]> git.proxmox.com Git - pve-qemu-kvm.git/blame - debian/patches/backup-add-vma-binary.patch
projects / pve-qemu-kvm.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
add vma-writer.o to block-obj-y
[pve-qemu-kvm.git] / debian / patches / backup-add-vma-binary.patch
CommitLineData
b668dc80
DM		 1From 6289a43696ca6f713a5d3bb9f95a5adb608a5e13 Mon Sep 17 00:00:00 2001
2From: Dietmar Maurer <dietmar@proxmox.com>
3Date: Tue, 13 Nov 2012 11:11:38 +0100
4Subject: [PATCH v5 4/6] introduce new vma archive format
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 ++
14 vma-reader.c | 799 ++++++++++++++++++++++++++++++++++++++++
15 vma-writer.c | 940 +++++++++++++++++++++++++++++++++++++++++++++++
16 vma.c | 561 ++++++++++++++++++++++++++++
17 vma.h | 145 ++++++++
18 8 files changed, 2476 insertions(+), 4 deletions(-)
19 create mode 100644 docs/specs/vma_spec.txt
20 create mode 100644 vma-reader.c
21 create mode 100644 vma-writer.c
22 create mode 100644 vma.c
23 create mode 100644 vma.h
24
25Index: new/Makefile
26===================================================================
09a78de1 27--- new.orig/Makefile 2013-11-29 08:45:24.000000000 +0100
2014bc7f 28+++ new/Makefile 2013-11-29 12:25:23.000000000 +0100
b668dc80
DM		 29@@ -128,7 +128,7 @@
30 include $(SRC_PATH)/libcacard/Makefile
31 endif
32
33-all: $(DOCS) $(TOOLS) $(HELPERS-y) recurse-all
34+all: $(DOCS) $(TOOLS) vma$(EXESUF) $(HELPERS-y) recurse-all
35
36 config-host.h: config-host.h-timestamp
37 config-host.h-timestamp: config-host.mak
38@@ -194,6 +194,7 @@
39 qemu-img$(EXESUF): qemu-img.o $(block-obj-y) libqemuutil.a libqemustub.a
40 qemu-nbd$(EXESUF): qemu-nbd.o $(block-obj-y) libqemuutil.a libqemustub.a
41 qemu-io$(EXESUF): qemu-io.o $(block-obj-y) libqemuutil.a libqemustub.a
2014bc7f 42+vma$(EXESUF): vma.o vma-reader.o $(block-obj-y) libqemuutil.a libqemustub.a
b668dc80
DM		 43
44 qemu-bridge-helper$(EXESUF): qemu-bridge-helper.o
45
46Index: new/docs/specs/vma_spec.txt
47===================================================================
48--- /dev/null 1970-01-01 00:00:00.000000000 +0000
09a78de1 49+++ new/docs/specs/vma_spec.txt 2013-11-29 10:08:41.000000000 +0100
b668dc80
DM		 50@@ -0,0 +1,24 @@
51+=Virtual Machine Archive format (VMA)=
52+
53+This format contains a header which includes the VM configuration as
54+binary blobs, and a list of devices (dev_id, name).
55+
56+The actual VM image data is stored inside extents. An extent contains
57+up to 64 clusters, and start with a 512 byte header containing
58+additional information for those clusters.
59+
60+We use a cluster size of 65536, and use 8 bytes for each
61+cluster in the header to store the following information:
62+
63+* 1 byte dev_id (to identity the drive)
64+* 1 byte not used (reserved)
65+* 2 bytes zero indicator (mark zero regions (16x4096))
66+* 4 bytes cluster number
67+
68+We only store non-zero blocks (such block is 4096 bytes).
69+
70+Each archive is marked with a uuid. The archive header and all
71+extent headers includes that uuid and a MD5 checksum (over header
72+data).
73+
74+
75Index: new/vma-reader.c
76===================================================================
77--- /dev/null 1970-01-01 00:00:00.000000000 +0000
2014bc7f 78+++ new/vma-reader.c 2013-11-29 12:10:53.000000000 +0100
b668dc80
DM		 79@@ -0,0 +1,799 @@
80+/*
81+ * VMA: Virtual Machine Archive
82+ *
83+ * Copyright (C) 2012 Proxmox Server Solutions
84+ *
85+ * Authors:
86+ * Dietmar Maurer (dietmar@proxmox.com)
87+ *
88+ * This work is licensed under the terms of the GNU GPL, version 2 or later.
89+ * See the COPYING file in the top-level directory.
90+ *
91+ */
92+
93+#include <stdio.h>
94+#include <errno.h>
95+#include <unistd.h>
96+#include <stdio.h>
97+#include <string.h>
98+#include <sys/types.h>
99+#include <sys/stat.h>
100+#include <fcntl.h>
101+#include <glib.h>
102+#include <uuid/uuid.h>
103+
104+#include "qemu-common.h"
105+#include "qemu/timer.h"
106+#include "qemu/ratelimit.h"
107+#include "vma.h"
108+#include "block/block.h"
109+
110+#define BITS_PER_LONG (sizeof(unsigned long) * CHAR_BIT)
111+
112+static unsigned char zero_vma_block[VMA_BLOCK_SIZE];
113+
114+typedef struct VmaRestoreState {
115+ BlockDriverState *bs;
116+ bool write_zeroes;
117+ unsigned long *bitmap;
118+ int bitmap_size;
119+} VmaRestoreState;
120+
121+struct VmaReader {
122+ int fd;
123+ GChecksum *md5csum;
124+ GHashTable *blob_hash;
125+ unsigned char *head_data;
126+ VmaDeviceInfo devinfo[256];
127+ VmaRestoreState rstate[256];
128+ GList *cdata_list;
129+ guint8 vmstate_stream;
130+ uint32_t vmstate_clusters;
131+ /* to show restore percentage if run with -v */
132+ time_t start_time;
133+ int64_t cluster_count;
134+ int64_t clusters_read;
135+ int clusters_read_per;
136+};
137+
138+static guint
139+g_int32_hash(gconstpointer v)
140+{
141+ return *(const uint32_t *)v;
142+}
143+
144+static gboolean
145+g_int32_equal(gconstpointer v1, gconstpointer v2)
146+{
147+ return *((const uint32_t *)v1) == *((const uint32_t *)v2);
148+}
149+
150+static int vma_reader_get_bitmap(VmaRestoreState *rstate, int64_t cluster_num)
151+{
152+ assert(rstate);
153+ assert(rstate->bitmap);
154+
155+ unsigned long val, idx, bit;
156+
157+ idx = cluster_num / BITS_PER_LONG;
158+
159+ assert(rstate->bitmap_size > idx);
160+
161+ bit = cluster_num % BITS_PER_LONG;
162+ val = rstate->bitmap[idx];
163+
164+ return !!(val & (1UL << bit));
165+}
166+
167+static void vma_reader_set_bitmap(VmaRestoreState *rstate, int64_t cluster_num,
168+ int dirty)
169+{
170+ assert(rstate);
171+ assert(rstate->bitmap);
172+
173+ unsigned long val, idx, bit;
174+
175+ idx = cluster_num / BITS_PER_LONG;
176+
177+ assert(rstate->bitmap_size > idx);
178+
179+ bit = cluster_num % BITS_PER_LONG;
180+ val = rstate->bitmap[idx];
181+ if (dirty) {
182+ if (!(val & (1UL << bit))) {
183+ val |= 1UL << bit;
184+ }
185+ } else {
186+ if (val & (1UL << bit)) {
187+ val &= ~(1UL << bit);
188+ }
189+ }
190+ rstate->bitmap[idx] = val;
191+}
192+
193+typedef struct VmaBlob {
194+ uint32_t start;
195+ uint32_t len;
196+ void *data;
197+} VmaBlob;
198+
199+static const VmaBlob *get_header_blob(VmaReader *vmar, uint32_t pos)
200+{
201+ assert(vmar);
202+ assert(vmar->blob_hash);
203+
204+ return g_hash_table_lookup(vmar->blob_hash, &pos);
205+}
206+
207+static const char *get_header_str(VmaReader *vmar, uint32_t pos)
208+{
209+ const VmaBlob *blob = get_header_blob(vmar, pos);
210+ if (!blob) {
211+ return NULL;
212+ }
213+ const char *res = (char *)blob->data;
214+ if (res[blob->len-1] != '\0') {
215+ return NULL;
216+ }
217+ return res;
218+}
219+
220+static ssize_t
221+safe_read(int fd, unsigned char *buf, size_t count)
222+{
223+ ssize_t n;
224+
225+ do {
226+ n = read(fd, buf, count);
227+ } while (n < 0 && errno == EINTR);
228+
229+ return n;
230+}
231+
232+static ssize_t
233+full_read(int fd, unsigned char *buf, size_t len)
234+{
235+ ssize_t n;
236+ size_t total;
237+
238+ total = 0;
239+
240+ while (len > 0) {
241+ n = safe_read(fd, buf, len);
242+
243+ if (n == 0) {
244+ return total;
245+ }
246+
247+ if (n <= 0) {
248+ break;
249+ }
250+
251+ buf += n;
252+ total += n;
253+ len -= n;
254+ }
255+
256+ if (len) {
257+ return -1;
258+ }
259+
260+ return total;
261+}
262+
263+void vma_reader_destroy(VmaReader *vmar)
264+{
265+ assert(vmar);
266+
267+ if (vmar->fd >= 0) {
268+ close(vmar->fd);
269+ }
270+
271+ if (vmar->cdata_list) {
272+ g_list_free(vmar->cdata_list);
273+ }
274+
275+ int i;
276+ for (i = 1; i < 256; i++) {
277+ if (vmar->rstate[i].bitmap) {
278+ g_free(vmar->rstate[i].bitmap);
279+ }
280+ }
281+
282+ if (vmar->md5csum) {
283+ g_checksum_free(vmar->md5csum);
284+ }
285+
286+ if (vmar->blob_hash) {
287+ g_hash_table_destroy(vmar->blob_hash);
288+ }
289+
290+ if (vmar->head_data) {
291+ g_free(vmar->head_data);
292+ }
293+
294+ g_free(vmar);
295+
296+};
297+
298+static int vma_reader_read_head(VmaReader *vmar, Error **errp)
299+{
300+ assert(vmar);
301+ assert(errp);
302+ assert(*errp == NULL);
303+
304+ unsigned char md5sum[16];
305+ int i;
306+ int ret = 0;
307+
308+ vmar->head_data = g_malloc(sizeof(VmaHeader));
309+
310+ if (full_read(vmar->fd, vmar->head_data, sizeof(VmaHeader)) !=
311+ sizeof(VmaHeader)) {
312+ error_setg(errp, "can't read vma header - %s",
313+ errno ? g_strerror(errno) : "got EOF");
314+ return -1;
315+ }
316+
317+ VmaHeader *h = (VmaHeader *)vmar->head_data;
318+
319+ if (h->magic != VMA_MAGIC) {
320+ error_setg(errp, "not a vma file - wrong magic number");
321+ return -1;
322+ }
323+
324+ uint32_t header_size = GUINT32_FROM_BE(h->header_size);
325+ int need = header_size - sizeof(VmaHeader);
326+ if (need <= 0) {
327+ error_setg(errp, "wrong vma header size %d", header_size);
328+ return -1;
329+ }
330+
331+ vmar->head_data = g_realloc(vmar->head_data, header_size);
332+ h = (VmaHeader *)vmar->head_data;
333+
334+ if (full_read(vmar->fd, vmar->head_data + sizeof(VmaHeader), need) !=
335+ need) {
336+ error_setg(errp, "can't read vma header data - %s",
337+ errno ? g_strerror(errno) : "got EOF");
338+ return -1;
339+ }
340+
341+ memcpy(md5sum, h->md5sum, 16);
342+ memset(h->md5sum, 0, 16);
343+
344+ g_checksum_reset(vmar->md5csum);
345+ g_checksum_update(vmar->md5csum, vmar->head_data, header_size);
346+ gsize csize = 16;
347+ g_checksum_get_digest(vmar->md5csum, (guint8 *)(h->md5sum), &csize);
348+
349+ if (memcmp(md5sum, h->md5sum, 16) != 0) {
350+ error_setg(errp, "wrong vma header chechsum");
351+ return -1;
352+ }
353+
354+ /* we can modify header data after checksum verify */
355+ h->header_size = header_size;
356+
357+ h->version = GUINT32_FROM_BE(h->version);
358+ if (h->version != 1) {
359+ error_setg(errp, "wrong vma version %d", h->version);
360+ return -1;
361+ }
362+
363+ h->ctime = GUINT64_FROM_BE(h->ctime);
364+ h->blob_buffer_offset = GUINT32_FROM_BE(h->blob_buffer_offset);
365+ h->blob_buffer_size = GUINT32_FROM_BE(h->blob_buffer_size);
366+
367+ uint32_t bstart = h->blob_buffer_offset + 1;
368+ uint32_t bend = h->blob_buffer_offset + h->blob_buffer_size;
369+
370+ if (bstart <= sizeof(VmaHeader)) {
371+ error_setg(errp, "wrong vma blob buffer offset %d",
372+ h->blob_buffer_offset);
373+ return -1;
374+ }
375+
376+ if (bend > header_size) {
377+ error_setg(errp, "wrong vma blob buffer size %d/%d",
378+ h->blob_buffer_offset, h->blob_buffer_size);
379+ return -1;
380+ }
381+
382+ while ((bstart + 2) <= bend) {
383+ uint32_t size = vmar->head_data[bstart] +
384+ (vmar->head_data[bstart+1] << 8);
385+ if ((bstart + size + 2) <= bend) {
386+ VmaBlob *blob = g_new0(VmaBlob, 1);
387+ blob->start = bstart - h->blob_buffer_offset;
388+ blob->len = size;
389+ blob->data = vmar->head_data + bstart + 2;
390+ g_hash_table_insert(vmar->blob_hash, &blob->start, blob);
391+ }
392+ bstart += size + 2;
393+ }
394+
395+
396+ int count = 0;
397+ for (i = 1; i < 256; i++) {
398+ VmaDeviceInfoHeader *dih = &h->dev_info[i];
399+ uint32_t devname_ptr = GUINT32_FROM_BE(dih->devname_ptr);
400+ uint64_t size = GUINT64_FROM_BE(dih->size);
401+ const char *devname = get_header_str(vmar, devname_ptr);
402+
403+ if (size && devname) {
404+ count++;
405+ vmar->devinfo[i].size = size;
406+ vmar->devinfo[i].devname = devname;
407+
408+ if (strcmp(devname, "vmstate") == 0) {
409+ vmar->vmstate_stream = i;
410+ }
411+ }
412+ }
413+
414+ if (!count) {
415+ error_setg(errp, "vma does not contain data");
416+ return -1;
417+ }
418+
419+ for (i = 0; i < VMA_MAX_CONFIGS; i++) {
420+ uint32_t name_ptr = GUINT32_FROM_BE(h->config_names[i]);
421+ uint32_t data_ptr = GUINT32_FROM_BE(h->config_data[i]);
422+
423+ if (!(name_ptr && data_ptr)) {
424+ continue;
425+ }
426+ const char *name = get_header_str(vmar, name_ptr);
427+ const VmaBlob *blob = get_header_blob(vmar, data_ptr);
428+
429+ if (!(name && blob)) {
430+ error_setg(errp, "vma contains invalid data pointers");
431+ return -1;
432+ }
433+
434+ VmaConfigData *cdata = g_new0(VmaConfigData, 1);
435+ cdata->name = name;
436+ cdata->data = blob->data;
437+ cdata->len = blob->len;
438+
439+ vmar->cdata_list = g_list_append(vmar->cdata_list, cdata);
440+ }
441+
442+ return ret;
443+};
444+
445+VmaReader *vma_reader_create(const char *filename, Error **errp)
446+{
447+ assert(filename);
448+ assert(errp);
449+
450+ VmaReader *vmar = g_new0(VmaReader, 1);
451+
452+ if (strcmp(filename, "-") == 0) {
453+ vmar->fd = dup(0);
454+ } else {
455+ vmar->fd = open(filename, O_RDONLY);
456+ }
457+
458+ if (vmar->fd < 0) {
459+ error_setg(errp, "can't open file %s - %s\n", filename,
460+ g_strerror(errno));
461+ goto err;
462+ }
463+
464+ vmar->md5csum = g_checksum_new(G_CHECKSUM_MD5);
465+ if (!vmar->md5csum) {
466+ error_setg(errp, "can't allocate cmsum\n");
467+ goto err;
468+ }
469+
470+ vmar->blob_hash = g_hash_table_new_full(g_int32_hash, g_int32_equal,
471+ NULL, g_free);
472+
473+ if (vma_reader_read_head(vmar, errp) < 0) {
474+ goto err;
475+ }
476+
477+ return vmar;
478+
479+err:
480+ if (vmar) {
481+ vma_reader_destroy(vmar);
482+ }
483+
484+ return NULL;
485+}
486+
487+VmaHeader *vma_reader_get_header(VmaReader *vmar)
488+{
489+ assert(vmar);
490+ assert(vmar->head_data);
491+
492+ return (VmaHeader *)(vmar->head_data);
493+}
494+
495+GList *vma_reader_get_config_data(VmaReader *vmar)
496+{
497+ assert(vmar);
498+ assert(vmar->head_data);
499+
500+ return vmar->cdata_list;
501+}
502+
503+VmaDeviceInfo *vma_reader_get_device_info(VmaReader *vmar, guint8 dev_id)
504+{
505+ assert(vmar);
506+ assert(dev_id);
507+
508+ if (vmar->devinfo[dev_id].size && vmar->devinfo[dev_id].devname) {
509+ return &vmar->devinfo[dev_id];
510+ }
511+
512+ return NULL;
513+}
514+
515+int vma_reader_register_bs(VmaReader *vmar, guint8 dev_id, BlockDriverState *bs,
516+ bool write_zeroes, Error **errp)
517+{
518+ assert(vmar);
519+ assert(bs != NULL);
520+ assert(dev_id);
521+ assert(vmar->rstate[dev_id].bs == NULL);
522+
523+ int64_t size = bdrv_getlength(bs);
524+ if (size != vmar->devinfo[dev_id].size) {
525+ error_setg(errp, "vma_reader_register_bs for stream %s failed - "
526+ "unexpected size %zd != %zd", vmar->devinfo[dev_id].devname,
527+ size, vmar->devinfo[dev_id].size);
528+ return -1;
529+ }
530+
531+ vmar->rstate[dev_id].bs = bs;
532+ vmar->rstate[dev_id].write_zeroes = write_zeroes;
533+
534+ int64_t bitmap_size = (size/BDRV_SECTOR_SIZE) +
535+ (VMA_CLUSTER_SIZE/BDRV_SECTOR_SIZE) * BITS_PER_LONG - 1;
536+ bitmap_size /= (VMA_CLUSTER_SIZE/BDRV_SECTOR_SIZE) * BITS_PER_LONG;
537+
538+ vmar->rstate[dev_id].bitmap_size = bitmap_size;
539+ vmar->rstate[dev_id].bitmap = g_new0(unsigned long, bitmap_size);
540+
541+ vmar->cluster_count += size/VMA_CLUSTER_SIZE;
542+
543+ return 0;
544+}
545+
546+static ssize_t safe_write(int fd, void *buf, size_t count)
547+{
548+ ssize_t n;
549+
550+ do {
551+ n = write(fd, buf, count);
552+ } while (n < 0 && errno == EINTR);
553+
554+ return n;
555+}
556+
557+static size_t full_write(int fd, void *buf, size_t len)
558+{
559+ ssize_t n;
560+ size_t total;
561+
562+ total = 0;
563+
564+ while (len > 0) {
565+ n = safe_write(fd, buf, len);
566+ if (n < 0) {
567+ return n;
568+ }
569+ buf += n;
570+ total += n;
571+ len -= n;
572+ }
573+
574+ if (len) {
575+ /* incomplete write ? */
576+ return -1;
577+ }
578+
579+ return total;
580+}
581+
582+static int restore_write_data(VmaReader *vmar, guint8 dev_id,
583+ BlockDriverState *bs, int vmstate_fd,
584+ unsigned char *buf, int64_t sector_num,
585+ int nb_sectors, Error **errp)
586+{
587+ assert(vmar);
588+
589+ if (dev_id == vmar->vmstate_stream) {
590+ if (vmstate_fd >= 0) {
591+ int len = nb_sectors * BDRV_SECTOR_SIZE;
592+ int res = full_write(vmstate_fd, buf, len);
593+ if (res < 0) {
594+ error_setg(errp, "write vmstate failed %d", res);
595+ return -1;
596+ }
597+ }
598+ } else {
599+ int res = bdrv_write(bs, sector_num, buf, nb_sectors);
600+ if (res < 0) {
601+ error_setg(errp, "bdrv_write to %s failed (%d)",
602+ bdrv_get_device_name(bs), res);
603+ return -1;
604+ }
605+ }
606+ return 0;
607+}
608+static int restore_extent(VmaReader *vmar, unsigned char *buf,
609+ int extent_size, int vmstate_fd,
610+ bool verbose, Error **errp)
611+{
612+ assert(vmar);
613+ assert(buf);
614+
615+ VmaExtentHeader *ehead = (VmaExtentHeader *)buf;
616+ int start = VMA_EXTENT_HEADER_SIZE;
617+ int i;
618+
619+ for (i = 0; i < VMA_BLOCKS_PER_EXTENT; i++) {
620+ uint64_t block_info = GUINT64_FROM_BE(ehead->blockinfo[i]);
621+ uint64_t cluster_num = block_info & 0xffffffff;
622+ uint8_t dev_id = (block_info >> 32) & 0xff;
623+ uint16_t mask = block_info >> (32+16);
624+ int64_t max_sector;
625+
626+ if (!dev_id) {
627+ continue;
628+ }
629+
630+ VmaRestoreState *rstate = &vmar->rstate[dev_id];
631+ BlockDriverState *bs = NULL;
632+
633+ if (dev_id != vmar->vmstate_stream) {
634+ bs = rstate->bs;
635+ if (!bs) {
636+ error_setg(errp, "got wrong dev id %d", dev_id);
637+ return -1;
638+ }
639+
640+ if (vma_reader_get_bitmap(rstate, cluster_num)) {
641+ error_setg(errp, "found duplicated cluster %zd for stream %s",
642+ cluster_num, vmar->devinfo[dev_id].devname);
643+ return -1;
644+ }
645+ vma_reader_set_bitmap(rstate, cluster_num, 1);
646+
647+ max_sector = vmar->devinfo[dev_id].size/BDRV_SECTOR_SIZE;
648+ } else {
649+ max_sector = G_MAXINT64;
650+ if (cluster_num != vmar->vmstate_clusters) {
651+ error_setg(errp, "found out of order vmstate data");
652+ return -1;
653+ }
654+ vmar->vmstate_clusters++;
655+ }
656+
657+ vmar->clusters_read++;
658+
659+ if (verbose) {
660+ time_t duration = time(NULL) - vmar->start_time;
661+ int percent = (vmar->clusters_read*100)/vmar->cluster_count;
662+ if (percent != vmar->clusters_read_per) {
663+ printf("progress %d%% (read %zd bytes, duration %zd sec)\n",
664+ percent, vmar->clusters_read*VMA_CLUSTER_SIZE,
665+ duration);
666+ fflush(stdout);
667+ vmar->clusters_read_per = percent;
668+ }
669+ }
670+
671+ /* try to write whole clusters to speedup restore */
672+ if (mask == 0xffff) {
673+ if ((start + VMA_CLUSTER_SIZE) > extent_size) {
674+ error_setg(errp, "short vma extent - too many blocks");
675+ return -1;
676+ }
677+ int64_t sector_num = (cluster_num * VMA_CLUSTER_SIZE) /
678+ BDRV_SECTOR_SIZE;
679+ int64_t end_sector = sector_num +
680+ VMA_CLUSTER_SIZE/BDRV_SECTOR_SIZE;
681+
682+ if (end_sector > max_sector) {
683+ end_sector = max_sector;
684+ }
685+
686+ if (end_sector <= sector_num) {
687+ error_setg(errp, "got wrong block address - write bejond end");
688+ return -1;
689+ }
690+
691+ int nb_sectors = end_sector - sector_num;
692+ if (restore_write_data(vmar, dev_id, bs, vmstate_fd, buf + start,
693+ sector_num, nb_sectors, errp) < 0) {
694+ return -1;
695+ }
696+
697+ start += VMA_CLUSTER_SIZE;
698+ } else {
699+ int j;
700+ int bit = 1;
701+
702+ for (j = 0; j < 16; j++) {
703+ int64_t sector_num = (cluster_num*VMA_CLUSTER_SIZE +
704+ j*VMA_BLOCK_SIZE)/BDRV_SECTOR_SIZE;
705+
706+ int64_t end_sector = sector_num +
707+ VMA_BLOCK_SIZE/BDRV_SECTOR_SIZE;
708+ if (end_sector > max_sector) {
709+ end_sector = max_sector;
710+ }
711+
712+ if (mask & bit) {
713+ if ((start + VMA_BLOCK_SIZE) > extent_size) {
714+ error_setg(errp, "short vma extent - too many blocks");
715+ return -1;
716+ }
717+
718+ if (end_sector <= sector_num) {
719+ error_setg(errp, "got wrong block address - "
720+ "write bejond end");
721+ return -1;
722+ }
723+
724+ int nb_sectors = end_sector - sector_num;
725+ if (restore_write_data(vmar, dev_id, bs, vmstate_fd,
726+ buf + start, sector_num,
727+ nb_sectors, errp) < 0) {
728+ return -1;
729+ }
730+
731+ start += VMA_BLOCK_SIZE;
732+
733+ } else {
734+
735+ if (rstate->write_zeroes && (end_sector > sector_num)) {
736+ /* Todo: use bdrv_co_write_zeroes (but that need to
737+ * be run inside coroutine?)
738+ */
739+ int nb_sectors = end_sector - sector_num;
740+ if (restore_write_data(vmar, dev_id, bs, vmstate_fd,
741+ zero_vma_block, sector_num,
742+ nb_sectors, errp) < 0) {
743+ return -1;
744+ }
745+ }
746+ }
747+
748+ bit = bit << 1;
749+ }
750+ }
751+ }
752+
753+ if (start != extent_size) {
754+ error_setg(errp, "vma extent error - missing blocks");
755+ return -1;
756+ }
757+
758+ return 0;
759+}
760+
761+int vma_reader_restore(VmaReader *vmar, int vmstate_fd, bool verbose,
762+ Error **errp)
763+{
764+ assert(vmar);
765+ assert(vmar->head_data);
766+
767+ int ret = 0;
768+ unsigned char buf[VMA_MAX_EXTENT_SIZE];
769+ int buf_pos = 0;
770+ unsigned char md5sum[16];
771+ VmaHeader *h = (VmaHeader *)vmar->head_data;
772+
773+ vmar->start_time = time(NULL);
774+
775+ while (1) {
776+ int bytes = full_read(vmar->fd, buf + buf_pos, sizeof(buf) - buf_pos);
777+ if (bytes < 0) {
778+ error_setg(errp, "read failed - %s", g_strerror(errno));
779+ return -1;
780+ }
781+
782+ buf_pos += bytes;
783+
784+ if (!buf_pos) {
785+ break; /* EOF */
786+ }
787+
788+ if (buf_pos < VMA_EXTENT_HEADER_SIZE) {
789+ error_setg(errp, "read short extent (%d bytes)", buf_pos);
790+ return -1;
791+ }
792+
793+ VmaExtentHeader *ehead = (VmaExtentHeader *)buf;
794+
795+ /* extract md5sum */
796+ memcpy(md5sum, ehead->md5sum, sizeof(ehead->md5sum));
797+ memset(ehead->md5sum, 0, sizeof(ehead->md5sum));
798+
799+ g_checksum_reset(vmar->md5csum);
800+ g_checksum_update(vmar->md5csum, buf, VMA_EXTENT_HEADER_SIZE);
801+ gsize csize = 16;
802+ g_checksum_get_digest(vmar->md5csum, ehead->md5sum, &csize);
803+
804+ if (memcmp(md5sum, ehead->md5sum, 16) != 0) {
805+ error_setg(errp, "wrong vma extent header chechsum");
806+ return -1;
807+ }
808+
809+ if (memcmp(h->uuid, ehead->uuid, sizeof(ehead->uuid)) != 0) {
810+ error_setg(errp, "wrong vma extent uuid");
811+ return -1;
812+ }
813+
814+ if (ehead->magic != VMA_EXTENT_MAGIC || ehead->reserved1 != 0) {
815+ error_setg(errp, "wrong vma extent header magic");
816+ return -1;
817+ }
818+
819+ int block_count = GUINT16_FROM_BE(ehead->block_count);
820+ int extent_size = VMA_EXTENT_HEADER_SIZE + block_count*VMA_BLOCK_SIZE;
821+
822+ if (buf_pos < extent_size) {
823+ error_setg(errp, "short vma extent (%d < %d)", buf_pos,
824+ extent_size);
825+ return -1;
826+ }
827+
828+ if (restore_extent(vmar, buf, extent_size, vmstate_fd, verbose,
829+ errp) < 0) {
830+ return -1;
831+ }
832+
833+ if (buf_pos > extent_size) {
834+ memmove(buf, buf + extent_size, buf_pos - extent_size);
835+ buf_pos = buf_pos - extent_size;
836+ } else {
837+ buf_pos = 0;
838+ }
839+ }
840+
841+ bdrv_drain_all();
842+
843+ int i;
844+ for (i = 1; i < 256; i++) {
845+ VmaRestoreState *rstate = &vmar->rstate[i];
846+ if (!rstate->bs) {
847+ continue;
848+ }
849+
850+ if (bdrv_flush(rstate->bs) < 0) {
851+ error_setg(errp, "vma bdrv_flush %s failed",
852+ vmar->devinfo[i].devname);
853+ return -1;
854+ }
855+
856+ if (vmar->devinfo[i].size &&
857+ (strcmp(vmar->devinfo[i].devname, "vmstate") != 0)) {
858+ assert(rstate->bitmap);
859+
860+ int64_t cluster_num, end;
861+
862+ end = (vmar->devinfo[i].size + VMA_CLUSTER_SIZE - 1) /
863+ VMA_CLUSTER_SIZE;
864+
865+ for (cluster_num = 0; cluster_num < end; cluster_num++) {
866+ if (!vma_reader_get_bitmap(rstate, cluster_num)) {
867+ error_setg(errp, "detected missing cluster %zd "
868+ "for stream %s", cluster_num,
869+ vmar->devinfo[i].devname);
870+ return -1;
871+ }
872+ }
873+ }
874+ }
875+
876+ return ret;
877+}
878+
879Index: new/vma-writer.c
880===================================================================
881--- /dev/null 1970-01-01 00:00:00.000000000 +0000
80b30085
DM		 882+++ new/vma-writer.c 2013-11-29 10:19:59.000000000 +0100
883@@ -0,0 +1,872 @@
b668dc80
DM		 884+/*
885+ * VMA: Virtual Machine Archive
886+ *
887+ * Copyright (C) 2012 Proxmox Server Solutions
888+ *
889+ * Authors:
890+ * Dietmar Maurer (dietmar@proxmox.com)
891+ *
892+ * This work is licensed under the terms of the GNU GPL, version 2 or later.
893+ * See the COPYING file in the top-level directory.
894+ *
895+ */
896+
897+#include <stdio.h>
898+#include <errno.h>
899+#include <unistd.h>
900+#include <stdio.h>
901+#include <string.h>
902+#include <sys/types.h>
903+#include <sys/stat.h>
904+#include <fcntl.h>
905+#include <glib.h>
906+#include <uuid/uuid.h>
907+
908+#include "qemu-common.h"
909+#include "vma.h"
910+#include "block/block.h"
911+#include "monitor/monitor.h"
912+
913+#define DEBUG_VMA 0
914+
915+#define DPRINTF(fmt, ...)\
916+ do { if (DEBUG_VMA) { printf("vma: " fmt, ## __VA_ARGS__); } } while (0)
917+
918+#define WRITE_BUFFERS 5
919+
920+typedef struct VmaAIOCB VmaAIOCB;
921+struct VmaAIOCB {
922+ unsigned char buffer[VMA_MAX_EXTENT_SIZE];
923+ VmaWriter *vmaw;
924+ size_t bytes;
925+ Coroutine *co;
926+};
927+
928+struct VmaWriter {
929+ int fd;
930+ FILE *cmd;
931+ int status;
932+ char errmsg[8192];
933+ uuid_t uuid;
934+ bool header_written;
935+ bool closed;
936+
937+ /* we always write extents */
938+ unsigned char outbuf[VMA_MAX_EXTENT_SIZE];
939+ int outbuf_pos; /* in bytes */
940+ int outbuf_count; /* in VMA_BLOCKS */
941+ uint64_t outbuf_block_info[VMA_BLOCKS_PER_EXTENT];
942+
943+ VmaAIOCB *aiocbs[WRITE_BUFFERS];
944+ CoQueue wqueue;
945+
946+ GChecksum *md5csum;
947+ CoMutex writer_lock;
948+ CoMutex flush_lock;
949+ Coroutine *co_writer;
950+
951+ /* drive informations */
952+ VmaStreamInfo stream_info[256];
953+ guint stream_count;
954+
955+ guint8 vmstate_stream;
956+ uint32_t vmstate_clusters;
957+
958+ /* header blob table */
959+ char *header_blob_table;
960+ uint32_t header_blob_table_size;
961+ uint32_t header_blob_table_pos;
962+
963+ /* store for config blobs */
964+ uint32_t config_names[VMA_MAX_CONFIGS]; /* offset into blob_buffer table */
965+ uint32_t config_data[VMA_MAX_CONFIGS]; /* offset into blob_buffer table */
966+ uint32_t config_count;
967+};
968+
969+void vma_writer_set_error(VmaWriter *vmaw, const char *fmt, ...)
970+{
971+ va_list ap;
972+
973+ if (vmaw->status < 0) {
974+ return;
975+ }
976+
977+ vmaw->status = -1;
978+
979+ va_start(ap, fmt);
980+ g_vsnprintf(vmaw->errmsg, sizeof(vmaw->errmsg), fmt, ap);
981+ va_end(ap);
982+
983+ DPRINTF("vma_writer_set_error: %s\n", vmaw->errmsg);
984+}
985+
986+static uint32_t allocate_header_blob(VmaWriter *vmaw, const char *data,
987+ size_t len)
988+{
989+ if (len > 65535) {
990+ return 0;
991+ }
992+
993+ if (!vmaw->header_blob_table ||
994+ (vmaw->header_blob_table_size <
995+ (vmaw->header_blob_table_pos + len + 2))) {
996+ int newsize = vmaw->header_blob_table_size + ((len + 2 + 511)/512)*512;
997+
998+ vmaw->header_blob_table = g_realloc(vmaw->header_blob_table, newsize);
999+ memset(vmaw->header_blob_table + vmaw->header_blob_table_size,
1000+ 0, newsize - vmaw->header_blob_table_size);
1001+ vmaw->header_blob_table_size = newsize;
1002+ }
1003+
1004+ uint32_t cpos = vmaw->header_blob_table_pos;
1005+ vmaw->header_blob_table[cpos] = len & 255;
1006+ vmaw->header_blob_table[cpos+1] = (len >> 8) & 255;
1007+ memcpy(vmaw->header_blob_table + cpos + 2, data, len);
1008+ vmaw->header_blob_table_pos += len + 2;
1009+ return cpos;
1010+}
1011+
1012+static uint32_t allocate_header_string(VmaWriter *vmaw, const char *str)
1013+{
1014+ assert(vmaw);
1015+
1016+ size_t len = strlen(str) + 1;
1017+
1018+ return allocate_header_blob(vmaw, str, len);
1019+}
1020+
1021+int vma_writer_add_config(VmaWriter *vmaw, const char *name, gpointer data,
1022+ gsize len)
1023+{
1024+ assert(vmaw);
1025+ assert(!vmaw->header_written);
1026+ assert(vmaw->config_count < VMA_MAX_CONFIGS);
1027+ assert(name);
1028+ assert(data);
1029+ assert(len);
1030+
80b30085
DM		 1031+ gchar *basename = g_path_get_basename(name);
1032+ uint32_t name_ptr = allocate_header_string(vmaw, basename);
1033+ g_free(basename);
1034+
b668dc80
DM		 1035+ if (!name_ptr) {
1036+ return -1;
1037+ }
1038+
1039+ uint32_t data_ptr = allocate_header_blob(vmaw, data, len);
1040+ if (!data_ptr) {
1041+ return -1;
1042+ }
1043+
1044+ vmaw->config_names[vmaw->config_count] = name_ptr;
1045+ vmaw->config_data[vmaw->config_count] = data_ptr;
1046+
1047+ vmaw->config_count++;
1048+
1049+ return 0;
1050+}
1051+
1052+int vma_writer_register_stream(VmaWriter *vmaw, const char *devname,
1053+ size_t size)
1054+{
1055+ assert(vmaw);
1056+ assert(devname);
1057+ assert(!vmaw->status);
1058+
1059+ if (vmaw->header_written) {
1060+ vma_writer_set_error(vmaw, "vma_writer_register_stream: header "
1061+ "already written");
1062+ return -1;
1063+ }
1064+
1065+ guint n = vmaw->stream_count + 1;
1066+
1067+ /* we can have dev_ids form 1 to 255 (0 reserved)
1068+ * 255(-1) reseverd for safety
1069+ */
1070+ if (n > 254) {
1071+ vma_writer_set_error(vmaw, "vma_writer_register_stream: "
1072+ "too many drives");
1073+ return -1;
1074+ }
1075+
1076+ if (size <= 0) {
1077+ vma_writer_set_error(vmaw, "vma_writer_register_stream: "
1078+ "got strange size %zd", size);
1079+ return -1;
1080+ }
1081+
1082+ DPRINTF("vma_writer_register_stream %s %zu %d\n", devname, size, n);
1083+
1084+ vmaw->stream_info[n].devname = g_strdup(devname);
1085+ vmaw->stream_info[n].size = size;
1086+
1087+ vmaw->stream_info[n].cluster_count = (size + VMA_CLUSTER_SIZE - 1) /
1088+ VMA_CLUSTER_SIZE;
1089+
1090+ vmaw->stream_count = n;
1091+
1092+ if (strcmp(devname, "vmstate") == 0) {
1093+ vmaw->vmstate_stream = n;
1094+ }
1095+
1096+ return n;
1097+}
1098+
1099+static void vma_co_continue_write(void *opaque)
1100+{
1101+ VmaWriter *vmaw = opaque;
1102+
1103+ DPRINTF("vma_co_continue_write\n");
1104+ qemu_coroutine_enter(vmaw->co_writer, NULL);
1105+}
1106+
1107+static ssize_t coroutine_fn
1108+vma_co_write(VmaWriter *vmaw, const void *buf, size_t bytes)
1109+{
1110+ size_t done = 0;
1111+ ssize_t ret;
1112+
1113+ /* atomic writes (we cannot interleave writes) */
1114+ qemu_co_mutex_lock(&vmaw->writer_lock);
1115+
1116+ DPRINTF("vma_co_write enter %zd\n", bytes);
1117+
1118+ assert(vmaw->co_writer == NULL);
1119+
1120+ vmaw->co_writer = qemu_coroutine_self();
1121+
1122+ qemu_aio_set_fd_handler(vmaw->fd, NULL, vma_co_continue_write, vmaw);
1123+
1124+ DPRINTF("vma_co_write wait until writable\n");
1125+ qemu_coroutine_yield();
1126+ DPRINTF("vma_co_write starting %zd\n", bytes);
1127+
1128+ while (done < bytes) {
1129+ ret = write(vmaw->fd, buf + done, bytes - done);
1130+ if (ret > 0) {
1131+ done += ret;
1132+ DPRINTF("vma_co_write written %zd %zd\n", done, ret);
1133+ } else if (ret < 0) {
1134+ if (errno == EAGAIN || errno == EWOULDBLOCK) {
1135+ DPRINTF("vma_co_write yield %zd\n", done);
1136+ qemu_coroutine_yield();
1137+ DPRINTF("vma_co_write restart %zd\n", done);
1138+ } else {
1139+ vma_writer_set_error(vmaw, "vma_co_write write error - %s",
1140+ g_strerror(errno));
1141+ done = -1; /* always return failure for partial writes */
1142+ break;
1143+ }
1144+ } else if (ret == 0) {
1145+ /* should not happen - simply try again */
1146+ }
1147+ }
1148+
1149+ qemu_aio_set_fd_handler(vmaw->fd, NULL, NULL, NULL);
1150+
1151+ vmaw->co_writer = NULL;
1152+
1153+ qemu_co_mutex_unlock(&vmaw->writer_lock);
1154+
1155+ DPRINTF("vma_co_write leave %zd\n", done);
1156+ return done;
1157+}
1158+
1159+static void coroutine_fn vma_co_writer_task(void *opaque)
1160+{
1161+ VmaAIOCB *cb = opaque;
1162+
1163+ DPRINTF("vma_co_writer_task start\n");
1164+
1165+ int64_t done = vma_co_write(cb->vmaw, cb->buffer, cb->bytes);
1166+ DPRINTF("vma_co_writer_task write done %zd\n", done);
1167+
1168+ if (done != cb->bytes) {
1169+ DPRINTF("vma_co_writer_task failed write %zd %zd", cb->bytes, done);
1170+ vma_writer_set_error(cb->vmaw, "vma_co_writer_task failed write %zd",
1171+ done);
1172+ }
1173+
1174+ cb->bytes = 0;
1175+
1176+ qemu_co_queue_next(&cb->vmaw->wqueue);
1177+
1178+ DPRINTF("vma_co_writer_task end\n");
1179+}
1180+
1181+static void coroutine_fn vma_queue_flush(VmaWriter *vmaw)
1182+{
1183+ DPRINTF("vma_queue_flush enter\n");
1184+
1185+ assert(vmaw);
1186+
1187+ while (1) {
1188+ int i;
1189+ VmaAIOCB *cb = NULL;
1190+ for (i = 0; i < WRITE_BUFFERS; i++) {
1191+ if (vmaw->aiocbs[i]->bytes) {
1192+ cb = vmaw->aiocbs[i];
1193+ DPRINTF("FOUND USED AIO BUFFER %d %zd\n", i,
1194+ vmaw->aiocbs[i]->bytes);
1195+ break;
1196+ }
1197+ }
1198+ if (!cb) {
1199+ break;
1200+ }
1201+ qemu_co_queue_wait(&vmaw->wqueue);
1202+ }
1203+
1204+ DPRINTF("vma_queue_flush leave\n");
1205+}
1206+
1207+/**
1208+ * NOTE: pipe buffer size in only 4096 bytes on linux (see 'ulimit -a')
1209+ * So we need to create a coroutione to allow 'parallel' execution.
1210+ */
1211+static ssize_t coroutine_fn
1212+vma_queue_write(VmaWriter *vmaw, const void *buf, size_t bytes)
1213+{
1214+ DPRINTF("vma_queue_write enter %zd\n", bytes);
1215+
1216+ assert(vmaw);
1217+ assert(buf);
1218+ assert(bytes <= VMA_MAX_EXTENT_SIZE);
1219+
1220+ VmaAIOCB *cb = NULL;
1221+ while (!cb) {
1222+ int i;
1223+ for (i = 0; i < WRITE_BUFFERS; i++) {
1224+ if (!vmaw->aiocbs[i]->bytes) {
1225+ cb = vmaw->aiocbs[i];
1226+ break;
1227+ }
1228+ }
1229+ if (!cb) {
1230+ qemu_co_queue_wait(&vmaw->wqueue);
1231+ }
1232+ }
1233+
1234+ memcpy(cb->buffer, buf, bytes);
1235+ cb->bytes = bytes;
1236+ cb->vmaw = vmaw;
1237+
1238+ DPRINTF("vma_queue_write start %zd\n", bytes);
1239+ cb->co = qemu_coroutine_create(vma_co_writer_task);
1240+ qemu_coroutine_enter(cb->co, cb);
1241+
1242+ DPRINTF("vma_queue_write leave\n");
1243+
1244+ return bytes;
1245+}
1246+
1247+VmaWriter *vma_writer_create(const char *filename, uuid_t uuid, Error **errp)
1248+{
1249+ const char *p;
1250+
1251+ assert(sizeof(VmaHeader) == (4096 + 8192));
1252+ assert(sizeof(VmaExtentHeader) == 512);
1253+
1254+ VmaWriter *vmaw = g_new0(VmaWriter, 1);
1255+ vmaw->fd = -1;
1256+
1257+ vmaw->md5csum = g_checksum_new(G_CHECKSUM_MD5);
1258+ if (!vmaw->md5csum) {
1259+ error_setg(errp, "can't allocate cmsum\n");
1260+ goto err;
1261+ }
1262+
1263+ if (strstart(filename, "exec:", &p)) {
1264+ vmaw->cmd = popen(p, "w");
1265+ if (vmaw->cmd == NULL) {
1266+ error_setg(errp, "can't popen command '%s' - %s\n", p,
1267+ g_strerror(errno));
1268+ goto err;
1269+ }
1270+ vmaw->fd = fileno(vmaw->cmd);
1271+
1272+ /* try to use O_NONBLOCK and O_DIRECT */
1273+ fcntl(vmaw->fd, F_SETFL, fcntl(vmaw->fd, F_GETFL)|O_NONBLOCK);
1274+ fcntl(vmaw->fd, F_SETFL, fcntl(vmaw->fd, F_GETFL)|O_DIRECT);
1275+
1276+ } else {
1277+ struct stat st;
1278+ int oflags;
1279+ const char *tmp_id_str;
1280+
1281+ if ((stat(filename, &st) == 0) && S_ISFIFO(st.st_mode)) {
1282+ oflags = O_NONBLOCK|O_DIRECT|O_WRONLY;
1283+ vmaw->fd = qemu_open(filename, oflags, 0644);
1284+ } else if (strstart(filename, "/dev/fdset/", &tmp_id_str)) {
1285+ oflags = O_NONBLOCK|O_DIRECT|O_WRONLY;
1286+ vmaw->fd = qemu_open(filename, oflags, 0644);
1287+ } else if (strstart(filename, "/dev/fdname/", &tmp_id_str)) {
1288+ vmaw->fd = monitor_get_fd(cur_mon, tmp_id_str, errp);
1289+ if (vmaw->fd < 0) {
1290+ goto err;
1291+ }
1292+ /* try to use O_NONBLOCK and O_DIRECT */
1293+ fcntl(vmaw->fd, F_SETFL, fcntl(vmaw->fd, F_GETFL)|O_NONBLOCK);
1294+ fcntl(vmaw->fd, F_SETFL, fcntl(vmaw->fd, F_GETFL)|O_DIRECT);
1295+ } else {
1296+ oflags = O_NONBLOCK|O_DIRECT|O_WRONLY|O_CREAT|O_EXCL;
1297+ vmaw->fd = qemu_open(filename, oflags, 0644);
1298+ }
1299+
1300+ if (vmaw->fd < 0) {
1301+ error_setg(errp, "can't open file %s - %s\n", filename,
1302+ g_strerror(errno));
1303+ goto err;
1304+ }
1305+ }
1306+
1307+ /* we use O_DIRECT, so we need to align IO buffers */
1308+ int i;
1309+ for (i = 0; i < WRITE_BUFFERS; i++) {
1310+ vmaw->aiocbs[i] = qemu_memalign(512, sizeof(VmaAIOCB));
1311+ memset(vmaw->aiocbs[i], 0, sizeof(VmaAIOCB));
1312+ }
1313+
1314+ vmaw->outbuf_count = 0;
1315+ vmaw->outbuf_pos = VMA_EXTENT_HEADER_SIZE;
1316+
1317+ vmaw->header_blob_table_pos = 1; /* start at pos 1 */
1318+
1319+ qemu_co_mutex_init(&vmaw->writer_lock);
1320+ qemu_co_mutex_init(&vmaw->flush_lock);
1321+ qemu_co_queue_init(&vmaw->wqueue);
1322+
1323+ uuid_copy(vmaw->uuid, uuid);
1324+
1325+ return vmaw;
1326+
1327+err:
1328+ if (vmaw) {
1329+ if (vmaw->cmd) {
1330+ pclose(vmaw->cmd);
1331+ } else if (vmaw->fd >= 0) {
1332+ close(vmaw->fd);
1333+ }
1334+
1335+ if (vmaw->md5csum) {
1336+ g_checksum_free(vmaw->md5csum);
1337+ }
1338+
1339+ g_free(vmaw);
1340+ }
1341+
1342+ return NULL;
1343+}
1344+
1345+static int coroutine_fn vma_write_header(VmaWriter *vmaw)
1346+{
1347+ assert(vmaw);
1348+ int header_clusters = 8;
1349+ char buf[65536*header_clusters];
1350+ VmaHeader *head = (VmaHeader *)buf;
1351+
1352+ int i;
1353+
1354+ DPRINTF("VMA WRITE HEADER\n");
1355+
1356+ if (vmaw->status < 0) {
1357+ return vmaw->status;
1358+ }
1359+
1360+ memset(buf, 0, sizeof(buf));
1361+
1362+ head->magic = VMA_MAGIC;
1363+ head->version = GUINT32_TO_BE(1); /* v1 */
1364+ memcpy(head->uuid, vmaw->uuid, 16);
1365+
1366+ time_t ctime = time(NULL);
1367+ head->ctime = GUINT64_TO_BE(ctime);
1368+
1369+ if (!vmaw->stream_count) {
1370+ return -1;
1371+ }
1372+
1373+ for (i = 0; i < VMA_MAX_CONFIGS; i++) {
1374+ head->config_names[i] = GUINT32_TO_BE(vmaw->config_names[i]);
1375+ head->config_data[i] = GUINT32_TO_BE(vmaw->config_data[i]);
1376+ }
1377+
1378+ /* 32 bytes per device (12 used currently) = 8192 bytes max */
1379+ for (i = 1; i <= 254; i++) {
1380+ VmaStreamInfo *si = &vmaw->stream_info[i];
1381+ if (si->size) {
1382+ assert(si->devname);
1383+ uint32_t devname_ptr = allocate_header_string(vmaw, si->devname);
1384+ if (!devname_ptr) {
1385+ return -1;
1386+ }
1387+ head->dev_info[i].devname_ptr = GUINT32_TO_BE(devname_ptr);
1388+ head->dev_info[i].size = GUINT64_TO_BE(si->size);
1389+ }
1390+ }
1391+
1392+ uint32_t header_size = sizeof(VmaHeader) + vmaw->header_blob_table_size;
1393+ head->header_size = GUINT32_TO_BE(header_size);
1394+
1395+ if (header_size > sizeof(buf)) {
1396+ return -1; /* just to be sure */
1397+ }
1398+
1399+ uint32_t blob_buffer_offset = sizeof(VmaHeader);
1400+ memcpy(buf + blob_buffer_offset, vmaw->header_blob_table,
1401+ vmaw->header_blob_table_size);
1402+ head->blob_buffer_offset = GUINT32_TO_BE(blob_buffer_offset);
1403+ head->blob_buffer_size = GUINT32_TO_BE(vmaw->header_blob_table_pos);
1404+
1405+ g_checksum_reset(vmaw->md5csum);
1406+ g_checksum_update(vmaw->md5csum, (const guchar *)buf, header_size);
1407+ gsize csize = 16;
1408+ g_checksum_get_digest(vmaw->md5csum, (guint8 *)(head->md5sum), &csize);
1409+
1410+ return vma_queue_write(vmaw, buf, header_size);
1411+}
1412+
1413+static int coroutine_fn vma_writer_flush(VmaWriter *vmaw)
1414+{
1415+ assert(vmaw);
1416+
1417+ int ret;
1418+ int i;
1419+
1420+ if (vmaw->status < 0) {
1421+ return vmaw->status;
1422+ }
1423+
1424+ if (!vmaw->header_written) {
1425+ vmaw->header_written = true;
1426+ ret = vma_write_header(vmaw);
1427+ if (ret < 0) {
1428+ vma_writer_set_error(vmaw, "vma_writer_flush: write header failed");
1429+ return ret;
1430+ }
1431+ }
1432+
1433+ DPRINTF("VMA WRITE FLUSH %d %d\n", vmaw->outbuf_count, vmaw->outbuf_pos);
1434+
1435+
1436+ VmaExtentHeader *ehead = (VmaExtentHeader *)vmaw->outbuf;
1437+
1438+ ehead->magic = VMA_EXTENT_MAGIC;
1439+ ehead->reserved1 = 0;
1440+
1441+ for (i = 0; i < VMA_BLOCKS_PER_EXTENT; i++) {
1442+ ehead->blockinfo[i] = GUINT64_TO_BE(vmaw->outbuf_block_info[i]);
1443+ }
1444+
1445+ guint16 block_count = (vmaw->outbuf_pos - VMA_EXTENT_HEADER_SIZE) /
1446+ VMA_BLOCK_SIZE;
1447+
1448+ ehead->block_count = GUINT16_TO_BE(block_count);
1449+
1450+ memcpy(ehead->uuid, vmaw->uuid, sizeof(ehead->uuid));
1451+ memset(ehead->md5sum, 0, sizeof(ehead->md5sum));
1452+
1453+ g_checksum_reset(vmaw->md5csum);
1454+ g_checksum_update(vmaw->md5csum, vmaw->outbuf, VMA_EXTENT_HEADER_SIZE);
1455+ gsize csize = 16;
1456+ g_checksum_get_digest(vmaw->md5csum, ehead->md5sum, &csize);
1457+
1458+ int bytes = vmaw->outbuf_pos;
1459+ ret = vma_queue_write(vmaw, vmaw->outbuf, bytes);
1460+ if (ret != bytes) {
1461+ vma_writer_set_error(vmaw, "vma_writer_flush: failed write");
1462+ }
1463+
1464+ vmaw->outbuf_count = 0;
1465+ vmaw->outbuf_pos = VMA_EXTENT_HEADER_SIZE;
1466+
1467+ for (i = 0; i < VMA_BLOCKS_PER_EXTENT; i++) {
1468+ vmaw->outbuf_block_info[i] = 0;
1469+ }
1470+
1471+ return vmaw->status;
1472+}
1473+
1474+static int vma_count_open_streams(VmaWriter *vmaw)
1475+{
1476+ g_assert(vmaw != NULL);
1477+
1478+ int i;
1479+ int open_drives = 0;
1480+ for (i = 0; i <= 255; i++) {
1481+ if (vmaw->stream_info[i].size && !vmaw->stream_info[i].finished) {
1482+ open_drives++;
1483+ }
1484+ }
1485+
1486+ return open_drives;
1487+}
1488+
1489+/**
1490+ * all jobs should call this when there is no more data
1491+ * Returns: number of remaining stream (0 ==> finished)
1492+ */
1493+int coroutine_fn
1494+vma_writer_close_stream(VmaWriter *vmaw, uint8_t dev_id)
1495+{
1496+ g_assert(vmaw != NULL);
1497+
1498+ DPRINTF("vma_writer_set_status %d\n", dev_id);
1499+ if (!vmaw->stream_info[dev_id].size) {
1500+ vma_writer_set_error(vmaw, "vma_writer_close_stream: "
1501+ "no such stream %d", dev_id);
1502+ return -1;
1503+ }
1504+ if (vmaw->stream_info[dev_id].finished) {
1505+ vma_writer_set_error(vmaw, "vma_writer_close_stream: "
1506+ "stream already closed %d", dev_id);
1507+ return -1;
1508+ }
1509+
1510+ vmaw->stream_info[dev_id].finished = true;
1511+
1512+ int open_drives = vma_count_open_streams(vmaw);
1513+
1514+ if (open_drives <= 0) {
1515+ DPRINTF("vma_writer_set_status all drives completed\n");
1516+ qemu_co_mutex_lock(&vmaw->flush_lock);
1517+ int ret = vma_writer_flush(vmaw);
1518+ qemu_co_mutex_unlock(&vmaw->flush_lock);
1519+ if (ret < 0) {
1520+ vma_writer_set_error(vmaw, "vma_writer_close_stream: flush failed");
1521+ }
1522+ }
1523+
1524+ return open_drives;
1525+}
1526+
1527+int vma_writer_get_status(VmaWriter *vmaw, VmaStatus *status)
1528+{
1529+ int i;
1530+
1531+ g_assert(vmaw != NULL);
1532+
1533+ if (status) {
1534+ status->status = vmaw->status;
1535+ g_strlcpy(status->errmsg, vmaw->errmsg, sizeof(status->errmsg));
1536+ for (i = 0; i <= 255; i++) {
1537+ status->stream_info[i] = vmaw->stream_info[i];
1538+ }
1539+
1540+ uuid_unparse_lower(vmaw->uuid, status->uuid_str);
1541+ }
1542+
1543+ status->closed = vmaw->closed;
1544+
1545+ return vmaw->status;
1546+}
1547+
1548+static int vma_writer_get_buffer(VmaWriter *vmaw)
1549+{
1550+ int ret = 0;
1551+
1552+ qemu_co_mutex_lock(&vmaw->flush_lock);
1553+
1554+ /* wait until buffer is available */
1555+ while (vmaw->outbuf_count >= (VMA_BLOCKS_PER_EXTENT - 1)) {
1556+ ret = vma_writer_flush(vmaw);
1557+ if (ret < 0) {
1558+ vma_writer_set_error(vmaw, "vma_writer_get_buffer: flush failed");
1559+ break;
1560+ }
1561+ }
1562+
1563+ qemu_co_mutex_unlock(&vmaw->flush_lock);
1564+
1565+ return ret;
1566+}
1567+
1568+
1569+int64_t coroutine_fn
1570+vma_writer_write(VmaWriter *vmaw, uint8_t dev_id, int64_t cluster_num,
1571+ unsigned char *buf, size_t *zero_bytes)
1572+{
1573+ g_assert(vmaw != NULL);
1574+ g_assert(zero_bytes != NULL);
1575+
1576+ *zero_bytes = 0;
1577+
1578+ if (vmaw->status < 0) {
1579+ return vmaw->status;
1580+ }
1581+
1582+ if (!dev_id || !vmaw->stream_info[dev_id].size) {
1583+ vma_writer_set_error(vmaw, "vma_writer_write: "
1584+ "no such stream %d", dev_id);
1585+ return -1;
1586+ }
1587+
1588+ if (vmaw->stream_info[dev_id].finished) {
1589+ vma_writer_set_error(vmaw, "vma_writer_write: "
1590+ "stream already closed %d", dev_id);
1591+ return -1;
1592+ }
1593+
1594+
1595+ if (cluster_num >= (((uint64_t)1)<<32)) {
1596+ vma_writer_set_error(vmaw, "vma_writer_write: "
1597+ "cluster number out of range");
1598+ return -1;
1599+ }
1600+
1601+ if (dev_id == vmaw->vmstate_stream) {
1602+ if (cluster_num != vmaw->vmstate_clusters) {
1603+ vma_writer_set_error(vmaw, "vma_writer_write: "
1604+ "non sequential vmstate write");
1605+ }
1606+ vmaw->vmstate_clusters++;
1607+ } else if (cluster_num >= vmaw->stream_info[dev_id].cluster_count) {
1608+ vma_writer_set_error(vmaw, "vma_writer_write: cluster number too big");
1609+ return -1;
1610+ }
1611+
1612+ /* wait until buffer is available */
1613+ if (vma_writer_get_buffer(vmaw) < 0) {
1614+ vma_writer_set_error(vmaw, "vma_writer_write: "
1615+ "vma_writer_get_buffer failed");
1616+ return -1;
1617+ }
1618+
1619+ DPRINTF("VMA WRITE %d %zd\n", dev_id, cluster_num);
1620+
1621+ uint16_t mask = 0;
1622+
1623+ if (buf) {
1624+ int i;
1625+ int bit = 1;
1626+ for (i = 0; i < 16; i++) {
1627+ unsigned char *vmablock = buf + (i*VMA_BLOCK_SIZE);
1628+ if (!buffer_is_zero(vmablock, VMA_BLOCK_SIZE)) {
1629+ mask |= bit;
1630+ memcpy(vmaw->outbuf + vmaw->outbuf_pos, vmablock,
1631+ VMA_BLOCK_SIZE);
1632+ vmaw->outbuf_pos += VMA_BLOCK_SIZE;
1633+ } else {
1634+ DPRINTF("VMA WRITE %zd ZERO BLOCK %d\n", cluster_num, i);
1635+ vmaw->stream_info[dev_id].zero_bytes += VMA_BLOCK_SIZE;
1636+ *zero_bytes += VMA_BLOCK_SIZE;
1637+ }
1638+
1639+ bit = bit << 1;
1640+ }
1641+ } else {
1642+ DPRINTF("VMA WRITE %zd ZERO CLUSTER\n", cluster_num);
1643+ vmaw->stream_info[dev_id].zero_bytes += VMA_CLUSTER_SIZE;
1644+ *zero_bytes += VMA_CLUSTER_SIZE;
1645+ }
1646+
1647+ uint64_t block_info = ((uint64_t)mask) << (32+16);
1648+ block_info |= ((uint64_t)dev_id) << 32;
1649+ block_info |= (cluster_num & 0xffffffff);
1650+ vmaw->outbuf_block_info[vmaw->outbuf_count] = block_info;
1651+
1652+ DPRINTF("VMA WRITE MASK %zd %zx\n", cluster_num, block_info);
1653+
1654+ vmaw->outbuf_count++;
1655+
1656+ /** NOTE: We allways write whole clusters, but we correctly set
1657+ * transferred bytes. So transferred == size when when everything
1658+ * went OK.
1659+ */
1660+ size_t transferred = VMA_CLUSTER_SIZE;
1661+
1662+ if (dev_id != vmaw->vmstate_stream) {
1663+ uint64_t last = (cluster_num + 1) * VMA_CLUSTER_SIZE;
1664+ if (last > vmaw->stream_info[dev_id].size) {
1665+ uint64_t diff = last - vmaw->stream_info[dev_id].size;
1666+ if (diff >= VMA_CLUSTER_SIZE) {
1667+ vma_writer_set_error(vmaw, "vma_writer_write: "
1668+ "read after last cluster");
1669+ return -1;
1670+ }
1671+ transferred -= diff;
1672+ }
1673+ }
1674+
1675+ vmaw->stream_info[dev_id].transferred += transferred;
1676+
1677+ return transferred;
1678+}
1679+
1680+int vma_writer_close(VmaWriter *vmaw, Error **errp)
1681+{
1682+ g_assert(vmaw != NULL);
1683+
1684+ int i;
1685+
1686+ vma_queue_flush(vmaw);
1687+
1688+ /* this should not happen - just to be sure */
1689+ while (!qemu_co_queue_empty(&vmaw->wqueue)) {
1690+ DPRINTF("vma_writer_close wait\n");
1691+ co_sleep_ns(QEMU_CLOCK_REALTIME, 1000000);
1692+ }
1693+
1694+ if (vmaw->cmd) {
1695+ if (pclose(vmaw->cmd) < 0) {
1696+ vma_writer_set_error(vmaw, "vma_writer_close: "
1697+ "pclose failed - %s", g_strerror(errno));
1698+ }
1699+ } else {
1700+ if (close(vmaw->fd) < 0) {
1701+ vma_writer_set_error(vmaw, "vma_writer_close: "
1702+ "close failed - %s", g_strerror(errno));
1703+ }
1704+ }
1705+
1706+ for (i = 0; i <= 255; i++) {
1707+ VmaStreamInfo *si = &vmaw->stream_info[i];
1708+ if (si->size) {
1709+ if (!si->finished) {
1710+ vma_writer_set_error(vmaw, "vma_writer_close: "
1711+ "detected open stream '%s'", si->devname);
1712+ } else if ((si->transferred != si->size) &&
1713+ (i != vmaw->vmstate_stream)) {
1714+ vma_writer_set_error(vmaw, "vma_writer_close: "
1715+ "incomplete stream '%s' (%zd != %zd)",
1716+ si->devname, si->transferred, si->size);
1717+ }
1718+ }
1719+ }
1720+
1721+ for (i = 0; i <= 255; i++) {
1722+ vmaw->stream_info[i].finished = 1; /* mark as closed */
1723+ }
1724+
1725+ vmaw->closed = 1;
1726+
1727+ if (vmaw->status < 0 && *errp == NULL) {
1728+ error_setg(errp, "%s", vmaw->errmsg);
1729+ }
1730+
1731+ return vmaw->status;
1732+}
1733+
1734+void vma_writer_destroy(VmaWriter *vmaw)
1735+{
1736+ assert(vmaw);
1737+
1738+ int i;
1739+
1740+ for (i = 0; i <= 255; i++) {
1741+ if (vmaw->stream_info[i].devname) {
1742+ g_free(vmaw->stream_info[i].devname);
1743+ }
1744+ }
1745+
1746+ if (vmaw->md5csum) {
1747+ g_checksum_free(vmaw->md5csum);
1748+ }
1749+
1750+ for (i = 0; i < WRITE_BUFFERS; i++) {
1751+ free(vmaw->aiocbs[i]);
1752+ }
1753+
1754+ g_free(vmaw);
1755+}
1756Index: new/vma.c
1757===================================================================
1758--- /dev/null 1970-01-01 00:00:00.000000000 +0000
2014bc7f 1759+++ new/vma.c 2013-11-29 12:10:53.000000000 +0100
b668dc80
DM		 1760@@ -0,0 +1,623 @@
1761+/*
1762+ * VMA: Virtual Machine Archive
1763+ *
1764+ * Copyright (C) 2012-2013 Proxmox Server Solutions
1765+ *
1766+ * Authors:
1767+ * Dietmar Maurer (dietmar@proxmox.com)
1768+ *
1769+ * This work is licensed under the terms of the GNU GPL, version 2 or later.
1770+ * See the COPYING file in the top-level directory.
1771+ *
1772+ */
1773+
1774+#include <stdio.h>
1775+#include <errno.h>
1776+#include <unistd.h>
1777+#include <stdio.h>
1778+#include <string.h>
1779+#include <sys/types.h>
1780+#include <sys/stat.h>
1781+#include <fcntl.h>
1782+#include <glib.h>
1783+
1784+#include "vma.h"
1785+#include "qemu-common.h"
1786+#include "qemu/error-report.h"
1787+#include "qemu/main-loop.h"
1788+
1789+static void help(void)
1790+{
1791+ const char *help_msg =
1792+ "usage: vma command [command options]\n"
1793+ "\n"
1794+ "vma list <filename>\n"
1795+ "vma create <filename> [-c config] <archive> pathname ...\n"
09a78de1 1796+ "vma extract <filename> [-r <fifo>] <targetdir>\n"
b668dc80
DM		 1797+ ;
1798+
1799+ printf("%s", help_msg);
1800+ exit(1);
1801+}
1802+
1803+static const char *extract_devname(const char *path, char **devname, int index)
1804+{
1805+ assert(path);
1806+
1807+ const char *sep = strchr(path, '=');
1808+
1809+ if (sep) {
1810+ *devname = g_strndup(path, sep - path);
1811+ path = sep + 1;
1812+ } else {
1813+ if (index >= 0) {
1814+ *devname = g_strdup_printf("disk%d", index);
1815+ } else {
1816+ *devname = NULL;
1817+ }
1818+ }
1819+
1820+ return path;
1821+}
1822+
1823+static void print_content(VmaReader *vmar)
1824+{
1825+ assert(vmar);
1826+
1827+ VmaHeader *head = vma_reader_get_header(vmar);
1828+
1829+ GList *l = vma_reader_get_config_data(vmar);
1830+ while (l && l->data) {
1831+ VmaConfigData *cdata = (VmaConfigData *)l->data;
1832+ l = g_list_next(l);
1833+ printf("CFG: size: %d name: %s\n", cdata->len, cdata->name);
1834+ }
1835+
1836+ int i;
1837+ VmaDeviceInfo *di;
1838+ for (i = 1; i < 255; i++) {
1839+ di = vma_reader_get_device_info(vmar, i);
1840+ if (di) {
1841+ if (strcmp(di->devname, "vmstate") == 0) {
1842+ printf("VMSTATE: dev_id=%d memory: %zd\n", i, di->size);
1843+ } else {
1844+ printf("DEV: dev_id=%d size: %zd devname: %s\n",
1845+ i, di->size, di->devname);
1846+ }
1847+ }
1848+ }
1849+ /* ctime is the last entry we print */
1850+ printf("CTIME: %s", ctime(&head->ctime));
1851+ fflush(stdout);
1852+}
1853+
1854+static int list_content(int argc, char **argv)
1855+{
1856+ int c, ret = 0;
1857+ const char *filename;
1858+
1859+ for (;;) {
1860+ c = getopt(argc, argv, "h");
1861+ if (c == -1) {
1862+ break;
1863+ }
1864+ switch (c) {
1865+ case '?':
1866+ case 'h':
1867+ help();
1868+ break;
1869+ default:
1870+ g_assert_not_reached();
1871+ }
1872+ }
1873+
1874+ /* Get the filename */
1875+ if ((optind + 1) != argc) {
1876+ help();
1877+ }
1878+ filename = argv[optind++];
1879+
1880+ Error *errp = NULL;
1881+ VmaReader *vmar = vma_reader_create(filename, &errp);
1882+
1883+ if (!vmar) {
1884+ g_error("%s", error_get_pretty(errp));
1885+ }
1886+
1887+ print_content(vmar);
1888+
1889+ vma_reader_destroy(vmar);
1890+
1891+ return ret;
1892+}
1893+
1894+typedef struct RestoreMap {
1895+ char *devname;
1896+ char *path;
1897+ bool write_zero;
1898+} RestoreMap;
1899+
1900+static int extract_content(int argc, char **argv)
1901+{
1902+ int c, ret = 0;
1903+ int verbose = 0;
1904+ const char *filename;
1905+ const char *dirname;
1906+ const char *readmap = NULL;
1907+
1908+ for (;;) {
1909+ c = getopt(argc, argv, "hvr:");
1910+ if (c == -1) {
1911+ break;
1912+ }
1913+ switch (c) {
1914+ case '?':
1915+ case 'h':
1916+ help();
1917+ break;
1918+ case 'r':
1919+ readmap = optarg;
1920+ break;
1921+ case 'v':
1922+ verbose = 1;
1923+ break;
1924+ default:
1925+ help();
1926+ }
1927+ }
1928+
1929+ /* Get the filename */
1930+ if ((optind + 2) != argc) {
1931+ help();
1932+ }
1933+ filename = argv[optind++];
1934+ dirname = argv[optind++];
1935+
1936+ Error *errp = NULL;
1937+ VmaReader *vmar = vma_reader_create(filename, &errp);
1938+
1939+ if (!vmar) {
1940+ g_error("%s", error_get_pretty(errp));
1941+ }
1942+
1943+ if (mkdir(dirname, 0777) < 0) {
1944+ g_error("unable to create target directory %s - %s",
1945+ dirname, g_strerror(errno));
1946+ }
1947+
1948+ GList *l = vma_reader_get_config_data(vmar);
1949+ while (l && l->data) {
1950+ VmaConfigData *cdata = (VmaConfigData *)l->data;
1951+ l = g_list_next(l);
1952+ char *cfgfn = g_strdup_printf("%s/%s", dirname, cdata->name);
1953+ GError *err = NULL;
1954+ if (!g_file_set_contents(cfgfn, (gchar *)cdata->data, cdata->len,
1955+ &err)) {
1956+ g_error("unable to write file: %s", err->message);
1957+ }
1958+ }
1959+
1960+ GHashTable *devmap = g_hash_table_new(g_str_hash, g_str_equal);
1961+
1962+ if (readmap) {
1963+ print_content(vmar);
1964+
1965+ FILE *map = fopen(readmap, "r");
1966+ if (!map) {
1967+ g_error("unable to open fifo %s - %s", readmap, g_strerror(errno));
1968+ }
1969+
1970+ while (1) {
1971+ char inbuf[8192];
1972+ char *line = fgets(inbuf, sizeof(inbuf), map);
1973+ if (!line || line[0] == '\0' || !strcmp(line, "done\n")) {
1974+ break;
1975+ }
1976+ int len = strlen(line);
1977+ if (line[len - 1] == '\n') {
1978+ line[len - 1] = '\0';
1979+ if (len == 1) {
1980+ break;
1981+ }
1982+ }
1983+
1984+ const char *path;
1985+ bool write_zero;
1986+ if (line[0] == '0' && line[1] == ':') {
1987+ path = inbuf + 2;
1988+ write_zero = false;
1989+ } else if (line[0] == '1' && line[1] == ':') {
1990+ path = inbuf + 2;
1991+ write_zero = true;
1992+ } else {
1993+ g_error("read map failed - parse error ('%s')", inbuf);
1994+ }
1995+
1996+ char *devname = NULL;
1997+ path = extract_devname(path, &devname, -1);
1998+ if (!devname) {
1999+ g_error("read map failed - no dev name specified ('%s')",
2000+ inbuf);
2001+ }
2002+
2003+ RestoreMap *map = g_new0(RestoreMap, 1);
2004+ map->devname = g_strdup(devname);
2005+ map->path = g_strdup(path);
2006+ map->write_zero = write_zero;
2007+
2008+ g_hash_table_insert(devmap, map->devname, map);
2009+
2010+ };
2011+ }
2012+
2013+ int i;
2014+ int vmstate_fd = -1;
2015+ guint8 vmstate_stream = 0;
2016+
2017+ for (i = 1; i < 255; i++) {
2018+ VmaDeviceInfo *di = vma_reader_get_device_info(vmar, i);
2019+ if (di && (strcmp(di->devname, "vmstate") == 0)) {
2020+ vmstate_stream = i;
2021+ char *statefn = g_strdup_printf("%s/vmstate.bin", dirname);
2022+ vmstate_fd = open(statefn, O_WRONLY|O_CREAT|O_EXCL, 0644);
2023+ if (vmstate_fd < 0) {
2024+ g_error("create vmstate file '%s' failed - %s", statefn,
2025+ g_strerror(errno));
2026+ }
2027+ g_free(statefn);
2028+ } else if (di) {
2029+ char *devfn = NULL;
2030+ int flags = BDRV_O_RDWR|BDRV_O_CACHE_WB;
2031+ bool write_zero = true;
2032+
2033+ if (readmap) {
2034+ RestoreMap *map;
2035+ map = (RestoreMap *)g_hash_table_lookup(devmap, di->devname);
2036+ if (map == NULL) {
2037+ g_error("no device name mapping for %s", di->devname);
2038+ }
2039+ devfn = map->path;
2040+ write_zero = map->write_zero;
2041+ } else {
2042+ devfn = g_strdup_printf("%s/tmp-disk-%s.raw",
2043+ dirname, di->devname);
2044+ printf("DEVINFO %s %zd\n", devfn, di->size);
2045+
2046+ bdrv_img_create(devfn, "raw", NULL, NULL, NULL, di->size,
2047+ flags, &errp, 0);
2048+ if (error_is_set(&errp)) {
2049+ g_error("can't create file %s: %s", devfn,
2050+ error_get_pretty(errp));
2051+ }
2052+
2053+ /* Note: we created an empty file above, so there is no
2054+ * need to write zeroes (so we generate a sparse file)
2055+ */
2056+ write_zero = false;
2057+ }
2058+
2059+ BlockDriverState *bs = bdrv_new(di->devname);
2060+ if (bdrv_open(bs, devfn, NULL, flags, NULL, &errp)) {
2061+ g_error("can't open file %s - %s", devfn,
2062+ error_get_pretty(errp));
2063+ }
2064+ if (vma_reader_register_bs(vmar, i, bs, write_zero, &errp) < 0) {
2065+ g_error("%s", error_get_pretty(errp));
2066+ }
2067+
2068+ if (!readmap) {
2069+ g_free(devfn);
2070+ }
2071+ }
2072+ }
2073+
2074+ if (vma_reader_restore(vmar, vmstate_fd, verbose, &errp) < 0) {
2075+ g_error("restore failed - %s", error_get_pretty(errp));
2076+ }
2077+
2078+ if (!readmap) {
2079+ for (i = 1; i < 255; i++) {
2080+ VmaDeviceInfo *di = vma_reader_get_device_info(vmar, i);
2081+ if (di && (i != vmstate_stream)) {
2082+ char *tmpfn = g_strdup_printf("%s/tmp-disk-%s.raw",
2083+ dirname, di->devname);
2084+ char *fn = g_strdup_printf("%s/disk-%s.raw",
2085+ dirname, di->devname);
2086+ if (rename(tmpfn, fn) != 0) {
2087+ g_error("rename %s to %s failed - %s",
2088+ tmpfn, fn, g_strerror(errno));
2089+ }
2090+ }
2091+ }
2092+ }
2093+
2094+ vma_reader_destroy(vmar);
2095+
2096+ bdrv_close_all();
2097+
2098+ return ret;
2099+}
2100+
2101+typedef struct BackupJob {
2102+ BlockDriverState *bs;
2103+ int64_t len;
2104+ VmaWriter *vmaw;
2105+ uint8_t dev_id;
2106+} BackupJob;
2107+
2108+/*
2109+static int backup_dump_cb(void *opaque, BlockDriverState *bs,
2110+ int64_t cluster_num, unsigned char *buf)
2111+{
2112+ BackupCB *bcb = opaque;
2113+ size_t zb = 0;
2114+ if (vma_writer_write(bcb->vmaw, bcb->dev_id, cluster_num, buf, &zb) < 0) {
2115+ g_warning("backup_dump_cb vma_writer_write failed");
2116+ return -1;
2117+ }
2118+
2119+ return 0;
2120+}
2121+
2122+static void backup_complete_cb(void *opaque, int ret)
2123+{
2124+ BackupCB *bcb = opaque;
2125+
2126+ if (ret < 0) {
2127+ vma_writer_set_error(bcb->vmaw, "backup_complete_cb %d", ret);
2128+ }
2129+
2130+ if (vma_writer_close_stream(bcb->vmaw, bcb->dev_id) <= 0) {
2131+ Error *err = NULL;
2132+ if (vma_writer_close(bcb->vmaw, &err) != 0) {
2133+ g_warning("vma_writer_close failed %s", error_get_pretty(err));
2134+ }
2135+ }
2136+}
2137+*/
2138+
2139+#define BACKUP_SECTORS_PER_CLUSTER (VMA_CLUSTER_SIZE / BDRV_SECTOR_SIZE)
2140+
2141+static void coroutine_fn backup_run(void *opaque)
2142+{
2143+ BackupJob *job = (BackupJob *)opaque;
2144+ struct iovec iov;
2145+ QEMUIOVector qiov;
2146+
2147+ int64_t start, end;
2148+ int ret = 0;
2149+
2150+ unsigned char *buf = qemu_blockalign(job->bs, VMA_CLUSTER_SIZE);
2151+
2152+ start = 0;
2153+ end = DIV_ROUND_UP(job->len / BDRV_SECTOR_SIZE,
2154+ BACKUP_SECTORS_PER_CLUSTER);
2155+
2156+ for (; start < end; start++) {
2157+ iov.iov_base = buf;
2158+ iov.iov_len = VMA_CLUSTER_SIZE;
2159+ qemu_iovec_init_external(&qiov, &iov, 1);
2160+
2161+ ret = bdrv_co_readv(job->bs, start * BACKUP_SECTORS_PER_CLUSTER,
2162+ BACKUP_SECTORS_PER_CLUSTER, &qiov);
2163+ if (ret < 0) {
2164+ vma_writer_set_error(job->vmaw, "read error", -1);
2165+ goto out;
2166+ }
2167+
2168+ size_t zb = 0;
2169+ if (vma_writer_write(job->vmaw, job->dev_id, start, buf, &zb) < 0) {
2170+ vma_writer_set_error(job->vmaw, "backup_dump_cb vma_writer_write failed", -1);
2171+ goto out;
2172+ }
2173+ }
2174+
2175+
2176+out:
2177+ if (vma_writer_close_stream(job->vmaw, job->dev_id) <= 0) {
2178+ Error *err = NULL;
2179+ if (vma_writer_close(job->vmaw, &err) != 0) {
2180+ g_warning("vma_writer_close failed %s", error_get_pretty(err));
2181+ }
2182+ }
2183+}
2184+
2185+static int create_archive(int argc, char **argv)
2186+{
2187+ int i, c, res;
2188+ int verbose = 0;
2189+ const char *archivename;
2190+ GList *config_files = NULL;
2191+
2192+ for (;;) {
2193+ c = getopt(argc, argv, "hvc:");
2194+ if (c == -1) {
2195+ break;
2196+ }
2197+ switch (c) {
2198+ case '?':
2199+ case 'h':
2200+ help();
2201+ break;
2202+ case 'c':
2203+ config_files = g_list_append(config_files, optarg);
2204+ break;
2205+ case 'v':
2206+ verbose = 1;
2207+ break;
2208+ default:
2209+ g_assert_not_reached();
2210+ }
2211+ }
2212+
2213+
2214+ /* make sure we have archive name and at least one path */
2215+ if ((optind + 2) > argc) {
2216+ help();
2217+ }
2218+
2219+ archivename = argv[optind++];
2220+
2221+ uuid_t uuid;
2222+ uuid_generate(uuid);
2223+
2224+ Error *local_err = NULL;
2225+ VmaWriter *vmaw = vma_writer_create(archivename, uuid, &local_err);
2226+
2227+ if (vmaw == NULL) {
2228+ g_error("%s", error_get_pretty(local_err));
2229+ }
2230+
2231+ GList *l = config_files;
2232+ while (l && l->data) {
2233+ char *name = l->data;
2234+ char *cdata = NULL;
2235+ gsize clen = 0;
2236+ GError *err = NULL;
2237+ if (!g_file_get_contents(name, &cdata, &clen, &err)) {
2238+ unlink(archivename);
2239+ g_error("Unable to read file: %s", err->message);
2240+ }
2241+
2242+ if (vma_writer_add_config(vmaw, name, cdata, clen) != 0) {
2243+ unlink(archivename);
2244+ g_error("Unable to append config data %s (len = %zd)",
2245+ name, clen);
2246+ }
2247+ l = g_list_next(l);
2248+ }
2249+
2250+ int ind = 0;
2251+ while (optind < argc) {
2252+ const char *path = argv[optind++];
2253+ char *devname = NULL;
2254+ path = extract_devname(path, &devname, ind++);
2255+
2256+ BlockDriver *drv = NULL;
2257+ BlockDriverState *bs = bdrv_new(devname);
2258+ Error *errp = NULL;
2259+
2260+ res = bdrv_open(bs, path, NULL, BDRV_O_CACHE_WB , drv, &errp);
2261+ if (res < 0) {
2262+ unlink(archivename);
2263+ g_error("bdrv_open '%s' failed - %s", path, error_get_pretty(errp));
2264+ }
2265+ int64_t size = bdrv_getlength(bs);
2266+ int dev_id = vma_writer_register_stream(vmaw, devname, size);
2267+ if (dev_id <= 0) {
2268+ unlink(archivename);
2269+ g_error("vma_writer_register_stream '%s' failed", devname);
2270+ }
2271+
2272+ BackupJob *job = g_new0(BackupJob, 1);
2273+ job->len = size;
2274+ job->bs = bs;
2275+ job->vmaw = vmaw;
2276+ job->dev_id = dev_id;
2277+
2278+ Coroutine *co = qemu_coroutine_create(backup_run);
2279+ qemu_coroutine_enter(co, job);
2280+
2281+ //BlockDriverState *target = NULL;
2282+ //backup_start(bs, target, 0, MIRROR_SYNC_MODE_FULL,
2283+ // BLOCKDEV_ON_ERROR_STOP, BLOCKDEV_ON_ERROR_STOP,
2284+ // NULL, bcb, &errp);
2285+
2286+ //if (backup_job_create(bs, backup_dump_cb, backup_complete_cb,
2287+ // bcb, 0) < 0) {
2288+ // unlink(archivename);
2289+ // g_error("backup_job_start failed");
2290+ //} else {
2291+ // backup_job_start(bs, false);
2292+ // }
2293+ }
2294+
2295+ VmaStatus vmastat;
2296+ int percent = 0;
2297+ int last_percent = -1;
2298+
2299+ while (1) {
2300+ main_loop_wait(false);
2301+ vma_writer_get_status(vmaw, &vmastat);
2302+
2303+ if (verbose) {
2304+
2305+ uint64_t total = 0;
2306+ uint64_t transferred = 0;
2307+ uint64_t zero_bytes = 0;
2308+
2309+ int i;
2310+ for (i = 0; i < 256; i++) {
2311+ if (vmastat.stream_info[i].size) {
2312+ total += vmastat.stream_info[i].size;
2313+ transferred += vmastat.stream_info[i].transferred;
2314+ zero_bytes += vmastat.stream_info[i].zero_bytes;
2315+ }
2316+ }
2317+ percent = (transferred*100)/total;
2318+ if (percent != last_percent) {
2319+ fprintf(stderr, "progress %d%% %zd/%zd %zd\n", percent,
2320+ transferred, total, zero_bytes);
2321+ fflush(stderr);
2322+
2323+ last_percent = percent;
2324+ }
2325+ }
2326+
2327+ if (vmastat.closed) {
2328+ break;
2329+ }
2330+ }
2331+
2332+ bdrv_drain_all();
2333+
2334+ vma_writer_get_status(vmaw, &vmastat);
2335+
2336+ if (verbose) {
2337+ for (i = 0; i < 256; i++) {
2338+ VmaStreamInfo *si = &vmastat.stream_info[i];
2339+ if (si->size) {
2340+ fprintf(stderr, "image %s: size=%zd zeros=%zd saved=%zd\n",
2341+ si->devname, si->size, si->zero_bytes,
2342+ si->size - si->zero_bytes);
2343+ }
2344+ }
2345+ }
2346+
2347+ if (vmastat.status < 0) {
2348+ unlink(archivename);
2349+ g_error("creating vma archive failed");
2350+ }
2351+
2352+ return 0;
2353+}
2354+
2355+int main(int argc, char **argv)
2356+{
2357+ const char *cmdname;
2358+
2359+ error_set_progname(argv[0]);
2360+
2361+ qemu_init_main_loop();
2362+
2363+ bdrv_init();
2364+
2365+ if (argc < 2) {
2366+ help();
2367+ }
2368+
2369+ cmdname = argv[1];
2370+ argc--; argv++;
2371+
2372+
2373+ if (!strcmp(cmdname, "list")) {
2374+ return list_content(argc, argv);
2375+ } else if (!strcmp(cmdname, "create")) {
2376+ return create_archive(argc, argv);
2377+ } else if (!strcmp(cmdname, "extract")) {
2378+ return extract_content(argc, argv);
2379+ }
2380+
2381+ help();
2382+ return 0;
2383+}
2384Index: new/vma.h
2385===================================================================
2386--- /dev/null 1970-01-01 00:00:00.000000000 +0000
2014bc7f 2387+++ new/vma.h 2013-11-29 12:10:53.000000000 +0100
b668dc80
DM		 2388@@ -0,0 +1,144 @@
2389+/*
2390+ * VMA: Virtual Machine Archive
2391+ *
2392+ * Copyright (C) Proxmox Server Solutions
2393+ *
2394+ * Authors:
2395+ * Dietmar Maurer (dietmar@proxmox.com)
2396+ *
2397+ * This work is licensed under the terms of the GNU GPL, version 2 or later.
2398+ * See the COPYING file in the top-level directory.
2399+ *
2400+ */
2401+
2402+#ifndef BACKUP_VMA_H
2403+#define BACKUP_VMA_H
2404+
2405+#include <uuid/uuid.h>
2406+#include "qapi/error.h"
2407+#include "block/block.h"
2408+
2409+#define VMA_BLOCK_BITS 12
2410+#define VMA_BLOCK_SIZE (1<<VMA_BLOCK_BITS)
2411+#define VMA_CLUSTER_BITS (VMA_BLOCK_BITS+4)
2412+#define VMA_CLUSTER_SIZE (1<<VMA_CLUSTER_BITS)
2413+
2414+#if VMA_CLUSTER_SIZE != 65536
2415+#error unexpected cluster size
2416+#endif
2417+
2418+#define VMA_EXTENT_HEADER_SIZE 512
2419+#define VMA_BLOCKS_PER_EXTENT 59
2420+#define VMA_MAX_CONFIGS 256
2421+
2422+#define VMA_MAX_EXTENT_SIZE \
2423+ (VMA_EXTENT_HEADER_SIZE+VMA_CLUSTER_SIZE*VMA_BLOCKS_PER_EXTENT)
2424+#if VMA_MAX_EXTENT_SIZE != 3867136
2425+#error unexpected VMA_EXTENT_SIZE
2426+#endif
2427+
2428+/* File Format Definitions */
2429+
2430+#define VMA_MAGIC (GUINT32_TO_BE(('V'<<24)|('M'<<16)|('A'<<8)|0x00))
2431+#define VMA_EXTENT_MAGIC (GUINT32_TO_BE(('V'<<24)|('M'<<16)|('A'<<8)|'E'))
2432+
2433+typedef struct VmaDeviceInfoHeader {
2434+ uint32_t devname_ptr; /* offset into blob_buffer table */
2435+ uint32_t reserved0;
2436+ uint64_t size; /* device size in bytes */
2437+ uint64_t reserved1;
2438+ uint64_t reserved2;
2439+} VmaDeviceInfoHeader;
2440+
2441+typedef struct VmaHeader {
2442+ uint32_t magic;
2443+ uint32_t version;
2444+ unsigned char uuid[16];
2445+ int64_t ctime;
2446+ unsigned char md5sum[16];
2447+
2448+ uint32_t blob_buffer_offset;
2449+ uint32_t blob_buffer_size;
2450+ uint32_t header_size;
2451+
2452+ unsigned char reserved[1984];
2453+
2454+ uint32_t config_names[VMA_MAX_CONFIGS]; /* offset into blob_buffer table */
2455+ uint32_t config_data[VMA_MAX_CONFIGS]; /* offset into blob_buffer table */
2456+
2457+ VmaDeviceInfoHeader dev_info[256];
2458+} VmaHeader;
2459+
2460+typedef struct VmaExtentHeader {
2461+ uint32_t magic;
2462+ uint16_t reserved1;
2463+ uint16_t block_count;
2464+ unsigned char uuid[16];
2465+ unsigned char md5sum[16];
2466+ uint64_t blockinfo[VMA_BLOCKS_PER_EXTENT];
2467+} VmaExtentHeader;
2468+
2469+/* functions/definitions to read/write vma files */
2470+
2471+typedef struct VmaReader VmaReader;
2472+
2473+typedef struct VmaWriter VmaWriter;
2474+
2475+typedef struct VmaConfigData {
2476+ const char *name;
2477+ const void *data;
2478+ uint32_t len;
2479+} VmaConfigData;
2480+
2481+typedef struct VmaStreamInfo {
2482+ uint64_t size;
2483+ uint64_t cluster_count;
2484+ uint64_t transferred;
2485+ uint64_t zero_bytes;
2486+ int finished;
2487+ char *devname;
2488+} VmaStreamInfo;
2489+
2490+typedef struct VmaStatus {
2491+ int status;
2492+ bool closed;
2493+ char errmsg[8192];
2494+ char uuid_str[37];
2495+ VmaStreamInfo stream_info[256];
2496+} VmaStatus;
2497+
2498+typedef struct VmaDeviceInfo {
2499+ uint64_t size; /* device size in bytes */
2500+ const char *devname;
2501+} VmaDeviceInfo;
2502+
2503+VmaWriter *vma_writer_create(const char *filename, uuid_t uuid, Error **errp);
2504+int vma_writer_close(VmaWriter *vmaw, Error **errp);
2505+void vma_writer_destroy(VmaWriter *vmaw);
2506+int vma_writer_add_config(VmaWriter *vmaw, const char *name, gpointer data,
2507+ size_t len);
2508+int vma_writer_register_stream(VmaWriter *vmaw, const char *devname,
2509+ size_t size);
2510+
2511+int64_t coroutine_fn vma_writer_write(VmaWriter *vmaw, uint8_t dev_id,
2512+ int64_t cluster_num, unsigned char *buf,
2513+ size_t *zero_bytes);
2514+
2515+int coroutine_fn vma_writer_close_stream(VmaWriter *vmaw, uint8_t dev_id);
2516+
2517+int vma_writer_get_status(VmaWriter *vmaw, VmaStatus *status);
2518+void vma_writer_set_error(VmaWriter *vmaw, const char *fmt, ...);
2519+
2520+
2521+VmaReader *vma_reader_create(const char *filename, Error **errp);
2522+void vma_reader_destroy(VmaReader *vmar);
2523+VmaHeader *vma_reader_get_header(VmaReader *vmar);
2524+GList *vma_reader_get_config_data(VmaReader *vmar);
2525+VmaDeviceInfo *vma_reader_get_device_info(VmaReader *vmar, guint8 dev_id);
2526+int vma_reader_register_bs(VmaReader *vmar, guint8 dev_id,
2527+ BlockDriverState *bs, bool write_zeroes,
2528+ Error **errp);
2529+int vma_reader_restore(VmaReader *vmar, int vmstate_fd, bool verbose,
2530+ Error **errp);
2531+
2532+#endif /* BACKUP_VMA_H */
2014bc7f
DM		 2533Index: new/Makefile.objs
2534===================================================================
2535--- new.orig/Makefile.objs 2013-11-29 12:10:45.000000000 +0100
2536+++ new/Makefile.objs 2013-11-29 12:24:55.000000000 +0100
2537@@ -14,6 +14,7 @@
2538 block-obj-y += block/
2539 block-obj-y += qapi-types.o qapi-visit.o
2540 block-obj-y += qemu-io-cmds.o
2541+block-obj-y += vma-writer.o
2542
2543 block-obj-y += qemu-coroutine.o qemu-coroutine-lock.o qemu-coroutine-io.o
2544 block-obj-y += qemu-coroutine-sleep.o
old QEMU for PVE