1 From 0000000000000000000000000000000000000000 Mon Sep 17 00:00:00 2001
2 From: Wolfgang Bumiller <w.bumiller@proxmox.com>
3 Date: Mon, 7 Aug 2017 08:51:16 +0200
4 Subject: [PATCH] adding old vma files
9 block/backup.c | 132 ++++---
10 block/replication.c | 1 +
11 blockdev.c | 249 +++++++++-----
13 include/block/block_int.h | 4 +
14 vma-reader.c | 857 ++++++++++++++++++++++++++++++++++++++++++++++
15 vma-writer.c | 771 +++++++++++++++++++++++++++++++++++++++++
16 vma.c | 757 ++++++++++++++++++++++++++++++++++++++++
18 11 files changed, 2802 insertions(+), 133 deletions(-)
19 create mode 100644 vma-reader.c
20 create mode 100644 vma-writer.c
21 create mode 100644 vma.c
22 create mode 100644 vma.h
24 diff --git a/Makefile b/Makefile
25 index 6c359b2f86..edbc8b50f0 100644
28 @@ -284,7 +284,7 @@ ifneq ($(wildcard config-host.mak),)
29 include $(SRC_PATH)/tests/Makefile.include
32 -all: $(DOCS) $(TOOLS) $(HELPERS-y) recurse-all modules
33 +all: $(DOCS) $(TOOLS) vma$(EXESUF) $(HELPERS-y) recurse-all modules
36 $(call quiet-command, \
37 @@ -377,6 +377,7 @@ qemu-img.o: qemu-img-cmds.h
38 qemu-img$(EXESUF): qemu-img.o $(block-obj-y) $(crypto-obj-y) $(io-obj-y) $(qom-obj-y) $(COMMON_LDADDS)
39 qemu-nbd$(EXESUF): qemu-nbd.o $(block-obj-y) $(crypto-obj-y) $(io-obj-y) $(qom-obj-y) $(COMMON_LDADDS)
40 qemu-io$(EXESUF): qemu-io.o $(block-obj-y) $(crypto-obj-y) $(io-obj-y) $(qom-obj-y) $(COMMON_LDADDS)
41 +vma$(EXESUF): vma.o vma-reader.o $(block-obj-y) $(crypto-obj-y) $(io-obj-y) $(qom-obj-y) $(COMMON_LDADDS)
43 qemu-bridge-helper$(EXESUF): qemu-bridge-helper.o $(COMMON_LDADDS)
45 diff --git a/Makefile.objs b/Makefile.objs
46 index fbfbbb7f70..f5f8dbab3b 100644
49 @@ -14,6 +14,7 @@ block-obj-y += block.o blockjob.o
51 block-obj-y += qemu-io-cmds.o
52 block-obj-$(CONFIG_REPLICATION) += replication.o
53 +block-obj-y += vma-writer.o
57 diff --git a/block/backup.c b/block/backup.c
58 index 1ede70c061..7c5febc434 100644
61 @@ -36,6 +36,7 @@ typedef struct BackupBlockJob {
62 BdrvDirtyBitmap *sync_bitmap;
63 MirrorSyncMode sync_mode;
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,
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);
78 + ret = job->dump_cb(job->common.opaque, job->target, start_sec, n, NULL);
81 + ret = blk_co_pwrite_zeroes(job->target, start * job->cluster_size,
82 + bounce_qiov.size, BDRV_REQ_MAY_UNMAP);
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);
89 + ret = job->dump_cb(job->common.opaque, job->target, start_sec, n, bounce_buffer);
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);
98 trace_backup_do_cow_write_fail(job, start, ret);
99 @@ -246,6 +258,9 @@ static void backup_abort(BlockJob *job)
100 static void backup_clean(BlockJob *job)
102 BackupBlockJob *s = container_of(job, BackupBlockJob, common);
107 blk_unref(s->target);
109 @@ -255,7 +270,9 @@ static void backup_attached_aio_context(BlockJob *job, AioContext *aio_context)
111 BackupBlockJob *s = container_of(job, BackupBlockJob, common);
113 - blk_set_aio_context(s->target, aio_context);
115 + blk_set_aio_context(s->target, aio_context);
119 void backup_do_checkpoint(BlockJob *job, Error **errp)
120 @@ -330,9 +347,11 @@ static BlockErrorAction backup_error_action(BackupBlockJob *job,
122 return block_job_error_action(&job->common, job->on_source_error,
125 + } else if (job->target) {
126 return block_job_error_action(&job->common, job->on_target_error,
129 + return BLOCK_ERROR_ACTION_REPORT;
133 @@ -557,6 +576,7 @@ BlockJob *backup_job_create(const char *job_id, BlockDriverState *bs,
134 BlockdevOnError on_source_error,
135 BlockdevOnError on_target_error,
137 + BackupDumpFunc *dump_cb,
138 BlockCompletionFunc *cb, void *opaque,
140 BlockJobTxn *txn, Error **errp)
141 @@ -567,7 +587,7 @@ BlockJob *backup_job_create(const char *job_id, BlockDriverState *bs,
146 + assert(target || dump_cb);
149 error_setg(errp, "Source and target cannot be the same");
150 @@ -580,13 +600,13 @@ BlockJob *backup_job_create(const char *job_id, BlockDriverState *bs,
154 - if (!bdrv_is_inserted(target)) {
155 + if (target && !bdrv_is_inserted(target)) {
156 error_setg(errp, "Device is not inserted: %s",
157 bdrv_get_device_name(target));
161 - if (compress && target->drv->bdrv_co_pwritev_compressed == NULL) {
162 + if (target && compress && target->drv->bdrv_co_pwritev_compressed == NULL) {
163 error_setg(errp, "Compression is not supported for this drive %s",
164 bdrv_get_device_name(target));
166 @@ -596,7 +616,7 @@ BlockJob *backup_job_create(const char *job_id, BlockDriverState *bs,
170 - if (bdrv_op_is_blocked(target, BLOCK_OP_TYPE_BACKUP_TARGET, errp)) {
171 + if (target && bdrv_op_is_blocked(target, BLOCK_OP_TYPE_BACKUP_TARGET, errp)) {
175 @@ -636,15 +656,18 @@ BlockJob *backup_job_create(const char *job_id, BlockDriverState *bs,
179 - /* The target must match the source in size, so no resize here either */
180 - job->target = blk_new(BLK_PERM_WRITE,
181 - BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE |
182 - BLK_PERM_WRITE_UNCHANGED | BLK_PERM_GRAPH_MOD);
183 - ret = blk_insert_bs(job->target, target, errp);
187 + /* The target must match the source in size, so no resize here either */
188 + job->target = blk_new(BLK_PERM_WRITE,
189 + BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE |
190 + BLK_PERM_WRITE_UNCHANGED | BLK_PERM_GRAPH_MOD);
191 + ret = blk_insert_bs(job->target, target, errp);
197 + job->dump_cb = dump_cb;
198 job->on_source_error = on_source_error;
199 job->on_target_error = on_target_error;
200 job->sync_mode = sync_mode;
201 @@ -652,38 +675,55 @@ BlockJob *backup_job_create(const char *job_id, BlockDriverState *bs,
203 job->compress = compress;
205 - /* If there is no backing file on the target, we cannot rely on COW if our
206 - * backup cluster size is smaller than the target cluster size. Even for
207 - * targets with a backing file, try to avoid COW if possible. */
208 - ret = bdrv_get_info(target, &bdi);
209 - if (ret == -ENOTSUP && !target->backing) {
210 - /* Cluster size is not defined */
211 - error_report("WARNING: The target block device doesn't provide "
212 - "information about the block size and it doesn't have a "
213 - "backing file. The default block size of %u bytes is "
214 - "used. If the actual block size of the target exceeds "
215 - "this default, the backup may be unusable",
216 - BACKUP_CLUSTER_SIZE_DEFAULT);
217 - job->cluster_size = BACKUP_CLUSTER_SIZE_DEFAULT;
218 - } else if (ret < 0 && !target->backing) {
219 - error_setg_errno(errp, -ret,
220 - "Couldn't determine the cluster size of the target image, "
221 - "which has no backing file");
222 - error_append_hint(errp,
223 - "Aborting, since this may create an unusable destination image\n");
225 - } else if (ret < 0 && target->backing) {
226 - /* Not fatal; just trudge on ahead. */
227 - job->cluster_size = BACKUP_CLUSTER_SIZE_DEFAULT;
229 + /* If there is no backing file on the target, we cannot rely on COW if our
230 + * backup cluster size is smaller than the target cluster size. Even for
231 + * targets with a backing file, try to avoid COW if possible. */
232 + ret = bdrv_get_info(target, &bdi);
233 + if (ret == -ENOTSUP && !target->backing) {
234 + /* Cluster size is not defined */
235 + error_report("WARNING: The target block device doesn't provide "
236 + "information about the block size and it doesn't have a "
237 + "backing file. The default block size of %u bytes is "
238 + "used. If the actual block size of the target exceeds "
239 + "this default, the backup may be unusable",
240 + BACKUP_CLUSTER_SIZE_DEFAULT);
241 + job->cluster_size = BACKUP_CLUSTER_SIZE_DEFAULT;
242 + } else if (ret < 0 && !target->backing) {
243 + error_setg_errno(errp, -ret,
244 + "Couldn't determine the cluster size of the target image, "
245 + "which has no backing file");
246 + error_append_hint(errp,
247 + "Aborting, since this may create an unusable destination image\n");
249 + } else if (ret < 0 && target->backing) {
250 + /* Not fatal; just trudge on ahead. */
251 + job->cluster_size = BACKUP_CLUSTER_SIZE_DEFAULT;
253 + job->cluster_size = MAX(BACKUP_CLUSTER_SIZE_DEFAULT, bdi.cluster_size);
256 - job->cluster_size = MAX(BACKUP_CLUSTER_SIZE_DEFAULT, bdi.cluster_size);
257 + ret = bdrv_get_info(bs, &bdi);
259 + job->cluster_size = BACKUP_CLUSTER_SIZE_DEFAULT;
261 + /* round down to nearest BACKUP_CLUSTER_SIZE_DEFAULT */
262 + job->cluster_size = (bdi.cluster_size / BACKUP_CLUSTER_SIZE_DEFAULT) * BACKUP_CLUSTER_SIZE_DEFAULT;
263 + if (job->cluster_size == 0) {
264 + /* but we can't go below it */
265 + job->cluster_size = BACKUP_CLUSTER_SIZE_DEFAULT;
270 - /* Required permissions are already taken with target's blk_new() */
271 - block_job_add_bdrv(&job->common, "target", target, 0, BLK_PERM_ALL,
274 + /* Required permissions are already taken with target's blk_new() */
275 + block_job_add_bdrv(&job->common, "target", target, 0, BLK_PERM_ALL,
278 + job->common.pause_count = pause_count;
280 job->common.len = len;
281 - job->common.pause_count = pause_count;
282 block_job_txn_add_job(txn, &job->common);
285 diff --git a/block/replication.c b/block/replication.c
286 index 1c41d9e6bf..60c6524417 100644
287 --- a/block/replication.c
288 +++ b/block/replication.c
289 @@ -531,6 +531,7 @@ static void replication_start(ReplicationState *rs, ReplicationMode mode,
290 0, MIRROR_SYNC_MODE_NONE, NULL, false,
291 BLOCKDEV_ON_ERROR_REPORT,
292 BLOCKDEV_ON_ERROR_REPORT, BLOCK_JOB_INTERNAL,
294 backup_job_completed, bs, 0, NULL, &local_err);
296 error_propagate(errp, local_err);
297 diff --git a/blockdev.c b/blockdev.c
298 index 534c00f5da..19a82e8774 100644
304 #include "qemu/osdep.h"
305 -#include "qemu/uuid.h"
306 #include "sysemu/block-backend.h"
307 #include "sysemu/blockdev.h"
308 #include "hw/block/block.h"
310 #include "qemu/cutils.h"
311 #include "qemu/help_option.h"
312 #include "qemu/throttle-options.h"
315 static QTAILQ_HEAD(, BlockDriverState) monitor_bdrv_states =
316 QTAILQ_HEAD_INITIALIZER(monitor_bdrv_states);
317 @@ -2934,20 +2934,44 @@ out:
318 aio_context_release(aio_context);
321 +void block_job_event_cancelled(BlockJob *job);
322 +void block_job_event_completed(BlockJob *job, const char *msg);
323 +static void block_job_cb(void *opaque, int ret)
325 + /* Note that this function may be executed from another AioContext besides
326 + * the QEMU main loop. If you need to access anything that assumes the
327 + * QEMU global mutex, use a BH or introduce a mutex.
330 + BlockDriverState *bs = opaque;
331 + const char *msg = NULL;
336 + msg = strerror(-ret);
339 + if (block_job_is_cancelled(bs->job)) {
340 + block_job_event_cancelled(bs->job);
342 + block_job_event_completed(bs->job, msg);
346 /* PVE backup related function */
348 static struct PVEBackupState {
365 @@ -2957,6 +2981,7 @@ typedef struct PVEBackupDevInfo {
366 BlockDriverState *bs;
371 char targetfile[PATH_MAX];
372 BlockDriverState *target;
373 @@ -2964,13 +2989,79 @@ typedef struct PVEBackupDevInfo {
375 static void pvebackup_run_next_job(void);
377 +static int pvebackup_dump_cb(void *opaque, BlockBackend *target,
378 + int64_t sector_num, int n_sectors,
379 + unsigned char *buf)
381 + PVEBackupDevInfo *di = opaque;
383 + int size = n_sectors * BDRV_SECTOR_SIZE;
384 + if (backup_state.cancel) {
385 + return size; // return success
388 + if (sector_num & 0x7f) {
389 + if (!backup_state.error) {
390 + error_setg(&backup_state.error,
391 + "got unaligned write inside backup dump "
392 + "callback (sector %ld)", sector_num);
394 + return -1; // not aligned to cluster size
397 + int64_t cluster_num = sector_num >> 7;
401 + if (backup_state.vmaw) {
402 + size_t zero_bytes = 0;
403 + int64_t remaining = n_sectors * BDRV_SECTOR_SIZE;
404 + while (remaining > 0) {
405 + ret = vma_writer_write(backup_state.vmaw, di->dev_id, cluster_num,
409 + buf += VMA_CLUSTER_SIZE;
412 + if (!backup_state.error) {
413 + vma_writer_error_propagate(backup_state.vmaw, &backup_state.error);
415 + if (di->bs && di->bs->job) {
416 + block_job_cancel(di->bs->job);
420 + backup_state.zero_bytes += zero_bytes;
421 + if (remaining >= VMA_CLUSTER_SIZE) {
422 + backup_state.transferred += VMA_CLUSTER_SIZE;
424 + backup_state.transferred += remaining;
426 + remaining -= VMA_CLUSTER_SIZE;
431 + backup_state.zero_bytes += size;
433 + backup_state.transferred += size;
436 + // Note: always return success, because we want that writes succeed anyways.
441 static void pvebackup_cleanup(void)
443 backup_state.end_time = time(NULL);
445 - if (backup_state.vmaobj) {
446 - object_unparent(backup_state.vmaobj);
447 - backup_state.vmaobj = NULL;
448 + if (backup_state.vmaw) {
449 + Error *local_err = NULL;
450 + vma_writer_close(backup_state.vmaw, &local_err);
451 + error_propagate(&backup_state.error, local_err);
452 + backup_state.vmaw = NULL;
455 if (backup_state.di_list) {
456 @@ -2985,6 +3076,13 @@ static void pvebackup_cleanup(void)
460 +static void coroutine_fn backup_close_vma_stream(void *opaque)
462 + PVEBackupDevInfo *di = opaque;
464 + vma_writer_close_stream(backup_state.vmaw, di->dev_id);
467 static void pvebackup_complete_cb(void *opaque, int ret)
469 PVEBackupDevInfo *di = opaque;
470 @@ -2996,14 +3094,18 @@ static void pvebackup_complete_cb(void *opaque, int ret)
471 ret, strerror(-ret));
474 + BlockDriverState *bs = di->bs;
479 - if (backup_state.vmaobj) {
480 - object_unparent(backup_state.vmaobj);
481 - backup_state.vmaobj = NULL;
482 + if (backup_state.vmaw) {
483 + Coroutine *co = qemu_coroutine_create(backup_close_vma_stream, di);
484 + qemu_coroutine_enter(co);
487 + block_job_cb(bs, ret);
489 if (!backup_state.cancel) {
490 pvebackup_run_next_job();
492 @@ -3017,14 +3119,9 @@ static void pvebackup_cancel(void *opaque)
493 error_setg(&backup_state.error, "backup cancelled");
496 - if (backup_state.vmaobj) {
498 + if (backup_state.vmaw) {
499 /* make sure vma writer does not block anymore */
500 - if (!object_set_props(backup_state.vmaobj, &err, "blocked", "yes", NULL)) {
502 - error_report_err(err);
505 + vma_writer_set_error(backup_state.vmaw, "backup cancelled");
508 GList *l = backup_state.di_list;
509 @@ -3049,19 +3146,15 @@ void qmp_backup_cancel(Error **errp)
510 Coroutine *co = qemu_coroutine_create(pvebackup_cancel, NULL);
511 qemu_coroutine_enter(co);
513 - while (backup_state.vmaobj) {
514 - /* FIXME: Find something better for this */
515 + while (backup_state.vmaw) {
516 + /* vma writer use main aio context */
517 aio_poll(qemu_get_aio_context(), true);
521 -void vma_object_add_config_file(Object *obj, const char *name,
522 - const char *contents, size_t len,
524 static int config_to_vma(const char *file, BackupFormat format,
526 - const char *backup_dir,
528 + const char *backup_dir, VmaWriter *vmaw,
533 @@ -3074,12 +3167,17 @@ static int config_to_vma(const char *file, BackupFormat format,
534 char *basename = g_path_get_basename(file);
536 if (format == BACKUP_FORMAT_VMA) {
537 - vma_object_add_config_file(vmaobj, basename, cdata, clen, errp);
538 + if (vma_writer_add_config(vmaw, basename, cdata, clen) != 0) {
539 + error_setg(errp, "unable to add %s config data to vma archive", file);
544 } else if (format == BACKUP_FORMAT_DIR) {
545 char config_path[PATH_MAX];
546 snprintf(config_path, PATH_MAX, "%s/%s", backup_dir, basename);
547 if (!g_file_set_contents(config_path, cdata, clen, &err)) {
548 - error_setg(errp, "unable to write config file '%s'", config_path);
549 + error_setg(errp, "unable to write config file '%s'", config_path);
553 @@ -3089,34 +3187,37 @@ static int config_to_vma(const char *file, BackupFormat format,
561 +bool block_job_should_pause(BlockJob *job);
562 static void pvebackup_run_next_job(void)
564 - bool cancel = backup_state.error || backup_state.cancel;
565 - GList *next = g_list_nth(backup_state.di_list, backup_state.next_job);
567 - PVEBackupDevInfo *di = (PVEBackupDevInfo *)next->data;
568 - backup_state.next_job++;
569 + GList *l = backup_state.di_list;
571 + PVEBackupDevInfo *di = (PVEBackupDevInfo *)l->data;
572 + l = g_list_next(l);
573 if (!di->completed && di->bs && di->bs->job) {
574 BlockJob *job = di->bs->job;
576 - block_job_cancel(job);
578 - block_job_resume(job);
579 + if (block_job_should_pause(job)) {
580 + bool cancel = backup_state.error || backup_state.cancel;
582 + block_job_cancel(job);
584 + block_job_resume(job);
589 - next = g_list_next(next);
595 UuidInfo *qmp_backup(const char *backup_file, bool has_format,
597 bool has_config_file, const char *config_file,
598 - bool has_firewall_file, const char *firewall_file,
599 + bool has_firewall_file, const char *firewall_file,
600 bool has_devlist, const char *devlist,
601 bool has_speed, int64_t speed, Error **errp)
603 @@ -3124,14 +3225,15 @@ UuidInfo *qmp_backup(const char *backup_file, bool has_format,
604 BlockDriverState *bs = NULL;
605 const char *backup_dir = NULL;
606 Error *local_err = NULL;
609 + VmaWriter *vmaw = NULL;
611 GList *di_list = NULL;
616 - if (backup_state.di_list || backup_state.vmaobj) {
617 + if (backup_state.di_list) {
618 error_set(errp, ERROR_CLASS_GENERIC_ERROR,
619 "previous backup not finished");
621 @@ -3206,40 +3308,28 @@ UuidInfo *qmp_backup(const char *backup_file, bool has_format,
625 - qemu_uuid_generate(&uuid);
626 + uuid_generate(uuid);
628 if (format == BACKUP_FORMAT_VMA) {
629 - char uuidstr[UUID_FMT_LEN+1];
630 - qemu_uuid_unparse(&uuid, uuidstr);
631 - uuidstr[UUID_FMT_LEN] = 0;
632 - backup_state.vmaobj =
633 - object_new_with_props("vma", object_get_objects_root(),
634 - "vma-backup-obj", &local_err,
635 - "filename", backup_file,
638 - if (!backup_state.vmaobj) {
639 + vmaw = vma_writer_create(backup_file, uuid, &local_err);
642 error_propagate(errp, local_err);
647 + /* register all devices for vma writer */
650 - QDict *options = qdict_new();
652 PVEBackupDevInfo *di = (PVEBackupDevInfo *)l->data;
655 const char *devname = bdrv_get_device_name(di->bs);
656 - snprintf(di->targetfile, PATH_MAX, "vma-backup-obj/%s.raw", devname);
658 - qdict_put(options, "driver", qstring_from_str("vma-drive"));
659 - qdict_put(options, "size", qint_from_int(di->size));
660 - di->target = bdrv_open(di->targetfile, NULL, options, BDRV_O_RDWR, &local_err);
662 - error_propagate(errp, local_err);
663 + di->dev_id = vma_writer_register_stream(vmaw, devname, di->size);
664 + if (di->dev_id <= 0) {
665 + error_set(errp, ERROR_CLASS_GENERIC_ERROR,
666 + "register_stream failed");
670 @@ -3280,15 +3370,15 @@ UuidInfo *qmp_backup(const char *backup_file, bool has_format,
672 /* add configuration file to archive */
673 if (has_config_file) {
674 - if(config_to_vma(config_file, format, backup_state.vmaobj, backup_dir, errp) != 0) {
676 + if(config_to_vma(config_file, format, backup_dir, vmaw, errp) != 0) {
681 /* add firewall file to archive */
682 if (has_firewall_file) {
683 - if(config_to_vma(firewall_file, format, backup_state.vmaobj, backup_dir, errp) != 0) {
685 + if(config_to_vma(firewall_file, format, backup_dir, vmaw, errp) != 0) {
689 /* initialize global backup_state now */
690 @@ -3310,11 +3400,12 @@ UuidInfo *qmp_backup(const char *backup_file, bool has_format,
692 backup_state.backup_file = g_strdup(backup_file);
694 - memcpy(&backup_state.uuid, &uuid, sizeof(uuid));
695 - qemu_uuid_unparse(&uuid, backup_state.uuid_str);
696 + backup_state.vmaw = vmaw;
698 + uuid_copy(backup_state.uuid, uuid);
699 + uuid_unparse_lower(uuid, backup_state.uuid_str);
701 backup_state.di_list = di_list;
702 - backup_state.next_job = 0;
704 backup_state.total = total;
705 backup_state.transferred = 0;
706 @@ -3325,21 +3416,16 @@ UuidInfo *qmp_backup(const char *backup_file, bool has_format,
708 PVEBackupDevInfo *di = (PVEBackupDevInfo *)l->data;
711 job = backup_job_create(NULL, di->bs, di->target, speed, MIRROR_SYNC_MODE_FULL, NULL,
712 false, BLOCKDEV_ON_ERROR_REPORT, BLOCKDEV_ON_ERROR_REPORT,
714 - pvebackup_complete_cb, di, 2, NULL, &local_err);
716 - bdrv_unref(di->target);
719 + pvebackup_dump_cb, pvebackup_complete_cb, di,
720 + 2, NULL, &local_err);
721 if (!job || local_err != NULL) {
722 error_setg(&backup_state.error, "backup_job_create failed");
723 pvebackup_cancel(NULL);
725 - block_job_start(job);
727 + block_job_start(job);
730 if (!backup_state.error) {
731 @@ -3373,9 +3459,10 @@ err:
735 - if (backup_state.vmaobj) {
736 - object_unparent(backup_state.vmaobj);
737 - backup_state.vmaobj = NULL;
740 + vma_writer_close(vmaw, &err);
741 + unlink(backup_file);
745 @@ -3760,7 +3847,7 @@ static BlockJob *do_drive_backup(DriveBackup *backup, BlockJobTxn *txn,
746 job = backup_job_create(backup->job_id, bs, target_bs, backup->speed,
747 backup->sync, bmap, backup->compress,
748 backup->on_source_error, backup->on_target_error,
749 - BLOCK_JOB_DEFAULT, NULL, NULL, 0, txn, &local_err);
750 + BLOCK_JOB_DEFAULT, NULL, NULL, NULL, 0, txn, &local_err);
751 bdrv_unref(target_bs);
752 if (local_err != NULL) {
753 error_propagate(errp, local_err);
754 @@ -3839,7 +3926,7 @@ BlockJob *do_blockdev_backup(BlockdevBackup *backup, BlockJobTxn *txn,
755 job = backup_job_create(backup->job_id, bs, target_bs, backup->speed,
756 backup->sync, NULL, backup->compress,
757 backup->on_source_error, backup->on_target_error,
758 - BLOCK_JOB_DEFAULT, NULL, NULL, 0, txn, &local_err);
759 + BLOCK_JOB_DEFAULT, NULL, NULL, NULL, 0, txn, &local_err);
760 if (local_err != NULL) {
761 error_propagate(errp, local_err);
763 diff --git a/blockjob.c b/blockjob.c
764 index 764d41863e..cb3741f6dd 100644
768 #include "qemu/timer.h"
769 #include "qapi-event.h"
771 -static void block_job_event_cancelled(BlockJob *job);
772 -static void block_job_event_completed(BlockJob *job, const char *msg);
773 +void block_job_event_cancelled(BlockJob *job);
774 +void block_job_event_completed(BlockJob *job, const char *msg);
776 /* Transactional group of block jobs */
778 @@ -473,7 +473,8 @@ void block_job_user_pause(BlockJob *job)
779 block_job_pause(job);
782 -static bool block_job_should_pause(BlockJob *job)
783 +bool block_job_should_pause(BlockJob *job);
784 +bool block_job_should_pause(BlockJob *job)
786 return job->pause_count > 0;
788 @@ -687,7 +688,7 @@ static void block_job_iostatus_set_err(BlockJob *job, int error)
792 -static void block_job_event_cancelled(BlockJob *job)
793 +void block_job_event_cancelled(BlockJob *job)
795 if (block_job_is_internal(job)) {
797 @@ -701,7 +702,7 @@ static void block_job_event_cancelled(BlockJob *job)
801 -static void block_job_event_completed(BlockJob *job, const char *msg)
802 +void block_job_event_completed(BlockJob *job, const char *msg)
804 if (block_job_is_internal(job)) {
806 diff --git a/include/block/block_int.h b/include/block/block_int.h
807 index 19b84b027f..fc1c53a059 100644
808 --- a/include/block/block_int.h
809 +++ b/include/block/block_int.h
812 #define BLOCK_PROBE_BUF_SIZE 512
814 +typedef int BackupDumpFunc(void *opaque, BlockBackend *be,
815 + int64_t sector_num, int n_sectors, unsigned char *buf);
817 enum BdrvTrackedRequestType {
820 @@ -878,6 +881,7 @@ BlockJob *backup_job_create(const char *job_id, BlockDriverState *bs,
821 BlockdevOnError on_source_error,
822 BlockdevOnError on_target_error,
824 + BackupDumpFunc *dump_cb,
825 BlockCompletionFunc *cb, void *opaque,
827 BlockJobTxn *txn, Error **errp);
828 diff --git a/vma-reader.c b/vma-reader.c
830 index 0000000000..2000889bd3
835 + * VMA: Virtual Machine Archive
837 + * Copyright (C) 2012 Proxmox Server Solutions
840 + * Dietmar Maurer (dietmar@proxmox.com)
842 + * This work is licensed under the terms of the GNU GPL, version 2 or later.
843 + * See the COPYING file in the top-level directory.
847 +#include "qemu/osdep.h"
849 +#include <uuid/uuid.h>
851 +#include "qemu-common.h"
852 +#include "qemu/timer.h"
853 +#include "qemu/ratelimit.h"
855 +#include "block/block.h"
856 +#include "sysemu/block-backend.h"
858 +static unsigned char zero_vma_block[VMA_BLOCK_SIZE];
860 +typedef struct VmaRestoreState {
861 + BlockBackend *target;
863 + unsigned long *bitmap;
869 + GChecksum *md5csum;
870 + GHashTable *blob_hash;
871 + unsigned char *head_data;
872 + VmaDeviceInfo devinfo[256];
873 + VmaRestoreState rstate[256];
875 + guint8 vmstate_stream;
876 + uint32_t vmstate_clusters;
877 + /* to show restore percentage if run with -v */
879 + int64_t cluster_count;
880 + int64_t clusters_read;
881 + int64_t zero_cluster_data;
882 + int64_t partial_zero_cluster_data;
883 + int clusters_read_per;
887 +g_int32_hash(gconstpointer v)
889 + return *(const uint32_t *)v;
893 +g_int32_equal(gconstpointer v1, gconstpointer v2)
895 + return *((const uint32_t *)v1) == *((const uint32_t *)v2);
898 +static int vma_reader_get_bitmap(VmaRestoreState *rstate, int64_t cluster_num)
901 + assert(rstate->bitmap);
903 + unsigned long val, idx, bit;
905 + idx = cluster_num / BITS_PER_LONG;
907 + assert(rstate->bitmap_size > idx);
909 + bit = cluster_num % BITS_PER_LONG;
910 + val = rstate->bitmap[idx];
912 + return !!(val & (1UL << bit));
915 +static void vma_reader_set_bitmap(VmaRestoreState *rstate, int64_t cluster_num,
919 + assert(rstate->bitmap);
921 + unsigned long val, idx, bit;
923 + idx = cluster_num / BITS_PER_LONG;
925 + assert(rstate->bitmap_size > idx);
927 + bit = cluster_num % BITS_PER_LONG;
928 + val = rstate->bitmap[idx];
930 + if (!(val & (1UL << bit))) {
934 + if (val & (1UL << bit)) {
935 + val &= ~(1UL << bit);
938 + rstate->bitmap[idx] = val;
941 +typedef struct VmaBlob {
947 +static const VmaBlob *get_header_blob(VmaReader *vmar, uint32_t pos)
950 + assert(vmar->blob_hash);
952 + return g_hash_table_lookup(vmar->blob_hash, &pos);
955 +static const char *get_header_str(VmaReader *vmar, uint32_t pos)
957 + const VmaBlob *blob = get_header_blob(vmar, pos);
961 + const char *res = (char *)blob->data;
962 + if (res[blob->len-1] != '\0') {
969 +safe_read(int fd, unsigned char *buf, size_t count)
974 + n = read(fd, buf, count);
975 + } while (n < 0 && errno == EINTR);
981 +full_read(int fd, unsigned char *buf, size_t len)
989 + n = safe_read(fd, buf, len);
1011 +void vma_reader_destroy(VmaReader *vmar)
1015 + if (vmar->fd >= 0) {
1019 + if (vmar->cdata_list) {
1020 + g_list_free(vmar->cdata_list);
1024 + for (i = 1; i < 256; i++) {
1025 + if (vmar->rstate[i].bitmap) {
1026 + g_free(vmar->rstate[i].bitmap);
1030 + if (vmar->md5csum) {
1031 + g_checksum_free(vmar->md5csum);
1034 + if (vmar->blob_hash) {
1035 + g_hash_table_destroy(vmar->blob_hash);
1038 + if (vmar->head_data) {
1039 + g_free(vmar->head_data);
1046 +static int vma_reader_read_head(VmaReader *vmar, Error **errp)
1050 + assert(*errp == NULL);
1052 + unsigned char md5sum[16];
1056 + vmar->head_data = g_malloc(sizeof(VmaHeader));
1058 + if (full_read(vmar->fd, vmar->head_data, sizeof(VmaHeader)) !=
1059 + sizeof(VmaHeader)) {
1060 + error_setg(errp, "can't read vma header - %s",
1061 + errno ? g_strerror(errno) : "got EOF");
1065 + VmaHeader *h = (VmaHeader *)vmar->head_data;
1067 + if (h->magic != VMA_MAGIC) {
1068 + error_setg(errp, "not a vma file - wrong magic number");
1072 + uint32_t header_size = GUINT32_FROM_BE(h->header_size);
1073 + int need = header_size - sizeof(VmaHeader);
1075 + error_setg(errp, "wrong vma header size %d", header_size);
1079 + vmar->head_data = g_realloc(vmar->head_data, header_size);
1080 + h = (VmaHeader *)vmar->head_data;
1082 + if (full_read(vmar->fd, vmar->head_data + sizeof(VmaHeader), need) !=
1084 + error_setg(errp, "can't read vma header data - %s",
1085 + errno ? g_strerror(errno) : "got EOF");
1089 + memcpy(md5sum, h->md5sum, 16);
1090 + memset(h->md5sum, 0, 16);
1092 + g_checksum_reset(vmar->md5csum);
1093 + g_checksum_update(vmar->md5csum, vmar->head_data, header_size);
1095 + g_checksum_get_digest(vmar->md5csum, (guint8 *)(h->md5sum), &csize);
1097 + if (memcmp(md5sum, h->md5sum, 16) != 0) {
1098 + error_setg(errp, "wrong vma header chechsum");
1102 + /* we can modify header data after checksum verify */
1103 + h->header_size = header_size;
1105 + h->version = GUINT32_FROM_BE(h->version);
1106 + if (h->version != 1) {
1107 + error_setg(errp, "wrong vma version %d", h->version);
1111 + h->ctime = GUINT64_FROM_BE(h->ctime);
1112 + h->blob_buffer_offset = GUINT32_FROM_BE(h->blob_buffer_offset);
1113 + h->blob_buffer_size = GUINT32_FROM_BE(h->blob_buffer_size);
1115 + uint32_t bstart = h->blob_buffer_offset + 1;
1116 + uint32_t bend = h->blob_buffer_offset + h->blob_buffer_size;
1118 + if (bstart <= sizeof(VmaHeader)) {
1119 + error_setg(errp, "wrong vma blob buffer offset %d",
1120 + h->blob_buffer_offset);
1124 + if (bend > header_size) {
1125 + error_setg(errp, "wrong vma blob buffer size %d/%d",
1126 + h->blob_buffer_offset, h->blob_buffer_size);
1130 + while ((bstart + 2) <= bend) {
1131 + uint32_t size = vmar->head_data[bstart] +
1132 + (vmar->head_data[bstart+1] << 8);
1133 + if ((bstart + size + 2) <= bend) {
1134 + VmaBlob *blob = g_new0(VmaBlob, 1);
1135 + blob->start = bstart - h->blob_buffer_offset;
1137 + blob->data = vmar->head_data + bstart + 2;
1138 + g_hash_table_insert(vmar->blob_hash, &blob->start, blob);
1140 + bstart += size + 2;
1145 + for (i = 1; i < 256; i++) {
1146 + VmaDeviceInfoHeader *dih = &h->dev_info[i];
1147 + uint32_t devname_ptr = GUINT32_FROM_BE(dih->devname_ptr);
1148 + uint64_t size = GUINT64_FROM_BE(dih->size);
1149 + const char *devname = get_header_str(vmar, devname_ptr);
1151 + if (size && devname) {
1153 + vmar->devinfo[i].size = size;
1154 + vmar->devinfo[i].devname = devname;
1156 + if (strcmp(devname, "vmstate") == 0) {
1157 + vmar->vmstate_stream = i;
1162 + for (i = 0; i < VMA_MAX_CONFIGS; i++) {
1163 + uint32_t name_ptr = GUINT32_FROM_BE(h->config_names[i]);
1164 + uint32_t data_ptr = GUINT32_FROM_BE(h->config_data[i]);
1166 + if (!(name_ptr && data_ptr)) {
1169 + const char *name = get_header_str(vmar, name_ptr);
1170 + const VmaBlob *blob = get_header_blob(vmar, data_ptr);
1172 + if (!(name && blob)) {
1173 + error_setg(errp, "vma contains invalid data pointers");
1177 + VmaConfigData *cdata = g_new0(VmaConfigData, 1);
1178 + cdata->name = name;
1179 + cdata->data = blob->data;
1180 + cdata->len = blob->len;
1182 + vmar->cdata_list = g_list_append(vmar->cdata_list, cdata);
1188 +VmaReader *vma_reader_create(const char *filename, Error **errp)
1193 + VmaReader *vmar = g_new0(VmaReader, 1);
1195 + if (strcmp(filename, "-") == 0) {
1196 + vmar->fd = dup(0);
1198 + vmar->fd = open(filename, O_RDONLY);
1201 + if (vmar->fd < 0) {
1202 + error_setg(errp, "can't open file %s - %s\n", filename,
1203 + g_strerror(errno));
1207 + vmar->md5csum = g_checksum_new(G_CHECKSUM_MD5);
1208 + if (!vmar->md5csum) {
1209 + error_setg(errp, "can't allocate cmsum\n");
1213 + vmar->blob_hash = g_hash_table_new_full(g_int32_hash, g_int32_equal,
1216 + if (vma_reader_read_head(vmar, errp) < 0) {
1224 + vma_reader_destroy(vmar);
1230 +VmaHeader *vma_reader_get_header(VmaReader *vmar)
1233 + assert(vmar->head_data);
1235 + return (VmaHeader *)(vmar->head_data);
1238 +GList *vma_reader_get_config_data(VmaReader *vmar)
1241 + assert(vmar->head_data);
1243 + return vmar->cdata_list;
1246 +VmaDeviceInfo *vma_reader_get_device_info(VmaReader *vmar, guint8 dev_id)
1251 + if (vmar->devinfo[dev_id].size && vmar->devinfo[dev_id].devname) {
1252 + return &vmar->devinfo[dev_id];
1258 +static void allocate_rstate(VmaReader *vmar, guint8 dev_id,
1259 + BlockBackend *target, bool write_zeroes)
1264 + vmar->rstate[dev_id].target = target;
1265 + vmar->rstate[dev_id].write_zeroes = write_zeroes;
1267 + int64_t size = vmar->devinfo[dev_id].size;
1269 + int64_t bitmap_size = (size/BDRV_SECTOR_SIZE) +
1270 + (VMA_CLUSTER_SIZE/BDRV_SECTOR_SIZE) * BITS_PER_LONG - 1;
1271 + bitmap_size /= (VMA_CLUSTER_SIZE/BDRV_SECTOR_SIZE) * BITS_PER_LONG;
1273 + vmar->rstate[dev_id].bitmap_size = bitmap_size;
1274 + vmar->rstate[dev_id].bitmap = g_new0(unsigned long, bitmap_size);
1276 + vmar->cluster_count += size/VMA_CLUSTER_SIZE;
1279 +int vma_reader_register_bs(VmaReader *vmar, guint8 dev_id, BlockBackend *target,
1280 + bool write_zeroes, Error **errp)
1283 + assert(target != NULL);
1285 + assert(vmar->rstate[dev_id].target == NULL);
1287 + int64_t size = blk_getlength(target);
1288 + int64_t size_diff = size - vmar->devinfo[dev_id].size;
1290 + /* storage types can have different size restrictions, so it
1291 + * is not always possible to create an image with exact size.
1292 + * So we tolerate a size difference up to 4MB.
1294 + if ((size_diff < 0) || (size_diff > 4*1024*1024)) {
1295 + error_setg(errp, "vma_reader_register_bs for stream %s failed - "
1296 + "unexpected size %zd != %zd", vmar->devinfo[dev_id].devname,
1297 + size, vmar->devinfo[dev_id].size);
1301 + allocate_rstate(vmar, dev_id, target, write_zeroes);
1306 +static ssize_t safe_write(int fd, void *buf, size_t count)
1311 + n = write(fd, buf, count);
1312 + } while (n < 0 && errno == EINTR);
1317 +static size_t full_write(int fd, void *buf, size_t len)
1325 + n = safe_write(fd, buf, len);
1335 + /* incomplete write ? */
1342 +static int restore_write_data(VmaReader *vmar, guint8 dev_id,
1343 + BlockBackend *target, int vmstate_fd,
1344 + unsigned char *buf, int64_t sector_num,
1345 + int nb_sectors, Error **errp)
1349 + if (dev_id == vmar->vmstate_stream) {
1350 + if (vmstate_fd >= 0) {
1351 + int len = nb_sectors * BDRV_SECTOR_SIZE;
1352 + int res = full_write(vmstate_fd, buf, len);
1354 + error_setg(errp, "write vmstate failed %d", res);
1359 + int res = blk_pwrite(target, sector_num * BDRV_SECTOR_SIZE, buf, nb_sectors * BDRV_SECTOR_SIZE, 0);
1361 + error_setg(errp, "blk_pwrite to %s failed (%d)",
1362 + bdrv_get_device_name(blk_bs(target)), res);
1369 +static int restore_extent(VmaReader *vmar, unsigned char *buf,
1370 + int extent_size, int vmstate_fd,
1371 + bool verbose, bool verify, Error **errp)
1376 + VmaExtentHeader *ehead = (VmaExtentHeader *)buf;
1377 + int start = VMA_EXTENT_HEADER_SIZE;
1380 + for (i = 0; i < VMA_BLOCKS_PER_EXTENT; i++) {
1381 + uint64_t block_info = GUINT64_FROM_BE(ehead->blockinfo[i]);
1382 + uint64_t cluster_num = block_info & 0xffffffff;
1383 + uint8_t dev_id = (block_info >> 32) & 0xff;
1384 + uint16_t mask = block_info >> (32+16);
1385 + int64_t max_sector;
1391 + VmaRestoreState *rstate = &vmar->rstate[dev_id];
1392 + BlockBackend *target = NULL;
1394 + if (dev_id != vmar->vmstate_stream) {
1395 + target = rstate->target;
1396 + if (!verify && !target) {
1397 + error_setg(errp, "got wrong dev id %d", dev_id);
1401 + if (vma_reader_get_bitmap(rstate, cluster_num)) {
1402 + error_setg(errp, "found duplicated cluster %zd for stream %s",
1403 + cluster_num, vmar->devinfo[dev_id].devname);
1406 + vma_reader_set_bitmap(rstate, cluster_num, 1);
1408 + max_sector = vmar->devinfo[dev_id].size/BDRV_SECTOR_SIZE;
1410 + max_sector = G_MAXINT64;
1411 + if (cluster_num != vmar->vmstate_clusters) {
1412 + error_setg(errp, "found out of order vmstate data");
1415 + vmar->vmstate_clusters++;
1418 + vmar->clusters_read++;
1421 + time_t duration = time(NULL) - vmar->start_time;
1422 + int percent = (vmar->clusters_read*100)/vmar->cluster_count;
1423 + if (percent != vmar->clusters_read_per) {
1424 + printf("progress %d%% (read %zd bytes, duration %zd sec)\n",
1425 + percent, vmar->clusters_read*VMA_CLUSTER_SIZE,
1428 + vmar->clusters_read_per = percent;
1432 + /* try to write whole clusters to speedup restore */
1433 + if (mask == 0xffff) {
1434 + if ((start + VMA_CLUSTER_SIZE) > extent_size) {
1435 + error_setg(errp, "short vma extent - too many blocks");
1438 + int64_t sector_num = (cluster_num * VMA_CLUSTER_SIZE) /
1440 + int64_t end_sector = sector_num +
1441 + VMA_CLUSTER_SIZE/BDRV_SECTOR_SIZE;
1443 + if (end_sector > max_sector) {
1444 + end_sector = max_sector;
1447 + if (end_sector <= sector_num) {
1448 + error_setg(errp, "got wrong block address - write bejond end");
1453 + int nb_sectors = end_sector - sector_num;
1454 + if (restore_write_data(vmar, dev_id, target, vmstate_fd,
1455 + buf + start, sector_num, nb_sectors,
1461 + start += VMA_CLUSTER_SIZE;
1466 + for (j = 0; j < 16; j++) {
1467 + int64_t sector_num = (cluster_num*VMA_CLUSTER_SIZE +
1468 + j*VMA_BLOCK_SIZE)/BDRV_SECTOR_SIZE;
1470 + int64_t end_sector = sector_num +
1471 + VMA_BLOCK_SIZE/BDRV_SECTOR_SIZE;
1472 + if (end_sector > max_sector) {
1473 + end_sector = max_sector;
1477 + if ((start + VMA_BLOCK_SIZE) > extent_size) {
1478 + error_setg(errp, "short vma extent - too many blocks");
1482 + if (end_sector <= sector_num) {
1483 + error_setg(errp, "got wrong block address - "
1484 + "write bejond end");
1489 + int nb_sectors = end_sector - sector_num;
1490 + if (restore_write_data(vmar, dev_id, target, vmstate_fd,
1491 + buf + start, sector_num,
1492 + nb_sectors, errp) < 0) {
1497 + start += VMA_BLOCK_SIZE;
1502 + if (end_sector > sector_num) {
1503 + /* Todo: use bdrv_co_write_zeroes (but that need to
1504 + * be run inside coroutine?)
1506 + int nb_sectors = end_sector - sector_num;
1507 + int zero_size = BDRV_SECTOR_SIZE*nb_sectors;
1508 + vmar->zero_cluster_data += zero_size;
1510 + vmar->partial_zero_cluster_data += zero_size;
1513 + if (rstate->write_zeroes && !verify) {
1514 + if (restore_write_data(vmar, dev_id, target, vmstate_fd,
1515 + zero_vma_block, sector_num,
1516 + nb_sectors, errp) < 0) {
1528 + if (start != extent_size) {
1529 + error_setg(errp, "vma extent error - missing blocks");
1536 +static int vma_reader_restore_full(VmaReader *vmar, int vmstate_fd,
1537 + bool verbose, bool verify,
1541 + assert(vmar->head_data);
1544 + unsigned char buf[VMA_MAX_EXTENT_SIZE];
1546 + unsigned char md5sum[16];
1547 + VmaHeader *h = (VmaHeader *)vmar->head_data;
1549 + vmar->start_time = time(NULL);
1552 + int bytes = full_read(vmar->fd, buf + buf_pos, sizeof(buf) - buf_pos);
1554 + error_setg(errp, "read failed - %s", g_strerror(errno));
1564 + if (buf_pos < VMA_EXTENT_HEADER_SIZE) {
1565 + error_setg(errp, "read short extent (%d bytes)", buf_pos);
1569 + VmaExtentHeader *ehead = (VmaExtentHeader *)buf;
1571 + /* extract md5sum */
1572 + memcpy(md5sum, ehead->md5sum, sizeof(ehead->md5sum));
1573 + memset(ehead->md5sum, 0, sizeof(ehead->md5sum));
1575 + g_checksum_reset(vmar->md5csum);
1576 + g_checksum_update(vmar->md5csum, buf, VMA_EXTENT_HEADER_SIZE);
1578 + g_checksum_get_digest(vmar->md5csum, ehead->md5sum, &csize);
1580 + if (memcmp(md5sum, ehead->md5sum, 16) != 0) {
1581 + error_setg(errp, "wrong vma extent header chechsum");
1585 + if (memcmp(h->uuid, ehead->uuid, sizeof(ehead->uuid)) != 0) {
1586 + error_setg(errp, "wrong vma extent uuid");
1590 + if (ehead->magic != VMA_EXTENT_MAGIC || ehead->reserved1 != 0) {
1591 + error_setg(errp, "wrong vma extent header magic");
1595 + int block_count = GUINT16_FROM_BE(ehead->block_count);
1596 + int extent_size = VMA_EXTENT_HEADER_SIZE + block_count*VMA_BLOCK_SIZE;
1598 + if (buf_pos < extent_size) {
1599 + error_setg(errp, "short vma extent (%d < %d)", buf_pos,
1604 + if (restore_extent(vmar, buf, extent_size, vmstate_fd, verbose,
1605 + verify, errp) < 0) {
1609 + if (buf_pos > extent_size) {
1610 + memmove(buf, buf + extent_size, buf_pos - extent_size);
1611 + buf_pos = buf_pos - extent_size;
1620 + for (i = 1; i < 256; i++) {
1621 + VmaRestoreState *rstate = &vmar->rstate[i];
1622 + if (!rstate->target) {
1626 + if (blk_flush(rstate->target) < 0) {
1627 + error_setg(errp, "vma blk_flush %s failed",
1628 + vmar->devinfo[i].devname);
1632 + if (vmar->devinfo[i].size &&
1633 + (strcmp(vmar->devinfo[i].devname, "vmstate") != 0)) {
1634 + assert(rstate->bitmap);
1636 + int64_t cluster_num, end;
1638 + end = (vmar->devinfo[i].size + VMA_CLUSTER_SIZE - 1) /
1641 + for (cluster_num = 0; cluster_num < end; cluster_num++) {
1642 + if (!vma_reader_get_bitmap(rstate, cluster_num)) {
1643 + error_setg(errp, "detected missing cluster %zd "
1644 + "for stream %s", cluster_num,
1645 + vmar->devinfo[i].devname);
1653 + if (vmar->clusters_read) {
1654 + printf("total bytes read %zd, sparse bytes %zd (%.3g%%)\n",
1655 + vmar->clusters_read*VMA_CLUSTER_SIZE,
1656 + vmar->zero_cluster_data,
1657 + (double)(100.0*vmar->zero_cluster_data)/
1658 + (vmar->clusters_read*VMA_CLUSTER_SIZE));
1660 + int64_t datasize = vmar->clusters_read*VMA_CLUSTER_SIZE-vmar->zero_cluster_data;
1661 + if (datasize) { // this does not make sense for empty files
1662 + printf("space reduction due to 4K zero blocks %.3g%%\n",
1663 + (double)(100.0*vmar->partial_zero_cluster_data) / datasize);
1666 + printf("vma archive contains no image data\n");
1672 +int vma_reader_restore(VmaReader *vmar, int vmstate_fd, bool verbose,
1675 + return vma_reader_restore_full(vmar, vmstate_fd, verbose, false, errp);
1678 +int vma_reader_verify(VmaReader *vmar, bool verbose, Error **errp)
1682 + for (dev_id = 1; dev_id < 255; dev_id++) {
1683 + if (vma_reader_get_device_info(vmar, dev_id)) {
1684 + allocate_rstate(vmar, dev_id, NULL, false);
1688 + return vma_reader_restore_full(vmar, -1, verbose, true, errp);
1691 diff --git a/vma-writer.c b/vma-writer.c
1692 new file mode 100644
1693 index 0000000000..9001cbdd2b
1698 + * VMA: Virtual Machine Archive
1700 + * Copyright (C) 2012 Proxmox Server Solutions
1703 + * Dietmar Maurer (dietmar@proxmox.com)
1705 + * This work is licensed under the terms of the GNU GPL, version 2 or later.
1706 + * See the COPYING file in the top-level directory.
1710 +#include "qemu/osdep.h"
1712 +#include <uuid/uuid.h>
1715 +#include "block/block.h"
1716 +#include "monitor/monitor.h"
1717 +#include "qemu/main-loop.h"
1718 +#include "qemu/coroutine.h"
1719 +#include "qemu/cutils.h"
1721 +#define DEBUG_VMA 0
1723 +#define DPRINTF(fmt, ...)\
1724 + do { if (DEBUG_VMA) { printf("vma: " fmt, ## __VA_ARGS__); } } while (0)
1726 +#define WRITE_BUFFERS 5
1727 +#define HEADER_CLUSTERS 8
1728 +#define HEADERBUF_SIZE (VMA_CLUSTER_SIZE*HEADER_CLUSTERS)
1734 + char errmsg[8192];
1736 + bool header_written;
1739 + /* we always write extents */
1740 + unsigned char *outbuf;
1741 + int outbuf_pos; /* in bytes */
1742 + int outbuf_count; /* in VMA_BLOCKS */
1743 + uint64_t outbuf_block_info[VMA_BLOCKS_PER_EXTENT];
1745 + unsigned char *headerbuf;
1747 + GChecksum *md5csum;
1748 + CoMutex flush_lock;
1749 + Coroutine *co_writer;
1751 + /* drive informations */
1752 + VmaStreamInfo stream_info[256];
1753 + guint stream_count;
1755 + guint8 vmstate_stream;
1756 + uint32_t vmstate_clusters;
1758 + /* header blob table */
1759 + char *header_blob_table;
1760 + uint32_t header_blob_table_size;
1761 + uint32_t header_blob_table_pos;
1763 + /* store for config blobs */
1764 + uint32_t config_names[VMA_MAX_CONFIGS]; /* offset into blob_buffer table */
1765 + uint32_t config_data[VMA_MAX_CONFIGS]; /* offset into blob_buffer table */
1766 + uint32_t config_count;
1769 +void vma_writer_set_error(VmaWriter *vmaw, const char *fmt, ...)
1773 + if (vmaw->status < 0) {
1777 + vmaw->status = -1;
1779 + va_start(ap, fmt);
1780 + g_vsnprintf(vmaw->errmsg, sizeof(vmaw->errmsg), fmt, ap);
1783 + DPRINTF("vma_writer_set_error: %s\n", vmaw->errmsg);
1786 +static uint32_t allocate_header_blob(VmaWriter *vmaw, const char *data,
1789 + if (len > 65535) {
1793 + if (!vmaw->header_blob_table ||
1794 + (vmaw->header_blob_table_size <
1795 + (vmaw->header_blob_table_pos + len + 2))) {
1796 + int newsize = vmaw->header_blob_table_size + ((len + 2 + 511)/512)*512;
1798 + vmaw->header_blob_table = g_realloc(vmaw->header_blob_table, newsize);
1799 + memset(vmaw->header_blob_table + vmaw->header_blob_table_size,
1800 + 0, newsize - vmaw->header_blob_table_size);
1801 + vmaw->header_blob_table_size = newsize;
1804 + uint32_t cpos = vmaw->header_blob_table_pos;
1805 + vmaw->header_blob_table[cpos] = len & 255;
1806 + vmaw->header_blob_table[cpos+1] = (len >> 8) & 255;
1807 + memcpy(vmaw->header_blob_table + cpos + 2, data, len);
1808 + vmaw->header_blob_table_pos += len + 2;
1812 +static uint32_t allocate_header_string(VmaWriter *vmaw, const char *str)
1816 + size_t len = strlen(str) + 1;
1818 + return allocate_header_blob(vmaw, str, len);
1821 +int vma_writer_add_config(VmaWriter *vmaw, const char *name, gpointer data,
1825 + assert(!vmaw->header_written);
1826 + assert(vmaw->config_count < VMA_MAX_CONFIGS);
1830 + gchar *basename = g_path_get_basename(name);
1831 + uint32_t name_ptr = allocate_header_string(vmaw, basename);
1838 + uint32_t data_ptr = allocate_header_blob(vmaw, data, len);
1843 + vmaw->config_names[vmaw->config_count] = name_ptr;
1844 + vmaw->config_data[vmaw->config_count] = data_ptr;
1846 + vmaw->config_count++;
1851 +int vma_writer_register_stream(VmaWriter *vmaw, const char *devname,
1856 + assert(!vmaw->status);
1858 + if (vmaw->header_written) {
1859 + vma_writer_set_error(vmaw, "vma_writer_register_stream: header "
1860 + "already written");
1864 + guint n = vmaw->stream_count + 1;
1866 + /* we can have dev_ids form 1 to 255 (0 reserved)
1867 + * 255(-1) reseverd for safety
1870 + vma_writer_set_error(vmaw, "vma_writer_register_stream: "
1871 + "too many drives");
1876 + vma_writer_set_error(vmaw, "vma_writer_register_stream: "
1877 + "got strange size %zd", size);
1881 + DPRINTF("vma_writer_register_stream %s %zu %d\n", devname, size, n);
1883 + vmaw->stream_info[n].devname = g_strdup(devname);
1884 + vmaw->stream_info[n].size = size;
1886 + vmaw->stream_info[n].cluster_count = (size + VMA_CLUSTER_SIZE - 1) /
1889 + vmaw->stream_count = n;
1891 + if (strcmp(devname, "vmstate") == 0) {
1892 + vmaw->vmstate_stream = n;
1898 +static void vma_co_continue_write(void *opaque)
1900 + VmaWriter *vmaw = opaque;
1902 + DPRINTF("vma_co_continue_write\n");
1903 + qemu_coroutine_enter(vmaw->co_writer);
1906 +static ssize_t coroutine_fn
1907 +vma_queue_write(VmaWriter *vmaw, const void *buf, size_t bytes)
1909 + DPRINTF("vma_queue_write enter %zd\n", bytes);
1913 + assert(bytes <= VMA_MAX_EXTENT_SIZE);
1918 + assert(vmaw->co_writer == NULL);
1920 + vmaw->co_writer = qemu_coroutine_self();
1922 + while (done < bytes) {
1923 + aio_set_fd_handler(qemu_get_aio_context(), vmaw->fd, false, NULL, vma_co_continue_write, NULL, vmaw);
1924 + qemu_coroutine_yield();
1925 + aio_set_fd_handler(qemu_get_aio_context(), vmaw->fd, false, NULL, NULL, NULL, NULL);
1926 + if (vmaw->status < 0) {
1927 + DPRINTF("vma_queue_write detected canceled backup\n");
1931 + ret = write(vmaw->fd, buf + done, bytes - done);
1934 + DPRINTF("vma_queue_write written %zd %zd\n", done, ret);
1935 + } else if (ret < 0) {
1936 + if (errno == EAGAIN || errno == EWOULDBLOCK) {
1939 + vma_writer_set_error(vmaw, "vma_queue_write: write error - %s",
1940 + g_strerror(errno));
1941 + done = -1; /* always return failure for partial writes */
1944 + } else if (ret == 0) {
1945 + /* should not happen - simply try again */
1949 + vmaw->co_writer = NULL;
1951 + return (done == bytes) ? bytes : -1;
1954 +VmaWriter *vma_writer_create(const char *filename, uuid_t uuid, Error **errp)
1958 + assert(sizeof(VmaHeader) == (4096 + 8192));
1959 + assert(G_STRUCT_OFFSET(VmaHeader, config_names) == 2044);
1960 + assert(G_STRUCT_OFFSET(VmaHeader, config_data) == 3068);
1961 + assert(G_STRUCT_OFFSET(VmaHeader, dev_info) == 4096);
1962 + assert(sizeof(VmaExtentHeader) == 512);
1964 + VmaWriter *vmaw = g_new0(VmaWriter, 1);
1967 + vmaw->md5csum = g_checksum_new(G_CHECKSUM_MD5);
1968 + if (!vmaw->md5csum) {
1969 + error_setg(errp, "can't allocate cmsum\n");
1973 + if (strstart(filename, "exec:", &p)) {
1974 + vmaw->cmd = popen(p, "w");
1975 + if (vmaw->cmd == NULL) {
1976 + error_setg(errp, "can't popen command '%s' - %s\n", p,
1977 + g_strerror(errno));
1980 + vmaw->fd = fileno(vmaw->cmd);
1982 + /* try to use O_NONBLOCK */
1983 + fcntl(vmaw->fd, F_SETFL, fcntl(vmaw->fd, F_GETFL)|O_NONBLOCK);
1988 + const char *tmp_id_str;
1990 + if ((stat(filename, &st) == 0) && S_ISFIFO(st.st_mode)) {
1991 + oflags = O_NONBLOCK|O_WRONLY;
1992 + vmaw->fd = qemu_open(filename, oflags, 0644);
1993 + } else if (strstart(filename, "/dev/fdset/", &tmp_id_str)) {
1994 + oflags = O_NONBLOCK|O_WRONLY;
1995 + vmaw->fd = qemu_open(filename, oflags, 0644);
1996 + } else if (strstart(filename, "/dev/fdname/", &tmp_id_str)) {
1997 + vmaw->fd = monitor_get_fd(cur_mon, tmp_id_str, errp);
1998 + if (vmaw->fd < 0) {
2001 + /* try to use O_NONBLOCK */
2002 + fcntl(vmaw->fd, F_SETFL, fcntl(vmaw->fd, F_GETFL)|O_NONBLOCK);
2004 + oflags = O_NONBLOCK|O_DIRECT|O_WRONLY|O_CREAT|O_EXCL;
2005 + vmaw->fd = qemu_open(filename, oflags, 0644);
2008 + if (vmaw->fd < 0) {
2009 + error_setg(errp, "can't open file %s - %s\n", filename,
2010 + g_strerror(errno));
2015 + /* we use O_DIRECT, so we need to align IO buffers */
2017 + vmaw->outbuf = qemu_memalign(512, VMA_MAX_EXTENT_SIZE);
2018 + vmaw->headerbuf = qemu_memalign(512, HEADERBUF_SIZE);
2020 + vmaw->outbuf_count = 0;
2021 + vmaw->outbuf_pos = VMA_EXTENT_HEADER_SIZE;
2023 + vmaw->header_blob_table_pos = 1; /* start at pos 1 */
2025 + qemu_co_mutex_init(&vmaw->flush_lock);
2027 + uuid_copy(vmaw->uuid, uuid);
2034 + pclose(vmaw->cmd);
2035 + } else if (vmaw->fd >= 0) {
2039 + if (vmaw->md5csum) {
2040 + g_checksum_free(vmaw->md5csum);
2049 +static int coroutine_fn vma_write_header(VmaWriter *vmaw)
2052 + unsigned char *buf = vmaw->headerbuf;
2053 + VmaHeader *head = (VmaHeader *)buf;
2057 + DPRINTF("VMA WRITE HEADER\n");
2059 + if (vmaw->status < 0) {
2060 + return vmaw->status;
2063 + memset(buf, 0, HEADERBUF_SIZE);
2065 + head->magic = VMA_MAGIC;
2066 + head->version = GUINT32_TO_BE(1); /* v1 */
2067 + memcpy(head->uuid, vmaw->uuid, 16);
2069 + time_t ctime = time(NULL);
2070 + head->ctime = GUINT64_TO_BE(ctime);
2072 + for (i = 0; i < VMA_MAX_CONFIGS; i++) {
2073 + head->config_names[i] = GUINT32_TO_BE(vmaw->config_names[i]);
2074 + head->config_data[i] = GUINT32_TO_BE(vmaw->config_data[i]);
2077 + /* 32 bytes per device (12 used currently) = 8192 bytes max */
2078 + for (i = 1; i <= 254; i++) {
2079 + VmaStreamInfo *si = &vmaw->stream_info[i];
2081 + assert(si->devname);
2082 + uint32_t devname_ptr = allocate_header_string(vmaw, si->devname);
2083 + if (!devname_ptr) {
2086 + head->dev_info[i].devname_ptr = GUINT32_TO_BE(devname_ptr);
2087 + head->dev_info[i].size = GUINT64_TO_BE(si->size);
2091 + uint32_t header_size = sizeof(VmaHeader) + vmaw->header_blob_table_size;
2092 + head->header_size = GUINT32_TO_BE(header_size);
2094 + if (header_size > HEADERBUF_SIZE) {
2095 + return -1; /* just to be sure */
2098 + uint32_t blob_buffer_offset = sizeof(VmaHeader);
2099 + memcpy(buf + blob_buffer_offset, vmaw->header_blob_table,
2100 + vmaw->header_blob_table_size);
2101 + head->blob_buffer_offset = GUINT32_TO_BE(blob_buffer_offset);
2102 + head->blob_buffer_size = GUINT32_TO_BE(vmaw->header_blob_table_pos);
2104 + g_checksum_reset(vmaw->md5csum);
2105 + g_checksum_update(vmaw->md5csum, (const guchar *)buf, header_size);
2107 + g_checksum_get_digest(vmaw->md5csum, (guint8 *)(head->md5sum), &csize);
2109 + return vma_queue_write(vmaw, buf, header_size);
2112 +static int coroutine_fn vma_writer_flush(VmaWriter *vmaw)
2119 + if (vmaw->status < 0) {
2120 + return vmaw->status;
2123 + if (!vmaw->header_written) {
2124 + vmaw->header_written = true;
2125 + ret = vma_write_header(vmaw);
2127 + vma_writer_set_error(vmaw, "vma_writer_flush: write header failed");
2132 + DPRINTF("VMA WRITE FLUSH %d %d\n", vmaw->outbuf_count, vmaw->outbuf_pos);
2135 + VmaExtentHeader *ehead = (VmaExtentHeader *)vmaw->outbuf;
2137 + ehead->magic = VMA_EXTENT_MAGIC;
2138 + ehead->reserved1 = 0;
2140 + for (i = 0; i < VMA_BLOCKS_PER_EXTENT; i++) {
2141 + ehead->blockinfo[i] = GUINT64_TO_BE(vmaw->outbuf_block_info[i]);
2144 + guint16 block_count = (vmaw->outbuf_pos - VMA_EXTENT_HEADER_SIZE) /
2147 + ehead->block_count = GUINT16_TO_BE(block_count);
2149 + memcpy(ehead->uuid, vmaw->uuid, sizeof(ehead->uuid));
2150 + memset(ehead->md5sum, 0, sizeof(ehead->md5sum));
2152 + g_checksum_reset(vmaw->md5csum);
2153 + g_checksum_update(vmaw->md5csum, vmaw->outbuf, VMA_EXTENT_HEADER_SIZE);
2155 + g_checksum_get_digest(vmaw->md5csum, ehead->md5sum, &csize);
2157 + int bytes = vmaw->outbuf_pos;
2158 + ret = vma_queue_write(vmaw, vmaw->outbuf, bytes);
2159 + if (ret != bytes) {
2160 + vma_writer_set_error(vmaw, "vma_writer_flush: failed write");
2163 + vmaw->outbuf_count = 0;
2164 + vmaw->outbuf_pos = VMA_EXTENT_HEADER_SIZE;
2166 + for (i = 0; i < VMA_BLOCKS_PER_EXTENT; i++) {
2167 + vmaw->outbuf_block_info[i] = 0;
2170 + return vmaw->status;
2173 +static int vma_count_open_streams(VmaWriter *vmaw)
2175 + g_assert(vmaw != NULL);
2178 + int open_drives = 0;
2179 + for (i = 0; i <= 255; i++) {
2180 + if (vmaw->stream_info[i].size && !vmaw->stream_info[i].finished) {
2185 + return open_drives;
2190 + * You need to call this if the vma archive does not contain
2191 + * any data stream.
2194 +vma_writer_flush_output(VmaWriter *vmaw)
2196 + qemu_co_mutex_lock(&vmaw->flush_lock);
2197 + int ret = vma_writer_flush(vmaw);
2198 + qemu_co_mutex_unlock(&vmaw->flush_lock);
2200 + vma_writer_set_error(vmaw, "vma_writer_flush_header failed");
2206 + * all jobs should call this when there is no more data
2207 + * Returns: number of remaining stream (0 ==> finished)
2210 +vma_writer_close_stream(VmaWriter *vmaw, uint8_t dev_id)
2212 + g_assert(vmaw != NULL);
2214 + DPRINTF("vma_writer_set_status %d\n", dev_id);
2215 + if (!vmaw->stream_info[dev_id].size) {
2216 + vma_writer_set_error(vmaw, "vma_writer_close_stream: "
2217 + "no such stream %d", dev_id);
2220 + if (vmaw->stream_info[dev_id].finished) {
2221 + vma_writer_set_error(vmaw, "vma_writer_close_stream: "
2222 + "stream already closed %d", dev_id);
2226 + vmaw->stream_info[dev_id].finished = true;
2228 + int open_drives = vma_count_open_streams(vmaw);
2230 + if (open_drives <= 0) {
2231 + DPRINTF("vma_writer_set_status all drives completed\n");
2232 + vma_writer_flush_output(vmaw);
2235 + return open_drives;
2238 +int vma_writer_get_status(VmaWriter *vmaw, VmaStatus *status)
2242 + g_assert(vmaw != NULL);
2245 + status->status = vmaw->status;
2246 + g_strlcpy(status->errmsg, vmaw->errmsg, sizeof(status->errmsg));
2247 + for (i = 0; i <= 255; i++) {
2248 + status->stream_info[i] = vmaw->stream_info[i];
2251 + uuid_unparse_lower(vmaw->uuid, status->uuid_str);
2254 + status->closed = vmaw->closed;
2256 + return vmaw->status;
2259 +static int vma_writer_get_buffer(VmaWriter *vmaw)
2263 + qemu_co_mutex_lock(&vmaw->flush_lock);
2265 + /* wait until buffer is available */
2266 + while (vmaw->outbuf_count >= (VMA_BLOCKS_PER_EXTENT - 1)) {
2267 + ret = vma_writer_flush(vmaw);
2269 + vma_writer_set_error(vmaw, "vma_writer_get_buffer: flush failed");
2274 + qemu_co_mutex_unlock(&vmaw->flush_lock);
2280 +int64_t coroutine_fn
2281 +vma_writer_write(VmaWriter *vmaw, uint8_t dev_id, int64_t cluster_num,
2282 + unsigned char *buf, size_t *zero_bytes)
2284 + g_assert(vmaw != NULL);
2285 + g_assert(zero_bytes != NULL);
2289 + if (vmaw->status < 0) {
2290 + return vmaw->status;
2293 + if (!dev_id || !vmaw->stream_info[dev_id].size) {
2294 + vma_writer_set_error(vmaw, "vma_writer_write: "
2295 + "no such stream %d", dev_id);
2299 + if (vmaw->stream_info[dev_id].finished) {
2300 + vma_writer_set_error(vmaw, "vma_writer_write: "
2301 + "stream already closed %d", dev_id);
2306 + if (cluster_num >= (((uint64_t)1)<<32)) {
2307 + vma_writer_set_error(vmaw, "vma_writer_write: "
2308 + "cluster number out of range");
2312 + if (dev_id == vmaw->vmstate_stream) {
2313 + if (cluster_num != vmaw->vmstate_clusters) {
2314 + vma_writer_set_error(vmaw, "vma_writer_write: "
2315 + "non sequential vmstate write");
2317 + vmaw->vmstate_clusters++;
2318 + } else if (cluster_num >= vmaw->stream_info[dev_id].cluster_count) {
2319 + vma_writer_set_error(vmaw, "vma_writer_write: cluster number too big");
2323 + /* wait until buffer is available */
2324 + if (vma_writer_get_buffer(vmaw) < 0) {
2325 + vma_writer_set_error(vmaw, "vma_writer_write: "
2326 + "vma_writer_get_buffer failed");
2330 + DPRINTF("VMA WRITE %d %zd\n", dev_id, cluster_num);
2332 + uint16_t mask = 0;
2337 + for (i = 0; i < 16; i++) {
2338 + unsigned char *vmablock = buf + (i*VMA_BLOCK_SIZE);
2339 + if (!buffer_is_zero(vmablock, VMA_BLOCK_SIZE)) {
2341 + memcpy(vmaw->outbuf + vmaw->outbuf_pos, vmablock,
2343 + vmaw->outbuf_pos += VMA_BLOCK_SIZE;
2345 + DPRINTF("VMA WRITE %zd ZERO BLOCK %d\n", cluster_num, i);
2346 + vmaw->stream_info[dev_id].zero_bytes += VMA_BLOCK_SIZE;
2347 + *zero_bytes += VMA_BLOCK_SIZE;
2353 + DPRINTF("VMA WRITE %zd ZERO CLUSTER\n", cluster_num);
2354 + vmaw->stream_info[dev_id].zero_bytes += VMA_CLUSTER_SIZE;
2355 + *zero_bytes += VMA_CLUSTER_SIZE;
2358 + uint64_t block_info = ((uint64_t)mask) << (32+16);
2359 + block_info |= ((uint64_t)dev_id) << 32;
2360 + block_info |= (cluster_num & 0xffffffff);
2361 + vmaw->outbuf_block_info[vmaw->outbuf_count] = block_info;
2363 + DPRINTF("VMA WRITE MASK %zd %zx\n", cluster_num, block_info);
2365 + vmaw->outbuf_count++;
2367 + /** NOTE: We allways write whole clusters, but we correctly set
2368 + * transferred bytes. So transferred == size when when everything
2371 + size_t transferred = VMA_CLUSTER_SIZE;
2373 + if (dev_id != vmaw->vmstate_stream) {
2374 + uint64_t last = (cluster_num + 1) * VMA_CLUSTER_SIZE;
2375 + if (last > vmaw->stream_info[dev_id].size) {
2376 + uint64_t diff = last - vmaw->stream_info[dev_id].size;
2377 + if (diff >= VMA_CLUSTER_SIZE) {
2378 + vma_writer_set_error(vmaw, "vma_writer_write: "
2379 + "read after last cluster");
2382 + transferred -= diff;
2386 + vmaw->stream_info[dev_id].transferred += transferred;
2388 + return transferred;
2391 +void vma_writer_error_propagate(VmaWriter *vmaw, Error **errp)
2393 + if (vmaw->status < 0 && *errp == NULL) {
2394 + error_setg(errp, "%s", vmaw->errmsg);
2398 +int vma_writer_close(VmaWriter *vmaw, Error **errp)
2400 + g_assert(vmaw != NULL);
2404 + while (vmaw->co_writer) {
2405 + aio_poll(qemu_get_aio_context(), true);
2408 + assert(vmaw->co_writer == NULL);
2411 + if (pclose(vmaw->cmd) < 0) {
2412 + vma_writer_set_error(vmaw, "vma_writer_close: "
2413 + "pclose failed - %s", g_strerror(errno));
2416 + if (close(vmaw->fd) < 0) {
2417 + vma_writer_set_error(vmaw, "vma_writer_close: "
2418 + "close failed - %s", g_strerror(errno));
2422 + for (i = 0; i <= 255; i++) {
2423 + VmaStreamInfo *si = &vmaw->stream_info[i];
2425 + if (!si->finished) {
2426 + vma_writer_set_error(vmaw, "vma_writer_close: "
2427 + "detected open stream '%s'", si->devname);
2428 + } else if ((si->transferred != si->size) &&
2429 + (i != vmaw->vmstate_stream)) {
2430 + vma_writer_set_error(vmaw, "vma_writer_close: "
2431 + "incomplete stream '%s' (%zd != %zd)",
2432 + si->devname, si->transferred, si->size);
2437 + for (i = 0; i <= 255; i++) {
2438 + vmaw->stream_info[i].finished = 1; /* mark as closed */
2443 + if (vmaw->status < 0 && *errp == NULL) {
2444 + error_setg(errp, "%s", vmaw->errmsg);
2447 + return vmaw->status;
2450 +void vma_writer_destroy(VmaWriter *vmaw)
2456 + for (i = 0; i <= 255; i++) {
2457 + if (vmaw->stream_info[i].devname) {
2458 + g_free(vmaw->stream_info[i].devname);
2462 + if (vmaw->md5csum) {
2463 + g_checksum_free(vmaw->md5csum);
2468 diff --git a/vma.c b/vma.c
2469 new file mode 100644
2470 index 0000000000..04915427c8
2475 + * VMA: Virtual Machine Archive
2477 + * Copyright (C) 2012-2013 Proxmox Server Solutions
2480 + * Dietmar Maurer (dietmar@proxmox.com)
2482 + * This work is licensed under the terms of the GNU GPL, version 2 or later.
2483 + * See the COPYING file in the top-level directory.
2487 +#include "qemu/osdep.h"
2491 +#include "qemu-common.h"
2492 +#include "qemu/error-report.h"
2493 +#include "qemu/main-loop.h"
2494 +#include "qapi/qmp/qstring.h"
2495 +#include "sysemu/char.h" /* qstring_from_str */
2496 +#include "sysemu/block-backend.h"
2498 +static void help(void)
2500 + const char *help_msg =
2501 + "usage: vma command [command options]\n"
2503 + "vma list <filename>\n"
2504 + "vma config <filename> [-c config]\n"
2505 + "vma create <filename> [-c config] pathname ...\n"
2506 + "vma extract <filename> [-r <fifo>] <targetdir>\n"
2507 + "vma verify <filename> [-v]\n"
2510 + printf("%s", help_msg);
2514 +static const char *extract_devname(const char *path, char **devname, int index)
2518 + const char *sep = strchr(path, '=');
2521 + *devname = g_strndup(path, sep - path);
2525 + *devname = g_strdup_printf("disk%d", index);
2534 +static void print_content(VmaReader *vmar)
2538 + VmaHeader *head = vma_reader_get_header(vmar);
2540 + GList *l = vma_reader_get_config_data(vmar);
2541 + while (l && l->data) {
2542 + VmaConfigData *cdata = (VmaConfigData *)l->data;
2543 + l = g_list_next(l);
2544 + printf("CFG: size: %d name: %s\n", cdata->len, cdata->name);
2548 + VmaDeviceInfo *di;
2549 + for (i = 1; i < 255; i++) {
2550 + di = vma_reader_get_device_info(vmar, i);
2552 + if (strcmp(di->devname, "vmstate") == 0) {
2553 + printf("VMSTATE: dev_id=%d memory: %zd\n", i, di->size);
2555 + printf("DEV: dev_id=%d size: %zd devname: %s\n",
2556 + i, di->size, di->devname);
2560 + /* ctime is the last entry we print */
2561 + printf("CTIME: %s", ctime(&head->ctime));
2565 +static int list_content(int argc, char **argv)
2568 + const char *filename;
2571 + c = getopt(argc, argv, "h");
2581 + g_assert_not_reached();
2585 + /* Get the filename */
2586 + if ((optind + 1) != argc) {
2589 + filename = argv[optind++];
2591 + Error *errp = NULL;
2592 + VmaReader *vmar = vma_reader_create(filename, &errp);
2595 + g_error("%s", error_get_pretty(errp));
2598 + print_content(vmar);
2600 + vma_reader_destroy(vmar);
2605 +typedef struct RestoreMap {
2612 +static int extract_content(int argc, char **argv)
2616 + const char *filename;
2617 + const char *dirname;
2618 + const char *readmap = NULL;
2621 + c = getopt(argc, argv, "hvr:");
2641 + /* Get the filename */
2642 + if ((optind + 2) != argc) {
2645 + filename = argv[optind++];
2646 + dirname = argv[optind++];
2648 + Error *errp = NULL;
2649 + VmaReader *vmar = vma_reader_create(filename, &errp);
2652 + g_error("%s", error_get_pretty(errp));
2655 + if (mkdir(dirname, 0777) < 0) {
2656 + g_error("unable to create target directory %s - %s",
2657 + dirname, g_strerror(errno));
2660 + GList *l = vma_reader_get_config_data(vmar);
2661 + while (l && l->data) {
2662 + VmaConfigData *cdata = (VmaConfigData *)l->data;
2663 + l = g_list_next(l);
2664 + char *cfgfn = g_strdup_printf("%s/%s", dirname, cdata->name);
2665 + GError *err = NULL;
2666 + if (!g_file_set_contents(cfgfn, (gchar *)cdata->data, cdata->len,
2668 + g_error("unable to write file: %s", err->message);
2672 + GHashTable *devmap = g_hash_table_new(g_str_hash, g_str_equal);
2675 + print_content(vmar);
2677 + FILE *map = fopen(readmap, "r");
2679 + g_error("unable to open fifo %s - %s", readmap, g_strerror(errno));
2684 + char *line = fgets(inbuf, sizeof(inbuf), map);
2685 + if (!line || line[0] == '\0' || !strcmp(line, "done\n")) {
2688 + int len = strlen(line);
2689 + if (line[len - 1] == '\n') {
2690 + line[len - 1] = '\0';
2696 + char *format = NULL;
2697 + if (strncmp(line, "format=", sizeof("format=")-1) == 0) {
2698 + format = line + sizeof("format=")-1;
2699 + char *colon = strchr(format, ':');
2701 + g_error("read map failed - found only a format ('%s')", inbuf);
2703 + format = g_strndup(format, colon - format);
2709 + if (line[0] == '0' && line[1] == ':') {
2711 + write_zero = false;
2712 + } else if (line[0] == '1' && line[1] == ':') {
2714 + write_zero = true;
2716 + g_error("read map failed - parse error ('%s')", inbuf);
2719 + char *devname = NULL;
2720 + path = extract_devname(path, &devname, -1);
2722 + g_error("read map failed - no dev name specified ('%s')",
2726 + RestoreMap *map = g_new0(RestoreMap, 1);
2727 + map->devname = g_strdup(devname);
2728 + map->path = g_strdup(path);
2729 + map->format = format;
2730 + map->write_zero = write_zero;
2732 + g_hash_table_insert(devmap, map->devname, map);
2738 + int vmstate_fd = -1;
2739 + guint8 vmstate_stream = 0;
2741 + BlockBackend *blk = NULL;
2743 + for (i = 1; i < 255; i++) {
2744 + VmaDeviceInfo *di = vma_reader_get_device_info(vmar, i);
2745 + if (di && (strcmp(di->devname, "vmstate") == 0)) {
2746 + vmstate_stream = i;
2747 + char *statefn = g_strdup_printf("%s/vmstate.bin", dirname);
2748 + vmstate_fd = open(statefn, O_WRONLY|O_CREAT|O_EXCL, 0644);
2749 + if (vmstate_fd < 0) {
2750 + g_error("create vmstate file '%s' failed - %s", statefn,
2751 + g_strerror(errno));
2755 + char *devfn = NULL;
2756 + const char *format = NULL;
2757 + int flags = BDRV_O_RDWR | BDRV_O_NO_FLUSH;
2758 + bool write_zero = true;
2762 + map = (RestoreMap *)g_hash_table_lookup(devmap, di->devname);
2763 + if (map == NULL) {
2764 + g_error("no device name mapping for %s", di->devname);
2766 + devfn = map->path;
2767 + format = map->format;
2768 + write_zero = map->write_zero;
2770 + devfn = g_strdup_printf("%s/tmp-disk-%s.raw",
2771 + dirname, di->devname);
2772 + printf("DEVINFO %s %zd\n", devfn, di->size);
2774 + bdrv_img_create(devfn, "raw", NULL, NULL, NULL, di->size,
2777 + g_error("can't create file %s: %s", devfn,
2778 + error_get_pretty(errp));
2781 + /* Note: we created an empty file above, so there is no
2782 + * need to write zeroes (so we generate a sparse file)
2784 + write_zero = false;
2787 + size_t devlen = strlen(devfn);
2788 + QDict *options = NULL;
2790 + /* explicit format from commandline */
2791 + options = qdict_new();
2792 + qdict_put(options, "driver", qstring_from_str(format));
2793 + } else if ((devlen > 4 && strcmp(devfn+devlen-4, ".raw") == 0) ||
2794 + strncmp(devfn, "/dev/", 5) == 0)
2796 + /* This part is now deprecated for PVE as well (just as qemu
2797 + * deprecated not specifying an explicit raw format, too.
2799 + /* explicit raw format */
2800 + options = qdict_new();
2801 + qdict_put(options, "driver", qstring_from_str("raw"));
2805 + if (errp || !(blk = blk_new_open(devfn, NULL, options, flags, &errp))) {
2806 + g_error("can't open file %s - %s", devfn,
2807 + error_get_pretty(errp));
2810 + if (vma_reader_register_bs(vmar, i, blk, write_zero, &errp) < 0) {
2811 + g_error("%s", error_get_pretty(errp));
2820 + if (vma_reader_restore(vmar, vmstate_fd, verbose, &errp) < 0) {
2821 + g_error("restore failed - %s", error_get_pretty(errp));
2825 + for (i = 1; i < 255; i++) {
2826 + VmaDeviceInfo *di = vma_reader_get_device_info(vmar, i);
2827 + if (di && (i != vmstate_stream)) {
2828 + char *tmpfn = g_strdup_printf("%s/tmp-disk-%s.raw",
2829 + dirname, di->devname);
2830 + char *fn = g_strdup_printf("%s/disk-%s.raw",
2831 + dirname, di->devname);
2832 + if (rename(tmpfn, fn) != 0) {
2833 + g_error("rename %s to %s failed - %s",
2834 + tmpfn, fn, g_strerror(errno));
2840 + vma_reader_destroy(vmar);
2849 +static int verify_content(int argc, char **argv)
2853 + const char *filename;
2856 + c = getopt(argc, argv, "hv");
2873 + /* Get the filename */
2874 + if ((optind + 1) != argc) {
2877 + filename = argv[optind++];
2879 + Error *errp = NULL;
2880 + VmaReader *vmar = vma_reader_create(filename, &errp);
2883 + g_error("%s", error_get_pretty(errp));
2887 + print_content(vmar);
2890 + if (vma_reader_verify(vmar, verbose, &errp) < 0) {
2891 + g_error("verify failed - %s", error_get_pretty(errp));
2894 + vma_reader_destroy(vmar);
2901 +typedef struct BackupJob {
2902 + BlockBackend *target;
2908 +#define BACKUP_SECTORS_PER_CLUSTER (VMA_CLUSTER_SIZE / BDRV_SECTOR_SIZE)
2910 +static void coroutine_fn backup_run_empty(void *opaque)
2912 + VmaWriter *vmaw = (VmaWriter *)opaque;
2914 + vma_writer_flush_output(vmaw);
2916 + Error *err = NULL;
2917 + if (vma_writer_close(vmaw, &err) != 0) {
2918 + g_warning("vma_writer_close failed %s", error_get_pretty(err));
2922 +static void coroutine_fn backup_run(void *opaque)
2924 + BackupJob *job = (BackupJob *)opaque;
2926 + QEMUIOVector qiov;
2928 + int64_t start, end;
2931 + unsigned char *buf = blk_blockalign(job->target, VMA_CLUSTER_SIZE);
2934 + end = DIV_ROUND_UP(job->len / BDRV_SECTOR_SIZE,
2935 + BACKUP_SECTORS_PER_CLUSTER);
2937 + for (; start < end; start++) {
2938 + iov.iov_base = buf;
2939 + iov.iov_len = VMA_CLUSTER_SIZE;
2940 + qemu_iovec_init_external(&qiov, &iov, 1);
2942 + ret = blk_co_preadv(job->target, start * VMA_CLUSTER_SIZE,
2943 + VMA_CLUSTER_SIZE, &qiov, 0);
2945 + vma_writer_set_error(job->vmaw, "read error", -1);
2950 + if (vma_writer_write(job->vmaw, job->dev_id, start, buf, &zb) < 0) {
2951 + vma_writer_set_error(job->vmaw, "backup_dump_cb vma_writer_write failed", -1);
2958 + if (vma_writer_close_stream(job->vmaw, job->dev_id) <= 0) {
2959 + Error *err = NULL;
2960 + if (vma_writer_close(job->vmaw, &err) != 0) {
2961 + g_warning("vma_writer_close failed %s", error_get_pretty(err));
2966 +static int create_archive(int argc, char **argv)
2970 + const char *archivename;
2971 + GList *config_files = NULL;
2974 + c = getopt(argc, argv, "hvc:");
2984 + config_files = g_list_append(config_files, optarg);
2990 + g_assert_not_reached();
2995 + /* make sure we an archive name */
2996 + if ((optind + 1) > argc) {
3000 + archivename = argv[optind++];
3003 + uuid_generate(uuid);
3005 + Error *local_err = NULL;
3006 + VmaWriter *vmaw = vma_writer_create(archivename, uuid, &local_err);
3008 + if (vmaw == NULL) {
3009 + g_error("%s", error_get_pretty(local_err));
3012 + GList *l = config_files;
3013 + while (l && l->data) {
3014 + char *name = l->data;
3015 + char *cdata = NULL;
3017 + GError *err = NULL;
3018 + if (!g_file_get_contents(name, &cdata, &clen, &err)) {
3019 + unlink(archivename);
3020 + g_error("Unable to read file: %s", err->message);
3023 + if (vma_writer_add_config(vmaw, name, cdata, clen) != 0) {
3024 + unlink(archivename);
3025 + g_error("Unable to append config data %s (len = %zd)",
3028 + l = g_list_next(l);
3032 + while (optind < argc) {
3033 + const char *path = argv[optind++];
3034 + char *devname = NULL;
3035 + path = extract_devname(path, &devname, devcount++);
3037 + Error *errp = NULL;
3038 + BlockBackend *target;
3040 + target = blk_new_open(path, NULL, NULL, 0, &errp);
3042 + unlink(archivename);
3043 + g_error("bdrv_open '%s' failed - %s", path, error_get_pretty(errp));
3045 + int64_t size = blk_getlength(target);
3046 + int dev_id = vma_writer_register_stream(vmaw, devname, size);
3047 + if (dev_id <= 0) {
3048 + unlink(archivename);
3049 + g_error("vma_writer_register_stream '%s' failed", devname);
3052 + BackupJob *job = g_new0(BackupJob, 1);
3054 + job->target = target;
3056 + job->dev_id = dev_id;
3058 + Coroutine *co = qemu_coroutine_create(backup_run, job);
3059 + qemu_coroutine_enter(co);
3062 + VmaStatus vmastat;
3064 + int last_percent = -1;
3068 + main_loop_wait(false);
3069 + vma_writer_get_status(vmaw, &vmastat);
3073 + uint64_t total = 0;
3074 + uint64_t transferred = 0;
3075 + uint64_t zero_bytes = 0;
3078 + for (i = 0; i < 256; i++) {
3079 + if (vmastat.stream_info[i].size) {
3080 + total += vmastat.stream_info[i].size;
3081 + transferred += vmastat.stream_info[i].transferred;
3082 + zero_bytes += vmastat.stream_info[i].zero_bytes;
3085 + percent = (transferred*100)/total;
3086 + if (percent != last_percent) {
3087 + fprintf(stderr, "progress %d%% %zd/%zd %zd\n", percent,
3088 + transferred, total, zero_bytes);
3091 + last_percent = percent;
3095 + if (vmastat.closed) {
3100 + Coroutine *co = qemu_coroutine_create(backup_run_empty, vmaw);
3101 + qemu_coroutine_enter(co);
3103 + main_loop_wait(false);
3104 + vma_writer_get_status(vmaw, &vmastat);
3105 + if (vmastat.closed) {
3113 + vma_writer_get_status(vmaw, &vmastat);
3116 + for (i = 0; i < 256; i++) {
3117 + VmaStreamInfo *si = &vmastat.stream_info[i];
3119 + fprintf(stderr, "image %s: size=%zd zeros=%zd saved=%zd\n",
3120 + si->devname, si->size, si->zero_bytes,
3121 + si->size - si->zero_bytes);
3126 + if (vmastat.status < 0) {
3127 + unlink(archivename);
3128 + g_error("creating vma archive failed");
3134 +static int dump_config(int argc, char **argv)
3137 + const char *filename;
3138 + const char *config_name = "qemu-server.conf";
3141 + c = getopt(argc, argv, "hc:");
3151 + config_name = optarg;
3158 + /* Get the filename */
3159 + if ((optind + 1) != argc) {
3162 + filename = argv[optind++];
3164 + Error *errp = NULL;
3165 + VmaReader *vmar = vma_reader_create(filename, &errp);
3168 + g_error("%s", error_get_pretty(errp));
3172 + GList *l = vma_reader_get_config_data(vmar);
3173 + while (l && l->data) {
3174 + VmaConfigData *cdata = (VmaConfigData *)l->data;
3175 + l = g_list_next(l);
3176 + if (strcmp(cdata->name, config_name) == 0) {
3178 + fwrite(cdata->data, cdata->len, 1, stdout);
3183 + vma_reader_destroy(vmar);
3188 + fprintf(stderr, "unable to find configuration data '%s'\n", config_name);
3195 +int main(int argc, char **argv)
3197 + const char *cmdname;
3198 + Error *main_loop_err = NULL;
3200 + error_set_progname(argv[0]);
3202 + if (qemu_init_main_loop(&main_loop_err)) {
3203 + g_error("%s", error_get_pretty(main_loop_err));
3212 + cmdname = argv[1];
3216 + if (!strcmp(cmdname, "list")) {
3217 + return list_content(argc, argv);
3218 + } else if (!strcmp(cmdname, "create")) {
3219 + return create_archive(argc, argv);
3220 + } else if (!strcmp(cmdname, "extract")) {
3221 + return extract_content(argc, argv);
3222 + } else if (!strcmp(cmdname, "verify")) {
3223 + return verify_content(argc, argv);
3224 + } else if (!strcmp(cmdname, "config")) {
3225 + return dump_config(argc, argv);
3231 diff --git a/vma.h b/vma.h
3232 new file mode 100644
3233 index 0000000000..fa6f4df7e7
3238 + * VMA: Virtual Machine Archive
3240 + * Copyright (C) Proxmox Server Solutions
3243 + * Dietmar Maurer (dietmar@proxmox.com)
3245 + * This work is licensed under the terms of the GNU GPL, version 2 or later.
3246 + * See the COPYING file in the top-level directory.
3250 +#ifndef BACKUP_VMA_H
3251 +#define BACKUP_VMA_H
3253 +#include <uuid/uuid.h>
3254 +#include "qapi/error.h"
3255 +#include "block/block.h"
3257 +#define VMA_BLOCK_BITS 12
3258 +#define VMA_BLOCK_SIZE (1<<VMA_BLOCK_BITS)
3259 +#define VMA_CLUSTER_BITS (VMA_BLOCK_BITS+4)
3260 +#define VMA_CLUSTER_SIZE (1<<VMA_CLUSTER_BITS)
3262 +#if VMA_CLUSTER_SIZE != 65536
3263 +#error unexpected cluster size
3266 +#define VMA_EXTENT_HEADER_SIZE 512
3267 +#define VMA_BLOCKS_PER_EXTENT 59
3268 +#define VMA_MAX_CONFIGS 256
3270 +#define VMA_MAX_EXTENT_SIZE \
3271 + (VMA_EXTENT_HEADER_SIZE+VMA_CLUSTER_SIZE*VMA_BLOCKS_PER_EXTENT)
3272 +#if VMA_MAX_EXTENT_SIZE != 3867136
3273 +#error unexpected VMA_EXTENT_SIZE
3276 +/* File Format Definitions */
3278 +#define VMA_MAGIC (GUINT32_TO_BE(('V'<<24)|('M'<<16)|('A'<<8)|0x00))
3279 +#define VMA_EXTENT_MAGIC (GUINT32_TO_BE(('V'<<24)|('M'<<16)|('A'<<8)|'E'))
3281 +typedef struct VmaDeviceInfoHeader {
3282 + uint32_t devname_ptr; /* offset into blob_buffer table */
3283 + uint32_t reserved0;
3284 + uint64_t size; /* device size in bytes */
3285 + uint64_t reserved1;
3286 + uint64_t reserved2;
3287 +} VmaDeviceInfoHeader;
3289 +typedef struct VmaHeader {
3292 + unsigned char uuid[16];
3294 + unsigned char md5sum[16];
3296 + uint32_t blob_buffer_offset;
3297 + uint32_t blob_buffer_size;
3298 + uint32_t header_size;
3300 + unsigned char reserved[1984];
3302 + uint32_t config_names[VMA_MAX_CONFIGS]; /* offset into blob_buffer table */
3303 + uint32_t config_data[VMA_MAX_CONFIGS]; /* offset into blob_buffer table */
3305 + uint32_t reserved1;
3307 + VmaDeviceInfoHeader dev_info[256];
3310 +typedef struct VmaExtentHeader {
3312 + uint16_t reserved1;
3313 + uint16_t block_count;
3314 + unsigned char uuid[16];
3315 + unsigned char md5sum[16];
3316 + uint64_t blockinfo[VMA_BLOCKS_PER_EXTENT];
3319 +/* functions/definitions to read/write vma files */
3321 +typedef struct VmaReader VmaReader;
3323 +typedef struct VmaWriter VmaWriter;
3325 +typedef struct VmaConfigData {
3331 +typedef struct VmaStreamInfo {
3333 + uint64_t cluster_count;
3334 + uint64_t transferred;
3335 + uint64_t zero_bytes;
3340 +typedef struct VmaStatus {
3343 + char errmsg[8192];
3344 + char uuid_str[37];
3345 + VmaStreamInfo stream_info[256];
3348 +typedef struct VmaDeviceInfo {
3349 + uint64_t size; /* device size in bytes */
3350 + const char *devname;
3353 +VmaWriter *vma_writer_create(const char *filename, uuid_t uuid, Error **errp);
3354 +int vma_writer_close(VmaWriter *vmaw, Error **errp);
3355 +void vma_writer_error_propagate(VmaWriter *vmaw, Error **errp);
3356 +void vma_writer_destroy(VmaWriter *vmaw);
3357 +int vma_writer_add_config(VmaWriter *vmaw, const char *name, gpointer data,
3359 +int vma_writer_register_stream(VmaWriter *vmaw, const char *devname,
3362 +int64_t coroutine_fn vma_writer_write(VmaWriter *vmaw, uint8_t dev_id,
3363 + int64_t cluster_num, unsigned char *buf,
3364 + size_t *zero_bytes);
3366 +int coroutine_fn vma_writer_close_stream(VmaWriter *vmaw, uint8_t dev_id);
3367 +int coroutine_fn vma_writer_flush_output(VmaWriter *vmaw);
3369 +int vma_writer_get_status(VmaWriter *vmaw, VmaStatus *status);
3370 +void vma_writer_set_error(VmaWriter *vmaw, const char *fmt, ...);
3373 +VmaReader *vma_reader_create(const char *filename, Error **errp);
3374 +void vma_reader_destroy(VmaReader *vmar);
3375 +VmaHeader *vma_reader_get_header(VmaReader *vmar);
3376 +GList *vma_reader_get_config_data(VmaReader *vmar);
3377 +VmaDeviceInfo *vma_reader_get_device_info(VmaReader *vmar, guint8 dev_id);
3378 +int vma_reader_register_bs(VmaReader *vmar, guint8 dev_id,
3379 + BlockBackend *target, bool write_zeroes,
3381 +int vma_reader_restore(VmaReader *vmar, int vmstate_fd, bool verbose,
3383 +int vma_reader_verify(VmaReader *vmar, bool verbose, Error **errp);
3385 +#endif /* BACKUP_VMA_H */