]> 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
rebased pve patches
[pve-qemu.git] / debian / patches / pve / 0028-adding-old-vma-files.patch
CommitLineData
67af0fa4
WB		 1From cf7632dba8a8b7191f5e0c2ba826c6827fecb412 Mon Sep 17 00:00:00 2001
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 +
11 blockdev.c | 245 ++++++++-----
12 blockjob.c | 11 +-
13 include/block/block_int.h | 4 +
14 vma-reader.c | 857 ++++++++++++++++++++++++++++++++++++++++++++++
15 vma-writer.c | 771 +++++++++++++++++++++++++++++++++++++++++
16 vma.c | 757 ++++++++++++++++++++++++++++++++++++++++
17 vma.h | 149 ++++++++
18 11 files changed, 2794 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
294index 981276692a..1eeffb0571 100644
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;
369@@ -2988,13 +3013,74 @@ typedef struct PVEBackupDevInfo {
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;
399+ n_sectors = n_sectors * BDRV_SECTOR_SIZE / VMA_BLOCK_SIZE;
400+ while (n_sectors >= VMA_CLUSTER_BITS) {
401+ ret = vma_writer_write(backup_state.vmaw, di->dev_id, cluster_num,
402+ buf, &zero_bytes);
403+ n_sectors -= VMA_CLUSTER_BITS;
404+ ++cluster_num;
405+ if (buf) {
406+ buf += VMA_CLUSTER_SIZE;
407+ }
408+ if (ret < 0) {
409+ if (!backup_state.error) {
410+ vma_writer_error_propagate(backup_state.vmaw, &backup_state.error);
411+ }
412+ if (di->bs && di->bs->job) {
413+ block_job_cancel(di->bs->job);
414+ }
415+ } else {
416+ backup_state.zero_bytes += zero_bytes;
417+ backup_state.transferred += VMA_CLUSTER_SIZE;
418+ }
419+ }
420+ } else {
421+ if (!buf) {
422+ backup_state.zero_bytes += size;
423+ }
424+ backup_state.transferred += size;
425+ }
426+
427+ // Note: always return success, because we want that writes succeed anyways.
428+
429+ return size;
430+}
431+
432 static void pvebackup_cleanup(void)
433 {
434 backup_state.end_time = time(NULL);
435
436- if (backup_state.vmaobj) {
437- object_unparent(backup_state.vmaobj);
438- backup_state.vmaobj = NULL;
439+ if (backup_state.vmaw) {
440+ Error *local_err = NULL;
441+ vma_writer_close(backup_state.vmaw, &local_err);
442+ error_propagate(&backup_state.error, local_err);
443+ backup_state.vmaw = NULL;
444 }
445
446 if (backup_state.di_list) {
447@@ -3009,6 +3095,13 @@ static void pvebackup_cleanup(void)
448 }
449 }
450
451+static void coroutine_fn backup_close_vma_stream(void *opaque)
452+{
453+ PVEBackupDevInfo *di = opaque;
454+
455+ vma_writer_close_stream(backup_state.vmaw, di->dev_id);
456+}
457+
458 static void pvebackup_complete_cb(void *opaque, int ret)
459 {
460 PVEBackupDevInfo *di = opaque;
461@@ -3020,14 +3113,18 @@ static void pvebackup_complete_cb(void *opaque, int ret)
462 ret, strerror(-ret));
463 }
464
465+ BlockDriverState *bs = di->bs;
466+
467 di->bs = NULL;
468 di->target = NULL;
469
470- if (backup_state.vmaobj) {
471- object_unparent(backup_state.vmaobj);
472- backup_state.vmaobj = NULL;
473+ if (backup_state.vmaw) {
474+ Coroutine *co = qemu_coroutine_create(backup_close_vma_stream, di);
475+ qemu_coroutine_enter(co);
476 }
477
478+ block_job_cb(bs, ret);
479+
480 if (!backup_state.cancel) {
481 pvebackup_run_next_job();
482 }
483@@ -3041,14 +3138,9 @@ static void pvebackup_cancel(void *opaque)
484 error_setg(&backup_state.error, "backup cancelled");
485 }
486
487- if (backup_state.vmaobj) {
488- Error *err;
489+ if (backup_state.vmaw) {
490 /* make sure vma writer does not block anymore */
491- if (!object_set_props(backup_state.vmaobj, &err, "blocked", "yes", NULL)) {
492- if (err) {
493- error_report_err(err);
494- }
495- }
496+ vma_writer_set_error(backup_state.vmaw, "backup cancelled");
497 }
498
499 GList *l = backup_state.di_list;
500@@ -3073,19 +3165,15 @@ void qmp_backup_cancel(Error **errp)
501 Coroutine *co = qemu_coroutine_create(pvebackup_cancel, NULL);
502 qemu_coroutine_enter(co);
503
504- while (backup_state.vmaobj) {
505- /* FIXME: Find something better for this */
506+ while (backup_state.vmaw) {
507+ /* vma writer use main aio context */
508 aio_poll(qemu_get_aio_context(), true);
509 }
510 }
511
512-void vma_object_add_config_file(Object *obj, const char *name,
513- const char *contents, size_t len,
514- Error **errp);
515 static int config_to_vma(const char *file, BackupFormat format,
516- Object *vmaobj,
517- const char *backup_dir,
518- Error **errp)
519+ const char *backup_dir, VmaWriter *vmaw,
520+ Error **errp)
521 {
522 char *cdata = NULL;
523 gsize clen = 0;
524@@ -3098,12 +3186,17 @@ static int config_to_vma(const char *file, BackupFormat format,
525 char *basename = g_path_get_basename(file);
526
527 if (format == BACKUP_FORMAT_VMA) {
528- vma_object_add_config_file(vmaobj, basename, cdata, clen, errp);
529+ if (vma_writer_add_config(vmaw, basename, cdata, clen) != 0) {
530+ error_setg(errp, "unable to add %s config data to vma archive", file);
531+ g_free(cdata);
532+ g_free(basename);
533+ return 1;
534+ }
535 } else if (format == BACKUP_FORMAT_DIR) {
536 char config_path[PATH_MAX];
537 snprintf(config_path, PATH_MAX, "%s/%s", backup_dir, basename);
538 if (!g_file_set_contents(config_path, cdata, clen, &err)) {
539- error_setg(errp, "unable to write config file '%s'", config_path);
540+ error_setg(errp, "unable to write config file '%s'", config_path);
541 g_free(cdata);
542 g_free(basename);
543 return 1;
544@@ -3113,35 +3206,37 @@ static int config_to_vma(const char *file, BackupFormat format,
545 g_free(basename);
546 g_free(cdata);
547
548- return 0;
549+ return 0;
550 }
551
552+bool block_job_should_pause(BlockJob *job);
553 static void pvebackup_run_next_job(void)
554 {
555- bool cancel = backup_state.error || backup_state.cancel;
556-fprintf(stderr, "run next job: %zu\n", backup_state.next_job);
557- GList *next = g_list_nth(backup_state.di_list, backup_state.next_job);
558- while (next) {
559- PVEBackupDevInfo *di = (PVEBackupDevInfo *)next->data;
560- backup_state.next_job++;
561+ GList *l = backup_state.di_list;
562+ while (l) {
563+ PVEBackupDevInfo *di = (PVEBackupDevInfo *)l->data;
564+ l = g_list_next(l);
565 if (!di->completed && di->bs && di->bs->job) {
566 BlockJob *job = di->bs->job;
567- if (cancel) {
568- block_job_cancel(job);
569- } else {
570- block_job_resume(job);
571+ if (block_job_should_pause(job)) {
572+ bool cancel = backup_state.error || backup_state.cancel;
573+ if (cancel) {
574+ block_job_cancel(job);
575+ } else {
576+ block_job_resume(job);
577+ }
578 }
579 return;
580 }
581- next = g_list_next(next);
582 }
583+
584 pvebackup_cleanup();
585 }
586
587 UuidInfo *qmp_backup(const char *backup_file, bool has_format,
588 BackupFormat format,
589 bool has_config_file, const char *config_file,
590- bool has_firewall_file, const char *firewall_file,
591+ bool has_firewall_file, const char *firewall_file,
592 bool has_devlist, const char *devlist,
593 bool has_speed, int64_t speed, Error **errp)
594 {
595@@ -3149,14 +3244,15 @@ UuidInfo *qmp_backup(const char *backup_file, bool has_format,
596 BlockDriverState *bs = NULL;
597 const char *backup_dir = NULL;
598 Error *local_err = NULL;
599- QemuUUID uuid;
600+ uuid_t uuid;
601+ VmaWriter *vmaw = NULL;
602 gchar **devs = NULL;
603 GList *di_list = NULL;
604 GList *l;
605 UuidInfo *uuid_info;
606 BlockJob *job;
607
608- if (backup_state.di_list || backup_state.vmaobj) {
609+ if (backup_state.di_list) {
610 error_set(errp, ERROR_CLASS_GENERIC_ERROR,
611 "previous backup not finished");
612 return NULL;
613@@ -3231,40 +3327,28 @@ UuidInfo *qmp_backup(const char *backup_file, bool has_format,
614 total += size;
615 }
616
617- qemu_uuid_generate(&uuid);
618+ uuid_generate(uuid);
619
620 if (format == BACKUP_FORMAT_VMA) {
621- char uuidstr[UUID_FMT_LEN+1];
622- qemu_uuid_unparse(&uuid, uuidstr);
623- uuidstr[UUID_FMT_LEN] = 0;
624- backup_state.vmaobj =
625- object_new_with_props("vma", object_get_objects_root(),
626- "vma-backup-obj", &local_err,
627- "filename", backup_file,
628- "uuid", uuidstr,
629- NULL);
630- if (!backup_state.vmaobj) {
631+ vmaw = vma_writer_create(backup_file, uuid, &local_err);
632+ if (!vmaw) {
633 if (local_err) {
634 error_propagate(errp, local_err);
635 }
636 goto err;
637 }
638
639+ /* register all devices for vma writer */
640 l = di_list;
641 while (l) {
642- QDict *options = qdict_new();
643-
644 PVEBackupDevInfo *di = (PVEBackupDevInfo *)l->data;
645 l = g_list_next(l);
646
647 const char *devname = bdrv_get_device_name(di->bs);
648- snprintf(di->targetfile, PATH_MAX, "vma-backup-obj/%s.raw", devname);
649-
650- qdict_put(options, "driver", qstring_from_str("vma-drive"));
651- qdict_put(options, "size", qint_from_int(di->size));
652- di->target = bdrv_open(di->targetfile, NULL, options, BDRV_O_RDWR, &local_err);
653- if (!di->target) {
654- error_propagate(errp, local_err);
655+ di->dev_id = vma_writer_register_stream(vmaw, devname, di->size);
656+ if (di->dev_id <= 0) {
657+ error_set(errp, ERROR_CLASS_GENERIC_ERROR,
658+ "register_stream failed");
659 goto err;
660 }
661 }
662@@ -3305,15 +3389,15 @@ UuidInfo *qmp_backup(const char *backup_file, bool has_format,
663
664 /* add configuration file to archive */
665 if (has_config_file) {
666- if(config_to_vma(config_file, format, backup_state.vmaobj, backup_dir, errp) != 0) {
667- goto err;
668+ if(config_to_vma(config_file, format, backup_dir, vmaw, errp) != 0) {
669+ goto err;
670 }
671 }
672
673 /* add firewall file to archive */
674 if (has_firewall_file) {
675- if(config_to_vma(firewall_file, format, backup_state.vmaobj, backup_dir, errp) != 0) {
676- goto err;
677+ if(config_to_vma(firewall_file, format, backup_dir, vmaw, errp) != 0) {
678+ goto err;
679 }
680 }
681 /* initialize global backup_state now */
682@@ -3335,11 +3419,12 @@ UuidInfo *qmp_backup(const char *backup_file, bool has_format,
683 }
684 backup_state.backup_file = g_strdup(backup_file);
685
686- memcpy(&backup_state.uuid, &uuid, sizeof(uuid));
687- qemu_uuid_unparse(&uuid, backup_state.uuid_str);
688+ backup_state.vmaw = vmaw;
689+
690+ uuid_copy(backup_state.uuid, uuid);
691+ uuid_unparse_lower(uuid, backup_state.uuid_str);
692
693 backup_state.di_list = di_list;
694- backup_state.next_job = 0;
695
696 backup_state.total = total;
697 backup_state.transferred = 0;
698@@ -3350,21 +3435,16 @@ UuidInfo *qmp_backup(const char *backup_file, bool has_format,
699 while (l) {
700 PVEBackupDevInfo *di = (PVEBackupDevInfo *)l->data;
701 l = g_list_next(l);
702-
703 job = backup_job_create(NULL, di->bs, di->target, speed, MIRROR_SYNC_MODE_FULL, NULL,
704 false, BLOCKDEV_ON_ERROR_REPORT, BLOCKDEV_ON_ERROR_REPORT,
705 BLOCK_JOB_DEFAULT,
706- pvebackup_complete_cb, di, 2, NULL, &local_err);
707- if (di->target) {
708- bdrv_unref(di->target);
709- di->target = NULL;
710- }
711+ pvebackup_dump_cb, pvebackup_complete_cb, di,
712+ 2, NULL, &local_err);
713 if (!job || local_err != NULL) {
714 error_setg(&backup_state.error, "backup_job_create failed");
715 pvebackup_cancel(NULL);
716- } else {
717- block_job_start(job);
718 }
719+ block_job_start(job);
720 }
721
722 if (!backup_state.error) {
723@@ -3398,9 +3478,10 @@ err:
724 g_strfreev(devs);
725 }
726
727- if (backup_state.vmaobj) {
728- object_unparent(backup_state.vmaobj);
729- backup_state.vmaobj = NULL;
730+ if (vmaw) {
731+ Error *err = NULL;
732+ vma_writer_close(vmaw, &err);
733+ unlink(backup_file);
734 }
735
736 if (backup_dir) {
737@@ -3772,7 +3853,7 @@ static BlockJob *do_drive_backup(DriveBackup *backup, BlockJobTxn *txn,
738 job = backup_job_create(backup->job_id, bs, target_bs, backup->speed,
739 backup->sync, bmap, backup->compress,
740 backup->on_source_error, backup->on_target_error,
741- BLOCK_JOB_DEFAULT, NULL, NULL, 0, txn, &local_err);
742+ BLOCK_JOB_DEFAULT, NULL, NULL, NULL, 0, txn, &local_err);
743 bdrv_unref(target_bs);
744 if (local_err != NULL) {
745 error_propagate(errp, local_err);
746@@ -3851,7 +3932,7 @@ BlockJob *do_blockdev_backup(BlockdevBackup *backup, BlockJobTxn *txn,
747 job = backup_job_create(backup->job_id, bs, target_bs, backup->speed,
748 backup->sync, NULL, backup->compress,
749 backup->on_source_error, backup->on_target_error,
750- BLOCK_JOB_DEFAULT, NULL, NULL, 0, txn, &local_err);
751+ BLOCK_JOB_DEFAULT, NULL, NULL, NULL, 0, txn, &local_err);
752 if (local_err != NULL) {
753 error_propagate(errp, local_err);
754 }
755diff --git a/blockjob.c b/blockjob.c
756index 764d41863e..cb3741f6dd 100644
757--- a/blockjob.c
758+++ b/blockjob.c
759@@ -37,8 +37,8 @@
760 #include "qemu/timer.h"
761 #include "qapi-event.h"
762
763-static void block_job_event_cancelled(BlockJob *job);
764-static void block_job_event_completed(BlockJob *job, const char *msg);
765+void block_job_event_cancelled(BlockJob *job);
766+void block_job_event_completed(BlockJob *job, const char *msg);
767
768 /* Transactional group of block jobs */
769 struct BlockJobTxn {
770@@ -473,7 +473,8 @@ void block_job_user_pause(BlockJob *job)
771 block_job_pause(job);
772 }
773
774-static bool block_job_should_pause(BlockJob *job)
775+bool block_job_should_pause(BlockJob *job);
776+bool block_job_should_pause(BlockJob *job)
777 {
778 return job->pause_count > 0;
779 }
780@@ -687,7 +688,7 @@ static void block_job_iostatus_set_err(BlockJob *job, int error)
781 }
782 }
783
784-static void block_job_event_cancelled(BlockJob *job)
785+void block_job_event_cancelled(BlockJob *job)
786 {
787 if (block_job_is_internal(job)) {
788 return;
789@@ -701,7 +702,7 @@ static void block_job_event_cancelled(BlockJob *job)
790 &error_abort);
791 }
792
793-static void block_job_event_completed(BlockJob *job, const char *msg)
794+void block_job_event_completed(BlockJob *job, const char *msg)
795 {
796 if (block_job_is_internal(job)) {
797 return;
798diff --git a/include/block/block_int.h b/include/block/block_int.h
799index 2b3ecd0575..278da161fb 100644
800--- a/include/block/block_int.h
801+++ b/include/block/block_int.h
802@@ -59,6 +59,9 @@
803
804 #define BLOCK_PROBE_BUF_SIZE 512
805
806+typedef int BackupDumpFunc(void *opaque, BlockBackend *be,
807+ int64_t sector_num, int n_sectors, unsigned char *buf);
808+
809 enum BdrvTrackedRequestType {
810 BDRV_TRACKED_READ,
811 BDRV_TRACKED_WRITE,
812@@ -877,6 +880,7 @@ BlockJob *backup_job_create(const char *job_id, BlockDriverState *bs,
813 BlockdevOnError on_source_error,
814 BlockdevOnError on_target_error,
815 int creation_flags,
816+ BackupDumpFunc *dump_cb,
817 BlockCompletionFunc *cb, void *opaque,
818 int pause_count,
819 BlockJobTxn *txn, Error **errp);
95259824
WB		 820diff --git a/vma-reader.c b/vma-reader.c
821new file mode 100644
67af0fa4 822index 0000000000..2000889bd3
95259824
WB		 823--- /dev/null
824+++ b/vma-reader.c
67af0fa4 825@@ -0,0 +1,857 @@
95259824
WB		 826+/*
827+ * VMA: Virtual Machine Archive
828+ *
829+ * Copyright (C) 2012 Proxmox Server Solutions
830+ *
831+ * Authors:
832+ * Dietmar Maurer (dietmar@proxmox.com)
833+ *
834+ * This work is licensed under the terms of the GNU GPL, version 2 or later.
835+ * See the COPYING file in the top-level directory.
836+ *
837+ */
838+
839+#include "qemu/osdep.h"
840+#include <glib.h>
841+#include <uuid/uuid.h>
842+
843+#include "qemu-common.h"
844+#include "qemu/timer.h"
845+#include "qemu/ratelimit.h"
846+#include "vma.h"
847+#include "block/block.h"
848+#include "sysemu/block-backend.h"
849+
850+static unsigned char zero_vma_block[VMA_BLOCK_SIZE];
851+
852+typedef struct VmaRestoreState {
67af0fa4 853+ BlockBackend *target;
95259824
WB		 854+ bool write_zeroes;
855+ unsigned long *bitmap;
856+ int bitmap_size;
857+} VmaRestoreState;
858+
859+struct VmaReader {
860+ int fd;
861+ GChecksum *md5csum;
862+ GHashTable *blob_hash;
863+ unsigned char *head_data;
864+ VmaDeviceInfo devinfo[256];
865+ VmaRestoreState rstate[256];
866+ GList *cdata_list;
867+ guint8 vmstate_stream;
868+ uint32_t vmstate_clusters;
869+ /* to show restore percentage if run with -v */
870+ time_t start_time;
871+ int64_t cluster_count;
872+ int64_t clusters_read;
67af0fa4
WB		 873+ int64_t zero_cluster_data;
874+ int64_t partial_zero_cluster_data;
95259824
WB		 875+ int clusters_read_per;
876+};
877+
878+static guint
879+g_int32_hash(gconstpointer v)
880+{
881+ return *(const uint32_t *)v;
882+}
883+
884+static gboolean
885+g_int32_equal(gconstpointer v1, gconstpointer v2)
886+{
887+ return *((const uint32_t *)v1) == *((const uint32_t *)v2);
888+}
889+
890+static int vma_reader_get_bitmap(VmaRestoreState *rstate, int64_t cluster_num)
891+{
892+ assert(rstate);
893+ assert(rstate->bitmap);
894+
895+ unsigned long val, idx, bit;
896+
897+ idx = cluster_num / BITS_PER_LONG;
898+
899+ assert(rstate->bitmap_size > idx);
900+
901+ bit = cluster_num % BITS_PER_LONG;
902+ val = rstate->bitmap[idx];
903+
904+ return !!(val & (1UL << bit));
905+}
906+
907+static void vma_reader_set_bitmap(VmaRestoreState *rstate, int64_t cluster_num,
908+ int dirty)
909+{
910+ assert(rstate);
911+ assert(rstate->bitmap);
912+
913+ unsigned long val, idx, bit;
914+
915+ idx = cluster_num / BITS_PER_LONG;
916+
917+ assert(rstate->bitmap_size > idx);
918+
919+ bit = cluster_num % BITS_PER_LONG;
920+ val = rstate->bitmap[idx];
921+ if (dirty) {
922+ if (!(val & (1UL << bit))) {
923+ val |= 1UL << bit;
924+ }
925+ } else {
926+ if (val & (1UL << bit)) {
927+ val &= ~(1UL << bit);
928+ }
929+ }
930+ rstate->bitmap[idx] = val;
931+}
932+
933+typedef struct VmaBlob {
934+ uint32_t start;
935+ uint32_t len;
936+ void *data;
937+} VmaBlob;
938+
939+static const VmaBlob *get_header_blob(VmaReader *vmar, uint32_t pos)
940+{
941+ assert(vmar);
942+ assert(vmar->blob_hash);
943+
944+ return g_hash_table_lookup(vmar->blob_hash, &pos);
945+}
946+
947+static const char *get_header_str(VmaReader *vmar, uint32_t pos)
948+{
949+ const VmaBlob *blob = get_header_blob(vmar, pos);
950+ if (!blob) {
951+ return NULL;
952+ }
953+ const char *res = (char *)blob->data;
954+ if (res[blob->len-1] != '\0') {
955+ return NULL;
956+ }
957+ return res;
958+}
959+
960+static ssize_t
961+safe_read(int fd, unsigned char *buf, size_t count)
962+{
963+ ssize_t n;
964+
965+ do {
966+ n = read(fd, buf, count);
967+ } while (n < 0 && errno == EINTR);
968+
969+ return n;
970+}
971+
972+static ssize_t
973+full_read(int fd, unsigned char *buf, size_t len)
974+{
975+ ssize_t n;
976+ size_t total;
977+
978+ total = 0;
979+
980+ while (len > 0) {
981+ n = safe_read(fd, buf, len);
982+
983+ if (n == 0) {
984+ return total;
985+ }
986+
987+ if (n <= 0) {
988+ break;
989+ }
990+
991+ buf += n;
992+ total += n;
993+ len -= n;
994+ }
995+
996+ if (len) {
997+ return -1;
998+ }
999+
1000+ return total;
1001+}
1002+
1003+void vma_reader_destroy(VmaReader *vmar)
1004+{
1005+ assert(vmar);
1006+
1007+ if (vmar->fd >= 0) {
1008+ close(vmar->fd);
1009+ }
1010+
1011+ if (vmar->cdata_list) {
1012+ g_list_free(vmar->cdata_list);
1013+ }
1014+
1015+ int i;
1016+ for (i = 1; i < 256; i++) {
1017+ if (vmar->rstate[i].bitmap) {
1018+ g_free(vmar->rstate[i].bitmap);
1019+ }
1020+ }
1021+
1022+ if (vmar->md5csum) {
1023+ g_checksum_free(vmar->md5csum);
1024+ }
1025+
1026+ if (vmar->blob_hash) {
1027+ g_hash_table_destroy(vmar->blob_hash);
1028+ }
1029+
1030+ if (vmar->head_data) {
1031+ g_free(vmar->head_data);
1032+ }
1033+
1034+ g_free(vmar);
1035+
1036+};
1037+
1038+static int vma_reader_read_head(VmaReader *vmar, Error **errp)
1039+{
1040+ assert(vmar);
1041+ assert(errp);
1042+ assert(*errp == NULL);
1043+
1044+ unsigned char md5sum[16];
1045+ int i;
1046+ int ret = 0;
1047+
1048+ vmar->head_data = g_malloc(sizeof(VmaHeader));
1049+
1050+ if (full_read(vmar->fd, vmar->head_data, sizeof(VmaHeader)) !=
1051+ sizeof(VmaHeader)) {
1052+ error_setg(errp, "can't read vma header - %s",
1053+ errno ? g_strerror(errno) : "got EOF");
1054+ return -1;
1055+ }
1056+
1057+ VmaHeader *h = (VmaHeader *)vmar->head_data;
1058+
1059+ if (h->magic != VMA_MAGIC) {
1060+ error_setg(errp, "not a vma file - wrong magic number");
1061+ return -1;
1062+ }
1063+
1064+ uint32_t header_size = GUINT32_FROM_BE(h->header_size);
1065+ int need = header_size - sizeof(VmaHeader);
1066+ if (need <= 0) {
1067+ error_setg(errp, "wrong vma header size %d", header_size);
1068+ return -1;
1069+ }
1070+
1071+ vmar->head_data = g_realloc(vmar->head_data, header_size);
1072+ h = (VmaHeader *)vmar->head_data;
1073+
1074+ if (full_read(vmar->fd, vmar->head_data + sizeof(VmaHeader), need) !=
1075+ need) {
1076+ error_setg(errp, "can't read vma header data - %s",
1077+ errno ? g_strerror(errno) : "got EOF");
1078+ return -1;
1079+ }
1080+
1081+ memcpy(md5sum, h->md5sum, 16);
1082+ memset(h->md5sum, 0, 16);
1083+
1084+ g_checksum_reset(vmar->md5csum);
1085+ g_checksum_update(vmar->md5csum, vmar->head_data, header_size);
1086+ gsize csize = 16;
1087+ g_checksum_get_digest(vmar->md5csum, (guint8 *)(h->md5sum), &csize);
1088+
1089+ if (memcmp(md5sum, h->md5sum, 16) != 0) {
1090+ error_setg(errp, "wrong vma header chechsum");
1091+ return -1;
1092+ }
1093+
1094+ /* we can modify header data after checksum verify */
1095+ h->header_size = header_size;
1096+
1097+ h->version = GUINT32_FROM_BE(h->version);
1098+ if (h->version != 1) {
1099+ error_setg(errp, "wrong vma version %d", h->version);
1100+ return -1;
1101+ }
1102+
1103+ h->ctime = GUINT64_FROM_BE(h->ctime);
1104+ h->blob_buffer_offset = GUINT32_FROM_BE(h->blob_buffer_offset);
1105+ h->blob_buffer_size = GUINT32_FROM_BE(h->blob_buffer_size);
1106+
1107+ uint32_t bstart = h->blob_buffer_offset + 1;
1108+ uint32_t bend = h->blob_buffer_offset + h->blob_buffer_size;
1109+
1110+ if (bstart <= sizeof(VmaHeader)) {
1111+ error_setg(errp, "wrong vma blob buffer offset %d",
1112+ h->blob_buffer_offset);
1113+ return -1;
1114+ }
1115+
1116+ if (bend > header_size) {
1117+ error_setg(errp, "wrong vma blob buffer size %d/%d",
1118+ h->blob_buffer_offset, h->blob_buffer_size);
1119+ return -1;
1120+ }
1121+
1122+ while ((bstart + 2) <= bend) {
1123+ uint32_t size = vmar->head_data[bstart] +
1124+ (vmar->head_data[bstart+1] << 8);
1125+ if ((bstart + size + 2) <= bend) {
1126+ VmaBlob *blob = g_new0(VmaBlob, 1);
1127+ blob->start = bstart - h->blob_buffer_offset;
1128+ blob->len = size;
1129+ blob->data = vmar->head_data + bstart + 2;
1130+ g_hash_table_insert(vmar->blob_hash, &blob->start, blob);
1131+ }
1132+ bstart += size + 2;
1133+ }
1134+
1135+
1136+ int count = 0;
1137+ for (i = 1; i < 256; i++) {
1138+ VmaDeviceInfoHeader *dih = &h->dev_info[i];
1139+ uint32_t devname_ptr = GUINT32_FROM_BE(dih->devname_ptr);
1140+ uint64_t size = GUINT64_FROM_BE(dih->size);
1141+ const char *devname = get_header_str(vmar, devname_ptr);
1142+
1143+ if (size && devname) {
1144+ count++;
1145+ vmar->devinfo[i].size = size;
1146+ vmar->devinfo[i].devname = devname;
1147+
1148+ if (strcmp(devname, "vmstate") == 0) {
1149+ vmar->vmstate_stream = i;
1150+ }
1151+ }
1152+ }
1153+
95259824
WB		 1154+ for (i = 0; i < VMA_MAX_CONFIGS; i++) {
1155+ uint32_t name_ptr = GUINT32_FROM_BE(h->config_names[i]);
1156+ uint32_t data_ptr = GUINT32_FROM_BE(h->config_data[i]);
1157+
1158+ if (!(name_ptr && data_ptr)) {
1159+ continue;
1160+ }
1161+ const char *name = get_header_str(vmar, name_ptr);
1162+ const VmaBlob *blob = get_header_blob(vmar, data_ptr);
1163+
1164+ if (!(name && blob)) {
1165+ error_setg(errp, "vma contains invalid data pointers");
1166+ return -1;
1167+ }
1168+
1169+ VmaConfigData *cdata = g_new0(VmaConfigData, 1);
1170+ cdata->name = name;
1171+ cdata->data = blob->data;
1172+ cdata->len = blob->len;
1173+
1174+ vmar->cdata_list = g_list_append(vmar->cdata_list, cdata);
1175+ }
1176+
1177+ return ret;
1178+};
1179+
1180+VmaReader *vma_reader_create(const char *filename, Error **errp)
1181+{
1182+ assert(filename);
1183+ assert(errp);
1184+
1185+ VmaReader *vmar = g_new0(VmaReader, 1);
1186+
1187+ if (strcmp(filename, "-") == 0) {
1188+ vmar->fd = dup(0);
1189+ } else {
1190+ vmar->fd = open(filename, O_RDONLY);
1191+ }
1192+
1193+ if (vmar->fd < 0) {
1194+ error_setg(errp, "can't open file %s - %s\n", filename,
1195+ g_strerror(errno));
1196+ goto err;
1197+ }
1198+
1199+ vmar->md5csum = g_checksum_new(G_CHECKSUM_MD5);
1200+ if (!vmar->md5csum) {
1201+ error_setg(errp, "can't allocate cmsum\n");
1202+ goto err;
1203+ }
1204+
1205+ vmar->blob_hash = g_hash_table_new_full(g_int32_hash, g_int32_equal,
1206+ NULL, g_free);
1207+
1208+ if (vma_reader_read_head(vmar, errp) < 0) {
1209+ goto err;
1210+ }
1211+
1212+ return vmar;
1213+
1214+err:
1215+ if (vmar) {
1216+ vma_reader_destroy(vmar);
1217+ }
1218+
1219+ return NULL;
1220+}
1221+
1222+VmaHeader *vma_reader_get_header(VmaReader *vmar)
1223+{
1224+ assert(vmar);
1225+ assert(vmar->head_data);
1226+
1227+ return (VmaHeader *)(vmar->head_data);
1228+}
1229+
1230+GList *vma_reader_get_config_data(VmaReader *vmar)
1231+{
1232+ assert(vmar);
1233+ assert(vmar->head_data);
1234+
1235+ return vmar->cdata_list;
1236+}
1237+
1238+VmaDeviceInfo *vma_reader_get_device_info(VmaReader *vmar, guint8 dev_id)
1239+{
1240+ assert(vmar);
1241+ assert(dev_id);
1242+
1243+ if (vmar->devinfo[dev_id].size && vmar->devinfo[dev_id].devname) {
1244+ return &vmar->devinfo[dev_id];
1245+ }
1246+
1247+ return NULL;
1248+}
1249+
67af0fa4
WB		 1250+static void allocate_rstate(VmaReader *vmar, guint8 dev_id,
1251+ BlockBackend *target, bool write_zeroes)
1252+{
1253+ assert(vmar);
1254+ assert(dev_id);
1255+
1256+ vmar->rstate[dev_id].target = target;
1257+ vmar->rstate[dev_id].write_zeroes = write_zeroes;
1258+
1259+ int64_t size = vmar->devinfo[dev_id].size;
1260+
1261+ int64_t bitmap_size = (size/BDRV_SECTOR_SIZE) +
1262+ (VMA_CLUSTER_SIZE/BDRV_SECTOR_SIZE) * BITS_PER_LONG - 1;
1263+ bitmap_size /= (VMA_CLUSTER_SIZE/BDRV_SECTOR_SIZE) * BITS_PER_LONG;
1264+
1265+ vmar->rstate[dev_id].bitmap_size = bitmap_size;
1266+ vmar->rstate[dev_id].bitmap = g_new0(unsigned long, bitmap_size);
1267+
1268+ vmar->cluster_count += size/VMA_CLUSTER_SIZE;
1269+}
1270+
1271+int vma_reader_register_bs(VmaReader *vmar, guint8 dev_id, BlockBackend *target,
95259824
WB		 1272+ bool write_zeroes, Error **errp)
1273+{
1274+ assert(vmar);
67af0fa4 1275+ assert(target != NULL);
95259824 1276+ assert(dev_id);
67af0fa4 1277+ assert(vmar->rstate[dev_id].target == NULL);
95259824 1278+
67af0fa4 1279+ int64_t size = blk_getlength(target);
95259824
WB		 1280+ int64_t size_diff = size - vmar->devinfo[dev_id].size;
1281+
1282+ /* storage types can have different size restrictions, so it
1283+ * is not always possible to create an image with exact size.
1284+ * So we tolerate a size difference up to 4MB.
1285+ */
1286+ if ((size_diff < 0) || (size_diff > 4*1024*1024)) {
1287+ error_setg(errp, "vma_reader_register_bs for stream %s failed - "
1288+ "unexpected size %zd != %zd", vmar->devinfo[dev_id].devname,
1289+ size, vmar->devinfo[dev_id].size);
1290+ return -1;
1291+ }
1292+
67af0fa4 1293+ allocate_rstate(vmar, dev_id, target, write_zeroes);
95259824
WB		 1294+
1295+ return 0;
1296+}
1297+
1298+static ssize_t safe_write(int fd, void *buf, size_t count)
1299+{
1300+ ssize_t n;
1301+
1302+ do {
1303+ n = write(fd, buf, count);
1304+ } while (n < 0 && errno == EINTR);
1305+
1306+ return n;
1307+}
1308+
1309+static size_t full_write(int fd, void *buf, size_t len)
1310+{
1311+ ssize_t n;
1312+ size_t total;
1313+
1314+ total = 0;
1315+
1316+ while (len > 0) {
1317+ n = safe_write(fd, buf, len);
1318+ if (n < 0) {
1319+ return n;
1320+ }
1321+ buf += n;
1322+ total += n;
1323+ len -= n;
1324+ }
1325+
1326+ if (len) {
1327+ /* incomplete write ? */
1328+ return -1;
1329+ }
1330+
1331+ return total;
1332+}
1333+
1334+static int restore_write_data(VmaReader *vmar, guint8 dev_id,
67af0fa4 1335+ BlockBackend *target, int vmstate_fd,
95259824
WB		 1336+ unsigned char *buf, int64_t sector_num,
1337+ int nb_sectors, Error **errp)
1338+{
1339+ assert(vmar);
1340+
1341+ if (dev_id == vmar->vmstate_stream) {
1342+ if (vmstate_fd >= 0) {
1343+ int len = nb_sectors * BDRV_SECTOR_SIZE;
1344+ int res = full_write(vmstate_fd, buf, len);
1345+ if (res < 0) {
1346+ error_setg(errp, "write vmstate failed %d", res);
1347+ return -1;
1348+ }
1349+ }
1350+ } else {
67af0fa4 1351+ int res = blk_pwrite(target, sector_num * BDRV_SECTOR_SIZE, buf, nb_sectors * BDRV_SECTOR_SIZE, 0);
95259824 1352+ if (res < 0) {
67af0fa4
WB		 1353+ error_setg(errp, "blk_pwrite to %s failed (%d)",
1354+ bdrv_get_device_name(blk_bs(target)), res);
95259824
WB		 1355+ return -1;
1356+ }
1357+ }
1358+ return 0;
1359+}
67af0fa4 1360+
95259824
WB		 1361+static int restore_extent(VmaReader *vmar, unsigned char *buf,
1362+ int extent_size, int vmstate_fd,
67af0fa4 1363+ bool verbose, bool verify, Error **errp)
95259824
WB		 1364+{
1365+ assert(vmar);
1366+ assert(buf);
1367+
1368+ VmaExtentHeader *ehead = (VmaExtentHeader *)buf;
1369+ int start = VMA_EXTENT_HEADER_SIZE;
1370+ int i;
1371+
1372+ for (i = 0; i < VMA_BLOCKS_PER_EXTENT; i++) {
1373+ uint64_t block_info = GUINT64_FROM_BE(ehead->blockinfo[i]);
1374+ uint64_t cluster_num = block_info & 0xffffffff;
1375+ uint8_t dev_id = (block_info >> 32) & 0xff;
1376+ uint16_t mask = block_info >> (32+16);
1377+ int64_t max_sector;
1378+
1379+ if (!dev_id) {
1380+ continue;
1381+ }
1382+
1383+ VmaRestoreState *rstate = &vmar->rstate[dev_id];
67af0fa4 1384+ BlockBackend *target = NULL;
95259824
WB		 1385+
1386+ if (dev_id != vmar->vmstate_stream) {
67af0fa4
WB		 1387+ target = rstate->target;
1388+ if (!verify && !target) {
95259824
WB		 1389+ error_setg(errp, "got wrong dev id %d", dev_id);
1390+ return -1;
1391+ }
1392+
1393+ if (vma_reader_get_bitmap(rstate, cluster_num)) {
1394+ error_setg(errp, "found duplicated cluster %zd for stream %s",
1395+ cluster_num, vmar->devinfo[dev_id].devname);
1396+ return -1;
1397+ }
1398+ vma_reader_set_bitmap(rstate, cluster_num, 1);
1399+
1400+ max_sector = vmar->devinfo[dev_id].size/BDRV_SECTOR_SIZE;
1401+ } else {
1402+ max_sector = G_MAXINT64;
1403+ if (cluster_num != vmar->vmstate_clusters) {
1404+ error_setg(errp, "found out of order vmstate data");
1405+ return -1;
1406+ }
1407+ vmar->vmstate_clusters++;
1408+ }
1409+
1410+ vmar->clusters_read++;
1411+
1412+ if (verbose) {
1413+ time_t duration = time(NULL) - vmar->start_time;
1414+ int percent = (vmar->clusters_read*100)/vmar->cluster_count;
1415+ if (percent != vmar->clusters_read_per) {
1416+ printf("progress %d%% (read %zd bytes, duration %zd sec)\n",
1417+ percent, vmar->clusters_read*VMA_CLUSTER_SIZE,
1418+ duration);
1419+ fflush(stdout);
1420+ vmar->clusters_read_per = percent;
1421+ }
1422+ }
1423+
1424+ /* try to write whole clusters to speedup restore */
1425+ if (mask == 0xffff) {
1426+ if ((start + VMA_CLUSTER_SIZE) > extent_size) {
1427+ error_setg(errp, "short vma extent - too many blocks");
1428+ return -1;
1429+ }
1430+ int64_t sector_num = (cluster_num * VMA_CLUSTER_SIZE) /
1431+ BDRV_SECTOR_SIZE;
1432+ int64_t end_sector = sector_num +
1433+ VMA_CLUSTER_SIZE/BDRV_SECTOR_SIZE;
1434+
1435+ if (end_sector > max_sector) {
1436+ end_sector = max_sector;
1437+ }
1438+
1439+ if (end_sector <= sector_num) {
1440+ error_setg(errp, "got wrong block address - write bejond end");
1441+ return -1;
1442+ }
1443+
67af0fa4
WB		 1444+ if (!verify) {
1445+ int nb_sectors = end_sector - sector_num;
1446+ if (restore_write_data(vmar, dev_id, target, vmstate_fd,
1447+ buf + start, sector_num, nb_sectors,
1448+ errp) < 0) {
1449+ return -1;
1450+ }
95259824
WB		 1451+ }
1452+
1453+ start += VMA_CLUSTER_SIZE;
1454+ } else {
1455+ int j;
1456+ int bit = 1;
1457+
1458+ for (j = 0; j < 16; j++) {
1459+ int64_t sector_num = (cluster_num*VMA_CLUSTER_SIZE +
1460+ j*VMA_BLOCK_SIZE)/BDRV_SECTOR_SIZE;
1461+
1462+ int64_t end_sector = sector_num +
1463+ VMA_BLOCK_SIZE/BDRV_SECTOR_SIZE;
1464+ if (end_sector > max_sector) {
1465+ end_sector = max_sector;
1466+ }
1467+
1468+ if (mask & bit) {
1469+ if ((start + VMA_BLOCK_SIZE) > extent_size) {
1470+ error_setg(errp, "short vma extent - too many blocks");
1471+ return -1;
1472+ }
1473+
1474+ if (end_sector <= sector_num) {
1475+ error_setg(errp, "got wrong block address - "
1476+ "write bejond end");
1477+ return -1;
1478+ }
1479+
67af0fa4
WB		 1480+ if (!verify) {
1481+ int nb_sectors = end_sector - sector_num;
1482+ if (restore_write_data(vmar, dev_id, target, vmstate_fd,
1483+ buf + start, sector_num,
1484+ nb_sectors, errp) < 0) {
1485+ return -1;
1486+ }
95259824
WB		 1487+ }
1488+
1489+ start += VMA_BLOCK_SIZE;
1490+
1491+ } else {
1492+
67af0fa4
WB		 1493+
1494+ if (end_sector > sector_num) {
95259824
WB		 1495+ /* Todo: use bdrv_co_write_zeroes (but that need to
1496+ * be run inside coroutine?)
1497+ */
1498+ int nb_sectors = end_sector - sector_num;
67af0fa4
WB		 1499+ int zero_size = BDRV_SECTOR_SIZE*nb_sectors;
1500+ vmar->zero_cluster_data += zero_size;
1501+ if (mask != 0) {
1502+ vmar->partial_zero_cluster_data += zero_size;
1503+ }
1504+
1505+ if (rstate->write_zeroes && !verify) {
1506+ if (restore_write_data(vmar, dev_id, target, vmstate_fd,
1507+ zero_vma_block, sector_num,
1508+ nb_sectors, errp) < 0) {
1509+ return -1;
1510+ }
95259824
WB		 1511+ }
1512+ }
1513+ }
1514+
1515+ bit = bit << 1;
1516+ }
1517+ }
1518+ }
1519+
1520+ if (start != extent_size) {
1521+ error_setg(errp, "vma extent error - missing blocks");
1522+ return -1;
1523+ }
1524+
1525+ return 0;
1526+}
1527+
67af0fa4
WB		 1528+static int vma_reader_restore_full(VmaReader *vmar, int vmstate_fd,
1529+ bool verbose, bool verify,
1530+ Error **errp)
95259824
WB		 1531+{
1532+ assert(vmar);
1533+ assert(vmar->head_data);
1534+
1535+ int ret = 0;
1536+ unsigned char buf[VMA_MAX_EXTENT_SIZE];
1537+ int buf_pos = 0;
1538+ unsigned char md5sum[16];
1539+ VmaHeader *h = (VmaHeader *)vmar->head_data;
1540+
1541+ vmar->start_time = time(NULL);
1542+
1543+ while (1) {
1544+ int bytes = full_read(vmar->fd, buf + buf_pos, sizeof(buf) - buf_pos);
1545+ if (bytes < 0) {
1546+ error_setg(errp, "read failed - %s", g_strerror(errno));
1547+ return -1;
1548+ }
1549+
1550+ buf_pos += bytes;
1551+
1552+ if (!buf_pos) {
1553+ break; /* EOF */
1554+ }
1555+
1556+ if (buf_pos < VMA_EXTENT_HEADER_SIZE) {
1557+ error_setg(errp, "read short extent (%d bytes)", buf_pos);
1558+ return -1;
1559+ }
1560+
1561+ VmaExtentHeader *ehead = (VmaExtentHeader *)buf;
1562+
1563+ /* extract md5sum */
1564+ memcpy(md5sum, ehead->md5sum, sizeof(ehead->md5sum));
1565+ memset(ehead->md5sum, 0, sizeof(ehead->md5sum));
1566+
1567+ g_checksum_reset(vmar->md5csum);
1568+ g_checksum_update(vmar->md5csum, buf, VMA_EXTENT_HEADER_SIZE);
1569+ gsize csize = 16;
1570+ g_checksum_get_digest(vmar->md5csum, ehead->md5sum, &csize);
1571+
1572+ if (memcmp(md5sum, ehead->md5sum, 16) != 0) {
1573+ error_setg(errp, "wrong vma extent header chechsum");
1574+ return -1;
1575+ }
1576+
1577+ if (memcmp(h->uuid, ehead->uuid, sizeof(ehead->uuid)) != 0) {
1578+ error_setg(errp, "wrong vma extent uuid");
1579+ return -1;
1580+ }
1581+
1582+ if (ehead->magic != VMA_EXTENT_MAGIC || ehead->reserved1 != 0) {
1583+ error_setg(errp, "wrong vma extent header magic");
1584+ return -1;
1585+ }
1586+
1587+ int block_count = GUINT16_FROM_BE(ehead->block_count);
1588+ int extent_size = VMA_EXTENT_HEADER_SIZE + block_count*VMA_BLOCK_SIZE;
1589+
1590+ if (buf_pos < extent_size) {
1591+ error_setg(errp, "short vma extent (%d < %d)", buf_pos,
1592+ extent_size);
1593+ return -1;
1594+ }
1595+
1596+ if (restore_extent(vmar, buf, extent_size, vmstate_fd, verbose,
67af0fa4 1597+ verify, errp) < 0) {
95259824
WB		 1598+ return -1;
1599+ }
1600+
1601+ if (buf_pos > extent_size) {
1602+ memmove(buf, buf + extent_size, buf_pos - extent_size);
1603+ buf_pos = buf_pos - extent_size;
1604+ } else {
1605+ buf_pos = 0;
1606+ }
1607+ }
1608+
1609+ bdrv_drain_all();
1610+
1611+ int i;
1612+ for (i = 1; i < 256; i++) {
1613+ VmaRestoreState *rstate = &vmar->rstate[i];
67af0fa4 1614+ if (!rstate->target) {
95259824
WB		 1615+ continue;
1616+ }
1617+
67af0fa4
WB		 1618+ if (blk_flush(rstate->target) < 0) {
1619+ error_setg(errp, "vma blk_flush %s failed",
95259824
WB		 1620+ vmar->devinfo[i].devname);
1621+ return -1;
1622+ }
1623+
1624+ if (vmar->devinfo[i].size &&
1625+ (strcmp(vmar->devinfo[i].devname, "vmstate") != 0)) {
1626+ assert(rstate->bitmap);
1627+
1628+ int64_t cluster_num, end;
1629+
1630+ end = (vmar->devinfo[i].size + VMA_CLUSTER_SIZE - 1) /
1631+ VMA_CLUSTER_SIZE;
1632+
1633+ for (cluster_num = 0; cluster_num < end; cluster_num++) {
1634+ if (!vma_reader_get_bitmap(rstate, cluster_num)) {
1635+ error_setg(errp, "detected missing cluster %zd "
1636+ "for stream %s", cluster_num,
1637+ vmar->devinfo[i].devname);
1638+ return -1;
1639+ }
1640+ }
1641+ }
1642+ }
1643+
67af0fa4
WB		 1644+ if (verbose) {
1645+ if (vmar->clusters_read) {
1646+ printf("total bytes read %zd, sparse bytes %zd (%.3g%%)\n",
1647+ vmar->clusters_read*VMA_CLUSTER_SIZE,
1648+ vmar->zero_cluster_data,
1649+ (double)(100.0*vmar->zero_cluster_data)/
1650+ (vmar->clusters_read*VMA_CLUSTER_SIZE));
1651+
1652+ int64_t datasize = vmar->clusters_read*VMA_CLUSTER_SIZE-vmar->zero_cluster_data;
1653+ if (datasize) { // this does not make sense for empty files
1654+ printf("space reduction due to 4K zero blocks %.3g%%\n",
1655+ (double)(100.0*vmar->partial_zero_cluster_data) / datasize);
1656+ }
1657+ } else {
1658+ printf("vma archive contains no image data\n");
1659+ }
1660+ }
95259824
WB		 1661+ return ret;
1662+}
1663+
67af0fa4
WB		 1664+int vma_reader_restore(VmaReader *vmar, int vmstate_fd, bool verbose,
1665+ Error **errp)
1666+{
1667+ return vma_reader_restore_full(vmar, vmstate_fd, verbose, false, errp);
1668+}
1669+
1670+int vma_reader_verify(VmaReader *vmar, bool verbose, Error **errp)
1671+{
1672+ guint8 dev_id;
1673+
1674+ for (dev_id = 1; dev_id < 255; dev_id++) {
1675+ if (vma_reader_get_device_info(vmar, dev_id)) {
1676+ allocate_rstate(vmar, dev_id, NULL, false);
1677+ }
1678+ }
1679+
1680+ return vma_reader_restore_full(vmar, -1, verbose, true, errp);
1681+}
1682+
95259824
WB		 1683diff --git a/vma-writer.c b/vma-writer.c
1684new file mode 100644
67af0fa4 1685index 0000000000..9001cbdd2b
95259824
WB		 1686--- /dev/null
1687+++ b/vma-writer.c
67af0fa4 1688@@ -0,0 +1,771 @@
95259824
WB		 1689+/*
1690+ * VMA: Virtual Machine Archive
1691+ *
1692+ * Copyright (C) 2012 Proxmox Server Solutions
1693+ *
1694+ * Authors:
1695+ * Dietmar Maurer (dietmar@proxmox.com)
1696+ *
1697+ * This work is licensed under the terms of the GNU GPL, version 2 or later.
1698+ * See the COPYING file in the top-level directory.
1699+ *
1700+ */
1701+
1702+#include "qemu/osdep.h"
1703+#include <glib.h>
1704+#include <uuid/uuid.h>
1705+
1706+#include "vma.h"
1707+#include "block/block.h"
1708+#include "monitor/monitor.h"
1709+#include "qemu/main-loop.h"
1710+#include "qemu/coroutine.h"
1711+#include "qemu/cutils.h"
1712+
1713+#define DEBUG_VMA 0
1714+
1715+#define DPRINTF(fmt, ...)\
1716+ do { if (DEBUG_VMA) { printf("vma: " fmt, ## __VA_ARGS__); } } while (0)
1717+
1718+#define WRITE_BUFFERS 5
67af0fa4
WB		 1719+#define HEADER_CLUSTERS 8
1720+#define HEADERBUF_SIZE (VMA_CLUSTER_SIZE*HEADER_CLUSTERS)
95259824
WB		 1721+
1722+struct VmaWriter {
1723+ int fd;
1724+ FILE *cmd;
1725+ int status;
1726+ char errmsg[8192];
1727+ uuid_t uuid;
1728+ bool header_written;
1729+ bool closed;
1730+
1731+ /* we always write extents */
67af0fa4 1732+ unsigned char *outbuf;
95259824
WB		 1733+ int outbuf_pos; /* in bytes */
1734+ int outbuf_count; /* in VMA_BLOCKS */
1735+ uint64_t outbuf_block_info[VMA_BLOCKS_PER_EXTENT];
1736+
67af0fa4 1737+ unsigned char *headerbuf;
95259824
WB		 1738+
1739+ GChecksum *md5csum;
95259824
WB		 1740+ CoMutex flush_lock;
1741+ Coroutine *co_writer;
1742+
1743+ /* drive informations */
1744+ VmaStreamInfo stream_info[256];
1745+ guint stream_count;
1746+
1747+ guint8 vmstate_stream;
1748+ uint32_t vmstate_clusters;
1749+
1750+ /* header blob table */
1751+ char *header_blob_table;
1752+ uint32_t header_blob_table_size;
1753+ uint32_t header_blob_table_pos;
1754+
1755+ /* store for config blobs */
1756+ uint32_t config_names[VMA_MAX_CONFIGS]; /* offset into blob_buffer table */
1757+ uint32_t config_data[VMA_MAX_CONFIGS]; /* offset into blob_buffer table */
1758+ uint32_t config_count;
1759+};
1760+
1761+void vma_writer_set_error(VmaWriter *vmaw, const char *fmt, ...)
1762+{
1763+ va_list ap;
1764+
1765+ if (vmaw->status < 0) {
1766+ return;
1767+ }
1768+
1769+ vmaw->status = -1;
1770+
1771+ va_start(ap, fmt);
1772+ g_vsnprintf(vmaw->errmsg, sizeof(vmaw->errmsg), fmt, ap);
1773+ va_end(ap);
1774+
1775+ DPRINTF("vma_writer_set_error: %s\n", vmaw->errmsg);
1776+}
1777+
1778+static uint32_t allocate_header_blob(VmaWriter *vmaw, const char *data,
1779+ size_t len)
1780+{
1781+ if (len > 65535) {
1782+ return 0;
1783+ }
1784+
1785+ if (!vmaw->header_blob_table ||
1786+ (vmaw->header_blob_table_size <
1787+ (vmaw->header_blob_table_pos + len + 2))) {
1788+ int newsize = vmaw->header_blob_table_size + ((len + 2 + 511)/512)*512;
1789+
1790+ vmaw->header_blob_table = g_realloc(vmaw->header_blob_table, newsize);
1791+ memset(vmaw->header_blob_table + vmaw->header_blob_table_size,
1792+ 0, newsize - vmaw->header_blob_table_size);
1793+ vmaw->header_blob_table_size = newsize;
1794+ }
1795+
1796+ uint32_t cpos = vmaw->header_blob_table_pos;
1797+ vmaw->header_blob_table[cpos] = len & 255;
1798+ vmaw->header_blob_table[cpos+1] = (len >> 8) & 255;
1799+ memcpy(vmaw->header_blob_table + cpos + 2, data, len);
1800+ vmaw->header_blob_table_pos += len + 2;
1801+ return cpos;
1802+}
1803+
1804+static uint32_t allocate_header_string(VmaWriter *vmaw, const char *str)
1805+{
1806+ assert(vmaw);
1807+
1808+ size_t len = strlen(str) + 1;
1809+
1810+ return allocate_header_blob(vmaw, str, len);
1811+}
1812+
1813+int vma_writer_add_config(VmaWriter *vmaw, const char *name, gpointer data,
1814+ gsize len)
1815+{
1816+ assert(vmaw);
1817+ assert(!vmaw->header_written);
1818+ assert(vmaw->config_count < VMA_MAX_CONFIGS);
1819+ assert(name);
1820+ assert(data);
95259824
WB		 1821+
1822+ gchar *basename = g_path_get_basename(name);
1823+ uint32_t name_ptr = allocate_header_string(vmaw, basename);
1824+ g_free(basename);
1825+
1826+ if (!name_ptr) {
1827+ return -1;
1828+ }
1829+
1830+ uint32_t data_ptr = allocate_header_blob(vmaw, data, len);
1831+ if (!data_ptr) {
1832+ return -1;
1833+ }
1834+
1835+ vmaw->config_names[vmaw->config_count] = name_ptr;
1836+ vmaw->config_data[vmaw->config_count] = data_ptr;
1837+
1838+ vmaw->config_count++;
1839+
1840+ return 0;
1841+}
1842+
1843+int vma_writer_register_stream(VmaWriter *vmaw, const char *devname,
1844+ size_t size)
1845+{
1846+ assert(vmaw);
1847+ assert(devname);
1848+ assert(!vmaw->status);
1849+
1850+ if (vmaw->header_written) {
1851+ vma_writer_set_error(vmaw, "vma_writer_register_stream: header "
1852+ "already written");
1853+ return -1;
1854+ }
1855+
1856+ guint n = vmaw->stream_count + 1;
1857+
1858+ /* we can have dev_ids form 1 to 255 (0 reserved)
1859+ * 255(-1) reseverd for safety
1860+ */
1861+ if (n > 254) {
1862+ vma_writer_set_error(vmaw, "vma_writer_register_stream: "
1863+ "too many drives");
1864+ return -1;
1865+ }
1866+
1867+ if (size <= 0) {
1868+ vma_writer_set_error(vmaw, "vma_writer_register_stream: "
1869+ "got strange size %zd", size);
1870+ return -1;
1871+ }
1872+
1873+ DPRINTF("vma_writer_register_stream %s %zu %d\n", devname, size, n);
1874+
1875+ vmaw->stream_info[n].devname = g_strdup(devname);
1876+ vmaw->stream_info[n].size = size;
1877+
1878+ vmaw->stream_info[n].cluster_count = (size + VMA_CLUSTER_SIZE - 1) /
1879+ VMA_CLUSTER_SIZE;
1880+
1881+ vmaw->stream_count = n;
1882+
1883+ if (strcmp(devname, "vmstate") == 0) {
1884+ vmaw->vmstate_stream = n;
1885+ }
1886+
1887+ return n;
1888+}
1889+
1890+static void vma_co_continue_write(void *opaque)
1891+{
1892+ VmaWriter *vmaw = opaque;
1893+
1894+ DPRINTF("vma_co_continue_write\n");
1895+ qemu_coroutine_enter(vmaw->co_writer);
1896+}
1897+
1898+static ssize_t coroutine_fn
67af0fa4 1899+vma_queue_write(VmaWriter *vmaw, const void *buf, size_t bytes)
95259824 1900+{
67af0fa4 1901+ DPRINTF("vma_queue_write enter %zd\n", bytes);
95259824 1902+
67af0fa4
WB		 1903+ assert(vmaw);
1904+ assert(buf);
1905+ assert(bytes <= VMA_MAX_EXTENT_SIZE);
95259824 1906+
67af0fa4
WB		 1907+ size_t done = 0;
1908+ ssize_t ret;
95259824
WB		 1909+
1910+ assert(vmaw->co_writer == NULL);
1911+
1912+ vmaw->co_writer = qemu_coroutine_self();
1913+
95259824 1914+ while (done < bytes) {
67af0fa4
WB		 1915+ aio_set_fd_handler(qemu_get_aio_context(), vmaw->fd, false, NULL, vma_co_continue_write, NULL, vmaw);
1916+ qemu_coroutine_yield();
1917+ aio_set_fd_handler(qemu_get_aio_context(), vmaw->fd, false, NULL, NULL, NULL, NULL);
1918+ if (vmaw->status < 0) {
1919+ DPRINTF("vma_queue_write detected canceled backup\n");
1920+ done = -1;
1921+ break;
1922+ }
95259824
WB		 1923+ ret = write(vmaw->fd, buf + done, bytes - done);
1924+ if (ret > 0) {
1925+ done += ret;
67af0fa4 1926+ DPRINTF("vma_queue_write written %zd %zd\n", done, ret);
95259824
WB		 1927+ } else if (ret < 0) {
1928+ if (errno == EAGAIN || errno == EWOULDBLOCK) {
67af0fa4
WB		 1929+ /* try again */
1930+ } else {
1931+ vma_writer_set_error(vmaw, "vma_queue_write: write error - %s",
95259824
WB		 1932+ g_strerror(errno));
1933+ done = -1; /* always return failure for partial writes */
1934+ break;
1935+ }
1936+ } else if (ret == 0) {
1937+ /* should not happen - simply try again */
1938+ }
1939+ }
1940+
95259824
WB		 1941+ vmaw->co_writer = NULL;
1942+
67af0fa4 1943+ return (done == bytes) ? bytes : -1;
95259824
WB		 1944+}
1945+
1946+VmaWriter *vma_writer_create(const char *filename, uuid_t uuid, Error **errp)
1947+{
1948+ const char *p;
1949+
1950+ assert(sizeof(VmaHeader) == (4096 + 8192));
1951+ assert(G_STRUCT_OFFSET(VmaHeader, config_names) == 2044);
1952+ assert(G_STRUCT_OFFSET(VmaHeader, config_data) == 3068);
1953+ assert(G_STRUCT_OFFSET(VmaHeader, dev_info) == 4096);
1954+ assert(sizeof(VmaExtentHeader) == 512);
1955+
1956+ VmaWriter *vmaw = g_new0(VmaWriter, 1);
1957+ vmaw->fd = -1;
1958+
1959+ vmaw->md5csum = g_checksum_new(G_CHECKSUM_MD5);
1960+ if (!vmaw->md5csum) {
1961+ error_setg(errp, "can't allocate cmsum\n");
1962+ goto err;
1963+ }
1964+
1965+ if (strstart(filename, "exec:", &p)) {
1966+ vmaw->cmd = popen(p, "w");
1967+ if (vmaw->cmd == NULL) {
1968+ error_setg(errp, "can't popen command '%s' - %s\n", p,
1969+ g_strerror(errno));
1970+ goto err;
1971+ }
1972+ vmaw->fd = fileno(vmaw->cmd);
1973+
67af0fa4 1974+ /* try to use O_NONBLOCK */
95259824 1975+ fcntl(vmaw->fd, F_SETFL, fcntl(vmaw->fd, F_GETFL)|O_NONBLOCK);
95259824
WB		 1976+
1977+ } else {
1978+ struct stat st;
1979+ int oflags;
1980+ const char *tmp_id_str;
1981+
1982+ if ((stat(filename, &st) == 0) && S_ISFIFO(st.st_mode)) {
67af0fa4 1983+ oflags = O_NONBLOCK|O_WRONLY;
95259824
WB		 1984+ vmaw->fd = qemu_open(filename, oflags, 0644);
1985+ } else if (strstart(filename, "/dev/fdset/", &tmp_id_str)) {
67af0fa4 1986+ oflags = O_NONBLOCK|O_WRONLY;
95259824
WB		 1987+ vmaw->fd = qemu_open(filename, oflags, 0644);
1988+ } else if (strstart(filename, "/dev/fdname/", &tmp_id_str)) {
1989+ vmaw->fd = monitor_get_fd(cur_mon, tmp_id_str, errp);
1990+ if (vmaw->fd < 0) {
1991+ goto err;
1992+ }
67af0fa4 1993+ /* try to use O_NONBLOCK */
95259824 1994+ fcntl(vmaw->fd, F_SETFL, fcntl(vmaw->fd, F_GETFL)|O_NONBLOCK);
95259824
WB		 1995+ } else {
1996+ oflags = O_NONBLOCK|O_DIRECT|O_WRONLY|O_CREAT|O_EXCL;
1997+ vmaw->fd = qemu_open(filename, oflags, 0644);
1998+ }
1999+
2000+ if (vmaw->fd < 0) {
2001+ error_setg(errp, "can't open file %s - %s\n", filename,
2002+ g_strerror(errno));
2003+ goto err;
2004+ }
2005+ }
2006+
2007+ /* we use O_DIRECT, so we need to align IO buffers */
67af0fa4
WB		 2008+
2009+ vmaw->outbuf = qemu_memalign(512, VMA_MAX_EXTENT_SIZE);
2010+ vmaw->headerbuf = qemu_memalign(512, HEADERBUF_SIZE);
95259824
WB		 2011+
2012+ vmaw->outbuf_count = 0;
2013+ vmaw->outbuf_pos = VMA_EXTENT_HEADER_SIZE;
2014+
2015+ vmaw->header_blob_table_pos = 1; /* start at pos 1 */
2016+
95259824 2017+ qemu_co_mutex_init(&vmaw->flush_lock);
95259824
WB		 2018+
2019+ uuid_copy(vmaw->uuid, uuid);
2020+
2021+ return vmaw;
2022+
2023+err:
2024+ if (vmaw) {
2025+ if (vmaw->cmd) {
2026+ pclose(vmaw->cmd);
2027+ } else if (vmaw->fd >= 0) {
2028+ close(vmaw->fd);
2029+ }
2030+
2031+ if (vmaw->md5csum) {
2032+ g_checksum_free(vmaw->md5csum);
2033+ }
2034+
2035+ g_free(vmaw);
2036+ }
2037+
2038+ return NULL;
2039+}
2040+
2041+static int coroutine_fn vma_write_header(VmaWriter *vmaw)
2042+{
2043+ assert(vmaw);
67af0fa4 2044+ unsigned char *buf = vmaw->headerbuf;
95259824
WB		 2045+ VmaHeader *head = (VmaHeader *)buf;
2046+
2047+ int i;
2048+
2049+ DPRINTF("VMA WRITE HEADER\n");
2050+
2051+ if (vmaw->status < 0) {
2052+ return vmaw->status;
2053+ }
2054+
67af0fa4 2055+ memset(buf, 0, HEADERBUF_SIZE);
95259824
WB		 2056+
2057+ head->magic = VMA_MAGIC;
2058+ head->version = GUINT32_TO_BE(1); /* v1 */
2059+ memcpy(head->uuid, vmaw->uuid, 16);
2060+
2061+ time_t ctime = time(NULL);
2062+ head->ctime = GUINT64_TO_BE(ctime);
2063+
95259824
WB		 2064+ for (i = 0; i < VMA_MAX_CONFIGS; i++) {
2065+ head->config_names[i] = GUINT32_TO_BE(vmaw->config_names[i]);
2066+ head->config_data[i] = GUINT32_TO_BE(vmaw->config_data[i]);
2067+ }
2068+
2069+ /* 32 bytes per device (12 used currently) = 8192 bytes max */
2070+ for (i = 1; i <= 254; i++) {
2071+ VmaStreamInfo *si = &vmaw->stream_info[i];
2072+ if (si->size) {
2073+ assert(si->devname);
2074+ uint32_t devname_ptr = allocate_header_string(vmaw, si->devname);
2075+ if (!devname_ptr) {
2076+ return -1;
2077+ }
2078+ head->dev_info[i].devname_ptr = GUINT32_TO_BE(devname_ptr);
2079+ head->dev_info[i].size = GUINT64_TO_BE(si->size);
2080+ }
2081+ }
2082+
2083+ uint32_t header_size = sizeof(VmaHeader) + vmaw->header_blob_table_size;
2084+ head->header_size = GUINT32_TO_BE(header_size);
2085+
67af0fa4 2086+ if (header_size > HEADERBUF_SIZE) {
95259824
WB		 2087+ return -1; /* just to be sure */
2088+ }
2089+
2090+ uint32_t blob_buffer_offset = sizeof(VmaHeader);
2091+ memcpy(buf + blob_buffer_offset, vmaw->header_blob_table,
2092+ vmaw->header_blob_table_size);
2093+ head->blob_buffer_offset = GUINT32_TO_BE(blob_buffer_offset);
2094+ head->blob_buffer_size = GUINT32_TO_BE(vmaw->header_blob_table_pos);
2095+
2096+ g_checksum_reset(vmaw->md5csum);
2097+ g_checksum_update(vmaw->md5csum, (const guchar *)buf, header_size);
2098+ gsize csize = 16;
2099+ g_checksum_get_digest(vmaw->md5csum, (guint8 *)(head->md5sum), &csize);
2100+
2101+ return vma_queue_write(vmaw, buf, header_size);
2102+}
2103+
2104+static int coroutine_fn vma_writer_flush(VmaWriter *vmaw)
2105+{
2106+ assert(vmaw);
2107+
2108+ int ret;
2109+ int i;
2110+
2111+ if (vmaw->status < 0) {
2112+ return vmaw->status;
2113+ }
2114+
2115+ if (!vmaw->header_written) {
2116+ vmaw->header_written = true;
2117+ ret = vma_write_header(vmaw);
2118+ if (ret < 0) {
2119+ vma_writer_set_error(vmaw, "vma_writer_flush: write header failed");
2120+ return ret;
2121+ }
2122+ }
2123+
2124+ DPRINTF("VMA WRITE FLUSH %d %d\n", vmaw->outbuf_count, vmaw->outbuf_pos);
2125+
2126+
2127+ VmaExtentHeader *ehead = (VmaExtentHeader *)vmaw->outbuf;
2128+
2129+ ehead->magic = VMA_EXTENT_MAGIC;
2130+ ehead->reserved1 = 0;
2131+
2132+ for (i = 0; i < VMA_BLOCKS_PER_EXTENT; i++) {
2133+ ehead->blockinfo[i] = GUINT64_TO_BE(vmaw->outbuf_block_info[i]);
2134+ }
2135+
2136+ guint16 block_count = (vmaw->outbuf_pos - VMA_EXTENT_HEADER_SIZE) /
2137+ VMA_BLOCK_SIZE;
2138+
2139+ ehead->block_count = GUINT16_TO_BE(block_count);
2140+
2141+ memcpy(ehead->uuid, vmaw->uuid, sizeof(ehead->uuid));
2142+ memset(ehead->md5sum, 0, sizeof(ehead->md5sum));
2143+
2144+ g_checksum_reset(vmaw->md5csum);
2145+ g_checksum_update(vmaw->md5csum, vmaw->outbuf, VMA_EXTENT_HEADER_SIZE);
2146+ gsize csize = 16;
2147+ g_checksum_get_digest(vmaw->md5csum, ehead->md5sum, &csize);
2148+
2149+ int bytes = vmaw->outbuf_pos;
2150+ ret = vma_queue_write(vmaw, vmaw->outbuf, bytes);
2151+ if (ret != bytes) {
2152+ vma_writer_set_error(vmaw, "vma_writer_flush: failed write");
2153+ }
2154+
2155+ vmaw->outbuf_count = 0;
2156+ vmaw->outbuf_pos = VMA_EXTENT_HEADER_SIZE;
2157+
2158+ for (i = 0; i < VMA_BLOCKS_PER_EXTENT; i++) {
2159+ vmaw->outbuf_block_info[i] = 0;
2160+ }
2161+
2162+ return vmaw->status;
2163+}
2164+
2165+static int vma_count_open_streams(VmaWriter *vmaw)
2166+{
2167+ g_assert(vmaw != NULL);
2168+
2169+ int i;
2170+ int open_drives = 0;
2171+ for (i = 0; i <= 255; i++) {
2172+ if (vmaw->stream_info[i].size && !vmaw->stream_info[i].finished) {
2173+ open_drives++;
2174+ }
2175+ }
2176+
2177+ return open_drives;
2178+}
2179+
67af0fa4
WB		 2180+
2181+/**
2182+ * You need to call this if the vma archive does not contain
2183+ * any data stream.
2184+ */
2185+int coroutine_fn
2186+vma_writer_flush_output(VmaWriter *vmaw)
2187+{
2188+ qemu_co_mutex_lock(&vmaw->flush_lock);
2189+ int ret = vma_writer_flush(vmaw);
2190+ qemu_co_mutex_unlock(&vmaw->flush_lock);
2191+ if (ret < 0) {
2192+ vma_writer_set_error(vmaw, "vma_writer_flush_header failed");
2193+ }
2194+ return ret;
2195+}
2196+
95259824
WB		 2197+/**
2198+ * all jobs should call this when there is no more data
2199+ * Returns: number of remaining stream (0 ==> finished)
2200+ */
2201+int coroutine_fn
2202+vma_writer_close_stream(VmaWriter *vmaw, uint8_t dev_id)
2203+{
2204+ g_assert(vmaw != NULL);
2205+
2206+ DPRINTF("vma_writer_set_status %d\n", dev_id);
2207+ if (!vmaw->stream_info[dev_id].size) {
2208+ vma_writer_set_error(vmaw, "vma_writer_close_stream: "
2209+ "no such stream %d", dev_id);
2210+ return -1;
2211+ }
2212+ if (vmaw->stream_info[dev_id].finished) {
2213+ vma_writer_set_error(vmaw, "vma_writer_close_stream: "
2214+ "stream already closed %d", dev_id);
2215+ return -1;
2216+ }
2217+
2218+ vmaw->stream_info[dev_id].finished = true;
2219+
2220+ int open_drives = vma_count_open_streams(vmaw);
2221+
2222+ if (open_drives <= 0) {
2223+ DPRINTF("vma_writer_set_status all drives completed\n");
67af0fa4 2224+ vma_writer_flush_output(vmaw);
95259824
WB		 2225+ }
2226+
2227+ return open_drives;
2228+}
2229+
2230+int vma_writer_get_status(VmaWriter *vmaw, VmaStatus *status)
2231+{
2232+ int i;
2233+
2234+ g_assert(vmaw != NULL);
2235+
2236+ if (status) {
2237+ status->status = vmaw->status;
2238+ g_strlcpy(status->errmsg, vmaw->errmsg, sizeof(status->errmsg));
2239+ for (i = 0; i <= 255; i++) {
2240+ status->stream_info[i] = vmaw->stream_info[i];
2241+ }
2242+
2243+ uuid_unparse_lower(vmaw->uuid, status->uuid_str);
2244+ }
2245+
2246+ status->closed = vmaw->closed;
2247+
2248+ return vmaw->status;
2249+}
2250+
2251+static int vma_writer_get_buffer(VmaWriter *vmaw)
2252+{
2253+ int ret = 0;
2254+
2255+ qemu_co_mutex_lock(&vmaw->flush_lock);
2256+
2257+ /* wait until buffer is available */
2258+ while (vmaw->outbuf_count >= (VMA_BLOCKS_PER_EXTENT - 1)) {
2259+ ret = vma_writer_flush(vmaw);
2260+ if (ret < 0) {
2261+ vma_writer_set_error(vmaw, "vma_writer_get_buffer: flush failed");
2262+ break;
2263+ }
2264+ }
2265+
2266+ qemu_co_mutex_unlock(&vmaw->flush_lock);
2267+
2268+ return ret;
2269+}
2270+
2271+
2272+int64_t coroutine_fn
2273+vma_writer_write(VmaWriter *vmaw, uint8_t dev_id, int64_t cluster_num,
2274+ unsigned char *buf, size_t *zero_bytes)
2275+{
2276+ g_assert(vmaw != NULL);
2277+ g_assert(zero_bytes != NULL);
2278+
2279+ *zero_bytes = 0;
2280+
2281+ if (vmaw->status < 0) {
2282+ return vmaw->status;
2283+ }
2284+
2285+ if (!dev_id || !vmaw->stream_info[dev_id].size) {
2286+ vma_writer_set_error(vmaw, "vma_writer_write: "
2287+ "no such stream %d", dev_id);
2288+ return -1;
2289+ }
2290+
2291+ if (vmaw->stream_info[dev_id].finished) {
2292+ vma_writer_set_error(vmaw, "vma_writer_write: "
2293+ "stream already closed %d", dev_id);
2294+ return -1;
2295+ }
2296+
2297+
2298+ if (cluster_num >= (((uint64_t)1)<<32)) {
2299+ vma_writer_set_error(vmaw, "vma_writer_write: "
2300+ "cluster number out of range");
2301+ return -1;
2302+ }
2303+
2304+ if (dev_id == vmaw->vmstate_stream) {
2305+ if (cluster_num != vmaw->vmstate_clusters) {
2306+ vma_writer_set_error(vmaw, "vma_writer_write: "
2307+ "non sequential vmstate write");
2308+ }
2309+ vmaw->vmstate_clusters++;
2310+ } else if (cluster_num >= vmaw->stream_info[dev_id].cluster_count) {
2311+ vma_writer_set_error(vmaw, "vma_writer_write: cluster number too big");
2312+ return -1;
2313+ }
2314+
2315+ /* wait until buffer is available */
2316+ if (vma_writer_get_buffer(vmaw) < 0) {
2317+ vma_writer_set_error(vmaw, "vma_writer_write: "
2318+ "vma_writer_get_buffer failed");
2319+ return -1;
2320+ }
2321+
2322+ DPRINTF("VMA WRITE %d %zd\n", dev_id, cluster_num);
2323+
2324+ uint16_t mask = 0;
2325+
2326+ if (buf) {
2327+ int i;
2328+ int bit = 1;
2329+ for (i = 0; i < 16; i++) {
2330+ unsigned char *vmablock = buf + (i*VMA_BLOCK_SIZE);
2331+ if (!buffer_is_zero(vmablock, VMA_BLOCK_SIZE)) {
2332+ mask |= bit;
2333+ memcpy(vmaw->outbuf + vmaw->outbuf_pos, vmablock,
2334+ VMA_BLOCK_SIZE);
2335+ vmaw->outbuf_pos += VMA_BLOCK_SIZE;
2336+ } else {
2337+ DPRINTF("VMA WRITE %zd ZERO BLOCK %d\n", cluster_num, i);
2338+ vmaw->stream_info[dev_id].zero_bytes += VMA_BLOCK_SIZE;
2339+ *zero_bytes += VMA_BLOCK_SIZE;
2340+ }
2341+
2342+ bit = bit << 1;
2343+ }
2344+ } else {
2345+ DPRINTF("VMA WRITE %zd ZERO CLUSTER\n", cluster_num);
2346+ vmaw->stream_info[dev_id].zero_bytes += VMA_CLUSTER_SIZE;
2347+ *zero_bytes += VMA_CLUSTER_SIZE;
2348+ }
2349+
2350+ uint64_t block_info = ((uint64_t)mask) << (32+16);
2351+ block_info |= ((uint64_t)dev_id) << 32;
2352+ block_info |= (cluster_num & 0xffffffff);
2353+ vmaw->outbuf_block_info[vmaw->outbuf_count] = block_info;
2354+
2355+ DPRINTF("VMA WRITE MASK %zd %zx\n", cluster_num, block_info);
2356+
2357+ vmaw->outbuf_count++;
2358+
2359+ /** NOTE: We allways write whole clusters, but we correctly set
2360+ * transferred bytes. So transferred == size when when everything
2361+ * went OK.
2362+ */
2363+ size_t transferred = VMA_CLUSTER_SIZE;
2364+
2365+ if (dev_id != vmaw->vmstate_stream) {
2366+ uint64_t last = (cluster_num + 1) * VMA_CLUSTER_SIZE;
2367+ if (last > vmaw->stream_info[dev_id].size) {
2368+ uint64_t diff = last - vmaw->stream_info[dev_id].size;
2369+ if (diff >= VMA_CLUSTER_SIZE) {
2370+ vma_writer_set_error(vmaw, "vma_writer_write: "
2371+ "read after last cluster");
2372+ return -1;
2373+ }
2374+ transferred -= diff;
2375+ }
2376+ }
2377+
2378+ vmaw->stream_info[dev_id].transferred += transferred;
2379+
2380+ return transferred;
2381+}
2382+
67af0fa4
WB		 2383+void vma_writer_error_propagate(VmaWriter *vmaw, Error **errp)
2384+{
2385+ if (vmaw->status < 0 && *errp == NULL) {
2386+ error_setg(errp, "%s", vmaw->errmsg);
2387+ }
2388+}
2389+
95259824
WB		 2390+int vma_writer_close(VmaWriter *vmaw, Error **errp)
2391+{
2392+ g_assert(vmaw != NULL);
2393+
2394+ int i;
2395+
67af0fa4
WB		 2396+ while (vmaw->co_writer) {
2397+ aio_poll(qemu_get_aio_context(), true);
95259824
WB		 2398+ }
2399+
67af0fa4
WB		 2400+ assert(vmaw->co_writer == NULL);
2401+
95259824
WB		 2402+ if (vmaw->cmd) {
2403+ if (pclose(vmaw->cmd) < 0) {
2404+ vma_writer_set_error(vmaw, "vma_writer_close: "
2405+ "pclose failed - %s", g_strerror(errno));
2406+ }
2407+ } else {
2408+ if (close(vmaw->fd) < 0) {
2409+ vma_writer_set_error(vmaw, "vma_writer_close: "
2410+ "close failed - %s", g_strerror(errno));
2411+ }
2412+ }
2413+
2414+ for (i = 0; i <= 255; i++) {
2415+ VmaStreamInfo *si = &vmaw->stream_info[i];
2416+ if (si->size) {
2417+ if (!si->finished) {
2418+ vma_writer_set_error(vmaw, "vma_writer_close: "
2419+ "detected open stream '%s'", si->devname);
2420+ } else if ((si->transferred != si->size) &&
2421+ (i != vmaw->vmstate_stream)) {
2422+ vma_writer_set_error(vmaw, "vma_writer_close: "
2423+ "incomplete stream '%s' (%zd != %zd)",
2424+ si->devname, si->transferred, si->size);
2425+ }
2426+ }
2427+ }
2428+
2429+ for (i = 0; i <= 255; i++) {
2430+ vmaw->stream_info[i].finished = 1; /* mark as closed */
2431+ }
2432+
2433+ vmaw->closed = 1;
2434+
2435+ if (vmaw->status < 0 && *errp == NULL) {
2436+ error_setg(errp, "%s", vmaw->errmsg);
2437+ }
2438+
2439+ return vmaw->status;
2440+}
2441+
2442+void vma_writer_destroy(VmaWriter *vmaw)
2443+{
2444+ assert(vmaw);
2445+
2446+ int i;
2447+
2448+ for (i = 0; i <= 255; i++) {
2449+ if (vmaw->stream_info[i].devname) {
2450+ g_free(vmaw->stream_info[i].devname);
2451+ }
2452+ }
2453+
2454+ if (vmaw->md5csum) {
2455+ g_checksum_free(vmaw->md5csum);
2456+ }
2457+
95259824
WB		 2458+ g_free(vmaw);
2459+}
2460diff --git a/vma.c b/vma.c
2461new file mode 100644
67af0fa4 2462index 0000000000..04915427c8
95259824
WB		 2463--- /dev/null
2464+++ b/vma.c
67af0fa4 2465@@ -0,0 +1,757 @@
95259824
WB		 2466+/*
2467+ * VMA: Virtual Machine Archive
2468+ *
2469+ * Copyright (C) 2012-2013 Proxmox Server Solutions
2470+ *
2471+ * Authors:
2472+ * Dietmar Maurer (dietmar@proxmox.com)
2473+ *
2474+ * This work is licensed under the terms of the GNU GPL, version 2 or later.
2475+ * See the COPYING file in the top-level directory.
2476+ *
2477+ */
2478+
2479+#include "qemu/osdep.h"
2480+#include <glib.h>
2481+
2482+#include "vma.h"
2483+#include "qemu-common.h"
2484+#include "qemu/error-report.h"
2485+#include "qemu/main-loop.h"
a544966d 2486+#include "qapi/qmp/qstring.h"
95259824 2487+#include "sysemu/char.h" /* qstring_from_str */
67af0fa4 2488+#include "sysemu/block-backend.h"
95259824
WB		 2489+
2490+static void help(void)
2491+{
2492+ const char *help_msg =
2493+ "usage: vma command [command options]\n"
2494+ "\n"
2495+ "vma list <filename>\n"
67af0fa4
WB		 2496+ "vma config <filename> [-c config]\n"
2497+ "vma create <filename> [-c config] pathname ...\n"
95259824 2498+ "vma extract <filename> [-r <fifo>] <targetdir>\n"
67af0fa4 2499+ "vma verify <filename> [-v]\n"
95259824
WB		 2500+ ;
2501+
2502+ printf("%s", help_msg);
2503+ exit(1);
2504+}
2505+
2506+static const char *extract_devname(const char *path, char **devname, int index)
2507+{
2508+ assert(path);
2509+
2510+ const char *sep = strchr(path, '=');
2511+
2512+ if (sep) {
2513+ *devname = g_strndup(path, sep - path);
2514+ path = sep + 1;
2515+ } else {
2516+ if (index >= 0) {
2517+ *devname = g_strdup_printf("disk%d", index);
2518+ } else {
2519+ *devname = NULL;
2520+ }
2521+ }
2522+
2523+ return path;
2524+}
2525+
2526+static void print_content(VmaReader *vmar)
2527+{
2528+ assert(vmar);
2529+
2530+ VmaHeader *head = vma_reader_get_header(vmar);
2531+
2532+ GList *l = vma_reader_get_config_data(vmar);
2533+ while (l && l->data) {
2534+ VmaConfigData *cdata = (VmaConfigData *)l->data;
2535+ l = g_list_next(l);
2536+ printf("CFG: size: %d name: %s\n", cdata->len, cdata->name);
2537+ }
2538+
2539+ int i;
2540+ VmaDeviceInfo *di;
2541+ for (i = 1; i < 255; i++) {
2542+ di = vma_reader_get_device_info(vmar, i);
2543+ if (di) {
2544+ if (strcmp(di->devname, "vmstate") == 0) {
2545+ printf("VMSTATE: dev_id=%d memory: %zd\n", i, di->size);
2546+ } else {
2547+ printf("DEV: dev_id=%d size: %zd devname: %s\n",
2548+ i, di->size, di->devname);
2549+ }
2550+ }
2551+ }
2552+ /* ctime is the last entry we print */
2553+ printf("CTIME: %s", ctime(&head->ctime));
2554+ fflush(stdout);
2555+}
2556+
2557+static int list_content(int argc, char **argv)
2558+{
2559+ int c, ret = 0;
2560+ const char *filename;
2561+
2562+ for (;;) {
2563+ c = getopt(argc, argv, "h");
2564+ if (c == -1) {
2565+ break;
2566+ }
2567+ switch (c) {
2568+ case '?':
2569+ case 'h':
2570+ help();
2571+ break;
2572+ default:
2573+ g_assert_not_reached();
2574+ }
2575+ }
2576+
2577+ /* Get the filename */
2578+ if ((optind + 1) != argc) {
2579+ help();
2580+ }
2581+ filename = argv[optind++];
2582+
2583+ Error *errp = NULL;
2584+ VmaReader *vmar = vma_reader_create(filename, &errp);
2585+
2586+ if (!vmar) {
2587+ g_error("%s", error_get_pretty(errp));
2588+ }
2589+
2590+ print_content(vmar);
2591+
2592+ vma_reader_destroy(vmar);
2593+
2594+ return ret;
2595+}
2596+
2597+typedef struct RestoreMap {
2598+ char *devname;
2599+ char *path;
67af0fa4 2600+ char *format;
95259824
WB		 2601+ bool write_zero;
2602+} RestoreMap;
2603+
2604+static int extract_content(int argc, char **argv)
2605+{
2606+ int c, ret = 0;
2607+ int verbose = 0;
2608+ const char *filename;
2609+ const char *dirname;
2610+ const char *readmap = NULL;
2611+
2612+ for (;;) {
2613+ c = getopt(argc, argv, "hvr:");
2614+ if (c == -1) {
2615+ break;
2616+ }
2617+ switch (c) {
2618+ case '?':
2619+ case 'h':
2620+ help();
2621+ break;
2622+ case 'r':
2623+ readmap = optarg;
2624+ break;
2625+ case 'v':
2626+ verbose = 1;
2627+ break;
2628+ default:
2629+ help();
2630+ }
2631+ }
2632+
2633+ /* Get the filename */
2634+ if ((optind + 2) != argc) {
2635+ help();
2636+ }
2637+ filename = argv[optind++];
2638+ dirname = argv[optind++];
2639+
2640+ Error *errp = NULL;
2641+ VmaReader *vmar = vma_reader_create(filename, &errp);
2642+
2643+ if (!vmar) {
2644+ g_error("%s", error_get_pretty(errp));
2645+ }
2646+
2647+ if (mkdir(dirname, 0777) < 0) {
2648+ g_error("unable to create target directory %s - %s",
2649+ dirname, g_strerror(errno));
2650+ }
2651+
2652+ GList *l = vma_reader_get_config_data(vmar);
2653+ while (l && l->data) {
2654+ VmaConfigData *cdata = (VmaConfigData *)l->data;
2655+ l = g_list_next(l);
2656+ char *cfgfn = g_strdup_printf("%s/%s", dirname, cdata->name);
2657+ GError *err = NULL;
2658+ if (!g_file_set_contents(cfgfn, (gchar *)cdata->data, cdata->len,
2659+ &err)) {
2660+ g_error("unable to write file: %s", err->message);
2661+ }
2662+ }
2663+
2664+ GHashTable *devmap = g_hash_table_new(g_str_hash, g_str_equal);
2665+
2666+ if (readmap) {
2667+ print_content(vmar);
2668+
2669+ FILE *map = fopen(readmap, "r");
2670+ if (!map) {
2671+ g_error("unable to open fifo %s - %s", readmap, g_strerror(errno));
2672+ }
2673+
2674+ while (1) {
2675+ char inbuf[8192];
2676+ char *line = fgets(inbuf, sizeof(inbuf), map);
2677+ if (!line || line[0] == '\0' || !strcmp(line, "done\n")) {
2678+ break;
2679+ }
2680+ int len = strlen(line);
2681+ if (line[len - 1] == '\n') {
2682+ line[len - 1] = '\0';
2683+ if (len == 1) {
2684+ break;
2685+ }
2686+ }
2687+
67af0fa4
WB		 2688+ char *format = NULL;
2689+ if (strncmp(line, "format=", sizeof("format=")-1) == 0) {
2690+ format = line + sizeof("format=")-1;
2691+ char *colon = strchr(format, ':');
2692+ if (!colon) {
2693+ g_error("read map failed - found only a format ('%s')", inbuf);
2694+ }
2695+ format = g_strndup(format, colon - format);
2696+ line = colon+1;
2697+ }
2698+
95259824
WB		 2699+ const char *path;
2700+ bool write_zero;
2701+ if (line[0] == '0' && line[1] == ':') {
67af0fa4 2702+ path = line + 2;
95259824
WB		 2703+ write_zero = false;
2704+ } else if (line[0] == '1' && line[1] == ':') {
67af0fa4 2705+ path = line + 2;
95259824
WB		 2706+ write_zero = true;
2707+ } else {
2708+ g_error("read map failed - parse error ('%s')", inbuf);
2709+ }
2710+
2711+ char *devname = NULL;
2712+ path = extract_devname(path, &devname, -1);
2713+ if (!devname) {
2714+ g_error("read map failed - no dev name specified ('%s')",
2715+ inbuf);
2716+ }
2717+
2718+ RestoreMap *map = g_new0(RestoreMap, 1);
2719+ map->devname = g_strdup(devname);
2720+ map->path = g_strdup(path);
67af0fa4 2721+ map->format = format;
95259824
WB		 2722+ map->write_zero = write_zero;
2723+
2724+ g_hash_table_insert(devmap, map->devname, map);
2725+
2726+ };
2727+ }
2728+
2729+ int i;
2730+ int vmstate_fd = -1;
2731+ guint8 vmstate_stream = 0;
2732+
67af0fa4
WB		 2733+ BlockBackend *blk = NULL;
2734+
95259824
WB		 2735+ for (i = 1; i < 255; i++) {
2736+ VmaDeviceInfo *di = vma_reader_get_device_info(vmar, i);
2737+ if (di && (strcmp(di->devname, "vmstate") == 0)) {
2738+ vmstate_stream = i;
2739+ char *statefn = g_strdup_printf("%s/vmstate.bin", dirname);
2740+ vmstate_fd = open(statefn, O_WRONLY|O_CREAT|O_EXCL, 0644);
2741+ if (vmstate_fd < 0) {
2742+ g_error("create vmstate file '%s' failed - %s", statefn,
2743+ g_strerror(errno));
2744+ }
2745+ g_free(statefn);
2746+ } else if (di) {
2747+ char *devfn = NULL;
67af0fa4
WB		 2748+ const char *format = NULL;
2749+ int flags = BDRV_O_RDWR | BDRV_O_NO_FLUSH;
95259824
WB		 2750+ bool write_zero = true;
2751+
2752+ if (readmap) {
2753+ RestoreMap *map;
2754+ map = (RestoreMap *)g_hash_table_lookup(devmap, di->devname);
2755+ if (map == NULL) {
2756+ g_error("no device name mapping for %s", di->devname);
2757+ }
2758+ devfn = map->path;
67af0fa4 2759+ format = map->format;
95259824
WB		 2760+ write_zero = map->write_zero;
2761+ } else {
2762+ devfn = g_strdup_printf("%s/tmp-disk-%s.raw",
2763+ dirname, di->devname);
2764+ printf("DEVINFO %s %zd\n", devfn, di->size);
2765+
2766+ bdrv_img_create(devfn, "raw", NULL, NULL, NULL, di->size,
2767+ flags, &errp, 0);
2768+ if (errp) {
2769+ g_error("can't create file %s: %s", devfn,
2770+ error_get_pretty(errp));
2771+ }
2772+
2773+ /* Note: we created an empty file above, so there is no
2774+ * need to write zeroes (so we generate a sparse file)
2775+ */
2776+ write_zero = false;
2777+ }
2778+
67af0fa4
WB		 2779+ size_t devlen = strlen(devfn);
2780+ QDict *options = NULL;
2781+ if (format) {
2782+ /* explicit format from commandline */
2783+ options = qdict_new();
2784+ qdict_put(options, "driver", qstring_from_str(format));
2785+ } else if ((devlen > 4 && strcmp(devfn+devlen-4, ".raw") == 0) ||
2786+ strncmp(devfn, "/dev/", 5) == 0)
2787+ {
2788+ /* This part is now deprecated for PVE as well (just as qemu
2789+ * deprecated not specifying an explicit raw format, too.
2790+ */
2791+ /* explicit raw format */
2792+ options = qdict_new();
2793+ qdict_put(options, "driver", qstring_from_str("raw"));
2794+ }
2795+
2796+
2797+ if (errp || !(blk = blk_new_open(devfn, NULL, options, flags, &errp))) {
95259824
WB		 2798+ g_error("can't open file %s - %s", devfn,
2799+ error_get_pretty(errp));
2800+ }
67af0fa4
WB		 2801+
2802+ if (vma_reader_register_bs(vmar, i, blk, write_zero, &errp) < 0) {
95259824
WB		 2803+ g_error("%s", error_get_pretty(errp));
2804+ }
2805+
2806+ if (!readmap) {
2807+ g_free(devfn);
2808+ }
2809+ }
2810+ }
2811+
2812+ if (vma_reader_restore(vmar, vmstate_fd, verbose, &errp) < 0) {
2813+ g_error("restore failed - %s", error_get_pretty(errp));
2814+ }
2815+
2816+ if (!readmap) {
2817+ for (i = 1; i < 255; i++) {
2818+ VmaDeviceInfo *di = vma_reader_get_device_info(vmar, i);
2819+ if (di && (i != vmstate_stream)) {
2820+ char *tmpfn = g_strdup_printf("%s/tmp-disk-%s.raw",
2821+ dirname, di->devname);
2822+ char *fn = g_strdup_printf("%s/disk-%s.raw",
2823+ dirname, di->devname);
2824+ if (rename(tmpfn, fn) != 0) {
2825+ g_error("rename %s to %s failed - %s",
2826+ tmpfn, fn, g_strerror(errno));
2827+ }
2828+ }
2829+ }
2830+ }
2831+
2832+ vma_reader_destroy(vmar);
2833+
67af0fa4
WB		 2834+ blk_unref(blk);
2835+
2836+ bdrv_close_all();
2837+
2838+ return ret;
2839+}
2840+
2841+static int verify_content(int argc, char **argv)
2842+{
2843+ int c, ret = 0;
2844+ int verbose = 0;
2845+ const char *filename;
2846+
2847+ for (;;) {
2848+ c = getopt(argc, argv, "hv");
2849+ if (c == -1) {
2850+ break;
2851+ }
2852+ switch (c) {
2853+ case '?':
2854+ case 'h':
2855+ help();
2856+ break;
2857+ case 'v':
2858+ verbose = 1;
2859+ break;
2860+ default:
2861+ help();
2862+ }
2863+ }
2864+
2865+ /* Get the filename */
2866+ if ((optind + 1) != argc) {
2867+ help();
2868+ }
2869+ filename = argv[optind++];
2870+
2871+ Error *errp = NULL;
2872+ VmaReader *vmar = vma_reader_create(filename, &errp);
2873+
2874+ if (!vmar) {
2875+ g_error("%s", error_get_pretty(errp));
2876+ }
2877+
2878+ if (verbose) {
2879+ print_content(vmar);
2880+ }
2881+
2882+ if (vma_reader_verify(vmar, verbose, &errp) < 0) {
2883+ g_error("verify failed - %s", error_get_pretty(errp));
2884+ }
2885+
2886+ vma_reader_destroy(vmar);
2887+
95259824
WB		 2888+ bdrv_close_all();
2889+
2890+ return ret;
2891+}
2892+
2893+typedef struct BackupJob {
67af0fa4 2894+ BlockBackend *target;
95259824
WB		 2895+ int64_t len;
2896+ VmaWriter *vmaw;
2897+ uint8_t dev_id;
2898+} BackupJob;
2899+
2900+#define BACKUP_SECTORS_PER_CLUSTER (VMA_CLUSTER_SIZE / BDRV_SECTOR_SIZE)
2901+
67af0fa4
WB		 2902+static void coroutine_fn backup_run_empty(void *opaque)
2903+{
2904+ VmaWriter *vmaw = (VmaWriter *)opaque;
2905+
2906+ vma_writer_flush_output(vmaw);
2907+
2908+ Error *err = NULL;
2909+ if (vma_writer_close(vmaw, &err) != 0) {
2910+ g_warning("vma_writer_close failed %s", error_get_pretty(err));
2911+ }
2912+}
2913+
95259824
WB		 2914+static void coroutine_fn backup_run(void *opaque)
2915+{
2916+ BackupJob *job = (BackupJob *)opaque;
2917+ struct iovec iov;
2918+ QEMUIOVector qiov;
2919+
2920+ int64_t start, end;
2921+ int ret = 0;
2922+
67af0fa4 2923+ unsigned char *buf = blk_blockalign(job->target, VMA_CLUSTER_SIZE);
95259824
WB		 2924+
2925+ start = 0;
2926+ end = DIV_ROUND_UP(job->len / BDRV_SECTOR_SIZE,
2927+ BACKUP_SECTORS_PER_CLUSTER);
2928+
2929+ for (; start < end; start++) {
2930+ iov.iov_base = buf;
2931+ iov.iov_len = VMA_CLUSTER_SIZE;
2932+ qemu_iovec_init_external(&qiov, &iov, 1);
2933+
67af0fa4
WB		 2934+ ret = blk_co_preadv(job->target, start * VMA_CLUSTER_SIZE,
2935+ VMA_CLUSTER_SIZE, &qiov, 0);
95259824
WB		 2936+ if (ret < 0) {
2937+ vma_writer_set_error(job->vmaw, "read error", -1);
2938+ goto out;
2939+ }
2940+
2941+ size_t zb = 0;
2942+ if (vma_writer_write(job->vmaw, job->dev_id, start, buf, &zb) < 0) {
2943+ vma_writer_set_error(job->vmaw, "backup_dump_cb vma_writer_write failed", -1);
2944+ goto out;
2945+ }
2946+ }
2947+
2948+
2949+out:
2950+ if (vma_writer_close_stream(job->vmaw, job->dev_id) <= 0) {
2951+ Error *err = NULL;
2952+ if (vma_writer_close(job->vmaw, &err) != 0) {
2953+ g_warning("vma_writer_close failed %s", error_get_pretty(err));
2954+ }
2955+ }
2956+}
2957+
2958+static int create_archive(int argc, char **argv)
2959+{
2960+ int i, c;
2961+ int verbose = 0;
2962+ const char *archivename;
2963+ GList *config_files = NULL;
2964+
2965+ for (;;) {
2966+ c = getopt(argc, argv, "hvc:");
2967+ if (c == -1) {
2968+ break;
2969+ }
2970+ switch (c) {
2971+ case '?':
2972+ case 'h':
2973+ help();
2974+ break;
2975+ case 'c':
2976+ config_files = g_list_append(config_files, optarg);
2977+ break;
2978+ case 'v':
2979+ verbose = 1;
2980+ break;
2981+ default:
2982+ g_assert_not_reached();
2983+ }
2984+ }
2985+
2986+
67af0fa4
WB		 2987+ /* make sure we an archive name */
2988+ if ((optind + 1) > argc) {
95259824
WB		 2989+ help();
2990+ }
2991+
2992+ archivename = argv[optind++];
2993+
2994+ uuid_t uuid;
2995+ uuid_generate(uuid);
2996+
2997+ Error *local_err = NULL;
2998+ VmaWriter *vmaw = vma_writer_create(archivename, uuid, &local_err);
2999+
3000+ if (vmaw == NULL) {
3001+ g_error("%s", error_get_pretty(local_err));
3002+ }
3003+
3004+ GList *l = config_files;
3005+ while (l && l->data) {
3006+ char *name = l->data;
3007+ char *cdata = NULL;
3008+ gsize clen = 0;
3009+ GError *err = NULL;
3010+ if (!g_file_get_contents(name, &cdata, &clen, &err)) {
3011+ unlink(archivename);
3012+ g_error("Unable to read file: %s", err->message);
3013+ }
3014+
3015+ if (vma_writer_add_config(vmaw, name, cdata, clen) != 0) {
3016+ unlink(archivename);
3017+ g_error("Unable to append config data %s (len = %zd)",
3018+ name, clen);
3019+ }
3020+ l = g_list_next(l);
3021+ }
3022+
67af0fa4 3023+ int devcount = 0;
95259824
WB		 3024+ while (optind < argc) {
3025+ const char *path = argv[optind++];
3026+ char *devname = NULL;
67af0fa4 3027+ path = extract_devname(path, &devname, devcount++);
95259824
WB		 3028+
3029+ Error *errp = NULL;
67af0fa4 3030+ BlockBackend *target;
95259824 3031+
67af0fa4
WB		 3032+ target = blk_new_open(path, NULL, NULL, 0, &errp);
3033+ if (!target) {
95259824
WB		 3034+ unlink(archivename);
3035+ g_error("bdrv_open '%s' failed - %s", path, error_get_pretty(errp));
3036+ }
67af0fa4 3037+ int64_t size = blk_getlength(target);
95259824
WB		 3038+ int dev_id = vma_writer_register_stream(vmaw, devname, size);
3039+ if (dev_id <= 0) {
3040+ unlink(archivename);
3041+ g_error("vma_writer_register_stream '%s' failed", devname);
3042+ }
3043+
3044+ BackupJob *job = g_new0(BackupJob, 1);
3045+ job->len = size;
67af0fa4 3046+ job->target = target;
95259824
WB		 3047+ job->vmaw = vmaw;
3048+ job->dev_id = dev_id;
3049+
3050+ Coroutine *co = qemu_coroutine_create(backup_run, job);
3051+ qemu_coroutine_enter(co);
3052+ }
3053+
3054+ VmaStatus vmastat;
3055+ int percent = 0;
3056+ int last_percent = -1;
3057+
67af0fa4
WB		 3058+ if (devcount) {
3059+ while (1) {
3060+ main_loop_wait(false);
3061+ vma_writer_get_status(vmaw, &vmastat);
95259824 3062+
67af0fa4 3063+ if (verbose) {
95259824 3064+
67af0fa4
WB		 3065+ uint64_t total = 0;
3066+ uint64_t transferred = 0;
3067+ uint64_t zero_bytes = 0;
95259824 3068+
67af0fa4
WB		 3069+ int i;
3070+ for (i = 0; i < 256; i++) {
3071+ if (vmastat.stream_info[i].size) {
3072+ total += vmastat.stream_info[i].size;
3073+ transferred += vmastat.stream_info[i].transferred;
3074+ zero_bytes += vmastat.stream_info[i].zero_bytes;
3075+ }
95259824 3076+ }
67af0fa4
WB		 3077+ percent = (transferred*100)/total;
3078+ if (percent != last_percent) {
3079+ fprintf(stderr, "progress %d%% %zd/%zd %zd\n", percent,
3080+ transferred, total, zero_bytes);
3081+ fflush(stderr);
95259824 3082+
67af0fa4
WB		 3083+ last_percent = percent;
3084+ }
95259824 3085+ }
95259824 3086+
67af0fa4
WB		 3087+ if (vmastat.closed) {
3088+ break;
3089+ }
95259824
WB		 3090+ }
3091+ } else {
3092+ Coroutine *co = qemu_coroutine_create(backup_run_empty, vmaw);
3093+ qemu_coroutine_enter(co);
3094+ while (1) {
3095+ main_loop_wait(false);
3096+ vma_writer_get_status(vmaw, &vmastat);
3097+ if (vmastat.closed) {
3098+ break;
3099+ }
3100+ }
3101+ }
3102+
3103+ bdrv_drain_all();
3104+
3105+ vma_writer_get_status(vmaw, &vmastat);
3106+
3107+ if (verbose) {
3108+ for (i = 0; i < 256; i++) {
3109+ VmaStreamInfo *si = &vmastat.stream_info[i];
3110+ if (si->size) {
3111+ fprintf(stderr, "image %s: size=%zd zeros=%zd saved=%zd\n",
3112+ si->devname, si->size, si->zero_bytes,
3113+ si->size - si->zero_bytes);
3114+ }
3115+ }
3116+ }
3117+
3118+ if (vmastat.status < 0) {
3119+ unlink(archivename);
3120+ g_error("creating vma archive failed");
3121+ }
3122+
3123+ return 0;
3124+}
3125+
67af0fa4
WB		 3126+static int dump_config(int argc, char **argv)
3127+{
3128+ int c, ret = 0;
3129+ const char *filename;
3130+ const char *config_name = "qemu-server.conf";
3131+
3132+ for (;;) {
3133+ c = getopt(argc, argv, "hc:");
3134+ if (c == -1) {
3135+ break;
3136+ }
3137+ switch (c) {
3138+ case '?':
3139+ case 'h':
3140+ help();
3141+ break;
3142+ case 'c':
3143+ config_name = optarg;
3144+ break;
3145+ default:
3146+ help();
3147+ }
3148+ }
3149+
3150+ /* Get the filename */
3151+ if ((optind + 1) != argc) {
3152+ help();
3153+ }
3154+ filename = argv[optind++];
3155+
3156+ Error *errp = NULL;
3157+ VmaReader *vmar = vma_reader_create(filename, &errp);
3158+
3159+ if (!vmar) {
3160+ g_error("%s", error_get_pretty(errp));
3161+ }
3162+
3163+ int found = 0;
3164+ GList *l = vma_reader_get_config_data(vmar);
3165+ while (l && l->data) {
3166+ VmaConfigData *cdata = (VmaConfigData *)l->data;
3167+ l = g_list_next(l);
3168+ if (strcmp(cdata->name, config_name) == 0) {
3169+ found = 1;
3170+ fwrite(cdata->data, cdata->len, 1, stdout);
3171+ break;
3172+ }
3173+ }
3174+
3175+ vma_reader_destroy(vmar);
3176+
3177+ bdrv_close_all();
3178+
3179+ if (!found) {
3180+ fprintf(stderr, "unable to find configuration data '%s'\n", config_name);
3181+ return -1;
3182+ }
3183+
3184+ return ret;
3185+}
3186+
95259824
WB		 3187+int main(int argc, char **argv)
3188+{
3189+ const char *cmdname;
3190+ Error *main_loop_err = NULL;
3191+
3192+ error_set_progname(argv[0]);
3193+
3194+ if (qemu_init_main_loop(&main_loop_err)) {
3195+ g_error("%s", error_get_pretty(main_loop_err));
3196+ }
3197+
3198+ bdrv_init();
3199+
3200+ if (argc < 2) {
3201+ help();
3202+ }
3203+
3204+ cmdname = argv[1];
3205+ argc--; argv++;
3206+
3207+
3208+ if (!strcmp(cmdname, "list")) {
3209+ return list_content(argc, argv);
3210+ } else if (!strcmp(cmdname, "create")) {
3211+ return create_archive(argc, argv);
3212+ } else if (!strcmp(cmdname, "extract")) {
3213+ return extract_content(argc, argv);
67af0fa4
WB		 3214+ } else if (!strcmp(cmdname, "verify")) {
3215+ return verify_content(argc, argv);
3216+ } else if (!strcmp(cmdname, "config")) {
3217+ return dump_config(argc, argv);
95259824
WB		 3218+ }
3219+
3220+ help();
3221+ return 0;
3222+}
3223diff --git a/vma.h b/vma.h
3224new file mode 100644
67af0fa4 3225index 0000000000..fa6f4df7e7
95259824
WB		 3226--- /dev/null
3227+++ b/vma.h
67af0fa4 3228@@ -0,0 +1,149 @@
95259824
WB		 3229+/*
3230+ * VMA: Virtual Machine Archive
3231+ *
3232+ * Copyright (C) Proxmox Server Solutions
3233+ *
3234+ * Authors:
3235+ * Dietmar Maurer (dietmar@proxmox.com)
3236+ *
3237+ * This work is licensed under the terms of the GNU GPL, version 2 or later.
3238+ * See the COPYING file in the top-level directory.
3239+ *
3240+ */
3241+
3242+#ifndef BACKUP_VMA_H
3243+#define BACKUP_VMA_H
3244+
3245+#include <uuid/uuid.h>
3246+#include "qapi/error.h"
3247+#include "block/block.h"
3248+
3249+#define VMA_BLOCK_BITS 12
3250+#define VMA_BLOCK_SIZE (1<<VMA_BLOCK_BITS)
3251+#define VMA_CLUSTER_BITS (VMA_BLOCK_BITS+4)
3252+#define VMA_CLUSTER_SIZE (1<<VMA_CLUSTER_BITS)
3253+
3254+#if VMA_CLUSTER_SIZE != 65536
3255+#error unexpected cluster size
3256+#endif
3257+
3258+#define VMA_EXTENT_HEADER_SIZE 512
3259+#define VMA_BLOCKS_PER_EXTENT 59
3260+#define VMA_MAX_CONFIGS 256
3261+
3262+#define VMA_MAX_EXTENT_SIZE \
3263+ (VMA_EXTENT_HEADER_SIZE+VMA_CLUSTER_SIZE*VMA_BLOCKS_PER_EXTENT)
3264+#if VMA_MAX_EXTENT_SIZE != 3867136
3265+#error unexpected VMA_EXTENT_SIZE
3266+#endif
3267+
3268+/* File Format Definitions */
3269+
3270+#define VMA_MAGIC (GUINT32_TO_BE(('V'<<24)|('M'<<16)|('A'<<8)|0x00))
3271+#define VMA_EXTENT_MAGIC (GUINT32_TO_BE(('V'<<24)|('M'<<16)|('A'<<8)|'E'))
3272+
3273+typedef struct VmaDeviceInfoHeader {
3274+ uint32_t devname_ptr; /* offset into blob_buffer table */
3275+ uint32_t reserved0;
3276+ uint64_t size; /* device size in bytes */
3277+ uint64_t reserved1;
3278+ uint64_t reserved2;
3279+} VmaDeviceInfoHeader;
3280+
3281+typedef struct VmaHeader {
3282+ uint32_t magic;
3283+ uint32_t version;
3284+ unsigned char uuid[16];
3285+ int64_t ctime;
3286+ unsigned char md5sum[16];
3287+
3288+ uint32_t blob_buffer_offset;
3289+ uint32_t blob_buffer_size;
3290+ uint32_t header_size;
3291+
3292+ unsigned char reserved[1984];
3293+
3294+ uint32_t config_names[VMA_MAX_CONFIGS]; /* offset into blob_buffer table */
3295+ uint32_t config_data[VMA_MAX_CONFIGS]; /* offset into blob_buffer table */
3296+
3297+ uint32_t reserved1;
3298+
3299+ VmaDeviceInfoHeader dev_info[256];
3300+} VmaHeader;
3301+
3302+typedef struct VmaExtentHeader {
3303+ uint32_t magic;
3304+ uint16_t reserved1;
3305+ uint16_t block_count;
3306+ unsigned char uuid[16];
3307+ unsigned char md5sum[16];
3308+ uint64_t blockinfo[VMA_BLOCKS_PER_EXTENT];
3309+} VmaExtentHeader;
3310+
3311+/* functions/definitions to read/write vma files */
3312+
3313+typedef struct VmaReader VmaReader;
3314+
3315+typedef struct VmaWriter VmaWriter;
3316+
3317+typedef struct VmaConfigData {
3318+ const char *name;
3319+ const void *data;
3320+ uint32_t len;
3321+} VmaConfigData;
3322+
3323+typedef struct VmaStreamInfo {
3324+ uint64_t size;
3325+ uint64_t cluster_count;
3326+ uint64_t transferred;
3327+ uint64_t zero_bytes;
3328+ int finished;
3329+ char *devname;
3330+} VmaStreamInfo;
3331+
3332+typedef struct VmaStatus {
3333+ int status;
3334+ bool closed;
3335+ char errmsg[8192];
3336+ char uuid_str[37];
3337+ VmaStreamInfo stream_info[256];
3338+} VmaStatus;
3339+
3340+typedef struct VmaDeviceInfo {
3341+ uint64_t size; /* device size in bytes */
3342+ const char *devname;
3343+} VmaDeviceInfo;
3344+
3345+VmaWriter *vma_writer_create(const char *filename, uuid_t uuid, Error **errp);
3346+int vma_writer_close(VmaWriter *vmaw, Error **errp);
67af0fa4 3347+void vma_writer_error_propagate(VmaWriter *vmaw, Error **errp);
95259824
WB		 3348+void vma_writer_destroy(VmaWriter *vmaw);
3349+int vma_writer_add_config(VmaWriter *vmaw, const char *name, gpointer data,
3350+ size_t len);
3351+int vma_writer_register_stream(VmaWriter *vmaw, const char *devname,
3352+ size_t size);
3353+
3354+int64_t coroutine_fn vma_writer_write(VmaWriter *vmaw, uint8_t dev_id,
3355+ int64_t cluster_num, unsigned char *buf,
3356+ size_t *zero_bytes);
3357+
3358+int coroutine_fn vma_writer_close_stream(VmaWriter *vmaw, uint8_t dev_id);
67af0fa4 3359+int coroutine_fn vma_writer_flush_output(VmaWriter *vmaw);
95259824
WB		 3360+
3361+int vma_writer_get_status(VmaWriter *vmaw, VmaStatus *status);
3362+void vma_writer_set_error(VmaWriter *vmaw, const char *fmt, ...);
3363+
3364+
3365+VmaReader *vma_reader_create(const char *filename, Error **errp);
3366+void vma_reader_destroy(VmaReader *vmar);
3367+VmaHeader *vma_reader_get_header(VmaReader *vmar);
3368+GList *vma_reader_get_config_data(VmaReader *vmar);
3369+VmaDeviceInfo *vma_reader_get_device_info(VmaReader *vmar, guint8 dev_id);
3370+int vma_reader_register_bs(VmaReader *vmar, guint8 dev_id,
67af0fa4 3371+ BlockBackend *target, bool write_zeroes,
95259824
WB		 3372+ Error **errp);
3373+int vma_reader_restore(VmaReader *vmar, int vmstate_fd, bool verbose,
3374+ Error **errp);
67af0fa4 3375+int vma_reader_verify(VmaReader *vmar, bool verbose, Error **errp);
95259824
WB		 3376+
3377+#endif /* BACKUP_VMA_H */
3378--
45169293 33792.11.0
95259824 3380
QEMU for PVE