]> 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
2dfd543c 1From 2b6ca0c6087ed51bc8318731713bb3aa83f606db Mon Sep 17 00:00:00 2001
5ad5891c
DM		 2From: Dietmar Maurer <dietmar@proxmox.com>
3Date: Tue, 13 Nov 2012 11:11:38 +0100
55827521 4Subject: [PATCH v3 4/6] 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 ++
2dfd543c
DM		 14 vma-reader.c | 801 ++++++++++++++++++++++++++++++++++++++++
15 vma-writer.c | 933 +++++++++++++++++++++++++++++++++++++++++++++++
16 vma.c | 561 ++++++++++++++++++++++++++++
55827521 17 vma.h | 145 ++++++++
2dfd543c 18 8 files changed, 2471 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
2dfd543c 60index ba28654..a030a13 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
2dfd543c 71@@ -1515,10 +1516,11 @@ 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
2dfd543c 924index 0000000..08c4ee3
5ad5891c
DM		 925--- /dev/null
926+++ b/vma-writer.c
2dfd543c 927@@ -0,0 +1,933 @@
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"
5ad5891c
DM		 956+#include "vma.h"
957+#include "block.h"
3f350e0d 958+#include "monitor.h"
5ad5891c
DM		 959+
960+#define DEBUG_VMA 0
961+
962+#define DPRINTF(fmt, ...)\
963+ do { if (DEBUG_VMA) { printf("vma: " fmt, ## __VA_ARGS__); } } while (0)
964+
965+#define WRITE_BUFFERS 5
966+
967+typedef struct VmaAIOCB VmaAIOCB;
968+struct VmaAIOCB {
5ad5891c 969+ unsigned char buffer[VMA_MAX_EXTENT_SIZE];
2dfd543c 970+ VmaWriter *vmaw;
5ad5891c
DM		 971+ size_t bytes;
972+ Coroutine *co;
973+};
974+
975+struct VmaWriter {
976+ int fd;
977+ FILE *cmd;
978+ int status;
979+ char errmsg[8192];
980+ uuid_t uuid;
981+ bool header_written;
982+ bool closed;
983+
984+ /* we always write extents */
985+ unsigned char outbuf[VMA_MAX_EXTENT_SIZE];
986+ int outbuf_pos; /* in bytes */
987+ int outbuf_count; /* in VMA_BLOCKS */
988+ uint64_t outbuf_block_info[VMA_BLOCKS_PER_EXTENT];
989+
2dfd543c 990+ VmaAIOCB *aiocbs[WRITE_BUFFERS];
5ad5891c
DM		 991+ CoQueue wqueue;
992+
993+ GChecksum *md5csum;
994+ CoMutex writer_lock;
995+ CoMutex flush_lock;
996+ Coroutine *co_writer;
5ad5891c
DM		 997+
998+ /* drive informations */
999+ VmaStreamInfo stream_info[256];
1000+ guint stream_count;
1001+
1002+ guint8 vmstate_stream;
1003+ uint32_t vmstate_clusters;
1004+
1005+ /* header blob table */
1006+ char *header_blob_table;
1007+ uint32_t header_blob_table_size;
1008+ uint32_t header_blob_table_pos;
1009+
1010+ /* store for config blobs */
1011+ uint32_t config_names[VMA_MAX_CONFIGS]; /* offset into blob_buffer table */
1012+ uint32_t config_data[VMA_MAX_CONFIGS]; /* offset into blob_buffer table */
1013+ uint32_t config_count;
1014+};
1015+
1016+void vma_writer_set_error(VmaWriter *vmaw, const char *fmt, ...)
1017+{
1018+ va_list ap;
1019+
1020+ if (vmaw->status < 0) {
1021+ return;
1022+ }
1023+
1024+ vmaw->status = -1;
1025+
1026+ va_start(ap, fmt);
1027+ g_vsnprintf(vmaw->errmsg, sizeof(vmaw->errmsg), fmt, ap);
1028+ va_end(ap);
1029+
1030+ DPRINTF("vma_writer_set_error: %s\n", vmaw->errmsg);
1031+}
1032+
1033+static uint32_t allocate_header_blob(VmaWriter *vmaw, const char *data,
1034+ size_t len)
1035+{
1036+ if (len > 65535) {
1037+ return 0;
1038+ }
1039+
1040+ if (!vmaw->header_blob_table ||
1041+ (vmaw->header_blob_table_size <
1042+ (vmaw->header_blob_table_pos + len + 2))) {
1043+ int newsize = vmaw->header_blob_table_size + ((len + 2 + 511)/512)*512;
1044+
1045+ vmaw->header_blob_table = g_realloc(vmaw->header_blob_table, newsize);
1046+ memset(vmaw->header_blob_table + vmaw->header_blob_table_size,
1047+ 0, newsize - vmaw->header_blob_table_size);
1048+ vmaw->header_blob_table_size = newsize;
1049+ }
1050+
1051+ uint32_t cpos = vmaw->header_blob_table_pos;
1052+ vmaw->header_blob_table[cpos] = len & 255;
1053+ vmaw->header_blob_table[cpos+1] = (len >> 8) & 255;
1054+ memcpy(vmaw->header_blob_table + cpos + 2, data, len);
1055+ vmaw->header_blob_table_pos += len + 2;
1056+ return cpos;
1057+}
1058+
1059+static uint32_t allocate_header_string(VmaWriter *vmaw, const char *str)
1060+{
1061+ assert(vmaw);
1062+
1063+ size_t len = strlen(str) + 1;
1064+
1065+ return allocate_header_blob(vmaw, str, len);
1066+}
1067+
1068+int vma_writer_add_config(VmaWriter *vmaw, const char *name, gpointer data,
1069+ gsize len)
1070+{
1071+ assert(vmaw);
1072+ assert(!vmaw->header_written);
1073+ assert(vmaw->config_count < VMA_MAX_CONFIGS);
1074+ assert(name);
1075+ assert(data);
1076+ assert(len);
1077+
1078+ uint32_t name_ptr = allocate_header_string(vmaw, name);
1079+ if (!name_ptr) {
1080+ return -1;
1081+ }
1082+
1083+ uint32_t data_ptr = allocate_header_blob(vmaw, data, len);
1084+ if (!data_ptr) {
1085+ return -1;
1086+ }
1087+
1088+ vmaw->config_names[vmaw->config_count] = name_ptr;
1089+ vmaw->config_data[vmaw->config_count] = data_ptr;
1090+
1091+ vmaw->config_count++;
1092+
1093+ return 0;
1094+}
1095+
1096+int vma_writer_register_stream(VmaWriter *vmaw, const char *devname,
1097+ size_t size)
1098+{
1099+ assert(vmaw);
1100+ assert(devname);
1101+ assert(!vmaw->status);
1102+
1103+ if (vmaw->header_written) {
1104+ vma_writer_set_error(vmaw, "vma_writer_register_stream: header "
1105+ "already written");
1106+ return -1;
1107+ }
1108+
1109+ guint n = vmaw->stream_count + 1;
1110+
1111+ /* we can have dev_ids form 1 to 255 (0 reserved)
1112+ * 255(-1) reseverd for safety
1113+ */
1114+ if (n > 254) {
1115+ vma_writer_set_error(vmaw, "vma_writer_register_stream: "
1116+ "too many drives");
1117+ return -1;
1118+ }
1119+
1120+ if (size <= 0) {
1121+ vma_writer_set_error(vmaw, "vma_writer_register_stream: "
1122+ "got strange size %zd", size);
1123+ return -1;
1124+ }
1125+
1126+ DPRINTF("vma_writer_register_stream %s %zu %d\n", devname, size, n);
1127+
1128+ vmaw->stream_info[n].devname = g_strdup(devname);
1129+ vmaw->stream_info[n].size = size;
1130+
1131+ vmaw->stream_info[n].cluster_count = (size + VMA_CLUSTER_SIZE - 1) /
1132+ VMA_CLUSTER_SIZE;
1133+
1134+ vmaw->stream_count = n;
1135+
1136+ if (strcmp(devname, "vmstate") == 0) {
1137+ vmaw->vmstate_stream = n;
1138+ }
1139+
1140+ return n;
1141+}
1142+
1143+static void vma_co_continue_write(void *opaque)
1144+{
1145+ VmaWriter *vmaw = opaque;
1146+
1147+ qemu_aio_set_fd_handler(vmaw->fd, NULL, NULL, NULL, NULL);
1148+
1149+ DPRINTF("vma_co_continue_write\n");
1150+ qemu_coroutine_enter(vmaw->co_writer, NULL);
1151+}
1152+
2dfd543c
DM		 1153+static int vma_co_write_finished(void *opaque)
1154+{
1155+ VmaWriter *vmaw = opaque;
1156+
1157+ return (vmaw->co_writer != 0);
1158+}
1159+
5ad5891c
DM		 1160+static ssize_t coroutine_fn
1161+vma_co_write(VmaWriter *vmaw, const void *buf, size_t bytes)
1162+{
1163+ size_t done = 0;
1164+ ssize_t ret;
1165+
1166+ /* atomic writes (we cannot interleave writes) */
1167+ qemu_co_mutex_lock(&vmaw->writer_lock);
1168+
1169+ DPRINTF("vma_co_write enter %zd\n", bytes);
1170+
2dfd543c
DM		 1171+ assert(vmaw->co_writer == NULL);
1172+
5ad5891c 1173+ while (done < bytes) {
2dfd543c
DM		 1174+ /* Note: we limit maximal write size - else VM gets slow */
1175+ ret = write(vmaw->fd, buf + done,
1176+ (bytes - done) > 4096 ? 4096 : bytes - done);
5ad5891c
DM		 1177+ if (ret > 0) {
1178+ done += ret;
1179+ DPRINTF("vma_co_write written %zd %zd\n", done, ret);
1180+ } else if (ret < 0) {
1181+ if (errno == EAGAIN || errno == EWOULDBLOCK) {
1182+ DPRINTF("vma_co_write yield %zd\n", done);
1183+
1184+ vmaw->co_writer = qemu_coroutine_self();
1185+ qemu_aio_set_fd_handler(vmaw->fd, NULL, vma_co_continue_write,
2dfd543c 1186+ vma_co_write_finished, vmaw);
5ad5891c
DM		 1187+
1188+ qemu_coroutine_yield();
1189+ DPRINTF("vma_co_write restart %zd\n", done);
1190+ } else {
1191+ vma_writer_set_error(vmaw, "vma_co_write write error - %s",
1192+ strerror(errno));
1193+ done = -1; /* always return failure for partial writes */
1194+ break;
1195+ }
1196+ } else if (ret == 0) {
1197+ /* should not happen - simply try again */
1198+ }
1199+ }
1200+
2dfd543c
DM		 1201+ vmaw->co_writer = NULL;
1202+
5ad5891c
DM		 1203+ qemu_co_mutex_unlock(&vmaw->writer_lock);
1204+
1205+ DPRINTF("vma_co_write leave %zd\n", done);
1206+ return done;
1207+}
1208+
1209+static void coroutine_fn vma_co_writer_task(void *opaque)
1210+{
1211+ VmaAIOCB *cb = opaque;
1212+
1213+ DPRINTF("vma_co_writer_task start\n");
1214+
1215+ int64_t done = vma_co_write(cb->vmaw, cb->buffer, cb->bytes);
1216+ DPRINTF("vma_co_writer_task write done %zd\n", done);
1217+
1218+ if (done != cb->bytes) {
1219+ DPRINTF("vma_co_writer_task failed write %zd %zd", cb->bytes, done);
1220+ vma_writer_set_error(cb->vmaw, "vma_co_writer_task failed write %zd",
1221+ done);
1222+ }
1223+
1224+ cb->bytes = 0;
1225+
1226+ qemu_co_queue_next(&cb->vmaw->wqueue);
1227+
1228+ DPRINTF("vma_co_writer_task end\n");
1229+}
1230+
1231+static void coroutine_fn vma_queue_flush(VmaWriter *vmaw)
1232+{
1233+ DPRINTF("vma_queue_flush enter\n");
1234+
1235+ assert(vmaw);
1236+
1237+ while (1) {
1238+ int i;
1239+ VmaAIOCB *cb = NULL;
1240+ for (i = 0; i < WRITE_BUFFERS; i++) {
2dfd543c
DM		 1241+ if (vmaw->aiocbs[i]->bytes) {
1242+ cb = vmaw->aiocbs[i];
5ad5891c 1243+ DPRINTF("FOUND USED AIO BUFFER %d %zd\n", i,
2dfd543c 1244+ vmaw->aiocbs[i]->bytes);
5ad5891c
DM		 1245+ break;
1246+ }
1247+ }
1248+ if (!cb) {
1249+ break;
1250+ }
1251+ qemu_co_queue_wait(&vmaw->wqueue);
1252+ }
1253+
1254+ DPRINTF("vma_queue_flush leave\n");
1255+}
1256+
1257+/**
1258+ * NOTE: pipe buffer size in only 4096 bytes on linux (see 'ulimit -a')
1259+ * So we need to create a coroutione to allow 'parallel' execution.
1260+ */
1261+static ssize_t coroutine_fn
1262+vma_queue_write(VmaWriter *vmaw, const void *buf, size_t bytes)
1263+{
1264+ DPRINTF("vma_queue_write enter %zd\n", bytes);
1265+
1266+ assert(vmaw);
1267+ assert(buf);
1268+ assert(bytes <= VMA_MAX_EXTENT_SIZE);
1269+
1270+ VmaAIOCB *cb = NULL;
1271+ while (!cb) {
1272+ int i;
1273+ for (i = 0; i < WRITE_BUFFERS; i++) {
2dfd543c
DM		 1274+ if (!vmaw->aiocbs[i]->bytes) {
1275+ cb = vmaw->aiocbs[i];
5ad5891c
DM		 1276+ break;
1277+ }
1278+ }
1279+ if (!cb) {
1280+ qemu_co_queue_wait(&vmaw->wqueue);
1281+ }
1282+ }
1283+
1284+ memcpy(cb->buffer, buf, bytes);
1285+ cb->bytes = bytes;
1286+ cb->vmaw = vmaw;
1287+
1288+ DPRINTF("vma_queue_write start %zd\n", bytes);
1289+ cb->co = qemu_coroutine_create(vma_co_writer_task);
1290+ qemu_coroutine_enter(cb->co, cb);
1291+
1292+ DPRINTF("vma_queue_write leave\n");
1293+
1294+ return bytes;
1295+}
1296+
55827521 1297+VmaWriter *vma_writer_create(const char *filename, uuid_t uuid, Error **errp)
5ad5891c
DM		 1298+{
1299+ const char *p;
1300+
1301+ assert(sizeof(VmaHeader) == (4096 + 8192));
1302+ assert(sizeof(VmaExtentHeader) == 512);
1303+
1304+ VmaWriter *vmaw = g_new0(VmaWriter, 1);
1305+ vmaw->fd = -1;
1306+
1307+ vmaw->md5csum = g_checksum_new(G_CHECKSUM_MD5);
1308+ if (!vmaw->md5csum) {
1309+ error_setg(errp, "can't allocate cmsum\n");
1310+ goto err;
1311+ }
1312+
1313+ if (strstart(filename, "exec:", &p)) {
1314+ vmaw->cmd = popen(p, "w");
1315+ if (vmaw->cmd == NULL) {
1316+ error_setg(errp, "can't popen command '%s' - %s\n", p,
1317+ strerror(errno));
1318+ goto err;
1319+ }
1320+ vmaw->fd = fileno(vmaw->cmd);
2dfd543c
DM		 1321+
1322+ /* try to use O_NONBLOCK and O_DIRECT */
1323+ fcntl(vmaw->fd, F_SETFL, fcntl(vmaw->fd, F_GETFL)|O_NONBLOCK);
1324+ fcntl(vmaw->fd, F_SETFL, fcntl(vmaw->fd, F_GETFL)|O_DIRECT);
5ad5891c
DM		 1325+
1326+ } else {
efa8e5de
DM		 1327+ struct stat st;
1328+ int oflags;
309874bd 1329+ const char *tmp_id_str;
3f350e0d 1330+
efa8e5de 1331+ if ((stat(filename, &st) == 0) && S_ISFIFO(st.st_mode)) {
2dfd543c 1332+ oflags = O_NONBLOCK|O_DIRECT|O_WRONLY;
309874bd 1333+ vmaw->fd = qemu_open(filename, oflags, 0644);
3f350e0d 1334+ } else if (strstart(filename, "/dev/fdset/", &tmp_id_str)) {
2dfd543c 1335+ oflags = O_NONBLOCK|O_DIRECT|O_WRONLY;
3f350e0d 1336+ vmaw->fd = qemu_open(filename, oflags, 0644);
309874bd 1337+ } else if (strstart(filename, "/dev/fdname/", &tmp_id_str)) {
3f350e0d
DM		 1338+ vmaw->fd = monitor_get_fd(cur_mon, tmp_id_str, errp);
1339+ if (vmaw->fd < 0) {
1340+ goto err;
1341+ }
2dfd543c
DM		 1342+ /* try to use O_NONBLOCK and O_DIRECT */
1343+ fcntl(vmaw->fd, F_SETFL, fcntl(vmaw->fd, F_GETFL)|O_NONBLOCK);
1344+ fcntl(vmaw->fd, F_SETFL, fcntl(vmaw->fd, F_GETFL)|O_DIRECT);
309874bd 1345+ } else {
2dfd543c 1346+ oflags = O_NONBLOCK|O_DIRECT|O_WRONLY|O_CREAT|O_EXCL;
309874bd 1347+ vmaw->fd = qemu_open(filename, oflags, 0644);
efa8e5de 1348+ }
3f350e0d 1349+
5ad5891c
DM		 1350+ if (vmaw->fd < 0) {
1351+ error_setg(errp, "can't open file %s - %s\n", filename,
1352+ strerror(errno));
1353+ goto err;
1354+ }
1355+ }
1356+
2dfd543c
DM		 1357+ /* we use O_DIRECT, so we need to align IO buffers */
1358+ int i;
1359+ for (i = 0; i < WRITE_BUFFERS; i++) {
1360+ vmaw->aiocbs[i] = qemu_memalign(512, sizeof(VmaAIOCB));
1361+ memset(vmaw->aiocbs[i], 0, sizeof(VmaAIOCB));
1362+ }
1363+
5ad5891c
DM		 1364+ vmaw->outbuf_count = 0;
1365+ vmaw->outbuf_pos = VMA_EXTENT_HEADER_SIZE;
1366+
1367+ vmaw->header_blob_table_pos = 1; /* start at pos 1 */
1368+
1369+ qemu_co_mutex_init(&vmaw->writer_lock);
1370+ qemu_co_mutex_init(&vmaw->flush_lock);
1371+ qemu_co_queue_init(&vmaw->wqueue);
1372+
1373+ uuid_copy(vmaw->uuid, uuid);
1374+
5ad5891c
DM		 1375+ return vmaw;
1376+
1377+err:
1378+ if (vmaw) {
1379+ if (vmaw->cmd) {
1380+ pclose(vmaw->cmd);
1381+ } else if (vmaw->fd >= 0) {
1382+ close(vmaw->fd);
1383+ }
1384+
1385+ if (vmaw->md5csum) {
1386+ g_checksum_free(vmaw->md5csum);
1387+ }
1388+
1389+ g_free(vmaw);
1390+ }
1391+
1392+ return NULL;
1393+}
1394+
1395+static int coroutine_fn vma_write_header(VmaWriter *vmaw)
1396+{
1397+ assert(vmaw);
1398+ int header_clusters = 8;
1399+ char buf[65536*header_clusters];
1400+ VmaHeader *head = (VmaHeader *)buf;
1401+
1402+ int i;
1403+
1404+ DPRINTF("VMA WRITE HEADER\n");
1405+
1406+ if (vmaw->status < 0) {
1407+ return vmaw->status;
1408+ }
1409+
1410+ memset(buf, 0, sizeof(buf));
1411+
1412+ head->magic = VMA_MAGIC;
1413+ head->version = GUINT32_TO_BE(1); /* v1 */
1414+ memcpy(head->uuid, vmaw->uuid, 16);
1415+
1416+ time_t ctime = time(NULL);
1417+ head->ctime = GUINT64_TO_BE(ctime);
1418+
1419+ if (!vmaw->stream_count) {
1420+ return -1;
1421+ }
1422+
1423+ for (i = 0; i < VMA_MAX_CONFIGS; i++) {
1424+ head->config_names[i] = GUINT32_TO_BE(vmaw->config_names[i]);
1425+ head->config_data[i] = GUINT32_TO_BE(vmaw->config_data[i]);
1426+ }
1427+
1428+ /* 32 bytes per device (12 used currently) = 8192 bytes max */
1429+ for (i = 1; i <= 254; i++) {
1430+ VmaStreamInfo *si = &vmaw->stream_info[i];
1431+ if (si->size) {
1432+ assert(si->devname);
1433+ uint32_t devname_ptr = allocate_header_string(vmaw, si->devname);
1434+ if (!devname_ptr) {
1435+ return -1;
1436+ }
1437+ head->dev_info[i].devname_ptr = GUINT32_TO_BE(devname_ptr);
1438+ head->dev_info[i].size = GUINT64_TO_BE(si->size);
1439+ }
1440+ }
1441+
1442+ uint32_t header_size = sizeof(VmaHeader) + vmaw->header_blob_table_size;
1443+ head->header_size = GUINT32_TO_BE(header_size);
1444+
1445+ if (header_size > sizeof(buf)) {
1446+ return -1; /* just to be sure */
1447+ }
1448+
1449+ uint32_t blob_buffer_offset = sizeof(VmaHeader);
1450+ memcpy(buf + blob_buffer_offset, vmaw->header_blob_table,
1451+ vmaw->header_blob_table_size);
1452+ head->blob_buffer_offset = GUINT32_TO_BE(blob_buffer_offset);
1453+ head->blob_buffer_size = GUINT32_TO_BE(vmaw->header_blob_table_pos);
1454+
1455+ g_checksum_reset(vmaw->md5csum);
1456+ g_checksum_update(vmaw->md5csum, (const guchar *)buf, header_size);
1457+ gsize csize = 16;
1458+ g_checksum_get_digest(vmaw->md5csum, (guint8 *)(head->md5sum), &csize);
1459+
1460+ return vma_queue_write(vmaw, buf, header_size);
1461+}
1462+
1463+static int coroutine_fn vma_writer_flush(VmaWriter *vmaw)
1464+{
1465+ assert(vmaw);
1466+
1467+ int ret;
1468+ int i;
1469+
1470+ if (vmaw->status < 0) {
1471+ return vmaw->status;
1472+ }
1473+
1474+ if (!vmaw->header_written) {
1475+ vmaw->header_written = true;
1476+ ret = vma_write_header(vmaw);
1477+ if (ret < 0) {
1478+ vma_writer_set_error(vmaw, "vma_writer_flush: write header failed");
1479+ return ret;
1480+ }
1481+ }
1482+
1483+ DPRINTF("VMA WRITE FLUSH %d %d\n", vmaw->outbuf_count, vmaw->outbuf_pos);
1484+
1485+
1486+ VmaExtentHeader *ehead = (VmaExtentHeader *)vmaw->outbuf;
1487+
1488+ ehead->magic = VMA_EXTENT_MAGIC;
1489+ ehead->reserved1 = 0;
1490+
1491+ for (i = 0; i < VMA_BLOCKS_PER_EXTENT; i++) {
1492+ ehead->blockinfo[i] = GUINT64_TO_BE(vmaw->outbuf_block_info[i]);
1493+ }
1494+
1495+ guint16 block_count = (vmaw->outbuf_pos - VMA_EXTENT_HEADER_SIZE) /
1496+ VMA_BLOCK_SIZE;
1497+
1498+ ehead->block_count = GUINT16_TO_BE(block_count);
1499+
1500+ memcpy(ehead->uuid, vmaw->uuid, sizeof(ehead->uuid));
1501+ memset(ehead->md5sum, 0, sizeof(ehead->md5sum));
1502+
1503+ g_checksum_reset(vmaw->md5csum);
1504+ g_checksum_update(vmaw->md5csum, vmaw->outbuf, VMA_EXTENT_HEADER_SIZE);
1505+ gsize csize = 16;
1506+ g_checksum_get_digest(vmaw->md5csum, ehead->md5sum, &csize);
1507+
1508+ int bytes = vmaw->outbuf_pos;
1509+ ret = vma_queue_write(vmaw, vmaw->outbuf, bytes);
1510+ if (ret != bytes) {
1511+ vma_writer_set_error(vmaw, "vma_writer_flush: failed write");
1512+ }
1513+
1514+ vmaw->outbuf_count = 0;
1515+ vmaw->outbuf_pos = VMA_EXTENT_HEADER_SIZE;
1516+
1517+ for (i = 0; i < VMA_BLOCKS_PER_EXTENT; i++) {
1518+ vmaw->outbuf_block_info[i] = 0;
1519+ }
1520+
1521+ return vmaw->status;
1522+}
1523+
1524+static int vma_count_open_streams(VmaWriter *vmaw)
1525+{
1526+ g_assert(vmaw != NULL);
1527+
1528+ int i;
1529+ int open_drives = 0;
1530+ for (i = 0; i <= 255; i++) {
1531+ if (vmaw->stream_info[i].size && !vmaw->stream_info[i].finished) {
1532+ open_drives++;
1533+ }
1534+ }
1535+
1536+ return open_drives;
1537+}
1538+
1539+/**
1540+ * all jobs should call this when there is no more data
1541+ * Returns: number of remaining stream (0 ==> finished)
1542+ */
1543+int coroutine_fn
1544+vma_writer_close_stream(VmaWriter *vmaw, uint8_t dev_id)
1545+{
1546+ g_assert(vmaw != NULL);
1547+
1548+ DPRINTF("vma_writer_set_status %d\n", dev_id);
1549+ if (!vmaw->stream_info[dev_id].size) {
1550+ vma_writer_set_error(vmaw, "vma_writer_close_stream: "
1551+ "no such stream %d", dev_id);
1552+ return -1;
1553+ }
1554+ if (vmaw->stream_info[dev_id].finished) {
1555+ vma_writer_set_error(vmaw, "vma_writer_close_stream: "
1556+ "stream already closed %d", dev_id);
1557+ return -1;
1558+ }
1559+
1560+ vmaw->stream_info[dev_id].finished = true;
1561+
1562+ int open_drives = vma_count_open_streams(vmaw);
1563+
1564+ if (open_drives <= 0) {
1565+ DPRINTF("vma_writer_set_status all drives completed\n");
1566+ qemu_co_mutex_lock(&vmaw->flush_lock);
1567+ int ret = vma_writer_flush(vmaw);
1568+ qemu_co_mutex_unlock(&vmaw->flush_lock);
1569+ if (ret < 0) {
1570+ vma_writer_set_error(vmaw, "vma_writer_close_stream: flush failed");
1571+ }
1572+ }
1573+
1574+ return open_drives;
1575+}
1576+
1577+int vma_writer_get_status(VmaWriter *vmaw, VmaStatus *status)
1578+{
1579+ int i;
1580+
1581+ g_assert(vmaw != NULL);
1582+
1583+ if (status) {
1584+ status->status = vmaw->status;
1585+ g_strlcpy(status->errmsg, vmaw->errmsg, sizeof(status->errmsg));
1586+ for (i = 0; i <= 255; i++) {
1587+ status->stream_info[i] = vmaw->stream_info[i];
1588+ }
1589+
1590+ uuid_unparse_lower(vmaw->uuid, status->uuid_str);
1591+ }
1592+
1593+ status->closed = vmaw->closed;
1594+
1595+ return vmaw->status;
1596+}
1597+
2dfd543c 1598+static int vma_writer_get_buffer(VmaWriter *vmaw)
5ad5891c 1599+{
309874bd 1600+ int ret = 0;
5ad5891c 1601+
309874bd 1602+ qemu_co_mutex_lock(&vmaw->flush_lock);
5ad5891c 1603+
309874bd
DM		 1604+ /* wait until buffer is available */
1605+ while (vmaw->outbuf_count >= (VMA_BLOCKS_PER_EXTENT - 1)) {
1606+ ret = vma_writer_flush(vmaw);
1607+ if (ret < 0) {
5ad5891c 1608+ vma_writer_set_error(vmaw, "vma_writer_get_buffer: flush failed");
309874bd 1609+ break;
5ad5891c
DM		 1610+ }
1611+ }
1612+
309874bd
DM		 1613+ qemu_co_mutex_unlock(&vmaw->flush_lock);
1614+
1615+ return ret;
5ad5891c
DM		 1616+}
1617+
1618+
1619+int64_t coroutine_fn
1620+vma_writer_write(VmaWriter *vmaw, uint8_t dev_id, int64_t cluster_num,
1621+ unsigned char *buf, size_t *zero_bytes)
1622+{
1623+ g_assert(vmaw != NULL);
1624+ g_assert(zero_bytes != NULL);
1625+
1626+ *zero_bytes = 0;
1627+
1628+ if (vmaw->status < 0) {
1629+ return vmaw->status;
1630+ }
1631+
1632+ if (!dev_id || !vmaw->stream_info[dev_id].size) {
1633+ vma_writer_set_error(vmaw, "vma_writer_write: "
1634+ "no such stream %d", dev_id);
1635+ return -1;
1636+ }
1637+
1638+ if (vmaw->stream_info[dev_id].finished) {
1639+ vma_writer_set_error(vmaw, "vma_writer_write: "
1640+ "stream already closed %d", dev_id);
1641+ return -1;
1642+ }
1643+
1644+
1645+ if (cluster_num >= (((uint64_t)1)<<32)) {
1646+ vma_writer_set_error(vmaw, "vma_writer_write: "
1647+ "cluster number out of range");
1648+ return -1;
1649+ }
1650+
1651+ if (dev_id == vmaw->vmstate_stream) {
1652+ if (cluster_num != vmaw->vmstate_clusters) {
1653+ vma_writer_set_error(vmaw, "vma_writer_write: "
1654+ "non sequential vmstate write");
1655+ }
1656+ vmaw->vmstate_clusters++;
1657+ } else if (cluster_num >= vmaw->stream_info[dev_id].cluster_count) {
1658+ vma_writer_set_error(vmaw, "vma_writer_write: cluster number too big");
1659+ return -1;
1660+ }
1661+
1662+ /* wait until buffer is available */
2dfd543c 1663+ if (vma_writer_get_buffer(vmaw) < 0) {
5ad5891c
DM		 1664+ vma_writer_set_error(vmaw, "vma_writer_write: "
1665+ "vma_writer_get_buffer failed");
1666+ return -1;
1667+ }
1668+
309874bd 1669+ DPRINTF("VMA WRITE %d %zd\n", dev_id, cluster_num);
5ad5891c 1670+
2dfd543c
DM		 1671+ int i;
1672+ int bit = 1;
1673+ uint16_t mask = 0;
5ad5891c
DM		 1674+ for (i = 0; i < 16; i++) {
1675+ unsigned char *vmablock = buf + (i*VMA_BLOCK_SIZE);
2dfd543c
DM		 1676+ if (!buffer_is_zero(vmablock, VMA_BLOCK_SIZE)) {
1677+ mask |= bit;
309874bd
DM		 1678+ memcpy(vmaw->outbuf + vmaw->outbuf_pos, vmablock, VMA_BLOCK_SIZE);
1679+ vmaw->outbuf_pos += VMA_BLOCK_SIZE;
1680+ } else {
5ad5891c
DM		 1681+ DPRINTF("VMA WRITE %zd ZERO BLOCK %d\n", cluster_num, i);
1682+ vmaw->stream_info[dev_id].zero_bytes += VMA_BLOCK_SIZE;
1683+ *zero_bytes += VMA_BLOCK_SIZE;
5ad5891c
DM		 1684+ }
1685+
1686+ bit = bit << 1;
1687+ }
1688+
1689+ uint64_t block_info = ((uint64_t)mask) << (32+16);
1690+ block_info |= ((uint64_t)dev_id) << 32;
1691+ block_info |= (cluster_num & 0xffffffff);
309874bd 1692+ vmaw->outbuf_block_info[vmaw->outbuf_count] = block_info;
5ad5891c
DM		 1693+
1694+ DPRINTF("VMA WRITE MASK %zd %zx\n", cluster_num, block_info);
1695+
1696+ vmaw->outbuf_count++;
1697+
1698+ /** NOTE: We allways write whole clusters, but we correctly set
1699+ * transferred bytes. So transferred == size when when everything
1700+ * went OK.
1701+ */
1702+ size_t transferred = VMA_CLUSTER_SIZE;
1703+
1704+ if (dev_id != vmaw->vmstate_stream) {
1705+ uint64_t last = (cluster_num + 1) * VMA_CLUSTER_SIZE;
1706+ if (last > vmaw->stream_info[dev_id].size) {
1707+ uint64_t diff = last - vmaw->stream_info[dev_id].size;
1708+ if (diff >= VMA_CLUSTER_SIZE) {
1709+ vma_writer_set_error(vmaw, "vma_writer_write: "
1710+ "read after last cluster");
1711+ return -1;
1712+ }
1713+ transferred -= diff;
1714+ }
1715+ }
1716+
1717+ vmaw->stream_info[dev_id].transferred += transferred;
1718+
1719+ return transferred;
1720+}
1721+
1722+int vma_writer_close(VmaWriter *vmaw, Error **errp)
1723+{
1724+ g_assert(vmaw != NULL);
1725+
1726+ int i;
1727+
1728+ vma_queue_flush(vmaw);
1729+
1730+ /* this should not happen - just to be sure */
1731+ while (!qemu_co_queue_empty(&vmaw->wqueue)) {
1732+ DPRINTF("vma_writer_close wait\n");
1733+ co_sleep_ns(rt_clock, 1000000);
1734+ }
1735+
1736+ if (vmaw->cmd) {
1737+ if (pclose(vmaw->cmd) < 0) {
1738+ vma_writer_set_error(vmaw, "vma_writer_close: "
1739+ "pclose failed - %s", strerror(errno));
1740+ }
1741+ } else {
1742+ if (close(vmaw->fd) < 0) {
1743+ vma_writer_set_error(vmaw, "vma_writer_close: "
1744+ "close failed - %s", strerror(errno));
1745+ }
1746+ }
1747+
1748+ for (i = 0; i <= 255; i++) {
1749+ VmaStreamInfo *si = &vmaw->stream_info[i];
1750+ if (si->size) {
1751+ if (!si->finished) {
1752+ vma_writer_set_error(vmaw, "vma_writer_close: "
1753+ "detected open stream '%s'", si->devname);
1754+ } else if ((si->transferred != si->size) &&
1755+ (i != vmaw->vmstate_stream)) {
1756+ vma_writer_set_error(vmaw, "vma_writer_close: "
1757+ "incomplete stream '%s' (%zd != %zd)",
1758+ si->devname, si->transferred, si->size);
1759+ }
1760+ }
1761+ }
1762+
1763+ for (i = 0; i <= 255; i++) {
1764+ vmaw->stream_info[i].finished = 1; /* mark as closed */
1765+ }
1766+
1767+ vmaw->closed = 1;
1768+
1769+ if (vmaw->status < 0 && *errp == NULL) {
1770+ error_setg(errp, "%s", vmaw->errmsg);
1771+ }
1772+
1773+ return vmaw->status;
1774+}
1775+
1776+void vma_writer_destroy(VmaWriter *vmaw)
1777+{
1778+ assert(vmaw);
1779+
1780+ int i;
1781+
1782+ for (i = 0; i <= 255; i++) {
1783+ if (vmaw->stream_info[i].devname) {
1784+ g_free(vmaw->stream_info[i].devname);
1785+ }
1786+ }
1787+
1788+ if (vmaw->md5csum) {
1789+ g_checksum_free(vmaw->md5csum);
1790+ }
1791+
2dfd543c
DM		 1792+ for (i = 0; i < WRITE_BUFFERS; i++) {
1793+ free(vmaw->aiocbs[i]);
1794+ }
1795+
5ad5891c
DM		 1796+ g_free(vmaw);
1797+}
1798+
1799+/* backup driver plugin */
1800+
1801+static int vma_dump_cb(void *opaque, uint8_t dev_id, int64_t cluster_num,
1802+ unsigned char *buf, size_t *zero_bytes)
1803+{
1804+ VmaWriter *vmaw = opaque;
1805+
1806+ return vma_writer_write(vmaw, dev_id, cluster_num, buf, zero_bytes);
1807+}
1808+
1809+static int vma_close_cb(void *opaque, Error **errp)
1810+{
1811+ VmaWriter *vmaw = opaque;
1812+
1813+ int res = vma_writer_close(vmaw, errp);
1814+ vma_writer_destroy(vmaw);
1815+
1816+ return res;
1817+}
1818+
1819+static int vma_complete_cb(void *opaque, uint8_t dev_id, int ret)
1820+{
1821+ VmaWriter *vmaw = opaque;
1822+
1823+ if (ret < 0) {
1824+ vma_writer_set_error(vmaw, "backup_complete_cb %d", ret);
1825+ }
1826+
1827+ return vma_writer_close_stream(vmaw, dev_id);
1828+}
1829+
1830+static int vma_register_stream_cb(void *opaque, const char *devname,
1831+ size_t size)
1832+{
1833+ VmaWriter *vmaw = opaque;
1834+
1835+ return vma_writer_register_stream(vmaw, devname, size);
1836+}
1837+
1838+static int vma_register_config_cb(void *opaque, const char *name,
1839+ gpointer data, size_t data_len)
1840+{
1841+ VmaWriter *vmaw = opaque;
1842+
1843+ return vma_writer_add_config(vmaw, name, data, data_len);
1844+}
1845+
55827521 1846+static void *vma_open_cb(const char *filename, uuid_t uuid, Error **errp)
5ad5891c 1847+{
55827521 1848+ return vma_writer_create(filename, uuid, errp);
5ad5891c
DM		 1849+}
1850+
1851+const BackupDriver backup_vma_driver = {
1852+ .format = "vma",
1853+ .open_cb = vma_open_cb,
1854+ .close_cb = vma_close_cb,
1855+ .register_config_cb = vma_register_config_cb,
1856+ .register_stream_cb = vma_register_stream_cb,
1857+ .dump_cb = vma_dump_cb,
1858+ .complete_cb = vma_complete_cb,
1859+};
1860+
1861diff --git a/vma.c b/vma.c
1862new file mode 100644
55827521 1863index 0000000..5d13906
5ad5891c
DM		 1864--- /dev/null
1865+++ b/vma.c
309874bd 1866@@ -0,0 +1,561 @@
5ad5891c
DM		 1867+/*
1868+ * VMA: Virtual Machine Archive
1869+ *
1870+ * Copyright (C) 2012 Proxmox Server Solutions
1871+ *
1872+ * Authors:
1873+ * Dietmar Maurer (dietmar@proxmox.com)
1874+ *
1875+ * This work is licensed under the terms of the GNU GPL, version 2 or later.
1876+ * See the COPYING file in the top-level directory.
1877+ *
1878+ */
1879+
1880+#include <stdio.h>
1881+#include <errno.h>
1882+#include <unistd.h>
1883+#include <stdio.h>
1884+#include <string.h>
1885+#include <sys/types.h>
1886+#include <sys/stat.h>
1887+#include <fcntl.h>
1888+#include <glib.h>
1889+
1890+#include "qemu-common.h"
1891+#include "qemu-option.h"
1892+#include "qemu-error.h"
1893+#include "osdep.h"
1894+#include "sysemu.h"
1895+#include "block_int.h"
1896+#include <stdio.h>
1897+#include "vma.h"
1898+
1899+static void help(void)
1900+{
1901+ const char *help_msg =
1902+ "usage: vma command [command options]\n"
1903+ "\n"
1904+ "vma list <filename>\n"
1905+ "vma create <filename> [-c config] <archive> pathname ...\n"
1906+ "vma extract <filename> [-r] <targetdir>\n"
1907+ ;
1908+
1909+ printf("%s", help_msg);
1910+ exit(1);
1911+}
1912+
1913+static const char *extract_devname(const char *path, char **devname, int index)
1914+{
1915+ assert(path);
1916+
1917+ const char *sep = strchr(path, '=');
1918+
1919+ if (sep) {
1920+ *devname = g_strndup(path, sep - path);
1921+ path = sep + 1;
1922+ } else {
1923+ if (index >= 0) {
1924+ *devname = g_strdup_printf("disk%d", index);
1925+ } else {
1926+ *devname = NULL;
1927+ }
1928+ }
1929+
1930+ return path;
1931+}
1932+
1933+static void print_content(VmaReader *vmar)
1934+{
1935+ assert(vmar);
1936+
1937+ VmaHeader *head = vma_reader_get_header(vmar);
1938+
5ad5891c
DM		 1939+ GList *l = vma_reader_get_config_data(vmar);
1940+ while (l && l->data) {
1941+ VmaConfigData *cdata = (VmaConfigData *)l->data;
1942+ l = g_list_next(l);
1943+ printf("CFG: size: %d name: %s\n", cdata->len, cdata->name);
1944+ }
1945+
1946+ int i;
1947+ VmaDeviceInfo *di;
1948+ for (i = 1; i < 255; i++) {
1949+ di = vma_reader_get_device_info(vmar, i);
1950+ if (di) {
1951+ if (strcmp(di->devname, "vmstate") == 0) {
1952+ printf("VMSTATE: dev_id=%d memory: %zd\n", i, di->size);
1953+ } else {
1954+ printf("DEV: dev_id=%d size: %zd devname: %s\n",
1955+ i, di->size, di->devname);
1956+ }
1957+ }
1958+ }
d95c8c31
DM		 1959+ /* ctime is the last entry we print */
1960+ printf("CTIME: %s", ctime(&head->ctime));
1961+ fflush(stdout);
5ad5891c
DM		 1962+}
1963+
1964+static int list_content(int argc, char **argv)
1965+{
1966+ int c, ret = 0;
1967+ const char *filename;
1968+
1969+ for (;;) {
1970+ c = getopt(argc, argv, "h");
1971+ if (c == -1) {
1972+ break;
1973+ }
1974+ switch (c) {
1975+ case '?':
1976+ case 'h':
1977+ help();
1978+ break;
1979+ default:
1980+ g_assert_not_reached();
1981+ }
1982+ }
1983+
1984+ /* Get the filename */
1985+ if ((optind + 1) != argc) {
1986+ help();
1987+ }
1988+ filename = argv[optind++];
1989+
1990+ Error *errp = NULL;
1991+ VmaReader *vmar = vma_reader_create(filename, &errp);
1992+
1993+ if (!vmar) {
1994+ g_error("%s", error_get_pretty(errp));
1995+ }
1996+
1997+ print_content(vmar);
1998+
1999+ vma_reader_destroy(vmar);
2000+
2001+ return ret;
2002+}
2003+
2004+typedef struct RestoreMap {
2005+ char *devname;
2006+ char *path;
2007+ bool write_zero;
2008+} RestoreMap;
2009+
2010+static int extract_content(int argc, char **argv)
2011+{
2012+ int c, ret = 0;
4244016d 2013+ int verbose = 0;
5ad5891c
DM		 2014+ const char *filename;
2015+ const char *dirname;
3f350e0d 2016+ const char *readmap = NULL;
5ad5891c
DM		 2017+
2018+ for (;;) {
4244016d 2019+ c = getopt(argc, argv, "hvr:");
5ad5891c
DM		 2020+ if (c == -1) {
2021+ break;
2022+ }
2023+ switch (c) {
2024+ case '?':
2025+ case 'h':
2026+ help();
2027+ break;
2028+ case 'r':
3f350e0d 2029+ readmap = optarg;
5ad5891c 2030+ break;
309874bd 2031+ case 'v':
4244016d
DM		 2032+ verbose = 1;
2033+ break;
5ad5891c
DM		 2034+ default:
2035+ help();
2036+ }
2037+ }
2038+
2039+ /* Get the filename */
2040+ if ((optind + 2) != argc) {
2041+ help();
2042+ }
2043+ filename = argv[optind++];
2044+ dirname = argv[optind++];
2045+
2046+ Error *errp = NULL;
2047+ VmaReader *vmar = vma_reader_create(filename, &errp);
2048+
2049+ if (!vmar) {
2050+ g_error("%s", error_get_pretty(errp));
2051+ }
2052+
2053+ if (mkdir(dirname, 0777) < 0) {
2054+ g_error("unable to create target directory %s - %s",
2055+ dirname, strerror(errno));
2056+ }
2057+
2058+ GList *l = vma_reader_get_config_data(vmar);
2059+ while (l && l->data) {
2060+ VmaConfigData *cdata = (VmaConfigData *)l->data;
2061+ l = g_list_next(l);
2062+ char *cfgfn = g_strdup_printf("%s/%s", dirname, cdata->name);
2063+ GError *err = NULL;
2064+ if (!g_file_set_contents(cfgfn, (gchar *)cdata->data, cdata->len,
2065+ &err)) {
3f350e0d 2066+ g_error("unable to write file: %s", err->message);
5ad5891c
DM		 2067+ }
2068+ }
2069+
2070+ GHashTable *devmap = g_hash_table_new(g_str_hash, g_str_equal);
2071+
2072+ if (readmap) {
2073+ print_content(vmar);
2074+
3f350e0d
DM		 2075+ FILE *map = fopen(readmap, "r");
2076+ if (!map) {
2077+ g_error("unable to open fifo %s - %s", readmap, strerror(errno));
2078+ }
2079+
5ad5891c
DM		 2080+ while (1) {
2081+ char inbuf[8192];
3f350e0d 2082+ char *line = fgets(inbuf, sizeof(inbuf), map);
5ad5891c
DM		 2083+ if (!line || line[0] == '\0' || !strcmp(line, "done\n")) {
2084+ break;
2085+ }
2086+ int len = strlen(line);
2087+ if (line[len - 1] == '\n') {
2088+ line[len - 1] = '\0';
2089+ if (len == 1) {
2090+ break;
2091+ }
2092+ }
2093+
2094+ const char *path;
2095+ bool write_zero;
2096+ if (line[0] == '0' && line[1] == ':') {
2097+ path = inbuf + 2;
2098+ write_zero = false;
2099+ } else if (line[0] == '1' && line[1] == ':') {
2100+ path = inbuf + 2;
2101+ write_zero = true;
3f350e0d 2102+ } else {
5ad5891c
DM		 2103+ g_error("read map failed - parse error ('%s')", inbuf);
2104+ }
2105+
2106+ char *devname = NULL;
2107+ path = extract_devname(path, &devname, -1);
2108+ if (!devname) {
2109+ g_error("read map failed - no dev name specified ('%s')",
2110+ inbuf);
2111+ }
2112+
5ad5891c
DM		 2113+ RestoreMap *map = g_new0(RestoreMap, 1);
2114+ map->devname = g_strdup(devname);
2115+ map->path = g_strdup(path);
2116+ map->write_zero = write_zero;
2117+
2118+ g_hash_table_insert(devmap, map->devname, map);
2119+
2120+ };
2121+ }
2122+
2123+ int i;
2124+ int vmstate_fd = -1;
2125+ guint8 vmstate_stream = 0;
2126+
2127+ for (i = 1; i < 255; i++) {
2128+ VmaDeviceInfo *di = vma_reader_get_device_info(vmar, i);
2129+ if (di && (strcmp(di->devname, "vmstate") == 0)) {
2130+ vmstate_stream = i;
2131+ char *statefn = g_strdup_printf("%s/vmstate.bin", dirname);
2132+ vmstate_fd = open(statefn, O_WRONLY|O_CREAT|O_EXCL, 0644);
2133+ if (vmstate_fd < 0) {
2134+ g_error("create vmstate file '%s' failed - %s", statefn,
2135+ strerror(errno));
2136+ }
2137+ g_free(statefn);
2138+ } else if (di) {
2139+ char *devfn = NULL;
2140+ int flags = BDRV_O_RDWR|BDRV_O_CACHE_WB;
2141+ bool write_zero = true;
2142+
2143+ if (readmap) {
2144+ RestoreMap *map;
2145+ map = (RestoreMap *)g_hash_table_lookup(devmap, di->devname);
2146+ if (map == NULL) {
2147+ g_error("no device name mapping for %s", di->devname);
2148+ }
2149+ devfn = map->path;
2150+ write_zero = map->write_zero;
2151+ } else {
2152+ devfn = g_strdup_printf("%s/tmp-disk-%s.raw",
2153+ dirname, di->devname);
2154+ printf("DEVINFO %s %zd\n", devfn, di->size);
2155+
2156+ if (bdrv_img_create(devfn, "raw", NULL, NULL, NULL,
2157+ di->size, flags)) {
2158+ g_error("can't create file %s", devfn);
2159+ }
2160+
2161+ /* Note: we created an empty file above, so there is no
2162+ * need to write zeroes (so we generate a sparse file)
2163+ */
2164+ write_zero = false;
2165+ }
2166+
9de54cd3
DM		 2167+ BlockDriverState *bs = bdrv_new(di->devname);
2168+ if (bdrv_open(bs, devfn, flags, NULL)) {
5ad5891c
DM		 2169+ g_error("can't open file %s", devfn);
2170+ }
2171+ if (vma_reader_register_bs(vmar, i, bs, write_zero, &errp) < 0) {
2172+ g_error("%s", error_get_pretty(errp));
2173+ }
2174+
2175+ if (!readmap) {
2176+ g_free(devfn);
2177+ }
2178+ }
2179+ }
2180+
4244016d 2181+ if (vma_reader_restore(vmar, vmstate_fd, verbose, &errp) < 0) {
5ad5891c
DM		 2182+ g_error("restore failed - %s", error_get_pretty(errp));
2183+ }
2184+
2185+ if (!readmap) {
2186+ for (i = 1; i < 255; i++) {
2187+ VmaDeviceInfo *di = vma_reader_get_device_info(vmar, i);
2188+ if (di && (i != vmstate_stream)) {
2189+ char *tmpfn = g_strdup_printf("%s/tmp-disk-%s.raw",
2190+ dirname, di->devname);
2191+ char *fn = g_strdup_printf("%s/disk-%s.raw",
2192+ dirname, di->devname);
2193+ if (rename(tmpfn, fn) != 0) {
2194+ g_error("rename %s to %s failed - %s",
2195+ tmpfn, fn, strerror(errno));
2196+ }
2197+ }
2198+ }
2199+ }
2200+
2201+ vma_reader_destroy(vmar);
2202+
2203+ bdrv_close_all();
2204+
2205+ return ret;
2206+}
2207+
2208+typedef struct BackupCB {
2209+ VmaWriter *vmaw;
2210+ uint8_t dev_id;
2211+} BackupCB;
2212+
2213+static int backup_dump_cb(void *opaque, BlockDriverState *bs,
2214+ int64_t cluster_num, unsigned char *buf)
2215+{
2216+ BackupCB *bcb = opaque;
2217+ size_t zb = 0;
2218+ if (vma_writer_write(bcb->vmaw, bcb->dev_id, cluster_num, buf, &zb) < 0) {
2219+ g_warning("backup_dump_cb vma_writer_write failed");
2220+ return -1;
2221+ }
2222+
2223+ return 0;
2224+}
2225+
2226+static void backup_complete_cb(void *opaque, int ret)
2227+{
2228+ BackupCB *bcb = opaque;
2229+
2230+ if (ret < 0) {
2231+ vma_writer_set_error(bcb->vmaw, "backup_complete_cb %d", ret);
2232+ }
2233+
2234+ if (vma_writer_close_stream(bcb->vmaw, bcb->dev_id) <= 0) {
2235+ Error *err = NULL;
2236+ if (vma_writer_close(bcb->vmaw, &err) != 0) {
2237+ g_warning("vma_writer_close failed %s", error_get_pretty(err));
2238+ }
2239+ }
2240+}
2241+
2242+static int create_archive(int argc, char **argv)
2243+{
2244+ int i, c, res;
2245+ int verbose = 0;
2246+ const char *archivename;
2247+ GList *config_files = NULL;
2248+
2249+ for (;;) {
2250+ c = getopt(argc, argv, "hvc:");
2251+ if (c == -1) {
2252+ break;
2253+ }
2254+ switch (c) {
2255+ case '?':
2256+ case 'h':
2257+ help();
2258+ break;
2259+ case 'c':
2260+ config_files = g_list_append(config_files, optarg);
2261+ break;
2262+ case 'v':
2263+ verbose = 1;
2264+ break;
2265+ default:
2266+ g_assert_not_reached();
2267+ }
2268+ }
2269+
2270+
2271+ /* make sure we have archive name and at least one path */
2272+ if ((optind + 2) > argc) {
2273+ help();
2274+ }
2275+
2276+ archivename = argv[optind++];
2277+
2278+ uuid_t uuid;
2279+ uuid_generate(uuid);
2280+
2281+ Error *local_err = NULL;
55827521 2282+ VmaWriter *vmaw = vma_writer_create(archivename, uuid, &local_err);
5ad5891c
DM		 2283+
2284+ if (vmaw == NULL) {
2285+ g_error("%s", error_get_pretty(local_err));
2286+ }
2287+
2288+ GList *l = config_files;
2289+ while (l && l->data) {
2290+ char *name = l->data;
2291+ char *cdata = NULL;
2292+ gsize clen = 0;
2293+ GError *err = NULL;
2294+ if (!g_file_get_contents(name, &cdata, &clen, &err)) {
2295+ unlink(archivename);
2296+ g_error("Unable to read file: %s", err->message);
2297+ }
2298+
2299+ if (vma_writer_add_config(vmaw, name, cdata, clen) != 0) {
2300+ unlink(archivename);
2301+ g_error("Unable to append config data %s (len = %zd)",
2302+ name, clen);
2303+ }
2304+ l = g_list_next(l);
2305+ }
2306+
2307+ int ind = 0;
2308+ while (optind < argc) {
2309+ const char *path = argv[optind++];
2310+ char *devname = NULL;
2311+ path = extract_devname(path, &devname, ind++);
2312+
2313+ BlockDriver *drv = NULL;
2314+ BlockDriverState *bs = bdrv_new(devname);
2315+
2316+ res = bdrv_open(bs, path, BDRV_O_CACHE_WB , drv);
2317+ if (res < 0) {
2318+ unlink(archivename);
2319+ g_error("bdrv_open '%s' failed", path);
2320+ }
2321+ int64_t size = bdrv_getlength(bs);
2322+ int dev_id = vma_writer_register_stream(vmaw, devname, size);
2323+ if (dev_id <= 0) {
2324+ unlink(archivename);
2325+ g_error("vma_writer_register_stream '%s' failed", devname);
2326+ }
2327+
2328+ BackupCB *bcb = g_new0(BackupCB, 1);
2329+ bcb->vmaw = vmaw;
2330+ bcb->dev_id = dev_id;
2331+
309874bd 2332+ if (backup_job_create(bs, backup_dump_cb, backup_complete_cb,
55827521 2333+ bcb, 0) < 0) {
5ad5891c
DM		 2334+ unlink(archivename);
2335+ g_error("backup_job_start failed");
309874bd
DM		 2336+ } else {
2337+ backup_job_start(bs);
5ad5891c
DM		 2338+ }
2339+ }
2340+
2341+ VmaStatus vmastat;
2342+ int percent = 0;
2343+ int last_percent = -1;
2344+
2345+ while (1) {
2346+ main_loop_wait(false);
2347+ vma_writer_get_status(vmaw, &vmastat);
2348+
2349+ if (verbose) {
2350+
2351+ uint64_t total = 0;
2352+ uint64_t transferred = 0;
2353+ uint64_t zero_bytes = 0;
2354+
2355+ int i;
2356+ for (i = 0; i < 256; i++) {
2357+ if (vmastat.stream_info[i].size) {
2358+ total += vmastat.stream_info[i].size;
2359+ transferred += vmastat.stream_info[i].transferred;
2360+ zero_bytes += vmastat.stream_info[i].zero_bytes;
2361+ }
2362+ }
2363+ percent = (transferred*100)/total;
2364+ if (percent != last_percent) {
2365+ printf("progress %d%% %zd/%zd %zd\n", percent,
2366+ transferred, total, zero_bytes);
2367+
2368+ last_percent = percent;
2369+ }
2370+ }
2371+
2372+ if (vmastat.closed) {
2373+ break;
2374+ }
2375+ }
2376+
2377+ bdrv_drain_all();
2378+
2379+ vma_writer_get_status(vmaw, &vmastat);
2380+
2381+ if (verbose) {
2382+ for (i = 0; i < 256; i++) {
2383+ VmaStreamInfo *si = &vmastat.stream_info[i];
2384+ if (si->size) {
2385+ printf("image %s: size=%zd zeros=%zd saved=%zd\n", si->devname,
2386+ si->size, si->zero_bytes, si->size - si->zero_bytes);
2387+ }
2388+ }
2389+ }
2390+
2391+ if (vmastat.status < 0) {
2392+ unlink(archivename);
2393+ g_error("creating vma archive failed");
2394+ }
2395+
2396+ return 0;
2397+}
2398+
2399+int main(int argc, char **argv)
2400+{
2401+ const char *cmdname;
2402+
2403+ error_set_progname(argv[0]);
2404+
2405+ qemu_init_main_loop();
2406+
2407+ bdrv_init();
2408+
2409+ if (argc < 2) {
2410+ help();
2411+ }
2412+
2413+ cmdname = argv[1];
2414+ argc--; argv++;
2415+
2416+
2417+ if (!strcmp(cmdname, "list")) {
2418+ return list_content(argc, argv);
2419+ } else if (!strcmp(cmdname, "create")) {
2420+ return create_archive(argc, argv);
2421+ } else if (!strcmp(cmdname, "extract")) {
2422+ return extract_content(argc, argv);
2423+ }
2424+
2425+ help();
2426+ return 0;
2427+}
2428diff --git a/vma.h b/vma.h
2429new file mode 100644
55827521 2430index 0000000..76d0dc8
5ad5891c
DM		 2431--- /dev/null
2432+++ b/vma.h
55827521 2433@@ -0,0 +1,145 @@
5ad5891c
DM		 2434+/*
2435+ * VMA: Virtual Machine Archive
2436+ *
2437+ * Copyright (C) Proxmox Server Solutions
2438+ *
2439+ * Authors:
2440+ * Dietmar Maurer (dietmar@proxmox.com)
2441+ *
2442+ * This work is licensed under the terms of the GNU GPL, version 2 or later.
2443+ * See the COPYING file in the top-level directory.
2444+ *
2445+ */
2446+
2447+#ifndef BACKUP_VMA_H
2448+#define BACKUP_VMA_H
2449+
2450+#include "backup.h"
2451+#include "error.h"
2452+
2453+#define VMA_BLOCK_BITS 12
2454+#define VMA_BLOCK_SIZE (1<<VMA_BLOCK_BITS)
2455+#define VMA_CLUSTER_BITS (VMA_BLOCK_BITS+4)
2456+#define VMA_CLUSTER_SIZE (1<<VMA_CLUSTER_BITS)
2457+
2458+#if VMA_CLUSTER_SIZE != 65536
2459+#error unexpected cluster size
2460+#endif
2461+
2462+#define VMA_EXTENT_HEADER_SIZE 512
2463+#define VMA_BLOCKS_PER_EXTENT 59
2464+#define VMA_MAX_CONFIGS 256
2465+
2466+#define VMA_MAX_EXTENT_SIZE \
2467+ (VMA_EXTENT_HEADER_SIZE+VMA_CLUSTER_SIZE*VMA_BLOCKS_PER_EXTENT)
2468+#if VMA_MAX_EXTENT_SIZE != 3867136
2469+#error unexpected VMA_EXTENT_SIZE
2470+#endif
2471+
2472+/* File Format Definitions */
2473+
2474+#define VMA_MAGIC (GUINT32_TO_BE(('V'<<24)|('M'<<16)|('A'<<8)|0x00))
2475+#define VMA_EXTENT_MAGIC (GUINT32_TO_BE(('V'<<24)|('M'<<16)|('A'<<8)|'E'))
2476+
2477+typedef struct VmaDeviceInfoHeader {
2478+ uint32_t devname_ptr; /* offset into blob_buffer table */
2479+ uint32_t reserved0;
2480+ uint64_t size; /* device size in bytes */
2481+ uint64_t reserved1;
2482+ uint64_t reserved2;
2483+} VmaDeviceInfoHeader;
2484+
2485+typedef struct VmaHeader {
2486+ uint32_t magic;
2487+ uint32_t version;
2488+ unsigned char uuid[16];
2489+ int64_t ctime;
2490+ unsigned char md5sum[16];
2491+
2492+ uint32_t blob_buffer_offset;
2493+ uint32_t blob_buffer_size;
2494+ uint32_t header_size;
2495+
2496+ unsigned char reserved[1984];
2497+
2498+ uint32_t config_names[VMA_MAX_CONFIGS]; /* offset into blob_buffer table */
2499+ uint32_t config_data[VMA_MAX_CONFIGS]; /* offset into blob_buffer table */
2500+
2501+ VmaDeviceInfoHeader dev_info[256];
2502+} VmaHeader;
2503+
2504+typedef struct VmaExtentHeader {
2505+ uint32_t magic;
2506+ uint16_t reserved1;
2507+ uint16_t block_count;
2508+ unsigned char uuid[16];
2509+ unsigned char md5sum[16];
2510+ uint64_t blockinfo[VMA_BLOCKS_PER_EXTENT];
2511+} VmaExtentHeader;
2512+
2513+/* functions/definitions to read/write vma files */
2514+
2515+typedef struct VmaReader VmaReader;
2516+
2517+typedef struct VmaWriter VmaWriter;
2518+
2519+typedef struct VmaConfigData {
2520+ const char *name;
2521+ const void *data;
2522+ uint32_t len;
2523+} VmaConfigData;
2524+
2525+typedef struct VmaStreamInfo {
2526+ uint64_t size;
2527+ uint64_t cluster_count;
2528+ uint64_t transferred;
2529+ uint64_t zero_bytes;
2530+ int finished;
2531+ char *devname;
2532+} VmaStreamInfo;
2533+
2534+typedef struct VmaStatus {
2535+ int status;
2536+ bool closed;
2537+ char errmsg[8192];
2538+ char uuid_str[37];
2539+ VmaStreamInfo stream_info[256];
2540+} VmaStatus;
2541+
2542+typedef struct VmaDeviceInfo {
2543+ uint64_t size; /* device size in bytes */
2544+ const char *devname;
2545+} VmaDeviceInfo;
2546+
2547+extern const BackupDriver backup_vma_driver;
2548+
55827521 2549+VmaWriter *vma_writer_create(const char *filename, uuid_t uuid, Error **errp);
5ad5891c
DM		 2550+int vma_writer_close(VmaWriter *vmaw, Error **errp);
2551+void vma_writer_destroy(VmaWriter *vmaw);
2552+int vma_writer_add_config(VmaWriter *vmaw, const char *name, gpointer data,
2553+ size_t len);
2554+int vma_writer_register_stream(VmaWriter *vmaw, const char *devname,
2555+ size_t size);
2556+
2557+int64_t coroutine_fn vma_writer_write(VmaWriter *vmaw, uint8_t dev_id,
2558+ int64_t cluster_num, unsigned char *buf,
2559+ size_t *zero_bytes);
2560+
2561+int coroutine_fn vma_writer_close_stream(VmaWriter *vmaw, uint8_t dev_id);
2562+
2563+int vma_writer_get_status(VmaWriter *vmaw, VmaStatus *status);
2564+void vma_writer_set_error(VmaWriter *vmaw, const char *fmt, ...);
2565+
2566+
2567+VmaReader *vma_reader_create(const char *filename, Error **errp);
2568+void vma_reader_destroy(VmaReader *vmar);
2569+VmaHeader *vma_reader_get_header(VmaReader *vmar);
2570+GList *vma_reader_get_config_data(VmaReader *vmar);
2571+VmaDeviceInfo *vma_reader_get_device_info(VmaReader *vmar, guint8 dev_id);
2572+int vma_reader_register_bs(VmaReader *vmar, guint8 dev_id,
2573+ BlockDriverState *bs, bool write_zeroes,
2574+ Error **errp);
309874bd
DM		 2575+int vma_reader_restore(VmaReader *vmar, int vmstate_fd, bool verbose,
2576+ Error **errp);
5ad5891c
DM		 2577+
2578+#endif /* BACKUP_VMA_H */
2579--
25801.7.2.5
2581
old QEMU for PVE