Merge remote-tracking branch 'remotes/alistair23/tags/pull-riscv-to-apply-20211022...
[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
69 const unsigned int postcopy_ram_discard_version;
70
71 /* Subcommands for QEMU_VM_COMMAND */
72 enum qemu_vm_cmd {
73 MIG_CMD_INVALID = 0, /* Must be 0 */
74 MIG_CMD_OPEN_RETURN_PATH, /* Tell the dest to open the Return path */
75 MIG_CMD_PING, /* Request a PONG on the RP */
76
77 MIG_CMD_POSTCOPY_ADVISE, /* Prior to any page transfers, just
78 warn we might want to do PC */
79 MIG_CMD_POSTCOPY_LISTEN, /* Start listening for incoming
80 pages as it's running. */
81 MIG_CMD_POSTCOPY_RUN, /* Start execution */
82
83 MIG_CMD_POSTCOPY_RAM_DISCARD, /* A list of pages to discard that
84 were previously sent during
85 precopy but are dirty. */
86 MIG_CMD_PACKAGED, /* Send a wrapped stream within this stream */
87 MIG_CMD_ENABLE_COLO, /* Enable COLO */
88 MIG_CMD_POSTCOPY_RESUME, /* resume postcopy on dest */
89 MIG_CMD_RECV_BITMAP, /* Request for recved bitmap on dst */
90 MIG_CMD_MAX
91 };
92
93 #define MAX_VM_CMD_PACKAGED_SIZE UINT32_MAX
94 static struct mig_cmd_args {
95 ssize_t len; /* -1 = variable */
96 const char *name;
97 } mig_cmd_args[] = {
98 [MIG_CMD_INVALID] = { .len = -1, .name = "INVALID" },
99 [MIG_CMD_OPEN_RETURN_PATH] = { .len = 0, .name = "OPEN_RETURN_PATH" },
100 [MIG_CMD_PING] = { .len = sizeof(uint32_t), .name = "PING" },
101 [MIG_CMD_POSTCOPY_ADVISE] = { .len = -1, .name = "POSTCOPY_ADVISE" },
102 [MIG_CMD_POSTCOPY_LISTEN] = { .len = 0, .name = "POSTCOPY_LISTEN" },
103 [MIG_CMD_POSTCOPY_RUN] = { .len = 0, .name = "POSTCOPY_RUN" },
104 [MIG_CMD_POSTCOPY_RAM_DISCARD] = {
105 .len = -1, .name = "POSTCOPY_RAM_DISCARD" },
106 [MIG_CMD_POSTCOPY_RESUME] = { .len = 0, .name = "POSTCOPY_RESUME" },
107 [MIG_CMD_PACKAGED] = { .len = 4, .name = "PACKAGED" },
108 [MIG_CMD_RECV_BITMAP] = { .len = -1, .name = "RECV_BITMAP" },
109 [MIG_CMD_MAX] = { .len = -1, .name = "MAX" },
110 };
111
112 /* Note for MIG_CMD_POSTCOPY_ADVISE:
113 * The format of arguments is depending on postcopy mode:
114 * - postcopy RAM only
115 * uint64_t host page size
116 * uint64_t taget page size
117 *
118 * - postcopy RAM and postcopy dirty bitmaps
119 * format is the same as for postcopy RAM only
120 *
121 * - postcopy dirty bitmaps only
122 * Nothing. Command length field is 0.
123 *
124 * Be careful: adding a new postcopy entity with some other parameters should
125 * not break format self-description ability. Good way is to introduce some
126 * generic extendable format with an exception for two old entities.
127 */
128
129 /***********************************************************/
130 /* savevm/loadvm support */
131
132 static ssize_t block_writev_buffer(void *opaque, struct iovec *iov, int iovcnt,
133 int64_t pos, Error **errp)
134 {
135 int ret;
136 QEMUIOVector qiov;
137
138 qemu_iovec_init_external(&qiov, iov, iovcnt);
139 ret = bdrv_writev_vmstate(opaque, &qiov, pos);
140 if (ret < 0) {
141 return ret;
142 }
143
144 return qiov.size;
145 }
146
147 static ssize_t block_get_buffer(void *opaque, uint8_t *buf, int64_t pos,
148 size_t size, Error **errp)
149 {
150 return bdrv_load_vmstate(opaque, buf, pos, size);
151 }
152
153 static int bdrv_fclose(void *opaque, Error **errp)
154 {
155 return bdrv_flush(opaque);
156 }
157
158 static const QEMUFileOps bdrv_read_ops = {
159 .get_buffer = block_get_buffer,
160 .close = bdrv_fclose
161 };
162
163 static const QEMUFileOps bdrv_write_ops = {
164 .writev_buffer = block_writev_buffer,
165 .close = bdrv_fclose
166 };
167
168 static QEMUFile *qemu_fopen_bdrv(BlockDriverState *bs, int is_writable)
169 {
170 if (is_writable) {
171 return qemu_fopen_ops(bs, &bdrv_write_ops);
172 }
173 return qemu_fopen_ops(bs, &bdrv_read_ops);
174 }
175
176
177 /* QEMUFile timer support.
178 * Not in qemu-file.c to not add qemu-timer.c as dependency to qemu-file.c
179 */
180
181 void timer_put(QEMUFile *f, QEMUTimer *ts)
182 {
183 uint64_t expire_time;
184
185 expire_time = timer_expire_time_ns(ts);
186 qemu_put_be64(f, expire_time);
187 }
188
189 void timer_get(QEMUFile *f, QEMUTimer *ts)
190 {
191 uint64_t expire_time;
192
193 expire_time = qemu_get_be64(f);
194 if (expire_time != -1) {
195 timer_mod_ns(ts, expire_time);
196 } else {
197 timer_del(ts);
198 }
199 }
200
201
202 /* VMState timer support.
203 * Not in vmstate.c to not add qemu-timer.c as dependency to vmstate.c
204 */
205
206 static int get_timer(QEMUFile *f, void *pv, size_t size,
207 const VMStateField *field)
208 {
209 QEMUTimer *v = pv;
210 timer_get(f, v);
211 return 0;
212 }
213
214 static int put_timer(QEMUFile *f, void *pv, size_t size,
215 const VMStateField *field, JSONWriter *vmdesc)
216 {
217 QEMUTimer *v = pv;
218 timer_put(f, v);
219
220 return 0;
221 }
222
223 const VMStateInfo vmstate_info_timer = {
224 .name = "timer",
225 .get = get_timer,
226 .put = put_timer,
227 };
228
229
230 typedef struct CompatEntry {
231 char idstr[256];
232 int instance_id;
233 } CompatEntry;
234
235 typedef struct SaveStateEntry {
236 QTAILQ_ENTRY(SaveStateEntry) entry;
237 char idstr[256];
238 uint32_t instance_id;
239 int alias_id;
240 int version_id;
241 /* version id read from the stream */
242 int load_version_id;
243 int section_id;
244 /* section id read from the stream */
245 int load_section_id;
246 const SaveVMHandlers *ops;
247 const VMStateDescription *vmsd;
248 void *opaque;
249 CompatEntry *compat;
250 int is_ram;
251 } SaveStateEntry;
252
253 typedef struct SaveState {
254 QTAILQ_HEAD(, SaveStateEntry) handlers;
255 SaveStateEntry *handler_pri_head[MIG_PRI_MAX + 1];
256 int global_section_id;
257 uint32_t len;
258 const char *name;
259 uint32_t target_page_bits;
260 uint32_t caps_count;
261 MigrationCapability *capabilities;
262 QemuUUID uuid;
263 } SaveState;
264
265 static SaveState savevm_state = {
266 .handlers = QTAILQ_HEAD_INITIALIZER(savevm_state.handlers),
267 .handler_pri_head = { [MIG_PRI_DEFAULT ... MIG_PRI_MAX] = NULL },
268 .global_section_id = 0,
269 };
270
271 static bool should_validate_capability(int capability)
272 {
273 assert(capability >= 0 && capability < MIGRATION_CAPABILITY__MAX);
274 /* Validate only new capabilities to keep compatibility. */
275 switch (capability) {
276 case MIGRATION_CAPABILITY_X_IGNORE_SHARED:
277 return true;
278 default:
279 return false;
280 }
281 }
282
283 static uint32_t get_validatable_capabilities_count(void)
284 {
285 MigrationState *s = migrate_get_current();
286 uint32_t result = 0;
287 int i;
288 for (i = 0; i < MIGRATION_CAPABILITY__MAX; i++) {
289 if (should_validate_capability(i) && s->enabled_capabilities[i]) {
290 result++;
291 }
292 }
293 return result;
294 }
295
296 static int configuration_pre_save(void *opaque)
297 {
298 SaveState *state = opaque;
299 const char *current_name = MACHINE_GET_CLASS(current_machine)->name;
300 MigrationState *s = migrate_get_current();
301 int i, j;
302
303 state->len = strlen(current_name);
304 state->name = current_name;
305 state->target_page_bits = qemu_target_page_bits();
306
307 state->caps_count = get_validatable_capabilities_count();
308 state->capabilities = g_renew(MigrationCapability, state->capabilities,
309 state->caps_count);
310 for (i = j = 0; i < MIGRATION_CAPABILITY__MAX; i++) {
311 if (should_validate_capability(i) && s->enabled_capabilities[i]) {
312 state->capabilities[j++] = i;
313 }
314 }
315 state->uuid = qemu_uuid;
316
317 return 0;
318 }
319
320 static int configuration_post_save(void *opaque)
321 {
322 SaveState *state = opaque;
323
324 g_free(state->capabilities);
325 state->capabilities = NULL;
326 state->caps_count = 0;
327 return 0;
328 }
329
330 static int configuration_pre_load(void *opaque)
331 {
332 SaveState *state = opaque;
333
334 /* If there is no target-page-bits subsection it means the source
335 * predates the variable-target-page-bits support and is using the
336 * minimum possible value for this CPU.
337 */
338 state->target_page_bits = qemu_target_page_bits_min();
339 return 0;
340 }
341
342 static bool configuration_validate_capabilities(SaveState *state)
343 {
344 bool ret = true;
345 MigrationState *s = migrate_get_current();
346 unsigned long *source_caps_bm;
347 int i;
348
349 source_caps_bm = bitmap_new(MIGRATION_CAPABILITY__MAX);
350 for (i = 0; i < state->caps_count; i++) {
351 MigrationCapability capability = state->capabilities[i];
352 set_bit(capability, source_caps_bm);
353 }
354
355 for (i = 0; i < MIGRATION_CAPABILITY__MAX; i++) {
356 bool source_state, target_state;
357 if (!should_validate_capability(i)) {
358 continue;
359 }
360 source_state = test_bit(i, source_caps_bm);
361 target_state = s->enabled_capabilities[i];
362 if (source_state != target_state) {
363 error_report("Capability %s is %s, but received capability is %s",
364 MigrationCapability_str(i),
365 target_state ? "on" : "off",
366 source_state ? "on" : "off");
367 ret = false;
368 /* Don't break here to report all failed capabilities */
369 }
370 }
371
372 g_free(source_caps_bm);
373 return ret;
374 }
375
376 static int configuration_post_load(void *opaque, int version_id)
377 {
378 SaveState *state = opaque;
379 const char *current_name = MACHINE_GET_CLASS(current_machine)->name;
380 int ret = 0;
381
382 if (strncmp(state->name, current_name, state->len) != 0) {
383 error_report("Machine type received is '%.*s' and local is '%s'",
384 (int) state->len, state->name, current_name);
385 ret = -EINVAL;
386 goto out;
387 }
388
389 if (state->target_page_bits != qemu_target_page_bits()) {
390 error_report("Received TARGET_PAGE_BITS is %d but local is %d",
391 state->target_page_bits, qemu_target_page_bits());
392 ret = -EINVAL;
393 goto out;
394 }
395
396 if (!configuration_validate_capabilities(state)) {
397 ret = -EINVAL;
398 goto out;
399 }
400
401 out:
402 g_free((void *)state->name);
403 state->name = NULL;
404 state->len = 0;
405 g_free(state->capabilities);
406 state->capabilities = NULL;
407 state->caps_count = 0;
408
409 return ret;
410 }
411
412 static int get_capability(QEMUFile *f, void *pv, size_t size,
413 const VMStateField *field)
414 {
415 MigrationCapability *capability = pv;
416 char capability_str[UINT8_MAX + 1];
417 uint8_t len;
418 int i;
419
420 len = qemu_get_byte(f);
421 qemu_get_buffer(f, (uint8_t *)capability_str, len);
422 capability_str[len] = '\0';
423 for (i = 0; i < MIGRATION_CAPABILITY__MAX; i++) {
424 if (!strcmp(MigrationCapability_str(i), capability_str)) {
425 *capability = i;
426 return 0;
427 }
428 }
429 error_report("Received unknown capability %s", capability_str);
430 return -EINVAL;
431 }
432
433 static int put_capability(QEMUFile *f, void *pv, size_t size,
434 const VMStateField *field, JSONWriter *vmdesc)
435 {
436 MigrationCapability *capability = pv;
437 const char *capability_str = MigrationCapability_str(*capability);
438 size_t len = strlen(capability_str);
439 assert(len <= UINT8_MAX);
440
441 qemu_put_byte(f, len);
442 qemu_put_buffer(f, (uint8_t *)capability_str, len);
443 return 0;
444 }
445
446 static const VMStateInfo vmstate_info_capability = {
447 .name = "capability",
448 .get = get_capability,
449 .put = put_capability,
450 };
451
452 /* The target-page-bits subsection is present only if the
453 * target page size is not the same as the default (ie the
454 * minimum page size for a variable-page-size guest CPU).
455 * If it is present then it contains the actual target page
456 * bits for the machine, and migration will fail if the
457 * two ends don't agree about it.
458 */
459 static bool vmstate_target_page_bits_needed(void *opaque)
460 {
461 return qemu_target_page_bits()
462 > qemu_target_page_bits_min();
463 }
464
465 static const VMStateDescription vmstate_target_page_bits = {
466 .name = "configuration/target-page-bits",
467 .version_id = 1,
468 .minimum_version_id = 1,
469 .needed = vmstate_target_page_bits_needed,
470 .fields = (VMStateField[]) {
471 VMSTATE_UINT32(target_page_bits, SaveState),
472 VMSTATE_END_OF_LIST()
473 }
474 };
475
476 static bool vmstate_capabilites_needed(void *opaque)
477 {
478 return get_validatable_capabilities_count() > 0;
479 }
480
481 static const VMStateDescription vmstate_capabilites = {
482 .name = "configuration/capabilities",
483 .version_id = 1,
484 .minimum_version_id = 1,
485 .needed = vmstate_capabilites_needed,
486 .fields = (VMStateField[]) {
487 VMSTATE_UINT32_V(caps_count, SaveState, 1),
488 VMSTATE_VARRAY_UINT32_ALLOC(capabilities, SaveState, caps_count, 1,
489 vmstate_info_capability,
490 MigrationCapability),
491 VMSTATE_END_OF_LIST()
492 }
493 };
494
495 static bool vmstate_uuid_needed(void *opaque)
496 {
497 return qemu_uuid_set && migrate_validate_uuid();
498 }
499
500 static int vmstate_uuid_post_load(void *opaque, int version_id)
501 {
502 SaveState *state = opaque;
503 char uuid_src[UUID_FMT_LEN + 1];
504 char uuid_dst[UUID_FMT_LEN + 1];
505
506 if (!qemu_uuid_set) {
507 /*
508 * It's warning because user might not know UUID in some cases,
509 * e.g. load an old snapshot
510 */
511 qemu_uuid_unparse(&state->uuid, uuid_src);
512 warn_report("UUID is received %s, but local uuid isn't set",
513 uuid_src);
514 return 0;
515 }
516 if (!qemu_uuid_is_equal(&state->uuid, &qemu_uuid)) {
517 qemu_uuid_unparse(&state->uuid, uuid_src);
518 qemu_uuid_unparse(&qemu_uuid, uuid_dst);
519 error_report("UUID received is %s and local is %s", uuid_src, uuid_dst);
520 return -EINVAL;
521 }
522 return 0;
523 }
524
525 static const VMStateDescription vmstate_uuid = {
526 .name = "configuration/uuid",
527 .version_id = 1,
528 .minimum_version_id = 1,
529 .needed = vmstate_uuid_needed,
530 .post_load = vmstate_uuid_post_load,
531 .fields = (VMStateField[]) {
532 VMSTATE_UINT8_ARRAY_V(uuid.data, SaveState, sizeof(QemuUUID), 1),
533 VMSTATE_END_OF_LIST()
534 }
535 };
536
537 static const VMStateDescription vmstate_configuration = {
538 .name = "configuration",
539 .version_id = 1,
540 .pre_load = configuration_pre_load,
541 .post_load = configuration_post_load,
542 .pre_save = configuration_pre_save,
543 .post_save = configuration_post_save,
544 .fields = (VMStateField[]) {
545 VMSTATE_UINT32(len, SaveState),
546 VMSTATE_VBUFFER_ALLOC_UINT32(name, SaveState, 0, NULL, len),
547 VMSTATE_END_OF_LIST()
548 },
549 .subsections = (const VMStateDescription *[]) {
550 &vmstate_target_page_bits,
551 &vmstate_capabilites,
552 &vmstate_uuid,
553 NULL
554 }
555 };
556
557 static void dump_vmstate_vmsd(FILE *out_file,
558 const VMStateDescription *vmsd, int indent,
559 bool is_subsection);
560
561 static void dump_vmstate_vmsf(FILE *out_file, const VMStateField *field,
562 int indent)
563 {
564 fprintf(out_file, "%*s{\n", indent, "");
565 indent += 2;
566 fprintf(out_file, "%*s\"field\": \"%s\",\n", indent, "", field->name);
567 fprintf(out_file, "%*s\"version_id\": %d,\n", indent, "",
568 field->version_id);
569 fprintf(out_file, "%*s\"field_exists\": %s,\n", indent, "",
570 field->field_exists ? "true" : "false");
571 fprintf(out_file, "%*s\"size\": %zu", indent, "", field->size);
572 if (field->vmsd != NULL) {
573 fprintf(out_file, ",\n");
574 dump_vmstate_vmsd(out_file, field->vmsd, indent, false);
575 }
576 fprintf(out_file, "\n%*s}", indent - 2, "");
577 }
578
579 static void dump_vmstate_vmss(FILE *out_file,
580 const VMStateDescription **subsection,
581 int indent)
582 {
583 if (*subsection != NULL) {
584 dump_vmstate_vmsd(out_file, *subsection, indent, true);
585 }
586 }
587
588 static void dump_vmstate_vmsd(FILE *out_file,
589 const VMStateDescription *vmsd, int indent,
590 bool is_subsection)
591 {
592 if (is_subsection) {
593 fprintf(out_file, "%*s{\n", indent, "");
594 } else {
595 fprintf(out_file, "%*s\"%s\": {\n", indent, "", "Description");
596 }
597 indent += 2;
598 fprintf(out_file, "%*s\"name\": \"%s\",\n", indent, "", vmsd->name);
599 fprintf(out_file, "%*s\"version_id\": %d,\n", indent, "",
600 vmsd->version_id);
601 fprintf(out_file, "%*s\"minimum_version_id\": %d", indent, "",
602 vmsd->minimum_version_id);
603 if (vmsd->fields != NULL) {
604 const VMStateField *field = vmsd->fields;
605 bool first;
606
607 fprintf(out_file, ",\n%*s\"Fields\": [\n", indent, "");
608 first = true;
609 while (field->name != NULL) {
610 if (field->flags & VMS_MUST_EXIST) {
611 /* Ignore VMSTATE_VALIDATE bits; these don't get migrated */
612 field++;
613 continue;
614 }
615 if (!first) {
616 fprintf(out_file, ",\n");
617 }
618 dump_vmstate_vmsf(out_file, field, indent + 2);
619 field++;
620 first = false;
621 }
622 fprintf(out_file, "\n%*s]", indent, "");
623 }
624 if (vmsd->subsections != NULL) {
625 const VMStateDescription **subsection = vmsd->subsections;
626 bool first;
627
628 fprintf(out_file, ",\n%*s\"Subsections\": [\n", indent, "");
629 first = true;
630 while (*subsection != NULL) {
631 if (!first) {
632 fprintf(out_file, ",\n");
633 }
634 dump_vmstate_vmss(out_file, subsection, indent + 2);
635 subsection++;
636 first = false;
637 }
638 fprintf(out_file, "\n%*s]", indent, "");
639 }
640 fprintf(out_file, "\n%*s}", indent - 2, "");
641 }
642
643 static void dump_machine_type(FILE *out_file)
644 {
645 MachineClass *mc;
646
647 mc = MACHINE_GET_CLASS(current_machine);
648
649 fprintf(out_file, " \"vmschkmachine\": {\n");
650 fprintf(out_file, " \"Name\": \"%s\"\n", mc->name);
651 fprintf(out_file, " },\n");
652 }
653
654 void dump_vmstate_json_to_file(FILE *out_file)
655 {
656 GSList *list, *elt;
657 bool first;
658
659 fprintf(out_file, "{\n");
660 dump_machine_type(out_file);
661
662 first = true;
663 list = object_class_get_list(TYPE_DEVICE, true);
664 for (elt = list; elt; elt = elt->next) {
665 DeviceClass *dc = OBJECT_CLASS_CHECK(DeviceClass, elt->data,
666 TYPE_DEVICE);
667 const char *name;
668 int indent = 2;
669
670 if (!dc->vmsd) {
671 continue;
672 }
673
674 if (!first) {
675 fprintf(out_file, ",\n");
676 }
677 name = object_class_get_name(OBJECT_CLASS(dc));
678 fprintf(out_file, "%*s\"%s\": {\n", indent, "", name);
679 indent += 2;
680 fprintf(out_file, "%*s\"Name\": \"%s\",\n", indent, "", name);
681 fprintf(out_file, "%*s\"version_id\": %d,\n", indent, "",
682 dc->vmsd->version_id);
683 fprintf(out_file, "%*s\"minimum_version_id\": %d,\n", indent, "",
684 dc->vmsd->minimum_version_id);
685
686 dump_vmstate_vmsd(out_file, dc->vmsd, indent, false);
687
688 fprintf(out_file, "\n%*s}", indent - 2, "");
689 first = false;
690 }
691 fprintf(out_file, "\n}\n");
692 fclose(out_file);
693 g_slist_free(list);
694 }
695
696 static uint32_t calculate_new_instance_id(const char *idstr)
697 {
698 SaveStateEntry *se;
699 uint32_t instance_id = 0;
700
701 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
702 if (strcmp(idstr, se->idstr) == 0
703 && instance_id <= se->instance_id) {
704 instance_id = se->instance_id + 1;
705 }
706 }
707 /* Make sure we never loop over without being noticed */
708 assert(instance_id != VMSTATE_INSTANCE_ID_ANY);
709 return instance_id;
710 }
711
712 static int calculate_compat_instance_id(const char *idstr)
713 {
714 SaveStateEntry *se;
715 int instance_id = 0;
716
717 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
718 if (!se->compat) {
719 continue;
720 }
721
722 if (strcmp(idstr, se->compat->idstr) == 0
723 && instance_id <= se->compat->instance_id) {
724 instance_id = se->compat->instance_id + 1;
725 }
726 }
727 return instance_id;
728 }
729
730 static inline MigrationPriority save_state_priority(SaveStateEntry *se)
731 {
732 if (se->vmsd) {
733 return se->vmsd->priority;
734 }
735 return MIG_PRI_DEFAULT;
736 }
737
738 static void savevm_state_handler_insert(SaveStateEntry *nse)
739 {
740 MigrationPriority priority = save_state_priority(nse);
741 SaveStateEntry *se;
742 int i;
743
744 assert(priority <= MIG_PRI_MAX);
745
746 for (i = priority - 1; i >= 0; i--) {
747 se = savevm_state.handler_pri_head[i];
748 if (se != NULL) {
749 assert(save_state_priority(se) < priority);
750 break;
751 }
752 }
753
754 if (i >= 0) {
755 QTAILQ_INSERT_BEFORE(se, nse, entry);
756 } else {
757 QTAILQ_INSERT_TAIL(&savevm_state.handlers, nse, entry);
758 }
759
760 if (savevm_state.handler_pri_head[priority] == NULL) {
761 savevm_state.handler_pri_head[priority] = nse;
762 }
763 }
764
765 static void savevm_state_handler_remove(SaveStateEntry *se)
766 {
767 SaveStateEntry *next;
768 MigrationPriority priority = save_state_priority(se);
769
770 if (se == savevm_state.handler_pri_head[priority]) {
771 next = QTAILQ_NEXT(se, entry);
772 if (next != NULL && save_state_priority(next) == priority) {
773 savevm_state.handler_pri_head[priority] = next;
774 } else {
775 savevm_state.handler_pri_head[priority] = NULL;
776 }
777 }
778 QTAILQ_REMOVE(&savevm_state.handlers, se, entry);
779 }
780
781 /* TODO: Individual devices generally have very little idea about the rest
782 of the system, so instance_id should be removed/replaced.
783 Meanwhile pass -1 as instance_id if you do not already have a clearly
784 distinguishing id for all instances of your device class. */
785 int register_savevm_live(const char *idstr,
786 uint32_t instance_id,
787 int version_id,
788 const SaveVMHandlers *ops,
789 void *opaque)
790 {
791 SaveStateEntry *se;
792
793 se = g_new0(SaveStateEntry, 1);
794 se->version_id = version_id;
795 se->section_id = savevm_state.global_section_id++;
796 se->ops = ops;
797 se->opaque = opaque;
798 se->vmsd = NULL;
799 /* if this is a live_savem then set is_ram */
800 if (ops->save_setup != NULL) {
801 se->is_ram = 1;
802 }
803
804 pstrcat(se->idstr, sizeof(se->idstr), idstr);
805
806 if (instance_id == VMSTATE_INSTANCE_ID_ANY) {
807 se->instance_id = calculate_new_instance_id(se->idstr);
808 } else {
809 se->instance_id = instance_id;
810 }
811 assert(!se->compat || se->instance_id == 0);
812 savevm_state_handler_insert(se);
813 return 0;
814 }
815
816 void unregister_savevm(VMStateIf *obj, const char *idstr, void *opaque)
817 {
818 SaveStateEntry *se, *new_se;
819 char id[256] = "";
820
821 if (obj) {
822 char *oid = vmstate_if_get_id(obj);
823 if (oid) {
824 pstrcpy(id, sizeof(id), oid);
825 pstrcat(id, sizeof(id), "/");
826 g_free(oid);
827 }
828 }
829 pstrcat(id, sizeof(id), idstr);
830
831 QTAILQ_FOREACH_SAFE(se, &savevm_state.handlers, entry, new_se) {
832 if (strcmp(se->idstr, id) == 0 && se->opaque == opaque) {
833 savevm_state_handler_remove(se);
834 g_free(se->compat);
835 g_free(se);
836 }
837 }
838 }
839
840 int vmstate_register_with_alias_id(VMStateIf *obj, uint32_t instance_id,
841 const VMStateDescription *vmsd,
842 void *opaque, int alias_id,
843 int required_for_version,
844 Error **errp)
845 {
846 SaveStateEntry *se;
847
848 /* If this triggers, alias support can be dropped for the vmsd. */
849 assert(alias_id == -1 || required_for_version >= vmsd->minimum_version_id);
850
851 se = g_new0(SaveStateEntry, 1);
852 se->version_id = vmsd->version_id;
853 se->section_id = savevm_state.global_section_id++;
854 se->opaque = opaque;
855 se->vmsd = vmsd;
856 se->alias_id = alias_id;
857
858 if (obj) {
859 char *id = vmstate_if_get_id(obj);
860 if (id) {
861 if (snprintf(se->idstr, sizeof(se->idstr), "%s/", id) >=
862 sizeof(se->idstr)) {
863 error_setg(errp, "Path too long for VMState (%s)", id);
864 g_free(id);
865 g_free(se);
866
867 return -1;
868 }
869 g_free(id);
870
871 se->compat = g_new0(CompatEntry, 1);
872 pstrcpy(se->compat->idstr, sizeof(se->compat->idstr), vmsd->name);
873 se->compat->instance_id = instance_id == VMSTATE_INSTANCE_ID_ANY ?
874 calculate_compat_instance_id(vmsd->name) : instance_id;
875 instance_id = VMSTATE_INSTANCE_ID_ANY;
876 }
877 }
878 pstrcat(se->idstr, sizeof(se->idstr), vmsd->name);
879
880 if (instance_id == VMSTATE_INSTANCE_ID_ANY) {
881 se->instance_id = calculate_new_instance_id(se->idstr);
882 } else {
883 se->instance_id = instance_id;
884 }
885 assert(!se->compat || se->instance_id == 0);
886 savevm_state_handler_insert(se);
887 return 0;
888 }
889
890 void vmstate_unregister(VMStateIf *obj, const VMStateDescription *vmsd,
891 void *opaque)
892 {
893 SaveStateEntry *se, *new_se;
894
895 QTAILQ_FOREACH_SAFE(se, &savevm_state.handlers, entry, new_se) {
896 if (se->vmsd == vmsd && se->opaque == opaque) {
897 savevm_state_handler_remove(se);
898 g_free(se->compat);
899 g_free(se);
900 }
901 }
902 }
903
904 static int vmstate_load(QEMUFile *f, SaveStateEntry *se)
905 {
906 trace_vmstate_load(se->idstr, se->vmsd ? se->vmsd->name : "(old)");
907 if (!se->vmsd) { /* Old style */
908 return se->ops->load_state(f, se->opaque, se->load_version_id);
909 }
910 return vmstate_load_state(f, se->vmsd, se->opaque, se->load_version_id);
911 }
912
913 static void vmstate_save_old_style(QEMUFile *f, SaveStateEntry *se,
914 JSONWriter *vmdesc)
915 {
916 int64_t old_offset, size;
917
918 old_offset = qemu_ftell_fast(f);
919 se->ops->save_state(f, se->opaque);
920 size = qemu_ftell_fast(f) - old_offset;
921
922 if (vmdesc) {
923 json_writer_int64(vmdesc, "size", size);
924 json_writer_start_array(vmdesc, "fields");
925 json_writer_start_object(vmdesc, NULL);
926 json_writer_str(vmdesc, "name", "data");
927 json_writer_int64(vmdesc, "size", size);
928 json_writer_str(vmdesc, "type", "buffer");
929 json_writer_end_object(vmdesc);
930 json_writer_end_array(vmdesc);
931 }
932 }
933
934 static int vmstate_save(QEMUFile *f, SaveStateEntry *se,
935 JSONWriter *vmdesc)
936 {
937 trace_vmstate_save(se->idstr, se->vmsd ? se->vmsd->name : "(old)");
938 if (!se->vmsd) {
939 vmstate_save_old_style(f, se, vmdesc);
940 return 0;
941 }
942 return vmstate_save_state(f, se->vmsd, se->opaque, vmdesc);
943 }
944
945 /*
946 * Write the header for device section (QEMU_VM_SECTION START/END/PART/FULL)
947 */
948 static void save_section_header(QEMUFile *f, SaveStateEntry *se,
949 uint8_t section_type)
950 {
951 qemu_put_byte(f, section_type);
952 qemu_put_be32(f, se->section_id);
953
954 if (section_type == QEMU_VM_SECTION_FULL ||
955 section_type == QEMU_VM_SECTION_START) {
956 /* ID string */
957 size_t len = strlen(se->idstr);
958 qemu_put_byte(f, len);
959 qemu_put_buffer(f, (uint8_t *)se->idstr, len);
960
961 qemu_put_be32(f, se->instance_id);
962 qemu_put_be32(f, se->version_id);
963 }
964 }
965
966 /*
967 * Write a footer onto device sections that catches cases misformatted device
968 * sections.
969 */
970 static void save_section_footer(QEMUFile *f, SaveStateEntry *se)
971 {
972 if (migrate_get_current()->send_section_footer) {
973 qemu_put_byte(f, QEMU_VM_SECTION_FOOTER);
974 qemu_put_be32(f, se->section_id);
975 }
976 }
977
978 /**
979 * qemu_savevm_command_send: Send a 'QEMU_VM_COMMAND' type element with the
980 * command and associated data.
981 *
982 * @f: File to send command on
983 * @command: Command type to send
984 * @len: Length of associated data
985 * @data: Data associated with command.
986 */
987 static void qemu_savevm_command_send(QEMUFile *f,
988 enum qemu_vm_cmd command,
989 uint16_t len,
990 uint8_t *data)
991 {
992 trace_savevm_command_send(command, len);
993 qemu_put_byte(f, QEMU_VM_COMMAND);
994 qemu_put_be16(f, (uint16_t)command);
995 qemu_put_be16(f, len);
996 qemu_put_buffer(f, data, len);
997 qemu_fflush(f);
998 }
999
1000 void qemu_savevm_send_colo_enable(QEMUFile *f)
1001 {
1002 trace_savevm_send_colo_enable();
1003 qemu_savevm_command_send(f, MIG_CMD_ENABLE_COLO, 0, NULL);
1004 }
1005
1006 void qemu_savevm_send_ping(QEMUFile *f, uint32_t value)
1007 {
1008 uint32_t buf;
1009
1010 trace_savevm_send_ping(value);
1011 buf = cpu_to_be32(value);
1012 qemu_savevm_command_send(f, MIG_CMD_PING, sizeof(value), (uint8_t *)&buf);
1013 }
1014
1015 void qemu_savevm_send_open_return_path(QEMUFile *f)
1016 {
1017 trace_savevm_send_open_return_path();
1018 qemu_savevm_command_send(f, MIG_CMD_OPEN_RETURN_PATH, 0, NULL);
1019 }
1020
1021 /* We have a buffer of data to send; we don't want that all to be loaded
1022 * by the command itself, so the command contains just the length of the
1023 * extra buffer that we then send straight after it.
1024 * TODO: Must be a better way to organise that
1025 *
1026 * Returns:
1027 * 0 on success
1028 * -ve on error
1029 */
1030 int qemu_savevm_send_packaged(QEMUFile *f, const uint8_t *buf, size_t len)
1031 {
1032 uint32_t tmp;
1033
1034 if (len > MAX_VM_CMD_PACKAGED_SIZE) {
1035 error_report("%s: Unreasonably large packaged state: %zu",
1036 __func__, len);
1037 return -1;
1038 }
1039
1040 tmp = cpu_to_be32(len);
1041
1042 trace_qemu_savevm_send_packaged();
1043 qemu_savevm_command_send(f, MIG_CMD_PACKAGED, 4, (uint8_t *)&tmp);
1044
1045 qemu_put_buffer(f, buf, len);
1046
1047 return 0;
1048 }
1049
1050 /* Send prior to any postcopy transfer */
1051 void qemu_savevm_send_postcopy_advise(QEMUFile *f)
1052 {
1053 if (migrate_postcopy_ram()) {
1054 uint64_t tmp[2];
1055 tmp[0] = cpu_to_be64(ram_pagesize_summary());
1056 tmp[1] = cpu_to_be64(qemu_target_page_size());
1057
1058 trace_qemu_savevm_send_postcopy_advise();
1059 qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_ADVISE,
1060 16, (uint8_t *)tmp);
1061 } else {
1062 qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_ADVISE, 0, NULL);
1063 }
1064 }
1065
1066 /* Sent prior to starting the destination running in postcopy, discard pages
1067 * that have already been sent but redirtied on the source.
1068 * CMD_POSTCOPY_RAM_DISCARD consist of:
1069 * byte version (0)
1070 * byte Length of name field (not including 0)
1071 * n x byte RAM block name
1072 * byte 0 terminator (just for safety)
1073 * n x Byte ranges within the named RAMBlock
1074 * be64 Start of the range
1075 * be64 Length
1076 *
1077 * name: RAMBlock name that these entries are part of
1078 * len: Number of page entries
1079 * start_list: 'len' addresses
1080 * length_list: 'len' addresses
1081 *
1082 */
1083 void qemu_savevm_send_postcopy_ram_discard(QEMUFile *f, const char *name,
1084 uint16_t len,
1085 uint64_t *start_list,
1086 uint64_t *length_list)
1087 {
1088 uint8_t *buf;
1089 uint16_t tmplen;
1090 uint16_t t;
1091 size_t name_len = strlen(name);
1092
1093 trace_qemu_savevm_send_postcopy_ram_discard(name, len);
1094 assert(name_len < 256);
1095 buf = g_malloc0(1 + 1 + name_len + 1 + (8 + 8) * len);
1096 buf[0] = postcopy_ram_discard_version;
1097 buf[1] = name_len;
1098 memcpy(buf + 2, name, name_len);
1099 tmplen = 2 + name_len;
1100 buf[tmplen++] = '\0';
1101
1102 for (t = 0; t < len; t++) {
1103 stq_be_p(buf + tmplen, start_list[t]);
1104 tmplen += 8;
1105 stq_be_p(buf + tmplen, length_list[t]);
1106 tmplen += 8;
1107 }
1108 qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_RAM_DISCARD, tmplen, buf);
1109 g_free(buf);
1110 }
1111
1112 /* Get the destination into a state where it can receive postcopy data. */
1113 void qemu_savevm_send_postcopy_listen(QEMUFile *f)
1114 {
1115 trace_savevm_send_postcopy_listen();
1116 qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_LISTEN, 0, NULL);
1117 }
1118
1119 /* Kick the destination into running */
1120 void qemu_savevm_send_postcopy_run(QEMUFile *f)
1121 {
1122 trace_savevm_send_postcopy_run();
1123 qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_RUN, 0, NULL);
1124 }
1125
1126 void qemu_savevm_send_postcopy_resume(QEMUFile *f)
1127 {
1128 trace_savevm_send_postcopy_resume();
1129 qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_RESUME, 0, NULL);
1130 }
1131
1132 void qemu_savevm_send_recv_bitmap(QEMUFile *f, char *block_name)
1133 {
1134 size_t len;
1135 char buf[256];
1136
1137 trace_savevm_send_recv_bitmap(block_name);
1138
1139 buf[0] = len = strlen(block_name);
1140 memcpy(buf + 1, block_name, len);
1141
1142 qemu_savevm_command_send(f, MIG_CMD_RECV_BITMAP, len + 1, (uint8_t *)buf);
1143 }
1144
1145 bool qemu_savevm_state_blocked(Error **errp)
1146 {
1147 SaveStateEntry *se;
1148
1149 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1150 if (se->vmsd && se->vmsd->unmigratable) {
1151 error_setg(errp, "State blocked by non-migratable device '%s'",
1152 se->idstr);
1153 return true;
1154 }
1155 }
1156 return false;
1157 }
1158
1159 void qemu_savevm_non_migratable_list(strList **reasons)
1160 {
1161 SaveStateEntry *se;
1162
1163 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1164 if (se->vmsd && se->vmsd->unmigratable) {
1165 QAPI_LIST_PREPEND(*reasons,
1166 g_strdup_printf("non-migratable device: %s",
1167 se->idstr));
1168 }
1169 }
1170 }
1171
1172 void qemu_savevm_state_header(QEMUFile *f)
1173 {
1174 trace_savevm_state_header();
1175 qemu_put_be32(f, QEMU_VM_FILE_MAGIC);
1176 qemu_put_be32(f, QEMU_VM_FILE_VERSION);
1177
1178 if (migrate_get_current()->send_configuration) {
1179 qemu_put_byte(f, QEMU_VM_CONFIGURATION);
1180 vmstate_save_state(f, &vmstate_configuration, &savevm_state, 0);
1181 }
1182 }
1183
1184 bool qemu_savevm_state_guest_unplug_pending(void)
1185 {
1186 SaveStateEntry *se;
1187
1188 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1189 if (se->vmsd && se->vmsd->dev_unplug_pending &&
1190 se->vmsd->dev_unplug_pending(se->opaque)) {
1191 return true;
1192 }
1193 }
1194
1195 return false;
1196 }
1197
1198 void qemu_savevm_state_setup(QEMUFile *f)
1199 {
1200 SaveStateEntry *se;
1201 Error *local_err = NULL;
1202 int ret;
1203
1204 trace_savevm_state_setup();
1205 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1206 if (!se->ops || !se->ops->save_setup) {
1207 continue;
1208 }
1209 if (se->ops->is_active) {
1210 if (!se->ops->is_active(se->opaque)) {
1211 continue;
1212 }
1213 }
1214 save_section_header(f, se, QEMU_VM_SECTION_START);
1215
1216 ret = se->ops->save_setup(f, se->opaque);
1217 save_section_footer(f, se);
1218 if (ret < 0) {
1219 qemu_file_set_error(f, ret);
1220 break;
1221 }
1222 }
1223
1224 if (precopy_notify(PRECOPY_NOTIFY_SETUP, &local_err)) {
1225 error_report_err(local_err);
1226 }
1227 }
1228
1229 int qemu_savevm_state_resume_prepare(MigrationState *s)
1230 {
1231 SaveStateEntry *se;
1232 int ret;
1233
1234 trace_savevm_state_resume_prepare();
1235
1236 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1237 if (!se->ops || !se->ops->resume_prepare) {
1238 continue;
1239 }
1240 if (se->ops->is_active) {
1241 if (!se->ops->is_active(se->opaque)) {
1242 continue;
1243 }
1244 }
1245 ret = se->ops->resume_prepare(s, se->opaque);
1246 if (ret < 0) {
1247 return ret;
1248 }
1249 }
1250
1251 return 0;
1252 }
1253
1254 /*
1255 * this function has three return values:
1256 * negative: there was one error, and we have -errno.
1257 * 0 : We haven't finished, caller have to go again
1258 * 1 : We have finished, we can go to complete phase
1259 */
1260 int qemu_savevm_state_iterate(QEMUFile *f, bool postcopy)
1261 {
1262 SaveStateEntry *se;
1263 int ret = 1;
1264
1265 trace_savevm_state_iterate();
1266 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1267 if (!se->ops || !se->ops->save_live_iterate) {
1268 continue;
1269 }
1270 if (se->ops->is_active &&
1271 !se->ops->is_active(se->opaque)) {
1272 continue;
1273 }
1274 if (se->ops->is_active_iterate &&
1275 !se->ops->is_active_iterate(se->opaque)) {
1276 continue;
1277 }
1278 /*
1279 * In the postcopy phase, any device that doesn't know how to
1280 * do postcopy should have saved it's state in the _complete
1281 * call that's already run, it might get confused if we call
1282 * iterate afterwards.
1283 */
1284 if (postcopy &&
1285 !(se->ops->has_postcopy && se->ops->has_postcopy(se->opaque))) {
1286 continue;
1287 }
1288 if (qemu_file_rate_limit(f)) {
1289 return 0;
1290 }
1291 trace_savevm_section_start(se->idstr, se->section_id);
1292
1293 save_section_header(f, se, QEMU_VM_SECTION_PART);
1294
1295 ret = se->ops->save_live_iterate(f, se->opaque);
1296 trace_savevm_section_end(se->idstr, se->section_id, ret);
1297 save_section_footer(f, se);
1298
1299 if (ret < 0) {
1300 error_report("failed to save SaveStateEntry with id(name): %d(%s)",
1301 se->section_id, se->idstr);
1302 qemu_file_set_error(f, ret);
1303 }
1304 if (ret <= 0) {
1305 /* Do not proceed to the next vmstate before this one reported
1306 completion of the current stage. This serializes the migration
1307 and reduces the probability that a faster changing state is
1308 synchronized over and over again. */
1309 break;
1310 }
1311 }
1312 return ret;
1313 }
1314
1315 static bool should_send_vmdesc(void)
1316 {
1317 MachineState *machine = MACHINE(qdev_get_machine());
1318 bool in_postcopy = migration_in_postcopy();
1319 return !machine->suppress_vmdesc && !in_postcopy;
1320 }
1321
1322 /*
1323 * Calls the save_live_complete_postcopy methods
1324 * causing the last few pages to be sent immediately and doing any associated
1325 * cleanup.
1326 * Note postcopy also calls qemu_savevm_state_complete_precopy to complete
1327 * all the other devices, but that happens at the point we switch to postcopy.
1328 */
1329 void qemu_savevm_state_complete_postcopy(QEMUFile *f)
1330 {
1331 SaveStateEntry *se;
1332 int ret;
1333
1334 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1335 if (!se->ops || !se->ops->save_live_complete_postcopy) {
1336 continue;
1337 }
1338 if (se->ops->is_active) {
1339 if (!se->ops->is_active(se->opaque)) {
1340 continue;
1341 }
1342 }
1343 trace_savevm_section_start(se->idstr, se->section_id);
1344 /* Section type */
1345 qemu_put_byte(f, QEMU_VM_SECTION_END);
1346 qemu_put_be32(f, se->section_id);
1347
1348 ret = se->ops->save_live_complete_postcopy(f, se->opaque);
1349 trace_savevm_section_end(se->idstr, se->section_id, ret);
1350 save_section_footer(f, se);
1351 if (ret < 0) {
1352 qemu_file_set_error(f, ret);
1353 return;
1354 }
1355 }
1356
1357 qemu_put_byte(f, QEMU_VM_EOF);
1358 qemu_fflush(f);
1359 }
1360
1361 static
1362 int qemu_savevm_state_complete_precopy_iterable(QEMUFile *f, bool in_postcopy)
1363 {
1364 SaveStateEntry *se;
1365 int ret;
1366
1367 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1368 if (!se->ops ||
1369 (in_postcopy && se->ops->has_postcopy &&
1370 se->ops->has_postcopy(se->opaque)) ||
1371 !se->ops->save_live_complete_precopy) {
1372 continue;
1373 }
1374
1375 if (se->ops->is_active) {
1376 if (!se->ops->is_active(se->opaque)) {
1377 continue;
1378 }
1379 }
1380 trace_savevm_section_start(se->idstr, se->section_id);
1381
1382 save_section_header(f, se, QEMU_VM_SECTION_END);
1383
1384 ret = se->ops->save_live_complete_precopy(f, se->opaque);
1385 trace_savevm_section_end(se->idstr, se->section_id, ret);
1386 save_section_footer(f, se);
1387 if (ret < 0) {
1388 qemu_file_set_error(f, ret);
1389 return -1;
1390 }
1391 }
1392
1393 return 0;
1394 }
1395
1396 int qemu_savevm_state_complete_precopy_non_iterable(QEMUFile *f,
1397 bool in_postcopy,
1398 bool inactivate_disks)
1399 {
1400 g_autoptr(JSONWriter) vmdesc = NULL;
1401 int vmdesc_len;
1402 SaveStateEntry *se;
1403 int ret;
1404
1405 vmdesc = json_writer_new(false);
1406 json_writer_start_object(vmdesc, NULL);
1407 json_writer_int64(vmdesc, "page_size", qemu_target_page_size());
1408 json_writer_start_array(vmdesc, "devices");
1409 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1410
1411 if ((!se->ops || !se->ops->save_state) && !se->vmsd) {
1412 continue;
1413 }
1414 if (se->vmsd && !vmstate_save_needed(se->vmsd, se->opaque)) {
1415 trace_savevm_section_skip(se->idstr, se->section_id);
1416 continue;
1417 }
1418
1419 trace_savevm_section_start(se->idstr, se->section_id);
1420
1421 json_writer_start_object(vmdesc, NULL);
1422 json_writer_str(vmdesc, "name", se->idstr);
1423 json_writer_int64(vmdesc, "instance_id", se->instance_id);
1424
1425 save_section_header(f, se, QEMU_VM_SECTION_FULL);
1426 ret = vmstate_save(f, se, vmdesc);
1427 if (ret) {
1428 qemu_file_set_error(f, ret);
1429 return ret;
1430 }
1431 trace_savevm_section_end(se->idstr, se->section_id, 0);
1432 save_section_footer(f, se);
1433
1434 json_writer_end_object(vmdesc);
1435 }
1436
1437 if (inactivate_disks) {
1438 /* Inactivate before sending QEMU_VM_EOF so that the
1439 * bdrv_invalidate_cache_all() on the other end won't fail. */
1440 ret = bdrv_inactivate_all();
1441 if (ret) {
1442 error_report("%s: bdrv_inactivate_all() failed (%d)",
1443 __func__, ret);
1444 qemu_file_set_error(f, ret);
1445 return ret;
1446 }
1447 }
1448 if (!in_postcopy) {
1449 /* Postcopy stream will still be going */
1450 qemu_put_byte(f, QEMU_VM_EOF);
1451 }
1452
1453 json_writer_end_array(vmdesc);
1454 json_writer_end_object(vmdesc);
1455 vmdesc_len = strlen(json_writer_get(vmdesc));
1456
1457 if (should_send_vmdesc()) {
1458 qemu_put_byte(f, QEMU_VM_VMDESCRIPTION);
1459 qemu_put_be32(f, vmdesc_len);
1460 qemu_put_buffer(f, (uint8_t *)json_writer_get(vmdesc), vmdesc_len);
1461 }
1462
1463 return 0;
1464 }
1465
1466 int qemu_savevm_state_complete_precopy(QEMUFile *f, bool iterable_only,
1467 bool inactivate_disks)
1468 {
1469 int ret;
1470 Error *local_err = NULL;
1471 bool in_postcopy = migration_in_postcopy();
1472
1473 if (precopy_notify(PRECOPY_NOTIFY_COMPLETE, &local_err)) {
1474 error_report_err(local_err);
1475 }
1476
1477 trace_savevm_state_complete_precopy();
1478
1479 cpu_synchronize_all_states();
1480
1481 if (!in_postcopy || iterable_only) {
1482 ret = qemu_savevm_state_complete_precopy_iterable(f, in_postcopy);
1483 if (ret) {
1484 return ret;
1485 }
1486 }
1487
1488 if (iterable_only) {
1489 goto flush;
1490 }
1491
1492 ret = qemu_savevm_state_complete_precopy_non_iterable(f, in_postcopy,
1493 inactivate_disks);
1494 if (ret) {
1495 return ret;
1496 }
1497
1498 flush:
1499 qemu_fflush(f);
1500 return 0;
1501 }
1502
1503 /* Give an estimate of the amount left to be transferred,
1504 * the result is split into the amount for units that can and
1505 * for units that can't do postcopy.
1506 */
1507 void qemu_savevm_state_pending(QEMUFile *f, uint64_t threshold_size,
1508 uint64_t *res_precopy_only,
1509 uint64_t *res_compatible,
1510 uint64_t *res_postcopy_only)
1511 {
1512 SaveStateEntry *se;
1513
1514 *res_precopy_only = 0;
1515 *res_compatible = 0;
1516 *res_postcopy_only = 0;
1517
1518
1519 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1520 if (!se->ops || !se->ops->save_live_pending) {
1521 continue;
1522 }
1523 if (se->ops->is_active) {
1524 if (!se->ops->is_active(se->opaque)) {
1525 continue;
1526 }
1527 }
1528 se->ops->save_live_pending(f, se->opaque, threshold_size,
1529 res_precopy_only, res_compatible,
1530 res_postcopy_only);
1531 }
1532 }
1533
1534 void qemu_savevm_state_cleanup(void)
1535 {
1536 SaveStateEntry *se;
1537 Error *local_err = NULL;
1538
1539 if (precopy_notify(PRECOPY_NOTIFY_CLEANUP, &local_err)) {
1540 error_report_err(local_err);
1541 }
1542
1543 trace_savevm_state_cleanup();
1544 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1545 if (se->ops && se->ops->save_cleanup) {
1546 se->ops->save_cleanup(se->opaque);
1547 }
1548 }
1549 }
1550
1551 static int qemu_savevm_state(QEMUFile *f, Error **errp)
1552 {
1553 int ret;
1554 MigrationState *ms = migrate_get_current();
1555 MigrationStatus status;
1556
1557 if (migration_is_running(ms->state)) {
1558 error_setg(errp, QERR_MIGRATION_ACTIVE);
1559 return -EINVAL;
1560 }
1561
1562 if (migrate_use_block()) {
1563 error_setg(errp, "Block migration and snapshots are incompatible");
1564 return -EINVAL;
1565 }
1566
1567 migrate_init(ms);
1568 memset(&ram_counters, 0, sizeof(ram_counters));
1569 ms->to_dst_file = f;
1570
1571 qemu_mutex_unlock_iothread();
1572 qemu_savevm_state_header(f);
1573 qemu_savevm_state_setup(f);
1574 qemu_mutex_lock_iothread();
1575
1576 while (qemu_file_get_error(f) == 0) {
1577 if (qemu_savevm_state_iterate(f, false) > 0) {
1578 break;
1579 }
1580 }
1581
1582 ret = qemu_file_get_error(f);
1583 if (ret == 0) {
1584 qemu_savevm_state_complete_precopy(f, false, false);
1585 ret = qemu_file_get_error(f);
1586 }
1587 qemu_savevm_state_cleanup();
1588 if (ret != 0) {
1589 error_setg_errno(errp, -ret, "Error while writing VM state");
1590 }
1591
1592 if (ret != 0) {
1593 status = MIGRATION_STATUS_FAILED;
1594 } else {
1595 status = MIGRATION_STATUS_COMPLETED;
1596 }
1597 migrate_set_state(&ms->state, MIGRATION_STATUS_SETUP, status);
1598
1599 /* f is outer parameter, it should not stay in global migration state after
1600 * this function finished */
1601 ms->to_dst_file = NULL;
1602
1603 return ret;
1604 }
1605
1606 void qemu_savevm_live_state(QEMUFile *f)
1607 {
1608 /* save QEMU_VM_SECTION_END section */
1609 qemu_savevm_state_complete_precopy(f, true, false);
1610 qemu_put_byte(f, QEMU_VM_EOF);
1611 }
1612
1613 int qemu_save_device_state(QEMUFile *f)
1614 {
1615 SaveStateEntry *se;
1616
1617 if (!migration_in_colo_state()) {
1618 qemu_put_be32(f, QEMU_VM_FILE_MAGIC);
1619 qemu_put_be32(f, QEMU_VM_FILE_VERSION);
1620 }
1621 cpu_synchronize_all_states();
1622
1623 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1624 int ret;
1625
1626 if (se->is_ram) {
1627 continue;
1628 }
1629 if ((!se->ops || !se->ops->save_state) && !se->vmsd) {
1630 continue;
1631 }
1632 if (se->vmsd && !vmstate_save_needed(se->vmsd, se->opaque)) {
1633 continue;
1634 }
1635
1636 save_section_header(f, se, QEMU_VM_SECTION_FULL);
1637
1638 ret = vmstate_save(f, se, NULL);
1639 if (ret) {
1640 return ret;
1641 }
1642
1643 save_section_footer(f, se);
1644 }
1645
1646 qemu_put_byte(f, QEMU_VM_EOF);
1647
1648 return qemu_file_get_error(f);
1649 }
1650
1651 static SaveStateEntry *find_se(const char *idstr, uint32_t instance_id)
1652 {
1653 SaveStateEntry *se;
1654
1655 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1656 if (!strcmp(se->idstr, idstr) &&
1657 (instance_id == se->instance_id ||
1658 instance_id == se->alias_id))
1659 return se;
1660 /* Migrating from an older version? */
1661 if (strstr(se->idstr, idstr) && se->compat) {
1662 if (!strcmp(se->compat->idstr, idstr) &&
1663 (instance_id == se->compat->instance_id ||
1664 instance_id == se->alias_id))
1665 return se;
1666 }
1667 }
1668 return NULL;
1669 }
1670
1671 enum LoadVMExitCodes {
1672 /* Allow a command to quit all layers of nested loadvm loops */
1673 LOADVM_QUIT = 1,
1674 };
1675
1676 /* ------ incoming postcopy messages ------ */
1677 /* 'advise' arrives before any transfers just to tell us that a postcopy
1678 * *might* happen - it might be skipped if precopy transferred everything
1679 * quickly.
1680 */
1681 static int loadvm_postcopy_handle_advise(MigrationIncomingState *mis,
1682 uint16_t len)
1683 {
1684 PostcopyState ps = postcopy_state_set(POSTCOPY_INCOMING_ADVISE);
1685 uint64_t remote_pagesize_summary, local_pagesize_summary, remote_tps;
1686 Error *local_err = NULL;
1687
1688 trace_loadvm_postcopy_handle_advise();
1689 if (ps != POSTCOPY_INCOMING_NONE) {
1690 error_report("CMD_POSTCOPY_ADVISE in wrong postcopy state (%d)", ps);
1691 return -1;
1692 }
1693
1694 switch (len) {
1695 case 0:
1696 if (migrate_postcopy_ram()) {
1697 error_report("RAM postcopy is enabled but have 0 byte advise");
1698 return -EINVAL;
1699 }
1700 return 0;
1701 case 8 + 8:
1702 if (!migrate_postcopy_ram()) {
1703 error_report("RAM postcopy is disabled but have 16 byte advise");
1704 return -EINVAL;
1705 }
1706 break;
1707 default:
1708 error_report("CMD_POSTCOPY_ADVISE invalid length (%d)", len);
1709 return -EINVAL;
1710 }
1711
1712 if (!postcopy_ram_supported_by_host(mis)) {
1713 postcopy_state_set(POSTCOPY_INCOMING_NONE);
1714 return -1;
1715 }
1716
1717 remote_pagesize_summary = qemu_get_be64(mis->from_src_file);
1718 local_pagesize_summary = ram_pagesize_summary();
1719
1720 if (remote_pagesize_summary != local_pagesize_summary) {
1721 /*
1722 * This detects two potential causes of mismatch:
1723 * a) A mismatch in host page sizes
1724 * Some combinations of mismatch are probably possible but it gets
1725 * a bit more complicated. In particular we need to place whole
1726 * host pages on the dest at once, and we need to ensure that we
1727 * handle dirtying to make sure we never end up sending part of
1728 * a hostpage on it's own.
1729 * b) The use of different huge page sizes on source/destination
1730 * a more fine grain test is performed during RAM block migration
1731 * but this test here causes a nice early clear failure, and
1732 * also fails when passed to an older qemu that doesn't
1733 * do huge pages.
1734 */
1735 error_report("Postcopy needs matching RAM page sizes (s=%" PRIx64
1736 " d=%" PRIx64 ")",
1737 remote_pagesize_summary, local_pagesize_summary);
1738 return -1;
1739 }
1740
1741 remote_tps = qemu_get_be64(mis->from_src_file);
1742 if (remote_tps != qemu_target_page_size()) {
1743 /*
1744 * Again, some differences could be dealt with, but for now keep it
1745 * simple.
1746 */
1747 error_report("Postcopy needs matching target page sizes (s=%d d=%zd)",
1748 (int)remote_tps, qemu_target_page_size());
1749 return -1;
1750 }
1751
1752 if (postcopy_notify(POSTCOPY_NOTIFY_INBOUND_ADVISE, &local_err)) {
1753 error_report_err(local_err);
1754 return -1;
1755 }
1756
1757 if (ram_postcopy_incoming_init(mis)) {
1758 return -1;
1759 }
1760
1761 return 0;
1762 }
1763
1764 /* After postcopy we will be told to throw some pages away since they're
1765 * dirty and will have to be demand fetched. Must happen before CPU is
1766 * started.
1767 * There can be 0..many of these messages, each encoding multiple pages.
1768 */
1769 static int loadvm_postcopy_ram_handle_discard(MigrationIncomingState *mis,
1770 uint16_t len)
1771 {
1772 int tmp;
1773 char ramid[256];
1774 PostcopyState ps = postcopy_state_get();
1775
1776 trace_loadvm_postcopy_ram_handle_discard();
1777
1778 switch (ps) {
1779 case POSTCOPY_INCOMING_ADVISE:
1780 /* 1st discard */
1781 tmp = postcopy_ram_prepare_discard(mis);
1782 if (tmp) {
1783 return tmp;
1784 }
1785 break;
1786
1787 case POSTCOPY_INCOMING_DISCARD:
1788 /* Expected state */
1789 break;
1790
1791 default:
1792 error_report("CMD_POSTCOPY_RAM_DISCARD in wrong postcopy state (%d)",
1793 ps);
1794 return -1;
1795 }
1796 /* We're expecting a
1797 * Version (0)
1798 * a RAM ID string (length byte, name, 0 term)
1799 * then at least 1 16 byte chunk
1800 */
1801 if (len < (1 + 1 + 1 + 1 + 2 * 8)) {
1802 error_report("CMD_POSTCOPY_RAM_DISCARD invalid length (%d)", len);
1803 return -1;
1804 }
1805
1806 tmp = qemu_get_byte(mis->from_src_file);
1807 if (tmp != postcopy_ram_discard_version) {
1808 error_report("CMD_POSTCOPY_RAM_DISCARD invalid version (%d)", tmp);
1809 return -1;
1810 }
1811
1812 if (!qemu_get_counted_string(mis->from_src_file, ramid)) {
1813 error_report("CMD_POSTCOPY_RAM_DISCARD Failed to read RAMBlock ID");
1814 return -1;
1815 }
1816 tmp = qemu_get_byte(mis->from_src_file);
1817 if (tmp != 0) {
1818 error_report("CMD_POSTCOPY_RAM_DISCARD missing nil (%d)", tmp);
1819 return -1;
1820 }
1821
1822 len -= 3 + strlen(ramid);
1823 if (len % 16) {
1824 error_report("CMD_POSTCOPY_RAM_DISCARD invalid length (%d)", len);
1825 return -1;
1826 }
1827 trace_loadvm_postcopy_ram_handle_discard_header(ramid, len);
1828 while (len) {
1829 uint64_t start_addr, block_length;
1830 start_addr = qemu_get_be64(mis->from_src_file);
1831 block_length = qemu_get_be64(mis->from_src_file);
1832
1833 len -= 16;
1834 int ret = ram_discard_range(ramid, start_addr, block_length);
1835 if (ret) {
1836 return ret;
1837 }
1838 }
1839 trace_loadvm_postcopy_ram_handle_discard_end();
1840
1841 return 0;
1842 }
1843
1844 /*
1845 * Triggered by a postcopy_listen command; this thread takes over reading
1846 * the input stream, leaving the main thread free to carry on loading the rest
1847 * of the device state (from RAM).
1848 * (TODO:This could do with being in a postcopy file - but there again it's
1849 * just another input loop, not that postcopy specific)
1850 */
1851 static void *postcopy_ram_listen_thread(void *opaque)
1852 {
1853 MigrationIncomingState *mis = migration_incoming_get_current();
1854 QEMUFile *f = mis->from_src_file;
1855 int load_res;
1856 MigrationState *migr = migrate_get_current();
1857
1858 object_ref(OBJECT(migr));
1859
1860 migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE,
1861 MIGRATION_STATUS_POSTCOPY_ACTIVE);
1862 qemu_sem_post(&mis->listen_thread_sem);
1863 trace_postcopy_ram_listen_thread_start();
1864
1865 rcu_register_thread();
1866 /*
1867 * Because we're a thread and not a coroutine we can't yield
1868 * in qemu_file, and thus we must be blocking now.
1869 */
1870 qemu_file_set_blocking(f, true);
1871 load_res = qemu_loadvm_state_main(f, mis);
1872
1873 /*
1874 * This is tricky, but, mis->from_src_file can change after it
1875 * returns, when postcopy recovery happened. In the future, we may
1876 * want a wrapper for the QEMUFile handle.
1877 */
1878 f = mis->from_src_file;
1879
1880 /* And non-blocking again so we don't block in any cleanup */
1881 qemu_file_set_blocking(f, false);
1882
1883 trace_postcopy_ram_listen_thread_exit();
1884 if (load_res < 0) {
1885 qemu_file_set_error(f, load_res);
1886 dirty_bitmap_mig_cancel_incoming();
1887 if (postcopy_state_get() == POSTCOPY_INCOMING_RUNNING &&
1888 !migrate_postcopy_ram() && migrate_dirty_bitmaps())
1889 {
1890 error_report("%s: loadvm failed during postcopy: %d. All states "
1891 "are migrated except dirty bitmaps. Some dirty "
1892 "bitmaps may be lost, and present migrated dirty "
1893 "bitmaps are correctly migrated and valid.",
1894 __func__, load_res);
1895 load_res = 0; /* prevent further exit() */
1896 } else {
1897 error_report("%s: loadvm failed: %d", __func__, load_res);
1898 migrate_set_state(&mis->state, MIGRATION_STATUS_POSTCOPY_ACTIVE,
1899 MIGRATION_STATUS_FAILED);
1900 }
1901 }
1902 if (load_res >= 0) {
1903 /*
1904 * This looks good, but it's possible that the device loading in the
1905 * main thread hasn't finished yet, and so we might not be in 'RUN'
1906 * state yet; wait for the end of the main thread.
1907 */
1908 qemu_event_wait(&mis->main_thread_load_event);
1909 }
1910 postcopy_ram_incoming_cleanup(mis);
1911
1912 if (load_res < 0) {
1913 /*
1914 * If something went wrong then we have a bad state so exit;
1915 * depending how far we got it might be possible at this point
1916 * to leave the guest running and fire MCEs for pages that never
1917 * arrived as a desperate recovery step.
1918 */
1919 rcu_unregister_thread();
1920 exit(EXIT_FAILURE);
1921 }
1922
1923 migrate_set_state(&mis->state, MIGRATION_STATUS_POSTCOPY_ACTIVE,
1924 MIGRATION_STATUS_COMPLETED);
1925 /*
1926 * If everything has worked fine, then the main thread has waited
1927 * for us to start, and we're the last use of the mis.
1928 * (If something broke then qemu will have to exit anyway since it's
1929 * got a bad migration state).
1930 */
1931 migration_incoming_state_destroy();
1932 qemu_loadvm_state_cleanup();
1933
1934 rcu_unregister_thread();
1935 mis->have_listen_thread = false;
1936 postcopy_state_set(POSTCOPY_INCOMING_END);
1937
1938 object_unref(OBJECT(migr));
1939
1940 return NULL;
1941 }
1942
1943 /* After this message we must be able to immediately receive postcopy data */
1944 static int loadvm_postcopy_handle_listen(MigrationIncomingState *mis)
1945 {
1946 PostcopyState ps = postcopy_state_set(POSTCOPY_INCOMING_LISTENING);
1947 trace_loadvm_postcopy_handle_listen();
1948 Error *local_err = NULL;
1949
1950 if (ps != POSTCOPY_INCOMING_ADVISE && ps != POSTCOPY_INCOMING_DISCARD) {
1951 error_report("CMD_POSTCOPY_LISTEN in wrong postcopy state (%d)", ps);
1952 return -1;
1953 }
1954 if (ps == POSTCOPY_INCOMING_ADVISE) {
1955 /*
1956 * A rare case, we entered listen without having to do any discards,
1957 * so do the setup that's normally done at the time of the 1st discard.
1958 */
1959 if (migrate_postcopy_ram()) {
1960 postcopy_ram_prepare_discard(mis);
1961 }
1962 }
1963
1964 /*
1965 * Sensitise RAM - can now generate requests for blocks that don't exist
1966 * However, at this point the CPU shouldn't be running, and the IO
1967 * shouldn't be doing anything yet so don't actually expect requests
1968 */
1969 if (migrate_postcopy_ram()) {
1970 if (postcopy_ram_incoming_setup(mis)) {
1971 postcopy_ram_incoming_cleanup(mis);
1972 return -1;
1973 }
1974 }
1975
1976 if (postcopy_notify(POSTCOPY_NOTIFY_INBOUND_LISTEN, &local_err)) {
1977 error_report_err(local_err);
1978 return -1;
1979 }
1980
1981 mis->have_listen_thread = true;
1982 /* Start up the listening thread and wait for it to signal ready */
1983 qemu_sem_init(&mis->listen_thread_sem, 0);
1984 qemu_thread_create(&mis->listen_thread, "postcopy/listen",
1985 postcopy_ram_listen_thread, NULL,
1986 QEMU_THREAD_DETACHED);
1987 qemu_sem_wait(&mis->listen_thread_sem);
1988 qemu_sem_destroy(&mis->listen_thread_sem);
1989
1990 return 0;
1991 }
1992
1993 static void loadvm_postcopy_handle_run_bh(void *opaque)
1994 {
1995 Error *local_err = NULL;
1996 MigrationIncomingState *mis = opaque;
1997
1998 /* TODO we should move all of this lot into postcopy_ram.c or a shared code
1999 * in migration.c
2000 */
2001 cpu_synchronize_all_post_init();
2002
2003 qemu_announce_self(&mis->announce_timer, migrate_announce_params());
2004
2005 /* Make sure all file formats flush their mutable metadata.
2006 * If we get an error here, just don't restart the VM yet. */
2007 bdrv_invalidate_cache_all(&local_err);
2008 if (local_err) {
2009 error_report_err(local_err);
2010 local_err = NULL;
2011 autostart = false;
2012 }
2013
2014 trace_loadvm_postcopy_handle_run_cpu_sync();
2015
2016 trace_loadvm_postcopy_handle_run_vmstart();
2017
2018 dirty_bitmap_mig_before_vm_start();
2019
2020 if (autostart) {
2021 /* Hold onto your hats, starting the CPU */
2022 vm_start();
2023 } else {
2024 /* leave it paused and let management decide when to start the CPU */
2025 runstate_set(RUN_STATE_PAUSED);
2026 }
2027
2028 qemu_bh_delete(mis->bh);
2029 }
2030
2031 /* After all discards we can start running and asking for pages */
2032 static int loadvm_postcopy_handle_run(MigrationIncomingState *mis)
2033 {
2034 PostcopyState ps = postcopy_state_get();
2035
2036 trace_loadvm_postcopy_handle_run();
2037 if (ps != POSTCOPY_INCOMING_LISTENING) {
2038 error_report("CMD_POSTCOPY_RUN in wrong postcopy state (%d)", ps);
2039 return -1;
2040 }
2041
2042 postcopy_state_set(POSTCOPY_INCOMING_RUNNING);
2043 mis->bh = qemu_bh_new(loadvm_postcopy_handle_run_bh, mis);
2044 qemu_bh_schedule(mis->bh);
2045
2046 /* We need to finish reading the stream from the package
2047 * and also stop reading anything more from the stream that loaded the
2048 * package (since it's now being read by the listener thread).
2049 * LOADVM_QUIT will quit all the layers of nested loadvm loops.
2050 */
2051 return LOADVM_QUIT;
2052 }
2053
2054 /* We must be with page_request_mutex held */
2055 static gboolean postcopy_sync_page_req(gpointer key, gpointer value,
2056 gpointer data)
2057 {
2058 MigrationIncomingState *mis = data;
2059 void *host_addr = (void *) key;
2060 ram_addr_t rb_offset;
2061 RAMBlock *rb;
2062 int ret;
2063
2064 rb = qemu_ram_block_from_host(host_addr, true, &rb_offset);
2065 if (!rb) {
2066 /*
2067 * This should _never_ happen. However be nice for a migrating VM to
2068 * not crash/assert. Post an error (note: intended to not use *_once
2069 * because we do want to see all the illegal addresses; and this can
2070 * never be triggered by the guest so we're safe) and move on next.
2071 */
2072 error_report("%s: illegal host addr %p", __func__, host_addr);
2073 /* Try the next entry */
2074 return FALSE;
2075 }
2076
2077 ret = migrate_send_rp_message_req_pages(mis, rb, rb_offset);
2078 if (ret) {
2079 /* Please refer to above comment. */
2080 error_report("%s: send rp message failed for addr %p",
2081 __func__, host_addr);
2082 return FALSE;
2083 }
2084
2085 trace_postcopy_page_req_sync(host_addr);
2086
2087 return FALSE;
2088 }
2089
2090 static void migrate_send_rp_req_pages_pending(MigrationIncomingState *mis)
2091 {
2092 WITH_QEMU_LOCK_GUARD(&mis->page_request_mutex) {
2093 g_tree_foreach(mis->page_requested, postcopy_sync_page_req, mis);
2094 }
2095 }
2096
2097 static int loadvm_postcopy_handle_resume(MigrationIncomingState *mis)
2098 {
2099 if (mis->state != MIGRATION_STATUS_POSTCOPY_RECOVER) {
2100 error_report("%s: illegal resume received", __func__);
2101 /* Don't fail the load, only for this. */
2102 return 0;
2103 }
2104
2105 /*
2106 * Reset the last_rb before we resend any page req to source again, since
2107 * the source should have it reset already.
2108 */
2109 mis->last_rb = NULL;
2110
2111 /*
2112 * This means source VM is ready to resume the postcopy migration.
2113 */
2114 migrate_set_state(&mis->state, MIGRATION_STATUS_POSTCOPY_RECOVER,
2115 MIGRATION_STATUS_POSTCOPY_ACTIVE);
2116
2117 trace_loadvm_postcopy_handle_resume();
2118
2119 /* Tell source that "we are ready" */
2120 migrate_send_rp_resume_ack(mis, MIGRATION_RESUME_ACK_VALUE);
2121
2122 /*
2123 * After a postcopy recovery, the source should have lost the postcopy
2124 * queue, or potentially the requested pages could have been lost during
2125 * the network down phase. Let's re-sync with the source VM by re-sending
2126 * all the pending pages that we eagerly need, so these threads won't get
2127 * blocked too long due to the recovery.
2128 *
2129 * Without this procedure, the faulted destination VM threads (waiting for
2130 * page requests right before the postcopy is interrupted) can keep hanging
2131 * until the pages are sent by the source during the background copying of
2132 * pages, or another thread faulted on the same address accidentally.
2133 */
2134 migrate_send_rp_req_pages_pending(mis);
2135
2136 /*
2137 * It's time to switch state and release the fault thread to continue
2138 * service page faults. Note that this should be explicitly after the
2139 * above call to migrate_send_rp_req_pages_pending(). In short:
2140 * migrate_send_rp_message_req_pages() is not thread safe, yet.
2141 */
2142 qemu_sem_post(&mis->postcopy_pause_sem_fault);
2143
2144 return 0;
2145 }
2146
2147 /**
2148 * Immediately following this command is a blob of data containing an embedded
2149 * chunk of migration stream; read it and load it.
2150 *
2151 * @mis: Incoming state
2152 * @length: Length of packaged data to read
2153 *
2154 * Returns: Negative values on error
2155 *
2156 */
2157 static int loadvm_handle_cmd_packaged(MigrationIncomingState *mis)
2158 {
2159 int ret;
2160 size_t length;
2161 QIOChannelBuffer *bioc;
2162
2163 length = qemu_get_be32(mis->from_src_file);
2164 trace_loadvm_handle_cmd_packaged(length);
2165
2166 if (length > MAX_VM_CMD_PACKAGED_SIZE) {
2167 error_report("Unreasonably large packaged state: %zu", length);
2168 return -1;
2169 }
2170
2171 bioc = qio_channel_buffer_new(length);
2172 qio_channel_set_name(QIO_CHANNEL(bioc), "migration-loadvm-buffer");
2173 ret = qemu_get_buffer(mis->from_src_file,
2174 bioc->data,
2175 length);
2176 if (ret != length) {
2177 object_unref(OBJECT(bioc));
2178 error_report("CMD_PACKAGED: Buffer receive fail ret=%d length=%zu",
2179 ret, length);
2180 return (ret < 0) ? ret : -EAGAIN;
2181 }
2182 bioc->usage += length;
2183 trace_loadvm_handle_cmd_packaged_received(ret);
2184
2185 QEMUFile *packf = qemu_fopen_channel_input(QIO_CHANNEL(bioc));
2186
2187 ret = qemu_loadvm_state_main(packf, mis);
2188 trace_loadvm_handle_cmd_packaged_main(ret);
2189 qemu_fclose(packf);
2190 object_unref(OBJECT(bioc));
2191
2192 return ret;
2193 }
2194
2195 /*
2196 * Handle request that source requests for recved_bitmap on
2197 * destination. Payload format:
2198 *
2199 * len (1 byte) + ramblock_name (<255 bytes)
2200 */
2201 static int loadvm_handle_recv_bitmap(MigrationIncomingState *mis,
2202 uint16_t len)
2203 {
2204 QEMUFile *file = mis->from_src_file;
2205 RAMBlock *rb;
2206 char block_name[256];
2207 size_t cnt;
2208
2209 cnt = qemu_get_counted_string(file, block_name);
2210 if (!cnt) {
2211 error_report("%s: failed to read block name", __func__);
2212 return -EINVAL;
2213 }
2214
2215 /* Validate before using the data */
2216 if (qemu_file_get_error(file)) {
2217 return qemu_file_get_error(file);
2218 }
2219
2220 if (len != cnt + 1) {
2221 error_report("%s: invalid payload length (%d)", __func__, len);
2222 return -EINVAL;
2223 }
2224
2225 rb = qemu_ram_block_by_name(block_name);
2226 if (!rb) {
2227 error_report("%s: block '%s' not found", __func__, block_name);
2228 return -EINVAL;
2229 }
2230
2231 migrate_send_rp_recv_bitmap(mis, block_name);
2232
2233 trace_loadvm_handle_recv_bitmap(block_name);
2234
2235 return 0;
2236 }
2237
2238 static int loadvm_process_enable_colo(MigrationIncomingState *mis)
2239 {
2240 int ret = migration_incoming_enable_colo();
2241
2242 if (!ret) {
2243 ret = colo_init_ram_cache();
2244 if (ret) {
2245 migration_incoming_disable_colo();
2246 }
2247 }
2248 return ret;
2249 }
2250
2251 /*
2252 * Process an incoming 'QEMU_VM_COMMAND'
2253 * 0 just a normal return
2254 * LOADVM_QUIT All good, but exit the loop
2255 * <0 Error
2256 */
2257 static int loadvm_process_command(QEMUFile *f)
2258 {
2259 MigrationIncomingState *mis = migration_incoming_get_current();
2260 uint16_t cmd;
2261 uint16_t len;
2262 uint32_t tmp32;
2263
2264 cmd = qemu_get_be16(f);
2265 len = qemu_get_be16(f);
2266
2267 /* Check validity before continue processing of cmds */
2268 if (qemu_file_get_error(f)) {
2269 return qemu_file_get_error(f);
2270 }
2271
2272 trace_loadvm_process_command(cmd, len);
2273 if (cmd >= MIG_CMD_MAX || cmd == MIG_CMD_INVALID) {
2274 error_report("MIG_CMD 0x%x unknown (len 0x%x)", cmd, len);
2275 return -EINVAL;
2276 }
2277
2278 if (mig_cmd_args[cmd].len != -1 && mig_cmd_args[cmd].len != len) {
2279 error_report("%s received with bad length - expecting %zu, got %d",
2280 mig_cmd_args[cmd].name,
2281 (size_t)mig_cmd_args[cmd].len, len);
2282 return -ERANGE;
2283 }
2284
2285 switch (cmd) {
2286 case MIG_CMD_OPEN_RETURN_PATH:
2287 if (mis->to_src_file) {
2288 error_report("CMD_OPEN_RETURN_PATH called when RP already open");
2289 /* Not really a problem, so don't give up */
2290 return 0;
2291 }
2292 mis->to_src_file = qemu_file_get_return_path(f);
2293 if (!mis->to_src_file) {
2294 error_report("CMD_OPEN_RETURN_PATH failed");
2295 return -1;
2296 }
2297 break;
2298
2299 case MIG_CMD_PING:
2300 tmp32 = qemu_get_be32(f);
2301 trace_loadvm_process_command_ping(tmp32);
2302 if (!mis->to_src_file) {
2303 error_report("CMD_PING (0x%x) received with no return path",
2304 tmp32);
2305 return -1;
2306 }
2307 migrate_send_rp_pong(mis, tmp32);
2308 break;
2309
2310 case MIG_CMD_PACKAGED:
2311 return loadvm_handle_cmd_packaged(mis);
2312
2313 case MIG_CMD_POSTCOPY_ADVISE:
2314 return loadvm_postcopy_handle_advise(mis, len);
2315
2316 case MIG_CMD_POSTCOPY_LISTEN:
2317 return loadvm_postcopy_handle_listen(mis);
2318
2319 case MIG_CMD_POSTCOPY_RUN:
2320 return loadvm_postcopy_handle_run(mis);
2321
2322 case MIG_CMD_POSTCOPY_RAM_DISCARD:
2323 return loadvm_postcopy_ram_handle_discard(mis, len);
2324
2325 case MIG_CMD_POSTCOPY_RESUME:
2326 return loadvm_postcopy_handle_resume(mis);
2327
2328 case MIG_CMD_RECV_BITMAP:
2329 return loadvm_handle_recv_bitmap(mis, len);
2330
2331 case MIG_CMD_ENABLE_COLO:
2332 return loadvm_process_enable_colo(mis);
2333 }
2334
2335 return 0;
2336 }
2337
2338 /*
2339 * Read a footer off the wire and check that it matches the expected section
2340 *
2341 * Returns: true if the footer was good
2342 * false if there is a problem (and calls error_report to say why)
2343 */
2344 static bool check_section_footer(QEMUFile *f, SaveStateEntry *se)
2345 {
2346 int ret;
2347 uint8_t read_mark;
2348 uint32_t read_section_id;
2349
2350 if (!migrate_get_current()->send_section_footer) {
2351 /* No footer to check */
2352 return true;
2353 }
2354
2355 read_mark = qemu_get_byte(f);
2356
2357 ret = qemu_file_get_error(f);
2358 if (ret) {
2359 error_report("%s: Read section footer failed: %d",
2360 __func__, ret);
2361 return false;
2362 }
2363
2364 if (read_mark != QEMU_VM_SECTION_FOOTER) {
2365 error_report("Missing section footer for %s", se->idstr);
2366 return false;
2367 }
2368
2369 read_section_id = qemu_get_be32(f);
2370 if (read_section_id != se->load_section_id) {
2371 error_report("Mismatched section id in footer for %s -"
2372 " read 0x%x expected 0x%x",
2373 se->idstr, read_section_id, se->load_section_id);
2374 return false;
2375 }
2376
2377 /* All good */
2378 return true;
2379 }
2380
2381 static int
2382 qemu_loadvm_section_start_full(QEMUFile *f, MigrationIncomingState *mis)
2383 {
2384 uint32_t instance_id, version_id, section_id;
2385 SaveStateEntry *se;
2386 char idstr[256];
2387 int ret;
2388
2389 /* Read section start */
2390 section_id = qemu_get_be32(f);
2391 if (!qemu_get_counted_string(f, idstr)) {
2392 error_report("Unable to read ID string for section %u",
2393 section_id);
2394 return -EINVAL;
2395 }
2396 instance_id = qemu_get_be32(f);
2397 version_id = qemu_get_be32(f);
2398
2399 ret = qemu_file_get_error(f);
2400 if (ret) {
2401 error_report("%s: Failed to read instance/version ID: %d",
2402 __func__, ret);
2403 return ret;
2404 }
2405
2406 trace_qemu_loadvm_state_section_startfull(section_id, idstr,
2407 instance_id, version_id);
2408 /* Find savevm section */
2409 se = find_se(idstr, instance_id);
2410 if (se == NULL) {
2411 error_report("Unknown savevm section or instance '%s' %"PRIu32". "
2412 "Make sure that your current VM setup matches your "
2413 "saved VM setup, including any hotplugged devices",
2414 idstr, instance_id);
2415 return -EINVAL;
2416 }
2417
2418 /* Validate version */
2419 if (version_id > se->version_id) {
2420 error_report("savevm: unsupported version %d for '%s' v%d",
2421 version_id, idstr, se->version_id);
2422 return -EINVAL;
2423 }
2424 se->load_version_id = version_id;
2425 se->load_section_id = section_id;
2426
2427 /* Validate if it is a device's state */
2428 if (xen_enabled() && se->is_ram) {
2429 error_report("loadvm: %s RAM loading not allowed on Xen", idstr);
2430 return -EINVAL;
2431 }
2432
2433 ret = vmstate_load(f, se);
2434 if (ret < 0) {
2435 error_report("error while loading state for instance 0x%"PRIx32" of"
2436 " device '%s'", instance_id, idstr);
2437 return ret;
2438 }
2439 if (!check_section_footer(f, se)) {
2440 return -EINVAL;
2441 }
2442
2443 return 0;
2444 }
2445
2446 static int
2447 qemu_loadvm_section_part_end(QEMUFile *f, MigrationIncomingState *mis)
2448 {
2449 uint32_t section_id;
2450 SaveStateEntry *se;
2451 int ret;
2452
2453 section_id = qemu_get_be32(f);
2454
2455 ret = qemu_file_get_error(f);
2456 if (ret) {
2457 error_report("%s: Failed to read section ID: %d",
2458 __func__, ret);
2459 return ret;
2460 }
2461
2462 trace_qemu_loadvm_state_section_partend(section_id);
2463 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
2464 if (se->load_section_id == section_id) {
2465 break;
2466 }
2467 }
2468 if (se == NULL) {
2469 error_report("Unknown savevm section %d", section_id);
2470 return -EINVAL;
2471 }
2472
2473 ret = vmstate_load(f, se);
2474 if (ret < 0) {
2475 error_report("error while loading state section id %d(%s)",
2476 section_id, se->idstr);
2477 return ret;
2478 }
2479 if (!check_section_footer(f, se)) {
2480 return -EINVAL;
2481 }
2482
2483 return 0;
2484 }
2485
2486 static int qemu_loadvm_state_header(QEMUFile *f)
2487 {
2488 unsigned int v;
2489 int ret;
2490
2491 v = qemu_get_be32(f);
2492 if (v != QEMU_VM_FILE_MAGIC) {
2493 error_report("Not a migration stream");
2494 return -EINVAL;
2495 }
2496
2497 v = qemu_get_be32(f);
2498 if (v == QEMU_VM_FILE_VERSION_COMPAT) {
2499 error_report("SaveVM v2 format is obsolete and don't work anymore");
2500 return -ENOTSUP;
2501 }
2502 if (v != QEMU_VM_FILE_VERSION) {
2503 error_report("Unsupported migration stream version");
2504 return -ENOTSUP;
2505 }
2506
2507 if (migrate_get_current()->send_configuration) {
2508 if (qemu_get_byte(f) != QEMU_VM_CONFIGURATION) {
2509 error_report("Configuration section missing");
2510 qemu_loadvm_state_cleanup();
2511 return -EINVAL;
2512 }
2513 ret = vmstate_load_state(f, &vmstate_configuration, &savevm_state, 0);
2514
2515 if (ret) {
2516 qemu_loadvm_state_cleanup();
2517 return ret;
2518 }
2519 }
2520 return 0;
2521 }
2522
2523 static int qemu_loadvm_state_setup(QEMUFile *f)
2524 {
2525 SaveStateEntry *se;
2526 int ret;
2527
2528 trace_loadvm_state_setup();
2529 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
2530 if (!se->ops || !se->ops->load_setup) {
2531 continue;
2532 }
2533 if (se->ops->is_active) {
2534 if (!se->ops->is_active(se->opaque)) {
2535 continue;
2536 }
2537 }
2538
2539 ret = se->ops->load_setup(f, se->opaque);
2540 if (ret < 0) {
2541 qemu_file_set_error(f, ret);
2542 error_report("Load state of device %s failed", se->idstr);
2543 return ret;
2544 }
2545 }
2546 return 0;
2547 }
2548
2549 void qemu_loadvm_state_cleanup(void)
2550 {
2551 SaveStateEntry *se;
2552
2553 trace_loadvm_state_cleanup();
2554 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
2555 if (se->ops && se->ops->load_cleanup) {
2556 se->ops->load_cleanup(se->opaque);
2557 }
2558 }
2559 }
2560
2561 /* Return true if we should continue the migration, or false. */
2562 static bool postcopy_pause_incoming(MigrationIncomingState *mis)
2563 {
2564 trace_postcopy_pause_incoming();
2565
2566 assert(migrate_postcopy_ram());
2567
2568 /* Clear the triggered bit to allow one recovery */
2569 mis->postcopy_recover_triggered = false;
2570
2571 assert(mis->from_src_file);
2572 qemu_file_shutdown(mis->from_src_file);
2573 qemu_fclose(mis->from_src_file);
2574 mis->from_src_file = NULL;
2575
2576 assert(mis->to_src_file);
2577 qemu_file_shutdown(mis->to_src_file);
2578 qemu_mutex_lock(&mis->rp_mutex);
2579 qemu_fclose(mis->to_src_file);
2580 mis->to_src_file = NULL;
2581 qemu_mutex_unlock(&mis->rp_mutex);
2582
2583 migrate_set_state(&mis->state, MIGRATION_STATUS_POSTCOPY_ACTIVE,
2584 MIGRATION_STATUS_POSTCOPY_PAUSED);
2585
2586 /* Notify the fault thread for the invalidated file handle */
2587 postcopy_fault_thread_notify(mis);
2588
2589 error_report("Detected IO failure for postcopy. "
2590 "Migration paused.");
2591
2592 while (mis->state == MIGRATION_STATUS_POSTCOPY_PAUSED) {
2593 qemu_sem_wait(&mis->postcopy_pause_sem_dst);
2594 }
2595
2596 trace_postcopy_pause_incoming_continued();
2597
2598 return true;
2599 }
2600
2601 int qemu_loadvm_state_main(QEMUFile *f, MigrationIncomingState *mis)
2602 {
2603 uint8_t section_type;
2604 int ret = 0;
2605
2606 retry:
2607 while (true) {
2608 section_type = qemu_get_byte(f);
2609
2610 if (qemu_file_get_error(f)) {
2611 ret = qemu_file_get_error(f);
2612 break;
2613 }
2614
2615 trace_qemu_loadvm_state_section(section_type);
2616 switch (section_type) {
2617 case QEMU_VM_SECTION_START:
2618 case QEMU_VM_SECTION_FULL:
2619 ret = qemu_loadvm_section_start_full(f, mis);
2620 if (ret < 0) {
2621 goto out;
2622 }
2623 break;
2624 case QEMU_VM_SECTION_PART:
2625 case QEMU_VM_SECTION_END:
2626 ret = qemu_loadvm_section_part_end(f, mis);
2627 if (ret < 0) {
2628 goto out;
2629 }
2630 break;
2631 case QEMU_VM_COMMAND:
2632 ret = loadvm_process_command(f);
2633 trace_qemu_loadvm_state_section_command(ret);
2634 if ((ret < 0) || (ret == LOADVM_QUIT)) {
2635 goto out;
2636 }
2637 break;
2638 case QEMU_VM_EOF:
2639 /* This is the end of migration */
2640 goto out;
2641 default:
2642 error_report("Unknown savevm section type %d", section_type);
2643 ret = -EINVAL;
2644 goto out;
2645 }
2646 }
2647
2648 out:
2649 if (ret < 0) {
2650 qemu_file_set_error(f, ret);
2651
2652 /* Cancel bitmaps incoming regardless of recovery */
2653 dirty_bitmap_mig_cancel_incoming();
2654
2655 /*
2656 * If we are during an active postcopy, then we pause instead
2657 * of bail out to at least keep the VM's dirty data. Note
2658 * that POSTCOPY_INCOMING_LISTENING stage is still not enough,
2659 * during which we're still receiving device states and we
2660 * still haven't yet started the VM on destination.
2661 *
2662 * Only RAM postcopy supports recovery. Still, if RAM postcopy is
2663 * enabled, canceled bitmaps postcopy will not affect RAM postcopy
2664 * recovering.
2665 */
2666 if (postcopy_state_get() == POSTCOPY_INCOMING_RUNNING &&
2667 migrate_postcopy_ram() && postcopy_pause_incoming(mis)) {
2668 /* Reset f to point to the newly created channel */
2669 f = mis->from_src_file;
2670 goto retry;
2671 }
2672 }
2673 return ret;
2674 }
2675
2676 int qemu_loadvm_state(QEMUFile *f)
2677 {
2678 MigrationIncomingState *mis = migration_incoming_get_current();
2679 Error *local_err = NULL;
2680 int ret;
2681
2682 if (qemu_savevm_state_blocked(&local_err)) {
2683 error_report_err(local_err);
2684 return -EINVAL;
2685 }
2686
2687 ret = qemu_loadvm_state_header(f);
2688 if (ret) {
2689 return ret;
2690 }
2691
2692 if (qemu_loadvm_state_setup(f) != 0) {
2693 return -EINVAL;
2694 }
2695
2696 cpu_synchronize_all_pre_loadvm();
2697
2698 ret = qemu_loadvm_state_main(f, mis);
2699 qemu_event_set(&mis->main_thread_load_event);
2700
2701 trace_qemu_loadvm_state_post_main(ret);
2702
2703 if (mis->have_listen_thread) {
2704 /* Listen thread still going, can't clean up yet */
2705 return ret;
2706 }
2707
2708 if (ret == 0) {
2709 ret = qemu_file_get_error(f);
2710 }
2711
2712 /*
2713 * Try to read in the VMDESC section as well, so that dumping tools that
2714 * intercept our migration stream have the chance to see it.
2715 */
2716
2717 /* We've got to be careful; if we don't read the data and just shut the fd
2718 * then the sender can error if we close while it's still sending.
2719 * We also mustn't read data that isn't there; some transports (RDMA)
2720 * will stall waiting for that data when the source has already closed.
2721 */
2722 if (ret == 0 && should_send_vmdesc()) {
2723 uint8_t *buf;
2724 uint32_t size;
2725 uint8_t section_type = qemu_get_byte(f);
2726
2727 if (section_type != QEMU_VM_VMDESCRIPTION) {
2728 error_report("Expected vmdescription section, but got %d",
2729 section_type);
2730 /*
2731 * It doesn't seem worth failing at this point since
2732 * we apparently have an otherwise valid VM state
2733 */
2734 } else {
2735 buf = g_malloc(0x1000);
2736 size = qemu_get_be32(f);
2737
2738 while (size > 0) {
2739 uint32_t read_chunk = MIN(size, 0x1000);
2740 qemu_get_buffer(f, buf, read_chunk);
2741 size -= read_chunk;
2742 }
2743 g_free(buf);
2744 }
2745 }
2746
2747 qemu_loadvm_state_cleanup();
2748 cpu_synchronize_all_post_init();
2749
2750 return ret;
2751 }
2752
2753 int qemu_load_device_state(QEMUFile *f)
2754 {
2755 MigrationIncomingState *mis = migration_incoming_get_current();
2756 int ret;
2757
2758 /* Load QEMU_VM_SECTION_FULL section */
2759 ret = qemu_loadvm_state_main(f, mis);
2760 if (ret < 0) {
2761 error_report("Failed to load device state: %d", ret);
2762 return ret;
2763 }
2764
2765 cpu_synchronize_all_post_init();
2766 return 0;
2767 }
2768
2769 bool save_snapshot(const char *name, bool overwrite, const char *vmstate,
2770 bool has_devices, strList *devices, Error **errp)
2771 {
2772 BlockDriverState *bs;
2773 QEMUSnapshotInfo sn1, *sn = &sn1;
2774 int ret = -1, ret2;
2775 QEMUFile *f;
2776 int saved_vm_running;
2777 uint64_t vm_state_size;
2778 g_autoptr(GDateTime) now = g_date_time_new_now_local();
2779 AioContext *aio_context;
2780
2781 if (migration_is_blocked(errp)) {
2782 return false;
2783 }
2784
2785 if (!replay_can_snapshot()) {
2786 error_setg(errp, "Record/replay does not allow making snapshot "
2787 "right now. Try once more later.");
2788 return false;
2789 }
2790
2791 if (!bdrv_all_can_snapshot(has_devices, devices, errp)) {
2792 return false;
2793 }
2794
2795 /* Delete old snapshots of the same name */
2796 if (name) {
2797 if (overwrite) {
2798 if (bdrv_all_delete_snapshot(name, has_devices,
2799 devices, errp) < 0) {
2800 return false;
2801 }
2802 } else {
2803 ret2 = bdrv_all_has_snapshot(name, has_devices, devices, errp);
2804 if (ret2 < 0) {
2805 return false;
2806 }
2807 if (ret2 == 1) {
2808 error_setg(errp,
2809 "Snapshot '%s' already exists in one or more devices",
2810 name);
2811 return false;
2812 }
2813 }
2814 }
2815
2816 bs = bdrv_all_find_vmstate_bs(vmstate, has_devices, devices, errp);
2817 if (bs == NULL) {
2818 return false;
2819 }
2820 aio_context = bdrv_get_aio_context(bs);
2821
2822 saved_vm_running = runstate_is_running();
2823
2824 ret = global_state_store();
2825 if (ret) {
2826 error_setg(errp, "Error saving global state");
2827 return false;
2828 }
2829 vm_stop(RUN_STATE_SAVE_VM);
2830
2831 bdrv_drain_all_begin();
2832
2833 aio_context_acquire(aio_context);
2834
2835 memset(sn, 0, sizeof(*sn));
2836
2837 /* fill auxiliary fields */
2838 sn->date_sec = g_date_time_to_unix(now);
2839 sn->date_nsec = g_date_time_get_microsecond(now) * 1000;
2840 sn->vm_clock_nsec = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL);
2841 if (replay_mode != REPLAY_MODE_NONE) {
2842 sn->icount = replay_get_current_icount();
2843 } else {
2844 sn->icount = -1ULL;
2845 }
2846
2847 if (name) {
2848 pstrcpy(sn->name, sizeof(sn->name), name);
2849 } else {
2850 g_autofree char *autoname = g_date_time_format(now, "vm-%Y%m%d%H%M%S");
2851 pstrcpy(sn->name, sizeof(sn->name), autoname);
2852 }
2853
2854 /* save the VM state */
2855 f = qemu_fopen_bdrv(bs, 1);
2856 if (!f) {
2857 error_setg(errp, "Could not open VM state file");
2858 goto the_end;
2859 }
2860 ret = qemu_savevm_state(f, errp);
2861 vm_state_size = qemu_ftell(f);
2862 ret2 = qemu_fclose(f);
2863 if (ret < 0) {
2864 goto the_end;
2865 }
2866 if (ret2 < 0) {
2867 ret = ret2;
2868 goto the_end;
2869 }
2870
2871 /* The bdrv_all_create_snapshot() call that follows acquires the AioContext
2872 * for itself. BDRV_POLL_WHILE() does not support nested locking because
2873 * it only releases the lock once. Therefore synchronous I/O will deadlock
2874 * unless we release the AioContext before bdrv_all_create_snapshot().
2875 */
2876 aio_context_release(aio_context);
2877 aio_context = NULL;
2878
2879 ret = bdrv_all_create_snapshot(sn, bs, vm_state_size,
2880 has_devices, devices, errp);
2881 if (ret < 0) {
2882 bdrv_all_delete_snapshot(sn->name, has_devices, devices, NULL);
2883 goto the_end;
2884 }
2885
2886 ret = 0;
2887
2888 the_end:
2889 if (aio_context) {
2890 aio_context_release(aio_context);
2891 }
2892
2893 bdrv_drain_all_end();
2894
2895 if (saved_vm_running) {
2896 vm_start();
2897 }
2898 return ret == 0;
2899 }
2900
2901 void qmp_xen_save_devices_state(const char *filename, bool has_live, bool live,
2902 Error **errp)
2903 {
2904 QEMUFile *f;
2905 QIOChannelFile *ioc;
2906 int saved_vm_running;
2907 int ret;
2908
2909 if (!has_live) {
2910 /* live default to true so old version of Xen tool stack can have a
2911 * successful live migration */
2912 live = true;
2913 }
2914
2915 saved_vm_running = runstate_is_running();
2916 vm_stop(RUN_STATE_SAVE_VM);
2917 global_state_store_running();
2918
2919 ioc = qio_channel_file_new_path(filename, O_WRONLY | O_CREAT | O_TRUNC,
2920 0660, errp);
2921 if (!ioc) {
2922 goto the_end;
2923 }
2924 qio_channel_set_name(QIO_CHANNEL(ioc), "migration-xen-save-state");
2925 f = qemu_fopen_channel_output(QIO_CHANNEL(ioc));
2926 object_unref(OBJECT(ioc));
2927 ret = qemu_save_device_state(f);
2928 if (ret < 0 || qemu_fclose(f) < 0) {
2929 error_setg(errp, QERR_IO_ERROR);
2930 } else {
2931 /* libxl calls the QMP command "stop" before calling
2932 * "xen-save-devices-state" and in case of migration failure, libxl
2933 * would call "cont".
2934 * So call bdrv_inactivate_all (release locks) here to let the other
2935 * side of the migration take control of the images.
2936 */
2937 if (live && !saved_vm_running) {
2938 ret = bdrv_inactivate_all();
2939 if (ret) {
2940 error_setg(errp, "%s: bdrv_inactivate_all() failed (%d)",
2941 __func__, ret);
2942 }
2943 }
2944 }
2945
2946 the_end:
2947 if (saved_vm_running) {
2948 vm_start();
2949 }
2950 }
2951
2952 void qmp_xen_load_devices_state(const char *filename, Error **errp)
2953 {
2954 QEMUFile *f;
2955 QIOChannelFile *ioc;
2956 int ret;
2957
2958 /* Guest must be paused before loading the device state; the RAM state
2959 * will already have been loaded by xc
2960 */
2961 if (runstate_is_running()) {
2962 error_setg(errp, "Cannot update device state while vm is running");
2963 return;
2964 }
2965 vm_stop(RUN_STATE_RESTORE_VM);
2966
2967 ioc = qio_channel_file_new_path(filename, O_RDONLY | O_BINARY, 0, errp);
2968 if (!ioc) {
2969 return;
2970 }
2971 qio_channel_set_name(QIO_CHANNEL(ioc), "migration-xen-load-state");
2972 f = qemu_fopen_channel_input(QIO_CHANNEL(ioc));
2973 object_unref(OBJECT(ioc));
2974
2975 ret = qemu_loadvm_state(f);
2976 qemu_fclose(f);
2977 if (ret < 0) {
2978 error_setg(errp, QERR_IO_ERROR);
2979 }
2980 migration_incoming_state_destroy();
2981 }
2982
2983 bool load_snapshot(const char *name, const char *vmstate,
2984 bool has_devices, strList *devices, Error **errp)
2985 {
2986 BlockDriverState *bs_vm_state;
2987 QEMUSnapshotInfo sn;
2988 QEMUFile *f;
2989 int ret;
2990 AioContext *aio_context;
2991 MigrationIncomingState *mis = migration_incoming_get_current();
2992
2993 if (!bdrv_all_can_snapshot(has_devices, devices, errp)) {
2994 return false;
2995 }
2996 ret = bdrv_all_has_snapshot(name, has_devices, devices, errp);
2997 if (ret < 0) {
2998 return false;
2999 }
3000 if (ret == 0) {
3001 error_setg(errp, "Snapshot '%s' does not exist in one or more devices",
3002 name);
3003 return false;
3004 }
3005
3006 bs_vm_state = bdrv_all_find_vmstate_bs(vmstate, has_devices, devices, errp);
3007 if (!bs_vm_state) {
3008 return false;
3009 }
3010 aio_context = bdrv_get_aio_context(bs_vm_state);
3011
3012 /* Don't even try to load empty VM states */
3013 aio_context_acquire(aio_context);
3014 ret = bdrv_snapshot_find(bs_vm_state, &sn, name);
3015 aio_context_release(aio_context);
3016 if (ret < 0) {
3017 return false;
3018 } else if (sn.vm_state_size == 0) {
3019 error_setg(errp, "This is a disk-only snapshot. Revert to it "
3020 " offline using qemu-img");
3021 return false;
3022 }
3023
3024 /*
3025 * Flush the record/replay queue. Now the VM state is going
3026 * to change. Therefore we don't need to preserve its consistency
3027 */
3028 replay_flush_events();
3029
3030 /* Flush all IO requests so they don't interfere with the new state. */
3031 bdrv_drain_all_begin();
3032
3033 ret = bdrv_all_goto_snapshot(name, has_devices, devices, errp);
3034 if (ret < 0) {
3035 goto err_drain;
3036 }
3037
3038 /* restore the VM state */
3039 f = qemu_fopen_bdrv(bs_vm_state, 0);
3040 if (!f) {
3041 error_setg(errp, "Could not open VM state file");
3042 goto err_drain;
3043 }
3044
3045 qemu_system_reset(SHUTDOWN_CAUSE_NONE);
3046 mis->from_src_file = f;
3047
3048 if (!yank_register_instance(MIGRATION_YANK_INSTANCE, errp)) {
3049 ret = -EINVAL;
3050 goto err_drain;
3051 }
3052 aio_context_acquire(aio_context);
3053 ret = qemu_loadvm_state(f);
3054 migration_incoming_state_destroy();
3055 aio_context_release(aio_context);
3056
3057 bdrv_drain_all_end();
3058
3059 if (ret < 0) {
3060 error_setg(errp, "Error %d while loading VM state", ret);
3061 return false;
3062 }
3063
3064 return true;
3065
3066 err_drain:
3067 bdrv_drain_all_end();
3068 return false;
3069 }
3070
3071 bool delete_snapshot(const char *name, bool has_devices,
3072 strList *devices, Error **errp)
3073 {
3074 if (!bdrv_all_can_snapshot(has_devices, devices, errp)) {
3075 return false;
3076 }
3077
3078 if (bdrv_all_delete_snapshot(name, has_devices, devices, errp) < 0) {
3079 return false;
3080 }
3081
3082 return true;
3083 }
3084
3085 void vmstate_register_ram(MemoryRegion *mr, DeviceState *dev)
3086 {
3087 qemu_ram_set_idstr(mr->ram_block,
3088 memory_region_name(mr), dev);
3089 qemu_ram_set_migratable(mr->ram_block);
3090 }
3091
3092 void vmstate_unregister_ram(MemoryRegion *mr, DeviceState *dev)
3093 {
3094 qemu_ram_unset_idstr(mr->ram_block);
3095 qemu_ram_unset_migratable(mr->ram_block);
3096 }
3097
3098 void vmstate_register_ram_global(MemoryRegion *mr)
3099 {
3100 vmstate_register_ram(mr, NULL);
3101 }
3102
3103 bool vmstate_check_only_migratable(const VMStateDescription *vmsd)
3104 {
3105 /* check needed if --only-migratable is specified */
3106 if (!only_migratable) {
3107 return true;
3108 }
3109
3110 return !(vmsd && vmsd->unmigratable);
3111 }
3112
3113 typedef struct SnapshotJob {
3114 Job common;
3115 char *tag;
3116 char *vmstate;
3117 strList *devices;
3118 Coroutine *co;
3119 Error **errp;
3120 bool ret;
3121 } SnapshotJob;
3122
3123 static void qmp_snapshot_job_free(SnapshotJob *s)
3124 {
3125 g_free(s->tag);
3126 g_free(s->vmstate);
3127 qapi_free_strList(s->devices);
3128 }
3129
3130
3131 static void snapshot_load_job_bh(void *opaque)
3132 {
3133 Job *job = opaque;
3134 SnapshotJob *s = container_of(job, SnapshotJob, common);
3135 int orig_vm_running;
3136
3137 job_progress_set_remaining(&s->common, 1);
3138
3139 orig_vm_running = runstate_is_running();
3140 vm_stop(RUN_STATE_RESTORE_VM);
3141
3142 s->ret = load_snapshot(s->tag, s->vmstate, true, s->devices, s->errp);
3143 if (s->ret && orig_vm_running) {
3144 vm_start();
3145 }
3146
3147 job_progress_update(&s->common, 1);
3148
3149 qmp_snapshot_job_free(s);
3150 aio_co_wake(s->co);
3151 }
3152
3153 static void snapshot_save_job_bh(void *opaque)
3154 {
3155 Job *job = opaque;
3156 SnapshotJob *s = container_of(job, SnapshotJob, common);
3157
3158 job_progress_set_remaining(&s->common, 1);
3159 s->ret = save_snapshot(s->tag, false, s->vmstate,
3160 true, s->devices, s->errp);
3161 job_progress_update(&s->common, 1);
3162
3163 qmp_snapshot_job_free(s);
3164 aio_co_wake(s->co);
3165 }
3166
3167 static void snapshot_delete_job_bh(void *opaque)
3168 {
3169 Job *job = opaque;
3170 SnapshotJob *s = container_of(job, SnapshotJob, common);
3171
3172 job_progress_set_remaining(&s->common, 1);
3173 s->ret = delete_snapshot(s->tag, true, s->devices, s->errp);
3174 job_progress_update(&s->common, 1);
3175
3176 qmp_snapshot_job_free(s);
3177 aio_co_wake(s->co);
3178 }
3179
3180 static int coroutine_fn snapshot_save_job_run(Job *job, Error **errp)
3181 {
3182 SnapshotJob *s = container_of(job, SnapshotJob, common);
3183 s->errp = errp;
3184 s->co = qemu_coroutine_self();
3185 aio_bh_schedule_oneshot(qemu_get_aio_context(),
3186 snapshot_save_job_bh, job);
3187 qemu_coroutine_yield();
3188 return s->ret ? 0 : -1;
3189 }
3190
3191 static int coroutine_fn snapshot_load_job_run(Job *job, Error **errp)
3192 {
3193 SnapshotJob *s = container_of(job, SnapshotJob, common);
3194 s->errp = errp;
3195 s->co = qemu_coroutine_self();
3196 aio_bh_schedule_oneshot(qemu_get_aio_context(),
3197 snapshot_load_job_bh, job);
3198 qemu_coroutine_yield();
3199 return s->ret ? 0 : -1;
3200 }
3201
3202 static int coroutine_fn snapshot_delete_job_run(Job *job, Error **errp)
3203 {
3204 SnapshotJob *s = container_of(job, SnapshotJob, common);
3205 s->errp = errp;
3206 s->co = qemu_coroutine_self();
3207 aio_bh_schedule_oneshot(qemu_get_aio_context(),
3208 snapshot_delete_job_bh, job);
3209 qemu_coroutine_yield();
3210 return s->ret ? 0 : -1;
3211 }
3212
3213
3214 static const JobDriver snapshot_load_job_driver = {
3215 .instance_size = sizeof(SnapshotJob),
3216 .job_type = JOB_TYPE_SNAPSHOT_LOAD,
3217 .run = snapshot_load_job_run,
3218 };
3219
3220 static const JobDriver snapshot_save_job_driver = {
3221 .instance_size = sizeof(SnapshotJob),
3222 .job_type = JOB_TYPE_SNAPSHOT_SAVE,
3223 .run = snapshot_save_job_run,
3224 };
3225
3226 static const JobDriver snapshot_delete_job_driver = {
3227 .instance_size = sizeof(SnapshotJob),
3228 .job_type = JOB_TYPE_SNAPSHOT_DELETE,
3229 .run = snapshot_delete_job_run,
3230 };
3231
3232
3233 void qmp_snapshot_save(const char *job_id,
3234 const char *tag,
3235 const char *vmstate,
3236 strList *devices,
3237 Error **errp)
3238 {
3239 SnapshotJob *s;
3240
3241 s = job_create(job_id, &snapshot_save_job_driver, NULL,
3242 qemu_get_aio_context(), JOB_MANUAL_DISMISS,
3243 NULL, NULL, errp);
3244 if (!s) {
3245 return;
3246 }
3247
3248 s->tag = g_strdup(tag);
3249 s->vmstate = g_strdup(vmstate);
3250 s->devices = QAPI_CLONE(strList, devices);
3251
3252 job_start(&s->common);
3253 }
3254
3255 void qmp_snapshot_load(const char *job_id,
3256 const char *tag,
3257 const char *vmstate,
3258 strList *devices,
3259 Error **errp)
3260 {
3261 SnapshotJob *s;
3262
3263 s = job_create(job_id, &snapshot_load_job_driver, NULL,
3264 qemu_get_aio_context(), JOB_MANUAL_DISMISS,
3265 NULL, NULL, errp);
3266 if (!s) {
3267 return;
3268 }
3269
3270 s->tag = g_strdup(tag);
3271 s->vmstate = g_strdup(vmstate);
3272 s->devices = QAPI_CLONE(strList, devices);
3273
3274 job_start(&s->common);
3275 }
3276
3277 void qmp_snapshot_delete(const char *job_id,
3278 const char *tag,
3279 strList *devices,
3280 Error **errp)
3281 {
3282 SnapshotJob *s;
3283
3284 s = job_create(job_id, &snapshot_delete_job_driver, NULL,
3285 qemu_get_aio_context(), JOB_MANUAL_DISMISS,
3286 NULL, NULL, errp);
3287 if (!s) {
3288 return;
3289 }
3290
3291 s->tag = g_strdup(tag);
3292 s->devices = QAPI_CLONE(strList, devices);
3293
3294 job_start(&s->common);
3295 }