3c7cfcb9f8e37ad53bb2cce422f898f0c5c9d62c
[pve-qemu-kvm.git] / debian / patches / pve / 0022-internal-snapshot-async.patch
1 From ddfc29076293a794f0d9cc74c0c822c144e7ecbc Mon Sep 17 00:00:00 2001
2 From: Wolfgang Bumiller <w.bumiller@proxmox.com>
3 Date: Wed, 9 Dec 2015 16:04:32 +0100
4 Subject: [PATCH 22/47] internal snapshot async
5
6 ---
7  Makefile.objs           |   1 +
8  block.c                 |   2 +-
9  hmp-commands-info.hx    |  13 ++
10  hmp-commands.hx         |  32 +++
11  hmp.c                   |  57 ++++++
12  hmp.h                   |   5 +
13  include/block/block.h   |   1 +
14  include/sysemu/sysemu.h |   5 +-
15  migration/savevm.c      |  12 +-
16  qapi-schema.json        |  46 +++++
17  qemu-options.hx         |  13 ++
18  qmp-commands.hx         |  30 +++
19  savevm-async.c          | 526 ++++++++++++++++++++++++++++++++++++++++++++++++
20  vl.c                    |   8 +
21  14 files changed, 743 insertions(+), 8 deletions(-)
22  create mode 100644 savevm-async.c
23
24 diff --git a/Makefile.objs b/Makefile.objs
25 index 845edd0..7d9d2d7 100644
26 --- a/Makefile.objs
27 +++ b/Makefile.objs
28 @@ -53,6 +53,7 @@ common-obj-$(CONFIG_LINUX) += fsdev/
29  common-obj-y += migration/
30  common-obj-y += qemu-char.o #aio.o
31  common-obj-y += page_cache.o
32 +common-obj-y += savevm-async.o
33  
34  common-obj-$(CONFIG_SPICE) += spice-qemu-char.o
35  
36 diff --git a/block.c b/block.c
37 index 30d64e6..95c1d32 100644
38 --- a/block.c
39 +++ b/block.c
40 @@ -2288,7 +2288,7 @@ void bdrv_replace_in_backing_chain(BlockDriverState *old, BlockDriverState *new)
41      bdrv_unref(old);
42  }
43  
44 -static void bdrv_delete(BlockDriverState *bs)
45 +void bdrv_delete(BlockDriverState *bs)
46  {
47      assert(!bs->job);
48      assert(bdrv_op_blocker_is_empty(bs));
49 diff --git a/hmp-commands-info.hx b/hmp-commands-info.hx
50 index 7616fe2..3046f9d 100644
51 --- a/hmp-commands-info.hx
52 +++ b/hmp-commands-info.hx
53 @@ -588,6 +588,19 @@ Show current migration xbzrle cache size.
54  ETEXI
55  
56      {
57 +        .name       = "savevm",
58 +        .args_type  = "",
59 +        .params     = "",
60 +        .help       = "show savevm status",
61 +        .mhandler.cmd = hmp_info_savevm,
62 +    },
63 +
64 +STEXI
65 +@item info savevm
66 +show savevm status
67 +ETEXI
68 +
69 +    {
70          .name       = "balloon",
71          .args_type  = "",
72          .params     = "",
73 diff --git a/hmp-commands.hx b/hmp-commands.hx
74 index 0e20ef9..4d735cb 100644
75 --- a/hmp-commands.hx
76 +++ b/hmp-commands.hx
77 @@ -1791,3 +1791,35 @@ ETEXI
78  STEXI
79  @end table
80  ETEXI
81 +
82 +    {
83 +        .name       = "savevm-start",
84 +        .args_type  = "statefile:s?",
85 +        .params     = "[statefile]",
86 +        .help       = "Prepare for snapshot and halt VM. Save VM state to statefile.",
87 +        .mhandler.cmd = hmp_savevm_start,
88 +    },
89 +
90 +    {
91 +        .name       = "snapshot-drive",
92 +        .args_type  = "device:s,name:s",
93 +        .params     = "device name",
94 +        .help       = "Create internal snapshot.",
95 +        .mhandler.cmd = hmp_snapshot_drive,
96 +    },
97 +
98 +    {
99 +        .name       = "delete-drive-snapshot",
100 +        .args_type  = "device:s,name:s",
101 +        .params     = "device name",
102 +        .help       = "Delete internal snapshot.",
103 +        .mhandler.cmd = hmp_delete_drive_snapshot,
104 +    },
105 +
106 +    {
107 +        .name       = "savevm-end",
108 +        .args_type  = "",
109 +        .params     = "",
110 +        .help       = "Resume VM after snaphot.",
111 +        .mhandler.cmd = hmp_savevm_end,
112 +    },
113 diff --git a/hmp.c b/hmp.c
114 index c23cf2f..030fd97 100644
115 --- a/hmp.c
116 +++ b/hmp.c
117 @@ -2117,6 +2117,63 @@ void hmp_info_memory_devices(Monitor *mon, const QDict *qdict)
118      qapi_free_MemoryDeviceInfoList(info_list);
119  }
120  
121 +void hmp_savevm_start(Monitor *mon, const QDict *qdict)
122 +{
123 +    Error *errp = NULL;
124 +    const char *statefile = qdict_get_try_str(qdict, "statefile");
125 +
126 +    qmp_savevm_start(statefile != NULL, statefile, &errp);
127 +    hmp_handle_error(mon, &errp);
128 +}
129 +
130 +void hmp_snapshot_drive(Monitor *mon, const QDict *qdict)
131 +{
132 +    Error *errp = NULL;
133 +    const char *name = qdict_get_str(qdict, "name");
134 +    const char *device = qdict_get_str(qdict, "device");
135 +
136 +    qmp_snapshot_drive(device, name, &errp);
137 +    hmp_handle_error(mon, &errp);
138 +}
139 +
140 +void hmp_delete_drive_snapshot(Monitor *mon, const QDict *qdict)
141 +{
142 +    Error *errp = NULL;
143 +    const char *name = qdict_get_str(qdict, "name");
144 +    const char *device = qdict_get_str(qdict, "device");
145 +
146 +    qmp_delete_drive_snapshot(device, name, &errp);
147 +    hmp_handle_error(mon, &errp);
148 +}
149 +
150 +void hmp_savevm_end(Monitor *mon, const QDict *qdict)
151 +{
152 +    Error *errp = NULL;
153 +
154 +    qmp_savevm_end(&errp);
155 +    hmp_handle_error(mon, &errp);
156 +}
157 +
158 +void hmp_info_savevm(Monitor *mon, const QDict *qdict)
159 +{
160 +    SaveVMInfo *info;
161 +    info = qmp_query_savevm(NULL);
162 +
163 +    if (info->has_status) {
164 +        monitor_printf(mon, "savevm status: %s\n", info->status);
165 +        monitor_printf(mon, "total time: %" PRIu64 " milliseconds\n",
166 +                       info->total_time);
167 +    } else {
168 +        monitor_printf(mon, "savevm status: not running\n");
169 +    }
170 +    if (info->has_bytes) {
171 +        monitor_printf(mon, "Bytes saved: %"PRIu64"\n", info->bytes);
172 +    }
173 +    if (info->has_error) {
174 +        monitor_printf(mon, "Error: %s\n", info->error);
175 +    }
176 +}
177 +
178  void hmp_info_iothreads(Monitor *mon, const QDict *qdict)
179  {
180      IOThreadInfoList *info_list = qmp_query_iothreads(NULL);
181 diff --git a/hmp.h b/hmp.h
182 index 9a4c1f6..b74ddbf 100644
183 --- a/hmp.h
184 +++ b/hmp.h
185 @@ -26,6 +26,7 @@ void hmp_info_status(Monitor *mon, const QDict *qdict);
186  void hmp_info_uuid(Monitor *mon, const QDict *qdict);
187  void hmp_info_chardev(Monitor *mon, const QDict *qdict);
188  void hmp_info_mice(Monitor *mon, const QDict *qdict);
189 +void hmp_info_savevm(Monitor *mon, const QDict *qdict);
190  void hmp_info_migrate(Monitor *mon, const QDict *qdict);
191  void hmp_info_migrate_capabilities(Monitor *mon, const QDict *qdict);
192  void hmp_info_migrate_parameters(Monitor *mon, const QDict *qdict);
193 @@ -92,6 +93,10 @@ void hmp_netdev_add(Monitor *mon, const QDict *qdict);
194  void hmp_netdev_del(Monitor *mon, const QDict *qdict);
195  void hmp_getfd(Monitor *mon, const QDict *qdict);
196  void hmp_closefd(Monitor *mon, const QDict *qdict);
197 +void hmp_savevm_start(Monitor *mon, const QDict *qdict);
198 +void hmp_snapshot_drive(Monitor *mon, const QDict *qdict);
199 +void hmp_delete_drive_snapshot(Monitor *mon, const QDict *qdict);
200 +void hmp_savevm_end(Monitor *mon, const QDict *qdict);
201  void hmp_sendkey(Monitor *mon, const QDict *qdict);
202  void hmp_screendump(Monitor *mon, const QDict *qdict);
203  void hmp_nbd_server_start(Monitor *mon, const QDict *qdict);
204 diff --git a/include/block/block.h b/include/block/block.h
205 index acddf3b..0f70a9d 100644
206 --- a/include/block/block.h
207 +++ b/include/block/block.h
208 @@ -256,6 +256,7 @@ BlockDriverState *bdrv_find_backing_image(BlockDriverState *bs,
209  int bdrv_get_backing_file_depth(BlockDriverState *bs);
210  void bdrv_refresh_filename(BlockDriverState *bs);
211  int bdrv_truncate(BlockDriverState *bs, int64_t offset);
212 +void bdrv_delete(BlockDriverState *bs);
213  int64_t bdrv_nb_sectors(BlockDriverState *bs);
214  int64_t bdrv_getlength(BlockDriverState *bs);
215  int64_t bdrv_get_allocated_file_size(BlockDriverState *bs);
216 diff --git a/include/sysemu/sysemu.h b/include/sysemu/sysemu.h
217 index ee7c760..4875441 100644
218 --- a/include/sysemu/sysemu.h
219 +++ b/include/sysemu/sysemu.h
220 @@ -79,6 +79,7 @@ void qemu_remove_machine_init_done_notifier(Notifier *notify);
221  
222  void hmp_savevm(Monitor *mon, const QDict *qdict);
223  int load_vmstate(const char *name);
224 +int load_state_from_blockdev(const char *filename);
225  void hmp_delvm(Monitor *mon, const QDict *qdict);
226  void hmp_info_snapshots(Monitor *mon, const QDict *qdict);
227  
228 @@ -106,13 +107,13 @@ enum qemu_vm_cmd {
229  #define MAX_VM_CMD_PACKAGED_SIZE (1ul << 24)
230  
231  bool qemu_savevm_state_blocked(Error **errp);
232 -void qemu_savevm_state_begin(QEMUFile *f,
233 +int qemu_savevm_state_begin(QEMUFile *f,
234                               const MigrationParams *params);
235  void qemu_savevm_state_header(QEMUFile *f);
236  int qemu_savevm_state_iterate(QEMUFile *f, bool postcopy);
237  void qemu_savevm_state_cleanup(void);
238  void qemu_savevm_state_complete_postcopy(QEMUFile *f);
239 -void qemu_savevm_state_complete_precopy(QEMUFile *f, bool iterable_only);
240 +int qemu_savevm_state_complete_precopy(QEMUFile *f, bool iterable_only);
241  void qemu_savevm_state_pending(QEMUFile *f, uint64_t max_size,
242                                 uint64_t *res_non_postcopiable,
243                                 uint64_t *res_postcopiable);
244 diff --git a/migration/savevm.c b/migration/savevm.c
245 index 33a2911..b1bdfb6 100644
246 --- a/migration/savevm.c
247 +++ b/migration/savevm.c
248 @@ -879,11 +879,11 @@ void qemu_savevm_state_header(QEMUFile *f)
249  
250  }
251  
252 -void qemu_savevm_state_begin(QEMUFile *f,
253 +int qemu_savevm_state_begin(QEMUFile *f,
254                               const MigrationParams *params)
255  {
256      SaveStateEntry *se;
257 -    int ret;
258 +    int ret = 0;
259  
260      trace_savevm_state_begin();
261      QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
262 @@ -911,6 +911,7 @@ void qemu_savevm_state_begin(QEMUFile *f,
263              break;
264          }
265      }
266 +    return ret;
267  }
268  
269  /*
270 @@ -1014,7 +1015,7 @@ void qemu_savevm_state_complete_postcopy(QEMUFile *f)
271      qemu_fflush(f);
272  }
273  
274 -void qemu_savevm_state_complete_precopy(QEMUFile *f, bool iterable_only)
275 +int qemu_savevm_state_complete_precopy(QEMUFile *f, bool iterable_only)
276  {
277      QJSON *vmdesc;
278      int vmdesc_len;
279 @@ -1048,12 +1049,12 @@ void qemu_savevm_state_complete_precopy(QEMUFile *f, bool iterable_only)
280          save_section_footer(f, se);
281          if (ret < 0) {
282              qemu_file_set_error(f, ret);
283 -            return;
284 +            return ret;
285          }
286      }
287  
288      if (iterable_only) {
289 -        return;
290 +        return ret;
291      }
292  
293      vmdesc = qjson_new();
294 @@ -1100,6 +1101,7 @@ void qemu_savevm_state_complete_precopy(QEMUFile *f, bool iterable_only)
295      qjson_destroy(vmdesc);
296  
297      qemu_fflush(f);
298 +    return qemu_file_get_error(f);
299  }
300  
301  /* Give an estimate of the amount left to be transferred,
302 diff --git a/qapi-schema.json b/qapi-schema.json
303 index 147137d..0c0faf7 100644
304 --- a/qapi-schema.json
305 +++ b/qapi-schema.json
306 @@ -594,6 +594,42 @@
307             '*cpu-throttle-percentage': 'int',
308             '*error-desc': 'str'} }
309  
310 +
311 +# @SaveVMInfo
312 +#
313 +# Information about current migration process.
314 +#
315 +# @status: #optional string describing the current savevm status.
316 +#          This can be 'active', 'completed', 'failed'.
317 +#          If this field is not returned, no savevm process
318 +#          has been initiated
319 +#
320 +# @error: #optional string containing error message is status is failed.
321 +#
322 +# @total-time: #optional total amount of milliseconds since savevm started.
323 +#        If savevm has ended, it returns the total save time
324 +#
325 +# @bytes: #optional total amount of data transfered
326 +#
327 +# Since: 1.3
328 +##
329 +{ 'struct': 'SaveVMInfo',
330 +  'data': {'*status': 'str', '*error': 'str',
331 +           '*total-time': 'int', '*bytes': 'int'} }
332 +
333 +##
334 +# @query-savevm
335 +#
336 +# Returns information about current savevm process.
337 +#
338 +# Returns: @SaveVMInfo
339 +#
340 +# Since: 1.3
341 +##
342 +{ 'command': 'query-savevm', 'returns': 'SaveVMInfo' }
343 +
344 +##
345 +
346  ##
347  # @query-migrate
348  #
349 @@ -3286,8 +3322,18 @@
350  #
351  # Since: 1.2.0
352  ##
353 +
354  { 'command': 'query-target', 'returns': 'TargetInfo' }
355  
356 +{ 'command': 'savevm-start', 'data': { '*statefile': 'str' } }
357 +
358 +{ 'command': 'snapshot-drive', 'data': { 'device': 'str', 'name': 'str' } }
359 +
360 +{ 'command': 'delete-drive-snapshot', 'data': { 'device': 'str', 'name': 'str' } }
361 +
362 +{ 'command': 'savevm-end' }
363 +
364 +
365  ##
366  # @QKeyCode:
367  #
368 diff --git a/qemu-options.hx b/qemu-options.hx
369 index a71aaf8..37fad3b 100644
370 --- a/qemu-options.hx
371 +++ b/qemu-options.hx
372 @@ -3302,6 +3302,19 @@ STEXI
373  Start right away with a saved state (@code{loadvm} in monitor)
374  ETEXI
375  
376 +DEF("loadstate", HAS_ARG, QEMU_OPTION_loadstate, \
377 +    "-loadstate file\n" \
378 +    "                start right away with a saved state\n",
379 +    QEMU_ARCH_ALL)
380 +STEXI
381 +@item -loadstate @var{file}
382 +@findex -loadstate
383 +Start right away with a saved state. This option does not rollback
384 +disk state like @code{loadvm}, so user must make sure that disk
385 +have correct state. @var{file} can be any valid device URL. See the section
386 +for "Device URL Syntax" for more information.
387 +ETEXI
388 +
389  #ifndef _WIN32
390  DEF("daemonize", 0, QEMU_OPTION_daemonize, \
391      "-daemonize      daemonize QEMU after initializing\n", QEMU_ARCH_ALL)
392 diff --git a/qmp-commands.hx b/qmp-commands.hx
393 index a8e8522..6342cd2 100644
394 --- a/qmp-commands.hx
395 +++ b/qmp-commands.hx
396 @@ -4904,6 +4904,36 @@ Example:
397  EQMP
398  
399      {
400 +        .name       = "savevm-start",
401 +        .args_type  = "statefile:s?",
402 +        .mhandler.cmd_new = qmp_marshal_savevm_start,
403 +    },
404 +
405 +    {
406 +        .name       = "snapshot-drive",
407 +        .args_type  = "device:s,name:s",
408 +        .mhandler.cmd_new = qmp_marshal_snapshot_drive,
409 +    },
410 +
411 +    {
412 +        .name       = "delete-drive-snapshot",
413 +        .args_type  = "device:s,name:s",
414 +        .mhandler.cmd_new = qmp_marshal_delete_drive_snapshot,
415 +    },
416 +
417 +    {
418 +        .name       = "savevm-end",
419 +        .args_type  = "",
420 +        .mhandler.cmd_new = qmp_marshal_savevm_end,
421 +    },
422 +
423 +    {
424 +        .name       = "query-savevm",
425 +        .args_type  = "",
426 +        .mhandler.cmd_new = qmp_marshal_query_savevm,
427 +    },
428 +
429 +    {
430          .name       = "query-rocker",
431          .args_type  = "name:s",
432          .mhandler.cmd_new = qmp_marshal_query_rocker,
433 diff --git a/savevm-async.c b/savevm-async.c
434 new file mode 100644
435 index 0000000..ae7ea84
436 --- /dev/null
437 +++ b/savevm-async.c
438 @@ -0,0 +1,526 @@
439 +#include "qemu/osdep.h"
440 +#include "qemu-common.h"
441 +#include "qapi/qmp/qerror.h"
442 +#include "qemu/error-report.h"
443 +#include "sysemu/sysemu.h"
444 +#include "qmp-commands.h"
445 +#include "qemu-options.h"
446 +#include "migration/qemu-file.h"
447 +#include "qom/qom-qobject.h"
448 +#include "migration/migration.h"
449 +#include "block/snapshot.h"
450 +#include "block/qapi.h"
451 +#include "block/block.h"
452 +#include "qemu/timer.h"
453 +#include "sysemu/block-backend.h"
454 +#include "qapi/qmp/qstring.h"
455 +#include "qemu/rcu.h"
456 +#include "qemu/thread.h"
457 +#include "qemu/cutils.h"
458 +
459 +/* #define DEBUG_SAVEVM_STATE */
460 +
461 +#ifdef DEBUG_SAVEVM_STATE
462 +#define DPRINTF(fmt, ...) \
463 +    do { printf("savevm-async: " fmt, ## __VA_ARGS__); } while (0)
464 +#else
465 +#define DPRINTF(fmt, ...) \
466 +    do { } while (0)
467 +#endif
468 +
469 +enum {
470 +    SAVE_STATE_DONE,
471 +    SAVE_STATE_ERROR,
472 +    SAVE_STATE_ACTIVE,
473 +    SAVE_STATE_COMPLETED,
474 +    SAVE_STATE_CANCELLED
475 +};
476 +
477 +
478 +static struct SnapshotState {
479 +    BlockDriverState *bs;
480 +    size_t bs_pos;
481 +    int state;
482 +    Error *error;
483 +    Error *blocker;
484 +    int saved_vm_running;
485 +    QEMUFile *file;
486 +    int64_t total_time;
487 +} snap_state;
488 +
489 +SaveVMInfo *qmp_query_savevm(Error **errp)
490 +{
491 +    SaveVMInfo *info = g_malloc0(sizeof(*info));
492 +    struct SnapshotState *s = &snap_state;
493 +
494 +    if (s->state != SAVE_STATE_DONE) {
495 +        info->has_bytes = true;
496 +        info->bytes = s->bs_pos;
497 +        switch (s->state) {
498 +        case SAVE_STATE_ERROR:
499 +            info->has_status = true;
500 +            info->status = g_strdup("failed");
501 +            info->has_total_time = true;
502 +            info->total_time = s->total_time;
503 +            if (s->error) {
504 +                info->has_error = true;
505 +                info->error = g_strdup(error_get_pretty(s->error));
506 +            }
507 +            break;
508 +        case SAVE_STATE_ACTIVE:
509 +            info->has_status = true;
510 +            info->status = g_strdup("active");
511 +            info->has_total_time = true;
512 +            info->total_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME)
513 +                - s->total_time;
514 +            break;
515 +        case SAVE_STATE_COMPLETED:
516 +            info->has_status = true;
517 +            info->status = g_strdup("completed");
518 +            info->has_total_time = true;
519 +            info->total_time = s->total_time;
520 +            break;
521 +        }
522 +    }
523 +
524 +    return info;
525 +}
526 +
527 +static int save_snapshot_cleanup(void)
528 +{
529 +    int ret = 0;
530 +
531 +    DPRINTF("save_snapshot_cleanup\n");
532 +
533 +    snap_state.total_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME) -
534 +        snap_state.total_time;
535 +
536 +    if (snap_state.file) {
537 +        ret = qemu_fclose(snap_state.file);
538 +    }
539 +
540 +    if (snap_state.bs) {
541 +        /* try to truncate, but ignore errors (will fail on block devices).
542 +         * note: bdrv_read() need whole blocks, so we round up
543 +         */
544 +        size_t size = (snap_state.bs_pos + BDRV_SECTOR_SIZE) & BDRV_SECTOR_MASK;
545 +        bdrv_truncate(snap_state.bs, size);
546 +        bdrv_op_unblock_all(snap_state.bs, snap_state.blocker);
547 +        error_free(snap_state.blocker);
548 +        snap_state.blocker = NULL;
549 +        bdrv_unref(snap_state.bs);
550 +        snap_state.bs = NULL;
551 +    }
552 +
553 +    return ret;
554 +}
555 +
556 +static void save_snapshot_error(const char *fmt, ...)
557 +{
558 +    va_list ap;
559 +    char *msg;
560 +
561 +    va_start(ap, fmt);
562 +    msg = g_strdup_vprintf(fmt, ap);
563 +    va_end(ap);
564 +
565 +    DPRINTF("save_snapshot_error: %s\n", msg);
566 +
567 +    if (!snap_state.error) {
568 +        error_set(&snap_state.error, ERROR_CLASS_GENERIC_ERROR, "%s", msg);
569 +    }
570 +
571 +    g_free (msg);
572 +
573 +    snap_state.state = SAVE_STATE_ERROR;
574 +
575 +    save_snapshot_cleanup();
576 +}
577 +
578 +static void save_snapshot_completed(void)
579 +{
580 +    DPRINTF("save_snapshot_completed\n");
581 +
582 +    if (save_snapshot_cleanup() < 0) {
583 +        snap_state.state = SAVE_STATE_ERROR;
584 +    } else {
585 +        snap_state.state = SAVE_STATE_COMPLETED;
586 +    }
587 +}
588 +
589 +static int block_state_close(void *opaque)
590 +{
591 +    snap_state.file = NULL;
592 +    return bdrv_flush(snap_state.bs);
593 +}
594 +
595 +static int block_state_put_buffer(void *opaque, const uint8_t *buf,
596 +                                  int64_t pos, int size)
597 +{
598 +    int ret;
599 +
600 +    assert(pos == snap_state.bs_pos);
601 +
602 +    if ((ret = bdrv_pwrite(snap_state.bs, snap_state.bs_pos, buf, size)) > 0) {
603 +        snap_state.bs_pos += ret;
604 +    }
605 +
606 +    return ret;
607 +}
608 +
609 +static int store_and_stop(void) {
610 +    if (global_state_store()) {
611 +        save_snapshot_error("Error saving global state");
612 +        return 1;
613 +    }
614 +    if (runstate_is_running()) {
615 +        vm_stop(RUN_STATE_SAVE_VM);
616 +    }
617 +    return 0;
618 +}
619 +
620 +static void process_savevm_co(void *opaque)
621 +{
622 +    int ret;
623 +    int64_t maxlen;
624 +    MigrationParams params = {
625 +        .blk = 0,
626 +        .shared = 0
627 +    };
628 +
629 +    snap_state.state = SAVE_STATE_ACTIVE;
630 +
631 +    qemu_mutex_unlock_iothread();
632 +    qemu_savevm_state_header(snap_state.file);
633 +    ret = qemu_savevm_state_begin(snap_state.file, &params);
634 +    qemu_mutex_lock_iothread();
635 +
636 +    if (ret < 0) {
637 +        save_snapshot_error("qemu_savevm_state_begin failed");
638 +        return;
639 +    }
640 +
641 +    while (snap_state.state == SAVE_STATE_ACTIVE) {
642 +        uint64_t pending_size;
643 +
644 +        pending_size = qemu_savevm_state_pending(snap_state.file, 0);
645 +
646 +        if (pending_size) {
647 +                ret = qemu_savevm_state_iterate(snap_state.file);
648 +                if (ret < 0) {
649 +                    save_snapshot_error("qemu_savevm_state_iterate error %d", ret);
650 +                    break;
651 +                }
652 +                DPRINTF("savevm inerate pending size %lu ret %d\n", pending_size, ret);
653 +        } else {
654 +            DPRINTF("done iterating\n");
655 +            if (store_and_stop())
656 +                break;
657 +            DPRINTF("savevm inerate finished\n");
658 +            qemu_savevm_state_complete_precopy(snap_state.file);
659 +            DPRINTF("save complete\n");
660 +            save_snapshot_completed();
661 +            break;
662 +        }
663 +
664 +        /* stop the VM if we get to the end of available space,
665 +         * or if pending_size is just a few MB
666 +         */
667 +        maxlen = bdrv_getlength(snap_state.bs) - 30*1024*1024;
668 +        if ((pending_size < 100000) ||
669 +            ((snap_state.bs_pos + pending_size) >= maxlen)) {
670 +            if (store_and_stop())
671 +                break;
672 +        }
673 +    }
674 +
675 +    if(snap_state.state == SAVE_STATE_CANCELLED) {
676 +        save_snapshot_completed();
677 +        Error *errp = NULL;
678 +        qmp_savevm_end(&errp);
679 +    }
680 +
681 +}
682 +
683 +static const QEMUFileOps block_file_ops = {
684 +    .put_buffer =     block_state_put_buffer,
685 +    .close =          block_state_close,
686 +};
687 +
688 +
689 +void qmp_savevm_start(bool has_statefile, const char *statefile, Error **errp)
690 +{
691 +    BlockDriver *drv = NULL;
692 +    Error *local_err = NULL;
693 +
694 +    int bdrv_oflags = BDRV_O_RDWR;
695 +    int ret;
696 +
697 +    if (snap_state.state != SAVE_STATE_DONE) {
698 +        error_set(errp, ERROR_CLASS_GENERIC_ERROR,
699 +                  "VM snapshot already started\n");
700 +        return;
701 +    }
702 +
703 +    /* initialize snapshot info */
704 +    snap_state.saved_vm_running = runstate_is_running();
705 +    snap_state.bs_pos = 0;
706 +    snap_state.total_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
707 +    snap_state.blocker = NULL;
708 +
709 +    if (snap_state.error) {
710 +        error_free(snap_state.error);
711 +        snap_state.error = NULL;
712 +    }
713 +
714 +    if (!has_statefile) {
715 +        vm_stop(RUN_STATE_SAVE_VM);
716 +        snap_state.state = SAVE_STATE_COMPLETED;
717 +        return;
718 +    }
719 +
720 +    if (qemu_savevm_state_blocked(errp)) {
721 +        return;
722 +    }
723 +
724 +    /* Open the image */
725 +    snap_state.bs = bdrv_new();
726
727 +    QDict *options = NULL;
728 +    options = qdict_new();
729 +    qdict_put(options, "driver", qstring_from_str("raw"));
730 +    ret = bdrv_open(&snap_state.bs, statefile, NULL, options, bdrv_oflags, drv, &local_err);
731 +    if (ret < 0) {
732 +        error_set(errp, ERROR_CLASS_GENERIC_ERROR, "failed to open '%s'", statefile);
733 +        goto restart;
734 +    }
735 +
736 +    snap_state.file = qemu_fopen_ops(&snap_state, &block_file_ops);
737 +
738 +    if (!snap_state.file) {
739 +        error_set(errp, ERROR_CLASS_GENERIC_ERROR, "failed to open '%s'", statefile);
740 +        goto restart;
741 +    }
742 +
743 +
744 +    error_setg(&snap_state.blocker, "block device is in use by savevm");
745 +    bdrv_op_block_all(snap_state.bs, snap_state.blocker);
746 +
747 +    Coroutine *co = qemu_coroutine_create(process_savevm_co);
748 +    qemu_coroutine_enter(co);
749 +
750 +    return;
751 +
752 +restart:
753 +
754 +    save_snapshot_error("setup failed");
755 +
756 +    if (snap_state.saved_vm_running) {
757 +        vm_start();
758 +    }
759 +}
760 +
761 +void qmp_savevm_end(Error **errp)
762 +{
763 +    if (snap_state.state == SAVE_STATE_DONE) {
764 +        error_set(errp, ERROR_CLASS_GENERIC_ERROR,
765 +                  "VM snapshot not started\n");
766 +        return;
767 +    }
768 +
769 +    if (snap_state.state == SAVE_STATE_ACTIVE) {
770 +        snap_state.state = SAVE_STATE_CANCELLED;
771 +        return;
772 +    }
773 +
774 +    if (snap_state.saved_vm_running) {
775 +        vm_start();
776 +    }
777 +
778 +    snap_state.state = SAVE_STATE_DONE;
779 +}
780 +
781 +void qmp_snapshot_drive(const char *device, const char *name, Error **errp)
782 +{
783 +    BlockBackend *blk;
784 +    BlockDriverState *bs;
785 +    QEMUSnapshotInfo sn1, *sn = &sn1;
786 +    int ret;
787 +#ifdef _WIN32
788 +    struct _timeb tb;
789 +#else
790 +    struct timeval tv;
791 +#endif
792 +
793 +    if (snap_state.state != SAVE_STATE_COMPLETED) {
794 +        error_set(errp, ERROR_CLASS_GENERIC_ERROR,
795 +                  "VM snapshot not ready/started\n");
796 +        return;
797 +    }
798 +
799 +    blk = blk_by_name(device);
800 +    if (!blk) {
801 +        error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
802 +                  "Device '%s' not found", device);
803 +        return;
804 +    }
805 +
806 +    bs = blk_bs(blk);
807 +    if (!bdrv_is_inserted(bs)) {
808 +        error_setg(errp, QERR_DEVICE_HAS_NO_MEDIUM, device);
809 +        return;
810 +    }
811 +
812 +    if (bdrv_is_read_only(bs)) {
813 +        error_setg(errp, "Node '%s' is read only", device);
814 +        return;
815 +    }
816 +
817 +    if (!bdrv_can_snapshot(bs)) {
818 +        error_setg(errp, QERR_UNSUPPORTED);
819 +        return;
820 +    }
821 +
822 +    if (bdrv_snapshot_find(bs, sn, name) >= 0) {
823 +        error_set(errp, ERROR_CLASS_GENERIC_ERROR,
824 +                  "snapshot '%s' already exists", name);
825 +        return;
826 +    }
827 +
828 +    sn = &sn1;
829 +    memset(sn, 0, sizeof(*sn));
830 +
831 +#ifdef _WIN32
832 +    _ftime(&tb);
833 +    sn->date_sec = tb.time;
834 +    sn->date_nsec = tb.millitm * 1000000;
835 +#else
836 +    gettimeofday(&tv, NULL);
837 +    sn->date_sec = tv.tv_sec;
838 +    sn->date_nsec = tv.tv_usec * 1000;
839 +#endif
840 +    sn->vm_clock_nsec = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL);
841 +
842 +    pstrcpy(sn->name, sizeof(sn->name), name);
843 +
844 +    sn->vm_state_size = 0; /* do not save state */
845 +
846 +    ret = bdrv_snapshot_create(bs, sn);
847 +    if (ret < 0) {
848 +        error_set(errp, ERROR_CLASS_GENERIC_ERROR,
849 +                  "Error while creating snapshot on '%s'\n", device);
850 +        return;
851 +    }
852 +}
853 +
854 +void qmp_delete_drive_snapshot(const char *device, const char *name,
855 +                               Error **errp)
856 +{
857 +    BlockBackend *blk;
858 +    BlockDriverState *bs;
859 +    QEMUSnapshotInfo sn1, *sn = &sn1;
860 +    Error *local_err = NULL;
861 +
862 +    int ret;
863 +
864 +    blk = blk_by_name(device);
865 +    if (!blk) {
866 +        error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
867 +                  "Device '%s' not found", device);
868 +        return;
869 +    }
870 +
871 +    bs = blk_bs(blk);
872 +    if (bdrv_is_read_only(bs)) {
873 +        error_setg(errp, "Node '%s' is read only", device);
874 +        return;
875 +    }
876 +
877 +    if (!bdrv_can_snapshot(bs)) {
878 +        error_setg(errp, QERR_UNSUPPORTED);
879 +        return;
880 +    }
881 +
882 +    if (bdrv_snapshot_find(bs, sn, name) < 0) {
883 +        /* return success if snapshot does not exists */
884 +        return;
885 +    }
886 +
887 +    ret = bdrv_snapshot_delete(bs, NULL, name, &local_err);
888 +    if (ret < 0) {
889 +        error_set(errp, ERROR_CLASS_GENERIC_ERROR,
890 +                  "Error while deleting snapshot on '%s'\n", device);
891 +        return;
892 +    }
893 +}
894 +
895 +static int loadstate_get_buffer(void *opaque, uint8_t *buf, int64_t pos,
896 +                                int size)
897 +{
898 +    BlockDriverState *bs = (BlockDriverState *)opaque;
899 +    int64_t maxlen = bdrv_getlength(bs);
900 +    if (pos > maxlen) {
901 +        return -EIO;
902 +    }
903 +    if ((pos + size) > maxlen) {
904 +        size = maxlen - pos - 1;
905 +    }
906 +    if (size == 0) {
907 +        return 0;
908 +    }
909 +    return bdrv_pread(bs, pos, buf, size);
910 +}
911 +
912 +static const QEMUFileOps loadstate_file_ops = {
913 +    .get_buffer = loadstate_get_buffer,
914 +};
915 +
916 +int load_state_from_blockdev(const char *filename)
917 +{
918 +    BlockDriverState *bs = NULL;
919 +    BlockDriver *drv = NULL;
920 +    Error *local_err = NULL;
921 +    Error *blocker = NULL;
922 +
923 +    QEMUFile *f;
924 +    int ret;
925 +
926 +    bs = bdrv_new();
927 +    ret = bdrv_open(&bs, filename, NULL, NULL, 0, drv, &local_err);
928 +    error_setg(&blocker, "block device is in use by load state");
929 +    bdrv_op_block_all(bs, blocker);
930 +
931 +    if (ret < 0) {
932 +        error_report("Could not open VM state file");
933 +        goto the_end;
934 +    }
935 +
936 +    /* restore the VM state */
937 +    f = qemu_fopen_ops(bs, &loadstate_file_ops);
938 +    if (!f) {
939 +        error_report("Could not open VM state file");
940 +        ret = -EINVAL;
941 +        goto the_end;
942 +    }
943 +
944 +    qemu_system_reset(VMRESET_SILENT);
945 +    migration_incoming_state_new(f);
946 +    ret = qemu_loadvm_state(f);
947 +
948 +    qemu_fclose(f);
949 +    migration_incoming_state_destroy();
950 +    if (ret < 0) {
951 +        error_report("Error %d while loading VM state", ret);
952 +        goto the_end;
953 +    }
954 +
955 +    ret = 0;
956 +
957 + the_end:
958 +    if (bs) {
959 +        bdrv_op_unblock_all(bs, blocker);
960 +        error_free(blocker);
961 +        bdrv_unref(bs);
962 +    }
963 +    return ret;
964 +}
965 diff --git a/vl.c b/vl.c
966 index b226e0b..c01b1b5 100644
967 --- a/vl.c
968 +++ b/vl.c
969 @@ -2962,6 +2962,7 @@ int main(int argc, char **argv, char **envp)
970      int optind;
971      const char *optarg;
972      const char *loadvm = NULL;
973 +    const char *loadstate = NULL;
974      MachineClass *machine_class;
975      const char *cpu_model;
976      const char *vga_model = NULL;
977 @@ -3603,6 +3604,9 @@ int main(int argc, char **argv, char **envp)
978              case QEMU_OPTION_loadvm:
979                  loadvm = optarg;
980                  break;
981 +            case QEMU_OPTION_loadstate:
982 +                loadstate = optarg;
983 +                break;
984              case QEMU_OPTION_full_screen:
985                  full_screen = 1;
986                  break;
987 @@ -4597,6 +4601,10 @@ int main(int argc, char **argv, char **envp)
988          if (load_vmstate(loadvm) < 0) {
989              autostart = 0;
990          }
991 +    } else if (loadstate) {
992 +        if (load_state_from_blockdev(loadstate) < 0) {
993 +            autostart = 0;
994 +        }
995      }
996  
997      qdev_prop_check_globals();
998 -- 
999 2.1.4
1000