]> git.proxmox.com Git - pve-qemu.git/blob - debian/patches/pve/0028-adding-old-vma-files.patch
projects / pve-qemu.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
fix #1420: fix stop mode backup with virtio-blk
[pve-qemu.git] / debian / patches / pve / 0028-adding-old-vma-files.patch
1 From 741b9f52069e043e93f0fc47e820ddbfa0bff7a3 Mon Sep 17 00:00:00 2001
2 From: Wolfgang Bumiller <w.bumiller@proxmox.com>
3 Date: Mon, 7 Aug 2017 08:51:16 +0200
4 Subject: [PATCH 28/28] adding old vma files
5
6 ---
7 Makefile | 3 +-
8 Makefile.objs | 1 +
9 block/backup.c | 128 ++++---
10 block/replication.c | 1 +
11 blockdev.c | 250 +++++++++-----
12 blockjob.c | 11 +-
13 include/block/block_int.h | 4 +
14 vma-reader.c | 857 ++++++++++++++++++++++++++++++++++++++++++++++
15 vma-writer.c | 771 +++++++++++++++++++++++++++++++++++++++++
16 vma.c | 757 ++++++++++++++++++++++++++++++++++++++++
17 vma.h | 149 ++++++++
18 11 files changed, 2799 insertions(+), 133 deletions(-)
19 create mode 100644 vma-reader.c
20 create mode 100644 vma-writer.c
21 create mode 100644 vma.c
22 create mode 100644 vma.h
23
24 diff --git a/Makefile b/Makefile
25 index 6c359b2f86..edbc8b50f0 100644
26 --- a/Makefile
27 +++ b/Makefile
28 @@ -284,7 +284,7 @@ ifneq ($(wildcard config-host.mak),)
29 include $(SRC_PATH)/tests/Makefile.include
30 endif
31
32 -all: $(DOCS) $(TOOLS) $(HELPERS-y) recurse-all modules
33 +all: $(DOCS) $(TOOLS) vma$(EXESUF) $(HELPERS-y) recurse-all modules
34
35 qemu-version.h: FORCE
36 $(call quiet-command, \
37 @@ -377,6 +377,7 @@ qemu-img.o: qemu-img-cmds.h
38 qemu-img$(EXESUF): qemu-img.o $(block-obj-y) $(crypto-obj-y) $(io-obj-y) $(qom-obj-y) $(COMMON_LDADDS)
39 qemu-nbd$(EXESUF): qemu-nbd.o $(block-obj-y) $(crypto-obj-y) $(io-obj-y) $(qom-obj-y) $(COMMON_LDADDS)
40 qemu-io$(EXESUF): qemu-io.o $(block-obj-y) $(crypto-obj-y) $(io-obj-y) $(qom-obj-y) $(COMMON_LDADDS)
41 +vma$(EXESUF): vma.o vma-reader.o $(block-obj-y) $(crypto-obj-y) $(io-obj-y) $(qom-obj-y) $(COMMON_LDADDS)
42
43 qemu-bridge-helper$(EXESUF): qemu-bridge-helper.o $(COMMON_LDADDS)
44
45 diff --git a/Makefile.objs b/Makefile.objs
46 index fbfbbb7f70..f5f8dbab3b 100644
47 --- a/Makefile.objs
48 +++ b/Makefile.objs
49 @@ -14,6 +14,7 @@ block-obj-y += block.o blockjob.o
50 block-obj-y += block/
51 block-obj-y += qemu-io-cmds.o
52 block-obj-$(CONFIG_REPLICATION) += replication.o
53 +block-obj-y += vma-writer.o
54
55 block-obj-m = block/
56
57 diff --git a/block/backup.c b/block/backup.c
58 index 1ede70c061..51b5ba6eda 100644
59 --- a/block/backup.c
60 +++ b/block/backup.c
61 @@ -36,6 +36,7 @@ typedef struct BackupBlockJob {
62 BdrvDirtyBitmap *sync_bitmap;
63 MirrorSyncMode sync_mode;
64 RateLimit limit;
65 + BackupDumpFunc *dump_cb;
66 BlockdevOnError on_source_error;
67 BlockdevOnError on_target_error;
68 CoRwlock flush_rwlock;
69 @@ -145,13 +146,24 @@ static int coroutine_fn backup_do_cow(BackupBlockJob *job,
70 goto out;
71 }
72
73 + int64_t start_sec = start * sectors_per_cluster;
74 if (buffer_is_zero(iov.iov_base, iov.iov_len)) {
75 - ret = blk_co_pwrite_zeroes(job->target, start * job->cluster_size,
76 - bounce_qiov.size, BDRV_REQ_MAY_UNMAP);
77 + if (job->dump_cb) {
78 + ret = job->dump_cb(job->common.opaque, job->target, start_sec, n, NULL);
79 + }
80 + if (job->target) {
81 + ret = blk_co_pwrite_zeroes(job->target, start * job->cluster_size,
82 + bounce_qiov.size, BDRV_REQ_MAY_UNMAP);
83 + }
84 } else {
85 - ret = blk_co_pwritev(job->target, start * job->cluster_size,
86 - bounce_qiov.size, &bounce_qiov,
87 - job->compress ? BDRV_REQ_WRITE_COMPRESSED : 0);
88 + if (job->dump_cb) {
89 + ret = job->dump_cb(job->common.opaque, job->target, start_sec, n, bounce_buffer);
90 + }
91 + if (job->target) {
92 + ret = blk_co_pwritev(job->target, start * job->cluster_size,
93 + bounce_qiov.size, &bounce_qiov,
94 + job->compress ? BDRV_REQ_WRITE_COMPRESSED : 0);
95 + }
96 }
97 if (ret < 0) {
98 trace_backup_do_cow_write_fail(job, start, ret);
99 @@ -246,6 +258,8 @@ static void backup_abort(BlockJob *job)
100 static void backup_clean(BlockJob *job)
101 {
102 BackupBlockJob *s = container_of(job, BackupBlockJob, common);
103 + if (!s->target)
104 + return;
105 assert(s->target);
106 blk_unref(s->target);
107 s->target = NULL;
108 @@ -255,7 +269,8 @@ static void backup_attached_aio_context(BlockJob *job, AioContext *aio_context)
109 {
110 BackupBlockJob *s = container_of(job, BackupBlockJob, common);
111
112 - blk_set_aio_context(s->target, aio_context);
113 + if (s->target)
114 + blk_set_aio_context(s->target, aio_context);
115 }
116
117 void backup_do_checkpoint(BlockJob *job, Error **errp)
118 @@ -330,9 +345,11 @@ static BlockErrorAction backup_error_action(BackupBlockJob *job,
119 if (read) {
120 return block_job_error_action(&job->common, job->on_source_error,
121 true, error);
122 - } else {
123 + } else if (job->target) {
124 return block_job_error_action(&job->common, job->on_target_error,
125 false, error);
126 + } else {
127 + return BLOCK_ERROR_ACTION_REPORT;
128 }
129 }
130
131 @@ -453,6 +469,7 @@ static void coroutine_fn backup_run(void *opaque)
132
133 job->done_bitmap = bitmap_new(end);
134
135 +
136 job->before_write.notify = backup_before_write_notify;
137 bdrv_add_before_write_notifier(bs, &job->before_write);
138
139 @@ -557,6 +574,7 @@ BlockJob *backup_job_create(const char *job_id, BlockDriverState *bs,
140 BlockdevOnError on_source_error,
141 BlockdevOnError on_target_error,
142 int creation_flags,
143 + BackupDumpFunc *dump_cb,
144 BlockCompletionFunc *cb, void *opaque,
145 int pause_count,
146 BlockJobTxn *txn, Error **errp)
147 @@ -567,7 +585,7 @@ BlockJob *backup_job_create(const char *job_id, BlockDriverState *bs,
148 int ret;
149
150 assert(bs);
151 - assert(target);
152 + assert(target || dump_cb);
153
154 if (bs == target) {
155 error_setg(errp, "Source and target cannot be the same");
156 @@ -580,13 +598,13 @@ BlockJob *backup_job_create(const char *job_id, BlockDriverState *bs,
157 return NULL;
158 }
159
160 - if (!bdrv_is_inserted(target)) {
161 + if (target && !bdrv_is_inserted(target)) {
162 error_setg(errp, "Device is not inserted: %s",
163 bdrv_get_device_name(target));
164 return NULL;
165 }
166
167 - if (compress && target->drv->bdrv_co_pwritev_compressed == NULL) {
168 + if (target && compress && target->drv->bdrv_co_pwritev_compressed == NULL) {
169 error_setg(errp, "Compression is not supported for this drive %s",
170 bdrv_get_device_name(target));
171 return NULL;
172 @@ -596,7 +614,7 @@ BlockJob *backup_job_create(const char *job_id, BlockDriverState *bs,
173 return NULL;
174 }
175
176 - if (bdrv_op_is_blocked(target, BLOCK_OP_TYPE_BACKUP_TARGET, errp)) {
177 + if (target && bdrv_op_is_blocked(target, BLOCK_OP_TYPE_BACKUP_TARGET, errp)) {
178 return NULL;
179 }
180
181 @@ -636,15 +654,18 @@ BlockJob *backup_job_create(const char *job_id, BlockDriverState *bs,
182 goto error;
183 }
184
185 - /* The target must match the source in size, so no resize here either */
186 - job->target = blk_new(BLK_PERM_WRITE,
187 - BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE |
188 - BLK_PERM_WRITE_UNCHANGED | BLK_PERM_GRAPH_MOD);
189 - ret = blk_insert_bs(job->target, target, errp);
190 - if (ret < 0) {
191 - goto error;
192 + if (target) {
193 + /* The target must match the source in size, so no resize here either */
194 + job->target = blk_new(BLK_PERM_WRITE,
195 + BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE |
196 + BLK_PERM_WRITE_UNCHANGED | BLK_PERM_GRAPH_MOD);
197 + ret = blk_insert_bs(job->target, target, errp);
198 + if (ret < 0) {
199 + goto error;
200 + }
201 }
202
203 + job->dump_cb = dump_cb;
204 job->on_source_error = on_source_error;
205 job->on_target_error = on_target_error;
206 job->sync_mode = sync_mode;
207 @@ -652,38 +673,55 @@ BlockJob *backup_job_create(const char *job_id, BlockDriverState *bs,
208 sync_bitmap : NULL;
209 job->compress = compress;
210
211 - /* If there is no backing file on the target, we cannot rely on COW if our
212 - * backup cluster size is smaller than the target cluster size. Even for
213 - * targets with a backing file, try to avoid COW if possible. */
214 - ret = bdrv_get_info(target, &bdi);
215 - if (ret == -ENOTSUP && !target->backing) {
216 - /* Cluster size is not defined */
217 - error_report("WARNING: The target block device doesn't provide "
218 - "information about the block size and it doesn't have a "
219 - "backing file. The default block size of %u bytes is "
220 - "used. If the actual block size of the target exceeds "
221 - "this default, the backup may be unusable",
222 - BACKUP_CLUSTER_SIZE_DEFAULT);
223 - job->cluster_size = BACKUP_CLUSTER_SIZE_DEFAULT;
224 - } else if (ret < 0 && !target->backing) {
225 - error_setg_errno(errp, -ret,
226 - "Couldn't determine the cluster size of the target image, "
227 - "which has no backing file");
228 - error_append_hint(errp,
229 - "Aborting, since this may create an unusable destination image\n");
230 - goto error;
231 - } else if (ret < 0 && target->backing) {
232 - /* Not fatal; just trudge on ahead. */
233 - job->cluster_size = BACKUP_CLUSTER_SIZE_DEFAULT;
234 + if (target) {
235 + /* If there is no backing file on the target, we cannot rely on COW if our
236 + * backup cluster size is smaller than the target cluster size. Even for
237 + * targets with a backing file, try to avoid COW if possible. */
238 + ret = bdrv_get_info(target, &bdi);
239 + if (ret == -ENOTSUP && !target->backing) {
240 + /* Cluster size is not defined */
241 + error_report("WARNING: The target block device doesn't provide "
242 + "information about the block size and it doesn't have a "
243 + "backing file. The default block size of %u bytes is "
244 + "used. If the actual block size of the target exceeds "
245 + "this default, the backup may be unusable",
246 + BACKUP_CLUSTER_SIZE_DEFAULT);
247 + job->cluster_size = BACKUP_CLUSTER_SIZE_DEFAULT;
248 + } else if (ret < 0 && !target->backing) {
249 + error_setg_errno(errp, -ret,
250 + "Couldn't determine the cluster size of the target image, "
251 + "which has no backing file");
252 + error_append_hint(errp,
253 + "Aborting, since this may create an unusable destination image\n");
254 + goto error;
255 + } else if (ret < 0 && target->backing) {
256 + /* Not fatal; just trudge on ahead. */
257 + job->cluster_size = BACKUP_CLUSTER_SIZE_DEFAULT;
258 + } else {
259 + job->cluster_size = MAX(BACKUP_CLUSTER_SIZE_DEFAULT, bdi.cluster_size);
260 + }
261 } else {
262 - job->cluster_size = MAX(BACKUP_CLUSTER_SIZE_DEFAULT, bdi.cluster_size);
263 + ret = bdrv_get_info(bs, &bdi);
264 + if (ret < 0) {
265 + job->cluster_size = BACKUP_CLUSTER_SIZE_DEFAULT;
266 + } else {
267 + /* round down to nearest BACKUP_CLUSTER_SIZE_DEFAULT */
268 + job->cluster_size = (bdi.cluster_size / BACKUP_CLUSTER_SIZE_DEFAULT) * BACKUP_CLUSTER_SIZE_DEFAULT;
269 + if (job->cluster_size == 0) {
270 + /* but we can't go below it */
271 + job->cluster_size = BACKUP_CLUSTER_SIZE_DEFAULT;
272 + }
273 + }
274 }
275
276 - /* Required permissions are already taken with target's blk_new() */
277 - block_job_add_bdrv(&job->common, "target", target, 0, BLK_PERM_ALL,
278 - &error_abort);
279 + if (target) {
280 + /* Required permissions are already taken with target's blk_new() */
281 + block_job_add_bdrv(&job->common, "target", target, 0, BLK_PERM_ALL,
282 + &error_abort);
283 + } else {
284 + job->common.pause_count = pause_count;
285 + }
286 job->common.len = len;
287 - job->common.pause_count = pause_count;
288 block_job_txn_add_job(txn, &job->common);
289
290 return &job->common;
291 diff --git a/block/replication.c b/block/replication.c
292 index 1c41d9e6bf..60c6524417 100644
293 --- a/block/replication.c
294 +++ b/block/replication.c
295 @@ -531,6 +531,7 @@ static void replication_start(ReplicationState *rs, ReplicationMode mode,
296 0, MIRROR_SYNC_MODE_NONE, NULL, false,
297 BLOCKDEV_ON_ERROR_REPORT,
298 BLOCKDEV_ON_ERROR_REPORT, BLOCK_JOB_INTERNAL,
299 + NULL,
300 backup_job_completed, bs, 0, NULL, &local_err);
301 if (local_err) {
302 error_propagate(errp, local_err);
303 diff --git a/blockdev.c b/blockdev.c
304 index 981276692a..76a7103743 100644
305 --- a/blockdev.c
306 +++ b/blockdev.c
307 @@ -31,7 +31,6 @@
308 */
309
310 #include "qemu/osdep.h"
311 -#include "qemu/uuid.h"
312 #include "sysemu/block-backend.h"
313 #include "sysemu/blockdev.h"
314 #include "hw/block/block.h"
315 @@ -55,6 +54,7 @@
316 #include "qemu/cutils.h"
317 #include "qemu/help_option.h"
318 #include "qemu/throttle-options.h"
319 +#include "vma.h"
320
321 static QTAILQ_HEAD(, BlockDriverState) monitor_bdrv_states =
322 QTAILQ_HEAD_INITIALIZER(monitor_bdrv_states);
323 @@ -2958,20 +2958,44 @@ out:
324 aio_context_release(aio_context);
325 }
326
327 +void block_job_event_cancelled(BlockJob *job);
328 +void block_job_event_completed(BlockJob *job, const char *msg);
329 +static void block_job_cb(void *opaque, int ret)
330 +{
331 + /* Note that this function may be executed from another AioContext besides
332 + * the QEMU main loop. If you need to access anything that assumes the
333 + * QEMU global mutex, use a BH or introduce a mutex.
334 + */
335 +
336 + BlockDriverState *bs = opaque;
337 + const char *msg = NULL;
338 +
339 + assert(bs->job);
340 +
341 + if (ret < 0) {
342 + msg = strerror(-ret);
343 + }
344 +
345 + if (block_job_is_cancelled(bs->job)) {
346 + block_job_event_cancelled(bs->job);
347 + } else {
348 + block_job_event_completed(bs->job, msg);
349 + }
350 +}
351 +
352 /* PVE backup related function */
353
354 static struct PVEBackupState {
355 Error *error;
356 bool cancel;
357 - QemuUUID uuid;
358 + uuid_t uuid;
359 char uuid_str[37];
360 int64_t speed;
361 time_t start_time;
362 time_t end_time;
363 char *backup_file;
364 - Object *vmaobj;
365 + VmaWriter *vmaw;
366 GList *di_list;
367 - size_t next_job;
368 size_t total;
369 size_t transferred;
370 size_t zero_bytes;
371 @@ -2981,6 +3005,7 @@ typedef struct PVEBackupDevInfo {
372 BlockDriverState *bs;
373 size_t size;
374 uint8_t dev_id;
375 + //bool started;
376 bool completed;
377 char targetfile[PATH_MAX];
378 BlockDriverState *target;
379 @@ -2988,13 +3013,79 @@ typedef struct PVEBackupDevInfo {
380
381 static void pvebackup_run_next_job(void);
382
383 +static int pvebackup_dump_cb(void *opaque, BlockBackend *target,
384 + int64_t sector_num, int n_sectors,
385 + unsigned char *buf)
386 +{
387 + PVEBackupDevInfo *di = opaque;
388 +
389 + int size = n_sectors * BDRV_SECTOR_SIZE;
390 + if (backup_state.cancel) {
391 + return size; // return success
392 + }
393 +
394 + if (sector_num & 0x7f) {
395 + if (!backup_state.error) {
396 + error_setg(&backup_state.error,
397 + "got unaligned write inside backup dump "
398 + "callback (sector %ld)", sector_num);
399 + }
400 + return -1; // not aligned to cluster size
401 + }
402 +
403 + int64_t cluster_num = sector_num >> 7;
404 +
405 + int ret = -1;
406 +
407 + if (backup_state.vmaw) {
408 + size_t zero_bytes = 0;
409 + int64_t remaining = n_sectors * BDRV_SECTOR_SIZE;
410 + while (remaining > 0) {
411 + ret = vma_writer_write(backup_state.vmaw, di->dev_id, cluster_num,
412 + buf, &zero_bytes);
413 + ++cluster_num;
414 + if (buf) {
415 + buf += VMA_CLUSTER_SIZE;
416 + }
417 + if (ret < 0) {
418 + if (!backup_state.error) {
419 + vma_writer_error_propagate(backup_state.vmaw, &backup_state.error);
420 + }
421 + if (di->bs && di->bs->job) {
422 + block_job_cancel(di->bs->job);
423 + }
424 + break;
425 + } else {
426 + backup_state.zero_bytes += zero_bytes;
427 + if (remaining >= VMA_CLUSTER_SIZE) {
428 + backup_state.transferred += VMA_CLUSTER_SIZE;
429 + } else {
430 + backup_state.transferred += remaining;
431 + }
432 + remaining -= VMA_CLUSTER_SIZE;
433 + }
434 + }
435 + } else {
436 + if (!buf) {
437 + backup_state.zero_bytes += size;
438 + }
439 + backup_state.transferred += size;
440 + }
441 +
442 + // Note: always return success, because we want that writes succeed anyways.
443 +
444 + return size;
445 +}
446 +
447 static void pvebackup_cleanup(void)
448 {
449 backup_state.end_time = time(NULL);
450
451 - if (backup_state.vmaobj) {
452 - object_unparent(backup_state.vmaobj);
453 - backup_state.vmaobj = NULL;
454 + if (backup_state.vmaw) {
455 + Error *local_err = NULL;
456 + vma_writer_close(backup_state.vmaw, &local_err);
457 + error_propagate(&backup_state.error, local_err);
458 + backup_state.vmaw = NULL;
459 }
460
461 if (backup_state.di_list) {
462 @@ -3009,6 +3100,13 @@ static void pvebackup_cleanup(void)
463 }
464 }
465
466 +static void coroutine_fn backup_close_vma_stream(void *opaque)
467 +{
468 + PVEBackupDevInfo *di = opaque;
469 +
470 + vma_writer_close_stream(backup_state.vmaw, di->dev_id);
471 +}
472 +
473 static void pvebackup_complete_cb(void *opaque, int ret)
474 {
475 PVEBackupDevInfo *di = opaque;
476 @@ -3020,14 +3118,18 @@ static void pvebackup_complete_cb(void *opaque, int ret)
477 ret, strerror(-ret));
478 }
479
480 + BlockDriverState *bs = di->bs;
481 +
482 di->bs = NULL;
483 di->target = NULL;
484
485 - if (backup_state.vmaobj) {
486 - object_unparent(backup_state.vmaobj);
487 - backup_state.vmaobj = NULL;
488 + if (backup_state.vmaw) {
489 + Coroutine *co = qemu_coroutine_create(backup_close_vma_stream, di);
490 + qemu_coroutine_enter(co);
491 }
492
493 + block_job_cb(bs, ret);
494 +
495 if (!backup_state.cancel) {
496 pvebackup_run_next_job();
497 }
498 @@ -3041,14 +3143,9 @@ static void pvebackup_cancel(void *opaque)
499 error_setg(&backup_state.error, "backup cancelled");
500 }
501
502 - if (backup_state.vmaobj) {
503 - Error *err;
504 + if (backup_state.vmaw) {
505 /* make sure vma writer does not block anymore */
506 - if (!object_set_props(backup_state.vmaobj, &err, "blocked", "yes", NULL)) {
507 - if (err) {
508 - error_report_err(err);
509 - }
510 - }
511 + vma_writer_set_error(backup_state.vmaw, "backup cancelled");
512 }
513
514 GList *l = backup_state.di_list;
515 @@ -3073,19 +3170,15 @@ void qmp_backup_cancel(Error **errp)
516 Coroutine *co = qemu_coroutine_create(pvebackup_cancel, NULL);
517 qemu_coroutine_enter(co);
518
519 - while (backup_state.vmaobj) {
520 - /* FIXME: Find something better for this */
521 + while (backup_state.vmaw) {
522 + /* vma writer use main aio context */
523 aio_poll(qemu_get_aio_context(), true);
524 }
525 }
526
527 -void vma_object_add_config_file(Object *obj, const char *name,
528 - const char *contents, size_t len,
529 - Error **errp);
530 static int config_to_vma(const char *file, BackupFormat format,
531 - Object *vmaobj,
532 - const char *backup_dir,
533 - Error **errp)
534 + const char *backup_dir, VmaWriter *vmaw,
535 + Error **errp)
536 {
537 char *cdata = NULL;
538 gsize clen = 0;
539 @@ -3098,12 +3191,17 @@ static int config_to_vma(const char *file, BackupFormat format,
540 char *basename = g_path_get_basename(file);
541
542 if (format == BACKUP_FORMAT_VMA) {
543 - vma_object_add_config_file(vmaobj, basename, cdata, clen, errp);
544 + if (vma_writer_add_config(vmaw, basename, cdata, clen) != 0) {
545 + error_setg(errp, "unable to add %s config data to vma archive", file);
546 + g_free(cdata);
547 + g_free(basename);
548 + return 1;
549 + }
550 } else if (format == BACKUP_FORMAT_DIR) {
551 char config_path[PATH_MAX];
552 snprintf(config_path, PATH_MAX, "%s/%s", backup_dir, basename);
553 if (!g_file_set_contents(config_path, cdata, clen, &err)) {
554 - error_setg(errp, "unable to write config file '%s'", config_path);
555 + error_setg(errp, "unable to write config file '%s'", config_path);
556 g_free(cdata);
557 g_free(basename);
558 return 1;
559 @@ -3113,35 +3211,37 @@ static int config_to_vma(const char *file, BackupFormat format,
560 g_free(basename);
561 g_free(cdata);
562
563 - return 0;
564 + return 0;
565 }
566
567 +bool block_job_should_pause(BlockJob *job);
568 static void pvebackup_run_next_job(void)
569 {
570 - bool cancel = backup_state.error || backup_state.cancel;
571 -fprintf(stderr, "run next job: %zu\n", backup_state.next_job);
572 - GList *next = g_list_nth(backup_state.di_list, backup_state.next_job);
573 - while (next) {
574 - PVEBackupDevInfo *di = (PVEBackupDevInfo *)next->data;
575 - backup_state.next_job++;
576 + GList *l = backup_state.di_list;
577 + while (l) {
578 + PVEBackupDevInfo *di = (PVEBackupDevInfo *)l->data;
579 + l = g_list_next(l);
580 if (!di->completed && di->bs && di->bs->job) {
581 BlockJob *job = di->bs->job;
582 - if (cancel) {
583 - block_job_cancel(job);
584 - } else {
585 - block_job_resume(job);
586 + if (block_job_should_pause(job)) {
587 + bool cancel = backup_state.error || backup_state.cancel;
588 + if (cancel) {
589 + block_job_cancel(job);
590 + } else {
591 + block_job_resume(job);
592 + }
593 }
594 return;
595 }
596 - next = g_list_next(next);
597 }
598 +
599 pvebackup_cleanup();
600 }
601
602 UuidInfo *qmp_backup(const char *backup_file, bool has_format,
603 BackupFormat format,
604 bool has_config_file, const char *config_file,
605 - bool has_firewall_file, const char *firewall_file,
606 + bool has_firewall_file, const char *firewall_file,
607 bool has_devlist, const char *devlist,
608 bool has_speed, int64_t speed, Error **errp)
609 {
610 @@ -3149,14 +3249,15 @@ UuidInfo *qmp_backup(const char *backup_file, bool has_format,
611 BlockDriverState *bs = NULL;
612 const char *backup_dir = NULL;
613 Error *local_err = NULL;
614 - QemuUUID uuid;
615 + uuid_t uuid;
616 + VmaWriter *vmaw = NULL;
617 gchar **devs = NULL;
618 GList *di_list = NULL;
619 GList *l;
620 UuidInfo *uuid_info;
621 BlockJob *job;
622
623 - if (backup_state.di_list || backup_state.vmaobj) {
624 + if (backup_state.di_list) {
625 error_set(errp, ERROR_CLASS_GENERIC_ERROR,
626 "previous backup not finished");
627 return NULL;
628 @@ -3231,40 +3332,28 @@ UuidInfo *qmp_backup(const char *backup_file, bool has_format,
629 total += size;
630 }
631
632 - qemu_uuid_generate(&uuid);
633 + uuid_generate(uuid);
634
635 if (format == BACKUP_FORMAT_VMA) {
636 - char uuidstr[UUID_FMT_LEN+1];
637 - qemu_uuid_unparse(&uuid, uuidstr);
638 - uuidstr[UUID_FMT_LEN] = 0;
639 - backup_state.vmaobj =
640 - object_new_with_props("vma", object_get_objects_root(),
641 - "vma-backup-obj", &local_err,
642 - "filename", backup_file,
643 - "uuid", uuidstr,
644 - NULL);
645 - if (!backup_state.vmaobj) {
646 + vmaw = vma_writer_create(backup_file, uuid, &local_err);
647 + if (!vmaw) {
648 if (local_err) {
649 error_propagate(errp, local_err);
650 }
651 goto err;
652 }
653
654 + /* register all devices for vma writer */
655 l = di_list;
656 while (l) {
657 - QDict *options = qdict_new();
658 -
659 PVEBackupDevInfo *di = (PVEBackupDevInfo *)l->data;
660 l = g_list_next(l);
661
662 const char *devname = bdrv_get_device_name(di->bs);
663 - snprintf(di->targetfile, PATH_MAX, "vma-backup-obj/%s.raw", devname);
664 -
665 - qdict_put(options, "driver", qstring_from_str("vma-drive"));
666 - qdict_put(options, "size", qint_from_int(di->size));
667 - di->target = bdrv_open(di->targetfile, NULL, options, BDRV_O_RDWR, &local_err);
668 - if (!di->target) {
669 - error_propagate(errp, local_err);
670 + di->dev_id = vma_writer_register_stream(vmaw, devname, di->size);
671 + if (di->dev_id <= 0) {
672 + error_set(errp, ERROR_CLASS_GENERIC_ERROR,
673 + "register_stream failed");
674 goto err;
675 }
676 }
677 @@ -3305,15 +3394,15 @@ UuidInfo *qmp_backup(const char *backup_file, bool has_format,
678
679 /* add configuration file to archive */
680 if (has_config_file) {
681 - if(config_to_vma(config_file, format, backup_state.vmaobj, backup_dir, errp) != 0) {
682 - goto err;
683 + if(config_to_vma(config_file, format, backup_dir, vmaw, errp) != 0) {
684 + goto err;
685 }
686 }
687
688 /* add firewall file to archive */
689 if (has_firewall_file) {
690 - if(config_to_vma(firewall_file, format, backup_state.vmaobj, backup_dir, errp) != 0) {
691 - goto err;
692 + if(config_to_vma(firewall_file, format, backup_dir, vmaw, errp) != 0) {
693 + goto err;
694 }
695 }
696 /* initialize global backup_state now */
697 @@ -3335,11 +3424,12 @@ UuidInfo *qmp_backup(const char *backup_file, bool has_format,
698 }
699 backup_state.backup_file = g_strdup(backup_file);
700
701 - memcpy(&backup_state.uuid, &uuid, sizeof(uuid));
702 - qemu_uuid_unparse(&uuid, backup_state.uuid_str);
703 + backup_state.vmaw = vmaw;
704 +
705 + uuid_copy(backup_state.uuid, uuid);
706 + uuid_unparse_lower(uuid, backup_state.uuid_str);
707
708 backup_state.di_list = di_list;
709 - backup_state.next_job = 0;
710
711 backup_state.total = total;
712 backup_state.transferred = 0;
713 @@ -3350,21 +3440,16 @@ UuidInfo *qmp_backup(const char *backup_file, bool has_format,
714 while (l) {
715 PVEBackupDevInfo *di = (PVEBackupDevInfo *)l->data;
716 l = g_list_next(l);
717 -
718 job = backup_job_create(NULL, di->bs, di->target, speed, MIRROR_SYNC_MODE_FULL, NULL,
719 false, BLOCKDEV_ON_ERROR_REPORT, BLOCKDEV_ON_ERROR_REPORT,
720 BLOCK_JOB_DEFAULT,
721 - pvebackup_complete_cb, di, 2, NULL, &local_err);
722 - if (di->target) {
723 - bdrv_unref(di->target);
724 - di->target = NULL;
725 - }
726 + pvebackup_dump_cb, pvebackup_complete_cb, di,
727 + 2, NULL, &local_err);
728 if (!job || local_err != NULL) {
729 error_setg(&backup_state.error, "backup_job_create failed");
730 pvebackup_cancel(NULL);
731 - } else {
732 - block_job_start(job);
733 }
734 + block_job_start(job);
735 }
736
737 if (!backup_state.error) {
738 @@ -3398,9 +3483,10 @@ err:
739 g_strfreev(devs);
740 }
741
742 - if (backup_state.vmaobj) {
743 - object_unparent(backup_state.vmaobj);
744 - backup_state.vmaobj = NULL;
745 + if (vmaw) {
746 + Error *err = NULL;
747 + vma_writer_close(vmaw, &err);
748 + unlink(backup_file);
749 }
750
751 if (backup_dir) {
752 @@ -3772,7 +3858,7 @@ static BlockJob *do_drive_backup(DriveBackup *backup, BlockJobTxn *txn,
753 job = backup_job_create(backup->job_id, bs, target_bs, backup->speed,
754 backup->sync, bmap, backup->compress,
755 backup->on_source_error, backup->on_target_error,
756 - BLOCK_JOB_DEFAULT, NULL, NULL, 0, txn, &local_err);
757 + BLOCK_JOB_DEFAULT, NULL, NULL, NULL, 0, txn, &local_err);
758 bdrv_unref(target_bs);
759 if (local_err != NULL) {
760 error_propagate(errp, local_err);
761 @@ -3851,7 +3937,7 @@ BlockJob *do_blockdev_backup(BlockdevBackup *backup, BlockJobTxn *txn,
762 job = backup_job_create(backup->job_id, bs, target_bs, backup->speed,
763 backup->sync, NULL, backup->compress,
764 backup->on_source_error, backup->on_target_error,
765 - BLOCK_JOB_DEFAULT, NULL, NULL, 0, txn, &local_err);
766 + BLOCK_JOB_DEFAULT, NULL, NULL, NULL, 0, txn, &local_err);
767 if (local_err != NULL) {
768 error_propagate(errp, local_err);
769 }
770 diff --git a/blockjob.c b/blockjob.c
771 index 764d41863e..cb3741f6dd 100644
772 --- a/blockjob.c
773 +++ b/blockjob.c
774 @@ -37,8 +37,8 @@
775 #include "qemu/timer.h"
776 #include "qapi-event.h"
777
778 -static void block_job_event_cancelled(BlockJob *job);
779 -static void block_job_event_completed(BlockJob *job, const char *msg);
780 +void block_job_event_cancelled(BlockJob *job);
781 +void block_job_event_completed(BlockJob *job, const char *msg);
782
783 /* Transactional group of block jobs */
784 struct BlockJobTxn {
785 @@ -473,7 +473,8 @@ void block_job_user_pause(BlockJob *job)
786 block_job_pause(job);
787 }
788
789 -static bool block_job_should_pause(BlockJob *job)
790 +bool block_job_should_pause(BlockJob *job);
791 +bool block_job_should_pause(BlockJob *job)
792 {
793 return job->pause_count > 0;
794 }
795 @@ -687,7 +688,7 @@ static void block_job_iostatus_set_err(BlockJob *job, int error)
796 }
797 }
798
799 -static void block_job_event_cancelled(BlockJob *job)
800 +void block_job_event_cancelled(BlockJob *job)
801 {
802 if (block_job_is_internal(job)) {
803 return;
804 @@ -701,7 +702,7 @@ static void block_job_event_cancelled(BlockJob *job)
805 &error_abort);
806 }
807
808 -static void block_job_event_completed(BlockJob *job, const char *msg)
809 +void block_job_event_completed(BlockJob *job, const char *msg)
810 {
811 if (block_job_is_internal(job)) {
812 return;
813 diff --git a/include/block/block_int.h b/include/block/block_int.h
814 index 2b3ecd0575..278da161fb 100644
815 --- a/include/block/block_int.h
816 +++ b/include/block/block_int.h
817 @@ -59,6 +59,9 @@
818
819 #define BLOCK_PROBE_BUF_SIZE 512
820
821 +typedef int BackupDumpFunc(void *opaque, BlockBackend *be,
822 + int64_t sector_num, int n_sectors, unsigned char *buf);
823 +
824 enum BdrvTrackedRequestType {
825 BDRV_TRACKED_READ,
826 BDRV_TRACKED_WRITE,
827 @@ -877,6 +880,7 @@ BlockJob *backup_job_create(const char *job_id, BlockDriverState *bs,
828 BlockdevOnError on_source_error,
829 BlockdevOnError on_target_error,
830 int creation_flags,
831 + BackupDumpFunc *dump_cb,
832 BlockCompletionFunc *cb, void *opaque,
833 int pause_count,
834 BlockJobTxn *txn, Error **errp);
835 diff --git a/vma-reader.c b/vma-reader.c
836 new file mode 100644
837 index 0000000000..2000889bd3
838 --- /dev/null
839 +++ b/vma-reader.c
840 @@ -0,0 +1,857 @@
841 +/*
842 + * VMA: Virtual Machine Archive
843 + *
844 + * Copyright (C) 2012 Proxmox Server Solutions
845 + *
846 + * Authors:
847 + * Dietmar Maurer (dietmar@proxmox.com)
848 + *
849 + * This work is licensed under the terms of the GNU GPL, version 2 or later.
850 + * See the COPYING file in the top-level directory.
851 + *
852 + */
853 +
854 +#include "qemu/osdep.h"
855 +#include <glib.h>
856 +#include <uuid/uuid.h>
857 +
858 +#include "qemu-common.h"
859 +#include "qemu/timer.h"
860 +#include "qemu/ratelimit.h"
861 +#include "vma.h"
862 +#include "block/block.h"
863 +#include "sysemu/block-backend.h"
864 +
865 +static unsigned char zero_vma_block[VMA_BLOCK_SIZE];
866 +
867 +typedef struct VmaRestoreState {
868 + BlockBackend *target;
869 + bool write_zeroes;
870 + unsigned long *bitmap;
871 + int bitmap_size;
872 +} VmaRestoreState;
873 +
874 +struct VmaReader {
875 + int fd;
876 + GChecksum *md5csum;
877 + GHashTable *blob_hash;
878 + unsigned char *head_data;
879 + VmaDeviceInfo devinfo[256];
880 + VmaRestoreState rstate[256];
881 + GList *cdata_list;
882 + guint8 vmstate_stream;
883 + uint32_t vmstate_clusters;
884 + /* to show restore percentage if run with -v */
885 + time_t start_time;
886 + int64_t cluster_count;
887 + int64_t clusters_read;
888 + int64_t zero_cluster_data;
889 + int64_t partial_zero_cluster_data;
890 + int clusters_read_per;
891 +};
892 +
893 +static guint
894 +g_int32_hash(gconstpointer v)
895 +{
896 + return *(const uint32_t *)v;
897 +}
898 +
899 +static gboolean
900 +g_int32_equal(gconstpointer v1, gconstpointer v2)
901 +{
902 + return *((const uint32_t *)v1) == *((const uint32_t *)v2);
903 +}
904 +
905 +static int vma_reader_get_bitmap(VmaRestoreState *rstate, int64_t cluster_num)
906 +{
907 + assert(rstate);
908 + assert(rstate->bitmap);
909 +
910 + unsigned long val, idx, bit;
911 +
912 + idx = cluster_num / BITS_PER_LONG;
913 +
914 + assert(rstate->bitmap_size > idx);
915 +
916 + bit = cluster_num % BITS_PER_LONG;
917 + val = rstate->bitmap[idx];
918 +
919 + return !!(val & (1UL << bit));
920 +}
921 +
922 +static void vma_reader_set_bitmap(VmaRestoreState *rstate, int64_t cluster_num,
923 + int dirty)
924 +{
925 + assert(rstate);
926 + assert(rstate->bitmap);
927 +
928 + unsigned long val, idx, bit;
929 +
930 + idx = cluster_num / BITS_PER_LONG;
931 +
932 + assert(rstate->bitmap_size > idx);
933 +
934 + bit = cluster_num % BITS_PER_LONG;
935 + val = rstate->bitmap[idx];
936 + if (dirty) {
937 + if (!(val & (1UL << bit))) {
938 + val |= 1UL << bit;
939 + }
940 + } else {
941 + if (val & (1UL << bit)) {
942 + val &= ~(1UL << bit);
943 + }
944 + }
945 + rstate->bitmap[idx] = val;
946 +}
947 +
948 +typedef struct VmaBlob {
949 + uint32_t start;
950 + uint32_t len;
951 + void *data;
952 +} VmaBlob;
953 +
954 +static const VmaBlob *get_header_blob(VmaReader *vmar, uint32_t pos)
955 +{
956 + assert(vmar);
957 + assert(vmar->blob_hash);
958 +
959 + return g_hash_table_lookup(vmar->blob_hash, &pos);
960 +}
961 +
962 +static const char *get_header_str(VmaReader *vmar, uint32_t pos)
963 +{
964 + const VmaBlob *blob = get_header_blob(vmar, pos);
965 + if (!blob) {
966 + return NULL;
967 + }
968 + const char *res = (char *)blob->data;
969 + if (res[blob->len-1] != '\0') {
970 + return NULL;
971 + }
972 + return res;
973 +}
974 +
975 +static ssize_t
976 +safe_read(int fd, unsigned char *buf, size_t count)
977 +{
978 + ssize_t n;
979 +
980 + do {
981 + n = read(fd, buf, count);
982 + } while (n < 0 && errno == EINTR);
983 +
984 + return n;
985 +}
986 +
987 +static ssize_t
988 +full_read(int fd, unsigned char *buf, size_t len)
989 +{
990 + ssize_t n;
991 + size_t total;
992 +
993 + total = 0;
994 +
995 + while (len > 0) {
996 + n = safe_read(fd, buf, len);
997 +
998 + if (n == 0) {
999 + return total;
1000 + }
1001 +
1002 + if (n <= 0) {
1003 + break;
1004 + }
1005 +
1006 + buf += n;
1007 + total += n;
1008 + len -= n;
1009 + }
1010 +
1011 + if (len) {
1012 + return -1;
1013 + }
1014 +
1015 + return total;
1016 +}
1017 +
1018 +void vma_reader_destroy(VmaReader *vmar)
1019 +{
1020 + assert(vmar);
1021 +
1022 + if (vmar->fd >= 0) {
1023 + close(vmar->fd);
1024 + }
1025 +
1026 + if (vmar->cdata_list) {
1027 + g_list_free(vmar->cdata_list);
1028 + }
1029 +
1030 + int i;
1031 + for (i = 1; i < 256; i++) {
1032 + if (vmar->rstate[i].bitmap) {
1033 + g_free(vmar->rstate[i].bitmap);
1034 + }
1035 + }
1036 +
1037 + if (vmar->md5csum) {
1038 + g_checksum_free(vmar->md5csum);
1039 + }
1040 +
1041 + if (vmar->blob_hash) {
1042 + g_hash_table_destroy(vmar->blob_hash);
1043 + }
1044 +
1045 + if (vmar->head_data) {
1046 + g_free(vmar->head_data);
1047 + }
1048 +
1049 + g_free(vmar);
1050 +
1051 +};
1052 +
1053 +static int vma_reader_read_head(VmaReader *vmar, Error **errp)
1054 +{
1055 + assert(vmar);
1056 + assert(errp);
1057 + assert(*errp == NULL);
1058 +
1059 + unsigned char md5sum[16];
1060 + int i;
1061 + int ret = 0;
1062 +
1063 + vmar->head_data = g_malloc(sizeof(VmaHeader));
1064 +
1065 + if (full_read(vmar->fd, vmar->head_data, sizeof(VmaHeader)) !=
1066 + sizeof(VmaHeader)) {
1067 + error_setg(errp, "can't read vma header - %s",
1068 + errno ? g_strerror(errno) : "got EOF");
1069 + return -1;
1070 + }
1071 +
1072 + VmaHeader *h = (VmaHeader *)vmar->head_data;
1073 +
1074 + if (h->magic != VMA_MAGIC) {
1075 + error_setg(errp, "not a vma file - wrong magic number");
1076 + return -1;
1077 + }
1078 +
1079 + uint32_t header_size = GUINT32_FROM_BE(h->header_size);
1080 + int need = header_size - sizeof(VmaHeader);
1081 + if (need <= 0) {
1082 + error_setg(errp, "wrong vma header size %d", header_size);
1083 + return -1;
1084 + }
1085 +
1086 + vmar->head_data = g_realloc(vmar->head_data, header_size);
1087 + h = (VmaHeader *)vmar->head_data;
1088 +
1089 + if (full_read(vmar->fd, vmar->head_data + sizeof(VmaHeader), need) !=
1090 + need) {
1091 + error_setg(errp, "can't read vma header data - %s",
1092 + errno ? g_strerror(errno) : "got EOF");
1093 + return -1;
1094 + }
1095 +
1096 + memcpy(md5sum, h->md5sum, 16);
1097 + memset(h->md5sum, 0, 16);
1098 +
1099 + g_checksum_reset(vmar->md5csum);
1100 + g_checksum_update(vmar->md5csum, vmar->head_data, header_size);
1101 + gsize csize = 16;
1102 + g_checksum_get_digest(vmar->md5csum, (guint8 *)(h->md5sum), &csize);
1103 +
1104 + if (memcmp(md5sum, h->md5sum, 16) != 0) {
1105 + error_setg(errp, "wrong vma header chechsum");
1106 + return -1;
1107 + }
1108 +
1109 + /* we can modify header data after checksum verify */
1110 + h->header_size = header_size;
1111 +
1112 + h->version = GUINT32_FROM_BE(h->version);
1113 + if (h->version != 1) {
1114 + error_setg(errp, "wrong vma version %d", h->version);
1115 + return -1;
1116 + }
1117 +
1118 + h->ctime = GUINT64_FROM_BE(h->ctime);
1119 + h->blob_buffer_offset = GUINT32_FROM_BE(h->blob_buffer_offset);
1120 + h->blob_buffer_size = GUINT32_FROM_BE(h->blob_buffer_size);
1121 +
1122 + uint32_t bstart = h->blob_buffer_offset + 1;
1123 + uint32_t bend = h->blob_buffer_offset + h->blob_buffer_size;
1124 +
1125 + if (bstart <= sizeof(VmaHeader)) {
1126 + error_setg(errp, "wrong vma blob buffer offset %d",
1127 + h->blob_buffer_offset);
1128 + return -1;
1129 + }
1130 +
1131 + if (bend > header_size) {
1132 + error_setg(errp, "wrong vma blob buffer size %d/%d",
1133 + h->blob_buffer_offset, h->blob_buffer_size);
1134 + return -1;
1135 + }
1136 +
1137 + while ((bstart + 2) <= bend) {
1138 + uint32_t size = vmar->head_data[bstart] +
1139 + (vmar->head_data[bstart+1] << 8);
1140 + if ((bstart + size + 2) <= bend) {
1141 + VmaBlob *blob = g_new0(VmaBlob, 1);
1142 + blob->start = bstart - h->blob_buffer_offset;
1143 + blob->len = size;
1144 + blob->data = vmar->head_data + bstart + 2;
1145 + g_hash_table_insert(vmar->blob_hash, &blob->start, blob);
1146 + }
1147 + bstart += size + 2;
1148 + }
1149 +
1150 +
1151 + int count = 0;
1152 + for (i = 1; i < 256; i++) {
1153 + VmaDeviceInfoHeader *dih = &h->dev_info[i];
1154 + uint32_t devname_ptr = GUINT32_FROM_BE(dih->devname_ptr);
1155 + uint64_t size = GUINT64_FROM_BE(dih->size);
1156 + const char *devname = get_header_str(vmar, devname_ptr);
1157 +
1158 + if (size && devname) {
1159 + count++;
1160 + vmar->devinfo[i].size = size;
1161 + vmar->devinfo[i].devname = devname;
1162 +
1163 + if (strcmp(devname, "vmstate") == 0) {
1164 + vmar->vmstate_stream = i;
1165 + }
1166 + }
1167 + }
1168 +
1169 + for (i = 0; i < VMA_MAX_CONFIGS; i++) {
1170 + uint32_t name_ptr = GUINT32_FROM_BE(h->config_names[i]);
1171 + uint32_t data_ptr = GUINT32_FROM_BE(h->config_data[i]);
1172 +
1173 + if (!(name_ptr && data_ptr)) {
1174 + continue;
1175 + }
1176 + const char *name = get_header_str(vmar, name_ptr);
1177 + const VmaBlob *blob = get_header_blob(vmar, data_ptr);
1178 +
1179 + if (!(name && blob)) {
1180 + error_setg(errp, "vma contains invalid data pointers");
1181 + return -1;
1182 + }
1183 +
1184 + VmaConfigData *cdata = g_new0(VmaConfigData, 1);
1185 + cdata->name = name;
1186 + cdata->data = blob->data;
1187 + cdata->len = blob->len;
1188 +
1189 + vmar->cdata_list = g_list_append(vmar->cdata_list, cdata);
1190 + }
1191 +
1192 + return ret;
1193 +};
1194 +
1195 +VmaReader *vma_reader_create(const char *filename, Error **errp)
1196 +{
1197 + assert(filename);
1198 + assert(errp);
1199 +
1200 + VmaReader *vmar = g_new0(VmaReader, 1);
1201 +
1202 + if (strcmp(filename, "-") == 0) {
1203 + vmar->fd = dup(0);
1204 + } else {
1205 + vmar->fd = open(filename, O_RDONLY);
1206 + }
1207 +
1208 + if (vmar->fd < 0) {
1209 + error_setg(errp, "can't open file %s - %s\n", filename,
1210 + g_strerror(errno));
1211 + goto err;
1212 + }
1213 +
1214 + vmar->md5csum = g_checksum_new(G_CHECKSUM_MD5);
1215 + if (!vmar->md5csum) {
1216 + error_setg(errp, "can't allocate cmsum\n");
1217 + goto err;
1218 + }
1219 +
1220 + vmar->blob_hash = g_hash_table_new_full(g_int32_hash, g_int32_equal,
1221 + NULL, g_free);
1222 +
1223 + if (vma_reader_read_head(vmar, errp) < 0) {
1224 + goto err;
1225 + }
1226 +
1227 + return vmar;
1228 +
1229 +err:
1230 + if (vmar) {
1231 + vma_reader_destroy(vmar);
1232 + }
1233 +
1234 + return NULL;
1235 +}
1236 +
1237 +VmaHeader *vma_reader_get_header(VmaReader *vmar)
1238 +{
1239 + assert(vmar);
1240 + assert(vmar->head_data);
1241 +
1242 + return (VmaHeader *)(vmar->head_data);
1243 +}
1244 +
1245 +GList *vma_reader_get_config_data(VmaReader *vmar)
1246 +{
1247 + assert(vmar);
1248 + assert(vmar->head_data);
1249 +
1250 + return vmar->cdata_list;
1251 +}
1252 +
1253 +VmaDeviceInfo *vma_reader_get_device_info(VmaReader *vmar, guint8 dev_id)
1254 +{
1255 + assert(vmar);
1256 + assert(dev_id);
1257 +
1258 + if (vmar->devinfo[dev_id].size && vmar->devinfo[dev_id].devname) {
1259 + return &vmar->devinfo[dev_id];
1260 + }
1261 +
1262 + return NULL;
1263 +}
1264 +
1265 +static void allocate_rstate(VmaReader *vmar, guint8 dev_id,
1266 + BlockBackend *target, bool write_zeroes)
1267 +{
1268 + assert(vmar);
1269 + assert(dev_id);
1270 +
1271 + vmar->rstate[dev_id].target = target;
1272 + vmar->rstate[dev_id].write_zeroes = write_zeroes;
1273 +
1274 + int64_t size = vmar->devinfo[dev_id].size;
1275 +
1276 + int64_t bitmap_size = (size/BDRV_SECTOR_SIZE) +
1277 + (VMA_CLUSTER_SIZE/BDRV_SECTOR_SIZE) * BITS_PER_LONG - 1;
1278 + bitmap_size /= (VMA_CLUSTER_SIZE/BDRV_SECTOR_SIZE) * BITS_PER_LONG;
1279 +
1280 + vmar->rstate[dev_id].bitmap_size = bitmap_size;
1281 + vmar->rstate[dev_id].bitmap = g_new0(unsigned long, bitmap_size);
1282 +
1283 + vmar->cluster_count += size/VMA_CLUSTER_SIZE;
1284 +}
1285 +
1286 +int vma_reader_register_bs(VmaReader *vmar, guint8 dev_id, BlockBackend *target,
1287 + bool write_zeroes, Error **errp)
1288 +{
1289 + assert(vmar);
1290 + assert(target != NULL);
1291 + assert(dev_id);
1292 + assert(vmar->rstate[dev_id].target == NULL);
1293 +
1294 + int64_t size = blk_getlength(target);
1295 + int64_t size_diff = size - vmar->devinfo[dev_id].size;
1296 +
1297 + /* storage types can have different size restrictions, so it
1298 + * is not always possible to create an image with exact size.
1299 + * So we tolerate a size difference up to 4MB.
1300 + */
1301 + if ((size_diff < 0) || (size_diff > 4*1024*1024)) {
1302 + error_setg(errp, "vma_reader_register_bs for stream %s failed - "
1303 + "unexpected size %zd != %zd", vmar->devinfo[dev_id].devname,
1304 + size, vmar->devinfo[dev_id].size);
1305 + return -1;
1306 + }
1307 +
1308 + allocate_rstate(vmar, dev_id, target, write_zeroes);
1309 +
1310 + return 0;
1311 +}
1312 +
1313 +static ssize_t safe_write(int fd, void *buf, size_t count)
1314 +{
1315 + ssize_t n;
1316 +
1317 + do {
1318 + n = write(fd, buf, count);
1319 + } while (n < 0 && errno == EINTR);
1320 +
1321 + return n;
1322 +}
1323 +
1324 +static size_t full_write(int fd, void *buf, size_t len)
1325 +{
1326 + ssize_t n;
1327 + size_t total;
1328 +
1329 + total = 0;
1330 +
1331 + while (len > 0) {
1332 + n = safe_write(fd, buf, len);
1333 + if (n < 0) {
1334 + return n;
1335 + }
1336 + buf += n;
1337 + total += n;
1338 + len -= n;
1339 + }
1340 +
1341 + if (len) {
1342 + /* incomplete write ? */
1343 + return -1;
1344 + }
1345 +
1346 + return total;
1347 +}
1348 +
1349 +static int restore_write_data(VmaReader *vmar, guint8 dev_id,
1350 + BlockBackend *target, int vmstate_fd,
1351 + unsigned char *buf, int64_t sector_num,
1352 + int nb_sectors, Error **errp)
1353 +{
1354 + assert(vmar);
1355 +
1356 + if (dev_id == vmar->vmstate_stream) {
1357 + if (vmstate_fd >= 0) {
1358 + int len = nb_sectors * BDRV_SECTOR_SIZE;
1359 + int res = full_write(vmstate_fd, buf, len);
1360 + if (res < 0) {
1361 + error_setg(errp, "write vmstate failed %d", res);
1362 + return -1;
1363 + }
1364 + }
1365 + } else {
1366 + int res = blk_pwrite(target, sector_num * BDRV_SECTOR_SIZE, buf, nb_sectors * BDRV_SECTOR_SIZE, 0);
1367 + if (res < 0) {
1368 + error_setg(errp, "blk_pwrite to %s failed (%d)",
1369 + bdrv_get_device_name(blk_bs(target)), res);
1370 + return -1;
1371 + }
1372 + }
1373 + return 0;
1374 +}
1375 +
1376 +static int restore_extent(VmaReader *vmar, unsigned char *buf,
1377 + int extent_size, int vmstate_fd,
1378 + bool verbose, bool verify, Error **errp)
1379 +{
1380 + assert(vmar);
1381 + assert(buf);
1382 +
1383 + VmaExtentHeader *ehead = (VmaExtentHeader *)buf;
1384 + int start = VMA_EXTENT_HEADER_SIZE;
1385 + int i;
1386 +
1387 + for (i = 0; i < VMA_BLOCKS_PER_EXTENT; i++) {
1388 + uint64_t block_info = GUINT64_FROM_BE(ehead->blockinfo[i]);
1389 + uint64_t cluster_num = block_info & 0xffffffff;
1390 + uint8_t dev_id = (block_info >> 32) & 0xff;
1391 + uint16_t mask = block_info >> (32+16);
1392 + int64_t max_sector;
1393 +
1394 + if (!dev_id) {
1395 + continue;
1396 + }
1397 +
1398 + VmaRestoreState *rstate = &vmar->rstate[dev_id];
1399 + BlockBackend *target = NULL;
1400 +
1401 + if (dev_id != vmar->vmstate_stream) {
1402 + target = rstate->target;
1403 + if (!verify && !target) {
1404 + error_setg(errp, "got wrong dev id %d", dev_id);
1405 + return -1;
1406 + }
1407 +
1408 + if (vma_reader_get_bitmap(rstate, cluster_num)) {
1409 + error_setg(errp, "found duplicated cluster %zd for stream %s",
1410 + cluster_num, vmar->devinfo[dev_id].devname);
1411 + return -1;
1412 + }
1413 + vma_reader_set_bitmap(rstate, cluster_num, 1);
1414 +
1415 + max_sector = vmar->devinfo[dev_id].size/BDRV_SECTOR_SIZE;
1416 + } else {
1417 + max_sector = G_MAXINT64;
1418 + if (cluster_num != vmar->vmstate_clusters) {
1419 + error_setg(errp, "found out of order vmstate data");
1420 + return -1;
1421 + }
1422 + vmar->vmstate_clusters++;
1423 + }
1424 +
1425 + vmar->clusters_read++;
1426 +
1427 + if (verbose) {
1428 + time_t duration = time(NULL) - vmar->start_time;
1429 + int percent = (vmar->clusters_read*100)/vmar->cluster_count;
1430 + if (percent != vmar->clusters_read_per) {
1431 + printf("progress %d%% (read %zd bytes, duration %zd sec)\n",
1432 + percent, vmar->clusters_read*VMA_CLUSTER_SIZE,
1433 + duration);
1434 + fflush(stdout);
1435 + vmar->clusters_read_per = percent;
1436 + }
1437 + }
1438 +
1439 + /* try to write whole clusters to speedup restore */
1440 + if (mask == 0xffff) {
1441 + if ((start + VMA_CLUSTER_SIZE) > extent_size) {
1442 + error_setg(errp, "short vma extent - too many blocks");
1443 + return -1;
1444 + }
1445 + int64_t sector_num = (cluster_num * VMA_CLUSTER_SIZE) /
1446 + BDRV_SECTOR_SIZE;
1447 + int64_t end_sector = sector_num +
1448 + VMA_CLUSTER_SIZE/BDRV_SECTOR_SIZE;
1449 +
1450 + if (end_sector > max_sector) {
1451 + end_sector = max_sector;
1452 + }
1453 +
1454 + if (end_sector <= sector_num) {
1455 + error_setg(errp, "got wrong block address - write bejond end");
1456 + return -1;
1457 + }
1458 +
1459 + if (!verify) {
1460 + int nb_sectors = end_sector - sector_num;
1461 + if (restore_write_data(vmar, dev_id, target, vmstate_fd,
1462 + buf + start, sector_num, nb_sectors,
1463 + errp) < 0) {
1464 + return -1;
1465 + }
1466 + }
1467 +
1468 + start += VMA_CLUSTER_SIZE;
1469 + } else {
1470 + int j;
1471 + int bit = 1;
1472 +
1473 + for (j = 0; j < 16; j++) {
1474 + int64_t sector_num = (cluster_num*VMA_CLUSTER_SIZE +
1475 + j*VMA_BLOCK_SIZE)/BDRV_SECTOR_SIZE;
1476 +
1477 + int64_t end_sector = sector_num +
1478 + VMA_BLOCK_SIZE/BDRV_SECTOR_SIZE;
1479 + if (end_sector > max_sector) {
1480 + end_sector = max_sector;
1481 + }
1482 +
1483 + if (mask & bit) {
1484 + if ((start + VMA_BLOCK_SIZE) > extent_size) {
1485 + error_setg(errp, "short vma extent - too many blocks");
1486 + return -1;
1487 + }
1488 +
1489 + if (end_sector <= sector_num) {
1490 + error_setg(errp, "got wrong block address - "
1491 + "write bejond end");
1492 + return -1;
1493 + }
1494 +
1495 + if (!verify) {
1496 + int nb_sectors = end_sector - sector_num;
1497 + if (restore_write_data(vmar, dev_id, target, vmstate_fd,
1498 + buf + start, sector_num,
1499 + nb_sectors, errp) < 0) {
1500 + return -1;
1501 + }
1502 + }
1503 +
1504 + start += VMA_BLOCK_SIZE;
1505 +
1506 + } else {
1507 +
1508 +
1509 + if (end_sector > sector_num) {
1510 + /* Todo: use bdrv_co_write_zeroes (but that need to
1511 + * be run inside coroutine?)
1512 + */
1513 + int nb_sectors = end_sector - sector_num;
1514 + int zero_size = BDRV_SECTOR_SIZE*nb_sectors;
1515 + vmar->zero_cluster_data += zero_size;
1516 + if (mask != 0) {
1517 + vmar->partial_zero_cluster_data += zero_size;
1518 + }
1519 +
1520 + if (rstate->write_zeroes && !verify) {
1521 + if (restore_write_data(vmar, dev_id, target, vmstate_fd,
1522 + zero_vma_block, sector_num,
1523 + nb_sectors, errp) < 0) {
1524 + return -1;
1525 + }
1526 + }
1527 + }
1528 + }
1529 +
1530 + bit = bit << 1;
1531 + }
1532 + }
1533 + }
1534 +
1535 + if (start != extent_size) {
1536 + error_setg(errp, "vma extent error - missing blocks");
1537 + return -1;
1538 + }
1539 +
1540 + return 0;
1541 +}
1542 +
1543 +static int vma_reader_restore_full(VmaReader *vmar, int vmstate_fd,
1544 + bool verbose, bool verify,
1545 + Error **errp)
1546 +{
1547 + assert(vmar);
1548 + assert(vmar->head_data);
1549 +
1550 + int ret = 0;
1551 + unsigned char buf[VMA_MAX_EXTENT_SIZE];
1552 + int buf_pos = 0;
1553 + unsigned char md5sum[16];
1554 + VmaHeader *h = (VmaHeader *)vmar->head_data;
1555 +
1556 + vmar->start_time = time(NULL);
1557 +
1558 + while (1) {
1559 + int bytes = full_read(vmar->fd, buf + buf_pos, sizeof(buf) - buf_pos);
1560 + if (bytes < 0) {
1561 + error_setg(errp, "read failed - %s", g_strerror(errno));
1562 + return -1;
1563 + }
1564 +
1565 + buf_pos += bytes;
1566 +
1567 + if (!buf_pos) {
1568 + break; /* EOF */
1569 + }
1570 +
1571 + if (buf_pos < VMA_EXTENT_HEADER_SIZE) {
1572 + error_setg(errp, "read short extent (%d bytes)", buf_pos);
1573 + return -1;
1574 + }
1575 +
1576 + VmaExtentHeader *ehead = (VmaExtentHeader *)buf;
1577 +
1578 + /* extract md5sum */
1579 + memcpy(md5sum, ehead->md5sum, sizeof(ehead->md5sum));
1580 + memset(ehead->md5sum, 0, sizeof(ehead->md5sum));
1581 +
1582 + g_checksum_reset(vmar->md5csum);
1583 + g_checksum_update(vmar->md5csum, buf, VMA_EXTENT_HEADER_SIZE);
1584 + gsize csize = 16;
1585 + g_checksum_get_digest(vmar->md5csum, ehead->md5sum, &csize);
1586 +
1587 + if (memcmp(md5sum, ehead->md5sum, 16) != 0) {
1588 + error_setg(errp, "wrong vma extent header chechsum");
1589 + return -1;
1590 + }
1591 +
1592 + if (memcmp(h->uuid, ehead->uuid, sizeof(ehead->uuid)) != 0) {
1593 + error_setg(errp, "wrong vma extent uuid");
1594 + return -1;
1595 + }
1596 +
1597 + if (ehead->magic != VMA_EXTENT_MAGIC || ehead->reserved1 != 0) {
1598 + error_setg(errp, "wrong vma extent header magic");
1599 + return -1;
1600 + }
1601 +
1602 + int block_count = GUINT16_FROM_BE(ehead->block_count);
1603 + int extent_size = VMA_EXTENT_HEADER_SIZE + block_count*VMA_BLOCK_SIZE;
1604 +
1605 + if (buf_pos < extent_size) {
1606 + error_setg(errp, "short vma extent (%d < %d)", buf_pos,
1607 + extent_size);
1608 + return -1;
1609 + }
1610 +
1611 + if (restore_extent(vmar, buf, extent_size, vmstate_fd, verbose,
1612 + verify, errp) < 0) {
1613 + return -1;
1614 + }
1615 +
1616 + if (buf_pos > extent_size) {
1617 + memmove(buf, buf + extent_size, buf_pos - extent_size);
1618 + buf_pos = buf_pos - extent_size;
1619 + } else {
1620 + buf_pos = 0;
1621 + }
1622 + }
1623 +
1624 + bdrv_drain_all();
1625 +
1626 + int i;
1627 + for (i = 1; i < 256; i++) {
1628 + VmaRestoreState *rstate = &vmar->rstate[i];
1629 + if (!rstate->target) {
1630 + continue;
1631 + }
1632 +
1633 + if (blk_flush(rstate->target) < 0) {
1634 + error_setg(errp, "vma blk_flush %s failed",
1635 + vmar->devinfo[i].devname);
1636 + return -1;
1637 + }
1638 +
1639 + if (vmar->devinfo[i].size &&
1640 + (strcmp(vmar->devinfo[i].devname, "vmstate") != 0)) {
1641 + assert(rstate->bitmap);
1642 +
1643 + int64_t cluster_num, end;
1644 +
1645 + end = (vmar->devinfo[i].size + VMA_CLUSTER_SIZE - 1) /
1646 + VMA_CLUSTER_SIZE;
1647 +
1648 + for (cluster_num = 0; cluster_num < end; cluster_num++) {
1649 + if (!vma_reader_get_bitmap(rstate, cluster_num)) {
1650 + error_setg(errp, "detected missing cluster %zd "
1651 + "for stream %s", cluster_num,
1652 + vmar->devinfo[i].devname);
1653 + return -1;
1654 + }
1655 + }
1656 + }
1657 + }
1658 +
1659 + if (verbose) {
1660 + if (vmar->clusters_read) {
1661 + printf("total bytes read %zd, sparse bytes %zd (%.3g%%)\n",
1662 + vmar->clusters_read*VMA_CLUSTER_SIZE,
1663 + vmar->zero_cluster_data,
1664 + (double)(100.0*vmar->zero_cluster_data)/
1665 + (vmar->clusters_read*VMA_CLUSTER_SIZE));
1666 +
1667 + int64_t datasize = vmar->clusters_read*VMA_CLUSTER_SIZE-vmar->zero_cluster_data;
1668 + if (datasize) { // this does not make sense for empty files
1669 + printf("space reduction due to 4K zero blocks %.3g%%\n",
1670 + (double)(100.0*vmar->partial_zero_cluster_data) / datasize);
1671 + }
1672 + } else {
1673 + printf("vma archive contains no image data\n");
1674 + }
1675 + }
1676 + return ret;
1677 +}
1678 +
1679 +int vma_reader_restore(VmaReader *vmar, int vmstate_fd, bool verbose,
1680 + Error **errp)
1681 +{
1682 + return vma_reader_restore_full(vmar, vmstate_fd, verbose, false, errp);
1683 +}
1684 +
1685 +int vma_reader_verify(VmaReader *vmar, bool verbose, Error **errp)
1686 +{
1687 + guint8 dev_id;
1688 +
1689 + for (dev_id = 1; dev_id < 255; dev_id++) {
1690 + if (vma_reader_get_device_info(vmar, dev_id)) {
1691 + allocate_rstate(vmar, dev_id, NULL, false);
1692 + }
1693 + }
1694 +
1695 + return vma_reader_restore_full(vmar, -1, verbose, true, errp);
1696 +}
1697 +
1698 diff --git a/vma-writer.c b/vma-writer.c
1699 new file mode 100644
1700 index 0000000000..9001cbdd2b
1701 --- /dev/null
1702 +++ b/vma-writer.c
1703 @@ -0,0 +1,771 @@
1704 +/*
1705 + * VMA: Virtual Machine Archive
1706 + *
1707 + * Copyright (C) 2012 Proxmox Server Solutions
1708 + *
1709 + * Authors:
1710 + * Dietmar Maurer (dietmar@proxmox.com)
1711 + *
1712 + * This work is licensed under the terms of the GNU GPL, version 2 or later.
1713 + * See the COPYING file in the top-level directory.
1714 + *
1715 + */
1716 +
1717 +#include "qemu/osdep.h"
1718 +#include <glib.h>
1719 +#include <uuid/uuid.h>
1720 +
1721 +#include "vma.h"
1722 +#include "block/block.h"
1723 +#include "monitor/monitor.h"
1724 +#include "qemu/main-loop.h"
1725 +#include "qemu/coroutine.h"
1726 +#include "qemu/cutils.h"
1727 +
1728 +#define DEBUG_VMA 0
1729 +
1730 +#define DPRINTF(fmt, ...)\
1731 + do { if (DEBUG_VMA) { printf("vma: " fmt, ## __VA_ARGS__); } } while (0)
1732 +
1733 +#define WRITE_BUFFERS 5
1734 +#define HEADER_CLUSTERS 8
1735 +#define HEADERBUF_SIZE (VMA_CLUSTER_SIZE*HEADER_CLUSTERS)
1736 +
1737 +struct VmaWriter {
1738 + int fd;
1739 + FILE *cmd;
1740 + int status;
1741 + char errmsg[8192];
1742 + uuid_t uuid;
1743 + bool header_written;
1744 + bool closed;
1745 +
1746 + /* we always write extents */
1747 + unsigned char *outbuf;
1748 + int outbuf_pos; /* in bytes */
1749 + int outbuf_count; /* in VMA_BLOCKS */
1750 + uint64_t outbuf_block_info[VMA_BLOCKS_PER_EXTENT];
1751 +
1752 + unsigned char *headerbuf;
1753 +
1754 + GChecksum *md5csum;
1755 + CoMutex flush_lock;
1756 + Coroutine *co_writer;
1757 +
1758 + /* drive informations */
1759 + VmaStreamInfo stream_info[256];
1760 + guint stream_count;
1761 +
1762 + guint8 vmstate_stream;
1763 + uint32_t vmstate_clusters;
1764 +
1765 + /* header blob table */
1766 + char *header_blob_table;
1767 + uint32_t header_blob_table_size;
1768 + uint32_t header_blob_table_pos;
1769 +
1770 + /* store for config blobs */
1771 + uint32_t config_names[VMA_MAX_CONFIGS]; /* offset into blob_buffer table */
1772 + uint32_t config_data[VMA_MAX_CONFIGS]; /* offset into blob_buffer table */
1773 + uint32_t config_count;
1774 +};
1775 +
1776 +void vma_writer_set_error(VmaWriter *vmaw, const char *fmt, ...)
1777 +{
1778 + va_list ap;
1779 +
1780 + if (vmaw->status < 0) {
1781 + return;
1782 + }
1783 +
1784 + vmaw->status = -1;
1785 +
1786 + va_start(ap, fmt);
1787 + g_vsnprintf(vmaw->errmsg, sizeof(vmaw->errmsg), fmt, ap);
1788 + va_end(ap);
1789 +
1790 + DPRINTF("vma_writer_set_error: %s\n", vmaw->errmsg);
1791 +}
1792 +
1793 +static uint32_t allocate_header_blob(VmaWriter *vmaw, const char *data,
1794 + size_t len)
1795 +{
1796 + if (len > 65535) {
1797 + return 0;
1798 + }
1799 +
1800 + if (!vmaw->header_blob_table ||
1801 + (vmaw->header_blob_table_size <
1802 + (vmaw->header_blob_table_pos + len + 2))) {
1803 + int newsize = vmaw->header_blob_table_size + ((len + 2 + 511)/512)*512;
1804 +
1805 + vmaw->header_blob_table = g_realloc(vmaw->header_blob_table, newsize);
1806 + memset(vmaw->header_blob_table + vmaw->header_blob_table_size,
1807 + 0, newsize - vmaw->header_blob_table_size);
1808 + vmaw->header_blob_table_size = newsize;
1809 + }
1810 +
1811 + uint32_t cpos = vmaw->header_blob_table_pos;
1812 + vmaw->header_blob_table[cpos] = len & 255;
1813 + vmaw->header_blob_table[cpos+1] = (len >> 8) & 255;
1814 + memcpy(vmaw->header_blob_table + cpos + 2, data, len);
1815 + vmaw->header_blob_table_pos += len + 2;
1816 + return cpos;
1817 +}
1818 +
1819 +static uint32_t allocate_header_string(VmaWriter *vmaw, const char *str)
1820 +{
1821 + assert(vmaw);
1822 +
1823 + size_t len = strlen(str) + 1;
1824 +
1825 + return allocate_header_blob(vmaw, str, len);
1826 +}
1827 +
1828 +int vma_writer_add_config(VmaWriter *vmaw, const char *name, gpointer data,
1829 + gsize len)
1830 +{
1831 + assert(vmaw);
1832 + assert(!vmaw->header_written);
1833 + assert(vmaw->config_count < VMA_MAX_CONFIGS);
1834 + assert(name);
1835 + assert(data);
1836 +
1837 + gchar *basename = g_path_get_basename(name);
1838 + uint32_t name_ptr = allocate_header_string(vmaw, basename);
1839 + g_free(basename);
1840 +
1841 + if (!name_ptr) {
1842 + return -1;
1843 + }
1844 +
1845 + uint32_t data_ptr = allocate_header_blob(vmaw, data, len);
1846 + if (!data_ptr) {
1847 + return -1;
1848 + }
1849 +
1850 + vmaw->config_names[vmaw->config_count] = name_ptr;
1851 + vmaw->config_data[vmaw->config_count] = data_ptr;
1852 +
1853 + vmaw->config_count++;
1854 +
1855 + return 0;
1856 +}
1857 +
1858 +int vma_writer_register_stream(VmaWriter *vmaw, const char *devname,
1859 + size_t size)
1860 +{
1861 + assert(vmaw);
1862 + assert(devname);
1863 + assert(!vmaw->status);
1864 +
1865 + if (vmaw->header_written) {
1866 + vma_writer_set_error(vmaw, "vma_writer_register_stream: header "
1867 + "already written");
1868 + return -1;
1869 + }
1870 +
1871 + guint n = vmaw->stream_count + 1;
1872 +
1873 + /* we can have dev_ids form 1 to 255 (0 reserved)
1874 + * 255(-1) reseverd for safety
1875 + */
1876 + if (n > 254) {
1877 + vma_writer_set_error(vmaw, "vma_writer_register_stream: "
1878 + "too many drives");
1879 + return -1;
1880 + }
1881 +
1882 + if (size <= 0) {
1883 + vma_writer_set_error(vmaw, "vma_writer_register_stream: "
1884 + "got strange size %zd", size);
1885 + return -1;
1886 + }
1887 +
1888 + DPRINTF("vma_writer_register_stream %s %zu %d\n", devname, size, n);
1889 +
1890 + vmaw->stream_info[n].devname = g_strdup(devname);
1891 + vmaw->stream_info[n].size = size;
1892 +
1893 + vmaw->stream_info[n].cluster_count = (size + VMA_CLUSTER_SIZE - 1) /
1894 + VMA_CLUSTER_SIZE;
1895 +
1896 + vmaw->stream_count = n;
1897 +
1898 + if (strcmp(devname, "vmstate") == 0) {
1899 + vmaw->vmstate_stream = n;
1900 + }
1901 +
1902 + return n;
1903 +}
1904 +
1905 +static void vma_co_continue_write(void *opaque)
1906 +{
1907 + VmaWriter *vmaw = opaque;
1908 +
1909 + DPRINTF("vma_co_continue_write\n");
1910 + qemu_coroutine_enter(vmaw->co_writer);
1911 +}
1912 +
1913 +static ssize_t coroutine_fn
1914 +vma_queue_write(VmaWriter *vmaw, const void *buf, size_t bytes)
1915 +{
1916 + DPRINTF("vma_queue_write enter %zd\n", bytes);
1917 +
1918 + assert(vmaw);
1919 + assert(buf);
1920 + assert(bytes <= VMA_MAX_EXTENT_SIZE);
1921 +
1922 + size_t done = 0;
1923 + ssize_t ret;
1924 +
1925 + assert(vmaw->co_writer == NULL);
1926 +
1927 + vmaw->co_writer = qemu_coroutine_self();
1928 +
1929 + while (done < bytes) {
1930 + aio_set_fd_handler(qemu_get_aio_context(), vmaw->fd, false, NULL, vma_co_continue_write, NULL, vmaw);
1931 + qemu_coroutine_yield();
1932 + aio_set_fd_handler(qemu_get_aio_context(), vmaw->fd, false, NULL, NULL, NULL, NULL);
1933 + if (vmaw->status < 0) {
1934 + DPRINTF("vma_queue_write detected canceled backup\n");
1935 + done = -1;
1936 + break;
1937 + }
1938 + ret = write(vmaw->fd, buf + done, bytes - done);
1939 + if (ret > 0) {
1940 + done += ret;
1941 + DPRINTF("vma_queue_write written %zd %zd\n", done, ret);
1942 + } else if (ret < 0) {
1943 + if (errno == EAGAIN || errno == EWOULDBLOCK) {
1944 + /* try again */
1945 + } else {
1946 + vma_writer_set_error(vmaw, "vma_queue_write: write error - %s",
1947 + g_strerror(errno));
1948 + done = -1; /* always return failure for partial writes */
1949 + break;
1950 + }
1951 + } else if (ret == 0) {
1952 + /* should not happen - simply try again */
1953 + }
1954 + }
1955 +
1956 + vmaw->co_writer = NULL;
1957 +
1958 + return (done == bytes) ? bytes : -1;
1959 +}
1960 +
1961 +VmaWriter *vma_writer_create(const char *filename, uuid_t uuid, Error **errp)
1962 +{
1963 + const char *p;
1964 +
1965 + assert(sizeof(VmaHeader) == (4096 + 8192));
1966 + assert(G_STRUCT_OFFSET(VmaHeader, config_names) == 2044);
1967 + assert(G_STRUCT_OFFSET(VmaHeader, config_data) == 3068);
1968 + assert(G_STRUCT_OFFSET(VmaHeader, dev_info) == 4096);
1969 + assert(sizeof(VmaExtentHeader) == 512);
1970 +
1971 + VmaWriter *vmaw = g_new0(VmaWriter, 1);
1972 + vmaw->fd = -1;
1973 +
1974 + vmaw->md5csum = g_checksum_new(G_CHECKSUM_MD5);
1975 + if (!vmaw->md5csum) {
1976 + error_setg(errp, "can't allocate cmsum\n");
1977 + goto err;
1978 + }
1979 +
1980 + if (strstart(filename, "exec:", &p)) {
1981 + vmaw->cmd = popen(p, "w");
1982 + if (vmaw->cmd == NULL) {
1983 + error_setg(errp, "can't popen command '%s' - %s\n", p,
1984 + g_strerror(errno));
1985 + goto err;
1986 + }
1987 + vmaw->fd = fileno(vmaw->cmd);
1988 +
1989 + /* try to use O_NONBLOCK */
1990 + fcntl(vmaw->fd, F_SETFL, fcntl(vmaw->fd, F_GETFL)|O_NONBLOCK);
1991 +
1992 + } else {
1993 + struct stat st;
1994 + int oflags;
1995 + const char *tmp_id_str;
1996 +
1997 + if ((stat(filename, &st) == 0) && S_ISFIFO(st.st_mode)) {
1998 + oflags = O_NONBLOCK|O_WRONLY;
1999 + vmaw->fd = qemu_open(filename, oflags, 0644);
2000 + } else if (strstart(filename, "/dev/fdset/", &tmp_id_str)) {
2001 + oflags = O_NONBLOCK|O_WRONLY;
2002 + vmaw->fd = qemu_open(filename, oflags, 0644);
2003 + } else if (strstart(filename, "/dev/fdname/", &tmp_id_str)) {
2004 + vmaw->fd = monitor_get_fd(cur_mon, tmp_id_str, errp);
2005 + if (vmaw->fd < 0) {
2006 + goto err;
2007 + }
2008 + /* try to use O_NONBLOCK */
2009 + fcntl(vmaw->fd, F_SETFL, fcntl(vmaw->fd, F_GETFL)|O_NONBLOCK);
2010 + } else {
2011 + oflags = O_NONBLOCK|O_DIRECT|O_WRONLY|O_CREAT|O_EXCL;
2012 + vmaw->fd = qemu_open(filename, oflags, 0644);
2013 + }
2014 +
2015 + if (vmaw->fd < 0) {
2016 + error_setg(errp, "can't open file %s - %s\n", filename,
2017 + g_strerror(errno));
2018 + goto err;
2019 + }
2020 + }
2021 +
2022 + /* we use O_DIRECT, so we need to align IO buffers */
2023 +
2024 + vmaw->outbuf = qemu_memalign(512, VMA_MAX_EXTENT_SIZE);
2025 + vmaw->headerbuf = qemu_memalign(512, HEADERBUF_SIZE);
2026 +
2027 + vmaw->outbuf_count = 0;
2028 + vmaw->outbuf_pos = VMA_EXTENT_HEADER_SIZE;
2029 +
2030 + vmaw->header_blob_table_pos = 1; /* start at pos 1 */
2031 +
2032 + qemu_co_mutex_init(&vmaw->flush_lock);
2033 +
2034 + uuid_copy(vmaw->uuid, uuid);
2035 +
2036 + return vmaw;
2037 +
2038 +err:
2039 + if (vmaw) {
2040 + if (vmaw->cmd) {
2041 + pclose(vmaw->cmd);
2042 + } else if (vmaw->fd >= 0) {
2043 + close(vmaw->fd);
2044 + }
2045 +
2046 + if (vmaw->md5csum) {
2047 + g_checksum_free(vmaw->md5csum);
2048 + }
2049 +
2050 + g_free(vmaw);
2051 + }
2052 +
2053 + return NULL;
2054 +}
2055 +
2056 +static int coroutine_fn vma_write_header(VmaWriter *vmaw)
2057 +{
2058 + assert(vmaw);
2059 + unsigned char *buf = vmaw->headerbuf;
2060 + VmaHeader *head = (VmaHeader *)buf;
2061 +
2062 + int i;
2063 +
2064 + DPRINTF("VMA WRITE HEADER\n");
2065 +
2066 + if (vmaw->status < 0) {
2067 + return vmaw->status;
2068 + }
2069 +
2070 + memset(buf, 0, HEADERBUF_SIZE);
2071 +
2072 + head->magic = VMA_MAGIC;
2073 + head->version = GUINT32_TO_BE(1); /* v1 */
2074 + memcpy(head->uuid, vmaw->uuid, 16);
2075 +
2076 + time_t ctime = time(NULL);
2077 + head->ctime = GUINT64_TO_BE(ctime);
2078 +
2079 + for (i = 0; i < VMA_MAX_CONFIGS; i++) {
2080 + head->config_names[i] = GUINT32_TO_BE(vmaw->config_names[i]);
2081 + head->config_data[i] = GUINT32_TO_BE(vmaw->config_data[i]);
2082 + }
2083 +
2084 + /* 32 bytes per device (12 used currently) = 8192 bytes max */
2085 + for (i = 1; i <= 254; i++) {
2086 + VmaStreamInfo *si = &vmaw->stream_info[i];
2087 + if (si->size) {
2088 + assert(si->devname);
2089 + uint32_t devname_ptr = allocate_header_string(vmaw, si->devname);
2090 + if (!devname_ptr) {
2091 + return -1;
2092 + }
2093 + head->dev_info[i].devname_ptr = GUINT32_TO_BE(devname_ptr);
2094 + head->dev_info[i].size = GUINT64_TO_BE(si->size);
2095 + }
2096 + }
2097 +
2098 + uint32_t header_size = sizeof(VmaHeader) + vmaw->header_blob_table_size;
2099 + head->header_size = GUINT32_TO_BE(header_size);
2100 +
2101 + if (header_size > HEADERBUF_SIZE) {
2102 + return -1; /* just to be sure */
2103 + }
2104 +
2105 + uint32_t blob_buffer_offset = sizeof(VmaHeader);
2106 + memcpy(buf + blob_buffer_offset, vmaw->header_blob_table,
2107 + vmaw->header_blob_table_size);
2108 + head->blob_buffer_offset = GUINT32_TO_BE(blob_buffer_offset);
2109 + head->blob_buffer_size = GUINT32_TO_BE(vmaw->header_blob_table_pos);
2110 +
2111 + g_checksum_reset(vmaw->md5csum);
2112 + g_checksum_update(vmaw->md5csum, (const guchar *)buf, header_size);
2113 + gsize csize = 16;
2114 + g_checksum_get_digest(vmaw->md5csum, (guint8 *)(head->md5sum), &csize);
2115 +
2116 + return vma_queue_write(vmaw, buf, header_size);
2117 +}
2118 +
2119 +static int coroutine_fn vma_writer_flush(VmaWriter *vmaw)
2120 +{
2121 + assert(vmaw);
2122 +
2123 + int ret;
2124 + int i;
2125 +
2126 + if (vmaw->status < 0) {
2127 + return vmaw->status;
2128 + }
2129 +
2130 + if (!vmaw->header_written) {
2131 + vmaw->header_written = true;
2132 + ret = vma_write_header(vmaw);
2133 + if (ret < 0) {
2134 + vma_writer_set_error(vmaw, "vma_writer_flush: write header failed");
2135 + return ret;
2136 + }
2137 + }
2138 +
2139 + DPRINTF("VMA WRITE FLUSH %d %d\n", vmaw->outbuf_count, vmaw->outbuf_pos);
2140 +
2141 +
2142 + VmaExtentHeader *ehead = (VmaExtentHeader *)vmaw->outbuf;
2143 +
2144 + ehead->magic = VMA_EXTENT_MAGIC;
2145 + ehead->reserved1 = 0;
2146 +
2147 + for (i = 0; i < VMA_BLOCKS_PER_EXTENT; i++) {
2148 + ehead->blockinfo[i] = GUINT64_TO_BE(vmaw->outbuf_block_info[i]);
2149 + }
2150 +
2151 + guint16 block_count = (vmaw->outbuf_pos - VMA_EXTENT_HEADER_SIZE) /
2152 + VMA_BLOCK_SIZE;
2153 +
2154 + ehead->block_count = GUINT16_TO_BE(block_count);
2155 +
2156 + memcpy(ehead->uuid, vmaw->uuid, sizeof(ehead->uuid));
2157 + memset(ehead->md5sum, 0, sizeof(ehead->md5sum));
2158 +
2159 + g_checksum_reset(vmaw->md5csum);
2160 + g_checksum_update(vmaw->md5csum, vmaw->outbuf, VMA_EXTENT_HEADER_SIZE);
2161 + gsize csize = 16;
2162 + g_checksum_get_digest(vmaw->md5csum, ehead->md5sum, &csize);
2163 +
2164 + int bytes = vmaw->outbuf_pos;
2165 + ret = vma_queue_write(vmaw, vmaw->outbuf, bytes);
2166 + if (ret != bytes) {
2167 + vma_writer_set_error(vmaw, "vma_writer_flush: failed write");
2168 + }
2169 +
2170 + vmaw->outbuf_count = 0;
2171 + vmaw->outbuf_pos = VMA_EXTENT_HEADER_SIZE;
2172 +
2173 + for (i = 0; i < VMA_BLOCKS_PER_EXTENT; i++) {
2174 + vmaw->outbuf_block_info[i] = 0;
2175 + }
2176 +
2177 + return vmaw->status;
2178 +}
2179 +
2180 +static int vma_count_open_streams(VmaWriter *vmaw)
2181 +{
2182 + g_assert(vmaw != NULL);
2183 +
2184 + int i;
2185 + int open_drives = 0;
2186 + for (i = 0; i <= 255; i++) {
2187 + if (vmaw->stream_info[i].size && !vmaw->stream_info[i].finished) {
2188 + open_drives++;
2189 + }
2190 + }
2191 +
2192 + return open_drives;
2193 +}
2194 +
2195 +
2196 +/**
2197 + * You need to call this if the vma archive does not contain
2198 + * any data stream.
2199 + */
2200 +int coroutine_fn
2201 +vma_writer_flush_output(VmaWriter *vmaw)
2202 +{
2203 + qemu_co_mutex_lock(&vmaw->flush_lock);
2204 + int ret = vma_writer_flush(vmaw);
2205 + qemu_co_mutex_unlock(&vmaw->flush_lock);
2206 + if (ret < 0) {
2207 + vma_writer_set_error(vmaw, "vma_writer_flush_header failed");
2208 + }
2209 + return ret;
2210 +}
2211 +
2212 +/**
2213 + * all jobs should call this when there is no more data
2214 + * Returns: number of remaining stream (0 ==> finished)
2215 + */
2216 +int coroutine_fn
2217 +vma_writer_close_stream(VmaWriter *vmaw, uint8_t dev_id)
2218 +{
2219 + g_assert(vmaw != NULL);
2220 +
2221 + DPRINTF("vma_writer_set_status %d\n", dev_id);
2222 + if (!vmaw->stream_info[dev_id].size) {
2223 + vma_writer_set_error(vmaw, "vma_writer_close_stream: "
2224 + "no such stream %d", dev_id);
2225 + return -1;
2226 + }
2227 + if (vmaw->stream_info[dev_id].finished) {
2228 + vma_writer_set_error(vmaw, "vma_writer_close_stream: "
2229 + "stream already closed %d", dev_id);
2230 + return -1;
2231 + }
2232 +
2233 + vmaw->stream_info[dev_id].finished = true;
2234 +
2235 + int open_drives = vma_count_open_streams(vmaw);
2236 +
2237 + if (open_drives <= 0) {
2238 + DPRINTF("vma_writer_set_status all drives completed\n");
2239 + vma_writer_flush_output(vmaw);
2240 + }
2241 +
2242 + return open_drives;
2243 +}
2244 +
2245 +int vma_writer_get_status(VmaWriter *vmaw, VmaStatus *status)
2246 +{
2247 + int i;
2248 +
2249 + g_assert(vmaw != NULL);
2250 +
2251 + if (status) {
2252 + status->status = vmaw->status;
2253 + g_strlcpy(status->errmsg, vmaw->errmsg, sizeof(status->errmsg));
2254 + for (i = 0; i <= 255; i++) {
2255 + status->stream_info[i] = vmaw->stream_info[i];
2256 + }
2257 +
2258 + uuid_unparse_lower(vmaw->uuid, status->uuid_str);
2259 + }
2260 +
2261 + status->closed = vmaw->closed;
2262 +
2263 + return vmaw->status;
2264 +}
2265 +
2266 +static int vma_writer_get_buffer(VmaWriter *vmaw)
2267 +{
2268 + int ret = 0;
2269 +
2270 + qemu_co_mutex_lock(&vmaw->flush_lock);
2271 +
2272 + /* wait until buffer is available */
2273 + while (vmaw->outbuf_count >= (VMA_BLOCKS_PER_EXTENT - 1)) {
2274 + ret = vma_writer_flush(vmaw);
2275 + if (ret < 0) {
2276 + vma_writer_set_error(vmaw, "vma_writer_get_buffer: flush failed");
2277 + break;
2278 + }
2279 + }
2280 +
2281 + qemu_co_mutex_unlock(&vmaw->flush_lock);
2282 +
2283 + return ret;
2284 +}
2285 +
2286 +
2287 +int64_t coroutine_fn
2288 +vma_writer_write(VmaWriter *vmaw, uint8_t dev_id, int64_t cluster_num,
2289 + unsigned char *buf, size_t *zero_bytes)
2290 +{
2291 + g_assert(vmaw != NULL);
2292 + g_assert(zero_bytes != NULL);
2293 +
2294 + *zero_bytes = 0;
2295 +
2296 + if (vmaw->status < 0) {
2297 + return vmaw->status;
2298 + }
2299 +
2300 + if (!dev_id || !vmaw->stream_info[dev_id].size) {
2301 + vma_writer_set_error(vmaw, "vma_writer_write: "
2302 + "no such stream %d", dev_id);
2303 + return -1;
2304 + }
2305 +
2306 + if (vmaw->stream_info[dev_id].finished) {
2307 + vma_writer_set_error(vmaw, "vma_writer_write: "
2308 + "stream already closed %d", dev_id);
2309 + return -1;
2310 + }
2311 +
2312 +
2313 + if (cluster_num >= (((uint64_t)1)<<32)) {
2314 + vma_writer_set_error(vmaw, "vma_writer_write: "
2315 + "cluster number out of range");
2316 + return -1;
2317 + }
2318 +
2319 + if (dev_id == vmaw->vmstate_stream) {
2320 + if (cluster_num != vmaw->vmstate_clusters) {
2321 + vma_writer_set_error(vmaw, "vma_writer_write: "
2322 + "non sequential vmstate write");
2323 + }
2324 + vmaw->vmstate_clusters++;
2325 + } else if (cluster_num >= vmaw->stream_info[dev_id].cluster_count) {
2326 + vma_writer_set_error(vmaw, "vma_writer_write: cluster number too big");
2327 + return -1;
2328 + }
2329 +
2330 + /* wait until buffer is available */
2331 + if (vma_writer_get_buffer(vmaw) < 0) {
2332 + vma_writer_set_error(vmaw, "vma_writer_write: "
2333 + "vma_writer_get_buffer failed");
2334 + return -1;
2335 + }
2336 +
2337 + DPRINTF("VMA WRITE %d %zd\n", dev_id, cluster_num);
2338 +
2339 + uint16_t mask = 0;
2340 +
2341 + if (buf) {
2342 + int i;
2343 + int bit = 1;
2344 + for (i = 0; i < 16; i++) {
2345 + unsigned char *vmablock = buf + (i*VMA_BLOCK_SIZE);
2346 + if (!buffer_is_zero(vmablock, VMA_BLOCK_SIZE)) {
2347 + mask |= bit;
2348 + memcpy(vmaw->outbuf + vmaw->outbuf_pos, vmablock,
2349 + VMA_BLOCK_SIZE);
2350 + vmaw->outbuf_pos += VMA_BLOCK_SIZE;
2351 + } else {
2352 + DPRINTF("VMA WRITE %zd ZERO BLOCK %d\n", cluster_num, i);
2353 + vmaw->stream_info[dev_id].zero_bytes += VMA_BLOCK_SIZE;
2354 + *zero_bytes += VMA_BLOCK_SIZE;
2355 + }
2356 +
2357 + bit = bit << 1;
2358 + }
2359 + } else {
2360 + DPRINTF("VMA WRITE %zd ZERO CLUSTER\n", cluster_num);
2361 + vmaw->stream_info[dev_id].zero_bytes += VMA_CLUSTER_SIZE;
2362 + *zero_bytes += VMA_CLUSTER_SIZE;
2363 + }
2364 +
2365 + uint64_t block_info = ((uint64_t)mask) << (32+16);
2366 + block_info |= ((uint64_t)dev_id) << 32;
2367 + block_info |= (cluster_num & 0xffffffff);
2368 + vmaw->outbuf_block_info[vmaw->outbuf_count] = block_info;
2369 +
2370 + DPRINTF("VMA WRITE MASK %zd %zx\n", cluster_num, block_info);
2371 +
2372 + vmaw->outbuf_count++;
2373 +
2374 + /** NOTE: We allways write whole clusters, but we correctly set
2375 + * transferred bytes. So transferred == size when when everything
2376 + * went OK.
2377 + */
2378 + size_t transferred = VMA_CLUSTER_SIZE;
2379 +
2380 + if (dev_id != vmaw->vmstate_stream) {
2381 + uint64_t last = (cluster_num + 1) * VMA_CLUSTER_SIZE;
2382 + if (last > vmaw->stream_info[dev_id].size) {
2383 + uint64_t diff = last - vmaw->stream_info[dev_id].size;
2384 + if (diff >= VMA_CLUSTER_SIZE) {
2385 + vma_writer_set_error(vmaw, "vma_writer_write: "
2386 + "read after last cluster");
2387 + return -1;
2388 + }
2389 + transferred -= diff;
2390 + }
2391 + }
2392 +
2393 + vmaw->stream_info[dev_id].transferred += transferred;
2394 +
2395 + return transferred;
2396 +}
2397 +
2398 +void vma_writer_error_propagate(VmaWriter *vmaw, Error **errp)
2399 +{
2400 + if (vmaw->status < 0 && *errp == NULL) {
2401 + error_setg(errp, "%s", vmaw->errmsg);
2402 + }
2403 +}
2404 +
2405 +int vma_writer_close(VmaWriter *vmaw, Error **errp)
2406 +{
2407 + g_assert(vmaw != NULL);
2408 +
2409 + int i;
2410 +
2411 + while (vmaw->co_writer) {
2412 + aio_poll(qemu_get_aio_context(), true);
2413 + }
2414 +
2415 + assert(vmaw->co_writer == NULL);
2416 +
2417 + if (vmaw->cmd) {
2418 + if (pclose(vmaw->cmd) < 0) {
2419 + vma_writer_set_error(vmaw, "vma_writer_close: "
2420 + "pclose failed - %s", g_strerror(errno));
2421 + }
2422 + } else {
2423 + if (close(vmaw->fd) < 0) {
2424 + vma_writer_set_error(vmaw, "vma_writer_close: "
2425 + "close failed - %s", g_strerror(errno));
2426 + }
2427 + }
2428 +
2429 + for (i = 0; i <= 255; i++) {
2430 + VmaStreamInfo *si = &vmaw->stream_info[i];
2431 + if (si->size) {
2432 + if (!si->finished) {
2433 + vma_writer_set_error(vmaw, "vma_writer_close: "
2434 + "detected open stream '%s'", si->devname);
2435 + } else if ((si->transferred != si->size) &&
2436 + (i != vmaw->vmstate_stream)) {
2437 + vma_writer_set_error(vmaw, "vma_writer_close: "
2438 + "incomplete stream '%s' (%zd != %zd)",
2439 + si->devname, si->transferred, si->size);
2440 + }
2441 + }
2442 + }
2443 +
2444 + for (i = 0; i <= 255; i++) {
2445 + vmaw->stream_info[i].finished = 1; /* mark as closed */
2446 + }
2447 +
2448 + vmaw->closed = 1;
2449 +
2450 + if (vmaw->status < 0 && *errp == NULL) {
2451 + error_setg(errp, "%s", vmaw->errmsg);
2452 + }
2453 +
2454 + return vmaw->status;
2455 +}
2456 +
2457 +void vma_writer_destroy(VmaWriter *vmaw)
2458 +{
2459 + assert(vmaw);
2460 +
2461 + int i;
2462 +
2463 + for (i = 0; i <= 255; i++) {
2464 + if (vmaw->stream_info[i].devname) {
2465 + g_free(vmaw->stream_info[i].devname);
2466 + }
2467 + }
2468 +
2469 + if (vmaw->md5csum) {
2470 + g_checksum_free(vmaw->md5csum);
2471 + }
2472 +
2473 + g_free(vmaw);
2474 +}
2475 diff --git a/vma.c b/vma.c
2476 new file mode 100644
2477 index 0000000000..04915427c8
2478 --- /dev/null
2479 +++ b/vma.c
2480 @@ -0,0 +1,757 @@
2481 +/*
2482 + * VMA: Virtual Machine Archive
2483 + *
2484 + * Copyright (C) 2012-2013 Proxmox Server Solutions
2485 + *
2486 + * Authors:
2487 + * Dietmar Maurer (dietmar@proxmox.com)
2488 + *
2489 + * This work is licensed under the terms of the GNU GPL, version 2 or later.
2490 + * See the COPYING file in the top-level directory.
2491 + *
2492 + */
2493 +
2494 +#include "qemu/osdep.h"
2495 +#include <glib.h>
2496 +
2497 +#include "vma.h"
2498 +#include "qemu-common.h"
2499 +#include "qemu/error-report.h"
2500 +#include "qemu/main-loop.h"
2501 +#include "qapi/qmp/qstring.h"
2502 +#include "sysemu/char.h" /* qstring_from_str */
2503 +#include "sysemu/block-backend.h"
2504 +
2505 +static void help(void)
2506 +{
2507 + const char *help_msg =
2508 + "usage: vma command [command options]\n"
2509 + "\n"
2510 + "vma list <filename>\n"
2511 + "vma config <filename> [-c config]\n"
2512 + "vma create <filename> [-c config] pathname ...\n"
2513 + "vma extract <filename> [-r <fifo>] <targetdir>\n"
2514 + "vma verify <filename> [-v]\n"
2515 + ;
2516 +
2517 + printf("%s", help_msg);
2518 + exit(1);
2519 +}
2520 +
2521 +static const char *extract_devname(const char *path, char **devname, int index)
2522 +{
2523 + assert(path);
2524 +
2525 + const char *sep = strchr(path, '=');
2526 +
2527 + if (sep) {
2528 + *devname = g_strndup(path, sep - path);
2529 + path = sep + 1;
2530 + } else {
2531 + if (index >= 0) {
2532 + *devname = g_strdup_printf("disk%d", index);
2533 + } else {
2534 + *devname = NULL;
2535 + }
2536 + }
2537 +
2538 + return path;
2539 +}
2540 +
2541 +static void print_content(VmaReader *vmar)
2542 +{
2543 + assert(vmar);
2544 +
2545 + VmaHeader *head = vma_reader_get_header(vmar);
2546 +
2547 + GList *l = vma_reader_get_config_data(vmar);
2548 + while (l && l->data) {
2549 + VmaConfigData *cdata = (VmaConfigData *)l->data;
2550 + l = g_list_next(l);
2551 + printf("CFG: size: %d name: %s\n", cdata->len, cdata->name);
2552 + }
2553 +
2554 + int i;
2555 + VmaDeviceInfo *di;
2556 + for (i = 1; i < 255; i++) {
2557 + di = vma_reader_get_device_info(vmar, i);
2558 + if (di) {
2559 + if (strcmp(di->devname, "vmstate") == 0) {
2560 + printf("VMSTATE: dev_id=%d memory: %zd\n", i, di->size);
2561 + } else {
2562 + printf("DEV: dev_id=%d size: %zd devname: %s\n",
2563 + i, di->size, di->devname);
2564 + }
2565 + }
2566 + }
2567 + /* ctime is the last entry we print */
2568 + printf("CTIME: %s", ctime(&head->ctime));
2569 + fflush(stdout);
2570 +}
2571 +
2572 +static int list_content(int argc, char **argv)
2573 +{
2574 + int c, ret = 0;
2575 + const char *filename;
2576 +
2577 + for (;;) {
2578 + c = getopt(argc, argv, "h");
2579 + if (c == -1) {
2580 + break;
2581 + }
2582 + switch (c) {
2583 + case '?':
2584 + case 'h':
2585 + help();
2586 + break;
2587 + default:
2588 + g_assert_not_reached();
2589 + }
2590 + }
2591 +
2592 + /* Get the filename */
2593 + if ((optind + 1) != argc) {
2594 + help();
2595 + }
2596 + filename = argv[optind++];
2597 +
2598 + Error *errp = NULL;
2599 + VmaReader *vmar = vma_reader_create(filename, &errp);
2600 +
2601 + if (!vmar) {
2602 + g_error("%s", error_get_pretty(errp));
2603 + }
2604 +
2605 + print_content(vmar);
2606 +
2607 + vma_reader_destroy(vmar);
2608 +
2609 + return ret;
2610 +}
2611 +
2612 +typedef struct RestoreMap {
2613 + char *devname;
2614 + char *path;
2615 + char *format;
2616 + bool write_zero;
2617 +} RestoreMap;
2618 +
2619 +static int extract_content(int argc, char **argv)
2620 +{
2621 + int c, ret = 0;
2622 + int verbose = 0;
2623 + const char *filename;
2624 + const char *dirname;
2625 + const char *readmap = NULL;
2626 +
2627 + for (;;) {
2628 + c = getopt(argc, argv, "hvr:");
2629 + if (c == -1) {
2630 + break;
2631 + }
2632 + switch (c) {
2633 + case '?':
2634 + case 'h':
2635 + help();
2636 + break;
2637 + case 'r':
2638 + readmap = optarg;
2639 + break;
2640 + case 'v':
2641 + verbose = 1;
2642 + break;
2643 + default:
2644 + help();
2645 + }
2646 + }
2647 +
2648 + /* Get the filename */
2649 + if ((optind + 2) != argc) {
2650 + help();
2651 + }
2652 + filename = argv[optind++];
2653 + dirname = argv[optind++];
2654 +
2655 + Error *errp = NULL;
2656 + VmaReader *vmar = vma_reader_create(filename, &errp);
2657 +
2658 + if (!vmar) {
2659 + g_error("%s", error_get_pretty(errp));
2660 + }
2661 +
2662 + if (mkdir(dirname, 0777) < 0) {
2663 + g_error("unable to create target directory %s - %s",
2664 + dirname, g_strerror(errno));
2665 + }
2666 +
2667 + GList *l = vma_reader_get_config_data(vmar);
2668 + while (l && l->data) {
2669 + VmaConfigData *cdata = (VmaConfigData *)l->data;
2670 + l = g_list_next(l);
2671 + char *cfgfn = g_strdup_printf("%s/%s", dirname, cdata->name);
2672 + GError *err = NULL;
2673 + if (!g_file_set_contents(cfgfn, (gchar *)cdata->data, cdata->len,
2674 + &err)) {
2675 + g_error("unable to write file: %s", err->message);
2676 + }
2677 + }
2678 +
2679 + GHashTable *devmap = g_hash_table_new(g_str_hash, g_str_equal);
2680 +
2681 + if (readmap) {
2682 + print_content(vmar);
2683 +
2684 + FILE *map = fopen(readmap, "r");
2685 + if (!map) {
2686 + g_error("unable to open fifo %s - %s", readmap, g_strerror(errno));
2687 + }
2688 +
2689 + while (1) {
2690 + char inbuf[8192];
2691 + char *line = fgets(inbuf, sizeof(inbuf), map);
2692 + if (!line || line[0] == '\0' || !strcmp(line, "done\n")) {
2693 + break;
2694 + }
2695 + int len = strlen(line);
2696 + if (line[len - 1] == '\n') {
2697 + line[len - 1] = '\0';
2698 + if (len == 1) {
2699 + break;
2700 + }
2701 + }
2702 +
2703 + char *format = NULL;
2704 + if (strncmp(line, "format=", sizeof("format=")-1) == 0) {
2705 + format = line + sizeof("format=")-1;
2706 + char *colon = strchr(format, ':');
2707 + if (!colon) {
2708 + g_error("read map failed - found only a format ('%s')", inbuf);
2709 + }
2710 + format = g_strndup(format, colon - format);
2711 + line = colon+1;
2712 + }
2713 +
2714 + const char *path;
2715 + bool write_zero;
2716 + if (line[0] == '0' && line[1] == ':') {
2717 + path = line + 2;
2718 + write_zero = false;
2719 + } else if (line[0] == '1' && line[1] == ':') {
2720 + path = line + 2;
2721 + write_zero = true;
2722 + } else {
2723 + g_error("read map failed - parse error ('%s')", inbuf);
2724 + }
2725 +
2726 + char *devname = NULL;
2727 + path = extract_devname(path, &devname, -1);
2728 + if (!devname) {
2729 + g_error("read map failed - no dev name specified ('%s')",
2730 + inbuf);
2731 + }
2732 +
2733 + RestoreMap *map = g_new0(RestoreMap, 1);
2734 + map->devname = g_strdup(devname);
2735 + map->path = g_strdup(path);
2736 + map->format = format;
2737 + map->write_zero = write_zero;
2738 +
2739 + g_hash_table_insert(devmap, map->devname, map);
2740 +
2741 + };
2742 + }
2743 +
2744 + int i;
2745 + int vmstate_fd = -1;
2746 + guint8 vmstate_stream = 0;
2747 +
2748 + BlockBackend *blk = NULL;
2749 +
2750 + for (i = 1; i < 255; i++) {
2751 + VmaDeviceInfo *di = vma_reader_get_device_info(vmar, i);
2752 + if (di && (strcmp(di->devname, "vmstate") == 0)) {
2753 + vmstate_stream = i;
2754 + char *statefn = g_strdup_printf("%s/vmstate.bin", dirname);
2755 + vmstate_fd = open(statefn, O_WRONLY|O_CREAT|O_EXCL, 0644);
2756 + if (vmstate_fd < 0) {
2757 + g_error("create vmstate file '%s' failed - %s", statefn,
2758 + g_strerror(errno));
2759 + }
2760 + g_free(statefn);
2761 + } else if (di) {
2762 + char *devfn = NULL;
2763 + const char *format = NULL;
2764 + int flags = BDRV_O_RDWR | BDRV_O_NO_FLUSH;
2765 + bool write_zero = true;
2766 +
2767 + if (readmap) {
2768 + RestoreMap *map;
2769 + map = (RestoreMap *)g_hash_table_lookup(devmap, di->devname);
2770 + if (map == NULL) {
2771 + g_error("no device name mapping for %s", di->devname);
2772 + }
2773 + devfn = map->path;
2774 + format = map->format;
2775 + write_zero = map->write_zero;
2776 + } else {
2777 + devfn = g_strdup_printf("%s/tmp-disk-%s.raw",
2778 + dirname, di->devname);
2779 + printf("DEVINFO %s %zd\n", devfn, di->size);
2780 +
2781 + bdrv_img_create(devfn, "raw", NULL, NULL, NULL, di->size,
2782 + flags, &errp, 0);
2783 + if (errp) {
2784 + g_error("can't create file %s: %s", devfn,
2785 + error_get_pretty(errp));
2786 + }
2787 +
2788 + /* Note: we created an empty file above, so there is no
2789 + * need to write zeroes (so we generate a sparse file)
2790 + */
2791 + write_zero = false;
2792 + }
2793 +
2794 + size_t devlen = strlen(devfn);
2795 + QDict *options = NULL;
2796 + if (format) {
2797 + /* explicit format from commandline */
2798 + options = qdict_new();
2799 + qdict_put(options, "driver", qstring_from_str(format));
2800 + } else if ((devlen > 4 && strcmp(devfn+devlen-4, ".raw") == 0) ||
2801 + strncmp(devfn, "/dev/", 5) == 0)
2802 + {
2803 + /* This part is now deprecated for PVE as well (just as qemu
2804 + * deprecated not specifying an explicit raw format, too.
2805 + */
2806 + /* explicit raw format */
2807 + options = qdict_new();
2808 + qdict_put(options, "driver", qstring_from_str("raw"));
2809 + }
2810 +
2811 +
2812 + if (errp || !(blk = blk_new_open(devfn, NULL, options, flags, &errp))) {
2813 + g_error("can't open file %s - %s", devfn,
2814 + error_get_pretty(errp));
2815 + }
2816 +
2817 + if (vma_reader_register_bs(vmar, i, blk, write_zero, &errp) < 0) {
2818 + g_error("%s", error_get_pretty(errp));
2819 + }
2820 +
2821 + if (!readmap) {
2822 + g_free(devfn);
2823 + }
2824 + }
2825 + }
2826 +
2827 + if (vma_reader_restore(vmar, vmstate_fd, verbose, &errp) < 0) {
2828 + g_error("restore failed - %s", error_get_pretty(errp));
2829 + }
2830 +
2831 + if (!readmap) {
2832 + for (i = 1; i < 255; i++) {
2833 + VmaDeviceInfo *di = vma_reader_get_device_info(vmar, i);
2834 + if (di && (i != vmstate_stream)) {
2835 + char *tmpfn = g_strdup_printf("%s/tmp-disk-%s.raw",
2836 + dirname, di->devname);
2837 + char *fn = g_strdup_printf("%s/disk-%s.raw",
2838 + dirname, di->devname);
2839 + if (rename(tmpfn, fn) != 0) {
2840 + g_error("rename %s to %s failed - %s",
2841 + tmpfn, fn, g_strerror(errno));
2842 + }
2843 + }
2844 + }
2845 + }
2846 +
2847 + vma_reader_destroy(vmar);
2848 +
2849 + blk_unref(blk);
2850 +
2851 + bdrv_close_all();
2852 +
2853 + return ret;
2854 +}
2855 +
2856 +static int verify_content(int argc, char **argv)
2857 +{
2858 + int c, ret = 0;
2859 + int verbose = 0;
2860 + const char *filename;
2861 +
2862 + for (;;) {
2863 + c = getopt(argc, argv, "hv");
2864 + if (c == -1) {
2865 + break;
2866 + }
2867 + switch (c) {
2868 + case '?':
2869 + case 'h':
2870 + help();
2871 + break;
2872 + case 'v':
2873 + verbose = 1;
2874 + break;
2875 + default:
2876 + help();
2877 + }
2878 + }
2879 +
2880 + /* Get the filename */
2881 + if ((optind + 1) != argc) {
2882 + help();
2883 + }
2884 + filename = argv[optind++];
2885 +
2886 + Error *errp = NULL;
2887 + VmaReader *vmar = vma_reader_create(filename, &errp);
2888 +
2889 + if (!vmar) {
2890 + g_error("%s", error_get_pretty(errp));
2891 + }
2892 +
2893 + if (verbose) {
2894 + print_content(vmar);
2895 + }
2896 +
2897 + if (vma_reader_verify(vmar, verbose, &errp) < 0) {
2898 + g_error("verify failed - %s", error_get_pretty(errp));
2899 + }
2900 +
2901 + vma_reader_destroy(vmar);
2902 +
2903 + bdrv_close_all();
2904 +
2905 + return ret;
2906 +}
2907 +
2908 +typedef struct BackupJob {
2909 + BlockBackend *target;
2910 + int64_t len;
2911 + VmaWriter *vmaw;
2912 + uint8_t dev_id;
2913 +} BackupJob;
2914 +
2915 +#define BACKUP_SECTORS_PER_CLUSTER (VMA_CLUSTER_SIZE / BDRV_SECTOR_SIZE)
2916 +
2917 +static void coroutine_fn backup_run_empty(void *opaque)
2918 +{
2919 + VmaWriter *vmaw = (VmaWriter *)opaque;
2920 +
2921 + vma_writer_flush_output(vmaw);
2922 +
2923 + Error *err = NULL;
2924 + if (vma_writer_close(vmaw, &err) != 0) {
2925 + g_warning("vma_writer_close failed %s", error_get_pretty(err));
2926 + }
2927 +}
2928 +
2929 +static void coroutine_fn backup_run(void *opaque)
2930 +{
2931 + BackupJob *job = (BackupJob *)opaque;
2932 + struct iovec iov;
2933 + QEMUIOVector qiov;
2934 +
2935 + int64_t start, end;
2936 + int ret = 0;
2937 +
2938 + unsigned char *buf = blk_blockalign(job->target, VMA_CLUSTER_SIZE);
2939 +
2940 + start = 0;
2941 + end = DIV_ROUND_UP(job->len / BDRV_SECTOR_SIZE,
2942 + BACKUP_SECTORS_PER_CLUSTER);
2943 +
2944 + for (; start < end; start++) {
2945 + iov.iov_base = buf;
2946 + iov.iov_len = VMA_CLUSTER_SIZE;
2947 + qemu_iovec_init_external(&qiov, &iov, 1);
2948 +
2949 + ret = blk_co_preadv(job->target, start * VMA_CLUSTER_SIZE,
2950 + VMA_CLUSTER_SIZE, &qiov, 0);
2951 + if (ret < 0) {
2952 + vma_writer_set_error(job->vmaw, "read error", -1);
2953 + goto out;
2954 + }
2955 +
2956 + size_t zb = 0;
2957 + if (vma_writer_write(job->vmaw, job->dev_id, start, buf, &zb) < 0) {
2958 + vma_writer_set_error(job->vmaw, "backup_dump_cb vma_writer_write failed", -1);
2959 + goto out;
2960 + }
2961 + }
2962 +
2963 +
2964 +out:
2965 + if (vma_writer_close_stream(job->vmaw, job->dev_id) <= 0) {
2966 + Error *err = NULL;
2967 + if (vma_writer_close(job->vmaw, &err) != 0) {
2968 + g_warning("vma_writer_close failed %s", error_get_pretty(err));
2969 + }
2970 + }
2971 +}
2972 +
2973 +static int create_archive(int argc, char **argv)
2974 +{
2975 + int i, c;
2976 + int verbose = 0;
2977 + const char *archivename;
2978 + GList *config_files = NULL;
2979 +
2980 + for (;;) {
2981 + c = getopt(argc, argv, "hvc:");
2982 + if (c == -1) {
2983 + break;
2984 + }
2985 + switch (c) {
2986 + case '?':
2987 + case 'h':
2988 + help();
2989 + break;
2990 + case 'c':
2991 + config_files = g_list_append(config_files, optarg);
2992 + break;
2993 + case 'v':
2994 + verbose = 1;
2995 + break;
2996 + default:
2997 + g_assert_not_reached();
2998 + }
2999 + }
3000 +
3001 +
3002 + /* make sure we an archive name */
3003 + if ((optind + 1) > argc) {
3004 + help();
3005 + }
3006 +
3007 + archivename = argv[optind++];
3008 +
3009 + uuid_t uuid;
3010 + uuid_generate(uuid);
3011 +
3012 + Error *local_err = NULL;
3013 + VmaWriter *vmaw = vma_writer_create(archivename, uuid, &local_err);
3014 +
3015 + if (vmaw == NULL) {
3016 + g_error("%s", error_get_pretty(local_err));
3017 + }
3018 +
3019 + GList *l = config_files;
3020 + while (l && l->data) {
3021 + char *name = l->data;
3022 + char *cdata = NULL;
3023 + gsize clen = 0;
3024 + GError *err = NULL;
3025 + if (!g_file_get_contents(name, &cdata, &clen, &err)) {
3026 + unlink(archivename);
3027 + g_error("Unable to read file: %s", err->message);
3028 + }
3029 +
3030 + if (vma_writer_add_config(vmaw, name, cdata, clen) != 0) {
3031 + unlink(archivename);
3032 + g_error("Unable to append config data %s (len = %zd)",
3033 + name, clen);
3034 + }
3035 + l = g_list_next(l);
3036 + }
3037 +
3038 + int devcount = 0;
3039 + while (optind < argc) {
3040 + const char *path = argv[optind++];
3041 + char *devname = NULL;
3042 + path = extract_devname(path, &devname, devcount++);
3043 +
3044 + Error *errp = NULL;
3045 + BlockBackend *target;
3046 +
3047 + target = blk_new_open(path, NULL, NULL, 0, &errp);
3048 + if (!target) {
3049 + unlink(archivename);
3050 + g_error("bdrv_open '%s' failed - %s", path, error_get_pretty(errp));
3051 + }
3052 + int64_t size = blk_getlength(target);
3053 + int dev_id = vma_writer_register_stream(vmaw, devname, size);
3054 + if (dev_id <= 0) {
3055 + unlink(archivename);
3056 + g_error("vma_writer_register_stream '%s' failed", devname);
3057 + }
3058 +
3059 + BackupJob *job = g_new0(BackupJob, 1);
3060 + job->len = size;
3061 + job->target = target;
3062 + job->vmaw = vmaw;
3063 + job->dev_id = dev_id;
3064 +
3065 + Coroutine *co = qemu_coroutine_create(backup_run, job);
3066 + qemu_coroutine_enter(co);
3067 + }
3068 +
3069 + VmaStatus vmastat;
3070 + int percent = 0;
3071 + int last_percent = -1;
3072 +
3073 + if (devcount) {
3074 + while (1) {
3075 + main_loop_wait(false);
3076 + vma_writer_get_status(vmaw, &vmastat);
3077 +
3078 + if (verbose) {
3079 +
3080 + uint64_t total = 0;
3081 + uint64_t transferred = 0;
3082 + uint64_t zero_bytes = 0;
3083 +
3084 + int i;
3085 + for (i = 0; i < 256; i++) {
3086 + if (vmastat.stream_info[i].size) {
3087 + total += vmastat.stream_info[i].size;
3088 + transferred += vmastat.stream_info[i].transferred;
3089 + zero_bytes += vmastat.stream_info[i].zero_bytes;
3090 + }
3091 + }
3092 + percent = (transferred*100)/total;
3093 + if (percent != last_percent) {
3094 + fprintf(stderr, "progress %d%% %zd/%zd %zd\n", percent,
3095 + transferred, total, zero_bytes);
3096 + fflush(stderr);
3097 +
3098 + last_percent = percent;
3099 + }
3100 + }
3101 +
3102 + if (vmastat.closed) {
3103 + break;
3104 + }
3105 + }
3106 + } else {
3107 + Coroutine *co = qemu_coroutine_create(backup_run_empty, vmaw);
3108 + qemu_coroutine_enter(co);
3109 + while (1) {
3110 + main_loop_wait(false);
3111 + vma_writer_get_status(vmaw, &vmastat);
3112 + if (vmastat.closed) {
3113 + break;
3114 + }
3115 + }
3116 + }
3117 +
3118 + bdrv_drain_all();
3119 +
3120 + vma_writer_get_status(vmaw, &vmastat);
3121 +
3122 + if (verbose) {
3123 + for (i = 0; i < 256; i++) {
3124 + VmaStreamInfo *si = &vmastat.stream_info[i];
3125 + if (si->size) {
3126 + fprintf(stderr, "image %s: size=%zd zeros=%zd saved=%zd\n",
3127 + si->devname, si->size, si->zero_bytes,
3128 + si->size - si->zero_bytes);
3129 + }
3130 + }
3131 + }
3132 +
3133 + if (vmastat.status < 0) {
3134 + unlink(archivename);
3135 + g_error("creating vma archive failed");
3136 + }
3137 +
3138 + return 0;
3139 +}
3140 +
3141 +static int dump_config(int argc, char **argv)
3142 +{
3143 + int c, ret = 0;
3144 + const char *filename;
3145 + const char *config_name = "qemu-server.conf";
3146 +
3147 + for (;;) {
3148 + c = getopt(argc, argv, "hc:");
3149 + if (c == -1) {
3150 + break;
3151 + }
3152 + switch (c) {
3153 + case '?':
3154 + case 'h':
3155 + help();
3156 + break;
3157 + case 'c':
3158 + config_name = optarg;
3159 + break;
3160 + default:
3161 + help();
3162 + }
3163 + }
3164 +
3165 + /* Get the filename */
3166 + if ((optind + 1) != argc) {
3167 + help();
3168 + }
3169 + filename = argv[optind++];
3170 +
3171 + Error *errp = NULL;
3172 + VmaReader *vmar = vma_reader_create(filename, &errp);
3173 +
3174 + if (!vmar) {
3175 + g_error("%s", error_get_pretty(errp));
3176 + }
3177 +
3178 + int found = 0;
3179 + GList *l = vma_reader_get_config_data(vmar);
3180 + while (l && l->data) {
3181 + VmaConfigData *cdata = (VmaConfigData *)l->data;
3182 + l = g_list_next(l);
3183 + if (strcmp(cdata->name, config_name) == 0) {
3184 + found = 1;
3185 + fwrite(cdata->data, cdata->len, 1, stdout);
3186 + break;
3187 + }
3188 + }
3189 +
3190 + vma_reader_destroy(vmar);
3191 +
3192 + bdrv_close_all();
3193 +
3194 + if (!found) {
3195 + fprintf(stderr, "unable to find configuration data '%s'\n", config_name);
3196 + return -1;
3197 + }
3198 +
3199 + return ret;
3200 +}
3201 +
3202 +int main(int argc, char **argv)
3203 +{
3204 + const char *cmdname;
3205 + Error *main_loop_err = NULL;
3206 +
3207 + error_set_progname(argv[0]);
3208 +
3209 + if (qemu_init_main_loop(&main_loop_err)) {
3210 + g_error("%s", error_get_pretty(main_loop_err));
3211 + }
3212 +
3213 + bdrv_init();
3214 +
3215 + if (argc < 2) {
3216 + help();
3217 + }
3218 +
3219 + cmdname = argv[1];
3220 + argc--; argv++;
3221 +
3222 +
3223 + if (!strcmp(cmdname, "list")) {
3224 + return list_content(argc, argv);
3225 + } else if (!strcmp(cmdname, "create")) {
3226 + return create_archive(argc, argv);
3227 + } else if (!strcmp(cmdname, "extract")) {
3228 + return extract_content(argc, argv);
3229 + } else if (!strcmp(cmdname, "verify")) {
3230 + return verify_content(argc, argv);
3231 + } else if (!strcmp(cmdname, "config")) {
3232 + return dump_config(argc, argv);
3233 + }
3234 +
3235 + help();
3236 + return 0;
3237 +}
3238 diff --git a/vma.h b/vma.h
3239 new file mode 100644
3240 index 0000000000..fa6f4df7e7
3241 --- /dev/null
3242 +++ b/vma.h
3243 @@ -0,0 +1,149 @@
3244 +/*
3245 + * VMA: Virtual Machine Archive
3246 + *
3247 + * Copyright (C) Proxmox Server Solutions
3248 + *
3249 + * Authors:
3250 + * Dietmar Maurer (dietmar@proxmox.com)
3251 + *
3252 + * This work is licensed under the terms of the GNU GPL, version 2 or later.
3253 + * See the COPYING file in the top-level directory.
3254 + *
3255 + */
3256 +
3257 +#ifndef BACKUP_VMA_H
3258 +#define BACKUP_VMA_H
3259 +
3260 +#include <uuid/uuid.h>
3261 +#include "qapi/error.h"
3262 +#include "block/block.h"
3263 +
3264 +#define VMA_BLOCK_BITS 12
3265 +#define VMA_BLOCK_SIZE (1<<VMA_BLOCK_BITS)
3266 +#define VMA_CLUSTER_BITS (VMA_BLOCK_BITS+4)
3267 +#define VMA_CLUSTER_SIZE (1<<VMA_CLUSTER_BITS)
3268 +
3269 +#if VMA_CLUSTER_SIZE != 65536
3270 +#error unexpected cluster size
3271 +#endif
3272 +
3273 +#define VMA_EXTENT_HEADER_SIZE 512
3274 +#define VMA_BLOCKS_PER_EXTENT 59
3275 +#define VMA_MAX_CONFIGS 256
3276 +
3277 +#define VMA_MAX_EXTENT_SIZE \
3278 + (VMA_EXTENT_HEADER_SIZE+VMA_CLUSTER_SIZE*VMA_BLOCKS_PER_EXTENT)
3279 +#if VMA_MAX_EXTENT_SIZE != 3867136
3280 +#error unexpected VMA_EXTENT_SIZE
3281 +#endif
3282 +
3283 +/* File Format Definitions */
3284 +
3285 +#define VMA_MAGIC (GUINT32_TO_BE(('V'<<24)|('M'<<16)|('A'<<8)|0x00))
3286 +#define VMA_EXTENT_MAGIC (GUINT32_TO_BE(('V'<<24)|('M'<<16)|('A'<<8)|'E'))
3287 +
3288 +typedef struct VmaDeviceInfoHeader {
3289 + uint32_t devname_ptr; /* offset into blob_buffer table */
3290 + uint32_t reserved0;
3291 + uint64_t size; /* device size in bytes */
3292 + uint64_t reserved1;
3293 + uint64_t reserved2;
3294 +} VmaDeviceInfoHeader;
3295 +
3296 +typedef struct VmaHeader {
3297 + uint32_t magic;
3298 + uint32_t version;
3299 + unsigned char uuid[16];
3300 + int64_t ctime;
3301 + unsigned char md5sum[16];
3302 +
3303 + uint32_t blob_buffer_offset;
3304 + uint32_t blob_buffer_size;
3305 + uint32_t header_size;
3306 +
3307 + unsigned char reserved[1984];
3308 +
3309 + uint32_t config_names[VMA_MAX_CONFIGS]; /* offset into blob_buffer table */
3310 + uint32_t config_data[VMA_MAX_CONFIGS]; /* offset into blob_buffer table */
3311 +
3312 + uint32_t reserved1;
3313 +
3314 + VmaDeviceInfoHeader dev_info[256];
3315 +} VmaHeader;
3316 +
3317 +typedef struct VmaExtentHeader {
3318 + uint32_t magic;
3319 + uint16_t reserved1;
3320 + uint16_t block_count;
3321 + unsigned char uuid[16];
3322 + unsigned char md5sum[16];
3323 + uint64_t blockinfo[VMA_BLOCKS_PER_EXTENT];
3324 +} VmaExtentHeader;
3325 +
3326 +/* functions/definitions to read/write vma files */
3327 +
3328 +typedef struct VmaReader VmaReader;
3329 +
3330 +typedef struct VmaWriter VmaWriter;
3331 +
3332 +typedef struct VmaConfigData {
3333 + const char *name;
3334 + const void *data;
3335 + uint32_t len;
3336 +} VmaConfigData;
3337 +
3338 +typedef struct VmaStreamInfo {
3339 + uint64_t size;
3340 + uint64_t cluster_count;
3341 + uint64_t transferred;
3342 + uint64_t zero_bytes;
3343 + int finished;
3344 + char *devname;
3345 +} VmaStreamInfo;
3346 +
3347 +typedef struct VmaStatus {
3348 + int status;
3349 + bool closed;
3350 + char errmsg[8192];
3351 + char uuid_str[37];
3352 + VmaStreamInfo stream_info[256];
3353 +} VmaStatus;
3354 +
3355 +typedef struct VmaDeviceInfo {
3356 + uint64_t size; /* device size in bytes */
3357 + const char *devname;
3358 +} VmaDeviceInfo;
3359 +
3360 +VmaWriter *vma_writer_create(const char *filename, uuid_t uuid, Error **errp);
3361 +int vma_writer_close(VmaWriter *vmaw, Error **errp);
3362 +void vma_writer_error_propagate(VmaWriter *vmaw, Error **errp);
3363 +void vma_writer_destroy(VmaWriter *vmaw);
3364 +int vma_writer_add_config(VmaWriter *vmaw, const char *name, gpointer data,
3365 + size_t len);
3366 +int vma_writer_register_stream(VmaWriter *vmaw, const char *devname,
3367 + size_t size);
3368 +
3369 +int64_t coroutine_fn vma_writer_write(VmaWriter *vmaw, uint8_t dev_id,
3370 + int64_t cluster_num, unsigned char *buf,
3371 + size_t *zero_bytes);
3372 +
3373 +int coroutine_fn vma_writer_close_stream(VmaWriter *vmaw, uint8_t dev_id);
3374 +int coroutine_fn vma_writer_flush_output(VmaWriter *vmaw);
3375 +
3376 +int vma_writer_get_status(VmaWriter *vmaw, VmaStatus *status);
3377 +void vma_writer_set_error(VmaWriter *vmaw, const char *fmt, ...);
3378 +
3379 +
3380 +VmaReader *vma_reader_create(const char *filename, Error **errp);
3381 +void vma_reader_destroy(VmaReader *vmar);
3382 +VmaHeader *vma_reader_get_header(VmaReader *vmar);
3383 +GList *vma_reader_get_config_data(VmaReader *vmar);
3384 +VmaDeviceInfo *vma_reader_get_device_info(VmaReader *vmar, guint8 dev_id);
3385 +int vma_reader_register_bs(VmaReader *vmar, guint8 dev_id,
3386 + BlockBackend *target, bool write_zeroes,
3387 + Error **errp);
3388 +int vma_reader_restore(VmaReader *vmar, int vmstate_fd, bool verbose,
3389 + Error **errp);
3390 +int vma_reader_verify(VmaReader *vmar, bool verbose, Error **errp);
3391 +
3392 +#endif /* BACKUP_VMA_H */
3393 --
3394 2.11.0
3395
QEMU for PVE