]> git.proxmox.com Git - pve-qemu-kvm.git/blame - debian/patches/0004-introduce-new-vma-archive-format.patch
projects / pve-qemu-kvm.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
update backup patches
[pve-qemu-kvm.git] / debian / patches / 0004-introduce-new-vma-archive-format.patch
CommitLineData
309874bd 1From 77a6564c9985c37990c3e90f558f2ef48d6d5f15 Mon Sep 17 00:00:00 2001
5ad5891c
DM		 2From: Dietmar Maurer <dietmar@proxmox.com>
3Date: Tue, 13 Nov 2012 11:11:38 +0100
309874bd 4Subject: [PATCH v3 4/7] introduce new vma archive format
5ad5891c
DM		 5
6This is a very simple archive format, see docs/specs/vma_spec.txt
7
8Signed-off-by: Dietmar Maurer <dietmar@proxmox.com>
9---
10 Makefile | 3 +-
11 Makefile.objs | 2 +-
12 blockdev.c | 6 +-
13 docs/specs/vma_spec.txt | 24 ++
309874bd
DM		 14 vma-reader.c | 801 ++++++++++++++++++++++++++++++++++++++++
15 vma-writer.c | 931 +++++++++++++++++++++++++++++++++++++++++++++++
16 vma.c | 561 ++++++++++++++++++++++++++++
4244016d 17 vma.h | 146 ++++++++
309874bd 18 8 files changed, 2470 insertions(+), 4 deletions(-)
5ad5891c
DM		 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
24
25diff --git a/Makefile b/Makefile
26index 9ecbcbb..30a9268 100644
27--- a/Makefile
28+++ b/Makefile
29@@ -100,7 +100,7 @@ defconfig:
30
31 -include config-all-devices.mak
32
33-all: $(DOCS) $(TOOLS) $(HELPERS-y) recurse-all
34+all: $(DOCS) $(TOOLS) vma$(EXESUF) $(HELPERS-y) recurse-all
35
36 config-host.h: config-host.h-timestamp
37 config-host.h-timestamp: config-host.mak
38@@ -194,6 +194,7 @@ tools-obj-$(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) $(block-obj-y) libqemustub.a
43
44 qemu-bridge-helper$(EXESUF): qemu-bridge-helper.o
45
46diff --git a/Makefile.objs b/Makefile.objs
47index cb46be5..b5732e2 100644
48--- a/Makefile.objs
49+++ b/Makefile.objs
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
59diff --git a/blockdev.c b/blockdev.c
309874bd 60index 68a8c55..80cb04d 100644
5ad5891c
DM		 61--- a/blockdev.c
62+++ b/blockdev.c
63@@ -21,6 +21,7 @@
64 #include "trace.h"
65 #include "arch_init.h"
66 #include "backup.h"
67+#include "vma.h"
68
69 static QTAILQ_HEAD(drivelist, DriveInfo) drives = QTAILQ_HEAD_INITIALIZER(drives);
70
309874bd 71@@ -1493,10 +1494,11 @@ char *qmp_backup(const char *backupfile, bool has_format, BackupFormat format,
5ad5891c 72 /* Todo: try to auto-detect format based on file name */
3055eeb4 73 format = has_format ? format : BACKUP_FORMAT_VMA;
5ad5891c
DM		 74
75- /* fixme: find driver for specifued format */
76 const BackupDriver *driver = NULL;
77
78- if (!driver) {
3055eeb4 79+ if (format == BACKUP_FORMAT_VMA) {
5ad5891c
DM		 80+ driver = &backup_vma_driver;
81+ } else {
3055eeb4 82 error_set(errp, ERROR_CLASS_GENERIC_ERROR, "unknown backup format");
5ad5891c 83 return NULL;
3055eeb4 84 }
5ad5891c
DM		 85diff --git a/docs/specs/vma_spec.txt b/docs/specs/vma_spec.txt
86new file mode 100644
87index 0000000..052c629
88--- /dev/null
89+++ b/docs/specs/vma_spec.txt
90@@ -0,0 +1,24 @@
91+=Virtual Machine Archive format (VMA)=
92+
93+This format contains a header which includes the VM configuration as
94+binary blobs, and a list of devices (dev_id, name).
95+
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.
99+
100+We use a cluster size of 65536, and use 8 bytes for each
101+cluster in the header to store the following information:
102+
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)
107+
108+We only store non-zero blocks (such block is 4096 bytes).
109+
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
112+data).
113+
114+
115diff --git a/vma-reader.c b/vma-reader.c
116new file mode 100644
309874bd 117index 0000000..2217a94
5ad5891c
DM		 118--- /dev/null
119+++ b/vma-reader.c
4244016d 120@@ -0,0 +1,801 @@
5ad5891c
DM		 121+/*
122+ * VMA: Virtual Machine Archive
123+ *
124+ * Copyright (C) 2012 Proxmox Server Solutions
125+ *
126+ * Authors:
127+ * Dietmar Maurer (dietmar@proxmox.com)
128+ *
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.
131+ *
132+ */
133+
134+#include <stdio.h>
135+#include <errno.h>
136+#include <unistd.h>
137+#include <stdio.h>
138+#include <string.h>
139+#include <sys/types.h>
140+#include <sys/stat.h>
141+#include <fcntl.h>
142+#include <glib.h>
143+#include <uuid/uuid.h>
144+
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"
150+#include "vma.h"
151+#include "block.h"
152+
153+#define BITS_PER_LONG (sizeof(unsigned long) * 8)
154+
155+static unsigned char zero_vma_block[VMA_BLOCK_SIZE];
156+
157+typedef struct VmaRestoreState {
158+ BlockDriverState *bs;
159+ bool write_zeroes;
160+ unsigned long *bitmap;
161+ int bitmap_size;
162+} VmaRestoreState;
163+
164+struct VmaReader {
165+ int fd;
166+ GChecksum *md5csum;
167+ GHashTable *blob_hash;
168+ unsigned char *head_data;
169+ VmaDeviceInfo devinfo[256];
170+ VmaRestoreState rstate[256];
171+ GList *cdata_list;
172+ guint8 vmstate_stream;
173+ uint32_t vmstate_clusters;
4244016d
DM		 174+ /* to show restore percentage if run with -v */
175+ time_t start_time;
176+ int64_t cluster_count;
177+ int64_t clusters_read;
178+ int clusters_read_per;
5ad5891c
DM		 179+};
180+
181+static guint
182+g_int32_hash(gconstpointer v)
183+{
184+ return *(const uint32_t *)v;
185+}
186+
187+static gboolean
188+g_int32_equal(gconstpointer v1, gconstpointer v2)
189+{
190+ return *((const uint32_t *)v1) == *((const uint32_t *)v2);
191+}
192+
193+static int vma_reader_get_bitmap(VmaRestoreState *rstate, int64_t cluster_num)
194+{
195+ assert(rstate);
196+ assert(rstate->bitmap);
197+
198+ unsigned long val, idx, bit;
199+
200+ idx = cluster_num / BITS_PER_LONG;
201+
202+ assert(rstate->bitmap_size > idx);
203+
204+ bit = cluster_num % BITS_PER_LONG;
205+ val = rstate->bitmap[idx];
206+
207+ return !!(val & (1UL << bit));
208+}
209+
210+static void vma_reader_set_bitmap(VmaRestoreState *rstate, int64_t cluster_num,
211+ int dirty)
212+{
213+ assert(rstate);
214+ assert(rstate->bitmap);
215+
216+ unsigned long val, idx, bit;
217+
218+ idx = cluster_num / BITS_PER_LONG;
219+
220+ assert(rstate->bitmap_size > idx);
221+
222+ bit = cluster_num % BITS_PER_LONG;
223+ val = rstate->bitmap[idx];
224+ if (dirty) {
225+ if (!(val & (1UL << bit))) {
226+ val |= 1UL << bit;
227+ }
228+ } else {
229+ if (val & (1UL << bit)) {
230+ val &= ~(1UL << bit);
231+ }
232+ }
233+ rstate->bitmap[idx] = val;
234+}
235+
236+typedef struct VmaBlob {
237+ uint32_t start;
238+ uint32_t len;
239+ void *data;
240+} VmaBlob;
241+
242+static const VmaBlob *get_header_blob(VmaReader *vmar, uint32_t pos)
243+{
244+ assert(vmar);
245+ assert(vmar->blob_hash);
246+
247+ return g_hash_table_lookup(vmar->blob_hash, &pos);
248+}
249+
250+static const char *get_header_str(VmaReader *vmar, uint32_t pos)
251+{
252+ const VmaBlob *blob = get_header_blob(vmar, pos);
253+ if (!blob) {
254+ return NULL;
255+ }
256+ const char *res = (char *)blob->data;
257+ if (res[blob->len-1] != '\0') {
258+ return NULL;
259+ }
260+ return res;
261+}
262+
263+static ssize_t
264+safe_read(int fd, unsigned char *buf, size_t count)
265+{
266+ ssize_t n;
267+
268+ do {
269+ n = read(fd, buf, count);
270+ } while (n < 0 && errno == EINTR);
271+
272+ return n;
273+}
274+
275+static ssize_t
276+full_read(int fd, unsigned char *buf, size_t len)
277+{
278+ ssize_t n;
279+ size_t total;
280+
281+ total = 0;
282+
283+ while (len > 0) {
284+ n = safe_read(fd, buf, len);
285+
286+ if (n == 0) {
287+ return total;
288+ }
289+
290+ if (n <= 0) {
291+ break;
292+ }
293+
294+ buf += n;
295+ total += n;
296+ len -= n;
297+ }
298+
299+ if (len) {
300+ return -1;
301+ }
302+
303+ return total;
304+}
305+
306+void vma_reader_destroy(VmaReader *vmar)
307+{
308+ assert(vmar);
309+
310+ if (vmar->fd >= 0) {
311+ close(vmar->fd);
312+ }
313+
314+ if (vmar->cdata_list) {
315+ g_list_free(vmar->cdata_list);
316+ }
317+
318+ int i;
319+ for (i = 1; i < 256; i++) {
320+ if (vmar->rstate[i].bitmap) {
321+ g_free(vmar->rstate[i].bitmap);
322+ }
323+ }
324+
325+ if (vmar->md5csum) {
326+ g_checksum_free(vmar->md5csum);
327+ }
328+
329+ if (vmar->blob_hash) {
330+ g_hash_table_destroy(vmar->blob_hash);
331+ }
332+
333+ if (vmar->head_data) {
334+ g_free(vmar->head_data);
335+ }
336+
337+ g_free(vmar);
338+
339+};
340+
341+static int vma_reader_read_head(VmaReader *vmar, Error **errp)
342+{
343+ assert(vmar);
344+ assert(errp);
345+ assert(*errp == NULL);
346+
347+ unsigned char md5sum[16];
348+ int i;
349+ int ret = 0;
350+
351+ vmar->head_data = g_malloc(sizeof(VmaHeader));
352+
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");
357+ return -1;
358+ }
359+
360+ VmaHeader *h = (VmaHeader *)vmar->head_data;
361+
362+ if (h->magic != VMA_MAGIC) {
363+ error_setg(errp, "not a vma file - wrong magic number");
364+ return -1;
365+ }
366+
367+ uint32_t header_size = GUINT32_FROM_BE(h->header_size);
368+ int need = header_size - sizeof(VmaHeader);
369+ if (need <= 0) {
370+ error_setg(errp, "wrong vma header size %d", header_size);
371+ return -1;
372+ }
373+
374+ vmar->head_data = g_realloc(vmar->head_data, header_size);
375+ h = (VmaHeader *)vmar->head_data;
376+
377+ if (full_read(vmar->fd, vmar->head_data + sizeof(VmaHeader), need) !=
378+ need) {
379+ error_setg(errp, "can't read vma header data - %s",
380+ errno ? strerror(errno) : "got EOF");
381+ return -1;
382+ }
383+
384+ memcpy(md5sum, h->md5sum, 16);
385+ memset(h->md5sum, 0, 16);
386+
387+ g_checksum_reset(vmar->md5csum);
388+ g_checksum_update(vmar->md5csum, vmar->head_data, header_size);
389+ gsize csize = 16;
390+ g_checksum_get_digest(vmar->md5csum, (guint8 *)(h->md5sum), &csize);
391+
392+ if (memcmp(md5sum, h->md5sum, 16) != 0) {
393+ error_setg(errp, "wrong vma header chechsum");
394+ return -1;
395+ }
396+
397+ /* we can modify header data after checksum verify */
398+ h->header_size = header_size;
399+
400+ h->version = GUINT32_FROM_BE(h->version);
401+ if (h->version != 1) {
402+ error_setg(errp, "wrong vma version %d", h->version);
403+ return -1;
404+ }
405+
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);
409+
410+ uint32_t bstart = h->blob_buffer_offset + 1;
411+ uint32_t bend = h->blob_buffer_offset + h->blob_buffer_size;
412+
413+ if (bstart <= sizeof(VmaHeader)) {
414+ error_setg(errp, "wrong vma blob buffer offset %d",
415+ h->blob_buffer_offset);
416+ return -1;
417+ }
418+
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);
422+ return -1;
423+ }
424+
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;
431+ blob->len = size;
432+ blob->data = vmar->head_data + bstart + 2;
433+ g_hash_table_insert(vmar->blob_hash, &blob->start, blob);
434+ }
435+ bstart += size + 2;
436+ }
437+
438+
439+ int count = 0;
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);
445+
446+ if (size && devname) {
447+ count++;
448+ vmar->devinfo[i].size = size;
449+ vmar->devinfo[i].devname = devname;
450+
451+ if (strcmp(devname, "vmstate") == 0) {
452+ vmar->vmstate_stream = i;
453+ }
454+ }
455+ }
456+
457+ if (!count) {
458+ error_setg(errp, "vma does not contain data");
459+ return -1;
460+ }
461+
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]);
465+
466+ if (!(name_ptr && data_ptr)) {
467+ continue;
468+ }
469+ const char *name = get_header_str(vmar, name_ptr);
470+ const VmaBlob *blob = get_header_blob(vmar, data_ptr);
471+
472+ if (!(name && blob)) {
473+ error_setg(errp, "vma contains invalid data pointers");
474+ return -1;
475+ }
476+
477+ VmaConfigData *cdata = g_new0(VmaConfigData, 1);
478+ cdata->name = name;
479+ cdata->data = blob->data;
480+ cdata->len = blob->len;
481+
482+ vmar->cdata_list = g_list_append(vmar->cdata_list, cdata);
483+ }
484+
485+ return ret;
486+};
487+
488+VmaReader *vma_reader_create(const char *filename, Error **errp)
489+{
490+ assert(filename);
491+ assert(errp);
492+
493+ VmaReader *vmar = g_new0(VmaReader, 1);
494+
3f350e0d
DM		 495+ if (strcmp(filename, "-") == 0) {
496+ vmar->fd = dup(0);
497+ } else {
498+ vmar->fd = open(filename, O_RDONLY);
499+ }
5ad5891c
DM		 500+
501+ if (vmar->fd < 0) {
502+ error_setg(errp, "can't open file %s - %s\n", filename,
503+ strerror(errno));
504+ goto err;
505+ }
506+
507+ vmar->md5csum = g_checksum_new(G_CHECKSUM_MD5);
508+ if (!vmar->md5csum) {
509+ error_setg(errp, "can't allocate cmsum\n");
510+ goto err;
511+ }
512+
513+ vmar->blob_hash = g_hash_table_new_full(g_int32_hash, g_int32_equal,
514+ NULL, g_free);
515+
516+ if (vma_reader_read_head(vmar, errp) < 0) {
517+ goto err;
518+ }
519+
520+ return vmar;
521+
522+err:
523+ if (vmar) {
524+ vma_reader_destroy(vmar);
525+ }
526+
527+ return NULL;
528+}
529+
530+VmaHeader *vma_reader_get_header(VmaReader *vmar)
531+{
532+ assert(vmar);
533+ assert(vmar->head_data);
534+
535+ return (VmaHeader *)(vmar->head_data);
536+}
537+
538+GList *vma_reader_get_config_data(VmaReader *vmar)
539+{
540+ assert(vmar);
541+ assert(vmar->head_data);
542+
543+ return vmar->cdata_list;
544+}
545+
546+VmaDeviceInfo *vma_reader_get_device_info(VmaReader *vmar, guint8 dev_id)
547+{
548+ assert(vmar);
549+ assert(dev_id);
550+
551+ if (vmar->devinfo[dev_id].size && vmar->devinfo[dev_id].devname) {
552+ return &vmar->devinfo[dev_id];
553+ }
554+
555+ return NULL;
556+}
557+
558+int vma_reader_register_bs(VmaReader *vmar, guint8 dev_id, BlockDriverState *bs,
559+ bool write_zeroes, Error **errp)
560+{
561+ assert(vmar);
562+ assert(bs != NULL);
563+ assert(dev_id);
564+ assert(vmar->rstate[dev_id].bs == NULL);
565+
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);
571+ return -1;
572+ }
573+
574+ vmar->rstate[dev_id].bs = bs;
575+ vmar->rstate[dev_id].write_zeroes = write_zeroes;
576+
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;
580+
581+ vmar->rstate[dev_id].bitmap_size = bitmap_size;
582+ vmar->rstate[dev_id].bitmap = g_new0(unsigned long, bitmap_size);
583+
4244016d
DM		 584+ vmar->cluster_count += size/VMA_CLUSTER_SIZE;
585+
5ad5891c
DM		 586+ return 0;
587+}
588+
589+static ssize_t safe_write(int fd, void *buf, size_t count)
590+{
591+ ssize_t n;
592+
593+ do {
594+ n = write(fd, buf, count);
595+ } while (n < 0 && errno == EINTR);
596+
597+ return n;
598+}
599+
600+static size_t full_write(int fd, void *buf, size_t len)
601+{
602+ ssize_t n;
603+ size_t total;
604+
605+ total = 0;
606+
607+ while (len > 0) {
608+ n = safe_write(fd, buf, len);
609+ if (n < 0) {
610+ return n;
611+ }
612+ buf += n;
613+ total += n;
614+ len -= n;
615+ }
616+
617+ if (len) {
618+ /* incomplete write ? */
619+ return -1;
620+ }
621+
622+ return total;
623+}
624+
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)
629+{
630+ assert(vmar);
631+
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);
636+ if (res < 0) {
637+ error_setg(errp, "write vmstate failed %d", res);
638+ return -1;
639+ }
640+ }
641+ } else {
642+ int res = bdrv_write(bs, sector_num, buf, nb_sectors);
643+ if (res < 0) {
644+ error_setg(errp, "bdrv_write to %s failed (%d)",
645+ bdrv_get_device_name(bs), res);
646+ return -1;
647+ }
648+ }
649+ return 0;
650+}
651+static int restore_extent(VmaReader *vmar, unsigned char *buf,
309874bd 652+ int extent_size, int vmstate_fd,
4244016d 653+ bool verbose, Error **errp)
5ad5891c
DM		 654+{
655+ assert(vmar);
656+ assert(buf);
657+
658+ VmaExtentHeader *ehead = (VmaExtentHeader *)buf;
659+ int start = VMA_EXTENT_HEADER_SIZE;
660+ int i;
661+
662+ for (i = 0; i < VMA_BLOCKS_PER_EXTENT; i++) {
663+ uint64_t block_info = GUINT64_FROM_BE(ehead->blockinfo[i]);
4244016d
DM		 664+ uint64_t cluster_num = block_info & 0xffffffff;
665+ uint8_t dev_id = (block_info >> 32) & 0xff;
5ad5891c
DM		 666+ uint16_t mask = block_info >> (32+16);
667+ int64_t max_sector;
668+
669+ if (!dev_id) {
670+ continue;
671+ }
672+
673+ VmaRestoreState *rstate = &vmar->rstate[dev_id];
674+ BlockDriverState *bs = NULL;
675+
676+ if (dev_id != vmar->vmstate_stream) {
677+ bs = rstate->bs;
678+ if (!bs) {
679+ error_setg(errp, "got wrong dev id %d", dev_id);
680+ return -1;
681+ }
682+
683+ if (vma_reader_get_bitmap(rstate, cluster_num)) {
4244016d 684+ error_setg(errp, "found duplicated cluster %zd for stream %s",
5ad5891c
DM		 685+ cluster_num, vmar->devinfo[dev_id].devname);
686+ return -1;
687+ }
688+ vma_reader_set_bitmap(rstate, cluster_num, 1);
689+
690+ max_sector = vmar->devinfo[dev_id].size/BDRV_SECTOR_SIZE;
691+ } else {
692+ max_sector = G_MAXINT64;
693+ if (cluster_num != vmar->vmstate_clusters) {
694+ error_setg(errp, "found out of order vmstate data");
695+ return -1;
696+ }
697+ vmar->vmstate_clusters++;
698+ }
699+
4244016d
DM		 700+ vmar->clusters_read++;
701+
702+ if (verbose) {
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) {
309874bd 706+ printf("progress %d%% (read %zd bytes, duration %zd sec)\n",
4244016d
DM		 707+ percent, vmar->clusters_read*VMA_CLUSTER_SIZE,
708+ duration);
709+ fflush(stdout);
710+ vmar->clusters_read_per = percent;
711+ }
712+ }
713+
5ad5891c
DM		 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");
718+ return -1;
719+ }
720+ int64_t sector_num = (cluster_num * VMA_CLUSTER_SIZE) /
721+ BDRV_SECTOR_SIZE;
722+ int64_t end_sector = sector_num +
723+ VMA_CLUSTER_SIZE/BDRV_SECTOR_SIZE;
724+
725+ if (end_sector > max_sector) {
726+ end_sector = max_sector;
727+ }
728+
729+ if (end_sector <= sector_num) {
730+ error_setg(errp, "got wrong block address - write bejond end");
731+ return -1;
732+ }
733+
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) {
737+ return -1;
738+ }
739+
740+ start += VMA_CLUSTER_SIZE;
741+ } else {
742+ int j;
743+ int bit = 1;
744+
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;
748+
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;
753+ }
754+
755+ if (mask & bit) {
756+ if ((start + VMA_BLOCK_SIZE) > extent_size) {
757+ error_setg(errp, "short vma extent - too many blocks");
758+ return -1;
759+ }
760+
761+ if (end_sector <= sector_num) {
762+ error_setg(errp, "got wrong block address - "
763+ "write bejond end");
764+ return -1;
765+ }
766+
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) {
771+ return -1;
772+ }
773+
774+ start += VMA_BLOCK_SIZE;
775+
776+ } else {
777+
4244016d 778+ if (rstate->write_zeroes && (end_sector > sector_num)) {
5ad5891c
DM		 779+ /* Todo: use bdrv_co_write_zeroes (but that need to
780+ * be run inside coroutine?)
781+ */
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) {
786+ return -1;
787+ }
788+ }
789+ }
790+
791+ bit = bit << 1;
792+ }
793+ }
794+ }
795+
796+ if (start != extent_size) {
797+ error_setg(errp, "vma extent error - missing blocks");
798+ return -1;
799+ }
800+
801+ return 0;
802+}
803+
4244016d 804+int vma_reader_restore(VmaReader *vmar, int vmstate_fd, bool verbose,
309874bd 805+ Error **errp)
5ad5891c
DM		 806+{
807+ assert(vmar);
808+ assert(vmar->head_data);
809+
810+ int ret = 0;
811+ unsigned char buf[VMA_MAX_EXTENT_SIZE];
812+ int buf_pos = 0;
813+ unsigned char md5sum[16];
814+ VmaHeader *h = (VmaHeader *)vmar->head_data;
815+
4244016d 816+ vmar->start_time = time(NULL);
5ad5891c
DM		 817+
818+ while (1) {
819+ int bytes = full_read(vmar->fd, buf + buf_pos, sizeof(buf) - buf_pos);
820+ if (bytes < 0) {
821+ error_setg(errp, "read failed - %s", strerror(errno));
822+ return -1;
823+ }
824+
825+ buf_pos += bytes;
826+
827+ if (!buf_pos) {
828+ break; /* EOF */
829+ }
830+
831+ if (buf_pos < VMA_EXTENT_HEADER_SIZE) {
832+ error_setg(errp, "read short extent (%d bytes)", buf_pos);
833+ return -1;
834+ }
835+
836+ VmaExtentHeader *ehead = (VmaExtentHeader *)buf;
837+
838+ /* extract md5sum */
839+ memcpy(md5sum, ehead->md5sum, sizeof(ehead->md5sum));
840+ memset(ehead->md5sum, 0, sizeof(ehead->md5sum));
841+
842+ g_checksum_reset(vmar->md5csum);
843+ g_checksum_update(vmar->md5csum, buf, VMA_EXTENT_HEADER_SIZE);
844+ gsize csize = 16;
845+ g_checksum_get_digest(vmar->md5csum, ehead->md5sum, &csize);
846+
847+ if (memcmp(md5sum, ehead->md5sum, 16) != 0) {
848+ error_setg(errp, "wrong vma extent header chechsum");
849+ return -1;
850+ }
851+
852+ if (memcmp(h->uuid, ehead->uuid, sizeof(ehead->uuid)) != 0) {
853+ error_setg(errp, "wrong vma extent uuid");
854+ return -1;
855+ }
856+
857+ if (ehead->magic != VMA_EXTENT_MAGIC || ehead->reserved1 != 0) {
858+ error_setg(errp, "wrong vma extent header magic");
859+ return -1;
860+ }
861+
862+ int block_count = GUINT16_FROM_BE(ehead->block_count);
863+ int extent_size = VMA_EXTENT_HEADER_SIZE + block_count*VMA_BLOCK_SIZE;
864+
865+ if (buf_pos < extent_size) {
866+ error_setg(errp, "short vma extent (%d < %d)", buf_pos,
867+ extent_size);
868+ return -1;
869+ }
870+
309874bd 871+ if (restore_extent(vmar, buf, extent_size, vmstate_fd, verbose,
4244016d 872+ errp) < 0) {
5ad5891c
DM		 873+ return -1;
874+ }
875+
876+ if (buf_pos > extent_size) {
877+ memmove(buf, buf + extent_size, buf_pos - extent_size);
878+ buf_pos = buf_pos - extent_size;
879+ } else {
880+ buf_pos = 0;
881+ }
882+ }
883+
884+ bdrv_drain_all();
885+
886+ int i;
887+ for (i = 1; i < 256; i++) {
888+ VmaRestoreState *rstate = &vmar->rstate[i];
889+ if (!rstate->bs) {
890+ continue;
891+ }
892+
893+ if (bdrv_flush(rstate->bs) < 0) {
894+ error_setg(errp, "vma bdrv_flush %s failed",
895+ vmar->devinfo[i].devname);
896+ return -1;
897+ }
898+
899+ if (vmar->devinfo[i].size &&
900+ (strcmp(vmar->devinfo[i].devname, "vmstate") != 0)) {
901+ assert(rstate->bitmap);
902+
903+ int64_t cluster_num, end;
904+
905+ end = (vmar->devinfo[i].size + VMA_CLUSTER_SIZE - 1) /
906+ VMA_CLUSTER_SIZE;
907+
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);
913+ return -1;
914+ }
915+ }
916+ }
917+ }
918+
919+ return ret;
920+}
921+
922diff --git a/vma-writer.c b/vma-writer.c
923new file mode 100644
309874bd 924index 0000000..688af4b
5ad5891c
DM		 925--- /dev/null
926+++ b/vma-writer.c
309874bd 927@@ -0,0 +1,931 @@
5ad5891c
DM		 928+/*
929+ * VMA: Virtual Machine Archive
930+ *
931+ * Copyright (C) 2012 Proxmox Server Solutions
932+ *
933+ * Authors:
934+ * Dietmar Maurer (dietmar@proxmox.com)
935+ *
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.
938+ *
939+ */
940+
941+#include <stdio.h>
942+#include <errno.h>
943+#include <unistd.h>
944+#include <stdio.h>
945+#include <string.h>
946+#include <sys/types.h>
947+#include <sys/stat.h>
948+#include <fcntl.h>
949+#include <glib.h>
950+#include <uuid/uuid.h>
951+
952+#include "qemu-common.h"
953+#include "qemu_socket.h"
954+#include "qemu-coroutine.h"
955+#include "qemu-aio.h"
956+#include "qemu/ratelimit.h"
957+#include "vma.h"
958+#include "block.h"
3f350e0d 959+#include "monitor.h"
5ad5891c
DM		 960+
961+#define DEBUG_VMA 0
962+
963+#define DPRINTF(fmt, ...)\
964+ do { if (DEBUG_VMA) { printf("vma: " fmt, ## __VA_ARGS__); } } while (0)
965+
966+#define WRITE_BUFFERS 5
967+
968+typedef struct VmaAIOCB VmaAIOCB;
969+struct VmaAIOCB {
970+ VmaWriter *vmaw;
971+ unsigned char buffer[VMA_MAX_EXTENT_SIZE];
972+ size_t bytes;
973+ Coroutine *co;
974+};
975+
976+struct VmaWriter {
977+ int fd;
978+ FILE *cmd;
979+ int status;
980+ char errmsg[8192];
981+ uuid_t uuid;
982+ bool header_written;
983+ bool closed;
984+
985+ /* we always write extents */
986+ unsigned char outbuf[VMA_MAX_EXTENT_SIZE];
987+ int outbuf_pos; /* in bytes */
988+ int outbuf_count; /* in VMA_BLOCKS */
989+ uint64_t outbuf_block_info[VMA_BLOCKS_PER_EXTENT];
990+
991+ VmaAIOCB aiocbs[WRITE_BUFFERS];
992+ CoQueue wqueue;
993+
994+ GChecksum *md5csum;
995+ CoMutex writer_lock;
996+ CoMutex flush_lock;
997+ Coroutine *co_writer;
998+ RateLimit limit;
5ad5891c
DM		 999+
1000+ /* drive informations */
1001+ VmaStreamInfo stream_info[256];
1002+ guint stream_count;
1003+
1004+ guint8 vmstate_stream;
1005+ uint32_t vmstate_clusters;
1006+
1007+ /* header blob table */
1008+ char *header_blob_table;
1009+ uint32_t header_blob_table_size;
1010+ uint32_t header_blob_table_pos;
1011+
1012+ /* store for config blobs */
1013+ uint32_t config_names[VMA_MAX_CONFIGS]; /* offset into blob_buffer table */
1014+ uint32_t config_data[VMA_MAX_CONFIGS]; /* offset into blob_buffer table */
1015+ uint32_t config_count;
1016+};
1017+
1018+void vma_writer_set_error(VmaWriter *vmaw, const char *fmt, ...)
1019+{
1020+ va_list ap;
1021+
1022+ if (vmaw->status < 0) {
1023+ return;
1024+ }
1025+
1026+ vmaw->status = -1;
1027+
1028+ va_start(ap, fmt);
1029+ g_vsnprintf(vmaw->errmsg, sizeof(vmaw->errmsg), fmt, ap);
1030+ va_end(ap);
1031+
1032+ DPRINTF("vma_writer_set_error: %s\n", vmaw->errmsg);
1033+}
1034+
1035+static uint32_t allocate_header_blob(VmaWriter *vmaw, const char *data,
1036+ size_t len)
1037+{
1038+ if (len > 65535) {
1039+ return 0;
1040+ }
1041+
1042+ if (!vmaw->header_blob_table ||
1043+ (vmaw->header_blob_table_size <
1044+ (vmaw->header_blob_table_pos + len + 2))) {
1045+ int newsize = vmaw->header_blob_table_size + ((len + 2 + 511)/512)*512;
1046+
1047+ vmaw->header_blob_table = g_realloc(vmaw->header_blob_table, newsize);
1048+ memset(vmaw->header_blob_table + vmaw->header_blob_table_size,
1049+ 0, newsize - vmaw->header_blob_table_size);
1050+ vmaw->header_blob_table_size = newsize;
1051+ }
1052+
1053+ uint32_t cpos = vmaw->header_blob_table_pos;
1054+ vmaw->header_blob_table[cpos] = len & 255;
1055+ vmaw->header_blob_table[cpos+1] = (len >> 8) & 255;
1056+ memcpy(vmaw->header_blob_table + cpos + 2, data, len);
1057+ vmaw->header_blob_table_pos += len + 2;
1058+ return cpos;
1059+}
1060+
1061+static uint32_t allocate_header_string(VmaWriter *vmaw, const char *str)
1062+{
1063+ assert(vmaw);
1064+
1065+ size_t len = strlen(str) + 1;
1066+
1067+ return allocate_header_blob(vmaw, str, len);
1068+}
1069+
1070+int vma_writer_add_config(VmaWriter *vmaw, const char *name, gpointer data,
1071+ gsize len)
1072+{
1073+ assert(vmaw);
1074+ assert(!vmaw->header_written);
1075+ assert(vmaw->config_count < VMA_MAX_CONFIGS);
1076+ assert(name);
1077+ assert(data);
1078+ assert(len);
1079+
1080+ uint32_t name_ptr = allocate_header_string(vmaw, name);
1081+ if (!name_ptr) {
1082+ return -1;
1083+ }
1084+
1085+ uint32_t data_ptr = allocate_header_blob(vmaw, data, len);
1086+ if (!data_ptr) {
1087+ return -1;
1088+ }
1089+
1090+ vmaw->config_names[vmaw->config_count] = name_ptr;
1091+ vmaw->config_data[vmaw->config_count] = data_ptr;
1092+
1093+ vmaw->config_count++;
1094+
1095+ return 0;
1096+}
1097+
1098+int vma_writer_register_stream(VmaWriter *vmaw, const char *devname,
1099+ size_t size)
1100+{
1101+ assert(vmaw);
1102+ assert(devname);
1103+ assert(!vmaw->status);
1104+
1105+ if (vmaw->header_written) {
1106+ vma_writer_set_error(vmaw, "vma_writer_register_stream: header "
1107+ "already written");
1108+ return -1;
1109+ }
1110+
1111+ guint n = vmaw->stream_count + 1;
1112+
1113+ /* we can have dev_ids form 1 to 255 (0 reserved)
1114+ * 255(-1) reseverd for safety
1115+ */
1116+ if (n > 254) {
1117+ vma_writer_set_error(vmaw, "vma_writer_register_stream: "
1118+ "too many drives");
1119+ return -1;
1120+ }
1121+
1122+ if (size <= 0) {
1123+ vma_writer_set_error(vmaw, "vma_writer_register_stream: "
1124+ "got strange size %zd", size);
1125+ return -1;
1126+ }
1127+
1128+ DPRINTF("vma_writer_register_stream %s %zu %d\n", devname, size, n);
1129+
1130+ vmaw->stream_info[n].devname = g_strdup(devname);
1131+ vmaw->stream_info[n].size = size;
1132+
1133+ vmaw->stream_info[n].cluster_count = (size + VMA_CLUSTER_SIZE - 1) /
1134+ VMA_CLUSTER_SIZE;
1135+
1136+ vmaw->stream_count = n;
1137+
1138+ if (strcmp(devname, "vmstate") == 0) {
1139+ vmaw->vmstate_stream = n;
1140+ }
1141+
1142+ return n;
1143+}
1144+
1145+static void vma_co_continue_write(void *opaque)
1146+{
1147+ VmaWriter *vmaw = opaque;
1148+
1149+ qemu_aio_set_fd_handler(vmaw->fd, NULL, NULL, NULL, NULL);
1150+
1151+ DPRINTF("vma_co_continue_write\n");
1152+ qemu_coroutine_enter(vmaw->co_writer, NULL);
1153+}
1154+
1155+static ssize_t coroutine_fn
1156+vma_co_write(VmaWriter *vmaw, const void *buf, size_t bytes)
1157+{
1158+ size_t done = 0;
1159+ ssize_t ret;
1160+
1161+ /* atomic writes (we cannot interleave writes) */
1162+ qemu_co_mutex_lock(&vmaw->writer_lock);
1163+
1164+ DPRINTF("vma_co_write enter %zd\n", bytes);
1165+
1166+ while (done < bytes) {
1167+ ret = write(vmaw->fd, buf + done, bytes - done);
1168+ if (ret > 0) {
1169+ done += ret;
1170+ DPRINTF("vma_co_write written %zd %zd\n", done, ret);
1171+ } else if (ret < 0) {
1172+ if (errno == EAGAIN || errno == EWOULDBLOCK) {
1173+ DPRINTF("vma_co_write yield %zd\n", done);
1174+
1175+ vmaw->co_writer = qemu_coroutine_self();
1176+ qemu_aio_set_fd_handler(vmaw->fd, NULL, vma_co_continue_write,
1177+ NULL, vmaw);
1178+
1179+ qemu_coroutine_yield();
1180+ DPRINTF("vma_co_write restart %zd\n", done);
1181+ } else {
1182+ vma_writer_set_error(vmaw, "vma_co_write write error - %s",
1183+ strerror(errno));
1184+ done = -1; /* always return failure for partial writes */
1185+ break;
1186+ }
1187+ } else if (ret == 0) {
1188+ /* should not happen - simply try again */
1189+ }
1190+ }
1191+
1192+ qemu_co_mutex_unlock(&vmaw->writer_lock);
1193+
1194+ DPRINTF("vma_co_write leave %zd\n", done);
1195+ return done;
1196+}
1197+
1198+static void coroutine_fn vma_co_writer_task(void *opaque)
1199+{
1200+ VmaAIOCB *cb = opaque;
1201+
1202+ DPRINTF("vma_co_writer_task start\n");
1203+
1204+ int64_t done = vma_co_write(cb->vmaw, cb->buffer, cb->bytes);
1205+ DPRINTF("vma_co_writer_task write done %zd\n", done);
1206+
1207+ if (done != cb->bytes) {
1208+ DPRINTF("vma_co_writer_task failed write %zd %zd", cb->bytes, done);
1209+ vma_writer_set_error(cb->vmaw, "vma_co_writer_task failed write %zd",
1210+ done);
1211+ }
1212+
1213+ cb->bytes = 0;
1214+
1215+ qemu_co_queue_next(&cb->vmaw->wqueue);
1216+
1217+ DPRINTF("vma_co_writer_task end\n");
1218+}
1219+
1220+static void coroutine_fn vma_queue_flush(VmaWriter *vmaw)
1221+{
1222+ DPRINTF("vma_queue_flush enter\n");
1223+
1224+ assert(vmaw);
1225+
1226+ while (1) {
1227+ int i;
1228+ VmaAIOCB *cb = NULL;
1229+ for (i = 0; i < WRITE_BUFFERS; i++) {
1230+ if (vmaw->aiocbs[i].bytes) {
1231+ cb = &vmaw->aiocbs[i];
1232+ DPRINTF("FOUND USED AIO BUFFER %d %zd\n", i,
1233+ vmaw->aiocbs[i].bytes);
1234+ break;
1235+ }
1236+ }
1237+ if (!cb) {
1238+ break;
1239+ }
1240+ qemu_co_queue_wait(&vmaw->wqueue);
1241+ }
1242+
1243+ DPRINTF("vma_queue_flush leave\n");
1244+}
1245+
1246+/**
1247+ * NOTE: pipe buffer size in only 4096 bytes on linux (see 'ulimit -a')
1248+ * So we need to create a coroutione to allow 'parallel' execution.
1249+ */
1250+static ssize_t coroutine_fn
1251+vma_queue_write(VmaWriter *vmaw, const void *buf, size_t bytes)
1252+{
1253+ DPRINTF("vma_queue_write enter %zd\n", bytes);
1254+
1255+ assert(vmaw);
1256+ assert(buf);
1257+ assert(bytes <= VMA_MAX_EXTENT_SIZE);
1258+
1259+ VmaAIOCB *cb = NULL;
1260+ while (!cb) {
1261+ int i;
1262+ for (i = 0; i < WRITE_BUFFERS; i++) {
1263+ if (!vmaw->aiocbs[i].bytes) {
1264+ cb = &vmaw->aiocbs[i];
1265+ break;
1266+ }
1267+ }
1268+ if (!cb) {
1269+ qemu_co_queue_wait(&vmaw->wqueue);
1270+ }
1271+ }
1272+
1273+ memcpy(cb->buffer, buf, bytes);
1274+ cb->bytes = bytes;
1275+ cb->vmaw = vmaw;
1276+
1277+ DPRINTF("vma_queue_write start %zd\n", bytes);
1278+ cb->co = qemu_coroutine_create(vma_co_writer_task);
1279+ qemu_coroutine_enter(cb->co, cb);
1280+
1281+ DPRINTF("vma_queue_write leave\n");
1282+
1283+ return bytes;
1284+}
1285+
1286+VmaWriter *vma_writer_create(const char *filename, uuid_t uuid, int64_t speed,
1287+ Error **errp)
1288+{
1289+ const char *p;
1290+
1291+ assert(sizeof(VmaHeader) == (4096 + 8192));
1292+ assert(sizeof(VmaExtentHeader) == 512);
1293+
1294+ VmaWriter *vmaw = g_new0(VmaWriter, 1);
1295+ vmaw->fd = -1;
1296+
1297+ vmaw->md5csum = g_checksum_new(G_CHECKSUM_MD5);
1298+ if (!vmaw->md5csum) {
1299+ error_setg(errp, "can't allocate cmsum\n");
1300+ goto err;
1301+ }
1302+
1303+ if (strstart(filename, "exec:", &p)) {
1304+ vmaw->cmd = popen(p, "w");
1305+ if (vmaw->cmd == NULL) {
1306+ error_setg(errp, "can't popen command '%s' - %s\n", p,
1307+ strerror(errno));
1308+ goto err;
1309+ }
1310+ vmaw->fd = fileno(vmaw->cmd);
1311+ socket_set_nonblock(vmaw->fd);
1312+
1313+ } else {
efa8e5de
DM		 1314+ struct stat st;
1315+ int oflags;
309874bd 1316+ const char *tmp_id_str;
3f350e0d 1317+
efa8e5de
DM		 1318+ if ((stat(filename, &st) == 0) && S_ISFIFO(st.st_mode)) {
1319+ oflags = O_NONBLOCK|O_WRONLY;
309874bd 1320+ vmaw->fd = qemu_open(filename, oflags, 0644);
3f350e0d
DM		 1321+ } else if (strstart(filename, "/dev/fdset/", &tmp_id_str)) {
1322+ oflags = O_NONBLOCK|O_WRONLY;
1323+ vmaw->fd = qemu_open(filename, oflags, 0644);
309874bd 1324+ } else if (strstart(filename, "/dev/fdname/", &tmp_id_str)) {
3f350e0d
DM		 1325+ vmaw->fd = monitor_get_fd(cur_mon, tmp_id_str, errp);
1326+ if (vmaw->fd < 0) {
1327+ goto err;
1328+ }
309874bd 1329+ } else {
efa8e5de 1330+ oflags = O_NONBLOCK|O_WRONLY|O_CREAT|O_EXCL;
309874bd 1331+ vmaw->fd = qemu_open(filename, oflags, 0644);
efa8e5de 1332+ }
3f350e0d 1333+
5ad5891c
DM		 1334+ if (vmaw->fd < 0) {
1335+ error_setg(errp, "can't open file %s - %s\n", filename,
1336+ strerror(errno));
1337+ goto err;
1338+ }
1339+ }
1340+
1341+ vmaw->outbuf_count = 0;
1342+ vmaw->outbuf_pos = VMA_EXTENT_HEADER_SIZE;
1343+
1344+ vmaw->header_blob_table_pos = 1; /* start at pos 1 */
1345+
1346+ qemu_co_mutex_init(&vmaw->writer_lock);
1347+ qemu_co_mutex_init(&vmaw->flush_lock);
1348+ qemu_co_queue_init(&vmaw->wqueue);
1349+
1350+ uuid_copy(vmaw->uuid, uuid);
1351+
1352+ if (speed <= 0) {
1353+ speed = 10*1024*1024*1024LLU; /* default 10GB/s */
1354+ }
1355+
1356+ ratelimit_set_speed(&vmaw->limit, speed, 100000000ULL /* 0.1 sec */);
1357+
1358+ return vmaw;
1359+
1360+err:
1361+ if (vmaw) {
1362+ if (vmaw->cmd) {
1363+ pclose(vmaw->cmd);
1364+ } else if (vmaw->fd >= 0) {
1365+ close(vmaw->fd);
1366+ }
1367+
1368+ if (vmaw->md5csum) {
1369+ g_checksum_free(vmaw->md5csum);
1370+ }
1371+
1372+ g_free(vmaw);
1373+ }
1374+
1375+ return NULL;
1376+}
1377+
1378+static int coroutine_fn vma_write_header(VmaWriter *vmaw)
1379+{
1380+ assert(vmaw);
1381+ int header_clusters = 8;
1382+ char buf[65536*header_clusters];
1383+ VmaHeader *head = (VmaHeader *)buf;
1384+
1385+ int i;
1386+
1387+ DPRINTF("VMA WRITE HEADER\n");
1388+
1389+ if (vmaw->status < 0) {
1390+ return vmaw->status;
1391+ }
1392+
1393+ memset(buf, 0, sizeof(buf));
1394+
1395+ head->magic = VMA_MAGIC;
1396+ head->version = GUINT32_TO_BE(1); /* v1 */
1397+ memcpy(head->uuid, vmaw->uuid, 16);
1398+
1399+ time_t ctime = time(NULL);
1400+ head->ctime = GUINT64_TO_BE(ctime);
1401+
1402+ if (!vmaw->stream_count) {
1403+ return -1;
1404+ }
1405+
1406+ for (i = 0; i < VMA_MAX_CONFIGS; i++) {
1407+ head->config_names[i] = GUINT32_TO_BE(vmaw->config_names[i]);
1408+ head->config_data[i] = GUINT32_TO_BE(vmaw->config_data[i]);
1409+ }
1410+
1411+ /* 32 bytes per device (12 used currently) = 8192 bytes max */
1412+ for (i = 1; i <= 254; i++) {
1413+ VmaStreamInfo *si = &vmaw->stream_info[i];
1414+ if (si->size) {
1415+ assert(si->devname);
1416+ uint32_t devname_ptr = allocate_header_string(vmaw, si->devname);
1417+ if (!devname_ptr) {
1418+ return -1;
1419+ }
1420+ head->dev_info[i].devname_ptr = GUINT32_TO_BE(devname_ptr);
1421+ head->dev_info[i].size = GUINT64_TO_BE(si->size);
1422+ }
1423+ }
1424+
1425+ uint32_t header_size = sizeof(VmaHeader) + vmaw->header_blob_table_size;
1426+ head->header_size = GUINT32_TO_BE(header_size);
1427+
1428+ if (header_size > sizeof(buf)) {
1429+ return -1; /* just to be sure */
1430+ }
1431+
1432+ uint32_t blob_buffer_offset = sizeof(VmaHeader);
1433+ memcpy(buf + blob_buffer_offset, vmaw->header_blob_table,
1434+ vmaw->header_blob_table_size);
1435+ head->blob_buffer_offset = GUINT32_TO_BE(blob_buffer_offset);
1436+ head->blob_buffer_size = GUINT32_TO_BE(vmaw->header_blob_table_pos);
1437+
1438+ g_checksum_reset(vmaw->md5csum);
1439+ g_checksum_update(vmaw->md5csum, (const guchar *)buf, header_size);
1440+ gsize csize = 16;
1441+ g_checksum_get_digest(vmaw->md5csum, (guint8 *)(head->md5sum), &csize);
1442+
1443+ return vma_queue_write(vmaw, buf, header_size);
1444+}
1445+
1446+static int coroutine_fn vma_writer_flush(VmaWriter *vmaw)
1447+{
1448+ assert(vmaw);
1449+
1450+ int ret;
1451+ int i;
1452+
1453+ if (vmaw->status < 0) {
1454+ return vmaw->status;
1455+ }
1456+
1457+ if (!vmaw->header_written) {
1458+ vmaw->header_written = true;
1459+ ret = vma_write_header(vmaw);
1460+ if (ret < 0) {
1461+ vma_writer_set_error(vmaw, "vma_writer_flush: write header failed");
1462+ return ret;
1463+ }
1464+ }
1465+
1466+ DPRINTF("VMA WRITE FLUSH %d %d\n", vmaw->outbuf_count, vmaw->outbuf_pos);
1467+
1468+
1469+ VmaExtentHeader *ehead = (VmaExtentHeader *)vmaw->outbuf;
1470+
1471+ ehead->magic = VMA_EXTENT_MAGIC;
1472+ ehead->reserved1 = 0;
1473+
1474+ for (i = 0; i < VMA_BLOCKS_PER_EXTENT; i++) {
1475+ ehead->blockinfo[i] = GUINT64_TO_BE(vmaw->outbuf_block_info[i]);
1476+ }
1477+
1478+ guint16 block_count = (vmaw->outbuf_pos - VMA_EXTENT_HEADER_SIZE) /
1479+ VMA_BLOCK_SIZE;
1480+
1481+ ehead->block_count = GUINT16_TO_BE(block_count);
1482+
1483+ memcpy(ehead->uuid, vmaw->uuid, sizeof(ehead->uuid));
1484+ memset(ehead->md5sum, 0, sizeof(ehead->md5sum));
1485+
1486+ g_checksum_reset(vmaw->md5csum);
1487+ g_checksum_update(vmaw->md5csum, vmaw->outbuf, VMA_EXTENT_HEADER_SIZE);
1488+ gsize csize = 16;
1489+ g_checksum_get_digest(vmaw->md5csum, ehead->md5sum, &csize);
1490+
1491+ int bytes = vmaw->outbuf_pos;
1492+ ret = vma_queue_write(vmaw, vmaw->outbuf, bytes);
1493+ if (ret != bytes) {
1494+ vma_writer_set_error(vmaw, "vma_writer_flush: failed write");
1495+ }
1496+
1497+ vmaw->outbuf_count = 0;
1498+ vmaw->outbuf_pos = VMA_EXTENT_HEADER_SIZE;
1499+
1500+ for (i = 0; i < VMA_BLOCKS_PER_EXTENT; i++) {
1501+ vmaw->outbuf_block_info[i] = 0;
1502+ }
1503+
1504+ return vmaw->status;
1505+}
1506+
1507+static int vma_count_open_streams(VmaWriter *vmaw)
1508+{
1509+ g_assert(vmaw != NULL);
1510+
1511+ int i;
1512+ int open_drives = 0;
1513+ for (i = 0; i <= 255; i++) {
1514+ if (vmaw->stream_info[i].size && !vmaw->stream_info[i].finished) {
1515+ open_drives++;
1516+ }
1517+ }
1518+
1519+ return open_drives;
1520+}
1521+
1522+/**
1523+ * all jobs should call this when there is no more data
1524+ * Returns: number of remaining stream (0 ==> finished)
1525+ */
1526+int coroutine_fn
1527+vma_writer_close_stream(VmaWriter *vmaw, uint8_t dev_id)
1528+{
1529+ g_assert(vmaw != NULL);
1530+
1531+ DPRINTF("vma_writer_set_status %d\n", dev_id);
1532+ if (!vmaw->stream_info[dev_id].size) {
1533+ vma_writer_set_error(vmaw, "vma_writer_close_stream: "
1534+ "no such stream %d", dev_id);
1535+ return -1;
1536+ }
1537+ if (vmaw->stream_info[dev_id].finished) {
1538+ vma_writer_set_error(vmaw, "vma_writer_close_stream: "
1539+ "stream already closed %d", dev_id);
1540+ return -1;
1541+ }
1542+
1543+ vmaw->stream_info[dev_id].finished = true;
1544+
1545+ int open_drives = vma_count_open_streams(vmaw);
1546+
1547+ if (open_drives <= 0) {
1548+ DPRINTF("vma_writer_set_status all drives completed\n");
1549+ qemu_co_mutex_lock(&vmaw->flush_lock);
1550+ int ret = vma_writer_flush(vmaw);
1551+ qemu_co_mutex_unlock(&vmaw->flush_lock);
1552+ if (ret < 0) {
1553+ vma_writer_set_error(vmaw, "vma_writer_close_stream: flush failed");
1554+ }
1555+ }
1556+
1557+ return open_drives;
1558+}
1559+
1560+int vma_writer_get_status(VmaWriter *vmaw, VmaStatus *status)
1561+{
1562+ int i;
1563+
1564+ g_assert(vmaw != NULL);
1565+
1566+ if (status) {
1567+ status->status = vmaw->status;
1568+ g_strlcpy(status->errmsg, vmaw->errmsg, sizeof(status->errmsg));
1569+ for (i = 0; i <= 255; i++) {
1570+ status->stream_info[i] = vmaw->stream_info[i];
1571+ }
1572+
1573+ uuid_unparse_lower(vmaw->uuid, status->uuid_str);
1574+ }
1575+
1576+ status->closed = vmaw->closed;
1577+
1578+ return vmaw->status;
1579+}
1580+
309874bd 1581+static int vma_writer_get_buffer(VmaWriter *vmaw, size_t bytes)
5ad5891c 1582+{
309874bd 1583+ int ret = 0;
5ad5891c 1584+
309874bd 1585+ qemu_co_mutex_lock(&vmaw->flush_lock);
5ad5891c 1586+
309874bd
DM		 1587+ /* rate limit */
1588+ uint64_t delay_ns = ratelimit_calculate_delay(&vmaw->limit, bytes);
1589+ if (delay_ns) {
1590+ DPRINTF("DELAY %zd\n", delay_ns);
1591+ co_sleep_ns(rt_clock, delay_ns);
1592+ }
5ad5891c 1593+
309874bd
DM		 1594+ /* wait until buffer is available */
1595+ while (vmaw->outbuf_count >= (VMA_BLOCKS_PER_EXTENT - 1)) {
1596+ ret = vma_writer_flush(vmaw);
1597+ if (ret < 0) {
5ad5891c 1598+ vma_writer_set_error(vmaw, "vma_writer_get_buffer: flush failed");
309874bd 1599+ break;
5ad5891c
DM		 1600+ }
1601+ }
1602+
309874bd
DM		 1603+ qemu_co_mutex_unlock(&vmaw->flush_lock);
1604+
1605+ return ret;
5ad5891c
DM		 1606+}
1607+
1608+
1609+int64_t coroutine_fn
1610+vma_writer_write(VmaWriter *vmaw, uint8_t dev_id, int64_t cluster_num,
1611+ unsigned char *buf, size_t *zero_bytes)
1612+{
1613+ g_assert(vmaw != NULL);
1614+ g_assert(zero_bytes != NULL);
1615+
1616+ *zero_bytes = 0;
1617+
1618+ if (vmaw->status < 0) {
1619+ return vmaw->status;
1620+ }
1621+
1622+ if (!dev_id || !vmaw->stream_info[dev_id].size) {
1623+ vma_writer_set_error(vmaw, "vma_writer_write: "
1624+ "no such stream %d", dev_id);
1625+ return -1;
1626+ }
1627+
1628+ if (vmaw->stream_info[dev_id].finished) {
1629+ vma_writer_set_error(vmaw, "vma_writer_write: "
1630+ "stream already closed %d", dev_id);
1631+ return -1;
1632+ }
1633+
1634+
1635+ if (cluster_num >= (((uint64_t)1)<<32)) {
1636+ vma_writer_set_error(vmaw, "vma_writer_write: "
1637+ "cluster number out of range");
1638+ return -1;
1639+ }
1640+
1641+ if (dev_id == vmaw->vmstate_stream) {
1642+ if (cluster_num != vmaw->vmstate_clusters) {
1643+ vma_writer_set_error(vmaw, "vma_writer_write: "
1644+ "non sequential vmstate write");
1645+ }
1646+ vmaw->vmstate_clusters++;
1647+ } else if (cluster_num >= vmaw->stream_info[dev_id].cluster_count) {
1648+ vma_writer_set_error(vmaw, "vma_writer_write: cluster number too big");
1649+ return -1;
1650+ }
1651+
309874bd
DM		 1652+ /* detect block containing zeroes */
1653+ int i;
1654+ int bit = 1;
1655+ uint16_t mask = 0;
1656+ size_t real_size = 0;
1657+ for (i = 0; i < 16; i++) {
1658+ unsigned char *vmablock = buf + (i*VMA_BLOCK_SIZE);
1659+ if (!buffer_is_zero(vmablock, VMA_BLOCK_SIZE)) {
1660+ mask |= bit;
1661+ real_size += VMA_BLOCK_SIZE;
1662+ }
1663+ bit = bit << 1;
1664+ }
1665+
5ad5891c 1666+ /* wait until buffer is available */
309874bd 1667+ if (vma_writer_get_buffer(vmaw, real_size) < 0) {
5ad5891c
DM		 1668+ vma_writer_set_error(vmaw, "vma_writer_write: "
1669+ "vma_writer_get_buffer failed");
1670+ return -1;
1671+ }
1672+
309874bd 1673+ DPRINTF("VMA WRITE %d %zd\n", dev_id, cluster_num);
5ad5891c 1674+
309874bd 1675+ bit = 1;
5ad5891c
DM		 1676+ for (i = 0; i < 16; i++) {
1677+ unsigned char *vmablock = buf + (i*VMA_BLOCK_SIZE);
309874bd
DM		 1678+ if (mask & bit) {
1679+ memcpy(vmaw->outbuf + vmaw->outbuf_pos, vmablock, VMA_BLOCK_SIZE);
1680+ vmaw->outbuf_pos += VMA_BLOCK_SIZE;
1681+ } else {
5ad5891c
DM		 1682+ DPRINTF("VMA WRITE %zd ZERO BLOCK %d\n", cluster_num, i);
1683+ vmaw->stream_info[dev_id].zero_bytes += VMA_BLOCK_SIZE;
1684+ *zero_bytes += VMA_BLOCK_SIZE;
5ad5891c
DM		 1685+ }
1686+
1687+ bit = bit << 1;
1688+ }
1689+
1690+ uint64_t block_info = ((uint64_t)mask) << (32+16);
1691+ block_info |= ((uint64_t)dev_id) << 32;
1692+ block_info |= (cluster_num & 0xffffffff);
309874bd 1693+ vmaw->outbuf_block_info[vmaw->outbuf_count] = block_info;
5ad5891c
DM		 1694+
1695+ DPRINTF("VMA WRITE MASK %zd %zx\n", cluster_num, block_info);
1696+
1697+ vmaw->outbuf_count++;
1698+
1699+ /** NOTE: We allways write whole clusters, but we correctly set
1700+ * transferred bytes. So transferred == size when when everything
1701+ * went OK.
1702+ */
1703+ size_t transferred = VMA_CLUSTER_SIZE;
1704+
1705+ if (dev_id != vmaw->vmstate_stream) {
1706+ uint64_t last = (cluster_num + 1) * VMA_CLUSTER_SIZE;
1707+ if (last > vmaw->stream_info[dev_id].size) {
1708+ uint64_t diff = last - vmaw->stream_info[dev_id].size;
1709+ if (diff >= VMA_CLUSTER_SIZE) {
1710+ vma_writer_set_error(vmaw, "vma_writer_write: "
1711+ "read after last cluster");
1712+ return -1;
1713+ }
1714+ transferred -= diff;
1715+ }
1716+ }
1717+
1718+ vmaw->stream_info[dev_id].transferred += transferred;
1719+
1720+ return transferred;
1721+}
1722+
1723+int vma_writer_close(VmaWriter *vmaw, Error **errp)
1724+{
1725+ g_assert(vmaw != NULL);
1726+
1727+ int i;
1728+
1729+ vma_queue_flush(vmaw);
1730+
1731+ /* this should not happen - just to be sure */
1732+ while (!qemu_co_queue_empty(&vmaw->wqueue)) {
1733+ DPRINTF("vma_writer_close wait\n");
1734+ co_sleep_ns(rt_clock, 1000000);
1735+ }
1736+
1737+ if (vmaw->cmd) {
1738+ if (pclose(vmaw->cmd) < 0) {
1739+ vma_writer_set_error(vmaw, "vma_writer_close: "
1740+ "pclose failed - %s", strerror(errno));
1741+ }
1742+ } else {
1743+ if (close(vmaw->fd) < 0) {
1744+ vma_writer_set_error(vmaw, "vma_writer_close: "
1745+ "close failed - %s", strerror(errno));
1746+ }
1747+ }
1748+
1749+ for (i = 0; i <= 255; i++) {
1750+ VmaStreamInfo *si = &vmaw->stream_info[i];
1751+ if (si->size) {
1752+ if (!si->finished) {
1753+ vma_writer_set_error(vmaw, "vma_writer_close: "
1754+ "detected open stream '%s'", si->devname);
1755+ } else if ((si->transferred != si->size) &&
1756+ (i != vmaw->vmstate_stream)) {
1757+ vma_writer_set_error(vmaw, "vma_writer_close: "
1758+ "incomplete stream '%s' (%zd != %zd)",
1759+ si->devname, si->transferred, si->size);
1760+ }
1761+ }
1762+ }
1763+
1764+ for (i = 0; i <= 255; i++) {
1765+ vmaw->stream_info[i].finished = 1; /* mark as closed */
1766+ }
1767+
1768+ vmaw->closed = 1;
1769+
1770+ if (vmaw->status < 0 && *errp == NULL) {
1771+ error_setg(errp, "%s", vmaw->errmsg);
1772+ }
1773+
1774+ return vmaw->status;
1775+}
1776+
1777+void vma_writer_destroy(VmaWriter *vmaw)
1778+{
1779+ assert(vmaw);
1780+
1781+ int i;
1782+
1783+ for (i = 0; i <= 255; i++) {
1784+ if (vmaw->stream_info[i].devname) {
1785+ g_free(vmaw->stream_info[i].devname);
1786+ }
1787+ }
1788+
1789+ if (vmaw->md5csum) {
1790+ g_checksum_free(vmaw->md5csum);
1791+ }
1792+
1793+ g_free(vmaw);
1794+}
1795+
1796+/* backup driver plugin */
1797+
1798+static int vma_dump_cb(void *opaque, uint8_t dev_id, int64_t cluster_num,
1799+ unsigned char *buf, size_t *zero_bytes)
1800+{
1801+ VmaWriter *vmaw = opaque;
1802+
1803+ return vma_writer_write(vmaw, dev_id, cluster_num, buf, zero_bytes);
1804+}
1805+
1806+static int vma_close_cb(void *opaque, Error **errp)
1807+{
1808+ VmaWriter *vmaw = opaque;
1809+
1810+ int res = vma_writer_close(vmaw, errp);
1811+ vma_writer_destroy(vmaw);
1812+
1813+ return res;
1814+}
1815+
1816+static int vma_complete_cb(void *opaque, uint8_t dev_id, int ret)
1817+{
1818+ VmaWriter *vmaw = opaque;
1819+
1820+ if (ret < 0) {
1821+ vma_writer_set_error(vmaw, "backup_complete_cb %d", ret);
1822+ }
1823+
1824+ return vma_writer_close_stream(vmaw, dev_id);
1825+}
1826+
1827+static int vma_register_stream_cb(void *opaque, const char *devname,
1828+ size_t size)
1829+{
1830+ VmaWriter *vmaw = opaque;
1831+
1832+ return vma_writer_register_stream(vmaw, devname, size);
1833+}
1834+
1835+static int vma_register_config_cb(void *opaque, const char *name,
1836+ gpointer data, size_t data_len)
1837+{
1838+ VmaWriter *vmaw = opaque;
1839+
1840+ return vma_writer_add_config(vmaw, name, data, data_len);
1841+}
1842+
1843+static void *vma_open_cb(const char *filename, uuid_t uuid, int64_t speed,
1844+ Error **errp)
1845+{
1846+ return vma_writer_create(filename, uuid, speed, errp);
1847+}
1848+
1849+const BackupDriver backup_vma_driver = {
1850+ .format = "vma",
1851+ .open_cb = vma_open_cb,
1852+ .close_cb = vma_close_cb,
1853+ .register_config_cb = vma_register_config_cb,
1854+ .register_stream_cb = vma_register_stream_cb,
1855+ .dump_cb = vma_dump_cb,
1856+ .complete_cb = vma_complete_cb,
1857+};
1858+
1859diff --git a/vma.c b/vma.c
1860new file mode 100644
309874bd 1861index 0000000..9b47b92
5ad5891c
DM		 1862--- /dev/null
1863+++ b/vma.c
309874bd 1864@@ -0,0 +1,561 @@
5ad5891c
DM		 1865+/*
1866+ * VMA: Virtual Machine Archive
1867+ *
1868+ * Copyright (C) 2012 Proxmox Server Solutions
1869+ *
1870+ * Authors:
1871+ * Dietmar Maurer (dietmar@proxmox.com)
1872+ *
1873+ * This work is licensed under the terms of the GNU GPL, version 2 or later.
1874+ * See the COPYING file in the top-level directory.
1875+ *
1876+ */
1877+
1878+#include <stdio.h>
1879+#include <errno.h>
1880+#include <unistd.h>
1881+#include <stdio.h>
1882+#include <string.h>
1883+#include <sys/types.h>
1884+#include <sys/stat.h>
1885+#include <fcntl.h>
1886+#include <glib.h>
1887+
1888+#include "qemu-common.h"
1889+#include "qemu-option.h"
1890+#include "qemu-error.h"
1891+#include "osdep.h"
1892+#include "sysemu.h"
1893+#include "block_int.h"
1894+#include <stdio.h>
1895+#include "vma.h"
1896+
1897+static void help(void)
1898+{
1899+ const char *help_msg =
1900+ "usage: vma command [command options]\n"
1901+ "\n"
1902+ "vma list <filename>\n"
1903+ "vma create <filename> [-c config] <archive> pathname ...\n"
1904+ "vma extract <filename> [-r] <targetdir>\n"
1905+ ;
1906+
1907+ printf("%s", help_msg);
1908+ exit(1);
1909+}
1910+
1911+static const char *extract_devname(const char *path, char **devname, int index)
1912+{
1913+ assert(path);
1914+
1915+ const char *sep = strchr(path, '=');
1916+
1917+ if (sep) {
1918+ *devname = g_strndup(path, sep - path);
1919+ path = sep + 1;
1920+ } else {
1921+ if (index >= 0) {
1922+ *devname = g_strdup_printf("disk%d", index);
1923+ } else {
1924+ *devname = NULL;
1925+ }
1926+ }
1927+
1928+ return path;
1929+}
1930+
1931+static void print_content(VmaReader *vmar)
1932+{
1933+ assert(vmar);
1934+
1935+ VmaHeader *head = vma_reader_get_header(vmar);
1936+
5ad5891c
DM		 1937+ GList *l = vma_reader_get_config_data(vmar);
1938+ while (l && l->data) {
1939+ VmaConfigData *cdata = (VmaConfigData *)l->data;
1940+ l = g_list_next(l);
1941+ printf("CFG: size: %d name: %s\n", cdata->len, cdata->name);
1942+ }
1943+
1944+ int i;
1945+ VmaDeviceInfo *di;
1946+ for (i = 1; i < 255; i++) {
1947+ di = vma_reader_get_device_info(vmar, i);
1948+ if (di) {
1949+ if (strcmp(di->devname, "vmstate") == 0) {
1950+ printf("VMSTATE: dev_id=%d memory: %zd\n", i, di->size);
1951+ } else {
1952+ printf("DEV: dev_id=%d size: %zd devname: %s\n",
1953+ i, di->size, di->devname);
1954+ }
1955+ }
1956+ }
d95c8c31
DM		 1957+ /* ctime is the last entry we print */
1958+ printf("CTIME: %s", ctime(&head->ctime));
1959+ fflush(stdout);
5ad5891c
DM		 1960+}
1961+
1962+static int list_content(int argc, char **argv)
1963+{
1964+ int c, ret = 0;
1965+ const char *filename;
1966+
1967+ for (;;) {
1968+ c = getopt(argc, argv, "h");
1969+ if (c == -1) {
1970+ break;
1971+ }
1972+ switch (c) {
1973+ case '?':
1974+ case 'h':
1975+ help();
1976+ break;
1977+ default:
1978+ g_assert_not_reached();
1979+ }
1980+ }
1981+
1982+ /* Get the filename */
1983+ if ((optind + 1) != argc) {
1984+ help();
1985+ }
1986+ filename = argv[optind++];
1987+
1988+ Error *errp = NULL;
1989+ VmaReader *vmar = vma_reader_create(filename, &errp);
1990+
1991+ if (!vmar) {
1992+ g_error("%s", error_get_pretty(errp));
1993+ }
1994+
1995+ print_content(vmar);
1996+
1997+ vma_reader_destroy(vmar);
1998+
1999+ return ret;
2000+}
2001+
2002+typedef struct RestoreMap {
2003+ char *devname;
2004+ char *path;
2005+ bool write_zero;
2006+} RestoreMap;
2007+
2008+static int extract_content(int argc, char **argv)
2009+{
2010+ int c, ret = 0;
4244016d 2011+ int verbose = 0;
5ad5891c
DM		 2012+ const char *filename;
2013+ const char *dirname;
3f350e0d 2014+ const char *readmap = NULL;
5ad5891c
DM		 2015+
2016+ for (;;) {
4244016d 2017+ c = getopt(argc, argv, "hvr:");
5ad5891c
DM		 2018+ if (c == -1) {
2019+ break;
2020+ }
2021+ switch (c) {
2022+ case '?':
2023+ case 'h':
2024+ help();
2025+ break;
2026+ case 'r':
3f350e0d 2027+ readmap = optarg;
5ad5891c 2028+ break;
309874bd 2029+ case 'v':
4244016d
DM		 2030+ verbose = 1;
2031+ break;
5ad5891c
DM		 2032+ default:
2033+ help();
2034+ }
2035+ }
2036+
2037+ /* Get the filename */
2038+ if ((optind + 2) != argc) {
2039+ help();
2040+ }
2041+ filename = argv[optind++];
2042+ dirname = argv[optind++];
2043+
2044+ Error *errp = NULL;
2045+ VmaReader *vmar = vma_reader_create(filename, &errp);
2046+
2047+ if (!vmar) {
2048+ g_error("%s", error_get_pretty(errp));
2049+ }
2050+
2051+ if (mkdir(dirname, 0777) < 0) {
2052+ g_error("unable to create target directory %s - %s",
2053+ dirname, strerror(errno));
2054+ }
2055+
2056+ GList *l = vma_reader_get_config_data(vmar);
2057+ while (l && l->data) {
2058+ VmaConfigData *cdata = (VmaConfigData *)l->data;
2059+ l = g_list_next(l);
2060+ char *cfgfn = g_strdup_printf("%s/%s", dirname, cdata->name);
2061+ GError *err = NULL;
2062+ if (!g_file_set_contents(cfgfn, (gchar *)cdata->data, cdata->len,
2063+ &err)) {
3f350e0d 2064+ g_error("unable to write file: %s", err->message);
5ad5891c
DM		 2065+ }
2066+ }
2067+
2068+ GHashTable *devmap = g_hash_table_new(g_str_hash, g_str_equal);
2069+
2070+ if (readmap) {
2071+ print_content(vmar);
2072+
3f350e0d
DM		 2073+ FILE *map = fopen(readmap, "r");
2074+ if (!map) {
2075+ g_error("unable to open fifo %s - %s", readmap, strerror(errno));
2076+ }
2077+
5ad5891c
DM		 2078+ while (1) {
2079+ char inbuf[8192];
3f350e0d 2080+ char *line = fgets(inbuf, sizeof(inbuf), map);
5ad5891c
DM		 2081+ if (!line || line[0] == '\0' || !strcmp(line, "done\n")) {
2082+ break;
2083+ }
2084+ int len = strlen(line);
2085+ if (line[len - 1] == '\n') {
2086+ line[len - 1] = '\0';
2087+ if (len == 1) {
2088+ break;
2089+ }
2090+ }
2091+
2092+ const char *path;
2093+ bool write_zero;
2094+ if (line[0] == '0' && line[1] == ':') {
2095+ path = inbuf + 2;
2096+ write_zero = false;
2097+ } else if (line[0] == '1' && line[1] == ':') {
2098+ path = inbuf + 2;
2099+ write_zero = true;
3f350e0d 2100+ } else {
5ad5891c
DM		 2101+ g_error("read map failed - parse error ('%s')", inbuf);
2102+ }
2103+
2104+ char *devname = NULL;
2105+ path = extract_devname(path, &devname, -1);
2106+ if (!devname) {
2107+ g_error("read map failed - no dev name specified ('%s')",
2108+ inbuf);
2109+ }
2110+
5ad5891c
DM		 2111+ RestoreMap *map = g_new0(RestoreMap, 1);
2112+ map->devname = g_strdup(devname);
2113+ map->path = g_strdup(path);
2114+ map->write_zero = write_zero;
2115+
2116+ g_hash_table_insert(devmap, map->devname, map);
2117+
2118+ };
2119+ }
2120+
2121+ int i;
2122+ int vmstate_fd = -1;
2123+ guint8 vmstate_stream = 0;
2124+
2125+ for (i = 1; i < 255; i++) {
2126+ VmaDeviceInfo *di = vma_reader_get_device_info(vmar, i);
2127+ if (di && (strcmp(di->devname, "vmstate") == 0)) {
2128+ vmstate_stream = i;
2129+ char *statefn = g_strdup_printf("%s/vmstate.bin", dirname);
2130+ vmstate_fd = open(statefn, O_WRONLY|O_CREAT|O_EXCL, 0644);
2131+ if (vmstate_fd < 0) {
2132+ g_error("create vmstate file '%s' failed - %s", statefn,
2133+ strerror(errno));
2134+ }
2135+ g_free(statefn);
2136+ } else if (di) {
2137+ char *devfn = NULL;
2138+ int flags = BDRV_O_RDWR|BDRV_O_CACHE_WB;
2139+ bool write_zero = true;
2140+
2141+ if (readmap) {
2142+ RestoreMap *map;
2143+ map = (RestoreMap *)g_hash_table_lookup(devmap, di->devname);
2144+ if (map == NULL) {
2145+ g_error("no device name mapping for %s", di->devname);
2146+ }
2147+ devfn = map->path;
2148+ write_zero = map->write_zero;
2149+ } else {
2150+ devfn = g_strdup_printf("%s/tmp-disk-%s.raw",
2151+ dirname, di->devname);
2152+ printf("DEVINFO %s %zd\n", devfn, di->size);
2153+
2154+ if (bdrv_img_create(devfn, "raw", NULL, NULL, NULL,
2155+ di->size, flags)) {
2156+ g_error("can't create file %s", devfn);
2157+ }
2158+
2159+ /* Note: we created an empty file above, so there is no
2160+ * need to write zeroes (so we generate a sparse file)
2161+ */
2162+ write_zero = false;
2163+ }
2164+
9de54cd3
DM		 2165+ BlockDriverState *bs = bdrv_new(di->devname);
2166+ if (bdrv_open(bs, devfn, flags, NULL)) {
5ad5891c
DM		 2167+ g_error("can't open file %s", devfn);
2168+ }
2169+ if (vma_reader_register_bs(vmar, i, bs, write_zero, &errp) < 0) {
2170+ g_error("%s", error_get_pretty(errp));
2171+ }
2172+
2173+ if (!readmap) {
2174+ g_free(devfn);
2175+ }
2176+ }
2177+ }
2178+
4244016d 2179+ if (vma_reader_restore(vmar, vmstate_fd, verbose, &errp) < 0) {
5ad5891c
DM		 2180+ g_error("restore failed - %s", error_get_pretty(errp));
2181+ }
2182+
2183+ if (!readmap) {
2184+ for (i = 1; i < 255; i++) {
2185+ VmaDeviceInfo *di = vma_reader_get_device_info(vmar, i);
2186+ if (di && (i != vmstate_stream)) {
2187+ char *tmpfn = g_strdup_printf("%s/tmp-disk-%s.raw",
2188+ dirname, di->devname);
2189+ char *fn = g_strdup_printf("%s/disk-%s.raw",
2190+ dirname, di->devname);
2191+ if (rename(tmpfn, fn) != 0) {
2192+ g_error("rename %s to %s failed - %s",
2193+ tmpfn, fn, strerror(errno));
2194+ }
2195+ }
2196+ }
2197+ }
2198+
2199+ vma_reader_destroy(vmar);
2200+
2201+ bdrv_close_all();
2202+
2203+ return ret;
2204+}
2205+
2206+typedef struct BackupCB {
2207+ VmaWriter *vmaw;
2208+ uint8_t dev_id;
2209+} BackupCB;
2210+
2211+static int backup_dump_cb(void *opaque, BlockDriverState *bs,
2212+ int64_t cluster_num, unsigned char *buf)
2213+{
2214+ BackupCB *bcb = opaque;
2215+ size_t zb = 0;
2216+ if (vma_writer_write(bcb->vmaw, bcb->dev_id, cluster_num, buf, &zb) < 0) {
2217+ g_warning("backup_dump_cb vma_writer_write failed");
2218+ return -1;
2219+ }
2220+
2221+ return 0;
2222+}
2223+
2224+static void backup_complete_cb(void *opaque, int ret)
2225+{
2226+ BackupCB *bcb = opaque;
2227+
2228+ if (ret < 0) {
2229+ vma_writer_set_error(bcb->vmaw, "backup_complete_cb %d", ret);
2230+ }
2231+
2232+ if (vma_writer_close_stream(bcb->vmaw, bcb->dev_id) <= 0) {
2233+ Error *err = NULL;
2234+ if (vma_writer_close(bcb->vmaw, &err) != 0) {
2235+ g_warning("vma_writer_close failed %s", error_get_pretty(err));
2236+ }
2237+ }
2238+}
2239+
2240+static int create_archive(int argc, char **argv)
2241+{
2242+ int i, c, res;
2243+ int verbose = 0;
2244+ const char *archivename;
2245+ GList *config_files = NULL;
2246+
2247+ for (;;) {
2248+ c = getopt(argc, argv, "hvc:");
2249+ if (c == -1) {
2250+ break;
2251+ }
2252+ switch (c) {
2253+ case '?':
2254+ case 'h':
2255+ help();
2256+ break;
2257+ case 'c':
2258+ config_files = g_list_append(config_files, optarg);
2259+ break;
2260+ case 'v':
2261+ verbose = 1;
2262+ break;
2263+ default:
2264+ g_assert_not_reached();
2265+ }
2266+ }
2267+
2268+
2269+ /* make sure we have archive name and at least one path */
2270+ if ((optind + 2) > argc) {
2271+ help();
2272+ }
2273+
2274+ archivename = argv[optind++];
2275+
2276+ uuid_t uuid;
2277+ uuid_generate(uuid);
2278+
2279+ Error *local_err = NULL;
2280+ VmaWriter *vmaw = vma_writer_create(archivename, uuid, 0, &local_err);
2281+
2282+ if (vmaw == NULL) {
2283+ g_error("%s", error_get_pretty(local_err));
2284+ }
2285+
2286+ GList *l = config_files;
2287+ while (l && l->data) {
2288+ char *name = l->data;
2289+ char *cdata = NULL;
2290+ gsize clen = 0;
2291+ GError *err = NULL;
2292+ if (!g_file_get_contents(name, &cdata, &clen, &err)) {
2293+ unlink(archivename);
2294+ g_error("Unable to read file: %s", err->message);
2295+ }
2296+
2297+ if (vma_writer_add_config(vmaw, name, cdata, clen) != 0) {
2298+ unlink(archivename);
2299+ g_error("Unable to append config data %s (len = %zd)",
2300+ name, clen);
2301+ }
2302+ l = g_list_next(l);
2303+ }
2304+
2305+ int ind = 0;
2306+ while (optind < argc) {
2307+ const char *path = argv[optind++];
2308+ char *devname = NULL;
2309+ path = extract_devname(path, &devname, ind++);
2310+
2311+ BlockDriver *drv = NULL;
2312+ BlockDriverState *bs = bdrv_new(devname);
2313+
2314+ res = bdrv_open(bs, path, BDRV_O_CACHE_WB , drv);
2315+ if (res < 0) {
2316+ unlink(archivename);
2317+ g_error("bdrv_open '%s' failed", path);
2318+ }
2319+ int64_t size = bdrv_getlength(bs);
2320+ int dev_id = vma_writer_register_stream(vmaw, devname, size);
2321+ if (dev_id <= 0) {
2322+ unlink(archivename);
2323+ g_error("vma_writer_register_stream '%s' failed", devname);
2324+ }
2325+
2326+ BackupCB *bcb = g_new0(BackupCB, 1);
2327+ bcb->vmaw = vmaw;
2328+ bcb->dev_id = dev_id;
2329+
309874bd
DM		 2330+ if (backup_job_create(bs, backup_dump_cb, backup_complete_cb,
2331+ bcb) < 0) {
5ad5891c
DM		 2332+ unlink(archivename);
2333+ g_error("backup_job_start failed");
309874bd
DM		 2334+ } else {
2335+ backup_job_start(bs);
5ad5891c
DM		 2336+ }
2337+ }
2338+
2339+ VmaStatus vmastat;
2340+ int percent = 0;
2341+ int last_percent = -1;
2342+
2343+ while (1) {
2344+ main_loop_wait(false);
2345+ vma_writer_get_status(vmaw, &vmastat);
2346+
2347+ if (verbose) {
2348+
2349+ uint64_t total = 0;
2350+ uint64_t transferred = 0;
2351+ uint64_t zero_bytes = 0;
2352+
2353+ int i;
2354+ for (i = 0; i < 256; i++) {
2355+ if (vmastat.stream_info[i].size) {
2356+ total += vmastat.stream_info[i].size;
2357+ transferred += vmastat.stream_info[i].transferred;
2358+ zero_bytes += vmastat.stream_info[i].zero_bytes;
2359+ }
2360+ }
2361+ percent = (transferred*100)/total;
2362+ if (percent != last_percent) {
2363+ printf("progress %d%% %zd/%zd %zd\n", percent,
2364+ transferred, total, zero_bytes);
2365+
2366+ last_percent = percent;
2367+ }
2368+ }
2369+
2370+ if (vmastat.closed) {
2371+ break;
2372+ }
2373+ }
2374+
2375+ bdrv_drain_all();
2376+
2377+ vma_writer_get_status(vmaw, &vmastat);
2378+
2379+ if (verbose) {
2380+ for (i = 0; i < 256; i++) {
2381+ VmaStreamInfo *si = &vmastat.stream_info[i];
2382+ if (si->size) {
2383+ printf("image %s: size=%zd zeros=%zd saved=%zd\n", si->devname,
2384+ si->size, si->zero_bytes, si->size - si->zero_bytes);
2385+ }
2386+ }
2387+ }
2388+
2389+ if (vmastat.status < 0) {
2390+ unlink(archivename);
2391+ g_error("creating vma archive failed");
2392+ }
2393+
2394+ return 0;
2395+}
2396+
2397+int main(int argc, char **argv)
2398+{
2399+ const char *cmdname;
2400+
2401+ error_set_progname(argv[0]);
2402+
2403+ qemu_init_main_loop();
2404+
2405+ bdrv_init();
2406+
2407+ if (argc < 2) {
2408+ help();
2409+ }
2410+
2411+ cmdname = argv[1];
2412+ argc--; argv++;
2413+
2414+
2415+ if (!strcmp(cmdname, "list")) {
2416+ return list_content(argc, argv);
2417+ } else if (!strcmp(cmdname, "create")) {
2418+ return create_archive(argc, argv);
2419+ } else if (!strcmp(cmdname, "extract")) {
2420+ return extract_content(argc, argv);
2421+ }
2422+
2423+ help();
2424+ return 0;
2425+}
2426diff --git a/vma.h b/vma.h
2427new file mode 100644
309874bd 2428index 0000000..689e639
5ad5891c
DM		 2429--- /dev/null
2430+++ b/vma.h
4244016d 2431@@ -0,0 +1,146 @@
5ad5891c
DM		 2432+/*
2433+ * VMA: Virtual Machine Archive
2434+ *
2435+ * Copyright (C) Proxmox Server Solutions
2436+ *
2437+ * Authors:
2438+ * Dietmar Maurer (dietmar@proxmox.com)
2439+ *
2440+ * This work is licensed under the terms of the GNU GPL, version 2 or later.
2441+ * See the COPYING file in the top-level directory.
2442+ *
2443+ */
2444+
2445+#ifndef BACKUP_VMA_H
2446+#define BACKUP_VMA_H
2447+
2448+#include "backup.h"
2449+#include "error.h"
2450+
2451+#define VMA_BLOCK_BITS 12
2452+#define VMA_BLOCK_SIZE (1<<VMA_BLOCK_BITS)
2453+#define VMA_CLUSTER_BITS (VMA_BLOCK_BITS+4)
2454+#define VMA_CLUSTER_SIZE (1<<VMA_CLUSTER_BITS)
2455+
2456+#if VMA_CLUSTER_SIZE != 65536
2457+#error unexpected cluster size
2458+#endif
2459+
2460+#define VMA_EXTENT_HEADER_SIZE 512
2461+#define VMA_BLOCKS_PER_EXTENT 59
2462+#define VMA_MAX_CONFIGS 256
2463+
2464+#define VMA_MAX_EXTENT_SIZE \
2465+ (VMA_EXTENT_HEADER_SIZE+VMA_CLUSTER_SIZE*VMA_BLOCKS_PER_EXTENT)
2466+#if VMA_MAX_EXTENT_SIZE != 3867136
2467+#error unexpected VMA_EXTENT_SIZE
2468+#endif
2469+
2470+/* File Format Definitions */
2471+
2472+#define VMA_MAGIC (GUINT32_TO_BE(('V'<<24)|('M'<<16)|('A'<<8)|0x00))
2473+#define VMA_EXTENT_MAGIC (GUINT32_TO_BE(('V'<<24)|('M'<<16)|('A'<<8)|'E'))
2474+
2475+typedef struct VmaDeviceInfoHeader {
2476+ uint32_t devname_ptr; /* offset into blob_buffer table */
2477+ uint32_t reserved0;
2478+ uint64_t size; /* device size in bytes */
2479+ uint64_t reserved1;
2480+ uint64_t reserved2;
2481+} VmaDeviceInfoHeader;
2482+
2483+typedef struct VmaHeader {
2484+ uint32_t magic;
2485+ uint32_t version;
2486+ unsigned char uuid[16];
2487+ int64_t ctime;
2488+ unsigned char md5sum[16];
2489+
2490+ uint32_t blob_buffer_offset;
2491+ uint32_t blob_buffer_size;
2492+ uint32_t header_size;
2493+
2494+ unsigned char reserved[1984];
2495+
2496+ uint32_t config_names[VMA_MAX_CONFIGS]; /* offset into blob_buffer table */
2497+ uint32_t config_data[VMA_MAX_CONFIGS]; /* offset into blob_buffer table */
2498+
2499+ VmaDeviceInfoHeader dev_info[256];
2500+} VmaHeader;
2501+
2502+typedef struct VmaExtentHeader {
2503+ uint32_t magic;
2504+ uint16_t reserved1;
2505+ uint16_t block_count;
2506+ unsigned char uuid[16];
2507+ unsigned char md5sum[16];
2508+ uint64_t blockinfo[VMA_BLOCKS_PER_EXTENT];
2509+} VmaExtentHeader;
2510+
2511+/* functions/definitions to read/write vma files */
2512+
2513+typedef struct VmaReader VmaReader;
2514+
2515+typedef struct VmaWriter VmaWriter;
2516+
2517+typedef struct VmaConfigData {
2518+ const char *name;
2519+ const void *data;
2520+ uint32_t len;
2521+} VmaConfigData;
2522+
2523+typedef struct VmaStreamInfo {
2524+ uint64_t size;
2525+ uint64_t cluster_count;
2526+ uint64_t transferred;
2527+ uint64_t zero_bytes;
2528+ int finished;
2529+ char *devname;
2530+} VmaStreamInfo;
2531+
2532+typedef struct VmaStatus {
2533+ int status;
2534+ bool closed;
2535+ char errmsg[8192];
2536+ char uuid_str[37];
2537+ VmaStreamInfo stream_info[256];
2538+} VmaStatus;
2539+
2540+typedef struct VmaDeviceInfo {
2541+ uint64_t size; /* device size in bytes */
2542+ const char *devname;
2543+} VmaDeviceInfo;
2544+
2545+extern const BackupDriver backup_vma_driver;
2546+
2547+VmaWriter *vma_writer_create(const char *filename, uuid_t uuid, int64_t speed,
2548+ Error **errp);
2549+int vma_writer_close(VmaWriter *vmaw, Error **errp);
2550+void vma_writer_destroy(VmaWriter *vmaw);
2551+int vma_writer_add_config(VmaWriter *vmaw, const char *name, gpointer data,
2552+ size_t len);
2553+int vma_writer_register_stream(VmaWriter *vmaw, const char *devname,
2554+ size_t size);
2555+
2556+int64_t coroutine_fn vma_writer_write(VmaWriter *vmaw, uint8_t dev_id,
2557+ int64_t cluster_num, unsigned char *buf,
2558+ size_t *zero_bytes);
2559+
2560+int coroutine_fn vma_writer_close_stream(VmaWriter *vmaw, uint8_t dev_id);
2561+
2562+int vma_writer_get_status(VmaWriter *vmaw, VmaStatus *status);
2563+void vma_writer_set_error(VmaWriter *vmaw, const char *fmt, ...);
2564+
2565+
2566+VmaReader *vma_reader_create(const char *filename, Error **errp);
2567+void vma_reader_destroy(VmaReader *vmar);
2568+VmaHeader *vma_reader_get_header(VmaReader *vmar);
2569+GList *vma_reader_get_config_data(VmaReader *vmar);
2570+VmaDeviceInfo *vma_reader_get_device_info(VmaReader *vmar, guint8 dev_id);
2571+int vma_reader_register_bs(VmaReader *vmar, guint8 dev_id,
2572+ BlockDriverState *bs, bool write_zeroes,
2573+ Error **errp);
309874bd
DM		 2574+int vma_reader_restore(VmaReader *vmar, int vmstate_fd, bool verbose,
2575+ Error **errp);
5ad5891c
DM		 2576+
2577+#endif /* BACKUP_VMA_H */
2578--
25791.7.2.5
2580
old QEMU for PVE