]> git.proxmox.com Git - pve-qemu.git/blob - debian/patches/pve/0026-PVE-Backup-add-vma-backup-format-code.patch
projects / pve-qemu.git / blob
? search:


ced8e74af871e15ab84da7fdf2ab067d27e0a2a2
[pve-qemu.git] / debian / patches / pve / 0026-PVE-Backup-add-vma-backup-format-code.patch
1 From 0000000000000000000000000000000000000000 Mon Sep 17 00:00:00 2001
2 From: Dietmar Maurer <dietmar@proxmox.com>
3 Date: Mon, 6 Apr 2020 12:16:57 +0200
4 Subject: [PATCH] PVE-Backup: add vma backup format code
5
6 Signed-off-by: Thomas Lamprecht <t.lamprecht@proxmox.com>
7 [FE: create: register all streams before entering coroutines]
8 Signed-off-by: Fabian Ebner <f.ebner@proxmox.com>
9 ---
10 block/meson.build | 2 +
11 meson.build | 5 +
12 vma-reader.c | 860 ++++++++++++++++++++++++++++++++++++++++++++++
13 vma-writer.c | 791 ++++++++++++++++++++++++++++++++++++++++++
14 vma.c | 850 +++++++++++++++++++++++++++++++++++++++++++++
15 vma.h | 150 ++++++++
16 6 files changed, 2658 insertions(+)
17 create mode 100644 vma-reader.c
18 create mode 100644 vma-writer.c
19 create mode 100644 vma.c
20 create mode 100644 vma.h
21
22 diff --git a/block/meson.build b/block/meson.build
23 index 9451fc3ccd..819eb80951 100644
24 --- a/block/meson.build
25 +++ b/block/meson.build
26 @@ -46,6 +46,8 @@ block_ss.add(files(
27 'zeroinit.c',
28 ), zstd, zlib, gnutls)
29
30 +block_ss.add(files('../vma-writer.c'), libuuid)
31 +
32 softmmu_ss.add(when: 'CONFIG_TCG', if_true: files('blkreplay.c'))
33
34 if get_option('qcow1').allowed()
35 diff --git a/meson.build b/meson.build
36 index 861de93c4f..96a91b95e4 100644
37 --- a/meson.build
38 +++ b/meson.build
39 @@ -1267,6 +1267,8 @@ keyutils = dependency('libkeyutils', required: false,
40
41 has_gettid = cc.has_function('gettid')
42
43 +libuuid = cc.find_library('uuid', required: true)
44 +
45 # libselinux
46 selinux = dependency('libselinux',
47 required: get_option('selinux'),
48 @@ -3359,6 +3361,9 @@ if have_tools
49 dependencies: [blockdev, qemuutil, gnutls, selinux],
50 install: true)
51
52 + vma = executable('vma', files('vma.c', 'vma-reader.c') + genh,
53 + dependencies: [authz, block, crypto, io, qom], install: true)
54 +
55 subdir('storage-daemon')
56 subdir('contrib/rdmacm-mux')
57 subdir('contrib/elf2dmp')
58 diff --git a/vma-reader.c b/vma-reader.c
59 new file mode 100644
60 index 0000000000..4f4ee2b47b
61 --- /dev/null
62 +++ b/vma-reader.c
63 @@ -0,0 +1,860 @@
64 +/*
65 + * VMA: Virtual Machine Archive
66 + *
67 + * Copyright (C) 2012 Proxmox Server Solutions
68 + *
69 + * Authors:
70 + * Dietmar Maurer (dietmar@proxmox.com)
71 + *
72 + * This work is licensed under the terms of the GNU GPL, version 2 or later.
73 + * See the COPYING file in the top-level directory.
74 + *
75 + */
76 +
77 +#include "qemu/osdep.h"
78 +#include <glib.h>
79 +#include <uuid/uuid.h>
80 +
81 +#include "qemu-common.h"
82 +#include "qemu/timer.h"
83 +#include "qemu/ratelimit.h"
84 +#include "vma.h"
85 +#include "block/block.h"
86 +#include "sysemu/block-backend.h"
87 +
88 +static unsigned char zero_vma_block[VMA_BLOCK_SIZE];
89 +
90 +typedef struct VmaRestoreState {
91 + BlockBackend *target;
92 + bool write_zeroes;
93 + unsigned long *bitmap;
94 + int bitmap_size;
95 +} VmaRestoreState;
96 +
97 +struct VmaReader {
98 + int fd;
99 + GChecksum *md5csum;
100 + GHashTable *blob_hash;
101 + unsigned char *head_data;
102 + VmaDeviceInfo devinfo[256];
103 + VmaRestoreState rstate[256];
104 + GList *cdata_list;
105 + guint8 vmstate_stream;
106 + uint32_t vmstate_clusters;
107 + /* to show restore percentage if run with -v */
108 + time_t start_time;
109 + int64_t cluster_count;
110 + int64_t clusters_read;
111 + int64_t zero_cluster_data;
112 + int64_t partial_zero_cluster_data;
113 + int clusters_read_per;
114 +};
115 +
116 +static guint
117 +g_int32_hash(gconstpointer v)
118 +{
119 + return *(const uint32_t *)v;
120 +}
121 +
122 +static gboolean
123 +g_int32_equal(gconstpointer v1, gconstpointer v2)
124 +{
125 + return *((const uint32_t *)v1) == *((const uint32_t *)v2);
126 +}
127 +
128 +static int vma_reader_get_bitmap(VmaRestoreState *rstate, int64_t cluster_num)
129 +{
130 + assert(rstate);
131 + assert(rstate->bitmap);
132 +
133 + unsigned long val, idx, bit;
134 +
135 + idx = cluster_num / BITS_PER_LONG;
136 +
137 + assert(rstate->bitmap_size > idx);
138 +
139 + bit = cluster_num % BITS_PER_LONG;
140 + val = rstate->bitmap[idx];
141 +
142 + return !!(val & (1UL << bit));
143 +}
144 +
145 +static void vma_reader_set_bitmap(VmaRestoreState *rstate, int64_t cluster_num,
146 + int dirty)
147 +{
148 + assert(rstate);
149 + assert(rstate->bitmap);
150 +
151 + unsigned long val, idx, bit;
152 +
153 + idx = cluster_num / BITS_PER_LONG;
154 +
155 + assert(rstate->bitmap_size > idx);
156 +
157 + bit = cluster_num % BITS_PER_LONG;
158 + val = rstate->bitmap[idx];
159 + if (dirty) {
160 + if (!(val & (1UL << bit))) {
161 + val |= 1UL << bit;
162 + }
163 + } else {
164 + if (val & (1UL << bit)) {
165 + val &= ~(1UL << bit);
166 + }
167 + }
168 + rstate->bitmap[idx] = val;
169 +}
170 +
171 +typedef struct VmaBlob {
172 + uint32_t start;
173 + uint32_t len;
174 + void *data;
175 +} VmaBlob;
176 +
177 +static const VmaBlob *get_header_blob(VmaReader *vmar, uint32_t pos)
178 +{
179 + assert(vmar);
180 + assert(vmar->blob_hash);
181 +
182 + return g_hash_table_lookup(vmar->blob_hash, &pos);
183 +}
184 +
185 +static const char *get_header_str(VmaReader *vmar, uint32_t pos)
186 +{
187 + const VmaBlob *blob = get_header_blob(vmar, pos);
188 + if (!blob) {
189 + return NULL;
190 + }
191 + const char *res = (char *)blob->data;
192 + if (res[blob->len-1] != '\0') {
193 + return NULL;
194 + }
195 + return res;
196 +}
197 +
198 +static ssize_t
199 +safe_read(int fd, unsigned char *buf, size_t count)
200 +{
201 + ssize_t n;
202 +
203 + do {
204 + n = read(fd, buf, count);
205 + } while (n < 0 && errno == EINTR);
206 +
207 + return n;
208 +}
209 +
210 +static ssize_t
211 +full_read(int fd, unsigned char *buf, size_t len)
212 +{
213 + ssize_t n;
214 + size_t total;
215 +
216 + total = 0;
217 +
218 + while (len > 0) {
219 + n = safe_read(fd, buf, len);
220 +
221 + if (n == 0) {
222 + return total;
223 + }
224 +
225 + if (n <= 0) {
226 + break;
227 + }
228 +
229 + buf += n;
230 + total += n;
231 + len -= n;
232 + }
233 +
234 + if (len) {
235 + return -1;
236 + }
237 +
238 + return total;
239 +}
240 +
241 +void vma_reader_destroy(VmaReader *vmar)
242 +{
243 + assert(vmar);
244 +
245 + if (vmar->fd >= 0) {
246 + close(vmar->fd);
247 + }
248 +
249 + if (vmar->cdata_list) {
250 + g_list_free(vmar->cdata_list);
251 + }
252 +
253 + int i;
254 + for (i = 1; i < 256; i++) {
255 + if (vmar->rstate[i].bitmap) {
256 + g_free(vmar->rstate[i].bitmap);
257 + }
258 + if (vmar->rstate[i].target) {
259 + blk_unref(vmar->rstate[i].target);
260 + }
261 + }
262 +
263 + if (vmar->md5csum) {
264 + g_checksum_free(vmar->md5csum);
265 + }
266 +
267 + if (vmar->blob_hash) {
268 + g_hash_table_destroy(vmar->blob_hash);
269 + }
270 +
271 + if (vmar->head_data) {
272 + g_free(vmar->head_data);
273 + }
274 +
275 + g_free(vmar);
276 +
277 +};
278 +
279 +static int vma_reader_read_head(VmaReader *vmar, Error **errp)
280 +{
281 + assert(vmar);
282 + assert(errp);
283 + assert(*errp == NULL);
284 +
285 + unsigned char md5sum[16];
286 + int i;
287 + int ret = 0;
288 +
289 + vmar->head_data = g_malloc(sizeof(VmaHeader));
290 +
291 + if (full_read(vmar->fd, vmar->head_data, sizeof(VmaHeader)) !=
292 + sizeof(VmaHeader)) {
293 + error_setg(errp, "can't read vma header - %s",
294 + errno ? g_strerror(errno) : "got EOF");
295 + return -1;
296 + }
297 +
298 + VmaHeader *h = (VmaHeader *)vmar->head_data;
299 +
300 + if (h->magic != VMA_MAGIC) {
301 + error_setg(errp, "not a vma file - wrong magic number");
302 + return -1;
303 + }
304 +
305 + uint32_t header_size = GUINT32_FROM_BE(h->header_size);
306 + int need = header_size - sizeof(VmaHeader);
307 + if (need <= 0) {
308 + error_setg(errp, "wrong vma header size %d", header_size);
309 + return -1;
310 + }
311 +
312 + vmar->head_data = g_realloc(vmar->head_data, header_size);
313 + h = (VmaHeader *)vmar->head_data;
314 +
315 + if (full_read(vmar->fd, vmar->head_data + sizeof(VmaHeader), need) !=
316 + need) {
317 + error_setg(errp, "can't read vma header data - %s",
318 + errno ? g_strerror(errno) : "got EOF");
319 + return -1;
320 + }
321 +
322 + memcpy(md5sum, h->md5sum, 16);
323 + memset(h->md5sum, 0, 16);
324 +
325 + g_checksum_reset(vmar->md5csum);
326 + g_checksum_update(vmar->md5csum, vmar->head_data, header_size);
327 + gsize csize = 16;
328 + g_checksum_get_digest(vmar->md5csum, (guint8 *)(h->md5sum), &csize);
329 +
330 + if (memcmp(md5sum, h->md5sum, 16) != 0) {
331 + error_setg(errp, "wrong vma header chechsum");
332 + return -1;
333 + }
334 +
335 + /* we can modify header data after checksum verify */
336 + h->header_size = header_size;
337 +
338 + h->version = GUINT32_FROM_BE(h->version);
339 + if (h->version != 1) {
340 + error_setg(errp, "wrong vma version %d", h->version);
341 + return -1;
342 + }
343 +
344 + h->ctime = GUINT64_FROM_BE(h->ctime);
345 + h->blob_buffer_offset = GUINT32_FROM_BE(h->blob_buffer_offset);
346 + h->blob_buffer_size = GUINT32_FROM_BE(h->blob_buffer_size);
347 +
348 + uint32_t bstart = h->blob_buffer_offset + 1;
349 + uint32_t bend = h->blob_buffer_offset + h->blob_buffer_size;
350 +
351 + if (bstart <= sizeof(VmaHeader)) {
352 + error_setg(errp, "wrong vma blob buffer offset %d",
353 + h->blob_buffer_offset);
354 + return -1;
355 + }
356 +
357 + if (bend > header_size) {
358 + error_setg(errp, "wrong vma blob buffer size %d/%d",
359 + h->blob_buffer_offset, h->blob_buffer_size);
360 + return -1;
361 + }
362 +
363 + while ((bstart + 2) <= bend) {
364 + uint32_t size = vmar->head_data[bstart] +
365 + (vmar->head_data[bstart+1] << 8);
366 + if ((bstart + size + 2) <= bend) {
367 + VmaBlob *blob = g_new0(VmaBlob, 1);
368 + blob->start = bstart - h->blob_buffer_offset;
369 + blob->len = size;
370 + blob->data = vmar->head_data + bstart + 2;
371 + g_hash_table_insert(vmar->blob_hash, &blob->start, blob);
372 + }
373 + bstart += size + 2;
374 + }
375 +
376 +
377 + int count = 0;
378 + for (i = 1; i < 256; i++) {
379 + VmaDeviceInfoHeader *dih = &h->dev_info[i];
380 + uint32_t devname_ptr = GUINT32_FROM_BE(dih->devname_ptr);
381 + uint64_t size = GUINT64_FROM_BE(dih->size);
382 + const char *devname = get_header_str(vmar, devname_ptr);
383 +
384 + if (size && devname) {
385 + count++;
386 + vmar->devinfo[i].size = size;
387 + vmar->devinfo[i].devname = devname;
388 +
389 + if (strcmp(devname, "vmstate") == 0) {
390 + vmar->vmstate_stream = i;
391 + }
392 + }
393 + }
394 +
395 + for (i = 0; i < VMA_MAX_CONFIGS; i++) {
396 + uint32_t name_ptr = GUINT32_FROM_BE(h->config_names[i]);
397 + uint32_t data_ptr = GUINT32_FROM_BE(h->config_data[i]);
398 +
399 + if (!(name_ptr && data_ptr)) {
400 + continue;
401 + }
402 + const char *name = get_header_str(vmar, name_ptr);
403 + const VmaBlob *blob = get_header_blob(vmar, data_ptr);
404 +
405 + if (!(name && blob)) {
406 + error_setg(errp, "vma contains invalid data pointers");
407 + return -1;
408 + }
409 +
410 + VmaConfigData *cdata = g_new0(VmaConfigData, 1);
411 + cdata->name = name;
412 + cdata->data = blob->data;
413 + cdata->len = blob->len;
414 +
415 + vmar->cdata_list = g_list_append(vmar->cdata_list, cdata);
416 + }
417 +
418 + return ret;
419 +};
420 +
421 +VmaReader *vma_reader_create(const char *filename, Error **errp)
422 +{
423 + assert(filename);
424 + assert(errp);
425 +
426 + VmaReader *vmar = g_new0(VmaReader, 1);
427 +
428 + if (strcmp(filename, "-") == 0) {
429 + vmar->fd = dup(0);
430 + } else {
431 + vmar->fd = open(filename, O_RDONLY);
432 + }
433 +
434 + if (vmar->fd < 0) {
435 + error_setg(errp, "can't open file %s - %s\n", filename,
436 + g_strerror(errno));
437 + goto err;
438 + }
439 +
440 + vmar->md5csum = g_checksum_new(G_CHECKSUM_MD5);
441 + if (!vmar->md5csum) {
442 + error_setg(errp, "can't allocate cmsum\n");
443 + goto err;
444 + }
445 +
446 + vmar->blob_hash = g_hash_table_new_full(g_int32_hash, g_int32_equal,
447 + NULL, g_free);
448 +
449 + if (vma_reader_read_head(vmar, errp) < 0) {
450 + goto err;
451 + }
452 +
453 + return vmar;
454 +
455 +err:
456 + if (vmar) {
457 + vma_reader_destroy(vmar);
458 + }
459 +
460 + return NULL;
461 +}
462 +
463 +VmaHeader *vma_reader_get_header(VmaReader *vmar)
464 +{
465 + assert(vmar);
466 + assert(vmar->head_data);
467 +
468 + return (VmaHeader *)(vmar->head_data);
469 +}
470 +
471 +GList *vma_reader_get_config_data(VmaReader *vmar)
472 +{
473 + assert(vmar);
474 + assert(vmar->head_data);
475 +
476 + return vmar->cdata_list;
477 +}
478 +
479 +VmaDeviceInfo *vma_reader_get_device_info(VmaReader *vmar, guint8 dev_id)
480 +{
481 + assert(vmar);
482 + assert(dev_id);
483 +
484 + if (vmar->devinfo[dev_id].size && vmar->devinfo[dev_id].devname) {
485 + return &vmar->devinfo[dev_id];
486 + }
487 +
488 + return NULL;
489 +}
490 +
491 +static void allocate_rstate(VmaReader *vmar, guint8 dev_id,
492 + BlockBackend *target, bool write_zeroes)
493 +{
494 + assert(vmar);
495 + assert(dev_id);
496 +
497 + vmar->rstate[dev_id].target = target;
498 + vmar->rstate[dev_id].write_zeroes = write_zeroes;
499 +
500 + int64_t size = vmar->devinfo[dev_id].size;
501 +
502 + int64_t bitmap_size = (size/BDRV_SECTOR_SIZE) +
503 + (VMA_CLUSTER_SIZE/BDRV_SECTOR_SIZE) * BITS_PER_LONG - 1;
504 + bitmap_size /= (VMA_CLUSTER_SIZE/BDRV_SECTOR_SIZE) * BITS_PER_LONG;
505 +
506 + vmar->rstate[dev_id].bitmap_size = bitmap_size;
507 + vmar->rstate[dev_id].bitmap = g_new0(unsigned long, bitmap_size);
508 +
509 + vmar->cluster_count += size/VMA_CLUSTER_SIZE;
510 +}
511 +
512 +int vma_reader_register_bs(VmaReader *vmar, guint8 dev_id, BlockBackend *target,
513 + bool write_zeroes, Error **errp)
514 +{
515 + assert(vmar);
516 + assert(target != NULL);
517 + assert(dev_id);
518 + assert(vmar->rstate[dev_id].target == NULL);
519 +
520 + int64_t size = blk_getlength(target);
521 + int64_t size_diff = size - vmar->devinfo[dev_id].size;
522 +
523 + /* storage types can have different size restrictions, so it
524 + * is not always possible to create an image with exact size.
525 + * So we tolerate a size difference up to 4MB.
526 + */
527 + if ((size_diff < 0) || (size_diff > 4*1024*1024)) {
528 + error_setg(errp, "vma_reader_register_bs for stream %s failed - "
529 + "unexpected size %zd != %zd", vmar->devinfo[dev_id].devname,
530 + size, vmar->devinfo[dev_id].size);
531 + return -1;
532 + }
533 +
534 + allocate_rstate(vmar, dev_id, target, write_zeroes);
535 +
536 + return 0;
537 +}
538 +
539 +static ssize_t safe_write(int fd, void *buf, size_t count)
540 +{
541 + ssize_t n;
542 +
543 + do {
544 + n = write(fd, buf, count);
545 + } while (n < 0 && errno == EINTR);
546 +
547 + return n;
548 +}
549 +
550 +static size_t full_write(int fd, void *buf, size_t len)
551 +{
552 + ssize_t n;
553 + size_t total;
554 +
555 + total = 0;
556 +
557 + while (len > 0) {
558 + n = safe_write(fd, buf, len);
559 + if (n < 0) {
560 + return n;
561 + }
562 + buf += n;
563 + total += n;
564 + len -= n;
565 + }
566 +
567 + if (len) {
568 + /* incomplete write ? */
569 + return -1;
570 + }
571 +
572 + return total;
573 +}
574 +
575 +static int restore_write_data(VmaReader *vmar, guint8 dev_id,
576 + BlockBackend *target, int vmstate_fd,
577 + unsigned char *buf, int64_t sector_num,
578 + int nb_sectors, Error **errp)
579 +{
580 + assert(vmar);
581 +
582 + if (dev_id == vmar->vmstate_stream) {
583 + if (vmstate_fd >= 0) {
584 + int len = nb_sectors * BDRV_SECTOR_SIZE;
585 + int res = full_write(vmstate_fd, buf, len);
586 + if (res < 0) {
587 + error_setg(errp, "write vmstate failed %d", res);
588 + return -1;
589 + }
590 + }
591 + } else {
592 + int res = blk_pwrite(target, sector_num * BDRV_SECTOR_SIZE, buf, nb_sectors * BDRV_SECTOR_SIZE, 0);
593 + if (res < 0) {
594 + error_setg(errp, "blk_pwrite to %s failed (%d)",
595 + bdrv_get_device_name(blk_bs(target)), res);
596 + return -1;
597 + }
598 + }
599 + return 0;
600 +}
601 +
602 +static int restore_extent(VmaReader *vmar, unsigned char *buf,
603 + int extent_size, int vmstate_fd,
604 + bool verbose, bool verify, Error **errp)
605 +{
606 + assert(vmar);
607 + assert(buf);
608 +
609 + VmaExtentHeader *ehead = (VmaExtentHeader *)buf;
610 + int start = VMA_EXTENT_HEADER_SIZE;
611 + int i;
612 +
613 + for (i = 0; i < VMA_BLOCKS_PER_EXTENT; i++) {
614 + uint64_t block_info = GUINT64_FROM_BE(ehead->blockinfo[i]);
615 + uint64_t cluster_num = block_info & 0xffffffff;
616 + uint8_t dev_id = (block_info >> 32) & 0xff;
617 + uint16_t mask = block_info >> (32+16);
618 + int64_t max_sector;
619 +
620 + if (!dev_id) {
621 + continue;
622 + }
623 +
624 + VmaRestoreState *rstate = &vmar->rstate[dev_id];
625 + BlockBackend *target = NULL;
626 +
627 + if (dev_id != vmar->vmstate_stream) {
628 + target = rstate->target;
629 + if (!verify && !target) {
630 + error_setg(errp, "got wrong dev id %d", dev_id);
631 + return -1;
632 + }
633 +
634 + if (vma_reader_get_bitmap(rstate, cluster_num)) {
635 + error_setg(errp, "found duplicated cluster %zd for stream %s",
636 + cluster_num, vmar->devinfo[dev_id].devname);
637 + return -1;
638 + }
639 + vma_reader_set_bitmap(rstate, cluster_num, 1);
640 +
641 + max_sector = vmar->devinfo[dev_id].size/BDRV_SECTOR_SIZE;
642 + } else {
643 + max_sector = G_MAXINT64;
644 + if (cluster_num != vmar->vmstate_clusters) {
645 + error_setg(errp, "found out of order vmstate data");
646 + return -1;
647 + }
648 + vmar->vmstate_clusters++;
649 + }
650 +
651 + vmar->clusters_read++;
652 +
653 + if (verbose) {
654 + time_t duration = time(NULL) - vmar->start_time;
655 + int percent = (vmar->clusters_read*100)/vmar->cluster_count;
656 + if (percent != vmar->clusters_read_per) {
657 + printf("progress %d%% (read %zd bytes, duration %zd sec)\n",
658 + percent, vmar->clusters_read*VMA_CLUSTER_SIZE,
659 + duration);
660 + fflush(stdout);
661 + vmar->clusters_read_per = percent;
662 + }
663 + }
664 +
665 + /* try to write whole clusters to speedup restore */
666 + if (mask == 0xffff) {
667 + if ((start + VMA_CLUSTER_SIZE) > extent_size) {
668 + error_setg(errp, "short vma extent - too many blocks");
669 + return -1;
670 + }
671 + int64_t sector_num = (cluster_num * VMA_CLUSTER_SIZE) /
672 + BDRV_SECTOR_SIZE;
673 + int64_t end_sector = sector_num +
674 + VMA_CLUSTER_SIZE/BDRV_SECTOR_SIZE;
675 +
676 + if (end_sector > max_sector) {
677 + end_sector = max_sector;
678 + }
679 +
680 + if (end_sector <= sector_num) {
681 + error_setg(errp, "got wrong block address - write beyond end");
682 + return -1;
683 + }
684 +
685 + if (!verify) {
686 + int nb_sectors = end_sector - sector_num;
687 + if (restore_write_data(vmar, dev_id, target, vmstate_fd,
688 + buf + start, sector_num, nb_sectors,
689 + errp) < 0) {
690 + return -1;
691 + }
692 + }
693 +
694 + start += VMA_CLUSTER_SIZE;
695 + } else {
696 + int j;
697 + int bit = 1;
698 +
699 + for (j = 0; j < 16; j++) {
700 + int64_t sector_num = (cluster_num*VMA_CLUSTER_SIZE +
701 + j*VMA_BLOCK_SIZE)/BDRV_SECTOR_SIZE;
702 +
703 + int64_t end_sector = sector_num +
704 + VMA_BLOCK_SIZE/BDRV_SECTOR_SIZE;
705 + if (end_sector > max_sector) {
706 + end_sector = max_sector;
707 + }
708 +
709 + if (mask & bit) {
710 + if ((start + VMA_BLOCK_SIZE) > extent_size) {
711 + error_setg(errp, "short vma extent - too many blocks");
712 + return -1;
713 + }
714 +
715 + if (end_sector <= sector_num) {
716 + error_setg(errp, "got wrong block address - "
717 + "write beyond end");
718 + return -1;
719 + }
720 +
721 + if (!verify) {
722 + int nb_sectors = end_sector - sector_num;
723 + if (restore_write_data(vmar, dev_id, target, vmstate_fd,
724 + buf + start, sector_num,
725 + nb_sectors, errp) < 0) {
726 + return -1;
727 + }
728 + }
729 +
730 + start += VMA_BLOCK_SIZE;
731 +
732 + } else {
733 +
734 +
735 + if (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 + int zero_size = BDRV_SECTOR_SIZE*nb_sectors;
741 + vmar->zero_cluster_data += zero_size;
742 + if (mask != 0) {
743 + vmar->partial_zero_cluster_data += zero_size;
744 + }
745 +
746 + if (rstate->write_zeroes && !verify) {
747 + if (restore_write_data(vmar, dev_id, target, vmstate_fd,
748 + zero_vma_block, sector_num,
749 + nb_sectors, errp) < 0) {
750 + return -1;
751 + }
752 + }
753 + }
754 + }
755 +
756 + bit = bit << 1;
757 + }
758 + }
759 + }
760 +
761 + if (start != extent_size) {
762 + error_setg(errp, "vma extent error - missing blocks");
763 + return -1;
764 + }
765 +
766 + return 0;
767 +}
768 +
769 +static int vma_reader_restore_full(VmaReader *vmar, int vmstate_fd,
770 + bool verbose, bool verify,
771 + Error **errp)
772 +{
773 + assert(vmar);
774 + assert(vmar->head_data);
775 +
776 + int ret = 0;
777 + unsigned char buf[VMA_MAX_EXTENT_SIZE];
778 + int buf_pos = 0;
779 + unsigned char md5sum[16];
780 + VmaHeader *h = (VmaHeader *)vmar->head_data;
781 +
782 + vmar->start_time = time(NULL);
783 +
784 + while (1) {
785 + int bytes = full_read(vmar->fd, buf + buf_pos, sizeof(buf) - buf_pos);
786 + if (bytes < 0) {
787 + error_setg(errp, "read failed - %s", g_strerror(errno));
788 + return -1;
789 + }
790 +
791 + buf_pos += bytes;
792 +
793 + if (!buf_pos) {
794 + break; /* EOF */
795 + }
796 +
797 + if (buf_pos < VMA_EXTENT_HEADER_SIZE) {
798 + error_setg(errp, "read short extent (%d bytes)", buf_pos);
799 + return -1;
800 + }
801 +
802 + VmaExtentHeader *ehead = (VmaExtentHeader *)buf;
803 +
804 + /* extract md5sum */
805 + memcpy(md5sum, ehead->md5sum, sizeof(ehead->md5sum));
806 + memset(ehead->md5sum, 0, sizeof(ehead->md5sum));
807 +
808 + g_checksum_reset(vmar->md5csum);
809 + g_checksum_update(vmar->md5csum, buf, VMA_EXTENT_HEADER_SIZE);
810 + gsize csize = 16;
811 + g_checksum_get_digest(vmar->md5csum, ehead->md5sum, &csize);
812 +
813 + if (memcmp(md5sum, ehead->md5sum, 16) != 0) {
814 + error_setg(errp, "wrong vma extent header chechsum");
815 + return -1;
816 + }
817 +
818 + if (memcmp(h->uuid, ehead->uuid, sizeof(ehead->uuid)) != 0) {
819 + error_setg(errp, "wrong vma extent uuid");
820 + return -1;
821 + }
822 +
823 + if (ehead->magic != VMA_EXTENT_MAGIC || ehead->reserved1 != 0) {
824 + error_setg(errp, "wrong vma extent header magic");
825 + return -1;
826 + }
827 +
828 + int block_count = GUINT16_FROM_BE(ehead->block_count);
829 + int extent_size = VMA_EXTENT_HEADER_SIZE + block_count*VMA_BLOCK_SIZE;
830 +
831 + if (buf_pos < extent_size) {
832 + error_setg(errp, "short vma extent (%d < %d)", buf_pos,
833 + extent_size);
834 + return -1;
835 + }
836 +
837 + if (restore_extent(vmar, buf, extent_size, vmstate_fd, verbose,
838 + verify, errp) < 0) {
839 + return -1;
840 + }
841 +
842 + if (buf_pos > extent_size) {
843 + memmove(buf, buf + extent_size, buf_pos - extent_size);
844 + buf_pos = buf_pos - extent_size;
845 + } else {
846 + buf_pos = 0;
847 + }
848 + }
849 +
850 + bdrv_drain_all();
851 +
852 + int i;
853 + for (i = 1; i < 256; i++) {
854 + VmaRestoreState *rstate = &vmar->rstate[i];
855 + if (!rstate->target) {
856 + continue;
857 + }
858 +
859 + if (blk_flush(rstate->target) < 0) {
860 + error_setg(errp, "vma blk_flush %s failed",
861 + vmar->devinfo[i].devname);
862 + return -1;
863 + }
864 +
865 + if (vmar->devinfo[i].size &&
866 + (strcmp(vmar->devinfo[i].devname, "vmstate") != 0)) {
867 + assert(rstate->bitmap);
868 +
869 + int64_t cluster_num, end;
870 +
871 + end = (vmar->devinfo[i].size + VMA_CLUSTER_SIZE - 1) /
872 + VMA_CLUSTER_SIZE;
873 +
874 + for (cluster_num = 0; cluster_num < end; cluster_num++) {
875 + if (!vma_reader_get_bitmap(rstate, cluster_num)) {
876 + error_setg(errp, "detected missing cluster %zd "
877 + "for stream %s", cluster_num,
878 + vmar->devinfo[i].devname);
879 + return -1;
880 + }
881 + }
882 + }
883 + }
884 +
885 + if (verbose) {
886 + if (vmar->clusters_read) {
887 + printf("total bytes read %zd, sparse bytes %zd (%.3g%%)\n",
888 + vmar->clusters_read*VMA_CLUSTER_SIZE,
889 + vmar->zero_cluster_data,
890 + (double)(100.0*vmar->zero_cluster_data)/
891 + (vmar->clusters_read*VMA_CLUSTER_SIZE));
892 +
893 + int64_t datasize = vmar->clusters_read*VMA_CLUSTER_SIZE-vmar->zero_cluster_data;
894 + if (datasize) { // this does not make sense for empty files
895 + printf("space reduction due to 4K zero blocks %.3g%%\n",
896 + (double)(100.0*vmar->partial_zero_cluster_data) / datasize);
897 + }
898 + } else {
899 + printf("vma archive contains no image data\n");
900 + }
901 + }
902 + return ret;
903 +}
904 +
905 +int vma_reader_restore(VmaReader *vmar, int vmstate_fd, bool verbose,
906 + Error **errp)
907 +{
908 + return vma_reader_restore_full(vmar, vmstate_fd, verbose, false, errp);
909 +}
910 +
911 +int vma_reader_verify(VmaReader *vmar, bool verbose, Error **errp)
912 +{
913 + guint8 dev_id;
914 +
915 + for (dev_id = 1; dev_id < 255; dev_id++) {
916 + if (vma_reader_get_device_info(vmar, dev_id)) {
917 + allocate_rstate(vmar, dev_id, NULL, false);
918 + }
919 + }
920 +
921 + return vma_reader_restore_full(vmar, -1, verbose, true, errp);
922 +}
923 +
924 diff --git a/vma-writer.c b/vma-writer.c
925 new file mode 100644
926 index 0000000000..df4b20793d
927 --- /dev/null
928 +++ b/vma-writer.c
929 @@ -0,0 +1,791 @@
930 +/*
931 + * VMA: Virtual Machine Archive
932 + *
933 + * Copyright (C) 2012 Proxmox Server Solutions
934 + *
935 + * Authors:
936 + * Dietmar Maurer (dietmar@proxmox.com)
937 + *
938 + * This work is licensed under the terms of the GNU GPL, version 2 or later.
939 + * See the COPYING file in the top-level directory.
940 + *
941 + */
942 +
943 +#include "qemu/osdep.h"
944 +#include <glib.h>
945 +#include <uuid/uuid.h>
946 +
947 +#include "vma.h"
948 +#include "block/block.h"
949 +#include "monitor/monitor.h"
950 +#include "qemu/main-loop.h"
951 +#include "qemu/coroutine.h"
952 +#include "qemu/cutils.h"
953 +#include "qemu/memalign.h"
954 +
955 +#define DEBUG_VMA 0
956 +
957 +#define DPRINTF(fmt, ...)\
958 + do { if (DEBUG_VMA) { printf("vma: " fmt, ## __VA_ARGS__); } } while (0)
959 +
960 +#define WRITE_BUFFERS 5
961 +#define HEADER_CLUSTERS 8
962 +#define HEADERBUF_SIZE (VMA_CLUSTER_SIZE*HEADER_CLUSTERS)
963 +
964 +struct VmaWriter {
965 + int fd;
966 + FILE *cmd;
967 + int status;
968 + char errmsg[8192];
969 + uuid_t uuid;
970 + bool header_written;
971 + bool closed;
972 +
973 + /* we always write extents */
974 + unsigned char *outbuf;
975 + int outbuf_pos; /* in bytes */
976 + int outbuf_count; /* in VMA_BLOCKS */
977 + uint64_t outbuf_block_info[VMA_BLOCKS_PER_EXTENT];
978 +
979 + unsigned char *headerbuf;
980 +
981 + GChecksum *md5csum;
982 + CoMutex flush_lock;
983 + Coroutine *co_writer;
984 +
985 + /* drive informations */
986 + VmaStreamInfo stream_info[256];
987 + guint stream_count;
988 +
989 + guint8 vmstate_stream;
990 + uint32_t vmstate_clusters;
991 +
992 + /* header blob table */
993 + char *header_blob_table;
994 + uint32_t header_blob_table_size;
995 + uint32_t header_blob_table_pos;
996 +
997 + /* store for config blobs */
998 + uint32_t config_names[VMA_MAX_CONFIGS]; /* offset into blob_buffer table */
999 + uint32_t config_data[VMA_MAX_CONFIGS]; /* offset into blob_buffer table */
1000 + uint32_t config_count;
1001 +};
1002 +
1003 +void vma_writer_set_error(VmaWriter *vmaw, const char *fmt, ...)
1004 +{
1005 + va_list ap;
1006 +
1007 + if (vmaw->status < 0) {
1008 + return;
1009 + }
1010 +
1011 + vmaw->status = -1;
1012 +
1013 + va_start(ap, fmt);
1014 + g_vsnprintf(vmaw->errmsg, sizeof(vmaw->errmsg), fmt, ap);
1015 + va_end(ap);
1016 +
1017 + DPRINTF("vma_writer_set_error: %s\n", vmaw->errmsg);
1018 +}
1019 +
1020 +static uint32_t allocate_header_blob(VmaWriter *vmaw, const char *data,
1021 + size_t len)
1022 +{
1023 + if (len > 65535) {
1024 + return 0;
1025 + }
1026 +
1027 + if (!vmaw->header_blob_table ||
1028 + (vmaw->header_blob_table_size <
1029 + (vmaw->header_blob_table_pos + len + 2))) {
1030 + int newsize = vmaw->header_blob_table_size + ((len + 2 + 511)/512)*512;
1031 +
1032 + vmaw->header_blob_table = g_realloc(vmaw->header_blob_table, newsize);
1033 + memset(vmaw->header_blob_table + vmaw->header_blob_table_size,
1034 + 0, newsize - vmaw->header_blob_table_size);
1035 + vmaw->header_blob_table_size = newsize;
1036 + }
1037 +
1038 + uint32_t cpos = vmaw->header_blob_table_pos;
1039 + vmaw->header_blob_table[cpos] = len & 255;
1040 + vmaw->header_blob_table[cpos+1] = (len >> 8) & 255;
1041 + memcpy(vmaw->header_blob_table + cpos + 2, data, len);
1042 + vmaw->header_blob_table_pos += len + 2;
1043 + return cpos;
1044 +}
1045 +
1046 +static uint32_t allocate_header_string(VmaWriter *vmaw, const char *str)
1047 +{
1048 + assert(vmaw);
1049 +
1050 + size_t len = strlen(str) + 1;
1051 +
1052 + return allocate_header_blob(vmaw, str, len);
1053 +}
1054 +
1055 +int vma_writer_add_config(VmaWriter *vmaw, const char *name, gpointer data,
1056 + gsize len)
1057 +{
1058 + assert(vmaw);
1059 + assert(!vmaw->header_written);
1060 + assert(vmaw->config_count < VMA_MAX_CONFIGS);
1061 + assert(name);
1062 + assert(data);
1063 +
1064 + gchar *basename = g_path_get_basename(name);
1065 + uint32_t name_ptr = allocate_header_string(vmaw, basename);
1066 + g_free(basename);
1067 +
1068 + if (!name_ptr) {
1069 + return -1;
1070 + }
1071 +
1072 + uint32_t data_ptr = allocate_header_blob(vmaw, data, len);
1073 + if (!data_ptr) {
1074 + return -1;
1075 + }
1076 +
1077 + vmaw->config_names[vmaw->config_count] = name_ptr;
1078 + vmaw->config_data[vmaw->config_count] = data_ptr;
1079 +
1080 + vmaw->config_count++;
1081 +
1082 + return 0;
1083 +}
1084 +
1085 +int vma_writer_register_stream(VmaWriter *vmaw, const char *devname,
1086 + size_t size)
1087 +{
1088 + assert(vmaw);
1089 + assert(devname);
1090 + assert(!vmaw->status);
1091 +
1092 + if (vmaw->header_written) {
1093 + vma_writer_set_error(vmaw, "vma_writer_register_stream: header "
1094 + "already written");
1095 + return -1;
1096 + }
1097 +
1098 + guint n = vmaw->stream_count + 1;
1099 +
1100 + /* we can have dev_ids form 1 to 255 (0 reserved)
1101 + * 255(-1) reseverd for safety
1102 + */
1103 + if (n > 254) {
1104 + vma_writer_set_error(vmaw, "vma_writer_register_stream: "
1105 + "too many drives");
1106 + return -1;
1107 + }
1108 +
1109 + if (size <= 0) {
1110 + vma_writer_set_error(vmaw, "vma_writer_register_stream: "
1111 + "got strange size %zd", size);
1112 + return -1;
1113 + }
1114 +
1115 + DPRINTF("vma_writer_register_stream %s %zu %d\n", devname, size, n);
1116 +
1117 + vmaw->stream_info[n].devname = g_strdup(devname);
1118 + vmaw->stream_info[n].size = size;
1119 +
1120 + vmaw->stream_info[n].cluster_count = (size + VMA_CLUSTER_SIZE - 1) /
1121 + VMA_CLUSTER_SIZE;
1122 +
1123 + vmaw->stream_count = n;
1124 +
1125 + if (strcmp(devname, "vmstate") == 0) {
1126 + vmaw->vmstate_stream = n;
1127 + }
1128 +
1129 + return n;
1130 +}
1131 +
1132 +static void coroutine_fn yield_until_fd_writable(int fd)
1133 +{
1134 + assert(qemu_in_coroutine());
1135 + AioContext *ctx = qemu_get_current_aio_context();
1136 + aio_set_fd_handler(ctx, fd, false, NULL, (IOHandler *)qemu_coroutine_enter,
1137 + NULL, NULL, qemu_coroutine_self());
1138 + qemu_coroutine_yield();
1139 + aio_set_fd_handler(ctx, fd, false, NULL, NULL, NULL, NULL, NULL);
1140 +}
1141 +
1142 +static ssize_t coroutine_fn
1143 +vma_queue_write(VmaWriter *vmaw, const void *buf, size_t bytes)
1144 +{
1145 + DPRINTF("vma_queue_write enter %zd\n", bytes);
1146 +
1147 + assert(vmaw);
1148 + assert(buf);
1149 + assert(bytes <= VMA_MAX_EXTENT_SIZE);
1150 +
1151 + size_t done = 0;
1152 + ssize_t ret;
1153 +
1154 + assert(vmaw->co_writer == NULL);
1155 +
1156 + vmaw->co_writer = qemu_coroutine_self();
1157 +
1158 + while (done < bytes) {
1159 + if (vmaw->status < 0) {
1160 + DPRINTF("vma_queue_write detected canceled backup\n");
1161 + done = -1;
1162 + break;
1163 + }
1164 + yield_until_fd_writable(vmaw->fd);
1165 + ret = write(vmaw->fd, buf + done, bytes - done);
1166 + if (ret > 0) {
1167 + done += ret;
1168 + DPRINTF("vma_queue_write written %zd %zd\n", done, ret);
1169 + } else if (ret < 0) {
1170 + if (errno == EAGAIN || errno == EWOULDBLOCK) {
1171 + /* try again */
1172 + } else {
1173 + vma_writer_set_error(vmaw, "vma_queue_write: write error - %s",
1174 + g_strerror(errno));
1175 + done = -1; /* always return failure for partial writes */
1176 + break;
1177 + }
1178 + } else if (ret == 0) {
1179 + /* should not happen - simply try again */
1180 + }
1181 + }
1182 +
1183 + vmaw->co_writer = NULL;
1184 +
1185 + return (done == bytes) ? bytes : -1;
1186 +}
1187 +
1188 +VmaWriter *vma_writer_create(const char *filename, uuid_t uuid, Error **errp)
1189 +{
1190 + const char *p;
1191 +
1192 + assert(sizeof(VmaHeader) == (4096 + 8192));
1193 + assert(G_STRUCT_OFFSET(VmaHeader, config_names) == 2044);
1194 + assert(G_STRUCT_OFFSET(VmaHeader, config_data) == 3068);
1195 + assert(G_STRUCT_OFFSET(VmaHeader, dev_info) == 4096);
1196 + assert(sizeof(VmaExtentHeader) == 512);
1197 +
1198 + VmaWriter *vmaw = g_new0(VmaWriter, 1);
1199 + vmaw->fd = -1;
1200 +
1201 + vmaw->md5csum = g_checksum_new(G_CHECKSUM_MD5);
1202 + if (!vmaw->md5csum) {
1203 + error_setg(errp, "can't allocate cmsum\n");
1204 + goto err;
1205 + }
1206 +
1207 + if (strstart(filename, "exec:", &p)) {
1208 + vmaw->cmd = popen(p, "w");
1209 + if (vmaw->cmd == NULL) {
1210 + error_setg(errp, "can't popen command '%s' - %s\n", p,
1211 + g_strerror(errno));
1212 + goto err;
1213 + }
1214 + vmaw->fd = fileno(vmaw->cmd);
1215 +
1216 + /* try to use O_NONBLOCK */
1217 + fcntl(vmaw->fd, F_SETFL, fcntl(vmaw->fd, F_GETFL)|O_NONBLOCK);
1218 +
1219 + } else {
1220 + struct stat st;
1221 + int oflags;
1222 + const char *tmp_id_str;
1223 +
1224 + if ((stat(filename, &st) == 0) && S_ISFIFO(st.st_mode)) {
1225 + oflags = O_NONBLOCK|O_WRONLY;
1226 + vmaw->fd = qemu_open(filename, oflags, errp);
1227 + } else if (strstart(filename, "/dev/fdset/", &tmp_id_str)) {
1228 + oflags = O_NONBLOCK|O_WRONLY;
1229 + vmaw->fd = qemu_open(filename, oflags, errp);
1230 + } else if (strstart(filename, "/dev/fdname/", &tmp_id_str)) {
1231 + vmaw->fd = monitor_get_fd(monitor_cur(), tmp_id_str, errp);
1232 + if (vmaw->fd < 0) {
1233 + goto err;
1234 + }
1235 + /* try to use O_NONBLOCK */
1236 + fcntl(vmaw->fd, F_SETFL, fcntl(vmaw->fd, F_GETFL)|O_NONBLOCK);
1237 + } else {
1238 + oflags = O_NONBLOCK|O_DIRECT|O_WRONLY|O_EXCL;
1239 + vmaw->fd = qemu_create(filename, oflags, 0644, errp);
1240 + }
1241 +
1242 + if (vmaw->fd < 0) {
1243 + error_setg(errp, "can't open file %s - %s\n", filename,
1244 + g_strerror(errno));
1245 + goto err;
1246 + }
1247 + }
1248 +
1249 + /* we use O_DIRECT, so we need to align IO buffers */
1250 +
1251 + vmaw->outbuf = qemu_memalign(512, VMA_MAX_EXTENT_SIZE);
1252 + vmaw->headerbuf = qemu_memalign(512, HEADERBUF_SIZE);
1253 +
1254 + vmaw->outbuf_count = 0;
1255 + vmaw->outbuf_pos = VMA_EXTENT_HEADER_SIZE;
1256 +
1257 + vmaw->header_blob_table_pos = 1; /* start at pos 1 */
1258 +
1259 + qemu_co_mutex_init(&vmaw->flush_lock);
1260 +
1261 + uuid_copy(vmaw->uuid, uuid);
1262 +
1263 + return vmaw;
1264 +
1265 +err:
1266 + if (vmaw) {
1267 + if (vmaw->cmd) {
1268 + pclose(vmaw->cmd);
1269 + } else if (vmaw->fd >= 0) {
1270 + close(vmaw->fd);
1271 + }
1272 +
1273 + if (vmaw->md5csum) {
1274 + g_checksum_free(vmaw->md5csum);
1275 + }
1276 +
1277 + g_free(vmaw);
1278 + }
1279 +
1280 + return NULL;
1281 +}
1282 +
1283 +static int coroutine_fn vma_write_header(VmaWriter *vmaw)
1284 +{
1285 + assert(vmaw);
1286 + unsigned char *buf = vmaw->headerbuf;
1287 + VmaHeader *head = (VmaHeader *)buf;
1288 +
1289 + int i;
1290 +
1291 + DPRINTF("VMA WRITE HEADER\n");
1292 +
1293 + if (vmaw->status < 0) {
1294 + return vmaw->status;
1295 + }
1296 +
1297 + memset(buf, 0, HEADERBUF_SIZE);
1298 +
1299 + head->magic = VMA_MAGIC;
1300 + head->version = GUINT32_TO_BE(1); /* v1 */
1301 + memcpy(head->uuid, vmaw->uuid, 16);
1302 +
1303 + time_t ctime = time(NULL);
1304 + head->ctime = GUINT64_TO_BE(ctime);
1305 +
1306 + for (i = 0; i < VMA_MAX_CONFIGS; i++) {
1307 + head->config_names[i] = GUINT32_TO_BE(vmaw->config_names[i]);
1308 + head->config_data[i] = GUINT32_TO_BE(vmaw->config_data[i]);
1309 + }
1310 +
1311 + /* 32 bytes per device (12 used currently) = 8192 bytes max */
1312 + for (i = 1; i <= 254; i++) {
1313 + VmaStreamInfo *si = &vmaw->stream_info[i];
1314 + if (si->size) {
1315 + assert(si->devname);
1316 + uint32_t devname_ptr = allocate_header_string(vmaw, si->devname);
1317 + if (!devname_ptr) {
1318 + return -1;
1319 + }
1320 + head->dev_info[i].devname_ptr = GUINT32_TO_BE(devname_ptr);
1321 + head->dev_info[i].size = GUINT64_TO_BE(si->size);
1322 + }
1323 + }
1324 +
1325 + uint32_t header_size = sizeof(VmaHeader) + vmaw->header_blob_table_size;
1326 + head->header_size = GUINT32_TO_BE(header_size);
1327 +
1328 + if (header_size > HEADERBUF_SIZE) {
1329 + return -1; /* just to be sure */
1330 + }
1331 +
1332 + uint32_t blob_buffer_offset = sizeof(VmaHeader);
1333 + memcpy(buf + blob_buffer_offset, vmaw->header_blob_table,
1334 + vmaw->header_blob_table_size);
1335 + head->blob_buffer_offset = GUINT32_TO_BE(blob_buffer_offset);
1336 + head->blob_buffer_size = GUINT32_TO_BE(vmaw->header_blob_table_pos);
1337 +
1338 + g_checksum_reset(vmaw->md5csum);
1339 + g_checksum_update(vmaw->md5csum, (const guchar *)buf, header_size);
1340 + gsize csize = 16;
1341 + g_checksum_get_digest(vmaw->md5csum, (guint8 *)(head->md5sum), &csize);
1342 +
1343 + return vma_queue_write(vmaw, buf, header_size);
1344 +}
1345 +
1346 +static int coroutine_fn vma_writer_flush(VmaWriter *vmaw)
1347 +{
1348 + assert(vmaw);
1349 +
1350 + int ret;
1351 + int i;
1352 +
1353 + if (vmaw->status < 0) {
1354 + return vmaw->status;
1355 + }
1356 +
1357 + if (!vmaw->header_written) {
1358 + vmaw->header_written = true;
1359 + ret = vma_write_header(vmaw);
1360 + if (ret < 0) {
1361 + vma_writer_set_error(vmaw, "vma_writer_flush: write header failed");
1362 + return ret;
1363 + }
1364 + }
1365 +
1366 + DPRINTF("VMA WRITE FLUSH %d %d\n", vmaw->outbuf_count, vmaw->outbuf_pos);
1367 +
1368 +
1369 + VmaExtentHeader *ehead = (VmaExtentHeader *)vmaw->outbuf;
1370 +
1371 + ehead->magic = VMA_EXTENT_MAGIC;
1372 + ehead->reserved1 = 0;
1373 +
1374 + for (i = 0; i < VMA_BLOCKS_PER_EXTENT; i++) {
1375 + ehead->blockinfo[i] = GUINT64_TO_BE(vmaw->outbuf_block_info[i]);
1376 + }
1377 +
1378 + guint16 block_count = (vmaw->outbuf_pos - VMA_EXTENT_HEADER_SIZE) /
1379 + VMA_BLOCK_SIZE;
1380 +
1381 + ehead->block_count = GUINT16_TO_BE(block_count);
1382 +
1383 + memcpy(ehead->uuid, vmaw->uuid, sizeof(ehead->uuid));
1384 + memset(ehead->md5sum, 0, sizeof(ehead->md5sum));
1385 +
1386 + g_checksum_reset(vmaw->md5csum);
1387 + g_checksum_update(vmaw->md5csum, vmaw->outbuf, VMA_EXTENT_HEADER_SIZE);
1388 + gsize csize = 16;
1389 + g_checksum_get_digest(vmaw->md5csum, ehead->md5sum, &csize);
1390 +
1391 + int bytes = vmaw->outbuf_pos;
1392 + ret = vma_queue_write(vmaw, vmaw->outbuf, bytes);
1393 + if (ret != bytes) {
1394 + vma_writer_set_error(vmaw, "vma_writer_flush: failed write");
1395 + }
1396 +
1397 + vmaw->outbuf_count = 0;
1398 + vmaw->outbuf_pos = VMA_EXTENT_HEADER_SIZE;
1399 +
1400 + for (i = 0; i < VMA_BLOCKS_PER_EXTENT; i++) {
1401 + vmaw->outbuf_block_info[i] = 0;
1402 + }
1403 +
1404 + return vmaw->status;
1405 +}
1406 +
1407 +static int vma_count_open_streams(VmaWriter *vmaw)
1408 +{
1409 + g_assert(vmaw != NULL);
1410 +
1411 + int i;
1412 + int open_drives = 0;
1413 + for (i = 0; i <= 255; i++) {
1414 + if (vmaw->stream_info[i].size && !vmaw->stream_info[i].finished) {
1415 + open_drives++;
1416 + }
1417 + }
1418 +
1419 + return open_drives;
1420 +}
1421 +
1422 +
1423 +/**
1424 + * You need to call this if the vma archive does not contain
1425 + * any data stream.
1426 + */
1427 +int coroutine_fn
1428 +vma_writer_flush_output(VmaWriter *vmaw)
1429 +{
1430 + qemu_co_mutex_lock(&vmaw->flush_lock);
1431 + int ret = vma_writer_flush(vmaw);
1432 + qemu_co_mutex_unlock(&vmaw->flush_lock);
1433 + if (ret < 0) {
1434 + vma_writer_set_error(vmaw, "vma_writer_flush_header failed");
1435 + }
1436 + return ret;
1437 +}
1438 +
1439 +/**
1440 + * all jobs should call this when there is no more data
1441 + * Returns: number of remaining stream (0 ==> finished)
1442 + */
1443 +int coroutine_fn
1444 +vma_writer_close_stream(VmaWriter *vmaw, uint8_t dev_id)
1445 +{
1446 + g_assert(vmaw != NULL);
1447 +
1448 + DPRINTF("vma_writer_set_status %d\n", dev_id);
1449 + if (!vmaw->stream_info[dev_id].size) {
1450 + vma_writer_set_error(vmaw, "vma_writer_close_stream: "
1451 + "no such stream %d", dev_id);
1452 + return -1;
1453 + }
1454 + if (vmaw->stream_info[dev_id].finished) {
1455 + vma_writer_set_error(vmaw, "vma_writer_close_stream: "
1456 + "stream already closed %d", dev_id);
1457 + return -1;
1458 + }
1459 +
1460 + vmaw->stream_info[dev_id].finished = true;
1461 +
1462 + int open_drives = vma_count_open_streams(vmaw);
1463 +
1464 + if (open_drives <= 0) {
1465 + DPRINTF("vma_writer_set_status all drives completed\n");
1466 + vma_writer_flush_output(vmaw);
1467 + }
1468 +
1469 + return open_drives;
1470 +}
1471 +
1472 +int vma_writer_get_status(VmaWriter *vmaw, VmaStatus *status)
1473 +{
1474 + int i;
1475 +
1476 + g_assert(vmaw != NULL);
1477 +
1478 + if (status) {
1479 + status->status = vmaw->status;
1480 + g_strlcpy(status->errmsg, vmaw->errmsg, sizeof(status->errmsg));
1481 + for (i = 0; i <= 255; i++) {
1482 + status->stream_info[i] = vmaw->stream_info[i];
1483 + }
1484 +
1485 + uuid_unparse_lower(vmaw->uuid, status->uuid_str);
1486 + }
1487 +
1488 + status->closed = vmaw->closed;
1489 +
1490 + return vmaw->status;
1491 +}
1492 +
1493 +static int vma_writer_get_buffer(VmaWriter *vmaw)
1494 +{
1495 + int ret = 0;
1496 +
1497 + qemu_co_mutex_lock(&vmaw->flush_lock);
1498 +
1499 + /* wait until buffer is available */
1500 + while (vmaw->outbuf_count >= (VMA_BLOCKS_PER_EXTENT - 1)) {
1501 + ret = vma_writer_flush(vmaw);
1502 + if (ret < 0) {
1503 + vma_writer_set_error(vmaw, "vma_writer_get_buffer: flush failed");
1504 + break;
1505 + }
1506 + }
1507 +
1508 + qemu_co_mutex_unlock(&vmaw->flush_lock);
1509 +
1510 + return ret;
1511 +}
1512 +
1513 +
1514 +int64_t coroutine_fn
1515 +vma_writer_write(VmaWriter *vmaw, uint8_t dev_id, int64_t cluster_num,
1516 + const unsigned char *buf, size_t *zero_bytes)
1517 +{
1518 + g_assert(vmaw != NULL);
1519 + g_assert(zero_bytes != NULL);
1520 +
1521 + *zero_bytes = 0;
1522 +
1523 + if (vmaw->status < 0) {
1524 + return vmaw->status;
1525 + }
1526 +
1527 + if (!dev_id || !vmaw->stream_info[dev_id].size) {
1528 + vma_writer_set_error(vmaw, "vma_writer_write: "
1529 + "no such stream %d", dev_id);
1530 + return -1;
1531 + }
1532 +
1533 + if (vmaw->stream_info[dev_id].finished) {
1534 + vma_writer_set_error(vmaw, "vma_writer_write: "
1535 + "stream already closed %d", dev_id);
1536 + return -1;
1537 + }
1538 +
1539 +
1540 + if (cluster_num >= (((uint64_t)1)<<32)) {
1541 + vma_writer_set_error(vmaw, "vma_writer_write: "
1542 + "cluster number out of range");
1543 + return -1;
1544 + }
1545 +
1546 + if (dev_id == vmaw->vmstate_stream) {
1547 + if (cluster_num != vmaw->vmstate_clusters) {
1548 + vma_writer_set_error(vmaw, "vma_writer_write: "
1549 + "non sequential vmstate write");
1550 + }
1551 + vmaw->vmstate_clusters++;
1552 + } else if (cluster_num >= vmaw->stream_info[dev_id].cluster_count) {
1553 + vma_writer_set_error(vmaw, "vma_writer_write: cluster number too big");
1554 + return -1;
1555 + }
1556 +
1557 + /* wait until buffer is available */
1558 + if (vma_writer_get_buffer(vmaw) < 0) {
1559 + vma_writer_set_error(vmaw, "vma_writer_write: "
1560 + "vma_writer_get_buffer failed");
1561 + return -1;
1562 + }
1563 +
1564 + DPRINTF("VMA WRITE %d %zd\n", dev_id, cluster_num);
1565 +
1566 + uint64_t dev_size = vmaw->stream_info[dev_id].size;
1567 + uint16_t mask = 0;
1568 +
1569 + if (buf) {
1570 + int i;
1571 + int bit = 1;
1572 + uint64_t byte_offset = cluster_num * VMA_CLUSTER_SIZE;
1573 + for (i = 0; i < 16; i++) {
1574 + const unsigned char *vmablock = buf + (i*VMA_BLOCK_SIZE);
1575 +
1576 + // Note: If the source is not 64k-aligned, we might reach 4k blocks
1577 + // after the end of the device. Always mark these as zero in the
1578 + // mask, so the restore handles them correctly.
1579 + if (byte_offset < dev_size &&
1580 + !buffer_is_zero(vmablock, VMA_BLOCK_SIZE))
1581 + {
1582 + mask |= bit;
1583 + memcpy(vmaw->outbuf + vmaw->outbuf_pos, vmablock,
1584 + VMA_BLOCK_SIZE);
1585 +
1586 + // prevent memory leakage on unaligned last block
1587 + if (byte_offset + VMA_BLOCK_SIZE > dev_size) {
1588 + uint64_t real_data_in_block = dev_size - byte_offset;
1589 + memset(vmaw->outbuf + vmaw->outbuf_pos + real_data_in_block,
1590 + 0, VMA_BLOCK_SIZE - real_data_in_block);
1591 + }
1592 +
1593 + vmaw->outbuf_pos += VMA_BLOCK_SIZE;
1594 + } else {
1595 + DPRINTF("VMA WRITE %zd ZERO BLOCK %d\n", cluster_num, i);
1596 + vmaw->stream_info[dev_id].zero_bytes += VMA_BLOCK_SIZE;
1597 + *zero_bytes += VMA_BLOCK_SIZE;
1598 + }
1599 +
1600 + byte_offset += VMA_BLOCK_SIZE;
1601 + bit = bit << 1;
1602 + }
1603 + } else {
1604 + DPRINTF("VMA WRITE %zd ZERO CLUSTER\n", cluster_num);
1605 + vmaw->stream_info[dev_id].zero_bytes += VMA_CLUSTER_SIZE;
1606 + *zero_bytes += VMA_CLUSTER_SIZE;
1607 + }
1608 +
1609 + uint64_t block_info = ((uint64_t)mask) << (32+16);
1610 + block_info |= ((uint64_t)dev_id) << 32;
1611 + block_info |= (cluster_num & 0xffffffff);
1612 + vmaw->outbuf_block_info[vmaw->outbuf_count] = block_info;
1613 +
1614 + DPRINTF("VMA WRITE MASK %zd %zx\n", cluster_num, block_info);
1615 +
1616 + vmaw->outbuf_count++;
1617 +
1618 + /** NOTE: We allways write whole clusters, but we correctly set
1619 + * transferred bytes. So transferred == size when when everything
1620 + * went OK.
1621 + */
1622 + size_t transferred = VMA_CLUSTER_SIZE;
1623 +
1624 + if (dev_id != vmaw->vmstate_stream) {
1625 + uint64_t last = (cluster_num + 1) * VMA_CLUSTER_SIZE;
1626 + if (last > dev_size) {
1627 + uint64_t diff = last - dev_size;
1628 + if (diff >= VMA_CLUSTER_SIZE) {
1629 + vma_writer_set_error(vmaw, "vma_writer_write: "
1630 + "read after last cluster");
1631 + return -1;
1632 + }
1633 + transferred -= diff;
1634 + }
1635 + }
1636 +
1637 + vmaw->stream_info[dev_id].transferred += transferred;
1638 +
1639 + return transferred;
1640 +}
1641 +
1642 +void vma_writer_error_propagate(VmaWriter *vmaw, Error **errp)
1643 +{
1644 + if (vmaw->status < 0 && *errp == NULL) {
1645 + error_setg(errp, "%s", vmaw->errmsg);
1646 + }
1647 +}
1648 +
1649 +int vma_writer_close(VmaWriter *vmaw, Error **errp)
1650 +{
1651 + g_assert(vmaw != NULL);
1652 +
1653 + int i;
1654 +
1655 + qemu_co_mutex_lock(&vmaw->flush_lock); // wait for pending writes
1656 +
1657 + assert(vmaw->co_writer == NULL);
1658 +
1659 + if (vmaw->cmd) {
1660 + if (pclose(vmaw->cmd) < 0) {
1661 + vma_writer_set_error(vmaw, "vma_writer_close: "
1662 + "pclose failed - %s", g_strerror(errno));
1663 + }
1664 + } else {
1665 + if (close(vmaw->fd) < 0) {
1666 + vma_writer_set_error(vmaw, "vma_writer_close: "
1667 + "close failed - %s", g_strerror(errno));
1668 + }
1669 + }
1670 +
1671 + for (i = 0; i <= 255; i++) {
1672 + VmaStreamInfo *si = &vmaw->stream_info[i];
1673 + if (si->size) {
1674 + if (!si->finished) {
1675 + vma_writer_set_error(vmaw, "vma_writer_close: "
1676 + "detected open stream '%s'", si->devname);
1677 + } else if ((si->transferred != si->size) &&
1678 + (i != vmaw->vmstate_stream)) {
1679 + vma_writer_set_error(vmaw, "vma_writer_close: "
1680 + "incomplete stream '%s' (%zd != %zd)",
1681 + si->devname, si->transferred, si->size);
1682 + }
1683 + }
1684 + }
1685 +
1686 + for (i = 0; i <= 255; i++) {
1687 + vmaw->stream_info[i].finished = 1; /* mark as closed */
1688 + }
1689 +
1690 + vmaw->closed = 1;
1691 +
1692 + if (vmaw->status < 0 && *errp == NULL) {
1693 + error_setg(errp, "%s", vmaw->errmsg);
1694 + }
1695 +
1696 + qemu_co_mutex_unlock(&vmaw->flush_lock);
1697 +
1698 + return vmaw->status;
1699 +}
1700 +
1701 +void vma_writer_destroy(VmaWriter *vmaw)
1702 +{
1703 + assert(vmaw);
1704 +
1705 + int i;
1706 +
1707 + for (i = 0; i <= 255; i++) {
1708 + if (vmaw->stream_info[i].devname) {
1709 + g_free(vmaw->stream_info[i].devname);
1710 + }
1711 + }
1712 +
1713 + if (vmaw->md5csum) {
1714 + g_checksum_free(vmaw->md5csum);
1715 + }
1716 +
1717 + qemu_vfree(vmaw->headerbuf);
1718 + qemu_vfree(vmaw->outbuf);
1719 + g_free(vmaw);
1720 +}
1721 diff --git a/vma.c b/vma.c
1722 new file mode 100644
1723 index 0000000000..91612d50a2
1724 --- /dev/null
1725 +++ b/vma.c
1726 @@ -0,0 +1,850 @@
1727 +/*
1728 + * VMA: Virtual Machine Archive
1729 + *
1730 + * Copyright (C) 2012-2013 Proxmox Server Solutions
1731 + *
1732 + * Authors:
1733 + * Dietmar Maurer (dietmar@proxmox.com)
1734 + *
1735 + * This work is licensed under the terms of the GNU GPL, version 2 or later.
1736 + * See the COPYING file in the top-level directory.
1737 + *
1738 + */
1739 +
1740 +#include "qemu/osdep.h"
1741 +#include <glib.h>
1742 +
1743 +#include "vma.h"
1744 +#include "qemu-common.h"
1745 +#include "qemu/module.h"
1746 +#include "qemu/error-report.h"
1747 +#include "qemu/main-loop.h"
1748 +#include "qemu/cutils.h"
1749 +#include "qemu/memalign.h"
1750 +#include "qapi/qmp/qdict.h"
1751 +#include "sysemu/block-backend.h"
1752 +
1753 +static void help(void)
1754 +{
1755 + const char *help_msg =
1756 + "usage: vma command [command options]\n"
1757 + "\n"
1758 + "vma list <filename>\n"
1759 + "vma config <filename> [-c config]\n"
1760 + "vma create <filename> [-c config] pathname ...\n"
1761 + "vma extract <filename> [-r <fifo>] <targetdir>\n"
1762 + "vma verify <filename> [-v]\n"
1763 + ;
1764 +
1765 + printf("%s", help_msg);
1766 + exit(1);
1767 +}
1768 +
1769 +static const char *extract_devname(const char *path, char **devname, int index)
1770 +{
1771 + assert(path);
1772 +
1773 + const char *sep = strchr(path, '=');
1774 +
1775 + if (sep) {
1776 + *devname = g_strndup(path, sep - path);
1777 + path = sep + 1;
1778 + } else {
1779 + if (index >= 0) {
1780 + *devname = g_strdup_printf("disk%d", index);
1781 + } else {
1782 + *devname = NULL;
1783 + }
1784 + }
1785 +
1786 + return path;
1787 +}
1788 +
1789 +static void print_content(VmaReader *vmar)
1790 +{
1791 + assert(vmar);
1792 +
1793 + VmaHeader *head = vma_reader_get_header(vmar);
1794 +
1795 + GList *l = vma_reader_get_config_data(vmar);
1796 + while (l && l->data) {
1797 + VmaConfigData *cdata = (VmaConfigData *)l->data;
1798 + l = g_list_next(l);
1799 + printf("CFG: size: %d name: %s\n", cdata->len, cdata->name);
1800 + }
1801 +
1802 + int i;
1803 + VmaDeviceInfo *di;
1804 + for (i = 1; i < 255; i++) {
1805 + di = vma_reader_get_device_info(vmar, i);
1806 + if (di) {
1807 + if (strcmp(di->devname, "vmstate") == 0) {
1808 + printf("VMSTATE: dev_id=%d memory: %zd\n", i, di->size);
1809 + } else {
1810 + printf("DEV: dev_id=%d size: %zd devname: %s\n",
1811 + i, di->size, di->devname);
1812 + }
1813 + }
1814 + }
1815 + /* ctime is the last entry we print */
1816 + printf("CTIME: %s", ctime(&head->ctime));
1817 + fflush(stdout);
1818 +}
1819 +
1820 +static int list_content(int argc, char **argv)
1821 +{
1822 + int c, ret = 0;
1823 + const char *filename;
1824 +
1825 + for (;;) {
1826 + c = getopt(argc, argv, "h");
1827 + if (c == -1) {
1828 + break;
1829 + }
1830 + switch (c) {
1831 + case '?':
1832 + case 'h':
1833 + help();
1834 + break;
1835 + default:
1836 + g_assert_not_reached();
1837 + }
1838 + }
1839 +
1840 + /* Get the filename */
1841 + if ((optind + 1) != argc) {
1842 + help();
1843 + }
1844 + filename = argv[optind++];
1845 +
1846 + Error *errp = NULL;
1847 + VmaReader *vmar = vma_reader_create(filename, &errp);
1848 +
1849 + if (!vmar) {
1850 + g_error("%s", error_get_pretty(errp));
1851 + }
1852 +
1853 + print_content(vmar);
1854 +
1855 + vma_reader_destroy(vmar);
1856 +
1857 + return ret;
1858 +}
1859 +
1860 +typedef struct RestoreMap {
1861 + char *devname;
1862 + char *path;
1863 + char *format;
1864 + uint64_t throttling_bps;
1865 + char *throttling_group;
1866 + char *cache;
1867 + bool write_zero;
1868 +} RestoreMap;
1869 +
1870 +static bool try_parse_option(char **line, const char *optname, char **out, const char *inbuf) {
1871 + size_t optlen = strlen(optname);
1872 + if (strncmp(*line, optname, optlen) != 0 || (*line)[optlen] != '=') {
1873 + return false;
1874 + }
1875 + if (*out) {
1876 + g_error("read map failed - duplicate value for option '%s'", optname);
1877 + }
1878 + char *value = (*line) + optlen + 1; /* including a '=' */
1879 + char *colon = strchr(value, ':');
1880 + if (!colon) {
1881 + g_error("read map failed - option '%s' not terminated ('%s')",
1882 + optname, inbuf);
1883 + }
1884 + *line = colon+1;
1885 + *out = g_strndup(value, colon - value);
1886 + return true;
1887 +}
1888 +
1889 +static uint64_t verify_u64(const char *text) {
1890 + uint64_t value;
1891 + const char *endptr = NULL;
1892 + if (qemu_strtou64(text, &endptr, 0, &value) != 0 || !endptr || *endptr) {
1893 + g_error("read map failed - not a number: %s", text);
1894 + }
1895 + return value;
1896 +}
1897 +
1898 +static int extract_content(int argc, char **argv)
1899 +{
1900 + int c, ret = 0;
1901 + int verbose = 0;
1902 + const char *filename;
1903 + const char *dirname;
1904 + const char *readmap = NULL;
1905 +
1906 + for (;;) {
1907 + c = getopt(argc, argv, "hvr:");
1908 + if (c == -1) {
1909 + break;
1910 + }
1911 + switch (c) {
1912 + case '?':
1913 + case 'h':
1914 + help();
1915 + break;
1916 + case 'r':
1917 + readmap = optarg;
1918 + break;
1919 + case 'v':
1920 + verbose = 1;
1921 + break;
1922 + default:
1923 + help();
1924 + }
1925 + }
1926 +
1927 + /* Get the filename */
1928 + if ((optind + 2) != argc) {
1929 + help();
1930 + }
1931 + filename = argv[optind++];
1932 + dirname = argv[optind++];
1933 +
1934 + Error *errp = NULL;
1935 + VmaReader *vmar = vma_reader_create(filename, &errp);
1936 +
1937 + if (!vmar) {
1938 + g_error("%s", error_get_pretty(errp));
1939 + }
1940 +
1941 + if (mkdir(dirname, 0777) < 0) {
1942 + g_error("unable to create target directory %s - %s",
1943 + dirname, g_strerror(errno));
1944 + }
1945 +
1946 + GList *l = vma_reader_get_config_data(vmar);
1947 + while (l && l->data) {
1948 + VmaConfigData *cdata = (VmaConfigData *)l->data;
1949 + l = g_list_next(l);
1950 + char *cfgfn = g_strdup_printf("%s/%s", dirname, cdata->name);
1951 + GError *err = NULL;
1952 + if (!g_file_set_contents(cfgfn, (gchar *)cdata->data, cdata->len,
1953 + &err)) {
1954 + g_error("unable to write file: %s", err->message);
1955 + }
1956 + }
1957 +
1958 + GHashTable *devmap = g_hash_table_new(g_str_hash, g_str_equal);
1959 +
1960 + if (readmap) {
1961 + print_content(vmar);
1962 +
1963 + FILE *map = fopen(readmap, "r");
1964 + if (!map) {
1965 + g_error("unable to open fifo %s - %s", readmap, g_strerror(errno));
1966 + }
1967 +
1968 + while (1) {
1969 + char inbuf[8192];
1970 + char *line = fgets(inbuf, sizeof(inbuf), map);
1971 + char *format = NULL;
1972 + char *bps = NULL;
1973 + char *group = NULL;
1974 + char *cache = NULL;
1975 + if (!line || line[0] == '\0' || !strcmp(line, "done\n")) {
1976 + break;
1977 + }
1978 + int len = strlen(line);
1979 + if (line[len - 1] == '\n') {
1980 + line[len - 1] = '\0';
1981 + if (len == 1) {
1982 + break;
1983 + }
1984 + }
1985 +
1986 + while (1) {
1987 + if (!try_parse_option(&line, "format", &format, inbuf) &&
1988 + !try_parse_option(&line, "throttling.bps", &bps, inbuf) &&
1989 + !try_parse_option(&line, "throttling.group", &group, inbuf) &&
1990 + !try_parse_option(&line, "cache", &cache, inbuf))
1991 + {
1992 + break;
1993 + }
1994 + }
1995 +
1996 + uint64_t bps_value = 0;
1997 + if (bps) {
1998 + bps_value = verify_u64(bps);
1999 + g_free(bps);
2000 + }
2001 +
2002 + const char *path;
2003 + bool write_zero;
2004 + if (line[0] == '0' && line[1] == ':') {
2005 + path = line + 2;
2006 + write_zero = false;
2007 + } else if (line[0] == '1' && line[1] == ':') {
2008 + path = line + 2;
2009 + write_zero = true;
2010 + } else {
2011 + g_error("read map failed - parse error ('%s')", inbuf);
2012 + }
2013 +
2014 + char *devname = NULL;
2015 + path = extract_devname(path, &devname, -1);
2016 + if (!devname) {
2017 + g_error("read map failed - no dev name specified ('%s')",
2018 + inbuf);
2019 + }
2020 +
2021 + RestoreMap *map = g_new0(RestoreMap, 1);
2022 + map->devname = g_strdup(devname);
2023 + map->path = g_strdup(path);
2024 + map->format = format;
2025 + map->throttling_bps = bps_value;
2026 + map->throttling_group = group;
2027 + map->cache = cache;
2028 + map->write_zero = write_zero;
2029 +
2030 + g_hash_table_insert(devmap, map->devname, map);
2031 +
2032 + };
2033 + }
2034 +
2035 + int i;
2036 + int vmstate_fd = -1;
2037 + guint8 vmstate_stream = 0;
2038 +
2039 + for (i = 1; i < 255; i++) {
2040 + VmaDeviceInfo *di = vma_reader_get_device_info(vmar, i);
2041 + if (di && (strcmp(di->devname, "vmstate") == 0)) {
2042 + vmstate_stream = i;
2043 + char *statefn = g_strdup_printf("%s/vmstate.bin", dirname);
2044 + vmstate_fd = open(statefn, O_WRONLY|O_CREAT|O_EXCL, 0644);
2045 + if (vmstate_fd < 0) {
2046 + g_error("create vmstate file '%s' failed - %s", statefn,
2047 + g_strerror(errno));
2048 + }
2049 + g_free(statefn);
2050 + } else if (di) {
2051 + char *devfn = NULL;
2052 + const char *format = NULL;
2053 + uint64_t throttling_bps = 0;
2054 + const char *throttling_group = NULL;
2055 + const char *cache = NULL;
2056 + int flags = BDRV_O_RDWR;
2057 + bool write_zero = true;
2058 +
2059 + BlockBackend *blk = NULL;
2060 +
2061 + if (readmap) {
2062 + RestoreMap *map;
2063 + map = (RestoreMap *)g_hash_table_lookup(devmap, di->devname);
2064 + if (map == NULL) {
2065 + g_error("no device name mapping for %s", di->devname);
2066 + }
2067 + devfn = map->path;
2068 + format = map->format;
2069 + throttling_bps = map->throttling_bps;
2070 + throttling_group = map->throttling_group;
2071 + cache = map->cache;
2072 + write_zero = map->write_zero;
2073 + } else {
2074 + devfn = g_strdup_printf("%s/tmp-disk-%s.raw",
2075 + dirname, di->devname);
2076 + printf("DEVINFO %s %zd\n", devfn, di->size);
2077 +
2078 + bdrv_img_create(devfn, "raw", NULL, NULL, NULL, di->size,
2079 + flags, true, &errp);
2080 + if (errp) {
2081 + g_error("can't create file %s: %s", devfn,
2082 + error_get_pretty(errp));
2083 + }
2084 +
2085 + /* Note: we created an empty file above, so there is no
2086 + * need to write zeroes (so we generate a sparse file)
2087 + */
2088 + write_zero = false;
2089 + }
2090 +
2091 + size_t devlen = strlen(devfn);
2092 + QDict *options = NULL;
2093 + bool writethrough;
2094 + if (format) {
2095 + /* explicit format from commandline */
2096 + options = qdict_new();
2097 + qdict_put_str(options, "driver", format);
2098 + } else if ((devlen > 4 && strcmp(devfn+devlen-4, ".raw") == 0) ||
2099 + strncmp(devfn, "/dev/", 5) == 0)
2100 + {
2101 + /* This part is now deprecated for PVE as well (just as qemu
2102 + * deprecated not specifying an explicit raw format, too.
2103 + */
2104 + /* explicit raw format */
2105 + options = qdict_new();
2106 + qdict_put_str(options, "driver", "raw");
2107 + }
2108 + if (cache && bdrv_parse_cache_mode(cache, &flags, &writethrough)) {
2109 + g_error("invalid cache option: %s\n", cache);
2110 + }
2111 +
2112 + if (errp || !(blk = blk_new_open(devfn, NULL, options, flags, &errp))) {
2113 + g_error("can't open file %s - %s", devfn,
2114 + error_get_pretty(errp));
2115 + }
2116 +
2117 + if (cache) {
2118 + blk_set_enable_write_cache(blk, !writethrough);
2119 + }
2120 +
2121 + if (throttling_group) {
2122 + blk_io_limits_enable(blk, throttling_group);
2123 + }
2124 +
2125 + if (throttling_bps) {
2126 + if (!throttling_group) {
2127 + blk_io_limits_enable(blk, devfn);
2128 + }
2129 +
2130 + ThrottleConfig cfg;
2131 + throttle_config_init(&cfg);
2132 + cfg.buckets[THROTTLE_BPS_WRITE].avg = throttling_bps;
2133 + Error *err = NULL;
2134 + if (!throttle_is_valid(&cfg, &err)) {
2135 + error_report_err(err);
2136 + g_error("failed to apply throttling");
2137 + }
2138 + blk_set_io_limits(blk, &cfg);
2139 + }
2140 +
2141 + if (vma_reader_register_bs(vmar, i, blk, write_zero, &errp) < 0) {
2142 + g_error("%s", error_get_pretty(errp));
2143 + }
2144 +
2145 + if (!readmap) {
2146 + g_free(devfn);
2147 + }
2148 + }
2149 + }
2150 +
2151 + if (vma_reader_restore(vmar, vmstate_fd, verbose, &errp) < 0) {
2152 + g_error("restore failed - %s", error_get_pretty(errp));
2153 + }
2154 +
2155 + if (!readmap) {
2156 + for (i = 1; i < 255; i++) {
2157 + VmaDeviceInfo *di = vma_reader_get_device_info(vmar, i);
2158 + if (di && (i != vmstate_stream)) {
2159 + char *tmpfn = g_strdup_printf("%s/tmp-disk-%s.raw",
2160 + dirname, di->devname);
2161 + char *fn = g_strdup_printf("%s/disk-%s.raw",
2162 + dirname, di->devname);
2163 + if (rename(tmpfn, fn) != 0) {
2164 + g_error("rename %s to %s failed - %s",
2165 + tmpfn, fn, g_strerror(errno));
2166 + }
2167 + }
2168 + }
2169 + }
2170 +
2171 + vma_reader_destroy(vmar);
2172 +
2173 + bdrv_close_all();
2174 +
2175 + return ret;
2176 +}
2177 +
2178 +static int verify_content(int argc, char **argv)
2179 +{
2180 + int c, ret = 0;
2181 + int verbose = 0;
2182 + const char *filename;
2183 +
2184 + for (;;) {
2185 + c = getopt(argc, argv, "hv");
2186 + if (c == -1) {
2187 + break;
2188 + }
2189 + switch (c) {
2190 + case '?':
2191 + case 'h':
2192 + help();
2193 + break;
2194 + case 'v':
2195 + verbose = 1;
2196 + break;
2197 + default:
2198 + help();
2199 + }
2200 + }
2201 +
2202 + /* Get the filename */
2203 + if ((optind + 1) != argc) {
2204 + help();
2205 + }
2206 + filename = argv[optind++];
2207 +
2208 + Error *errp = NULL;
2209 + VmaReader *vmar = vma_reader_create(filename, &errp);
2210 +
2211 + if (!vmar) {
2212 + g_error("%s", error_get_pretty(errp));
2213 + }
2214 +
2215 + if (verbose) {
2216 + print_content(vmar);
2217 + }
2218 +
2219 + if (vma_reader_verify(vmar, verbose, &errp) < 0) {
2220 + g_error("verify failed - %s", error_get_pretty(errp));
2221 + }
2222 +
2223 + vma_reader_destroy(vmar);
2224 +
2225 + bdrv_close_all();
2226 +
2227 + return ret;
2228 +}
2229 +
2230 +typedef struct BackupJob {
2231 + BlockBackend *target;
2232 + int64_t len;
2233 + VmaWriter *vmaw;
2234 + uint8_t dev_id;
2235 +} BackupJob;
2236 +
2237 +#define BACKUP_SECTORS_PER_CLUSTER (VMA_CLUSTER_SIZE / BDRV_SECTOR_SIZE)
2238 +
2239 +static void coroutine_fn backup_run_empty(void *opaque)
2240 +{
2241 + VmaWriter *vmaw = (VmaWriter *)opaque;
2242 +
2243 + vma_writer_flush_output(vmaw);
2244 +
2245 + Error *err = NULL;
2246 + if (vma_writer_close(vmaw, &err) != 0) {
2247 + g_warning("vma_writer_close failed %s", error_get_pretty(err));
2248 + }
2249 +}
2250 +
2251 +static void coroutine_fn backup_run(void *opaque)
2252 +{
2253 + BackupJob *job = (BackupJob *)opaque;
2254 + struct iovec iov;
2255 + QEMUIOVector qiov;
2256 +
2257 + int64_t start, end;
2258 + int ret = 0;
2259 +
2260 + unsigned char *buf = blk_blockalign(job->target, VMA_CLUSTER_SIZE);
2261 +
2262 + start = 0;
2263 + end = DIV_ROUND_UP(job->len / BDRV_SECTOR_SIZE,
2264 + BACKUP_SECTORS_PER_CLUSTER);
2265 +
2266 + for (; start < end; start++) {
2267 + iov.iov_base = buf;
2268 + iov.iov_len = VMA_CLUSTER_SIZE;
2269 + qemu_iovec_init_external(&qiov, &iov, 1);
2270 +
2271 + ret = blk_co_preadv(job->target, start * VMA_CLUSTER_SIZE,
2272 + VMA_CLUSTER_SIZE, &qiov, 0);
2273 + if (ret < 0) {
2274 + vma_writer_set_error(job->vmaw, "read error", -1);
2275 + goto out;
2276 + }
2277 +
2278 + size_t zb = 0;
2279 + if (vma_writer_write(job->vmaw, job->dev_id, start, buf, &zb) < 0) {
2280 + vma_writer_set_error(job->vmaw, "backup_dump_cb vma_writer_write failed", -1);
2281 + goto out;
2282 + }
2283 + }
2284 +
2285 +
2286 +out:
2287 + if (vma_writer_close_stream(job->vmaw, job->dev_id) <= 0) {
2288 + Error *err = NULL;
2289 + if (vma_writer_close(job->vmaw, &err) != 0) {
2290 + g_warning("vma_writer_close failed %s", error_get_pretty(err));
2291 + }
2292 + }
2293 + qemu_vfree(buf);
2294 +}
2295 +
2296 +static int create_archive(int argc, char **argv)
2297 +{
2298 + int i, c;
2299 + int verbose = 0;
2300 + const char *archivename;
2301 + GList *backup_coroutines = NULL;
2302 + GList *config_files = NULL;
2303 +
2304 + for (;;) {
2305 + c = getopt(argc, argv, "hvc:");
2306 + if (c == -1) {
2307 + break;
2308 + }
2309 + switch (c) {
2310 + case '?':
2311 + case 'h':
2312 + help();
2313 + break;
2314 + case 'c':
2315 + config_files = g_list_append(config_files, optarg);
2316 + break;
2317 + case 'v':
2318 + verbose = 1;
2319 + break;
2320 + default:
2321 + g_assert_not_reached();
2322 + }
2323 + }
2324 +
2325 +
2326 + /* make sure we an archive name */
2327 + if ((optind + 1) > argc) {
2328 + help();
2329 + }
2330 +
2331 + archivename = argv[optind++];
2332 +
2333 + uuid_t uuid;
2334 + uuid_generate(uuid);
2335 +
2336 + Error *local_err = NULL;
2337 + VmaWriter *vmaw = vma_writer_create(archivename, uuid, &local_err);
2338 +
2339 + if (vmaw == NULL) {
2340 + g_error("%s", error_get_pretty(local_err));
2341 + }
2342 +
2343 + GList *l = config_files;
2344 + while (l && l->data) {
2345 + char *name = l->data;
2346 + char *cdata = NULL;
2347 + gsize clen = 0;
2348 + GError *err = NULL;
2349 + if (!g_file_get_contents(name, &cdata, &clen, &err)) {
2350 + unlink(archivename);
2351 + g_error("Unable to read file: %s", err->message);
2352 + }
2353 +
2354 + if (vma_writer_add_config(vmaw, name, cdata, clen) != 0) {
2355 + unlink(archivename);
2356 + g_error("Unable to append config data %s (len = %zd)",
2357 + name, clen);
2358 + }
2359 + l = g_list_next(l);
2360 + }
2361 +
2362 + int devcount = 0;
2363 + while (optind < argc) {
2364 + const char *path = argv[optind++];
2365 + char *devname = NULL;
2366 + path = extract_devname(path, &devname, devcount++);
2367 +
2368 + Error *errp = NULL;
2369 + BlockBackend *target;
2370 +
2371 + target = blk_new_open(path, NULL, NULL, 0, &errp);
2372 + if (!target) {
2373 + unlink(archivename);
2374 + g_error("bdrv_open '%s' failed - %s", path, error_get_pretty(errp));
2375 + }
2376 + int64_t size = blk_getlength(target);
2377 + int dev_id = vma_writer_register_stream(vmaw, devname, size);
2378 + if (dev_id <= 0) {
2379 + unlink(archivename);
2380 + g_error("vma_writer_register_stream '%s' failed", devname);
2381 + }
2382 +
2383 + BackupJob *job = g_new0(BackupJob, 1);
2384 + job->len = size;
2385 + job->target = target;
2386 + job->vmaw = vmaw;
2387 + job->dev_id = dev_id;
2388 +
2389 + Coroutine *co = qemu_coroutine_create(backup_run, job);
2390 + // Don't enter coroutine yet, because it might write the header before
2391 + // all streams can be registered.
2392 + backup_coroutines = g_list_append(backup_coroutines, co);
2393 + }
2394 +
2395 + VmaStatus vmastat;
2396 + int percent = 0;
2397 + int last_percent = -1;
2398 +
2399 + if (devcount) {
2400 + GList *entry = backup_coroutines;
2401 + while (entry && entry->data) {
2402 + Coroutine *co = entry->data;
2403 + qemu_coroutine_enter(co);
2404 + entry = g_list_next(entry);
2405 + }
2406 +
2407 + while (1) {
2408 + main_loop_wait(false);
2409 + vma_writer_get_status(vmaw, &vmastat);
2410 +
2411 + if (verbose) {
2412 +
2413 + uint64_t total = 0;
2414 + uint64_t transferred = 0;
2415 + uint64_t zero_bytes = 0;
2416 +
2417 + int i;
2418 + for (i = 0; i < 256; i++) {
2419 + if (vmastat.stream_info[i].size) {
2420 + total += vmastat.stream_info[i].size;
2421 + transferred += vmastat.stream_info[i].transferred;
2422 + zero_bytes += vmastat.stream_info[i].zero_bytes;
2423 + }
2424 + }
2425 + percent = (transferred*100)/total;
2426 + if (percent != last_percent) {
2427 + fprintf(stderr, "progress %d%% %zd/%zd %zd\n", percent,
2428 + transferred, total, zero_bytes);
2429 + fflush(stderr);
2430 +
2431 + last_percent = percent;
2432 + }
2433 + }
2434 +
2435 + if (vmastat.closed) {
2436 + break;
2437 + }
2438 + }
2439 + } else {
2440 + Coroutine *co = qemu_coroutine_create(backup_run_empty, vmaw);
2441 + qemu_coroutine_enter(co);
2442 + while (1) {
2443 + main_loop_wait(false);
2444 + vma_writer_get_status(vmaw, &vmastat);
2445 + if (vmastat.closed) {
2446 + break;
2447 + }
2448 + }
2449 + }
2450 +
2451 + bdrv_drain_all();
2452 +
2453 + vma_writer_get_status(vmaw, &vmastat);
2454 +
2455 + if (verbose) {
2456 + for (i = 0; i < 256; i++) {
2457 + VmaStreamInfo *si = &vmastat.stream_info[i];
2458 + if (si->size) {
2459 + fprintf(stderr, "image %s: size=%zd zeros=%zd saved=%zd\n",
2460 + si->devname, si->size, si->zero_bytes,
2461 + si->size - si->zero_bytes);
2462 + }
2463 + }
2464 + }
2465 +
2466 + if (vmastat.status < 0) {
2467 + unlink(archivename);
2468 + g_error("creating vma archive failed");
2469 + }
2470 +
2471 + g_list_free(backup_coroutines);
2472 + g_list_free(config_files);
2473 + vma_writer_destroy(vmaw);
2474 + return 0;
2475 +}
2476 +
2477 +static int dump_config(int argc, char **argv)
2478 +{
2479 + int c, ret = 0;
2480 + const char *filename;
2481 + const char *config_name = "qemu-server.conf";
2482 +
2483 + for (;;) {
2484 + c = getopt(argc, argv, "hc:");
2485 + if (c == -1) {
2486 + break;
2487 + }
2488 + switch (c) {
2489 + case '?':
2490 + case 'h':
2491 + help();
2492 + break;
2493 + case 'c':
2494 + config_name = optarg;
2495 + break;
2496 + default:
2497 + help();
2498 + }
2499 + }
2500 +
2501 + /* Get the filename */
2502 + if ((optind + 1) != argc) {
2503 + help();
2504 + }
2505 + filename = argv[optind++];
2506 +
2507 + Error *errp = NULL;
2508 + VmaReader *vmar = vma_reader_create(filename, &errp);
2509 +
2510 + if (!vmar) {
2511 + g_error("%s", error_get_pretty(errp));
2512 + }
2513 +
2514 + int found = 0;
2515 + GList *l = vma_reader_get_config_data(vmar);
2516 + while (l && l->data) {
2517 + VmaConfigData *cdata = (VmaConfigData *)l->data;
2518 + l = g_list_next(l);
2519 + if (strcmp(cdata->name, config_name) == 0) {
2520 + found = 1;
2521 + fwrite(cdata->data, cdata->len, 1, stdout);
2522 + break;
2523 + }
2524 + }
2525 +
2526 + vma_reader_destroy(vmar);
2527 +
2528 + bdrv_close_all();
2529 +
2530 + if (!found) {
2531 + fprintf(stderr, "unable to find configuration data '%s'\n", config_name);
2532 + return -1;
2533 + }
2534 +
2535 + return ret;
2536 +}
2537 +
2538 +int main(int argc, char **argv)
2539 +{
2540 + const char *cmdname;
2541 + Error *main_loop_err = NULL;
2542 +
2543 + error_init(argv[0]);
2544 + module_call_init(MODULE_INIT_TRACE);
2545 + qemu_init_exec_dir(argv[0]);
2546 +
2547 + if (qemu_init_main_loop(&main_loop_err)) {
2548 + g_error("%s", error_get_pretty(main_loop_err));
2549 + }
2550 +
2551 + bdrv_init();
2552 + module_call_init(MODULE_INIT_QOM);
2553 +
2554 + if (argc < 2) {
2555 + help();
2556 + }
2557 +
2558 + cmdname = argv[1];
2559 + argc--; argv++;
2560 +
2561 +
2562 + if (!strcmp(cmdname, "list")) {
2563 + return list_content(argc, argv);
2564 + } else if (!strcmp(cmdname, "create")) {
2565 + return create_archive(argc, argv);
2566 + } else if (!strcmp(cmdname, "extract")) {
2567 + return extract_content(argc, argv);
2568 + } else if (!strcmp(cmdname, "verify")) {
2569 + return verify_content(argc, argv);
2570 + } else if (!strcmp(cmdname, "config")) {
2571 + return dump_config(argc, argv);
2572 + }
2573 +
2574 + help();
2575 + return 0;
2576 +}
2577 diff --git a/vma.h b/vma.h
2578 new file mode 100644
2579 index 0000000000..c895c97f6d
2580 --- /dev/null
2581 +++ b/vma.h
2582 @@ -0,0 +1,150 @@
2583 +/*
2584 + * VMA: Virtual Machine Archive
2585 + *
2586 + * Copyright (C) Proxmox Server Solutions
2587 + *
2588 + * Authors:
2589 + * Dietmar Maurer (dietmar@proxmox.com)
2590 + *
2591 + * This work is licensed under the terms of the GNU GPL, version 2 or later.
2592 + * See the COPYING file in the top-level directory.
2593 + *
2594 + */
2595 +
2596 +#ifndef BACKUP_VMA_H
2597 +#define BACKUP_VMA_H
2598 +
2599 +#include <uuid/uuid.h>
2600 +#include "qapi/error.h"
2601 +#include "block/block.h"
2602 +
2603 +#define VMA_BLOCK_BITS 12
2604 +#define VMA_BLOCK_SIZE (1<<VMA_BLOCK_BITS)
2605 +#define VMA_CLUSTER_BITS (VMA_BLOCK_BITS+4)
2606 +#define VMA_CLUSTER_SIZE (1<<VMA_CLUSTER_BITS)
2607 +
2608 +#if VMA_CLUSTER_SIZE != 65536
2609 +#error unexpected cluster size
2610 +#endif
2611 +
2612 +#define VMA_EXTENT_HEADER_SIZE 512
2613 +#define VMA_BLOCKS_PER_EXTENT 59
2614 +#define VMA_MAX_CONFIGS 256
2615 +
2616 +#define VMA_MAX_EXTENT_SIZE \
2617 + (VMA_EXTENT_HEADER_SIZE+VMA_CLUSTER_SIZE*VMA_BLOCKS_PER_EXTENT)
2618 +#if VMA_MAX_EXTENT_SIZE != 3867136
2619 +#error unexpected VMA_EXTENT_SIZE
2620 +#endif
2621 +
2622 +/* File Format Definitions */
2623 +
2624 +#define VMA_MAGIC (GUINT32_TO_BE(('V'<<24)|('M'<<16)|('A'<<8)|0x00))
2625 +#define VMA_EXTENT_MAGIC (GUINT32_TO_BE(('V'<<24)|('M'<<16)|('A'<<8)|'E'))
2626 +
2627 +typedef struct VmaDeviceInfoHeader {
2628 + uint32_t devname_ptr; /* offset into blob_buffer table */
2629 + uint32_t reserved0;
2630 + uint64_t size; /* device size in bytes */
2631 + uint64_t reserved1;
2632 + uint64_t reserved2;
2633 +} VmaDeviceInfoHeader;
2634 +
2635 +typedef struct VmaHeader {
2636 + uint32_t magic;
2637 + uint32_t version;
2638 + unsigned char uuid[16];
2639 + int64_t ctime;
2640 + unsigned char md5sum[16];
2641 +
2642 + uint32_t blob_buffer_offset;
2643 + uint32_t blob_buffer_size;
2644 + uint32_t header_size;
2645 +
2646 + unsigned char reserved[1984];
2647 +
2648 + uint32_t config_names[VMA_MAX_CONFIGS]; /* offset into blob_buffer table */
2649 + uint32_t config_data[VMA_MAX_CONFIGS]; /* offset into blob_buffer table */
2650 +
2651 + uint32_t reserved1;
2652 +
2653 + VmaDeviceInfoHeader dev_info[256];
2654 +} VmaHeader;
2655 +
2656 +typedef struct VmaExtentHeader {
2657 + uint32_t magic;
2658 + uint16_t reserved1;
2659 + uint16_t block_count;
2660 + unsigned char uuid[16];
2661 + unsigned char md5sum[16];
2662 + uint64_t blockinfo[VMA_BLOCKS_PER_EXTENT];
2663 +} VmaExtentHeader;
2664 +
2665 +/* functions/definitions to read/write vma files */
2666 +
2667 +typedef struct VmaReader VmaReader;
2668 +
2669 +typedef struct VmaWriter VmaWriter;
2670 +
2671 +typedef struct VmaConfigData {
2672 + const char *name;
2673 + const void *data;
2674 + uint32_t len;
2675 +} VmaConfigData;
2676 +
2677 +typedef struct VmaStreamInfo {
2678 + uint64_t size;
2679 + uint64_t cluster_count;
2680 + uint64_t transferred;
2681 + uint64_t zero_bytes;
2682 + int finished;
2683 + char *devname;
2684 +} VmaStreamInfo;
2685 +
2686 +typedef struct VmaStatus {
2687 + int status;
2688 + bool closed;
2689 + char errmsg[8192];
2690 + char uuid_str[37];
2691 + VmaStreamInfo stream_info[256];
2692 +} VmaStatus;
2693 +
2694 +typedef struct VmaDeviceInfo {
2695 + uint64_t size; /* device size in bytes */
2696 + const char *devname;
2697 +} VmaDeviceInfo;
2698 +
2699 +VmaWriter *vma_writer_create(const char *filename, uuid_t uuid, Error **errp);
2700 +int vma_writer_close(VmaWriter *vmaw, Error **errp);
2701 +void vma_writer_error_propagate(VmaWriter *vmaw, Error **errp);
2702 +void vma_writer_destroy(VmaWriter *vmaw);
2703 +int vma_writer_add_config(VmaWriter *vmaw, const char *name, gpointer data,
2704 + size_t len);
2705 +int vma_writer_register_stream(VmaWriter *vmaw, const char *devname,
2706 + size_t size);
2707 +
2708 +int64_t coroutine_fn vma_writer_write(VmaWriter *vmaw, uint8_t dev_id,
2709 + int64_t cluster_num,
2710 + const unsigned char *buf,
2711 + size_t *zero_bytes);
2712 +
2713 +int coroutine_fn vma_writer_close_stream(VmaWriter *vmaw, uint8_t dev_id);
2714 +int coroutine_fn vma_writer_flush_output(VmaWriter *vmaw);
2715 +
2716 +int vma_writer_get_status(VmaWriter *vmaw, VmaStatus *status);
2717 +void vma_writer_set_error(VmaWriter *vmaw, const char *fmt, ...);
2718 +
2719 +
2720 +VmaReader *vma_reader_create(const char *filename, Error **errp);
2721 +void vma_reader_destroy(VmaReader *vmar);
2722 +VmaHeader *vma_reader_get_header(VmaReader *vmar);
2723 +GList *vma_reader_get_config_data(VmaReader *vmar);
2724 +VmaDeviceInfo *vma_reader_get_device_info(VmaReader *vmar, guint8 dev_id);
2725 +int vma_reader_register_bs(VmaReader *vmar, guint8 dev_id,
2726 + BlockBackend *target, bool write_zeroes,
2727 + Error **errp);
2728 +int vma_reader_restore(VmaReader *vmar, int vmstate_fd, bool verbose,
2729 + Error **errp);
2730 +int vma_reader_verify(VmaReader *vmar, bool verbose, Error **errp);
2731 +
2732 +#endif /* BACKUP_VMA_H */
QEMU for PVE