| blob | 6fbd7adfc26c61e4e44b4fdffa4bbe8e130d7ab0 |
1 From 6289a43696ca6f713a5d3bb9f95a5adb608a5e13 Mon Sep 17 00:00:00 2001
2 From: Dietmar Maurer <dietmar@proxmox.com>
3 Date: Tue, 13 Nov 2012 11:11:38 +0100
4 Subject: [PATCH v5 4/6] introduce new vma archive format
6 This is a very simple archive format, see docs/specs/vma_spec.txt
8 Signed-off-by: Dietmar Maurer <dietmar@proxmox.com>
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
25 Index: new/Makefile
26 ===================================================================
27 --- new.orig/Makefile 2014-11-20 08:12:54.000000000 +0100
28 +++ new/Makefile 2014-11-20 08:15:12.000000000 +0100
29 @@ -156,7 +156,7 @@
30 include $(SRC_PATH)/libcacard/Makefile
31 endif
33 -all: $(DOCS) $(TOOLS) $(HELPERS-y) recurse-all modules
34 +all: $(DOCS) $(TOOLS) vma$(EXESUF) $(HELPERS-y) recurse-all modules
36 config-host.h: config-host.h-timestamp
37 config-host.h-timestamp: config-host.mak
38 @@ -224,6 +224,7 @@
39 qemu-img$(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 $(block-obj-y) libqemuutil.a libqemustub.a
42 +vma$(EXESUF): vma.o vma-reader.o $(block-obj-y) libqemuutil.a libqemustub.a
44 qemu-bridge-helper$(EXESUF): qemu-bridge-helper.o
46 Index: new/vma-reader.c
47 ===================================================================
48 --- /dev/null 1970-01-01 00:00:00.000000000 +0000
49 +++ new/vma-reader.c 2014-11-20 08:15:12.000000000 +0100
50 @@ -0,0 +1,799 @@
51 +/*
52 + * VMA: Virtual Machine Archive
53 + *
54 + * Copyright (C) 2012 Proxmox Server Solutions
55 + *
56 + * Authors:
57 + * Dietmar Maurer (dietmar@proxmox.com)
58 + *
59 + * This work is licensed under the terms of the GNU GPL, version 2 or later.
60 + * See the COPYING file in the top-level directory.
61 + *
62 + */
63 +
64 +#include <stdio.h>
65 +#include <errno.h>
66 +#include <unistd.h>
67 +#include <stdio.h>
68 +#include <string.h>
69 +#include <sys/types.h>
70 +#include <sys/stat.h>
71 +#include <fcntl.h>
72 +#include <glib.h>
73 +#include <uuid/uuid.h>
74 +
75 +#include "qemu-common.h"
76 +#include "qemu/timer.h"
77 +#include "qemu/ratelimit.h"
78 +#include "vma.h"
79 +#include "block/block.h"
80 +
81 +#define BITS_PER_LONG (sizeof(unsigned long) * CHAR_BIT)
82 +
83 +static unsigned char zero_vma_block[VMA_BLOCK_SIZE];
84 +
85 +typedef struct VmaRestoreState {
86 + BlockDriverState *bs;
87 + bool write_zeroes;
88 + unsigned long *bitmap;
89 + int bitmap_size;
90 +} VmaRestoreState;
91 +
92 +struct VmaReader {
93 + int fd;
94 + GChecksum *md5csum;
95 + GHashTable *blob_hash;
96 + unsigned char *head_data;
97 + VmaDeviceInfo devinfo[256];
98 + VmaRestoreState rstate[256];
99 + GList *cdata_list;
100 + guint8 vmstate_stream;
101 + uint32_t vmstate_clusters;
102 + /* to show restore percentage if run with -v */
103 + time_t start_time;
104 + int64_t cluster_count;
105 + int64_t clusters_read;
106 + int clusters_read_per;
107 +};
108 +
109 +static guint
110 +g_int32_hash(gconstpointer v)
111 +{
112 + return *(const uint32_t *)v;
113 +}
114 +
115 +static gboolean
116 +g_int32_equal(gconstpointer v1, gconstpointer v2)
117 +{
118 + return *((const uint32_t *)v1) == *((const uint32_t *)v2);
119 +}
120 +
121 +static int vma_reader_get_bitmap(VmaRestoreState *rstate, int64_t cluster_num)
122 +{
123 + assert(rstate);
124 + assert(rstate->bitmap);
125 +
126 + unsigned long val, idx, bit;
127 +
128 + idx = cluster_num / BITS_PER_LONG;
129 +
130 + assert(rstate->bitmap_size > idx);
131 +
132 + bit = cluster_num % BITS_PER_LONG;
133 + val = rstate->bitmap[idx];
134 +
135 + return !!(val & (1UL << bit));
136 +}
137 +
138 +static void vma_reader_set_bitmap(VmaRestoreState *rstate, int64_t cluster_num,
139 + int dirty)
140 +{
141 + assert(rstate);
142 + assert(rstate->bitmap);
143 +
144 + unsigned long val, idx, bit;
145 +
146 + idx = cluster_num / BITS_PER_LONG;
147 +
148 + assert(rstate->bitmap_size > idx);
149 +
150 + bit = cluster_num % BITS_PER_LONG;
151 + val = rstate->bitmap[idx];
152 + if (dirty) {
153 + if (!(val & (1UL << bit))) {
154 + val |= 1UL << bit;
155 + }
156 + } else {
157 + if (val & (1UL << bit)) {
158 + val &= ~(1UL << bit);
159 + }
160 + }
161 + rstate->bitmap[idx] = val;
162 +}
163 +
164 +typedef struct VmaBlob {
165 + uint32_t start;
166 + uint32_t len;
167 + void *data;
168 +} VmaBlob;
169 +
170 +static const VmaBlob *get_header_blob(VmaReader *vmar, uint32_t pos)
171 +{
172 + assert(vmar);
173 + assert(vmar->blob_hash);
174 +
175 + return g_hash_table_lookup(vmar->blob_hash, &pos);
176 +}
177 +
178 +static const char *get_header_str(VmaReader *vmar, uint32_t pos)
179 +{
180 + const VmaBlob *blob = get_header_blob(vmar, pos);
181 + if (!blob) {
182 + return NULL;
183 + }
184 + const char *res = (char *)blob->data;
185 + if (res[blob->len-1] != '\0') {
186 + return NULL;
187 + }
188 + return res;
189 +}
190 +
191 +static ssize_t
192 +safe_read(int fd, unsigned char *buf, size_t count)
193 +{
194 + ssize_t n;
195 +
196 + do {
197 + n = read(fd, buf, count);
198 + } while (n < 0 && errno == EINTR);
199 +
200 + return n;
201 +}
202 +
203 +static ssize_t
204 +full_read(int fd, unsigned char *buf, size_t len)
205 +{
206 + ssize_t n;
207 + size_t total;
208 +
209 + total = 0;
210 +
211 + while (len > 0) {
212 + n = safe_read(fd, buf, len);
213 +
214 + if (n == 0) {
215 + return total;
216 + }
217 +
218 + if (n <= 0) {
219 + break;
220 + }
221 +
222 + buf += n;
223 + total += n;
224 + len -= n;
225 + }
226 +
227 + if (len) {
228 + return -1;
229 + }
230 +
231 + return total;
232 +}
233 +
234 +void vma_reader_destroy(VmaReader *vmar)
235 +{
236 + assert(vmar);
237 +
238 + if (vmar->fd >= 0) {
239 + close(vmar->fd);
240 + }
241 +
242 + if (vmar->cdata_list) {
243 + g_list_free(vmar->cdata_list);
244 + }
245 +
246 + int i;
247 + for (i = 1; i < 256; i++) {
248 + if (vmar->rstate[i].bitmap) {
249 + g_free(vmar->rstate[i].bitmap);
250 + }
251 + }
252 +
253 + if (vmar->md5csum) {
254 + g_checksum_free(vmar->md5csum);
255 + }
256 +
257 + if (vmar->blob_hash) {
258 + g_hash_table_destroy(vmar->blob_hash);
259 + }
260 +
261 + if (vmar->head_data) {
262 + g_free(vmar->head_data);
263 + }
264 +
265 + g_free(vmar);
266 +
267 +};
268 +
269 +static int vma_reader_read_head(VmaReader *vmar, Error **errp)
270 +{
271 + assert(vmar);
272 + assert(errp);
273 + assert(*errp == NULL);
274 +
275 + unsigned char md5sum[16];
276 + int i;
277 + int ret = 0;
278 +
279 + vmar->head_data = g_malloc(sizeof(VmaHeader));
280 +
281 + if (full_read(vmar->fd, vmar->head_data, sizeof(VmaHeader)) !=
282 + sizeof(VmaHeader)) {
283 + error_setg(errp, "can't read vma header - %s",
284 + errno ? g_strerror(errno) : "got EOF");
285 + return -1;
286 + }
287 +
288 + VmaHeader *h = (VmaHeader *)vmar->head_data;
289 +
290 + if (h->magic != VMA_MAGIC) {
291 + error_setg(errp, "not a vma file - wrong magic number");
292 + return -1;
293 + }
294 +
295 + uint32_t header_size = GUINT32_FROM_BE(h->header_size);
296 + int need = header_size - sizeof(VmaHeader);
297 + if (need <= 0) {
298 + error_setg(errp, "wrong vma header size %d", header_size);
299 + return -1;
300 + }
301 +
302 + vmar->head_data = g_realloc(vmar->head_data, header_size);
303 + h = (VmaHeader *)vmar->head_data;
304 +
305 + if (full_read(vmar->fd, vmar->head_data + sizeof(VmaHeader), need) !=
306 + need) {
307 + error_setg(errp, "can't read vma header data - %s",
308 + errno ? g_strerror(errno) : "got EOF");
309 + return -1;
310 + }
311 +
312 + memcpy(md5sum, h->md5sum, 16);
313 + memset(h->md5sum, 0, 16);
314 +
315 + g_checksum_reset(vmar->md5csum);
316 + g_checksum_update(vmar->md5csum, vmar->head_data, header_size);
317 + gsize csize = 16;
318 + g_checksum_get_digest(vmar->md5csum, (guint8 *)(h->md5sum), &csize);
319 +
320 + if (memcmp(md5sum, h->md5sum, 16) != 0) {
321 + error_setg(errp, "wrong vma header chechsum");
322 + return -1;
323 + }
324 +
325 + /* we can modify header data after checksum verify */
326 + h->header_size = header_size;
327 +
328 + h->version = GUINT32_FROM_BE(h->version);
329 + if (h->version != 1) {
330 + error_setg(errp, "wrong vma version %d", h->version);
331 + return -1;
332 + }
333 +
334 + h->ctime = GUINT64_FROM_BE(h->ctime);
335 + h->blob_buffer_offset = GUINT32_FROM_BE(h->blob_buffer_offset);
336 + h->blob_buffer_size = GUINT32_FROM_BE(h->blob_buffer_size);
337 +
338 + uint32_t bstart = h->blob_buffer_offset + 1;
339 + uint32_t bend = h->blob_buffer_offset + h->blob_buffer_size;
340 +
341 + if (bstart <= sizeof(VmaHeader)) {
342 + error_setg(errp, "wrong vma blob buffer offset %d",
343 + h->blob_buffer_offset);
344 + return -1;
345 + }
346 +
347 + if (bend > header_size) {
348 + error_setg(errp, "wrong vma blob buffer size %d/%d",
349 + h->blob_buffer_offset, h->blob_buffer_size);
350 + return -1;
351 + }
352 +
353 + while ((bstart + 2) <= bend) {
354 + uint32_t size = vmar->head_data[bstart] +
355 + (vmar->head_data[bstart+1] << 8);
356 + if ((bstart + size + 2) <= bend) {
357 + VmaBlob *blob = g_new0(VmaBlob, 1);
358 + blob->start = bstart - h->blob_buffer_offset;
359 + blob->len = size;
360 + blob->data = vmar->head_data + bstart + 2;
361 + g_hash_table_insert(vmar->blob_hash, &blob->start, blob);
362 + }
363 + bstart += size + 2;
364 + }
365 +
366 +
367 + int count = 0;
368 + for (i = 1; i < 256; i++) {
369 + VmaDeviceInfoHeader *dih = &h->dev_info[i];
370 + uint32_t devname_ptr = GUINT32_FROM_BE(dih->devname_ptr);
371 + uint64_t size = GUINT64_FROM_BE(dih->size);
372 + const char *devname = get_header_str(vmar, devname_ptr);
373 +
374 + if (size && devname) {
375 + count++;
376 + vmar->devinfo[i].size = size;
377 + vmar->devinfo[i].devname = devname;
378 +
379 + if (strcmp(devname, "vmstate") == 0) {
380 + vmar->vmstate_stream = i;
381 + }
382 + }
383 + }
384 +
385 + if (!count) {
386 + error_setg(errp, "vma does not contain data");
387 + return -1;
388 + }
389 +
390 + for (i = 0; i < VMA_MAX_CONFIGS; i++) {
391 + uint32_t name_ptr = GUINT32_FROM_BE(h->config_names[i]);
392 + uint32_t data_ptr = GUINT32_FROM_BE(h->config_data[i]);
393 +
394 + if (!(name_ptr && data_ptr)) {
395 + continue;
396 + }
397 + const char *name = get_header_str(vmar, name_ptr);
398 + const VmaBlob *blob = get_header_blob(vmar, data_ptr);
399 +
400 + if (!(name && blob)) {
401 + error_setg(errp, "vma contains invalid data pointers");
402 + return -1;
403 + }
404 +
405 + VmaConfigData *cdata = g_new0(VmaConfigData, 1);
406 + cdata->name = name;
407 + cdata->data = blob->data;
408 + cdata->len = blob->len;
409 +
410 + vmar->cdata_list = g_list_append(vmar->cdata_list, cdata);
411 + }
412 +
413 + return ret;
414 +};
415 +
416 +VmaReader *vma_reader_create(const char *filename, Error **errp)
417 +{
418 + assert(filename);
419 + assert(errp);
420 +
421 + VmaReader *vmar = g_new0(VmaReader, 1);
422 +
423 + if (strcmp(filename, "-") == 0) {
424 + vmar->fd = dup(0);
425 + } else {
426 + vmar->fd = open(filename, O_RDONLY);
427 + }
428 +
429 + if (vmar->fd < 0) {
430 + error_setg(errp, "can't open file %s - %s\n", filename,
431 + g_strerror(errno));
432 + goto err;
433 + }
434 +
435 + vmar->md5csum = g_checksum_new(G_CHECKSUM_MD5);
436 + if (!vmar->md5csum) {
437 + error_setg(errp, "can't allocate cmsum\n");
438 + goto err;
439 + }
440 +
441 + vmar->blob_hash = g_hash_table_new_full(g_int32_hash, g_int32_equal,
442 + NULL, g_free);
443 +
444 + if (vma_reader_read_head(vmar, errp) < 0) {
445 + goto err;
446 + }
447 +
448 + return vmar;
449 +
450 +err:
451 + if (vmar) {
452 + vma_reader_destroy(vmar);
453 + }
454 +
455 + return NULL;
456 +}
457 +
458 +VmaHeader *vma_reader_get_header(VmaReader *vmar)
459 +{
460 + assert(vmar);
461 + assert(vmar->head_data);
462 +
463 + return (VmaHeader *)(vmar->head_data);
464 +}
465 +
466 +GList *vma_reader_get_config_data(VmaReader *vmar)
467 +{
468 + assert(vmar);
469 + assert(vmar->head_data);
470 +
471 + return vmar->cdata_list;
472 +}
473 +
474 +VmaDeviceInfo *vma_reader_get_device_info(VmaReader *vmar, guint8 dev_id)
475 +{
476 + assert(vmar);
477 + assert(dev_id);
478 +
479 + if (vmar->devinfo[dev_id].size && vmar->devinfo[dev_id].devname) {
480 + return &vmar->devinfo[dev_id];
481 + }
482 +
483 + return NULL;
484 +}
485 +
486 +int vma_reader_register_bs(VmaReader *vmar, guint8 dev_id, BlockDriverState *bs,
487 + bool write_zeroes, Error **errp)
488 +{
489 + assert(vmar);
490 + assert(bs != NULL);
491 + assert(dev_id);
492 + assert(vmar->rstate[dev_id].bs == NULL);
493 +
494 + int64_t size = bdrv_getlength(bs);
495 + if (size != vmar->devinfo[dev_id].size) {
496 + error_setg(errp, "vma_reader_register_bs for stream %s failed - "
497 + "unexpected size %zd != %zd", vmar->devinfo[dev_id].devname,
498 + size, vmar->devinfo[dev_id].size);
499 + return -1;
500 + }
501 +
502 + vmar->rstate[dev_id].bs = bs;
503 + vmar->rstate[dev_id].write_zeroes = write_zeroes;
504 +
505 + int64_t bitmap_size = (size/BDRV_SECTOR_SIZE) +
506 + (VMA_CLUSTER_SIZE/BDRV_SECTOR_SIZE) * BITS_PER_LONG - 1;
507 + bitmap_size /= (VMA_CLUSTER_SIZE/BDRV_SECTOR_SIZE) * BITS_PER_LONG;
508 +
509 + vmar->rstate[dev_id].bitmap_size = bitmap_size;
510 + vmar->rstate[dev_id].bitmap = g_new0(unsigned long, bitmap_size);
511 +
512 + vmar->cluster_count += size/VMA_CLUSTER_SIZE;
513 +
514 + return 0;
515 +}
516 +
517 +static ssize_t safe_write(int fd, void *buf, size_t count)
518 +{
519 + ssize_t n;
520 +
521 + do {
522 + n = write(fd, buf, count);
523 + } while (n < 0 && errno == EINTR);
524 +
525 + return n;
526 +}
527 +
528 +static size_t full_write(int fd, void *buf, size_t len)
529 +{
530 + ssize_t n;
531 + size_t total;
532 +
533 + total = 0;
534 +
535 + while (len > 0) {
536 + n = safe_write(fd, buf, len);
537 + if (n < 0) {
538 + return n;
539 + }
540 + buf += n;
541 + total += n;
542 + len -= n;
543 + }
544 +
545 + if (len) {
546 + /* incomplete write ? */
547 + return -1;
548 + }
549 +
550 + return total;
551 +}
552 +
553 +static int restore_write_data(VmaReader *vmar, guint8 dev_id,
554 + BlockDriverState *bs, int vmstate_fd,
555 + unsigned char *buf, int64_t sector_num,
556 + int nb_sectors, Error **errp)
557 +{
558 + assert(vmar);
559 +
560 + if (dev_id == vmar->vmstate_stream) {
561 + if (vmstate_fd >= 0) {
562 + int len = nb_sectors * BDRV_SECTOR_SIZE;
563 + int res = full_write(vmstate_fd, buf, len);
564 + if (res < 0) {
565 + error_setg(errp, "write vmstate failed %d", res);
566 + return -1;
567 + }
568 + }
569 + } else {
570 + int res = bdrv_write(bs, sector_num, buf, nb_sectors);
571 + if (res < 0) {
572 + error_setg(errp, "bdrv_write to %s failed (%d)",
573 + bdrv_get_device_name(bs), res);
574 + return -1;
575 + }
576 + }
577 + return 0;
578 +}
579 +static int restore_extent(VmaReader *vmar, unsigned char *buf,
580 + int extent_size, int vmstate_fd,
581 + bool verbose, Error **errp)
582 +{
583 + assert(vmar);
584 + assert(buf);
585 +
586 + VmaExtentHeader *ehead = (VmaExtentHeader *)buf;
587 + int start = VMA_EXTENT_HEADER_SIZE;
588 + int i;
589 +
590 + for (i = 0; i < VMA_BLOCKS_PER_EXTENT; i++) {
591 + uint64_t block_info = GUINT64_FROM_BE(ehead->blockinfo[i]);
592 + uint64_t cluster_num = block_info & 0xffffffff;
593 + uint8_t dev_id = (block_info >> 32) & 0xff;
594 + uint16_t mask = block_info >> (32+16);
595 + int64_t max_sector;
596 +
597 + if (!dev_id) {
598 + continue;
599 + }
600 +
601 + VmaRestoreState *rstate = &vmar->rstate[dev_id];
602 + BlockDriverState *bs = NULL;
603 +
604 + if (dev_id != vmar->vmstate_stream) {
605 + bs = rstate->bs;
606 + if (!bs) {
607 + error_setg(errp, "got wrong dev id %d", dev_id);
608 + return -1;
609 + }
610 +
611 + if (vma_reader_get_bitmap(rstate, cluster_num)) {
612 + error_setg(errp, "found duplicated cluster %zd for stream %s",
613 + cluster_num, vmar->devinfo[dev_id].devname);
614 + return -1;
615 + }
616 + vma_reader_set_bitmap(rstate, cluster_num, 1);
617 +
618 + max_sector = vmar->devinfo[dev_id].size/BDRV_SECTOR_SIZE;
619 + } else {
620 + max_sector = G_MAXINT64;
621 + if (cluster_num != vmar->vmstate_clusters) {
622 + error_setg(errp, "found out of order vmstate data");
623 + return -1;
624 + }
625 + vmar->vmstate_clusters++;
626 + }
627 +
628 + vmar->clusters_read++;
629 +
630 + if (verbose) {
631 + time_t duration = time(NULL) - vmar->start_time;
632 + int percent = (vmar->clusters_read*100)/vmar->cluster_count;
633 + if (percent != vmar->clusters_read_per) {
634 + printf("progress %d%% (read %zd bytes, duration %zd sec)\n",
635 + percent, vmar->clusters_read*VMA_CLUSTER_SIZE,
636 + duration);
637 + fflush(stdout);
638 + vmar->clusters_read_per = percent;
639 + }
640 + }
641 +
642 + /* try to write whole clusters to speedup restore */
643 + if (mask == 0xffff) {
644 + if ((start + VMA_CLUSTER_SIZE) > extent_size) {
645 + error_setg(errp, "short vma extent - too many blocks");
646 + return -1;
647 + }
648 + int64_t sector_num = (cluster_num * VMA_CLUSTER_SIZE) /
649 + BDRV_SECTOR_SIZE;
650 + int64_t end_sector = sector_num +
651 + VMA_CLUSTER_SIZE/BDRV_SECTOR_SIZE;
652 +
653 + if (end_sector > max_sector) {
654 + end_sector = max_sector;
655 + }
656 +
657 + if (end_sector <= sector_num) {
658 + error_setg(errp, "got wrong block address - write bejond end");
659 + return -1;
660 + }
661 +
662 + int nb_sectors = end_sector - sector_num;
663 + if (restore_write_data(vmar, dev_id, bs, vmstate_fd, buf + start,
664 + sector_num, nb_sectors, errp) < 0) {
665 + return -1;
666 + }
667 +
668 + start += VMA_CLUSTER_SIZE;
669 + } else {
670 + int j;
671 + int bit = 1;
672 +
673 + for (j = 0; j < 16; j++) {
674 + int64_t sector_num = (cluster_num*VMA_CLUSTER_SIZE +
675 + j*VMA_BLOCK_SIZE)/BDRV_SECTOR_SIZE;
676 +
677 + int64_t end_sector = sector_num +
678 + VMA_BLOCK_SIZE/BDRV_SECTOR_SIZE;
679 + if (end_sector > max_sector) {
680 + end_sector = max_sector;
681 + }
682 +
683 + if (mask & bit) {
684 + if ((start + VMA_BLOCK_SIZE) > extent_size) {
685 + error_setg(errp, "short vma extent - too many blocks");
686 + return -1;
687 + }
688 +
689 + if (end_sector <= sector_num) {
690 + error_setg(errp, "got wrong block address - "
691 + "write bejond end");
692 + return -1;
693 + }
694 +
695 + int nb_sectors = end_sector - sector_num;
696 + if (restore_write_data(vmar, dev_id, bs, vmstate_fd,
697 + buf + start, sector_num,
698 + nb_sectors, errp) < 0) {
699 + return -1;
700 + }
701 +
702 + start += VMA_BLOCK_SIZE;
703 +
704 + } else {
705 +
706 + if (rstate->write_zeroes && (end_sector > sector_num)) {
707 + /* Todo: use bdrv_co_write_zeroes (but that need to
708 + * be run inside coroutine?)
709 + */
710 + int nb_sectors = end_sector - sector_num;
711 + if (restore_write_data(vmar, dev_id, bs, vmstate_fd,
712 + zero_vma_block, sector_num,
713 + nb_sectors, errp) < 0) {
714 + return -1;
715 + }
716 + }
717 + }
718 +
719 + bit = bit << 1;
720 + }
721 + }
722 + }
723 +
724 + if (start != extent_size) {
725 + error_setg(errp, "vma extent error - missing blocks");
726 + return -1;
727 + }
728 +
729 + return 0;
730 +}
731 +
732 +int vma_reader_restore(VmaReader *vmar, int vmstate_fd, bool verbose,
733 + Error **errp)
734 +{
735 + assert(vmar);
736 + assert(vmar->head_data);
737 +
738 + int ret = 0;
739 + unsigned char buf[VMA_MAX_EXTENT_SIZE];
740 + int buf_pos = 0;
741 + unsigned char md5sum[16];
742 + VmaHeader *h = (VmaHeader *)vmar->head_data;
743 +
744 + vmar->start_time = time(NULL);
745 +
746 + while (1) {
747 + int bytes = full_read(vmar->fd, buf + buf_pos, sizeof(buf) - buf_pos);
748 + if (bytes < 0) {
749 + error_setg(errp, "read failed - %s", g_strerror(errno));
750 + return -1;
751 + }
752 +
753 + buf_pos += bytes;
754 +
755 + if (!buf_pos) {
756 + break; /* EOF */
757 + }
758 +
759 + if (buf_pos < VMA_EXTENT_HEADER_SIZE) {
760 + error_setg(errp, "read short extent (%d bytes)", buf_pos);
761 + return -1;
762 + }
763 +
764 + VmaExtentHeader *ehead = (VmaExtentHeader *)buf;
765 +
766 + /* extract md5sum */
767 + memcpy(md5sum, ehead->md5sum, sizeof(ehead->md5sum));
768 + memset(ehead->md5sum, 0, sizeof(ehead->md5sum));
769 +
770 + g_checksum_reset(vmar->md5csum);
771 + g_checksum_update(vmar->md5csum, buf, VMA_EXTENT_HEADER_SIZE);
772 + gsize csize = 16;
773 + g_checksum_get_digest(vmar->md5csum, ehead->md5sum, &csize);
774 +
775 + if (memcmp(md5sum, ehead->md5sum, 16) != 0) {
776 + error_setg(errp, "wrong vma extent header chechsum");
777 + return -1;
778 + }
779 +
780 + if (memcmp(h->uuid, ehead->uuid, sizeof(ehead->uuid)) != 0) {
781 + error_setg(errp, "wrong vma extent uuid");
782 + return -1;
783 + }
784 +
785 + if (ehead->magic != VMA_EXTENT_MAGIC || ehead->reserved1 != 0) {
786 + error_setg(errp, "wrong vma extent header magic");
787 + return -1;
788 + }
789 +
790 + int block_count = GUINT16_FROM_BE(ehead->block_count);
791 + int extent_size = VMA_EXTENT_HEADER_SIZE + block_count*VMA_BLOCK_SIZE;
792 +
793 + if (buf_pos < extent_size) {
794 + error_setg(errp, "short vma extent (%d < %d)", buf_pos,
795 + extent_size);
796 + return -1;
797 + }
798 +
799 + if (restore_extent(vmar, buf, extent_size, vmstate_fd, verbose,
800 + errp) < 0) {
801 + return -1;
802 + }
803 +
804 + if (buf_pos > extent_size) {
805 + memmove(buf, buf + extent_size, buf_pos - extent_size);
806 + buf_pos = buf_pos - extent_size;
807 + } else {
808 + buf_pos = 0;
809 + }
810 + }
811 +
812 + bdrv_drain_all();
813 +
814 + int i;
815 + for (i = 1; i < 256; i++) {
816 + VmaRestoreState *rstate = &vmar->rstate[i];
817 + if (!rstate->bs) {
818 + continue;
819 + }
820 +
821 + if (bdrv_flush(rstate->bs) < 0) {
822 + error_setg(errp, "vma bdrv_flush %s failed",
823 + vmar->devinfo[i].devname);
824 + return -1;
825 + }
826 +
827 + if (vmar->devinfo[i].size &&
828 + (strcmp(vmar->devinfo[i].devname, "vmstate") != 0)) {
829 + assert(rstate->bitmap);
830 +
831 + int64_t cluster_num, end;
832 +
833 + end = (vmar->devinfo[i].size + VMA_CLUSTER_SIZE - 1) /
834 + VMA_CLUSTER_SIZE;
835 +
836 + for (cluster_num = 0; cluster_num < end; cluster_num++) {
837 + if (!vma_reader_get_bitmap(rstate, cluster_num)) {
838 + error_setg(errp, "detected missing cluster %zd "
839 + "for stream %s", cluster_num,
840 + vmar->devinfo[i].devname);
841 + return -1;
842 + }
843 + }
844 + }
845 + }
846 +
847 + return ret;
848 +}
849 +
850 Index: new/vma-writer.c
851 ===================================================================
852 --- /dev/null 1970-01-01 00:00:00.000000000 +0000
853 +++ new/vma-writer.c 2014-11-20 08:31:06.000000000 +0100
854 @@ -0,0 +1,876 @@
855 +/*
856 + * VMA: Virtual Machine Archive
857 + *
858 + * Copyright (C) 2012 Proxmox Server Solutions
859 + *
860 + * Authors:
861 + * Dietmar Maurer (dietmar@proxmox.com)
862 + *
863 + * This work is licensed under the terms of the GNU GPL, version 2 or later.
864 + * See the COPYING file in the top-level directory.
865 + *
866 + */
867 +
868 +#include <stdio.h>
869 +#include <errno.h>
870 +#include <unistd.h>
871 +#include <stdio.h>
872 +#include <string.h>
873 +#include <sys/types.h>
874 +#include <sys/stat.h>
875 +#include <fcntl.h>
876 +#include <glib.h>
877 +#include <uuid/uuid.h>
878 +
879 +#include "qemu-common.h"
880 +#include "vma.h"
881 +#include "block/block.h"
882 +#include "monitor/monitor.h"
883 +#include "qemu/main-loop.h"
884 +
885 +#define DEBUG_VMA 0
886 +
887 +#define DPRINTF(fmt, ...)\
888 + do { if (DEBUG_VMA) { printf("vma: " fmt, ## __VA_ARGS__); } } while (0)
889 +
890 +#define WRITE_BUFFERS 5
891 +
892 +typedef struct VmaAIOCB VmaAIOCB;
893 +struct VmaAIOCB {
894 + unsigned char buffer[VMA_MAX_EXTENT_SIZE];
895 + VmaWriter *vmaw;
896 + size_t bytes;
897 + Coroutine *co;
898 +};
899 +
900 +struct VmaWriter {
901 + int fd;
902 + FILE *cmd;
903 + int status;
904 + char errmsg[8192];
905 + uuid_t uuid;
906 + bool header_written;
907 + bool closed;
908 +
909 + /* we always write extents */
910 + unsigned char outbuf[VMA_MAX_EXTENT_SIZE];
911 + int outbuf_pos; /* in bytes */
912 + int outbuf_count; /* in VMA_BLOCKS */
913 + uint64_t outbuf_block_info[VMA_BLOCKS_PER_EXTENT];
914 +
915 + VmaAIOCB *aiocbs[WRITE_BUFFERS];
916 + CoQueue wqueue;
917 +
918 + GChecksum *md5csum;
919 + CoMutex writer_lock;
920 + CoMutex flush_lock;
921 + Coroutine *co_writer;
922 +
923 + /* drive informations */
924 + VmaStreamInfo stream_info[256];
925 + guint stream_count;
926 +
927 + guint8 vmstate_stream;
928 + uint32_t vmstate_clusters;
929 +
930 + /* header blob table */
931 + char *header_blob_table;
932 + uint32_t header_blob_table_size;
933 + uint32_t header_blob_table_pos;
934 +
935 + /* store for config blobs */
936 + uint32_t config_names[VMA_MAX_CONFIGS]; /* offset into blob_buffer table */
937 + uint32_t config_data[VMA_MAX_CONFIGS]; /* offset into blob_buffer table */
938 + uint32_t config_count;
939 +};
940 +
941 +void vma_writer_set_error(VmaWriter *vmaw, const char *fmt, ...)
942 +{
943 + va_list ap;
944 +
945 + if (vmaw->status < 0) {
946 + return;
947 + }
948 +
949 + vmaw->status = -1;
950 +
951 + va_start(ap, fmt);
952 + g_vsnprintf(vmaw->errmsg, sizeof(vmaw->errmsg), fmt, ap);
953 + va_end(ap);
954 +
955 + DPRINTF("vma_writer_set_error: %s\n", vmaw->errmsg);
956 +}
957 +
958 +static uint32_t allocate_header_blob(VmaWriter *vmaw, const char *data,
959 + size_t len)
960 +{
961 + if (len > 65535) {
962 + return 0;
963 + }
964 +
965 + if (!vmaw->header_blob_table ||
966 + (vmaw->header_blob_table_size <
967 + (vmaw->header_blob_table_pos + len + 2))) {
968 + int newsize = vmaw->header_blob_table_size + ((len + 2 + 511)/512)*512;
969 +
970 + vmaw->header_blob_table = g_realloc(vmaw->header_blob_table, newsize);
971 + memset(vmaw->header_blob_table + vmaw->header_blob_table_size,
972 + 0, newsize - vmaw->header_blob_table_size);
973 + vmaw->header_blob_table_size = newsize;
974 + }
975 +
976 + uint32_t cpos = vmaw->header_blob_table_pos;
977 + vmaw->header_blob_table[cpos] = len & 255;
978 + vmaw->header_blob_table[cpos+1] = (len >> 8) & 255;
979 + memcpy(vmaw->header_blob_table + cpos + 2, data, len);
980 + vmaw->header_blob_table_pos += len + 2;
981 + return cpos;
982 +}
983 +
984 +static uint32_t allocate_header_string(VmaWriter *vmaw, const char *str)
985 +{
986 + assert(vmaw);
987 +
988 + size_t len = strlen(str) + 1;
989 +
990 + return allocate_header_blob(vmaw, str, len);
991 +}
992 +
993 +int vma_writer_add_config(VmaWriter *vmaw, const char *name, gpointer data,
994 + gsize len)
995 +{
996 + assert(vmaw);
997 + assert(!vmaw->header_written);
998 + assert(vmaw->config_count < VMA_MAX_CONFIGS);
999 + assert(name);
1000 + assert(data);
1001 + assert(len);
1002 +
1003 + gchar *basename = g_path_get_basename(name);
1004 + uint32_t name_ptr = allocate_header_string(vmaw, basename);
1005 + g_free(basename);
1006 +
1007 + if (!name_ptr) {
1008 + return -1;
1009 + }
1010 +
1011 + uint32_t data_ptr = allocate_header_blob(vmaw, data, len);
1012 + if (!data_ptr) {
1013 + return -1;
1014 + }
1015 +
1016 + vmaw->config_names[vmaw->config_count] = name_ptr;
1017 + vmaw->config_data[vmaw->config_count] = data_ptr;
1018 +
1019 + vmaw->config_count++;
1020 +
1021 + return 0;
1022 +}
1023 +
1024 +int vma_writer_register_stream(VmaWriter *vmaw, const char *devname,
1025 + size_t size)
1026 +{
1027 + assert(vmaw);
1028 + assert(devname);
1029 + assert(!vmaw->status);
1030 +
1031 + if (vmaw->header_written) {
1032 + vma_writer_set_error(vmaw, "vma_writer_register_stream: header "
1033 + "already written");
1034 + return -1;
1035 + }
1036 +
1037 + guint n = vmaw->stream_count + 1;
1038 +
1039 + /* we can have dev_ids form 1 to 255 (0 reserved)
1040 + * 255(-1) reseverd for safety
1041 + */
1042 + if (n > 254) {
1043 + vma_writer_set_error(vmaw, "vma_writer_register_stream: "
1044 + "too many drives");
1045 + return -1;
1046 + }
1047 +
1048 + if (size <= 0) {
1049 + vma_writer_set_error(vmaw, "vma_writer_register_stream: "
1050 + "got strange size %zd", size);
1051 + return -1;
1052 + }
1053 +
1054 + DPRINTF("vma_writer_register_stream %s %zu %d\n", devname, size, n);
1055 +
1056 + vmaw->stream_info[n].devname = g_strdup(devname);
1057 + vmaw->stream_info[n].size = size;
1058 +
1059 + vmaw->stream_info[n].cluster_count = (size + VMA_CLUSTER_SIZE - 1) /
1060 + VMA_CLUSTER_SIZE;
1061 +
1062 + vmaw->stream_count = n;
1063 +
1064 + if (strcmp(devname, "vmstate") == 0) {
1065 + vmaw->vmstate_stream = n;
1066 + }
1067 +
1068 + return n;
1069 +}
1070 +
1071 +static void vma_co_continue_write(void *opaque)
1072 +{
1073 + VmaWriter *vmaw = opaque;
1074 +
1075 + DPRINTF("vma_co_continue_write\n");
1076 + qemu_coroutine_enter(vmaw->co_writer, NULL);
1077 +}
1078 +
1079 +static ssize_t coroutine_fn
1080 +vma_co_write(VmaWriter *vmaw, const void *buf, size_t bytes)
1081 +{
1082 + size_t done = 0;
1083 + ssize_t ret;
1084 +
1085 + /* atomic writes (we cannot interleave writes) */
1086 + qemu_co_mutex_lock(&vmaw->writer_lock);
1087 +
1088 + DPRINTF("vma_co_write enter %zd\n", bytes);
1089 +
1090 + assert(vmaw->co_writer == NULL);
1091 +
1092 + vmaw->co_writer = qemu_coroutine_self();
1093 +
1094 + aio_set_fd_handler(qemu_get_aio_context(), vmaw->fd, NULL, vma_co_continue_write, vmaw);
1095 +
1096 + DPRINTF("vma_co_write wait until writable\n");
1097 + qemu_coroutine_yield();
1098 + DPRINTF("vma_co_write starting %zd\n", bytes);
1099 +
1100 + while (done < bytes) {
1101 + ret = write(vmaw->fd, buf + done, bytes - done);
1102 + if (ret > 0) {
1103 + done += ret;
1104 + DPRINTF("vma_co_write written %zd %zd\n", done, ret);
1105 + } else if (ret < 0) {
1106 + if (errno == EAGAIN || errno == EWOULDBLOCK) {
1107 + DPRINTF("vma_co_write yield %zd\n", done);
1108 + qemu_coroutine_yield();
1109 + DPRINTF("vma_co_write restart %zd\n", done);
1110 + } else {
1111 + vma_writer_set_error(vmaw, "vma_co_write write error - %s",
1112 + g_strerror(errno));
1113 + done = -1; /* always return failure for partial writes */
1114 + break;
1115 + }
1116 + } else if (ret == 0) {
1117 + /* should not happen - simply try again */
1118 + }
1119 + }
1120 +
1121 + aio_set_fd_handler(qemu_get_aio_context(), vmaw->fd, NULL, NULL, NULL);
1122 +
1123 + vmaw->co_writer = NULL;
1124 +
1125 + qemu_co_mutex_unlock(&vmaw->writer_lock);
1126 +
1127 + DPRINTF("vma_co_write leave %zd\n", done);
1128 + return done;
1129 +}
1130 +
1131 +static void coroutine_fn vma_co_writer_task(void *opaque)
1132 +{
1133 + VmaAIOCB *cb = opaque;
1134 +
1135 + DPRINTF("vma_co_writer_task start\n");
1136 +
1137 + int64_t done = vma_co_write(cb->vmaw, cb->buffer, cb->bytes);
1138 + DPRINTF("vma_co_writer_task write done %zd\n", done);
1139 +
1140 + if (done != cb->bytes) {
1141 + DPRINTF("vma_co_writer_task failed write %zd %zd", cb->bytes, done);
1142 + vma_writer_set_error(cb->vmaw, "vma_co_writer_task failed write %zd",
1143 + done);
1144 + }
1145 +
1146 + cb->bytes = 0;
1147 +
1148 + qemu_co_queue_next(&cb->vmaw->wqueue);
1149 +
1150 + DPRINTF("vma_co_writer_task end\n");
1151 +}
1152 +
1153 +static void coroutine_fn vma_queue_flush(VmaWriter *vmaw)
1154 +{
1155 + DPRINTF("vma_queue_flush enter\n");
1156 +
1157 + assert(vmaw);
1158 +
1159 + while (1) {
1160 + int i;
1161 + VmaAIOCB *cb = NULL;
1162 + for (i = 0; i < WRITE_BUFFERS; i++) {
1163 + if (vmaw->aiocbs[i]->bytes) {
1164 + cb = vmaw->aiocbs[i];
1165 + DPRINTF("FOUND USED AIO BUFFER %d %zd\n", i,
1166 + vmaw->aiocbs[i]->bytes);
1167 + break;
1168 + }
1169 + }
1170 + if (!cb) {
1171 + break;
1172 + }
1173 + qemu_co_queue_wait(&vmaw->wqueue);
1174 + }
1175 +
1176 + DPRINTF("vma_queue_flush leave\n");
1177 +}
1178 +
1179 +/**
1180 + * NOTE: pipe buffer size in only 4096 bytes on linux (see 'ulimit -a')
1181 + * So we need to create a coroutione to allow 'parallel' execution.
1182 + */
1183 +static ssize_t coroutine_fn
1184 +vma_queue_write(VmaWriter *vmaw, const void *buf, size_t bytes)
1185 +{
1186 + DPRINTF("vma_queue_write enter %zd\n", bytes);
1187 +
1188 + assert(vmaw);
1189 + assert(buf);
1190 + assert(bytes <= VMA_MAX_EXTENT_SIZE);
1191 +
1192 + VmaAIOCB *cb = NULL;
1193 + while (!cb) {
1194 + int i;
1195 + for (i = 0; i < WRITE_BUFFERS; i++) {
1196 + if (!vmaw->aiocbs[i]->bytes) {
1197 + cb = vmaw->aiocbs[i];
1198 + break;
1199 + }
1200 + }
1201 + if (!cb) {
1202 + qemu_co_queue_wait(&vmaw->wqueue);
1203 + }
1204 + }
1205 +
1206 + memcpy(cb->buffer, buf, bytes);
1207 + cb->bytes = bytes;
1208 + cb->vmaw = vmaw;
1209 +
1210 + DPRINTF("vma_queue_write start %zd\n", bytes);
1211 + cb->co = qemu_coroutine_create(vma_co_writer_task);
1212 + qemu_coroutine_enter(cb->co, cb);
1213 +
1214 + DPRINTF("vma_queue_write leave\n");
1215 +
1216 + return bytes;
1217 +}
1218 +
1219 +VmaWriter *vma_writer_create(const char *filename, uuid_t uuid, Error **errp)
1220 +{
1221 + const char *p;
1222 +
1223 + assert(sizeof(VmaHeader) == (4096 + 8192));
1224 + assert(G_STRUCT_OFFSET(VmaHeader, config_names) == 2044);
1225 + assert(G_STRUCT_OFFSET(VmaHeader, config_data) == 3068);
1226 + assert(G_STRUCT_OFFSET(VmaHeader, dev_info) == 4096);
1227 + assert(sizeof(VmaExtentHeader) == 512);
1228 +
1229 + VmaWriter *vmaw = g_new0(VmaWriter, 1);
1230 + vmaw->fd = -1;
1231 +
1232 + vmaw->md5csum = g_checksum_new(G_CHECKSUM_MD5);
1233 + if (!vmaw->md5csum) {
1234 + error_setg(errp, "can't allocate cmsum\n");
1235 + goto err;
1236 + }
1237 +
1238 + if (strstart(filename, "exec:", &p)) {
1239 + vmaw->cmd = popen(p, "w");
1240 + if (vmaw->cmd == NULL) {
1241 + error_setg(errp, "can't popen command '%s' - %s\n", p,
1242 + g_strerror(errno));
1243 + goto err;
1244 + }
1245 + vmaw->fd = fileno(vmaw->cmd);
1246 +
1247 + /* try to use O_NONBLOCK and O_DIRECT */
1248 + fcntl(vmaw->fd, F_SETFL, fcntl(vmaw->fd, F_GETFL)|O_NONBLOCK);
1249 + fcntl(vmaw->fd, F_SETFL, fcntl(vmaw->fd, F_GETFL)|O_DIRECT);
1250 +
1251 + } else {
1252 + struct stat st;
1253 + int oflags;
1254 + const char *tmp_id_str;
1255 +
1256 + if ((stat(filename, &st) == 0) && S_ISFIFO(st.st_mode)) {
1257 + oflags = O_NONBLOCK|O_DIRECT|O_WRONLY;
1258 + vmaw->fd = qemu_open(filename, oflags, 0644);
1259 + } else if (strstart(filename, "/dev/fdset/", &tmp_id_str)) {
1260 + oflags = O_NONBLOCK|O_DIRECT|O_WRONLY;
1261 + vmaw->fd = qemu_open(filename, oflags, 0644);
1262 + } else if (strstart(filename, "/dev/fdname/", &tmp_id_str)) {
1263 + vmaw->fd = monitor_get_fd(cur_mon, tmp_id_str, errp);
1264 + if (vmaw->fd < 0) {
1265 + goto err;
1266 + }
1267 + /* try to use O_NONBLOCK and O_DIRECT */
1268 + fcntl(vmaw->fd, F_SETFL, fcntl(vmaw->fd, F_GETFL)|O_NONBLOCK);
1269 + fcntl(vmaw->fd, F_SETFL, fcntl(vmaw->fd, F_GETFL)|O_DIRECT);
1270 + } else {
1271 + oflags = O_NONBLOCK|O_DIRECT|O_WRONLY|O_CREAT|O_EXCL;
1272 + vmaw->fd = qemu_open(filename, oflags, 0644);
1273 + }
1274 +
1275 + if (vmaw->fd < 0) {
1276 + error_setg(errp, "can't open file %s - %s\n", filename,
1277 + g_strerror(errno));
1278 + goto err;
1279 + }
1280 + }
1281 +
1282 + /* we use O_DIRECT, so we need to align IO buffers */
1283 + int i;
1284 + for (i = 0; i < WRITE_BUFFERS; i++) {
1285 + vmaw->aiocbs[i] = qemu_memalign(512, sizeof(VmaAIOCB));
1286 + memset(vmaw->aiocbs[i], 0, sizeof(VmaAIOCB));
1287 + }
1288 +
1289 + vmaw->outbuf_count = 0;
1290 + vmaw->outbuf_pos = VMA_EXTENT_HEADER_SIZE;
1291 +
1292 + vmaw->header_blob_table_pos = 1; /* start at pos 1 */
1293 +
1294 + qemu_co_mutex_init(&vmaw->writer_lock);
1295 + qemu_co_mutex_init(&vmaw->flush_lock);
1296 + qemu_co_queue_init(&vmaw->wqueue);
1297 +
1298 + uuid_copy(vmaw->uuid, uuid);
1299 +
1300 + return vmaw;
1301 +
1302 +err:
1303 + if (vmaw) {
1304 + if (vmaw->cmd) {
1305 + pclose(vmaw->cmd);
1306 + } else if (vmaw->fd >= 0) {
1307 + close(vmaw->fd);
1308 + }
1309 +
1310 + if (vmaw->md5csum) {
1311 + g_checksum_free(vmaw->md5csum);
1312 + }
1313 +
1314 + g_free(vmaw);
1315 + }
1316 +
1317 + return NULL;
1318 +}
1319 +
1320 +static int coroutine_fn vma_write_header(VmaWriter *vmaw)
1321 +{
1322 + assert(vmaw);
1323 + int header_clusters = 8;
1324 + char buf[65536*header_clusters];
1325 + VmaHeader *head = (VmaHeader *)buf;
1326 +
1327 + int i;
1328 +
1329 + DPRINTF("VMA WRITE HEADER\n");
1330 +
1331 + if (vmaw->status < 0) {
1332 + return vmaw->status;
1333 + }
1334 +
1335 + memset(buf, 0, sizeof(buf));
1336 +
1337 + head->magic = VMA_MAGIC;
1338 + head->version = GUINT32_TO_BE(1); /* v1 */
1339 + memcpy(head->uuid, vmaw->uuid, 16);
1340 +
1341 + time_t ctime = time(NULL);
1342 + head->ctime = GUINT64_TO_BE(ctime);
1343 +
1344 + if (!vmaw->stream_count) {
1345 + return -1;
1346 + }
1347 +
1348 + for (i = 0; i < VMA_MAX_CONFIGS; i++) {
1349 + head->config_names[i] = GUINT32_TO_BE(vmaw->config_names[i]);
1350 + head->config_data[i] = GUINT32_TO_BE(vmaw->config_data[i]);
1351 + }
1352 +
1353 + /* 32 bytes per device (12 used currently) = 8192 bytes max */
1354 + for (i = 1; i <= 254; i++) {
1355 + VmaStreamInfo *si = &vmaw->stream_info[i];
1356 + if (si->size) {
1357 + assert(si->devname);
1358 + uint32_t devname_ptr = allocate_header_string(vmaw, si->devname);
1359 + if (!devname_ptr) {
1360 + return -1;
1361 + }
1362 + head->dev_info[i].devname_ptr = GUINT32_TO_BE(devname_ptr);
1363 + head->dev_info[i].size = GUINT64_TO_BE(si->size);
1364 + }
1365 + }
1366 +
1367 + uint32_t header_size = sizeof(VmaHeader) + vmaw->header_blob_table_size;
1368 + head->header_size = GUINT32_TO_BE(header_size);
1369 +
1370 + if (header_size > sizeof(buf)) {
1371 + return -1; /* just to be sure */
1372 + }
1373 +
1374 + uint32_t blob_buffer_offset = sizeof(VmaHeader);
1375 + memcpy(buf + blob_buffer_offset, vmaw->header_blob_table,
1376 + vmaw->header_blob_table_size);
1377 + head->blob_buffer_offset = GUINT32_TO_BE(blob_buffer_offset);
1378 + head->blob_buffer_size = GUINT32_TO_BE(vmaw->header_blob_table_pos);
1379 +
1380 + g_checksum_reset(vmaw->md5csum);
1381 + g_checksum_update(vmaw->md5csum, (const guchar *)buf, header_size);
1382 + gsize csize = 16;
1383 + g_checksum_get_digest(vmaw->md5csum, (guint8 *)(head->md5sum), &csize);
1384 +
1385 + return vma_queue_write(vmaw, buf, header_size);
1386 +}
1387 +
1388 +static int coroutine_fn vma_writer_flush(VmaWriter *vmaw)
1389 +{
1390 + assert(vmaw);
1391 +
1392 + int ret;
1393 + int i;
1394 +
1395 + if (vmaw->status < 0) {
1396 + return vmaw->status;
1397 + }
1398 +
1399 + if (!vmaw->header_written) {
1400 + vmaw->header_written = true;
1401 + ret = vma_write_header(vmaw);
1402 + if (ret < 0) {
1403 + vma_writer_set_error(vmaw, "vma_writer_flush: write header failed");
1404 + return ret;
1405 + }
1406 + }
1407 +
1408 + DPRINTF("VMA WRITE FLUSH %d %d\n", vmaw->outbuf_count, vmaw->outbuf_pos);
1409 +
1410 +
1411 + VmaExtentHeader *ehead = (VmaExtentHeader *)vmaw->outbuf;
1412 +
1413 + ehead->magic = VMA_EXTENT_MAGIC;
1414 + ehead->reserved1 = 0;
1415 +
1416 + for (i = 0; i < VMA_BLOCKS_PER_EXTENT; i++) {
1417 + ehead->blockinfo[i] = GUINT64_TO_BE(vmaw->outbuf_block_info[i]);
1418 + }
1419 +
1420 + guint16 block_count = (vmaw->outbuf_pos - VMA_EXTENT_HEADER_SIZE) /
1421 + VMA_BLOCK_SIZE;
1422 +
1423 + ehead->block_count = GUINT16_TO_BE(block_count);
1424 +
1425 + memcpy(ehead->uuid, vmaw->uuid, sizeof(ehead->uuid));
1426 + memset(ehead->md5sum, 0, sizeof(ehead->md5sum));
1427 +
1428 + g_checksum_reset(vmaw->md5csum);
1429 + g_checksum_update(vmaw->md5csum, vmaw->outbuf, VMA_EXTENT_HEADER_SIZE);
1430 + gsize csize = 16;
1431 + g_checksum_get_digest(vmaw->md5csum, ehead->md5sum, &csize);
1432 +
1433 + int bytes = vmaw->outbuf_pos;
1434 + ret = vma_queue_write(vmaw, vmaw->outbuf, bytes);
1435 + if (ret != bytes) {
1436 + vma_writer_set_error(vmaw, "vma_writer_flush: failed write");
1437 + }
1438 +
1439 + vmaw->outbuf_count = 0;
1440 + vmaw->outbuf_pos = VMA_EXTENT_HEADER_SIZE;
1441 +
1442 + for (i = 0; i < VMA_BLOCKS_PER_EXTENT; i++) {
1443 + vmaw->outbuf_block_info[i] = 0;
1444 + }
1445 +
1446 + return vmaw->status;
1447 +}
1448 +
1449 +static int vma_count_open_streams(VmaWriter *vmaw)
1450 +{
1451 + g_assert(vmaw != NULL);
1452 +
1453 + int i;
1454 + int open_drives = 0;
1455 + for (i = 0; i <= 255; i++) {
1456 + if (vmaw->stream_info[i].size && !vmaw->stream_info[i].finished) {
1457 + open_drives++;
1458 + }
1459 + }
1460 +
1461 + return open_drives;
1462 +}
1463 +
1464 +/**
1465 + * all jobs should call this when there is no more data
1466 + * Returns: number of remaining stream (0 ==> finished)
1467 + */
1468 +int coroutine_fn
1469 +vma_writer_close_stream(VmaWriter *vmaw, uint8_t dev_id)
1470 +{
1471 + g_assert(vmaw != NULL);
1472 +
1473 + DPRINTF("vma_writer_set_status %d\n", dev_id);
1474 + if (!vmaw->stream_info[dev_id].size) {
1475 + vma_writer_set_error(vmaw, "vma_writer_close_stream: "
1476 + "no such stream %d", dev_id);
1477 + return -1;
1478 + }
1479 + if (vmaw->stream_info[dev_id].finished) {
1480 + vma_writer_set_error(vmaw, "vma_writer_close_stream: "
1481 + "stream already closed %d", dev_id);
1482 + return -1;
1483 + }
1484 +
1485 + vmaw->stream_info[dev_id].finished = true;
1486 +
1487 + int open_drives = vma_count_open_streams(vmaw);
1488 +
1489 + if (open_drives <= 0) {
1490 + DPRINTF("vma_writer_set_status all drives completed\n");
1491 + qemu_co_mutex_lock(&vmaw->flush_lock);
1492 + int ret = vma_writer_flush(vmaw);
1493 + qemu_co_mutex_unlock(&vmaw->flush_lock);
1494 + if (ret < 0) {
1495 + vma_writer_set_error(vmaw, "vma_writer_close_stream: flush failed");
1496 + }
1497 + }
1498 +
1499 + return open_drives;
1500 +}
1501 +
1502 +int vma_writer_get_status(VmaWriter *vmaw, VmaStatus *status)
1503 +{
1504 + int i;
1505 +
1506 + g_assert(vmaw != NULL);
1507 +
1508 + if (status) {
1509 + status->status = vmaw->status;
1510 + g_strlcpy(status->errmsg, vmaw->errmsg, sizeof(status->errmsg));
1511 + for (i = 0; i <= 255; i++) {
1512 + status->stream_info[i] = vmaw->stream_info[i];
1513 + }
1514 +
1515 + uuid_unparse_lower(vmaw->uuid, status->uuid_str);
1516 + }
1517 +
1518 + status->closed = vmaw->closed;
1519 +
1520 + return vmaw->status;
1521 +}
1522 +
1523 +static int vma_writer_get_buffer(VmaWriter *vmaw)
1524 +{
1525 + int ret = 0;
1526 +
1527 + qemu_co_mutex_lock(&vmaw->flush_lock);
1528 +
1529 + /* wait until buffer is available */
1530 + while (vmaw->outbuf_count >= (VMA_BLOCKS_PER_EXTENT - 1)) {
1531 + ret = vma_writer_flush(vmaw);
1532 + if (ret < 0) {
1533 + vma_writer_set_error(vmaw, "vma_writer_get_buffer: flush failed");
1534 + break;
1535 + }
1536 + }
1537 +
1538 + qemu_co_mutex_unlock(&vmaw->flush_lock);
1539 +
1540 + return ret;
1541 +}
1542 +
1543 +
1544 +int64_t coroutine_fn
1545 +vma_writer_write(VmaWriter *vmaw, uint8_t dev_id, int64_t cluster_num,
1546 + unsigned char *buf, size_t *zero_bytes)
1547 +{
1548 + g_assert(vmaw != NULL);
1549 + g_assert(zero_bytes != NULL);
1550 +
1551 + *zero_bytes = 0;
1552 +
1553 + if (vmaw->status < 0) {
1554 + return vmaw->status;
1555 + }
1556 +
1557 + if (!dev_id || !vmaw->stream_info[dev_id].size) {
1558 + vma_writer_set_error(vmaw, "vma_writer_write: "
1559 + "no such stream %d", dev_id);
1560 + return -1;
1561 + }
1562 +
1563 + if (vmaw->stream_info[dev_id].finished) {
1564 + vma_writer_set_error(vmaw, "vma_writer_write: "
1565 + "stream already closed %d", dev_id);
1566 + return -1;
1567 + }
1568 +
1569 +
1570 + if (cluster_num >= (((uint64_t)1)<<32)) {
1571 + vma_writer_set_error(vmaw, "vma_writer_write: "
1572 + "cluster number out of range");
1573 + return -1;
1574 + }
1575 +
1576 + if (dev_id == vmaw->vmstate_stream) {
1577 + if (cluster_num != vmaw->vmstate_clusters) {
1578 + vma_writer_set_error(vmaw, "vma_writer_write: "
1579 + "non sequential vmstate write");
1580 + }
1581 + vmaw->vmstate_clusters++;
1582 + } else if (cluster_num >= vmaw->stream_info[dev_id].cluster_count) {
1583 + vma_writer_set_error(vmaw, "vma_writer_write: cluster number too big");
1584 + return -1;
1585 + }
1586 +
1587 + /* wait until buffer is available */
1588 + if (vma_writer_get_buffer(vmaw) < 0) {
1589 + vma_writer_set_error(vmaw, "vma_writer_write: "
1590 + "vma_writer_get_buffer failed");
1591 + return -1;
1592 + }
1593 +
1594 + DPRINTF("VMA WRITE %d %zd\n", dev_id, cluster_num);
1595 +
1596 + uint16_t mask = 0;
1597 +
1598 + if (buf) {
1599 + int i;
1600 + int bit = 1;
1601 + for (i = 0; i < 16; i++) {
1602 + unsigned char *vmablock = buf + (i*VMA_BLOCK_SIZE);
1603 + if (!buffer_is_zero(vmablock, VMA_BLOCK_SIZE)) {
1604 + mask |= bit;
1605 + memcpy(vmaw->outbuf + vmaw->outbuf_pos, vmablock,
1606 + VMA_BLOCK_SIZE);
1607 + vmaw->outbuf_pos += VMA_BLOCK_SIZE;
1608 + } else {
1609 + DPRINTF("VMA WRITE %zd ZERO BLOCK %d\n", cluster_num, i);
1610 + vmaw->stream_info[dev_id].zero_bytes += VMA_BLOCK_SIZE;
1611 + *zero_bytes += VMA_BLOCK_SIZE;
1612 + }
1613 +
1614 + bit = bit << 1;
1615 + }
1616 + } else {
1617 + DPRINTF("VMA WRITE %zd ZERO CLUSTER\n", cluster_num);
1618 + vmaw->stream_info[dev_id].zero_bytes += VMA_CLUSTER_SIZE;
1619 + *zero_bytes += VMA_CLUSTER_SIZE;
1620 + }
1621 +
1622 + uint64_t block_info = ((uint64_t)mask) << (32+16);
1623 + block_info |= ((uint64_t)dev_id) << 32;
1624 + block_info |= (cluster_num & 0xffffffff);
1625 + vmaw->outbuf_block_info[vmaw->outbuf_count] = block_info;
1626 +
1627 + DPRINTF("VMA WRITE MASK %zd %zx\n", cluster_num, block_info);
1628 +
1629 + vmaw->outbuf_count++;
1630 +
1631 + /** NOTE: We allways write whole clusters, but we correctly set
1632 + * transferred bytes. So transferred == size when when everything
1633 + * went OK.
1634 + */
1635 + size_t transferred = VMA_CLUSTER_SIZE;
1636 +
1637 + if (dev_id != vmaw->vmstate_stream) {
1638 + uint64_t last = (cluster_num + 1) * VMA_CLUSTER_SIZE;
1639 + if (last > vmaw->stream_info[dev_id].size) {
1640 + uint64_t diff = last - vmaw->stream_info[dev_id].size;
1641 + if (diff >= VMA_CLUSTER_SIZE) {
1642 + vma_writer_set_error(vmaw, "vma_writer_write: "
1643 + "read after last cluster");
1644 + return -1;
1645 + }
1646 + transferred -= diff;
1647 + }
1648 + }
1649 +
1650 + vmaw->stream_info[dev_id].transferred += transferred;
1651 +
1652 + return transferred;
1653 +}
1654 +
1655 +int vma_writer_close(VmaWriter *vmaw, Error **errp)
1656 +{
1657 + g_assert(vmaw != NULL);
1658 +
1659 + int i;
1660 +
1661 + vma_queue_flush(vmaw);
1662 +
1663 + /* this should not happen - just to be sure */
1664 + while (!qemu_co_queue_empty(&vmaw->wqueue)) {
1665 + DPRINTF("vma_writer_close wait\n");
1666 + co_aio_sleep_ns(qemu_get_aio_context(), QEMU_CLOCK_REALTIME, 1000000);
1667 + }
1668 +
1669 + if (vmaw->cmd) {
1670 + if (pclose(vmaw->cmd) < 0) {
1671 + vma_writer_set_error(vmaw, "vma_writer_close: "
1672 + "pclose failed - %s", g_strerror(errno));
1673 + }
1674 + } else {
1675 + if (close(vmaw->fd) < 0) {
1676 + vma_writer_set_error(vmaw, "vma_writer_close: "
1677 + "close failed - %s", g_strerror(errno));
1678 + }
1679 + }
1680 +
1681 + for (i = 0; i <= 255; i++) {
1682 + VmaStreamInfo *si = &vmaw->stream_info[i];
1683 + if (si->size) {
1684 + if (!si->finished) {
1685 + vma_writer_set_error(vmaw, "vma_writer_close: "
1686 + "detected open stream '%s'", si->devname);
1687 + } else if ((si->transferred != si->size) &&
1688 + (i != vmaw->vmstate_stream)) {
1689 + vma_writer_set_error(vmaw, "vma_writer_close: "
1690 + "incomplete stream '%s' (%zd != %zd)",
1691 + si->devname, si->transferred, si->size);
1692 + }
1693 + }
1694 + }
1695 +
1696 + for (i = 0; i <= 255; i++) {
1697 + vmaw->stream_info[i].finished = 1; /* mark as closed */
1698 + }
1699 +
1700 + vmaw->closed = 1;
1701 +
1702 + if (vmaw->status < 0 && *errp == NULL) {
1703 + error_setg(errp, "%s", vmaw->errmsg);
1704 + }
1705 +
1706 + return vmaw->status;
1707 +}
1708 +
1709 +void vma_writer_destroy(VmaWriter *vmaw)
1710 +{
1711 + assert(vmaw);
1712 +
1713 + int i;
1714 +
1715 + for (i = 0; i <= 255; i++) {
1716 + if (vmaw->stream_info[i].devname) {
1717 + g_free(vmaw->stream_info[i].devname);
1718 + }
1719 + }
1720 +
1721 + if (vmaw->md5csum) {
1722 + g_checksum_free(vmaw->md5csum);
1723 + }
1724 +
1725 + for (i = 0; i < WRITE_BUFFERS; i++) {
1726 + free(vmaw->aiocbs[i]);
1727 + }
1728 +
1729 + g_free(vmaw);
1730 +}
1731 Index: new/vma.c
1732 ===================================================================
1733 --- /dev/null 1970-01-01 00:00:00.000000000 +0000
1734 +++ new/vma.c 2014-11-20 08:47:23.000000000 +0100
1735 @@ -0,0 +1,582 @@
1736 +/*
1737 + * VMA: Virtual Machine Archive
1738 + *
1739 + * Copyright (C) 2012-2013 Proxmox Server Solutions
1740 + *
1741 + * Authors:
1742 + * Dietmar Maurer (dietmar@proxmox.com)
1743 + *
1744 + * This work is licensed under the terms of the GNU GPL, version 2 or later.
1745 + * See the COPYING file in the top-level directory.
1746 + *
1747 + */
1748 +
1749 +#include <stdio.h>
1750 +#include <errno.h>
1751 +#include <unistd.h>
1752 +#include <stdio.h>
1753 +#include <string.h>
1754 +#include <sys/types.h>
1755 +#include <sys/stat.h>
1756 +#include <fcntl.h>
1757 +#include <glib.h>
1758 +
1759 +#include "vma.h"
1760 +#include "qemu-common.h"
1761 +#include "qemu/error-report.h"
1762 +#include "qemu/main-loop.h"
1763 +
1764 +static void help(void)
1765 +{
1766 + const char *help_msg =
1767 + "usage: vma command [command options]\n"
1768 + "\n"
1769 + "vma list <filename>\n"
1770 + "vma create <filename> [-c config] <archive> pathname ...\n"
1771 + "vma extract <filename> [-r <fifo>] <targetdir>\n"
1772 + ;
1773 +
1774 + printf("%s", help_msg);
1775 + exit(1);
1776 +}
1777 +
1778 +static const char *extract_devname(const char *path, char **devname, int index)
1779 +{
1780 + assert(path);
1781 +
1782 + const char *sep = strchr(path, '=');
1783 +
1784 + if (sep) {
1785 + *devname = g_strndup(path, sep - path);
1786 + path = sep + 1;
1787 + } else {
1788 + if (index >= 0) {
1789 + *devname = g_strdup_printf("disk%d", index);
1790 + } else {
1791 + *devname = NULL;
1792 + }
1793 + }
1794 +
1795 + return path;
1796 +}
1797 +
1798 +static void print_content(VmaReader *vmar)
1799 +{
1800 + assert(vmar);
1801 +
1802 + VmaHeader *head = vma_reader_get_header(vmar);
1803 +
1804 + GList *l = vma_reader_get_config_data(vmar);
1805 + while (l && l->data) {
1806 + VmaConfigData *cdata = (VmaConfigData *)l->data;
1807 + l = g_list_next(l);
1808 + printf("CFG: size: %d name: %s\n", cdata->len, cdata->name);
1809 + }
1810 +
1811 + int i;
1812 + VmaDeviceInfo *di;
1813 + for (i = 1; i < 255; i++) {
1814 + di = vma_reader_get_device_info(vmar, i);
1815 + if (di) {
1816 + if (strcmp(di->devname, "vmstate") == 0) {
1817 + printf("VMSTATE: dev_id=%d memory: %zd\n", i, di->size);
1818 + } else {
1819 + printf("DEV: dev_id=%d size: %zd devname: %s\n",
1820 + i, di->size, di->devname);
1821 + }
1822 + }
1823 + }
1824 + /* ctime is the last entry we print */
1825 + printf("CTIME: %s", ctime(&head->ctime));
1826 + fflush(stdout);
1827 +}
1828 +
1829 +static int list_content(int argc, char **argv)
1830 +{
1831 + int c, ret = 0;
1832 + const char *filename;
1833 +
1834 + for (;;) {
1835 + c = getopt(argc, argv, "h");
1836 + if (c == -1) {
1837 + break;
1838 + }
1839 + switch (c) {
1840 + case '?':
1841 + case 'h':
1842 + help();
1843 + break;
1844 + default:
1845 + g_assert_not_reached();
1846 + }
1847 + }
1848 +
1849 + /* Get the filename */
1850 + if ((optind + 1) != argc) {
1851 + help();
1852 + }
1853 + filename = argv[optind++];
1854 +
1855 + Error *errp = NULL;
1856 + VmaReader *vmar = vma_reader_create(filename, &errp);
1857 +
1858 + if (!vmar) {
1859 + g_error("%s", error_get_pretty(errp));
1860 + }
1861 +
1862 + print_content(vmar);
1863 +
1864 + vma_reader_destroy(vmar);
1865 +
1866 + return ret;
1867 +}
1868 +
1869 +typedef struct RestoreMap {
1870 + char *devname;
1871 + char *path;
1872 + bool write_zero;
1873 +} RestoreMap;
1874 +
1875 +static int extract_content(int argc, char **argv)
1876 +{
1877 + int c, ret = 0;
1878 + int verbose = 0;
1879 + const char *filename;
1880 + const char *dirname;
1881 + const char *readmap = NULL;
1882 +
1883 + for (;;) {
1884 + c = getopt(argc, argv, "hvr:");
1885 + if (c == -1) {
1886 + break;
1887 + }
1888 + switch (c) {
1889 + case '?':
1890 + case 'h':
1891 + help();
1892 + break;
1893 + case 'r':
1894 + readmap = optarg;
1895 + break;
1896 + case 'v':
1897 + verbose = 1;
1898 + break;
1899 + default:
1900 + help();
1901 + }
1902 + }
1903 +
1904 + /* Get the filename */
1905 + if ((optind + 2) != argc) {
1906 + help();
1907 + }
1908 + filename = argv[optind++];
1909 + dirname = argv[optind++];
1910 +
1911 + Error *errp = NULL;
1912 + VmaReader *vmar = vma_reader_create(filename, &errp);
1913 +
1914 + if (!vmar) {
1915 + g_error("%s", error_get_pretty(errp));
1916 + }
1917 +
1918 + if (mkdir(dirname, 0777) < 0) {
1919 + g_error("unable to create target directory %s - %s",
1920 + dirname, g_strerror(errno));
1921 + }
1922 +
1923 + GList *l = vma_reader_get_config_data(vmar);
1924 + while (l && l->data) {
1925 + VmaConfigData *cdata = (VmaConfigData *)l->data;
1926 + l = g_list_next(l);
1927 + char *cfgfn = g_strdup_printf("%s/%s", dirname, cdata->name);
1928 + GError *err = NULL;
1929 + if (!g_file_set_contents(cfgfn, (gchar *)cdata->data, cdata->len,
1930 + &err)) {
1931 + g_error("unable to write file: %s", err->message);
1932 + }
1933 + }
1934 +
1935 + GHashTable *devmap = g_hash_table_new(g_str_hash, g_str_equal);
1936 +
1937 + if (readmap) {
1938 + print_content(vmar);
1939 +
1940 + FILE *map = fopen(readmap, "r");
1941 + if (!map) {
1942 + g_error("unable to open fifo %s - %s", readmap, g_strerror(errno));
1943 + }
1944 +
1945 + while (1) {
1946 + char inbuf[8192];
1947 + char *line = fgets(inbuf, sizeof(inbuf), map);
1948 + if (!line || line[0] == '\0' || !strcmp(line, "done\n")) {
1949 + break;
1950 + }
1951 + int len = strlen(line);
1952 + if (line[len - 1] == '\n') {
1953 + line[len - 1] = '\0';
1954 + if (len == 1) {
1955 + break;
1956 + }
1957 + }
1958 +
1959 + const char *path;
1960 + bool write_zero;
1961 + if (line[0] == '0' && line[1] == ':') {
1962 + path = inbuf + 2;
1963 + write_zero = false;
1964 + } else if (line[0] == '1' && line[1] == ':') {
1965 + path = inbuf + 2;
1966 + write_zero = true;
1967 + } else {
1968 + g_error("read map failed - parse error ('%s')", inbuf);
1969 + }
1970 +
1971 + char *devname = NULL;
1972 + path = extract_devname(path, &devname, -1);
1973 + if (!devname) {
1974 + g_error("read map failed - no dev name specified ('%s')",
1975 + inbuf);
1976 + }
1977 +
1978 + RestoreMap *map = g_new0(RestoreMap, 1);
1979 + map->devname = g_strdup(devname);
1980 + map->path = g_strdup(path);
1981 + map->write_zero = write_zero;
1982 +
1983 + g_hash_table_insert(devmap, map->devname, map);
1984 +
1985 + };
1986 + }
1987 +
1988 + int i;
1989 + int vmstate_fd = -1;
1990 + guint8 vmstate_stream = 0;
1991 +
1992 + for (i = 1; i < 255; i++) {
1993 + VmaDeviceInfo *di = vma_reader_get_device_info(vmar, i);
1994 + if (di && (strcmp(di->devname, "vmstate") == 0)) {
1995 + vmstate_stream = i;
1996 + char *statefn = g_strdup_printf("%s/vmstate.bin", dirname);
1997 + vmstate_fd = open(statefn, O_WRONLY|O_CREAT|O_EXCL, 0644);
1998 + if (vmstate_fd < 0) {
1999 + g_error("create vmstate file '%s' failed - %s", statefn,
2000 + g_strerror(errno));
2001 + }
2002 + g_free(statefn);
2003 + } else if (di) {
2004 + char *devfn = NULL;
2005 + int flags = BDRV_O_RDWR|BDRV_O_CACHE_WB;
2006 + bool write_zero = true;
2007 +
2008 + if (readmap) {
2009 + RestoreMap *map;
2010 + map = (RestoreMap *)g_hash_table_lookup(devmap, di->devname);
2011 + if (map == NULL) {
2012 + g_error("no device name mapping for %s", di->devname);
2013 + }
2014 + devfn = map->path;
2015 + write_zero = map->write_zero;
2016 + } else {
2017 + devfn = g_strdup_printf("%s/tmp-disk-%s.raw",
2018 + dirname, di->devname);
2019 + printf("DEVINFO %s %zd\n", devfn, di->size);
2020 +
2021 + bdrv_img_create(devfn, "raw", NULL, NULL, NULL, di->size,
2022 + flags, &errp, 0);
2023 + if (errp) {
2024 + g_error("can't create file %s: %s", devfn,
2025 + error_get_pretty(errp));
2026 + }
2027 +
2028 + /* Note: we created an empty file above, so there is no
2029 + * need to write zeroes (so we generate a sparse file)
2030 + */
2031 + write_zero = false;
2032 + }
2033 +
2034 + BlockDriverState *bs = bdrv_new();
2035 + if (errp || bdrv_open(&bs, devfn, NULL, NULL, flags, NULL, &errp)) {
2036 + g_error("can't open file %s - %s", devfn,
2037 + error_get_pretty(errp));
2038 + }
2039 + if (vma_reader_register_bs(vmar, i, bs, write_zero, &errp) < 0) {
2040 + g_error("%s", error_get_pretty(errp));
2041 + }
2042 +
2043 + if (!readmap) {
2044 + g_free(devfn);
2045 + }
2046 + }
2047 + }
2048 +
2049 + if (vma_reader_restore(vmar, vmstate_fd, verbose, &errp) < 0) {
2050 + g_error("restore failed - %s", error_get_pretty(errp));
2051 + }
2052 +
2053 + if (!readmap) {
2054 + for (i = 1; i < 255; i++) {
2055 + VmaDeviceInfo *di = vma_reader_get_device_info(vmar, i);
2056 + if (di && (i != vmstate_stream)) {
2057 + char *tmpfn = g_strdup_printf("%s/tmp-disk-%s.raw",
2058 + dirname, di->devname);
2059 + char *fn = g_strdup_printf("%s/disk-%s.raw",
2060 + dirname, di->devname);
2061 + if (rename(tmpfn, fn) != 0) {
2062 + g_error("rename %s to %s failed - %s",
2063 + tmpfn, fn, g_strerror(errno));
2064 + }
2065 + }
2066 + }
2067 + }
2068 +
2069 + vma_reader_destroy(vmar);
2070 +
2071 + bdrv_close_all();
2072 +
2073 + return ret;
2074 +}
2075 +
2076 +typedef struct BackupJob {
2077 + BlockDriverState *bs;
2078 + int64_t len;
2079 + VmaWriter *vmaw;
2080 + uint8_t dev_id;
2081 +} BackupJob;
2082 +
2083 +#define BACKUP_SECTORS_PER_CLUSTER (VMA_CLUSTER_SIZE / BDRV_SECTOR_SIZE)
2084 +
2085 +static void coroutine_fn backup_run(void *opaque)
2086 +{
2087 + BackupJob *job = (BackupJob *)opaque;
2088 + struct iovec iov;
2089 + QEMUIOVector qiov;
2090 +
2091 + int64_t start, end;
2092 + int ret = 0;
2093 +
2094 + unsigned char *buf = qemu_blockalign(job->bs, VMA_CLUSTER_SIZE);
2095 +
2096 + start = 0;
2097 + end = DIV_ROUND_UP(job->len / BDRV_SECTOR_SIZE,
2098 + BACKUP_SECTORS_PER_CLUSTER);
2099 +
2100 + for (; start < end; start++) {
2101 + iov.iov_base = buf;
2102 + iov.iov_len = VMA_CLUSTER_SIZE;
2103 + qemu_iovec_init_external(&qiov, &iov, 1);
2104 +
2105 + ret = bdrv_co_readv(job->bs, start * BACKUP_SECTORS_PER_CLUSTER,
2106 + BACKUP_SECTORS_PER_CLUSTER, &qiov);
2107 + if (ret < 0) {
2108 + vma_writer_set_error(job->vmaw, "read error", -1);
2109 + goto out;
2110 + }
2111 +
2112 + size_t zb = 0;
2113 + if (vma_writer_write(job->vmaw, job->dev_id, start, buf, &zb) < 0) {
2114 + vma_writer_set_error(job->vmaw, "backup_dump_cb vma_writer_write failed", -1);
2115 + goto out;
2116 + }
2117 + }
2118 +
2119 +
2120 +out:
2121 + if (vma_writer_close_stream(job->vmaw, job->dev_id) <= 0) {
2122 + Error *err = NULL;
2123 + if (vma_writer_close(job->vmaw, &err) != 0) {
2124 + g_warning("vma_writer_close failed %s", error_get_pretty(err));
2125 + }
2126 + }
2127 +}
2128 +
2129 +static int create_archive(int argc, char **argv)
2130 +{
2131 + int i, c, res;
2132 + int verbose = 0;
2133 + const char *archivename;
2134 + GList *config_files = NULL;
2135 +
2136 + for (;;) {
2137 + c = getopt(argc, argv, "hvc:");
2138 + if (c == -1) {
2139 + break;
2140 + }
2141 + switch (c) {
2142 + case '?':
2143 + case 'h':
2144 + help();
2145 + break;
2146 + case 'c':
2147 + config_files = g_list_append(config_files, optarg);
2148 + break;
2149 + case 'v':
2150 + verbose = 1;
2151 + break;
2152 + default:
2153 + g_assert_not_reached();
2154 + }
2155 + }
2156 +
2157 +
2158 + /* make sure we have archive name and at least one path */
2159 + if ((optind + 2) > argc) {
2160 + help();
2161 + }
2162 +
2163 + archivename = argv[optind++];
2164 +
2165 + uuid_t uuid;
2166 + uuid_generate(uuid);
2167 +
2168 + Error *local_err = NULL;
2169 + VmaWriter *vmaw = vma_writer_create(archivename, uuid, &local_err);
2170 +
2171 + if (vmaw == NULL) {
2172 + g_error("%s", error_get_pretty(local_err));
2173 + }
2174 +
2175 + GList *l = config_files;
2176 + while (l && l->data) {
2177 + char *name = l->data;
2178 + char *cdata = NULL;
2179 + gsize clen = 0;
2180 + GError *err = NULL;
2181 + if (!g_file_get_contents(name, &cdata, &clen, &err)) {
2182 + unlink(archivename);
2183 + g_error("Unable to read file: %s", err->message);
2184 + }
2185 +
2186 + if (vma_writer_add_config(vmaw, name, cdata, clen) != 0) {
2187 + unlink(archivename);
2188 + g_error("Unable to append config data %s (len = %zd)",
2189 + name, clen);
2190 + }
2191 + l = g_list_next(l);
2192 + }
2193 +
2194 + int ind = 0;
2195 + while (optind < argc) {
2196 + const char *path = argv[optind++];
2197 + char *devname = NULL;
2198 + path = extract_devname(path, &devname, ind++);
2199 +
2200 + BlockDriver *drv = NULL;
2201 + Error *errp = NULL;
2202 + BlockDriverState *bs = bdrv_new();
2203 +
2204 + res = bdrv_open(&bs, path, NULL, NULL, BDRV_O_CACHE_WB , drv, &errp);
2205 + if (res < 0) {
2206 + unlink(archivename);
2207 + g_error("bdrv_open '%s' failed - %s", path, error_get_pretty(errp));
2208 + }
2209 + int64_t size = bdrv_getlength(bs);
2210 + int dev_id = vma_writer_register_stream(vmaw, devname, size);
2211 + if (dev_id <= 0) {
2212 + unlink(archivename);
2213 + g_error("vma_writer_register_stream '%s' failed", devname);
2214 + }
2215 +
2216 + BackupJob *job = g_new0(BackupJob, 1);
2217 + job->len = size;
2218 + job->bs = bs;
2219 + job->vmaw = vmaw;
2220 + job->dev_id = dev_id;
2221 +
2222 + Coroutine *co = qemu_coroutine_create(backup_run);
2223 + qemu_coroutine_enter(co, job);
2224 + }
2225 +
2226 + VmaStatus vmastat;
2227 + int percent = 0;
2228 + int last_percent = -1;
2229 +
2230 + while (1) {
2231 + main_loop_wait(false);
2232 + vma_writer_get_status(vmaw, &vmastat);
2233 +
2234 + if (verbose) {
2235 +
2236 + uint64_t total = 0;
2237 + uint64_t transferred = 0;
2238 + uint64_t zero_bytes = 0;
2239 +
2240 + int i;
2241 + for (i = 0; i < 256; i++) {
2242 + if (vmastat.stream_info[i].size) {
2243 + total += vmastat.stream_info[i].size;
2244 + transferred += vmastat.stream_info[i].transferred;
2245 + zero_bytes += vmastat.stream_info[i].zero_bytes;
2246 + }
2247 + }
2248 + percent = (transferred*100)/total;
2249 + if (percent != last_percent) {
2250 + fprintf(stderr, "progress %d%% %zd/%zd %zd\n", percent,
2251 + transferred, total, zero_bytes);
2252 + fflush(stderr);
2253 +
2254 + last_percent = percent;
2255 + }
2256 + }
2257 +
2258 + if (vmastat.closed) {
2259 + break;
2260 + }
2261 + }
2262 +
2263 + bdrv_drain_all();
2264 +
2265 + vma_writer_get_status(vmaw, &vmastat);
2266 +
2267 + if (verbose) {
2268 + for (i = 0; i < 256; i++) {
2269 + VmaStreamInfo *si = &vmastat.stream_info[i];
2270 + if (si->size) {
2271 + fprintf(stderr, "image %s: size=%zd zeros=%zd saved=%zd\n",
2272 + si->devname, si->size, si->zero_bytes,
2273 + si->size - si->zero_bytes);
2274 + }
2275 + }
2276 + }
2277 +
2278 + if (vmastat.status < 0) {
2279 + unlink(archivename);
2280 + g_error("creating vma archive failed");
2281 + }
2282 +
2283 + return 0;
2284 +}
2285 +
2286 +int main(int argc, char **argv)
2287 +{
2288 + const char *cmdname;
2289 + Error *main_loop_err = NULL;
2290 +
2291 + error_set_progname(argv[0]);
2292 +
2293 + if (qemu_init_main_loop(&main_loop_err)) {
2294 + g_error("%s", error_get_pretty(main_loop_err));
2295 + }
2296 +
2297 + bdrv_init();
2298 +
2299 + if (argc < 2) {
2300 + help();
2301 + }
2302 +
2303 + cmdname = argv[1];
2304 + argc--; argv++;
2305 +
2306 +
2307 + if (!strcmp(cmdname, "list")) {
2308 + return list_content(argc, argv);
2309 + } else if (!strcmp(cmdname, "create")) {
2310 + return create_archive(argc, argv);
2311 + } else if (!strcmp(cmdname, "extract")) {
2312 + return extract_content(argc, argv);
2313 + }
2314 +
2315 + help();
2316 + return 0;
2317 +}
2318 Index: new/vma.h
2319 ===================================================================
2320 --- /dev/null 1970-01-01 00:00:00.000000000 +0000
2321 +++ new/vma.h 2014-11-20 08:15:12.000000000 +0100
2322 @@ -0,0 +1,146 @@
2323 +/*
2324 + * VMA: Virtual Machine Archive
2325 + *
2326 + * Copyright (C) Proxmox Server Solutions
2327 + *
2328 + * Authors:
2329 + * Dietmar Maurer (dietmar@proxmox.com)
2330 + *
2331 + * This work is licensed under the terms of the GNU GPL, version 2 or later.
2332 + * See the COPYING file in the top-level directory.
2333 + *
2334 + */
2335 +
2336 +#ifndef BACKUP_VMA_H
2337 +#define BACKUP_VMA_H
2338 +
2339 +#include <uuid/uuid.h>
2340 +#include "qapi/error.h"
2341 +#include "block/block.h"
2342 +
2343 +#define VMA_BLOCK_BITS 12
2344 +#define VMA_BLOCK_SIZE (1<<VMA_BLOCK_BITS)
2345 +#define VMA_CLUSTER_BITS (VMA_BLOCK_BITS+4)
2346 +#define VMA_CLUSTER_SIZE (1<<VMA_CLUSTER_BITS)
2347 +
2348 +#if VMA_CLUSTER_SIZE != 65536
2349 +#error unexpected cluster size
2350 +#endif
2351 +
2352 +#define VMA_EXTENT_HEADER_SIZE 512
2353 +#define VMA_BLOCKS_PER_EXTENT 59
2354 +#define VMA_MAX_CONFIGS 256
2355 +
2356 +#define VMA_MAX_EXTENT_SIZE \
2357 + (VMA_EXTENT_HEADER_SIZE+VMA_CLUSTER_SIZE*VMA_BLOCKS_PER_EXTENT)
2358 +#if VMA_MAX_EXTENT_SIZE != 3867136
2359 +#error unexpected VMA_EXTENT_SIZE
2360 +#endif
2361 +
2362 +/* File Format Definitions */
2363 +
2364 +#define VMA_MAGIC (GUINT32_TO_BE(('V'<<24)|('M'<<16)|('A'<<8)|0x00))
2365 +#define VMA_EXTENT_MAGIC (GUINT32_TO_BE(('V'<<24)|('M'<<16)|('A'<<8)|'E'))
2366 +
2367 +typedef struct VmaDeviceInfoHeader {
2368 + uint32_t devname_ptr; /* offset into blob_buffer table */
2369 + uint32_t reserved0;
2370 + uint64_t size; /* device size in bytes */
2371 + uint64_t reserved1;
2372 + uint64_t reserved2;
2373 +} VmaDeviceInfoHeader;
2374 +
2375 +typedef struct VmaHeader {
2376 + uint32_t magic;
2377 + uint32_t version;
2378 + unsigned char uuid[16];
2379 + int64_t ctime;
2380 + unsigned char md5sum[16];
2381 +
2382 + uint32_t blob_buffer_offset;
2383 + uint32_t blob_buffer_size;
2384 + uint32_t header_size;
2385 +
2386 + unsigned char reserved[1984];
2387 +
2388 + uint32_t config_names[VMA_MAX_CONFIGS]; /* offset into blob_buffer table */
2389 + uint32_t config_data[VMA_MAX_CONFIGS]; /* offset into blob_buffer table */
2390 +
2391 + uint32_t reserved1;
2392 +
2393 + VmaDeviceInfoHeader dev_info[256];
2394 +} VmaHeader;
2395 +
2396 +typedef struct VmaExtentHeader {
2397 + uint32_t magic;
2398 + uint16_t reserved1;
2399 + uint16_t block_count;
2400 + unsigned char uuid[16];
2401 + unsigned char md5sum[16];
2402 + uint64_t blockinfo[VMA_BLOCKS_PER_EXTENT];
2403 +} VmaExtentHeader;
2404 +
2405 +/* functions/definitions to read/write vma files */
2406 +
2407 +typedef struct VmaReader VmaReader;
2408 +
2409 +typedef struct VmaWriter VmaWriter;
2410 +
2411 +typedef struct VmaConfigData {
2412 + const char *name;
2413 + const void *data;
2414 + uint32_t len;
2415 +} VmaConfigData;
2416 +
2417 +typedef struct VmaStreamInfo {
2418 + uint64_t size;
2419 + uint64_t cluster_count;
2420 + uint64_t transferred;
2421 + uint64_t zero_bytes;
2422 + int finished;
2423 + char *devname;
2424 +} VmaStreamInfo;
2425 +
2426 +typedef struct VmaStatus {
2427 + int status;
2428 + bool closed;
2429 + char errmsg[8192];
2430 + char uuid_str[37];
2431 + VmaStreamInfo stream_info[256];
2432 +} VmaStatus;
2433 +
2434 +typedef struct VmaDeviceInfo {
2435 + uint64_t size; /* device size in bytes */
2436 + const char *devname;
2437 +} VmaDeviceInfo;
2438 +
2439 +VmaWriter *vma_writer_create(const char *filename, uuid_t uuid, Error **errp);
2440 +int vma_writer_close(VmaWriter *vmaw, Error **errp);
2441 +void vma_writer_destroy(VmaWriter *vmaw);
2442 +int vma_writer_add_config(VmaWriter *vmaw, const char *name, gpointer data,
2443 + size_t len);
2444 +int vma_writer_register_stream(VmaWriter *vmaw, const char *devname,
2445 + size_t size);
2446 +
2447 +int64_t coroutine_fn vma_writer_write(VmaWriter *vmaw, uint8_t dev_id,
2448 + int64_t cluster_num, unsigned char *buf,
2449 + size_t *zero_bytes);
2450 +
2451 +int coroutine_fn vma_writer_close_stream(VmaWriter *vmaw, uint8_t dev_id);
2452 +
2453 +int vma_writer_get_status(VmaWriter *vmaw, VmaStatus *status);
2454 +void vma_writer_set_error(VmaWriter *vmaw, const char *fmt, ...);
2455 +
2456 +
2457 +VmaReader *vma_reader_create(const char *filename, Error **errp);
2458 +void vma_reader_destroy(VmaReader *vmar);
2459 +VmaHeader *vma_reader_get_header(VmaReader *vmar);
2460 +GList *vma_reader_get_config_data(VmaReader *vmar);
2461 +VmaDeviceInfo *vma_reader_get_device_info(VmaReader *vmar, guint8 dev_id);
2462 +int vma_reader_register_bs(VmaReader *vmar, guint8 dev_id,
2463 + BlockDriverState *bs, bool write_zeroes,
2464 + Error **errp);
2465 +int vma_reader_restore(VmaReader *vmar, int vmstate_fd, bool verbose,
2466 + Error **errp);
2467 +
2468 +#endif /* BACKUP_VMA_H */
2469 Index: new/Makefile.objs
2470 ===================================================================
2471 --- new.orig/Makefile.objs 2014-11-20 08:12:54.000000000 +0100
2472 +++ new/Makefile.objs 2014-11-20 08:15:12.000000000 +0100
2473 @@ -13,6 +13,7 @@
2474 block-obj-$(CONFIG_WIN32) += aio-win32.o
2475 block-obj-y += block/
2476 block-obj-y += qemu-io-cmds.o
2477 +block-obj-y += vma-writer.o
2479 block-obj-y += qemu-coroutine.o qemu-coroutine-lock.o qemu-coroutine-io.o
2480 block-obj-y += qemu-coroutine-sleep.o