]> git.proxmox.com Git - pve-qemu.git/blob - debian/patches/pve/0025-PVE-Backup-add-vma-backup-format-code.patch
projects / pve-qemu.git / blob
? search:


c4ed5bb51e01405a5ba9cbee428e78e6960da4d3
[pve-qemu.git] / debian / patches / pve / 0025-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 | 857 ++++++++++++++++++++++++++++++++++++++++++++++
13 vma-writer.c | 790 ++++++++++++++++++++++++++++++++++++++++++
14 vma.c | 851 +++++++++++++++++++++++++++++++++++++++++++++
15 vma.h | 150 ++++++++
16 6 files changed, 2655 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 c9d1fdca7d..72081a9974 100644
24 --- a/block/meson.build
25 +++ b/block/meson.build
26 @@ -44,6 +44,8 @@ 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 block_ss.add(when: 'CONFIG_QCOW1', if_true: files('qcow.c'))
35 diff --git a/meson.build b/meson.build
36 index 96de1a6ef9..54c23b9567 100644
37 --- a/meson.build
38 +++ b/meson.build
39 @@ -1202,6 +1202,8 @@ 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 @@ -3070,6 +3072,9 @@ 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..2b1d1cdab3
61 --- /dev/null
62 +++ b/vma-reader.c
63 @@ -0,0 +1,857 @@
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 + }
259 +
260 + if (vmar->md5csum) {
261 + g_checksum_free(vmar->md5csum);
262 + }
263 +
264 + if (vmar->blob_hash) {
265 + g_hash_table_destroy(vmar->blob_hash);
266 + }
267 +
268 + if (vmar->head_data) {
269 + g_free(vmar->head_data);
270 + }
271 +
272 + g_free(vmar);
273 +
274 +};
275 +
276 +static int vma_reader_read_head(VmaReader *vmar, Error **errp)
277 +{
278 + assert(vmar);
279 + assert(errp);
280 + assert(*errp == NULL);
281 +
282 + unsigned char md5sum[16];
283 + int i;
284 + int ret = 0;
285 +
286 + vmar->head_data = g_malloc(sizeof(VmaHeader));
287 +
288 + if (full_read(vmar->fd, vmar->head_data, sizeof(VmaHeader)) !=
289 + sizeof(VmaHeader)) {
290 + error_setg(errp, "can't read vma header - %s",
291 + errno ? g_strerror(errno) : "got EOF");
292 + return -1;
293 + }
294 +
295 + VmaHeader *h = (VmaHeader *)vmar->head_data;
296 +
297 + if (h->magic != VMA_MAGIC) {
298 + error_setg(errp, "not a vma file - wrong magic number");
299 + return -1;
300 + }
301 +
302 + uint32_t header_size = GUINT32_FROM_BE(h->header_size);
303 + int need = header_size - sizeof(VmaHeader);
304 + if (need <= 0) {
305 + error_setg(errp, "wrong vma header size %d", header_size);
306 + return -1;
307 + }
308 +
309 + vmar->head_data = g_realloc(vmar->head_data, header_size);
310 + h = (VmaHeader *)vmar->head_data;
311 +
312 + if (full_read(vmar->fd, vmar->head_data + sizeof(VmaHeader), need) !=
313 + need) {
314 + error_setg(errp, "can't read vma header data - %s",
315 + errno ? g_strerror(errno) : "got EOF");
316 + return -1;
317 + }
318 +
319 + memcpy(md5sum, h->md5sum, 16);
320 + memset(h->md5sum, 0, 16);
321 +
322 + g_checksum_reset(vmar->md5csum);
323 + g_checksum_update(vmar->md5csum, vmar->head_data, header_size);
324 + gsize csize = 16;
325 + g_checksum_get_digest(vmar->md5csum, (guint8 *)(h->md5sum), &csize);
326 +
327 + if (memcmp(md5sum, h->md5sum, 16) != 0) {
328 + error_setg(errp, "wrong vma header chechsum");
329 + return -1;
330 + }
331 +
332 + /* we can modify header data after checksum verify */
333 + h->header_size = header_size;
334 +
335 + h->version = GUINT32_FROM_BE(h->version);
336 + if (h->version != 1) {
337 + error_setg(errp, "wrong vma version %d", h->version);
338 + return -1;
339 + }
340 +
341 + h->ctime = GUINT64_FROM_BE(h->ctime);
342 + h->blob_buffer_offset = GUINT32_FROM_BE(h->blob_buffer_offset);
343 + h->blob_buffer_size = GUINT32_FROM_BE(h->blob_buffer_size);
344 +
345 + uint32_t bstart = h->blob_buffer_offset + 1;
346 + uint32_t bend = h->blob_buffer_offset + h->blob_buffer_size;
347 +
348 + if (bstart <= sizeof(VmaHeader)) {
349 + error_setg(errp, "wrong vma blob buffer offset %d",
350 + h->blob_buffer_offset);
351 + return -1;
352 + }
353 +
354 + if (bend > header_size) {
355 + error_setg(errp, "wrong vma blob buffer size %d/%d",
356 + h->blob_buffer_offset, h->blob_buffer_size);
357 + return -1;
358 + }
359 +
360 + while ((bstart + 2) <= bend) {
361 + uint32_t size = vmar->head_data[bstart] +
362 + (vmar->head_data[bstart+1] << 8);
363 + if ((bstart + size + 2) <= bend) {
364 + VmaBlob *blob = g_new0(VmaBlob, 1);
365 + blob->start = bstart - h->blob_buffer_offset;
366 + blob->len = size;
367 + blob->data = vmar->head_data + bstart + 2;
368 + g_hash_table_insert(vmar->blob_hash, &blob->start, blob);
369 + }
370 + bstart += size + 2;
371 + }
372 +
373 +
374 + int count = 0;
375 + for (i = 1; i < 256; i++) {
376 + VmaDeviceInfoHeader *dih = &h->dev_info[i];
377 + uint32_t devname_ptr = GUINT32_FROM_BE(dih->devname_ptr);
378 + uint64_t size = GUINT64_FROM_BE(dih->size);
379 + const char *devname = get_header_str(vmar, devname_ptr);
380 +
381 + if (size && devname) {
382 + count++;
383 + vmar->devinfo[i].size = size;
384 + vmar->devinfo[i].devname = devname;
385 +
386 + if (strcmp(devname, "vmstate") == 0) {
387 + vmar->vmstate_stream = i;
388 + }
389 + }
390 + }
391 +
392 + for (i = 0; i < VMA_MAX_CONFIGS; i++) {
393 + uint32_t name_ptr = GUINT32_FROM_BE(h->config_names[i]);
394 + uint32_t data_ptr = GUINT32_FROM_BE(h->config_data[i]);
395 +
396 + if (!(name_ptr && data_ptr)) {
397 + continue;
398 + }
399 + const char *name = get_header_str(vmar, name_ptr);
400 + const VmaBlob *blob = get_header_blob(vmar, data_ptr);
401 +
402 + if (!(name && blob)) {
403 + error_setg(errp, "vma contains invalid data pointers");
404 + return -1;
405 + }
406 +
407 + VmaConfigData *cdata = g_new0(VmaConfigData, 1);
408 + cdata->name = name;
409 + cdata->data = blob->data;
410 + cdata->len = blob->len;
411 +
412 + vmar->cdata_list = g_list_append(vmar->cdata_list, cdata);
413 + }
414 +
415 + return ret;
416 +};
417 +
418 +VmaReader *vma_reader_create(const char *filename, Error **errp)
419 +{
420 + assert(filename);
421 + assert(errp);
422 +
423 + VmaReader *vmar = g_new0(VmaReader, 1);
424 +
425 + if (strcmp(filename, "-") == 0) {
426 + vmar->fd = dup(0);
427 + } else {
428 + vmar->fd = open(filename, O_RDONLY);
429 + }
430 +
431 + if (vmar->fd < 0) {
432 + error_setg(errp, "can't open file %s - %s\n", filename,
433 + g_strerror(errno));
434 + goto err;
435 + }
436 +
437 + vmar->md5csum = g_checksum_new(G_CHECKSUM_MD5);
438 + if (!vmar->md5csum) {
439 + error_setg(errp, "can't allocate cmsum\n");
440 + goto err;
441 + }
442 +
443 + vmar->blob_hash = g_hash_table_new_full(g_int32_hash, g_int32_equal,
444 + NULL, g_free);
445 +
446 + if (vma_reader_read_head(vmar, errp) < 0) {
447 + goto err;
448 + }
449 +
450 + return vmar;
451 +
452 +err:
453 + if (vmar) {
454 + vma_reader_destroy(vmar);
455 + }
456 +
457 + return NULL;
458 +}
459 +
460 +VmaHeader *vma_reader_get_header(VmaReader *vmar)
461 +{
462 + assert(vmar);
463 + assert(vmar->head_data);
464 +
465 + return (VmaHeader *)(vmar->head_data);
466 +}
467 +
468 +GList *vma_reader_get_config_data(VmaReader *vmar)
469 +{
470 + assert(vmar);
471 + assert(vmar->head_data);
472 +
473 + return vmar->cdata_list;
474 +}
475 +
476 +VmaDeviceInfo *vma_reader_get_device_info(VmaReader *vmar, guint8 dev_id)
477 +{
478 + assert(vmar);
479 + assert(dev_id);
480 +
481 + if (vmar->devinfo[dev_id].size && vmar->devinfo[dev_id].devname) {
482 + return &vmar->devinfo[dev_id];
483 + }
484 +
485 + return NULL;
486 +}
487 +
488 +static void allocate_rstate(VmaReader *vmar, guint8 dev_id,
489 + BlockBackend *target, bool write_zeroes)
490 +{
491 + assert(vmar);
492 + assert(dev_id);
493 +
494 + vmar->rstate[dev_id].target = target;
495 + vmar->rstate[dev_id].write_zeroes = write_zeroes;
496 +
497 + int64_t size = vmar->devinfo[dev_id].size;
498 +
499 + int64_t bitmap_size = (size/BDRV_SECTOR_SIZE) +
500 + (VMA_CLUSTER_SIZE/BDRV_SECTOR_SIZE) * BITS_PER_LONG - 1;
501 + bitmap_size /= (VMA_CLUSTER_SIZE/BDRV_SECTOR_SIZE) * BITS_PER_LONG;
502 +
503 + vmar->rstate[dev_id].bitmap_size = bitmap_size;
504 + vmar->rstate[dev_id].bitmap = g_new0(unsigned long, bitmap_size);
505 +
506 + vmar->cluster_count += size/VMA_CLUSTER_SIZE;
507 +}
508 +
509 +int vma_reader_register_bs(VmaReader *vmar, guint8 dev_id, BlockBackend *target,
510 + bool write_zeroes, Error **errp)
511 +{
512 + assert(vmar);
513 + assert(target != NULL);
514 + assert(dev_id);
515 + assert(vmar->rstate[dev_id].target == NULL);
516 +
517 + int64_t size = blk_getlength(target);
518 + int64_t size_diff = size - vmar->devinfo[dev_id].size;
519 +
520 + /* storage types can have different size restrictions, so it
521 + * is not always possible to create an image with exact size.
522 + * So we tolerate a size difference up to 4MB.
523 + */
524 + if ((size_diff < 0) || (size_diff > 4*1024*1024)) {
525 + error_setg(errp, "vma_reader_register_bs for stream %s failed - "
526 + "unexpected size %zd != %zd", vmar->devinfo[dev_id].devname,
527 + size, vmar->devinfo[dev_id].size);
528 + return -1;
529 + }
530 +
531 + allocate_rstate(vmar, dev_id, target, write_zeroes);
532 +
533 + return 0;
534 +}
535 +
536 +static ssize_t safe_write(int fd, void *buf, size_t count)
537 +{
538 + ssize_t n;
539 +
540 + do {
541 + n = write(fd, buf, count);
542 + } while (n < 0 && errno == EINTR);
543 +
544 + return n;
545 +}
546 +
547 +static size_t full_write(int fd, void *buf, size_t len)
548 +{
549 + ssize_t n;
550 + size_t total;
551 +
552 + total = 0;
553 +
554 + while (len > 0) {
555 + n = safe_write(fd, buf, len);
556 + if (n < 0) {
557 + return n;
558 + }
559 + buf += n;
560 + total += n;
561 + len -= n;
562 + }
563 +
564 + if (len) {
565 + /* incomplete write ? */
566 + return -1;
567 + }
568 +
569 + return total;
570 +}
571 +
572 +static int restore_write_data(VmaReader *vmar, guint8 dev_id,
573 + BlockBackend *target, int vmstate_fd,
574 + unsigned char *buf, int64_t sector_num,
575 + int nb_sectors, Error **errp)
576 +{
577 + assert(vmar);
578 +
579 + if (dev_id == vmar->vmstate_stream) {
580 + if (vmstate_fd >= 0) {
581 + int len = nb_sectors * BDRV_SECTOR_SIZE;
582 + int res = full_write(vmstate_fd, buf, len);
583 + if (res < 0) {
584 + error_setg(errp, "write vmstate failed %d", res);
585 + return -1;
586 + }
587 + }
588 + } else {
589 + int res = blk_pwrite(target, sector_num * BDRV_SECTOR_SIZE, buf, nb_sectors * BDRV_SECTOR_SIZE, 0);
590 + if (res < 0) {
591 + error_setg(errp, "blk_pwrite to %s failed (%d)",
592 + bdrv_get_device_name(blk_bs(target)), res);
593 + return -1;
594 + }
595 + }
596 + return 0;
597 +}
598 +
599 +static int restore_extent(VmaReader *vmar, unsigned char *buf,
600 + int extent_size, int vmstate_fd,
601 + bool verbose, bool verify, Error **errp)
602 +{
603 + assert(vmar);
604 + assert(buf);
605 +
606 + VmaExtentHeader *ehead = (VmaExtentHeader *)buf;
607 + int start = VMA_EXTENT_HEADER_SIZE;
608 + int i;
609 +
610 + for (i = 0; i < VMA_BLOCKS_PER_EXTENT; i++) {
611 + uint64_t block_info = GUINT64_FROM_BE(ehead->blockinfo[i]);
612 + uint64_t cluster_num = block_info & 0xffffffff;
613 + uint8_t dev_id = (block_info >> 32) & 0xff;
614 + uint16_t mask = block_info >> (32+16);
615 + int64_t max_sector;
616 +
617 + if (!dev_id) {
618 + continue;
619 + }
620 +
621 + VmaRestoreState *rstate = &vmar->rstate[dev_id];
622 + BlockBackend *target = NULL;
623 +
624 + if (dev_id != vmar->vmstate_stream) {
625 + target = rstate->target;
626 + if (!verify && !target) {
627 + error_setg(errp, "got wrong dev id %d", dev_id);
628 + return -1;
629 + }
630 +
631 + if (vma_reader_get_bitmap(rstate, cluster_num)) {
632 + error_setg(errp, "found duplicated cluster %zd for stream %s",
633 + cluster_num, vmar->devinfo[dev_id].devname);
634 + return -1;
635 + }
636 + vma_reader_set_bitmap(rstate, cluster_num, 1);
637 +
638 + max_sector = vmar->devinfo[dev_id].size/BDRV_SECTOR_SIZE;
639 + } else {
640 + max_sector = G_MAXINT64;
641 + if (cluster_num != vmar->vmstate_clusters) {
642 + error_setg(errp, "found out of order vmstate data");
643 + return -1;
644 + }
645 + vmar->vmstate_clusters++;
646 + }
647 +
648 + vmar->clusters_read++;
649 +
650 + if (verbose) {
651 + time_t duration = time(NULL) - vmar->start_time;
652 + int percent = (vmar->clusters_read*100)/vmar->cluster_count;
653 + if (percent != vmar->clusters_read_per) {
654 + printf("progress %d%% (read %zd bytes, duration %zd sec)\n",
655 + percent, vmar->clusters_read*VMA_CLUSTER_SIZE,
656 + duration);
657 + fflush(stdout);
658 + vmar->clusters_read_per = percent;
659 + }
660 + }
661 +
662 + /* try to write whole clusters to speedup restore */
663 + if (mask == 0xffff) {
664 + if ((start + VMA_CLUSTER_SIZE) > extent_size) {
665 + error_setg(errp, "short vma extent - too many blocks");
666 + return -1;
667 + }
668 + int64_t sector_num = (cluster_num * VMA_CLUSTER_SIZE) /
669 + BDRV_SECTOR_SIZE;
670 + int64_t end_sector = sector_num +
671 + VMA_CLUSTER_SIZE/BDRV_SECTOR_SIZE;
672 +
673 + if (end_sector > max_sector) {
674 + end_sector = max_sector;
675 + }
676 +
677 + if (end_sector <= sector_num) {
678 + error_setg(errp, "got wrong block address - write beyond end");
679 + return -1;
680 + }
681 +
682 + if (!verify) {
683 + int nb_sectors = end_sector - sector_num;
684 + if (restore_write_data(vmar, dev_id, target, vmstate_fd,
685 + buf + start, sector_num, nb_sectors,
686 + errp) < 0) {
687 + return -1;
688 + }
689 + }
690 +
691 + start += VMA_CLUSTER_SIZE;
692 + } else {
693 + int j;
694 + int bit = 1;
695 +
696 + for (j = 0; j < 16; j++) {
697 + int64_t sector_num = (cluster_num*VMA_CLUSTER_SIZE +
698 + j*VMA_BLOCK_SIZE)/BDRV_SECTOR_SIZE;
699 +
700 + int64_t end_sector = sector_num +
701 + VMA_BLOCK_SIZE/BDRV_SECTOR_SIZE;
702 + if (end_sector > max_sector) {
703 + end_sector = max_sector;
704 + }
705 +
706 + if (mask & bit) {
707 + if ((start + VMA_BLOCK_SIZE) > extent_size) {
708 + error_setg(errp, "short vma extent - too many blocks");
709 + return -1;
710 + }
711 +
712 + if (end_sector <= sector_num) {
713 + error_setg(errp, "got wrong block address - "
714 + "write beyond end");
715 + return -1;
716 + }
717 +
718 + if (!verify) {
719 + int nb_sectors = end_sector - sector_num;
720 + if (restore_write_data(vmar, dev_id, target, vmstate_fd,
721 + buf + start, sector_num,
722 + nb_sectors, errp) < 0) {
723 + return -1;
724 + }
725 + }
726 +
727 + start += VMA_BLOCK_SIZE;
728 +
729 + } else {
730 +
731 +
732 + if (end_sector > sector_num) {
733 + /* Todo: use bdrv_co_write_zeroes (but that need to
734 + * be run inside coroutine?)
735 + */
736 + int nb_sectors = end_sector - sector_num;
737 + int zero_size = BDRV_SECTOR_SIZE*nb_sectors;
738 + vmar->zero_cluster_data += zero_size;
739 + if (mask != 0) {
740 + vmar->partial_zero_cluster_data += zero_size;
741 + }
742 +
743 + if (rstate->write_zeroes && !verify) {
744 + if (restore_write_data(vmar, dev_id, target, vmstate_fd,
745 + zero_vma_block, sector_num,
746 + nb_sectors, errp) < 0) {
747 + return -1;
748 + }
749 + }
750 + }
751 + }
752 +
753 + bit = bit << 1;
754 + }
755 + }
756 + }
757 +
758 + if (start != extent_size) {
759 + error_setg(errp, "vma extent error - missing blocks");
760 + return -1;
761 + }
762 +
763 + return 0;
764 +}
765 +
766 +static int vma_reader_restore_full(VmaReader *vmar, int vmstate_fd,
767 + bool verbose, bool verify,
768 + Error **errp)
769 +{
770 + assert(vmar);
771 + assert(vmar->head_data);
772 +
773 + int ret = 0;
774 + unsigned char buf[VMA_MAX_EXTENT_SIZE];
775 + int buf_pos = 0;
776 + unsigned char md5sum[16];
777 + VmaHeader *h = (VmaHeader *)vmar->head_data;
778 +
779 + vmar->start_time = time(NULL);
780 +
781 + while (1) {
782 + int bytes = full_read(vmar->fd, buf + buf_pos, sizeof(buf) - buf_pos);
783 + if (bytes < 0) {
784 + error_setg(errp, "read failed - %s", g_strerror(errno));
785 + return -1;
786 + }
787 +
788 + buf_pos += bytes;
789 +
790 + if (!buf_pos) {
791 + break; /* EOF */
792 + }
793 +
794 + if (buf_pos < VMA_EXTENT_HEADER_SIZE) {
795 + error_setg(errp, "read short extent (%d bytes)", buf_pos);
796 + return -1;
797 + }
798 +
799 + VmaExtentHeader *ehead = (VmaExtentHeader *)buf;
800 +
801 + /* extract md5sum */
802 + memcpy(md5sum, ehead->md5sum, sizeof(ehead->md5sum));
803 + memset(ehead->md5sum, 0, sizeof(ehead->md5sum));
804 +
805 + g_checksum_reset(vmar->md5csum);
806 + g_checksum_update(vmar->md5csum, buf, VMA_EXTENT_HEADER_SIZE);
807 + gsize csize = 16;
808 + g_checksum_get_digest(vmar->md5csum, ehead->md5sum, &csize);
809 +
810 + if (memcmp(md5sum, ehead->md5sum, 16) != 0) {
811 + error_setg(errp, "wrong vma extent header chechsum");
812 + return -1;
813 + }
814 +
815 + if (memcmp(h->uuid, ehead->uuid, sizeof(ehead->uuid)) != 0) {
816 + error_setg(errp, "wrong vma extent uuid");
817 + return -1;
818 + }
819 +
820 + if (ehead->magic != VMA_EXTENT_MAGIC || ehead->reserved1 != 0) {
821 + error_setg(errp, "wrong vma extent header magic");
822 + return -1;
823 + }
824 +
825 + int block_count = GUINT16_FROM_BE(ehead->block_count);
826 + int extent_size = VMA_EXTENT_HEADER_SIZE + block_count*VMA_BLOCK_SIZE;
827 +
828 + if (buf_pos < extent_size) {
829 + error_setg(errp, "short vma extent (%d < %d)", buf_pos,
830 + extent_size);
831 + return -1;
832 + }
833 +
834 + if (restore_extent(vmar, buf, extent_size, vmstate_fd, verbose,
835 + verify, errp) < 0) {
836 + return -1;
837 + }
838 +
839 + if (buf_pos > extent_size) {
840 + memmove(buf, buf + extent_size, buf_pos - extent_size);
841 + buf_pos = buf_pos - extent_size;
842 + } else {
843 + buf_pos = 0;
844 + }
845 + }
846 +
847 + bdrv_drain_all();
848 +
849 + int i;
850 + for (i = 1; i < 256; i++) {
851 + VmaRestoreState *rstate = &vmar->rstate[i];
852 + if (!rstate->target) {
853 + continue;
854 + }
855 +
856 + if (blk_flush(rstate->target) < 0) {
857 + error_setg(errp, "vma blk_flush %s failed",
858 + vmar->devinfo[i].devname);
859 + return -1;
860 + }
861 +
862 + if (vmar->devinfo[i].size &&
863 + (strcmp(vmar->devinfo[i].devname, "vmstate") != 0)) {
864 + assert(rstate->bitmap);
865 +
866 + int64_t cluster_num, end;
867 +
868 + end = (vmar->devinfo[i].size + VMA_CLUSTER_SIZE - 1) /
869 + VMA_CLUSTER_SIZE;
870 +
871 + for (cluster_num = 0; cluster_num < end; cluster_num++) {
872 + if (!vma_reader_get_bitmap(rstate, cluster_num)) {
873 + error_setg(errp, "detected missing cluster %zd "
874 + "for stream %s", cluster_num,
875 + vmar->devinfo[i].devname);
876 + return -1;
877 + }
878 + }
879 + }
880 + }
881 +
882 + if (verbose) {
883 + if (vmar->clusters_read) {
884 + printf("total bytes read %zd, sparse bytes %zd (%.3g%%)\n",
885 + vmar->clusters_read*VMA_CLUSTER_SIZE,
886 + vmar->zero_cluster_data,
887 + (double)(100.0*vmar->zero_cluster_data)/
888 + (vmar->clusters_read*VMA_CLUSTER_SIZE));
889 +
890 + int64_t datasize = vmar->clusters_read*VMA_CLUSTER_SIZE-vmar->zero_cluster_data;
891 + if (datasize) { // this does not make sense for empty files
892 + printf("space reduction due to 4K zero blocks %.3g%%\n",
893 + (double)(100.0*vmar->partial_zero_cluster_data) / datasize);
894 + }
895 + } else {
896 + printf("vma archive contains no image data\n");
897 + }
898 + }
899 + return ret;
900 +}
901 +
902 +int vma_reader_restore(VmaReader *vmar, int vmstate_fd, bool verbose,
903 + Error **errp)
904 +{
905 + return vma_reader_restore_full(vmar, vmstate_fd, verbose, false, errp);
906 +}
907 +
908 +int vma_reader_verify(VmaReader *vmar, bool verbose, Error **errp)
909 +{
910 + guint8 dev_id;
911 +
912 + for (dev_id = 1; dev_id < 255; dev_id++) {
913 + if (vma_reader_get_device_info(vmar, dev_id)) {
914 + allocate_rstate(vmar, dev_id, NULL, false);
915 + }
916 + }
917 +
918 + return vma_reader_restore_full(vmar, -1, verbose, true, errp);
919 +}
920 +
921 diff --git a/vma-writer.c b/vma-writer.c
922 new file mode 100644
923 index 0000000000..11d8321ffd
924 --- /dev/null
925 +++ b/vma-writer.c
926 @@ -0,0 +1,790 @@
927 +/*
928 + * VMA: Virtual Machine Archive
929 + *
930 + * Copyright (C) 2012 Proxmox Server Solutions
931 + *
932 + * Authors:
933 + * Dietmar Maurer (dietmar@proxmox.com)
934 + *
935 + * This work is licensed under the terms of the GNU GPL, version 2 or later.
936 + * See the COPYING file in the top-level directory.
937 + *
938 + */
939 +
940 +#include "qemu/osdep.h"
941 +#include <glib.h>
942 +#include <uuid/uuid.h>
943 +
944 +#include "vma.h"
945 +#include "block/block.h"
946 +#include "monitor/monitor.h"
947 +#include "qemu/main-loop.h"
948 +#include "qemu/coroutine.h"
949 +#include "qemu/cutils.h"
950 +
951 +#define DEBUG_VMA 0
952 +
953 +#define DPRINTF(fmt, ...)\
954 + do { if (DEBUG_VMA) { printf("vma: " fmt, ## __VA_ARGS__); } } while (0)
955 +
956 +#define WRITE_BUFFERS 5
957 +#define HEADER_CLUSTERS 8
958 +#define HEADERBUF_SIZE (VMA_CLUSTER_SIZE*HEADER_CLUSTERS)
959 +
960 +struct VmaWriter {
961 + int fd;
962 + FILE *cmd;
963 + int status;
964 + char errmsg[8192];
965 + uuid_t uuid;
966 + bool header_written;
967 + bool closed;
968 +
969 + /* we always write extents */
970 + unsigned char *outbuf;
971 + int outbuf_pos; /* in bytes */
972 + int outbuf_count; /* in VMA_BLOCKS */
973 + uint64_t outbuf_block_info[VMA_BLOCKS_PER_EXTENT];
974 +
975 + unsigned char *headerbuf;
976 +
977 + GChecksum *md5csum;
978 + CoMutex flush_lock;
979 + Coroutine *co_writer;
980 +
981 + /* drive informations */
982 + VmaStreamInfo stream_info[256];
983 + guint stream_count;
984 +
985 + guint8 vmstate_stream;
986 + uint32_t vmstate_clusters;
987 +
988 + /* header blob table */
989 + char *header_blob_table;
990 + uint32_t header_blob_table_size;
991 + uint32_t header_blob_table_pos;
992 +
993 + /* store for config blobs */
994 + uint32_t config_names[VMA_MAX_CONFIGS]; /* offset into blob_buffer table */
995 + uint32_t config_data[VMA_MAX_CONFIGS]; /* offset into blob_buffer table */
996 + uint32_t config_count;
997 +};
998 +
999 +void vma_writer_set_error(VmaWriter *vmaw, const char *fmt, ...)
1000 +{
1001 + va_list ap;
1002 +
1003 + if (vmaw->status < 0) {
1004 + return;
1005 + }
1006 +
1007 + vmaw->status = -1;
1008 +
1009 + va_start(ap, fmt);
1010 + g_vsnprintf(vmaw->errmsg, sizeof(vmaw->errmsg), fmt, ap);
1011 + va_end(ap);
1012 +
1013 + DPRINTF("vma_writer_set_error: %s\n", vmaw->errmsg);
1014 +}
1015 +
1016 +static uint32_t allocate_header_blob(VmaWriter *vmaw, const char *data,
1017 + size_t len)
1018 +{
1019 + if (len > 65535) {
1020 + return 0;
1021 + }
1022 +
1023 + if (!vmaw->header_blob_table ||
1024 + (vmaw->header_blob_table_size <
1025 + (vmaw->header_blob_table_pos + len + 2))) {
1026 + int newsize = vmaw->header_blob_table_size + ((len + 2 + 511)/512)*512;
1027 +
1028 + vmaw->header_blob_table = g_realloc(vmaw->header_blob_table, newsize);
1029 + memset(vmaw->header_blob_table + vmaw->header_blob_table_size,
1030 + 0, newsize - vmaw->header_blob_table_size);
1031 + vmaw->header_blob_table_size = newsize;
1032 + }
1033 +
1034 + uint32_t cpos = vmaw->header_blob_table_pos;
1035 + vmaw->header_blob_table[cpos] = len & 255;
1036 + vmaw->header_blob_table[cpos+1] = (len >> 8) & 255;
1037 + memcpy(vmaw->header_blob_table + cpos + 2, data, len);
1038 + vmaw->header_blob_table_pos += len + 2;
1039 + return cpos;
1040 +}
1041 +
1042 +static uint32_t allocate_header_string(VmaWriter *vmaw, const char *str)
1043 +{
1044 + assert(vmaw);
1045 +
1046 + size_t len = strlen(str) + 1;
1047 +
1048 + return allocate_header_blob(vmaw, str, len);
1049 +}
1050 +
1051 +int vma_writer_add_config(VmaWriter *vmaw, const char *name, gpointer data,
1052 + gsize len)
1053 +{
1054 + assert(vmaw);
1055 + assert(!vmaw->header_written);
1056 + assert(vmaw->config_count < VMA_MAX_CONFIGS);
1057 + assert(name);
1058 + assert(data);
1059 +
1060 + gchar *basename = g_path_get_basename(name);
1061 + uint32_t name_ptr = allocate_header_string(vmaw, basename);
1062 + g_free(basename);
1063 +
1064 + if (!name_ptr) {
1065 + return -1;
1066 + }
1067 +
1068 + uint32_t data_ptr = allocate_header_blob(vmaw, data, len);
1069 + if (!data_ptr) {
1070 + return -1;
1071 + }
1072 +
1073 + vmaw->config_names[vmaw->config_count] = name_ptr;
1074 + vmaw->config_data[vmaw->config_count] = data_ptr;
1075 +
1076 + vmaw->config_count++;
1077 +
1078 + return 0;
1079 +}
1080 +
1081 +int vma_writer_register_stream(VmaWriter *vmaw, const char *devname,
1082 + size_t size)
1083 +{
1084 + assert(vmaw);
1085 + assert(devname);
1086 + assert(!vmaw->status);
1087 +
1088 + if (vmaw->header_written) {
1089 + vma_writer_set_error(vmaw, "vma_writer_register_stream: header "
1090 + "already written");
1091 + return -1;
1092 + }
1093 +
1094 + guint n = vmaw->stream_count + 1;
1095 +
1096 + /* we can have dev_ids form 1 to 255 (0 reserved)
1097 + * 255(-1) reseverd for safety
1098 + */
1099 + if (n > 254) {
1100 + vma_writer_set_error(vmaw, "vma_writer_register_stream: "
1101 + "too many drives");
1102 + return -1;
1103 + }
1104 +
1105 + if (size <= 0) {
1106 + vma_writer_set_error(vmaw, "vma_writer_register_stream: "
1107 + "got strange size %zd", size);
1108 + return -1;
1109 + }
1110 +
1111 + DPRINTF("vma_writer_register_stream %s %zu %d\n", devname, size, n);
1112 +
1113 + vmaw->stream_info[n].devname = g_strdup(devname);
1114 + vmaw->stream_info[n].size = size;
1115 +
1116 + vmaw->stream_info[n].cluster_count = (size + VMA_CLUSTER_SIZE - 1) /
1117 + VMA_CLUSTER_SIZE;
1118 +
1119 + vmaw->stream_count = n;
1120 +
1121 + if (strcmp(devname, "vmstate") == 0) {
1122 + vmaw->vmstate_stream = n;
1123 + }
1124 +
1125 + return n;
1126 +}
1127 +
1128 +static void coroutine_fn yield_until_fd_writable(int fd)
1129 +{
1130 + assert(qemu_in_coroutine());
1131 + AioContext *ctx = qemu_get_current_aio_context();
1132 + aio_set_fd_handler(ctx, fd, false, NULL, (IOHandler *)qemu_coroutine_enter,
1133 + NULL, qemu_coroutine_self());
1134 + qemu_coroutine_yield();
1135 + aio_set_fd_handler(ctx, fd, false, NULL, NULL, NULL, NULL);
1136 +}
1137 +
1138 +static ssize_t coroutine_fn
1139 +vma_queue_write(VmaWriter *vmaw, const void *buf, size_t bytes)
1140 +{
1141 + DPRINTF("vma_queue_write enter %zd\n", bytes);
1142 +
1143 + assert(vmaw);
1144 + assert(buf);
1145 + assert(bytes <= VMA_MAX_EXTENT_SIZE);
1146 +
1147 + size_t done = 0;
1148 + ssize_t ret;
1149 +
1150 + assert(vmaw->co_writer == NULL);
1151 +
1152 + vmaw->co_writer = qemu_coroutine_self();
1153 +
1154 + while (done < bytes) {
1155 + if (vmaw->status < 0) {
1156 + DPRINTF("vma_queue_write detected canceled backup\n");
1157 + done = -1;
1158 + break;
1159 + }
1160 + yield_until_fd_writable(vmaw->fd);
1161 + ret = write(vmaw->fd, buf + done, bytes - done);
1162 + if (ret > 0) {
1163 + done += ret;
1164 + DPRINTF("vma_queue_write written %zd %zd\n", done, ret);
1165 + } else if (ret < 0) {
1166 + if (errno == EAGAIN || errno == EWOULDBLOCK) {
1167 + /* try again */
1168 + } else {
1169 + vma_writer_set_error(vmaw, "vma_queue_write: write error - %s",
1170 + g_strerror(errno));
1171 + done = -1; /* always return failure for partial writes */
1172 + break;
1173 + }
1174 + } else if (ret == 0) {
1175 + /* should not happen - simply try again */
1176 + }
1177 + }
1178 +
1179 + vmaw->co_writer = NULL;
1180 +
1181 + return (done == bytes) ? bytes : -1;
1182 +}
1183 +
1184 +VmaWriter *vma_writer_create(const char *filename, uuid_t uuid, Error **errp)
1185 +{
1186 + const char *p;
1187 +
1188 + assert(sizeof(VmaHeader) == (4096 + 8192));
1189 + assert(G_STRUCT_OFFSET(VmaHeader, config_names) == 2044);
1190 + assert(G_STRUCT_OFFSET(VmaHeader, config_data) == 3068);
1191 + assert(G_STRUCT_OFFSET(VmaHeader, dev_info) == 4096);
1192 + assert(sizeof(VmaExtentHeader) == 512);
1193 +
1194 + VmaWriter *vmaw = g_new0(VmaWriter, 1);
1195 + vmaw->fd = -1;
1196 +
1197 + vmaw->md5csum = g_checksum_new(G_CHECKSUM_MD5);
1198 + if (!vmaw->md5csum) {
1199 + error_setg(errp, "can't allocate cmsum\n");
1200 + goto err;
1201 + }
1202 +
1203 + if (strstart(filename, "exec:", &p)) {
1204 + vmaw->cmd = popen(p, "w");
1205 + if (vmaw->cmd == NULL) {
1206 + error_setg(errp, "can't popen command '%s' - %s\n", p,
1207 + g_strerror(errno));
1208 + goto err;
1209 + }
1210 + vmaw->fd = fileno(vmaw->cmd);
1211 +
1212 + /* try to use O_NONBLOCK */
1213 + fcntl(vmaw->fd, F_SETFL, fcntl(vmaw->fd, F_GETFL)|O_NONBLOCK);
1214 +
1215 + } else {
1216 + struct stat st;
1217 + int oflags;
1218 + const char *tmp_id_str;
1219 +
1220 + if ((stat(filename, &st) == 0) && S_ISFIFO(st.st_mode)) {
1221 + oflags = O_NONBLOCK|O_WRONLY;
1222 + vmaw->fd = qemu_open(filename, oflags, errp);
1223 + } else if (strstart(filename, "/dev/fdset/", &tmp_id_str)) {
1224 + oflags = O_NONBLOCK|O_WRONLY;
1225 + vmaw->fd = qemu_open(filename, oflags, errp);
1226 + } else if (strstart(filename, "/dev/fdname/", &tmp_id_str)) {
1227 + vmaw->fd = monitor_get_fd(monitor_cur(), tmp_id_str, errp);
1228 + if (vmaw->fd < 0) {
1229 + goto err;
1230 + }
1231 + /* try to use O_NONBLOCK */
1232 + fcntl(vmaw->fd, F_SETFL, fcntl(vmaw->fd, F_GETFL)|O_NONBLOCK);
1233 + } else {
1234 + oflags = O_NONBLOCK|O_DIRECT|O_WRONLY|O_EXCL;
1235 + vmaw->fd = qemu_create(filename, oflags, 0644, errp);
1236 + }
1237 +
1238 + if (vmaw->fd < 0) {
1239 + error_setg(errp, "can't open file %s - %s\n", filename,
1240 + g_strerror(errno));
1241 + goto err;
1242 + }
1243 + }
1244 +
1245 + /* we use O_DIRECT, so we need to align IO buffers */
1246 +
1247 + vmaw->outbuf = qemu_memalign(512, VMA_MAX_EXTENT_SIZE);
1248 + vmaw->headerbuf = qemu_memalign(512, HEADERBUF_SIZE);
1249 +
1250 + vmaw->outbuf_count = 0;
1251 + vmaw->outbuf_pos = VMA_EXTENT_HEADER_SIZE;
1252 +
1253 + vmaw->header_blob_table_pos = 1; /* start at pos 1 */
1254 +
1255 + qemu_co_mutex_init(&vmaw->flush_lock);
1256 +
1257 + uuid_copy(vmaw->uuid, uuid);
1258 +
1259 + return vmaw;
1260 +
1261 +err:
1262 + if (vmaw) {
1263 + if (vmaw->cmd) {
1264 + pclose(vmaw->cmd);
1265 + } else if (vmaw->fd >= 0) {
1266 + close(vmaw->fd);
1267 + }
1268 +
1269 + if (vmaw->md5csum) {
1270 + g_checksum_free(vmaw->md5csum);
1271 + }
1272 +
1273 + g_free(vmaw);
1274 + }
1275 +
1276 + return NULL;
1277 +}
1278 +
1279 +static int coroutine_fn vma_write_header(VmaWriter *vmaw)
1280 +{
1281 + assert(vmaw);
1282 + unsigned char *buf = vmaw->headerbuf;
1283 + VmaHeader *head = (VmaHeader *)buf;
1284 +
1285 + int i;
1286 +
1287 + DPRINTF("VMA WRITE HEADER\n");
1288 +
1289 + if (vmaw->status < 0) {
1290 + return vmaw->status;
1291 + }
1292 +
1293 + memset(buf, 0, HEADERBUF_SIZE);
1294 +
1295 + head->magic = VMA_MAGIC;
1296 + head->version = GUINT32_TO_BE(1); /* v1 */
1297 + memcpy(head->uuid, vmaw->uuid, 16);
1298 +
1299 + time_t ctime = time(NULL);
1300 + head->ctime = GUINT64_TO_BE(ctime);
1301 +
1302 + for (i = 0; i < VMA_MAX_CONFIGS; i++) {
1303 + head->config_names[i] = GUINT32_TO_BE(vmaw->config_names[i]);
1304 + head->config_data[i] = GUINT32_TO_BE(vmaw->config_data[i]);
1305 + }
1306 +
1307 + /* 32 bytes per device (12 used currently) = 8192 bytes max */
1308 + for (i = 1; i <= 254; i++) {
1309 + VmaStreamInfo *si = &vmaw->stream_info[i];
1310 + if (si->size) {
1311 + assert(si->devname);
1312 + uint32_t devname_ptr = allocate_header_string(vmaw, si->devname);
1313 + if (!devname_ptr) {
1314 + return -1;
1315 + }
1316 + head->dev_info[i].devname_ptr = GUINT32_TO_BE(devname_ptr);
1317 + head->dev_info[i].size = GUINT64_TO_BE(si->size);
1318 + }
1319 + }
1320 +
1321 + uint32_t header_size = sizeof(VmaHeader) + vmaw->header_blob_table_size;
1322 + head->header_size = GUINT32_TO_BE(header_size);
1323 +
1324 + if (header_size > HEADERBUF_SIZE) {
1325 + return -1; /* just to be sure */
1326 + }
1327 +
1328 + uint32_t blob_buffer_offset = sizeof(VmaHeader);
1329 + memcpy(buf + blob_buffer_offset, vmaw->header_blob_table,
1330 + vmaw->header_blob_table_size);
1331 + head->blob_buffer_offset = GUINT32_TO_BE(blob_buffer_offset);
1332 + head->blob_buffer_size = GUINT32_TO_BE(vmaw->header_blob_table_pos);
1333 +
1334 + g_checksum_reset(vmaw->md5csum);
1335 + g_checksum_update(vmaw->md5csum, (const guchar *)buf, header_size);
1336 + gsize csize = 16;
1337 + g_checksum_get_digest(vmaw->md5csum, (guint8 *)(head->md5sum), &csize);
1338 +
1339 + return vma_queue_write(vmaw, buf, header_size);
1340 +}
1341 +
1342 +static int coroutine_fn vma_writer_flush(VmaWriter *vmaw)
1343 +{
1344 + assert(vmaw);
1345 +
1346 + int ret;
1347 + int i;
1348 +
1349 + if (vmaw->status < 0) {
1350 + return vmaw->status;
1351 + }
1352 +
1353 + if (!vmaw->header_written) {
1354 + vmaw->header_written = true;
1355 + ret = vma_write_header(vmaw);
1356 + if (ret < 0) {
1357 + vma_writer_set_error(vmaw, "vma_writer_flush: write header failed");
1358 + return ret;
1359 + }
1360 + }
1361 +
1362 + DPRINTF("VMA WRITE FLUSH %d %d\n", vmaw->outbuf_count, vmaw->outbuf_pos);
1363 +
1364 +
1365 + VmaExtentHeader *ehead = (VmaExtentHeader *)vmaw->outbuf;
1366 +
1367 + ehead->magic = VMA_EXTENT_MAGIC;
1368 + ehead->reserved1 = 0;
1369 +
1370 + for (i = 0; i < VMA_BLOCKS_PER_EXTENT; i++) {
1371 + ehead->blockinfo[i] = GUINT64_TO_BE(vmaw->outbuf_block_info[i]);
1372 + }
1373 +
1374 + guint16 block_count = (vmaw->outbuf_pos - VMA_EXTENT_HEADER_SIZE) /
1375 + VMA_BLOCK_SIZE;
1376 +
1377 + ehead->block_count = GUINT16_TO_BE(block_count);
1378 +
1379 + memcpy(ehead->uuid, vmaw->uuid, sizeof(ehead->uuid));
1380 + memset(ehead->md5sum, 0, sizeof(ehead->md5sum));
1381 +
1382 + g_checksum_reset(vmaw->md5csum);
1383 + g_checksum_update(vmaw->md5csum, vmaw->outbuf, VMA_EXTENT_HEADER_SIZE);
1384 + gsize csize = 16;
1385 + g_checksum_get_digest(vmaw->md5csum, ehead->md5sum, &csize);
1386 +
1387 + int bytes = vmaw->outbuf_pos;
1388 + ret = vma_queue_write(vmaw, vmaw->outbuf, bytes);
1389 + if (ret != bytes) {
1390 + vma_writer_set_error(vmaw, "vma_writer_flush: failed write");
1391 + }
1392 +
1393 + vmaw->outbuf_count = 0;
1394 + vmaw->outbuf_pos = VMA_EXTENT_HEADER_SIZE;
1395 +
1396 + for (i = 0; i < VMA_BLOCKS_PER_EXTENT; i++) {
1397 + vmaw->outbuf_block_info[i] = 0;
1398 + }
1399 +
1400 + return vmaw->status;
1401 +}
1402 +
1403 +static int vma_count_open_streams(VmaWriter *vmaw)
1404 +{
1405 + g_assert(vmaw != NULL);
1406 +
1407 + int i;
1408 + int open_drives = 0;
1409 + for (i = 0; i <= 255; i++) {
1410 + if (vmaw->stream_info[i].size && !vmaw->stream_info[i].finished) {
1411 + open_drives++;
1412 + }
1413 + }
1414 +
1415 + return open_drives;
1416 +}
1417 +
1418 +
1419 +/**
1420 + * You need to call this if the vma archive does not contain
1421 + * any data stream.
1422 + */
1423 +int coroutine_fn
1424 +vma_writer_flush_output(VmaWriter *vmaw)
1425 +{
1426 + qemu_co_mutex_lock(&vmaw->flush_lock);
1427 + int ret = vma_writer_flush(vmaw);
1428 + qemu_co_mutex_unlock(&vmaw->flush_lock);
1429 + if (ret < 0) {
1430 + vma_writer_set_error(vmaw, "vma_writer_flush_header failed");
1431 + }
1432 + return ret;
1433 +}
1434 +
1435 +/**
1436 + * all jobs should call this when there is no more data
1437 + * Returns: number of remaining stream (0 ==> finished)
1438 + */
1439 +int coroutine_fn
1440 +vma_writer_close_stream(VmaWriter *vmaw, uint8_t dev_id)
1441 +{
1442 + g_assert(vmaw != NULL);
1443 +
1444 + DPRINTF("vma_writer_set_status %d\n", dev_id);
1445 + if (!vmaw->stream_info[dev_id].size) {
1446 + vma_writer_set_error(vmaw, "vma_writer_close_stream: "
1447 + "no such stream %d", dev_id);
1448 + return -1;
1449 + }
1450 + if (vmaw->stream_info[dev_id].finished) {
1451 + vma_writer_set_error(vmaw, "vma_writer_close_stream: "
1452 + "stream already closed %d", dev_id);
1453 + return -1;
1454 + }
1455 +
1456 + vmaw->stream_info[dev_id].finished = true;
1457 +
1458 + int open_drives = vma_count_open_streams(vmaw);
1459 +
1460 + if (open_drives <= 0) {
1461 + DPRINTF("vma_writer_set_status all drives completed\n");
1462 + vma_writer_flush_output(vmaw);
1463 + }
1464 +
1465 + return open_drives;
1466 +}
1467 +
1468 +int vma_writer_get_status(VmaWriter *vmaw, VmaStatus *status)
1469 +{
1470 + int i;
1471 +
1472 + g_assert(vmaw != NULL);
1473 +
1474 + if (status) {
1475 + status->status = vmaw->status;
1476 + g_strlcpy(status->errmsg, vmaw->errmsg, sizeof(status->errmsg));
1477 + for (i = 0; i <= 255; i++) {
1478 + status->stream_info[i] = vmaw->stream_info[i];
1479 + }
1480 +
1481 + uuid_unparse_lower(vmaw->uuid, status->uuid_str);
1482 + }
1483 +
1484 + status->closed = vmaw->closed;
1485 +
1486 + return vmaw->status;
1487 +}
1488 +
1489 +static int vma_writer_get_buffer(VmaWriter *vmaw)
1490 +{
1491 + int ret = 0;
1492 +
1493 + qemu_co_mutex_lock(&vmaw->flush_lock);
1494 +
1495 + /* wait until buffer is available */
1496 + while (vmaw->outbuf_count >= (VMA_BLOCKS_PER_EXTENT - 1)) {
1497 + ret = vma_writer_flush(vmaw);
1498 + if (ret < 0) {
1499 + vma_writer_set_error(vmaw, "vma_writer_get_buffer: flush failed");
1500 + break;
1501 + }
1502 + }
1503 +
1504 + qemu_co_mutex_unlock(&vmaw->flush_lock);
1505 +
1506 + return ret;
1507 +}
1508 +
1509 +
1510 +int64_t coroutine_fn
1511 +vma_writer_write(VmaWriter *vmaw, uint8_t dev_id, int64_t cluster_num,
1512 + const unsigned char *buf, size_t *zero_bytes)
1513 +{
1514 + g_assert(vmaw != NULL);
1515 + g_assert(zero_bytes != NULL);
1516 +
1517 + *zero_bytes = 0;
1518 +
1519 + if (vmaw->status < 0) {
1520 + return vmaw->status;
1521 + }
1522 +
1523 + if (!dev_id || !vmaw->stream_info[dev_id].size) {
1524 + vma_writer_set_error(vmaw, "vma_writer_write: "
1525 + "no such stream %d", dev_id);
1526 + return -1;
1527 + }
1528 +
1529 + if (vmaw->stream_info[dev_id].finished) {
1530 + vma_writer_set_error(vmaw, "vma_writer_write: "
1531 + "stream already closed %d", dev_id);
1532 + return -1;
1533 + }
1534 +
1535 +
1536 + if (cluster_num >= (((uint64_t)1)<<32)) {
1537 + vma_writer_set_error(vmaw, "vma_writer_write: "
1538 + "cluster number out of range");
1539 + return -1;
1540 + }
1541 +
1542 + if (dev_id == vmaw->vmstate_stream) {
1543 + if (cluster_num != vmaw->vmstate_clusters) {
1544 + vma_writer_set_error(vmaw, "vma_writer_write: "
1545 + "non sequential vmstate write");
1546 + }
1547 + vmaw->vmstate_clusters++;
1548 + } else if (cluster_num >= vmaw->stream_info[dev_id].cluster_count) {
1549 + vma_writer_set_error(vmaw, "vma_writer_write: cluster number too big");
1550 + return -1;
1551 + }
1552 +
1553 + /* wait until buffer is available */
1554 + if (vma_writer_get_buffer(vmaw) < 0) {
1555 + vma_writer_set_error(vmaw, "vma_writer_write: "
1556 + "vma_writer_get_buffer failed");
1557 + return -1;
1558 + }
1559 +
1560 + DPRINTF("VMA WRITE %d %zd\n", dev_id, cluster_num);
1561 +
1562 + uint64_t dev_size = vmaw->stream_info[dev_id].size;
1563 + uint16_t mask = 0;
1564 +
1565 + if (buf) {
1566 + int i;
1567 + int bit = 1;
1568 + uint64_t byte_offset = cluster_num * VMA_CLUSTER_SIZE;
1569 + for (i = 0; i < 16; i++) {
1570 + const unsigned char *vmablock = buf + (i*VMA_BLOCK_SIZE);
1571 +
1572 + // Note: If the source is not 64k-aligned, we might reach 4k blocks
1573 + // after the end of the device. Always mark these as zero in the
1574 + // mask, so the restore handles them correctly.
1575 + if (byte_offset < dev_size &&
1576 + !buffer_is_zero(vmablock, VMA_BLOCK_SIZE))
1577 + {
1578 + mask |= bit;
1579 + memcpy(vmaw->outbuf + vmaw->outbuf_pos, vmablock,
1580 + VMA_BLOCK_SIZE);
1581 +
1582 + // prevent memory leakage on unaligned last block
1583 + if (byte_offset + VMA_BLOCK_SIZE > dev_size) {
1584 + uint64_t real_data_in_block = dev_size - byte_offset;
1585 + memset(vmaw->outbuf + vmaw->outbuf_pos + real_data_in_block,
1586 + 0, VMA_BLOCK_SIZE - real_data_in_block);
1587 + }
1588 +
1589 + vmaw->outbuf_pos += VMA_BLOCK_SIZE;
1590 + } else {
1591 + DPRINTF("VMA WRITE %zd ZERO BLOCK %d\n", cluster_num, i);
1592 + vmaw->stream_info[dev_id].zero_bytes += VMA_BLOCK_SIZE;
1593 + *zero_bytes += VMA_BLOCK_SIZE;
1594 + }
1595 +
1596 + byte_offset += VMA_BLOCK_SIZE;
1597 + bit = bit << 1;
1598 + }
1599 + } else {
1600 + DPRINTF("VMA WRITE %zd ZERO CLUSTER\n", cluster_num);
1601 + vmaw->stream_info[dev_id].zero_bytes += VMA_CLUSTER_SIZE;
1602 + *zero_bytes += VMA_CLUSTER_SIZE;
1603 + }
1604 +
1605 + uint64_t block_info = ((uint64_t)mask) << (32+16);
1606 + block_info |= ((uint64_t)dev_id) << 32;
1607 + block_info |= (cluster_num & 0xffffffff);
1608 + vmaw->outbuf_block_info[vmaw->outbuf_count] = block_info;
1609 +
1610 + DPRINTF("VMA WRITE MASK %zd %zx\n", cluster_num, block_info);
1611 +
1612 + vmaw->outbuf_count++;
1613 +
1614 + /** NOTE: We allways write whole clusters, but we correctly set
1615 + * transferred bytes. So transferred == size when when everything
1616 + * went OK.
1617 + */
1618 + size_t transferred = VMA_CLUSTER_SIZE;
1619 +
1620 + if (dev_id != vmaw->vmstate_stream) {
1621 + uint64_t last = (cluster_num + 1) * VMA_CLUSTER_SIZE;
1622 + if (last > dev_size) {
1623 + uint64_t diff = last - dev_size;
1624 + if (diff >= VMA_CLUSTER_SIZE) {
1625 + vma_writer_set_error(vmaw, "vma_writer_write: "
1626 + "read after last cluster");
1627 + return -1;
1628 + }
1629 + transferred -= diff;
1630 + }
1631 + }
1632 +
1633 + vmaw->stream_info[dev_id].transferred += transferred;
1634 +
1635 + return transferred;
1636 +}
1637 +
1638 +void vma_writer_error_propagate(VmaWriter *vmaw, Error **errp)
1639 +{
1640 + if (vmaw->status < 0 && *errp == NULL) {
1641 + error_setg(errp, "%s", vmaw->errmsg);
1642 + }
1643 +}
1644 +
1645 +int vma_writer_close(VmaWriter *vmaw, Error **errp)
1646 +{
1647 + g_assert(vmaw != NULL);
1648 +
1649 + int i;
1650 +
1651 + qemu_co_mutex_lock(&vmaw->flush_lock); // wait for pending writes
1652 +
1653 + assert(vmaw->co_writer == NULL);
1654 +
1655 + if (vmaw->cmd) {
1656 + if (pclose(vmaw->cmd) < 0) {
1657 + vma_writer_set_error(vmaw, "vma_writer_close: "
1658 + "pclose failed - %s", g_strerror(errno));
1659 + }
1660 + } else {
1661 + if (close(vmaw->fd) < 0) {
1662 + vma_writer_set_error(vmaw, "vma_writer_close: "
1663 + "close failed - %s", g_strerror(errno));
1664 + }
1665 + }
1666 +
1667 + for (i = 0; i <= 255; i++) {
1668 + VmaStreamInfo *si = &vmaw->stream_info[i];
1669 + if (si->size) {
1670 + if (!si->finished) {
1671 + vma_writer_set_error(vmaw, "vma_writer_close: "
1672 + "detected open stream '%s'", si->devname);
1673 + } else if ((si->transferred != si->size) &&
1674 + (i != vmaw->vmstate_stream)) {
1675 + vma_writer_set_error(vmaw, "vma_writer_close: "
1676 + "incomplete stream '%s' (%zd != %zd)",
1677 + si->devname, si->transferred, si->size);
1678 + }
1679 + }
1680 + }
1681 +
1682 + for (i = 0; i <= 255; i++) {
1683 + vmaw->stream_info[i].finished = 1; /* mark as closed */
1684 + }
1685 +
1686 + vmaw->closed = 1;
1687 +
1688 + if (vmaw->status < 0 && *errp == NULL) {
1689 + error_setg(errp, "%s", vmaw->errmsg);
1690 + }
1691 +
1692 + qemu_co_mutex_unlock(&vmaw->flush_lock);
1693 +
1694 + return vmaw->status;
1695 +}
1696 +
1697 +void vma_writer_destroy(VmaWriter *vmaw)
1698 +{
1699 + assert(vmaw);
1700 +
1701 + int i;
1702 +
1703 + for (i = 0; i <= 255; i++) {
1704 + if (vmaw->stream_info[i].devname) {
1705 + g_free(vmaw->stream_info[i].devname);
1706 + }
1707 + }
1708 +
1709 + if (vmaw->md5csum) {
1710 + g_checksum_free(vmaw->md5csum);
1711 + }
1712 +
1713 + qemu_vfree(vmaw->headerbuf);
1714 + qemu_vfree(vmaw->outbuf);
1715 + g_free(vmaw);
1716 +}
1717 diff --git a/vma.c b/vma.c
1718 new file mode 100644
1719 index 0000000000..df542b7732
1720 --- /dev/null
1721 +++ b/vma.c
1722 @@ -0,0 +1,851 @@
1723 +/*
1724 + * VMA: Virtual Machine Archive
1725 + *
1726 + * Copyright (C) 2012-2013 Proxmox Server Solutions
1727 + *
1728 + * Authors:
1729 + * Dietmar Maurer (dietmar@proxmox.com)
1730 + *
1731 + * This work is licensed under the terms of the GNU GPL, version 2 or later.
1732 + * See the COPYING file in the top-level directory.
1733 + *
1734 + */
1735 +
1736 +#include "qemu/osdep.h"
1737 +#include <glib.h>
1738 +
1739 +#include "vma.h"
1740 +#include "qemu-common.h"
1741 +#include "qemu/module.h"
1742 +#include "qemu/error-report.h"
1743 +#include "qemu/main-loop.h"
1744 +#include "qemu/cutils.h"
1745 +#include "qapi/qmp/qdict.h"
1746 +#include "sysemu/block-backend.h"
1747 +
1748 +static void help(void)
1749 +{
1750 + const char *help_msg =
1751 + "usage: vma command [command options]\n"
1752 + "\n"
1753 + "vma list <filename>\n"
1754 + "vma config <filename> [-c config]\n"
1755 + "vma create <filename> [-c config] pathname ...\n"
1756 + "vma extract <filename> [-r <fifo>] <targetdir>\n"
1757 + "vma verify <filename> [-v]\n"
1758 + ;
1759 +
1760 + printf("%s", help_msg);
1761 + exit(1);
1762 +}
1763 +
1764 +static const char *extract_devname(const char *path, char **devname, int index)
1765 +{
1766 + assert(path);
1767 +
1768 + const char *sep = strchr(path, '=');
1769 +
1770 + if (sep) {
1771 + *devname = g_strndup(path, sep - path);
1772 + path = sep + 1;
1773 + } else {
1774 + if (index >= 0) {
1775 + *devname = g_strdup_printf("disk%d", index);
1776 + } else {
1777 + *devname = NULL;
1778 + }
1779 + }
1780 +
1781 + return path;
1782 +}
1783 +
1784 +static void print_content(VmaReader *vmar)
1785 +{
1786 + assert(vmar);
1787 +
1788 + VmaHeader *head = vma_reader_get_header(vmar);
1789 +
1790 + GList *l = vma_reader_get_config_data(vmar);
1791 + while (l && l->data) {
1792 + VmaConfigData *cdata = (VmaConfigData *)l->data;
1793 + l = g_list_next(l);
1794 + printf("CFG: size: %d name: %s\n", cdata->len, cdata->name);
1795 + }
1796 +
1797 + int i;
1798 + VmaDeviceInfo *di;
1799 + for (i = 1; i < 255; i++) {
1800 + di = vma_reader_get_device_info(vmar, i);
1801 + if (di) {
1802 + if (strcmp(di->devname, "vmstate") == 0) {
1803 + printf("VMSTATE: dev_id=%d memory: %zd\n", i, di->size);
1804 + } else {
1805 + printf("DEV: dev_id=%d size: %zd devname: %s\n",
1806 + i, di->size, di->devname);
1807 + }
1808 + }
1809 + }
1810 + /* ctime is the last entry we print */
1811 + printf("CTIME: %s", ctime(&head->ctime));
1812 + fflush(stdout);
1813 +}
1814 +
1815 +static int list_content(int argc, char **argv)
1816 +{
1817 + int c, ret = 0;
1818 + const char *filename;
1819 +
1820 + for (;;) {
1821 + c = getopt(argc, argv, "h");
1822 + if (c == -1) {
1823 + break;
1824 + }
1825 + switch (c) {
1826 + case '?':
1827 + case 'h':
1828 + help();
1829 + break;
1830 + default:
1831 + g_assert_not_reached();
1832 + }
1833 + }
1834 +
1835 + /* Get the filename */
1836 + if ((optind + 1) != argc) {
1837 + help();
1838 + }
1839 + filename = argv[optind++];
1840 +
1841 + Error *errp = NULL;
1842 + VmaReader *vmar = vma_reader_create(filename, &errp);
1843 +
1844 + if (!vmar) {
1845 + g_error("%s", error_get_pretty(errp));
1846 + }
1847 +
1848 + print_content(vmar);
1849 +
1850 + vma_reader_destroy(vmar);
1851 +
1852 + return ret;
1853 +}
1854 +
1855 +typedef struct RestoreMap {
1856 + char *devname;
1857 + char *path;
1858 + char *format;
1859 + uint64_t throttling_bps;
1860 + char *throttling_group;
1861 + char *cache;
1862 + bool write_zero;
1863 +} RestoreMap;
1864 +
1865 +static bool try_parse_option(char **line, const char *optname, char **out, const char *inbuf) {
1866 + size_t optlen = strlen(optname);
1867 + if (strncmp(*line, optname, optlen) != 0 || (*line)[optlen] != '=') {
1868 + return false;
1869 + }
1870 + if (*out) {
1871 + g_error("read map failed - duplicate value for option '%s'", optname);
1872 + }
1873 + char *value = (*line) + optlen + 1; /* including a '=' */
1874 + char *colon = strchr(value, ':');
1875 + if (!colon) {
1876 + g_error("read map failed - option '%s' not terminated ('%s')",
1877 + optname, inbuf);
1878 + }
1879 + *line = colon+1;
1880 + *out = g_strndup(value, colon - value);
1881 + return true;
1882 +}
1883 +
1884 +static uint64_t verify_u64(const char *text) {
1885 + uint64_t value;
1886 + const char *endptr = NULL;
1887 + if (qemu_strtou64(text, &endptr, 0, &value) != 0 || !endptr || *endptr) {
1888 + g_error("read map failed - not a number: %s", text);
1889 + }
1890 + return value;
1891 +}
1892 +
1893 +static int extract_content(int argc, char **argv)
1894 +{
1895 + int c, ret = 0;
1896 + int verbose = 0;
1897 + const char *filename;
1898 + const char *dirname;
1899 + const char *readmap = NULL;
1900 +
1901 + for (;;) {
1902 + c = getopt(argc, argv, "hvr:");
1903 + if (c == -1) {
1904 + break;
1905 + }
1906 + switch (c) {
1907 + case '?':
1908 + case 'h':
1909 + help();
1910 + break;
1911 + case 'r':
1912 + readmap = optarg;
1913 + break;
1914 + case 'v':
1915 + verbose = 1;
1916 + break;
1917 + default:
1918 + help();
1919 + }
1920 + }
1921 +
1922 + /* Get the filename */
1923 + if ((optind + 2) != argc) {
1924 + help();
1925 + }
1926 + filename = argv[optind++];
1927 + dirname = argv[optind++];
1928 +
1929 + Error *errp = NULL;
1930 + VmaReader *vmar = vma_reader_create(filename, &errp);
1931 +
1932 + if (!vmar) {
1933 + g_error("%s", error_get_pretty(errp));
1934 + }
1935 +
1936 + if (mkdir(dirname, 0777) < 0) {
1937 + g_error("unable to create target directory %s - %s",
1938 + dirname, g_strerror(errno));
1939 + }
1940 +
1941 + GList *l = vma_reader_get_config_data(vmar);
1942 + while (l && l->data) {
1943 + VmaConfigData *cdata = (VmaConfigData *)l->data;
1944 + l = g_list_next(l);
1945 + char *cfgfn = g_strdup_printf("%s/%s", dirname, cdata->name);
1946 + GError *err = NULL;
1947 + if (!g_file_set_contents(cfgfn, (gchar *)cdata->data, cdata->len,
1948 + &err)) {
1949 + g_error("unable to write file: %s", err->message);
1950 + }
1951 + }
1952 +
1953 + GHashTable *devmap = g_hash_table_new(g_str_hash, g_str_equal);
1954 +
1955 + if (readmap) {
1956 + print_content(vmar);
1957 +
1958 + FILE *map = fopen(readmap, "r");
1959 + if (!map) {
1960 + g_error("unable to open fifo %s - %s", readmap, g_strerror(errno));
1961 + }
1962 +
1963 + while (1) {
1964 + char inbuf[8192];
1965 + char *line = fgets(inbuf, sizeof(inbuf), map);
1966 + char *format = NULL;
1967 + char *bps = NULL;
1968 + char *group = NULL;
1969 + char *cache = NULL;
1970 + if (!line || line[0] == '\0' || !strcmp(line, "done\n")) {
1971 + break;
1972 + }
1973 + int len = strlen(line);
1974 + if (line[len - 1] == '\n') {
1975 + line[len - 1] = '\0';
1976 + if (len == 1) {
1977 + break;
1978 + }
1979 + }
1980 +
1981 + while (1) {
1982 + if (!try_parse_option(&line, "format", &format, inbuf) &&
1983 + !try_parse_option(&line, "throttling.bps", &bps, inbuf) &&
1984 + !try_parse_option(&line, "throttling.group", &group, inbuf) &&
1985 + !try_parse_option(&line, "cache", &cache, inbuf))
1986 + {
1987 + break;
1988 + }
1989 + }
1990 +
1991 + uint64_t bps_value = 0;
1992 + if (bps) {
1993 + bps_value = verify_u64(bps);
1994 + g_free(bps);
1995 + }
1996 +
1997 + const char *path;
1998 + bool write_zero;
1999 + if (line[0] == '0' && line[1] == ':') {
2000 + path = line + 2;
2001 + write_zero = false;
2002 + } else if (line[0] == '1' && line[1] == ':') {
2003 + path = line + 2;
2004 + write_zero = true;
2005 + } else {
2006 + g_error("read map failed - parse error ('%s')", inbuf);
2007 + }
2008 +
2009 + char *devname = NULL;
2010 + path = extract_devname(path, &devname, -1);
2011 + if (!devname) {
2012 + g_error("read map failed - no dev name specified ('%s')",
2013 + inbuf);
2014 + }
2015 +
2016 + RestoreMap *map = g_new0(RestoreMap, 1);
2017 + map->devname = g_strdup(devname);
2018 + map->path = g_strdup(path);
2019 + map->format = format;
2020 + map->throttling_bps = bps_value;
2021 + map->throttling_group = group;
2022 + map->cache = cache;
2023 + map->write_zero = write_zero;
2024 +
2025 + g_hash_table_insert(devmap, map->devname, map);
2026 +
2027 + };
2028 + }
2029 +
2030 + int i;
2031 + int vmstate_fd = -1;
2032 + guint8 vmstate_stream = 0;
2033 +
2034 + BlockBackend *blk = NULL;
2035 +
2036 + for (i = 1; i < 255; i++) {
2037 + VmaDeviceInfo *di = vma_reader_get_device_info(vmar, i);
2038 + if (di && (strcmp(di->devname, "vmstate") == 0)) {
2039 + vmstate_stream = i;
2040 + char *statefn = g_strdup_printf("%s/vmstate.bin", dirname);
2041 + vmstate_fd = open(statefn, O_WRONLY|O_CREAT|O_EXCL, 0644);
2042 + if (vmstate_fd < 0) {
2043 + g_error("create vmstate file '%s' failed - %s", statefn,
2044 + g_strerror(errno));
2045 + }
2046 + g_free(statefn);
2047 + } else if (di) {
2048 + char *devfn = NULL;
2049 + const char *format = NULL;
2050 + uint64_t throttling_bps = 0;
2051 + const char *throttling_group = NULL;
2052 + const char *cache = NULL;
2053 + int flags = BDRV_O_RDWR;
2054 + bool write_zero = true;
2055 +
2056 + if (readmap) {
2057 + RestoreMap *map;
2058 + map = (RestoreMap *)g_hash_table_lookup(devmap, di->devname);
2059 + if (map == NULL) {
2060 + g_error("no device name mapping for %s", di->devname);
2061 + }
2062 + devfn = map->path;
2063 + format = map->format;
2064 + throttling_bps = map->throttling_bps;
2065 + throttling_group = map->throttling_group;
2066 + cache = map->cache;
2067 + write_zero = map->write_zero;
2068 + } else {
2069 + devfn = g_strdup_printf("%s/tmp-disk-%s.raw",
2070 + dirname, di->devname);
2071 + printf("DEVINFO %s %zd\n", devfn, di->size);
2072 +
2073 + bdrv_img_create(devfn, "raw", NULL, NULL, NULL, di->size,
2074 + flags, true, &errp);
2075 + if (errp) {
2076 + g_error("can't create file %s: %s", devfn,
2077 + error_get_pretty(errp));
2078 + }
2079 +
2080 + /* Note: we created an empty file above, so there is no
2081 + * need to write zeroes (so we generate a sparse file)
2082 + */
2083 + write_zero = false;
2084 + }
2085 +
2086 + size_t devlen = strlen(devfn);
2087 + QDict *options = NULL;
2088 + bool writethrough;
2089 + if (format) {
2090 + /* explicit format from commandline */
2091 + options = qdict_new();
2092 + qdict_put_str(options, "driver", format);
2093 + } else if ((devlen > 4 && strcmp(devfn+devlen-4, ".raw") == 0) ||
2094 + strncmp(devfn, "/dev/", 5) == 0)
2095 + {
2096 + /* This part is now deprecated for PVE as well (just as qemu
2097 + * deprecated not specifying an explicit raw format, too.
2098 + */
2099 + /* explicit raw format */
2100 + options = qdict_new();
2101 + qdict_put_str(options, "driver", "raw");
2102 + }
2103 + if (cache && bdrv_parse_cache_mode(cache, &flags, &writethrough)) {
2104 + g_error("invalid cache option: %s\n", cache);
2105 + }
2106 +
2107 + if (errp || !(blk = blk_new_open(devfn, NULL, options, flags, &errp))) {
2108 + g_error("can't open file %s - %s", devfn,
2109 + error_get_pretty(errp));
2110 + }
2111 +
2112 + if (cache) {
2113 + blk_set_enable_write_cache(blk, !writethrough);
2114 + }
2115 +
2116 + if (throttling_group) {
2117 + blk_io_limits_enable(blk, throttling_group);
2118 + }
2119 +
2120 + if (throttling_bps) {
2121 + if (!throttling_group) {
2122 + blk_io_limits_enable(blk, devfn);
2123 + }
2124 +
2125 + ThrottleConfig cfg;
2126 + throttle_config_init(&cfg);
2127 + cfg.buckets[THROTTLE_BPS_WRITE].avg = throttling_bps;
2128 + Error *err = NULL;
2129 + if (!throttle_is_valid(&cfg, &err)) {
2130 + error_report_err(err);
2131 + g_error("failed to apply throttling");
2132 + }
2133 + blk_set_io_limits(blk, &cfg);
2134 + }
2135 +
2136 + if (vma_reader_register_bs(vmar, i, blk, write_zero, &errp) < 0) {
2137 + g_error("%s", error_get_pretty(errp));
2138 + }
2139 +
2140 + if (!readmap) {
2141 + g_free(devfn);
2142 + }
2143 + }
2144 + }
2145 +
2146 + if (vma_reader_restore(vmar, vmstate_fd, verbose, &errp) < 0) {
2147 + g_error("restore failed - %s", error_get_pretty(errp));
2148 + }
2149 +
2150 + if (!readmap) {
2151 + for (i = 1; i < 255; i++) {
2152 + VmaDeviceInfo *di = vma_reader_get_device_info(vmar, i);
2153 + if (di && (i != vmstate_stream)) {
2154 + char *tmpfn = g_strdup_printf("%s/tmp-disk-%s.raw",
2155 + dirname, di->devname);
2156 + char *fn = g_strdup_printf("%s/disk-%s.raw",
2157 + dirname, di->devname);
2158 + if (rename(tmpfn, fn) != 0) {
2159 + g_error("rename %s to %s failed - %s",
2160 + tmpfn, fn, g_strerror(errno));
2161 + }
2162 + }
2163 + }
2164 + }
2165 +
2166 + vma_reader_destroy(vmar);
2167 +
2168 + blk_unref(blk);
2169 +
2170 + bdrv_close_all();
2171 +
2172 + return ret;
2173 +}
2174 +
2175 +static int verify_content(int argc, char **argv)
2176 +{
2177 + int c, ret = 0;
2178 + int verbose = 0;
2179 + const char *filename;
2180 +
2181 + for (;;) {
2182 + c = getopt(argc, argv, "hv");
2183 + if (c == -1) {
2184 + break;
2185 + }
2186 + switch (c) {
2187 + case '?':
2188 + case 'h':
2189 + help();
2190 + break;
2191 + case 'v':
2192 + verbose = 1;
2193 + break;
2194 + default:
2195 + help();
2196 + }
2197 + }
2198 +
2199 + /* Get the filename */
2200 + if ((optind + 1) != argc) {
2201 + help();
2202 + }
2203 + filename = argv[optind++];
2204 +
2205 + Error *errp = NULL;
2206 + VmaReader *vmar = vma_reader_create(filename, &errp);
2207 +
2208 + if (!vmar) {
2209 + g_error("%s", error_get_pretty(errp));
2210 + }
2211 +
2212 + if (verbose) {
2213 + print_content(vmar);
2214 + }
2215 +
2216 + if (vma_reader_verify(vmar, verbose, &errp) < 0) {
2217 + g_error("verify failed - %s", error_get_pretty(errp));
2218 + }
2219 +
2220 + vma_reader_destroy(vmar);
2221 +
2222 + bdrv_close_all();
2223 +
2224 + return ret;
2225 +}
2226 +
2227 +typedef struct BackupJob {
2228 + BlockBackend *target;
2229 + int64_t len;
2230 + VmaWriter *vmaw;
2231 + uint8_t dev_id;
2232 +} BackupJob;
2233 +
2234 +#define BACKUP_SECTORS_PER_CLUSTER (VMA_CLUSTER_SIZE / BDRV_SECTOR_SIZE)
2235 +
2236 +static void coroutine_fn backup_run_empty(void *opaque)
2237 +{
2238 + VmaWriter *vmaw = (VmaWriter *)opaque;
2239 +
2240 + vma_writer_flush_output(vmaw);
2241 +
2242 + Error *err = NULL;
2243 + if (vma_writer_close(vmaw, &err) != 0) {
2244 + g_warning("vma_writer_close failed %s", error_get_pretty(err));
2245 + }
2246 +}
2247 +
2248 +static void coroutine_fn backup_run(void *opaque)
2249 +{
2250 + BackupJob *job = (BackupJob *)opaque;
2251 + struct iovec iov;
2252 + QEMUIOVector qiov;
2253 +
2254 + int64_t start, end;
2255 + int ret = 0;
2256 +
2257 + unsigned char *buf = blk_blockalign(job->target, VMA_CLUSTER_SIZE);
2258 +
2259 + start = 0;
2260 + end = DIV_ROUND_UP(job->len / BDRV_SECTOR_SIZE,
2261 + BACKUP_SECTORS_PER_CLUSTER);
2262 +
2263 + for (; start < end; start++) {
2264 + iov.iov_base = buf;
2265 + iov.iov_len = VMA_CLUSTER_SIZE;
2266 + qemu_iovec_init_external(&qiov, &iov, 1);
2267 +
2268 + ret = blk_co_preadv(job->target, start * VMA_CLUSTER_SIZE,
2269 + VMA_CLUSTER_SIZE, &qiov, 0);
2270 + if (ret < 0) {
2271 + vma_writer_set_error(job->vmaw, "read error", -1);
2272 + goto out;
2273 + }
2274 +
2275 + size_t zb = 0;
2276 + if (vma_writer_write(job->vmaw, job->dev_id, start, buf, &zb) < 0) {
2277 + vma_writer_set_error(job->vmaw, "backup_dump_cb vma_writer_write failed", -1);
2278 + goto out;
2279 + }
2280 + }
2281 +
2282 +
2283 +out:
2284 + if (vma_writer_close_stream(job->vmaw, job->dev_id) <= 0) {
2285 + Error *err = NULL;
2286 + if (vma_writer_close(job->vmaw, &err) != 0) {
2287 + g_warning("vma_writer_close failed %s", error_get_pretty(err));
2288 + }
2289 + }
2290 + qemu_vfree(buf);
2291 +}
2292 +
2293 +static int create_archive(int argc, char **argv)
2294 +{
2295 + int i, c;
2296 + int verbose = 0;
2297 + const char *archivename;
2298 + GList *backup_coroutines = NULL;
2299 + GList *config_files = NULL;
2300 +
2301 + for (;;) {
2302 + c = getopt(argc, argv, "hvc:");
2303 + if (c == -1) {
2304 + break;
2305 + }
2306 + switch (c) {
2307 + case '?':
2308 + case 'h':
2309 + help();
2310 + break;
2311 + case 'c':
2312 + config_files = g_list_append(config_files, optarg);
2313 + break;
2314 + case 'v':
2315 + verbose = 1;
2316 + break;
2317 + default:
2318 + g_assert_not_reached();
2319 + }
2320 + }
2321 +
2322 +
2323 + /* make sure we an archive name */
2324 + if ((optind + 1) > argc) {
2325 + help();
2326 + }
2327 +
2328 + archivename = argv[optind++];
2329 +
2330 + uuid_t uuid;
2331 + uuid_generate(uuid);
2332 +
2333 + Error *local_err = NULL;
2334 + VmaWriter *vmaw = vma_writer_create(archivename, uuid, &local_err);
2335 +
2336 + if (vmaw == NULL) {
2337 + g_error("%s", error_get_pretty(local_err));
2338 + }
2339 +
2340 + GList *l = config_files;
2341 + while (l && l->data) {
2342 + char *name = l->data;
2343 + char *cdata = NULL;
2344 + gsize clen = 0;
2345 + GError *err = NULL;
2346 + if (!g_file_get_contents(name, &cdata, &clen, &err)) {
2347 + unlink(archivename);
2348 + g_error("Unable to read file: %s", err->message);
2349 + }
2350 +
2351 + if (vma_writer_add_config(vmaw, name, cdata, clen) != 0) {
2352 + unlink(archivename);
2353 + g_error("Unable to append config data %s (len = %zd)",
2354 + name, clen);
2355 + }
2356 + l = g_list_next(l);
2357 + }
2358 +
2359 + int devcount = 0;
2360 + while (optind < argc) {
2361 + const char *path = argv[optind++];
2362 + char *devname = NULL;
2363 + path = extract_devname(path, &devname, devcount++);
2364 +
2365 + Error *errp = NULL;
2366 + BlockBackend *target;
2367 +
2368 + target = blk_new_open(path, NULL, NULL, 0, &errp);
2369 + if (!target) {
2370 + unlink(archivename);
2371 + g_error("bdrv_open '%s' failed - %s", path, error_get_pretty(errp));
2372 + }
2373 + int64_t size = blk_getlength(target);
2374 + int dev_id = vma_writer_register_stream(vmaw, devname, size);
2375 + if (dev_id <= 0) {
2376 + unlink(archivename);
2377 + g_error("vma_writer_register_stream '%s' failed", devname);
2378 + }
2379 +
2380 + BackupJob *job = g_new0(BackupJob, 1);
2381 + job->len = size;
2382 + job->target = target;
2383 + job->vmaw = vmaw;
2384 + job->dev_id = dev_id;
2385 +
2386 + Coroutine *co = qemu_coroutine_create(backup_run, job);
2387 + // Don't enter coroutine yet, because it might write the header before
2388 + // all streams can be registered.
2389 + backup_coroutines = g_list_append(backup_coroutines, co);
2390 + }
2391 +
2392 + VmaStatus vmastat;
2393 + int percent = 0;
2394 + int last_percent = -1;
2395 +
2396 + if (devcount) {
2397 + GList *entry = backup_coroutines;
2398 + while (entry && entry->data) {
2399 + Coroutine *co = entry->data;
2400 + qemu_coroutine_enter(co);
2401 + entry = g_list_next(entry);
2402 + }
2403 +
2404 + while (1) {
2405 + main_loop_wait(false);
2406 + vma_writer_get_status(vmaw, &vmastat);
2407 +
2408 + if (verbose) {
2409 +
2410 + uint64_t total = 0;
2411 + uint64_t transferred = 0;
2412 + uint64_t zero_bytes = 0;
2413 +
2414 + int i;
2415 + for (i = 0; i < 256; i++) {
2416 + if (vmastat.stream_info[i].size) {
2417 + total += vmastat.stream_info[i].size;
2418 + transferred += vmastat.stream_info[i].transferred;
2419 + zero_bytes += vmastat.stream_info[i].zero_bytes;
2420 + }
2421 + }
2422 + percent = (transferred*100)/total;
2423 + if (percent != last_percent) {
2424 + fprintf(stderr, "progress %d%% %zd/%zd %zd\n", percent,
2425 + transferred, total, zero_bytes);
2426 + fflush(stderr);
2427 +
2428 + last_percent = percent;
2429 + }
2430 + }
2431 +
2432 + if (vmastat.closed) {
2433 + break;
2434 + }
2435 + }
2436 + } else {
2437 + Coroutine *co = qemu_coroutine_create(backup_run_empty, vmaw);
2438 + qemu_coroutine_enter(co);
2439 + while (1) {
2440 + main_loop_wait(false);
2441 + vma_writer_get_status(vmaw, &vmastat);
2442 + if (vmastat.closed) {
2443 + break;
2444 + }
2445 + }
2446 + }
2447 +
2448 + bdrv_drain_all();
2449 +
2450 + vma_writer_get_status(vmaw, &vmastat);
2451 +
2452 + if (verbose) {
2453 + for (i = 0; i < 256; i++) {
2454 + VmaStreamInfo *si = &vmastat.stream_info[i];
2455 + if (si->size) {
2456 + fprintf(stderr, "image %s: size=%zd zeros=%zd saved=%zd\n",
2457 + si->devname, si->size, si->zero_bytes,
2458 + si->size - si->zero_bytes);
2459 + }
2460 + }
2461 + }
2462 +
2463 + if (vmastat.status < 0) {
2464 + unlink(archivename);
2465 + g_error("creating vma archive failed");
2466 + }
2467 +
2468 + g_list_free(backup_coroutines);
2469 + g_list_free(config_files);
2470 + vma_writer_destroy(vmaw);
2471 + return 0;
2472 +}
2473 +
2474 +static int dump_config(int argc, char **argv)
2475 +{
2476 + int c, ret = 0;
2477 + const char *filename;
2478 + const char *config_name = "qemu-server.conf";
2479 +
2480 + for (;;) {
2481 + c = getopt(argc, argv, "hc:");
2482 + if (c == -1) {
2483 + break;
2484 + }
2485 + switch (c) {
2486 + case '?':
2487 + case 'h':
2488 + help();
2489 + break;
2490 + case 'c':
2491 + config_name = optarg;
2492 + break;
2493 + default:
2494 + help();
2495 + }
2496 + }
2497 +
2498 + /* Get the filename */
2499 + if ((optind + 1) != argc) {
2500 + help();
2501 + }
2502 + filename = argv[optind++];
2503 +
2504 + Error *errp = NULL;
2505 + VmaReader *vmar = vma_reader_create(filename, &errp);
2506 +
2507 + if (!vmar) {
2508 + g_error("%s", error_get_pretty(errp));
2509 + }
2510 +
2511 + int found = 0;
2512 + GList *l = vma_reader_get_config_data(vmar);
2513 + while (l && l->data) {
2514 + VmaConfigData *cdata = (VmaConfigData *)l->data;
2515 + l = g_list_next(l);
2516 + if (strcmp(cdata->name, config_name) == 0) {
2517 + found = 1;
2518 + fwrite(cdata->data, cdata->len, 1, stdout);
2519 + break;
2520 + }
2521 + }
2522 +
2523 + vma_reader_destroy(vmar);
2524 +
2525 + bdrv_close_all();
2526 +
2527 + if (!found) {
2528 + fprintf(stderr, "unable to find configuration data '%s'\n", config_name);
2529 + return -1;
2530 + }
2531 +
2532 + return ret;
2533 +}
2534 +
2535 +int main(int argc, char **argv)
2536 +{
2537 + const char *cmdname;
2538 + Error *main_loop_err = NULL;
2539 +
2540 + error_init(argv[0]);
2541 + module_call_init(MODULE_INIT_TRACE);
2542 + qemu_init_exec_dir(argv[0]);
2543 +
2544 + if (qemu_init_main_loop(&main_loop_err)) {
2545 + g_error("%s", error_get_pretty(main_loop_err));
2546 + }
2547 +
2548 + bdrv_init();
2549 + module_call_init(MODULE_INIT_QOM);
2550 +
2551 + if (argc < 2) {
2552 + help();
2553 + }
2554 +
2555 + cmdname = argv[1];
2556 + argc--; argv++;
2557 +
2558 +
2559 + if (!strcmp(cmdname, "list")) {
2560 + return list_content(argc, argv);
2561 + } else if (!strcmp(cmdname, "create")) {
2562 + return create_archive(argc, argv);
2563 + } else if (!strcmp(cmdname, "extract")) {
2564 + return extract_content(argc, argv);
2565 + } else if (!strcmp(cmdname, "verify")) {
2566 + return verify_content(argc, argv);
2567 + } else if (!strcmp(cmdname, "config")) {
2568 + return dump_config(argc, argv);
2569 + }
2570 +
2571 + help();
2572 + return 0;
2573 +}
2574 diff --git a/vma.h b/vma.h
2575 new file mode 100644
2576 index 0000000000..c895c97f6d
2577 --- /dev/null
2578 +++ b/vma.h
2579 @@ -0,0 +1,150 @@
2580 +/*
2581 + * VMA: Virtual Machine Archive
2582 + *
2583 + * Copyright (C) Proxmox Server Solutions
2584 + *
2585 + * Authors:
2586 + * Dietmar Maurer (dietmar@proxmox.com)
2587 + *
2588 + * This work is licensed under the terms of the GNU GPL, version 2 or later.
2589 + * See the COPYING file in the top-level directory.
2590 + *
2591 + */
2592 +
2593 +#ifndef BACKUP_VMA_H
2594 +#define BACKUP_VMA_H
2595 +
2596 +#include <uuid/uuid.h>
2597 +#include "qapi/error.h"
2598 +#include "block/block.h"
2599 +
2600 +#define VMA_BLOCK_BITS 12
2601 +#define VMA_BLOCK_SIZE (1<<VMA_BLOCK_BITS)
2602 +#define VMA_CLUSTER_BITS (VMA_BLOCK_BITS+4)
2603 +#define VMA_CLUSTER_SIZE (1<<VMA_CLUSTER_BITS)
2604 +
2605 +#if VMA_CLUSTER_SIZE != 65536
2606 +#error unexpected cluster size
2607 +#endif
2608 +
2609 +#define VMA_EXTENT_HEADER_SIZE 512
2610 +#define VMA_BLOCKS_PER_EXTENT 59
2611 +#define VMA_MAX_CONFIGS 256
2612 +
2613 +#define VMA_MAX_EXTENT_SIZE \
2614 + (VMA_EXTENT_HEADER_SIZE+VMA_CLUSTER_SIZE*VMA_BLOCKS_PER_EXTENT)
2615 +#if VMA_MAX_EXTENT_SIZE != 3867136
2616 +#error unexpected VMA_EXTENT_SIZE
2617 +#endif
2618 +
2619 +/* File Format Definitions */
2620 +
2621 +#define VMA_MAGIC (GUINT32_TO_BE(('V'<<24)|('M'<<16)|('A'<<8)|0x00))
2622 +#define VMA_EXTENT_MAGIC (GUINT32_TO_BE(('V'<<24)|('M'<<16)|('A'<<8)|'E'))
2623 +
2624 +typedef struct VmaDeviceInfoHeader {
2625 + uint32_t devname_ptr; /* offset into blob_buffer table */
2626 + uint32_t reserved0;
2627 + uint64_t size; /* device size in bytes */
2628 + uint64_t reserved1;
2629 + uint64_t reserved2;
2630 +} VmaDeviceInfoHeader;
2631 +
2632 +typedef struct VmaHeader {
2633 + uint32_t magic;
2634 + uint32_t version;
2635 + unsigned char uuid[16];
2636 + int64_t ctime;
2637 + unsigned char md5sum[16];
2638 +
2639 + uint32_t blob_buffer_offset;
2640 + uint32_t blob_buffer_size;
2641 + uint32_t header_size;
2642 +
2643 + unsigned char reserved[1984];
2644 +
2645 + uint32_t config_names[VMA_MAX_CONFIGS]; /* offset into blob_buffer table */
2646 + uint32_t config_data[VMA_MAX_CONFIGS]; /* offset into blob_buffer table */
2647 +
2648 + uint32_t reserved1;
2649 +
2650 + VmaDeviceInfoHeader dev_info[256];
2651 +} VmaHeader;
2652 +
2653 +typedef struct VmaExtentHeader {
2654 + uint32_t magic;
2655 + uint16_t reserved1;
2656 + uint16_t block_count;
2657 + unsigned char uuid[16];
2658 + unsigned char md5sum[16];
2659 + uint64_t blockinfo[VMA_BLOCKS_PER_EXTENT];
2660 +} VmaExtentHeader;
2661 +
2662 +/* functions/definitions to read/write vma files */
2663 +
2664 +typedef struct VmaReader VmaReader;
2665 +
2666 +typedef struct VmaWriter VmaWriter;
2667 +
2668 +typedef struct VmaConfigData {
2669 + const char *name;
2670 + const void *data;
2671 + uint32_t len;
2672 +} VmaConfigData;
2673 +
2674 +typedef struct VmaStreamInfo {
2675 + uint64_t size;
2676 + uint64_t cluster_count;
2677 + uint64_t transferred;
2678 + uint64_t zero_bytes;
2679 + int finished;
2680 + char *devname;
2681 +} VmaStreamInfo;
2682 +
2683 +typedef struct VmaStatus {
2684 + int status;
2685 + bool closed;
2686 + char errmsg[8192];
2687 + char uuid_str[37];
2688 + VmaStreamInfo stream_info[256];
2689 +} VmaStatus;
2690 +
2691 +typedef struct VmaDeviceInfo {
2692 + uint64_t size; /* device size in bytes */
2693 + const char *devname;
2694 +} VmaDeviceInfo;
2695 +
2696 +VmaWriter *vma_writer_create(const char *filename, uuid_t uuid, Error **errp);
2697 +int vma_writer_close(VmaWriter *vmaw, Error **errp);
2698 +void vma_writer_error_propagate(VmaWriter *vmaw, Error **errp);
2699 +void vma_writer_destroy(VmaWriter *vmaw);
2700 +int vma_writer_add_config(VmaWriter *vmaw, const char *name, gpointer data,
2701 + size_t len);
2702 +int vma_writer_register_stream(VmaWriter *vmaw, const char *devname,
2703 + size_t size);
2704 +
2705 +int64_t coroutine_fn vma_writer_write(VmaWriter *vmaw, uint8_t dev_id,
2706 + int64_t cluster_num,
2707 + const unsigned char *buf,
2708 + size_t *zero_bytes);
2709 +
2710 +int coroutine_fn vma_writer_close_stream(VmaWriter *vmaw, uint8_t dev_id);
2711 +int coroutine_fn vma_writer_flush_output(VmaWriter *vmaw);
2712 +
2713 +int vma_writer_get_status(VmaWriter *vmaw, VmaStatus *status);
2714 +void vma_writer_set_error(VmaWriter *vmaw, const char *fmt, ...);
2715 +
2716 +
2717 +VmaReader *vma_reader_create(const char *filename, Error **errp);
2718 +void vma_reader_destroy(VmaReader *vmar);
2719 +VmaHeader *vma_reader_get_header(VmaReader *vmar);
2720 +GList *vma_reader_get_config_data(VmaReader *vmar);
2721 +VmaDeviceInfo *vma_reader_get_device_info(VmaReader *vmar, guint8 dev_id);
2722 +int vma_reader_register_bs(VmaReader *vmar, guint8 dev_id,
2723 + BlockBackend *target, bool write_zeroes,
2724 + Error **errp);
2725 +int vma_reader_restore(VmaReader *vmar, int vmstate_fd, bool verbose,
2726 + Error **errp);
2727 +int vma_reader_verify(VmaReader *vmar, bool verbose, Error **errp);
2728 +
2729 +#endif /* BACKUP_VMA_H */
QEMU for PVE