Merge tag 'linux-user-for-7.1-pull-request' of https://gitlab.com/laurent_vivier...
[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 "hw/boards.h"
31 #include "net/net.h"
32 #include "migration.h"
33 #include "migration/snapshot.h"
34 #include "migration/vmstate.h"
35 #include "migration/misc.h"
36 #include "migration/register.h"
37 #include "migration/global_state.h"
38 #include "ram.h"
39 #include "qemu-file-channel.h"
40 #include "qemu-file.h"
41 #include "savevm.h"
42 #include "postcopy-ram.h"
43 #include "qapi/error.h"
44 #include "qapi/qapi-commands-migration.h"
45 #include "qapi/qmp/json-writer.h"
46 #include "qapi/clone-visitor.h"
47 #include "qapi/qapi-builtin-visit.h"
48 #include "qapi/qmp/qerror.h"
49 #include "qemu/error-report.h"
50 #include "sysemu/cpus.h"
51 #include "exec/memory.h"
52 #include "exec/target_page.h"
53 #include "trace.h"
54 #include "qemu/iov.h"
55 #include "qemu/main-loop.h"
56 #include "block/snapshot.h"
57 #include "qemu/cutils.h"
58 #include "io/channel-buffer.h"
59 #include "io/channel-file.h"
60 #include "sysemu/replay.h"
61 #include "sysemu/runstate.h"
62 #include "sysemu/sysemu.h"
63 #include "sysemu/xen.h"
64 #include "migration/colo.h"
65 #include "qemu/bitmap.h"
66 #include "net/announce.h"
67 #include "qemu/yank.h"
68 #include "yank_functions.h"
69
70 const unsigned int postcopy_ram_discard_version;
71
72 /* Subcommands for QEMU_VM_COMMAND */
73 enum qemu_vm_cmd {
74 MIG_CMD_INVALID = 0, /* Must be 0 */
75 MIG_CMD_OPEN_RETURN_PATH, /* Tell the dest to open the Return path */
76 MIG_CMD_PING, /* Request a PONG on the RP */
77
78 MIG_CMD_POSTCOPY_ADVISE, /* Prior to any page transfers, just
79 warn we might want to do PC */
80 MIG_CMD_POSTCOPY_LISTEN, /* Start listening for incoming
81 pages as it's running. */
82 MIG_CMD_POSTCOPY_RUN, /* Start execution */
83
84 MIG_CMD_POSTCOPY_RAM_DISCARD, /* A list of pages to discard that
85 were previously sent during
86 precopy but are dirty. */
87 MIG_CMD_PACKAGED, /* Send a wrapped stream within this stream */
88 MIG_CMD_ENABLE_COLO, /* Enable COLO */
89 MIG_CMD_POSTCOPY_RESUME, /* resume postcopy on dest */
90 MIG_CMD_RECV_BITMAP, /* Request for recved bitmap on dst */
91 MIG_CMD_MAX
92 };
93
94 #define MAX_VM_CMD_PACKAGED_SIZE UINT32_MAX
95 static struct mig_cmd_args {
96 ssize_t len; /* -1 = variable */
97 const char *name;
98 } mig_cmd_args[] = {
99 [MIG_CMD_INVALID] = { .len = -1, .name = "INVALID" },
100 [MIG_CMD_OPEN_RETURN_PATH] = { .len = 0, .name = "OPEN_RETURN_PATH" },
101 [MIG_CMD_PING] = { .len = sizeof(uint32_t), .name = "PING" },
102 [MIG_CMD_POSTCOPY_ADVISE] = { .len = -1, .name = "POSTCOPY_ADVISE" },
103 [MIG_CMD_POSTCOPY_LISTEN] = { .len = 0, .name = "POSTCOPY_LISTEN" },
104 [MIG_CMD_POSTCOPY_RUN] = { .len = 0, .name = "POSTCOPY_RUN" },
105 [MIG_CMD_POSTCOPY_RAM_DISCARD] = {
106 .len = -1, .name = "POSTCOPY_RAM_DISCARD" },
107 [MIG_CMD_POSTCOPY_RESUME] = { .len = 0, .name = "POSTCOPY_RESUME" },
108 [MIG_CMD_PACKAGED] = { .len = 4, .name = "PACKAGED" },
109 [MIG_CMD_RECV_BITMAP] = { .len = -1, .name = "RECV_BITMAP" },
110 [MIG_CMD_MAX] = { .len = -1, .name = "MAX" },
111 };
112
113 /* Note for MIG_CMD_POSTCOPY_ADVISE:
114 * The format of arguments is depending on postcopy mode:
115 * - postcopy RAM only
116 * uint64_t host page size
117 * uint64_t taget page size
118 *
119 * - postcopy RAM and postcopy dirty bitmaps
120 * format is the same as for postcopy RAM only
121 *
122 * - postcopy dirty bitmaps only
123 * Nothing. Command length field is 0.
124 *
125 * Be careful: adding a new postcopy entity with some other parameters should
126 * not break format self-description ability. Good way is to introduce some
127 * generic extendable format with an exception for two old entities.
128 */
129
130 /***********************************************************/
131 /* savevm/loadvm support */
132
133 static ssize_t block_writev_buffer(void *opaque, struct iovec *iov, int iovcnt,
134 int64_t pos, Error **errp)
135 {
136 int ret;
137 QEMUIOVector qiov;
138
139 qemu_iovec_init_external(&qiov, iov, iovcnt);
140 ret = bdrv_writev_vmstate(opaque, &qiov, pos);
141 if (ret < 0) {
142 return ret;
143 }
144
145 return qiov.size;
146 }
147
148 static ssize_t block_get_buffer(void *opaque, uint8_t *buf, int64_t pos,
149 size_t size, Error **errp)
150 {
151 return bdrv_load_vmstate(opaque, buf, pos, size);
152 }
153
154 static int bdrv_fclose(void *opaque, Error **errp)
155 {
156 return bdrv_flush(opaque);
157 }
158
159 static const QEMUFileOps bdrv_read_ops = {
160 .get_buffer = block_get_buffer,
161 .close = bdrv_fclose
162 };
163
164 static const QEMUFileOps bdrv_write_ops = {
165 .writev_buffer = block_writev_buffer,
166 .close = bdrv_fclose
167 };
168
169 static QEMUFile *qemu_fopen_bdrv(BlockDriverState *bs, int is_writable)
170 {
171 if (is_writable) {
172 return qemu_fopen_ops(bs, &bdrv_write_ops, false);
173 }
174 return qemu_fopen_ops(bs, &bdrv_read_ops, false);
175 }
176
177
178 /* QEMUFile timer support.
179 * Not in qemu-file.c to not add qemu-timer.c as dependency to qemu-file.c
180 */
181
182 void timer_put(QEMUFile *f, QEMUTimer *ts)
183 {
184 uint64_t expire_time;
185
186 expire_time = timer_expire_time_ns(ts);
187 qemu_put_be64(f, expire_time);
188 }
189
190 void timer_get(QEMUFile *f, QEMUTimer *ts)
191 {
192 uint64_t expire_time;
193
194 expire_time = qemu_get_be64(f);
195 if (expire_time != -1) {
196 timer_mod_ns(ts, expire_time);
197 } else {
198 timer_del(ts);
199 }
200 }
201
202
203 /* VMState timer support.
204 * Not in vmstate.c to not add qemu-timer.c as dependency to vmstate.c
205 */
206
207 static int get_timer(QEMUFile *f, void *pv, size_t size,
208 const VMStateField *field)
209 {
210 QEMUTimer *v = pv;
211 timer_get(f, v);
212 return 0;
213 }
214
215 static int put_timer(QEMUFile *f, void *pv, size_t size,
216 const VMStateField *field, JSONWriter *vmdesc)
217 {
218 QEMUTimer *v = pv;
219 timer_put(f, v);
220
221 return 0;
222 }
223
224 const VMStateInfo vmstate_info_timer = {
225 .name = "timer",
226 .get = get_timer,
227 .put = put_timer,
228 };
229
230
231 typedef struct CompatEntry {
232 char idstr[256];
233 int instance_id;
234 } CompatEntry;
235
236 typedef struct SaveStateEntry {
237 QTAILQ_ENTRY(SaveStateEntry) entry;
238 char idstr[256];
239 uint32_t instance_id;
240 int alias_id;
241 int version_id;
242 /* version id read from the stream */
243 int load_version_id;
244 int section_id;
245 /* section id read from the stream */
246 int load_section_id;
247 const SaveVMHandlers *ops;
248 const VMStateDescription *vmsd;
249 void *opaque;
250 CompatEntry *compat;
251 int is_ram;
252 } SaveStateEntry;
253
254 typedef struct SaveState {
255 QTAILQ_HEAD(, SaveStateEntry) handlers;
256 SaveStateEntry *handler_pri_head[MIG_PRI_MAX + 1];
257 int global_section_id;
258 uint32_t len;
259 const char *name;
260 uint32_t target_page_bits;
261 uint32_t caps_count;
262 MigrationCapability *capabilities;
263 QemuUUID uuid;
264 } SaveState;
265
266 static SaveState savevm_state = {
267 .handlers = QTAILQ_HEAD_INITIALIZER(savevm_state.handlers),
268 .handler_pri_head = { [MIG_PRI_DEFAULT ... MIG_PRI_MAX] = NULL },
269 .global_section_id = 0,
270 };
271
272 static bool should_validate_capability(int capability)
273 {
274 assert(capability >= 0 && capability < MIGRATION_CAPABILITY__MAX);
275 /* Validate only new capabilities to keep compatibility. */
276 switch (capability) {
277 case MIGRATION_CAPABILITY_X_IGNORE_SHARED:
278 return true;
279 default:
280 return false;
281 }
282 }
283
284 static uint32_t get_validatable_capabilities_count(void)
285 {
286 MigrationState *s = migrate_get_current();
287 uint32_t result = 0;
288 int i;
289 for (i = 0; i < MIGRATION_CAPABILITY__MAX; i++) {
290 if (should_validate_capability(i) && s->enabled_capabilities[i]) {
291 result++;
292 }
293 }
294 return result;
295 }
296
297 static int configuration_pre_save(void *opaque)
298 {
299 SaveState *state = opaque;
300 const char *current_name = MACHINE_GET_CLASS(current_machine)->name;
301 MigrationState *s = migrate_get_current();
302 int i, j;
303
304 state->len = strlen(current_name);
305 state->name = current_name;
306 state->target_page_bits = qemu_target_page_bits();
307
308 state->caps_count = get_validatable_capabilities_count();
309 state->capabilities = g_renew(MigrationCapability, state->capabilities,
310 state->caps_count);
311 for (i = j = 0; i < MIGRATION_CAPABILITY__MAX; i++) {
312 if (should_validate_capability(i) && s->enabled_capabilities[i]) {
313 state->capabilities[j++] = i;
314 }
315 }
316 state->uuid = qemu_uuid;
317
318 return 0;
319 }
320
321 static int configuration_post_save(void *opaque)
322 {
323 SaveState *state = opaque;
324
325 g_free(state->capabilities);
326 state->capabilities = NULL;
327 state->caps_count = 0;
328 return 0;
329 }
330
331 static int configuration_pre_load(void *opaque)
332 {
333 SaveState *state = opaque;
334
335 /* If there is no target-page-bits subsection it means the source
336 * predates the variable-target-page-bits support and is using the
337 * minimum possible value for this CPU.
338 */
339 state->target_page_bits = qemu_target_page_bits_min();
340 return 0;
341 }
342
343 static bool configuration_validate_capabilities(SaveState *state)
344 {
345 bool ret = true;
346 MigrationState *s = migrate_get_current();
347 unsigned long *source_caps_bm;
348 int i;
349
350 source_caps_bm = bitmap_new(MIGRATION_CAPABILITY__MAX);
351 for (i = 0; i < state->caps_count; i++) {
352 MigrationCapability capability = state->capabilities[i];
353 set_bit(capability, source_caps_bm);
354 }
355
356 for (i = 0; i < MIGRATION_CAPABILITY__MAX; i++) {
357 bool source_state, target_state;
358 if (!should_validate_capability(i)) {
359 continue;
360 }
361 source_state = test_bit(i, source_caps_bm);
362 target_state = s->enabled_capabilities[i];
363 if (source_state != target_state) {
364 error_report("Capability %s is %s, but received capability is %s",
365 MigrationCapability_str(i),
366 target_state ? "on" : "off",
367 source_state ? "on" : "off");
368 ret = false;
369 /* Don't break here to report all failed capabilities */
370 }
371 }
372
373 g_free(source_caps_bm);
374 return ret;
375 }
376
377 static int configuration_post_load(void *opaque, int version_id)
378 {
379 SaveState *state = opaque;
380 const char *current_name = MACHINE_GET_CLASS(current_machine)->name;
381 int ret = 0;
382
383 if (strncmp(state->name, current_name, state->len) != 0) {
384 error_report("Machine type received is '%.*s' and local is '%s'",
385 (int) state->len, state->name, current_name);
386 ret = -EINVAL;
387 goto out;
388 }
389
390 if (state->target_page_bits != qemu_target_page_bits()) {
391 error_report("Received TARGET_PAGE_BITS is %d but local is %d",
392 state->target_page_bits, qemu_target_page_bits());
393 ret = -EINVAL;
394 goto out;
395 }
396
397 if (!configuration_validate_capabilities(state)) {
398 ret = -EINVAL;
399 goto out;
400 }
401
402 out:
403 g_free((void *)state->name);
404 state->name = NULL;
405 state->len = 0;
406 g_free(state->capabilities);
407 state->capabilities = NULL;
408 state->caps_count = 0;
409
410 return ret;
411 }
412
413 static int get_capability(QEMUFile *f, void *pv, size_t size,
414 const VMStateField *field)
415 {
416 MigrationCapability *capability = pv;
417 char capability_str[UINT8_MAX + 1];
418 uint8_t len;
419 int i;
420
421 len = qemu_get_byte(f);
422 qemu_get_buffer(f, (uint8_t *)capability_str, len);
423 capability_str[len] = '\0';
424 for (i = 0; i < MIGRATION_CAPABILITY__MAX; i++) {
425 if (!strcmp(MigrationCapability_str(i), capability_str)) {
426 *capability = i;
427 return 0;
428 }
429 }
430 error_report("Received unknown capability %s", capability_str);
431 return -EINVAL;
432 }
433
434 static int put_capability(QEMUFile *f, void *pv, size_t size,
435 const VMStateField *field, JSONWriter *vmdesc)
436 {
437 MigrationCapability *capability = pv;
438 const char *capability_str = MigrationCapability_str(*capability);
439 size_t len = strlen(capability_str);
440 assert(len <= UINT8_MAX);
441
442 qemu_put_byte(f, len);
443 qemu_put_buffer(f, (uint8_t *)capability_str, len);
444 return 0;
445 }
446
447 static const VMStateInfo vmstate_info_capability = {
448 .name = "capability",
449 .get = get_capability,
450 .put = put_capability,
451 };
452
453 /* The target-page-bits subsection is present only if the
454 * target page size is not the same as the default (ie the
455 * minimum page size for a variable-page-size guest CPU).
456 * If it is present then it contains the actual target page
457 * bits for the machine, and migration will fail if the
458 * two ends don't agree about it.
459 */
460 static bool vmstate_target_page_bits_needed(void *opaque)
461 {
462 return qemu_target_page_bits()
463 > qemu_target_page_bits_min();
464 }
465
466 static const VMStateDescription vmstate_target_page_bits = {
467 .name = "configuration/target-page-bits",
468 .version_id = 1,
469 .minimum_version_id = 1,
470 .needed = vmstate_target_page_bits_needed,
471 .fields = (VMStateField[]) {
472 VMSTATE_UINT32(target_page_bits, SaveState),
473 VMSTATE_END_OF_LIST()
474 }
475 };
476
477 static bool vmstate_capabilites_needed(void *opaque)
478 {
479 return get_validatable_capabilities_count() > 0;
480 }
481
482 static const VMStateDescription vmstate_capabilites = {
483 .name = "configuration/capabilities",
484 .version_id = 1,
485 .minimum_version_id = 1,
486 .needed = vmstate_capabilites_needed,
487 .fields = (VMStateField[]) {
488 VMSTATE_UINT32_V(caps_count, SaveState, 1),
489 VMSTATE_VARRAY_UINT32_ALLOC(capabilities, SaveState, caps_count, 1,
490 vmstate_info_capability,
491 MigrationCapability),
492 VMSTATE_END_OF_LIST()
493 }
494 };
495
496 static bool vmstate_uuid_needed(void *opaque)
497 {
498 return qemu_uuid_set && migrate_validate_uuid();
499 }
500
501 static int vmstate_uuid_post_load(void *opaque, int version_id)
502 {
503 SaveState *state = opaque;
504 char uuid_src[UUID_FMT_LEN + 1];
505 char uuid_dst[UUID_FMT_LEN + 1];
506
507 if (!qemu_uuid_set) {
508 /*
509 * It's warning because user might not know UUID in some cases,
510 * e.g. load an old snapshot
511 */
512 qemu_uuid_unparse(&state->uuid, uuid_src);
513 warn_report("UUID is received %s, but local uuid isn't set",
514 uuid_src);
515 return 0;
516 }
517 if (!qemu_uuid_is_equal(&state->uuid, &qemu_uuid)) {
518 qemu_uuid_unparse(&state->uuid, uuid_src);
519 qemu_uuid_unparse(&qemu_uuid, uuid_dst);
520 error_report("UUID received is %s and local is %s", uuid_src, uuid_dst);
521 return -EINVAL;
522 }
523 return 0;
524 }
525
526 static const VMStateDescription vmstate_uuid = {
527 .name = "configuration/uuid",
528 .version_id = 1,
529 .minimum_version_id = 1,
530 .needed = vmstate_uuid_needed,
531 .post_load = vmstate_uuid_post_load,
532 .fields = (VMStateField[]) {
533 VMSTATE_UINT8_ARRAY_V(uuid.data, SaveState, sizeof(QemuUUID), 1),
534 VMSTATE_END_OF_LIST()
535 }
536 };
537
538 static const VMStateDescription vmstate_configuration = {
539 .name = "configuration",
540 .version_id = 1,
541 .pre_load = configuration_pre_load,
542 .post_load = configuration_post_load,
543 .pre_save = configuration_pre_save,
544 .post_save = configuration_post_save,
545 .fields = (VMStateField[]) {
546 VMSTATE_UINT32(len, SaveState),
547 VMSTATE_VBUFFER_ALLOC_UINT32(name, SaveState, 0, NULL, len),
548 VMSTATE_END_OF_LIST()
549 },
550 .subsections = (const VMStateDescription *[]) {
551 &vmstate_target_page_bits,
552 &vmstate_capabilites,
553 &vmstate_uuid,
554 NULL
555 }
556 };
557
558 static void dump_vmstate_vmsd(FILE *out_file,
559 const VMStateDescription *vmsd, int indent,
560 bool is_subsection);
561
562 static void dump_vmstate_vmsf(FILE *out_file, const VMStateField *field,
563 int indent)
564 {
565 fprintf(out_file, "%*s{\n", indent, "");
566 indent += 2;
567 fprintf(out_file, "%*s\"field\": \"%s\",\n", indent, "", field->name);
568 fprintf(out_file, "%*s\"version_id\": %d,\n", indent, "",
569 field->version_id);
570 fprintf(out_file, "%*s\"field_exists\": %s,\n", indent, "",
571 field->field_exists ? "true" : "false");
572 fprintf(out_file, "%*s\"size\": %zu", indent, "", field->size);
573 if (field->vmsd != NULL) {
574 fprintf(out_file, ",\n");
575 dump_vmstate_vmsd(out_file, field->vmsd, indent, false);
576 }
577 fprintf(out_file, "\n%*s}", indent - 2, "");
578 }
579
580 static void dump_vmstate_vmss(FILE *out_file,
581 const VMStateDescription **subsection,
582 int indent)
583 {
584 if (*subsection != NULL) {
585 dump_vmstate_vmsd(out_file, *subsection, indent, true);
586 }
587 }
588
589 static void dump_vmstate_vmsd(FILE *out_file,
590 const VMStateDescription *vmsd, int indent,
591 bool is_subsection)
592 {
593 if (is_subsection) {
594 fprintf(out_file, "%*s{\n", indent, "");
595 } else {
596 fprintf(out_file, "%*s\"%s\": {\n", indent, "", "Description");
597 }
598 indent += 2;
599 fprintf(out_file, "%*s\"name\": \"%s\",\n", indent, "", vmsd->name);
600 fprintf(out_file, "%*s\"version_id\": %d,\n", indent, "",
601 vmsd->version_id);
602 fprintf(out_file, "%*s\"minimum_version_id\": %d", indent, "",
603 vmsd->minimum_version_id);
604 if (vmsd->fields != NULL) {
605 const VMStateField *field = vmsd->fields;
606 bool first;
607
608 fprintf(out_file, ",\n%*s\"Fields\": [\n", indent, "");
609 first = true;
610 while (field->name != NULL) {
611 if (field->flags & VMS_MUST_EXIST) {
612 /* Ignore VMSTATE_VALIDATE bits; these don't get migrated */
613 field++;
614 continue;
615 }
616 if (!first) {
617 fprintf(out_file, ",\n");
618 }
619 dump_vmstate_vmsf(out_file, field, indent + 2);
620 field++;
621 first = false;
622 }
623 fprintf(out_file, "\n%*s]", indent, "");
624 }
625 if (vmsd->subsections != NULL) {
626 const VMStateDescription **subsection = vmsd->subsections;
627 bool first;
628
629 fprintf(out_file, ",\n%*s\"Subsections\": [\n", indent, "");
630 first = true;
631 while (*subsection != NULL) {
632 if (!first) {
633 fprintf(out_file, ",\n");
634 }
635 dump_vmstate_vmss(out_file, subsection, indent + 2);
636 subsection++;
637 first = false;
638 }
639 fprintf(out_file, "\n%*s]", indent, "");
640 }
641 fprintf(out_file, "\n%*s}", indent - 2, "");
642 }
643
644 static void dump_machine_type(FILE *out_file)
645 {
646 MachineClass *mc;
647
648 mc = MACHINE_GET_CLASS(current_machine);
649
650 fprintf(out_file, " \"vmschkmachine\": {\n");
651 fprintf(out_file, " \"Name\": \"%s\"\n", mc->name);
652 fprintf(out_file, " },\n");
653 }
654
655 void dump_vmstate_json_to_file(FILE *out_file)
656 {
657 GSList *list, *elt;
658 bool first;
659
660 fprintf(out_file, "{\n");
661 dump_machine_type(out_file);
662
663 first = true;
664 list = object_class_get_list(TYPE_DEVICE, true);
665 for (elt = list; elt; elt = elt->next) {
666 DeviceClass *dc = OBJECT_CLASS_CHECK(DeviceClass, elt->data,
667 TYPE_DEVICE);
668 const char *name;
669 int indent = 2;
670
671 if (!dc->vmsd) {
672 continue;
673 }
674
675 if (!first) {
676 fprintf(out_file, ",\n");
677 }
678 name = object_class_get_name(OBJECT_CLASS(dc));
679 fprintf(out_file, "%*s\"%s\": {\n", indent, "", name);
680 indent += 2;
681 fprintf(out_file, "%*s\"Name\": \"%s\",\n", indent, "", name);
682 fprintf(out_file, "%*s\"version_id\": %d,\n", indent, "",
683 dc->vmsd->version_id);
684 fprintf(out_file, "%*s\"minimum_version_id\": %d,\n", indent, "",
685 dc->vmsd->minimum_version_id);
686
687 dump_vmstate_vmsd(out_file, dc->vmsd, indent, false);
688
689 fprintf(out_file, "\n%*s}", indent - 2, "");
690 first = false;
691 }
692 fprintf(out_file, "\n}\n");
693 fclose(out_file);
694 g_slist_free(list);
695 }
696
697 static uint32_t calculate_new_instance_id(const char *idstr)
698 {
699 SaveStateEntry *se;
700 uint32_t instance_id = 0;
701
702 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
703 if (strcmp(idstr, se->idstr) == 0
704 && instance_id <= se->instance_id) {
705 instance_id = se->instance_id + 1;
706 }
707 }
708 /* Make sure we never loop over without being noticed */
709 assert(instance_id != VMSTATE_INSTANCE_ID_ANY);
710 return instance_id;
711 }
712
713 static int calculate_compat_instance_id(const char *idstr)
714 {
715 SaveStateEntry *se;
716 int instance_id = 0;
717
718 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
719 if (!se->compat) {
720 continue;
721 }
722
723 if (strcmp(idstr, se->compat->idstr) == 0
724 && instance_id <= se->compat->instance_id) {
725 instance_id = se->compat->instance_id + 1;
726 }
727 }
728 return instance_id;
729 }
730
731 static inline MigrationPriority save_state_priority(SaveStateEntry *se)
732 {
733 if (se->vmsd) {
734 return se->vmsd->priority;
735 }
736 return MIG_PRI_DEFAULT;
737 }
738
739 static void savevm_state_handler_insert(SaveStateEntry *nse)
740 {
741 MigrationPriority priority = save_state_priority(nse);
742 SaveStateEntry *se;
743 int i;
744
745 assert(priority <= MIG_PRI_MAX);
746
747 for (i = priority - 1; i >= 0; i--) {
748 se = savevm_state.handler_pri_head[i];
749 if (se != NULL) {
750 assert(save_state_priority(se) < priority);
751 break;
752 }
753 }
754
755 if (i >= 0) {
756 QTAILQ_INSERT_BEFORE(se, nse, entry);
757 } else {
758 QTAILQ_INSERT_TAIL(&savevm_state.handlers, nse, entry);
759 }
760
761 if (savevm_state.handler_pri_head[priority] == NULL) {
762 savevm_state.handler_pri_head[priority] = nse;
763 }
764 }
765
766 static void savevm_state_handler_remove(SaveStateEntry *se)
767 {
768 SaveStateEntry *next;
769 MigrationPriority priority = save_state_priority(se);
770
771 if (se == savevm_state.handler_pri_head[priority]) {
772 next = QTAILQ_NEXT(se, entry);
773 if (next != NULL && save_state_priority(next) == priority) {
774 savevm_state.handler_pri_head[priority] = next;
775 } else {
776 savevm_state.handler_pri_head[priority] = NULL;
777 }
778 }
779 QTAILQ_REMOVE(&savevm_state.handlers, se, entry);
780 }
781
782 /* TODO: Individual devices generally have very little idea about the rest
783 of the system, so instance_id should be removed/replaced.
784 Meanwhile pass -1 as instance_id if you do not already have a clearly
785 distinguishing id for all instances of your device class. */
786 int register_savevm_live(const char *idstr,
787 uint32_t instance_id,
788 int version_id,
789 const SaveVMHandlers *ops,
790 void *opaque)
791 {
792 SaveStateEntry *se;
793
794 se = g_new0(SaveStateEntry, 1);
795 se->version_id = version_id;
796 se->section_id = savevm_state.global_section_id++;
797 se->ops = ops;
798 se->opaque = opaque;
799 se->vmsd = NULL;
800 /* if this is a live_savem then set is_ram */
801 if (ops->save_setup != NULL) {
802 se->is_ram = 1;
803 }
804
805 pstrcat(se->idstr, sizeof(se->idstr), idstr);
806
807 if (instance_id == VMSTATE_INSTANCE_ID_ANY) {
808 se->instance_id = calculate_new_instance_id(se->idstr);
809 } else {
810 se->instance_id = instance_id;
811 }
812 assert(!se->compat || se->instance_id == 0);
813 savevm_state_handler_insert(se);
814 return 0;
815 }
816
817 void unregister_savevm(VMStateIf *obj, const char *idstr, void *opaque)
818 {
819 SaveStateEntry *se, *new_se;
820 char id[256] = "";
821
822 if (obj) {
823 char *oid = vmstate_if_get_id(obj);
824 if (oid) {
825 pstrcpy(id, sizeof(id), oid);
826 pstrcat(id, sizeof(id), "/");
827 g_free(oid);
828 }
829 }
830 pstrcat(id, sizeof(id), idstr);
831
832 QTAILQ_FOREACH_SAFE(se, &savevm_state.handlers, entry, new_se) {
833 if (strcmp(se->idstr, id) == 0 && se->opaque == opaque) {
834 savevm_state_handler_remove(se);
835 g_free(se->compat);
836 g_free(se);
837 }
838 }
839 }
840
841 int vmstate_register_with_alias_id(VMStateIf *obj, uint32_t instance_id,
842 const VMStateDescription *vmsd,
843 void *opaque, int alias_id,
844 int required_for_version,
845 Error **errp)
846 {
847 SaveStateEntry *se;
848
849 /* If this triggers, alias support can be dropped for the vmsd. */
850 assert(alias_id == -1 || required_for_version >= vmsd->minimum_version_id);
851
852 se = g_new0(SaveStateEntry, 1);
853 se->version_id = vmsd->version_id;
854 se->section_id = savevm_state.global_section_id++;
855 se->opaque = opaque;
856 se->vmsd = vmsd;
857 se->alias_id = alias_id;
858
859 if (obj) {
860 char *id = vmstate_if_get_id(obj);
861 if (id) {
862 if (snprintf(se->idstr, sizeof(se->idstr), "%s/", id) >=
863 sizeof(se->idstr)) {
864 error_setg(errp, "Path too long for VMState (%s)", id);
865 g_free(id);
866 g_free(se);
867
868 return -1;
869 }
870 g_free(id);
871
872 se->compat = g_new0(CompatEntry, 1);
873 pstrcpy(se->compat->idstr, sizeof(se->compat->idstr), vmsd->name);
874 se->compat->instance_id = instance_id == VMSTATE_INSTANCE_ID_ANY ?
875 calculate_compat_instance_id(vmsd->name) : instance_id;
876 instance_id = VMSTATE_INSTANCE_ID_ANY;
877 }
878 }
879 pstrcat(se->idstr, sizeof(se->idstr), vmsd->name);
880
881 if (instance_id == VMSTATE_INSTANCE_ID_ANY) {
882 se->instance_id = calculate_new_instance_id(se->idstr);
883 } else {
884 se->instance_id = instance_id;
885 }
886 assert(!se->compat || se->instance_id == 0);
887 savevm_state_handler_insert(se);
888 return 0;
889 }
890
891 void vmstate_unregister(VMStateIf *obj, const VMStateDescription *vmsd,
892 void *opaque)
893 {
894 SaveStateEntry *se, *new_se;
895
896 QTAILQ_FOREACH_SAFE(se, &savevm_state.handlers, entry, new_se) {
897 if (se->vmsd == vmsd && se->opaque == opaque) {
898 savevm_state_handler_remove(se);
899 g_free(se->compat);
900 g_free(se);
901 }
902 }
903 }
904
905 static int vmstate_load(QEMUFile *f, SaveStateEntry *se)
906 {
907 trace_vmstate_load(se->idstr, se->vmsd ? se->vmsd->name : "(old)");
908 if (!se->vmsd) { /* Old style */
909 return se->ops->load_state(f, se->opaque, se->load_version_id);
910 }
911 return vmstate_load_state(f, se->vmsd, se->opaque, se->load_version_id);
912 }
913
914 static void vmstate_save_old_style(QEMUFile *f, SaveStateEntry *se,
915 JSONWriter *vmdesc)
916 {
917 int64_t old_offset, size;
918
919 old_offset = qemu_ftell_fast(f);
920 se->ops->save_state(f, se->opaque);
921 size = qemu_ftell_fast(f) - old_offset;
922
923 if (vmdesc) {
924 json_writer_int64(vmdesc, "size", size);
925 json_writer_start_array(vmdesc, "fields");
926 json_writer_start_object(vmdesc, NULL);
927 json_writer_str(vmdesc, "name", "data");
928 json_writer_int64(vmdesc, "size", size);
929 json_writer_str(vmdesc, "type", "buffer");
930 json_writer_end_object(vmdesc);
931 json_writer_end_array(vmdesc);
932 }
933 }
934
935 static int vmstate_save(QEMUFile *f, SaveStateEntry *se,
936 JSONWriter *vmdesc)
937 {
938 trace_vmstate_save(se->idstr, se->vmsd ? se->vmsd->name : "(old)");
939 if (!se->vmsd) {
940 vmstate_save_old_style(f, se, vmdesc);
941 return 0;
942 }
943 return vmstate_save_state(f, se->vmsd, se->opaque, vmdesc);
944 }
945
946 /*
947 * Write the header for device section (QEMU_VM_SECTION START/END/PART/FULL)
948 */
949 static void save_section_header(QEMUFile *f, SaveStateEntry *se,
950 uint8_t section_type)
951 {
952 qemu_put_byte(f, section_type);
953 qemu_put_be32(f, se->section_id);
954
955 if (section_type == QEMU_VM_SECTION_FULL ||
956 section_type == QEMU_VM_SECTION_START) {
957 /* ID string */
958 size_t len = strlen(se->idstr);
959 qemu_put_byte(f, len);
960 qemu_put_buffer(f, (uint8_t *)se->idstr, len);
961
962 qemu_put_be32(f, se->instance_id);
963 qemu_put_be32(f, se->version_id);
964 }
965 }
966
967 /*
968 * Write a footer onto device sections that catches cases misformatted device
969 * sections.
970 */
971 static void save_section_footer(QEMUFile *f, SaveStateEntry *se)
972 {
973 if (migrate_get_current()->send_section_footer) {
974 qemu_put_byte(f, QEMU_VM_SECTION_FOOTER);
975 qemu_put_be32(f, se->section_id);
976 }
977 }
978
979 /**
980 * qemu_savevm_command_send: Send a 'QEMU_VM_COMMAND' type element with the
981 * command and associated data.
982 *
983 * @f: File to send command on
984 * @command: Command type to send
985 * @len: Length of associated data
986 * @data: Data associated with command.
987 */
988 static void qemu_savevm_command_send(QEMUFile *f,
989 enum qemu_vm_cmd command,
990 uint16_t len,
991 uint8_t *data)
992 {
993 trace_savevm_command_send(command, len);
994 qemu_put_byte(f, QEMU_VM_COMMAND);
995 qemu_put_be16(f, (uint16_t)command);
996 qemu_put_be16(f, len);
997 qemu_put_buffer(f, data, len);
998 qemu_fflush(f);
999 }
1000
1001 void qemu_savevm_send_colo_enable(QEMUFile *f)
1002 {
1003 trace_savevm_send_colo_enable();
1004 qemu_savevm_command_send(f, MIG_CMD_ENABLE_COLO, 0, NULL);
1005 }
1006
1007 void qemu_savevm_send_ping(QEMUFile *f, uint32_t value)
1008 {
1009 uint32_t buf;
1010
1011 trace_savevm_send_ping(value);
1012 buf = cpu_to_be32(value);
1013 qemu_savevm_command_send(f, MIG_CMD_PING, sizeof(value), (uint8_t *)&buf);
1014 }
1015
1016 void qemu_savevm_send_open_return_path(QEMUFile *f)
1017 {
1018 trace_savevm_send_open_return_path();
1019 qemu_savevm_command_send(f, MIG_CMD_OPEN_RETURN_PATH, 0, NULL);
1020 }
1021
1022 /* We have a buffer of data to send; we don't want that all to be loaded
1023 * by the command itself, so the command contains just the length of the
1024 * extra buffer that we then send straight after it.
1025 * TODO: Must be a better way to organise that
1026 *
1027 * Returns:
1028 * 0 on success
1029 * -ve on error
1030 */
1031 int qemu_savevm_send_packaged(QEMUFile *f, const uint8_t *buf, size_t len)
1032 {
1033 uint32_t tmp;
1034
1035 if (len > MAX_VM_CMD_PACKAGED_SIZE) {
1036 error_report("%s: Unreasonably large packaged state: %zu",
1037 __func__, len);
1038 return -1;
1039 }
1040
1041 tmp = cpu_to_be32(len);
1042
1043 trace_qemu_savevm_send_packaged();
1044 qemu_savevm_command_send(f, MIG_CMD_PACKAGED, 4, (uint8_t *)&tmp);
1045
1046 qemu_put_buffer(f, buf, len);
1047
1048 return 0;
1049 }
1050
1051 /* Send prior to any postcopy transfer */
1052 void qemu_savevm_send_postcopy_advise(QEMUFile *f)
1053 {
1054 if (migrate_postcopy_ram()) {
1055 uint64_t tmp[2];
1056 tmp[0] = cpu_to_be64(ram_pagesize_summary());
1057 tmp[1] = cpu_to_be64(qemu_target_page_size());
1058
1059 trace_qemu_savevm_send_postcopy_advise();
1060 qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_ADVISE,
1061 16, (uint8_t *)tmp);
1062 } else {
1063 qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_ADVISE, 0, NULL);
1064 }
1065 }
1066
1067 /* Sent prior to starting the destination running in postcopy, discard pages
1068 * that have already been sent but redirtied on the source.
1069 * CMD_POSTCOPY_RAM_DISCARD consist of:
1070 * byte version (0)
1071 * byte Length of name field (not including 0)
1072 * n x byte RAM block name
1073 * byte 0 terminator (just for safety)
1074 * n x Byte ranges within the named RAMBlock
1075 * be64 Start of the range
1076 * be64 Length
1077 *
1078 * name: RAMBlock name that these entries are part of
1079 * len: Number of page entries
1080 * start_list: 'len' addresses
1081 * length_list: 'len' addresses
1082 *
1083 */
1084 void qemu_savevm_send_postcopy_ram_discard(QEMUFile *f, const char *name,
1085 uint16_t len,
1086 uint64_t *start_list,
1087 uint64_t *length_list)
1088 {
1089 uint8_t *buf;
1090 uint16_t tmplen;
1091 uint16_t t;
1092 size_t name_len = strlen(name);
1093
1094 trace_qemu_savevm_send_postcopy_ram_discard(name, len);
1095 assert(name_len < 256);
1096 buf = g_malloc0(1 + 1 + name_len + 1 + (8 + 8) * len);
1097 buf[0] = postcopy_ram_discard_version;
1098 buf[1] = name_len;
1099 memcpy(buf + 2, name, name_len);
1100 tmplen = 2 + name_len;
1101 buf[tmplen++] = '\0';
1102
1103 for (t = 0; t < len; t++) {
1104 stq_be_p(buf + tmplen, start_list[t]);
1105 tmplen += 8;
1106 stq_be_p(buf + tmplen, length_list[t]);
1107 tmplen += 8;
1108 }
1109 qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_RAM_DISCARD, tmplen, buf);
1110 g_free(buf);
1111 }
1112
1113 /* Get the destination into a state where it can receive postcopy data. */
1114 void qemu_savevm_send_postcopy_listen(QEMUFile *f)
1115 {
1116 trace_savevm_send_postcopy_listen();
1117 qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_LISTEN, 0, NULL);
1118 }
1119
1120 /* Kick the destination into running */
1121 void qemu_savevm_send_postcopy_run(QEMUFile *f)
1122 {
1123 trace_savevm_send_postcopy_run();
1124 qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_RUN, 0, NULL);
1125 }
1126
1127 void qemu_savevm_send_postcopy_resume(QEMUFile *f)
1128 {
1129 trace_savevm_send_postcopy_resume();
1130 qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_RESUME, 0, NULL);
1131 }
1132
1133 void qemu_savevm_send_recv_bitmap(QEMUFile *f, char *block_name)
1134 {
1135 size_t len;
1136 char buf[256];
1137
1138 trace_savevm_send_recv_bitmap(block_name);
1139
1140 buf[0] = len = strlen(block_name);
1141 memcpy(buf + 1, block_name, len);
1142
1143 qemu_savevm_command_send(f, MIG_CMD_RECV_BITMAP, len + 1, (uint8_t *)buf);
1144 }
1145
1146 bool qemu_savevm_state_blocked(Error **errp)
1147 {
1148 SaveStateEntry *se;
1149
1150 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1151 if (se->vmsd && se->vmsd->unmigratable) {
1152 error_setg(errp, "State blocked by non-migratable device '%s'",
1153 se->idstr);
1154 return true;
1155 }
1156 }
1157 return false;
1158 }
1159
1160 void qemu_savevm_non_migratable_list(strList **reasons)
1161 {
1162 SaveStateEntry *se;
1163
1164 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1165 if (se->vmsd && se->vmsd->unmigratable) {
1166 QAPI_LIST_PREPEND(*reasons,
1167 g_strdup_printf("non-migratable device: %s",
1168 se->idstr));
1169 }
1170 }
1171 }
1172
1173 void qemu_savevm_state_header(QEMUFile *f)
1174 {
1175 trace_savevm_state_header();
1176 qemu_put_be32(f, QEMU_VM_FILE_MAGIC);
1177 qemu_put_be32(f, QEMU_VM_FILE_VERSION);
1178
1179 if (migrate_get_current()->send_configuration) {
1180 qemu_put_byte(f, QEMU_VM_CONFIGURATION);
1181 vmstate_save_state(f, &vmstate_configuration, &savevm_state, 0);
1182 }
1183 }
1184
1185 bool qemu_savevm_state_guest_unplug_pending(void)
1186 {
1187 SaveStateEntry *se;
1188
1189 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1190 if (se->vmsd && se->vmsd->dev_unplug_pending &&
1191 se->vmsd->dev_unplug_pending(se->opaque)) {
1192 return true;
1193 }
1194 }
1195
1196 return false;
1197 }
1198
1199 void qemu_savevm_state_setup(QEMUFile *f)
1200 {
1201 SaveStateEntry *se;
1202 Error *local_err = NULL;
1203 int ret;
1204
1205 trace_savevm_state_setup();
1206 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1207 if (!se->ops || !se->ops->save_setup) {
1208 continue;
1209 }
1210 if (se->ops->is_active) {
1211 if (!se->ops->is_active(se->opaque)) {
1212 continue;
1213 }
1214 }
1215 save_section_header(f, se, QEMU_VM_SECTION_START);
1216
1217 ret = se->ops->save_setup(f, se->opaque);
1218 save_section_footer(f, se);
1219 if (ret < 0) {
1220 qemu_file_set_error(f, ret);
1221 break;
1222 }
1223 }
1224
1225 if (precopy_notify(PRECOPY_NOTIFY_SETUP, &local_err)) {
1226 error_report_err(local_err);
1227 }
1228 }
1229
1230 int qemu_savevm_state_resume_prepare(MigrationState *s)
1231 {
1232 SaveStateEntry *se;
1233 int ret;
1234
1235 trace_savevm_state_resume_prepare();
1236
1237 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1238 if (!se->ops || !se->ops->resume_prepare) {
1239 continue;
1240 }
1241 if (se->ops->is_active) {
1242 if (!se->ops->is_active(se->opaque)) {
1243 continue;
1244 }
1245 }
1246 ret = se->ops->resume_prepare(s, se->opaque);
1247 if (ret < 0) {
1248 return ret;
1249 }
1250 }
1251
1252 return 0;
1253 }
1254
1255 /*
1256 * this function has three return values:
1257 * negative: there was one error, and we have -errno.
1258 * 0 : We haven't finished, caller have to go again
1259 * 1 : We have finished, we can go to complete phase
1260 */
1261 int qemu_savevm_state_iterate(QEMUFile *f, bool postcopy)
1262 {
1263 SaveStateEntry *se;
1264 int ret = 1;
1265
1266 trace_savevm_state_iterate();
1267 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1268 if (!se->ops || !se->ops->save_live_iterate) {
1269 continue;
1270 }
1271 if (se->ops->is_active &&
1272 !se->ops->is_active(se->opaque)) {
1273 continue;
1274 }
1275 if (se->ops->is_active_iterate &&
1276 !se->ops->is_active_iterate(se->opaque)) {
1277 continue;
1278 }
1279 /*
1280 * In the postcopy phase, any device that doesn't know how to
1281 * do postcopy should have saved it's state in the _complete
1282 * call that's already run, it might get confused if we call
1283 * iterate afterwards.
1284 */
1285 if (postcopy &&
1286 !(se->ops->has_postcopy && se->ops->has_postcopy(se->opaque))) {
1287 continue;
1288 }
1289 if (qemu_file_rate_limit(f)) {
1290 return 0;
1291 }
1292 trace_savevm_section_start(se->idstr, se->section_id);
1293
1294 save_section_header(f, se, QEMU_VM_SECTION_PART);
1295
1296 ret = se->ops->save_live_iterate(f, se->opaque);
1297 trace_savevm_section_end(se->idstr, se->section_id, ret);
1298 save_section_footer(f, se);
1299
1300 if (ret < 0) {
1301 error_report("failed to save SaveStateEntry with id(name): %d(%s)",
1302 se->section_id, se->idstr);
1303 qemu_file_set_error(f, ret);
1304 }
1305 if (ret <= 0) {
1306 /* Do not proceed to the next vmstate before this one reported
1307 completion of the current stage. This serializes the migration
1308 and reduces the probability that a faster changing state is
1309 synchronized over and over again. */
1310 break;
1311 }
1312 }
1313 return ret;
1314 }
1315
1316 static bool should_send_vmdesc(void)
1317 {
1318 MachineState *machine = MACHINE(qdev_get_machine());
1319 bool in_postcopy = migration_in_postcopy();
1320 return !machine->suppress_vmdesc && !in_postcopy;
1321 }
1322
1323 /*
1324 * Calls the save_live_complete_postcopy methods
1325 * causing the last few pages to be sent immediately and doing any associated
1326 * cleanup.
1327 * Note postcopy also calls qemu_savevm_state_complete_precopy to complete
1328 * all the other devices, but that happens at the point we switch to postcopy.
1329 */
1330 void qemu_savevm_state_complete_postcopy(QEMUFile *f)
1331 {
1332 SaveStateEntry *se;
1333 int ret;
1334
1335 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1336 if (!se->ops || !se->ops->save_live_complete_postcopy) {
1337 continue;
1338 }
1339 if (se->ops->is_active) {
1340 if (!se->ops->is_active(se->opaque)) {
1341 continue;
1342 }
1343 }
1344 trace_savevm_section_start(se->idstr, se->section_id);
1345 /* Section type */
1346 qemu_put_byte(f, QEMU_VM_SECTION_END);
1347 qemu_put_be32(f, se->section_id);
1348
1349 ret = se->ops->save_live_complete_postcopy(f, se->opaque);
1350 trace_savevm_section_end(se->idstr, se->section_id, ret);
1351 save_section_footer(f, se);
1352 if (ret < 0) {
1353 qemu_file_set_error(f, ret);
1354 return;
1355 }
1356 }
1357
1358 qemu_put_byte(f, QEMU_VM_EOF);
1359 qemu_fflush(f);
1360 }
1361
1362 static
1363 int qemu_savevm_state_complete_precopy_iterable(QEMUFile *f, bool in_postcopy)
1364 {
1365 SaveStateEntry *se;
1366 int ret;
1367
1368 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1369 if (!se->ops ||
1370 (in_postcopy && se->ops->has_postcopy &&
1371 se->ops->has_postcopy(se->opaque)) ||
1372 !se->ops->save_live_complete_precopy) {
1373 continue;
1374 }
1375
1376 if (se->ops->is_active) {
1377 if (!se->ops->is_active(se->opaque)) {
1378 continue;
1379 }
1380 }
1381 trace_savevm_section_start(se->idstr, se->section_id);
1382
1383 save_section_header(f, se, QEMU_VM_SECTION_END);
1384
1385 ret = se->ops->save_live_complete_precopy(f, se->opaque);
1386 trace_savevm_section_end(se->idstr, se->section_id, ret);
1387 save_section_footer(f, se);
1388 if (ret < 0) {
1389 qemu_file_set_error(f, ret);
1390 return -1;
1391 }
1392 }
1393
1394 return 0;
1395 }
1396
1397 int qemu_savevm_state_complete_precopy_non_iterable(QEMUFile *f,
1398 bool in_postcopy,
1399 bool inactivate_disks)
1400 {
1401 g_autoptr(JSONWriter) vmdesc = NULL;
1402 int vmdesc_len;
1403 SaveStateEntry *se;
1404 int ret;
1405
1406 vmdesc = json_writer_new(false);
1407 json_writer_start_object(vmdesc, NULL);
1408 json_writer_int64(vmdesc, "page_size", qemu_target_page_size());
1409 json_writer_start_array(vmdesc, "devices");
1410 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1411
1412 if ((!se->ops || !se->ops->save_state) && !se->vmsd) {
1413 continue;
1414 }
1415 if (se->vmsd && !vmstate_save_needed(se->vmsd, se->opaque)) {
1416 trace_savevm_section_skip(se->idstr, se->section_id);
1417 continue;
1418 }
1419
1420 trace_savevm_section_start(se->idstr, se->section_id);
1421
1422 json_writer_start_object(vmdesc, NULL);
1423 json_writer_str(vmdesc, "name", se->idstr);
1424 json_writer_int64(vmdesc, "instance_id", se->instance_id);
1425
1426 save_section_header(f, se, QEMU_VM_SECTION_FULL);
1427 ret = vmstate_save(f, se, vmdesc);
1428 if (ret) {
1429 qemu_file_set_error(f, ret);
1430 return ret;
1431 }
1432 trace_savevm_section_end(se->idstr, se->section_id, 0);
1433 save_section_footer(f, se);
1434
1435 json_writer_end_object(vmdesc);
1436 }
1437
1438 if (inactivate_disks) {
1439 /* Inactivate before sending QEMU_VM_EOF so that the
1440 * bdrv_invalidate_cache_all() on the other end won't fail. */
1441 ret = bdrv_inactivate_all();
1442 if (ret) {
1443 error_report("%s: bdrv_inactivate_all() failed (%d)",
1444 __func__, ret);
1445 qemu_file_set_error(f, ret);
1446 return ret;
1447 }
1448 }
1449 if (!in_postcopy) {
1450 /* Postcopy stream will still be going */
1451 qemu_put_byte(f, QEMU_VM_EOF);
1452 }
1453
1454 json_writer_end_array(vmdesc);
1455 json_writer_end_object(vmdesc);
1456 vmdesc_len = strlen(json_writer_get(vmdesc));
1457
1458 if (should_send_vmdesc()) {
1459 qemu_put_byte(f, QEMU_VM_VMDESCRIPTION);
1460 qemu_put_be32(f, vmdesc_len);
1461 qemu_put_buffer(f, (uint8_t *)json_writer_get(vmdesc), vmdesc_len);
1462 }
1463
1464 return 0;
1465 }
1466
1467 int qemu_savevm_state_complete_precopy(QEMUFile *f, bool iterable_only,
1468 bool inactivate_disks)
1469 {
1470 int ret;
1471 Error *local_err = NULL;
1472 bool in_postcopy = migration_in_postcopy();
1473
1474 if (precopy_notify(PRECOPY_NOTIFY_COMPLETE, &local_err)) {
1475 error_report_err(local_err);
1476 }
1477
1478 trace_savevm_state_complete_precopy();
1479
1480 cpu_synchronize_all_states();
1481
1482 if (!in_postcopy || iterable_only) {
1483 ret = qemu_savevm_state_complete_precopy_iterable(f, in_postcopy);
1484 if (ret) {
1485 return ret;
1486 }
1487 }
1488
1489 if (iterable_only) {
1490 goto flush;
1491 }
1492
1493 ret = qemu_savevm_state_complete_precopy_non_iterable(f, in_postcopy,
1494 inactivate_disks);
1495 if (ret) {
1496 return ret;
1497 }
1498
1499 flush:
1500 qemu_fflush(f);
1501 return 0;
1502 }
1503
1504 /* Give an estimate of the amount left to be transferred,
1505 * the result is split into the amount for units that can and
1506 * for units that can't do postcopy.
1507 */
1508 void qemu_savevm_state_pending(QEMUFile *f, uint64_t threshold_size,
1509 uint64_t *res_precopy_only,
1510 uint64_t *res_compatible,
1511 uint64_t *res_postcopy_only)
1512 {
1513 SaveStateEntry *se;
1514
1515 *res_precopy_only = 0;
1516 *res_compatible = 0;
1517 *res_postcopy_only = 0;
1518
1519
1520 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1521 if (!se->ops || !se->ops->save_live_pending) {
1522 continue;
1523 }
1524 if (se->ops->is_active) {
1525 if (!se->ops->is_active(se->opaque)) {
1526 continue;
1527 }
1528 }
1529 se->ops->save_live_pending(f, se->opaque, threshold_size,
1530 res_precopy_only, res_compatible,
1531 res_postcopy_only);
1532 }
1533 }
1534
1535 void qemu_savevm_state_cleanup(void)
1536 {
1537 SaveStateEntry *se;
1538 Error *local_err = NULL;
1539
1540 if (precopy_notify(PRECOPY_NOTIFY_CLEANUP, &local_err)) {
1541 error_report_err(local_err);
1542 }
1543
1544 trace_savevm_state_cleanup();
1545 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1546 if (se->ops && se->ops->save_cleanup) {
1547 se->ops->save_cleanup(se->opaque);
1548 }
1549 }
1550 }
1551
1552 static int qemu_savevm_state(QEMUFile *f, Error **errp)
1553 {
1554 int ret;
1555 MigrationState *ms = migrate_get_current();
1556 MigrationStatus status;
1557
1558 if (migration_is_running(ms->state)) {
1559 error_setg(errp, QERR_MIGRATION_ACTIVE);
1560 return -EINVAL;
1561 }
1562
1563 if (migrate_use_block()) {
1564 error_setg(errp, "Block migration and snapshots are incompatible");
1565 return -EINVAL;
1566 }
1567
1568 migrate_init(ms);
1569 memset(&ram_counters, 0, sizeof(ram_counters));
1570 memset(&compression_counters, 0, sizeof(compression_counters));
1571 ms->to_dst_file = f;
1572
1573 qemu_mutex_unlock_iothread();
1574 qemu_savevm_state_header(f);
1575 qemu_savevm_state_setup(f);
1576 qemu_mutex_lock_iothread();
1577
1578 while (qemu_file_get_error(f) == 0) {
1579 if (qemu_savevm_state_iterate(f, false) > 0) {
1580 break;
1581 }
1582 }
1583
1584 ret = qemu_file_get_error(f);
1585 if (ret == 0) {
1586 qemu_savevm_state_complete_precopy(f, false, false);
1587 ret = qemu_file_get_error(f);
1588 }
1589 qemu_savevm_state_cleanup();
1590 if (ret != 0) {
1591 error_setg_errno(errp, -ret, "Error while writing VM state");
1592 }
1593
1594 if (ret != 0) {
1595 status = MIGRATION_STATUS_FAILED;
1596 } else {
1597 status = MIGRATION_STATUS_COMPLETED;
1598 }
1599 migrate_set_state(&ms->state, MIGRATION_STATUS_SETUP, status);
1600
1601 /* f is outer parameter, it should not stay in global migration state after
1602 * this function finished */
1603 ms->to_dst_file = NULL;
1604
1605 return ret;
1606 }
1607
1608 void qemu_savevm_live_state(QEMUFile *f)
1609 {
1610 /* save QEMU_VM_SECTION_END section */
1611 qemu_savevm_state_complete_precopy(f, true, false);
1612 qemu_put_byte(f, QEMU_VM_EOF);
1613 }
1614
1615 int qemu_save_device_state(QEMUFile *f)
1616 {
1617 SaveStateEntry *se;
1618
1619 if (!migration_in_colo_state()) {
1620 qemu_put_be32(f, QEMU_VM_FILE_MAGIC);
1621 qemu_put_be32(f, QEMU_VM_FILE_VERSION);
1622 }
1623 cpu_synchronize_all_states();
1624
1625 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1626 int ret;
1627
1628 if (se->is_ram) {
1629 continue;
1630 }
1631 if ((!se->ops || !se->ops->save_state) && !se->vmsd) {
1632 continue;
1633 }
1634 if (se->vmsd && !vmstate_save_needed(se->vmsd, se->opaque)) {
1635 continue;
1636 }
1637
1638 save_section_header(f, se, QEMU_VM_SECTION_FULL);
1639
1640 ret = vmstate_save(f, se, NULL);
1641 if (ret) {
1642 return ret;
1643 }
1644
1645 save_section_footer(f, se);
1646 }
1647
1648 qemu_put_byte(f, QEMU_VM_EOF);
1649
1650 return qemu_file_get_error(f);
1651 }
1652
1653 static SaveStateEntry *find_se(const char *idstr, uint32_t instance_id)
1654 {
1655 SaveStateEntry *se;
1656
1657 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1658 if (!strcmp(se->idstr, idstr) &&
1659 (instance_id == se->instance_id ||
1660 instance_id == se->alias_id))
1661 return se;
1662 /* Migrating from an older version? */
1663 if (strstr(se->idstr, idstr) && se->compat) {
1664 if (!strcmp(se->compat->idstr, idstr) &&
1665 (instance_id == se->compat->instance_id ||
1666 instance_id == se->alias_id))
1667 return se;
1668 }
1669 }
1670 return NULL;
1671 }
1672
1673 enum LoadVMExitCodes {
1674 /* Allow a command to quit all layers of nested loadvm loops */
1675 LOADVM_QUIT = 1,
1676 };
1677
1678 /* ------ incoming postcopy messages ------ */
1679 /* 'advise' arrives before any transfers just to tell us that a postcopy
1680 * *might* happen - it might be skipped if precopy transferred everything
1681 * quickly.
1682 */
1683 static int loadvm_postcopy_handle_advise(MigrationIncomingState *mis,
1684 uint16_t len)
1685 {
1686 PostcopyState ps = postcopy_state_set(POSTCOPY_INCOMING_ADVISE);
1687 uint64_t remote_pagesize_summary, local_pagesize_summary, remote_tps;
1688 size_t page_size = qemu_target_page_size();
1689 Error *local_err = NULL;
1690
1691 trace_loadvm_postcopy_handle_advise();
1692 if (ps != POSTCOPY_INCOMING_NONE) {
1693 error_report("CMD_POSTCOPY_ADVISE in wrong postcopy state (%d)", ps);
1694 return -1;
1695 }
1696
1697 switch (len) {
1698 case 0:
1699 if (migrate_postcopy_ram()) {
1700 error_report("RAM postcopy is enabled but have 0 byte advise");
1701 return -EINVAL;
1702 }
1703 return 0;
1704 case 8 + 8:
1705 if (!migrate_postcopy_ram()) {
1706 error_report("RAM postcopy is disabled but have 16 byte advise");
1707 return -EINVAL;
1708 }
1709 break;
1710 default:
1711 error_report("CMD_POSTCOPY_ADVISE invalid length (%d)", len);
1712 return -EINVAL;
1713 }
1714
1715 if (!postcopy_ram_supported_by_host(mis)) {
1716 postcopy_state_set(POSTCOPY_INCOMING_NONE);
1717 return -1;
1718 }
1719
1720 remote_pagesize_summary = qemu_get_be64(mis->from_src_file);
1721 local_pagesize_summary = ram_pagesize_summary();
1722
1723 if (remote_pagesize_summary != local_pagesize_summary) {
1724 /*
1725 * This detects two potential causes of mismatch:
1726 * a) A mismatch in host page sizes
1727 * Some combinations of mismatch are probably possible but it gets
1728 * a bit more complicated. In particular we need to place whole
1729 * host pages on the dest at once, and we need to ensure that we
1730 * handle dirtying to make sure we never end up sending part of
1731 * a hostpage on it's own.
1732 * b) The use of different huge page sizes on source/destination
1733 * a more fine grain test is performed during RAM block migration
1734 * but this test here causes a nice early clear failure, and
1735 * also fails when passed to an older qemu that doesn't
1736 * do huge pages.
1737 */
1738 error_report("Postcopy needs matching RAM page sizes (s=%" PRIx64
1739 " d=%" PRIx64 ")",
1740 remote_pagesize_summary, local_pagesize_summary);
1741 return -1;
1742 }
1743
1744 remote_tps = qemu_get_be64(mis->from_src_file);
1745 if (remote_tps != page_size) {
1746 /*
1747 * Again, some differences could be dealt with, but for now keep it
1748 * simple.
1749 */
1750 error_report("Postcopy needs matching target page sizes (s=%d d=%zd)",
1751 (int)remote_tps, page_size);
1752 return -1;
1753 }
1754
1755 if (postcopy_notify(POSTCOPY_NOTIFY_INBOUND_ADVISE, &local_err)) {
1756 error_report_err(local_err);
1757 return -1;
1758 }
1759
1760 if (ram_postcopy_incoming_init(mis)) {
1761 return -1;
1762 }
1763
1764 return 0;
1765 }
1766
1767 /* After postcopy we will be told to throw some pages away since they're
1768 * dirty and will have to be demand fetched. Must happen before CPU is
1769 * started.
1770 * There can be 0..many of these messages, each encoding multiple pages.
1771 */
1772 static int loadvm_postcopy_ram_handle_discard(MigrationIncomingState *mis,
1773 uint16_t len)
1774 {
1775 int tmp;
1776 char ramid[256];
1777 PostcopyState ps = postcopy_state_get();
1778
1779 trace_loadvm_postcopy_ram_handle_discard();
1780
1781 switch (ps) {
1782 case POSTCOPY_INCOMING_ADVISE:
1783 /* 1st discard */
1784 tmp = postcopy_ram_prepare_discard(mis);
1785 if (tmp) {
1786 return tmp;
1787 }
1788 break;
1789
1790 case POSTCOPY_INCOMING_DISCARD:
1791 /* Expected state */
1792 break;
1793
1794 default:
1795 error_report("CMD_POSTCOPY_RAM_DISCARD in wrong postcopy state (%d)",
1796 ps);
1797 return -1;
1798 }
1799 /* We're expecting a
1800 * Version (0)
1801 * a RAM ID string (length byte, name, 0 term)
1802 * then at least 1 16 byte chunk
1803 */
1804 if (len < (1 + 1 + 1 + 1 + 2 * 8)) {
1805 error_report("CMD_POSTCOPY_RAM_DISCARD invalid length (%d)", len);
1806 return -1;
1807 }
1808
1809 tmp = qemu_get_byte(mis->from_src_file);
1810 if (tmp != postcopy_ram_discard_version) {
1811 error_report("CMD_POSTCOPY_RAM_DISCARD invalid version (%d)", tmp);
1812 return -1;
1813 }
1814
1815 if (!qemu_get_counted_string(mis->from_src_file, ramid)) {
1816 error_report("CMD_POSTCOPY_RAM_DISCARD Failed to read RAMBlock ID");
1817 return -1;
1818 }
1819 tmp = qemu_get_byte(mis->from_src_file);
1820 if (tmp != 0) {
1821 error_report("CMD_POSTCOPY_RAM_DISCARD missing nil (%d)", tmp);
1822 return -1;
1823 }
1824
1825 len -= 3 + strlen(ramid);
1826 if (len % 16) {
1827 error_report("CMD_POSTCOPY_RAM_DISCARD invalid length (%d)", len);
1828 return -1;
1829 }
1830 trace_loadvm_postcopy_ram_handle_discard_header(ramid, len);
1831 while (len) {
1832 uint64_t start_addr, block_length;
1833 start_addr = qemu_get_be64(mis->from_src_file);
1834 block_length = qemu_get_be64(mis->from_src_file);
1835
1836 len -= 16;
1837 int ret = ram_discard_range(ramid, start_addr, block_length);
1838 if (ret) {
1839 return ret;
1840 }
1841 }
1842 trace_loadvm_postcopy_ram_handle_discard_end();
1843
1844 return 0;
1845 }
1846
1847 /*
1848 * Triggered by a postcopy_listen command; this thread takes over reading
1849 * the input stream, leaving the main thread free to carry on loading the rest
1850 * of the device state (from RAM).
1851 * (TODO:This could do with being in a postcopy file - but there again it's
1852 * just another input loop, not that postcopy specific)
1853 */
1854 static void *postcopy_ram_listen_thread(void *opaque)
1855 {
1856 MigrationIncomingState *mis = migration_incoming_get_current();
1857 QEMUFile *f = mis->from_src_file;
1858 int load_res;
1859 MigrationState *migr = migrate_get_current();
1860
1861 object_ref(OBJECT(migr));
1862
1863 migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE,
1864 MIGRATION_STATUS_POSTCOPY_ACTIVE);
1865 qemu_sem_post(&mis->listen_thread_sem);
1866 trace_postcopy_ram_listen_thread_start();
1867
1868 rcu_register_thread();
1869 /*
1870 * Because we're a thread and not a coroutine we can't yield
1871 * in qemu_file, and thus we must be blocking now.
1872 */
1873 qemu_file_set_blocking(f, true);
1874 load_res = qemu_loadvm_state_main(f, mis);
1875
1876 /*
1877 * This is tricky, but, mis->from_src_file can change after it
1878 * returns, when postcopy recovery happened. In the future, we may
1879 * want a wrapper for the QEMUFile handle.
1880 */
1881 f = mis->from_src_file;
1882
1883 /* And non-blocking again so we don't block in any cleanup */
1884 qemu_file_set_blocking(f, false);
1885
1886 trace_postcopy_ram_listen_thread_exit();
1887 if (load_res < 0) {
1888 qemu_file_set_error(f, load_res);
1889 dirty_bitmap_mig_cancel_incoming();
1890 if (postcopy_state_get() == POSTCOPY_INCOMING_RUNNING &&
1891 !migrate_postcopy_ram() && migrate_dirty_bitmaps())
1892 {
1893 error_report("%s: loadvm failed during postcopy: %d. All states "
1894 "are migrated except dirty bitmaps. Some dirty "
1895 "bitmaps may be lost, and present migrated dirty "
1896 "bitmaps are correctly migrated and valid.",
1897 __func__, load_res);
1898 load_res = 0; /* prevent further exit() */
1899 } else {
1900 error_report("%s: loadvm failed: %d", __func__, load_res);
1901 migrate_set_state(&mis->state, MIGRATION_STATUS_POSTCOPY_ACTIVE,
1902 MIGRATION_STATUS_FAILED);
1903 }
1904 }
1905 if (load_res >= 0) {
1906 /*
1907 * This looks good, but it's possible that the device loading in the
1908 * main thread hasn't finished yet, and so we might not be in 'RUN'
1909 * state yet; wait for the end of the main thread.
1910 */
1911 qemu_event_wait(&mis->main_thread_load_event);
1912 }
1913 postcopy_ram_incoming_cleanup(mis);
1914
1915 if (load_res < 0) {
1916 /*
1917 * If something went wrong then we have a bad state so exit;
1918 * depending how far we got it might be possible at this point
1919 * to leave the guest running and fire MCEs for pages that never
1920 * arrived as a desperate recovery step.
1921 */
1922 rcu_unregister_thread();
1923 exit(EXIT_FAILURE);
1924 }
1925
1926 migrate_set_state(&mis->state, MIGRATION_STATUS_POSTCOPY_ACTIVE,
1927 MIGRATION_STATUS_COMPLETED);
1928 /*
1929 * If everything has worked fine, then the main thread has waited
1930 * for us to start, and we're the last use of the mis.
1931 * (If something broke then qemu will have to exit anyway since it's
1932 * got a bad migration state).
1933 */
1934 migration_incoming_state_destroy();
1935 qemu_loadvm_state_cleanup();
1936
1937 rcu_unregister_thread();
1938 mis->have_listen_thread = false;
1939 postcopy_state_set(POSTCOPY_INCOMING_END);
1940
1941 object_unref(OBJECT(migr));
1942
1943 return NULL;
1944 }
1945
1946 /* After this message we must be able to immediately receive postcopy data */
1947 static int loadvm_postcopy_handle_listen(MigrationIncomingState *mis)
1948 {
1949 PostcopyState ps = postcopy_state_set(POSTCOPY_INCOMING_LISTENING);
1950 trace_loadvm_postcopy_handle_listen();
1951 Error *local_err = NULL;
1952
1953 if (ps != POSTCOPY_INCOMING_ADVISE && ps != POSTCOPY_INCOMING_DISCARD) {
1954 error_report("CMD_POSTCOPY_LISTEN in wrong postcopy state (%d)", ps);
1955 return -1;
1956 }
1957 if (ps == POSTCOPY_INCOMING_ADVISE) {
1958 /*
1959 * A rare case, we entered listen without having to do any discards,
1960 * so do the setup that's normally done at the time of the 1st discard.
1961 */
1962 if (migrate_postcopy_ram()) {
1963 postcopy_ram_prepare_discard(mis);
1964 }
1965 }
1966
1967 /*
1968 * Sensitise RAM - can now generate requests for blocks that don't exist
1969 * However, at this point the CPU shouldn't be running, and the IO
1970 * shouldn't be doing anything yet so don't actually expect requests
1971 */
1972 if (migrate_postcopy_ram()) {
1973 if (postcopy_ram_incoming_setup(mis)) {
1974 postcopy_ram_incoming_cleanup(mis);
1975 return -1;
1976 }
1977 }
1978
1979 if (postcopy_notify(POSTCOPY_NOTIFY_INBOUND_LISTEN, &local_err)) {
1980 error_report_err(local_err);
1981 return -1;
1982 }
1983
1984 mis->have_listen_thread = true;
1985 /* Start up the listening thread and wait for it to signal ready */
1986 qemu_sem_init(&mis->listen_thread_sem, 0);
1987 qemu_thread_create(&mis->listen_thread, "postcopy/listen",
1988 postcopy_ram_listen_thread, NULL,
1989 QEMU_THREAD_DETACHED);
1990 qemu_sem_wait(&mis->listen_thread_sem);
1991 qemu_sem_destroy(&mis->listen_thread_sem);
1992
1993 return 0;
1994 }
1995
1996 static void loadvm_postcopy_handle_run_bh(void *opaque)
1997 {
1998 Error *local_err = NULL;
1999 MigrationIncomingState *mis = opaque;
2000
2001 /* TODO we should move all of this lot into postcopy_ram.c or a shared code
2002 * in migration.c
2003 */
2004 cpu_synchronize_all_post_init();
2005
2006 qemu_announce_self(&mis->announce_timer, migrate_announce_params());
2007
2008 /* Make sure all file formats flush their mutable metadata.
2009 * If we get an error here, just don't restart the VM yet. */
2010 bdrv_invalidate_cache_all(&local_err);
2011 if (local_err) {
2012 error_report_err(local_err);
2013 local_err = NULL;
2014 autostart = false;
2015 }
2016
2017 trace_loadvm_postcopy_handle_run_cpu_sync();
2018
2019 trace_loadvm_postcopy_handle_run_vmstart();
2020
2021 dirty_bitmap_mig_before_vm_start();
2022
2023 if (autostart) {
2024 /* Hold onto your hats, starting the CPU */
2025 vm_start();
2026 } else {
2027 /* leave it paused and let management decide when to start the CPU */
2028 runstate_set(RUN_STATE_PAUSED);
2029 }
2030
2031 qemu_bh_delete(mis->bh);
2032 }
2033
2034 /* After all discards we can start running and asking for pages */
2035 static int loadvm_postcopy_handle_run(MigrationIncomingState *mis)
2036 {
2037 PostcopyState ps = postcopy_state_get();
2038
2039 trace_loadvm_postcopy_handle_run();
2040 if (ps != POSTCOPY_INCOMING_LISTENING) {
2041 error_report("CMD_POSTCOPY_RUN in wrong postcopy state (%d)", ps);
2042 return -1;
2043 }
2044
2045 postcopy_state_set(POSTCOPY_INCOMING_RUNNING);
2046 mis->bh = qemu_bh_new(loadvm_postcopy_handle_run_bh, mis);
2047 qemu_bh_schedule(mis->bh);
2048
2049 /* We need to finish reading the stream from the package
2050 * and also stop reading anything more from the stream that loaded the
2051 * package (since it's now being read by the listener thread).
2052 * LOADVM_QUIT will quit all the layers of nested loadvm loops.
2053 */
2054 return LOADVM_QUIT;
2055 }
2056
2057 /* We must be with page_request_mutex held */
2058 static gboolean postcopy_sync_page_req(gpointer key, gpointer value,
2059 gpointer data)
2060 {
2061 MigrationIncomingState *mis = data;
2062 void *host_addr = (void *) key;
2063 ram_addr_t rb_offset;
2064 RAMBlock *rb;
2065 int ret;
2066
2067 rb = qemu_ram_block_from_host(host_addr, true, &rb_offset);
2068 if (!rb) {
2069 /*
2070 * This should _never_ happen. However be nice for a migrating VM to
2071 * not crash/assert. Post an error (note: intended to not use *_once
2072 * because we do want to see all the illegal addresses; and this can
2073 * never be triggered by the guest so we're safe) and move on next.
2074 */
2075 error_report("%s: illegal host addr %p", __func__, host_addr);
2076 /* Try the next entry */
2077 return FALSE;
2078 }
2079
2080 ret = migrate_send_rp_message_req_pages(mis, rb, rb_offset);
2081 if (ret) {
2082 /* Please refer to above comment. */
2083 error_report("%s: send rp message failed for addr %p",
2084 __func__, host_addr);
2085 return FALSE;
2086 }
2087
2088 trace_postcopy_page_req_sync(host_addr);
2089
2090 return FALSE;
2091 }
2092
2093 static void migrate_send_rp_req_pages_pending(MigrationIncomingState *mis)
2094 {
2095 WITH_QEMU_LOCK_GUARD(&mis->page_request_mutex) {
2096 g_tree_foreach(mis->page_requested, postcopy_sync_page_req, mis);
2097 }
2098 }
2099
2100 static int loadvm_postcopy_handle_resume(MigrationIncomingState *mis)
2101 {
2102 if (mis->state != MIGRATION_STATUS_POSTCOPY_RECOVER) {
2103 error_report("%s: illegal resume received", __func__);
2104 /* Don't fail the load, only for this. */
2105 return 0;
2106 }
2107
2108 /*
2109 * Reset the last_rb before we resend any page req to source again, since
2110 * the source should have it reset already.
2111 */
2112 mis->last_rb = NULL;
2113
2114 /*
2115 * This means source VM is ready to resume the postcopy migration.
2116 */
2117 migrate_set_state(&mis->state, MIGRATION_STATUS_POSTCOPY_RECOVER,
2118 MIGRATION_STATUS_POSTCOPY_ACTIVE);
2119
2120 trace_loadvm_postcopy_handle_resume();
2121
2122 /* Tell source that "we are ready" */
2123 migrate_send_rp_resume_ack(mis, MIGRATION_RESUME_ACK_VALUE);
2124
2125 /*
2126 * After a postcopy recovery, the source should have lost the postcopy
2127 * queue, or potentially the requested pages could have been lost during
2128 * the network down phase. Let's re-sync with the source VM by re-sending
2129 * all the pending pages that we eagerly need, so these threads won't get
2130 * blocked too long due to the recovery.
2131 *
2132 * Without this procedure, the faulted destination VM threads (waiting for
2133 * page requests right before the postcopy is interrupted) can keep hanging
2134 * until the pages are sent by the source during the background copying of
2135 * pages, or another thread faulted on the same address accidentally.
2136 */
2137 migrate_send_rp_req_pages_pending(mis);
2138
2139 /*
2140 * It's time to switch state and release the fault thread to continue
2141 * service page faults. Note that this should be explicitly after the
2142 * above call to migrate_send_rp_req_pages_pending(). In short:
2143 * migrate_send_rp_message_req_pages() is not thread safe, yet.
2144 */
2145 qemu_sem_post(&mis->postcopy_pause_sem_fault);
2146
2147 return 0;
2148 }
2149
2150 /**
2151 * Immediately following this command is a blob of data containing an embedded
2152 * chunk of migration stream; read it and load it.
2153 *
2154 * @mis: Incoming state
2155 * @length: Length of packaged data to read
2156 *
2157 * Returns: Negative values on error
2158 *
2159 */
2160 static int loadvm_handle_cmd_packaged(MigrationIncomingState *mis)
2161 {
2162 int ret;
2163 size_t length;
2164 QIOChannelBuffer *bioc;
2165
2166 length = qemu_get_be32(mis->from_src_file);
2167 trace_loadvm_handle_cmd_packaged(length);
2168
2169 if (length > MAX_VM_CMD_PACKAGED_SIZE) {
2170 error_report("Unreasonably large packaged state: %zu", length);
2171 return -1;
2172 }
2173
2174 bioc = qio_channel_buffer_new(length);
2175 qio_channel_set_name(QIO_CHANNEL(bioc), "migration-loadvm-buffer");
2176 ret = qemu_get_buffer(mis->from_src_file,
2177 bioc->data,
2178 length);
2179 if (ret != length) {
2180 object_unref(OBJECT(bioc));
2181 error_report("CMD_PACKAGED: Buffer receive fail ret=%d length=%zu",
2182 ret, length);
2183 return (ret < 0) ? ret : -EAGAIN;
2184 }
2185 bioc->usage += length;
2186 trace_loadvm_handle_cmd_packaged_received(ret);
2187
2188 QEMUFile *packf = qemu_fopen_channel_input(QIO_CHANNEL(bioc));
2189
2190 ret = qemu_loadvm_state_main(packf, mis);
2191 trace_loadvm_handle_cmd_packaged_main(ret);
2192 qemu_fclose(packf);
2193 object_unref(OBJECT(bioc));
2194
2195 return ret;
2196 }
2197
2198 /*
2199 * Handle request that source requests for recved_bitmap on
2200 * destination. Payload format:
2201 *
2202 * len (1 byte) + ramblock_name (<255 bytes)
2203 */
2204 static int loadvm_handle_recv_bitmap(MigrationIncomingState *mis,
2205 uint16_t len)
2206 {
2207 QEMUFile *file = mis->from_src_file;
2208 RAMBlock *rb;
2209 char block_name[256];
2210 size_t cnt;
2211
2212 cnt = qemu_get_counted_string(file, block_name);
2213 if (!cnt) {
2214 error_report("%s: failed to read block name", __func__);
2215 return -EINVAL;
2216 }
2217
2218 /* Validate before using the data */
2219 if (qemu_file_get_error(file)) {
2220 return qemu_file_get_error(file);
2221 }
2222
2223 if (len != cnt + 1) {
2224 error_report("%s: invalid payload length (%d)", __func__, len);
2225 return -EINVAL;
2226 }
2227
2228 rb = qemu_ram_block_by_name(block_name);
2229 if (!rb) {
2230 error_report("%s: block '%s' not found", __func__, block_name);
2231 return -EINVAL;
2232 }
2233
2234 migrate_send_rp_recv_bitmap(mis, block_name);
2235
2236 trace_loadvm_handle_recv_bitmap(block_name);
2237
2238 return 0;
2239 }
2240
2241 static int loadvm_process_enable_colo(MigrationIncomingState *mis)
2242 {
2243 int ret = migration_incoming_enable_colo();
2244
2245 if (!ret) {
2246 ret = colo_init_ram_cache();
2247 if (ret) {
2248 migration_incoming_disable_colo();
2249 }
2250 }
2251 return ret;
2252 }
2253
2254 /*
2255 * Process an incoming 'QEMU_VM_COMMAND'
2256 * 0 just a normal return
2257 * LOADVM_QUIT All good, but exit the loop
2258 * <0 Error
2259 */
2260 static int loadvm_process_command(QEMUFile *f)
2261 {
2262 MigrationIncomingState *mis = migration_incoming_get_current();
2263 uint16_t cmd;
2264 uint16_t len;
2265 uint32_t tmp32;
2266
2267 cmd = qemu_get_be16(f);
2268 len = qemu_get_be16(f);
2269
2270 /* Check validity before continue processing of cmds */
2271 if (qemu_file_get_error(f)) {
2272 return qemu_file_get_error(f);
2273 }
2274
2275 trace_loadvm_process_command(cmd, len);
2276 if (cmd >= MIG_CMD_MAX || cmd == MIG_CMD_INVALID) {
2277 error_report("MIG_CMD 0x%x unknown (len 0x%x)", cmd, len);
2278 return -EINVAL;
2279 }
2280
2281 if (mig_cmd_args[cmd].len != -1 && mig_cmd_args[cmd].len != len) {
2282 error_report("%s received with bad length - expecting %zu, got %d",
2283 mig_cmd_args[cmd].name,
2284 (size_t)mig_cmd_args[cmd].len, len);
2285 return -ERANGE;
2286 }
2287
2288 switch (cmd) {
2289 case MIG_CMD_OPEN_RETURN_PATH:
2290 if (mis->to_src_file) {
2291 error_report("CMD_OPEN_RETURN_PATH called when RP already open");
2292 /* Not really a problem, so don't give up */
2293 return 0;
2294 }
2295 mis->to_src_file = qemu_file_get_return_path(f);
2296 if (!mis->to_src_file) {
2297 error_report("CMD_OPEN_RETURN_PATH failed");
2298 return -1;
2299 }
2300 break;
2301
2302 case MIG_CMD_PING:
2303 tmp32 = qemu_get_be32(f);
2304 trace_loadvm_process_command_ping(tmp32);
2305 if (!mis->to_src_file) {
2306 error_report("CMD_PING (0x%x) received with no return path",
2307 tmp32);
2308 return -1;
2309 }
2310 migrate_send_rp_pong(mis, tmp32);
2311 break;
2312
2313 case MIG_CMD_PACKAGED:
2314 return loadvm_handle_cmd_packaged(mis);
2315
2316 case MIG_CMD_POSTCOPY_ADVISE:
2317 return loadvm_postcopy_handle_advise(mis, len);
2318
2319 case MIG_CMD_POSTCOPY_LISTEN:
2320 return loadvm_postcopy_handle_listen(mis);
2321
2322 case MIG_CMD_POSTCOPY_RUN:
2323 return loadvm_postcopy_handle_run(mis);
2324
2325 case MIG_CMD_POSTCOPY_RAM_DISCARD:
2326 return loadvm_postcopy_ram_handle_discard(mis, len);
2327
2328 case MIG_CMD_POSTCOPY_RESUME:
2329 return loadvm_postcopy_handle_resume(mis);
2330
2331 case MIG_CMD_RECV_BITMAP:
2332 return loadvm_handle_recv_bitmap(mis, len);
2333
2334 case MIG_CMD_ENABLE_COLO:
2335 return loadvm_process_enable_colo(mis);
2336 }
2337
2338 return 0;
2339 }
2340
2341 /*
2342 * Read a footer off the wire and check that it matches the expected section
2343 *
2344 * Returns: true if the footer was good
2345 * false if there is a problem (and calls error_report to say why)
2346 */
2347 static bool check_section_footer(QEMUFile *f, SaveStateEntry *se)
2348 {
2349 int ret;
2350 uint8_t read_mark;
2351 uint32_t read_section_id;
2352
2353 if (!migrate_get_current()->send_section_footer) {
2354 /* No footer to check */
2355 return true;
2356 }
2357
2358 read_mark = qemu_get_byte(f);
2359
2360 ret = qemu_file_get_error(f);
2361 if (ret) {
2362 error_report("%s: Read section footer failed: %d",
2363 __func__, ret);
2364 return false;
2365 }
2366
2367 if (read_mark != QEMU_VM_SECTION_FOOTER) {
2368 error_report("Missing section footer for %s", se->idstr);
2369 return false;
2370 }
2371
2372 read_section_id = qemu_get_be32(f);
2373 if (read_section_id != se->load_section_id) {
2374 error_report("Mismatched section id in footer for %s -"
2375 " read 0x%x expected 0x%x",
2376 se->idstr, read_section_id, se->load_section_id);
2377 return false;
2378 }
2379
2380 /* All good */
2381 return true;
2382 }
2383
2384 static int
2385 qemu_loadvm_section_start_full(QEMUFile *f, MigrationIncomingState *mis)
2386 {
2387 uint32_t instance_id, version_id, section_id;
2388 SaveStateEntry *se;
2389 char idstr[256];
2390 int ret;
2391
2392 /* Read section start */
2393 section_id = qemu_get_be32(f);
2394 if (!qemu_get_counted_string(f, idstr)) {
2395 error_report("Unable to read ID string for section %u",
2396 section_id);
2397 return -EINVAL;
2398 }
2399 instance_id = qemu_get_be32(f);
2400 version_id = qemu_get_be32(f);
2401
2402 ret = qemu_file_get_error(f);
2403 if (ret) {
2404 error_report("%s: Failed to read instance/version ID: %d",
2405 __func__, ret);
2406 return ret;
2407 }
2408
2409 trace_qemu_loadvm_state_section_startfull(section_id, idstr,
2410 instance_id, version_id);
2411 /* Find savevm section */
2412 se = find_se(idstr, instance_id);
2413 if (se == NULL) {
2414 error_report("Unknown savevm section or instance '%s' %"PRIu32". "
2415 "Make sure that your current VM setup matches your "
2416 "saved VM setup, including any hotplugged devices",
2417 idstr, instance_id);
2418 return -EINVAL;
2419 }
2420
2421 /* Validate version */
2422 if (version_id > se->version_id) {
2423 error_report("savevm: unsupported version %d for '%s' v%d",
2424 version_id, idstr, se->version_id);
2425 return -EINVAL;
2426 }
2427 se->load_version_id = version_id;
2428 se->load_section_id = section_id;
2429
2430 /* Validate if it is a device's state */
2431 if (xen_enabled() && se->is_ram) {
2432 error_report("loadvm: %s RAM loading not allowed on Xen", idstr);
2433 return -EINVAL;
2434 }
2435
2436 ret = vmstate_load(f, se);
2437 if (ret < 0) {
2438 error_report("error while loading state for instance 0x%"PRIx32" of"
2439 " device '%s'", instance_id, idstr);
2440 return ret;
2441 }
2442 if (!check_section_footer(f, se)) {
2443 return -EINVAL;
2444 }
2445
2446 return 0;
2447 }
2448
2449 static int
2450 qemu_loadvm_section_part_end(QEMUFile *f, MigrationIncomingState *mis)
2451 {
2452 uint32_t section_id;
2453 SaveStateEntry *se;
2454 int ret;
2455
2456 section_id = qemu_get_be32(f);
2457
2458 ret = qemu_file_get_error(f);
2459 if (ret) {
2460 error_report("%s: Failed to read section ID: %d",
2461 __func__, ret);
2462 return ret;
2463 }
2464
2465 trace_qemu_loadvm_state_section_partend(section_id);
2466 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
2467 if (se->load_section_id == section_id) {
2468 break;
2469 }
2470 }
2471 if (se == NULL) {
2472 error_report("Unknown savevm section %d", section_id);
2473 return -EINVAL;
2474 }
2475
2476 ret = vmstate_load(f, se);
2477 if (ret < 0) {
2478 error_report("error while loading state section id %d(%s)",
2479 section_id, se->idstr);
2480 return ret;
2481 }
2482 if (!check_section_footer(f, se)) {
2483 return -EINVAL;
2484 }
2485
2486 return 0;
2487 }
2488
2489 static int qemu_loadvm_state_header(QEMUFile *f)
2490 {
2491 unsigned int v;
2492 int ret;
2493
2494 v = qemu_get_be32(f);
2495 if (v != QEMU_VM_FILE_MAGIC) {
2496 error_report("Not a migration stream");
2497 return -EINVAL;
2498 }
2499
2500 v = qemu_get_be32(f);
2501 if (v == QEMU_VM_FILE_VERSION_COMPAT) {
2502 error_report("SaveVM v2 format is obsolete and don't work anymore");
2503 return -ENOTSUP;
2504 }
2505 if (v != QEMU_VM_FILE_VERSION) {
2506 error_report("Unsupported migration stream version");
2507 return -ENOTSUP;
2508 }
2509
2510 if (migrate_get_current()->send_configuration) {
2511 if (qemu_get_byte(f) != QEMU_VM_CONFIGURATION) {
2512 error_report("Configuration section missing");
2513 qemu_loadvm_state_cleanup();
2514 return -EINVAL;
2515 }
2516 ret = vmstate_load_state(f, &vmstate_configuration, &savevm_state, 0);
2517
2518 if (ret) {
2519 qemu_loadvm_state_cleanup();
2520 return ret;
2521 }
2522 }
2523 return 0;
2524 }
2525
2526 static int qemu_loadvm_state_setup(QEMUFile *f)
2527 {
2528 SaveStateEntry *se;
2529 int ret;
2530
2531 trace_loadvm_state_setup();
2532 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
2533 if (!se->ops || !se->ops->load_setup) {
2534 continue;
2535 }
2536 if (se->ops->is_active) {
2537 if (!se->ops->is_active(se->opaque)) {
2538 continue;
2539 }
2540 }
2541
2542 ret = se->ops->load_setup(f, se->opaque);
2543 if (ret < 0) {
2544 qemu_file_set_error(f, ret);
2545 error_report("Load state of device %s failed", se->idstr);
2546 return ret;
2547 }
2548 }
2549 return 0;
2550 }
2551
2552 void qemu_loadvm_state_cleanup(void)
2553 {
2554 SaveStateEntry *se;
2555
2556 trace_loadvm_state_cleanup();
2557 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
2558 if (se->ops && se->ops->load_cleanup) {
2559 se->ops->load_cleanup(se->opaque);
2560 }
2561 }
2562 }
2563
2564 /* Return true if we should continue the migration, or false. */
2565 static bool postcopy_pause_incoming(MigrationIncomingState *mis)
2566 {
2567 trace_postcopy_pause_incoming();
2568
2569 assert(migrate_postcopy_ram());
2570
2571 /* Clear the triggered bit to allow one recovery */
2572 mis->postcopy_recover_triggered = false;
2573
2574 /*
2575 * Unregister yank with either from/to src would work, since ioc behind it
2576 * is the same
2577 */
2578 migration_ioc_unregister_yank_from_file(mis->from_src_file);
2579
2580 assert(mis->from_src_file);
2581 qemu_file_shutdown(mis->from_src_file);
2582 qemu_fclose(mis->from_src_file);
2583 mis->from_src_file = NULL;
2584
2585 assert(mis->to_src_file);
2586 qemu_file_shutdown(mis->to_src_file);
2587 qemu_mutex_lock(&mis->rp_mutex);
2588 qemu_fclose(mis->to_src_file);
2589 mis->to_src_file = NULL;
2590 qemu_mutex_unlock(&mis->rp_mutex);
2591
2592 migrate_set_state(&mis->state, MIGRATION_STATUS_POSTCOPY_ACTIVE,
2593 MIGRATION_STATUS_POSTCOPY_PAUSED);
2594
2595 /* Notify the fault thread for the invalidated file handle */
2596 postcopy_fault_thread_notify(mis);
2597
2598 error_report("Detected IO failure for postcopy. "
2599 "Migration paused.");
2600
2601 while (mis->state == MIGRATION_STATUS_POSTCOPY_PAUSED) {
2602 qemu_sem_wait(&mis->postcopy_pause_sem_dst);
2603 }
2604
2605 trace_postcopy_pause_incoming_continued();
2606
2607 return true;
2608 }
2609
2610 int qemu_loadvm_state_main(QEMUFile *f, MigrationIncomingState *mis)
2611 {
2612 uint8_t section_type;
2613 int ret = 0;
2614
2615 retry:
2616 while (true) {
2617 section_type = qemu_get_byte(f);
2618
2619 if (qemu_file_get_error(f)) {
2620 ret = qemu_file_get_error(f);
2621 break;
2622 }
2623
2624 trace_qemu_loadvm_state_section(section_type);
2625 switch (section_type) {
2626 case QEMU_VM_SECTION_START:
2627 case QEMU_VM_SECTION_FULL:
2628 ret = qemu_loadvm_section_start_full(f, mis);
2629 if (ret < 0) {
2630 goto out;
2631 }
2632 break;
2633 case QEMU_VM_SECTION_PART:
2634 case QEMU_VM_SECTION_END:
2635 ret = qemu_loadvm_section_part_end(f, mis);
2636 if (ret < 0) {
2637 goto out;
2638 }
2639 break;
2640 case QEMU_VM_COMMAND:
2641 ret = loadvm_process_command(f);
2642 trace_qemu_loadvm_state_section_command(ret);
2643 if ((ret < 0) || (ret == LOADVM_QUIT)) {
2644 goto out;
2645 }
2646 break;
2647 case QEMU_VM_EOF:
2648 /* This is the end of migration */
2649 goto out;
2650 default:
2651 error_report("Unknown savevm section type %d", section_type);
2652 ret = -EINVAL;
2653 goto out;
2654 }
2655 }
2656
2657 out:
2658 if (ret < 0) {
2659 qemu_file_set_error(f, ret);
2660
2661 /* Cancel bitmaps incoming regardless of recovery */
2662 dirty_bitmap_mig_cancel_incoming();
2663
2664 /*
2665 * If we are during an active postcopy, then we pause instead
2666 * of bail out to at least keep the VM's dirty data. Note
2667 * that POSTCOPY_INCOMING_LISTENING stage is still not enough,
2668 * during which we're still receiving device states and we
2669 * still haven't yet started the VM on destination.
2670 *
2671 * Only RAM postcopy supports recovery. Still, if RAM postcopy is
2672 * enabled, canceled bitmaps postcopy will not affect RAM postcopy
2673 * recovering.
2674 */
2675 if (postcopy_state_get() == POSTCOPY_INCOMING_RUNNING &&
2676 migrate_postcopy_ram() && postcopy_pause_incoming(mis)) {
2677 /* Reset f to point to the newly created channel */
2678 f = mis->from_src_file;
2679 goto retry;
2680 }
2681 }
2682 return ret;
2683 }
2684
2685 int qemu_loadvm_state(QEMUFile *f)
2686 {
2687 MigrationIncomingState *mis = migration_incoming_get_current();
2688 Error *local_err = NULL;
2689 int ret;
2690
2691 if (qemu_savevm_state_blocked(&local_err)) {
2692 error_report_err(local_err);
2693 return -EINVAL;
2694 }
2695
2696 ret = qemu_loadvm_state_header(f);
2697 if (ret) {
2698 return ret;
2699 }
2700
2701 if (qemu_loadvm_state_setup(f) != 0) {
2702 return -EINVAL;
2703 }
2704
2705 cpu_synchronize_all_pre_loadvm();
2706
2707 ret = qemu_loadvm_state_main(f, mis);
2708 qemu_event_set(&mis->main_thread_load_event);
2709
2710 trace_qemu_loadvm_state_post_main(ret);
2711
2712 if (mis->have_listen_thread) {
2713 /* Listen thread still going, can't clean up yet */
2714 return ret;
2715 }
2716
2717 if (ret == 0) {
2718 ret = qemu_file_get_error(f);
2719 }
2720
2721 /*
2722 * Try to read in the VMDESC section as well, so that dumping tools that
2723 * intercept our migration stream have the chance to see it.
2724 */
2725
2726 /* We've got to be careful; if we don't read the data and just shut the fd
2727 * then the sender can error if we close while it's still sending.
2728 * We also mustn't read data that isn't there; some transports (RDMA)
2729 * will stall waiting for that data when the source has already closed.
2730 */
2731 if (ret == 0 && should_send_vmdesc()) {
2732 uint8_t *buf;
2733 uint32_t size;
2734 uint8_t section_type = qemu_get_byte(f);
2735
2736 if (section_type != QEMU_VM_VMDESCRIPTION) {
2737 error_report("Expected vmdescription section, but got %d",
2738 section_type);
2739 /*
2740 * It doesn't seem worth failing at this point since
2741 * we apparently have an otherwise valid VM state
2742 */
2743 } else {
2744 buf = g_malloc(0x1000);
2745 size = qemu_get_be32(f);
2746
2747 while (size > 0) {
2748 uint32_t read_chunk = MIN(size, 0x1000);
2749 qemu_get_buffer(f, buf, read_chunk);
2750 size -= read_chunk;
2751 }
2752 g_free(buf);
2753 }
2754 }
2755
2756 qemu_loadvm_state_cleanup();
2757 cpu_synchronize_all_post_init();
2758
2759 return ret;
2760 }
2761
2762 int qemu_load_device_state(QEMUFile *f)
2763 {
2764 MigrationIncomingState *mis = migration_incoming_get_current();
2765 int ret;
2766
2767 /* Load QEMU_VM_SECTION_FULL section */
2768 ret = qemu_loadvm_state_main(f, mis);
2769 if (ret < 0) {
2770 error_report("Failed to load device state: %d", ret);
2771 return ret;
2772 }
2773
2774 cpu_synchronize_all_post_init();
2775 return 0;
2776 }
2777
2778 bool save_snapshot(const char *name, bool overwrite, const char *vmstate,
2779 bool has_devices, strList *devices, Error **errp)
2780 {
2781 BlockDriverState *bs;
2782 QEMUSnapshotInfo sn1, *sn = &sn1;
2783 int ret = -1, ret2;
2784 QEMUFile *f;
2785 int saved_vm_running;
2786 uint64_t vm_state_size;
2787 g_autoptr(GDateTime) now = g_date_time_new_now_local();
2788 AioContext *aio_context;
2789
2790 if (migration_is_blocked(errp)) {
2791 return false;
2792 }
2793
2794 if (!replay_can_snapshot()) {
2795 error_setg(errp, "Record/replay does not allow making snapshot "
2796 "right now. Try once more later.");
2797 return false;
2798 }
2799
2800 if (!bdrv_all_can_snapshot(has_devices, devices, errp)) {
2801 return false;
2802 }
2803
2804 /* Delete old snapshots of the same name */
2805 if (name) {
2806 if (overwrite) {
2807 if (bdrv_all_delete_snapshot(name, has_devices,
2808 devices, errp) < 0) {
2809 return false;
2810 }
2811 } else {
2812 ret2 = bdrv_all_has_snapshot(name, has_devices, devices, errp);
2813 if (ret2 < 0) {
2814 return false;
2815 }
2816 if (ret2 == 1) {
2817 error_setg(errp,
2818 "Snapshot '%s' already exists in one or more devices",
2819 name);
2820 return false;
2821 }
2822 }
2823 }
2824
2825 bs = bdrv_all_find_vmstate_bs(vmstate, has_devices, devices, errp);
2826 if (bs == NULL) {
2827 return false;
2828 }
2829 aio_context = bdrv_get_aio_context(bs);
2830
2831 saved_vm_running = runstate_is_running();
2832
2833 ret = global_state_store();
2834 if (ret) {
2835 error_setg(errp, "Error saving global state");
2836 return false;
2837 }
2838 vm_stop(RUN_STATE_SAVE_VM);
2839
2840 bdrv_drain_all_begin();
2841
2842 aio_context_acquire(aio_context);
2843
2844 memset(sn, 0, sizeof(*sn));
2845
2846 /* fill auxiliary fields */
2847 sn->date_sec = g_date_time_to_unix(now);
2848 sn->date_nsec = g_date_time_get_microsecond(now) * 1000;
2849 sn->vm_clock_nsec = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL);
2850 if (replay_mode != REPLAY_MODE_NONE) {
2851 sn->icount = replay_get_current_icount();
2852 } else {
2853 sn->icount = -1ULL;
2854 }
2855
2856 if (name) {
2857 pstrcpy(sn->name, sizeof(sn->name), name);
2858 } else {
2859 g_autofree char *autoname = g_date_time_format(now, "vm-%Y%m%d%H%M%S");
2860 pstrcpy(sn->name, sizeof(sn->name), autoname);
2861 }
2862
2863 /* save the VM state */
2864 f = qemu_fopen_bdrv(bs, 1);
2865 if (!f) {
2866 error_setg(errp, "Could not open VM state file");
2867 goto the_end;
2868 }
2869 ret = qemu_savevm_state(f, errp);
2870 vm_state_size = qemu_ftell(f);
2871 ret2 = qemu_fclose(f);
2872 if (ret < 0) {
2873 goto the_end;
2874 }
2875 if (ret2 < 0) {
2876 ret = ret2;
2877 goto the_end;
2878 }
2879
2880 /* The bdrv_all_create_snapshot() call that follows acquires the AioContext
2881 * for itself. BDRV_POLL_WHILE() does not support nested locking because
2882 * it only releases the lock once. Therefore synchronous I/O will deadlock
2883 * unless we release the AioContext before bdrv_all_create_snapshot().
2884 */
2885 aio_context_release(aio_context);
2886 aio_context = NULL;
2887
2888 ret = bdrv_all_create_snapshot(sn, bs, vm_state_size,
2889 has_devices, devices, errp);
2890 if (ret < 0) {
2891 bdrv_all_delete_snapshot(sn->name, has_devices, devices, NULL);
2892 goto the_end;
2893 }
2894
2895 ret = 0;
2896
2897 the_end:
2898 if (aio_context) {
2899 aio_context_release(aio_context);
2900 }
2901
2902 bdrv_drain_all_end();
2903
2904 if (saved_vm_running) {
2905 vm_start();
2906 }
2907 return ret == 0;
2908 }
2909
2910 void qmp_xen_save_devices_state(const char *filename, bool has_live, bool live,
2911 Error **errp)
2912 {
2913 QEMUFile *f;
2914 QIOChannelFile *ioc;
2915 int saved_vm_running;
2916 int ret;
2917
2918 if (!has_live) {
2919 /* live default to true so old version of Xen tool stack can have a
2920 * successful live migration */
2921 live = true;
2922 }
2923
2924 saved_vm_running = runstate_is_running();
2925 vm_stop(RUN_STATE_SAVE_VM);
2926 global_state_store_running();
2927
2928 ioc = qio_channel_file_new_path(filename, O_WRONLY | O_CREAT | O_TRUNC,
2929 0660, errp);
2930 if (!ioc) {
2931 goto the_end;
2932 }
2933 qio_channel_set_name(QIO_CHANNEL(ioc), "migration-xen-save-state");
2934 f = qemu_fopen_channel_output(QIO_CHANNEL(ioc));
2935 object_unref(OBJECT(ioc));
2936 ret = qemu_save_device_state(f);
2937 if (ret < 0 || qemu_fclose(f) < 0) {
2938 error_setg(errp, QERR_IO_ERROR);
2939 } else {
2940 /* libxl calls the QMP command "stop" before calling
2941 * "xen-save-devices-state" and in case of migration failure, libxl
2942 * would call "cont".
2943 * So call bdrv_inactivate_all (release locks) here to let the other
2944 * side of the migration take control of the images.
2945 */
2946 if (live && !saved_vm_running) {
2947 ret = bdrv_inactivate_all();
2948 if (ret) {
2949 error_setg(errp, "%s: bdrv_inactivate_all() failed (%d)",
2950 __func__, ret);
2951 }
2952 }
2953 }
2954
2955 the_end:
2956 if (saved_vm_running) {
2957 vm_start();
2958 }
2959 }
2960
2961 void qmp_xen_load_devices_state(const char *filename, Error **errp)
2962 {
2963 QEMUFile *f;
2964 QIOChannelFile *ioc;
2965 int ret;
2966
2967 /* Guest must be paused before loading the device state; the RAM state
2968 * will already have been loaded by xc
2969 */
2970 if (runstate_is_running()) {
2971 error_setg(errp, "Cannot update device state while vm is running");
2972 return;
2973 }
2974 vm_stop(RUN_STATE_RESTORE_VM);
2975
2976 ioc = qio_channel_file_new_path(filename, O_RDONLY | O_BINARY, 0, errp);
2977 if (!ioc) {
2978 return;
2979 }
2980 qio_channel_set_name(QIO_CHANNEL(ioc), "migration-xen-load-state");
2981 f = qemu_fopen_channel_input(QIO_CHANNEL(ioc));
2982 object_unref(OBJECT(ioc));
2983
2984 ret = qemu_loadvm_state(f);
2985 qemu_fclose(f);
2986 if (ret < 0) {
2987 error_setg(errp, QERR_IO_ERROR);
2988 }
2989 migration_incoming_state_destroy();
2990 }
2991
2992 bool load_snapshot(const char *name, const char *vmstate,
2993 bool has_devices, strList *devices, Error **errp)
2994 {
2995 BlockDriverState *bs_vm_state;
2996 QEMUSnapshotInfo sn;
2997 QEMUFile *f;
2998 int ret;
2999 AioContext *aio_context;
3000 MigrationIncomingState *mis = migration_incoming_get_current();
3001
3002 if (!bdrv_all_can_snapshot(has_devices, devices, errp)) {
3003 return false;
3004 }
3005 ret = bdrv_all_has_snapshot(name, has_devices, devices, errp);
3006 if (ret < 0) {
3007 return false;
3008 }
3009 if (ret == 0) {
3010 error_setg(errp, "Snapshot '%s' does not exist in one or more devices",
3011 name);
3012 return false;
3013 }
3014
3015 bs_vm_state = bdrv_all_find_vmstate_bs(vmstate, has_devices, devices, errp);
3016 if (!bs_vm_state) {
3017 return false;
3018 }
3019 aio_context = bdrv_get_aio_context(bs_vm_state);
3020
3021 /* Don't even try to load empty VM states */
3022 aio_context_acquire(aio_context);
3023 ret = bdrv_snapshot_find(bs_vm_state, &sn, name);
3024 aio_context_release(aio_context);
3025 if (ret < 0) {
3026 return false;
3027 } else if (sn.vm_state_size == 0) {
3028 error_setg(errp, "This is a disk-only snapshot. Revert to it "
3029 " offline using qemu-img");
3030 return false;
3031 }
3032
3033 /*
3034 * Flush the record/replay queue. Now the VM state is going
3035 * to change. Therefore we don't need to preserve its consistency
3036 */
3037 replay_flush_events();
3038
3039 /* Flush all IO requests so they don't interfere with the new state. */
3040 bdrv_drain_all_begin();
3041
3042 ret = bdrv_all_goto_snapshot(name, has_devices, devices, errp);
3043 if (ret < 0) {
3044 goto err_drain;
3045 }
3046
3047 /* restore the VM state */
3048 f = qemu_fopen_bdrv(bs_vm_state, 0);
3049 if (!f) {
3050 error_setg(errp, "Could not open VM state file");
3051 goto err_drain;
3052 }
3053
3054 qemu_system_reset(SHUTDOWN_CAUSE_NONE);
3055 mis->from_src_file = f;
3056
3057 if (!yank_register_instance(MIGRATION_YANK_INSTANCE, errp)) {
3058 ret = -EINVAL;
3059 goto err_drain;
3060 }
3061 aio_context_acquire(aio_context);
3062 ret = qemu_loadvm_state(f);
3063 migration_incoming_state_destroy();
3064 aio_context_release(aio_context);
3065
3066 bdrv_drain_all_end();
3067
3068 if (ret < 0) {
3069 error_setg(errp, "Error %d while loading VM state", ret);
3070 return false;
3071 }
3072
3073 return true;
3074
3075 err_drain:
3076 bdrv_drain_all_end();
3077 return false;
3078 }
3079
3080 bool delete_snapshot(const char *name, bool has_devices,
3081 strList *devices, Error **errp)
3082 {
3083 if (!bdrv_all_can_snapshot(has_devices, devices, errp)) {
3084 return false;
3085 }
3086
3087 if (bdrv_all_delete_snapshot(name, has_devices, devices, errp) < 0) {
3088 return false;
3089 }
3090
3091 return true;
3092 }
3093
3094 void vmstate_register_ram(MemoryRegion *mr, DeviceState *dev)
3095 {
3096 qemu_ram_set_idstr(mr->ram_block,
3097 memory_region_name(mr), dev);
3098 qemu_ram_set_migratable(mr->ram_block);
3099 }
3100
3101 void vmstate_unregister_ram(MemoryRegion *mr, DeviceState *dev)
3102 {
3103 qemu_ram_unset_idstr(mr->ram_block);
3104 qemu_ram_unset_migratable(mr->ram_block);
3105 }
3106
3107 void vmstate_register_ram_global(MemoryRegion *mr)
3108 {
3109 vmstate_register_ram(mr, NULL);
3110 }
3111
3112 bool vmstate_check_only_migratable(const VMStateDescription *vmsd)
3113 {
3114 /* check needed if --only-migratable is specified */
3115 if (!only_migratable) {
3116 return true;
3117 }
3118
3119 return !(vmsd && vmsd->unmigratable);
3120 }
3121
3122 typedef struct SnapshotJob {
3123 Job common;
3124 char *tag;
3125 char *vmstate;
3126 strList *devices;
3127 Coroutine *co;
3128 Error **errp;
3129 bool ret;
3130 } SnapshotJob;
3131
3132 static void qmp_snapshot_job_free(SnapshotJob *s)
3133 {
3134 g_free(s->tag);
3135 g_free(s->vmstate);
3136 qapi_free_strList(s->devices);
3137 }
3138
3139
3140 static void snapshot_load_job_bh(void *opaque)
3141 {
3142 Job *job = opaque;
3143 SnapshotJob *s = container_of(job, SnapshotJob, common);
3144 int orig_vm_running;
3145
3146 job_progress_set_remaining(&s->common, 1);
3147
3148 orig_vm_running = runstate_is_running();
3149 vm_stop(RUN_STATE_RESTORE_VM);
3150
3151 s->ret = load_snapshot(s->tag, s->vmstate, true, s->devices, s->errp);
3152 if (s->ret && orig_vm_running) {
3153 vm_start();
3154 }
3155
3156 job_progress_update(&s->common, 1);
3157
3158 qmp_snapshot_job_free(s);
3159 aio_co_wake(s->co);
3160 }
3161
3162 static void snapshot_save_job_bh(void *opaque)
3163 {
3164 Job *job = opaque;
3165 SnapshotJob *s = container_of(job, SnapshotJob, common);
3166
3167 job_progress_set_remaining(&s->common, 1);
3168 s->ret = save_snapshot(s->tag, false, s->vmstate,
3169 true, s->devices, s->errp);
3170 job_progress_update(&s->common, 1);
3171
3172 qmp_snapshot_job_free(s);
3173 aio_co_wake(s->co);
3174 }
3175
3176 static void snapshot_delete_job_bh(void *opaque)
3177 {
3178 Job *job = opaque;
3179 SnapshotJob *s = container_of(job, SnapshotJob, common);
3180
3181 job_progress_set_remaining(&s->common, 1);
3182 s->ret = delete_snapshot(s->tag, true, s->devices, s->errp);
3183 job_progress_update(&s->common, 1);
3184
3185 qmp_snapshot_job_free(s);
3186 aio_co_wake(s->co);
3187 }
3188
3189 static int coroutine_fn snapshot_save_job_run(Job *job, Error **errp)
3190 {
3191 SnapshotJob *s = container_of(job, SnapshotJob, common);
3192 s->errp = errp;
3193 s->co = qemu_coroutine_self();
3194 aio_bh_schedule_oneshot(qemu_get_aio_context(),
3195 snapshot_save_job_bh, job);
3196 qemu_coroutine_yield();
3197 return s->ret ? 0 : -1;
3198 }
3199
3200 static int coroutine_fn snapshot_load_job_run(Job *job, Error **errp)
3201 {
3202 SnapshotJob *s = container_of(job, SnapshotJob, common);
3203 s->errp = errp;
3204 s->co = qemu_coroutine_self();
3205 aio_bh_schedule_oneshot(qemu_get_aio_context(),
3206 snapshot_load_job_bh, job);
3207 qemu_coroutine_yield();
3208 return s->ret ? 0 : -1;
3209 }
3210
3211 static int coroutine_fn snapshot_delete_job_run(Job *job, Error **errp)
3212 {
3213 SnapshotJob *s = container_of(job, SnapshotJob, common);
3214 s->errp = errp;
3215 s->co = qemu_coroutine_self();
3216 aio_bh_schedule_oneshot(qemu_get_aio_context(),
3217 snapshot_delete_job_bh, job);
3218 qemu_coroutine_yield();
3219 return s->ret ? 0 : -1;
3220 }
3221
3222
3223 static const JobDriver snapshot_load_job_driver = {
3224 .instance_size = sizeof(SnapshotJob),
3225 .job_type = JOB_TYPE_SNAPSHOT_LOAD,
3226 .run = snapshot_load_job_run,
3227 };
3228
3229 static const JobDriver snapshot_save_job_driver = {
3230 .instance_size = sizeof(SnapshotJob),
3231 .job_type = JOB_TYPE_SNAPSHOT_SAVE,
3232 .run = snapshot_save_job_run,
3233 };
3234
3235 static const JobDriver snapshot_delete_job_driver = {
3236 .instance_size = sizeof(SnapshotJob),
3237 .job_type = JOB_TYPE_SNAPSHOT_DELETE,
3238 .run = snapshot_delete_job_run,
3239 };
3240
3241
3242 void qmp_snapshot_save(const char *job_id,
3243 const char *tag,
3244 const char *vmstate,
3245 strList *devices,
3246 Error **errp)
3247 {
3248 SnapshotJob *s;
3249
3250 s = job_create(job_id, &snapshot_save_job_driver, NULL,
3251 qemu_get_aio_context(), JOB_MANUAL_DISMISS,
3252 NULL, NULL, errp);
3253 if (!s) {
3254 return;
3255 }
3256
3257 s->tag = g_strdup(tag);
3258 s->vmstate = g_strdup(vmstate);
3259 s->devices = QAPI_CLONE(strList, devices);
3260
3261 job_start(&s->common);
3262 }
3263
3264 void qmp_snapshot_load(const char *job_id,
3265 const char *tag,
3266 const char *vmstate,
3267 strList *devices,
3268 Error **errp)
3269 {
3270 SnapshotJob *s;
3271
3272 s = job_create(job_id, &snapshot_load_job_driver, NULL,
3273 qemu_get_aio_context(), JOB_MANUAL_DISMISS,
3274 NULL, NULL, errp);
3275 if (!s) {
3276 return;
3277 }
3278
3279 s->tag = g_strdup(tag);
3280 s->vmstate = g_strdup(vmstate);
3281 s->devices = QAPI_CLONE(strList, devices);
3282
3283 job_start(&s->common);
3284 }
3285
3286 void qmp_snapshot_delete(const char *job_id,
3287 const char *tag,
3288 strList *devices,
3289 Error **errp)
3290 {
3291 SnapshotJob *s;
3292
3293 s = job_create(job_id, &snapshot_delete_job_driver, NULL,
3294 qemu_get_aio_context(), JOB_MANUAL_DISMISS,
3295 NULL, NULL, errp);
3296 if (!s) {
3297 return;
3298 }
3299
3300 s->tag = g_strdup(tag);
3301 s->devices = QAPI_CLONE(strList, devices);
3302
3303 job_start(&s->common);
3304 }