qapi2texi: Fix up output around #optional
[qemu.git] / migration / savevm.c
1 /*
2 * QEMU System Emulator
3 *
4 * Copyright (c) 2003-2008 Fabrice Bellard
5 * Copyright (c) 2009-2015 Red Hat Inc
6 *
7 * Authors:
8 * Juan Quintela <quintela@redhat.com>
9 *
10 * Permission is hereby granted, free of charge, to any person obtaining a copy
11 * of this software and associated documentation files (the "Software"), to deal
12 * in the Software without restriction, including without limitation the rights
13 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
14 * copies of the Software, and to permit persons to whom the Software is
15 * furnished to do so, subject to the following conditions:
16 *
17 * The above copyright notice and this permission notice shall be included in
18 * all copies or substantial portions of the Software.
19 *
20 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
21 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
22 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
23 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
24 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
25 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
26 * THE SOFTWARE.
27 */
28
29 #include "qemu/osdep.h"
30 #include "cpu.h"
31 #include "hw/boards.h"
32 #include "hw/hw.h"
33 #include "hw/qdev.h"
34 #include "hw/xen/xen.h"
35 #include "net/net.h"
36 #include "monitor/monitor.h"
37 #include "sysemu/sysemu.h"
38 #include "qemu/timer.h"
39 #include "audio/audio.h"
40 #include "migration/migration.h"
41 #include "migration/postcopy-ram.h"
42 #include "qapi/qmp/qerror.h"
43 #include "qemu/error-report.h"
44 #include "qemu/sockets.h"
45 #include "qemu/queue.h"
46 #include "sysemu/cpus.h"
47 #include "exec/memory.h"
48 #include "qmp-commands.h"
49 #include "trace.h"
50 #include "qemu/bitops.h"
51 #include "qemu/iov.h"
52 #include "block/snapshot.h"
53 #include "block/qapi.h"
54 #include "qemu/cutils.h"
55 #include "io/channel-buffer.h"
56 #include "io/channel-file.h"
57
58 #ifndef ETH_P_RARP
59 #define ETH_P_RARP 0x8035
60 #endif
61 #define ARP_HTYPE_ETH 0x0001
62 #define ARP_PTYPE_IP 0x0800
63 #define ARP_OP_REQUEST_REV 0x3
64
65 const unsigned int postcopy_ram_discard_version = 0;
66
67 static bool skip_section_footers;
68
69 static struct mig_cmd_args {
70 ssize_t len; /* -1 = variable */
71 const char *name;
72 } mig_cmd_args[] = {
73 [MIG_CMD_INVALID] = { .len = -1, .name = "INVALID" },
74 [MIG_CMD_OPEN_RETURN_PATH] = { .len = 0, .name = "OPEN_RETURN_PATH" },
75 [MIG_CMD_PING] = { .len = sizeof(uint32_t), .name = "PING" },
76 [MIG_CMD_POSTCOPY_ADVISE] = { .len = 16, .name = "POSTCOPY_ADVISE" },
77 [MIG_CMD_POSTCOPY_LISTEN] = { .len = 0, .name = "POSTCOPY_LISTEN" },
78 [MIG_CMD_POSTCOPY_RUN] = { .len = 0, .name = "POSTCOPY_RUN" },
79 [MIG_CMD_POSTCOPY_RAM_DISCARD] = {
80 .len = -1, .name = "POSTCOPY_RAM_DISCARD" },
81 [MIG_CMD_PACKAGED] = { .len = 4, .name = "PACKAGED" },
82 [MIG_CMD_MAX] = { .len = -1, .name = "MAX" },
83 };
84
85 static int announce_self_create(uint8_t *buf,
86 uint8_t *mac_addr)
87 {
88 /* Ethernet header. */
89 memset(buf, 0xff, 6); /* destination MAC addr */
90 memcpy(buf + 6, mac_addr, 6); /* source MAC addr */
91 *(uint16_t *)(buf + 12) = htons(ETH_P_RARP); /* ethertype */
92
93 /* RARP header. */
94 *(uint16_t *)(buf + 14) = htons(ARP_HTYPE_ETH); /* hardware addr space */
95 *(uint16_t *)(buf + 16) = htons(ARP_PTYPE_IP); /* protocol addr space */
96 *(buf + 18) = 6; /* hardware addr length (ethernet) */
97 *(buf + 19) = 4; /* protocol addr length (IPv4) */
98 *(uint16_t *)(buf + 20) = htons(ARP_OP_REQUEST_REV); /* opcode */
99 memcpy(buf + 22, mac_addr, 6); /* source hw addr */
100 memset(buf + 28, 0x00, 4); /* source protocol addr */
101 memcpy(buf + 32, mac_addr, 6); /* target hw addr */
102 memset(buf + 38, 0x00, 4); /* target protocol addr */
103
104 /* Padding to get up to 60 bytes (ethernet min packet size, minus FCS). */
105 memset(buf + 42, 0x00, 18);
106
107 return 60; /* len (FCS will be added by hardware) */
108 }
109
110 static void qemu_announce_self_iter(NICState *nic, void *opaque)
111 {
112 uint8_t buf[60];
113 int len;
114
115 trace_qemu_announce_self_iter(qemu_ether_ntoa(&nic->conf->macaddr));
116 len = announce_self_create(buf, nic->conf->macaddr.a);
117
118 qemu_send_packet_raw(qemu_get_queue(nic), buf, len);
119 }
120
121
122 static void qemu_announce_self_once(void *opaque)
123 {
124 static int count = SELF_ANNOUNCE_ROUNDS;
125 QEMUTimer *timer = *(QEMUTimer **)opaque;
126
127 qemu_foreach_nic(qemu_announce_self_iter, NULL);
128
129 if (--count) {
130 /* delay 50ms, 150ms, 250ms, ... */
131 timer_mod(timer, qemu_clock_get_ms(QEMU_CLOCK_REALTIME) +
132 self_announce_delay(count));
133 } else {
134 timer_del(timer);
135 timer_free(timer);
136 }
137 }
138
139 void qemu_announce_self(void)
140 {
141 static QEMUTimer *timer;
142 timer = timer_new_ms(QEMU_CLOCK_REALTIME, qemu_announce_self_once, &timer);
143 qemu_announce_self_once(&timer);
144 }
145
146 /***********************************************************/
147 /* savevm/loadvm support */
148
149 static ssize_t block_writev_buffer(void *opaque, struct iovec *iov, int iovcnt,
150 int64_t pos)
151 {
152 int ret;
153 QEMUIOVector qiov;
154
155 qemu_iovec_init_external(&qiov, iov, iovcnt);
156 ret = bdrv_writev_vmstate(opaque, &qiov, pos);
157 if (ret < 0) {
158 return ret;
159 }
160
161 return qiov.size;
162 }
163
164 static ssize_t block_get_buffer(void *opaque, uint8_t *buf, int64_t pos,
165 size_t size)
166 {
167 return bdrv_load_vmstate(opaque, buf, pos, size);
168 }
169
170 static int bdrv_fclose(void *opaque)
171 {
172 return bdrv_flush(opaque);
173 }
174
175 static const QEMUFileOps bdrv_read_ops = {
176 .get_buffer = block_get_buffer,
177 .close = bdrv_fclose
178 };
179
180 static const QEMUFileOps bdrv_write_ops = {
181 .writev_buffer = block_writev_buffer,
182 .close = bdrv_fclose
183 };
184
185 static QEMUFile *qemu_fopen_bdrv(BlockDriverState *bs, int is_writable)
186 {
187 if (is_writable) {
188 return qemu_fopen_ops(bs, &bdrv_write_ops);
189 }
190 return qemu_fopen_ops(bs, &bdrv_read_ops);
191 }
192
193
194 /* QEMUFile timer support.
195 * Not in qemu-file.c to not add qemu-timer.c as dependency to qemu-file.c
196 */
197
198 void timer_put(QEMUFile *f, QEMUTimer *ts)
199 {
200 uint64_t expire_time;
201
202 expire_time = timer_expire_time_ns(ts);
203 qemu_put_be64(f, expire_time);
204 }
205
206 void timer_get(QEMUFile *f, QEMUTimer *ts)
207 {
208 uint64_t expire_time;
209
210 expire_time = qemu_get_be64(f);
211 if (expire_time != -1) {
212 timer_mod_ns(ts, expire_time);
213 } else {
214 timer_del(ts);
215 }
216 }
217
218
219 /* VMState timer support.
220 * Not in vmstate.c to not add qemu-timer.c as dependency to vmstate.c
221 */
222
223 static int get_timer(QEMUFile *f, void *pv, size_t size, VMStateField *field)
224 {
225 QEMUTimer *v = pv;
226 timer_get(f, v);
227 return 0;
228 }
229
230 static int put_timer(QEMUFile *f, void *pv, size_t size, VMStateField *field,
231 QJSON *vmdesc)
232 {
233 QEMUTimer *v = pv;
234 timer_put(f, v);
235
236 return 0;
237 }
238
239 const VMStateInfo vmstate_info_timer = {
240 .name = "timer",
241 .get = get_timer,
242 .put = put_timer,
243 };
244
245
246 typedef struct CompatEntry {
247 char idstr[256];
248 int instance_id;
249 } CompatEntry;
250
251 typedef struct SaveStateEntry {
252 QTAILQ_ENTRY(SaveStateEntry) entry;
253 char idstr[256];
254 int instance_id;
255 int alias_id;
256 int version_id;
257 int section_id;
258 SaveVMHandlers *ops;
259 const VMStateDescription *vmsd;
260 void *opaque;
261 CompatEntry *compat;
262 int is_ram;
263 } SaveStateEntry;
264
265 typedef struct SaveState {
266 QTAILQ_HEAD(, SaveStateEntry) handlers;
267 int global_section_id;
268 bool skip_configuration;
269 uint32_t len;
270 const char *name;
271 uint32_t target_page_bits;
272 } SaveState;
273
274 static SaveState savevm_state = {
275 .handlers = QTAILQ_HEAD_INITIALIZER(savevm_state.handlers),
276 .global_section_id = 0,
277 .skip_configuration = false,
278 };
279
280 void savevm_skip_configuration(void)
281 {
282 savevm_state.skip_configuration = true;
283 }
284
285
286 static void configuration_pre_save(void *opaque)
287 {
288 SaveState *state = opaque;
289 const char *current_name = MACHINE_GET_CLASS(current_machine)->name;
290
291 state->len = strlen(current_name);
292 state->name = current_name;
293 state->target_page_bits = TARGET_PAGE_BITS;
294 }
295
296 static int configuration_pre_load(void *opaque)
297 {
298 SaveState *state = opaque;
299
300 /* If there is no target-page-bits subsection it means the source
301 * predates the variable-target-page-bits support and is using the
302 * minimum possible value for this CPU.
303 */
304 state->target_page_bits = TARGET_PAGE_BITS_MIN;
305 return 0;
306 }
307
308 static int configuration_post_load(void *opaque, int version_id)
309 {
310 SaveState *state = opaque;
311 const char *current_name = MACHINE_GET_CLASS(current_machine)->name;
312
313 if (strncmp(state->name, current_name, state->len) != 0) {
314 error_report("Machine type received is '%.*s' and local is '%s'",
315 (int) state->len, state->name, current_name);
316 return -EINVAL;
317 }
318
319 if (state->target_page_bits != TARGET_PAGE_BITS) {
320 error_report("Received TARGET_PAGE_BITS is %d but local is %d",
321 state->target_page_bits, TARGET_PAGE_BITS);
322 return -EINVAL;
323 }
324
325 return 0;
326 }
327
328 /* The target-page-bits subsection is present only if the
329 * target page size is not the same as the default (ie the
330 * minimum page size for a variable-page-size guest CPU).
331 * If it is present then it contains the actual target page
332 * bits for the machine, and migration will fail if the
333 * two ends don't agree about it.
334 */
335 static bool vmstate_target_page_bits_needed(void *opaque)
336 {
337 return TARGET_PAGE_BITS > TARGET_PAGE_BITS_MIN;
338 }
339
340 static const VMStateDescription vmstate_target_page_bits = {
341 .name = "configuration/target-page-bits",
342 .version_id = 1,
343 .minimum_version_id = 1,
344 .needed = vmstate_target_page_bits_needed,
345 .fields = (VMStateField[]) {
346 VMSTATE_UINT32(target_page_bits, SaveState),
347 VMSTATE_END_OF_LIST()
348 }
349 };
350
351 static const VMStateDescription vmstate_configuration = {
352 .name = "configuration",
353 .version_id = 1,
354 .pre_load = configuration_pre_load,
355 .post_load = configuration_post_load,
356 .pre_save = configuration_pre_save,
357 .fields = (VMStateField[]) {
358 VMSTATE_UINT32(len, SaveState),
359 VMSTATE_VBUFFER_ALLOC_UINT32(name, SaveState, 0, NULL, len),
360 VMSTATE_END_OF_LIST()
361 },
362 .subsections = (const VMStateDescription*[]) {
363 &vmstate_target_page_bits,
364 NULL
365 }
366 };
367
368 static void dump_vmstate_vmsd(FILE *out_file,
369 const VMStateDescription *vmsd, int indent,
370 bool is_subsection);
371
372 static void dump_vmstate_vmsf(FILE *out_file, const VMStateField *field,
373 int indent)
374 {
375 fprintf(out_file, "%*s{\n", indent, "");
376 indent += 2;
377 fprintf(out_file, "%*s\"field\": \"%s\",\n", indent, "", field->name);
378 fprintf(out_file, "%*s\"version_id\": %d,\n", indent, "",
379 field->version_id);
380 fprintf(out_file, "%*s\"field_exists\": %s,\n", indent, "",
381 field->field_exists ? "true" : "false");
382 fprintf(out_file, "%*s\"size\": %zu", indent, "", field->size);
383 if (field->vmsd != NULL) {
384 fprintf(out_file, ",\n");
385 dump_vmstate_vmsd(out_file, field->vmsd, indent, false);
386 }
387 fprintf(out_file, "\n%*s}", indent - 2, "");
388 }
389
390 static void dump_vmstate_vmss(FILE *out_file,
391 const VMStateDescription **subsection,
392 int indent)
393 {
394 if (*subsection != NULL) {
395 dump_vmstate_vmsd(out_file, *subsection, indent, true);
396 }
397 }
398
399 static void dump_vmstate_vmsd(FILE *out_file,
400 const VMStateDescription *vmsd, int indent,
401 bool is_subsection)
402 {
403 if (is_subsection) {
404 fprintf(out_file, "%*s{\n", indent, "");
405 } else {
406 fprintf(out_file, "%*s\"%s\": {\n", indent, "", "Description");
407 }
408 indent += 2;
409 fprintf(out_file, "%*s\"name\": \"%s\",\n", indent, "", vmsd->name);
410 fprintf(out_file, "%*s\"version_id\": %d,\n", indent, "",
411 vmsd->version_id);
412 fprintf(out_file, "%*s\"minimum_version_id\": %d", indent, "",
413 vmsd->minimum_version_id);
414 if (vmsd->fields != NULL) {
415 const VMStateField *field = vmsd->fields;
416 bool first;
417
418 fprintf(out_file, ",\n%*s\"Fields\": [\n", indent, "");
419 first = true;
420 while (field->name != NULL) {
421 if (field->flags & VMS_MUST_EXIST) {
422 /* Ignore VMSTATE_VALIDATE bits; these don't get migrated */
423 field++;
424 continue;
425 }
426 if (!first) {
427 fprintf(out_file, ",\n");
428 }
429 dump_vmstate_vmsf(out_file, field, indent + 2);
430 field++;
431 first = false;
432 }
433 fprintf(out_file, "\n%*s]", indent, "");
434 }
435 if (vmsd->subsections != NULL) {
436 const VMStateDescription **subsection = vmsd->subsections;
437 bool first;
438
439 fprintf(out_file, ",\n%*s\"Subsections\": [\n", indent, "");
440 first = true;
441 while (*subsection != NULL) {
442 if (!first) {
443 fprintf(out_file, ",\n");
444 }
445 dump_vmstate_vmss(out_file, subsection, indent + 2);
446 subsection++;
447 first = false;
448 }
449 fprintf(out_file, "\n%*s]", indent, "");
450 }
451 fprintf(out_file, "\n%*s}", indent - 2, "");
452 }
453
454 static void dump_machine_type(FILE *out_file)
455 {
456 MachineClass *mc;
457
458 mc = MACHINE_GET_CLASS(current_machine);
459
460 fprintf(out_file, " \"vmschkmachine\": {\n");
461 fprintf(out_file, " \"Name\": \"%s\"\n", mc->name);
462 fprintf(out_file, " },\n");
463 }
464
465 void dump_vmstate_json_to_file(FILE *out_file)
466 {
467 GSList *list, *elt;
468 bool first;
469
470 fprintf(out_file, "{\n");
471 dump_machine_type(out_file);
472
473 first = true;
474 list = object_class_get_list(TYPE_DEVICE, true);
475 for (elt = list; elt; elt = elt->next) {
476 DeviceClass *dc = OBJECT_CLASS_CHECK(DeviceClass, elt->data,
477 TYPE_DEVICE);
478 const char *name;
479 int indent = 2;
480
481 if (!dc->vmsd) {
482 continue;
483 }
484
485 if (!first) {
486 fprintf(out_file, ",\n");
487 }
488 name = object_class_get_name(OBJECT_CLASS(dc));
489 fprintf(out_file, "%*s\"%s\": {\n", indent, "", name);
490 indent += 2;
491 fprintf(out_file, "%*s\"Name\": \"%s\",\n", indent, "", name);
492 fprintf(out_file, "%*s\"version_id\": %d,\n", indent, "",
493 dc->vmsd->version_id);
494 fprintf(out_file, "%*s\"minimum_version_id\": %d,\n", indent, "",
495 dc->vmsd->minimum_version_id);
496
497 dump_vmstate_vmsd(out_file, dc->vmsd, indent, false);
498
499 fprintf(out_file, "\n%*s}", indent - 2, "");
500 first = false;
501 }
502 fprintf(out_file, "\n}\n");
503 fclose(out_file);
504 }
505
506 static int calculate_new_instance_id(const char *idstr)
507 {
508 SaveStateEntry *se;
509 int instance_id = 0;
510
511 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
512 if (strcmp(idstr, se->idstr) == 0
513 && instance_id <= se->instance_id) {
514 instance_id = se->instance_id + 1;
515 }
516 }
517 return instance_id;
518 }
519
520 static int calculate_compat_instance_id(const char *idstr)
521 {
522 SaveStateEntry *se;
523 int instance_id = 0;
524
525 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
526 if (!se->compat) {
527 continue;
528 }
529
530 if (strcmp(idstr, se->compat->idstr) == 0
531 && instance_id <= se->compat->instance_id) {
532 instance_id = se->compat->instance_id + 1;
533 }
534 }
535 return instance_id;
536 }
537
538 static inline MigrationPriority save_state_priority(SaveStateEntry *se)
539 {
540 if (se->vmsd) {
541 return se->vmsd->priority;
542 }
543 return MIG_PRI_DEFAULT;
544 }
545
546 static void savevm_state_handler_insert(SaveStateEntry *nse)
547 {
548 MigrationPriority priority = save_state_priority(nse);
549 SaveStateEntry *se;
550
551 assert(priority <= MIG_PRI_MAX);
552
553 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
554 if (save_state_priority(se) < priority) {
555 break;
556 }
557 }
558
559 if (se) {
560 QTAILQ_INSERT_BEFORE(se, nse, entry);
561 } else {
562 QTAILQ_INSERT_TAIL(&savevm_state.handlers, nse, entry);
563 }
564 }
565
566 /* TODO: Individual devices generally have very little idea about the rest
567 of the system, so instance_id should be removed/replaced.
568 Meanwhile pass -1 as instance_id if you do not already have a clearly
569 distinguishing id for all instances of your device class. */
570 int register_savevm_live(DeviceState *dev,
571 const char *idstr,
572 int instance_id,
573 int version_id,
574 SaveVMHandlers *ops,
575 void *opaque)
576 {
577 SaveStateEntry *se;
578
579 se = g_new0(SaveStateEntry, 1);
580 se->version_id = version_id;
581 se->section_id = savevm_state.global_section_id++;
582 se->ops = ops;
583 se->opaque = opaque;
584 se->vmsd = NULL;
585 /* if this is a live_savem then set is_ram */
586 if (ops->save_live_setup != NULL) {
587 se->is_ram = 1;
588 }
589
590 if (dev) {
591 char *id = qdev_get_dev_path(dev);
592 if (id) {
593 if (snprintf(se->idstr, sizeof(se->idstr), "%s/", id) >=
594 sizeof(se->idstr)) {
595 error_report("Path too long for VMState (%s)", id);
596 g_free(id);
597 g_free(se);
598
599 return -1;
600 }
601 g_free(id);
602
603 se->compat = g_new0(CompatEntry, 1);
604 pstrcpy(se->compat->idstr, sizeof(se->compat->idstr), idstr);
605 se->compat->instance_id = instance_id == -1 ?
606 calculate_compat_instance_id(idstr) : instance_id;
607 instance_id = -1;
608 }
609 }
610 pstrcat(se->idstr, sizeof(se->idstr), idstr);
611
612 if (instance_id == -1) {
613 se->instance_id = calculate_new_instance_id(se->idstr);
614 } else {
615 se->instance_id = instance_id;
616 }
617 assert(!se->compat || se->instance_id == 0);
618 savevm_state_handler_insert(se);
619 return 0;
620 }
621
622 int register_savevm(DeviceState *dev,
623 const char *idstr,
624 int instance_id,
625 int version_id,
626 SaveStateHandler *save_state,
627 LoadStateHandler *load_state,
628 void *opaque)
629 {
630 SaveVMHandlers *ops = g_new0(SaveVMHandlers, 1);
631 ops->save_state = save_state;
632 ops->load_state = load_state;
633 return register_savevm_live(dev, idstr, instance_id, version_id,
634 ops, opaque);
635 }
636
637 void unregister_savevm(DeviceState *dev, const char *idstr, void *opaque)
638 {
639 SaveStateEntry *se, *new_se;
640 char id[256] = "";
641
642 if (dev) {
643 char *path = qdev_get_dev_path(dev);
644 if (path) {
645 pstrcpy(id, sizeof(id), path);
646 pstrcat(id, sizeof(id), "/");
647 g_free(path);
648 }
649 }
650 pstrcat(id, sizeof(id), idstr);
651
652 QTAILQ_FOREACH_SAFE(se, &savevm_state.handlers, entry, new_se) {
653 if (strcmp(se->idstr, id) == 0 && se->opaque == opaque) {
654 QTAILQ_REMOVE(&savevm_state.handlers, se, entry);
655 g_free(se->compat);
656 g_free(se->ops);
657 g_free(se);
658 }
659 }
660 }
661
662 int vmstate_register_with_alias_id(DeviceState *dev, int instance_id,
663 const VMStateDescription *vmsd,
664 void *opaque, int alias_id,
665 int required_for_version,
666 Error **errp)
667 {
668 SaveStateEntry *se;
669
670 /* If this triggers, alias support can be dropped for the vmsd. */
671 assert(alias_id == -1 || required_for_version >= vmsd->minimum_version_id);
672
673 se = g_new0(SaveStateEntry, 1);
674 se->version_id = vmsd->version_id;
675 se->section_id = savevm_state.global_section_id++;
676 se->opaque = opaque;
677 se->vmsd = vmsd;
678 se->alias_id = alias_id;
679
680 if (dev) {
681 char *id = qdev_get_dev_path(dev);
682 if (id) {
683 if (snprintf(se->idstr, sizeof(se->idstr), "%s/", id) >=
684 sizeof(se->idstr)) {
685 error_setg(errp, "Path too long for VMState (%s)", id);
686 g_free(id);
687 g_free(se);
688
689 return -1;
690 }
691 g_free(id);
692
693 se->compat = g_new0(CompatEntry, 1);
694 pstrcpy(se->compat->idstr, sizeof(se->compat->idstr), vmsd->name);
695 se->compat->instance_id = instance_id == -1 ?
696 calculate_compat_instance_id(vmsd->name) : instance_id;
697 instance_id = -1;
698 }
699 }
700 pstrcat(se->idstr, sizeof(se->idstr), vmsd->name);
701
702 if (instance_id == -1) {
703 se->instance_id = calculate_new_instance_id(se->idstr);
704 } else {
705 se->instance_id = instance_id;
706 }
707 assert(!se->compat || se->instance_id == 0);
708 savevm_state_handler_insert(se);
709 return 0;
710 }
711
712 void vmstate_unregister(DeviceState *dev, const VMStateDescription *vmsd,
713 void *opaque)
714 {
715 SaveStateEntry *se, *new_se;
716
717 QTAILQ_FOREACH_SAFE(se, &savevm_state.handlers, entry, new_se) {
718 if (se->vmsd == vmsd && se->opaque == opaque) {
719 QTAILQ_REMOVE(&savevm_state.handlers, se, entry);
720 g_free(se->compat);
721 g_free(se);
722 }
723 }
724 }
725
726 static int vmstate_load(QEMUFile *f, SaveStateEntry *se, int version_id)
727 {
728 trace_vmstate_load(se->idstr, se->vmsd ? se->vmsd->name : "(old)");
729 if (!se->vmsd) { /* Old style */
730 return se->ops->load_state(f, se->opaque, version_id);
731 }
732 return vmstate_load_state(f, se->vmsd, se->opaque, version_id);
733 }
734
735 static void vmstate_save_old_style(QEMUFile *f, SaveStateEntry *se, QJSON *vmdesc)
736 {
737 int64_t old_offset, size;
738
739 old_offset = qemu_ftell_fast(f);
740 se->ops->save_state(f, se->opaque);
741 size = qemu_ftell_fast(f) - old_offset;
742
743 if (vmdesc) {
744 json_prop_int(vmdesc, "size", size);
745 json_start_array(vmdesc, "fields");
746 json_start_object(vmdesc, NULL);
747 json_prop_str(vmdesc, "name", "data");
748 json_prop_int(vmdesc, "size", size);
749 json_prop_str(vmdesc, "type", "buffer");
750 json_end_object(vmdesc);
751 json_end_array(vmdesc);
752 }
753 }
754
755 static void vmstate_save(QEMUFile *f, SaveStateEntry *se, QJSON *vmdesc)
756 {
757 trace_vmstate_save(se->idstr, se->vmsd ? se->vmsd->name : "(old)");
758 if (!se->vmsd) {
759 vmstate_save_old_style(f, se, vmdesc);
760 return;
761 }
762 vmstate_save_state(f, se->vmsd, se->opaque, vmdesc);
763 }
764
765 void savevm_skip_section_footers(void)
766 {
767 skip_section_footers = true;
768 }
769
770 /*
771 * Write the header for device section (QEMU_VM_SECTION START/END/PART/FULL)
772 */
773 static void save_section_header(QEMUFile *f, SaveStateEntry *se,
774 uint8_t section_type)
775 {
776 qemu_put_byte(f, section_type);
777 qemu_put_be32(f, se->section_id);
778
779 if (section_type == QEMU_VM_SECTION_FULL ||
780 section_type == QEMU_VM_SECTION_START) {
781 /* ID string */
782 size_t len = strlen(se->idstr);
783 qemu_put_byte(f, len);
784 qemu_put_buffer(f, (uint8_t *)se->idstr, len);
785
786 qemu_put_be32(f, se->instance_id);
787 qemu_put_be32(f, se->version_id);
788 }
789 }
790
791 /*
792 * Write a footer onto device sections that catches cases misformatted device
793 * sections.
794 */
795 static void save_section_footer(QEMUFile *f, SaveStateEntry *se)
796 {
797 if (!skip_section_footers) {
798 qemu_put_byte(f, QEMU_VM_SECTION_FOOTER);
799 qemu_put_be32(f, se->section_id);
800 }
801 }
802
803 /**
804 * qemu_savevm_command_send: Send a 'QEMU_VM_COMMAND' type element with the
805 * command and associated data.
806 *
807 * @f: File to send command on
808 * @command: Command type to send
809 * @len: Length of associated data
810 * @data: Data associated with command.
811 */
812 void qemu_savevm_command_send(QEMUFile *f,
813 enum qemu_vm_cmd command,
814 uint16_t len,
815 uint8_t *data)
816 {
817 trace_savevm_command_send(command, len);
818 qemu_put_byte(f, QEMU_VM_COMMAND);
819 qemu_put_be16(f, (uint16_t)command);
820 qemu_put_be16(f, len);
821 qemu_put_buffer(f, data, len);
822 qemu_fflush(f);
823 }
824
825 void qemu_savevm_send_ping(QEMUFile *f, uint32_t value)
826 {
827 uint32_t buf;
828
829 trace_savevm_send_ping(value);
830 buf = cpu_to_be32(value);
831 qemu_savevm_command_send(f, MIG_CMD_PING, sizeof(value), (uint8_t *)&buf);
832 }
833
834 void qemu_savevm_send_open_return_path(QEMUFile *f)
835 {
836 trace_savevm_send_open_return_path();
837 qemu_savevm_command_send(f, MIG_CMD_OPEN_RETURN_PATH, 0, NULL);
838 }
839
840 /* We have a buffer of data to send; we don't want that all to be loaded
841 * by the command itself, so the command contains just the length of the
842 * extra buffer that we then send straight after it.
843 * TODO: Must be a better way to organise that
844 *
845 * Returns:
846 * 0 on success
847 * -ve on error
848 */
849 int qemu_savevm_send_packaged(QEMUFile *f, const uint8_t *buf, size_t len)
850 {
851 uint32_t tmp;
852
853 if (len > MAX_VM_CMD_PACKAGED_SIZE) {
854 error_report("%s: Unreasonably large packaged state: %zu",
855 __func__, len);
856 return -1;
857 }
858
859 tmp = cpu_to_be32(len);
860
861 trace_qemu_savevm_send_packaged();
862 qemu_savevm_command_send(f, MIG_CMD_PACKAGED, 4, (uint8_t *)&tmp);
863
864 qemu_put_buffer(f, buf, len);
865
866 return 0;
867 }
868
869 /* Send prior to any postcopy transfer */
870 void qemu_savevm_send_postcopy_advise(QEMUFile *f)
871 {
872 uint64_t tmp[2];
873 tmp[0] = cpu_to_be64(ram_pagesize_summary());
874 tmp[1] = cpu_to_be64(1ul << qemu_target_page_bits());
875
876 trace_qemu_savevm_send_postcopy_advise();
877 qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_ADVISE, 16, (uint8_t *)tmp);
878 }
879
880 /* Sent prior to starting the destination running in postcopy, discard pages
881 * that have already been sent but redirtied on the source.
882 * CMD_POSTCOPY_RAM_DISCARD consist of:
883 * byte version (0)
884 * byte Length of name field (not including 0)
885 * n x byte RAM block name
886 * byte 0 terminator (just for safety)
887 * n x Byte ranges within the named RAMBlock
888 * be64 Start of the range
889 * be64 Length
890 *
891 * name: RAMBlock name that these entries are part of
892 * len: Number of page entries
893 * start_list: 'len' addresses
894 * length_list: 'len' addresses
895 *
896 */
897 void qemu_savevm_send_postcopy_ram_discard(QEMUFile *f, const char *name,
898 uint16_t len,
899 uint64_t *start_list,
900 uint64_t *length_list)
901 {
902 uint8_t *buf;
903 uint16_t tmplen;
904 uint16_t t;
905 size_t name_len = strlen(name);
906
907 trace_qemu_savevm_send_postcopy_ram_discard(name, len);
908 assert(name_len < 256);
909 buf = g_malloc0(1 + 1 + name_len + 1 + (8 + 8) * len);
910 buf[0] = postcopy_ram_discard_version;
911 buf[1] = name_len;
912 memcpy(buf + 2, name, name_len);
913 tmplen = 2 + name_len;
914 buf[tmplen++] = '\0';
915
916 for (t = 0; t < len; t++) {
917 stq_be_p(buf + tmplen, start_list[t]);
918 tmplen += 8;
919 stq_be_p(buf + tmplen, length_list[t]);
920 tmplen += 8;
921 }
922 qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_RAM_DISCARD, tmplen, buf);
923 g_free(buf);
924 }
925
926 /* Get the destination into a state where it can receive postcopy data. */
927 void qemu_savevm_send_postcopy_listen(QEMUFile *f)
928 {
929 trace_savevm_send_postcopy_listen();
930 qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_LISTEN, 0, NULL);
931 }
932
933 /* Kick the destination into running */
934 void qemu_savevm_send_postcopy_run(QEMUFile *f)
935 {
936 trace_savevm_send_postcopy_run();
937 qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_RUN, 0, NULL);
938 }
939
940 bool qemu_savevm_state_blocked(Error **errp)
941 {
942 SaveStateEntry *se;
943
944 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
945 if (se->vmsd && se->vmsd->unmigratable) {
946 error_setg(errp, "State blocked by non-migratable device '%s'",
947 se->idstr);
948 return true;
949 }
950 }
951 return false;
952 }
953
954 static bool enforce_config_section(void)
955 {
956 MachineState *machine = MACHINE(qdev_get_machine());
957 return machine->enforce_config_section;
958 }
959
960 void qemu_savevm_state_header(QEMUFile *f)
961 {
962 trace_savevm_state_header();
963 qemu_put_be32(f, QEMU_VM_FILE_MAGIC);
964 qemu_put_be32(f, QEMU_VM_FILE_VERSION);
965
966 if (!savevm_state.skip_configuration || enforce_config_section()) {
967 qemu_put_byte(f, QEMU_VM_CONFIGURATION);
968 vmstate_save_state(f, &vmstate_configuration, &savevm_state, 0);
969 }
970
971 }
972
973 void qemu_savevm_state_begin(QEMUFile *f,
974 const MigrationParams *params)
975 {
976 SaveStateEntry *se;
977 int ret;
978
979 trace_savevm_state_begin();
980 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
981 if (!se->ops || !se->ops->set_params) {
982 continue;
983 }
984 se->ops->set_params(params, se->opaque);
985 }
986
987 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
988 if (!se->ops || !se->ops->save_live_setup) {
989 continue;
990 }
991 if (se->ops && se->ops->is_active) {
992 if (!se->ops->is_active(se->opaque)) {
993 continue;
994 }
995 }
996 save_section_header(f, se, QEMU_VM_SECTION_START);
997
998 ret = se->ops->save_live_setup(f, se->opaque);
999 save_section_footer(f, se);
1000 if (ret < 0) {
1001 qemu_file_set_error(f, ret);
1002 break;
1003 }
1004 }
1005 }
1006
1007 /*
1008 * this function has three return values:
1009 * negative: there was one error, and we have -errno.
1010 * 0 : We haven't finished, caller have to go again
1011 * 1 : We have finished, we can go to complete phase
1012 */
1013 int qemu_savevm_state_iterate(QEMUFile *f, bool postcopy)
1014 {
1015 SaveStateEntry *se;
1016 int ret = 1;
1017
1018 trace_savevm_state_iterate();
1019 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1020 if (!se->ops || !se->ops->save_live_iterate) {
1021 continue;
1022 }
1023 if (se->ops && se->ops->is_active) {
1024 if (!se->ops->is_active(se->opaque)) {
1025 continue;
1026 }
1027 }
1028 /*
1029 * In the postcopy phase, any device that doesn't know how to
1030 * do postcopy should have saved it's state in the _complete
1031 * call that's already run, it might get confused if we call
1032 * iterate afterwards.
1033 */
1034 if (postcopy && !se->ops->save_live_complete_postcopy) {
1035 continue;
1036 }
1037 if (qemu_file_rate_limit(f)) {
1038 return 0;
1039 }
1040 trace_savevm_section_start(se->idstr, se->section_id);
1041
1042 save_section_header(f, se, QEMU_VM_SECTION_PART);
1043
1044 ret = se->ops->save_live_iterate(f, se->opaque);
1045 trace_savevm_section_end(se->idstr, se->section_id, ret);
1046 save_section_footer(f, se);
1047
1048 if (ret < 0) {
1049 qemu_file_set_error(f, ret);
1050 }
1051 if (ret <= 0) {
1052 /* Do not proceed to the next vmstate before this one reported
1053 completion of the current stage. This serializes the migration
1054 and reduces the probability that a faster changing state is
1055 synchronized over and over again. */
1056 break;
1057 }
1058 }
1059 return ret;
1060 }
1061
1062 static bool should_send_vmdesc(void)
1063 {
1064 MachineState *machine = MACHINE(qdev_get_machine());
1065 bool in_postcopy = migration_in_postcopy(migrate_get_current());
1066 return !machine->suppress_vmdesc && !in_postcopy;
1067 }
1068
1069 /*
1070 * Calls the save_live_complete_postcopy methods
1071 * causing the last few pages to be sent immediately and doing any associated
1072 * cleanup.
1073 * Note postcopy also calls qemu_savevm_state_complete_precopy to complete
1074 * all the other devices, but that happens at the point we switch to postcopy.
1075 */
1076 void qemu_savevm_state_complete_postcopy(QEMUFile *f)
1077 {
1078 SaveStateEntry *se;
1079 int ret;
1080
1081 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1082 if (!se->ops || !se->ops->save_live_complete_postcopy) {
1083 continue;
1084 }
1085 if (se->ops && se->ops->is_active) {
1086 if (!se->ops->is_active(se->opaque)) {
1087 continue;
1088 }
1089 }
1090 trace_savevm_section_start(se->idstr, se->section_id);
1091 /* Section type */
1092 qemu_put_byte(f, QEMU_VM_SECTION_END);
1093 qemu_put_be32(f, se->section_id);
1094
1095 ret = se->ops->save_live_complete_postcopy(f, se->opaque);
1096 trace_savevm_section_end(se->idstr, se->section_id, ret);
1097 save_section_footer(f, se);
1098 if (ret < 0) {
1099 qemu_file_set_error(f, ret);
1100 return;
1101 }
1102 }
1103
1104 qemu_put_byte(f, QEMU_VM_EOF);
1105 qemu_fflush(f);
1106 }
1107
1108 void qemu_savevm_state_complete_precopy(QEMUFile *f, bool iterable_only)
1109 {
1110 QJSON *vmdesc;
1111 int vmdesc_len;
1112 SaveStateEntry *se;
1113 int ret;
1114 bool in_postcopy = migration_in_postcopy(migrate_get_current());
1115
1116 trace_savevm_state_complete_precopy();
1117
1118 cpu_synchronize_all_states();
1119
1120 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1121 if (!se->ops ||
1122 (in_postcopy && se->ops->save_live_complete_postcopy) ||
1123 (in_postcopy && !iterable_only) ||
1124 !se->ops->save_live_complete_precopy) {
1125 continue;
1126 }
1127
1128 if (se->ops && se->ops->is_active) {
1129 if (!se->ops->is_active(se->opaque)) {
1130 continue;
1131 }
1132 }
1133 trace_savevm_section_start(se->idstr, se->section_id);
1134
1135 save_section_header(f, se, QEMU_VM_SECTION_END);
1136
1137 ret = se->ops->save_live_complete_precopy(f, se->opaque);
1138 trace_savevm_section_end(se->idstr, se->section_id, ret);
1139 save_section_footer(f, se);
1140 if (ret < 0) {
1141 qemu_file_set_error(f, ret);
1142 return;
1143 }
1144 }
1145
1146 if (iterable_only) {
1147 return;
1148 }
1149
1150 vmdesc = qjson_new();
1151 json_prop_int(vmdesc, "page_size", TARGET_PAGE_SIZE);
1152 json_start_array(vmdesc, "devices");
1153 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1154
1155 if ((!se->ops || !se->ops->save_state) && !se->vmsd) {
1156 continue;
1157 }
1158 if (se->vmsd && !vmstate_save_needed(se->vmsd, se->opaque)) {
1159 trace_savevm_section_skip(se->idstr, se->section_id);
1160 continue;
1161 }
1162
1163 trace_savevm_section_start(se->idstr, se->section_id);
1164
1165 json_start_object(vmdesc, NULL);
1166 json_prop_str(vmdesc, "name", se->idstr);
1167 json_prop_int(vmdesc, "instance_id", se->instance_id);
1168
1169 save_section_header(f, se, QEMU_VM_SECTION_FULL);
1170 vmstate_save(f, se, vmdesc);
1171 trace_savevm_section_end(se->idstr, se->section_id, 0);
1172 save_section_footer(f, se);
1173
1174 json_end_object(vmdesc);
1175 }
1176
1177 if (!in_postcopy) {
1178 /* Postcopy stream will still be going */
1179 qemu_put_byte(f, QEMU_VM_EOF);
1180 }
1181
1182 json_end_array(vmdesc);
1183 qjson_finish(vmdesc);
1184 vmdesc_len = strlen(qjson_get_str(vmdesc));
1185
1186 if (should_send_vmdesc()) {
1187 qemu_put_byte(f, QEMU_VM_VMDESCRIPTION);
1188 qemu_put_be32(f, vmdesc_len);
1189 qemu_put_buffer(f, (uint8_t *)qjson_get_str(vmdesc), vmdesc_len);
1190 }
1191 qjson_destroy(vmdesc);
1192
1193 qemu_fflush(f);
1194 }
1195
1196 /* Give an estimate of the amount left to be transferred,
1197 * the result is split into the amount for units that can and
1198 * for units that can't do postcopy.
1199 */
1200 void qemu_savevm_state_pending(QEMUFile *f, uint64_t max_size,
1201 uint64_t *res_non_postcopiable,
1202 uint64_t *res_postcopiable)
1203 {
1204 SaveStateEntry *se;
1205
1206 *res_non_postcopiable = 0;
1207 *res_postcopiable = 0;
1208
1209
1210 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1211 if (!se->ops || !se->ops->save_live_pending) {
1212 continue;
1213 }
1214 if (se->ops && se->ops->is_active) {
1215 if (!se->ops->is_active(se->opaque)) {
1216 continue;
1217 }
1218 }
1219 se->ops->save_live_pending(f, se->opaque, max_size,
1220 res_non_postcopiable, res_postcopiable);
1221 }
1222 }
1223
1224 void qemu_savevm_state_cleanup(void)
1225 {
1226 SaveStateEntry *se;
1227
1228 trace_savevm_state_cleanup();
1229 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1230 if (se->ops && se->ops->cleanup) {
1231 se->ops->cleanup(se->opaque);
1232 }
1233 }
1234 }
1235
1236 static int qemu_savevm_state(QEMUFile *f, Error **errp)
1237 {
1238 int ret;
1239 MigrationParams params = {
1240 .blk = 0,
1241 .shared = 0
1242 };
1243 MigrationState *ms = migrate_init(&params);
1244 MigrationStatus status;
1245 ms->to_dst_file = f;
1246
1247 if (migration_is_blocked(errp)) {
1248 ret = -EINVAL;
1249 goto done;
1250 }
1251
1252 qemu_mutex_unlock_iothread();
1253 qemu_savevm_state_header(f);
1254 qemu_savevm_state_begin(f, &params);
1255 qemu_mutex_lock_iothread();
1256
1257 while (qemu_file_get_error(f) == 0) {
1258 if (qemu_savevm_state_iterate(f, false) > 0) {
1259 break;
1260 }
1261 }
1262
1263 ret = qemu_file_get_error(f);
1264 if (ret == 0) {
1265 qemu_savevm_state_complete_precopy(f, false);
1266 ret = qemu_file_get_error(f);
1267 }
1268 qemu_savevm_state_cleanup();
1269 if (ret != 0) {
1270 error_setg_errno(errp, -ret, "Error while writing VM state");
1271 }
1272
1273 done:
1274 if (ret != 0) {
1275 status = MIGRATION_STATUS_FAILED;
1276 } else {
1277 status = MIGRATION_STATUS_COMPLETED;
1278 }
1279 migrate_set_state(&ms->state, MIGRATION_STATUS_SETUP, status);
1280
1281 /* f is outer parameter, it should not stay in global migration state after
1282 * this function finished */
1283 ms->to_dst_file = NULL;
1284
1285 return ret;
1286 }
1287
1288 static int qemu_save_device_state(QEMUFile *f)
1289 {
1290 SaveStateEntry *se;
1291
1292 qemu_put_be32(f, QEMU_VM_FILE_MAGIC);
1293 qemu_put_be32(f, QEMU_VM_FILE_VERSION);
1294
1295 cpu_synchronize_all_states();
1296
1297 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1298 if (se->is_ram) {
1299 continue;
1300 }
1301 if ((!se->ops || !se->ops->save_state) && !se->vmsd) {
1302 continue;
1303 }
1304 if (se->vmsd && !vmstate_save_needed(se->vmsd, se->opaque)) {
1305 continue;
1306 }
1307
1308 save_section_header(f, se, QEMU_VM_SECTION_FULL);
1309
1310 vmstate_save(f, se, NULL);
1311
1312 save_section_footer(f, se);
1313 }
1314
1315 qemu_put_byte(f, QEMU_VM_EOF);
1316
1317 return qemu_file_get_error(f);
1318 }
1319
1320 static SaveStateEntry *find_se(const char *idstr, int instance_id)
1321 {
1322 SaveStateEntry *se;
1323
1324 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1325 if (!strcmp(se->idstr, idstr) &&
1326 (instance_id == se->instance_id ||
1327 instance_id == se->alias_id))
1328 return se;
1329 /* Migrating from an older version? */
1330 if (strstr(se->idstr, idstr) && se->compat) {
1331 if (!strcmp(se->compat->idstr, idstr) &&
1332 (instance_id == se->compat->instance_id ||
1333 instance_id == se->alias_id))
1334 return se;
1335 }
1336 }
1337 return NULL;
1338 }
1339
1340 enum LoadVMExitCodes {
1341 /* Allow a command to quit all layers of nested loadvm loops */
1342 LOADVM_QUIT = 1,
1343 };
1344
1345 static int qemu_loadvm_state_main(QEMUFile *f, MigrationIncomingState *mis);
1346
1347 /* ------ incoming postcopy messages ------ */
1348 /* 'advise' arrives before any transfers just to tell us that a postcopy
1349 * *might* happen - it might be skipped if precopy transferred everything
1350 * quickly.
1351 */
1352 static int loadvm_postcopy_handle_advise(MigrationIncomingState *mis)
1353 {
1354 PostcopyState ps = postcopy_state_set(POSTCOPY_INCOMING_ADVISE);
1355 uint64_t remote_pagesize_summary, local_pagesize_summary, remote_tps;
1356
1357 trace_loadvm_postcopy_handle_advise();
1358 if (ps != POSTCOPY_INCOMING_NONE) {
1359 error_report("CMD_POSTCOPY_ADVISE in wrong postcopy state (%d)", ps);
1360 return -1;
1361 }
1362
1363 if (!postcopy_ram_supported_by_host()) {
1364 postcopy_state_set(POSTCOPY_INCOMING_NONE);
1365 return -1;
1366 }
1367
1368 remote_pagesize_summary = qemu_get_be64(mis->from_src_file);
1369 local_pagesize_summary = ram_pagesize_summary();
1370
1371 if (remote_pagesize_summary != local_pagesize_summary) {
1372 /*
1373 * This detects two potential causes of mismatch:
1374 * a) A mismatch in host page sizes
1375 * Some combinations of mismatch are probably possible but it gets
1376 * a bit more complicated. In particular we need to place whole
1377 * host pages on the dest at once, and we need to ensure that we
1378 * handle dirtying to make sure we never end up sending part of
1379 * a hostpage on it's own.
1380 * b) The use of different huge page sizes on source/destination
1381 * a more fine grain test is performed during RAM block migration
1382 * but this test here causes a nice early clear failure, and
1383 * also fails when passed to an older qemu that doesn't
1384 * do huge pages.
1385 */
1386 error_report("Postcopy needs matching RAM page sizes (s=%" PRIx64
1387 " d=%" PRIx64 ")",
1388 remote_pagesize_summary, local_pagesize_summary);
1389 return -1;
1390 }
1391
1392 remote_tps = qemu_get_be64(mis->from_src_file);
1393 if (remote_tps != (1ul << qemu_target_page_bits())) {
1394 /*
1395 * Again, some differences could be dealt with, but for now keep it
1396 * simple.
1397 */
1398 error_report("Postcopy needs matching target page sizes (s=%d d=%d)",
1399 (int)remote_tps, 1 << qemu_target_page_bits());
1400 return -1;
1401 }
1402
1403 if (ram_postcopy_incoming_init(mis)) {
1404 return -1;
1405 }
1406
1407 postcopy_state_set(POSTCOPY_INCOMING_ADVISE);
1408
1409 return 0;
1410 }
1411
1412 /* After postcopy we will be told to throw some pages away since they're
1413 * dirty and will have to be demand fetched. Must happen before CPU is
1414 * started.
1415 * There can be 0..many of these messages, each encoding multiple pages.
1416 */
1417 static int loadvm_postcopy_ram_handle_discard(MigrationIncomingState *mis,
1418 uint16_t len)
1419 {
1420 int tmp;
1421 char ramid[256];
1422 PostcopyState ps = postcopy_state_get();
1423
1424 trace_loadvm_postcopy_ram_handle_discard();
1425
1426 switch (ps) {
1427 case POSTCOPY_INCOMING_ADVISE:
1428 /* 1st discard */
1429 tmp = postcopy_ram_prepare_discard(mis);
1430 if (tmp) {
1431 return tmp;
1432 }
1433 break;
1434
1435 case POSTCOPY_INCOMING_DISCARD:
1436 /* Expected state */
1437 break;
1438
1439 default:
1440 error_report("CMD_POSTCOPY_RAM_DISCARD in wrong postcopy state (%d)",
1441 ps);
1442 return -1;
1443 }
1444 /* We're expecting a
1445 * Version (0)
1446 * a RAM ID string (length byte, name, 0 term)
1447 * then at least 1 16 byte chunk
1448 */
1449 if (len < (1 + 1 + 1 + 1 + 2 * 8)) {
1450 error_report("CMD_POSTCOPY_RAM_DISCARD invalid length (%d)", len);
1451 return -1;
1452 }
1453
1454 tmp = qemu_get_byte(mis->from_src_file);
1455 if (tmp != postcopy_ram_discard_version) {
1456 error_report("CMD_POSTCOPY_RAM_DISCARD invalid version (%d)", tmp);
1457 return -1;
1458 }
1459
1460 if (!qemu_get_counted_string(mis->from_src_file, ramid)) {
1461 error_report("CMD_POSTCOPY_RAM_DISCARD Failed to read RAMBlock ID");
1462 return -1;
1463 }
1464 tmp = qemu_get_byte(mis->from_src_file);
1465 if (tmp != 0) {
1466 error_report("CMD_POSTCOPY_RAM_DISCARD missing nil (%d)", tmp);
1467 return -1;
1468 }
1469
1470 len -= 3 + strlen(ramid);
1471 if (len % 16) {
1472 error_report("CMD_POSTCOPY_RAM_DISCARD invalid length (%d)", len);
1473 return -1;
1474 }
1475 trace_loadvm_postcopy_ram_handle_discard_header(ramid, len);
1476 while (len) {
1477 uint64_t start_addr, block_length;
1478 start_addr = qemu_get_be64(mis->from_src_file);
1479 block_length = qemu_get_be64(mis->from_src_file);
1480
1481 len -= 16;
1482 int ret = ram_discard_range(mis, ramid, start_addr,
1483 block_length);
1484 if (ret) {
1485 return ret;
1486 }
1487 }
1488 trace_loadvm_postcopy_ram_handle_discard_end();
1489
1490 return 0;
1491 }
1492
1493 /*
1494 * Triggered by a postcopy_listen command; this thread takes over reading
1495 * the input stream, leaving the main thread free to carry on loading the rest
1496 * of the device state (from RAM).
1497 * (TODO:This could do with being in a postcopy file - but there again it's
1498 * just another input loop, not that postcopy specific)
1499 */
1500 static void *postcopy_ram_listen_thread(void *opaque)
1501 {
1502 QEMUFile *f = opaque;
1503 MigrationIncomingState *mis = migration_incoming_get_current();
1504 int load_res;
1505
1506 migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE,
1507 MIGRATION_STATUS_POSTCOPY_ACTIVE);
1508 qemu_sem_post(&mis->listen_thread_sem);
1509 trace_postcopy_ram_listen_thread_start();
1510
1511 /*
1512 * Because we're a thread and not a coroutine we can't yield
1513 * in qemu_file, and thus we must be blocking now.
1514 */
1515 qemu_file_set_blocking(f, true);
1516 load_res = qemu_loadvm_state_main(f, mis);
1517 /* And non-blocking again so we don't block in any cleanup */
1518 qemu_file_set_blocking(f, false);
1519
1520 trace_postcopy_ram_listen_thread_exit();
1521 if (load_res < 0) {
1522 error_report("%s: loadvm failed: %d", __func__, load_res);
1523 qemu_file_set_error(f, load_res);
1524 migrate_set_state(&mis->state, MIGRATION_STATUS_POSTCOPY_ACTIVE,
1525 MIGRATION_STATUS_FAILED);
1526 } else {
1527 /*
1528 * This looks good, but it's possible that the device loading in the
1529 * main thread hasn't finished yet, and so we might not be in 'RUN'
1530 * state yet; wait for the end of the main thread.
1531 */
1532 qemu_event_wait(&mis->main_thread_load_event);
1533 }
1534 postcopy_ram_incoming_cleanup(mis);
1535
1536 if (load_res < 0) {
1537 /*
1538 * If something went wrong then we have a bad state so exit;
1539 * depending how far we got it might be possible at this point
1540 * to leave the guest running and fire MCEs for pages that never
1541 * arrived as a desperate recovery step.
1542 */
1543 exit(EXIT_FAILURE);
1544 }
1545
1546 migrate_set_state(&mis->state, MIGRATION_STATUS_POSTCOPY_ACTIVE,
1547 MIGRATION_STATUS_COMPLETED);
1548 /*
1549 * If everything has worked fine, then the main thread has waited
1550 * for us to start, and we're the last use of the mis.
1551 * (If something broke then qemu will have to exit anyway since it's
1552 * got a bad migration state).
1553 */
1554 migration_incoming_state_destroy();
1555
1556
1557 return NULL;
1558 }
1559
1560 /* After this message we must be able to immediately receive postcopy data */
1561 static int loadvm_postcopy_handle_listen(MigrationIncomingState *mis)
1562 {
1563 PostcopyState ps = postcopy_state_set(POSTCOPY_INCOMING_LISTENING);
1564 trace_loadvm_postcopy_handle_listen();
1565 if (ps != POSTCOPY_INCOMING_ADVISE && ps != POSTCOPY_INCOMING_DISCARD) {
1566 error_report("CMD_POSTCOPY_LISTEN in wrong postcopy state (%d)", ps);
1567 return -1;
1568 }
1569 if (ps == POSTCOPY_INCOMING_ADVISE) {
1570 /*
1571 * A rare case, we entered listen without having to do any discards,
1572 * so do the setup that's normally done at the time of the 1st discard.
1573 */
1574 postcopy_ram_prepare_discard(mis);
1575 }
1576
1577 /*
1578 * Sensitise RAM - can now generate requests for blocks that don't exist
1579 * However, at this point the CPU shouldn't be running, and the IO
1580 * shouldn't be doing anything yet so don't actually expect requests
1581 */
1582 if (postcopy_ram_enable_notify(mis)) {
1583 return -1;
1584 }
1585
1586 if (mis->have_listen_thread) {
1587 error_report("CMD_POSTCOPY_RAM_LISTEN already has a listen thread");
1588 return -1;
1589 }
1590
1591 mis->have_listen_thread = true;
1592 /* Start up the listening thread and wait for it to signal ready */
1593 qemu_sem_init(&mis->listen_thread_sem, 0);
1594 qemu_thread_create(&mis->listen_thread, "postcopy/listen",
1595 postcopy_ram_listen_thread, mis->from_src_file,
1596 QEMU_THREAD_DETACHED);
1597 qemu_sem_wait(&mis->listen_thread_sem);
1598 qemu_sem_destroy(&mis->listen_thread_sem);
1599
1600 return 0;
1601 }
1602
1603
1604 typedef struct {
1605 QEMUBH *bh;
1606 } HandleRunBhData;
1607
1608 static void loadvm_postcopy_handle_run_bh(void *opaque)
1609 {
1610 Error *local_err = NULL;
1611 HandleRunBhData *data = opaque;
1612
1613 /* TODO we should move all of this lot into postcopy_ram.c or a shared code
1614 * in migration.c
1615 */
1616 cpu_synchronize_all_post_init();
1617
1618 qemu_announce_self();
1619
1620 /* Make sure all file formats flush their mutable metadata */
1621 bdrv_invalidate_cache_all(&local_err);
1622 if (local_err) {
1623 error_report_err(local_err);
1624 }
1625
1626 trace_loadvm_postcopy_handle_run_cpu_sync();
1627 cpu_synchronize_all_post_init();
1628
1629 trace_loadvm_postcopy_handle_run_vmstart();
1630
1631 if (autostart) {
1632 /* Hold onto your hats, starting the CPU */
1633 vm_start();
1634 } else {
1635 /* leave it paused and let management decide when to start the CPU */
1636 runstate_set(RUN_STATE_PAUSED);
1637 }
1638
1639 qemu_bh_delete(data->bh);
1640 g_free(data);
1641 }
1642
1643 /* After all discards we can start running and asking for pages */
1644 static int loadvm_postcopy_handle_run(MigrationIncomingState *mis)
1645 {
1646 PostcopyState ps = postcopy_state_set(POSTCOPY_INCOMING_RUNNING);
1647 HandleRunBhData *data;
1648
1649 trace_loadvm_postcopy_handle_run();
1650 if (ps != POSTCOPY_INCOMING_LISTENING) {
1651 error_report("CMD_POSTCOPY_RUN in wrong postcopy state (%d)", ps);
1652 return -1;
1653 }
1654
1655 data = g_new(HandleRunBhData, 1);
1656 data->bh = qemu_bh_new(loadvm_postcopy_handle_run_bh, data);
1657 qemu_bh_schedule(data->bh);
1658
1659 /* We need to finish reading the stream from the package
1660 * and also stop reading anything more from the stream that loaded the
1661 * package (since it's now being read by the listener thread).
1662 * LOADVM_QUIT will quit all the layers of nested loadvm loops.
1663 */
1664 return LOADVM_QUIT;
1665 }
1666
1667 /**
1668 * Immediately following this command is a blob of data containing an embedded
1669 * chunk of migration stream; read it and load it.
1670 *
1671 * @mis: Incoming state
1672 * @length: Length of packaged data to read
1673 *
1674 * Returns: Negative values on error
1675 *
1676 */
1677 static int loadvm_handle_cmd_packaged(MigrationIncomingState *mis)
1678 {
1679 int ret;
1680 size_t length;
1681 QIOChannelBuffer *bioc;
1682
1683 length = qemu_get_be32(mis->from_src_file);
1684 trace_loadvm_handle_cmd_packaged(length);
1685
1686 if (length > MAX_VM_CMD_PACKAGED_SIZE) {
1687 error_report("Unreasonably large packaged state: %zu", length);
1688 return -1;
1689 }
1690
1691 bioc = qio_channel_buffer_new(length);
1692 qio_channel_set_name(QIO_CHANNEL(bioc), "migration-loadvm-buffer");
1693 ret = qemu_get_buffer(mis->from_src_file,
1694 bioc->data,
1695 length);
1696 if (ret != length) {
1697 object_unref(OBJECT(bioc));
1698 error_report("CMD_PACKAGED: Buffer receive fail ret=%d length=%zu",
1699 ret, length);
1700 return (ret < 0) ? ret : -EAGAIN;
1701 }
1702 bioc->usage += length;
1703 trace_loadvm_handle_cmd_packaged_received(ret);
1704
1705 QEMUFile *packf = qemu_fopen_channel_input(QIO_CHANNEL(bioc));
1706
1707 ret = qemu_loadvm_state_main(packf, mis);
1708 trace_loadvm_handle_cmd_packaged_main(ret);
1709 qemu_fclose(packf);
1710 object_unref(OBJECT(bioc));
1711
1712 return ret;
1713 }
1714
1715 /*
1716 * Process an incoming 'QEMU_VM_COMMAND'
1717 * 0 just a normal return
1718 * LOADVM_QUIT All good, but exit the loop
1719 * <0 Error
1720 */
1721 static int loadvm_process_command(QEMUFile *f)
1722 {
1723 MigrationIncomingState *mis = migration_incoming_get_current();
1724 uint16_t cmd;
1725 uint16_t len;
1726 uint32_t tmp32;
1727
1728 cmd = qemu_get_be16(f);
1729 len = qemu_get_be16(f);
1730
1731 trace_loadvm_process_command(cmd, len);
1732 if (cmd >= MIG_CMD_MAX || cmd == MIG_CMD_INVALID) {
1733 error_report("MIG_CMD 0x%x unknown (len 0x%x)", cmd, len);
1734 return -EINVAL;
1735 }
1736
1737 if (mig_cmd_args[cmd].len != -1 && mig_cmd_args[cmd].len != len) {
1738 error_report("%s received with bad length - expecting %zu, got %d",
1739 mig_cmd_args[cmd].name,
1740 (size_t)mig_cmd_args[cmd].len, len);
1741 return -ERANGE;
1742 }
1743
1744 switch (cmd) {
1745 case MIG_CMD_OPEN_RETURN_PATH:
1746 if (mis->to_src_file) {
1747 error_report("CMD_OPEN_RETURN_PATH called when RP already open");
1748 /* Not really a problem, so don't give up */
1749 return 0;
1750 }
1751 mis->to_src_file = qemu_file_get_return_path(f);
1752 if (!mis->to_src_file) {
1753 error_report("CMD_OPEN_RETURN_PATH failed");
1754 return -1;
1755 }
1756 break;
1757
1758 case MIG_CMD_PING:
1759 tmp32 = qemu_get_be32(f);
1760 trace_loadvm_process_command_ping(tmp32);
1761 if (!mis->to_src_file) {
1762 error_report("CMD_PING (0x%x) received with no return path",
1763 tmp32);
1764 return -1;
1765 }
1766 migrate_send_rp_pong(mis, tmp32);
1767 break;
1768
1769 case MIG_CMD_PACKAGED:
1770 return loadvm_handle_cmd_packaged(mis);
1771
1772 case MIG_CMD_POSTCOPY_ADVISE:
1773 return loadvm_postcopy_handle_advise(mis);
1774
1775 case MIG_CMD_POSTCOPY_LISTEN:
1776 return loadvm_postcopy_handle_listen(mis);
1777
1778 case MIG_CMD_POSTCOPY_RUN:
1779 return loadvm_postcopy_handle_run(mis);
1780
1781 case MIG_CMD_POSTCOPY_RAM_DISCARD:
1782 return loadvm_postcopy_ram_handle_discard(mis, len);
1783 }
1784
1785 return 0;
1786 }
1787
1788 struct LoadStateEntry {
1789 QLIST_ENTRY(LoadStateEntry) entry;
1790 SaveStateEntry *se;
1791 int section_id;
1792 int version_id;
1793 };
1794
1795 /*
1796 * Read a footer off the wire and check that it matches the expected section
1797 *
1798 * Returns: true if the footer was good
1799 * false if there is a problem (and calls error_report to say why)
1800 */
1801 static bool check_section_footer(QEMUFile *f, LoadStateEntry *le)
1802 {
1803 uint8_t read_mark;
1804 uint32_t read_section_id;
1805
1806 if (skip_section_footers) {
1807 /* No footer to check */
1808 return true;
1809 }
1810
1811 read_mark = qemu_get_byte(f);
1812
1813 if (read_mark != QEMU_VM_SECTION_FOOTER) {
1814 error_report("Missing section footer for %s", le->se->idstr);
1815 return false;
1816 }
1817
1818 read_section_id = qemu_get_be32(f);
1819 if (read_section_id != le->section_id) {
1820 error_report("Mismatched section id in footer for %s -"
1821 " read 0x%x expected 0x%x",
1822 le->se->idstr, read_section_id, le->section_id);
1823 return false;
1824 }
1825
1826 /* All good */
1827 return true;
1828 }
1829
1830 void loadvm_free_handlers(MigrationIncomingState *mis)
1831 {
1832 LoadStateEntry *le, *new_le;
1833
1834 QLIST_FOREACH_SAFE(le, &mis->loadvm_handlers, entry, new_le) {
1835 QLIST_REMOVE(le, entry);
1836 g_free(le);
1837 }
1838 }
1839
1840 static int
1841 qemu_loadvm_section_start_full(QEMUFile *f, MigrationIncomingState *mis)
1842 {
1843 uint32_t instance_id, version_id, section_id;
1844 SaveStateEntry *se;
1845 LoadStateEntry *le;
1846 char idstr[256];
1847 int ret;
1848
1849 /* Read section start */
1850 section_id = qemu_get_be32(f);
1851 if (!qemu_get_counted_string(f, idstr)) {
1852 error_report("Unable to read ID string for section %u",
1853 section_id);
1854 return -EINVAL;
1855 }
1856 instance_id = qemu_get_be32(f);
1857 version_id = qemu_get_be32(f);
1858
1859 trace_qemu_loadvm_state_section_startfull(section_id, idstr,
1860 instance_id, version_id);
1861 /* Find savevm section */
1862 se = find_se(idstr, instance_id);
1863 if (se == NULL) {
1864 error_report("Unknown savevm section or instance '%s' %d",
1865 idstr, instance_id);
1866 return -EINVAL;
1867 }
1868
1869 /* Validate version */
1870 if (version_id > se->version_id) {
1871 error_report("savevm: unsupported version %d for '%s' v%d",
1872 version_id, idstr, se->version_id);
1873 return -EINVAL;
1874 }
1875
1876 /* Validate if it is a device's state */
1877 if (xen_enabled() && se->is_ram) {
1878 error_report("loadvm: %s RAM loading not allowed on Xen", idstr);
1879 return -EINVAL;
1880 }
1881
1882 /* Add entry */
1883 le = g_malloc0(sizeof(*le));
1884
1885 le->se = se;
1886 le->section_id = section_id;
1887 le->version_id = version_id;
1888 QLIST_INSERT_HEAD(&mis->loadvm_handlers, le, entry);
1889
1890 ret = vmstate_load(f, le->se, le->version_id);
1891 if (ret < 0) {
1892 error_report("error while loading state for instance 0x%x of"
1893 " device '%s'", instance_id, idstr);
1894 return ret;
1895 }
1896 if (!check_section_footer(f, le)) {
1897 return -EINVAL;
1898 }
1899
1900 return 0;
1901 }
1902
1903 static int
1904 qemu_loadvm_section_part_end(QEMUFile *f, MigrationIncomingState *mis)
1905 {
1906 uint32_t section_id;
1907 LoadStateEntry *le;
1908 int ret;
1909
1910 section_id = qemu_get_be32(f);
1911
1912 trace_qemu_loadvm_state_section_partend(section_id);
1913 QLIST_FOREACH(le, &mis->loadvm_handlers, entry) {
1914 if (le->section_id == section_id) {
1915 break;
1916 }
1917 }
1918 if (le == NULL) {
1919 error_report("Unknown savevm section %d", section_id);
1920 return -EINVAL;
1921 }
1922
1923 ret = vmstate_load(f, le->se, le->version_id);
1924 if (ret < 0) {
1925 error_report("error while loading state section id %d(%s)",
1926 section_id, le->se->idstr);
1927 return ret;
1928 }
1929 if (!check_section_footer(f, le)) {
1930 return -EINVAL;
1931 }
1932
1933 return 0;
1934 }
1935
1936 static int qemu_loadvm_state_main(QEMUFile *f, MigrationIncomingState *mis)
1937 {
1938 uint8_t section_type;
1939 int ret = 0;
1940
1941 while ((section_type = qemu_get_byte(f)) != QEMU_VM_EOF) {
1942 ret = 0;
1943 trace_qemu_loadvm_state_section(section_type);
1944 switch (section_type) {
1945 case QEMU_VM_SECTION_START:
1946 case QEMU_VM_SECTION_FULL:
1947 ret = qemu_loadvm_section_start_full(f, mis);
1948 if (ret < 0) {
1949 goto out;
1950 }
1951 break;
1952 case QEMU_VM_SECTION_PART:
1953 case QEMU_VM_SECTION_END:
1954 ret = qemu_loadvm_section_part_end(f, mis);
1955 if (ret < 0) {
1956 goto out;
1957 }
1958 break;
1959 case QEMU_VM_COMMAND:
1960 ret = loadvm_process_command(f);
1961 trace_qemu_loadvm_state_section_command(ret);
1962 if ((ret < 0) || (ret & LOADVM_QUIT)) {
1963 goto out;
1964 }
1965 break;
1966 default:
1967 error_report("Unknown savevm section type %d", section_type);
1968 ret = -EINVAL;
1969 goto out;
1970 }
1971 }
1972
1973 out:
1974 if (ret < 0) {
1975 qemu_file_set_error(f, ret);
1976 }
1977 return ret;
1978 }
1979
1980 int qemu_loadvm_state(QEMUFile *f)
1981 {
1982 MigrationIncomingState *mis = migration_incoming_get_current();
1983 Error *local_err = NULL;
1984 unsigned int v;
1985 int ret;
1986
1987 if (qemu_savevm_state_blocked(&local_err)) {
1988 error_report_err(local_err);
1989 return -EINVAL;
1990 }
1991
1992 v = qemu_get_be32(f);
1993 if (v != QEMU_VM_FILE_MAGIC) {
1994 error_report("Not a migration stream");
1995 return -EINVAL;
1996 }
1997
1998 v = qemu_get_be32(f);
1999 if (v == QEMU_VM_FILE_VERSION_COMPAT) {
2000 error_report("SaveVM v2 format is obsolete and don't work anymore");
2001 return -ENOTSUP;
2002 }
2003 if (v != QEMU_VM_FILE_VERSION) {
2004 error_report("Unsupported migration stream version");
2005 return -ENOTSUP;
2006 }
2007
2008 if (!savevm_state.skip_configuration || enforce_config_section()) {
2009 if (qemu_get_byte(f) != QEMU_VM_CONFIGURATION) {
2010 error_report("Configuration section missing");
2011 return -EINVAL;
2012 }
2013 ret = vmstate_load_state(f, &vmstate_configuration, &savevm_state, 0);
2014
2015 if (ret) {
2016 return ret;
2017 }
2018 }
2019
2020 ret = qemu_loadvm_state_main(f, mis);
2021 qemu_event_set(&mis->main_thread_load_event);
2022
2023 trace_qemu_loadvm_state_post_main(ret);
2024
2025 if (mis->have_listen_thread) {
2026 /* Listen thread still going, can't clean up yet */
2027 return ret;
2028 }
2029
2030 if (ret == 0) {
2031 ret = qemu_file_get_error(f);
2032 }
2033
2034 /*
2035 * Try to read in the VMDESC section as well, so that dumping tools that
2036 * intercept our migration stream have the chance to see it.
2037 */
2038
2039 /* We've got to be careful; if we don't read the data and just shut the fd
2040 * then the sender can error if we close while it's still sending.
2041 * We also mustn't read data that isn't there; some transports (RDMA)
2042 * will stall waiting for that data when the source has already closed.
2043 */
2044 if (ret == 0 && should_send_vmdesc()) {
2045 uint8_t *buf;
2046 uint32_t size;
2047 uint8_t section_type = qemu_get_byte(f);
2048
2049 if (section_type != QEMU_VM_VMDESCRIPTION) {
2050 error_report("Expected vmdescription section, but got %d",
2051 section_type);
2052 /*
2053 * It doesn't seem worth failing at this point since
2054 * we apparently have an otherwise valid VM state
2055 */
2056 } else {
2057 buf = g_malloc(0x1000);
2058 size = qemu_get_be32(f);
2059
2060 while (size > 0) {
2061 uint32_t read_chunk = MIN(size, 0x1000);
2062 qemu_get_buffer(f, buf, read_chunk);
2063 size -= read_chunk;
2064 }
2065 g_free(buf);
2066 }
2067 }
2068
2069 cpu_synchronize_all_post_init();
2070
2071 return ret;
2072 }
2073
2074 int save_vmstate(Monitor *mon, const char *name)
2075 {
2076 BlockDriverState *bs, *bs1;
2077 QEMUSnapshotInfo sn1, *sn = &sn1, old_sn1, *old_sn = &old_sn1;
2078 int ret = -1;
2079 QEMUFile *f;
2080 int saved_vm_running;
2081 uint64_t vm_state_size;
2082 qemu_timeval tv;
2083 struct tm tm;
2084 Error *local_err = NULL;
2085 AioContext *aio_context;
2086
2087 if (!bdrv_all_can_snapshot(&bs)) {
2088 monitor_printf(mon, "Device '%s' is writable but does not "
2089 "support snapshots.\n", bdrv_get_device_name(bs));
2090 return ret;
2091 }
2092
2093 /* Delete old snapshots of the same name */
2094 if (name) {
2095 ret = bdrv_all_delete_snapshot(name, &bs1, &local_err);
2096 if (ret < 0) {
2097 error_reportf_err(local_err,
2098 "Error while deleting snapshot on device '%s': ",
2099 bdrv_get_device_name(bs1));
2100 return ret;
2101 }
2102 }
2103
2104 bs = bdrv_all_find_vmstate_bs();
2105 if (bs == NULL) {
2106 monitor_printf(mon, "No block device can accept snapshots\n");
2107 return ret;
2108 }
2109 aio_context = bdrv_get_aio_context(bs);
2110
2111 saved_vm_running = runstate_is_running();
2112
2113 ret = global_state_store();
2114 if (ret) {
2115 monitor_printf(mon, "Error saving global state\n");
2116 return ret;
2117 }
2118 vm_stop(RUN_STATE_SAVE_VM);
2119
2120 aio_context_acquire(aio_context);
2121
2122 memset(sn, 0, sizeof(*sn));
2123
2124 /* fill auxiliary fields */
2125 qemu_gettimeofday(&tv);
2126 sn->date_sec = tv.tv_sec;
2127 sn->date_nsec = tv.tv_usec * 1000;
2128 sn->vm_clock_nsec = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL);
2129
2130 if (name) {
2131 ret = bdrv_snapshot_find(bs, old_sn, name);
2132 if (ret >= 0) {
2133 pstrcpy(sn->name, sizeof(sn->name), old_sn->name);
2134 pstrcpy(sn->id_str, sizeof(sn->id_str), old_sn->id_str);
2135 } else {
2136 pstrcpy(sn->name, sizeof(sn->name), name);
2137 }
2138 } else {
2139 /* cast below needed for OpenBSD where tv_sec is still 'long' */
2140 localtime_r((const time_t *)&tv.tv_sec, &tm);
2141 strftime(sn->name, sizeof(sn->name), "vm-%Y%m%d%H%M%S", &tm);
2142 }
2143
2144 /* save the VM state */
2145 f = qemu_fopen_bdrv(bs, 1);
2146 if (!f) {
2147 monitor_printf(mon, "Could not open VM state file\n");
2148 goto the_end;
2149 }
2150 ret = qemu_savevm_state(f, &local_err);
2151 vm_state_size = qemu_ftell(f);
2152 qemu_fclose(f);
2153 if (ret < 0) {
2154 error_report_err(local_err);
2155 goto the_end;
2156 }
2157
2158 ret = bdrv_all_create_snapshot(sn, bs, vm_state_size, &bs);
2159 if (ret < 0) {
2160 monitor_printf(mon, "Error while creating snapshot on '%s'\n",
2161 bdrv_get_device_name(bs));
2162 goto the_end;
2163 }
2164
2165 ret = 0;
2166
2167 the_end:
2168 aio_context_release(aio_context);
2169 if (saved_vm_running) {
2170 vm_start();
2171 }
2172 return ret;
2173 }
2174
2175 void hmp_savevm(Monitor *mon, const QDict *qdict)
2176 {
2177 save_vmstate(mon, qdict_get_try_str(qdict, "name"));
2178 }
2179
2180 void qmp_xen_save_devices_state(const char *filename, Error **errp)
2181 {
2182 QEMUFile *f;
2183 QIOChannelFile *ioc;
2184 int saved_vm_running;
2185 int ret;
2186
2187 saved_vm_running = runstate_is_running();
2188 vm_stop(RUN_STATE_SAVE_VM);
2189 global_state_store_running();
2190
2191 ioc = qio_channel_file_new_path(filename, O_WRONLY | O_CREAT, 0660, errp);
2192 if (!ioc) {
2193 goto the_end;
2194 }
2195 qio_channel_set_name(QIO_CHANNEL(ioc), "migration-xen-save-state");
2196 f = qemu_fopen_channel_output(QIO_CHANNEL(ioc));
2197 ret = qemu_save_device_state(f);
2198 qemu_fclose(f);
2199 if (ret < 0) {
2200 error_setg(errp, QERR_IO_ERROR);
2201 }
2202
2203 the_end:
2204 if (saved_vm_running) {
2205 vm_start();
2206 }
2207 }
2208
2209 void qmp_xen_load_devices_state(const char *filename, Error **errp)
2210 {
2211 QEMUFile *f;
2212 QIOChannelFile *ioc;
2213 int ret;
2214
2215 /* Guest must be paused before loading the device state; the RAM state
2216 * will already have been loaded by xc
2217 */
2218 if (runstate_is_running()) {
2219 error_setg(errp, "Cannot update device state while vm is running");
2220 return;
2221 }
2222 vm_stop(RUN_STATE_RESTORE_VM);
2223
2224 ioc = qio_channel_file_new_path(filename, O_RDONLY | O_BINARY, 0, errp);
2225 if (!ioc) {
2226 return;
2227 }
2228 qio_channel_set_name(QIO_CHANNEL(ioc), "migration-xen-load-state");
2229 f = qemu_fopen_channel_input(QIO_CHANNEL(ioc));
2230
2231 ret = qemu_loadvm_state(f);
2232 qemu_fclose(f);
2233 if (ret < 0) {
2234 error_setg(errp, QERR_IO_ERROR);
2235 }
2236 migration_incoming_state_destroy();
2237 }
2238
2239 int load_vmstate(const char *name)
2240 {
2241 BlockDriverState *bs, *bs_vm_state;
2242 QEMUSnapshotInfo sn;
2243 QEMUFile *f;
2244 int ret;
2245 AioContext *aio_context;
2246 MigrationIncomingState *mis = migration_incoming_get_current();
2247
2248 if (!bdrv_all_can_snapshot(&bs)) {
2249 error_report("Device '%s' is writable but does not support snapshots.",
2250 bdrv_get_device_name(bs));
2251 return -ENOTSUP;
2252 }
2253 ret = bdrv_all_find_snapshot(name, &bs);
2254 if (ret < 0) {
2255 error_report("Device '%s' does not have the requested snapshot '%s'",
2256 bdrv_get_device_name(bs), name);
2257 return ret;
2258 }
2259
2260 bs_vm_state = bdrv_all_find_vmstate_bs();
2261 if (!bs_vm_state) {
2262 error_report("No block device supports snapshots");
2263 return -ENOTSUP;
2264 }
2265 aio_context = bdrv_get_aio_context(bs_vm_state);
2266
2267 /* Don't even try to load empty VM states */
2268 aio_context_acquire(aio_context);
2269 ret = bdrv_snapshot_find(bs_vm_state, &sn, name);
2270 aio_context_release(aio_context);
2271 if (ret < 0) {
2272 return ret;
2273 } else if (sn.vm_state_size == 0) {
2274 error_report("This is a disk-only snapshot. Revert to it offline "
2275 "using qemu-img.");
2276 return -EINVAL;
2277 }
2278
2279 /* Flush all IO requests so they don't interfere with the new state. */
2280 bdrv_drain_all();
2281
2282 ret = bdrv_all_goto_snapshot(name, &bs);
2283 if (ret < 0) {
2284 error_report("Error %d while activating snapshot '%s' on '%s'",
2285 ret, name, bdrv_get_device_name(bs));
2286 return ret;
2287 }
2288
2289 /* restore the VM state */
2290 f = qemu_fopen_bdrv(bs_vm_state, 0);
2291 if (!f) {
2292 error_report("Could not open VM state file");
2293 return -EINVAL;
2294 }
2295
2296 qemu_system_reset(VMRESET_SILENT);
2297 mis->from_src_file = f;
2298
2299 aio_context_acquire(aio_context);
2300 ret = qemu_loadvm_state(f);
2301 qemu_fclose(f);
2302 aio_context_release(aio_context);
2303
2304 migration_incoming_state_destroy();
2305 if (ret < 0) {
2306 error_report("Error %d while loading VM state", ret);
2307 return ret;
2308 }
2309
2310 return 0;
2311 }
2312
2313 void hmp_delvm(Monitor *mon, const QDict *qdict)
2314 {
2315 BlockDriverState *bs;
2316 Error *err;
2317 const char *name = qdict_get_str(qdict, "name");
2318
2319 if (bdrv_all_delete_snapshot(name, &bs, &err) < 0) {
2320 error_reportf_err(err,
2321 "Error while deleting snapshot on device '%s': ",
2322 bdrv_get_device_name(bs));
2323 }
2324 }
2325
2326 void hmp_info_snapshots(Monitor *mon, const QDict *qdict)
2327 {
2328 BlockDriverState *bs, *bs1;
2329 BdrvNextIterator it1;
2330 QEMUSnapshotInfo *sn_tab, *sn;
2331 bool no_snapshot = true;
2332 int nb_sns, i;
2333 int total;
2334 int *global_snapshots;
2335 AioContext *aio_context;
2336
2337 typedef struct SnapshotEntry {
2338 QEMUSnapshotInfo sn;
2339 QTAILQ_ENTRY(SnapshotEntry) next;
2340 } SnapshotEntry;
2341
2342 typedef struct ImageEntry {
2343 const char *imagename;
2344 QTAILQ_ENTRY(ImageEntry) next;
2345 QTAILQ_HEAD(, SnapshotEntry) snapshots;
2346 } ImageEntry;
2347
2348 QTAILQ_HEAD(, ImageEntry) image_list =
2349 QTAILQ_HEAD_INITIALIZER(image_list);
2350
2351 ImageEntry *image_entry, *next_ie;
2352 SnapshotEntry *snapshot_entry;
2353
2354 bs = bdrv_all_find_vmstate_bs();
2355 if (!bs) {
2356 monitor_printf(mon, "No available block device supports snapshots\n");
2357 return;
2358 }
2359 aio_context = bdrv_get_aio_context(bs);
2360
2361 aio_context_acquire(aio_context);
2362 nb_sns = bdrv_snapshot_list(bs, &sn_tab);
2363 aio_context_release(aio_context);
2364
2365 if (nb_sns < 0) {
2366 monitor_printf(mon, "bdrv_snapshot_list: error %d\n", nb_sns);
2367 return;
2368 }
2369
2370 for (bs1 = bdrv_first(&it1); bs1; bs1 = bdrv_next(&it1)) {
2371 int bs1_nb_sns = 0;
2372 ImageEntry *ie;
2373 SnapshotEntry *se;
2374 AioContext *ctx = bdrv_get_aio_context(bs1);
2375
2376 aio_context_acquire(ctx);
2377 if (bdrv_can_snapshot(bs1)) {
2378 sn = NULL;
2379 bs1_nb_sns = bdrv_snapshot_list(bs1, &sn);
2380 if (bs1_nb_sns > 0) {
2381 no_snapshot = false;
2382 ie = g_new0(ImageEntry, 1);
2383 ie->imagename = bdrv_get_device_name(bs1);
2384 QTAILQ_INIT(&ie->snapshots);
2385 QTAILQ_INSERT_TAIL(&image_list, ie, next);
2386 for (i = 0; i < bs1_nb_sns; i++) {
2387 se = g_new0(SnapshotEntry, 1);
2388 se->sn = sn[i];
2389 QTAILQ_INSERT_TAIL(&ie->snapshots, se, next);
2390 }
2391 }
2392 g_free(sn);
2393 }
2394 aio_context_release(ctx);
2395 }
2396
2397 if (no_snapshot) {
2398 monitor_printf(mon, "There is no snapshot available.\n");
2399 return;
2400 }
2401
2402 global_snapshots = g_new0(int, nb_sns);
2403 total = 0;
2404 for (i = 0; i < nb_sns; i++) {
2405 SnapshotEntry *next_sn;
2406 if (bdrv_all_find_snapshot(sn_tab[i].name, &bs1) == 0) {
2407 global_snapshots[total] = i;
2408 total++;
2409 QTAILQ_FOREACH(image_entry, &image_list, next) {
2410 QTAILQ_FOREACH_SAFE(snapshot_entry, &image_entry->snapshots,
2411 next, next_sn) {
2412 if (!strcmp(sn_tab[i].name, snapshot_entry->sn.name)) {
2413 QTAILQ_REMOVE(&image_entry->snapshots, snapshot_entry,
2414 next);
2415 g_free(snapshot_entry);
2416 }
2417 }
2418 }
2419 }
2420 }
2421
2422 monitor_printf(mon, "List of snapshots present on all disks:\n");
2423
2424 if (total > 0) {
2425 bdrv_snapshot_dump((fprintf_function)monitor_printf, mon, NULL);
2426 monitor_printf(mon, "\n");
2427 for (i = 0; i < total; i++) {
2428 sn = &sn_tab[global_snapshots[i]];
2429 /* The ID is not guaranteed to be the same on all images, so
2430 * overwrite it.
2431 */
2432 pstrcpy(sn->id_str, sizeof(sn->id_str), "--");
2433 bdrv_snapshot_dump((fprintf_function)monitor_printf, mon, sn);
2434 monitor_printf(mon, "\n");
2435 }
2436 } else {
2437 monitor_printf(mon, "None\n");
2438 }
2439
2440 QTAILQ_FOREACH(image_entry, &image_list, next) {
2441 if (QTAILQ_EMPTY(&image_entry->snapshots)) {
2442 continue;
2443 }
2444 monitor_printf(mon,
2445 "\nList of partial (non-loadable) snapshots on '%s':\n",
2446 image_entry->imagename);
2447 bdrv_snapshot_dump((fprintf_function)monitor_printf, mon, NULL);
2448 monitor_printf(mon, "\n");
2449 QTAILQ_FOREACH(snapshot_entry, &image_entry->snapshots, next) {
2450 bdrv_snapshot_dump((fprintf_function)monitor_printf, mon,
2451 &snapshot_entry->sn);
2452 monitor_printf(mon, "\n");
2453 }
2454 }
2455
2456 QTAILQ_FOREACH_SAFE(image_entry, &image_list, next, next_ie) {
2457 SnapshotEntry *next_sn;
2458 QTAILQ_FOREACH_SAFE(snapshot_entry, &image_entry->snapshots, next,
2459 next_sn) {
2460 g_free(snapshot_entry);
2461 }
2462 g_free(image_entry);
2463 }
2464 g_free(sn_tab);
2465 g_free(global_snapshots);
2466
2467 }
2468
2469 void vmstate_register_ram(MemoryRegion *mr, DeviceState *dev)
2470 {
2471 qemu_ram_set_idstr(mr->ram_block,
2472 memory_region_name(mr), dev);
2473 }
2474
2475 void vmstate_unregister_ram(MemoryRegion *mr, DeviceState *dev)
2476 {
2477 qemu_ram_unset_idstr(mr->ram_block);
2478 }
2479
2480 void vmstate_register_ram_global(MemoryRegion *mr)
2481 {
2482 vmstate_register_ram(mr, NULL);
2483 }