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/qerror.h"
46 #include "qemu/error-report.h"
47 #include "sysemu/cpus.h"
48 #include "exec/memory.h"
49 #include "exec/target_page.h"
50 #include "trace.h"
51 #include "qemu/iov.h"
52 #include "qemu/main-loop.h"
53 #include "block/snapshot.h"
54 #include "qemu/cutils.h"
55 #include "io/channel-buffer.h"
56 #include "io/channel-file.h"
57 #include "sysemu/replay.h"
58 #include "sysemu/runstate.h"
59 #include "sysemu/sysemu.h"
60 #include "sysemu/xen.h"
61 #include "qjson.h"
62 #include "migration/colo.h"
63 #include "qemu/bitmap.h"
64 #include "net/announce.h"
65
66 const unsigned int postcopy_ram_discard_version;
67
68 /* Subcommands for QEMU_VM_COMMAND */
69 enum qemu_vm_cmd {
70 MIG_CMD_INVALID = 0, /* Must be 0 */
71 MIG_CMD_OPEN_RETURN_PATH, /* Tell the dest to open the Return path */
72 MIG_CMD_PING, /* Request a PONG on the RP */
73
74 MIG_CMD_POSTCOPY_ADVISE, /* Prior to any page transfers, just
75 warn we might want to do PC */
76 MIG_CMD_POSTCOPY_LISTEN, /* Start listening for incoming
77 pages as it's running. */
78 MIG_CMD_POSTCOPY_RUN, /* Start execution */
79
80 MIG_CMD_POSTCOPY_RAM_DISCARD, /* A list of pages to discard that
81 were previously sent during
82 precopy but are dirty. */
83 MIG_CMD_PACKAGED, /* Send a wrapped stream within this stream */
84 MIG_CMD_ENABLE_COLO, /* Enable COLO */
85 MIG_CMD_POSTCOPY_RESUME, /* resume postcopy on dest */
86 MIG_CMD_RECV_BITMAP, /* Request for recved bitmap on dst */
87 MIG_CMD_MAX
88 };
89
90 #define MAX_VM_CMD_PACKAGED_SIZE UINT32_MAX
91 static struct mig_cmd_args {
92 ssize_t len; /* -1 = variable */
93 const char *name;
94 } mig_cmd_args[] = {
95 [MIG_CMD_INVALID] = { .len = -1, .name = "INVALID" },
96 [MIG_CMD_OPEN_RETURN_PATH] = { .len = 0, .name = "OPEN_RETURN_PATH" },
97 [MIG_CMD_PING] = { .len = sizeof(uint32_t), .name = "PING" },
98 [MIG_CMD_POSTCOPY_ADVISE] = { .len = -1, .name = "POSTCOPY_ADVISE" },
99 [MIG_CMD_POSTCOPY_LISTEN] = { .len = 0, .name = "POSTCOPY_LISTEN" },
100 [MIG_CMD_POSTCOPY_RUN] = { .len = 0, .name = "POSTCOPY_RUN" },
101 [MIG_CMD_POSTCOPY_RAM_DISCARD] = {
102 .len = -1, .name = "POSTCOPY_RAM_DISCARD" },
103 [MIG_CMD_POSTCOPY_RESUME] = { .len = 0, .name = "POSTCOPY_RESUME" },
104 [MIG_CMD_PACKAGED] = { .len = 4, .name = "PACKAGED" },
105 [MIG_CMD_RECV_BITMAP] = { .len = -1, .name = "RECV_BITMAP" },
106 [MIG_CMD_MAX] = { .len = -1, .name = "MAX" },
107 };
108
109 /* Note for MIG_CMD_POSTCOPY_ADVISE:
110 * The format of arguments is depending on postcopy mode:
111 * - postcopy RAM only
112 * uint64_t host page size
113 * uint64_t taget page size
114 *
115 * - postcopy RAM and postcopy dirty bitmaps
116 * format is the same as for postcopy RAM only
117 *
118 * - postcopy dirty bitmaps only
119 * Nothing. Command length field is 0.
120 *
121 * Be careful: adding a new postcopy entity with some other parameters should
122 * not break format self-description ability. Good way is to introduce some
123 * generic extendable format with an exception for two old entities.
124 */
125
126 /***********************************************************/
127 /* savevm/loadvm support */
128
block_writev_buffer(void * opaque,struct iovec * iov,int iovcnt,int64_t pos,Error ** errp)129 static ssize_t block_writev_buffer(void *opaque, struct iovec *iov, int iovcnt,
130 int64_t pos, Error **errp)
131 {
132 int ret;
133 QEMUIOVector qiov;
134
135 qemu_iovec_init_external(&qiov, iov, iovcnt);
136 ret = bdrv_writev_vmstate(opaque, &qiov, pos);
137 if (ret < 0) {
138 return ret;
139 }
140
141 return qiov.size;
142 }
143
block_get_buffer(void * opaque,uint8_t * buf,int64_t pos,size_t size,Error ** errp)144 static ssize_t block_get_buffer(void *opaque, uint8_t *buf, int64_t pos,
145 size_t size, Error **errp)
146 {
147 return bdrv_load_vmstate(opaque, buf, pos, size);
148 }
149
bdrv_fclose(void * opaque,Error ** errp)150 static int bdrv_fclose(void *opaque, Error **errp)
151 {
152 return bdrv_flush(opaque);
153 }
154
155 static const QEMUFileOps bdrv_read_ops = {
156 .get_buffer = block_get_buffer,
157 .close = bdrv_fclose
158 };
159
160 static const QEMUFileOps bdrv_write_ops = {
161 .writev_buffer = block_writev_buffer,
162 .close = bdrv_fclose
163 };
164
qemu_fopen_bdrv(BlockDriverState * bs,int is_writable)165 static QEMUFile *qemu_fopen_bdrv(BlockDriverState *bs, int is_writable)
166 {
167 if (is_writable) {
168 return qemu_fopen_ops(bs, &bdrv_write_ops);
169 }
170 return qemu_fopen_ops(bs, &bdrv_read_ops);
171 }
172
173
174 /* QEMUFile timer support.
175 * Not in qemu-file.c to not add qemu-timer.c as dependency to qemu-file.c
176 */
177
timer_put(QEMUFile * f,QEMUTimer * ts)178 void timer_put(QEMUFile *f, QEMUTimer *ts)
179 {
180 uint64_t expire_time;
181
182 expire_time = timer_expire_time_ns(ts);
183 qemu_put_be64(f, expire_time);
184 }
185
timer_get(QEMUFile * f,QEMUTimer * ts)186 void timer_get(QEMUFile *f, QEMUTimer *ts)
187 {
188 uint64_t expire_time;
189
190 expire_time = qemu_get_be64(f);
191 if (expire_time != -1) {
192 timer_mod_ns(ts, expire_time);
193 } else {
194 timer_del(ts);
195 }
196 }
197
198
199 /* VMState timer support.
200 * Not in vmstate.c to not add qemu-timer.c as dependency to vmstate.c
201 */
202
get_timer(QEMUFile * f,void * pv,size_t size,const VMStateField * field)203 static int get_timer(QEMUFile *f, void *pv, size_t size,
204 const VMStateField *field)
205 {
206 QEMUTimer *v = pv;
207 timer_get(f, v);
208 return 0;
209 }
210
put_timer(QEMUFile * f,void * pv,size_t size,const VMStateField * field,QJSON * vmdesc)211 static int put_timer(QEMUFile *f, void *pv, size_t size,
212 const VMStateField *field, QJSON *vmdesc)
213 {
214 QEMUTimer *v = pv;
215 timer_put(f, v);
216
217 return 0;
218 }
219
220 const VMStateInfo vmstate_info_timer = {
221 .name = "timer",
222 .get = get_timer,
223 .put = put_timer,
224 };
225
226
227 typedef struct CompatEntry {
228 char idstr[256];
229 int instance_id;
230 } CompatEntry;
231
232 typedef struct SaveStateEntry {
233 QTAILQ_ENTRY(SaveStateEntry) entry;
234 char idstr[256];
235 uint32_t instance_id;
236 int alias_id;
237 int version_id;
238 /* version id read from the stream */
239 int load_version_id;
240 int section_id;
241 /* section id read from the stream */
242 int load_section_id;
243 const SaveVMHandlers *ops;
244 const VMStateDescription *vmsd;
245 void *opaque;
246 CompatEntry *compat;
247 int is_ram;
248 } SaveStateEntry;
249
250 typedef struct SaveState {
251 QTAILQ_HEAD(, SaveStateEntry) handlers;
252 SaveStateEntry *handler_pri_head[MIG_PRI_MAX + 1];
253 int global_section_id;
254 uint32_t len;
255 const char *name;
256 uint32_t target_page_bits;
257 uint32_t caps_count;
258 MigrationCapability *capabilities;
259 QemuUUID uuid;
260 } SaveState;
261
262 static SaveState savevm_state = {
263 .handlers = QTAILQ_HEAD_INITIALIZER(savevm_state.handlers),
264 .handler_pri_head = { [MIG_PRI_DEFAULT ... MIG_PRI_MAX] = NULL },
265 .global_section_id = 0,
266 };
267
should_validate_capability(int capability)268 static bool should_validate_capability(int capability)
269 {
270 assert(capability >= 0 && capability < MIGRATION_CAPABILITY__MAX);
271 /* Validate only new capabilities to keep compatibility. */
272 switch (capability) {
273 case MIGRATION_CAPABILITY_X_IGNORE_SHARED:
274 return true;
275 default:
276 return false;
277 }
278 }
279
get_validatable_capabilities_count(void)280 static uint32_t get_validatable_capabilities_count(void)
281 {
282 MigrationState *s = migrate_get_current();
283 uint32_t result = 0;
284 int i;
285 for (i = 0; i < MIGRATION_CAPABILITY__MAX; i++) {
286 if (should_validate_capability(i) && s->enabled_capabilities[i]) {
287 result++;
288 }
289 }
290 return result;
291 }
292
configuration_pre_save(void * opaque)293 static int configuration_pre_save(void *opaque)
294 {
295 SaveState *state = opaque;
296 const char *current_name = MACHINE_GET_CLASS(current_machine)->name;
297 MigrationState *s = migrate_get_current();
298 int i, j;
299
300 state->len = strlen(current_name);
301 state->name = current_name;
302 state->target_page_bits = qemu_target_page_bits();
303
304 state->caps_count = get_validatable_capabilities_count();
305 state->capabilities = g_renew(MigrationCapability, state->capabilities,
306 state->caps_count);
307 for (i = j = 0; i < MIGRATION_CAPABILITY__MAX; i++) {
308 if (should_validate_capability(i) && s->enabled_capabilities[i]) {
309 state->capabilities[j++] = i;
310 }
311 }
312 state->uuid = qemu_uuid;
313
314 return 0;
315 }
316
configuration_pre_load(void * opaque)317 static int configuration_pre_load(void *opaque)
318 {
319 SaveState *state = opaque;
320
321 /* If there is no target-page-bits subsection it means the source
322 * predates the variable-target-page-bits support and is using the
323 * minimum possible value for this CPU.
324 */
325 state->target_page_bits = qemu_target_page_bits_min();
326 return 0;
327 }
328
configuration_validate_capabilities(SaveState * state)329 static bool configuration_validate_capabilities(SaveState *state)
330 {
331 bool ret = true;
332 MigrationState *s = migrate_get_current();
333 unsigned long *source_caps_bm;
334 int i;
335
336 source_caps_bm = bitmap_new(MIGRATION_CAPABILITY__MAX);
337 for (i = 0; i < state->caps_count; i++) {
338 MigrationCapability capability = state->capabilities[i];
339 set_bit(capability, source_caps_bm);
340 }
341
342 for (i = 0; i < MIGRATION_CAPABILITY__MAX; i++) {
343 bool source_state, target_state;
344 if (!should_validate_capability(i)) {
345 continue;
346 }
347 source_state = test_bit(i, source_caps_bm);
348 target_state = s->enabled_capabilities[i];
349 if (source_state != target_state) {
350 error_report("Capability %s is %s, but received capability is %s",
351 MigrationCapability_str(i),
352 target_state ? "on" : "off",
353 source_state ? "on" : "off");
354 ret = false;
355 /* Don't break here to report all failed capabilities */
356 }
357 }
358
359 g_free(source_caps_bm);
360 return ret;
361 }
362
configuration_post_load(void * opaque,int version_id)363 static int configuration_post_load(void *opaque, int version_id)
364 {
365 SaveState *state = opaque;
366 const char *current_name = MACHINE_GET_CLASS(current_machine)->name;
367
368 if (strncmp(state->name, current_name, state->len) != 0) {
369 error_report("Machine type received is '%.*s' and local is '%s'",
370 (int) state->len, state->name, current_name);
371 return -EINVAL;
372 }
373
374 if (state->target_page_bits != qemu_target_page_bits()) {
375 error_report("Received TARGET_PAGE_BITS is %d but local is %d",
376 state->target_page_bits, qemu_target_page_bits());
377 return -EINVAL;
378 }
379
380 if (!configuration_validate_capabilities(state)) {
381 return -EINVAL;
382 }
383
384 return 0;
385 }
386
get_capability(QEMUFile * f,void * pv,size_t size,const VMStateField * field)387 static int get_capability(QEMUFile *f, void *pv, size_t size,
388 const VMStateField *field)
389 {
390 MigrationCapability *capability = pv;
391 char capability_str[UINT8_MAX + 1];
392 uint8_t len;
393 int i;
394
395 len = qemu_get_byte(f);
396 qemu_get_buffer(f, (uint8_t *)capability_str, len);
397 capability_str[len] = '\0';
398 for (i = 0; i < MIGRATION_CAPABILITY__MAX; i++) {
399 if (!strcmp(MigrationCapability_str(i), capability_str)) {
400 *capability = i;
401 return 0;
402 }
403 }
404 error_report("Received unknown capability %s", capability_str);
405 return -EINVAL;
406 }
407
put_capability(QEMUFile * f,void * pv,size_t size,const VMStateField * field,QJSON * vmdesc)408 static int put_capability(QEMUFile *f, void *pv, size_t size,
409 const VMStateField *field, QJSON *vmdesc)
410 {
411 MigrationCapability *capability = pv;
412 const char *capability_str = MigrationCapability_str(*capability);
413 size_t len = strlen(capability_str);
414 assert(len <= UINT8_MAX);
415
416 qemu_put_byte(f, len);
417 qemu_put_buffer(f, (uint8_t *)capability_str, len);
418 return 0;
419 }
420
421 static const VMStateInfo vmstate_info_capability = {
422 .name = "capability",
423 .get = get_capability,
424 .put = put_capability,
425 };
426
427 /* The target-page-bits subsection is present only if the
428 * target page size is not the same as the default (ie the
429 * minimum page size for a variable-page-size guest CPU).
430 * If it is present then it contains the actual target page
431 * bits for the machine, and migration will fail if the
432 * two ends don't agree about it.
433 */
vmstate_target_page_bits_needed(void * opaque)434 static bool vmstate_target_page_bits_needed(void *opaque)
435 {
436 return qemu_target_page_bits()
437 > qemu_target_page_bits_min();
438 }
439
440 static const VMStateDescription vmstate_target_page_bits = {
441 .name = "configuration/target-page-bits",
442 .version_id = 1,
443 .minimum_version_id = 1,
444 .needed = vmstate_target_page_bits_needed,
445 .fields = (VMStateField[]) {
446 VMSTATE_UINT32(target_page_bits, SaveState),
447 VMSTATE_END_OF_LIST()
448 }
449 };
450
vmstate_capabilites_needed(void * opaque)451 static bool vmstate_capabilites_needed(void *opaque)
452 {
453 return get_validatable_capabilities_count() > 0;
454 }
455
456 static const VMStateDescription vmstate_capabilites = {
457 .name = "configuration/capabilities",
458 .version_id = 1,
459 .minimum_version_id = 1,
460 .needed = vmstate_capabilites_needed,
461 .fields = (VMStateField[]) {
462 VMSTATE_UINT32_V(caps_count, SaveState, 1),
463 VMSTATE_VARRAY_UINT32_ALLOC(capabilities, SaveState, caps_count, 1,
464 vmstate_info_capability,
465 MigrationCapability),
466 VMSTATE_END_OF_LIST()
467 }
468 };
469
vmstate_uuid_needed(void * opaque)470 static bool vmstate_uuid_needed(void *opaque)
471 {
472 return qemu_uuid_set && migrate_validate_uuid();
473 }
474
vmstate_uuid_post_load(void * opaque,int version_id)475 static int vmstate_uuid_post_load(void *opaque, int version_id)
476 {
477 SaveState *state = opaque;
478 char uuid_src[UUID_FMT_LEN + 1];
479 char uuid_dst[UUID_FMT_LEN + 1];
480
481 if (!qemu_uuid_set) {
482 /*
483 * It's warning because user might not know UUID in some cases,
484 * e.g. load an old snapshot
485 */
486 qemu_uuid_unparse(&state->uuid, uuid_src);
487 warn_report("UUID is received %s, but local uuid isn't set",
488 uuid_src);
489 return 0;
490 }
491 if (!qemu_uuid_is_equal(&state->uuid, &qemu_uuid)) {
492 qemu_uuid_unparse(&state->uuid, uuid_src);
493 qemu_uuid_unparse(&qemu_uuid, uuid_dst);
494 error_report("UUID received is %s and local is %s", uuid_src, uuid_dst);
495 return -EINVAL;
496 }
497 return 0;
498 }
499
500 static const VMStateDescription vmstate_uuid = {
501 .name = "configuration/uuid",
502 .version_id = 1,
503 .minimum_version_id = 1,
504 .needed = vmstate_uuid_needed,
505 .post_load = vmstate_uuid_post_load,
506 .fields = (VMStateField[]) {
507 VMSTATE_UINT8_ARRAY_V(uuid.data, SaveState, sizeof(QemuUUID), 1),
508 VMSTATE_END_OF_LIST()
509 }
510 };
511
512 static const VMStateDescription vmstate_configuration = {
513 .name = "configuration",
514 .version_id = 1,
515 .pre_load = configuration_pre_load,
516 .post_load = configuration_post_load,
517 .pre_save = configuration_pre_save,
518 .fields = (VMStateField[]) {
519 VMSTATE_UINT32(len, SaveState),
520 VMSTATE_VBUFFER_ALLOC_UINT32(name, SaveState, 0, NULL, len),
521 VMSTATE_END_OF_LIST()
522 },
523 .subsections = (const VMStateDescription *[]) {
524 &vmstate_target_page_bits,
525 &vmstate_capabilites,
526 &vmstate_uuid,
527 NULL
528 }
529 };
530
531 static void dump_vmstate_vmsd(FILE *out_file,
532 const VMStateDescription *vmsd, int indent,
533 bool is_subsection);
534
dump_vmstate_vmsf(FILE * out_file,const VMStateField * field,int indent)535 static void dump_vmstate_vmsf(FILE *out_file, const VMStateField *field,
536 int indent)
537 {
538 fprintf(out_file, "%*s{\n", indent, "");
539 indent += 2;
540 fprintf(out_file, "%*s\"field\": \"%s\",\n", indent, "", field->name);
541 fprintf(out_file, "%*s\"version_id\": %d,\n", indent, "",
542 field->version_id);
543 fprintf(out_file, "%*s\"field_exists\": %s,\n", indent, "",
544 field->field_exists ? "true" : "false");
545 fprintf(out_file, "%*s\"size\": %zu", indent, "", field->size);
546 if (field->vmsd != NULL) {
547 fprintf(out_file, ",\n");
548 dump_vmstate_vmsd(out_file, field->vmsd, indent, false);
549 }
550 fprintf(out_file, "\n%*s}", indent - 2, "");
551 }
552
dump_vmstate_vmss(FILE * out_file,const VMStateDescription ** subsection,int indent)553 static void dump_vmstate_vmss(FILE *out_file,
554 const VMStateDescription **subsection,
555 int indent)
556 {
557 if (*subsection != NULL) {
558 dump_vmstate_vmsd(out_file, *subsection, indent, true);
559 }
560 }
561
dump_vmstate_vmsd(FILE * out_file,const VMStateDescription * vmsd,int indent,bool is_subsection)562 static void dump_vmstate_vmsd(FILE *out_file,
563 const VMStateDescription *vmsd, int indent,
564 bool is_subsection)
565 {
566 if (is_subsection) {
567 fprintf(out_file, "%*s{\n", indent, "");
568 } else {
569 fprintf(out_file, "%*s\"%s\": {\n", indent, "", "Description");
570 }
571 indent += 2;
572 fprintf(out_file, "%*s\"name\": \"%s\",\n", indent, "", vmsd->name);
573 fprintf(out_file, "%*s\"version_id\": %d,\n", indent, "",
574 vmsd->version_id);
575 fprintf(out_file, "%*s\"minimum_version_id\": %d", indent, "",
576 vmsd->minimum_version_id);
577 if (vmsd->fields != NULL) {
578 const VMStateField *field = vmsd->fields;
579 bool first;
580
581 fprintf(out_file, ",\n%*s\"Fields\": [\n", indent, "");
582 first = true;
583 while (field->name != NULL) {
584 if (field->flags & VMS_MUST_EXIST) {
585 /* Ignore VMSTATE_VALIDATE bits; these don't get migrated */
586 field++;
587 continue;
588 }
589 if (!first) {
590 fprintf(out_file, ",\n");
591 }
592 dump_vmstate_vmsf(out_file, field, indent + 2);
593 field++;
594 first = false;
595 }
596 fprintf(out_file, "\n%*s]", indent, "");
597 }
598 if (vmsd->subsections != NULL) {
599 const VMStateDescription **subsection = vmsd->subsections;
600 bool first;
601
602 fprintf(out_file, ",\n%*s\"Subsections\": [\n", indent, "");
603 first = true;
604 while (*subsection != NULL) {
605 if (!first) {
606 fprintf(out_file, ",\n");
607 }
608 dump_vmstate_vmss(out_file, subsection, indent + 2);
609 subsection++;
610 first = false;
611 }
612 fprintf(out_file, "\n%*s]", indent, "");
613 }
614 fprintf(out_file, "\n%*s}", indent - 2, "");
615 }
616
dump_machine_type(FILE * out_file)617 static void dump_machine_type(FILE *out_file)
618 {
619 MachineClass *mc;
620
621 mc = MACHINE_GET_CLASS(current_machine);
622
623 fprintf(out_file, " \"vmschkmachine\": {\n");
624 fprintf(out_file, " \"Name\": \"%s\"\n", mc->name);
625 fprintf(out_file, " },\n");
626 }
627
dump_vmstate_json_to_file(FILE * out_file)628 void dump_vmstate_json_to_file(FILE *out_file)
629 {
630 GSList *list, *elt;
631 bool first;
632
633 fprintf(out_file, "{\n");
634 dump_machine_type(out_file);
635
636 first = true;
637 list = object_class_get_list(TYPE_DEVICE, true);
638 for (elt = list; elt; elt = elt->next) {
639 DeviceClass *dc = OBJECT_CLASS_CHECK(DeviceClass, elt->data,
640 TYPE_DEVICE);
641 const char *name;
642 int indent = 2;
643
644 if (!dc->vmsd) {
645 continue;
646 }
647
648 if (!first) {
649 fprintf(out_file, ",\n");
650 }
651 name = object_class_get_name(OBJECT_CLASS(dc));
652 fprintf(out_file, "%*s\"%s\": {\n", indent, "", name);
653 indent += 2;
654 fprintf(out_file, "%*s\"Name\": \"%s\",\n", indent, "", name);
655 fprintf(out_file, "%*s\"version_id\": %d,\n", indent, "",
656 dc->vmsd->version_id);
657 fprintf(out_file, "%*s\"minimum_version_id\": %d,\n", indent, "",
658 dc->vmsd->minimum_version_id);
659
660 dump_vmstate_vmsd(out_file, dc->vmsd, indent, false);
661
662 fprintf(out_file, "\n%*s}", indent - 2, "");
663 first = false;
664 }
665 fprintf(out_file, "\n}\n");
666 fclose(out_file);
667 g_slist_free(list);
668 }
669
calculate_new_instance_id(const char * idstr)670 static uint32_t calculate_new_instance_id(const char *idstr)
671 {
672 SaveStateEntry *se;
673 uint32_t instance_id = 0;
674
675 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
676 if (strcmp(idstr, se->idstr) == 0
677 && instance_id <= se->instance_id) {
678 instance_id = se->instance_id + 1;
679 }
680 }
681 /* Make sure we never loop over without being noticed */
682 assert(instance_id != VMSTATE_INSTANCE_ID_ANY);
683 return instance_id;
684 }
685
calculate_compat_instance_id(const char * idstr)686 static int calculate_compat_instance_id(const char *idstr)
687 {
688 SaveStateEntry *se;
689 int instance_id = 0;
690
691 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
692 if (!se->compat) {
693 continue;
694 }
695
696 if (strcmp(idstr, se->compat->idstr) == 0
697 && instance_id <= se->compat->instance_id) {
698 instance_id = se->compat->instance_id + 1;
699 }
700 }
701 return instance_id;
702 }
703
save_state_priority(SaveStateEntry * se)704 static inline MigrationPriority save_state_priority(SaveStateEntry *se)
705 {
706 if (se->vmsd) {
707 return se->vmsd->priority;
708 }
709 return MIG_PRI_DEFAULT;
710 }
711
savevm_state_handler_insert(SaveStateEntry * nse)712 static void savevm_state_handler_insert(SaveStateEntry *nse)
713 {
714 MigrationPriority priority = save_state_priority(nse);
715 SaveStateEntry *se;
716 int i;
717
718 assert(priority <= MIG_PRI_MAX);
719
720 for (i = priority - 1; i >= 0; i--) {
721 se = savevm_state.handler_pri_head[i];
722 if (se != NULL) {
723 assert(save_state_priority(se) < priority);
724 break;
725 }
726 }
727
728 if (i >= 0) {
729 QTAILQ_INSERT_BEFORE(se, nse, entry);
730 } else {
731 QTAILQ_INSERT_TAIL(&savevm_state.handlers, nse, entry);
732 }
733
734 if (savevm_state.handler_pri_head[priority] == NULL) {
735 savevm_state.handler_pri_head[priority] = nse;
736 }
737 }
738
savevm_state_handler_remove(SaveStateEntry * se)739 static void savevm_state_handler_remove(SaveStateEntry *se)
740 {
741 SaveStateEntry *next;
742 MigrationPriority priority = save_state_priority(se);
743
744 if (se == savevm_state.handler_pri_head[priority]) {
745 next = QTAILQ_NEXT(se, entry);
746 if (next != NULL && save_state_priority(next) == priority) {
747 savevm_state.handler_pri_head[priority] = next;
748 } else {
749 savevm_state.handler_pri_head[priority] = NULL;
750 }
751 }
752 QTAILQ_REMOVE(&savevm_state.handlers, se, entry);
753 }
754
755 /* TODO: Individual devices generally have very little idea about the rest
756 of the system, so instance_id should be removed/replaced.
757 Meanwhile pass -1 as instance_id if you do not already have a clearly
758 distinguishing id for all instances of your device class. */
register_savevm_live(const char * idstr,uint32_t instance_id,int version_id,const SaveVMHandlers * ops,void * opaque)759 int register_savevm_live(const char *idstr,
760 uint32_t instance_id,
761 int version_id,
762 const SaveVMHandlers *ops,
763 void *opaque)
764 {
765 SaveStateEntry *se;
766
767 se = g_new0(SaveStateEntry, 1);
768 se->version_id = version_id;
769 se->section_id = savevm_state.global_section_id++;
770 se->ops = ops;
771 se->opaque = opaque;
772 se->vmsd = NULL;
773 /* if this is a live_savem then set is_ram */
774 if (ops->save_setup != NULL) {
775 se->is_ram = 1;
776 }
777
778 pstrcat(se->idstr, sizeof(se->idstr), idstr);
779
780 if (instance_id == VMSTATE_INSTANCE_ID_ANY) {
781 se->instance_id = calculate_new_instance_id(se->idstr);
782 } else {
783 se->instance_id = instance_id;
784 }
785 assert(!se->compat || se->instance_id == 0);
786 savevm_state_handler_insert(se);
787 return 0;
788 }
789
unregister_savevm(VMStateIf * obj,const char * idstr,void * opaque)790 void unregister_savevm(VMStateIf *obj, const char *idstr, void *opaque)
791 {
792 SaveStateEntry *se, *new_se;
793 char id[256] = "";
794
795 if (obj) {
796 char *oid = vmstate_if_get_id(obj);
797 if (oid) {
798 pstrcpy(id, sizeof(id), oid);
799 pstrcat(id, sizeof(id), "/");
800 g_free(oid);
801 }
802 }
803 pstrcat(id, sizeof(id), idstr);
804
805 QTAILQ_FOREACH_SAFE(se, &savevm_state.handlers, entry, new_se) {
806 if (strcmp(se->idstr, id) == 0 && se->opaque == opaque) {
807 savevm_state_handler_remove(se);
808 g_free(se->compat);
809 g_free(se);
810 }
811 }
812 }
813
vmstate_register_with_alias_id(VMStateIf * obj,uint32_t instance_id,const VMStateDescription * vmsd,void * opaque,int alias_id,int required_for_version,Error ** errp)814 int vmstate_register_with_alias_id(VMStateIf *obj, uint32_t instance_id,
815 const VMStateDescription *vmsd,
816 void *opaque, int alias_id,
817 int required_for_version,
818 Error **errp)
819 {
820 SaveStateEntry *se;
821
822 /* If this triggers, alias support can be dropped for the vmsd. */
823 assert(alias_id == -1 || required_for_version >= vmsd->minimum_version_id);
824
825 se = g_new0(SaveStateEntry, 1);
826 se->version_id = vmsd->version_id;
827 se->section_id = savevm_state.global_section_id++;
828 se->opaque = opaque;
829 se->vmsd = vmsd;
830 se->alias_id = alias_id;
831
832 if (obj) {
833 char *id = vmstate_if_get_id(obj);
834 if (id) {
835 if (snprintf(se->idstr, sizeof(se->idstr), "%s/", id) >=
836 sizeof(se->idstr)) {
837 error_setg(errp, "Path too long for VMState (%s)", id);
838 g_free(id);
839 g_free(se);
840
841 return -1;
842 }
843 g_free(id);
844
845 se->compat = g_new0(CompatEntry, 1);
846 pstrcpy(se->compat->idstr, sizeof(se->compat->idstr), vmsd->name);
847 se->compat->instance_id = instance_id == VMSTATE_INSTANCE_ID_ANY ?
848 calculate_compat_instance_id(vmsd->name) : instance_id;
849 instance_id = VMSTATE_INSTANCE_ID_ANY;
850 }
851 }
852 pstrcat(se->idstr, sizeof(se->idstr), vmsd->name);
853
854 if (instance_id == VMSTATE_INSTANCE_ID_ANY) {
855 se->instance_id = calculate_new_instance_id(se->idstr);
856 } else {
857 se->instance_id = instance_id;
858 }
859 assert(!se->compat || se->instance_id == 0);
860 savevm_state_handler_insert(se);
861 return 0;
862 }
863
vmstate_unregister(VMStateIf * obj,const VMStateDescription * vmsd,void * opaque)864 void vmstate_unregister(VMStateIf *obj, const VMStateDescription *vmsd,
865 void *opaque)
866 {
867 SaveStateEntry *se, *new_se;
868
869 QTAILQ_FOREACH_SAFE(se, &savevm_state.handlers, entry, new_se) {
870 if (se->vmsd == vmsd && se->opaque == opaque) {
871 savevm_state_handler_remove(se);
872 g_free(se->compat);
873 g_free(se);
874 }
875 }
876 }
877
vmstate_load(QEMUFile * f,SaveStateEntry * se)878 static int vmstate_load(QEMUFile *f, SaveStateEntry *se)
879 {
880 trace_vmstate_load(se->idstr, se->vmsd ? se->vmsd->name : "(old)");
881 if (!se->vmsd) { /* Old style */
882 return se->ops->load_state(f, se->opaque, se->load_version_id);
883 }
884 return vmstate_load_state(f, se->vmsd, se->opaque, se->load_version_id);
885 }
886
vmstate_save_old_style(QEMUFile * f,SaveStateEntry * se,QJSON * vmdesc)887 static void vmstate_save_old_style(QEMUFile *f, SaveStateEntry *se, QJSON *vmdesc)
888 {
889 int64_t old_offset, size;
890
891 old_offset = qemu_ftell_fast(f);
892 se->ops->save_state(f, se->opaque);
893 size = qemu_ftell_fast(f) - old_offset;
894
895 if (vmdesc) {
896 json_prop_int(vmdesc, "size", size);
897 json_start_array(vmdesc, "fields");
898 json_start_object(vmdesc, NULL);
899 json_prop_str(vmdesc, "name", "data");
900 json_prop_int(vmdesc, "size", size);
901 json_prop_str(vmdesc, "type", "buffer");
902 json_end_object(vmdesc);
903 json_end_array(vmdesc);
904 }
905 }
906
vmstate_save(QEMUFile * f,SaveStateEntry * se,QJSON * vmdesc)907 static int vmstate_save(QEMUFile *f, SaveStateEntry *se, QJSON *vmdesc)
908 {
909 trace_vmstate_save(se->idstr, se->vmsd ? se->vmsd->name : "(old)");
910 if (!se->vmsd) {
911 vmstate_save_old_style(f, se, vmdesc);
912 return 0;
913 }
914 return vmstate_save_state(f, se->vmsd, se->opaque, vmdesc);
915 }
916
917 /*
918 * Write the header for device section (QEMU_VM_SECTION START/END/PART/FULL)
919 */
save_section_header(QEMUFile * f,SaveStateEntry * se,uint8_t section_type)920 static void save_section_header(QEMUFile *f, SaveStateEntry *se,
921 uint8_t section_type)
922 {
923 qemu_put_byte(f, section_type);
924 qemu_put_be32(f, se->section_id);
925
926 if (section_type == QEMU_VM_SECTION_FULL ||
927 section_type == QEMU_VM_SECTION_START) {
928 /* ID string */
929 size_t len = strlen(se->idstr);
930 qemu_put_byte(f, len);
931 qemu_put_buffer(f, (uint8_t *)se->idstr, len);
932
933 qemu_put_be32(f, se->instance_id);
934 qemu_put_be32(f, se->version_id);
935 }
936 }
937
938 /*
939 * Write a footer onto device sections that catches cases misformatted device
940 * sections.
941 */
save_section_footer(QEMUFile * f,SaveStateEntry * se)942 static void save_section_footer(QEMUFile *f, SaveStateEntry *se)
943 {
944 if (migrate_get_current()->send_section_footer) {
945 qemu_put_byte(f, QEMU_VM_SECTION_FOOTER);
946 qemu_put_be32(f, se->section_id);
947 }
948 }
949
950 /**
951 * qemu_savevm_command_send: Send a 'QEMU_VM_COMMAND' type element with the
952 * command and associated data.
953 *
954 * @f: File to send command on
955 * @command: Command type to send
956 * @len: Length of associated data
957 * @data: Data associated with command.
958 */
qemu_savevm_command_send(QEMUFile * f,enum qemu_vm_cmd command,uint16_t len,uint8_t * data)959 static void qemu_savevm_command_send(QEMUFile *f,
960 enum qemu_vm_cmd command,
961 uint16_t len,
962 uint8_t *data)
963 {
964 trace_savevm_command_send(command, len);
965 qemu_put_byte(f, QEMU_VM_COMMAND);
966 qemu_put_be16(f, (uint16_t)command);
967 qemu_put_be16(f, len);
968 qemu_put_buffer(f, data, len);
969 qemu_fflush(f);
970 }
971
qemu_savevm_send_colo_enable(QEMUFile * f)972 void qemu_savevm_send_colo_enable(QEMUFile *f)
973 {
974 trace_savevm_send_colo_enable();
975 qemu_savevm_command_send(f, MIG_CMD_ENABLE_COLO, 0, NULL);
976 }
977
qemu_savevm_send_ping(QEMUFile * f,uint32_t value)978 void qemu_savevm_send_ping(QEMUFile *f, uint32_t value)
979 {
980 uint32_t buf;
981
982 trace_savevm_send_ping(value);
983 buf = cpu_to_be32(value);
984 qemu_savevm_command_send(f, MIG_CMD_PING, sizeof(value), (uint8_t *)&buf);
985 }
986
qemu_savevm_send_open_return_path(QEMUFile * f)987 void qemu_savevm_send_open_return_path(QEMUFile *f)
988 {
989 trace_savevm_send_open_return_path();
990 qemu_savevm_command_send(f, MIG_CMD_OPEN_RETURN_PATH, 0, NULL);
991 }
992
993 /* We have a buffer of data to send; we don't want that all to be loaded
994 * by the command itself, so the command contains just the length of the
995 * extra buffer that we then send straight after it.
996 * TODO: Must be a better way to organise that
997 *
998 * Returns:
999 * 0 on success
1000 * -ve on error
1001 */
qemu_savevm_send_packaged(QEMUFile * f,const uint8_t * buf,size_t len)1002 int qemu_savevm_send_packaged(QEMUFile *f, const uint8_t *buf, size_t len)
1003 {
1004 uint32_t tmp;
1005
1006 if (len > MAX_VM_CMD_PACKAGED_SIZE) {
1007 error_report("%s: Unreasonably large packaged state: %zu",
1008 __func__, len);
1009 return -1;
1010 }
1011
1012 tmp = cpu_to_be32(len);
1013
1014 trace_qemu_savevm_send_packaged();
1015 qemu_savevm_command_send(f, MIG_CMD_PACKAGED, 4, (uint8_t *)&tmp);
1016
1017 qemu_put_buffer(f, buf, len);
1018
1019 return 0;
1020 }
1021
1022 /* Send prior to any postcopy transfer */
qemu_savevm_send_postcopy_advise(QEMUFile * f)1023 void qemu_savevm_send_postcopy_advise(QEMUFile *f)
1024 {
1025 if (migrate_postcopy_ram()) {
1026 uint64_t tmp[2];
1027 tmp[0] = cpu_to_be64(ram_pagesize_summary());
1028 tmp[1] = cpu_to_be64(qemu_target_page_size());
1029
1030 trace_qemu_savevm_send_postcopy_advise();
1031 qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_ADVISE,
1032 16, (uint8_t *)tmp);
1033 } else {
1034 qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_ADVISE, 0, NULL);
1035 }
1036 }
1037
1038 /* Sent prior to starting the destination running in postcopy, discard pages
1039 * that have already been sent but redirtied on the source.
1040 * CMD_POSTCOPY_RAM_DISCARD consist of:
1041 * byte version (0)
1042 * byte Length of name field (not including 0)
1043 * n x byte RAM block name
1044 * byte 0 terminator (just for safety)
1045 * n x Byte ranges within the named RAMBlock
1046 * be64 Start of the range
1047 * be64 Length
1048 *
1049 * name: RAMBlock name that these entries are part of
1050 * len: Number of page entries
1051 * start_list: 'len' addresses
1052 * length_list: 'len' addresses
1053 *
1054 */
qemu_savevm_send_postcopy_ram_discard(QEMUFile * f,const char * name,uint16_t len,uint64_t * start_list,uint64_t * length_list)1055 void qemu_savevm_send_postcopy_ram_discard(QEMUFile *f, const char *name,
1056 uint16_t len,
1057 uint64_t *start_list,
1058 uint64_t *length_list)
1059 {
1060 uint8_t *buf;
1061 uint16_t tmplen;
1062 uint16_t t;
1063 size_t name_len = strlen(name);
1064
1065 trace_qemu_savevm_send_postcopy_ram_discard(name, len);
1066 assert(name_len < 256);
1067 buf = g_malloc0(1 + 1 + name_len + 1 + (8 + 8) * len);
1068 buf[0] = postcopy_ram_discard_version;
1069 buf[1] = name_len;
1070 memcpy(buf + 2, name, name_len);
1071 tmplen = 2 + name_len;
1072 buf[tmplen++] = '\0';
1073
1074 for (t = 0; t < len; t++) {
1075 stq_be_p(buf + tmplen, start_list[t]);
1076 tmplen += 8;
1077 stq_be_p(buf + tmplen, length_list[t]);
1078 tmplen += 8;
1079 }
1080 qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_RAM_DISCARD, tmplen, buf);
1081 g_free(buf);
1082 }
1083
1084 /* Get the destination into a state where it can receive postcopy data. */
qemu_savevm_send_postcopy_listen(QEMUFile * f)1085 void qemu_savevm_send_postcopy_listen(QEMUFile *f)
1086 {
1087 trace_savevm_send_postcopy_listen();
1088 qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_LISTEN, 0, NULL);
1089 }
1090
1091 /* Kick the destination into running */
qemu_savevm_send_postcopy_run(QEMUFile * f)1092 void qemu_savevm_send_postcopy_run(QEMUFile *f)
1093 {
1094 trace_savevm_send_postcopy_run();
1095 qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_RUN, 0, NULL);
1096 }
1097
qemu_savevm_send_postcopy_resume(QEMUFile * f)1098 void qemu_savevm_send_postcopy_resume(QEMUFile *f)
1099 {
1100 trace_savevm_send_postcopy_resume();
1101 qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_RESUME, 0, NULL);
1102 }
1103
qemu_savevm_send_recv_bitmap(QEMUFile * f,char * block_name)1104 void qemu_savevm_send_recv_bitmap(QEMUFile *f, char *block_name)
1105 {
1106 size_t len;
1107 char buf[256];
1108
1109 trace_savevm_send_recv_bitmap(block_name);
1110
1111 buf[0] = len = strlen(block_name);
1112 memcpy(buf + 1, block_name, len);
1113
1114 qemu_savevm_command_send(f, MIG_CMD_RECV_BITMAP, len + 1, (uint8_t *)buf);
1115 }
1116
qemu_savevm_state_blocked(Error ** errp)1117 bool qemu_savevm_state_blocked(Error **errp)
1118 {
1119 SaveStateEntry *se;
1120
1121 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1122 if (se->vmsd && se->vmsd->unmigratable) {
1123 error_setg(errp, "State blocked by non-migratable device '%s'",
1124 se->idstr);
1125 return true;
1126 }
1127 }
1128 return false;
1129 }
1130
qemu_savevm_state_header(QEMUFile * f)1131 void qemu_savevm_state_header(QEMUFile *f)
1132 {
1133 trace_savevm_state_header();
1134 qemu_put_be32(f, QEMU_VM_FILE_MAGIC);
1135 qemu_put_be32(f, QEMU_VM_FILE_VERSION);
1136
1137 if (migrate_get_current()->send_configuration) {
1138 qemu_put_byte(f, QEMU_VM_CONFIGURATION);
1139 vmstate_save_state(f, &vmstate_configuration, &savevm_state, 0);
1140 }
1141 }
1142
qemu_savevm_state_guest_unplug_pending(void)1143 bool qemu_savevm_state_guest_unplug_pending(void)
1144 {
1145 SaveStateEntry *se;
1146
1147 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1148 if (se->vmsd && se->vmsd->dev_unplug_pending &&
1149 se->vmsd->dev_unplug_pending(se->opaque)) {
1150 return true;
1151 }
1152 }
1153
1154 return false;
1155 }
1156
qemu_savevm_state_setup(QEMUFile * f)1157 void qemu_savevm_state_setup(QEMUFile *f)
1158 {
1159 SaveStateEntry *se;
1160 Error *local_err = NULL;
1161 int ret;
1162
1163 trace_savevm_state_setup();
1164 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1165 if (!se->ops || !se->ops->save_setup) {
1166 continue;
1167 }
1168 if (se->ops->is_active) {
1169 if (!se->ops->is_active(se->opaque)) {
1170 continue;
1171 }
1172 }
1173 save_section_header(f, se, QEMU_VM_SECTION_START);
1174
1175 ret = se->ops->save_setup(f, se->opaque);
1176 save_section_footer(f, se);
1177 if (ret < 0) {
1178 qemu_file_set_error(f, ret);
1179 break;
1180 }
1181 }
1182
1183 if (precopy_notify(PRECOPY_NOTIFY_SETUP, &local_err)) {
1184 error_report_err(local_err);
1185 }
1186 }
1187
qemu_savevm_state_resume_prepare(MigrationState * s)1188 int qemu_savevm_state_resume_prepare(MigrationState *s)
1189 {
1190 SaveStateEntry *se;
1191 int ret;
1192
1193 trace_savevm_state_resume_prepare();
1194
1195 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1196 if (!se->ops || !se->ops->resume_prepare) {
1197 continue;
1198 }
1199 if (se->ops->is_active) {
1200 if (!se->ops->is_active(se->opaque)) {
1201 continue;
1202 }
1203 }
1204 ret = se->ops->resume_prepare(s, se->opaque);
1205 if (ret < 0) {
1206 return ret;
1207 }
1208 }
1209
1210 return 0;
1211 }
1212
1213 /*
1214 * this function has three return values:
1215 * negative: there was one error, and we have -errno.
1216 * 0 : We haven't finished, caller have to go again
1217 * 1 : We have finished, we can go to complete phase
1218 */
qemu_savevm_state_iterate(QEMUFile * f,bool postcopy)1219 int qemu_savevm_state_iterate(QEMUFile *f, bool postcopy)
1220 {
1221 SaveStateEntry *se;
1222 int ret = 1;
1223
1224 trace_savevm_state_iterate();
1225 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1226 if (!se->ops || !se->ops->save_live_iterate) {
1227 continue;
1228 }
1229 if (se->ops->is_active &&
1230 !se->ops->is_active(se->opaque)) {
1231 continue;
1232 }
1233 if (se->ops->is_active_iterate &&
1234 !se->ops->is_active_iterate(se->opaque)) {
1235 continue;
1236 }
1237 /*
1238 * In the postcopy phase, any device that doesn't know how to
1239 * do postcopy should have saved it's state in the _complete
1240 * call that's already run, it might get confused if we call
1241 * iterate afterwards.
1242 */
1243 if (postcopy &&
1244 !(se->ops->has_postcopy && se->ops->has_postcopy(se->opaque))) {
1245 continue;
1246 }
1247 if (qemu_file_rate_limit(f)) {
1248 return 0;
1249 }
1250 trace_savevm_section_start(se->idstr, se->section_id);
1251
1252 save_section_header(f, se, QEMU_VM_SECTION_PART);
1253
1254 ret = se->ops->save_live_iterate(f, se->opaque);
1255 trace_savevm_section_end(se->idstr, se->section_id, ret);
1256 save_section_footer(f, se);
1257
1258 if (ret < 0) {
1259 error_report("failed to save SaveStateEntry with id(name): %d(%s)",
1260 se->section_id, se->idstr);
1261 qemu_file_set_error(f, ret);
1262 }
1263 if (ret <= 0) {
1264 /* Do not proceed to the next vmstate before this one reported
1265 completion of the current stage. This serializes the migration
1266 and reduces the probability that a faster changing state is
1267 synchronized over and over again. */
1268 break;
1269 }
1270 }
1271 return ret;
1272 }
1273
should_send_vmdesc(void)1274 static bool should_send_vmdesc(void)
1275 {
1276 MachineState *machine = MACHINE(qdev_get_machine());
1277 bool in_postcopy = migration_in_postcopy();
1278 return !machine->suppress_vmdesc && !in_postcopy;
1279 }
1280
1281 /*
1282 * Calls the save_live_complete_postcopy methods
1283 * causing the last few pages to be sent immediately and doing any associated
1284 * cleanup.
1285 * Note postcopy also calls qemu_savevm_state_complete_precopy to complete
1286 * all the other devices, but that happens at the point we switch to postcopy.
1287 */
qemu_savevm_state_complete_postcopy(QEMUFile * f)1288 void qemu_savevm_state_complete_postcopy(QEMUFile *f)
1289 {
1290 SaveStateEntry *se;
1291 int ret;
1292
1293 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1294 if (!se->ops || !se->ops->save_live_complete_postcopy) {
1295 continue;
1296 }
1297 if (se->ops->is_active) {
1298 if (!se->ops->is_active(se->opaque)) {
1299 continue;
1300 }
1301 }
1302 trace_savevm_section_start(se->idstr, se->section_id);
1303 /* Section type */
1304 qemu_put_byte(f, QEMU_VM_SECTION_END);
1305 qemu_put_be32(f, se->section_id);
1306
1307 ret = se->ops->save_live_complete_postcopy(f, se->opaque);
1308 trace_savevm_section_end(se->idstr, se->section_id, ret);
1309 save_section_footer(f, se);
1310 if (ret < 0) {
1311 qemu_file_set_error(f, ret);
1312 return;
1313 }
1314 }
1315
1316 qemu_put_byte(f, QEMU_VM_EOF);
1317 qemu_fflush(f);
1318 }
1319
1320 static
qemu_savevm_state_complete_precopy_iterable(QEMUFile * f,bool in_postcopy)1321 int qemu_savevm_state_complete_precopy_iterable(QEMUFile *f, bool in_postcopy)
1322 {
1323 SaveStateEntry *se;
1324 int ret;
1325
1326 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1327 if (!se->ops ||
1328 (in_postcopy && se->ops->has_postcopy &&
1329 se->ops->has_postcopy(se->opaque)) ||
1330 !se->ops->save_live_complete_precopy) {
1331 continue;
1332 }
1333
1334 if (se->ops->is_active) {
1335 if (!se->ops->is_active(se->opaque)) {
1336 continue;
1337 }
1338 }
1339 trace_savevm_section_start(se->idstr, se->section_id);
1340
1341 save_section_header(f, se, QEMU_VM_SECTION_END);
1342
1343 ret = se->ops->save_live_complete_precopy(f, se->opaque);
1344 trace_savevm_section_end(se->idstr, se->section_id, ret);
1345 save_section_footer(f, se);
1346 if (ret < 0) {
1347 qemu_file_set_error(f, ret);
1348 return -1;
1349 }
1350 }
1351
1352 return 0;
1353 }
1354
1355 static
qemu_savevm_state_complete_precopy_non_iterable(QEMUFile * f,bool in_postcopy,bool inactivate_disks)1356 int qemu_savevm_state_complete_precopy_non_iterable(QEMUFile *f,
1357 bool in_postcopy,
1358 bool inactivate_disks)
1359 {
1360 g_autoptr(QJSON) vmdesc = NULL;
1361 int vmdesc_len;
1362 SaveStateEntry *se;
1363 int ret;
1364
1365 vmdesc = qjson_new();
1366 json_prop_int(vmdesc, "page_size", qemu_target_page_size());
1367 json_start_array(vmdesc, "devices");
1368 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1369
1370 if ((!se->ops || !se->ops->save_state) && !se->vmsd) {
1371 continue;
1372 }
1373 if (se->vmsd && !vmstate_save_needed(se->vmsd, se->opaque)) {
1374 trace_savevm_section_skip(se->idstr, se->section_id);
1375 continue;
1376 }
1377
1378 trace_savevm_section_start(se->idstr, se->section_id);
1379
1380 json_start_object(vmdesc, NULL);
1381 json_prop_str(vmdesc, "name", se->idstr);
1382 json_prop_int(vmdesc, "instance_id", se->instance_id);
1383
1384 save_section_header(f, se, QEMU_VM_SECTION_FULL);
1385 ret = vmstate_save(f, se, vmdesc);
1386 if (ret) {
1387 qemu_file_set_error(f, ret);
1388 return ret;
1389 }
1390 trace_savevm_section_end(se->idstr, se->section_id, 0);
1391 save_section_footer(f, se);
1392
1393 json_end_object(vmdesc);
1394 }
1395
1396 if (inactivate_disks) {
1397 /* Inactivate before sending QEMU_VM_EOF so that the
1398 * bdrv_invalidate_cache_all() on the other end won't fail. */
1399 ret = bdrv_inactivate_all();
1400 if (ret) {
1401 error_report("%s: bdrv_inactivate_all() failed (%d)",
1402 __func__, ret);
1403 qemu_file_set_error(f, ret);
1404 return ret;
1405 }
1406 }
1407 if (!in_postcopy) {
1408 /* Postcopy stream will still be going */
1409 qemu_put_byte(f, QEMU_VM_EOF);
1410 }
1411
1412 json_end_array(vmdesc);
1413 qjson_finish(vmdesc);
1414 vmdesc_len = strlen(qjson_get_str(vmdesc));
1415
1416 if (should_send_vmdesc()) {
1417 qemu_put_byte(f, QEMU_VM_VMDESCRIPTION);
1418 qemu_put_be32(f, vmdesc_len);
1419 qemu_put_buffer(f, (uint8_t *)qjson_get_str(vmdesc), vmdesc_len);
1420 }
1421
1422 return 0;
1423 }
1424
qemu_savevm_state_complete_precopy(QEMUFile * f,bool iterable_only,bool inactivate_disks)1425 int qemu_savevm_state_complete_precopy(QEMUFile *f, bool iterable_only,
1426 bool inactivate_disks)
1427 {
1428 int ret;
1429 Error *local_err = NULL;
1430 bool in_postcopy = migration_in_postcopy();
1431
1432 if (precopy_notify(PRECOPY_NOTIFY_COMPLETE, &local_err)) {
1433 error_report_err(local_err);
1434 }
1435
1436 trace_savevm_state_complete_precopy();
1437
1438 cpu_synchronize_all_states();
1439
1440 if (!in_postcopy || iterable_only) {
1441 ret = qemu_savevm_state_complete_precopy_iterable(f, in_postcopy);
1442 if (ret) {
1443 return ret;
1444 }
1445 }
1446
1447 if (iterable_only) {
1448 goto flush;
1449 }
1450
1451 ret = qemu_savevm_state_complete_precopy_non_iterable(f, in_postcopy,
1452 inactivate_disks);
1453 if (ret) {
1454 return ret;
1455 }
1456
1457 flush:
1458 qemu_fflush(f);
1459 return 0;
1460 }
1461
1462 /* Give an estimate of the amount left to be transferred,
1463 * the result is split into the amount for units that can and
1464 * for units that can't do postcopy.
1465 */
qemu_savevm_state_pending(QEMUFile * f,uint64_t threshold_size,uint64_t * res_precopy_only,uint64_t * res_compatible,uint64_t * res_postcopy_only)1466 void qemu_savevm_state_pending(QEMUFile *f, uint64_t threshold_size,
1467 uint64_t *res_precopy_only,
1468 uint64_t *res_compatible,
1469 uint64_t *res_postcopy_only)
1470 {
1471 SaveStateEntry *se;
1472
1473 *res_precopy_only = 0;
1474 *res_compatible = 0;
1475 *res_postcopy_only = 0;
1476
1477
1478 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1479 if (!se->ops || !se->ops->save_live_pending) {
1480 continue;
1481 }
1482 if (se->ops->is_active) {
1483 if (!se->ops->is_active(se->opaque)) {
1484 continue;
1485 }
1486 }
1487 se->ops->save_live_pending(f, se->opaque, threshold_size,
1488 res_precopy_only, res_compatible,
1489 res_postcopy_only);
1490 }
1491 }
1492
qemu_savevm_state_cleanup(void)1493 void qemu_savevm_state_cleanup(void)
1494 {
1495 SaveStateEntry *se;
1496 Error *local_err = NULL;
1497
1498 if (precopy_notify(PRECOPY_NOTIFY_CLEANUP, &local_err)) {
1499 error_report_err(local_err);
1500 }
1501
1502 trace_savevm_state_cleanup();
1503 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1504 if (se->ops && se->ops->save_cleanup) {
1505 se->ops->save_cleanup(se->opaque);
1506 }
1507 }
1508 }
1509
qemu_savevm_state(QEMUFile * f,Error ** errp)1510 static int qemu_savevm_state(QEMUFile *f, Error **errp)
1511 {
1512 int ret;
1513 MigrationState *ms = migrate_get_current();
1514 MigrationStatus status;
1515
1516 if (migration_is_running(ms->state)) {
1517 error_setg(errp, QERR_MIGRATION_ACTIVE);
1518 return -EINVAL;
1519 }
1520
1521 if (migrate_use_block()) {
1522 error_setg(errp, "Block migration and snapshots are incompatible");
1523 return -EINVAL;
1524 }
1525
1526 migrate_init(ms);
1527 memset(&ram_counters, 0, sizeof(ram_counters));
1528 ms->to_dst_file = f;
1529
1530 qemu_mutex_unlock_iothread();
1531 qemu_savevm_state_header(f);
1532 qemu_savevm_state_setup(f);
1533 qemu_mutex_lock_iothread();
1534
1535 while (qemu_file_get_error(f) == 0) {
1536 if (qemu_savevm_state_iterate(f, false) > 0) {
1537 break;
1538 }
1539 }
1540
1541 ret = qemu_file_get_error(f);
1542 if (ret == 0) {
1543 qemu_savevm_state_complete_precopy(f, false, false);
1544 ret = qemu_file_get_error(f);
1545 }
1546 qemu_savevm_state_cleanup();
1547 if (ret != 0) {
1548 error_setg_errno(errp, -ret, "Error while writing VM state");
1549 }
1550
1551 if (ret != 0) {
1552 status = MIGRATION_STATUS_FAILED;
1553 } else {
1554 status = MIGRATION_STATUS_COMPLETED;
1555 }
1556 migrate_set_state(&ms->state, MIGRATION_STATUS_SETUP, status);
1557
1558 /* f is outer parameter, it should not stay in global migration state after
1559 * this function finished */
1560 ms->to_dst_file = NULL;
1561
1562 return ret;
1563 }
1564
qemu_savevm_live_state(QEMUFile * f)1565 void qemu_savevm_live_state(QEMUFile *f)
1566 {
1567 /* save QEMU_VM_SECTION_END section */
1568 qemu_savevm_state_complete_precopy(f, true, false);
1569 qemu_put_byte(f, QEMU_VM_EOF);
1570 }
1571
qemu_save_device_state(QEMUFile * f)1572 int qemu_save_device_state(QEMUFile *f)
1573 {
1574 SaveStateEntry *se;
1575
1576 if (!migration_in_colo_state()) {
1577 qemu_put_be32(f, QEMU_VM_FILE_MAGIC);
1578 qemu_put_be32(f, QEMU_VM_FILE_VERSION);
1579 }
1580 cpu_synchronize_all_states();
1581
1582 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1583 int ret;
1584
1585 if (se->is_ram) {
1586 continue;
1587 }
1588 if ((!se->ops || !se->ops->save_state) && !se->vmsd) {
1589 continue;
1590 }
1591 if (se->vmsd && !vmstate_save_needed(se->vmsd, se->opaque)) {
1592 continue;
1593 }
1594
1595 save_section_header(f, se, QEMU_VM_SECTION_FULL);
1596
1597 ret = vmstate_save(f, se, NULL);
1598 if (ret) {
1599 return ret;
1600 }
1601
1602 save_section_footer(f, se);
1603 }
1604
1605 qemu_put_byte(f, QEMU_VM_EOF);
1606
1607 return qemu_file_get_error(f);
1608 }
1609
find_se(const char * idstr,uint32_t instance_id)1610 static SaveStateEntry *find_se(const char *idstr, uint32_t instance_id)
1611 {
1612 SaveStateEntry *se;
1613
1614 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1615 if (!strcmp(se->idstr, idstr) &&
1616 (instance_id == se->instance_id ||
1617 instance_id == se->alias_id))
1618 return se;
1619 /* Migrating from an older version? */
1620 if (strstr(se->idstr, idstr) && se->compat) {
1621 if (!strcmp(se->compat->idstr, idstr) &&
1622 (instance_id == se->compat->instance_id ||
1623 instance_id == se->alias_id))
1624 return se;
1625 }
1626 }
1627 return NULL;
1628 }
1629
1630 enum LoadVMExitCodes {
1631 /* Allow a command to quit all layers of nested loadvm loops */
1632 LOADVM_QUIT = 1,
1633 };
1634
1635 /* ------ incoming postcopy messages ------ */
1636 /* 'advise' arrives before any transfers just to tell us that a postcopy
1637 * *might* happen - it might be skipped if precopy transferred everything
1638 * quickly.
1639 */
loadvm_postcopy_handle_advise(MigrationIncomingState * mis,uint16_t len)1640 static int loadvm_postcopy_handle_advise(MigrationIncomingState *mis,
1641 uint16_t len)
1642 {
1643 PostcopyState ps = postcopy_state_set(POSTCOPY_INCOMING_ADVISE);
1644 uint64_t remote_pagesize_summary, local_pagesize_summary, remote_tps;
1645 Error *local_err = NULL;
1646
1647 trace_loadvm_postcopy_handle_advise();
1648 if (ps != POSTCOPY_INCOMING_NONE) {
1649 error_report("CMD_POSTCOPY_ADVISE in wrong postcopy state (%d)", ps);
1650 return -1;
1651 }
1652
1653 switch (len) {
1654 case 0:
1655 if (migrate_postcopy_ram()) {
1656 error_report("RAM postcopy is enabled but have 0 byte advise");
1657 return -EINVAL;
1658 }
1659 return 0;
1660 case 8 + 8:
1661 if (!migrate_postcopy_ram()) {
1662 error_report("RAM postcopy is disabled but have 16 byte advise");
1663 return -EINVAL;
1664 }
1665 break;
1666 default:
1667 error_report("CMD_POSTCOPY_ADVISE invalid length (%d)", len);
1668 return -EINVAL;
1669 }
1670
1671 if (!postcopy_ram_supported_by_host(mis)) {
1672 postcopy_state_set(POSTCOPY_INCOMING_NONE);
1673 return -1;
1674 }
1675
1676 remote_pagesize_summary = qemu_get_be64(mis->from_src_file);
1677 local_pagesize_summary = ram_pagesize_summary();
1678
1679 if (remote_pagesize_summary != local_pagesize_summary) {
1680 /*
1681 * This detects two potential causes of mismatch:
1682 * a) A mismatch in host page sizes
1683 * Some combinations of mismatch are probably possible but it gets
1684 * a bit more complicated. In particular we need to place whole
1685 * host pages on the dest at once, and we need to ensure that we
1686 * handle dirtying to make sure we never end up sending part of
1687 * a hostpage on it's own.
1688 * b) The use of different huge page sizes on source/destination
1689 * a more fine grain test is performed during RAM block migration
1690 * but this test here causes a nice early clear failure, and
1691 * also fails when passed to an older qemu that doesn't
1692 * do huge pages.
1693 */
1694 error_report("Postcopy needs matching RAM page sizes (s=%" PRIx64
1695 " d=%" PRIx64 ")",
1696 remote_pagesize_summary, local_pagesize_summary);
1697 return -1;
1698 }
1699
1700 remote_tps = qemu_get_be64(mis->from_src_file);
1701 if (remote_tps != qemu_target_page_size()) {
1702 /*
1703 * Again, some differences could be dealt with, but for now keep it
1704 * simple.
1705 */
1706 error_report("Postcopy needs matching target page sizes (s=%d d=%zd)",
1707 (int)remote_tps, qemu_target_page_size());
1708 return -1;
1709 }
1710
1711 if (postcopy_notify(POSTCOPY_NOTIFY_INBOUND_ADVISE, &local_err)) {
1712 error_report_err(local_err);
1713 return -1;
1714 }
1715
1716 if (ram_postcopy_incoming_init(mis)) {
1717 return -1;
1718 }
1719
1720 return 0;
1721 }
1722
1723 /* After postcopy we will be told to throw some pages away since they're
1724 * dirty and will have to be demand fetched. Must happen before CPU is
1725 * started.
1726 * There can be 0..many of these messages, each encoding multiple pages.
1727 */
loadvm_postcopy_ram_handle_discard(MigrationIncomingState * mis,uint16_t len)1728 static int loadvm_postcopy_ram_handle_discard(MigrationIncomingState *mis,
1729 uint16_t len)
1730 {
1731 int tmp;
1732 char ramid[256];
1733 PostcopyState ps = postcopy_state_get();
1734
1735 trace_loadvm_postcopy_ram_handle_discard();
1736
1737 switch (ps) {
1738 case POSTCOPY_INCOMING_ADVISE:
1739 /* 1st discard */
1740 tmp = postcopy_ram_prepare_discard(mis);
1741 if (tmp) {
1742 return tmp;
1743 }
1744 break;
1745
1746 case POSTCOPY_INCOMING_DISCARD:
1747 /* Expected state */
1748 break;
1749
1750 default:
1751 error_report("CMD_POSTCOPY_RAM_DISCARD in wrong postcopy state (%d)",
1752 ps);
1753 return -1;
1754 }
1755 /* We're expecting a
1756 * Version (0)
1757 * a RAM ID string (length byte, name, 0 term)
1758 * then at least 1 16 byte chunk
1759 */
1760 if (len < (1 + 1 + 1 + 1 + 2 * 8)) {
1761 error_report("CMD_POSTCOPY_RAM_DISCARD invalid length (%d)", len);
1762 return -1;
1763 }
1764
1765 tmp = qemu_get_byte(mis->from_src_file);
1766 if (tmp != postcopy_ram_discard_version) {
1767 error_report("CMD_POSTCOPY_RAM_DISCARD invalid version (%d)", tmp);
1768 return -1;
1769 }
1770
1771 if (!qemu_get_counted_string(mis->from_src_file, ramid)) {
1772 error_report("CMD_POSTCOPY_RAM_DISCARD Failed to read RAMBlock ID");
1773 return -1;
1774 }
1775 tmp = qemu_get_byte(mis->from_src_file);
1776 if (tmp != 0) {
1777 error_report("CMD_POSTCOPY_RAM_DISCARD missing nil (%d)", tmp);
1778 return -1;
1779 }
1780
1781 len -= 3 + strlen(ramid);
1782 if (len % 16) {
1783 error_report("CMD_POSTCOPY_RAM_DISCARD invalid length (%d)", len);
1784 return -1;
1785 }
1786 trace_loadvm_postcopy_ram_handle_discard_header(ramid, len);
1787 while (len) {
1788 uint64_t start_addr, block_length;
1789 start_addr = qemu_get_be64(mis->from_src_file);
1790 block_length = qemu_get_be64(mis->from_src_file);
1791
1792 len -= 16;
1793 int ret = ram_discard_range(ramid, start_addr, block_length);
1794 if (ret) {
1795 return ret;
1796 }
1797 }
1798 trace_loadvm_postcopy_ram_handle_discard_end();
1799
1800 return 0;
1801 }
1802
1803 /*
1804 * Triggered by a postcopy_listen command; this thread takes over reading
1805 * the input stream, leaving the main thread free to carry on loading the rest
1806 * of the device state (from RAM).
1807 * (TODO:This could do with being in a postcopy file - but there again it's
1808 * just another input loop, not that postcopy specific)
1809 */
postcopy_ram_listen_thread(void * opaque)1810 static void *postcopy_ram_listen_thread(void *opaque)
1811 {
1812 MigrationIncomingState *mis = migration_incoming_get_current();
1813 QEMUFile *f = mis->from_src_file;
1814 int load_res;
1815 MigrationState *migr = migrate_get_current();
1816
1817 object_ref(OBJECT(migr));
1818
1819 migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE,
1820 MIGRATION_STATUS_POSTCOPY_ACTIVE);
1821 qemu_sem_post(&mis->listen_thread_sem);
1822 trace_postcopy_ram_listen_thread_start();
1823
1824 rcu_register_thread();
1825 /*
1826 * Because we're a thread and not a coroutine we can't yield
1827 * in qemu_file, and thus we must be blocking now.
1828 */
1829 qemu_file_set_blocking(f, true);
1830 load_res = qemu_loadvm_state_main(f, mis);
1831
1832 /*
1833 * This is tricky, but, mis->from_src_file can change after it
1834 * returns, when postcopy recovery happened. In the future, we may
1835 * want a wrapper for the QEMUFile handle.
1836 */
1837 f = mis->from_src_file;
1838
1839 /* And non-blocking again so we don't block in any cleanup */
1840 qemu_file_set_blocking(f, false);
1841
1842 trace_postcopy_ram_listen_thread_exit();
1843 if (load_res < 0) {
1844 qemu_file_set_error(f, load_res);
1845 dirty_bitmap_mig_cancel_incoming();
1846 if (postcopy_state_get() == POSTCOPY_INCOMING_RUNNING &&
1847 !migrate_postcopy_ram() && migrate_dirty_bitmaps())
1848 {
1849 error_report("%s: loadvm failed during postcopy: %d. All states "
1850 "are migrated except dirty bitmaps. Some dirty "
1851 "bitmaps may be lost, and present migrated dirty "
1852 "bitmaps are correctly migrated and valid.",
1853 __func__, load_res);
1854 load_res = 0; /* prevent further exit() */
1855 } else {
1856 error_report("%s: loadvm failed: %d", __func__, load_res);
1857 migrate_set_state(&mis->state, MIGRATION_STATUS_POSTCOPY_ACTIVE,
1858 MIGRATION_STATUS_FAILED);
1859 }
1860 }
1861 if (load_res >= 0) {
1862 /*
1863 * This looks good, but it's possible that the device loading in the
1864 * main thread hasn't finished yet, and so we might not be in 'RUN'
1865 * state yet; wait for the end of the main thread.
1866 */
1867 qemu_event_wait(&mis->main_thread_load_event);
1868 }
1869 postcopy_ram_incoming_cleanup(mis);
1870
1871 if (load_res < 0) {
1872 /*
1873 * If something went wrong then we have a bad state so exit;
1874 * depending how far we got it might be possible at this point
1875 * to leave the guest running and fire MCEs for pages that never
1876 * arrived as a desperate recovery step.
1877 */
1878 rcu_unregister_thread();
1879 exit(EXIT_FAILURE);
1880 }
1881
1882 migrate_set_state(&mis->state, MIGRATION_STATUS_POSTCOPY_ACTIVE,
1883 MIGRATION_STATUS_COMPLETED);
1884 /*
1885 * If everything has worked fine, then the main thread has waited
1886 * for us to start, and we're the last use of the mis.
1887 * (If something broke then qemu will have to exit anyway since it's
1888 * got a bad migration state).
1889 */
1890 migration_incoming_state_destroy();
1891 qemu_loadvm_state_cleanup();
1892
1893 rcu_unregister_thread();
1894 mis->have_listen_thread = false;
1895 postcopy_state_set(POSTCOPY_INCOMING_END);
1896
1897 object_unref(OBJECT(migr));
1898
1899 return NULL;
1900 }
1901
1902 /* After this message we must be able to immediately receive postcopy data */
loadvm_postcopy_handle_listen(MigrationIncomingState * mis)1903 static int loadvm_postcopy_handle_listen(MigrationIncomingState *mis)
1904 {
1905 PostcopyState ps = postcopy_state_set(POSTCOPY_INCOMING_LISTENING);
1906 trace_loadvm_postcopy_handle_listen();
1907 Error *local_err = NULL;
1908
1909 if (ps != POSTCOPY_INCOMING_ADVISE && ps != POSTCOPY_INCOMING_DISCARD) {
1910 error_report("CMD_POSTCOPY_LISTEN in wrong postcopy state (%d)", ps);
1911 return -1;
1912 }
1913 if (ps == POSTCOPY_INCOMING_ADVISE) {
1914 /*
1915 * A rare case, we entered listen without having to do any discards,
1916 * so do the setup that's normally done at the time of the 1st discard.
1917 */
1918 if (migrate_postcopy_ram()) {
1919 postcopy_ram_prepare_discard(mis);
1920 }
1921 }
1922
1923 /*
1924 * Sensitise RAM - can now generate requests for blocks that don't exist
1925 * However, at this point the CPU shouldn't be running, and the IO
1926 * shouldn't be doing anything yet so don't actually expect requests
1927 */
1928 if (migrate_postcopy_ram()) {
1929 if (postcopy_ram_incoming_setup(mis)) {
1930 postcopy_ram_incoming_cleanup(mis);
1931 return -1;
1932 }
1933 }
1934
1935 if (postcopy_notify(POSTCOPY_NOTIFY_INBOUND_LISTEN, &local_err)) {
1936 error_report_err(local_err);
1937 return -1;
1938 }
1939
1940 mis->have_listen_thread = true;
1941 /* Start up the listening thread and wait for it to signal ready */
1942 qemu_sem_init(&mis->listen_thread_sem, 0);
1943 qemu_thread_create(&mis->listen_thread, "postcopy/listen",
1944 postcopy_ram_listen_thread, NULL,
1945 QEMU_THREAD_DETACHED);
1946 qemu_sem_wait(&mis->listen_thread_sem);
1947 qemu_sem_destroy(&mis->listen_thread_sem);
1948
1949 return 0;
1950 }
1951
loadvm_postcopy_handle_run_bh(void * opaque)1952 static void loadvm_postcopy_handle_run_bh(void *opaque)
1953 {
1954 Error *local_err = NULL;
1955 MigrationIncomingState *mis = opaque;
1956
1957 /* TODO we should move all of this lot into postcopy_ram.c or a shared code
1958 * in migration.c
1959 */
1960 cpu_synchronize_all_post_init();
1961
1962 qemu_announce_self(&mis->announce_timer, migrate_announce_params());
1963
1964 /* Make sure all file formats flush their mutable metadata.
1965 * If we get an error here, just don't restart the VM yet. */
1966 bdrv_invalidate_cache_all(&local_err);
1967 if (local_err) {
1968 error_report_err(local_err);
1969 local_err = NULL;
1970 autostart = false;
1971 }
1972
1973 trace_loadvm_postcopy_handle_run_cpu_sync();
1974
1975 trace_loadvm_postcopy_handle_run_vmstart();
1976
1977 dirty_bitmap_mig_before_vm_start();
1978
1979 if (autostart) {
1980 /* Hold onto your hats, starting the CPU */
1981 vm_start();
1982 } else {
1983 /* leave it paused and let management decide when to start the CPU */
1984 runstate_set(RUN_STATE_PAUSED);
1985 }
1986
1987 qemu_bh_delete(mis->bh);
1988 }
1989
1990 /* After all discards we can start running and asking for pages */
loadvm_postcopy_handle_run(MigrationIncomingState * mis)1991 static int loadvm_postcopy_handle_run(MigrationIncomingState *mis)
1992 {
1993 PostcopyState ps = postcopy_state_get();
1994
1995 trace_loadvm_postcopy_handle_run();
1996 if (ps != POSTCOPY_INCOMING_LISTENING) {
1997 error_report("CMD_POSTCOPY_RUN in wrong postcopy state (%d)", ps);
1998 return -1;
1999 }
2000
2001 postcopy_state_set(POSTCOPY_INCOMING_RUNNING);
2002 mis->bh = qemu_bh_new(loadvm_postcopy_handle_run_bh, mis);
2003 qemu_bh_schedule(mis->bh);
2004
2005 /* We need to finish reading the stream from the package
2006 * and also stop reading anything more from the stream that loaded the
2007 * package (since it's now being read by the listener thread).
2008 * LOADVM_QUIT will quit all the layers of nested loadvm loops.
2009 */
2010 return LOADVM_QUIT;
2011 }
2012
2013 /* We must be with page_request_mutex held */
postcopy_sync_page_req(gpointer key,gpointer value,gpointer data)2014 static gboolean postcopy_sync_page_req(gpointer key, gpointer value,
2015 gpointer data)
2016 {
2017 MigrationIncomingState *mis = data;
2018 void *host_addr = (void *) key;
2019 ram_addr_t rb_offset;
2020 RAMBlock *rb;
2021 int ret;
2022
2023 rb = qemu_ram_block_from_host(host_addr, true, &rb_offset);
2024 if (!rb) {
2025 /*
2026 * This should _never_ happen. However be nice for a migrating VM to
2027 * not crash/assert. Post an error (note: intended to not use *_once
2028 * because we do want to see all the illegal addresses; and this can
2029 * never be triggered by the guest so we're safe) and move on next.
2030 */
2031 error_report("%s: illegal host addr %p", __func__, host_addr);
2032 /* Try the next entry */
2033 return FALSE;
2034 }
2035
2036 ret = migrate_send_rp_message_req_pages(mis, rb, rb_offset);
2037 if (ret) {
2038 /* Please refer to above comment. */
2039 error_report("%s: send rp message failed for addr %p",
2040 __func__, host_addr);
2041 return FALSE;
2042 }
2043
2044 trace_postcopy_page_req_sync(host_addr);
2045
2046 return FALSE;
2047 }
2048
migrate_send_rp_req_pages_pending(MigrationIncomingState * mis)2049 static void migrate_send_rp_req_pages_pending(MigrationIncomingState *mis)
2050 {
2051 WITH_QEMU_LOCK_GUARD(&mis->page_request_mutex) {
2052 g_tree_foreach(mis->page_requested, postcopy_sync_page_req, mis);
2053 }
2054 }
2055
loadvm_postcopy_handle_resume(MigrationIncomingState * mis)2056 static int loadvm_postcopy_handle_resume(MigrationIncomingState *mis)
2057 {
2058 if (mis->state != MIGRATION_STATUS_POSTCOPY_RECOVER) {
2059 error_report("%s: illegal resume received", __func__);
2060 /* Don't fail the load, only for this. */
2061 return 0;
2062 }
2063
2064 /*
2065 * Reset the last_rb before we resend any page req to source again, since
2066 * the source should have it reset already.
2067 */
2068 mis->last_rb = NULL;
2069
2070 /*
2071 * This means source VM is ready to resume the postcopy migration.
2072 */
2073 migrate_set_state(&mis->state, MIGRATION_STATUS_POSTCOPY_RECOVER,
2074 MIGRATION_STATUS_POSTCOPY_ACTIVE);
2075
2076 trace_loadvm_postcopy_handle_resume();
2077
2078 /* Tell source that "we are ready" */
2079 migrate_send_rp_resume_ack(mis, MIGRATION_RESUME_ACK_VALUE);
2080
2081 /*
2082 * After a postcopy recovery, the source should have lost the postcopy
2083 * queue, or potentially the requested pages could have been lost during
2084 * the network down phase. Let's re-sync with the source VM by re-sending
2085 * all the pending pages that we eagerly need, so these threads won't get
2086 * blocked too long due to the recovery.
2087 *
2088 * Without this procedure, the faulted destination VM threads (waiting for
2089 * page requests right before the postcopy is interrupted) can keep hanging
2090 * until the pages are sent by the source during the background copying of
2091 * pages, or another thread faulted on the same address accidentally.
2092 */
2093 migrate_send_rp_req_pages_pending(mis);
2094
2095 /*
2096 * It's time to switch state and release the fault thread to continue
2097 * service page faults. Note that this should be explicitly after the
2098 * above call to migrate_send_rp_req_pages_pending(). In short:
2099 * migrate_send_rp_message_req_pages() is not thread safe, yet.
2100 */
2101 qemu_sem_post(&mis->postcopy_pause_sem_fault);
2102
2103 return 0;
2104 }
2105
2106 /**
2107 * Immediately following this command is a blob of data containing an embedded
2108 * chunk of migration stream; read it and load it.
2109 *
2110 * @mis: Incoming state
2111 * @length: Length of packaged data to read
2112 *
2113 * Returns: Negative values on error
2114 *
2115 */
loadvm_handle_cmd_packaged(MigrationIncomingState * mis)2116 static int loadvm_handle_cmd_packaged(MigrationIncomingState *mis)
2117 {
2118 int ret;
2119 size_t length;
2120 QIOChannelBuffer *bioc;
2121
2122 length = qemu_get_be32(mis->from_src_file);
2123 trace_loadvm_handle_cmd_packaged(length);
2124
2125 if (length > MAX_VM_CMD_PACKAGED_SIZE) {
2126 error_report("Unreasonably large packaged state: %zu", length);
2127 return -1;
2128 }
2129
2130 bioc = qio_channel_buffer_new(length);
2131 qio_channel_set_name(QIO_CHANNEL(bioc), "migration-loadvm-buffer");
2132 ret = qemu_get_buffer(mis->from_src_file,
2133 bioc->data,
2134 length);
2135 if (ret != length) {
2136 object_unref(OBJECT(bioc));
2137 error_report("CMD_PACKAGED: Buffer receive fail ret=%d length=%zu",
2138 ret, length);
2139 return (ret < 0) ? ret : -EAGAIN;
2140 }
2141 bioc->usage += length;
2142 trace_loadvm_handle_cmd_packaged_received(ret);
2143
2144 QEMUFile *packf = qemu_fopen_channel_input(QIO_CHANNEL(bioc));
2145
2146 ret = qemu_loadvm_state_main(packf, mis);
2147 trace_loadvm_handle_cmd_packaged_main(ret);
2148 qemu_fclose(packf);
2149 object_unref(OBJECT(bioc));
2150
2151 return ret;
2152 }
2153
2154 /*
2155 * Handle request that source requests for recved_bitmap on
2156 * destination. Payload format:
2157 *
2158 * len (1 byte) + ramblock_name (<255 bytes)
2159 */
loadvm_handle_recv_bitmap(MigrationIncomingState * mis,uint16_t len)2160 static int loadvm_handle_recv_bitmap(MigrationIncomingState *mis,
2161 uint16_t len)
2162 {
2163 QEMUFile *file = mis->from_src_file;
2164 RAMBlock *rb;
2165 char block_name[256];
2166 size_t cnt;
2167
2168 cnt = qemu_get_counted_string(file, block_name);
2169 if (!cnt) {
2170 error_report("%s: failed to read block name", __func__);
2171 return -EINVAL;
2172 }
2173
2174 /* Validate before using the data */
2175 if (qemu_file_get_error(file)) {
2176 return qemu_file_get_error(file);
2177 }
2178
2179 if (len != cnt + 1) {
2180 error_report("%s: invalid payload length (%d)", __func__, len);
2181 return -EINVAL;
2182 }
2183
2184 rb = qemu_ram_block_by_name(block_name);
2185 if (!rb) {
2186 error_report("%s: block '%s' not found", __func__, block_name);
2187 return -EINVAL;
2188 }
2189
2190 migrate_send_rp_recv_bitmap(mis, block_name);
2191
2192 trace_loadvm_handle_recv_bitmap(block_name);
2193
2194 return 0;
2195 }
2196
loadvm_process_enable_colo(MigrationIncomingState * mis)2197 static int loadvm_process_enable_colo(MigrationIncomingState *mis)
2198 {
2199 int ret = migration_incoming_enable_colo();
2200
2201 if (!ret) {
2202 ret = colo_init_ram_cache();
2203 if (ret) {
2204 migration_incoming_disable_colo();
2205 }
2206 }
2207 return ret;
2208 }
2209
2210 /*
2211 * Process an incoming 'QEMU_VM_COMMAND'
2212 * 0 just a normal return
2213 * LOADVM_QUIT All good, but exit the loop
2214 * <0 Error
2215 */
loadvm_process_command(QEMUFile * f)2216 static int loadvm_process_command(QEMUFile *f)
2217 {
2218 MigrationIncomingState *mis = migration_incoming_get_current();
2219 uint16_t cmd;
2220 uint16_t len;
2221 uint32_t tmp32;
2222
2223 cmd = qemu_get_be16(f);
2224 len = qemu_get_be16(f);
2225
2226 /* Check validity before continue processing of cmds */
2227 if (qemu_file_get_error(f)) {
2228 return qemu_file_get_error(f);
2229 }
2230
2231 trace_loadvm_process_command(cmd, len);
2232 if (cmd >= MIG_CMD_MAX || cmd == MIG_CMD_INVALID) {
2233 error_report("MIG_CMD 0x%x unknown (len 0x%x)", cmd, len);
2234 return -EINVAL;
2235 }
2236
2237 if (mig_cmd_args[cmd].len != -1 && mig_cmd_args[cmd].len != len) {
2238 error_report("%s received with bad length - expecting %zu, got %d",
2239 mig_cmd_args[cmd].name,
2240 (size_t)mig_cmd_args[cmd].len, len);
2241 return -ERANGE;
2242 }
2243
2244 switch (cmd) {
2245 case MIG_CMD_OPEN_RETURN_PATH:
2246 if (mis->to_src_file) {
2247 error_report("CMD_OPEN_RETURN_PATH called when RP already open");
2248 /* Not really a problem, so don't give up */
2249 return 0;
2250 }
2251 mis->to_src_file = qemu_file_get_return_path(f);
2252 if (!mis->to_src_file) {
2253 error_report("CMD_OPEN_RETURN_PATH failed");
2254 return -1;
2255 }
2256 break;
2257
2258 case MIG_CMD_PING:
2259 tmp32 = qemu_get_be32(f);
2260 trace_loadvm_process_command_ping(tmp32);
2261 if (!mis->to_src_file) {
2262 error_report("CMD_PING (0x%x) received with no return path",
2263 tmp32);
2264 return -1;
2265 }
2266 migrate_send_rp_pong(mis, tmp32);
2267 break;
2268
2269 case MIG_CMD_PACKAGED:
2270 return loadvm_handle_cmd_packaged(mis);
2271
2272 case MIG_CMD_POSTCOPY_ADVISE:
2273 return loadvm_postcopy_handle_advise(mis, len);
2274
2275 case MIG_CMD_POSTCOPY_LISTEN:
2276 return loadvm_postcopy_handle_listen(mis);
2277
2278 case MIG_CMD_POSTCOPY_RUN:
2279 return loadvm_postcopy_handle_run(mis);
2280
2281 case MIG_CMD_POSTCOPY_RAM_DISCARD:
2282 return loadvm_postcopy_ram_handle_discard(mis, len);
2283
2284 case MIG_CMD_POSTCOPY_RESUME:
2285 return loadvm_postcopy_handle_resume(mis);
2286
2287 case MIG_CMD_RECV_BITMAP:
2288 return loadvm_handle_recv_bitmap(mis, len);
2289
2290 case MIG_CMD_ENABLE_COLO:
2291 return loadvm_process_enable_colo(mis);
2292 }
2293
2294 return 0;
2295 }
2296
2297 /*
2298 * Read a footer off the wire and check that it matches the expected section
2299 *
2300 * Returns: true if the footer was good
2301 * false if there is a problem (and calls error_report to say why)
2302 */
check_section_footer(QEMUFile * f,SaveStateEntry * se)2303 static bool check_section_footer(QEMUFile *f, SaveStateEntry *se)
2304 {
2305 int ret;
2306 uint8_t read_mark;
2307 uint32_t read_section_id;
2308
2309 if (!migrate_get_current()->send_section_footer) {
2310 /* No footer to check */
2311 return true;
2312 }
2313
2314 read_mark = qemu_get_byte(f);
2315
2316 ret = qemu_file_get_error(f);
2317 if (ret) {
2318 error_report("%s: Read section footer failed: %d",
2319 __func__, ret);
2320 return false;
2321 }
2322
2323 if (read_mark != QEMU_VM_SECTION_FOOTER) {
2324 error_report("Missing section footer for %s", se->idstr);
2325 return false;
2326 }
2327
2328 read_section_id = qemu_get_be32(f);
2329 if (read_section_id != se->load_section_id) {
2330 error_report("Mismatched section id in footer for %s -"
2331 " read 0x%x expected 0x%x",
2332 se->idstr, read_section_id, se->load_section_id);
2333 return false;
2334 }
2335
2336 /* All good */
2337 return true;
2338 }
2339
2340 static int
qemu_loadvm_section_start_full(QEMUFile * f,MigrationIncomingState * mis)2341 qemu_loadvm_section_start_full(QEMUFile *f, MigrationIncomingState *mis)
2342 {
2343 uint32_t instance_id, version_id, section_id;
2344 SaveStateEntry *se;
2345 char idstr[256];
2346 int ret;
2347
2348 /* Read section start */
2349 section_id = qemu_get_be32(f);
2350 if (!qemu_get_counted_string(f, idstr)) {
2351 error_report("Unable to read ID string for section %u",
2352 section_id);
2353 return -EINVAL;
2354 }
2355 instance_id = qemu_get_be32(f);
2356 version_id = qemu_get_be32(f);
2357
2358 ret = qemu_file_get_error(f);
2359 if (ret) {
2360 error_report("%s: Failed to read instance/version ID: %d",
2361 __func__, ret);
2362 return ret;
2363 }
2364
2365 trace_qemu_loadvm_state_section_startfull(section_id, idstr,
2366 instance_id, version_id);
2367 /* Find savevm section */
2368 se = find_se(idstr, instance_id);
2369 if (se == NULL) {
2370 error_report("Unknown savevm section or instance '%s' %"PRIu32". "
2371 "Make sure that your current VM setup matches your "
2372 "saved VM setup, including any hotplugged devices",
2373 idstr, instance_id);
2374 return -EINVAL;
2375 }
2376
2377 /* Validate version */
2378 if (version_id > se->version_id) {
2379 error_report("savevm: unsupported version %d for '%s' v%d",
2380 version_id, idstr, se->version_id);
2381 return -EINVAL;
2382 }
2383 se->load_version_id = version_id;
2384 se->load_section_id = section_id;
2385
2386 /* Validate if it is a device's state */
2387 if (xen_enabled() && se->is_ram) {
2388 error_report("loadvm: %s RAM loading not allowed on Xen", idstr);
2389 return -EINVAL;
2390 }
2391
2392 ret = vmstate_load(f, se);
2393 if (ret < 0) {
2394 error_report("error while loading state for instance 0x%"PRIx32" of"
2395 " device '%s'", instance_id, idstr);
2396 return ret;
2397 }
2398 if (!check_section_footer(f, se)) {
2399 return -EINVAL;
2400 }
2401
2402 return 0;
2403 }
2404
2405 static int
qemu_loadvm_section_part_end(QEMUFile * f,MigrationIncomingState * mis)2406 qemu_loadvm_section_part_end(QEMUFile *f, MigrationIncomingState *mis)
2407 {
2408 uint32_t section_id;
2409 SaveStateEntry *se;
2410 int ret;
2411
2412 section_id = qemu_get_be32(f);
2413
2414 ret = qemu_file_get_error(f);
2415 if (ret) {
2416 error_report("%s: Failed to read section ID: %d",
2417 __func__, ret);
2418 return ret;
2419 }
2420
2421 trace_qemu_loadvm_state_section_partend(section_id);
2422 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
2423 if (se->load_section_id == section_id) {
2424 break;
2425 }
2426 }
2427 if (se == NULL) {
2428 error_report("Unknown savevm section %d", section_id);
2429 return -EINVAL;
2430 }
2431
2432 ret = vmstate_load(f, se);
2433 if (ret < 0) {
2434 error_report("error while loading state section id %d(%s)",
2435 section_id, se->idstr);
2436 return ret;
2437 }
2438 if (!check_section_footer(f, se)) {
2439 return -EINVAL;
2440 }
2441
2442 return 0;
2443 }
2444
qemu_loadvm_state_header(QEMUFile * f)2445 static int qemu_loadvm_state_header(QEMUFile *f)
2446 {
2447 unsigned int v;
2448 int ret;
2449
2450 v = qemu_get_be32(f);
2451 if (v != QEMU_VM_FILE_MAGIC) {
2452 error_report("Not a migration stream");
2453 return -EINVAL;
2454 }
2455
2456 v = qemu_get_be32(f);
2457 if (v == QEMU_VM_FILE_VERSION_COMPAT) {
2458 error_report("SaveVM v2 format is obsolete and don't work anymore");
2459 return -ENOTSUP;
2460 }
2461 if (v != QEMU_VM_FILE_VERSION) {
2462 error_report("Unsupported migration stream version");
2463 return -ENOTSUP;
2464 }
2465
2466 if (migrate_get_current()->send_configuration) {
2467 if (qemu_get_byte(f) != QEMU_VM_CONFIGURATION) {
2468 error_report("Configuration section missing");
2469 qemu_loadvm_state_cleanup();
2470 return -EINVAL;
2471 }
2472 ret = vmstate_load_state(f, &vmstate_configuration, &savevm_state, 0);
2473
2474 if (ret) {
2475 qemu_loadvm_state_cleanup();
2476 return ret;
2477 }
2478 }
2479 return 0;
2480 }
2481
qemu_loadvm_state_setup(QEMUFile * f)2482 static int qemu_loadvm_state_setup(QEMUFile *f)
2483 {
2484 SaveStateEntry *se;
2485 int ret;
2486
2487 trace_loadvm_state_setup();
2488 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
2489 if (!se->ops || !se->ops->load_setup) {
2490 continue;
2491 }
2492 if (se->ops->is_active) {
2493 if (!se->ops->is_active(se->opaque)) {
2494 continue;
2495 }
2496 }
2497
2498 ret = se->ops->load_setup(f, se->opaque);
2499 if (ret < 0) {
2500 qemu_file_set_error(f, ret);
2501 error_report("Load state of device %s failed", se->idstr);
2502 return ret;
2503 }
2504 }
2505 return 0;
2506 }
2507
qemu_loadvm_state_cleanup(void)2508 void qemu_loadvm_state_cleanup(void)
2509 {
2510 SaveStateEntry *se;
2511
2512 trace_loadvm_state_cleanup();
2513 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
2514 if (se->ops && se->ops->load_cleanup) {
2515 se->ops->load_cleanup(se->opaque);
2516 }
2517 }
2518 }
2519
2520 /* Return true if we should continue the migration, or false. */
postcopy_pause_incoming(MigrationIncomingState * mis)2521 static bool postcopy_pause_incoming(MigrationIncomingState *mis)
2522 {
2523 trace_postcopy_pause_incoming();
2524
2525 assert(migrate_postcopy_ram());
2526
2527 /* Clear the triggered bit to allow one recovery */
2528 mis->postcopy_recover_triggered = false;
2529
2530 assert(mis->from_src_file);
2531 qemu_file_shutdown(mis->from_src_file);
2532 qemu_fclose(mis->from_src_file);
2533 mis->from_src_file = NULL;
2534
2535 assert(mis->to_src_file);
2536 qemu_file_shutdown(mis->to_src_file);
2537 qemu_mutex_lock(&mis->rp_mutex);
2538 qemu_fclose(mis->to_src_file);
2539 mis->to_src_file = NULL;
2540 qemu_mutex_unlock(&mis->rp_mutex);
2541
2542 migrate_set_state(&mis->state, MIGRATION_STATUS_POSTCOPY_ACTIVE,
2543 MIGRATION_STATUS_POSTCOPY_PAUSED);
2544
2545 /* Notify the fault thread for the invalidated file handle */
2546 postcopy_fault_thread_notify(mis);
2547
2548 error_report("Detected IO failure for postcopy. "
2549 "Migration paused.");
2550
2551 while (mis->state == MIGRATION_STATUS_POSTCOPY_PAUSED) {
2552 qemu_sem_wait(&mis->postcopy_pause_sem_dst);
2553 }
2554
2555 trace_postcopy_pause_incoming_continued();
2556
2557 return true;
2558 }
2559
qemu_loadvm_state_main(QEMUFile * f,MigrationIncomingState * mis)2560 int qemu_loadvm_state_main(QEMUFile *f, MigrationIncomingState *mis)
2561 {
2562 uint8_t section_type;
2563 int ret = 0;
2564
2565 retry:
2566 while (true) {
2567 section_type = qemu_get_byte(f);
2568
2569 if (qemu_file_get_error(f)) {
2570 ret = qemu_file_get_error(f);
2571 break;
2572 }
2573
2574 trace_qemu_loadvm_state_section(section_type);
2575 switch (section_type) {
2576 case QEMU_VM_SECTION_START:
2577 case QEMU_VM_SECTION_FULL:
2578 ret = qemu_loadvm_section_start_full(f, mis);
2579 if (ret < 0) {
2580 goto out;
2581 }
2582 break;
2583 case QEMU_VM_SECTION_PART:
2584 case QEMU_VM_SECTION_END:
2585 ret = qemu_loadvm_section_part_end(f, mis);
2586 if (ret < 0) {
2587 goto out;
2588 }
2589 break;
2590 case QEMU_VM_COMMAND:
2591 ret = loadvm_process_command(f);
2592 trace_qemu_loadvm_state_section_command(ret);
2593 if ((ret < 0) || (ret == LOADVM_QUIT)) {
2594 goto out;
2595 }
2596 break;
2597 case QEMU_VM_EOF:
2598 /* This is the end of migration */
2599 goto out;
2600 default:
2601 error_report("Unknown savevm section type %d", section_type);
2602 ret = -EINVAL;
2603 goto out;
2604 }
2605 }
2606
2607 out:
2608 if (ret < 0) {
2609 qemu_file_set_error(f, ret);
2610
2611 /* Cancel bitmaps incoming regardless of recovery */
2612 dirty_bitmap_mig_cancel_incoming();
2613
2614 /*
2615 * If we are during an active postcopy, then we pause instead
2616 * of bail out to at least keep the VM's dirty data. Note
2617 * that POSTCOPY_INCOMING_LISTENING stage is still not enough,
2618 * during which we're still receiving device states and we
2619 * still haven't yet started the VM on destination.
2620 *
2621 * Only RAM postcopy supports recovery. Still, if RAM postcopy is
2622 * enabled, canceled bitmaps postcopy will not affect RAM postcopy
2623 * recovering.
2624 */
2625 if (postcopy_state_get() == POSTCOPY_INCOMING_RUNNING &&
2626 migrate_postcopy_ram() && postcopy_pause_incoming(mis)) {
2627 /* Reset f to point to the newly created channel */
2628 f = mis->from_src_file;
2629 goto retry;
2630 }
2631 }
2632 return ret;
2633 }
2634
qemu_loadvm_state(QEMUFile * f)2635 int qemu_loadvm_state(QEMUFile *f)
2636 {
2637 MigrationIncomingState *mis = migration_incoming_get_current();
2638 Error *local_err = NULL;
2639 int ret;
2640
2641 if (qemu_savevm_state_blocked(&local_err)) {
2642 error_report_err(local_err);
2643 return -EINVAL;
2644 }
2645
2646 ret = qemu_loadvm_state_header(f);
2647 if (ret) {
2648 return ret;
2649 }
2650
2651 if (qemu_loadvm_state_setup(f) != 0) {
2652 return -EINVAL;
2653 }
2654
2655 cpu_synchronize_all_pre_loadvm();
2656
2657 ret = qemu_loadvm_state_main(f, mis);
2658 qemu_event_set(&mis->main_thread_load_event);
2659
2660 trace_qemu_loadvm_state_post_main(ret);
2661
2662 if (mis->have_listen_thread) {
2663 /* Listen thread still going, can't clean up yet */
2664 return ret;
2665 }
2666
2667 if (ret == 0) {
2668 ret = qemu_file_get_error(f);
2669 }
2670
2671 /*
2672 * Try to read in the VMDESC section as well, so that dumping tools that
2673 * intercept our migration stream have the chance to see it.
2674 */
2675
2676 /* We've got to be careful; if we don't read the data and just shut the fd
2677 * then the sender can error if we close while it's still sending.
2678 * We also mustn't read data that isn't there; some transports (RDMA)
2679 * will stall waiting for that data when the source has already closed.
2680 */
2681 if (ret == 0 && should_send_vmdesc()) {
2682 uint8_t *buf;
2683 uint32_t size;
2684 uint8_t section_type = qemu_get_byte(f);
2685
2686 if (section_type != QEMU_VM_VMDESCRIPTION) {
2687 error_report("Expected vmdescription section, but got %d",
2688 section_type);
2689 /*
2690 * It doesn't seem worth failing at this point since
2691 * we apparently have an otherwise valid VM state
2692 */
2693 } else {
2694 buf = g_malloc(0x1000);
2695 size = qemu_get_be32(f);
2696
2697 while (size > 0) {
2698 uint32_t read_chunk = MIN(size, 0x1000);
2699 qemu_get_buffer(f, buf, read_chunk);
2700 size -= read_chunk;
2701 }
2702 g_free(buf);
2703 }
2704 }
2705
2706 qemu_loadvm_state_cleanup();
2707 cpu_synchronize_all_post_init();
2708
2709 return ret;
2710 }
2711
qemu_load_device_state(QEMUFile * f)2712 int qemu_load_device_state(QEMUFile *f)
2713 {
2714 MigrationIncomingState *mis = migration_incoming_get_current();
2715 int ret;
2716
2717 /* Load QEMU_VM_SECTION_FULL section */
2718 ret = qemu_loadvm_state_main(f, mis);
2719 if (ret < 0) {
2720 error_report("Failed to load device state: %d", ret);
2721 return ret;
2722 }
2723
2724 cpu_synchronize_all_post_init();
2725 return 0;
2726 }
2727
save_snapshot(const char * name,Error ** errp)2728 int save_snapshot(const char *name, Error **errp)
2729 {
2730 BlockDriverState *bs, *bs1;
2731 QEMUSnapshotInfo sn1, *sn = &sn1, old_sn1, *old_sn = &old_sn1;
2732 int ret = -1, ret2;
2733 QEMUFile *f;
2734 int saved_vm_running;
2735 uint64_t vm_state_size;
2736 qemu_timeval tv;
2737 struct tm tm;
2738 AioContext *aio_context;
2739
2740 if (migration_is_blocked(errp)) {
2741 return ret;
2742 }
2743
2744 if (!replay_can_snapshot()) {
2745 error_setg(errp, "Record/replay does not allow making snapshot "
2746 "right now. Try once more later.");
2747 return ret;
2748 }
2749
2750 if (!bdrv_all_can_snapshot(&bs)) {
2751 error_setg(errp, "Device '%s' is writable but does not support "
2752 "snapshots", bdrv_get_device_or_node_name(bs));
2753 return ret;
2754 }
2755
2756 /* Delete old snapshots of the same name */
2757 if (name) {
2758 ret = bdrv_all_delete_snapshot(name, &bs1, errp);
2759 if (ret < 0) {
2760 error_prepend(errp, "Error while deleting snapshot on device "
2761 "'%s': ", bdrv_get_device_or_node_name(bs1));
2762 return ret;
2763 }
2764 }
2765
2766 bs = bdrv_all_find_vmstate_bs();
2767 if (bs == NULL) {
2768 error_setg(errp, "No block device can accept snapshots");
2769 return ret;
2770 }
2771 aio_context = bdrv_get_aio_context(bs);
2772
2773 saved_vm_running = runstate_is_running();
2774
2775 ret = global_state_store();
2776 if (ret) {
2777 error_setg(errp, "Error saving global state");
2778 return ret;
2779 }
2780 vm_stop(RUN_STATE_SAVE_VM);
2781
2782 bdrv_drain_all_begin();
2783
2784 aio_context_acquire(aio_context);
2785
2786 memset(sn, 0, sizeof(*sn));
2787
2788 /* fill auxiliary fields */
2789 qemu_gettimeofday(&tv);
2790 sn->date_sec = tv.tv_sec;
2791 sn->date_nsec = tv.tv_usec * 1000;
2792 sn->vm_clock_nsec = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL);
2793 if (replay_mode != REPLAY_MODE_NONE) {
2794 sn->icount = replay_get_current_icount();
2795 } else {
2796 sn->icount = -1ULL;
2797 }
2798
2799 if (name) {
2800 ret = bdrv_snapshot_find(bs, old_sn, name);
2801 if (ret >= 0) {
2802 pstrcpy(sn->name, sizeof(sn->name), old_sn->name);
2803 pstrcpy(sn->id_str, sizeof(sn->id_str), old_sn->id_str);
2804 } else {
2805 pstrcpy(sn->name, sizeof(sn->name), name);
2806 }
2807 } else {
2808 /* cast below needed for OpenBSD where tv_sec is still 'long' */
2809 localtime_r((const time_t *)&tv.tv_sec, &tm);
2810 strftime(sn->name, sizeof(sn->name), "vm-%Y%m%d%H%M%S", &tm);
2811 }
2812
2813 /* save the VM state */
2814 f = qemu_fopen_bdrv(bs, 1);
2815 if (!f) {
2816 error_setg(errp, "Could not open VM state file");
2817 goto the_end;
2818 }
2819 ret = qemu_savevm_state(f, errp);
2820 vm_state_size = qemu_ftell(f);
2821 ret2 = qemu_fclose(f);
2822 if (ret < 0) {
2823 goto the_end;
2824 }
2825 if (ret2 < 0) {
2826 ret = ret2;
2827 goto the_end;
2828 }
2829
2830 /* The bdrv_all_create_snapshot() call that follows acquires the AioContext
2831 * for itself. BDRV_POLL_WHILE() does not support nested locking because
2832 * it only releases the lock once. Therefore synchronous I/O will deadlock
2833 * unless we release the AioContext before bdrv_all_create_snapshot().
2834 */
2835 aio_context_release(aio_context);
2836 aio_context = NULL;
2837
2838 ret = bdrv_all_create_snapshot(sn, bs, vm_state_size, &bs);
2839 if (ret < 0) {
2840 error_setg(errp, "Error while creating snapshot on '%s'",
2841 bdrv_get_device_or_node_name(bs));
2842 goto the_end;
2843 }
2844
2845 ret = 0;
2846
2847 the_end:
2848 if (aio_context) {
2849 aio_context_release(aio_context);
2850 }
2851
2852 bdrv_drain_all_end();
2853
2854 if (saved_vm_running) {
2855 vm_start();
2856 }
2857 return ret;
2858 }
2859
qmp_xen_save_devices_state(const char * filename,bool has_live,bool live,Error ** errp)2860 void qmp_xen_save_devices_state(const char *filename, bool has_live, bool live,
2861 Error **errp)
2862 {
2863 QEMUFile *f;
2864 QIOChannelFile *ioc;
2865 int saved_vm_running;
2866 int ret;
2867
2868 if (!has_live) {
2869 /* live default to true so old version of Xen tool stack can have a
2870 * successful live migration */
2871 live = true;
2872 }
2873
2874 saved_vm_running = runstate_is_running();
2875 vm_stop(RUN_STATE_SAVE_VM);
2876 global_state_store_running();
2877
2878 ioc = qio_channel_file_new_path(filename, O_WRONLY | O_CREAT | O_TRUNC,
2879 0660, errp);
2880 if (!ioc) {
2881 goto the_end;
2882 }
2883 qio_channel_set_name(QIO_CHANNEL(ioc), "migration-xen-save-state");
2884 f = qemu_fopen_channel_output(QIO_CHANNEL(ioc));
2885 object_unref(OBJECT(ioc));
2886 ret = qemu_save_device_state(f);
2887 if (ret < 0 || qemu_fclose(f) < 0) {
2888 error_setg(errp, QERR_IO_ERROR);
2889 } else {
2890 /* libxl calls the QMP command "stop" before calling
2891 * "xen-save-devices-state" and in case of migration failure, libxl
2892 * would call "cont".
2893 * So call bdrv_inactivate_all (release locks) here to let the other
2894 * side of the migration take control of the images.
2895 */
2896 if (live && !saved_vm_running) {
2897 ret = bdrv_inactivate_all();
2898 if (ret) {
2899 error_setg(errp, "%s: bdrv_inactivate_all() failed (%d)",
2900 __func__, ret);
2901 }
2902 }
2903 }
2904
2905 the_end:
2906 if (saved_vm_running) {
2907 vm_start();
2908 }
2909 }
2910
qmp_xen_load_devices_state(const char * filename,Error ** errp)2911 void qmp_xen_load_devices_state(const char *filename, Error **errp)
2912 {
2913 QEMUFile *f;
2914 QIOChannelFile *ioc;
2915 int ret;
2916
2917 /* Guest must be paused before loading the device state; the RAM state
2918 * will already have been loaded by xc
2919 */
2920 if (runstate_is_running()) {
2921 error_setg(errp, "Cannot update device state while vm is running");
2922 return;
2923 }
2924 vm_stop(RUN_STATE_RESTORE_VM);
2925
2926 ioc = qio_channel_file_new_path(filename, O_RDONLY | O_BINARY, 0, errp);
2927 if (!ioc) {
2928 return;
2929 }
2930 qio_channel_set_name(QIO_CHANNEL(ioc), "migration-xen-load-state");
2931 f = qemu_fopen_channel_input(QIO_CHANNEL(ioc));
2932 object_unref(OBJECT(ioc));
2933
2934 ret = qemu_loadvm_state(f);
2935 qemu_fclose(f);
2936 if (ret < 0) {
2937 error_setg(errp, QERR_IO_ERROR);
2938 }
2939 migration_incoming_state_destroy();
2940 }
2941
load_snapshot(const char * name,Error ** errp)2942 int load_snapshot(const char *name, Error **errp)
2943 {
2944 BlockDriverState *bs, *bs_vm_state;
2945 QEMUSnapshotInfo sn;
2946 QEMUFile *f;
2947 int ret;
2948 AioContext *aio_context;
2949 MigrationIncomingState *mis = migration_incoming_get_current();
2950
2951 if (!bdrv_all_can_snapshot(&bs)) {
2952 error_setg(errp,
2953 "Device '%s' is writable but does not support snapshots",
2954 bdrv_get_device_or_node_name(bs));
2955 return -ENOTSUP;
2956 }
2957 ret = bdrv_all_find_snapshot(name, &bs);
2958 if (ret < 0) {
2959 error_setg(errp,
2960 "Device '%s' does not have the requested snapshot '%s'",
2961 bdrv_get_device_or_node_name(bs), name);
2962 return ret;
2963 }
2964
2965 bs_vm_state = bdrv_all_find_vmstate_bs();
2966 if (!bs_vm_state) {
2967 error_setg(errp, "No block device supports snapshots");
2968 return -ENOTSUP;
2969 }
2970 aio_context = bdrv_get_aio_context(bs_vm_state);
2971
2972 /* Don't even try to load empty VM states */
2973 aio_context_acquire(aio_context);
2974 ret = bdrv_snapshot_find(bs_vm_state, &sn, name);
2975 aio_context_release(aio_context);
2976 if (ret < 0) {
2977 return ret;
2978 } else if (sn.vm_state_size == 0) {
2979 error_setg(errp, "This is a disk-only snapshot. Revert to it "
2980 " offline using qemu-img");
2981 return -EINVAL;
2982 }
2983
2984 /*
2985 * Flush the record/replay queue. Now the VM state is going
2986 * to change. Therefore we don't need to preserve its consistency
2987 */
2988 replay_flush_events();
2989
2990 /* Flush all IO requests so they don't interfere with the new state. */
2991 bdrv_drain_all_begin();
2992
2993 ret = bdrv_all_goto_snapshot(name, &bs, errp);
2994 if (ret < 0) {
2995 error_prepend(errp, "Could not load snapshot '%s' on '%s': ",
2996 name, bdrv_get_device_or_node_name(bs));
2997 goto err_drain;
2998 }
2999
3000 /* restore the VM state */
3001 f = qemu_fopen_bdrv(bs_vm_state, 0);
3002 if (!f) {
3003 error_setg(errp, "Could not open VM state file");
3004 ret = -EINVAL;
3005 goto err_drain;
3006 }
3007
3008 qemu_system_reset(SHUTDOWN_CAUSE_NONE);
3009 mis->from_src_file = f;
3010
3011 aio_context_acquire(aio_context);
3012 ret = qemu_loadvm_state(f);
3013 migration_incoming_state_destroy();
3014 aio_context_release(aio_context);
3015
3016 bdrv_drain_all_end();
3017
3018 if (ret < 0) {
3019 error_setg(errp, "Error %d while loading VM state", ret);
3020 return ret;
3021 }
3022
3023 return 0;
3024
3025 err_drain:
3026 bdrv_drain_all_end();
3027 return ret;
3028 }
3029
vmstate_register_ram(MemoryRegion * mr,DeviceState * dev)3030 void vmstate_register_ram(MemoryRegion *mr, DeviceState *dev)
3031 {
3032 qemu_ram_set_idstr(mr->ram_block,
3033 memory_region_name(mr), dev);
3034 qemu_ram_set_migratable(mr->ram_block);
3035 }
3036
vmstate_unregister_ram(MemoryRegion * mr,DeviceState * dev)3037 void vmstate_unregister_ram(MemoryRegion *mr, DeviceState *dev)
3038 {
3039 qemu_ram_unset_idstr(mr->ram_block);
3040 qemu_ram_unset_migratable(mr->ram_block);
3041 }
3042
vmstate_register_ram_global(MemoryRegion * mr)3043 void vmstate_register_ram_global(MemoryRegion *mr)
3044 {
3045 vmstate_register_ram(mr, NULL);
3046 }
3047
vmstate_check_only_migratable(const VMStateDescription * vmsd)3048 bool vmstate_check_only_migratable(const VMStateDescription *vmsd)
3049 {
3050 /* check needed if --only-migratable is specified */
3051 if (!only_migratable) {
3052 return true;
3053 }
3054
3055 return !(vmsd && vmsd->unmigratable);
3056 }
3057