xref: /qemu/migration/savevm.c (revision 6b8f40c61bbfef1abe77eeb9c716ec642927c12c)
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-stats.h"
35 #include "migration/vmstate.h"
36 #include "migration/misc.h"
37 #include "migration/register.h"
38 #include "migration/global_state.h"
39 #include "migration/channel-block.h"
40 #include "ram.h"
41 #include "qemu-file.h"
42 #include "savevm.h"
43 #include "postcopy-ram.h"
44 #include "qapi/error.h"
45 #include "qapi/qapi-commands-migration.h"
46 #include "qapi/clone-visitor.h"
47 #include "qapi/qapi-builtin-visit.h"
48 #include "qemu/error-report.h"
49 #include "system/cpus.h"
50 #include "exec/memory.h"
51 #include "exec/target_page.h"
52 #include "trace.h"
53 #include "qemu/iov.h"
54 #include "qemu/job.h"
55 #include "qemu/main-loop.h"
56 #include "block/snapshot.h"
57 #include "qemu/cutils.h"
58 #include "io/channel-buffer.h"
59 #include "io/channel-file.h"
60 #include "system/replay.h"
61 #include "system/runstate.h"
62 #include "system/system.h"
63 #include "system/xen.h"
64 #include "migration/colo.h"
65 #include "qemu/bitmap.h"
66 #include "net/announce.h"
67 #include "qemu/yank.h"
68 #include "yank_functions.h"
69 #include "system/qtest.h"
70 #include "options.h"
71 
72 const unsigned int postcopy_ram_discard_version;
73 
74 /* Subcommands for QEMU_VM_COMMAND */
75 enum qemu_vm_cmd {
76     MIG_CMD_INVALID = 0,   /* Must be 0 */
77     MIG_CMD_OPEN_RETURN_PATH,  /* Tell the dest to open the Return path */
78     MIG_CMD_PING,              /* Request a PONG on the RP */
79 
80     MIG_CMD_POSTCOPY_ADVISE,       /* Prior to any page transfers, just
81                                       warn we might want to do PC */
82     MIG_CMD_POSTCOPY_LISTEN,       /* Start listening for incoming
83                                       pages as it's running. */
84     MIG_CMD_POSTCOPY_RUN,          /* Start execution */
85 
86     MIG_CMD_POSTCOPY_RAM_DISCARD,  /* A list of pages to discard that
87                                       were previously sent during
88                                       precopy but are dirty. */
89     MIG_CMD_PACKAGED,          /* Send a wrapped stream within this stream */
90     MIG_CMD_ENABLE_COLO,       /* Enable COLO */
91     MIG_CMD_POSTCOPY_RESUME,   /* resume postcopy on dest */
92     MIG_CMD_RECV_BITMAP,       /* Request for recved bitmap on dst */
93     MIG_CMD_MAX
94 };
95 
96 #define MAX_VM_CMD_PACKAGED_SIZE UINT32_MAX
97 static struct mig_cmd_args {
98     ssize_t     len; /* -1 = variable */
99     const char *name;
100 } mig_cmd_args[] = {
101     [MIG_CMD_INVALID]          = { .len = -1, .name = "INVALID" },
102     [MIG_CMD_OPEN_RETURN_PATH] = { .len =  0, .name = "OPEN_RETURN_PATH" },
103     [MIG_CMD_PING]             = { .len = sizeof(uint32_t), .name = "PING" },
104     [MIG_CMD_POSTCOPY_ADVISE]  = { .len = -1, .name = "POSTCOPY_ADVISE" },
105     [MIG_CMD_POSTCOPY_LISTEN]  = { .len =  0, .name = "POSTCOPY_LISTEN" },
106     [MIG_CMD_POSTCOPY_RUN]     = { .len =  0, .name = "POSTCOPY_RUN" },
107     [MIG_CMD_POSTCOPY_RAM_DISCARD] = {
108                                    .len = -1, .name = "POSTCOPY_RAM_DISCARD" },
109     [MIG_CMD_POSTCOPY_RESUME]  = { .len =  0, .name = "POSTCOPY_RESUME" },
110     [MIG_CMD_PACKAGED]         = { .len =  4, .name = "PACKAGED" },
111     [MIG_CMD_RECV_BITMAP]      = { .len = -1, .name = "RECV_BITMAP" },
112     [MIG_CMD_MAX]              = { .len = -1, .name = "MAX" },
113 };
114 
115 /* Note for MIG_CMD_POSTCOPY_ADVISE:
116  * The format of arguments is depending on postcopy mode:
117  * - postcopy RAM only
118  *   uint64_t host page size
119  *   uint64_t target page size
120  *
121  * - postcopy RAM and postcopy dirty bitmaps
122  *   format is the same as for postcopy RAM only
123  *
124  * - postcopy dirty bitmaps only
125  *   Nothing. Command length field is 0.
126  *
127  * Be careful: adding a new postcopy entity with some other parameters should
128  * not break format self-description ability. Good way is to introduce some
129  * generic extendable format with an exception for two old entities.
130  */
131 
132 /***********************************************************/
133 /* savevm/loadvm support */
134 
135 static QEMUFile *qemu_fopen_bdrv(BlockDriverState *bs, int is_writable)
136 {
137     if (is_writable) {
138         return qemu_file_new_output(QIO_CHANNEL(qio_channel_block_new(bs)));
139     } else {
140         return qemu_file_new_input(QIO_CHANNEL(qio_channel_block_new(bs)));
141     }
142 }
143 
144 
145 /* QEMUFile timer support.
146  * Not in qemu-file.c to not add qemu-timer.c as dependency to qemu-file.c
147  */
148 
149 void timer_put(QEMUFile *f, QEMUTimer *ts)
150 {
151     uint64_t expire_time;
152 
153     expire_time = timer_expire_time_ns(ts);
154     qemu_put_be64(f, expire_time);
155 }
156 
157 void timer_get(QEMUFile *f, QEMUTimer *ts)
158 {
159     uint64_t expire_time;
160 
161     expire_time = qemu_get_be64(f);
162     if (expire_time != -1) {
163         timer_mod_ns(ts, expire_time);
164     } else {
165         timer_del(ts);
166     }
167 }
168 
169 
170 /* VMState timer support.
171  * Not in vmstate.c to not add qemu-timer.c as dependency to vmstate.c
172  */
173 
174 static int get_timer(QEMUFile *f, void *pv, size_t size,
175                      const VMStateField *field)
176 {
177     QEMUTimer *v = pv;
178     timer_get(f, v);
179     return 0;
180 }
181 
182 static int put_timer(QEMUFile *f, void *pv, size_t size,
183                      const VMStateField *field, JSONWriter *vmdesc)
184 {
185     QEMUTimer *v = pv;
186     timer_put(f, v);
187 
188     return 0;
189 }
190 
191 const VMStateInfo vmstate_info_timer = {
192     .name = "timer",
193     .get  = get_timer,
194     .put  = put_timer,
195 };
196 
197 
198 typedef struct CompatEntry {
199     char idstr[256];
200     int instance_id;
201 } CompatEntry;
202 
203 typedef struct SaveStateEntry {
204     QTAILQ_ENTRY(SaveStateEntry) entry;
205     char idstr[256];
206     uint32_t instance_id;
207     int alias_id;
208     int version_id;
209     /* version id read from the stream */
210     int load_version_id;
211     int section_id;
212     /* section id read from the stream */
213     int load_section_id;
214     const SaveVMHandlers *ops;
215     const VMStateDescription *vmsd;
216     void *opaque;
217     CompatEntry *compat;
218     int is_ram;
219 } SaveStateEntry;
220 
221 typedef struct SaveState {
222     QTAILQ_HEAD(, SaveStateEntry) handlers;
223     SaveStateEntry *handler_pri_head[MIG_PRI_MAX + 1];
224     int global_section_id;
225     uint32_t len;
226     const char *name;
227     uint32_t target_page_bits;
228     uint32_t caps_count;
229     MigrationCapability *capabilities;
230     QemuUUID uuid;
231 } SaveState;
232 
233 static SaveState savevm_state = {
234     .handlers = QTAILQ_HEAD_INITIALIZER(savevm_state.handlers),
235     .handler_pri_head = { [MIG_PRI_DEFAULT ... MIG_PRI_MAX] = NULL },
236     .global_section_id = 0,
237 };
238 
239 static SaveStateEntry *find_se(const char *idstr, uint32_t instance_id);
240 
241 static bool should_validate_capability(int capability)
242 {
243     assert(capability >= 0 && capability < MIGRATION_CAPABILITY__MAX);
244     /* Validate only new capabilities to keep compatibility. */
245     switch (capability) {
246     case MIGRATION_CAPABILITY_X_IGNORE_SHARED:
247     case MIGRATION_CAPABILITY_MAPPED_RAM:
248         return true;
249     default:
250         return false;
251     }
252 }
253 
254 static uint32_t get_validatable_capabilities_count(void)
255 {
256     MigrationState *s = migrate_get_current();
257     uint32_t result = 0;
258     int i;
259     for (i = 0; i < MIGRATION_CAPABILITY__MAX; i++) {
260         if (should_validate_capability(i) && s->capabilities[i]) {
261             result++;
262         }
263     }
264     return result;
265 }
266 
267 static int configuration_pre_save(void *opaque)
268 {
269     SaveState *state = opaque;
270     const char *current_name = MACHINE_GET_CLASS(current_machine)->name;
271     MigrationState *s = migrate_get_current();
272     int i, j;
273 
274     state->len = strlen(current_name);
275     state->name = current_name;
276     state->target_page_bits = qemu_target_page_bits();
277 
278     state->caps_count = get_validatable_capabilities_count();
279     state->capabilities = g_renew(MigrationCapability, state->capabilities,
280                                   state->caps_count);
281     for (i = j = 0; i < MIGRATION_CAPABILITY__MAX; i++) {
282         if (should_validate_capability(i) && s->capabilities[i]) {
283             state->capabilities[j++] = i;
284         }
285     }
286     state->uuid = qemu_uuid;
287 
288     return 0;
289 }
290 
291 static int configuration_post_save(void *opaque)
292 {
293     SaveState *state = opaque;
294 
295     g_free(state->capabilities);
296     state->capabilities = NULL;
297     state->caps_count = 0;
298     return 0;
299 }
300 
301 static int configuration_pre_load(void *opaque)
302 {
303     SaveState *state = opaque;
304 
305     /* If there is no target-page-bits subsection it means the source
306      * predates the variable-target-page-bits support and is using the
307      * minimum possible value for this CPU.
308      */
309     state->target_page_bits = qemu_target_page_bits_min();
310     return 0;
311 }
312 
313 static bool configuration_validate_capabilities(SaveState *state)
314 {
315     bool ret = true;
316     MigrationState *s = migrate_get_current();
317     unsigned long *source_caps_bm;
318     int i;
319 
320     source_caps_bm = bitmap_new(MIGRATION_CAPABILITY__MAX);
321     for (i = 0; i < state->caps_count; i++) {
322         MigrationCapability capability = state->capabilities[i];
323         set_bit(capability, source_caps_bm);
324     }
325 
326     for (i = 0; i < MIGRATION_CAPABILITY__MAX; i++) {
327         bool source_state, target_state;
328         if (!should_validate_capability(i)) {
329             continue;
330         }
331         source_state = test_bit(i, source_caps_bm);
332         target_state = s->capabilities[i];
333         if (source_state != target_state) {
334             error_report("Capability %s is %s, but received capability is %s",
335                          MigrationCapability_str(i),
336                          target_state ? "on" : "off",
337                          source_state ? "on" : "off");
338             ret = false;
339             /* Don't break here to report all failed capabilities */
340         }
341     }
342 
343     g_free(source_caps_bm);
344     return ret;
345 }
346 
347 static int configuration_post_load(void *opaque, int version_id)
348 {
349     SaveState *state = opaque;
350     const char *current_name = MACHINE_GET_CLASS(current_machine)->name;
351     int ret = 0;
352 
353     if (strncmp(state->name, current_name, state->len) != 0) {
354         error_report("Machine type received is '%.*s' and local is '%s'",
355                      (int) state->len, state->name, current_name);
356         ret = -EINVAL;
357         goto out;
358     }
359 
360     if (state->target_page_bits != qemu_target_page_bits()) {
361         error_report("Received TARGET_PAGE_BITS is %d but local is %d",
362                      state->target_page_bits, qemu_target_page_bits());
363         ret = -EINVAL;
364         goto out;
365     }
366 
367     if (!configuration_validate_capabilities(state)) {
368         ret = -EINVAL;
369         goto out;
370     }
371 
372 out:
373     g_free((void *)state->name);
374     state->name = NULL;
375     state->len = 0;
376     g_free(state->capabilities);
377     state->capabilities = NULL;
378     state->caps_count = 0;
379 
380     return ret;
381 }
382 
383 static int get_capability(QEMUFile *f, void *pv, size_t size,
384                           const VMStateField *field)
385 {
386     MigrationCapability *capability = pv;
387     char capability_str[UINT8_MAX + 1];
388     uint8_t len;
389     int i;
390 
391     len = qemu_get_byte(f);
392     qemu_get_buffer(f, (uint8_t *)capability_str, len);
393     capability_str[len] = '\0';
394     for (i = 0; i < MIGRATION_CAPABILITY__MAX; i++) {
395         if (!strcmp(MigrationCapability_str(i), capability_str)) {
396             *capability = i;
397             return 0;
398         }
399     }
400     error_report("Received unknown capability %s", capability_str);
401     return -EINVAL;
402 }
403 
404 static int put_capability(QEMUFile *f, void *pv, size_t size,
405                           const VMStateField *field, JSONWriter *vmdesc)
406 {
407     MigrationCapability *capability = pv;
408     const char *capability_str = MigrationCapability_str(*capability);
409     size_t len = strlen(capability_str);
410     assert(len <= UINT8_MAX);
411 
412     qemu_put_byte(f, len);
413     qemu_put_buffer(f, (uint8_t *)capability_str, len);
414     return 0;
415 }
416 
417 static const VMStateInfo vmstate_info_capability = {
418     .name = "capability",
419     .get  = get_capability,
420     .put  = put_capability,
421 };
422 
423 /* The target-page-bits subsection is present only if the
424  * target page size is not the same as the default (ie the
425  * minimum page size for a variable-page-size guest CPU).
426  * If it is present then it contains the actual target page
427  * bits for the machine, and migration will fail if the
428  * two ends don't agree about it.
429  */
430 static bool vmstate_target_page_bits_needed(void *opaque)
431 {
432     return qemu_target_page_bits()
433         > qemu_target_page_bits_min();
434 }
435 
436 static const VMStateDescription vmstate_target_page_bits = {
437     .name = "configuration/target-page-bits",
438     .version_id = 1,
439     .minimum_version_id = 1,
440     .needed = vmstate_target_page_bits_needed,
441     .fields = (const VMStateField[]) {
442         VMSTATE_UINT32(target_page_bits, SaveState),
443         VMSTATE_END_OF_LIST()
444     }
445 };
446 
447 static bool vmstate_capabilites_needed(void *opaque)
448 {
449     return get_validatable_capabilities_count() > 0;
450 }
451 
452 static const VMStateDescription vmstate_capabilites = {
453     .name = "configuration/capabilities",
454     .version_id = 1,
455     .minimum_version_id = 1,
456     .needed = vmstate_capabilites_needed,
457     .fields = (const VMStateField[]) {
458         VMSTATE_UINT32_V(caps_count, SaveState, 1),
459         VMSTATE_VARRAY_UINT32_ALLOC(capabilities, SaveState, caps_count, 1,
460                                     vmstate_info_capability,
461                                     MigrationCapability),
462         VMSTATE_END_OF_LIST()
463     }
464 };
465 
466 static bool vmstate_uuid_needed(void *opaque)
467 {
468     return qemu_uuid_set && migrate_validate_uuid();
469 }
470 
471 static int vmstate_uuid_post_load(void *opaque, int version_id)
472 {
473     SaveState *state = opaque;
474     char uuid_src[UUID_STR_LEN];
475     char uuid_dst[UUID_STR_LEN];
476 
477     if (!qemu_uuid_set) {
478         /*
479          * It's warning because user might not know UUID in some cases,
480          * e.g. load an old snapshot
481          */
482         qemu_uuid_unparse(&state->uuid, uuid_src);
483         warn_report("UUID is received %s, but local uuid isn't set",
484                      uuid_src);
485         return 0;
486     }
487     if (!qemu_uuid_is_equal(&state->uuid, &qemu_uuid)) {
488         qemu_uuid_unparse(&state->uuid, uuid_src);
489         qemu_uuid_unparse(&qemu_uuid, uuid_dst);
490         error_report("UUID received is %s and local is %s", uuid_src, uuid_dst);
491         return -EINVAL;
492     }
493     return 0;
494 }
495 
496 static const VMStateDescription vmstate_uuid = {
497     .name = "configuration/uuid",
498     .version_id = 1,
499     .minimum_version_id = 1,
500     .needed = vmstate_uuid_needed,
501     .post_load = vmstate_uuid_post_load,
502     .fields = (const VMStateField[]) {
503         VMSTATE_UINT8_ARRAY_V(uuid.data, SaveState, sizeof(QemuUUID), 1),
504         VMSTATE_END_OF_LIST()
505     }
506 };
507 
508 static const VMStateDescription vmstate_configuration = {
509     .name = "configuration",
510     .version_id = 1,
511     .pre_load = configuration_pre_load,
512     .post_load = configuration_post_load,
513     .pre_save = configuration_pre_save,
514     .post_save = configuration_post_save,
515     .fields = (const VMStateField[]) {
516         VMSTATE_UINT32(len, SaveState),
517         VMSTATE_VBUFFER_ALLOC_UINT32(name, SaveState, 0, NULL, len),
518         VMSTATE_END_OF_LIST()
519     },
520     .subsections = (const VMStateDescription * const []) {
521         &vmstate_target_page_bits,
522         &vmstate_capabilites,
523         &vmstate_uuid,
524         NULL
525     }
526 };
527 
528 static void dump_vmstate_vmsd(FILE *out_file,
529                               const VMStateDescription *vmsd, int indent,
530                               bool is_subsection);
531 
532 static void dump_vmstate_vmsf(FILE *out_file, const VMStateField *field,
533                               int indent)
534 {
535     fprintf(out_file, "%*s{\n", indent, "");
536     indent += 2;
537     fprintf(out_file, "%*s\"field\": \"%s\",\n", indent, "", field->name);
538     fprintf(out_file, "%*s\"version_id\": %d,\n", indent, "",
539             field->version_id);
540     fprintf(out_file, "%*s\"field_exists\": %s,\n", indent, "",
541             field->field_exists ? "true" : "false");
542     if (field->flags & VMS_ARRAY) {
543         fprintf(out_file, "%*s\"num\": %d,\n", indent, "", field->num);
544     }
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 
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 
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         assert(field->flags == VMS_END);
597         fprintf(out_file, "\n%*s]", indent, "");
598     }
599     if (vmsd->subsections != NULL) {
600         const VMStateDescription * const *subsection = vmsd->subsections;
601         bool first;
602 
603         fprintf(out_file, ",\n%*s\"Subsections\": [\n", indent, "");
604         first = true;
605         while (*subsection != NULL) {
606             if (!first) {
607                 fprintf(out_file, ",\n");
608             }
609             dump_vmstate_vmss(out_file, *subsection, indent + 2);
610             subsection++;
611             first = false;
612         }
613         fprintf(out_file, "\n%*s]", indent, "");
614     }
615     fprintf(out_file, "\n%*s}", indent - 2, "");
616 }
617 
618 static void dump_machine_type(FILE *out_file)
619 {
620     MachineClass *mc;
621 
622     mc = MACHINE_GET_CLASS(current_machine);
623 
624     fprintf(out_file, "  \"vmschkmachine\": {\n");
625     fprintf(out_file, "    \"Name\": \"%s\"\n", mc->name);
626     fprintf(out_file, "  },\n");
627 }
628 
629 void dump_vmstate_json_to_file(FILE *out_file)
630 {
631     GSList *list, *elt;
632     bool first;
633 
634     fprintf(out_file, "{\n");
635     dump_machine_type(out_file);
636 
637     first = true;
638     list = object_class_get_list(TYPE_DEVICE, true);
639     for (elt = list; elt; elt = elt->next) {
640         DeviceClass *dc = OBJECT_CLASS_CHECK(DeviceClass, elt->data,
641                                              TYPE_DEVICE);
642         const char *name;
643         int indent = 2;
644 
645         if (!dc->vmsd) {
646             continue;
647         }
648 
649         if (!first) {
650             fprintf(out_file, ",\n");
651         }
652         name = object_class_get_name(OBJECT_CLASS(dc));
653         fprintf(out_file, "%*s\"%s\": {\n", indent, "", name);
654         indent += 2;
655         fprintf(out_file, "%*s\"Name\": \"%s\",\n", indent, "", name);
656         fprintf(out_file, "%*s\"version_id\": %d,\n", indent, "",
657                 dc->vmsd->version_id);
658         fprintf(out_file, "%*s\"minimum_version_id\": %d,\n", indent, "",
659                 dc->vmsd->minimum_version_id);
660 
661         dump_vmstate_vmsd(out_file, dc->vmsd, indent, false);
662 
663         fprintf(out_file, "\n%*s}", indent - 2, "");
664         first = false;
665     }
666     fprintf(out_file, "\n}\n");
667     fclose(out_file);
668     g_slist_free(list);
669 }
670 
671 static uint32_t calculate_new_instance_id(const char *idstr)
672 {
673     SaveStateEntry *se;
674     uint32_t instance_id = 0;
675 
676     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
677         if (strcmp(idstr, se->idstr) == 0
678             && instance_id <= se->instance_id) {
679             instance_id = se->instance_id + 1;
680         }
681     }
682     /* Make sure we never loop over without being noticed */
683     assert(instance_id != VMSTATE_INSTANCE_ID_ANY);
684     return instance_id;
685 }
686 
687 static int calculate_compat_instance_id(const char *idstr)
688 {
689     SaveStateEntry *se;
690     int instance_id = 0;
691 
692     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
693         if (!se->compat) {
694             continue;
695         }
696 
697         if (strcmp(idstr, se->compat->idstr) == 0
698             && instance_id <= se->compat->instance_id) {
699             instance_id = se->compat->instance_id + 1;
700         }
701     }
702     return instance_id;
703 }
704 
705 static inline MigrationPriority save_state_priority(SaveStateEntry *se)
706 {
707     if (se->vmsd) {
708         return se->vmsd->priority;
709     }
710     return MIG_PRI_DEFAULT;
711 }
712 
713 static void savevm_state_handler_insert(SaveStateEntry *nse)
714 {
715     MigrationPriority priority = save_state_priority(nse);
716     SaveStateEntry *se;
717     int i;
718 
719     assert(priority <= MIG_PRI_MAX);
720 
721     /*
722      * This should never happen otherwise migration will probably fail
723      * silently somewhere because we can be wrongly applying one
724      * object properties upon another one.  Bail out ASAP.
725      */
726     if (find_se(nse->idstr, nse->instance_id)) {
727         error_report("%s: Detected duplicate SaveStateEntry: "
728                      "id=%s, instance_id=0x%"PRIx32, __func__,
729                      nse->idstr, nse->instance_id);
730         exit(EXIT_FAILURE);
731     }
732 
733     for (i = priority - 1; i >= 0; i--) {
734         se = savevm_state.handler_pri_head[i];
735         if (se != NULL) {
736             assert(save_state_priority(se) < priority);
737             break;
738         }
739     }
740 
741     if (i >= 0) {
742         QTAILQ_INSERT_BEFORE(se, nse, entry);
743     } else {
744         QTAILQ_INSERT_TAIL(&savevm_state.handlers, nse, entry);
745     }
746 
747     if (savevm_state.handler_pri_head[priority] == NULL) {
748         savevm_state.handler_pri_head[priority] = nse;
749     }
750 }
751 
752 static void savevm_state_handler_remove(SaveStateEntry *se)
753 {
754     SaveStateEntry *next;
755     MigrationPriority priority = save_state_priority(se);
756 
757     if (se == savevm_state.handler_pri_head[priority]) {
758         next = QTAILQ_NEXT(se, entry);
759         if (next != NULL && save_state_priority(next) == priority) {
760             savevm_state.handler_pri_head[priority] = next;
761         } else {
762             savevm_state.handler_pri_head[priority] = NULL;
763         }
764     }
765     QTAILQ_REMOVE(&savevm_state.handlers, se, entry);
766 }
767 
768 /* TODO: Individual devices generally have very little idea about the rest
769    of the system, so instance_id should be removed/replaced.
770    Meanwhile pass -1 as instance_id if you do not already have a clearly
771    distinguishing id for all instances of your device class. */
772 int register_savevm_live(const char *idstr,
773                          uint32_t instance_id,
774                          int version_id,
775                          const SaveVMHandlers *ops,
776                          void *opaque)
777 {
778     SaveStateEntry *se;
779 
780     se = g_new0(SaveStateEntry, 1);
781     se->version_id = version_id;
782     se->section_id = savevm_state.global_section_id++;
783     se->ops = ops;
784     se->opaque = opaque;
785     se->vmsd = NULL;
786     /* if this is a live_savem then set is_ram */
787     if (ops->save_setup != NULL) {
788         se->is_ram = 1;
789     }
790 
791     pstrcat(se->idstr, sizeof(se->idstr), idstr);
792 
793     if (instance_id == VMSTATE_INSTANCE_ID_ANY) {
794         se->instance_id = calculate_new_instance_id(se->idstr);
795     } else {
796         se->instance_id = instance_id;
797     }
798     assert(!se->compat || se->instance_id == 0);
799     savevm_state_handler_insert(se);
800     return 0;
801 }
802 
803 void unregister_savevm(VMStateIf *obj, const char *idstr, void *opaque)
804 {
805     SaveStateEntry *se, *new_se;
806     char id[256] = "";
807 
808     if (obj) {
809         char *oid = vmstate_if_get_id(obj);
810         if (oid) {
811             pstrcpy(id, sizeof(id), oid);
812             pstrcat(id, sizeof(id), "/");
813             g_free(oid);
814         }
815     }
816     pstrcat(id, sizeof(id), idstr);
817 
818     QTAILQ_FOREACH_SAFE(se, &savevm_state.handlers, entry, new_se) {
819         if (strcmp(se->idstr, id) == 0 && se->opaque == opaque) {
820             savevm_state_handler_remove(se);
821             g_free(se->compat);
822             g_free(se);
823         }
824     }
825 }
826 
827 /*
828  * Perform some basic checks on vmsd's at registration
829  * time.
830  */
831 static void vmstate_check(const VMStateDescription *vmsd)
832 {
833     const VMStateField *field = vmsd->fields;
834     const VMStateDescription * const *subsection = vmsd->subsections;
835 
836     if (field) {
837         while (field->name) {
838             if (field->flags & (VMS_STRUCT | VMS_VSTRUCT)) {
839                 /* Recurse to sub structures */
840                 vmstate_check(field->vmsd);
841             }
842             /* Carry on */
843             field++;
844         }
845         /* Check for the end of field list canary */
846         if (field->flags != VMS_END) {
847             error_report("VMSTATE not ending with VMS_END: %s", vmsd->name);
848             g_assert_not_reached();
849         }
850     }
851 
852     while (subsection && *subsection) {
853         /*
854          * The name of a subsection should start with the name of the
855          * current object.
856          */
857         assert(!strncmp(vmsd->name, (*subsection)->name, strlen(vmsd->name)));
858         vmstate_check(*subsection);
859         subsection++;
860     }
861 }
862 
863 
864 int vmstate_register_with_alias_id(VMStateIf *obj, uint32_t instance_id,
865                                    const VMStateDescription *vmsd,
866                                    void *opaque, int alias_id,
867                                    int required_for_version,
868                                    Error **errp)
869 {
870     SaveStateEntry *se;
871 
872     /* If this triggers, alias support can be dropped for the vmsd. */
873     assert(alias_id == -1 || required_for_version >= vmsd->minimum_version_id);
874 
875     se = g_new0(SaveStateEntry, 1);
876     se->version_id = vmsd->version_id;
877     se->section_id = savevm_state.global_section_id++;
878     se->opaque = opaque;
879     se->vmsd = vmsd;
880     se->alias_id = alias_id;
881 
882     if (obj) {
883         char *id = vmstate_if_get_id(obj);
884         if (id) {
885             if (snprintf(se->idstr, sizeof(se->idstr), "%s/", id) >=
886                 sizeof(se->idstr)) {
887                 error_setg(errp, "Path too long for VMState (%s)", id);
888                 g_free(id);
889                 g_free(se);
890 
891                 return -1;
892             }
893             g_free(id);
894 
895             se->compat = g_new0(CompatEntry, 1);
896             pstrcpy(se->compat->idstr, sizeof(se->compat->idstr), vmsd->name);
897             se->compat->instance_id = instance_id == VMSTATE_INSTANCE_ID_ANY ?
898                          calculate_compat_instance_id(vmsd->name) : instance_id;
899             instance_id = VMSTATE_INSTANCE_ID_ANY;
900         }
901     }
902     pstrcat(se->idstr, sizeof(se->idstr), vmsd->name);
903 
904     if (instance_id == VMSTATE_INSTANCE_ID_ANY) {
905         se->instance_id = calculate_new_instance_id(se->idstr);
906     } else {
907         se->instance_id = instance_id;
908     }
909 
910     /* Perform a recursive sanity check during the test runs */
911     if (qtest_enabled()) {
912         vmstate_check(vmsd);
913     }
914     assert(!se->compat || se->instance_id == 0);
915     savevm_state_handler_insert(se);
916     return 0;
917 }
918 
919 void vmstate_unregister(VMStateIf *obj, const VMStateDescription *vmsd,
920                         void *opaque)
921 {
922     SaveStateEntry *se, *new_se;
923 
924     QTAILQ_FOREACH_SAFE(se, &savevm_state.handlers, entry, new_se) {
925         if (se->vmsd == vmsd && se->opaque == opaque) {
926             savevm_state_handler_remove(se);
927             g_free(se->compat);
928             g_free(se);
929         }
930     }
931 }
932 
933 static int vmstate_load(QEMUFile *f, SaveStateEntry *se)
934 {
935     trace_vmstate_load(se->idstr, se->vmsd ? se->vmsd->name : "(old)");
936     if (!se->vmsd) {         /* Old style */
937         return se->ops->load_state(f, se->opaque, se->load_version_id);
938     }
939     return vmstate_load_state(f, se->vmsd, se->opaque, se->load_version_id);
940 }
941 
942 static void vmstate_save_old_style(QEMUFile *f, SaveStateEntry *se,
943                                    JSONWriter *vmdesc)
944 {
945     uint64_t old_offset = qemu_file_transferred(f);
946     se->ops->save_state(f, se->opaque);
947     uint64_t size = qemu_file_transferred(f) - old_offset;
948 
949     if (vmdesc) {
950         json_writer_int64(vmdesc, "size", size);
951         json_writer_start_array(vmdesc, "fields");
952         json_writer_start_object(vmdesc, NULL);
953         json_writer_str(vmdesc, "name", "data");
954         json_writer_int64(vmdesc, "size", size);
955         json_writer_str(vmdesc, "type", "buffer");
956         json_writer_end_object(vmdesc);
957         json_writer_end_array(vmdesc);
958     }
959 }
960 
961 /*
962  * Write the header for device section (QEMU_VM_SECTION START/END/PART/FULL)
963  */
964 static void save_section_header(QEMUFile *f, SaveStateEntry *se,
965                                 uint8_t section_type)
966 {
967     qemu_put_byte(f, section_type);
968     qemu_put_be32(f, se->section_id);
969 
970     if (section_type == QEMU_VM_SECTION_FULL ||
971         section_type == QEMU_VM_SECTION_START) {
972         /* ID string */
973         size_t len = strlen(se->idstr);
974         qemu_put_byte(f, len);
975         qemu_put_buffer(f, (uint8_t *)se->idstr, len);
976 
977         qemu_put_be32(f, se->instance_id);
978         qemu_put_be32(f, se->version_id);
979     }
980 }
981 
982 /*
983  * Write a footer onto device sections that catches cases misformatted device
984  * sections.
985  */
986 static void save_section_footer(QEMUFile *f, SaveStateEntry *se)
987 {
988     if (migrate_get_current()->send_section_footer) {
989         qemu_put_byte(f, QEMU_VM_SECTION_FOOTER);
990         qemu_put_be32(f, se->section_id);
991     }
992 }
993 
994 static int vmstate_save(QEMUFile *f, SaveStateEntry *se, JSONWriter *vmdesc,
995                         Error **errp)
996 {
997     int ret;
998 
999     if ((!se->ops || !se->ops->save_state) && !se->vmsd) {
1000         return 0;
1001     }
1002     if (se->vmsd && !vmstate_section_needed(se->vmsd, se->opaque)) {
1003         trace_savevm_section_skip(se->idstr, se->section_id);
1004         return 0;
1005     }
1006 
1007     trace_savevm_section_start(se->idstr, se->section_id);
1008     save_section_header(f, se, QEMU_VM_SECTION_FULL);
1009     if (vmdesc) {
1010         json_writer_start_object(vmdesc, NULL);
1011         json_writer_str(vmdesc, "name", se->idstr);
1012         json_writer_int64(vmdesc, "instance_id", se->instance_id);
1013     }
1014 
1015     trace_vmstate_save(se->idstr, se->vmsd ? se->vmsd->name : "(old)");
1016     if (!se->vmsd) {
1017         vmstate_save_old_style(f, se, vmdesc);
1018     } else {
1019         ret = vmstate_save_state_with_err(f, se->vmsd, se->opaque, vmdesc,
1020                                           errp);
1021         if (ret) {
1022             return ret;
1023         }
1024     }
1025 
1026     trace_savevm_section_end(se->idstr, se->section_id, 0);
1027     save_section_footer(f, se);
1028     if (vmdesc) {
1029         json_writer_end_object(vmdesc);
1030     }
1031     return 0;
1032 }
1033 /**
1034  * qemu_savevm_command_send: Send a 'QEMU_VM_COMMAND' type element with the
1035  *                           command and associated data.
1036  *
1037  * @f: File to send command on
1038  * @command: Command type to send
1039  * @len: Length of associated data
1040  * @data: Data associated with command.
1041  */
1042 static void qemu_savevm_command_send(QEMUFile *f,
1043                                      enum qemu_vm_cmd command,
1044                                      uint16_t len,
1045                                      uint8_t *data)
1046 {
1047     trace_savevm_command_send(command, len);
1048     qemu_put_byte(f, QEMU_VM_COMMAND);
1049     qemu_put_be16(f, (uint16_t)command);
1050     qemu_put_be16(f, len);
1051     qemu_put_buffer(f, data, len);
1052     qemu_fflush(f);
1053 }
1054 
1055 void qemu_savevm_send_colo_enable(QEMUFile *f)
1056 {
1057     trace_savevm_send_colo_enable();
1058     qemu_savevm_command_send(f, MIG_CMD_ENABLE_COLO, 0, NULL);
1059 }
1060 
1061 void qemu_savevm_send_ping(QEMUFile *f, uint32_t value)
1062 {
1063     uint32_t buf;
1064 
1065     trace_savevm_send_ping(value);
1066     buf = cpu_to_be32(value);
1067     qemu_savevm_command_send(f, MIG_CMD_PING, sizeof(value), (uint8_t *)&buf);
1068 }
1069 
1070 void qemu_savevm_send_open_return_path(QEMUFile *f)
1071 {
1072     trace_savevm_send_open_return_path();
1073     qemu_savevm_command_send(f, MIG_CMD_OPEN_RETURN_PATH, 0, NULL);
1074 }
1075 
1076 /* We have a buffer of data to send; we don't want that all to be loaded
1077  * by the command itself, so the command contains just the length of the
1078  * extra buffer that we then send straight after it.
1079  * TODO: Must be a better way to organise that
1080  *
1081  * Returns:
1082  *    0 on success
1083  *    -ve on error
1084  */
1085 int qemu_savevm_send_packaged(QEMUFile *f, const uint8_t *buf, size_t len)
1086 {
1087     uint32_t tmp;
1088     MigrationState *ms = migrate_get_current();
1089     Error *local_err = NULL;
1090 
1091     if (len > MAX_VM_CMD_PACKAGED_SIZE) {
1092         error_setg(&local_err, "%s: Unreasonably large packaged state: %zu",
1093                      __func__, len);
1094         migrate_set_error(ms, local_err);
1095         error_report_err(local_err);
1096         return -1;
1097     }
1098 
1099     tmp = cpu_to_be32(len);
1100 
1101     trace_qemu_savevm_send_packaged();
1102     qemu_savevm_command_send(f, MIG_CMD_PACKAGED, 4, (uint8_t *)&tmp);
1103 
1104     qemu_put_buffer(f, buf, len);
1105 
1106     return 0;
1107 }
1108 
1109 /* Send prior to any postcopy transfer */
1110 void qemu_savevm_send_postcopy_advise(QEMUFile *f)
1111 {
1112     if (migrate_postcopy_ram()) {
1113         uint64_t tmp[2];
1114         tmp[0] = cpu_to_be64(ram_pagesize_summary());
1115         tmp[1] = cpu_to_be64(qemu_target_page_size());
1116 
1117         trace_qemu_savevm_send_postcopy_advise();
1118         qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_ADVISE,
1119                                  16, (uint8_t *)tmp);
1120     } else {
1121         qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_ADVISE, 0, NULL);
1122     }
1123 }
1124 
1125 /* Sent prior to starting the destination running in postcopy, discard pages
1126  * that have already been sent but redirtied on the source.
1127  * CMD_POSTCOPY_RAM_DISCARD consist of:
1128  *      byte   version (0)
1129  *      byte   Length of name field (not including 0)
1130  *  n x byte   RAM block name
1131  *      byte   0 terminator (just for safety)
1132  *  n x        Byte ranges within the named RAMBlock
1133  *      be64   Start of the range
1134  *      be64   Length
1135  *
1136  *  name:  RAMBlock name that these entries are part of
1137  *  len: Number of page entries
1138  *  start_list: 'len' addresses
1139  *  length_list: 'len' addresses
1140  *
1141  */
1142 void qemu_savevm_send_postcopy_ram_discard(QEMUFile *f, const char *name,
1143                                            uint16_t len,
1144                                            uint64_t *start_list,
1145                                            uint64_t *length_list)
1146 {
1147     uint8_t *buf;
1148     uint16_t tmplen;
1149     uint16_t t;
1150     size_t name_len = strlen(name);
1151 
1152     trace_qemu_savevm_send_postcopy_ram_discard(name, len);
1153     assert(name_len < 256);
1154     buf = g_malloc0(1 + 1 + name_len + 1 + (8 + 8) * len);
1155     buf[0] = postcopy_ram_discard_version;
1156     buf[1] = name_len;
1157     memcpy(buf + 2, name, name_len);
1158     tmplen = 2 + name_len;
1159     buf[tmplen++] = '\0';
1160 
1161     for (t = 0; t < len; t++) {
1162         stq_be_p(buf + tmplen, start_list[t]);
1163         tmplen += 8;
1164         stq_be_p(buf + tmplen, length_list[t]);
1165         tmplen += 8;
1166     }
1167     qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_RAM_DISCARD, tmplen, buf);
1168     g_free(buf);
1169 }
1170 
1171 /* Get the destination into a state where it can receive postcopy data. */
1172 void qemu_savevm_send_postcopy_listen(QEMUFile *f)
1173 {
1174     trace_savevm_send_postcopy_listen();
1175     qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_LISTEN, 0, NULL);
1176 }
1177 
1178 /* Kick the destination into running */
1179 void qemu_savevm_send_postcopy_run(QEMUFile *f)
1180 {
1181     trace_savevm_send_postcopy_run();
1182     qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_RUN, 0, NULL);
1183 }
1184 
1185 void qemu_savevm_send_postcopy_resume(QEMUFile *f)
1186 {
1187     trace_savevm_send_postcopy_resume();
1188     qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_RESUME, 0, NULL);
1189 }
1190 
1191 void qemu_savevm_send_recv_bitmap(QEMUFile *f, char *block_name)
1192 {
1193     size_t len;
1194     char buf[256];
1195 
1196     trace_savevm_send_recv_bitmap(block_name);
1197 
1198     buf[0] = len = strlen(block_name);
1199     memcpy(buf + 1, block_name, len);
1200 
1201     qemu_savevm_command_send(f, MIG_CMD_RECV_BITMAP, len + 1, (uint8_t *)buf);
1202 }
1203 
1204 bool qemu_savevm_state_blocked(Error **errp)
1205 {
1206     SaveStateEntry *se;
1207 
1208     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1209         if (se->vmsd && se->vmsd->unmigratable) {
1210             error_setg(errp, "State blocked by non-migratable device '%s'",
1211                        se->idstr);
1212             return true;
1213         }
1214     }
1215     return false;
1216 }
1217 
1218 void qemu_savevm_non_migratable_list(strList **reasons)
1219 {
1220     SaveStateEntry *se;
1221 
1222     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1223         if (se->vmsd && se->vmsd->unmigratable) {
1224             QAPI_LIST_PREPEND(*reasons,
1225                               g_strdup_printf("non-migratable device: %s",
1226                                               se->idstr));
1227         }
1228     }
1229 }
1230 
1231 void qemu_savevm_state_header(QEMUFile *f)
1232 {
1233     MigrationState *s = migrate_get_current();
1234 
1235     s->vmdesc = json_writer_new(false);
1236 
1237     trace_savevm_state_header();
1238     qemu_put_be32(f, QEMU_VM_FILE_MAGIC);
1239     qemu_put_be32(f, QEMU_VM_FILE_VERSION);
1240 
1241     if (s->send_configuration) {
1242         qemu_put_byte(f, QEMU_VM_CONFIGURATION);
1243 
1244         /*
1245          * This starts the main json object and is paired with the
1246          * json_writer_end_object in
1247          * qemu_savevm_state_complete_precopy_non_iterable
1248          */
1249         json_writer_start_object(s->vmdesc, NULL);
1250 
1251         json_writer_start_object(s->vmdesc, "configuration");
1252         vmstate_save_state(f, &vmstate_configuration, &savevm_state, s->vmdesc);
1253         json_writer_end_object(s->vmdesc);
1254     }
1255 }
1256 
1257 bool qemu_savevm_state_guest_unplug_pending(void)
1258 {
1259     SaveStateEntry *se;
1260 
1261     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1262         if (se->vmsd && se->vmsd->dev_unplug_pending &&
1263             se->vmsd->dev_unplug_pending(se->opaque)) {
1264             return true;
1265         }
1266     }
1267 
1268     return false;
1269 }
1270 
1271 int qemu_savevm_state_prepare(Error **errp)
1272 {
1273     SaveStateEntry *se;
1274     int ret;
1275 
1276     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1277         if (!se->ops || !se->ops->save_prepare) {
1278             continue;
1279         }
1280         if (se->ops->is_active) {
1281             if (!se->ops->is_active(se->opaque)) {
1282                 continue;
1283             }
1284         }
1285 
1286         ret = se->ops->save_prepare(se->opaque, errp);
1287         if (ret < 0) {
1288             return ret;
1289         }
1290     }
1291 
1292     return 0;
1293 }
1294 
1295 int qemu_savevm_state_setup(QEMUFile *f, Error **errp)
1296 {
1297     ERRP_GUARD();
1298     MigrationState *ms = migrate_get_current();
1299     SaveStateEntry *se;
1300     int ret = 0;
1301 
1302     json_writer_int64(ms->vmdesc, "page_size", qemu_target_page_size());
1303     json_writer_start_array(ms->vmdesc, "devices");
1304 
1305     trace_savevm_state_setup();
1306     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1307         if (se->vmsd && se->vmsd->early_setup) {
1308             ret = vmstate_save(f, se, ms->vmdesc, errp);
1309             if (ret) {
1310                 migrate_set_error(ms, *errp);
1311                 qemu_file_set_error(f, ret);
1312                 break;
1313             }
1314             continue;
1315         }
1316 
1317         if (!se->ops || !se->ops->save_setup) {
1318             continue;
1319         }
1320         if (se->ops->is_active) {
1321             if (!se->ops->is_active(se->opaque)) {
1322                 continue;
1323             }
1324         }
1325         save_section_header(f, se, QEMU_VM_SECTION_START);
1326 
1327         ret = se->ops->save_setup(f, se->opaque, errp);
1328         save_section_footer(f, se);
1329         if (ret < 0) {
1330             qemu_file_set_error(f, ret);
1331             break;
1332         }
1333     }
1334 
1335     if (ret) {
1336         return ret;
1337     }
1338 
1339     /* TODO: Should we check that errp is set in case of failure ? */
1340     return precopy_notify(PRECOPY_NOTIFY_SETUP, errp);
1341 }
1342 
1343 int qemu_savevm_state_resume_prepare(MigrationState *s)
1344 {
1345     SaveStateEntry *se;
1346     int ret;
1347 
1348     trace_savevm_state_resume_prepare();
1349 
1350     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1351         if (!se->ops || !se->ops->resume_prepare) {
1352             continue;
1353         }
1354         if (se->ops->is_active) {
1355             if (!se->ops->is_active(se->opaque)) {
1356                 continue;
1357             }
1358         }
1359         ret = se->ops->resume_prepare(s, se->opaque);
1360         if (ret < 0) {
1361             return ret;
1362         }
1363     }
1364 
1365     return 0;
1366 }
1367 
1368 /*
1369  * this function has three return values:
1370  *   negative: there was one error, and we have -errno.
1371  *   0 : We haven't finished, caller have to go again
1372  *   1 : We have finished, we can go to complete phase
1373  */
1374 int qemu_savevm_state_iterate(QEMUFile *f, bool postcopy)
1375 {
1376     SaveStateEntry *se;
1377     bool all_finished = true;
1378     int ret;
1379 
1380     trace_savevm_state_iterate();
1381     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1382         if (!se->ops || !se->ops->save_live_iterate) {
1383             continue;
1384         }
1385         if (se->ops->is_active &&
1386             !se->ops->is_active(se->opaque)) {
1387             continue;
1388         }
1389         if (se->ops->is_active_iterate &&
1390             !se->ops->is_active_iterate(se->opaque)) {
1391             continue;
1392         }
1393         /*
1394          * In the postcopy phase, any device that doesn't know how to
1395          * do postcopy should have saved it's state in the _complete
1396          * call that's already run, it might get confused if we call
1397          * iterate afterwards.
1398          */
1399         if (postcopy &&
1400             !(se->ops->has_postcopy && se->ops->has_postcopy(se->opaque))) {
1401             continue;
1402         }
1403         if (migration_rate_exceeded(f)) {
1404             return 0;
1405         }
1406         trace_savevm_section_start(se->idstr, se->section_id);
1407 
1408         save_section_header(f, se, QEMU_VM_SECTION_PART);
1409 
1410         ret = se->ops->save_live_iterate(f, se->opaque);
1411         trace_savevm_section_end(se->idstr, se->section_id, ret);
1412         save_section_footer(f, se);
1413 
1414         if (ret < 0) {
1415             error_report("failed to save SaveStateEntry with id(name): "
1416                          "%d(%s): %d",
1417                          se->section_id, se->idstr, ret);
1418             qemu_file_set_error(f, ret);
1419             return ret;
1420         } else if (!ret) {
1421             all_finished = false;
1422         }
1423     }
1424     return all_finished;
1425 }
1426 
1427 static bool should_send_vmdesc(void)
1428 {
1429     MachineState *machine = MACHINE(qdev_get_machine());
1430     bool in_postcopy = migration_in_postcopy();
1431     return !machine->suppress_vmdesc && !in_postcopy;
1432 }
1433 
1434 /*
1435  * Calls the save_live_complete_postcopy methods
1436  * causing the last few pages to be sent immediately and doing any associated
1437  * cleanup.
1438  * Note postcopy also calls qemu_savevm_state_complete_precopy to complete
1439  * all the other devices, but that happens at the point we switch to postcopy.
1440  */
1441 void qemu_savevm_state_complete_postcopy(QEMUFile *f)
1442 {
1443     SaveStateEntry *se;
1444     int ret;
1445 
1446     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1447         if (!se->ops || !se->ops->save_live_complete_postcopy) {
1448             continue;
1449         }
1450         if (se->ops->is_active) {
1451             if (!se->ops->is_active(se->opaque)) {
1452                 continue;
1453             }
1454         }
1455         trace_savevm_section_start(se->idstr, se->section_id);
1456         /* Section type */
1457         qemu_put_byte(f, QEMU_VM_SECTION_END);
1458         qemu_put_be32(f, se->section_id);
1459 
1460         ret = se->ops->save_live_complete_postcopy(f, se->opaque);
1461         trace_savevm_section_end(se->idstr, se->section_id, ret);
1462         save_section_footer(f, se);
1463         if (ret < 0) {
1464             qemu_file_set_error(f, ret);
1465             return;
1466         }
1467     }
1468 
1469     qemu_put_byte(f, QEMU_VM_EOF);
1470     qemu_fflush(f);
1471 }
1472 
1473 static
1474 int qemu_savevm_state_complete_precopy_iterable(QEMUFile *f, bool in_postcopy)
1475 {
1476     int64_t start_ts_each, end_ts_each;
1477     SaveStateEntry *se;
1478     int ret;
1479 
1480     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1481         if (!se->ops ||
1482             (in_postcopy && se->ops->has_postcopy &&
1483              se->ops->has_postcopy(se->opaque)) ||
1484             !se->ops->save_live_complete_precopy) {
1485             continue;
1486         }
1487 
1488         if (se->ops->is_active) {
1489             if (!se->ops->is_active(se->opaque)) {
1490                 continue;
1491             }
1492         }
1493 
1494         start_ts_each = qemu_clock_get_us(QEMU_CLOCK_REALTIME);
1495         trace_savevm_section_start(se->idstr, se->section_id);
1496 
1497         save_section_header(f, se, QEMU_VM_SECTION_END);
1498 
1499         ret = se->ops->save_live_complete_precopy(f, se->opaque);
1500         trace_savevm_section_end(se->idstr, se->section_id, ret);
1501         save_section_footer(f, se);
1502         if (ret < 0) {
1503             qemu_file_set_error(f, ret);
1504             return -1;
1505         }
1506         end_ts_each = qemu_clock_get_us(QEMU_CLOCK_REALTIME);
1507         trace_vmstate_downtime_save("iterable", se->idstr, se->instance_id,
1508                                     end_ts_each - start_ts_each);
1509     }
1510 
1511     trace_vmstate_downtime_checkpoint("src-iterable-saved");
1512 
1513     return 0;
1514 }
1515 
1516 int qemu_savevm_state_complete_precopy_non_iterable(QEMUFile *f,
1517                                                     bool in_postcopy,
1518                                                     bool inactivate_disks)
1519 {
1520     MigrationState *ms = migrate_get_current();
1521     int64_t start_ts_each, end_ts_each;
1522     JSONWriter *vmdesc = ms->vmdesc;
1523     int vmdesc_len;
1524     SaveStateEntry *se;
1525     Error *local_err = NULL;
1526     int ret;
1527 
1528     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1529         if (se->vmsd && se->vmsd->early_setup) {
1530             /* Already saved during qemu_savevm_state_setup(). */
1531             continue;
1532         }
1533 
1534         start_ts_each = qemu_clock_get_us(QEMU_CLOCK_REALTIME);
1535 
1536         ret = vmstate_save(f, se, vmdesc, &local_err);
1537         if (ret) {
1538             migrate_set_error(ms, local_err);
1539             error_report_err(local_err);
1540             qemu_file_set_error(f, ret);
1541             return ret;
1542         }
1543 
1544         end_ts_each = qemu_clock_get_us(QEMU_CLOCK_REALTIME);
1545         trace_vmstate_downtime_save("non-iterable", se->idstr, se->instance_id,
1546                                     end_ts_each - start_ts_each);
1547     }
1548 
1549     if (inactivate_disks) {
1550         /*
1551          * Inactivate before sending QEMU_VM_EOF so that the
1552          * bdrv_activate_all() on the other end won't fail.
1553          */
1554         if (!migration_block_inactivate()) {
1555             error_setg(&local_err, "%s: bdrv_inactivate_all() failed",
1556                        __func__);
1557             migrate_set_error(ms, local_err);
1558             error_report_err(local_err);
1559             qemu_file_set_error(f, -EFAULT);
1560             return ret;
1561         }
1562     }
1563     if (!in_postcopy) {
1564         /* Postcopy stream will still be going */
1565         qemu_put_byte(f, QEMU_VM_EOF);
1566     }
1567 
1568     json_writer_end_array(vmdesc);
1569     json_writer_end_object(vmdesc);
1570     vmdesc_len = strlen(json_writer_get(vmdesc));
1571 
1572     if (should_send_vmdesc()) {
1573         qemu_put_byte(f, QEMU_VM_VMDESCRIPTION);
1574         qemu_put_be32(f, vmdesc_len);
1575         qemu_put_buffer(f, (uint8_t *)json_writer_get(vmdesc), vmdesc_len);
1576     }
1577 
1578     /* Free it now to detect any inconsistencies. */
1579     json_writer_free(vmdesc);
1580     ms->vmdesc = NULL;
1581 
1582     trace_vmstate_downtime_checkpoint("src-non-iterable-saved");
1583 
1584     return 0;
1585 }
1586 
1587 int qemu_savevm_state_complete_precopy(QEMUFile *f, bool iterable_only,
1588                                        bool inactivate_disks)
1589 {
1590     int ret;
1591     Error *local_err = NULL;
1592     bool in_postcopy = migration_in_postcopy();
1593 
1594     if (precopy_notify(PRECOPY_NOTIFY_COMPLETE, &local_err)) {
1595         error_report_err(local_err);
1596     }
1597 
1598     trace_savevm_state_complete_precopy();
1599 
1600     cpu_synchronize_all_states();
1601 
1602     if (!in_postcopy || iterable_only) {
1603         ret = qemu_savevm_state_complete_precopy_iterable(f, in_postcopy);
1604         if (ret) {
1605             return ret;
1606         }
1607     }
1608 
1609     if (iterable_only) {
1610         goto flush;
1611     }
1612 
1613     ret = qemu_savevm_state_complete_precopy_non_iterable(f, in_postcopy,
1614                                                           inactivate_disks);
1615     if (ret) {
1616         return ret;
1617     }
1618 
1619 flush:
1620     return qemu_fflush(f);
1621 }
1622 
1623 /* Give an estimate of the amount left to be transferred,
1624  * the result is split into the amount for units that can and
1625  * for units that can't do postcopy.
1626  */
1627 void qemu_savevm_state_pending_estimate(uint64_t *must_precopy,
1628                                         uint64_t *can_postcopy)
1629 {
1630     SaveStateEntry *se;
1631 
1632     *must_precopy = 0;
1633     *can_postcopy = 0;
1634 
1635     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1636         if (!se->ops || !se->ops->state_pending_estimate) {
1637             continue;
1638         }
1639         if (se->ops->is_active) {
1640             if (!se->ops->is_active(se->opaque)) {
1641                 continue;
1642             }
1643         }
1644         se->ops->state_pending_estimate(se->opaque, must_precopy, can_postcopy);
1645     }
1646 }
1647 
1648 void qemu_savevm_state_pending_exact(uint64_t *must_precopy,
1649                                      uint64_t *can_postcopy)
1650 {
1651     SaveStateEntry *se;
1652 
1653     *must_precopy = 0;
1654     *can_postcopy = 0;
1655 
1656     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1657         if (!se->ops || !se->ops->state_pending_exact) {
1658             continue;
1659         }
1660         if (se->ops->is_active) {
1661             if (!se->ops->is_active(se->opaque)) {
1662                 continue;
1663             }
1664         }
1665         se->ops->state_pending_exact(se->opaque, must_precopy, can_postcopy);
1666     }
1667 }
1668 
1669 void qemu_savevm_state_cleanup(void)
1670 {
1671     SaveStateEntry *se;
1672     Error *local_err = NULL;
1673 
1674     if (precopy_notify(PRECOPY_NOTIFY_CLEANUP, &local_err)) {
1675         error_report_err(local_err);
1676     }
1677 
1678     trace_savevm_state_cleanup();
1679     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1680         if (se->ops && se->ops->save_cleanup) {
1681             se->ops->save_cleanup(se->opaque);
1682         }
1683     }
1684 }
1685 
1686 static int qemu_savevm_state(QEMUFile *f, Error **errp)
1687 {
1688     int ret;
1689     MigrationState *ms = migrate_get_current();
1690     MigrationStatus status;
1691 
1692     if (migration_is_running()) {
1693         error_setg(errp, "There's a migration process in progress");
1694         return -EINVAL;
1695     }
1696 
1697     ret = migrate_init(ms, errp);
1698     if (ret) {
1699         return ret;
1700     }
1701     ms->to_dst_file = f;
1702 
1703     qemu_savevm_state_header(f);
1704     ret = qemu_savevm_state_setup(f, errp);
1705     if (ret) {
1706         goto cleanup;
1707     }
1708 
1709     while (qemu_file_get_error(f) == 0) {
1710         if (qemu_savevm_state_iterate(f, false) > 0) {
1711             break;
1712         }
1713     }
1714 
1715     ret = qemu_file_get_error(f);
1716     if (ret == 0) {
1717         qemu_savevm_state_complete_precopy(f, false, false);
1718         ret = qemu_file_get_error(f);
1719     }
1720     if (ret != 0) {
1721         error_setg_errno(errp, -ret, "Error while writing VM state");
1722     }
1723 cleanup:
1724     qemu_savevm_state_cleanup();
1725 
1726     if (ret != 0) {
1727         status = MIGRATION_STATUS_FAILED;
1728     } else {
1729         status = MIGRATION_STATUS_COMPLETED;
1730     }
1731     migrate_set_state(&ms->state, MIGRATION_STATUS_SETUP, status);
1732 
1733     /* f is outer parameter, it should not stay in global migration state after
1734      * this function finished */
1735     ms->to_dst_file = NULL;
1736 
1737     return ret;
1738 }
1739 
1740 void qemu_savevm_live_state(QEMUFile *f)
1741 {
1742     /* save QEMU_VM_SECTION_END section */
1743     qemu_savevm_state_complete_precopy(f, true, false);
1744     qemu_put_byte(f, QEMU_VM_EOF);
1745 }
1746 
1747 int qemu_save_device_state(QEMUFile *f)
1748 {
1749     MigrationState *ms = migrate_get_current();
1750     Error *local_err = NULL;
1751     SaveStateEntry *se;
1752 
1753     if (!migration_in_colo_state()) {
1754         qemu_put_be32(f, QEMU_VM_FILE_MAGIC);
1755         qemu_put_be32(f, QEMU_VM_FILE_VERSION);
1756     }
1757     cpu_synchronize_all_states();
1758 
1759     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1760         int ret;
1761 
1762         if (se->is_ram) {
1763             continue;
1764         }
1765         ret = vmstate_save(f, se, NULL, &local_err);
1766         if (ret) {
1767             migrate_set_error(ms, local_err);
1768             error_report_err(local_err);
1769             return ret;
1770         }
1771     }
1772 
1773     qemu_put_byte(f, QEMU_VM_EOF);
1774 
1775     return qemu_file_get_error(f);
1776 }
1777 
1778 static SaveStateEntry *find_se(const char *idstr, uint32_t instance_id)
1779 {
1780     SaveStateEntry *se;
1781 
1782     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1783         if (!strcmp(se->idstr, idstr) &&
1784             (instance_id == se->instance_id ||
1785              instance_id == se->alias_id))
1786             return se;
1787         /* Migrating from an older version? */
1788         if (strstr(se->idstr, idstr) && se->compat) {
1789             if (!strcmp(se->compat->idstr, idstr) &&
1790                 (instance_id == se->compat->instance_id ||
1791                  instance_id == se->alias_id))
1792                 return se;
1793         }
1794     }
1795     return NULL;
1796 }
1797 
1798 enum LoadVMExitCodes {
1799     /* Allow a command to quit all layers of nested loadvm loops */
1800     LOADVM_QUIT     =  1,
1801 };
1802 
1803 /* ------ incoming postcopy messages ------ */
1804 /* 'advise' arrives before any transfers just to tell us that a postcopy
1805  * *might* happen - it might be skipped if precopy transferred everything
1806  * quickly.
1807  */
1808 static int loadvm_postcopy_handle_advise(MigrationIncomingState *mis,
1809                                          uint16_t len)
1810 {
1811     PostcopyState ps = postcopy_state_set(POSTCOPY_INCOMING_ADVISE);
1812     uint64_t remote_pagesize_summary, local_pagesize_summary, remote_tps;
1813     size_t page_size = qemu_target_page_size();
1814     Error *local_err = NULL;
1815 
1816     trace_loadvm_postcopy_handle_advise();
1817     if (ps != POSTCOPY_INCOMING_NONE) {
1818         error_report("CMD_POSTCOPY_ADVISE in wrong postcopy state (%d)", ps);
1819         return -1;
1820     }
1821 
1822     switch (len) {
1823     case 0:
1824         if (migrate_postcopy_ram()) {
1825             error_report("RAM postcopy is enabled but have 0 byte advise");
1826             return -EINVAL;
1827         }
1828         return 0;
1829     case 8 + 8:
1830         if (!migrate_postcopy_ram()) {
1831             error_report("RAM postcopy is disabled but have 16 byte advise");
1832             return -EINVAL;
1833         }
1834         break;
1835     default:
1836         error_report("CMD_POSTCOPY_ADVISE invalid length (%d)", len);
1837         return -EINVAL;
1838     }
1839 
1840     if (!postcopy_ram_supported_by_host(mis, &local_err)) {
1841         error_report_err(local_err);
1842         postcopy_state_set(POSTCOPY_INCOMING_NONE);
1843         return -1;
1844     }
1845 
1846     remote_pagesize_summary = qemu_get_be64(mis->from_src_file);
1847     local_pagesize_summary = ram_pagesize_summary();
1848 
1849     if (remote_pagesize_summary != local_pagesize_summary)  {
1850         /*
1851          * This detects two potential causes of mismatch:
1852          *   a) A mismatch in host page sizes
1853          *      Some combinations of mismatch are probably possible but it gets
1854          *      a bit more complicated.  In particular we need to place whole
1855          *      host pages on the dest at once, and we need to ensure that we
1856          *      handle dirtying to make sure we never end up sending part of
1857          *      a hostpage on it's own.
1858          *   b) The use of different huge page sizes on source/destination
1859          *      a more fine grain test is performed during RAM block migration
1860          *      but this test here causes a nice early clear failure, and
1861          *      also fails when passed to an older qemu that doesn't
1862          *      do huge pages.
1863          */
1864         error_report("Postcopy needs matching RAM page sizes (s=%" PRIx64
1865                                                              " d=%" PRIx64 ")",
1866                      remote_pagesize_summary, local_pagesize_summary);
1867         return -1;
1868     }
1869 
1870     remote_tps = qemu_get_be64(mis->from_src_file);
1871     if (remote_tps != page_size) {
1872         /*
1873          * Again, some differences could be dealt with, but for now keep it
1874          * simple.
1875          */
1876         error_report("Postcopy needs matching target page sizes (s=%d d=%zd)",
1877                      (int)remote_tps, page_size);
1878         return -1;
1879     }
1880 
1881     if (postcopy_notify(POSTCOPY_NOTIFY_INBOUND_ADVISE, &local_err)) {
1882         error_report_err(local_err);
1883         return -1;
1884     }
1885 
1886     if (ram_postcopy_incoming_init(mis)) {
1887         return -1;
1888     }
1889 
1890     return 0;
1891 }
1892 
1893 /* After postcopy we will be told to throw some pages away since they're
1894  * dirty and will have to be demand fetched.  Must happen before CPU is
1895  * started.
1896  * There can be 0..many of these messages, each encoding multiple pages.
1897  */
1898 static int loadvm_postcopy_ram_handle_discard(MigrationIncomingState *mis,
1899                                               uint16_t len)
1900 {
1901     int tmp;
1902     char ramid[256];
1903     PostcopyState ps = postcopy_state_get();
1904 
1905     trace_loadvm_postcopy_ram_handle_discard();
1906 
1907     switch (ps) {
1908     case POSTCOPY_INCOMING_ADVISE:
1909         /* 1st discard */
1910         tmp = postcopy_ram_prepare_discard(mis);
1911         if (tmp) {
1912             return tmp;
1913         }
1914         break;
1915 
1916     case POSTCOPY_INCOMING_DISCARD:
1917         /* Expected state */
1918         break;
1919 
1920     default:
1921         error_report("CMD_POSTCOPY_RAM_DISCARD in wrong postcopy state (%d)",
1922                      ps);
1923         return -1;
1924     }
1925     /* We're expecting a
1926      *    Version (0)
1927      *    a RAM ID string (length byte, name, 0 term)
1928      *    then at least 1 16 byte chunk
1929     */
1930     if (len < (1 + 1 + 1 + 1 + 2 * 8)) {
1931         error_report("CMD_POSTCOPY_RAM_DISCARD invalid length (%d)", len);
1932         return -1;
1933     }
1934 
1935     tmp = qemu_get_byte(mis->from_src_file);
1936     if (tmp != postcopy_ram_discard_version) {
1937         error_report("CMD_POSTCOPY_RAM_DISCARD invalid version (%d)", tmp);
1938         return -1;
1939     }
1940 
1941     if (!qemu_get_counted_string(mis->from_src_file, ramid)) {
1942         error_report("CMD_POSTCOPY_RAM_DISCARD Failed to read RAMBlock ID");
1943         return -1;
1944     }
1945     tmp = qemu_get_byte(mis->from_src_file);
1946     if (tmp != 0) {
1947         error_report("CMD_POSTCOPY_RAM_DISCARD missing nil (%d)", tmp);
1948         return -1;
1949     }
1950 
1951     len -= 3 + strlen(ramid);
1952     if (len % 16) {
1953         error_report("CMD_POSTCOPY_RAM_DISCARD invalid length (%d)", len);
1954         return -1;
1955     }
1956     trace_loadvm_postcopy_ram_handle_discard_header(ramid, len);
1957     while (len) {
1958         uint64_t start_addr, block_length;
1959         start_addr = qemu_get_be64(mis->from_src_file);
1960         block_length = qemu_get_be64(mis->from_src_file);
1961 
1962         len -= 16;
1963         int ret = ram_discard_range(ramid, start_addr, block_length);
1964         if (ret) {
1965             return ret;
1966         }
1967     }
1968     trace_loadvm_postcopy_ram_handle_discard_end();
1969 
1970     return 0;
1971 }
1972 
1973 /*
1974  * Triggered by a postcopy_listen command; this thread takes over reading
1975  * the input stream, leaving the main thread free to carry on loading the rest
1976  * of the device state (from RAM).
1977  * (TODO:This could do with being in a postcopy file - but there again it's
1978  * just another input loop, not that postcopy specific)
1979  */
1980 static void *postcopy_ram_listen_thread(void *opaque)
1981 {
1982     MigrationIncomingState *mis = migration_incoming_get_current();
1983     QEMUFile *f = mis->from_src_file;
1984     int load_res;
1985     MigrationState *migr = migrate_get_current();
1986 
1987     object_ref(OBJECT(migr));
1988 
1989     migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE,
1990                                    MIGRATION_STATUS_POSTCOPY_ACTIVE);
1991     qemu_sem_post(&mis->thread_sync_sem);
1992     trace_postcopy_ram_listen_thread_start();
1993 
1994     rcu_register_thread();
1995     /*
1996      * Because we're a thread and not a coroutine we can't yield
1997      * in qemu_file, and thus we must be blocking now.
1998      */
1999     qemu_file_set_blocking(f, true);
2000     load_res = qemu_loadvm_state_main(f, mis);
2001 
2002     /*
2003      * This is tricky, but, mis->from_src_file can change after it
2004      * returns, when postcopy recovery happened. In the future, we may
2005      * want a wrapper for the QEMUFile handle.
2006      */
2007     f = mis->from_src_file;
2008 
2009     /* And non-blocking again so we don't block in any cleanup */
2010     qemu_file_set_blocking(f, false);
2011 
2012     trace_postcopy_ram_listen_thread_exit();
2013     if (load_res < 0) {
2014         qemu_file_set_error(f, load_res);
2015         dirty_bitmap_mig_cancel_incoming();
2016         if (postcopy_state_get() == POSTCOPY_INCOMING_RUNNING &&
2017             !migrate_postcopy_ram() && migrate_dirty_bitmaps())
2018         {
2019             error_report("%s: loadvm failed during postcopy: %d. All states "
2020                          "are migrated except dirty bitmaps. Some dirty "
2021                          "bitmaps may be lost, and present migrated dirty "
2022                          "bitmaps are correctly migrated and valid.",
2023                          __func__, load_res);
2024             load_res = 0; /* prevent further exit() */
2025         } else {
2026             error_report("%s: loadvm failed: %d", __func__, load_res);
2027             migrate_set_state(&mis->state, MIGRATION_STATUS_POSTCOPY_ACTIVE,
2028                                            MIGRATION_STATUS_FAILED);
2029         }
2030     }
2031     if (load_res >= 0) {
2032         /*
2033          * This looks good, but it's possible that the device loading in the
2034          * main thread hasn't finished yet, and so we might not be in 'RUN'
2035          * state yet; wait for the end of the main thread.
2036          */
2037         qemu_event_wait(&mis->main_thread_load_event);
2038     }
2039     postcopy_ram_incoming_cleanup(mis);
2040 
2041     if (load_res < 0) {
2042         /*
2043          * If something went wrong then we have a bad state so exit;
2044          * depending how far we got it might be possible at this point
2045          * to leave the guest running and fire MCEs for pages that never
2046          * arrived as a desperate recovery step.
2047          */
2048         rcu_unregister_thread();
2049         exit(EXIT_FAILURE);
2050     }
2051 
2052     migrate_set_state(&mis->state, MIGRATION_STATUS_POSTCOPY_ACTIVE,
2053                                    MIGRATION_STATUS_COMPLETED);
2054     /*
2055      * If everything has worked fine, then the main thread has waited
2056      * for us to start, and we're the last use of the mis.
2057      * (If something broke then qemu will have to exit anyway since it's
2058      * got a bad migration state).
2059      */
2060     migration_incoming_state_destroy();
2061 
2062     rcu_unregister_thread();
2063     mis->have_listen_thread = false;
2064     postcopy_state_set(POSTCOPY_INCOMING_END);
2065 
2066     object_unref(OBJECT(migr));
2067 
2068     return NULL;
2069 }
2070 
2071 /* After this message we must be able to immediately receive postcopy data */
2072 static int loadvm_postcopy_handle_listen(MigrationIncomingState *mis)
2073 {
2074     PostcopyState ps = postcopy_state_set(POSTCOPY_INCOMING_LISTENING);
2075     Error *local_err = NULL;
2076 
2077     trace_loadvm_postcopy_handle_listen("enter");
2078 
2079     if (ps != POSTCOPY_INCOMING_ADVISE && ps != POSTCOPY_INCOMING_DISCARD) {
2080         error_report("CMD_POSTCOPY_LISTEN in wrong postcopy state (%d)", ps);
2081         return -1;
2082     }
2083     if (ps == POSTCOPY_INCOMING_ADVISE) {
2084         /*
2085          * A rare case, we entered listen without having to do any discards,
2086          * so do the setup that's normally done at the time of the 1st discard.
2087          */
2088         if (migrate_postcopy_ram()) {
2089             postcopy_ram_prepare_discard(mis);
2090         }
2091     }
2092 
2093     trace_loadvm_postcopy_handle_listen("after discard");
2094 
2095     /*
2096      * Sensitise RAM - can now generate requests for blocks that don't exist
2097      * However, at this point the CPU shouldn't be running, and the IO
2098      * shouldn't be doing anything yet so don't actually expect requests
2099      */
2100     if (migrate_postcopy_ram()) {
2101         if (postcopy_ram_incoming_setup(mis)) {
2102             postcopy_ram_incoming_cleanup(mis);
2103             return -1;
2104         }
2105     }
2106 
2107     trace_loadvm_postcopy_handle_listen("after uffd");
2108 
2109     if (postcopy_notify(POSTCOPY_NOTIFY_INBOUND_LISTEN, &local_err)) {
2110         error_report_err(local_err);
2111         return -1;
2112     }
2113 
2114     mis->have_listen_thread = true;
2115     postcopy_thread_create(mis, &mis->listen_thread,
2116                            MIGRATION_THREAD_DST_LISTEN,
2117                            postcopy_ram_listen_thread, QEMU_THREAD_DETACHED);
2118     trace_loadvm_postcopy_handle_listen("return");
2119 
2120     return 0;
2121 }
2122 
2123 static void loadvm_postcopy_handle_run_bh(void *opaque)
2124 {
2125     MigrationIncomingState *mis = opaque;
2126 
2127     trace_vmstate_downtime_checkpoint("dst-postcopy-bh-enter");
2128 
2129     /* TODO we should move all of this lot into postcopy_ram.c or a shared code
2130      * in migration.c
2131      */
2132     cpu_synchronize_all_post_init();
2133 
2134     trace_vmstate_downtime_checkpoint("dst-postcopy-bh-cpu-synced");
2135 
2136     qemu_announce_self(&mis->announce_timer, migrate_announce_params());
2137 
2138     trace_vmstate_downtime_checkpoint("dst-postcopy-bh-announced");
2139 
2140     dirty_bitmap_mig_before_vm_start();
2141 
2142     if (autostart) {
2143         /*
2144          * Make sure all file formats throw away their mutable metadata.
2145          * If we get an error here, just don't restart the VM yet.
2146          */
2147         bool success = migration_block_activate(NULL);
2148 
2149         trace_vmstate_downtime_checkpoint("dst-postcopy-bh-cache-invalidated");
2150 
2151         if (success) {
2152             vm_start();
2153         }
2154     } else {
2155         /* leave it paused and let management decide when to start the CPU */
2156         runstate_set(RUN_STATE_PAUSED);
2157     }
2158 
2159     trace_vmstate_downtime_checkpoint("dst-postcopy-bh-vm-started");
2160 }
2161 
2162 /* After all discards we can start running and asking for pages */
2163 static int loadvm_postcopy_handle_run(MigrationIncomingState *mis)
2164 {
2165     PostcopyState ps = postcopy_state_get();
2166 
2167     trace_loadvm_postcopy_handle_run();
2168     if (ps != POSTCOPY_INCOMING_LISTENING) {
2169         error_report("CMD_POSTCOPY_RUN in wrong postcopy state (%d)", ps);
2170         return -1;
2171     }
2172 
2173     postcopy_state_set(POSTCOPY_INCOMING_RUNNING);
2174     migration_bh_schedule(loadvm_postcopy_handle_run_bh, mis);
2175 
2176     /* We need to finish reading the stream from the package
2177      * and also stop reading anything more from the stream that loaded the
2178      * package (since it's now being read by the listener thread).
2179      * LOADVM_QUIT will quit all the layers of nested loadvm loops.
2180      */
2181     return LOADVM_QUIT;
2182 }
2183 
2184 /* We must be with page_request_mutex held */
2185 static gboolean postcopy_sync_page_req(gpointer key, gpointer value,
2186                                        gpointer data)
2187 {
2188     MigrationIncomingState *mis = data;
2189     void *host_addr = (void *) key;
2190     ram_addr_t rb_offset;
2191     RAMBlock *rb;
2192     int ret;
2193 
2194     rb = qemu_ram_block_from_host(host_addr, true, &rb_offset);
2195     if (!rb) {
2196         /*
2197          * This should _never_ happen.  However be nice for a migrating VM to
2198          * not crash/assert.  Post an error (note: intended to not use *_once
2199          * because we do want to see all the illegal addresses; and this can
2200          * never be triggered by the guest so we're safe) and move on next.
2201          */
2202         error_report("%s: illegal host addr %p", __func__, host_addr);
2203         /* Try the next entry */
2204         return FALSE;
2205     }
2206 
2207     ret = migrate_send_rp_message_req_pages(mis, rb, rb_offset);
2208     if (ret) {
2209         /* Please refer to above comment. */
2210         error_report("%s: send rp message failed for addr %p",
2211                      __func__, host_addr);
2212         return FALSE;
2213     }
2214 
2215     trace_postcopy_page_req_sync(host_addr);
2216 
2217     return FALSE;
2218 }
2219 
2220 static void migrate_send_rp_req_pages_pending(MigrationIncomingState *mis)
2221 {
2222     WITH_QEMU_LOCK_GUARD(&mis->page_request_mutex) {
2223         g_tree_foreach(mis->page_requested, postcopy_sync_page_req, mis);
2224     }
2225 }
2226 
2227 static int loadvm_postcopy_handle_resume(MigrationIncomingState *mis)
2228 {
2229     if (mis->state != MIGRATION_STATUS_POSTCOPY_RECOVER) {
2230         error_report("%s: illegal resume received", __func__);
2231         /* Don't fail the load, only for this. */
2232         return 0;
2233     }
2234 
2235     /*
2236      * Reset the last_rb before we resend any page req to source again, since
2237      * the source should have it reset already.
2238      */
2239     mis->last_rb = NULL;
2240 
2241     /*
2242      * This means source VM is ready to resume the postcopy migration.
2243      */
2244     migrate_set_state(&mis->state, MIGRATION_STATUS_POSTCOPY_RECOVER,
2245                       MIGRATION_STATUS_POSTCOPY_ACTIVE);
2246 
2247     trace_loadvm_postcopy_handle_resume();
2248 
2249     /* Tell source that "we are ready" */
2250     migrate_send_rp_resume_ack(mis, MIGRATION_RESUME_ACK_VALUE);
2251 
2252     /*
2253      * After a postcopy recovery, the source should have lost the postcopy
2254      * queue, or potentially the requested pages could have been lost during
2255      * the network down phase.  Let's re-sync with the source VM by re-sending
2256      * all the pending pages that we eagerly need, so these threads won't get
2257      * blocked too long due to the recovery.
2258      *
2259      * Without this procedure, the faulted destination VM threads (waiting for
2260      * page requests right before the postcopy is interrupted) can keep hanging
2261      * until the pages are sent by the source during the background copying of
2262      * pages, or another thread faulted on the same address accidentally.
2263      */
2264     migrate_send_rp_req_pages_pending(mis);
2265 
2266     /*
2267      * It's time to switch state and release the fault thread to continue
2268      * service page faults.  Note that this should be explicitly after the
2269      * above call to migrate_send_rp_req_pages_pending().  In short:
2270      * migrate_send_rp_message_req_pages() is not thread safe, yet.
2271      */
2272     qemu_sem_post(&mis->postcopy_pause_sem_fault);
2273 
2274     if (migrate_postcopy_preempt()) {
2275         /*
2276          * The preempt channel will be created in async manner, now let's
2277          * wait for it and make sure it's created.
2278          */
2279         qemu_sem_wait(&mis->postcopy_qemufile_dst_done);
2280         assert(mis->postcopy_qemufile_dst);
2281         /* Kick the fast ram load thread too */
2282         qemu_sem_post(&mis->postcopy_pause_sem_fast_load);
2283     }
2284 
2285     return 0;
2286 }
2287 
2288 /**
2289  * Immediately following this command is a blob of data containing an embedded
2290  * chunk of migration stream; read it and load it.
2291  *
2292  * @mis: Incoming state
2293  * @length: Length of packaged data to read
2294  *
2295  * Returns: Negative values on error
2296  *
2297  */
2298 static int loadvm_handle_cmd_packaged(MigrationIncomingState *mis)
2299 {
2300     int ret;
2301     size_t length;
2302     QIOChannelBuffer *bioc;
2303 
2304     length = qemu_get_be32(mis->from_src_file);
2305     trace_loadvm_handle_cmd_packaged(length);
2306 
2307     if (length > MAX_VM_CMD_PACKAGED_SIZE) {
2308         error_report("Unreasonably large packaged state: %zu", length);
2309         return -1;
2310     }
2311 
2312     bioc = qio_channel_buffer_new(length);
2313     qio_channel_set_name(QIO_CHANNEL(bioc), "migration-loadvm-buffer");
2314     ret = qemu_get_buffer(mis->from_src_file,
2315                           bioc->data,
2316                           length);
2317     if (ret != length) {
2318         object_unref(OBJECT(bioc));
2319         error_report("CMD_PACKAGED: Buffer receive fail ret=%d length=%zu",
2320                      ret, length);
2321         return (ret < 0) ? ret : -EAGAIN;
2322     }
2323     bioc->usage += length;
2324     trace_loadvm_handle_cmd_packaged_received(ret);
2325 
2326     QEMUFile *packf = qemu_file_new_input(QIO_CHANNEL(bioc));
2327 
2328     /*
2329      * Before loading the guest states, ensure that the preempt channel has
2330      * been ready to use, as some of the states (e.g. via virtio_load) might
2331      * trigger page faults that will be handled through the preempt channel.
2332      * So yield to the main thread in the case that the channel create event
2333      * hasn't been dispatched.
2334      *
2335      * TODO: if we can move migration loadvm out of main thread, then we
2336      * won't block main thread from polling the accept() fds.  We can drop
2337      * this as a whole when that is done.
2338      */
2339     do {
2340         if (!migrate_postcopy_preempt() || !qemu_in_coroutine() ||
2341             mis->postcopy_qemufile_dst) {
2342             break;
2343         }
2344 
2345         aio_co_schedule(qemu_get_current_aio_context(), qemu_coroutine_self());
2346         qemu_coroutine_yield();
2347     } while (1);
2348 
2349     ret = qemu_loadvm_state_main(packf, mis);
2350     trace_loadvm_handle_cmd_packaged_main(ret);
2351     qemu_fclose(packf);
2352     object_unref(OBJECT(bioc));
2353 
2354     return ret;
2355 }
2356 
2357 /*
2358  * Handle request that source requests for recved_bitmap on
2359  * destination. Payload format:
2360  *
2361  * len (1 byte) + ramblock_name (<255 bytes)
2362  */
2363 static int loadvm_handle_recv_bitmap(MigrationIncomingState *mis,
2364                                      uint16_t len)
2365 {
2366     QEMUFile *file = mis->from_src_file;
2367     RAMBlock *rb;
2368     char block_name[256];
2369     size_t cnt;
2370 
2371     cnt = qemu_get_counted_string(file, block_name);
2372     if (!cnt) {
2373         error_report("%s: failed to read block name", __func__);
2374         return -EINVAL;
2375     }
2376 
2377     /* Validate before using the data */
2378     if (qemu_file_get_error(file)) {
2379         return qemu_file_get_error(file);
2380     }
2381 
2382     if (len != cnt + 1) {
2383         error_report("%s: invalid payload length (%d)", __func__, len);
2384         return -EINVAL;
2385     }
2386 
2387     rb = qemu_ram_block_by_name(block_name);
2388     if (!rb) {
2389         error_report("%s: block '%s' not found", __func__, block_name);
2390         return -EINVAL;
2391     }
2392 
2393     migrate_send_rp_recv_bitmap(mis, block_name);
2394 
2395     trace_loadvm_handle_recv_bitmap(block_name);
2396 
2397     return 0;
2398 }
2399 
2400 static int loadvm_process_enable_colo(MigrationIncomingState *mis)
2401 {
2402     int ret = migration_incoming_enable_colo();
2403 
2404     if (!ret) {
2405         ret = colo_init_ram_cache();
2406         if (ret) {
2407             migration_incoming_disable_colo();
2408         }
2409     }
2410     return ret;
2411 }
2412 
2413 /*
2414  * Process an incoming 'QEMU_VM_COMMAND'
2415  * 0           just a normal return
2416  * LOADVM_QUIT All good, but exit the loop
2417  * <0          Error
2418  */
2419 static int loadvm_process_command(QEMUFile *f)
2420 {
2421     MigrationIncomingState *mis = migration_incoming_get_current();
2422     uint16_t cmd;
2423     uint16_t len;
2424     uint32_t tmp32;
2425 
2426     cmd = qemu_get_be16(f);
2427     len = qemu_get_be16(f);
2428 
2429     /* Check validity before continue processing of cmds */
2430     if (qemu_file_get_error(f)) {
2431         return qemu_file_get_error(f);
2432     }
2433 
2434     if (cmd >= MIG_CMD_MAX || cmd == MIG_CMD_INVALID) {
2435         error_report("MIG_CMD 0x%x unknown (len 0x%x)", cmd, len);
2436         return -EINVAL;
2437     }
2438 
2439     trace_loadvm_process_command(mig_cmd_args[cmd].name, len);
2440 
2441     if (mig_cmd_args[cmd].len != -1 && mig_cmd_args[cmd].len != len) {
2442         error_report("%s received with bad length - expecting %zu, got %d",
2443                      mig_cmd_args[cmd].name,
2444                      (size_t)mig_cmd_args[cmd].len, len);
2445         return -ERANGE;
2446     }
2447 
2448     switch (cmd) {
2449     case MIG_CMD_OPEN_RETURN_PATH:
2450         if (mis->to_src_file) {
2451             error_report("CMD_OPEN_RETURN_PATH called when RP already open");
2452             /* Not really a problem, so don't give up */
2453             return 0;
2454         }
2455         mis->to_src_file = qemu_file_get_return_path(f);
2456         if (!mis->to_src_file) {
2457             error_report("CMD_OPEN_RETURN_PATH failed");
2458             return -1;
2459         }
2460 
2461         /*
2462          * Switchover ack is enabled but no device uses it, so send an ACK to
2463          * source that it's OK to switchover. Do it here, after return path has
2464          * been created.
2465          */
2466         if (migrate_switchover_ack() && !mis->switchover_ack_pending_num) {
2467             int ret = migrate_send_rp_switchover_ack(mis);
2468             if (ret) {
2469                 error_report(
2470                     "Could not send switchover ack RP MSG, err %d (%s)", ret,
2471                     strerror(-ret));
2472                 return ret;
2473             }
2474         }
2475         break;
2476 
2477     case MIG_CMD_PING:
2478         tmp32 = qemu_get_be32(f);
2479         trace_loadvm_process_command_ping(tmp32);
2480         if (!mis->to_src_file) {
2481             error_report("CMD_PING (0x%x) received with no return path",
2482                          tmp32);
2483             return -1;
2484         }
2485         migrate_send_rp_pong(mis, tmp32);
2486         break;
2487 
2488     case MIG_CMD_PACKAGED:
2489         return loadvm_handle_cmd_packaged(mis);
2490 
2491     case MIG_CMD_POSTCOPY_ADVISE:
2492         return loadvm_postcopy_handle_advise(mis, len);
2493 
2494     case MIG_CMD_POSTCOPY_LISTEN:
2495         return loadvm_postcopy_handle_listen(mis);
2496 
2497     case MIG_CMD_POSTCOPY_RUN:
2498         return loadvm_postcopy_handle_run(mis);
2499 
2500     case MIG_CMD_POSTCOPY_RAM_DISCARD:
2501         return loadvm_postcopy_ram_handle_discard(mis, len);
2502 
2503     case MIG_CMD_POSTCOPY_RESUME:
2504         return loadvm_postcopy_handle_resume(mis);
2505 
2506     case MIG_CMD_RECV_BITMAP:
2507         return loadvm_handle_recv_bitmap(mis, len);
2508 
2509     case MIG_CMD_ENABLE_COLO:
2510         return loadvm_process_enable_colo(mis);
2511     }
2512 
2513     return 0;
2514 }
2515 
2516 /*
2517  * Read a footer off the wire and check that it matches the expected section
2518  *
2519  * Returns: true if the footer was good
2520  *          false if there is a problem (and calls error_report to say why)
2521  */
2522 static bool check_section_footer(QEMUFile *f, SaveStateEntry *se)
2523 {
2524     int ret;
2525     uint8_t read_mark;
2526     uint32_t read_section_id;
2527 
2528     if (!migrate_get_current()->send_section_footer) {
2529         /* No footer to check */
2530         return true;
2531     }
2532 
2533     read_mark = qemu_get_byte(f);
2534 
2535     ret = qemu_file_get_error(f);
2536     if (ret) {
2537         error_report("%s: Read section footer failed: %d",
2538                      __func__, ret);
2539         return false;
2540     }
2541 
2542     if (read_mark != QEMU_VM_SECTION_FOOTER) {
2543         error_report("Missing section footer for %s", se->idstr);
2544         return false;
2545     }
2546 
2547     read_section_id = qemu_get_be32(f);
2548     if (read_section_id != se->load_section_id) {
2549         error_report("Mismatched section id in footer for %s -"
2550                      " read 0x%x expected 0x%x",
2551                      se->idstr, read_section_id, se->load_section_id);
2552         return false;
2553     }
2554 
2555     /* All good */
2556     return true;
2557 }
2558 
2559 static int
2560 qemu_loadvm_section_start_full(QEMUFile *f, uint8_t type)
2561 {
2562     bool trace_downtime = (type == QEMU_VM_SECTION_FULL);
2563     uint32_t instance_id, version_id, section_id;
2564     int64_t start_ts, end_ts;
2565     SaveStateEntry *se;
2566     char idstr[256];
2567     int ret;
2568 
2569     /* Read section start */
2570     section_id = qemu_get_be32(f);
2571     if (!qemu_get_counted_string(f, idstr)) {
2572         error_report("Unable to read ID string for section %u",
2573                      section_id);
2574         return -EINVAL;
2575     }
2576     instance_id = qemu_get_be32(f);
2577     version_id = qemu_get_be32(f);
2578 
2579     ret = qemu_file_get_error(f);
2580     if (ret) {
2581         error_report("%s: Failed to read instance/version ID: %d",
2582                      __func__, ret);
2583         return ret;
2584     }
2585 
2586     trace_qemu_loadvm_state_section_startfull(section_id, idstr,
2587             instance_id, version_id);
2588     /* Find savevm section */
2589     se = find_se(idstr, instance_id);
2590     if (se == NULL) {
2591         error_report("Unknown savevm section or instance '%s' %"PRIu32". "
2592                      "Make sure that your current VM setup matches your "
2593                      "saved VM setup, including any hotplugged devices",
2594                      idstr, instance_id);
2595         return -EINVAL;
2596     }
2597 
2598     /* Validate version */
2599     if (version_id > se->version_id) {
2600         error_report("savevm: unsupported version %d for '%s' v%d",
2601                      version_id, idstr, se->version_id);
2602         return -EINVAL;
2603     }
2604     se->load_version_id = version_id;
2605     se->load_section_id = section_id;
2606 
2607     /* Validate if it is a device's state */
2608     if (xen_enabled() && se->is_ram) {
2609         error_report("loadvm: %s RAM loading not allowed on Xen", idstr);
2610         return -EINVAL;
2611     }
2612 
2613     if (trace_downtime) {
2614         start_ts = qemu_clock_get_us(QEMU_CLOCK_REALTIME);
2615     }
2616 
2617     ret = vmstate_load(f, se);
2618     if (ret < 0) {
2619         error_report("error while loading state for instance 0x%"PRIx32" of"
2620                      " device '%s'", instance_id, idstr);
2621         return ret;
2622     }
2623 
2624     if (trace_downtime) {
2625         end_ts = qemu_clock_get_us(QEMU_CLOCK_REALTIME);
2626         trace_vmstate_downtime_load("non-iterable", se->idstr,
2627                                     se->instance_id, end_ts - start_ts);
2628     }
2629 
2630     if (!check_section_footer(f, se)) {
2631         return -EINVAL;
2632     }
2633 
2634     return 0;
2635 }
2636 
2637 static int
2638 qemu_loadvm_section_part_end(QEMUFile *f, uint8_t type)
2639 {
2640     bool trace_downtime = (type == QEMU_VM_SECTION_END);
2641     int64_t start_ts, end_ts;
2642     uint32_t section_id;
2643     SaveStateEntry *se;
2644     int ret;
2645 
2646     section_id = qemu_get_be32(f);
2647 
2648     ret = qemu_file_get_error(f);
2649     if (ret) {
2650         error_report("%s: Failed to read section ID: %d",
2651                      __func__, ret);
2652         return ret;
2653     }
2654 
2655     trace_qemu_loadvm_state_section_partend(section_id);
2656     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
2657         if (se->load_section_id == section_id) {
2658             break;
2659         }
2660     }
2661     if (se == NULL) {
2662         error_report("Unknown savevm section %d", section_id);
2663         return -EINVAL;
2664     }
2665 
2666     if (trace_downtime) {
2667         start_ts = qemu_clock_get_us(QEMU_CLOCK_REALTIME);
2668     }
2669 
2670     ret = vmstate_load(f, se);
2671     if (ret < 0) {
2672         error_report("error while loading state section id %d(%s)",
2673                      section_id, se->idstr);
2674         return ret;
2675     }
2676 
2677     if (trace_downtime) {
2678         end_ts = qemu_clock_get_us(QEMU_CLOCK_REALTIME);
2679         trace_vmstate_downtime_load("iterable", se->idstr,
2680                                     se->instance_id, end_ts - start_ts);
2681     }
2682 
2683     if (!check_section_footer(f, se)) {
2684         return -EINVAL;
2685     }
2686 
2687     return 0;
2688 }
2689 
2690 static int qemu_loadvm_state_header(QEMUFile *f)
2691 {
2692     unsigned int v;
2693     int ret;
2694 
2695     v = qemu_get_be32(f);
2696     if (v != QEMU_VM_FILE_MAGIC) {
2697         error_report("Not a migration stream");
2698         return -EINVAL;
2699     }
2700 
2701     v = qemu_get_be32(f);
2702     if (v == QEMU_VM_FILE_VERSION_COMPAT) {
2703         error_report("SaveVM v2 format is obsolete and don't work anymore");
2704         return -ENOTSUP;
2705     }
2706     if (v != QEMU_VM_FILE_VERSION) {
2707         error_report("Unsupported migration stream version");
2708         return -ENOTSUP;
2709     }
2710 
2711     if (migrate_get_current()->send_configuration) {
2712         if (qemu_get_byte(f) != QEMU_VM_CONFIGURATION) {
2713             error_report("Configuration section missing");
2714             return -EINVAL;
2715         }
2716         ret = vmstate_load_state(f, &vmstate_configuration, &savevm_state, 0);
2717 
2718         if (ret) {
2719             return ret;
2720         }
2721     }
2722     return 0;
2723 }
2724 
2725 static void qemu_loadvm_state_switchover_ack_needed(MigrationIncomingState *mis)
2726 {
2727     SaveStateEntry *se;
2728 
2729     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
2730         if (!se->ops || !se->ops->switchover_ack_needed) {
2731             continue;
2732         }
2733 
2734         if (se->ops->switchover_ack_needed(se->opaque)) {
2735             mis->switchover_ack_pending_num++;
2736         }
2737     }
2738 
2739     trace_loadvm_state_switchover_ack_needed(mis->switchover_ack_pending_num);
2740 }
2741 
2742 static int qemu_loadvm_state_setup(QEMUFile *f, Error **errp)
2743 {
2744     ERRP_GUARD();
2745     SaveStateEntry *se;
2746     int ret;
2747 
2748     trace_loadvm_state_setup();
2749     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
2750         if (!se->ops || !se->ops->load_setup) {
2751             continue;
2752         }
2753         if (se->ops->is_active) {
2754             if (!se->ops->is_active(se->opaque)) {
2755                 continue;
2756             }
2757         }
2758 
2759         ret = se->ops->load_setup(f, se->opaque, errp);
2760         if (ret < 0) {
2761             error_prepend(errp, "Load state of device %s failed: ",
2762                           se->idstr);
2763             qemu_file_set_error(f, ret);
2764             return ret;
2765         }
2766     }
2767     return 0;
2768 }
2769 
2770 void qemu_loadvm_state_cleanup(void)
2771 {
2772     SaveStateEntry *se;
2773 
2774     trace_loadvm_state_cleanup();
2775     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
2776         if (se->ops && se->ops->load_cleanup) {
2777             se->ops->load_cleanup(se->opaque);
2778         }
2779     }
2780 }
2781 
2782 /* Return true if we should continue the migration, or false. */
2783 static bool postcopy_pause_incoming(MigrationIncomingState *mis)
2784 {
2785     int i;
2786 
2787     trace_postcopy_pause_incoming();
2788 
2789     assert(migrate_postcopy_ram());
2790 
2791     /*
2792      * Unregister yank with either from/to src would work, since ioc behind it
2793      * is the same
2794      */
2795     migration_ioc_unregister_yank_from_file(mis->from_src_file);
2796 
2797     assert(mis->from_src_file);
2798     qemu_file_shutdown(mis->from_src_file);
2799     qemu_fclose(mis->from_src_file);
2800     mis->from_src_file = NULL;
2801 
2802     assert(mis->to_src_file);
2803     qemu_file_shutdown(mis->to_src_file);
2804     qemu_mutex_lock(&mis->rp_mutex);
2805     qemu_fclose(mis->to_src_file);
2806     mis->to_src_file = NULL;
2807     qemu_mutex_unlock(&mis->rp_mutex);
2808 
2809     /*
2810      * NOTE: this must happen before reset the PostcopyTmpPages below,
2811      * otherwise it's racy to reset those fields when the fast load thread
2812      * can be accessing it in parallel.
2813      */
2814     if (mis->postcopy_qemufile_dst) {
2815         qemu_file_shutdown(mis->postcopy_qemufile_dst);
2816         /* Take the mutex to make sure the fast ram load thread halted */
2817         qemu_mutex_lock(&mis->postcopy_prio_thread_mutex);
2818         migration_ioc_unregister_yank_from_file(mis->postcopy_qemufile_dst);
2819         qemu_fclose(mis->postcopy_qemufile_dst);
2820         mis->postcopy_qemufile_dst = NULL;
2821         qemu_mutex_unlock(&mis->postcopy_prio_thread_mutex);
2822     }
2823 
2824     /* Current state can be either ACTIVE or RECOVER */
2825     migrate_set_state(&mis->state, mis->state,
2826                       MIGRATION_STATUS_POSTCOPY_PAUSED);
2827 
2828     /* Notify the fault thread for the invalidated file handle */
2829     postcopy_fault_thread_notify(mis);
2830 
2831     /*
2832      * If network is interrupted, any temp page we received will be useless
2833      * because we didn't mark them as "received" in receivedmap.  After a
2834      * proper recovery later (which will sync src dirty bitmap with receivedmap
2835      * on dest) these cached small pages will be resent again.
2836      */
2837     for (i = 0; i < mis->postcopy_channels; i++) {
2838         postcopy_temp_page_reset(&mis->postcopy_tmp_pages[i]);
2839     }
2840 
2841     error_report("Detected IO failure for postcopy. "
2842                  "Migration paused.");
2843 
2844     do {
2845         qemu_sem_wait(&mis->postcopy_pause_sem_dst);
2846     } while (postcopy_is_paused(mis->state));
2847 
2848     trace_postcopy_pause_incoming_continued();
2849 
2850     return true;
2851 }
2852 
2853 int qemu_loadvm_state_main(QEMUFile *f, MigrationIncomingState *mis)
2854 {
2855     uint8_t section_type;
2856     int ret = 0;
2857 
2858 retry:
2859     while (true) {
2860         section_type = qemu_get_byte(f);
2861 
2862         ret = qemu_file_get_error_obj_any(f, mis->postcopy_qemufile_dst, NULL);
2863         if (ret) {
2864             break;
2865         }
2866 
2867         trace_qemu_loadvm_state_section(section_type);
2868         switch (section_type) {
2869         case QEMU_VM_SECTION_START:
2870         case QEMU_VM_SECTION_FULL:
2871             ret = qemu_loadvm_section_start_full(f, section_type);
2872             if (ret < 0) {
2873                 goto out;
2874             }
2875             break;
2876         case QEMU_VM_SECTION_PART:
2877         case QEMU_VM_SECTION_END:
2878             ret = qemu_loadvm_section_part_end(f, section_type);
2879             if (ret < 0) {
2880                 goto out;
2881             }
2882             break;
2883         case QEMU_VM_COMMAND:
2884             ret = loadvm_process_command(f);
2885             trace_qemu_loadvm_state_section_command(ret);
2886             if ((ret < 0) || (ret == LOADVM_QUIT)) {
2887                 goto out;
2888             }
2889             break;
2890         case QEMU_VM_EOF:
2891             /* This is the end of migration */
2892             goto out;
2893         default:
2894             error_report("Unknown savevm section type %d", section_type);
2895             ret = -EINVAL;
2896             goto out;
2897         }
2898     }
2899 
2900 out:
2901     if (ret < 0) {
2902         qemu_file_set_error(f, ret);
2903 
2904         /* Cancel bitmaps incoming regardless of recovery */
2905         dirty_bitmap_mig_cancel_incoming();
2906 
2907         /*
2908          * If we are during an active postcopy, then we pause instead
2909          * of bail out to at least keep the VM's dirty data.  Note
2910          * that POSTCOPY_INCOMING_LISTENING stage is still not enough,
2911          * during which we're still receiving device states and we
2912          * still haven't yet started the VM on destination.
2913          *
2914          * Only RAM postcopy supports recovery. Still, if RAM postcopy is
2915          * enabled, canceled bitmaps postcopy will not affect RAM postcopy
2916          * recovering.
2917          */
2918         if (postcopy_state_get() == POSTCOPY_INCOMING_RUNNING &&
2919             migrate_postcopy_ram() && postcopy_pause_incoming(mis)) {
2920             /* Reset f to point to the newly created channel */
2921             f = mis->from_src_file;
2922             goto retry;
2923         }
2924     }
2925     return ret;
2926 }
2927 
2928 int qemu_loadvm_state(QEMUFile *f)
2929 {
2930     MigrationIncomingState *mis = migration_incoming_get_current();
2931     Error *local_err = NULL;
2932     int ret;
2933 
2934     if (qemu_savevm_state_blocked(&local_err)) {
2935         error_report_err(local_err);
2936         return -EINVAL;
2937     }
2938 
2939     ret = qemu_loadvm_state_header(f);
2940     if (ret) {
2941         return ret;
2942     }
2943 
2944     if (qemu_loadvm_state_setup(f, &local_err) != 0) {
2945         error_report_err(local_err);
2946         return -EINVAL;
2947     }
2948 
2949     if (migrate_switchover_ack()) {
2950         qemu_loadvm_state_switchover_ack_needed(mis);
2951     }
2952 
2953     cpu_synchronize_all_pre_loadvm();
2954 
2955     ret = qemu_loadvm_state_main(f, mis);
2956     qemu_event_set(&mis->main_thread_load_event);
2957 
2958     trace_qemu_loadvm_state_post_main(ret);
2959 
2960     if (mis->have_listen_thread) {
2961         /*
2962          * Postcopy listen thread still going, don't synchronize the
2963          * cpus yet.
2964          */
2965         return ret;
2966     }
2967 
2968     if (ret == 0) {
2969         ret = qemu_file_get_error(f);
2970     }
2971 
2972     /*
2973      * Try to read in the VMDESC section as well, so that dumping tools that
2974      * intercept our migration stream have the chance to see it.
2975      */
2976 
2977     /* We've got to be careful; if we don't read the data and just shut the fd
2978      * then the sender can error if we close while it's still sending.
2979      * We also mustn't read data that isn't there; some transports (RDMA)
2980      * will stall waiting for that data when the source has already closed.
2981      */
2982     if (ret == 0 && should_send_vmdesc()) {
2983         uint8_t *buf;
2984         uint32_t size;
2985         uint8_t  section_type = qemu_get_byte(f);
2986 
2987         if (section_type != QEMU_VM_VMDESCRIPTION) {
2988             error_report("Expected vmdescription section, but got %d",
2989                          section_type);
2990             /*
2991              * It doesn't seem worth failing at this point since
2992              * we apparently have an otherwise valid VM state
2993              */
2994         } else {
2995             buf = g_malloc(0x1000);
2996             size = qemu_get_be32(f);
2997 
2998             while (size > 0) {
2999                 uint32_t read_chunk = MIN(size, 0x1000);
3000                 qemu_get_buffer(f, buf, read_chunk);
3001                 size -= read_chunk;
3002             }
3003             g_free(buf);
3004         }
3005     }
3006 
3007     cpu_synchronize_all_post_init();
3008 
3009     return ret;
3010 }
3011 
3012 int qemu_load_device_state(QEMUFile *f)
3013 {
3014     MigrationIncomingState *mis = migration_incoming_get_current();
3015     int ret;
3016 
3017     /* Load QEMU_VM_SECTION_FULL section */
3018     ret = qemu_loadvm_state_main(f, mis);
3019     if (ret < 0) {
3020         error_report("Failed to load device state: %d", ret);
3021         return ret;
3022     }
3023 
3024     cpu_synchronize_all_post_init();
3025     return 0;
3026 }
3027 
3028 int qemu_loadvm_approve_switchover(void)
3029 {
3030     MigrationIncomingState *mis = migration_incoming_get_current();
3031 
3032     if (!mis->switchover_ack_pending_num) {
3033         return -EINVAL;
3034     }
3035 
3036     mis->switchover_ack_pending_num--;
3037     trace_loadvm_approve_switchover(mis->switchover_ack_pending_num);
3038 
3039     if (mis->switchover_ack_pending_num) {
3040         return 0;
3041     }
3042 
3043     return migrate_send_rp_switchover_ack(mis);
3044 }
3045 
3046 bool save_snapshot(const char *name, bool overwrite, const char *vmstate,
3047                   bool has_devices, strList *devices, Error **errp)
3048 {
3049     BlockDriverState *bs;
3050     QEMUSnapshotInfo sn1, *sn = &sn1;
3051     int ret = -1, ret2;
3052     QEMUFile *f;
3053     RunState saved_state = runstate_get();
3054     uint64_t vm_state_size;
3055     g_autoptr(GDateTime) now = g_date_time_new_now_local();
3056 
3057     GLOBAL_STATE_CODE();
3058 
3059     if (migration_is_blocked(errp)) {
3060         return false;
3061     }
3062 
3063     if (!replay_can_snapshot()) {
3064         error_setg(errp, "Record/replay does not allow making snapshot "
3065                    "right now. Try once more later.");
3066         return false;
3067     }
3068 
3069     if (!bdrv_all_can_snapshot(has_devices, devices, errp)) {
3070         return false;
3071     }
3072 
3073     /* Delete old snapshots of the same name */
3074     if (name) {
3075         if (overwrite) {
3076             if (bdrv_all_delete_snapshot(name, has_devices,
3077                                          devices, errp) < 0) {
3078                 return false;
3079             }
3080         } else {
3081             ret2 = bdrv_all_has_snapshot(name, has_devices, devices, errp);
3082             if (ret2 < 0) {
3083                 return false;
3084             }
3085             if (ret2 == 1) {
3086                 error_setg(errp,
3087                            "Snapshot '%s' already exists in one or more devices",
3088                            name);
3089                 return false;
3090             }
3091         }
3092     }
3093 
3094     bs = bdrv_all_find_vmstate_bs(vmstate, has_devices, devices, errp);
3095     if (bs == NULL) {
3096         return false;
3097     }
3098 
3099     global_state_store();
3100     vm_stop(RUN_STATE_SAVE_VM);
3101 
3102     bdrv_drain_all_begin();
3103 
3104     memset(sn, 0, sizeof(*sn));
3105 
3106     /* fill auxiliary fields */
3107     sn->date_sec = g_date_time_to_unix(now);
3108     sn->date_nsec = g_date_time_get_microsecond(now) * 1000;
3109     sn->vm_clock_nsec = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL);
3110     if (replay_mode != REPLAY_MODE_NONE) {
3111         sn->icount = replay_get_current_icount();
3112     } else {
3113         sn->icount = -1ULL;
3114     }
3115 
3116     if (name) {
3117         pstrcpy(sn->name, sizeof(sn->name), name);
3118     } else {
3119         g_autofree char *autoname = g_date_time_format(now,  "vm-%Y%m%d%H%M%S");
3120         pstrcpy(sn->name, sizeof(sn->name), autoname);
3121     }
3122 
3123     /* save the VM state */
3124     f = qemu_fopen_bdrv(bs, 1);
3125     if (!f) {
3126         error_setg(errp, "Could not open VM state file");
3127         goto the_end;
3128     }
3129     ret = qemu_savevm_state(f, errp);
3130     vm_state_size = qemu_file_transferred(f);
3131     ret2 = qemu_fclose(f);
3132     if (ret < 0) {
3133         goto the_end;
3134     }
3135     if (ret2 < 0) {
3136         ret = ret2;
3137         goto the_end;
3138     }
3139 
3140     ret = bdrv_all_create_snapshot(sn, bs, vm_state_size,
3141                                    has_devices, devices, errp);
3142     if (ret < 0) {
3143         bdrv_all_delete_snapshot(sn->name, has_devices, devices, NULL);
3144         goto the_end;
3145     }
3146 
3147     ret = 0;
3148 
3149  the_end:
3150     bdrv_drain_all_end();
3151 
3152     vm_resume(saved_state);
3153     return ret == 0;
3154 }
3155 
3156 void qmp_xen_save_devices_state(const char *filename, bool has_live, bool live,
3157                                 Error **errp)
3158 {
3159     QEMUFile *f;
3160     QIOChannelFile *ioc;
3161     int saved_vm_running;
3162     int ret;
3163 
3164     if (!has_live) {
3165         /* live default to true so old version of Xen tool stack can have a
3166          * successful live migration */
3167         live = true;
3168     }
3169 
3170     saved_vm_running = runstate_is_running();
3171     vm_stop(RUN_STATE_SAVE_VM);
3172     global_state_store_running();
3173 
3174     ioc = qio_channel_file_new_path(filename, O_WRONLY | O_CREAT | O_TRUNC,
3175                                     0660, errp);
3176     if (!ioc) {
3177         goto the_end;
3178     }
3179     qio_channel_set_name(QIO_CHANNEL(ioc), "migration-xen-save-state");
3180     f = qemu_file_new_output(QIO_CHANNEL(ioc));
3181     object_unref(OBJECT(ioc));
3182     ret = qemu_save_device_state(f);
3183     if (ret < 0 || qemu_fclose(f) < 0) {
3184         error_setg(errp, "saving Xen device state failed");
3185     } else {
3186         /* libxl calls the QMP command "stop" before calling
3187          * "xen-save-devices-state" and in case of migration failure, libxl
3188          * would call "cont".
3189          * So call bdrv_inactivate_all (release locks) here to let the other
3190          * side of the migration take control of the images.
3191          */
3192         if (live && !saved_vm_running) {
3193             migration_block_inactivate();
3194         }
3195     }
3196 
3197  the_end:
3198     if (saved_vm_running) {
3199         vm_start();
3200     }
3201 }
3202 
3203 void qmp_xen_load_devices_state(const char *filename, Error **errp)
3204 {
3205     QEMUFile *f;
3206     QIOChannelFile *ioc;
3207     int ret;
3208 
3209     /* Guest must be paused before loading the device state; the RAM state
3210      * will already have been loaded by xc
3211      */
3212     if (runstate_is_running()) {
3213         error_setg(errp, "Cannot update device state while vm is running");
3214         return;
3215     }
3216     vm_stop(RUN_STATE_RESTORE_VM);
3217 
3218     ioc = qio_channel_file_new_path(filename, O_RDONLY | O_BINARY, 0, errp);
3219     if (!ioc) {
3220         return;
3221     }
3222     qio_channel_set_name(QIO_CHANNEL(ioc), "migration-xen-load-state");
3223     f = qemu_file_new_input(QIO_CHANNEL(ioc));
3224     object_unref(OBJECT(ioc));
3225 
3226     ret = qemu_loadvm_state(f);
3227     qemu_fclose(f);
3228     if (ret < 0) {
3229         error_setg(errp, "loading Xen device state failed");
3230     }
3231     migration_incoming_state_destroy();
3232 }
3233 
3234 bool load_snapshot(const char *name, const char *vmstate,
3235                    bool has_devices, strList *devices, Error **errp)
3236 {
3237     BlockDriverState *bs_vm_state;
3238     QEMUSnapshotInfo sn;
3239     QEMUFile *f;
3240     int ret;
3241     MigrationIncomingState *mis = migration_incoming_get_current();
3242 
3243     if (!bdrv_all_can_snapshot(has_devices, devices, errp)) {
3244         return false;
3245     }
3246     ret = bdrv_all_has_snapshot(name, has_devices, devices, errp);
3247     if (ret < 0) {
3248         return false;
3249     }
3250     if (ret == 0) {
3251         error_setg(errp, "Snapshot '%s' does not exist in one or more devices",
3252                    name);
3253         return false;
3254     }
3255 
3256     bs_vm_state = bdrv_all_find_vmstate_bs(vmstate, has_devices, devices, errp);
3257     if (!bs_vm_state) {
3258         return false;
3259     }
3260 
3261     /* Don't even try to load empty VM states */
3262     ret = bdrv_snapshot_find(bs_vm_state, &sn, name);
3263     if (ret < 0) {
3264         error_setg(errp, "Snapshot can not be found");
3265         return false;
3266     } else if (sn.vm_state_size == 0) {
3267         error_setg(errp, "This is a disk-only snapshot. Revert to it "
3268                    " offline using qemu-img");
3269         return false;
3270     }
3271 
3272     /*
3273      * Flush the record/replay queue. Now the VM state is going
3274      * to change. Therefore we don't need to preserve its consistency
3275      */
3276     replay_flush_events();
3277 
3278     /* Flush all IO requests so they don't interfere with the new state.  */
3279     bdrv_drain_all_begin();
3280 
3281     ret = bdrv_all_goto_snapshot(name, has_devices, devices, errp);
3282     if (ret < 0) {
3283         goto err_drain;
3284     }
3285 
3286     /* restore the VM state */
3287     f = qemu_fopen_bdrv(bs_vm_state, 0);
3288     if (!f) {
3289         error_setg(errp, "Could not open VM state file");
3290         goto err_drain;
3291     }
3292 
3293     qemu_system_reset(SHUTDOWN_CAUSE_SNAPSHOT_LOAD);
3294     mis->from_src_file = f;
3295 
3296     if (!yank_register_instance(MIGRATION_YANK_INSTANCE, errp)) {
3297         ret = -EINVAL;
3298         goto err_drain;
3299     }
3300     ret = qemu_loadvm_state(f);
3301     migration_incoming_state_destroy();
3302 
3303     bdrv_drain_all_end();
3304 
3305     if (ret < 0) {
3306         error_setg(errp, "Error %d while loading VM state", ret);
3307         return false;
3308     }
3309 
3310     return true;
3311 
3312 err_drain:
3313     bdrv_drain_all_end();
3314     return false;
3315 }
3316 
3317 void load_snapshot_resume(RunState state)
3318 {
3319     vm_resume(state);
3320     if (state == RUN_STATE_RUNNING && runstate_get() == RUN_STATE_SUSPENDED) {
3321         qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER, &error_abort);
3322     }
3323 }
3324 
3325 bool delete_snapshot(const char *name, bool has_devices,
3326                      strList *devices, Error **errp)
3327 {
3328     if (!bdrv_all_can_snapshot(has_devices, devices, errp)) {
3329         return false;
3330     }
3331 
3332     if (bdrv_all_delete_snapshot(name, has_devices, devices, errp) < 0) {
3333         return false;
3334     }
3335 
3336     return true;
3337 }
3338 
3339 void vmstate_register_ram(MemoryRegion *mr, DeviceState *dev)
3340 {
3341     qemu_ram_set_idstr(mr->ram_block,
3342                        memory_region_name(mr), dev);
3343     qemu_ram_set_migratable(mr->ram_block);
3344 }
3345 
3346 void vmstate_unregister_ram(MemoryRegion *mr, DeviceState *dev)
3347 {
3348     qemu_ram_unset_idstr(mr->ram_block);
3349     qemu_ram_unset_migratable(mr->ram_block);
3350 }
3351 
3352 void vmstate_register_ram_global(MemoryRegion *mr)
3353 {
3354     vmstate_register_ram(mr, NULL);
3355 }
3356 
3357 bool vmstate_check_only_migratable(const VMStateDescription *vmsd)
3358 {
3359     /* check needed if --only-migratable is specified */
3360     if (!only_migratable) {
3361         return true;
3362     }
3363 
3364     return !(vmsd && vmsd->unmigratable);
3365 }
3366 
3367 typedef struct SnapshotJob {
3368     Job common;
3369     char *tag;
3370     char *vmstate;
3371     strList *devices;
3372     Coroutine *co;
3373     Error **errp;
3374     bool ret;
3375 } SnapshotJob;
3376 
3377 static void qmp_snapshot_job_free(SnapshotJob *s)
3378 {
3379     g_free(s->tag);
3380     g_free(s->vmstate);
3381     qapi_free_strList(s->devices);
3382 }
3383 
3384 
3385 static void snapshot_load_job_bh(void *opaque)
3386 {
3387     Job *job = opaque;
3388     SnapshotJob *s = container_of(job, SnapshotJob, common);
3389     RunState orig_state = runstate_get();
3390 
3391     job_progress_set_remaining(&s->common, 1);
3392 
3393     vm_stop(RUN_STATE_RESTORE_VM);
3394 
3395     s->ret = load_snapshot(s->tag, s->vmstate, true, s->devices, s->errp);
3396     if (s->ret) {
3397         load_snapshot_resume(orig_state);
3398     }
3399 
3400     job_progress_update(&s->common, 1);
3401 
3402     qmp_snapshot_job_free(s);
3403     aio_co_wake(s->co);
3404 }
3405 
3406 static void snapshot_save_job_bh(void *opaque)
3407 {
3408     Job *job = opaque;
3409     SnapshotJob *s = container_of(job, SnapshotJob, common);
3410 
3411     job_progress_set_remaining(&s->common, 1);
3412     s->ret = save_snapshot(s->tag, false, s->vmstate,
3413                            true, s->devices, s->errp);
3414     job_progress_update(&s->common, 1);
3415 
3416     qmp_snapshot_job_free(s);
3417     aio_co_wake(s->co);
3418 }
3419 
3420 static void snapshot_delete_job_bh(void *opaque)
3421 {
3422     Job *job = opaque;
3423     SnapshotJob *s = container_of(job, SnapshotJob, common);
3424 
3425     job_progress_set_remaining(&s->common, 1);
3426     s->ret = delete_snapshot(s->tag, true, s->devices, s->errp);
3427     job_progress_update(&s->common, 1);
3428 
3429     qmp_snapshot_job_free(s);
3430     aio_co_wake(s->co);
3431 }
3432 
3433 static int coroutine_fn snapshot_save_job_run(Job *job, Error **errp)
3434 {
3435     SnapshotJob *s = container_of(job, SnapshotJob, common);
3436     s->errp = errp;
3437     s->co = qemu_coroutine_self();
3438     aio_bh_schedule_oneshot(qemu_get_aio_context(),
3439                             snapshot_save_job_bh, job);
3440     qemu_coroutine_yield();
3441     return s->ret ? 0 : -1;
3442 }
3443 
3444 static int coroutine_fn snapshot_load_job_run(Job *job, Error **errp)
3445 {
3446     SnapshotJob *s = container_of(job, SnapshotJob, common);
3447     s->errp = errp;
3448     s->co = qemu_coroutine_self();
3449     aio_bh_schedule_oneshot(qemu_get_aio_context(),
3450                             snapshot_load_job_bh, job);
3451     qemu_coroutine_yield();
3452     return s->ret ? 0 : -1;
3453 }
3454 
3455 static int coroutine_fn snapshot_delete_job_run(Job *job, Error **errp)
3456 {
3457     SnapshotJob *s = container_of(job, SnapshotJob, common);
3458     s->errp = errp;
3459     s->co = qemu_coroutine_self();
3460     aio_bh_schedule_oneshot(qemu_get_aio_context(),
3461                             snapshot_delete_job_bh, job);
3462     qemu_coroutine_yield();
3463     return s->ret ? 0 : -1;
3464 }
3465 
3466 
3467 static const JobDriver snapshot_load_job_driver = {
3468     .instance_size = sizeof(SnapshotJob),
3469     .job_type      = JOB_TYPE_SNAPSHOT_LOAD,
3470     .run           = snapshot_load_job_run,
3471 };
3472 
3473 static const JobDriver snapshot_save_job_driver = {
3474     .instance_size = sizeof(SnapshotJob),
3475     .job_type      = JOB_TYPE_SNAPSHOT_SAVE,
3476     .run           = snapshot_save_job_run,
3477 };
3478 
3479 static const JobDriver snapshot_delete_job_driver = {
3480     .instance_size = sizeof(SnapshotJob),
3481     .job_type      = JOB_TYPE_SNAPSHOT_DELETE,
3482     .run           = snapshot_delete_job_run,
3483 };
3484 
3485 
3486 void qmp_snapshot_save(const char *job_id,
3487                        const char *tag,
3488                        const char *vmstate,
3489                        strList *devices,
3490                        Error **errp)
3491 {
3492     SnapshotJob *s;
3493 
3494     s = job_create(job_id, &snapshot_save_job_driver, NULL,
3495                    qemu_get_aio_context(), JOB_MANUAL_DISMISS,
3496                    NULL, NULL, errp);
3497     if (!s) {
3498         return;
3499     }
3500 
3501     s->tag = g_strdup(tag);
3502     s->vmstate = g_strdup(vmstate);
3503     s->devices = QAPI_CLONE(strList, devices);
3504 
3505     job_start(&s->common);
3506 }
3507 
3508 void qmp_snapshot_load(const char *job_id,
3509                        const char *tag,
3510                        const char *vmstate,
3511                        strList *devices,
3512                        Error **errp)
3513 {
3514     SnapshotJob *s;
3515 
3516     s = job_create(job_id, &snapshot_load_job_driver, NULL,
3517                    qemu_get_aio_context(), JOB_MANUAL_DISMISS,
3518                    NULL, NULL, errp);
3519     if (!s) {
3520         return;
3521     }
3522 
3523     s->tag = g_strdup(tag);
3524     s->vmstate = g_strdup(vmstate);
3525     s->devices = QAPI_CLONE(strList, devices);
3526 
3527     job_start(&s->common);
3528 }
3529 
3530 void qmp_snapshot_delete(const char *job_id,
3531                          const char *tag,
3532                          strList *devices,
3533                          Error **errp)
3534 {
3535     SnapshotJob *s;
3536 
3537     s = job_create(job_id, &snapshot_delete_job_driver, NULL,
3538                    qemu_get_aio_context(), JOB_MANUAL_DISMISS,
3539                    NULL, NULL, errp);
3540     if (!s) {
3541         return;
3542     }
3543 
3544     s->tag = g_strdup(tag);
3545     s->devices = QAPI_CLONE(strList, devices);
3546 
3547     job_start(&s->common);
3548 }
3549