-helper.o
46 Index: new/vma-reader.c
47 ===================================================================
48 --- /dev/null 1970-01-01 00:00:00.000000000 +0000
49 +++ new/vma-reader.c 2014-06-17 06:15:02.000000000 +0200
52 + * VMA: Virtual Machine Archive
54 + * Copyright (C) 2012 Proxmox Server Solutions
57 + * Dietmar Maurer (dietmar@proxmox.com)
59 + * This work is licensed under the terms of the GNU GPL, version 2 or later.
60 + * See the COPYING file in the top-level directory.
69 +#include <sys/types.h>
70 +#include <sys/stat.h>
73 +#include <uuid/uuid.h>
75 +#include "qemu-common.h"
76 +#include "qemu/timer.h"
77 +#include "qemu/ratelimit.h"
79 +#include "block/block.h"
81 +#define BITS_PER_LONG (sizeof(unsigned long) * CHAR_BIT)
83 +static unsigned char zero_vma_block[VMA_BLOCK_SIZE];
85 +typedef struct VmaRestoreState {
86 + BlockDriverState *bs;
88 + unsigned long *bitmap;
95 + GHashTable *blob_hash;
96 + unsigned char *head_data;
97 + VmaDeviceInfo devinfo[256];
98 + VmaRestoreState rstate[256];
100 + guint8 vmstate_stream;
101 + uint32_t vmstate_clusters;
102 + /* to show restore percentage if run with -v */
104 + int64_t cluster_count;
105 + int64_t clusters_read;
106 + int clusters_read_per;
110 +g_int32_hash(gconstpointer v)
112 + return *(const uint32_t *)v;
116 +g_int32_equal(gconstpointer v1, gconstpointer v2)
118 + return *((const uint32_t *)v1) == *((const uint32_t *)v2);
121 +static int vma_reader_get_bitmap(VmaRestoreState *rstate, int64_t cluster_num)
124 + assert(rstate->bitmap);
126 + unsigned long val, idx, bit;
128 + idx = cluster_num / BITS_PER_LONG;
130 + assert(rstate->bitmap_size > idx);
132 + bit = cluster_num % BITS_PER_LONG;
133 + val = rstate->bitmap[idx];
135 + return !!(val & (1UL << bit));
138 +static void vma_reader_set_bitmap(VmaRestoreState *rstate, int64_t cluster_num,
142 + assert(rstate->bitmap);
144 + unsigned long val, idx, bit;
146 + idx = cluster_num / BITS_PER_LONG;
148 + assert(rstate->bitmap_size > idx);
150 + bit = cluster_num % BITS_PER_LONG;
151 + val = rstate->bitmap[idx];
153 + if (!(val & (1UL << bit))) {
157 + if (val & (1UL << bit)) {
158 + val &= ~(1UL << bit);
161 + rstate->bitmap[idx] = val;
164 +typedef struct VmaBlob {
170 +static const VmaBlob *get_header_blob(VmaReader *vmar, uint32_t pos)
173 + assert(vmar->blob_hash);
175 + return g_hash_table_lookup(vmar->blob_hash, &pos);
178 +static const char *get_header_str(VmaReader *vmar, uint32_t pos)
180 + const VmaBlob *blob = get_header_blob(vmar, pos);
184 + const char *res = (char *)blob->data;
185 + if (res[blob->len-1] != '\0') {
192 +safe_read(int fd, unsigned char *buf, size_t count)
197 + n = read(fd, buf, count);
198 + } while (n < 0 && errno == EINTR);
204 +full_read(int fd, unsigned char *buf, size_t len)
212 + n = safe_read(fd, buf, len);
234 +void vma_reader_destroy(VmaReader *vmar)
238 + if (vmar->fd >= 0) {
242 + if (vmar->cdata_list) {
243 + g_list_free(vmar->cdata_list);
247 + for (i = 1; i < 256; i++) {
248 + if (vmar->rstate[i].bitmap) {
249 + g_free(vmar->rstate[i].bitmap);
253 + if (vmar->md5csum) {
254 + g_checksum_free(vmar->md5csum);
257 + if (vmar->blob_hash) {
258 + g_hash_table_destroy(vmar->blob_hash);
261 + if (vmar->head_data) {
262 + g_free(vmar->head_data);
269 +static int vma_reader_read_head(VmaReader *vmar, Error **errp)
273 + assert(*errp == NULL);
275 + unsigned char md5sum[16];
279 + vmar->head_data = g_malloc(sizeof(VmaHeader));
281 + if (full_read(vmar->fd, vmar->head_data, sizeof(VmaHeader)) !=
282 + sizeof(VmaHeader)) {
283 + error_setg(errp, "can't read vma header - %s",
284 + errno ? g_strerror(errno) : "got EOF");
288 + VmaHeader *h = (VmaHeader *)vmar->head_data;
290 + if (h->magic != VMA_MAGIC) {
291 + error_setg(errp, "not a vma file - wrong magic number");
295 + uint32_t header_size = GUINT32_FROM_BE(h->header_size);
296 + int need = header_size - sizeof(VmaHeader);
298 + error_setg(errp, "wrong vma header size %d", header_size);
302 + vmar->head_data = g_realloc(vmar->head_data, header_size);
303 + h = (VmaHeader *)vmar->head_data;
305 + if (full_read(vmar->fd, vmar->head_data + sizeof(VmaHeader), need) !=
307 + error_setg(errp, "can't read vma header data - %s",
308 + errno ? g_strerror(errno) : "got EOF");
312 + memcpy(md5sum, h->md5sum, 16);
313 + memset(h->md5sum, 0, 16);
315 + g_checksum_reset(vmar->md5csum);
316 + g_checksum_update(vmar->md5csum, vmar->head_data, header_size);
318 + g_checksum_get_digest(vmar->md5csum, (guint8 *)(h->md5sum), &csize);
320 + if (memcmp(md5sum, h->md5sum, 16) != 0) {
321 + error_setg(errp, "wrong vma header chechsum");
325 + /* we can modify header data after checksum verify */
326 + h->header_size = header_size;
328 + h->version = GUINT32_FROM_BE(h->version);
329 + if (h->version != 1) {
330 + error_setg(errp, "wrong vma version %d", h->version);
334 + h->ctime = GUINT64_FROM_BE(h->ctime);
335 + h->blob_buffer_offset = GUINT32_FROM_BE(h->blob_buffer_offset);
336 + h->blob_buffer_size = GUINT32_FROM_BE(h->blob_buffer_size);
338 + uint32_t bstart = h->blob_buffer_offset + 1;
339 + uint32_t bend = h->blob_buffer_offset + h->blob_buffer_size;
341 + if (bstart <= sizeof(VmaHeader)) {
342 + error_setg(errp, "wrong vma blob buffer offset %d",
343 + h->blob_buffer_offset);
347 + if (bend > header_size) {
348 + error_setg(errp, "wrong vma blob buffer size %d/%d",
349 + h->blob_buffer_offset, h->blob_buffer_size);
353 + while ((bstart + 2) <= bend) {
354 + uint32_t size = vmar->head_data[bstart] +
355 + (vmar->head_data[bstart+1] << 8);
356 + if ((bstart + size + 2) <= bend) {
357 + VmaBlob *blob = g_new0(VmaBlob, 1);
358 + blob->start = bstart - h->blob_buffer_offset;
360 + blob->data = vmar->head_data + bstart + 2;
361 + g_hash_table_insert(vmar->blob_hash, &blob->start, blob);
363 + bstart += size + 2;
368 + for (i = 1; i < 256; i++) {
369 + VmaDeviceInfoHeader *dih = &h->dev_info[i];
370 + uint32_t devname_ptr = GUINT32_FROM_BE(dih->devname_ptr);
371 + uint64_t size = GUINT64_FROM_BE(dih->size);
372 + const char *devname = get_header_str(vmar, devname_ptr);
374 + if (size && devname) {
376 + vmar->devinfo[i].size = size;
377 + vmar->devinfo[i].devname = devname;
379 + if (strcmp(devname, "vmstate") == 0) {
380 + vmar->vmstate_stream = i;
386 + error_setg(errp, "vma does not contain data");
390 + for (i = 0; i < VMA_MAX_CONFIGS; i++) {
391 + uint32_t name_ptr = GUINT32_FROM_BE(h->config_names[i]);
392 + uint32_t data_ptr = GUINT32_FROM_BE(h->config_data[i]);
394 + if (!(name_ptr && data_ptr)) {
397 + const char *name = get_header_str(vmar, name_ptr);
398 + const VmaBlob *blob = get_header_blob(vmar, data_ptr);
400 + if (!(name && blob)) {
401 + error_setg(errp, "vma contains invalid data pointers");
405 + VmaConfigData *cdata = g_new0(VmaConfigData, 1);
406 + cdata->name = name;
407 + cdata->data = blob->data;
408 + cdata->len = blob->len;
410 + vmar->cdata_list = g_list_append(vmar->cdata_list, cdata);
416 +VmaReader *vma_reader_create(const char *filename, Error **errp)
421 + VmaReader *vmar = g_new0(VmaReader, 1);
423 + if (strcmp(filename, "-") == 0) {
426 + vmar->fd = open(filename, O_RDONLY);
429 + if (vmar->fd < 0) {
430 + error_setg(errp, "can't open file %s - %s\n", filename,
431 + g_strerror(errno));
435 + vmar->md5csum = g_checksum_new(G_CHECKSUM_MD5);
436 + if (!vmar->md5csum) {
437 + error_setg(errp, "can't allocate cmsum\n");
441 + vmar->blob_hash = g_hash_table_new_full(g_int32_hash, g_int32_equal,
444 + if (vma_reader_read_head(vmar, errp) < 0) {
452 + vma_reader_destroy(vmar);
458 +VmaHeader *vma_reader_get_header(VmaReader *vmar)
461 + assert(vmar->head_data);
463 + return (VmaHeader *)(vmar->head_data);
466 +GList *vma_reader_get_config_data(VmaReader *vmar)
469 + assert(vmar->head_data);
471 + return vmar->cdata_list;
474 +VmaDeviceInfo *vma_reader_get_device_info(VmaReader *vmar, guint8 dev_id)
479 + if (vmar->devinfo[dev_id].size && vmar->devinfo[dev_id].devname) {
480 + return &vmar->devinfo[dev_id];
486 +int vma_reader_register_bs(VmaReader *vmar, guint8 dev_id, BlockDriverState *bs,
487 + bool write_zeroes, Error **errp)
490 + assert(bs != NULL);
492 + assert(vmar->rstate[dev_id].bs == NULL);
494 + int64_t size = bdrv_getlength(bs);
495 + if (size != vmar->devinfo[dev_id].size) {
496 + error_setg(errp, "vma_reader_register_bs for stream %s failed - "
497 + "unexpected size %zd != %zd", vmar->devinfo[dev_id].devname,
498 + size, vmar->devinfo[dev_id].size);
502 + vmar->rstate[dev_id].bs = bs;
503 + vmar->rstate[dev_id].write_zeroes = write_zeroes;
505 + int64_t bitmap_size = (size/BDRV_SECTOR_SIZE) +
506 + (VMA_CLUSTER_SIZE/BDRV_SECTOR_SIZE) * BITS_PER_LONG - 1;
507 + bitmap_size /= (VMA_CLUSTER_SIZE/BDRV_SECTOR_SIZE) * BITS_PER_LONG;
509 + vmar->rstate[dev_id].bitmap_size = bitmap_size;
510 + vmar->rstate[dev_id].bitmap = g_new0(unsigned long, bitmap_size);
512 + vmar->cluster_count += size/VMA_CLUSTER_SIZE;
517 +static ssize_t safe_write(int fd, void *buf, size_t count)
522 + n = write(fd, buf, count);
523 + } while (n < 0 && errno == EINTR);
528 +static size_t full_write(int fd, void *buf, size_t len)
536 + n = safe_write(fd, buf, len);
546 + /* incomplete write ? */
553 +static int restore_write_data(VmaReader *vmar, guint8 dev_id,
554 + BlockDriverState *bs, int vmstate_fd,
555 + unsigned char *buf, int64_t sector_num,
556 + int nb_sectors, Error **errp)
560 + if (dev_id == vmar->vmstate_stream) {
561 + if (vmstate_fd >= 0) {
562 + int len = nb_sectors * BDRV_SECTOR_SIZE;
563 + int res = full_write(vmstate_fd, buf, len);
565 + error_setg(errp, "write vmstate failed %d", res);
570 + int res = bdrv_write(bs, sector_num, buf, nb_sectors);
572 + error_setg(errp, "bdrv_write to %s failed (%d)",
573 + bdrv_get_device_name(bs), res);
579 +static int restore_extent(VmaReader *vmar, unsigned char *buf,
580 + int extent_size, int vmstate_fd,
581 + bool verbose, Error **errp)
586 + VmaExtentHeader *ehead = (VmaExtentHeader *)buf;
587 + int start = VMA_EXTENT_HEADER_SIZE;
590 + for (i = 0; i < VMA_BLOCKS_PER_EXTENT; i++) {
591 + uint64_t block_info = GUINT64_FROM_BE(ehead->blockinfo[i]);
592 + uint64_t cluster_num = block_info & 0xffffffff;
593 + uint8_t dev_id = (block_info >> 32) & 0xff;
594 + uint16_t mask = block_info >> (32+16);
595 + int64_t max_sector;
601 + VmaRestoreState *rstate = &vmar->rstate[dev_id];
602 + BlockDriverState *bs = NULL;
604 + if (dev_id != vmar->vmstate_stream) {
607 + error_setg(errp, "got wrong dev id %d", dev_id);
611 + if (vma_reader_get_bitmap(rstate, cluster_num)) {
612 + error_setg(errp, "found duplicated cluster %zd for stream %s",
613 + cluster_num, vmar->devinfo[dev_id].devname);
616 + vma_reader_set_bitmap(rstate, cluster_num, 1);
618 + max_sector = vmar->devinfo[dev_id].size/BDRV_SECTOR_SIZE;
620 + max_sector = G_MAXINT64;
621 + if (cluster_num != vmar->vmstate_clusters) {
622 + error_setg(errp, "found out of order vmstate data");
625 + vmar->vmstate_clusters++;
628 + vmar->clusters_read++;
631 + time_t duration = time(NULL) - vmar->start_time;
632 + int percent = (vmar->clusters_read*100)/vmar->cluster_count;
633 + if (percent != vmar->clusters_read_per) {
634 + printf("progress %d%% (read %zd bytes, duration %zd sec)\n",
635 + percent, vmar->clusters_read*VMA_CLUSTER_SIZE,
638 + vmar->clusters_read_per = percent;
642 + /* try to write whole clusters to speedup restore */
643 + if (mask == 0xffff) {
644 + if ((start + VMA_CLUSTER_SIZE) > extent_size) {
645 + error_setg(errp, "short vma extent - too many blocks");
648 + int64_t sector_num = (cluster_num * VMA_CLUSTER_SIZE) /
650 + int64_t end_sector = sector_num +
651 + VMA_CLUSTER_SIZE/BDRV_SECTOR_SIZE;
653 + if (end_sector > max_sector) {
654 + end_sector = max_sector;
657 + if (end_sector <= sector_num) {
658 + error_setg(errp, "got wrong block address - write bejond end");
662 + int nb_sectors = end_sector - sector_num;
663 + if (restore_write_data(vmar, dev_id, bs, vmstate_fd, buf + start,
664 + sector_num, nb_sectors, errp) < 0) {
668 + start += VMA_CLUSTER_SIZE;
673 + for (j = 0; j < 16; j++) {
674 + int64_t sector_num = (cluster_num*VMA_CLUSTER_SIZE +
675 + j*VMA_BLOCK_SIZE)/BDRV_SECTOR_SIZE;
677 + int64_t end_sector = sector_num +
678 + VMA_BLOCK_SIZE/BDRV_SECTOR_SIZE;
679 + if (end_sector > max_sector) {
680 + end_sector = max_sector;
684 + if ((start + VMA_BLOCK_SIZE) > extent_size) {
685 + error_setg(errp, "short vma extent - too many blocks");
689 + if (end_sector <= sector_num) {
690 + error_setg(errp, "got wrong block address - "
691 + "write bejond end");
695 + int nb_sectors = end_sector - sector_num;
696 + if (restore_write_data(vmar, dev_id, bs, vmstate_fd,
697 + buf + start, sector_num,
698 + nb_sectors, errp) < 0) {
702 + start += VMA_BLOCK_SIZE;
706 + if (rstate->write_zeroes && (end_sector > sector_num)) {
707 + /* Todo: use bdrv_co_write_zeroes (but that need to
708 + * be run inside coroutine?)
710 + int nb_sectors = end_sector - sector_num;
711 + if (restore_write_data(vmar, dev_id, bs, vmstate_fd,
712 + zero_vma_block, sector_num,
713 + nb_sectors, errp) < 0) {
724 + if (start != extent_size) {
725 + error_setg(errp, "vma extent error - missing blocks");
732 +int vma_reader_restore(VmaReader *vmar, int vmstate_fd, bool verbose,
736 + assert(vmar->head_data);
739 + unsigned char buf[VMA_MAX_EXTENT_SIZE];
741 + unsigned char md5sum[16];
742 + VmaHeader *h = (VmaHeader *)vmar->head_data;
744 + vmar->start_time = time(NULL);
747 + int bytes = full_read(vmar->fd, buf + buf_pos, sizeof(buf) - buf_pos);
749 + error_setg(errp, "read failed - %s", g_strerror(errno));
759 + if (buf_pos < VMA_EXTENT_HEADER_SIZE) {
760 + error_setg(errp, "read short extent (%d bytes)", buf_pos);
764 + VmaExtentHeader *ehead = (VmaExtentHeader *)buf;
766 + /* extract md5sum */
767 + memcpy(md5sum, ehead->md5sum, sizeof(ehead->md5sum));
768 + memset(ehead->md5sum, 0, sizeof(ehead->md5sum));
770 + g_checksum_reset(vmar->md5csum);
771 + g_checksum_update(vmar->md5csum, buf, VMA_EXTENT_HEADER_SIZE);
773 + g_checksum_get_digest(vmar->md5csum, ehead->md5sum, &csize);
775 + if (memcmp(md5sum, ehead->md5sum, 16) != 0) {
776 + error_setg(errp, "wrong vma extent header chechsum");
780 + if (memcmp(h->uuid, ehead->uuid, sizeof(ehead->uuid)) != 0) {
781 + error_setg(errp, "wrong vma extent uuid");
785 + if (ehead->magic != VMA_EXTENT_MAGIC || ehead->reserved1 != 0) {
786 + error_setg(errp, "wrong vma extent header magic");
790 + int block_count = GUINT16_FROM_BE(ehead->block_count);
791 + int extent_size = VMA_EXTENT_HEADER_SIZE + block_count*VMA_BLOCK_SIZE;
793 + if (buf_pos < extent_size) {
794 + error_setg(errp, "short vma extent (%d < %d)", buf_pos,
799 + if (restore_extent(vmar, buf, extent_size, vmstate_fd, verbose,
804 + if (buf_pos > extent_size) {
805 + memmove(buf, buf + extent_size, buf_pos - extent_size);
806 + buf_pos = buf_pos - extent_size;
815 + for (i = 1; i < 256; i++) {
816 + VmaRestoreState *rstate = &vmar->rstate[i];
821 + if (bdrv_flush(rstate->bs) < 0) {
822 + error_setg(errp, "vma bdrv_flush %s failed",
823 + vmar->devinfo[i].devname);
827 + if (vmar->devinfo[i].size &&
828 + (strcmp(vmar->devinfo[i].devname, "vmstate") != 0)) {
829 + assert(rstate->bitmap);
831 + int64_t cluster_num, end;
833 + end = (vmar->devinfo[i].size + VMA_CLUSTER_SIZE - 1) /
836 + for (cluster_num = 0; cluster_num < end; cluster_num++) {
837 + if (!vma_reader_get_bitmap(rstate, cluster_num)) {
838 + error_setg(errp, "detected missing cluster %zd "
839 + "for stream %s", cluster_num,
840 + vmar->devinfo[i].devname);
850 Index: new/vma-writer.c
851 ===================================================================
852 --- /dev/null 1970-01-01 00:00:00.000000000 +0000
853 +++ new/vma-writer.c 2014-06-17 06:15:02.000000000 +0200
856 + * VMA: Virtual Machine Archive
858 + * Copyright (C) 2012 Proxmox Server Solutions
861 + * Dietmar Maurer (dietmar@proxmox.com)
863 + * This work is licensed under the terms of the GNU GPL, version 2 or later.
864 + * See the COPYING file in the top-level directory.
873 +#include <sys/types.h>
874 +#include <sys/stat.h>
877 +#include <uuid/uuid.h>
879 +#include "qemu-common.h"
881 +#include "block/block.h"
882 +#include "monitor/monitor.h"
886 +#define DPRINTF(fmt, ...)\
887 + do { if (DEBUG_VMA) { printf("vma: " fmt, ## __VA_ARGS__); } } while (0)
889 +#define WRITE_BUFFERS 5
891 +typedef struct VmaAIOCB VmaAIOCB;
893 + unsigned char buffer[VMA_MAX_EXTENT_SIZE];
905 + bool header_written;
908 + /* we always write extents */
909 + unsigned char outbuf[VMA_MAX_EXTENT_SIZE];
910 + int outbuf_pos; /* in bytes */
911 + int outbuf_count; /* in VMA_BLOCKS */
912 + uint64_t outbuf_block_info[VMA_BLOCKS_PER_EXTENT];
914 + VmaAIOCB *aiocbs[WRITE_BUFFERS];
917 + GChecksum *md5csum;
918 + CoMutex writer_lock;
919 + CoMutex flush_lock;
920 + Coroutine *co_writer;
922 + /* drive informations */
923 + VmaStreamInfo stream_info[256];
924 + guint stream_count;
926 + guint8 vmstate_stream;
927 + uint32_t vmstate_clusters;
929 + /* header blob table */
930 + char *header_blob_table;
931 + uint32_t header_blob_table_size;
932 + uint32_t header_blob_table_pos;
934 + /* store for config blobs */
935 + uint32_t config_names[VMA_MAX_CONFIGS]; /* offset into blob_buffer table */
936 + uint32_t config_data[VMA_MAX_CONFIGS]; /* offset into blob_buffer table */
937 + uint32_t config_count;
940 +void vma_writer_set_error(VmaWriter *vmaw, const char *fmt, ...)
944 + if (vmaw->status < 0) {
951 + g_vsnprintf(vmaw->errmsg, sizeof(vmaw->errmsg), fmt, ap);
954 + DPRINTF("vma_writer_set_error: %s\n", vmaw->errmsg);
957 +static uint32_t allocate_header_blob(VmaWriter *vmaw, const char *data,
964 + if (!vmaw->header_blob_table ||
965 + (vmaw->header_blob_table_size <
966 + (vmaw->header_blob_table_pos + len + 2))) {
967 + int newsize = vmaw->header_blob_table_size + ((len + 2 + 511)/512)*512;
969 + vmaw->header_blob_table = g_realloc(vmaw->header_blob_table, newsize);
970 + memset(vmaw->header_blob_table + vmaw->header_blob_table_size,
971 + 0, newsize - vmaw->header_blob_table_size);
972 + vmaw->header_blob_table_size = newsize;
975 + uint32_t cpos = vmaw->header_blob_table_pos;
976 + vmaw->header_blob_table[cpos] = len & 255;
977 + vmaw->header_blob_table[cpos+1] = (len >> 8) & 255;
978 + memcpy(vmaw->header_blob_table + cpos + 2, data, len);
979 + vmaw->header_blob_table_pos += len + 2;
983 +static uint32_t allocate_header_string(VmaWriter *vmaw, const char *str)
987 + size_t len = strlen(str) + 1;
989 + return allocate_header_blob(vmaw, str, len);
992 +int vma_writer_add_config(VmaWriter *vmaw, const char *name, gpointer data,
996 + assert(!vmaw->header_written);
997 + assert(vmaw->config_count < VMA_MAX_CONFIGS);
1002 + gchar *basename = g_path_get_basename(name);
1003 + uint32_t name_ptr = allocate_header_string(vmaw, basename);
1010 + uint32_t data_ptr = allocate_header_blob(vmaw, data, len);
1015 + vmaw->config_names[vmaw->config_count] = name_ptr;
1016 + vmaw->config_data[vmaw->config_count] = data_ptr;
1018 + vmaw->config_count++;
1023 +int vma_writer_register_stream(VmaWriter *vmaw, const char *devname,
1028 + assert(!vmaw->status);
1030 + if (vmaw->header_written) {
1031 + vma_writer_set_error(vmaw, "vma_writer_register_stream: header "
1032 + "already written");
1036 + guint n = vmaw->stream_count + 1;
1038 + /* we can have dev_ids form 1 to 255 (0 reserved)
1039 + * 255(-1) reseverd for safety
1042 + vma_writer_set_error(vmaw, "vma_writer_register_stream: "
1043 + "too many drives");
1048 + vma_writer_set_error(vmaw, "vma_writer_register_stream: "
1049 + "got strange size %zd", size);
1053 + DPRINTF("vma_writer_register_stream %s %zu %d\n", devname, size, n);
1055 + vmaw->stream_info[n].devname = g_strdup(devname);
1056 + vmaw->stream_info[n].size = size;
1058 + vmaw->stream_info[n].cluster_count = (size + VMA_CLUSTER_SIZE - 1) /
1061 + vmaw->stream_count = n;
1063 + if (strcmp(devname, "vmstate") == 0) {
1064 + vmaw->vmstate_stream = n;
1070 +static void vma_co_continue_write(void *opaque)
1072 + VmaWriter *vmaw = opaque;
1074 + DPRINTF("vma_co_continue_write\n");
1075 + qemu_coroutine_enter(vmaw->co_writer, NULL);
1078 +static ssize_t coroutine_fn
1079 +vma_co_write(VmaWriter *vmaw, const void *buf, size_t bytes)
1084 + /* atomic writes (we cannot interleave writes) */
1085 + qemu_co_mutex_lock(&vmaw->writer_lock);
1087 + DPRINTF("vma_co_write enter %zd\n", bytes);
1089 + assert(vmaw->co_writer == NULL);
1091 + vmaw->co_writer = qemu_coroutine_self();
1093 + qemu_aio_set_fd_handler(vmaw->fd, NULL, vma_co_continue_write, vmaw);
1095 + DPRINTF("vma_co_write wait until writable\n");
1096 + qemu_coroutine_yield();
1097 + DPRINTF("vma_co_write starting %zd\n", bytes);
1099 + while (done < bytes) {
1100 + ret = write(vmaw->fd, buf + done, bytes - done);
1103 + DPRINTF("vma_co_write written %zd %zd\n", done, ret);
1104 + } else if (ret < 0) {
1105 + if (errno == EAGAIN || errno == EWOULDBLOCK) {
1106 + DPRINTF("vma_co_write yield %zd\n", done);
1107 + qemu_coroutine_yield();
1108 + DPRINTF("vma_co_write restart %zd\n", done);
1110 + vma_writer_set_error(vmaw, "vma_co_write write error - %s",
1111 + g_strerror(errno));
1112 + done = -1; /* always return failure for partial writes */
1115 + } else if (ret == 0) {
1116 + /* should not happen - simply try again */
1120 + qemu_aio_set_fd_handler(vmaw->fd, NULL, NULL, NULL);
1122 + vmaw->co_writer = NULL;
1124 + qemu_co_mutex_unlock(&vmaw->writer_lock);
1126 + DPRINTF("vma_co_write leave %zd\n", done);
1130 +static void coroutine_fn vma_co_writer_task(void *opaque)
1132 + VmaAIOCB *cb = opaque;
1134 + DPRINTF("vma_co_writer_task start\n");
1136 + int64_t done = vma_co_write(cb->vmaw, cb->buffer, cb->bytes);
1137 + DPRINTF("vma_co_writer_task write done %zd\n", done);
1139 + if (done != cb->bytes) {
1140 + DPRINTF("vma_co_writer_task failed write %zd %zd", cb->bytes, done);
1141 + vma_writer_set_error(cb->vmaw, "vma_co_writer_task failed write %zd",
1147 + qemu_co_queue_next(&cb->vmaw->wqueue);
1149 + DPRINTF("vma_co_writer_task end\n");
1152 +static void coroutine_fn vma_queue_flush(VmaWriter *vmaw)
1154 + DPRINTF("vma_queue_flush enter\n");
1160 + VmaAIOCB *cb = NULL;
1161 + for (i = 0; i < WRITE_BUFFERS; i++) {
1162 + if (vmaw->aiocbs[i]->bytes) {
1163 + cb = vmaw->aiocbs[i];
1164 + DPRINTF("FOUND USED AIO BUFFER %d %zd\n", i,
1165 + vmaw->aiocbs[i]->bytes);
1172 + qemu_co_queue_wait(&vmaw->wqueue);
1175 + DPRINTF("vma_queue_flush leave\n");
1179 + * NOTE: pipe buffer size in only 4096 bytes on linux (see 'ulimit -a')
1180 + * So we need to create a coroutione to allow 'parallel' execution.
1182 +static ssize_t coroutine_fn
1183 +vma_queue_write(VmaWriter *vmaw, const void *buf, size_t bytes)
1185 + DPRINTF("vma_queue_write enter %zd\n", bytes);
1189 + assert(bytes <= VMA_MAX_EXTENT_SIZE);
1191 + VmaAIOCB *cb = NULL;
1194 + for (i = 0; i < WRITE_BUFFERS; i++) {
1195 + if (!vmaw->aiocbs[i]->bytes) {
1196 + cb = vmaw->aiocbs[i];
1201 + qemu_co_queue_wait(&vmaw->wqueue);
1205 + memcpy(cb->buffer, buf, bytes);
1206 + cb->bytes = bytes;
1209 + DPRINTF("vma_queue_write start %zd\n", bytes);
1210 + cb->co = qemu_coroutine_create(vma_co_writer_task);
1211 + qemu_coroutine_enter(cb->co, cb);
1213 + DPRINTF("vma_queue_write leave\n");
1218 +VmaWriter *vma_writer_create(const char *filename, uuid_t uuid, Error **errp)
1222 + assert(sizeof(VmaHeader) == (4096 + 8192));
1223 + assert(G_STRUCT_OFFSET(VmaHeader, config_names) == 2044);
1224 + assert(G_STRUCT_OFFSET(VmaHeader, config_data) == 3068);
1225 + assert(G_STRUCT_OFFSET(VmaHeader, dev_info) == 4096);
1226 + assert(sizeof(VmaExtentHeader) == 512);
1228 + VmaWriter *vmaw = g_new0(VmaWriter, 1);
1231 + vmaw->md5csum = g_checksum_new(G_CHECKSUM_MD5);
1232 + if (!vmaw->md5csum) {
1233 + error_setg(errp, "can't allocate cmsum\n");
1237 + if (strstart(filename, "exec:", &p)) {
1238 + vmaw->cmd = popen(p, "w");
1239 + if (vmaw->cmd == NULL) {
1240 + error_setg(errp, "can't popen command '%s' - %s\n", p,
1241 + g_strerror(errno));
1244 + vmaw->fd = fileno(vmaw->cmd);
1246 + /* try to use O_NONBLOCK and O_DIRECT */
1247 + fcntl(vmaw->fd, F_SETFL, fcntl(vmaw->fd, F_GETFL)|O_NONBLOCK);
1248 + fcntl(vmaw->fd, F_SETFL, fcntl(vmaw->fd, F_GETFL)|O_DIRECT);
1253 + const char *tmp_id_str;
1255 + if ((stat(filename, &st) == 0) && S_ISFIFO(st.st_mode)) {
1256 + oflags = O_NONBLOCK|O_DIRECT|O_WRONLY;
1257 + vmaw->fd = qemu_open(filename, oflags, 0644);
1258 + } else if (strstart(filename, "/dev/fdset/", &tmp_id_str)) {
1259 + oflags = O_NONBLOCK|O_DIRECT|O_WRONLY;
1260 + vmaw->fd = qemu_open(filename, oflags, 0644);
1261 + } else if (strstart(filename, "/dev/fdname/", &tmp_id_str)) {
1262 + vmaw->fd = monitor_get_fd(cur_mon, tmp_id_str, errp);
1263 + if (vmaw->fd < 0) {
1266 + /* try to use O_NONBLOCK and O_DIRECT */
1267 + fcntl(vmaw->fd, F_SETFL, fcntl(vmaw->fd, F_GETFL)|O_NONBLOCK);
1268 + fcntl(vmaw->fd, F_SETFL, fcntl(vmaw->fd, F_GETFL)|O_DIRECT);
1270 + oflags = O_NONBLOCK|O_DIRECT|O_WRONLY|O_CREAT|O_EXCL;
1271 + vmaw->fd = qemu_open(filename, oflags, 0644);
1274 + if (vmaw->fd < 0) {
1275 + error_setg(errp, "can't open file %s - %s\n", filename,
1276 + g_strerror(errno));
1281 + /* we use O_DIRECT, so we need to align IO buffers */
1283 + for (i = 0; i < WRITE_BUFFERS; i++) {
1284 + vmaw->aiocbs[i] = qemu_memalign(512, sizeof(VmaAIOCB));
1285 + memset(vmaw->aiocbs[i], 0, sizeof(VmaAIOCB));
1288 + vmaw->outbuf_count = 0;
1289 + vmaw->outbuf_pos = VMA_EXTENT_HEADER_SIZE;
1291 + vmaw->header_blob_table_pos = 1; /* start at pos 1 */
1293 + qemu_co_mutex_init(&vmaw->writer_lock);
1294 + qemu_co_mutex_init(&vmaw->flush_lock);
1295 + qemu_co_queue_init(&vmaw->wqueue);
1297 + uuid_copy(vmaw->uuid, uuid);
1304 + pclose(vmaw->cmd);
1305 + } else if (vmaw->fd >= 0) {
1309 + if (vmaw->md5csum) {
1310 + g_checksum_free(vmaw->md5csum);
1319 +static int coroutine_fn vma_write_header(VmaWriter *vmaw)
1322 + int header_clusters = 8;
1323 + char buf[65536*header_clusters];
1324 + VmaHeader *head = (VmaHeader *)buf;
1328 + DPRINTF("VMA WRITE HEADER\n");
1330 + if (vmaw->status < 0) {
1331 + return vmaw->status;
1334 + memset(buf, 0, sizeof(buf));
1336 + head->magic = VMA_MAGIC;
1337 + head->version = GUINT32_TO_BE(1); /* v1 */
1338 + memcpy(head->uuid, vmaw->uuid, 16);
1340 + time_t ctime = time(NULL);
1341 + head->ctime = GUINT64_TO_BE(ctime);
1343 + if (!vmaw->stream_count) {
1347 + for (i = 0; i < VMA_MAX_CONFIGS; i++) {
1348 + head->config_names[i] = GUINT32_TO_BE(vmaw->config_names[i]);
1349 + head->config_data[i] = GUINT32_TO_BE(vmaw->config_data[i]);
1352 + /* 32 bytes per device (12 used currently) = 8192 bytes max */
1353 + for (i = 1; i <= 254; i++) {
1354 + VmaStreamInfo *si = &vmaw->stream_info[i];
1356 + assert(si->devname);
1357 + uint32_t devname_ptr = allocate_header_string(vmaw, si->devname);
1358 + if (!devname_ptr) {
1361 + head->dev_info[i].devname_ptr = GUINT32_TO_BE(devname_ptr);
1362 + head->dev_info[i].size = GUINT64_TO_BE(si->size);
1366 + uint32_t header_size = sizeof(VmaHeader) + vmaw->header_blob_table_size;
1367 + head->header_size = GUINT32_TO_BE(header_size);
1369 + if (header_size > sizeof(buf)) {
1370 + return -1; /* just to be sure */
1373 + uint32_t blob_buffer_offset = sizeof(VmaHeader);
1374 + memcpy(buf + blob_buffer_offset, vmaw->header_blob_table,
1375 + vmaw->header_blob_table_size);
1376 + head->blob_buffer_offset = GUINT32_TO_BE(blob_buffer_offset);
1377 + head->blob_buffer_size = GUINT32_TO_BE(vmaw->header_blob_table_pos);
1379 + g_checksum_reset(vmaw->md5csum);
1380 + g_checksum_update(vmaw->md5csum, (const guchar *)buf, header_size);
1382 + g_checksum_get_digest(vmaw->md5csum, (guint8 *)(head->md5sum), &csize);
1384 + return vma_queue_write(vmaw, buf, header_size);
1387 +static int coroutine_fn vma_writer_flush(VmaWriter *vmaw)
1394 + if (vmaw->status < 0) {
1395 + return vmaw->status;
1398 + if (!vmaw->header_written) {
1399 + vmaw->header_written = true;
1400 + ret = vma_write_header(vmaw);
1402 + vma_writer_set_error(vmaw, "vma_writer_flush: write header failed");
1407 + DPRINTF("VMA WRITE FLUSH %d %d\n", vmaw->outbuf_count, vmaw->outbuf_pos);
1410 + VmaExtentHeader *ehead = (VmaExtentHeader *)vmaw->outbuf;
1412 + ehead->magic = VMA_EXTENT_MAGIC;
1413 + ehead->reserved1 = 0;
1415 + for (i = 0; i < VMA_BLOCKS_PER_EXTENT; i++) {
1416 + ehead->blockinfo[i] = GUINT64_TO_BE(vmaw->outbuf_block_info[i]);
1419 + guint16 block_count = (vmaw->outbuf_pos - VMA_EXTENT_HEADER_SIZE) /
1422 + ehead->block_count = GUINT16_TO_BE(block_count);
1424 + memcpy(ehead->uuid, vmaw->uuid, sizeof(ehead->uuid));
1425 + memset(ehead->md5sum, 0, sizeof(ehead->md5sum));
1427 + g_checksum_reset(vmaw->md5csum);
1428 + g_checksum_update(vmaw->md5csum, vmaw->outbuf, VMA_EXTENT_HEADER_SIZE);
1430 + g_checksum_get_digest(vmaw->md5csum, ehead->md5sum, &csize);
1432 + int bytes = vmaw->outbuf_pos;
1433 + ret = vma_queue_write(vmaw, vmaw->outbuf, bytes);
1434 + if (ret != bytes) {
1435 + vma_writer_set_error(vmaw, "vma_writer_flush: failed write");
1438 + vmaw->outbuf_count = 0;
1439 + vmaw->outbuf_pos = VMA_EXTENT_HEADER_SIZE;
1441 + for (i = 0; i < VMA_BLOCKS_PER_EXTENT; i++) {
1442 + vmaw->outbuf_block_info[i] = 0;
1445 + return vmaw->status;
1448 +static int vma_count_open_streams(VmaWriter *vmaw)
1450 + g_assert(vmaw != NULL);
1453 + int open_drives = 0;
1454 + for (i = 0; i <= 255; i++) {
1455 + if (vmaw->stream_info[i].size && !vmaw->stream_info[i].finished) {
1460 + return open_drives;
1464 + * all jobs should call this when there is no more data
1465 + * Returns: number of remaining stream (0 ==> finished)
1468 +vma_writer_close_stream(VmaWriter *vmaw, uint8_t dev_id)
1470 + g_assert(vmaw != NULL);
1472 + DPRINTF("vma_writer_set_status %d\n", dev_id);
1473 + if (!vmaw->stream_info[dev_id].size) {
1474 + vma_writer_set_error(vmaw, "vma_writer_close_stream: "
1475 + "no such stream %d", dev_id);
1478 + if (vmaw->stream_info[dev_id].finished) {
1479 + vma_writer_set_error(vmaw, "vma_writer_close_stream: "
1480 + "stream already closed %d", dev_id);
1484 + vmaw->stream_info[dev_id].finished = true;
1486 + int open_drives = vma_count_open_streams(vmaw);
1488 + if (open_drives <= 0) {
1489 + DPRINTF("vma_writer_set_status all drives completed\n");
1490 + qemu_co_mutex_lock(&vmaw->flush_lock);
1491 + int ret = vma_writer_flush(vmaw);
1492 + qemu_co_mutex_unlock(&vmaw->flush_lock);
1494 + vma_writer_set_error(vmaw, "vma_writer_close_stream: flush failed");
1498 + return open_drives;
1501 +int vma_writer_get_status(VmaWriter *vmaw, VmaStatus *status)
1505 + g_assert(vmaw != NULL);
1508 + status->status = vmaw->status;
1509 + g_strlcpy(status->errmsg, vmaw->errmsg, sizeof(status->errmsg));
1510 + for (i = 0; i <= 255; i++) {
1511 + status->stream_info[i] = vmaw->stream_info[i];
1514 + uuid_unparse_lower(vmaw->uuid, status->uuid_str);
1517 + status->closed = vmaw->closed;
1519 + return vmaw->status;
1522 +static int vma_writer_get_buffer(VmaWriter *vmaw)
1526 + qemu_co_mutex_lock(&vmaw->flush_lock);
1528 + /* wait until buffer is available */
1529 + while (vmaw->outbuf_count >= (VMA_BLOCKS_PER_EXTENT - 1)) {
1530 + ret = vma_writer_flush(vmaw);
1532 + vma_writer_set_error(vmaw, "vma_writer_get_buffer: flush failed");
1537 + qemu_co_mutex_unlock(&vmaw->flush_lock);
1543 +int64_t coroutine_fn
1544 +vma_writer_write(VmaWriter *vmaw, uint8_t dev_id, int64_t cluster_num,
1545 + unsigned char *buf, size_t *zero_bytes)
1547 + g_assert(vmaw != NULL);
1548 + g_assert(zero_bytes != NULL);
1552 + if (vmaw->status < 0) {
1553 + return vmaw->status;
1556 + if (!dev_id || !vmaw->stream_info[dev_id].size) {
1557 + vma_writer_set_error(vmaw, "vma_writer_write: "
1558 + "no such stream %d", dev_id);
1562 + if (vmaw->stream_info[dev_id].finished) {
1563 + vma_writer_set_error(vmaw, "vma_writer_write: "
1564 + "stream already closed %d", dev_id);
1569 + if (cluster_num >= (((uint64_t)1)<<32)) {
1570 + vma_writer_set_error(vmaw, "vma_writer_write: "
1571 + "cluster number out of range");
1575 + if (dev_id == vmaw->vmstate_stream) {
1576 + if (cluster_num != vmaw->vmstate_clusters) {
1577 + vma_writer_set_error(vmaw, "vma_writer_write: "
1578 + "non sequential vmstate write");
1580 + vmaw->vmstate_clusters++;
1581 + } else if (cluster_num >= vmaw->stream_info[dev_id].cluster_count) {
1582 + vma_writer_set_error(vmaw, "vma_writer_write: cluster number too big");
1586 + /* wait until buffer is available */
1587 + if (vma_writer_get_buffer(vmaw) < 0) {
1588 + vma_writer_set_error(vmaw, "vma_writer_write: "
1589 + "vma_writer_get_buffer failed");
1593 + DPRINTF("VMA WRITE %d %zd\n", dev_id, cluster_num);
1595 + uint16_t mask = 0;
1600 + for (i = 0; i < 16; i++) {
1601 + unsigned char *vmablock = buf + (i*VMA_BLOCK_SIZE);
1602 + if (!buffer_is_zero(vmablock, VMA_BLOCK_SIZE)) {
1604 + memcpy(vmaw->outbuf + vmaw->outbuf_pos, vmablock,
1606 + vmaw->outbuf_pos += VMA_BLOCK_SIZE;
1608 + DPRINTF("VMA WRITE %zd ZERO BLOCK %d\n", cluster_num, i);
1609 + vmaw->stream_info[dev_id].zero_bytes += VMA_BLOCK_SIZE;
1610 + *zero_bytes += VMA_BLOCK_SIZE;
1616 + DPRINTF("VMA WRITE %zd ZERO CLUSTER\n", cluster_num);
1617 + vmaw->stream_info[dev_id].zero_bytes += VMA_CLUSTER_SIZE;
1618 + *zero_bytes += VMA_CLUSTER_SIZE;
1621 + uint64_t block_info = ((uint64_t)mask) << (32+16);
1622 + block_info |= ((uint64_t)dev_id) << 32;
1623 + block_info |= (cluster_num & 0xffffffff);
1624 + vmaw->outbuf_block_info[vmaw->outbuf_count] = block_info;
1626 + DPRINTF("VMA WRITE MASK %zd %zx\n", cluster_num, block_info);
1628 + vmaw->outbuf_count++;
1630 + /** NOTE: We allways write whole clusters, but we correctly set
1631 + * transferred bytes. So transferred == size when when everything
1634 + size_t transferred = VMA_CLUSTER_SIZE;
1636 + if (dev_id != vmaw->vmstate_stream) {
1637 + uint64_t last = (cluster_num + 1) * VMA_CLUSTER_SIZE;
1638 + if (last > vmaw->stream_info[dev_id].size) {
1639 + uint64_t diff = last - vmaw->stream_info[dev_id].size;
1640 + if (diff >= VMA_CLUSTER_SIZE) {
1641 + vma_writer_set_error(vmaw, "vma_writer_write: "
1642 + "read after last cluster");
1645 + transferred -= diff;
1649 + vmaw->stream_info[dev_id].transferred += transferred;
1651 + return transferred;
1654 +int vma_writer_close(VmaWriter *vmaw, Error **errp)
1656 + g_assert(vmaw != NULL);
1660 + vma_queue_flush(vmaw);
1662 + /* this should not happen - just to be sure */
1663 + while (!qemu_co_queue_empty(&vmaw->wqueue)) {
1664 + DPRINTF("vma_writer_close wait\n");
1665 + co_sleep_ns(QEMU_CLOCK_REALTIME, 1000000);
1669 + if (pclose(vmaw->cmd) < 0) {
1670 + vma_writer_set_error(vmaw, "vma_writer_close: "
1671 + "pclose failed - %s", g_strerror(errno));
1674 + if (close(vmaw->fd) < 0) {
1675 + vma_writer_set_error(vmaw, "vma_writer_close: "
1676 + "close failed - %s", g_strerror(errno));
1680 + for (i = 0; i <= 255; i++) {
1681 + VmaStreamInfo *si = &vmaw->stream_info[i];
1683 + if (!si->finished) {
1684 + vma_writer_set_error(vmaw, "vma_writer_close: "
1685 + "detected open stream '%s'", si->devname);
1686 + } else if ((si->transferred != si->size) &&
1687 + (i != vmaw->vmstate_stream)) {
1688 + vma_writer_set_error(vmaw, "vma_writer_close: "
1689 + "incomplete stream '%s' (%zd != %zd)",
1690 + si->devname, si->transferred, si->size);
1695 + for (i = 0; i <= 255; i++) {
1696 + vmaw->stream_info[i].finished = 1; /* mark as closed */
1701 + if (vmaw->status < 0 && *errp == NULL) {
1702 + error_setg(errp, "%s", vmaw->errmsg);
1705 + return vmaw->status;
1708 +void vma_writer_destroy(VmaWriter *vmaw)
1714 + for (i = 0; i <= 255; i++) {
1715 + if (vmaw->stream_info[i].devname) {
1716 + g_free(vmaw->stream_info[i].devname);
1720 + if (vmaw->md5csum) {
1721 + g_checksum_free(vmaw->md5csum);
1724 + for (i = 0; i < WRITE_BUFFERS; i++) {
1725 + free(vmaw->aiocbs[i]);
1731 ===================================================================
1732 --- /dev/null 1970-01-01 00:00:00.000000000 +0000
1733 +++ new/vma.c 2014-06-17 06:15:02.000000000 +0200
1736 + * VMA: Virtual Machine Archive
1738 + * Copyright (C) 2012-2013 Proxmox Server Solutions
1741 + * Dietmar Maurer (dietmar@proxmox.com)
1743 + * This work is licensed under the terms of the GNU GPL, version 2 or later.
1744 + * See the COPYING file in the top-level directory.
1750 +#include <unistd.h>
1752 +#include <string.h>
1753 +#include <sys/types.h>
1754 +#include <sys/stat.h>
1759 +#include "qemu-common.h"
1760 +#include "qemu/error-report.h"
1761 +#include "qemu/main-loop.h"
1763 +static void help(void)
1765 + const char *help_msg =
1766 + "usage: vma command [command options]\n"
1768 + "vma list <filename>\n"
1769 + "vma create <filename> [-c config] <archive> pathname ...\n"
1770 + "vma extract <filename> [-r <fifo>] <targetdir>\n"
1773 + printf("%s", help_msg);
1777 +static const char *extract_devname(const char *path, char **devname, int index)
1781 + const char *sep = strchr(path, '=');
1784 + *devname = g_strndup(path, sep - path);
1788 + *devname = g_strdup_printf("disk%d", index);
1797 +static void print_content(VmaReader *vmar)
1801 + VmaHeader *head = vma_reader_get_header(vmar);
1803 + GList *l = vma_reader_get_config_data(vmar);
1804 + while (l && l->data) {
1805 + VmaConfigData *cdata = (VmaConfigData *)l->data;
1806 + l = g_list_next(l);
1807 + printf("CFG: size: %d name: %s\n", cdata->len, cdata->name);
1811 + VmaDeviceInfo *di;
1812 + for (i = 1; i < 255; i++) {
1813 + di = vma_reader_get_device_info(vmar, i);
1815 + if (strcmp(di->devname, "vmstate") == 0) {
1816 + printf("VMSTATE: dev_id=%d memory: %zd\n", i, di->size);
1818 + printf("DEV: dev_id=%d size: %zd devname: %s\n",
1819 + i, di->size, di->devname);
1823 + /* ctime is the last entry we print */
1824 + printf("CTIME: %s", ctime(&head->ctime));
1828 +static int list_content(int argc, char **argv)
1831 + const char *filename;
1834 + c = getopt(argc, argv, "h");
1844 + g_assert_not_reached();
1848 + /* Get the filename */
1849 + if ((optind + 1) != argc) {
1852 + filename = argv[optind++];
1854 + Error *errp = NULL;
1855 + VmaReader *vmar = vma_reader_create(filename, &errp);
1858 + g_error("%s", error_get_pretty(errp));
1861 + print_content(vmar);
1863 + vma_reader_destroy(vmar);
1868 +typedef struct RestoreMap {
1874 +static int extract_content(int argc, char **argv)
1878 + const char *filename;
1879 + const char *dirname;
1880 + const char *readmap = NULL;
1883 + c = getopt(argc, argv, "hvr:");
1903 + /* Get the filename */
1904 + if ((optind + 2) != argc) {
1907 + filename = argv[optind++];
1908 + dirname = argv[optind++];
1910 + Error *errp = NULL;
1911 + VmaReader *vmar = vma_reader_create(filename, &errp);
1914 + g_error("%s", error_get_pretty(errp));
1917 + if (mkdir(dirname, 0777) < 0) {
1918 + g_error("unable to create target directory %s - %s",
1919 + dirname, g_strerror(errno));
1922 + GList *l = vma_reader_get_config_data(vmar);
1923 + while (l && l->data) {
1924 + VmaConfigData *cdata = (VmaConfigData *)l->data;
1925 + l = g_list_next(l);
1926 + char *cfgfn = g_strdup_printf("%s/%s", dirname, cdata->name);
1927 + GError *err = NULL;
1928 + if (!g_file_set_contents(cfgfn, (gchar *)cdata->data, cdata->len,
1930 + g_error("unable to write file: %s", err->message);
1934 + GHashTable *devmap = g_hash_table_new(g_str_hash, g_str_equal);
1937 + print_content(vmar);
1939 + FILE *map = fopen(readmap, "r");
1941 + g_error("unable to open fifo %s - %s", readmap, g_strerror(errno));
1946 + char *line = fgets(inbuf, sizeof(inbuf), map);
1947 + if (!line || line[0] == '\0' || !strcmp(line, "done\n")) {
1950 + int len = strlen(line);
1951 + if (line[len - 1] == '\n') {
1952 + line[len - 1] = '\0';
1960 + if (line[0] == '0' && line[1] == ':') {
1962 + write_zero = false;
1963 + } else if (line[0] == '1' && line[1] == ':') {
1965 + write_zero = true;
1967 + g_error("read map failed - parse error ('%s')", inbuf);
1970 + char *devname = NULL;
1971 + path = extract_devname(path, &devname, -1);
1973 + g_error("read map failed - no dev name specified ('%s')",
1977 + RestoreMap *map = g_new0(RestoreMap, 1);
1978 + map->devname = g_strdup(devname);
1979 + map->path = g_strdup(path);
1980 + map->write_zero = write_zero;
1982 + g_hash_table_insert(devmap, map->devname, map);
1988 + int vmstate_fd = -1;
1989 + guint8 vmstate_stream = 0;
1991 + for (i = 1; i < 255; i++) {
1992 + VmaDeviceInfo *di = vma_reader_get_device_info(vmar, i);
1993 + if (di && (strcmp(di->devname, "vmstate") == 0)) {
1994 + vmstate_stream = i;
1995 + char *statefn = g_strdup_printf("%s/vmstate.bin", dirname);
1996 + vmstate_fd = open(statefn, O_WRONLY|O_CREAT|O_EXCL, 0644);
1997 + if (vmstate_fd < 0) {
1998 + g_error("create vmstate file '%s' failed - %s", statefn,
1999 + g_strerror(errno));
2003 + char *devfn = NULL;
2004 + int flags = BDRV_O_RDWR|BDRV_O_CACHE_WB;
2005 + bool write_zero = true;
2009 + map = (RestoreMap *)g_hash_table_lookup(devmap, di->devname);
2010 + if (map == NULL) {
2011 + g_error("no device name mapping for %s", di->devname);
2013 + devfn = map->path;
2014 + write_zero = map->write_zero;
2016 + devfn = g_strdup_printf("%s/tmp-disk-%s.raw",
2017 + dirname, di->devname);
2018 + printf("DEVINFO %s %zd\n", devfn, di->size);
2020 + bdrv_img_create(devfn, "raw", NULL, NULL, NULL, di->size,
2023 + g_error("can't create file %s: %s", devfn,
2024 + error_get_pretty(errp));
2027 + /* Note: we created an empty file above, so there is no
2028 + * need to write zeroes (so we generate a sparse file)
2030 + write_zero = false;
2033 + BlockDriverState *bs = bdrv_new(di->devname, &errp);
2034 + if (errp || bdrv_open(&bs, devfn, NULL, NULL, flags, NULL, &errp)) {
2035 + g_error("can't open file %s - %s", devfn,
2036 + error_get_pretty(errp));
2038 + if (vma_reader_register_bs(vmar, i, bs, write_zero, &errp) < 0) {
2039 + g_error("%s", error_get_pretty(errp));
2048 + if (vma_reader_restore(vmar, vmstate_fd, verbose, &errp) < 0) {
2049 + g_error("restore failed - %s", error_get_pretty(errp));
2053 + for (i = 1; i < 255; i++) {
2054 + VmaDeviceInfo *di = vma_reader_get_device_info(vmar, i);
2055 + if (di && (i != vmstate_stream)) {
2056 + char *tmpfn = g_strdup_printf("%s/tmp-disk-%s.raw",
2057 + dirname, di->devname);
2058 + char *fn = g_strdup_printf("%s/disk-%s.raw",
2059 + dirname, di->devname);
2060 + if (rename(tmpfn, fn) != 0) {
2061 + g_error("rename %s to %s failed - %s",
2062 + tmpfn, fn, g_strerror(errno));
2068 + vma_reader_destroy(vmar);
2075 +typedef struct BackupJob {
2076 + BlockDriverState *bs;
2082 +#define BACKUP_SECTORS_PER_CLUSTER (VMA_CLUSTER_SIZE / BDRV_SECTOR_SIZE)
2084 +static void coroutine_fn backup_run(void *opaque)
2086 + BackupJob *job = (BackupJob *)opaque;
2088 + QEMUIOVector qiov;
2090 + int64_t start, end;
2093 + unsigned char *buf = qemu_blockalign(job->bs, VMA_CLUSTER_SIZE);
2096 + end = DIV_ROUND_UP(job->len / BDRV_SECTOR_SIZE,
2097 + BACKUP_SECTORS_PER_CLUSTER);
2099 + for (; start < end; start++) {
2100 + iov.iov_base = buf;
2101 + iov.iov_len = VMA_CLUSTER_SIZE;
2102 + qemu_iovec_init_external(&qiov, &iov, 1);
2104 + ret = bdrv_co_readv(job->bs, start * BACKUP_SECTORS_PER_CLUSTER,
2105 + BACKUP_SECTORS_PER_CLUSTER, &qiov);
2107 + vma_writer_set_error(job->vmaw, "read error", -1);
2112 + if (vma_writer_write(job->vmaw, job->dev_id, start, buf, &zb) < 0) {
2113 + vma_writer_set_error(job->vmaw, "backup_dump_cb vma_writer_write failed", -1);
2120 + if (vma_writer_close_stream(job->vmaw, job->dev_id) <= 0) {
2121 + Error *err = NULL;
2122 + if (vma_writer_close(job->vmaw, &err) != 0) {
2123 + g_warning("vma_writer_close failed %s", error_get_pretty(err));
2128 +static int create_archive(int argc, char **argv)
2132 + const char *archivename;
2133 + GList *config_files = NULL;
2136 + c = getopt(argc, argv, "hvc:");
2146 + config_files = g_list_append(config_files, optarg);
2152 + g_assert_not_reached();
2157 + /* make sure we have archive name and at least one path */
2158 + if ((optind + 2) > argc) {
2162 + archivename = argv[optind++];
2165 + uuid_generate(uuid);
2167 + Error *local_err = NULL;
2168 + VmaWriter *vmaw = vma_writer_create(archivename, uuid, &local_err);
2170 + if (vmaw == NULL) {
2171 + g_error("%s", error_get_pretty(local_err));
2174 + GList *l = config_files;
2175 + while (l && l->data) {
2176 + char *name = l->data;
2177 + char *cdata = NULL;
2179 + GError *err = NULL;
2180 + if (!g_file_get_contents(name, &cdata, &clen, &err)) {
2181 + unlink(archivename);
2182 + g_error("Unable to read file: %s", err->message);
2185 + if (vma_writer_add_config(vmaw, name, cdata, clen) != 0) {
2186 + unlink(archivename);
2187 + g_error("Unable to append config data %s (len = %zd)",
2190 + l = g_list_next(l);
2194 + while (optind < argc) {
2195 + const char *path = argv[optind++];
2196 + char *devname = NULL;
2197 + path = extract_devname(path, &devname, ind++);
2199 + BlockDriver *drv = NULL;
2200 + Error *errp = NULL;
2201 + BlockDriverState *bs = bdrv_new(devname, &errp);
2204 + g_error("bdrv_new '%s' failed - %s", devname, error_get_pretty(errp));
2207 + res = bdrv_open(&bs, path, NULL, NULL, BDRV_O_CACHE_WB , drv, &errp);
2209 + unlink(archivename);
2210 + g_error("bdrv_open '%s' failed - %s", path, error_get_pretty(errp));
2212 + int64_t size = bdrv_getlength(bs);
2213 + int dev_id = vma_writer_register_stream(vmaw, devname, size);
2214 + if (dev_id <= 0) {
2215 + unlink(archivename);
2216 + g_error("vma_writer_register_stream '%s' failed", devname);
2219 + BackupJob *job = g_new0(BackupJob, 1);
2223 + job->dev_id = dev_id;
2225 + Coroutine *co = qemu_coroutine_create(backup_run);
2226 + qemu_coroutine_enter(co, job);
2229 + VmaStatus vmastat;
2231 + int last_percent = -1;
2234 + main_loop_wait(false);
2235 + vma_writer_get_status(vmaw, &vmastat);
2239 + uint64_t total = 0;
2240 + uint64_t transferred = 0;
2241 + uint64_t zero_bytes = 0;
2244 + for (i = 0; i < 256; i++) {
2245 + if (vmastat.stream_info[i].size) {
2246 + total += vmastat.stream_info[i].size;
2247 + transferred += vmastat.stream_info[i].transferred;
2248 + zero_bytes += vmastat.stream_info[i].zero_bytes;
2251 + percent = (transferred*100)/total;
2252 + if (percent != last_percent) {
2253 + fprintf(stderr, "progress %d%% %zd/%zd %zd\n", percent,
2254 + transferred, total, zero_bytes);
2257 + last_percent = percent;
2261 + if (vmastat.closed) {
2268 + vma_writer_get_status(vmaw, &vmastat);
2271 + for (i = 0; i < 256; i++) {
2272 + VmaStreamInfo *si = &vmastat.stream_info[i];
2274 + fprintf(stderr, "image %s: size=%zd zeros=%zd saved=%zd\n",
2275 + si->devname, si->size, si->zero_bytes,
2276 + si->size - si->zero_bytes);
2281 + if (vmastat.status < 0) {
2282 + unlink(archivename);
2283 + g_error("creating vma archive failed");
2289 +int main(int argc, char **argv)
2291 + const char *cmdname;
2293 + error_set_progname(argv[0]);
2295 + qemu_init_main_loop();
2303 + cmdname = argv[1];
2307 + if (!strcmp(cmdname, "list")) {
2308 + return list_content(argc, argv);
2309 + } else if (!strcmp(cmdname, "create")) {
2310 + return create_archive(argc, argv);
2311 + } else if (!strcmp(cmdname, "extract")) {
2312 + return extract_content(argc, argv);
2319 ===================================================================
2320 --- /dev/null 1970-01-01 00:00:00.000000000 +0000
2321 +++ new/vma.h 2014-06-17 06:15:02.000000000 +0200
2324 + * VMA: Virtual Machine Archive
2326 + * Copyright (C) Proxmox Server Solutions
2329 + * Dietmar Maurer (dietmar@proxmox.com)
2331 + * This work is licensed under the terms of the GNU GPL, version 2 or later.
2332 + * See the COPYING file in the top-level directory.
2336 +#ifndef BACKUP_VMA_H
2337 +#define BACKUP_VMA_H
2339 +#include <uuid/uuid.h>
2340 +#include "qapi/error.h"
2341 +#include "block/block.h"
2343 +#define VMA_BLOCK_BITS 12
2344 +#define VMA_BLOCK_SIZE (1<<VMA_BLOCK_BITS)
2345 +#define VMA_CLUSTER_BITS (VMA_BLOCK_BITS+4)
2346 +#define VMA_CLUSTER_SIZE (1<<VMA_CLUSTER_BITS)
2348 +#if VMA_CLUSTER_SIZE != 65536
2349 +#error unexpected cluster size
2352 +#define VMA_EXTENT_HEADER_SIZE 512
2353 +#define VMA_BLOCKS_PER_EXTENT 59
2354 +#define VMA_MAX_CONFIGS 256
2356 +#define VMA_MAX_EXTENT_SIZE \
2357 + (VMA_EXTENT_HEADER_SIZE+VMA_CLUSTER_SIZE*VMA_BLOCKS_PER_EXTENT)
2358 +#if VMA_MAX_EXTENT_SIZE != 3867136
2359 +#error unexpected VMA_EXTENT_SIZE
2362 +/* File Format Definitions */
2364 +#define VMA_MAGIC (GUINT32_TO_BE(('V'<<24)|('M'<<16)|('A'<<8)|0x00))
2365 +#define VMA_EXTENT_MAGIC (GUINT32_TO_BE(('V'<<24)|('M'<<16)|('A'<<8)|'E'))
2367 +typedef struct VmaDeviceInfoHeader {
2368 + uint32_t devname_ptr; /* offset into blob_buffer table */
2369 + uint32_t reserved0;
2370 + uint64_t size; /* device size in bytes */
2371 + uint64_t reserved1;
2372 + uint64_t reserved2;
2373 +} VmaDeviceInfoHeader;
2375 +typedef struct VmaHeader {
2378 + unsigned char uuid[16];
2380 + unsigned char md5sum[16];
2382 + uint32_t blob_buffer_offset;
2383 + uint32_t blob_buffer_size;
2384 + uint32_t header_size;
2386 + unsigned char reserved[1984];
2388 + uint32_t config_names[VMA_MAX_CONFIGS]; /* offset into blob_buffer table */
2389 + uint32_t config_data[VMA_MAX_CONFIGS]; /* offset into blob_buffer table */
2391 + uint32_t reserved1;
2393 + VmaDeviceInfoHeader dev_info[256];
2396 +typedef struct VmaExtentHeader {
2398 + uint16_t reserved1;
2399 + uint16_t block_count;
2400 + unsigned char uuid[16];
2401 + unsigned char md5sum[16];
2402 + uint64_t blockinfo[VMA_BLOCKS_PER_EXTENT];
2405 +/* functions/definitions to read/write vma files */
2407 +typedef struct VmaReader VmaReader;
2409 +typedef struct VmaWriter VmaWriter;
2411 +typedef struct VmaConfigData {
2417 +typedef struct VmaStreamInfo {
2419 + uint64_t cluster_count;
2420 + uint64_t transferred;
2421 + uint64_t zero_bytes;
2426 +typedef struct VmaStatus {
2429 + char errmsg[8192];
2430 + char uuid_str[37];
2431 + VmaStreamInfo stream_info[256];
2434 +typedef struct VmaDeviceInfo {
2435 + uint64_t size; /* device size in bytes */
2436 + const char *devname;
2439 +VmaWriter *vma_writer_create(const char *filename, uuid_t uuid, Error **errp);
2440 +int vma_writer_close(VmaWriter *vmaw, Error **errp);
2441 +void vma_writer_destroy(VmaWriter *vmaw);
2442 +int vma_writer_add_config(VmaWriter *vmaw, const char *name, gpointer data,
2444 +int vma_writer_register_stream(VmaWriter *vmaw, const char *devname,
2447 +int64_t coroutine_fn vma_writer_write(VmaWriter *vmaw, uint8_t dev_id,
2448 + int64_t cluster_num, unsigned char *buf,
2449 + size_t *zero_bytes);
2451 +int coroutine_fn vma_writer_close_stream(VmaWriter *vmaw, uint8_t dev_id);
2453 +int vma_writer_get_status(VmaWriter *vmaw, VmaStatus *status);
2454 +void vma_writer_set_error(VmaWriter *vmaw, const char *fmt, ...);
2457 +VmaReader *vma_reader_create(const char *filename, Error **errp);
2458 +void vma_reader_destroy(VmaReader *vmar);
2459 +VmaHeader *vma_reader_get_header(VmaReader *vmar);
2460 +GList *vma_reader_get_config_data(VmaReader *vmar);
2461 +VmaDeviceInfo *vma_reader_get_device_info(VmaReader *vmar, guint8 dev_id);
2462 +int vma_reader_register_bs(VmaReader *vmar, guint8 dev_id,
2463 + BlockDriverState *bs, bool write_zeroes,
2465 +int vma_reader_restore(VmaReader *vmar, int vmstate_fd, bool verbose,
2468 +#endif /* BACKUP_VMA_H */
2469 Index: new/Makefile.objs
2470 ===================================================================
2471 --- new.orig/Makefile.objs 2014-06-17 06:14:27.000000000 +0200
2472 +++ new/Makefile.objs 2014-06-17 06:15:02.000000000 +0200
2474 block-obj-y += block/
2475 block-obj-y += qapi-types.o qapi-visit.o
2476 block-obj-y += qemu-io-cmds.o
2477 +block-obj-y += vma-writer.o
2479 block-obj-y += qemu-coroutine.o qemu-coroutine-lock.o qemu-coroutine-io.o
2480 block-obj-y += qemu-coroutine-sleep.o