]> git.proxmox.com Git - pve-qemu.git/blame - debian/patches/pve/0028-adding-old-vma-files.patch
projects / pve-qemu.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
vma: fix regression in backup with non-cluster-aligned sizes
[pve-qemu.git] / debian / patches / pve / 0028-adding-old-vma-files.patch
CommitLineData
2ab9b48e 1From 741b9f52069e043e93f0fc47e820ddbfa0bff7a3 Mon Sep 17 00:00:00 2001
67af0fa4
WB		 2From: Wolfgang Bumiller <w.bumiller@proxmox.com>
3Date: Mon, 7 Aug 2017 08:51:16 +0200
4Subject: [PATCH 28/28] adding old vma files
95259824 5
95259824 6---
67af0fa4
WB		 7 Makefile | 3 +-
8 Makefile.objs | 1 +
9 block/backup.c | 128 ++++---
10 block/replication.c | 1 +
2ab9b48e 11 blockdev.c | 250 +++++++++-----
67af0fa4
WB		 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 ++++++++
2ab9b48e 18 11 files changed, 2799 insertions(+), 133 deletions(-)
95259824
WB		 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
24diff --git a/Makefile b/Makefile
45169293 25index 6c359b2f86..edbc8b50f0 100644
95259824
WB		 26--- a/Makefile
27+++ b/Makefile
a544966d 28@@ -284,7 +284,7 @@ ifneq ($(wildcard config-host.mak),)
95259824
WB		 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 $(call quiet-command, \
a544966d
WB		 37@@ -377,6 +377,7 @@ qemu-img.o: qemu-img-cmds.h
38 qemu-img$(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) $(COMMON_LDADDS)
95259824 42
a544966d 43 qemu-bridge-helper$(EXESUF): qemu-bridge-helper.o $(COMMON_LDADDS)
95259824
WB		 44
45diff --git a/Makefile.objs b/Makefile.objs
67af0fa4 46index fbfbbb7f70..f5f8dbab3b 100644
95259824
WB		 47--- a/Makefile.objs
48+++ b/Makefile.objs
a544966d 49@@ -14,6 +14,7 @@ block-obj-y += block.o blockjob.o
95259824
WB		 50 block-obj-y += block/
51 block-obj-y += qemu-io-cmds.o
a544966d 52 block-obj-$(CONFIG_REPLICATION) += replication.o
95259824
WB		 53+block-obj-y += vma-writer.o
54
55 block-obj-m = block/
56
67af0fa4
WB		 57diff --git a/block/backup.c b/block/backup.c
58index 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@@ -330,9 +344,11 @@ static BlockErrorAction backup_error_action(BackupBlockJob *job,
109 if (read) {
110 return block_job_error_action(&job->common, job->on_source_error,
111 true, error);
112- } else {
113+ } else if (job->target) {
114 return block_job_error_action(&job->common, job->on_target_error,
115 false, error);
116+ } else {
117+ return BLOCK_ERROR_ACTION_REPORT;
118 }
119 }
120
121@@ -453,6 +469,7 @@ static void coroutine_fn backup_run(void *opaque)
122
123 job->done_bitmap = bitmap_new(end);
124
125+
126 job->before_write.notify = backup_before_write_notify;
127 bdrv_add_before_write_notifier(bs, &job->before_write);
128
129@@ -557,6 +574,7 @@ BlockJob *backup_job_create(const char *job_id, BlockDriverState *bs,
130 BlockdevOnError on_source_error,
131 BlockdevOnError on_target_error,
132 int creation_flags,
133+ BackupDumpFunc *dump_cb,
134 BlockCompletionFunc *cb, void *opaque,
135 int pause_count,
136 BlockJobTxn *txn, Error **errp)
137@@ -567,7 +585,7 @@ BlockJob *backup_job_create(const char *job_id, BlockDriverState *bs,
138 int ret;
139
140 assert(bs);
141- assert(target);
142+ assert(target || dump_cb);
143
144 if (bs == target) {
145 error_setg(errp, "Source and target cannot be the same");
146@@ -580,13 +598,13 @@ BlockJob *backup_job_create(const char *job_id, BlockDriverState *bs,
147 return NULL;
148 }
149
150- if (!bdrv_is_inserted(target)) {
151+ if (target && !bdrv_is_inserted(target)) {
152 error_setg(errp, "Device is not inserted: %s",
153 bdrv_get_device_name(target));
154 return NULL;
155 }
156
157- if (compress && target->drv->bdrv_co_pwritev_compressed == NULL) {
158+ if (target && compress && target->drv->bdrv_co_pwritev_compressed == NULL) {
159 error_setg(errp, "Compression is not supported for this drive %s",
160 bdrv_get_device_name(target));
161 return NULL;
162@@ -596,7 +614,7 @@ BlockJob *backup_job_create(const char *job_id, BlockDriverState *bs,
163 return NULL;
164 }
165
166- if (bdrv_op_is_blocked(target, BLOCK_OP_TYPE_BACKUP_TARGET, errp)) {
167+ if (target && bdrv_op_is_blocked(target, BLOCK_OP_TYPE_BACKUP_TARGET, errp)) {
168 return NULL;
169 }
170
171@@ -636,15 +654,18 @@ BlockJob *backup_job_create(const char *job_id, BlockDriverState *bs,
172 goto error;
173 }
174
175- /* The target must match the source in size, so no resize here either */
176- job->target = blk_new(BLK_PERM_WRITE,
177- BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE |
178- BLK_PERM_WRITE_UNCHANGED | BLK_PERM_GRAPH_MOD);
179- ret = blk_insert_bs(job->target, target, errp);
180- if (ret < 0) {
181- goto error;
182+ if (target) {
183+ /* The target must match the source in size, so no resize here either */
184+ job->target = blk_new(BLK_PERM_WRITE,
185+ BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE |
186+ BLK_PERM_WRITE_UNCHANGED | BLK_PERM_GRAPH_MOD);
187+ ret = blk_insert_bs(job->target, target, errp);
188+ if (ret < 0) {
189+ goto error;
190+ }
191 }
192
193+ job->dump_cb = dump_cb;
194 job->on_source_error = on_source_error;
195 job->on_target_error = on_target_error;
196 job->sync_mode = sync_mode;
197@@ -652,38 +673,55 @@ BlockJob *backup_job_create(const char *job_id, BlockDriverState *bs,
198 sync_bitmap : NULL;
199 job->compress = compress;
200
201- /* If there is no backing file on the target, we cannot rely on COW if our
202- * backup cluster size is smaller than the target cluster size. Even for
203- * targets with a backing file, try to avoid COW if possible. */
204- ret = bdrv_get_info(target, &bdi);
205- if (ret == -ENOTSUP && !target->backing) {
206- /* Cluster size is not defined */
207- error_report("WARNING: The target block device doesn't provide "
208- "information about the block size and it doesn't have a "
209- "backing file. The default block size of %u bytes is "
210- "used. If the actual block size of the target exceeds "
211- "this default, the backup may be unusable",
212- BACKUP_CLUSTER_SIZE_DEFAULT);
213- job->cluster_size = BACKUP_CLUSTER_SIZE_DEFAULT;
214- } else if (ret < 0 && !target->backing) {
215- error_setg_errno(errp, -ret,
216- "Couldn't determine the cluster size of the target image, "
217- "which has no backing file");
218- error_append_hint(errp,
219- "Aborting, since this may create an unusable destination image\n");
220- goto error;
221- } else if (ret < 0 && target->backing) {
222- /* Not fatal; just trudge on ahead. */
223- job->cluster_size = BACKUP_CLUSTER_SIZE_DEFAULT;
224+ if (target) {
225+ /* If there is no backing file on the target, we cannot rely on COW if our
226+ * backup cluster size is smaller than the target cluster size. Even for
227+ * targets with a backing file, try to avoid COW if possible. */
228+ ret = bdrv_get_info(target, &bdi);
229+ if (ret == -ENOTSUP && !target->backing) {
230+ /* Cluster size is not defined */
231+ error_report("WARNING: The target block device doesn't provide "
232+ "information about the block size and it doesn't have a "
233+ "backing file. The default block size of %u bytes is "
234+ "used. If the actual block size of the target exceeds "
235+ "this default, the backup may be unusable",
236+ BACKUP_CLUSTER_SIZE_DEFAULT);
237+ job->cluster_size = BACKUP_CLUSTER_SIZE_DEFAULT;
238+ } else if (ret < 0 && !target->backing) {
239+ error_setg_errno(errp, -ret,
240+ "Couldn't determine the cluster size of the target image, "
241+ "which has no backing file");
242+ error_append_hint(errp,
243+ "Aborting, since this may create an unusable destination image\n");
244+ goto error;
245+ } else if (ret < 0 && target->backing) {
246+ /* Not fatal; just trudge on ahead. */
247+ job->cluster_size = BACKUP_CLUSTER_SIZE_DEFAULT;
248+ } else {
249+ job->cluster_size = MAX(BACKUP_CLUSTER_SIZE_DEFAULT, bdi.cluster_size);
250+ }
251 } else {
252- job->cluster_size = MAX(BACKUP_CLUSTER_SIZE_DEFAULT, bdi.cluster_size);
253+ ret = bdrv_get_info(bs, &bdi);
254+ if (ret < 0) {
255+ job->cluster_size = BACKUP_CLUSTER_SIZE_DEFAULT;
256+ } else {
257+ /* round down to nearest BACKUP_CLUSTER_SIZE_DEFAULT */
258+ job->cluster_size = (bdi.cluster_size / BACKUP_CLUSTER_SIZE_DEFAULT) * BACKUP_CLUSTER_SIZE_DEFAULT;
259+ if (job->cluster_size == 0) {
260+ /* but we can't go below it */
261+ job->cluster_size = BACKUP_CLUSTER_SIZE_DEFAULT;
262+ }
263+ }
264 }
265
266- /* Required permissions are already taken with target's blk_new() */
267- block_job_add_bdrv(&job->common, "target", target, 0, BLK_PERM_ALL,
268- &error_abort);
269+ if (target) {
270+ /* Required permissions are already taken with target's blk_new() */
271+ block_job_add_bdrv(&job->common, "target", target, 0, BLK_PERM_ALL,
272+ &error_abort);
273+ } else {
274+ job->common.pause_count = pause_count;
275+ }
276 job->common.len = len;
277- job->common.pause_count = pause_count;
278 block_job_txn_add_job(txn, &job->common);
279
280 return &job->common;
281diff --git a/block/replication.c b/block/replication.c
282index 1c41d9e6bf..60c6524417 100644
283--- a/block/replication.c
284+++ b/block/replication.c
285@@ -531,6 +531,7 @@ static void replication_start(ReplicationState *rs, ReplicationMode mode,
286 0, MIRROR_SYNC_MODE_NONE, NULL, false,
287 BLOCKDEV_ON_ERROR_REPORT,
288 BLOCKDEV_ON_ERROR_REPORT, BLOCK_JOB_INTERNAL,
289+ NULL,
290 backup_job_completed, bs, 0, NULL, &local_err);
291 if (local_err) {
292 error_propagate(errp, local_err);
293diff --git a/blockdev.c b/blockdev.c
2ab9b48e 294index 981276692a..76a7103743 100644
67af0fa4
WB		 295--- a/blockdev.c
296+++ b/blockdev.c
297@@ -31,7 +31,6 @@
298 */
299
300 #include "qemu/osdep.h"
301-#include "qemu/uuid.h"
302 #include "sysemu/block-backend.h"
303 #include "sysemu/blockdev.h"
304 #include "hw/block/block.h"
305@@ -55,6 +54,7 @@
306 #include "qemu/cutils.h"
307 #include "qemu/help_option.h"
308 #include "qemu/throttle-options.h"
309+#include "vma.h"
310
311 static QTAILQ_HEAD(, BlockDriverState) monitor_bdrv_states =
312 QTAILQ_HEAD_INITIALIZER(monitor_bdrv_states);
313@@ -2958,20 +2958,44 @@ out:
314 aio_context_release(aio_context);
315 }
316
317+void block_job_event_cancelled(BlockJob *job);
318+void block_job_event_completed(BlockJob *job, const char *msg);
319+static void block_job_cb(void *opaque, int ret)
320+{
321+ /* Note that this function may be executed from another AioContext besides
322+ * the QEMU main loop. If you need to access anything that assumes the
323+ * QEMU global mutex, use a BH or introduce a mutex.
324+ */
325+
326+ BlockDriverState *bs = opaque;
327+ const char *msg = NULL;
328+
329+ assert(bs->job);
330+
331+ if (ret < 0) {
332+ msg = strerror(-ret);
333+ }
334+
335+ if (block_job_is_cancelled(bs->job)) {
336+ block_job_event_cancelled(bs->job);
337+ } else {
338+ block_job_event_completed(bs->job, msg);
339+ }
340+}
341+
342 /* PVE backup related function */
343
344 static struct PVEBackupState {
345 Error *error;
346 bool cancel;
347- QemuUUID uuid;
348+ uuid_t uuid;
349 char uuid_str[37];
350 int64_t speed;
351 time_t start_time;
352 time_t end_time;
353 char *backup_file;
354- Object *vmaobj;
355+ VmaWriter *vmaw;
356 GList *di_list;
357- size_t next_job;
358 size_t total;
359 size_t transferred;
360 size_t zero_bytes;
361@@ -2981,6 +3005,7 @@ typedef struct PVEBackupDevInfo {
362 BlockDriverState *bs;
363 size_t size;
364 uint8_t dev_id;
365+ //bool started;
366 bool completed;
367 char targetfile[PATH_MAX];
368 BlockDriverState *target;
2ab9b48e 369@@ -2988,13 +3013,79 @@ typedef struct PVEBackupDevInfo {
67af0fa4
WB		 370
371 static void pvebackup_run_next_job(void);
372
373+static int pvebackup_dump_cb(void *opaque, BlockBackend *target,
374+ int64_t sector_num, int n_sectors,
375+ unsigned char *buf)
376+{
377+ PVEBackupDevInfo *di = opaque;
378+
379+ int size = n_sectors * BDRV_SECTOR_SIZE;
380+ if (backup_state.cancel) {
381+ return size; // return success
382+ }
383+
384+ if (sector_num & 0x7f) {
385+ if (!backup_state.error) {
386+ error_setg(&backup_state.error,
387+ "got unaligned write inside backup dump "
388+ "callback (sector %ld)", sector_num);
389+ }
390+ return -1; // not aligned to cluster size
391+ }
392+
393+ int64_t cluster_num = sector_num >> 7;
394+
395+ int ret = -1;
396+
397+ if (backup_state.vmaw) {
398+ size_t zero_bytes = 0;
2ab9b48e
WB		 399+ int64_t remaining = n_sectors * BDRV_SECTOR_SIZE;
400+ while (remaining > 0) {
67af0fa4
WB		 401+ ret = vma_writer_write(backup_state.vmaw, di->dev_id, cluster_num,
402+ buf, &zero_bytes);
67af0fa4
WB		 403+ ++cluster_num;
404+ if (buf) {
405+ buf += VMA_CLUSTER_SIZE;
406+ }
407+ if (ret < 0) {
408+ if (!backup_state.error) {
409+ vma_writer_error_propagate(backup_state.vmaw, &backup_state.error);
410+ }
411+ if (di->bs && di->bs->job) {
412+ block_job_cancel(di->bs->job);
413+ }
2ab9b48e 414+ break;
67af0fa4
WB		 415+ } else {
416+ backup_state.zero_bytes += zero_bytes;
2ab9b48e
WB		 417+ if (remaining >= VMA_CLUSTER_SIZE) {
418+ backup_state.transferred += VMA_CLUSTER_SIZE;
419+ } else {
420+ backup_state.transferred += remaining;
421+ }
422+ remaining -= VMA_CLUSTER_SIZE;
67af0fa4
WB		 423+ }
424+ }
425+ } else {
426+ if (!buf) {
427+ backup_state.zero_bytes += size;
428+ }
429+ backup_state.transferred += size;
430+ }
431+
432+ // Note: always return success, because we want that writes succeed anyways.
433+
434+ return size;
435+}
436+
437 static void pvebackup_cleanup(void)
438 {
439 backup_state.end_time = time(NULL);
440
441- if (backup_state.vmaobj) {
442- object_unparent(backup_state.vmaobj);
443- backup_state.vmaobj = NULL;
444+ if (backup_state.vmaw) {
445+ Error *local_err = NULL;
446+ vma_writer_close(backup_state.vmaw, &local_err);
447+ error_propagate(&backup_state.error, local_err);
448+ backup_state.vmaw = NULL;
449 }
450
451 if (backup_state.di_list) {
2ab9b48e 452@@ -3009,6 +3100,13 @@ static void pvebackup_cleanup(void)
67af0fa4
WB		 453 }
454 }
455
456+static void coroutine_fn backup_close_vma_stream(void *opaque)
457+{
458+ PVEBackupDevInfo *di = opaque;
459+
460+ vma_writer_close_stream(backup_state.vmaw, di->dev_id);
461+}
462+
463 static void pvebackup_complete_cb(void *opaque, int ret)
464 {
465 PVEBackupDevInfo *di = opaque;
2ab9b48e 466@@ -3020,14 +3118,18 @@ static void pvebackup_complete_cb(void *opaque, int ret)
67af0fa4
WB		 467 ret, strerror(-ret));
468 }
469
470+ BlockDriverState *bs = di->bs;
471+
472 di->bs = NULL;
473 di->target = NULL;
474
475- if (backup_state.vmaobj) {
476- object_unparent(backup_state.vmaobj);
477- backup_state.vmaobj = NULL;
478+ if (backup_state.vmaw) {
479+ Coroutine *co = qemu_coroutine_create(backup_close_vma_stream, di);
480+ qemu_coroutine_enter(co);
481 }
482
483+ block_job_cb(bs, ret);
484+
485 if (!backup_state.cancel) {
486 pvebackup_run_next_job();
487 }
2ab9b48e 488@@ -3041,14 +3143,9 @@ static void pvebackup_cancel(void *opaque)
67af0fa4
WB		 489 error_setg(&backup_state.error, "backup cancelled");
490 }
491
492- if (backup_state.vmaobj) {
493- Error *err;
494+ if (backup_state.vmaw) {
495 /* make sure vma writer does not block anymore */
496- if (!object_set_props(backup_state.vmaobj, &err, "blocked", "yes", NULL)) {
497- if (err) {
498- error_report_err(err);
499- }
500- }
501+ vma_writer_set_error(backup_state.vmaw, "backup cancelled");
502 }
503
504 GList *l = backup_state.di_list;
2ab9b48e 505@@ -3073,19 +3170,15 @@ void qmp_backup_cancel(Error **errp)
67af0fa4
WB		 506 Coroutine *co = qemu_coroutine_create(pvebackup_cancel, NULL);
507 qemu_coroutine_enter(co);
508
509- while (backup_state.vmaobj) {
510- /* FIXME: Find something better for this */
511+ while (backup_state.vmaw) {
512+ /* vma writer use main aio context */
513 aio_poll(qemu_get_aio_context(), true);
514 }
515 }
516
517-void vma_object_add_config_file(Object *obj, const char *name,
518- const char *contents, size_t len,
519- Error **errp);
520 static int config_to_vma(const char *file, BackupFormat format,
521- Object *vmaobj,
522- const char *backup_dir,
523- Error **errp)
524+ const char *backup_dir, VmaWriter *vmaw,
525+ Error **errp)
526 {
527 char *cdata = NULL;
528 gsize clen = 0;
2ab9b48e 529@@ -3098,12 +3191,17 @@ static int config_to_vma(const char *file, BackupFormat format,
67af0fa4
WB		 530 char *basename = g_path_get_basename(file);
531
532 if (format == BACKUP_FORMAT_VMA) {
533- vma_object_add_config_file(vmaobj, basename, cdata, clen, errp);
534+ if (vma_writer_add_config(vmaw, basename, cdata, clen) != 0) {
535+ error_setg(errp, "unable to add %s config data to vma archive", file);
536+ g_free(cdata);
537+ g_free(basename);
538+ return 1;
539+ }
540 } else if (format == BACKUP_FORMAT_DIR) {
541 char config_path[PATH_MAX];
542 snprintf(config_path, PATH_MAX, "%s/%s", backup_dir, basename);
543 if (!g_file_set_contents(config_path, cdata, clen, &err)) {
544- error_setg(errp, "unable to write config file '%s'", config_path);
545+ error_setg(errp, "unable to write config file '%s'", config_path);
546 g_free(cdata);
547 g_free(basename);
548 return 1;
2ab9b48e 549@@ -3113,35 +3211,37 @@ static int config_to_vma(const char *file, BackupFormat format,
67af0fa4
WB		 550 g_free(basename);
551 g_free(cdata);
552
553- return 0;
554+ return 0;
555 }
556
557+bool block_job_should_pause(BlockJob *job);
558 static void pvebackup_run_next_job(void)
559 {
560- bool cancel = backup_state.error || backup_state.cancel;
561-fprintf(stderr, "run next job: %zu\n", backup_state.next_job);
562- GList *next = g_list_nth(backup_state.di_list, backup_state.next_job);
563- while (next) {
564- PVEBackupDevInfo *di = (PVEBackupDevInfo *)next->data;
565- backup_state.next_job++;
566+ GList *l = backup_state.di_list;
567+ while (l) {
568+ PVEBackupDevInfo *di = (PVEBackupDevInfo *)l->data;
569+ l = g_list_next(l);
570 if (!di->completed && di->bs && di->bs->job) {
571 BlockJob *job = di->bs->job;
572- if (cancel) {
573- block_job_cancel(job);
574- } else {
575- block_job_resume(job);
576+ if (block_job_should_pause(job)) {
577+ bool cancel = backup_state.error || backup_state.cancel;
578+ if (cancel) {
579+ block_job_cancel(job);
580+ } else {
581+ block_job_resume(job);
582+ }
583 }
584 return;
585 }
586- next = g_list_next(next);
587 }
588+
589 pvebackup_cleanup();
590 }
591
592 UuidInfo *qmp_backup(const char *backup_file, bool has_format,
593 BackupFormat format,
594 bool has_config_file, const char *config_file,
595- bool has_firewall_file, const char *firewall_file,
596+ bool has_firewall_file, const char *firewall_file,
597 bool has_devlist, const char *devlist,
598 bool has_speed, int64_t speed, Error **errp)
599 {
2ab9b48e 600@@ -3149,14 +3249,15 @@ UuidInfo *qmp_backup(const char *backup_file, bool has_format,
67af0fa4
WB		 601 BlockDriverState *bs = NULL;
602 const char *backup_dir = NULL;
603 Error *local_err = NULL;
604- QemuUUID uuid;
605+ uuid_t uuid;
606+ VmaWriter *vmaw = NULL;
607 gchar **devs = NULL;
608 GList *di_list = NULL;
609 GList *l;
610 UuidInfo *uuid_info;
611 BlockJob *job;
612
613- if (backup_state.di_list || backup_state.vmaobj) {
614+ if (backup_state.di_list) {
615 error_set(errp, ERROR_CLASS_GENERIC_ERROR,
616 "previous backup not finished");
617 return NULL;
2ab9b48e 618@@ -3231,40 +3332,28 @@ UuidInfo *qmp_backup(const char *backup_file, bool has_format,
67af0fa4
WB		 619 total += size;
620 }
621
622- qemu_uuid_generate(&uuid);
623+ uuid_generate(uuid);
624
625 if (format == BACKUP_FORMAT_VMA) {
626- char uuidstr[UUID_FMT_LEN+1];
627- qemu_uuid_unparse(&uuid, uuidstr);
628- uuidstr[UUID_FMT_LEN] = 0;
629- backup_state.vmaobj =
630- object_new_with_props("vma", object_get_objects_root(),
631- "vma-backup-obj", &local_err,
632- "filename", backup_file,
633- "uuid", uuidstr,
634- NULL);
635- if (!backup_state.vmaobj) {
636+ vmaw = vma_writer_create(backup_file, uuid, &local_err);
637+ if (!vmaw) {
638 if (local_err) {
639 error_propagate(errp, local_err);
640 }
641 goto err;
642 }
643
644+ /* register all devices for vma writer */
645 l = di_list;
646 while (l) {
647- QDict *options = qdict_new();
648-
649 PVEBackupDevInfo *di = (PVEBackupDevInfo *)l->data;
650 l = g_list_next(l);
651
652 const char *devname = bdrv_get_device_name(di->bs);
653- snprintf(di->targetfile, PATH_MAX, "vma-backup-obj/%s.raw", devname);
654-
655- qdict_put(options, "driver", qstring_from_str("vma-drive"));
656- qdict_put(options, "size", qint_from_int(di->size));
657- di->target = bdrv_open(di->targetfile, NULL, options, BDRV_O_RDWR, &local_err);
658- if (!di->target) {
659- error_propagate(errp, local_err);
660+ di->dev_id = vma_writer_register_stream(vmaw, devname, di->size);
661+ if (di->dev_id <= 0) {
662+ error_set(errp, ERROR_CLASS_GENERIC_ERROR,
663+ "register_stream failed");
664 goto err;
665 }
666 }
2ab9b48e 667@@ -3305,15 +3394,15 @@ UuidInfo *qmp_backup(const char *backup_file, bool has_format,
67af0fa4
WB		 668
669 /* add configuration file to archive */
670 if (has_config_file) {
671- if(config_to_vma(config_file, format, backup_state.vmaobj, backup_dir, errp) != 0) {
672- goto err;
673+ if(config_to_vma(config_file, format, backup_dir, vmaw, errp) != 0) {
674+ goto err;
675 }
676 }
677
678 /* add firewall file to archive */
679 if (has_firewall_file) {
680- if(config_to_vma(firewall_file, format, backup_state.vmaobj, backup_dir, errp) != 0) {
681- goto err;
682+ if(config_to_vma(firewall_file, format, backup_dir, vmaw, errp) != 0) {
683+ goto err;
684 }
685 }
686 /* initialize global backup_state now */
2ab9b48e 687@@ -3335,11 +3424,12 @@ UuidInfo *qmp_backup(const char *backup_file, bool has_format,
67af0fa4
WB		 688 }
689 backup_state.backup_file = g_strdup(backup_file);
690
691- memcpy(&backup_state.uuid, &uuid, sizeof(uuid));
692- qemu_uuid_unparse(&uuid, backup_state.uuid_str);
693+ backup_state.vmaw = vmaw;
694+
695+ uuid_copy(backup_state.uuid, uuid);
696+ uuid_unparse_lower(uuid, backup_state.uuid_str);
697
698 backup_state.di_list = di_list;
699- backup_state.next_job = 0;
700
701 backup_state.total = total;
702 backup_state.transferred = 0;
2ab9b48e 703@@ -3350,21 +3440,16 @@ UuidInfo *qmp_backup(const char *backup_file, bool has_format,
67af0fa4
WB		 704 while (l) {
705 PVEBackupDevInfo *di = (PVEBackupDevInfo *)l->data;
706 l = g_list_next(l);
707-
708 job = backup_job_create(NULL, di->bs, di->target, speed, MIRROR_SYNC_MODE_FULL, NULL,
709 false, BLOCKDEV_ON_ERROR_REPORT, BLOCKDEV_ON_ERROR_REPORT,
710 BLOCK_JOB_DEFAULT,
711- pvebackup_complete_cb, di, 2, NULL, &local_err);
712- if (di->target) {
713- bdrv_unref(di->target);
714- di->target = NULL;
715- }
716+ pvebackup_dump_cb, pvebackup_complete_cb, di,
717+ 2, NULL, &local_err);
718 if (!job || local_err != NULL) {
719 error_setg(&backup_state.error, "backup_job_create failed");
720 pvebackup_cancel(NULL);
721- } else {
722- block_job_start(job);
723 }
724+ block_job_start(job);
725 }
726
727 if (!backup_state.error) {
2ab9b48e 728@@ -3398,9 +3483,10 @@ err:
67af0fa4
WB		 729 g_strfreev(devs);
730 }
731
732- if (backup_state.vmaobj) {
733- object_unparent(backup_state.vmaobj);
734- backup_state.vmaobj = NULL;
735+ if (vmaw) {
736+ Error *err = NULL;
737+ vma_writer_close(vmaw, &err);
738+ unlink(backup_file);
739 }
740
741 if (backup_dir) {
2ab9b48e 742@@ -3772,7 +3858,7 @@ static BlockJob *do_drive_backup(DriveBackup *backup, BlockJobTxn *txn,
67af0fa4
WB		 743 job = backup_job_create(backup->job_id, bs, target_bs, backup->speed,
744 backup->sync, bmap, backup->compress,
745 backup->on_source_error, backup->on_target_error,
746- BLOCK_JOB_DEFAULT, NULL, NULL, 0, txn, &local_err);
747+ BLOCK_JOB_DEFAULT, NULL, NULL, NULL, 0, txn, &local_err);
748 bdrv_unref(target_bs);
749 if (local_err != NULL) {
750 error_propagate(errp, local_err);
2ab9b48e 751@@ -3851,7 +3937,7 @@ BlockJob *do_blockdev_backup(BlockdevBackup *backup, BlockJobTxn *txn,
67af0fa4
WB		 752 job = backup_job_create(backup->job_id, bs, target_bs, backup->speed,
753 backup->sync, NULL, backup->compress,
754 backup->on_source_error, backup->on_target_error,
755- BLOCK_JOB_DEFAULT, NULL, NULL, 0, txn, &local_err);
756+ BLOCK_JOB_DEFAULT, NULL, NULL, NULL, 0, txn, &local_err);
757 if (local_err != NULL) {
758 error_propagate(errp, local_err);
759 }
760diff --git a/blockjob.c b/blockjob.c
761index 764d41863e..cb3741f6dd 100644
762--- a/blockjob.c
763+++ b/blockjob.c
764@@ -37,8 +37,8 @@
765 #include "qemu/timer.h"
766 #include "qapi-event.h"
767
768-static void block_job_event_cancelled(BlockJob *job);
769-static void block_job_event_completed(BlockJob *job, const char *msg);
770+void block_job_event_cancelled(BlockJob *job);
771+void block_job_event_completed(BlockJob *job, const char *msg);
772
773 /* Transactional group of block jobs */
774 struct BlockJobTxn {
775@@ -473,7 +473,8 @@ void block_job_user_pause(BlockJob *job)
776 block_job_pause(job);
777 }
778
779-static bool block_job_should_pause(BlockJob *job)
780+bool block_job_should_pause(BlockJob *job);
781+bool block_job_should_pause(BlockJob *job)
782 {
783 return job->pause_count > 0;
784 }
785@@ -687,7 +688,7 @@ static void block_job_iostatus_set_err(BlockJob *job, int error)
786 }
787 }
788
789-static void block_job_event_cancelled(BlockJob *job)
790+void block_job_event_cancelled(BlockJob *job)
791 {
792 if (block_job_is_internal(job)) {
793 return;
794@@ -701,7 +702,7 @@ static void block_job_event_cancelled(BlockJob *job)
795 &error_abort);
796 }
797
798-static void block_job_event_completed(BlockJob *job, const char *msg)
799+void block_job_event_completed(BlockJob *job, const char *msg)
800 {
801 if (block_job_is_internal(job)) {
802 return;
803diff --git a/include/block/block_int.h b/include/block/block_int.h
804index 2b3ecd0575..278da161fb 100644
805--- a/include/block/block_int.h
806+++ b/include/block/block_int.h
807@@ -59,6 +59,9 @@
808
809 #define BLOCK_PROBE_BUF_SIZE 512
810
811+typedef int BackupDumpFunc(void *opaque, BlockBackend *be,
812+ int64_t sector_num, int n_sectors, unsigned char *buf);
813+
814 enum BdrvTrackedRequestType {
815 BDRV_TRACKED_READ,
816 BDRV_TRACKED_WRITE,
817@@ -877,6 +880,7 @@ BlockJob *backup_job_create(const char *job_id, BlockDriverState *bs,
818 BlockdevOnError on_source_error,
819 BlockdevOnError on_target_error,
820 int creation_flags,
821+ BackupDumpFunc *dump_cb,
822 BlockCompletionFunc *cb, void *opaque,
823 int pause_count,
824 BlockJobTxn *txn, Error **errp);
95259824
WB		 825diff --git a/vma-reader.c b/vma-reader.c
826new file mode 100644
67af0fa4 827index 0000000000..2000889bd3
95259824
WB		 828--- /dev/null
829+++ b/vma-reader.c
67af0fa4 830@@ -0,0 +1,857 @@
95259824
WB		 831+/*
832+ * VMA: Virtual Machine Archive
833+ *
834+ * Copyright (C) 2012 Proxmox Server Solutions
835+ *
836+ * Authors:
837+ * Dietmar Maurer (dietmar@proxmox.com)
838+ *
839+ * This work is licensed under the terms of the GNU GPL, version 2 or later.
840+ * See the COPYING file in the top-level directory.
841+ *
842+ */
843+
844+#include "qemu/osdep.h"
845+#include <glib.h>
846+#include <uuid/uuid.h>
847+
848+#include "qemu-common.h"
849+#include "qemu/timer.h"
850+#include "qemu/ratelimit.h"
851+#include "vma.h"
852+#include "block/block.h"
853+#include "sysemu/block-backend.h"
854+
855+static unsigned char zero_vma_block[VMA_BLOCK_SIZE];
856+
857+typedef struct VmaRestoreState {
67af0fa4 858+ BlockBackend *target;
95259824
WB		 859+ bool write_zeroes;
860+ unsigned long *bitmap;
861+ int bitmap_size;
862+} VmaRestoreState;
863+
864+struct VmaReader {
865+ int fd;
866+ GChecksum *md5csum;
867+ GHashTable *blob_hash;
868+ unsigned char *head_data;
869+ VmaDeviceInfo devinfo[256];
870+ VmaRestoreState rstate[256];
871+ GList *cdata_list;
872+ guint8 vmstate_stream;
873+ uint32_t vmstate_clusters;
874+ /* to show restore percentage if run with -v */
875+ time_t start_time;
876+ int64_t cluster_count;
877+ int64_t clusters_read;
67af0fa4
WB		 878+ int64_t zero_cluster_data;
879+ int64_t partial_zero_cluster_data;
95259824
WB		 880+ int clusters_read_per;
881+};
882+
883+static guint
884+g_int32_hash(gconstpointer v)
885+{
886+ return *(const uint32_t *)v;
887+}
888+
889+static gboolean
890+g_int32_equal(gconstpointer v1, gconstpointer v2)
891+{
892+ return *((const uint32_t *)v1) == *((const uint32_t *)v2);
893+}
894+
895+static int vma_reader_get_bitmap(VmaRestoreState *rstate, int64_t cluster_num)
896+{
897+ assert(rstate);
898+ assert(rstate->bitmap);
899+
900+ unsigned long val, idx, bit;
901+
902+ idx = cluster_num / BITS_PER_LONG;
903+
904+ assert(rstate->bitmap_size > idx);
905+
906+ bit = cluster_num % BITS_PER_LONG;
907+ val = rstate->bitmap[idx];
908+
909+ return !!(val & (1UL << bit));
910+}
911+
912+static void vma_reader_set_bitmap(VmaRestoreState *rstate, int64_t cluster_num,
913+ int dirty)
914+{
915+ assert(rstate);
916+ assert(rstate->bitmap);
917+
918+ unsigned long val, idx, bit;
919+
920+ idx = cluster_num / BITS_PER_LONG;
921+
922+ assert(rstate->bitmap_size > idx);
923+
924+ bit = cluster_num % BITS_PER_LONG;
925+ val = rstate->bitmap[idx];
926+ if (dirty) {
927+ if (!(val & (1UL << bit))) {
928+ val |= 1UL << bit;
929+ }
930+ } else {
931+ if (val & (1UL << bit)) {
932+ val &= ~(1UL << bit);
933+ }
934+ }
935+ rstate->bitmap[idx] = val;
936+}
937+
938+typedef struct VmaBlob {
939+ uint32_t start;
940+ uint32_t len;
941+ void *data;
942+} VmaBlob;
943+
944+static const VmaBlob *get_header_blob(VmaReader *vmar, uint32_t pos)
945+{
946+ assert(vmar);
947+ assert(vmar->blob_hash);
948+
949+ return g_hash_table_lookup(vmar->blob_hash, &pos);
950+}
951+
952+static const char *get_header_str(VmaReader *vmar, uint32_t pos)
953+{
954+ const VmaBlob *blob = get_header_blob(vmar, pos);
955+ if (!blob) {
956+ return NULL;
957+ }
958+ const char *res = (char *)blob->data;
959+ if (res[blob->len-1] != '\0') {
960+ return NULL;
961+ }
962+ return res;
963+}
964+
965+static ssize_t
966+safe_read(int fd, unsigned char *buf, size_t count)
967+{
968+ ssize_t n;
969+
970+ do {
971+ n = read(fd, buf, count);
972+ } while (n < 0 && errno == EINTR);
973+
974+ return n;
975+}
976+
977+static ssize_t
978+full_read(int fd, unsigned char *buf, size_t len)
979+{
980+ ssize_t n;
981+ size_t total;
982+
983+ total = 0;
984+
985+ while (len > 0) {
986+ n = safe_read(fd, buf, len);
987+
988+ if (n == 0) {
989+ return total;
990+ }
991+
992+ if (n <= 0) {
993+ break;
994+ }
995+
996+ buf += n;
997+ total += n;
998+ len -= n;
999+ }
1000+
1001+ if (len) {
1002+ return -1;
1003+ }
1004+
1005+ return total;
1006+}
1007+
1008+void vma_reader_destroy(VmaReader *vmar)
1009+{
1010+ assert(vmar);
1011+
1012+ if (vmar->fd >= 0) {
1013+ close(vmar->fd);
1014+ }
1015+
1016+ if (vmar->cdata_list) {
1017+ g_list_free(vmar->cdata_list);
1018+ }
1019+
1020+ int i;
1021+ for (i = 1; i < 256; i++) {
1022+ if (vmar->rstate[i].bitmap) {
1023+ g_free(vmar->rstate[i].bitmap);
1024+ }
1025+ }
1026+
1027+ if (vmar->md5csum) {
1028+ g_checksum_free(vmar->md5csum);
1029+ }
1030+
1031+ if (vmar->blob_hash) {
1032+ g_hash_table_destroy(vmar->blob_hash);
1033+ }
1034+
1035+ if (vmar->head_data) {
1036+ g_free(vmar->head_data);
1037+ }
1038+
1039+ g_free(vmar);
1040+
1041+};
1042+
1043+static int vma_reader_read_head(VmaReader *vmar, Error **errp)
1044+{
1045+ assert(vmar);
1046+ assert(errp);
1047+ assert(*errp == NULL);
1048+
1049+ unsigned char md5sum[16];
1050+ int i;
1051+ int ret = 0;
1052+
1053+ vmar->head_data = g_malloc(sizeof(VmaHeader));
1054+
1055+ if (full_read(vmar->fd, vmar->head_data, sizeof(VmaHeader)) !=
1056+ sizeof(VmaHeader)) {
1057+ error_setg(errp, "can't read vma header - %s",
1058+ errno ? g_strerror(errno) : "got EOF");
1059+ return -1;
1060+ }
1061+
1062+ VmaHeader *h = (VmaHeader *)vmar->head_data;
1063+
1064+ if (h->magic != VMA_MAGIC) {
1065+ error_setg(errp, "not a vma file - wrong magic number");
1066+ return -1;
1067+ }
1068+
1069+ uint32_t header_size = GUINT32_FROM_BE(h->header_size);
1070+ int need = header_size - sizeof(VmaHeader);
1071+ if (need <= 0) {
1072+ error_setg(errp, "wrong vma header size %d", header_size);
1073+ return -1;
1074+ }
1075+
1076+ vmar->head_data = g_realloc(vmar->head_data, header_size);
1077+ h = (VmaHeader *)vmar->head_data;
1078+
1079+ if (full_read(vmar->fd, vmar->head_data + sizeof(VmaHeader), need) !=
1080+ need) {
1081+ error_setg(errp, "can't read vma header data - %s",
1082+ errno ? g_strerror(errno) : "got EOF");
1083+ return -1;
1084+ }
1085+
1086+ memcpy(md5sum, h->md5sum, 16);
1087+ memset(h->md5sum, 0, 16);
1088+
1089+ g_checksum_reset(vmar->md5csum);
1090+ g_checksum_update(vmar->md5csum, vmar->head_data, header_size);
1091+ gsize csize = 16;
1092+ g_checksum_get_digest(vmar->md5csum, (guint8 *)(h->md5sum), &csize);
1093+
1094+ if (memcmp(md5sum, h->md5sum, 16) != 0) {
1095+ error_setg(errp, "wrong vma header chechsum");
1096+ return -1;
1097+ }
1098+
1099+ /* we can modify header data after checksum verify */
1100+ h->header_size = header_size;
1101+
1102+ h->version = GUINT32_FROM_BE(h->version);
1103+ if (h->version != 1) {
1104+ error_setg(errp, "wrong vma version %d", h->version);
1105+ return -1;
1106+ }
1107+
1108+ h->ctime = GUINT64_FROM_BE(h->ctime);
1109+ h->blob_buffer_offset = GUINT32_FROM_BE(h->blob_buffer_offset);
1110+ h->blob_buffer_size = GUINT32_FROM_BE(h->blob_buffer_size);
1111+
1112+ uint32_t bstart = h->blob_buffer_offset + 1;
1113+ uint32_t bend = h->blob_buffer_offset + h->blob_buffer_size;
1114+
1115+ if (bstart <= sizeof(VmaHeader)) {
1116+ error_setg(errp, "wrong vma blob buffer offset %d",
1117+ h->blob_buffer_offset);
1118+ return -1;
1119+ }
1120+
1121+ if (bend > header_size) {
1122+ error_setg(errp, "wrong vma blob buffer size %d/%d",
1123+ h->blob_buffer_offset, h->blob_buffer_size);
1124+ return -1;
1125+ }
1126+
1127+ while ((bstart + 2) <= bend) {
1128+ uint32_t size = vmar->head_data[bstart] +
1129+ (vmar->head_data[bstart+1] << 8);
1130+ if ((bstart + size + 2) <= bend) {
1131+ VmaBlob *blob = g_new0(VmaBlob, 1);
1132+ blob->start = bstart - h->blob_buffer_offset;
1133+ blob->len = size;
1134+ blob->data = vmar->head_data + bstart + 2;
1135+ g_hash_table_insert(vmar->blob_hash, &blob->start, blob);
1136+ }
1137+ bstart += size + 2;
1138+ }
1139+
1140+
1141+ int count = 0;
1142+ for (i = 1; i < 256; i++) {
1143+ VmaDeviceInfoHeader *dih = &h->dev_info[i];
1144+ uint32_t devname_ptr = GUINT32_FROM_BE(dih->devname_ptr);
1145+ uint64_t size = GUINT64_FROM_BE(dih->size);
1146+ const char *devname = get_header_str(vmar, devname_ptr);
1147+
1148+ if (size && devname) {
1149+ count++;
1150+ vmar->devinfo[i].size = size;
1151+ vmar->devinfo[i].devname = devname;
1152+
1153+ if (strcmp(devname, "vmstate") == 0) {
1154+ vmar->vmstate_stream = i;
1155+ }
1156+ }
1157+ }
1158+
95259824
WB		 1159+ for (i = 0; i < VMA_MAX_CONFIGS; i++) {
1160+ uint32_t name_ptr = GUINT32_FROM_BE(h->config_names[i]);
1161+ uint32_t data_ptr = GUINT32_FROM_BE(h->config_data[i]);
1162+
1163+ if (!(name_ptr && data_ptr)) {
1164+ continue;
1165+ }
1166+ const char *name = get_header_str(vmar, name_ptr);
1167+ const VmaBlob *blob = get_header_blob(vmar, data_ptr);
1168+
1169+ if (!(name && blob)) {
1170+ error_setg(errp, "vma contains invalid data pointers");
1171+ return -1;
1172+ }
1173+
1174+ VmaConfigData *cdata = g_new0(VmaConfigData, 1);
1175+ cdata->name = name;
1176+ cdata->data = blob->data;
1177+ cdata->len = blob->len;
1178+
1179+ vmar->cdata_list = g_list_append(vmar->cdata_list, cdata);
1180+ }
1181+
1182+ return ret;
1183+};
1184+
1185+VmaReader *vma_reader_create(const char *filename, Error **errp)
1186+{
1187+ assert(filename);
1188+ assert(errp);
1189+
1190+ VmaReader *vmar = g_new0(VmaReader, 1);
1191+
1192+ if (strcmp(filename, "-") == 0) {
1193+ vmar->fd = dup(0);
1194+ } else {
1195+ vmar->fd = open(filename, O_RDONLY);
1196+ }
1197+
1198+ if (vmar->fd < 0) {
1199+ error_setg(errp, "can't open file %s - %s\n", filename,
1200+ g_strerror(errno));
1201+ goto err;
1202+ }
1203+
1204+ vmar->md5csum = g_checksum_new(G_CHECKSUM_MD5);
1205+ if (!vmar->md5csum) {
1206+ error_setg(errp, "can't allocate cmsum\n");
1207+ goto err;
1208+ }
1209+
1210+ vmar->blob_hash = g_hash_table_new_full(g_int32_hash, g_int32_equal,
1211+ NULL, g_free);
1212+
1213+ if (vma_reader_read_head(vmar, errp) < 0) {
1214+ goto err;
1215+ }
1216+
1217+ return vmar;
1218+
1219+err:
1220+ if (vmar) {
1221+ vma_reader_destroy(vmar);
1222+ }
1223+
1224+ return NULL;
1225+}
1226+
1227+VmaHeader *vma_reader_get_header(VmaReader *vmar)
1228+{
1229+ assert(vmar);
1230+ assert(vmar->head_data);
1231+
1232+ return (VmaHeader *)(vmar->head_data);
1233+}
1234+
1235+GList *vma_reader_get_config_data(VmaReader *vmar)
1236+{
1237+ assert(vmar);
1238+ assert(vmar->head_data);
1239+
1240+ return vmar->cdata_list;
1241+}
1242+
1243+VmaDeviceInfo *vma_reader_get_device_info(VmaReader *vmar, guint8 dev_id)
1244+{
1245+ assert(vmar);
1246+ assert(dev_id);
1247+
1248+ if (vmar->devinfo[dev_id].size && vmar->devinfo[dev_id].devname) {
1249+ return &vmar->devinfo[dev_id];
1250+ }
1251+
1252+ return NULL;
1253+}
1254+
67af0fa4
WB		 1255+static void allocate_rstate(VmaReader *vmar, guint8 dev_id,
1256+ BlockBackend *target, bool write_zeroes)
1257+{
1258+ assert(vmar);
1259+ assert(dev_id);
1260+
1261+ vmar->rstate[dev_id].target = target;
1262+ vmar->rstate[dev_id].write_zeroes = write_zeroes;
1263+
1264+ int64_t size = vmar->devinfo[dev_id].size;
1265+
1266+ int64_t bitmap_size = (size/BDRV_SECTOR_SIZE) +
1267+ (VMA_CLUSTER_SIZE/BDRV_SECTOR_SIZE) * BITS_PER_LONG - 1;
1268+ bitmap_size /= (VMA_CLUSTER_SIZE/BDRV_SECTOR_SIZE) * BITS_PER_LONG;
1269+
1270+ vmar->rstate[dev_id].bitmap_size = bitmap_size;
1271+ vmar->rstate[dev_id].bitmap = g_new0(unsigned long, bitmap_size);
1272+
1273+ vmar->cluster_count += size/VMA_CLUSTER_SIZE;
1274+}
1275+
1276+int vma_reader_register_bs(VmaReader *vmar, guint8 dev_id, BlockBackend *target,
95259824
WB		 1277+ bool write_zeroes, Error **errp)
1278+{
1279+ assert(vmar);
67af0fa4 1280+ assert(target != NULL);
95259824 1281+ assert(dev_id);
67af0fa4 1282+ assert(vmar->rstate[dev_id].target == NULL);
95259824 1283+
67af0fa4 1284+ int64_t size = blk_getlength(target);
95259824
WB		 1285+ int64_t size_diff = size - vmar->devinfo[dev_id].size;
1286+
1287+ /* storage types can have different size restrictions, so it
1288+ * is not always possible to create an image with exact size.
1289+ * So we tolerate a size difference up to 4MB.
1290+ */
1291+ if ((size_diff < 0) || (size_diff > 4*1024*1024)) {
1292+ error_setg(errp, "vma_reader_register_bs for stream %s failed - "
1293+ "unexpected size %zd != %zd", vmar->devinfo[dev_id].devname,
1294+ size, vmar->devinfo[dev_id].size);
1295+ return -1;
1296+ }
1297+
67af0fa4 1298+ allocate_rstate(vmar, dev_id, target, write_zeroes);
95259824
WB		 1299+
1300+ return 0;
1301+}
1302+
1303+static ssize_t safe_write(int fd, void *buf, size_t count)
1304+{
1305+ ssize_t n;
1306+
1307+ do {
1308+ n = write(fd, buf, count);
1309+ } while (n < 0 && errno == EINTR);
1310+
1311+ return n;
1312+}
1313+
1314+static size_t full_write(int fd, void *buf, size_t len)
1315+{
1316+ ssize_t n;
1317+ size_t total;
1318+
1319+ total = 0;
1320+
1321+ while (len > 0) {
1322+ n = safe_write(fd, buf, len);
1323+ if (n < 0) {
1324+ return n;
1325+ }
1326+ buf += n;
1327+ total += n;
1328+ len -= n;
1329+ }
1330+
1331+ if (len) {
1332+ /* incomplete write ? */
1333+ return -1;
1334+ }
1335+
1336+ return total;
1337+}
1338+
1339+static int restore_write_data(VmaReader *vmar, guint8 dev_id,
67af0fa4 1340+ BlockBackend *target, int vmstate_fd,
95259824
WB		 1341+ unsigned char *buf, int64_t sector_num,
1342+ int nb_sectors, Error **errp)
1343+{
1344+ assert(vmar);
1345+
1346+ if (dev_id == vmar->vmstate_stream) {
1347+ if (vmstate_fd >= 0) {
1348+ int len = nb_sectors * BDRV_SECTOR_SIZE;
1349+ int res = full_write(vmstate_fd, buf, len);
1350+ if (res < 0) {
1351+ error_setg(errp, "write vmstate failed %d", res);
1352+ return -1;
1353+ }
1354+ }
1355+ } else {
67af0fa4 1356+ int res = blk_pwrite(target, sector_num * BDRV_SECTOR_SIZE, buf, nb_sectors * BDRV_SECTOR_SIZE, 0);
95259824 1357+ if (res < 0) {
67af0fa4
WB		 1358+ error_setg(errp, "blk_pwrite to %s failed (%d)",
1359+ bdrv_get_device_name(blk_bs(target)), res);
95259824
WB		 1360+ return -1;
1361+ }
1362+ }
1363+ return 0;
1364+}
67af0fa4 1365+
95259824
WB		 1366+static int restore_extent(VmaReader *vmar, unsigned char *buf,
1367+ int extent_size, int vmstate_fd,
67af0fa4 1368+ bool verbose, bool verify, Error **errp)
95259824
WB		 1369+{
1370+ assert(vmar);
1371+ assert(buf);
1372+
1373+ VmaExtentHeader *ehead = (VmaExtentHeader *)buf;
1374+ int start = VMA_EXTENT_HEADER_SIZE;
1375+ int i;
1376+
1377+ for (i = 0; i < VMA_BLOCKS_PER_EXTENT; i++) {
1378+ uint64_t block_info = GUINT64_FROM_BE(ehead->blockinfo[i]);
1379+ uint64_t cluster_num = block_info & 0xffffffff;
1380+ uint8_t dev_id = (block_info >> 32) & 0xff;
1381+ uint16_t mask = block_info >> (32+16);
1382+ int64_t max_sector;
1383+
1384+ if (!dev_id) {
1385+ continue;
1386+ }
1387+
1388+ VmaRestoreState *rstate = &vmar->rstate[dev_id];
67af0fa4 1389+ BlockBackend *target = NULL;
95259824
WB		 1390+
1391+ if (dev_id != vmar->vmstate_stream) {
67af0fa4
WB		 1392+ target = rstate->target;
1393+ if (!verify && !target) {
95259824
WB		 1394+ error_setg(errp, "got wrong dev id %d", dev_id);
1395+ return -1;
1396+ }
1397+
1398+ if (vma_reader_get_bitmap(rstate, cluster_num)) {
1399+ error_setg(errp, "found duplicated cluster %zd for stream %s",
1400+ cluster_num, vmar->devinfo[dev_id].devname);
1401+ return -1;
1402+ }
1403+ vma_reader_set_bitmap(rstate, cluster_num, 1);
1404+
1405+ max_sector = vmar->devinfo[dev_id].size/BDRV_SECTOR_SIZE;
1406+ } else {
1407+ max_sector = G_MAXINT64;
1408+ if (cluster_num != vmar->vmstate_clusters) {
1409+ error_setg(errp, "found out of order vmstate data");
1410+ return -1;
1411+ }
1412+ vmar->vmstate_clusters++;
1413+ }
1414+
1415+ vmar->clusters_read++;
1416+
1417+ if (verbose) {
1418+ time_t duration = time(NULL) - vmar->start_time;
1419+ int percent = (vmar->clusters_read*100)/vmar->cluster_count;
1420+ if (percent != vmar->clusters_read_per) {
1421+ printf("progress %d%% (read %zd bytes, duration %zd sec)\n",
1422+ percent, vmar->clusters_read*VMA_CLUSTER_SIZE,
1423+ duration);
1424+ fflush(stdout);
1425+ vmar->clusters_read_per = percent;
1426+ }
1427+ }
1428+
1429+ /* try to write whole clusters to speedup restore */
1430+ if (mask == 0xffff) {
1431+ if ((start + VMA_CLUSTER_SIZE) > extent_size) {
1432+ error_setg(errp, "short vma extent - too many blocks");
1433+ return -1;
1434+ }
1435+ int64_t sector_num = (cluster_num * VMA_CLUSTER_SIZE) /
1436+ BDRV_SECTOR_SIZE;
1437+ int64_t end_sector = sector_num +
1438+ VMA_CLUSTER_SIZE/BDRV_SECTOR_SIZE;
1439+
1440+ if (end_sector > max_sector) {
1441+ end_sector = max_sector;
1442+ }
1443+
1444+ if (end_sector <= sector_num) {
1445+ error_setg(errp, "got wrong block address - write bejond end");
1446+ return -1;
1447+ }
1448+
67af0fa4
WB		 1449+ if (!verify) {
1450+ int nb_sectors = end_sector - sector_num;
1451+ if (restore_write_data(vmar, dev_id, target, vmstate_fd,
1452+ buf + start, sector_num, nb_sectors,
1453+ errp) < 0) {
1454+ return -1;
1455+ }
95259824
WB		 1456+ }
1457+
1458+ start += VMA_CLUSTER_SIZE;
1459+ } else {
1460+ int j;
1461+ int bit = 1;
1462+
1463+ for (j = 0; j < 16; j++) {
1464+ int64_t sector_num = (cluster_num*VMA_CLUSTER_SIZE +
1465+ j*VMA_BLOCK_SIZE)/BDRV_SECTOR_SIZE;
1466+
1467+ int64_t end_sector = sector_num +
1468+ VMA_BLOCK_SIZE/BDRV_SECTOR_SIZE;
1469+ if (end_sector > max_sector) {
1470+ end_sector = max_sector;
1471+ }
1472+
1473+ if (mask & bit) {
1474+ if ((start + VMA_BLOCK_SIZE) > extent_size) {
1475+ error_setg(errp, "short vma extent - too many blocks");
1476+ return -1;
1477+ }
1478+
1479+ if (end_sector <= sector_num) {
1480+ error_setg(errp, "got wrong block address - "
1481+ "write bejond end");
1482+ return -1;
1483+ }
1484+
67af0fa4
WB		 1485+ if (!verify) {
1486+ int nb_sectors = end_sector - sector_num;
1487+ if (restore_write_data(vmar, dev_id, target, vmstate_fd,
1488+ buf + start, sector_num,
1489+ nb_sectors, errp) < 0) {
1490+ return -1;
1491+ }
95259824
WB		 1492+ }
1493+
1494+ start += VMA_BLOCK_SIZE;
1495+
1496+ } else {
1497+
67af0fa4
WB		 1498+
1499+ if (end_sector > sector_num) {
95259824
WB		 1500+ /* Todo: use bdrv_co_write_zeroes (but that need to
1501+ * be run inside coroutine?)
1502+ */
1503+ int nb_sectors = end_sector - sector_num;
67af0fa4
WB		 1504+ int zero_size = BDRV_SECTOR_SIZE*nb_sectors;
1505+ vmar->zero_cluster_data += zero_size;
1506+ if (mask != 0) {
1507+ vmar->partial_zero_cluster_data += zero_size;
1508+ }
1509+
1510+ if (rstate->write_zeroes && !verify) {
1511+ if (restore_write_data(vmar, dev_id, target, vmstate_fd,
1512+ zero_vma_block, sector_num,
1513+ nb_sectors, errp) < 0) {
1514+ return -1;
1515+ }
95259824
WB		 1516+ }
1517+ }
1518+ }
1519+
1520+ bit = bit << 1;
1521+ }
1522+ }
1523+ }
1524+
1525+ if (start != extent_size) {
1526+ error_setg(errp, "vma extent error - missing blocks");
1527+ return -1;
1528+ }
1529+
1530+ return 0;
1531+}
1532+
67af0fa4
WB		 1533+static int vma_reader_restore_full(VmaReader *vmar, int vmstate_fd,
1534+ bool verbose, bool verify,
1535+ Error **errp)
95259824
WB		 1536+{
1537+ assert(vmar);
1538+ assert(vmar->head_data);
1539+
1540+ int ret = 0;
1541+ unsigned char buf[VMA_MAX_EXTENT_SIZE];
1542+ int buf_pos = 0;
1543+ unsigned char md5sum[16];
1544+ VmaHeader *h = (VmaHeader *)vmar->head_data;
1545+
1546+ vmar->start_time = time(NULL);
1547+
1548+ while (1) {
1549+ int bytes = full_read(vmar->fd, buf + buf_pos, sizeof(buf) - buf_pos);
1550+ if (bytes < 0) {
1551+ error_setg(errp, "read failed - %s", g_strerror(errno));
1552+ return -1;
1553+ }
1554+
1555+ buf_pos += bytes;
1556+
1557+ if (!buf_pos) {
1558+ break; /* EOF */
1559+ }
1560+
1561+ if (buf_pos < VMA_EXTENT_HEADER_SIZE) {
1562+ error_setg(errp, "read short extent (%d bytes)", buf_pos);
1563+ return -1;
1564+ }
1565+
1566+ VmaExtentHeader *ehead = (VmaExtentHeader *)buf;
1567+
1568+ /* extract md5sum */
1569+ memcpy(md5sum, ehead->md5sum, sizeof(ehead->md5sum));
1570+ memset(ehead->md5sum, 0, sizeof(ehead->md5sum));
1571+
1572+ g_checksum_reset(vmar->md5csum);
1573+ g_checksum_update(vmar->md5csum, buf, VMA_EXTENT_HEADER_SIZE);
1574+ gsize csize = 16;
1575+ g_checksum_get_digest(vmar->md5csum, ehead->md5sum, &csize);
1576+
1577+ if (memcmp(md5sum, ehead->md5sum, 16) != 0) {
1578+ error_setg(errp, "wrong vma extent header chechsum");
1579+ return -1;
1580+ }
1581+
1582+ if (memcmp(h->uuid, ehead->uuid, sizeof(ehead->uuid)) != 0) {
1583+ error_setg(errp, "wrong vma extent uuid");
1584+ return -1;
1585+ }
1586+
1587+ if (ehead->magic != VMA_EXTENT_MAGIC || ehead->reserved1 != 0) {
1588+ error_setg(errp, "wrong vma extent header magic");
1589+ return -1;
1590+ }
1591+
1592+ int block_count = GUINT16_FROM_BE(ehead->block_count);
1593+ int extent_size = VMA_EXTENT_HEADER_SIZE + block_count*VMA_BLOCK_SIZE;
1594+
1595+ if (buf_pos < extent_size) {
1596+ error_setg(errp, "short vma extent (%d < %d)", buf_pos,
1597+ extent_size);
1598+ return -1;
1599+ }
1600+
1601+ if (restore_extent(vmar, buf, extent_size, vmstate_fd, verbose,
67af0fa4 1602+ verify, errp) < 0) {
95259824
WB		 1603+ return -1;
1604+ }
1605+
1606+ if (buf_pos > extent_size) {
1607+ memmove(buf, buf + extent_size, buf_pos - extent_size);
1608+ buf_pos = buf_pos - extent_size;
1609+ } else {
1610+ buf_pos = 0;
1611+ }
1612+ }
1613+
1614+ bdrv_drain_all();
1615+
1616+ int i;
1617+ for (i = 1; i < 256; i++) {
1618+ VmaRestoreState *rstate = &vmar->rstate[i];
67af0fa4 1619+ if (!rstate->target) {
95259824
WB		 1620+ continue;
1621+ }
1622+
67af0fa4
WB		 1623+ if (blk_flush(rstate->target) < 0) {
1624+ error_setg(errp, "vma blk_flush %s failed",
95259824
WB		 1625+ vmar->devinfo[i].devname);
1626+ return -1;
1627+ }
1628+
1629+ if (vmar->devinfo[i].size &&
1630+ (strcmp(vmar->devinfo[i].devname, "vmstate") != 0)) {
1631+ assert(rstate->bitmap);
1632+
1633+ int64_t cluster_num, end;
1634+
1635+ end = (vmar->devinfo[i].size + VMA_CLUSTER_SIZE - 1) /
1636+ VMA_CLUSTER_SIZE;
1637+
1638+ for (cluster_num = 0; cluster_num < end; cluster_num++) {
1639+ if (!vma_reader_get_bitmap(rstate, cluster_num)) {
1640+ error_setg(errp, "detected missing cluster %zd "
1641+ "for stream %s", cluster_num,
1642+ vmar->devinfo[i].devname);
1643+ return -1;
1644+ }
1645+ }
1646+ }
1647+ }
1648+
67af0fa4
WB		 1649+ if (verbose) {
1650+ if (vmar->clusters_read) {
1651+ printf("total bytes read %zd, sparse bytes %zd (%.3g%%)\n",
1652+ vmar->clusters_read*VMA_CLUSTER_SIZE,
1653+ vmar->zero_cluster_data,
1654+ (double)(100.0*vmar->zero_cluster_data)/
1655+ (vmar->clusters_read*VMA_CLUSTER_SIZE));
1656+
1657+ int64_t datasize = vmar->clusters_read*VMA_CLUSTER_SIZE-vmar->zero_cluster_data;
1658+ if (datasize) { // this does not make sense for empty files
1659+ printf("space reduction due to 4K zero blocks %.3g%%\n",
1660+ (double)(100.0*vmar->partial_zero_cluster_data) / datasize);
1661+ }
1662+ } else {
1663+ printf("vma archive contains no image data\n");
1664+ }
1665+ }
95259824
WB		 1666+ return ret;
1667+}
1668+
67af0fa4
WB		 1669+int vma_reader_restore(VmaReader *vmar, int vmstate_fd, bool verbose,
1670+ Error **errp)
1671+{
1672+ return vma_reader_restore_full(vmar, vmstate_fd, verbose, false, errp);
1673+}
1674+
1675+int vma_reader_verify(VmaReader *vmar, bool verbose, Error **errp)
1676+{
1677+ guint8 dev_id;
1678+
1679+ for (dev_id = 1; dev_id < 255; dev_id++) {
1680+ if (vma_reader_get_device_info(vmar, dev_id)) {
1681+ allocate_rstate(vmar, dev_id, NULL, false);
1682+ }
1683+ }
1684+
1685+ return vma_reader_restore_full(vmar, -1, verbose, true, errp);
1686+}
1687+
95259824
WB		 1688diff --git a/vma-writer.c b/vma-writer.c
1689new file mode 100644
67af0fa4 1690index 0000000000..9001cbdd2b
95259824
WB		 1691--- /dev/null
1692+++ b/vma-writer.c
67af0fa4 1693@@ -0,0 +1,771 @@
95259824
WB		 1694+/*
1695+ * VMA: Virtual Machine Archive
1696+ *
1697+ * Copyright (C) 2012 Proxmox Server Solutions
1698+ *
1699+ * Authors:
1700+ * Dietmar Maurer (dietmar@proxmox.com)
1701+ *
1702+ * This work is licensed under the terms of the GNU GPL, version 2 or later.
1703+ * See the COPYING file in the top-level directory.
1704+ *
1705+ */
1706+
1707+#include "qemu/osdep.h"
1708+#include <glib.h>
1709+#include <uuid/uuid.h>
1710+
1711+#include "vma.h"
1712+#include "block/block.h"
1713+#include "monitor/monitor.h"
1714+#include "qemu/main-loop.h"
1715+#include "qemu/coroutine.h"
1716+#include "qemu/cutils.h"
1717+
1718+#define DEBUG_VMA 0
1719+
1720+#define DPRINTF(fmt, ...)\
1721+ do { if (DEBUG_VMA) { printf("vma: " fmt, ## __VA_ARGS__); } } while (0)
1722+
1723+#define WRITE_BUFFERS 5
67af0fa4
WB		 1724+#define HEADER_CLUSTERS 8
1725+#define HEADERBUF_SIZE (VMA_CLUSTER_SIZE*HEADER_CLUSTERS)
95259824
WB		 1726+
1727+struct VmaWriter {
1728+ int fd;
1729+ FILE *cmd;
1730+ int status;
1731+ char errmsg[8192];
1732+ uuid_t uuid;
1733+ bool header_written;
1734+ bool closed;
1735+
1736+ /* we always write extents */
67af0fa4 1737+ unsigned char *outbuf;
95259824
WB		 1738+ int outbuf_pos; /* in bytes */
1739+ int outbuf_count; /* in VMA_BLOCKS */
1740+ uint64_t outbuf_block_info[VMA_BLOCKS_PER_EXTENT];
1741+
67af0fa4 1742+ unsigned char *headerbuf;
95259824
WB		 1743+
1744+ GChecksum *md5csum;
95259824
WB		 1745+ CoMutex flush_lock;
1746+ Coroutine *co_writer;
1747+
1748+ /* drive informations */
1749+ VmaStreamInfo stream_info[256];
1750+ guint stream_count;
1751+
1752+ guint8 vmstate_stream;
1753+ uint32_t vmstate_clusters;
1754+
1755+ /* header blob table */
1756+ char *header_blob_table;
1757+ uint32_t header_blob_table_size;
1758+ uint32_t header_blob_table_pos;
1759+
1760+ /* store for config blobs */
1761+ uint32_t config_names[VMA_MAX_CONFIGS]; /* offset into blob_buffer table */
1762+ uint32_t config_data[VMA_MAX_CONFIGS]; /* offset into blob_buffer table */
1763+ uint32_t config_count;
1764+};
1765+
1766+void vma_writer_set_error(VmaWriter *vmaw, const char *fmt, ...)
1767+{
1768+ va_list ap;
1769+
1770+ if (vmaw->status < 0) {
1771+ return;
1772+ }
1773+
1774+ vmaw->status = -1;
1775+
1776+ va_start(ap, fmt);
1777+ g_vsnprintf(vmaw->errmsg, sizeof(vmaw->errmsg), fmt, ap);
1778+ va_end(ap);
1779+
1780+ DPRINTF("vma_writer_set_error: %s\n", vmaw->errmsg);
1781+}
1782+
1783+static uint32_t allocate_header_blob(VmaWriter *vmaw, const char *data,
1784+ size_t len)
1785+{
1786+ if (len > 65535) {
1787+ return 0;
1788+ }
1789+
1790+ if (!vmaw->header_blob_table ||
1791+ (vmaw->header_blob_table_size <
1792+ (vmaw->header_blob_table_pos + len + 2))) {
1793+ int newsize = vmaw->header_blob_table_size + ((len + 2 + 511)/512)*512;
1794+
1795+ vmaw->header_blob_table = g_realloc(vmaw->header_blob_table, newsize);
1796+ memset(vmaw->header_blob_table + vmaw->header_blob_table_size,
1797+ 0, newsize - vmaw->header_blob_table_size);
1798+ vmaw->header_blob_table_size = newsize;
1799+ }
1800+
1801+ uint32_t cpos = vmaw->header_blob_table_pos;
1802+ vmaw->header_blob_table[cpos] = len & 255;
1803+ vmaw->header_blob_table[cpos+1] = (len >> 8) & 255;
1804+ memcpy(vmaw->header_blob_table + cpos + 2, data, len);
1805+ vmaw->header_blob_table_pos += len + 2;
1806+ return cpos;
1807+}
1808+
1809+static uint32_t allocate_header_string(VmaWriter *vmaw, const char *str)
1810+{
1811+ assert(vmaw);
1812+
1813+ size_t len = strlen(str) + 1;
1814+
1815+ return allocate_header_blob(vmaw, str, len);
1816+}
1817+
1818+int vma_writer_add_config(VmaWriter *vmaw, const char *name, gpointer data,
1819+ gsize len)
1820+{
1821+ assert(vmaw);
1822+ assert(!vmaw->header_written);
1823+ assert(vmaw->config_count < VMA_MAX_CONFIGS);
1824+ assert(name);
1825+ assert(data);
95259824
WB		 1826+
1827+ gchar *basename = g_path_get_basename(name);
1828+ uint32_t name_ptr = allocate_header_string(vmaw, basename);
1829+ g_free(basename);
1830+
1831+ if (!name_ptr) {
1832+ return -1;
1833+ }
1834+
1835+ uint32_t data_ptr = allocate_header_blob(vmaw, data, len);
1836+ if (!data_ptr) {
1837+ return -1;
1838+ }
1839+
1840+ vmaw->config_names[vmaw->config_count] = name_ptr;
1841+ vmaw->config_data[vmaw->config_count] = data_ptr;
1842+
1843+ vmaw->config_count++;
1844+
1845+ return 0;
1846+}
1847+
1848+int vma_writer_register_stream(VmaWriter *vmaw, const char *devname,
1849+ size_t size)
1850+{
1851+ assert(vmaw);
1852+ assert(devname);
1853+ assert(!vmaw->status);
1854+
1855+ if (vmaw->header_written) {
1856+ vma_writer_set_error(vmaw, "vma_writer_register_stream: header "
1857+ "already written");
1858+ return -1;
1859+ }
1860+
1861+ guint n = vmaw->stream_count + 1;
1862+
1863+ /* we can have dev_ids form 1 to 255 (0 reserved)
1864+ * 255(-1) reseverd for safety
1865+ */
1866+ if (n > 254) {
1867+ vma_writer_set_error(vmaw, "vma_writer_register_stream: "
1868+ "too many drives");
1869+ return -1;
1870+ }
1871+
1872+ if (size <= 0) {
1873+ vma_writer_set_error(vmaw, "vma_writer_register_stream: "
1874+ "got strange size %zd", size);
1875+ return -1;
1876+ }
1877+
1878+ DPRINTF("vma_writer_register_stream %s %zu %d\n", devname, size, n);
1879+
1880+ vmaw->stream_info[n].devname = g_strdup(devname);
1881+ vmaw->stream_info[n].size = size;
1882+
1883+ vmaw->stream_info[n].cluster_count = (size + VMA_CLUSTER_SIZE - 1) /
1884+ VMA_CLUSTER_SIZE;
1885+
1886+ vmaw->stream_count = n;
1887+
1888+ if (strcmp(devname, "vmstate") == 0) {
1889+ vmaw->vmstate_stream = n;
1890+ }
1891+
1892+ return n;
1893+}
1894+
1895+static void vma_co_continue_write(void *opaque)
1896+{
1897+ VmaWriter *vmaw = opaque;
1898+
1899+ DPRINTF("vma_co_continue_write\n");
1900+ qemu_coroutine_enter(vmaw->co_writer);
1901+}
1902+
1903+static ssize_t coroutine_fn
67af0fa4 1904+vma_queue_write(VmaWriter *vmaw, const void *buf, size_t bytes)
95259824 1905+{
67af0fa4 1906+ DPRINTF("vma_queue_write enter %zd\n", bytes);
95259824 1907+
67af0fa4
WB		 1908+ assert(vmaw);
1909+ assert(buf);
1910+ assert(bytes <= VMA_MAX_EXTENT_SIZE);
95259824 1911+
67af0fa4
WB		 1912+ size_t done = 0;
1913+ ssize_t ret;
95259824
WB		 1914+
1915+ assert(vmaw->co_writer == NULL);
1916+
1917+ vmaw->co_writer = qemu_coroutine_self();
1918+
95259824 1919+ while (done < bytes) {
67af0fa4
WB		 1920+ aio_set_fd_handler(qemu_get_aio_context(), vmaw->fd, false, NULL, vma_co_continue_write, NULL, vmaw);
1921+ qemu_coroutine_yield();
1922+ aio_set_fd_handler(qemu_get_aio_context(), vmaw->fd, false, NULL, NULL, NULL, NULL);
1923+ if (vmaw->status < 0) {
1924+ DPRINTF("vma_queue_write detected canceled backup\n");
1925+ done = -1;
1926+ break;
1927+ }
95259824
WB		 1928+ ret = write(vmaw->fd, buf + done, bytes - done);
1929+ if (ret > 0) {
1930+ done += ret;
67af0fa4 1931+ DPRINTF("vma_queue_write written %zd %zd\n", done, ret);
95259824
WB		 1932+ } else if (ret < 0) {
1933+ if (errno == EAGAIN || errno == EWOULDBLOCK) {
67af0fa4
WB		 1934+ /* try again */
1935+ } else {
1936+ vma_writer_set_error(vmaw, "vma_queue_write: write error - %s",
95259824
WB		 1937+ g_strerror(errno));
1938+ done = -1; /* always return failure for partial writes */
1939+ break;
1940+ }
1941+ } else if (ret == 0) {
1942+ /* should not happen - simply try again */
1943+ }
1944+ }
1945+
95259824
WB		 1946+ vmaw->co_writer = NULL;
1947+
67af0fa4 1948+ return (done == bytes) ? bytes : -1;
95259824
WB		 1949+}
1950+
1951+VmaWriter *vma_writer_create(const char *filename, uuid_t uuid, Error **errp)
1952+{
1953+ const char *p;
1954+
1955+ assert(sizeof(VmaHeader) == (4096 + 8192));
1956+ assert(G_STRUCT_OFFSET(VmaHeader, config_names) == 2044);
1957+ assert(G_STRUCT_OFFSET(VmaHeader, config_data) == 3068);
1958+ assert(G_STRUCT_OFFSET(VmaHeader, dev_info) == 4096);
1959+ assert(sizeof(VmaExtentHeader) == 512);
1960+
1961+ VmaWriter *vmaw = g_new0(VmaWriter, 1);
1962+ vmaw->fd = -1;
1963+
1964+ vmaw->md5csum = g_checksum_new(G_CHECKSUM_MD5);
1965+ if (!vmaw->md5csum) {
1966+ error_setg(errp, "can't allocate cmsum\n");
1967+ goto err;
1968+ }
1969+
1970+ if (strstart(filename, "exec:", &p)) {
1971+ vmaw->cmd = popen(p, "w");
1972+ if (vmaw->cmd == NULL) {
1973+ error_setg(errp, "can't popen command '%s' - %s\n", p,
1974+ g_strerror(errno));
1975+ goto err;
1976+ }
1977+ vmaw->fd = fileno(vmaw->cmd);
1978+
67af0fa4 1979+ /* try to use O_NONBLOCK */
95259824 1980+ fcntl(vmaw->fd, F_SETFL, fcntl(vmaw->fd, F_GETFL)|O_NONBLOCK);
95259824
WB		 1981+
1982+ } else {
1983+ struct stat st;
1984+ int oflags;
1985+ const char *tmp_id_str;
1986+
1987+ if ((stat(filename, &st) == 0) && S_ISFIFO(st.st_mode)) {
67af0fa4 1988+ oflags = O_NONBLOCK|O_WRONLY;
95259824
WB		 1989+ vmaw->fd = qemu_open(filename, oflags, 0644);
1990+ } else if (strstart(filename, "/dev/fdset/", &tmp_id_str)) {
67af0fa4 1991+ oflags = O_NONBLOCK|O_WRONLY;
95259824
WB		 1992+ vmaw->fd = qemu_open(filename, oflags, 0644);
1993+ } else if (strstart(filename, "/dev/fdname/", &tmp_id_str)) {
1994+ vmaw->fd = monitor_get_fd(cur_mon, tmp_id_str, errp);
1995+ if (vmaw->fd < 0) {
1996+ goto err;
1997+ }
67af0fa4 1998+ /* try to use O_NONBLOCK */
95259824 1999+ fcntl(vmaw->fd, F_SETFL, fcntl(vmaw->fd, F_GETFL)|O_NONBLOCK);
95259824
WB		 2000+ } else {
2001+ oflags = O_NONBLOCK|O_DIRECT|O_WRONLY|O_CREAT|O_EXCL;
2002+ vmaw->fd = qemu_open(filename, oflags, 0644);
2003+ }
2004+
2005+ if (vmaw->fd < 0) {
2006+ error_setg(errp, "can't open file %s - %s\n", filename,
2007+ g_strerror(errno));
2008+ goto err;
2009+ }
2010+ }
2011+
2012+ /* we use O_DIRECT, so we need to align IO buffers */
67af0fa4
WB		 2013+
2014+ vmaw->outbuf = qemu_memalign(512, VMA_MAX_EXTENT_SIZE);
2015+ vmaw->headerbuf = qemu_memalign(512, HEADERBUF_SIZE);
95259824
WB		 2016+
2017+ vmaw->outbuf_count = 0;
2018+ vmaw->outbuf_pos = VMA_EXTENT_HEADER_SIZE;
2019+
2020+ vmaw->header_blob_table_pos = 1; /* start at pos 1 */
2021+
95259824 2022+ qemu_co_mutex_init(&vmaw->flush_lock);
95259824
WB		 2023+
2024+ uuid_copy(vmaw->uuid, uuid);
2025+
2026+ return vmaw;
2027+
2028+err:
2029+ if (vmaw) {
2030+ if (vmaw->cmd) {
2031+ pclose(vmaw->cmd);
2032+ } else if (vmaw->fd >= 0) {
2033+ close(vmaw->fd);
2034+ }
2035+
2036+ if (vmaw->md5csum) {
2037+ g_checksum_free(vmaw->md5csum);
2038+ }
2039+
2040+ g_free(vmaw);
2041+ }
2042+
2043+ return NULL;
2044+}
2045+
2046+static int coroutine_fn vma_write_header(VmaWriter *vmaw)
2047+{
2048+ assert(vmaw);
67af0fa4 2049+ unsigned char *buf = vmaw->headerbuf;
95259824
WB		 2050+ VmaHeader *head = (VmaHeader *)buf;
2051+
2052+ int i;
2053+
2054+ DPRINTF("VMA WRITE HEADER\n");
2055+
2056+ if (vmaw->status < 0) {
2057+ return vmaw->status;
2058+ }
2059+
67af0fa4 2060+ memset(buf, 0, HEADERBUF_SIZE);
95259824
WB		 2061+
2062+ head->magic = VMA_MAGIC;
2063+ head->version = GUINT32_TO_BE(1); /* v1 */
2064+ memcpy(head->uuid, vmaw->uuid, 16);
2065+
2066+ time_t ctime = time(NULL);
2067+ head->ctime = GUINT64_TO_BE(ctime);
2068+
95259824
WB		 2069+ for (i = 0; i < VMA_MAX_CONFIGS; i++) {
2070+ head->config_names[i] = GUINT32_TO_BE(vmaw->config_names[i]);
2071+ head->config_data[i] = GUINT32_TO_BE(vmaw->config_data[i]);
2072+ }
2073+
2074+ /* 32 bytes per device (12 used currently) = 8192 bytes max */
2075+ for (i = 1; i <= 254; i++) {
2076+ VmaStreamInfo *si = &vmaw->stream_info[i];
2077+ if (si->size) {
2078+ assert(si->devname);
2079+ uint32_t devname_ptr = allocate_header_string(vmaw, si->devname);
2080+ if (!devname_ptr) {
2081+ return -1;
2082+ }
2083+ head->dev_info[i].devname_ptr = GUINT32_TO_BE(devname_ptr);
2084+ head->dev_info[i].size = GUINT64_TO_BE(si->size);
2085+ }
2086+ }
2087+
2088+ uint32_t header_size = sizeof(VmaHeader) + vmaw->header_blob_table_size;
2089+ head->header_size = GUINT32_TO_BE(header_size);
2090+
67af0fa4 2091+ if (header_size > HEADERBUF_SIZE) {
95259824
WB		 2092+ return -1; /* just to be sure */
2093+ }
2094+
2095+ uint32_t blob_buffer_offset = sizeof(VmaHeader);
2096+ memcpy(buf + blob_buffer_offset, vmaw->header_blob_table,
2097+ vmaw->header_blob_table_size);
2098+ head->blob_buffer_offset = GUINT32_TO_BE(blob_buffer_offset);
2099+ head->blob_buffer_size = GUINT32_TO_BE(vmaw->header_blob_table_pos);
2100+
2101+ g_checksum_reset(vmaw->md5csum);
2102+ g_checksum_update(vmaw->md5csum, (const guchar *)buf, header_size);
2103+ gsize csize = 16;
2104+ g_checksum_get_digest(vmaw->md5csum, (guint8 *)(head->md5sum), &csize);
2105+
2106+ return vma_queue_write(vmaw, buf, header_size);
2107+}
2108+
2109+static int coroutine_fn vma_writer_flush(VmaWriter *vmaw)
2110+{
2111+ assert(vmaw);
2112+
2113+ int ret;
2114+ int i;
2115+
2116+ if (vmaw->status < 0) {
2117+ return vmaw->status;
2118+ }
2119+
2120+ if (!vmaw->header_written) {
2121+ vmaw->header_written = true;
2122+ ret = vma_write_header(vmaw);
2123+ if (ret < 0) {
2124+ vma_writer_set_error(vmaw, "vma_writer_flush: write header failed");
2125+ return ret;
2126+ }
2127+ }
2128+
2129+ DPRINTF("VMA WRITE FLUSH %d %d\n", vmaw->outbuf_count, vmaw->outbuf_pos);
2130+
2131+
2132+ VmaExtentHeader *ehead = (VmaExtentHeader *)vmaw->outbuf;
2133+
2134+ ehead->magic = VMA_EXTENT_MAGIC;
2135+ ehead->reserved1 = 0;
2136+
2137+ for (i = 0; i < VMA_BLOCKS_PER_EXTENT; i++) {
2138+ ehead->blockinfo[i] = GUINT64_TO_BE(vmaw->outbuf_block_info[i]);
2139+ }
2140+
2141+ guint16 block_count = (vmaw->outbuf_pos - VMA_EXTENT_HEADER_SIZE) /
2142+ VMA_BLOCK_SIZE;
2143+
2144+ ehead->block_count = GUINT16_TO_BE(block_count);
2145+
2146+ memcpy(ehead->uuid, vmaw->uuid, sizeof(ehead->uuid));
2147+ memset(ehead->md5sum, 0, sizeof(ehead->md5sum));
2148+
2149+ g_checksum_reset(vmaw->md5csum);
2150+ g_checksum_update(vmaw->md5csum, vmaw->outbuf, VMA_EXTENT_HEADER_SIZE);
2151+ gsize csize = 16;
2152+ g_checksum_get_digest(vmaw->md5csum, ehead->md5sum, &csize);
2153+
2154+ int bytes = vmaw->outbuf_pos;
2155+ ret = vma_queue_write(vmaw, vmaw->outbuf, bytes);
2156+ if (ret != bytes) {
2157+ vma_writer_set_error(vmaw, "vma_writer_flush: failed write");
2158+ }
2159+
2160+ vmaw->outbuf_count = 0;
2161+ vmaw->outbuf_pos = VMA_EXTENT_HEADER_SIZE;
2162+
2163+ for (i = 0; i < VMA_BLOCKS_PER_EXTENT; i++) {
2164+ vmaw->outbuf_block_info[i] = 0;
2165+ }
2166+
2167+ return vmaw->status;
2168+}
2169+
2170+static int vma_count_open_streams(VmaWriter *vmaw)
2171+{
2172+ g_assert(vmaw != NULL);
2173+
2174+ int i;
2175+ int open_drives = 0;
2176+ for (i = 0; i <= 255; i++) {
2177+ if (vmaw->stream_info[i].size && !vmaw->stream_info[i].finished) {
2178+ open_drives++;
2179+ }
2180+ }
2181+
2182+ return open_drives;
2183+}
2184+
67af0fa4
WB		 2185+
2186+/**
2187+ * You need to call this if the vma archive does not contain
2188+ * any data stream.
2189+ */
2190+int coroutine_fn
2191+vma_writer_flush_output(VmaWriter *vmaw)
2192+{
2193+ qemu_co_mutex_lock(&vmaw->flush_lock);
2194+ int ret = vma_writer_flush(vmaw);
2195+ qemu_co_mutex_unlock(&vmaw->flush_lock);
2196+ if (ret < 0) {
2197+ vma_writer_set_error(vmaw, "vma_writer_flush_header failed");
2198+ }
2199+ return ret;
2200+}
2201+
95259824
WB		 2202+/**
2203+ * all jobs should call this when there is no more data
2204+ * Returns: number of remaining stream (0 ==> finished)
2205+ */
2206+int coroutine_fn
2207+vma_writer_close_stream(VmaWriter *vmaw, uint8_t dev_id)
2208+{
2209+ g_assert(vmaw != NULL);
2210+
2211+ DPRINTF("vma_writer_set_status %d\n", dev_id);
2212+ if (!vmaw->stream_info[dev_id].size) {
2213+ vma_writer_set_error(vmaw, "vma_writer_close_stream: "
2214+ "no such stream %d", dev_id);
2215+ return -1;
2216+ }
2217+ if (vmaw->stream_info[dev_id].finished) {
2218+ vma_writer_set_error(vmaw, "vma_writer_close_stream: "
2219+ "stream already closed %d", dev_id);
2220+ return -1;
2221+ }
2222+
2223+ vmaw->stream_info[dev_id].finished = true;
2224+
2225+ int open_drives = vma_count_open_streams(vmaw);
2226+
2227+ if (open_drives <= 0) {
2228+ DPRINTF("vma_writer_set_status all drives completed\n");
67af0fa4 2229+ vma_writer_flush_output(vmaw);
95259824
WB		 2230+ }
2231+
2232+ return open_drives;
2233+}
2234+
2235+int vma_writer_get_status(VmaWriter *vmaw, VmaStatus *status)
2236+{
2237+ int i;
2238+
2239+ g_assert(vmaw != NULL);
2240+
2241+ if (status) {
2242+ status->status = vmaw->status;
2243+ g_strlcpy(status->errmsg, vmaw->errmsg, sizeof(status->errmsg));
2244+ for (i = 0; i <= 255; i++) {
2245+ status->stream_info[i] = vmaw->stream_info[i];
2246+ }
2247+
2248+ uuid_unparse_lower(vmaw->uuid, status->uuid_str);
2249+ }
2250+
2251+ status->closed = vmaw->closed;
2252+
2253+ return vmaw->status;
2254+}
2255+
2256+static int vma_writer_get_buffer(VmaWriter *vmaw)
2257+{
2258+ int ret = 0;
2259+
2260+ qemu_co_mutex_lock(&vmaw->flush_lock);
2261+
2262+ /* wait until buffer is available */
2263+ while (vmaw->outbuf_count >= (VMA_BLOCKS_PER_EXTENT - 1)) {
2264+ ret = vma_writer_flush(vmaw);
2265+ if (ret < 0) {
2266+ vma_writer_set_error(vmaw, "vma_writer_get_buffer: flush failed");
2267+ break;
2268+ }
2269+ }
2270+
2271+ qemu_co_mutex_unlock(&vmaw->flush_lock);
2272+
2273+ return ret;
2274+}
2275+
2276+
2277+int64_t coroutine_fn
2278+vma_writer_write(VmaWriter *vmaw, uint8_t dev_id, int64_t cluster_num,
2279+ unsigned char *buf, size_t *zero_bytes)
2280+{
2281+ g_assert(vmaw != NULL);
2282+ g_assert(zero_bytes != NULL);
2283+
2284+ *zero_bytes = 0;
2285+
2286+ if (vmaw->status < 0) {
2287+ return vmaw->status;
2288+ }
2289+
2290+ if (!dev_id || !vmaw->stream_info[dev_id].size) {
2291+ vma_writer_set_error(vmaw, "vma_writer_write: "
2292+ "no such stream %d", dev_id);
2293+ return -1;
2294+ }
2295+
2296+ if (vmaw->stream_info[dev_id].finished) {
2297+ vma_writer_set_error(vmaw, "vma_writer_write: "
2298+ "stream already closed %d", dev_id);
2299+ return -1;
2300+ }
2301+
2302+
2303+ if (cluster_num >= (((uint64_t)1)<<32)) {
2304+ vma_writer_set_error(vmaw, "vma_writer_write: "
2305+ "cluster number out of range");
2306+ return -1;
2307+ }
2308+
2309+ if (dev_id == vmaw->vmstate_stream) {
2310+ if (cluster_num != vmaw->vmstate_clusters) {
2311+ vma_writer_set_error(vmaw, "vma_writer_write: "
2312+ "non sequential vmstate write");
2313+ }
2314+ vmaw->vmstate_clusters++;
2315+ } else if (cluster_num >= vmaw->stream_info[dev_id].cluster_count) {
2316+ vma_writer_set_error(vmaw, "vma_writer_write: cluster number too big");
2317+ return -1;
2318+ }
2319+
2320+ /* wait until buffer is available */
2321+ if (vma_writer_get_buffer(vmaw) < 0) {
2322+ vma_writer_set_error(vmaw, "vma_writer_write: "
2323+ "vma_writer_get_buffer failed");
2324+ return -1;
2325+ }
2326+
2327+ DPRINTF("VMA WRITE %d %zd\n", dev_id, cluster_num);
2328+
2329+ uint16_t mask = 0;
2330+
2331+ if (buf) {
2332+ int i;
2333+ int bit = 1;
2334+ for (i = 0; i < 16; i++) {
2335+ unsigned char *vmablock = buf + (i*VMA_BLOCK_SIZE);
2336+ if (!buffer_is_zero(vmablock, VMA_BLOCK_SIZE)) {
2337+ mask |= bit;
2338+ memcpy(vmaw->outbuf + vmaw->outbuf_pos, vmablock,
2339+ VMA_BLOCK_SIZE);
2340+ vmaw->outbuf_pos += VMA_BLOCK_SIZE;
2341+ } else {
2342+ DPRINTF("VMA WRITE %zd ZERO BLOCK %d\n", cluster_num, i);
2343+ vmaw->stream_info[dev_id].zero_bytes += VMA_BLOCK_SIZE;
2344+ *zero_bytes += VMA_BLOCK_SIZE;
2345+ }
2346+
2347+ bit = bit << 1;
2348+ }
2349+ } else {
2350+ DPRINTF("VMA WRITE %zd ZERO CLUSTER\n", cluster_num);
2351+ vmaw->stream_info[dev_id].zero_bytes += VMA_CLUSTER_SIZE;
2352+ *zero_bytes += VMA_CLUSTER_SIZE;
2353+ }
2354+
2355+ uint64_t block_info = ((uint64_t)mask) << (32+16);
2356+ block_info |= ((uint64_t)dev_id) << 32;
2357+ block_info |= (cluster_num & 0xffffffff);
2358+ vmaw->outbuf_block_info[vmaw->outbuf_count] = block_info;
2359+
2360+ DPRINTF("VMA WRITE MASK %zd %zx\n", cluster_num, block_info);
2361+
2362+ vmaw->outbuf_count++;
2363+
2364+ /** NOTE: We allways write whole clusters, but we correctly set
2365+ * transferred bytes. So transferred == size when when everything
2366+ * went OK.
2367+ */
2368+ size_t transferred = VMA_CLUSTER_SIZE;
2369+
2370+ if (dev_id != vmaw->vmstate_stream) {
2371+ uint64_t last = (cluster_num + 1) * VMA_CLUSTER_SIZE;
2372+ if (last > vmaw->stream_info[dev_id].size) {
2373+ uint64_t diff = last - vmaw->stream_info[dev_id].size;
2374+ if (diff >= VMA_CLUSTER_SIZE) {
2375+ vma_writer_set_error(vmaw, "vma_writer_write: "
2376+ "read after last cluster");
2377+ return -1;
2378+ }
2379+ transferred -= diff;
2380+ }
2381+ }
2382+
2383+ vmaw->stream_info[dev_id].transferred += transferred;
2384+
2385+ return transferred;
2386+}
2387+
67af0fa4
WB		 2388+void vma_writer_error_propagate(VmaWriter *vmaw, Error **errp)
2389+{
2390+ if (vmaw->status < 0 && *errp == NULL) {
2391+ error_setg(errp, "%s", vmaw->errmsg);
2392+ }
2393+}
2394+
95259824
WB		 2395+int vma_writer_close(VmaWriter *vmaw, Error **errp)
2396+{
2397+ g_assert(vmaw != NULL);
2398+
2399+ int i;
2400+
67af0fa4
WB		 2401+ while (vmaw->co_writer) {
2402+ aio_poll(qemu_get_aio_context(), true);
95259824
WB		 2403+ }
2404+
67af0fa4
WB		 2405+ assert(vmaw->co_writer == NULL);
2406+
95259824
WB		 2407+ if (vmaw->cmd) {
2408+ if (pclose(vmaw->cmd) < 0) {
2409+ vma_writer_set_error(vmaw, "vma_writer_close: "
2410+ "pclose failed - %s", g_strerror(errno));
2411+ }
2412+ } else {
2413+ if (close(vmaw->fd) < 0) {
2414+ vma_writer_set_error(vmaw, "vma_writer_close: "
2415+ "close failed - %s", g_strerror(errno));
2416+ }
2417+ }
2418+
2419+ for (i = 0; i <= 255; i++) {
2420+ VmaStreamInfo *si = &vmaw->stream_info[i];
2421+ if (si->size) {
2422+ if (!si->finished) {
2423+ vma_writer_set_error(vmaw, "vma_writer_close: "
2424+ "detected open stream '%s'", si->devname);
2425+ } else if ((si->transferred != si->size) &&
2426+ (i != vmaw->vmstate_stream)) {
2427+ vma_writer_set_error(vmaw, "vma_writer_close: "
2428+ "incomplete stream '%s' (%zd != %zd)",
2429+ si->devname, si->transferred, si->size);
2430+ }
2431+ }
2432+ }
2433+
2434+ for (i = 0; i <= 255; i++) {
2435+ vmaw->stream_info[i].finished = 1; /* mark as closed */
2436+ }
2437+
2438+ vmaw->closed = 1;
2439+
2440+ if (vmaw->status < 0 && *errp == NULL) {
2441+ error_setg(errp, "%s", vmaw->errmsg);
2442+ }
2443+
2444+ return vmaw->status;
2445+}
2446+
2447+void vma_writer_destroy(VmaWriter *vmaw)
2448+{
2449+ assert(vmaw);
2450+
2451+ int i;
2452+
2453+ for (i = 0; i <= 255; i++) {
2454+ if (vmaw->stream_info[i].devname) {
2455+ g_free(vmaw->stream_info[i].devname);
2456+ }
2457+ }
2458+
2459+ if (vmaw->md5csum) {
2460+ g_checksum_free(vmaw->md5csum);
2461+ }
2462+
95259824
WB		 2463+ g_free(vmaw);
2464+}
2465diff --git a/vma.c b/vma.c
2466new file mode 100644
67af0fa4 2467index 0000000000..04915427c8
95259824
WB		 2468--- /dev/null
2469+++ b/vma.c
67af0fa4 2470@@ -0,0 +1,757 @@
95259824
WB		 2471+/*
2472+ * VMA: Virtual Machine Archive
2473+ *
2474+ * Copyright (C) 2012-2013 Proxmox Server Solutions
2475+ *
2476+ * Authors:
2477+ * Dietmar Maurer (dietmar@proxmox.com)
2478+ *
2479+ * This work is licensed under the terms of the GNU GPL, version 2 or later.
2480+ * See the COPYING file in the top-level directory.
2481+ *
2482+ */
2483+
2484+#include "qemu/osdep.h"
2485+#include <glib.h>
2486+
2487+#include "vma.h"
2488+#include "qemu-common.h"
2489+#include "qemu/error-report.h"
2490+#include "qemu/main-loop.h"
a544966d 2491+#include "qapi/qmp/qstring.h"
95259824 2492+#include "sysemu/char.h" /* qstring_from_str */
67af0fa4 2493+#include "sysemu/block-backend.h"
95259824
WB		 2494+
2495+static void help(void)
2496+{
2497+ const char *help_msg =
2498+ "usage: vma command [command options]\n"
2499+ "\n"
2500+ "vma list <filename>\n"
67af0fa4
WB		 2501+ "vma config <filename> [-c config]\n"
2502+ "vma create <filename> [-c config] pathname ...\n"
95259824 2503+ "vma extract <filename> [-r <fifo>] <targetdir>\n"
67af0fa4 2504+ "vma verify <filename> [-v]\n"
95259824
WB		 2505+ ;
2506+
2507+ printf("%s", help_msg);
2508+ exit(1);
2509+}
2510+
2511+static const char *extract_devname(const char *path, char **devname, int index)
2512+{
2513+ assert(path);
2514+
2515+ const char *sep = strchr(path, '=');
2516+
2517+ if (sep) {
2518+ *devname = g_strndup(path, sep - path);
2519+ path = sep + 1;
2520+ } else {
2521+ if (index >= 0) {
2522+ *devname = g_strdup_printf("disk%d", index);
2523+ } else {
2524+ *devname = NULL;
2525+ }
2526+ }
2527+
2528+ return path;
2529+}
2530+
2531+static void print_content(VmaReader *vmar)
2532+{
2533+ assert(vmar);
2534+
2535+ VmaHeader *head = vma_reader_get_header(vmar);
2536+
2537+ GList *l = vma_reader_get_config_data(vmar);
2538+ while (l && l->data) {
2539+ VmaConfigData *cdata = (VmaConfigData *)l->data;
2540+ l = g_list_next(l);
2541+ printf("CFG: size: %d name: %s\n", cdata->len, cdata->name);
2542+ }
2543+
2544+ int i;
2545+ VmaDeviceInfo *di;
2546+ for (i = 1; i < 255; i++) {
2547+ di = vma_reader_get_device_info(vmar, i);
2548+ if (di) {
2549+ if (strcmp(di->devname, "vmstate") == 0) {
2550+ printf("VMSTATE: dev_id=%d memory: %zd\n", i, di->size);
2551+ } else {
2552+ printf("DEV: dev_id=%d size: %zd devname: %s\n",
2553+ i, di->size, di->devname);
2554+ }
2555+ }
2556+ }
2557+ /* ctime is the last entry we print */
2558+ printf("CTIME: %s", ctime(&head->ctime));
2559+ fflush(stdout);
2560+}
2561+
2562+static int list_content(int argc, char **argv)
2563+{
2564+ int c, ret = 0;
2565+ const char *filename;
2566+
2567+ for (;;) {
2568+ c = getopt(argc, argv, "h");
2569+ if (c == -1) {
2570+ break;
2571+ }
2572+ switch (c) {
2573+ case '?':
2574+ case 'h':
2575+ help();
2576+ break;
2577+ default:
2578+ g_assert_not_reached();
2579+ }
2580+ }
2581+
2582+ /* Get the filename */
2583+ if ((optind + 1) != argc) {
2584+ help();
2585+ }
2586+ filename = argv[optind++];
2587+
2588+ Error *errp = NULL;
2589+ VmaReader *vmar = vma_reader_create(filename, &errp);
2590+
2591+ if (!vmar) {
2592+ g_error("%s", error_get_pretty(errp));
2593+ }
2594+
2595+ print_content(vmar);
2596+
2597+ vma_reader_destroy(vmar);
2598+
2599+ return ret;
2600+}
2601+
2602+typedef struct RestoreMap {
2603+ char *devname;
2604+ char *path;
67af0fa4 2605+ char *format;
95259824
WB		 2606+ bool write_zero;
2607+} RestoreMap;
2608+
2609+static int extract_content(int argc, char **argv)
2610+{
2611+ int c, ret = 0;
2612+ int verbose = 0;
2613+ const char *filename;
2614+ const char *dirname;
2615+ const char *readmap = NULL;
2616+
2617+ for (;;) {
2618+ c = getopt(argc, argv, "hvr:");
2619+ if (c == -1) {
2620+ break;
2621+ }
2622+ switch (c) {
2623+ case '?':
2624+ case 'h':
2625+ help();
2626+ break;
2627+ case 'r':
2628+ readmap = optarg;
2629+ break;
2630+ case 'v':
2631+ verbose = 1;
2632+ break;
2633+ default:
2634+ help();
2635+ }
2636+ }
2637+
2638+ /* Get the filename */
2639+ if ((optind + 2) != argc) {
2640+ help();
2641+ }
2642+ filename = argv[optind++];
2643+ dirname = argv[optind++];
2644+
2645+ Error *errp = NULL;
2646+ VmaReader *vmar = vma_reader_create(filename, &errp);
2647+
2648+ if (!vmar) {
2649+ g_error("%s", error_get_pretty(errp));
2650+ }
2651+
2652+ if (mkdir(dirname, 0777) < 0) {
2653+ g_error("unable to create target directory %s - %s",
2654+ dirname, g_strerror(errno));
2655+ }
2656+
2657+ GList *l = vma_reader_get_config_data(vmar);
2658+ while (l && l->data) {
2659+ VmaConfigData *cdata = (VmaConfigData *)l->data;
2660+ l = g_list_next(l);
2661+ char *cfgfn = g_strdup_printf("%s/%s", dirname, cdata->name);
2662+ GError *err = NULL;
2663+ if (!g_file_set_contents(cfgfn, (gchar *)cdata->data, cdata->len,
2664+ &err)) {
2665+ g_error("unable to write file: %s", err->message);
2666+ }
2667+ }
2668+
2669+ GHashTable *devmap = g_hash_table_new(g_str_hash, g_str_equal);
2670+
2671+ if (readmap) {
2672+ print_content(vmar);
2673+
2674+ FILE *map = fopen(readmap, "r");
2675+ if (!map) {
2676+ g_error("unable to open fifo %s - %s", readmap, g_strerror(errno));
2677+ }
2678+
2679+ while (1) {
2680+ char inbuf[8192];
2681+ char *line = fgets(inbuf, sizeof(inbuf), map);
2682+ if (!line || line[0] == '\0' || !strcmp(line, "done\n")) {
2683+ break;
2684+ }
2685+ int len = strlen(line);
2686+ if (line[len - 1] == '\n') {
2687+ line[len - 1] = '\0';
2688+ if (len == 1) {
2689+ break;
2690+ }
2691+ }
2692+
67af0fa4
WB		 2693+ char *format = NULL;
2694+ if (strncmp(line, "format=", sizeof("format=")-1) == 0) {
2695+ format = line + sizeof("format=")-1;
2696+ char *colon = strchr(format, ':');
2697+ if (!colon) {
2698+ g_error("read map failed - found only a format ('%s')", inbuf);
2699+ }
2700+ format = g_strndup(format, colon - format);
2701+ line = colon+1;
2702+ }
2703+
95259824
WB		 2704+ const char *path;
2705+ bool write_zero;
2706+ if (line[0] == '0' && line[1] == ':') {
67af0fa4 2707+ path = line + 2;
95259824
WB		 2708+ write_zero = false;
2709+ } else if (line[0] == '1' && line[1] == ':') {
67af0fa4 2710+ path = line + 2;
95259824
WB		 2711+ write_zero = true;
2712+ } else {
2713+ g_error("read map failed - parse error ('%s')", inbuf);
2714+ }
2715+
2716+ char *devname = NULL;
2717+ path = extract_devname(path, &devname, -1);
2718+ if (!devname) {
2719+ g_error("read map failed - no dev name specified ('%s')",
2720+ inbuf);
2721+ }
2722+
2723+ RestoreMap *map = g_new0(RestoreMap, 1);
2724+ map->devname = g_strdup(devname);
2725+ map->path = g_strdup(path);
67af0fa4 2726+ map->format = format;
95259824
WB		 2727+ map->write_zero = write_zero;
2728+
2729+ g_hash_table_insert(devmap, map->devname, map);
2730+
2731+ };
2732+ }
2733+
2734+ int i;
2735+ int vmstate_fd = -1;
2736+ guint8 vmstate_stream = 0;
2737+
67af0fa4
WB		 2738+ BlockBackend *blk = NULL;
2739+
95259824
WB		 2740+ for (i = 1; i < 255; i++) {
2741+ VmaDeviceInfo *di = vma_reader_get_device_info(vmar, i);
2742+ if (di && (strcmp(di->devname, "vmstate") == 0)) {
2743+ vmstate_stream = i;
2744+ char *statefn = g_strdup_printf("%s/vmstate.bin", dirname);
2745+ vmstate_fd = open(statefn, O_WRONLY|O_CREAT|O_EXCL, 0644);
2746+ if (vmstate_fd < 0) {
2747+ g_error("create vmstate file '%s' failed - %s", statefn,
2748+ g_strerror(errno));
2749+ }
2750+ g_free(statefn);
2751+ } else if (di) {
2752+ char *devfn = NULL;
67af0fa4
WB		 2753+ const char *format = NULL;
2754+ int flags = BDRV_O_RDWR | BDRV_O_NO_FLUSH;
95259824
WB		 2755+ bool write_zero = true;
2756+
2757+ if (readmap) {
2758+ RestoreMap *map;
2759+ map = (RestoreMap *)g_hash_table_lookup(devmap, di->devname);
2760+ if (map == NULL) {
2761+ g_error("no device name mapping for %s", di->devname);
2762+ }
2763+ devfn = map->path;
67af0fa4 2764+ format = map->format;
95259824
WB		 2765+ write_zero = map->write_zero;
2766+ } else {
2767+ devfn = g_strdup_printf("%s/tmp-disk-%s.raw",
2768+ dirname, di->devname);
2769+ printf("DEVINFO %s %zd\n", devfn, di->size);
2770+
2771+ bdrv_img_create(devfn, "raw", NULL, NULL, NULL, di->size,
2772+ flags, &errp, 0);
2773+ if (errp) {
2774+ g_error("can't create file %s: %s", devfn,
2775+ error_get_pretty(errp));
2776+ }
2777+
2778+ /* Note: we created an empty file above, so there is no
2779+ * need to write zeroes (so we generate a sparse file)
2780+ */
2781+ write_zero = false;
2782+ }
2783+
67af0fa4
WB		 2784+ size_t devlen = strlen(devfn);
2785+ QDict *options = NULL;
2786+ if (format) {
2787+ /* explicit format from commandline */
2788+ options = qdict_new();
2789+ qdict_put(options, "driver", qstring_from_str(format));
2790+ } else if ((devlen > 4 && strcmp(devfn+devlen-4, ".raw") == 0) ||
2791+ strncmp(devfn, "/dev/", 5) == 0)
2792+ {
2793+ /* This part is now deprecated for PVE as well (just as qemu
2794+ * deprecated not specifying an explicit raw format, too.
2795+ */
2796+ /* explicit raw format */
2797+ options = qdict_new();
2798+ qdict_put(options, "driver", qstring_from_str("raw"));
2799+ }
2800+
2801+
2802+ if (errp || !(blk = blk_new_open(devfn, NULL, options, flags, &errp))) {
95259824
WB		 2803+ g_error("can't open file %s - %s", devfn,
2804+ error_get_pretty(errp));
2805+ }
67af0fa4
WB		 2806+
2807+ if (vma_reader_register_bs(vmar, i, blk, write_zero, &errp) < 0) {
95259824
WB		 2808+ g_error("%s", error_get_pretty(errp));
2809+ }
2810+
2811+ if (!readmap) {
2812+ g_free(devfn);
2813+ }
2814+ }
2815+ }
2816+
2817+ if (vma_reader_restore(vmar, vmstate_fd, verbose, &errp) < 0) {
2818+ g_error("restore failed - %s", error_get_pretty(errp));
2819+ }
2820+
2821+ if (!readmap) {
2822+ for (i = 1; i < 255; i++) {
2823+ VmaDeviceInfo *di = vma_reader_get_device_info(vmar, i);
2824+ if (di && (i != vmstate_stream)) {
2825+ char *tmpfn = g_strdup_printf("%s/tmp-disk-%s.raw",
2826+ dirname, di->devname);
2827+ char *fn = g_strdup_printf("%s/disk-%s.raw",
2828+ dirname, di->devname);
2829+ if (rename(tmpfn, fn) != 0) {
2830+ g_error("rename %s to %s failed - %s",
2831+ tmpfn, fn, g_strerror(errno));
2832+ }
2833+ }
2834+ }
2835+ }
2836+
2837+ vma_reader_destroy(vmar);
2838+
67af0fa4
WB		 2839+ blk_unref(blk);
2840+
2841+ bdrv_close_all();
2842+
2843+ return ret;
2844+}
2845+
2846+static int verify_content(int argc, char **argv)
2847+{
2848+ int c, ret = 0;
2849+ int verbose = 0;
2850+ const char *filename;
2851+
2852+ for (;;) {
2853+ c = getopt(argc, argv, "hv");
2854+ if (c == -1) {
2855+ break;
2856+ }
2857+ switch (c) {
2858+ case '?':
2859+ case 'h':
2860+ help();
2861+ break;
2862+ case 'v':
2863+ verbose = 1;
2864+ break;
2865+ default:
2866+ help();
2867+ }
2868+ }
2869+
2870+ /* Get the filename */
2871+ if ((optind + 1) != argc) {
2872+ help();
2873+ }
2874+ filename = argv[optind++];
2875+
2876+ Error *errp = NULL;
2877+ VmaReader *vmar = vma_reader_create(filename, &errp);
2878+
2879+ if (!vmar) {
2880+ g_error("%s", error_get_pretty(errp));
2881+ }
2882+
2883+ if (verbose) {
2884+ print_content(vmar);
2885+ }
2886+
2887+ if (vma_reader_verify(vmar, verbose, &errp) < 0) {
2888+ g_error("verify failed - %s", error_get_pretty(errp));
2889+ }
2890+
2891+ vma_reader_destroy(vmar);
2892+
95259824
WB		 2893+ bdrv_close_all();
2894+
2895+ return ret;
2896+}
2897+
2898+typedef struct BackupJob {
67af0fa4 2899+ BlockBackend *target;
95259824
WB		 2900+ int64_t len;
2901+ VmaWriter *vmaw;
2902+ uint8_t dev_id;
2903+} BackupJob;
2904+
2905+#define BACKUP_SECTORS_PER_CLUSTER (VMA_CLUSTER_SIZE / BDRV_SECTOR_SIZE)
2906+
67af0fa4
WB		 2907+static void coroutine_fn backup_run_empty(void *opaque)
2908+{
2909+ VmaWriter *vmaw = (VmaWriter *)opaque;
2910+
2911+ vma_writer_flush_output(vmaw);
2912+
2913+ Error *err = NULL;
2914+ if (vma_writer_close(vmaw, &err) != 0) {
2915+ g_warning("vma_writer_close failed %s", error_get_pretty(err));
2916+ }
2917+}
2918+
95259824
WB		 2919+static void coroutine_fn backup_run(void *opaque)
2920+{
2921+ BackupJob *job = (BackupJob *)opaque;
2922+ struct iovec iov;
2923+ QEMUIOVector qiov;
2924+
2925+ int64_t start, end;
2926+ int ret = 0;
2927+
67af0fa4 2928+ unsigned char *buf = blk_blockalign(job->target, VMA_CLUSTER_SIZE);
95259824
WB		 2929+
2930+ start = 0;
2931+ end = DIV_ROUND_UP(job->len / BDRV_SECTOR_SIZE,
2932+ BACKUP_SECTORS_PER_CLUSTER);
2933+
2934+ for (; start < end; start++) {
2935+ iov.iov_base = buf;
2936+ iov.iov_len = VMA_CLUSTER_SIZE;
2937+ qemu_iovec_init_external(&qiov, &iov, 1);
2938+
67af0fa4
WB		 2939+ ret = blk_co_preadv(job->target, start * VMA_CLUSTER_SIZE,
2940+ VMA_CLUSTER_SIZE, &qiov, 0);
95259824
WB		 2941+ if (ret < 0) {
2942+ vma_writer_set_error(job->vmaw, "read error", -1);
2943+ goto out;
2944+ }
2945+
2946+ size_t zb = 0;
2947+ if (vma_writer_write(job->vmaw, job->dev_id, start, buf, &zb) < 0) {
2948+ vma_writer_set_error(job->vmaw, "backup_dump_cb vma_writer_write failed", -1);
2949+ goto out;
2950+ }
2951+ }
2952+
2953+
2954+out:
2955+ if (vma_writer_close_stream(job->vmaw, job->dev_id) <= 0) {
2956+ Error *err = NULL;
2957+ if (vma_writer_close(job->vmaw, &err) != 0) {
2958+ g_warning("vma_writer_close failed %s", error_get_pretty(err));
2959+ }
2960+ }
2961+}
2962+
2963+static int create_archive(int argc, char **argv)
2964+{
2965+ int i, c;
2966+ int verbose = 0;
2967+ const char *archivename;
2968+ GList *config_files = NULL;
2969+
2970+ for (;;) {
2971+ c = getopt(argc, argv, "hvc:");
2972+ if (c == -1) {
2973+ break;
2974+ }
2975+ switch (c) {
2976+ case '?':
2977+ case 'h':
2978+ help();
2979+ break;
2980+ case 'c':
2981+ config_files = g_list_append(config_files, optarg);
2982+ break;
2983+ case 'v':
2984+ verbose = 1;
2985+ break;
2986+ default:
2987+ g_assert_not_reached();
2988+ }
2989+ }
2990+
2991+
67af0fa4
WB		 2992+ /* make sure we an archive name */
2993+ if ((optind + 1) > argc) {
95259824
WB		 2994+ help();
2995+ }
2996+
2997+ archivename = argv[optind++];
2998+
2999+ uuid_t uuid;
3000+ uuid_generate(uuid);
3001+
3002+ Error *local_err = NULL;
3003+ VmaWriter *vmaw = vma_writer_create(archivename, uuid, &local_err);
3004+
3005+ if (vmaw == NULL) {
3006+ g_error("%s", error_get_pretty(local_err));
3007+ }
3008+
3009+ GList *l = config_files;
3010+ while (l && l->data) {
3011+ char *name = l->data;
3012+ char *cdata = NULL;
3013+ gsize clen = 0;
3014+ GError *err = NULL;
3015+ if (!g_file_get_contents(name, &cdata, &clen, &err)) {
3016+ unlink(archivename);
3017+ g_error("Unable to read file: %s", err->message);
3018+ }
3019+
3020+ if (vma_writer_add_config(vmaw, name, cdata, clen) != 0) {
3021+ unlink(archivename);
3022+ g_error("Unable to append config data %s (len = %zd)",
3023+ name, clen);
3024+ }
3025+ l = g_list_next(l);
3026+ }
3027+
67af0fa4 3028+ int devcount = 0;
95259824
WB		 3029+ while (optind < argc) {
3030+ const char *path = argv[optind++];
3031+ char *devname = NULL;
67af0fa4 3032+ path = extract_devname(path, &devname, devcount++);
95259824
WB		 3033+
3034+ Error *errp = NULL;
67af0fa4 3035+ BlockBackend *target;
95259824 3036+
67af0fa4
WB		 3037+ target = blk_new_open(path, NULL, NULL, 0, &errp);
3038+ if (!target) {
95259824
WB		 3039+ unlink(archivename);
3040+ g_error("bdrv_open '%s' failed - %s", path, error_get_pretty(errp));
3041+ }
67af0fa4 3042+ int64_t size = blk_getlength(target);
95259824
WB		 3043+ int dev_id = vma_writer_register_stream(vmaw, devname, size);
3044+ if (dev_id <= 0) {
3045+ unlink(archivename);
3046+ g_error("vma_writer_register_stream '%s' failed", devname);
3047+ }
3048+
3049+ BackupJob *job = g_new0(BackupJob, 1);
3050+ job->len = size;
67af0fa4 3051+ job->target = target;
95259824
WB		 3052+ job->vmaw = vmaw;
3053+ job->dev_id = dev_id;
3054+
3055+ Coroutine *co = qemu_coroutine_create(backup_run, job);
3056+ qemu_coroutine_enter(co);
3057+ }
3058+
3059+ VmaStatus vmastat;
3060+ int percent = 0;
3061+ int last_percent = -1;
3062+
67af0fa4
WB		 3063+ if (devcount) {
3064+ while (1) {
3065+ main_loop_wait(false);
3066+ vma_writer_get_status(vmaw, &vmastat);
95259824 3067+
67af0fa4 3068+ if (verbose) {
95259824 3069+
67af0fa4
WB		 3070+ uint64_t total = 0;
3071+ uint64_t transferred = 0;
3072+ uint64_t zero_bytes = 0;
95259824 3073+
67af0fa4
WB		 3074+ int i;
3075+ for (i = 0; i < 256; i++) {
3076+ if (vmastat.stream_info[i].size) {
3077+ total += vmastat.stream_info[i].size;
3078+ transferred += vmastat.stream_info[i].transferred;
3079+ zero_bytes += vmastat.stream_info[i].zero_bytes;
3080+ }
95259824 3081+ }
67af0fa4
WB		 3082+ percent = (transferred*100)/total;
3083+ if (percent != last_percent) {
3084+ fprintf(stderr, "progress %d%% %zd/%zd %zd\n", percent,
3085+ transferred, total, zero_bytes);
3086+ fflush(stderr);
95259824 3087+
67af0fa4
WB		 3088+ last_percent = percent;
3089+ }
95259824 3090+ }
95259824 3091+
67af0fa4
WB		 3092+ if (vmastat.closed) {
3093+ break;
3094+ }
95259824
WB		 3095+ }
3096+ } else {
3097+ Coroutine *co = qemu_coroutine_create(backup_run_empty, vmaw);
3098+ qemu_coroutine_enter(co);
3099+ while (1) {
3100+ main_loop_wait(false);
3101+ vma_writer_get_status(vmaw, &vmastat);
3102+ if (vmastat.closed) {
3103+ break;
3104+ }
3105+ }
3106+ }
3107+
3108+ bdrv_drain_all();
3109+
3110+ vma_writer_get_status(vmaw, &vmastat);
3111+
3112+ if (verbose) {
3113+ for (i = 0; i < 256; i++) {
3114+ VmaStreamInfo *si = &vmastat.stream_info[i];
3115+ if (si->size) {
3116+ fprintf(stderr, "image %s: size=%zd zeros=%zd saved=%zd\n",
3117+ si->devname, si->size, si->zero_bytes,
3118+ si->size - si->zero_bytes);
3119+ }
3120+ }
3121+ }
3122+
3123+ if (vmastat.status < 0) {
3124+ unlink(archivename);
3125+ g_error("creating vma archive failed");
3126+ }
3127+
3128+ return 0;
3129+}
3130+
67af0fa4
WB		 3131+static int dump_config(int argc, char **argv)
3132+{
3133+ int c, ret = 0;
3134+ const char *filename;
3135+ const char *config_name = "qemu-server.conf";
3136+
3137+ for (;;) {
3138+ c = getopt(argc, argv, "hc:");
3139+ if (c == -1) {
3140+ break;
3141+ }
3142+ switch (c) {
3143+ case '?':
3144+ case 'h':
3145+ help();
3146+ break;
3147+ case 'c':
3148+ config_name = optarg;
3149+ break;
3150+ default:
3151+ help();
3152+ }
3153+ }
3154+
3155+ /* Get the filename */
3156+ if ((optind + 1) != argc) {
3157+ help();
3158+ }
3159+ filename = argv[optind++];
3160+
3161+ Error *errp = NULL;
3162+ VmaReader *vmar = vma_reader_create(filename, &errp);
3163+
3164+ if (!vmar) {
3165+ g_error("%s", error_get_pretty(errp));
3166+ }
3167+
3168+ int found = 0;
3169+ GList *l = vma_reader_get_config_data(vmar);
3170+ while (l && l->data) {
3171+ VmaConfigData *cdata = (VmaConfigData *)l->data;
3172+ l = g_list_next(l);
3173+ if (strcmp(cdata->name, config_name) == 0) {
3174+ found = 1;
3175+ fwrite(cdata->data, cdata->len, 1, stdout);
3176+ break;
3177+ }
3178+ }
3179+
3180+ vma_reader_destroy(vmar);
3181+
3182+ bdrv_close_all();
3183+
3184+ if (!found) {
3185+ fprintf(stderr, "unable to find configuration data '%s'\n", config_name);
3186+ return -1;
3187+ }
3188+
3189+ return ret;
3190+}
3191+
95259824
WB		 3192+int main(int argc, char **argv)
3193+{
3194+ const char *cmdname;
3195+ Error *main_loop_err = NULL;
3196+
3197+ error_set_progname(argv[0]);
3198+
3199+ if (qemu_init_main_loop(&main_loop_err)) {
3200+ g_error("%s", error_get_pretty(main_loop_err));
3201+ }
3202+
3203+ bdrv_init();
3204+
3205+ if (argc < 2) {
3206+ help();
3207+ }
3208+
3209+ cmdname = argv[1];
3210+ argc--; argv++;
3211+
3212+
3213+ if (!strcmp(cmdname, "list")) {
3214+ return list_content(argc, argv);
3215+ } else if (!strcmp(cmdname, "create")) {
3216+ return create_archive(argc, argv);
3217+ } else if (!strcmp(cmdname, "extract")) {
3218+ return extract_content(argc, argv);
67af0fa4
WB		 3219+ } else if (!strcmp(cmdname, "verify")) {
3220+ return verify_content(argc, argv);
3221+ } else if (!strcmp(cmdname, "config")) {
3222+ return dump_config(argc, argv);
95259824
WB		 3223+ }
3224+
3225+ help();
3226+ return 0;
3227+}
3228diff --git a/vma.h b/vma.h
3229new file mode 100644
67af0fa4 3230index 0000000000..fa6f4df7e7
95259824
WB		 3231--- /dev/null
3232+++ b/vma.h
67af0fa4 3233@@ -0,0 +1,149 @@
95259824
WB		 3234+/*
3235+ * VMA: Virtual Machine Archive
3236+ *
3237+ * Copyright (C) Proxmox Server Solutions
3238+ *
3239+ * Authors:
3240+ * Dietmar Maurer (dietmar@proxmox.com)
3241+ *
3242+ * This work is licensed under the terms of the GNU GPL, version 2 or later.
3243+ * See the COPYING file in the top-level directory.
3244+ *
3245+ */
3246+
3247+#ifndef BACKUP_VMA_H
3248+#define BACKUP_VMA_H
3249+
3250+#include <uuid/uuid.h>
3251+#include "qapi/error.h"
3252+#include "block/block.h"
3253+
3254+#define VMA_BLOCK_BITS 12
3255+#define VMA_BLOCK_SIZE (1<<VMA_BLOCK_BITS)
3256+#define VMA_CLUSTER_BITS (VMA_BLOCK_BITS+4)
3257+#define VMA_CLUSTER_SIZE (1<<VMA_CLUSTER_BITS)
3258+
3259+#if VMA_CLUSTER_SIZE != 65536
3260+#error unexpected cluster size
3261+#endif
3262+
3263+#define VMA_EXTENT_HEADER_SIZE 512
3264+#define VMA_BLOCKS_PER_EXTENT 59
3265+#define VMA_MAX_CONFIGS 256
3266+
3267+#define VMA_MAX_EXTENT_SIZE \
3268+ (VMA_EXTENT_HEADER_SIZE+VMA_CLUSTER_SIZE*VMA_BLOCKS_PER_EXTENT)
3269+#if VMA_MAX_EXTENT_SIZE != 3867136
3270+#error unexpected VMA_EXTENT_SIZE
3271+#endif
3272+
3273+/* File Format Definitions */
3274+
3275+#define VMA_MAGIC (GUINT32_TO_BE(('V'<<24)|('M'<<16)|('A'<<8)|0x00))
3276+#define VMA_EXTENT_MAGIC (GUINT32_TO_BE(('V'<<24)|('M'<<16)|('A'<<8)|'E'))
3277+
3278+typedef struct VmaDeviceInfoHeader {
3279+ uint32_t devname_ptr; /* offset into blob_buffer table */
3280+ uint32_t reserved0;
3281+ uint64_t size; /* device size in bytes */
3282+ uint64_t reserved1;
3283+ uint64_t reserved2;
3284+} VmaDeviceInfoHeader;
3285+
3286+typedef struct VmaHeader {
3287+ uint32_t magic;
3288+ uint32_t version;
3289+ unsigned char uuid[16];
3290+ int64_t ctime;
3291+ unsigned char md5sum[16];
3292+
3293+ uint32_t blob_buffer_offset;
3294+ uint32_t blob_buffer_size;
3295+ uint32_t header_size;
3296+
3297+ unsigned char reserved[1984];
3298+
3299+ uint32_t config_names[VMA_MAX_CONFIGS]; /* offset into blob_buffer table */
3300+ uint32_t config_data[VMA_MAX_CONFIGS]; /* offset into blob_buffer table */
3301+
3302+ uint32_t reserved1;
3303+
3304+ VmaDeviceInfoHeader dev_info[256];
3305+} VmaHeader;
3306+
3307+typedef struct VmaExtentHeader {
3308+ uint32_t magic;
3309+ uint16_t reserved1;
3310+ uint16_t block_count;
3311+ unsigned char uuid[16];
3312+ unsigned char md5sum[16];
3313+ uint64_t blockinfo[VMA_BLOCKS_PER_EXTENT];
3314+} VmaExtentHeader;
3315+
3316+/* functions/definitions to read/write vma files */
3317+
3318+typedef struct VmaReader VmaReader;
3319+
3320+typedef struct VmaWriter VmaWriter;
3321+
3322+typedef struct VmaConfigData {
3323+ const char *name;
3324+ const void *data;
3325+ uint32_t len;
3326+} VmaConfigData;
3327+
3328+typedef struct VmaStreamInfo {
3329+ uint64_t size;
3330+ uint64_t cluster_count;
3331+ uint64_t transferred;
3332+ uint64_t zero_bytes;
3333+ int finished;
3334+ char *devname;
3335+} VmaStreamInfo;
3336+
3337+typedef struct VmaStatus {
3338+ int status;
3339+ bool closed;
3340+ char errmsg[8192];
3341+ char uuid_str[37];
3342+ VmaStreamInfo stream_info[256];
3343+} VmaStatus;
3344+
3345+typedef struct VmaDeviceInfo {
3346+ uint64_t size; /* device size in bytes */
3347+ const char *devname;
3348+} VmaDeviceInfo;
3349+
3350+VmaWriter *vma_writer_create(const char *filename, uuid_t uuid, Error **errp);
3351+int vma_writer_close(VmaWriter *vmaw, Error **errp);
67af0fa4 3352+void vma_writer_error_propagate(VmaWriter *vmaw, Error **errp);
95259824
WB		 3353+void vma_writer_destroy(VmaWriter *vmaw);
3354+int vma_writer_add_config(VmaWriter *vmaw, const char *name, gpointer data,
3355+ size_t len);
3356+int vma_writer_register_stream(VmaWriter *vmaw, const char *devname,
3357+ size_t size);
3358+
3359+int64_t coroutine_fn vma_writer_write(VmaWriter *vmaw, uint8_t dev_id,
3360+ int64_t cluster_num, unsigned char *buf,
3361+ size_t *zero_bytes);
3362+
3363+int coroutine_fn vma_writer_close_stream(VmaWriter *vmaw, uint8_t dev_id);
67af0fa4 3364+int coroutine_fn vma_writer_flush_output(VmaWriter *vmaw);
95259824
WB		 3365+
3366+int vma_writer_get_status(VmaWriter *vmaw, VmaStatus *status);
3367+void vma_writer_set_error(VmaWriter *vmaw, const char *fmt, ...);
3368+
3369+
3370+VmaReader *vma_reader_create(const char *filename, Error **errp);
3371+void vma_reader_destroy(VmaReader *vmar);
3372+VmaHeader *vma_reader_get_header(VmaReader *vmar);
3373+GList *vma_reader_get_config_data(VmaReader *vmar);
3374+VmaDeviceInfo *vma_reader_get_device_info(VmaReader *vmar, guint8 dev_id);
3375+int vma_reader_register_bs(VmaReader *vmar, guint8 dev_id,
67af0fa4 3376+ BlockBackend *target, bool write_zeroes,
95259824
WB		 3377+ Error **errp);
3378+int vma_reader_restore(VmaReader *vmar, int vmstate_fd, bool verbose,
3379+ Error **errp);
67af0fa4 3380+int vma_reader_verify(VmaReader *vmar, bool verbose, Error **errp);
95259824
WB		 3381+
3382+#endif /* BACKUP_VMA_H */
3383--
45169293 33842.11.0
95259824 3385
QEMU for PVE