xref: /openbmc/qemu/migration/savevm.c (revision d44f2f96)
1 /*
2  * QEMU System Emulator
3  *
4  * Copyright (c) 2003-2008 Fabrice Bellard
5  * Copyright (c) 2009-2015 Red Hat Inc
6  *
7  * Authors:
8  *  Juan Quintela <quintela@redhat.com>
9  *
10  * Permission is hereby granted, free of charge, to any person obtaining a copy
11  * of this software and associated documentation files (the "Software"), to deal
12  * in the Software without restriction, including without limitation the rights
13  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
14  * copies of the Software, and to permit persons to whom the Software is
15  * furnished to do so, subject to the following conditions:
16  *
17  * The above copyright notice and this permission notice shall be included in
18  * all copies or substantial portions of the Software.
19  *
20  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
21  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
22  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
23  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
24  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
25  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
26  * THE SOFTWARE.
27  */
28 
29 #include "qemu/osdep.h"
30 #include "hw/boards.h"
31 #include "net/net.h"
32 #include "migration.h"
33 #include "migration/snapshot.h"
34 #include "migration/vmstate.h"
35 #include "migration/misc.h"
36 #include "migration/register.h"
37 #include "migration/global_state.h"
38 #include "migration/channel-block.h"
39 #include "ram.h"
40 #include "qemu-file.h"
41 #include "savevm.h"
42 #include "postcopy-ram.h"
43 #include "qapi/error.h"
44 #include "qapi/qapi-commands-migration.h"
45 #include "qapi/qmp/json-writer.h"
46 #include "qapi/clone-visitor.h"
47 #include "qapi/qapi-builtin-visit.h"
48 #include "qapi/qmp/qerror.h"
49 #include "qemu/error-report.h"
50 #include "sysemu/cpus.h"
51 #include "exec/memory.h"
52 #include "exec/target_page.h"
53 #include "trace.h"
54 #include "qemu/iov.h"
55 #include "qemu/main-loop.h"
56 #include "block/snapshot.h"
57 #include "qemu/cutils.h"
58 #include "io/channel-buffer.h"
59 #include "io/channel-file.h"
60 #include "sysemu/replay.h"
61 #include "sysemu/runstate.h"
62 #include "sysemu/sysemu.h"
63 #include "sysemu/xen.h"
64 #include "migration/colo.h"
65 #include "qemu/bitmap.h"
66 #include "net/announce.h"
67 #include "qemu/yank.h"
68 #include "yank_functions.h"
69 
70 const unsigned int postcopy_ram_discard_version;
71 
72 /* Subcommands for QEMU_VM_COMMAND */
73 enum qemu_vm_cmd {
74     MIG_CMD_INVALID = 0,   /* Must be 0 */
75     MIG_CMD_OPEN_RETURN_PATH,  /* Tell the dest to open the Return path */
76     MIG_CMD_PING,              /* Request a PONG on the RP */
77 
78     MIG_CMD_POSTCOPY_ADVISE,       /* Prior to any page transfers, just
79                                       warn we might want to do PC */
80     MIG_CMD_POSTCOPY_LISTEN,       /* Start listening for incoming
81                                       pages as it's running. */
82     MIG_CMD_POSTCOPY_RUN,          /* Start execution */
83 
84     MIG_CMD_POSTCOPY_RAM_DISCARD,  /* A list of pages to discard that
85                                       were previously sent during
86                                       precopy but are dirty. */
87     MIG_CMD_PACKAGED,          /* Send a wrapped stream within this stream */
88     MIG_CMD_ENABLE_COLO,       /* Enable COLO */
89     MIG_CMD_POSTCOPY_RESUME,   /* resume postcopy on dest */
90     MIG_CMD_RECV_BITMAP,       /* Request for recved bitmap on dst */
91     MIG_CMD_MAX
92 };
93 
94 #define MAX_VM_CMD_PACKAGED_SIZE UINT32_MAX
95 static struct mig_cmd_args {
96     ssize_t     len; /* -1 = variable */
97     const char *name;
98 } mig_cmd_args[] = {
99     [MIG_CMD_INVALID]          = { .len = -1, .name = "INVALID" },
100     [MIG_CMD_OPEN_RETURN_PATH] = { .len =  0, .name = "OPEN_RETURN_PATH" },
101     [MIG_CMD_PING]             = { .len = sizeof(uint32_t), .name = "PING" },
102     [MIG_CMD_POSTCOPY_ADVISE]  = { .len = -1, .name = "POSTCOPY_ADVISE" },
103     [MIG_CMD_POSTCOPY_LISTEN]  = { .len =  0, .name = "POSTCOPY_LISTEN" },
104     [MIG_CMD_POSTCOPY_RUN]     = { .len =  0, .name = "POSTCOPY_RUN" },
105     [MIG_CMD_POSTCOPY_RAM_DISCARD] = {
106                                    .len = -1, .name = "POSTCOPY_RAM_DISCARD" },
107     [MIG_CMD_POSTCOPY_RESUME]  = { .len =  0, .name = "POSTCOPY_RESUME" },
108     [MIG_CMD_PACKAGED]         = { .len =  4, .name = "PACKAGED" },
109     [MIG_CMD_RECV_BITMAP]      = { .len = -1, .name = "RECV_BITMAP" },
110     [MIG_CMD_MAX]              = { .len = -1, .name = "MAX" },
111 };
112 
113 /* Note for MIG_CMD_POSTCOPY_ADVISE:
114  * The format of arguments is depending on postcopy mode:
115  * - postcopy RAM only
116  *   uint64_t host page size
117  *   uint64_t taget page size
118  *
119  * - postcopy RAM and postcopy dirty bitmaps
120  *   format is the same as for postcopy RAM only
121  *
122  * - postcopy dirty bitmaps only
123  *   Nothing. Command length field is 0.
124  *
125  * Be careful: adding a new postcopy entity with some other parameters should
126  * not break format self-description ability. Good way is to introduce some
127  * generic extendable format with an exception for two old entities.
128  */
129 
130 /***********************************************************/
131 /* savevm/loadvm support */
132 
133 static QEMUFile *qemu_fopen_bdrv(BlockDriverState *bs, int is_writable)
134 {
135     if (is_writable) {
136         return qemu_file_new_output(QIO_CHANNEL(qio_channel_block_new(bs)));
137     } else {
138         return qemu_file_new_input(QIO_CHANNEL(qio_channel_block_new(bs)));
139     }
140 }
141 
142 
143 /* QEMUFile timer support.
144  * Not in qemu-file.c to not add qemu-timer.c as dependency to qemu-file.c
145  */
146 
147 void timer_put(QEMUFile *f, QEMUTimer *ts)
148 {
149     uint64_t expire_time;
150 
151     expire_time = timer_expire_time_ns(ts);
152     qemu_put_be64(f, expire_time);
153 }
154 
155 void timer_get(QEMUFile *f, QEMUTimer *ts)
156 {
157     uint64_t expire_time;
158 
159     expire_time = qemu_get_be64(f);
160     if (expire_time != -1) {
161         timer_mod_ns(ts, expire_time);
162     } else {
163         timer_del(ts);
164     }
165 }
166 
167 
168 /* VMState timer support.
169  * Not in vmstate.c to not add qemu-timer.c as dependency to vmstate.c
170  */
171 
172 static int get_timer(QEMUFile *f, void *pv, size_t size,
173                      const VMStateField *field)
174 {
175     QEMUTimer *v = pv;
176     timer_get(f, v);
177     return 0;
178 }
179 
180 static int put_timer(QEMUFile *f, void *pv, size_t size,
181                      const VMStateField *field, JSONWriter *vmdesc)
182 {
183     QEMUTimer *v = pv;
184     timer_put(f, v);
185 
186     return 0;
187 }
188 
189 const VMStateInfo vmstate_info_timer = {
190     .name = "timer",
191     .get  = get_timer,
192     .put  = put_timer,
193 };
194 
195 
196 typedef struct CompatEntry {
197     char idstr[256];
198     int instance_id;
199 } CompatEntry;
200 
201 typedef struct SaveStateEntry {
202     QTAILQ_ENTRY(SaveStateEntry) entry;
203     char idstr[256];
204     uint32_t instance_id;
205     int alias_id;
206     int version_id;
207     /* version id read from the stream */
208     int load_version_id;
209     int section_id;
210     /* section id read from the stream */
211     int load_section_id;
212     const SaveVMHandlers *ops;
213     const VMStateDescription *vmsd;
214     void *opaque;
215     CompatEntry *compat;
216     int is_ram;
217 } SaveStateEntry;
218 
219 typedef struct SaveState {
220     QTAILQ_HEAD(, SaveStateEntry) handlers;
221     SaveStateEntry *handler_pri_head[MIG_PRI_MAX + 1];
222     int global_section_id;
223     uint32_t len;
224     const char *name;
225     uint32_t target_page_bits;
226     uint32_t caps_count;
227     MigrationCapability *capabilities;
228     QemuUUID uuid;
229 } SaveState;
230 
231 static SaveState savevm_state = {
232     .handlers = QTAILQ_HEAD_INITIALIZER(savevm_state.handlers),
233     .handler_pri_head = { [MIG_PRI_DEFAULT ... MIG_PRI_MAX] = NULL },
234     .global_section_id = 0,
235 };
236 
237 static bool should_validate_capability(int capability)
238 {
239     assert(capability >= 0 && capability < MIGRATION_CAPABILITY__MAX);
240     /* Validate only new capabilities to keep compatibility. */
241     switch (capability) {
242     case MIGRATION_CAPABILITY_X_IGNORE_SHARED:
243         return true;
244     default:
245         return false;
246     }
247 }
248 
249 static uint32_t get_validatable_capabilities_count(void)
250 {
251     MigrationState *s = migrate_get_current();
252     uint32_t result = 0;
253     int i;
254     for (i = 0; i < MIGRATION_CAPABILITY__MAX; i++) {
255         if (should_validate_capability(i) && s->enabled_capabilities[i]) {
256             result++;
257         }
258     }
259     return result;
260 }
261 
262 static int configuration_pre_save(void *opaque)
263 {
264     SaveState *state = opaque;
265     const char *current_name = MACHINE_GET_CLASS(current_machine)->name;
266     MigrationState *s = migrate_get_current();
267     int i, j;
268 
269     state->len = strlen(current_name);
270     state->name = current_name;
271     state->target_page_bits = qemu_target_page_bits();
272 
273     state->caps_count = get_validatable_capabilities_count();
274     state->capabilities = g_renew(MigrationCapability, state->capabilities,
275                                   state->caps_count);
276     for (i = j = 0; i < MIGRATION_CAPABILITY__MAX; i++) {
277         if (should_validate_capability(i) && s->enabled_capabilities[i]) {
278             state->capabilities[j++] = i;
279         }
280     }
281     state->uuid = qemu_uuid;
282 
283     return 0;
284 }
285 
286 static int configuration_post_save(void *opaque)
287 {
288     SaveState *state = opaque;
289 
290     g_free(state->capabilities);
291     state->capabilities = NULL;
292     state->caps_count = 0;
293     return 0;
294 }
295 
296 static int configuration_pre_load(void *opaque)
297 {
298     SaveState *state = opaque;
299 
300     /* If there is no target-page-bits subsection it means the source
301      * predates the variable-target-page-bits support and is using the
302      * minimum possible value for this CPU.
303      */
304     state->target_page_bits = qemu_target_page_bits_min();
305     return 0;
306 }
307 
308 static bool configuration_validate_capabilities(SaveState *state)
309 {
310     bool ret = true;
311     MigrationState *s = migrate_get_current();
312     unsigned long *source_caps_bm;
313     int i;
314 
315     source_caps_bm = bitmap_new(MIGRATION_CAPABILITY__MAX);
316     for (i = 0; i < state->caps_count; i++) {
317         MigrationCapability capability = state->capabilities[i];
318         set_bit(capability, source_caps_bm);
319     }
320 
321     for (i = 0; i < MIGRATION_CAPABILITY__MAX; i++) {
322         bool source_state, target_state;
323         if (!should_validate_capability(i)) {
324             continue;
325         }
326         source_state = test_bit(i, source_caps_bm);
327         target_state = s->enabled_capabilities[i];
328         if (source_state != target_state) {
329             error_report("Capability %s is %s, but received capability is %s",
330                          MigrationCapability_str(i),
331                          target_state ? "on" : "off",
332                          source_state ? "on" : "off");
333             ret = false;
334             /* Don't break here to report all failed capabilities */
335         }
336     }
337 
338     g_free(source_caps_bm);
339     return ret;
340 }
341 
342 static int configuration_post_load(void *opaque, int version_id)
343 {
344     SaveState *state = opaque;
345     const char *current_name = MACHINE_GET_CLASS(current_machine)->name;
346     int ret = 0;
347 
348     if (strncmp(state->name, current_name, state->len) != 0) {
349         error_report("Machine type received is '%.*s' and local is '%s'",
350                      (int) state->len, state->name, current_name);
351         ret = -EINVAL;
352         goto out;
353     }
354 
355     if (state->target_page_bits != qemu_target_page_bits()) {
356         error_report("Received TARGET_PAGE_BITS is %d but local is %d",
357                      state->target_page_bits, qemu_target_page_bits());
358         ret = -EINVAL;
359         goto out;
360     }
361 
362     if (!configuration_validate_capabilities(state)) {
363         ret = -EINVAL;
364         goto out;
365     }
366 
367 out:
368     g_free((void *)state->name);
369     state->name = NULL;
370     state->len = 0;
371     g_free(state->capabilities);
372     state->capabilities = NULL;
373     state->caps_count = 0;
374 
375     return ret;
376 }
377 
378 static int get_capability(QEMUFile *f, void *pv, size_t size,
379                           const VMStateField *field)
380 {
381     MigrationCapability *capability = pv;
382     char capability_str[UINT8_MAX + 1];
383     uint8_t len;
384     int i;
385 
386     len = qemu_get_byte(f);
387     qemu_get_buffer(f, (uint8_t *)capability_str, len);
388     capability_str[len] = '\0';
389     for (i = 0; i < MIGRATION_CAPABILITY__MAX; i++) {
390         if (!strcmp(MigrationCapability_str(i), capability_str)) {
391             *capability = i;
392             return 0;
393         }
394     }
395     error_report("Received unknown capability %s", capability_str);
396     return -EINVAL;
397 }
398 
399 static int put_capability(QEMUFile *f, void *pv, size_t size,
400                           const VMStateField *field, JSONWriter *vmdesc)
401 {
402     MigrationCapability *capability = pv;
403     const char *capability_str = MigrationCapability_str(*capability);
404     size_t len = strlen(capability_str);
405     assert(len <= UINT8_MAX);
406 
407     qemu_put_byte(f, len);
408     qemu_put_buffer(f, (uint8_t *)capability_str, len);
409     return 0;
410 }
411 
412 static const VMStateInfo vmstate_info_capability = {
413     .name = "capability",
414     .get  = get_capability,
415     .put  = put_capability,
416 };
417 
418 /* The target-page-bits subsection is present only if the
419  * target page size is not the same as the default (ie the
420  * minimum page size for a variable-page-size guest CPU).
421  * If it is present then it contains the actual target page
422  * bits for the machine, and migration will fail if the
423  * two ends don't agree about it.
424  */
425 static bool vmstate_target_page_bits_needed(void *opaque)
426 {
427     return qemu_target_page_bits()
428         > qemu_target_page_bits_min();
429 }
430 
431 static const VMStateDescription vmstate_target_page_bits = {
432     .name = "configuration/target-page-bits",
433     .version_id = 1,
434     .minimum_version_id = 1,
435     .needed = vmstate_target_page_bits_needed,
436     .fields = (VMStateField[]) {
437         VMSTATE_UINT32(target_page_bits, SaveState),
438         VMSTATE_END_OF_LIST()
439     }
440 };
441 
442 static bool vmstate_capabilites_needed(void *opaque)
443 {
444     return get_validatable_capabilities_count() > 0;
445 }
446 
447 static const VMStateDescription vmstate_capabilites = {
448     .name = "configuration/capabilities",
449     .version_id = 1,
450     .minimum_version_id = 1,
451     .needed = vmstate_capabilites_needed,
452     .fields = (VMStateField[]) {
453         VMSTATE_UINT32_V(caps_count, SaveState, 1),
454         VMSTATE_VARRAY_UINT32_ALLOC(capabilities, SaveState, caps_count, 1,
455                                     vmstate_info_capability,
456                                     MigrationCapability),
457         VMSTATE_END_OF_LIST()
458     }
459 };
460 
461 static bool vmstate_uuid_needed(void *opaque)
462 {
463     return qemu_uuid_set && migrate_validate_uuid();
464 }
465 
466 static int vmstate_uuid_post_load(void *opaque, int version_id)
467 {
468     SaveState *state = opaque;
469     char uuid_src[UUID_FMT_LEN + 1];
470     char uuid_dst[UUID_FMT_LEN + 1];
471 
472     if (!qemu_uuid_set) {
473         /*
474          * It's warning because user might not know UUID in some cases,
475          * e.g. load an old snapshot
476          */
477         qemu_uuid_unparse(&state->uuid, uuid_src);
478         warn_report("UUID is received %s, but local uuid isn't set",
479                      uuid_src);
480         return 0;
481     }
482     if (!qemu_uuid_is_equal(&state->uuid, &qemu_uuid)) {
483         qemu_uuid_unparse(&state->uuid, uuid_src);
484         qemu_uuid_unparse(&qemu_uuid, uuid_dst);
485         error_report("UUID received is %s and local is %s", uuid_src, uuid_dst);
486         return -EINVAL;
487     }
488     return 0;
489 }
490 
491 static const VMStateDescription vmstate_uuid = {
492     .name = "configuration/uuid",
493     .version_id = 1,
494     .minimum_version_id = 1,
495     .needed = vmstate_uuid_needed,
496     .post_load = vmstate_uuid_post_load,
497     .fields = (VMStateField[]) {
498         VMSTATE_UINT8_ARRAY_V(uuid.data, SaveState, sizeof(QemuUUID), 1),
499         VMSTATE_END_OF_LIST()
500     }
501 };
502 
503 static const VMStateDescription vmstate_configuration = {
504     .name = "configuration",
505     .version_id = 1,
506     .pre_load = configuration_pre_load,
507     .post_load = configuration_post_load,
508     .pre_save = configuration_pre_save,
509     .post_save = configuration_post_save,
510     .fields = (VMStateField[]) {
511         VMSTATE_UINT32(len, SaveState),
512         VMSTATE_VBUFFER_ALLOC_UINT32(name, SaveState, 0, NULL, len),
513         VMSTATE_END_OF_LIST()
514     },
515     .subsections = (const VMStateDescription *[]) {
516         &vmstate_target_page_bits,
517         &vmstate_capabilites,
518         &vmstate_uuid,
519         NULL
520     }
521 };
522 
523 static void dump_vmstate_vmsd(FILE *out_file,
524                               const VMStateDescription *vmsd, int indent,
525                               bool is_subsection);
526 
527 static void dump_vmstate_vmsf(FILE *out_file, const VMStateField *field,
528                               int indent)
529 {
530     fprintf(out_file, "%*s{\n", indent, "");
531     indent += 2;
532     fprintf(out_file, "%*s\"field\": \"%s\",\n", indent, "", field->name);
533     fprintf(out_file, "%*s\"version_id\": %d,\n", indent, "",
534             field->version_id);
535     fprintf(out_file, "%*s\"field_exists\": %s,\n", indent, "",
536             field->field_exists ? "true" : "false");
537     fprintf(out_file, "%*s\"size\": %zu", indent, "", field->size);
538     if (field->vmsd != NULL) {
539         fprintf(out_file, ",\n");
540         dump_vmstate_vmsd(out_file, field->vmsd, indent, false);
541     }
542     fprintf(out_file, "\n%*s}", indent - 2, "");
543 }
544 
545 static void dump_vmstate_vmss(FILE *out_file,
546                               const VMStateDescription **subsection,
547                               int indent)
548 {
549     if (*subsection != NULL) {
550         dump_vmstate_vmsd(out_file, *subsection, indent, true);
551     }
552 }
553 
554 static void dump_vmstate_vmsd(FILE *out_file,
555                               const VMStateDescription *vmsd, int indent,
556                               bool is_subsection)
557 {
558     if (is_subsection) {
559         fprintf(out_file, "%*s{\n", indent, "");
560     } else {
561         fprintf(out_file, "%*s\"%s\": {\n", indent, "", "Description");
562     }
563     indent += 2;
564     fprintf(out_file, "%*s\"name\": \"%s\",\n", indent, "", vmsd->name);
565     fprintf(out_file, "%*s\"version_id\": %d,\n", indent, "",
566             vmsd->version_id);
567     fprintf(out_file, "%*s\"minimum_version_id\": %d", indent, "",
568             vmsd->minimum_version_id);
569     if (vmsd->fields != NULL) {
570         const VMStateField *field = vmsd->fields;
571         bool first;
572 
573         fprintf(out_file, ",\n%*s\"Fields\": [\n", indent, "");
574         first = true;
575         while (field->name != NULL) {
576             if (field->flags & VMS_MUST_EXIST) {
577                 /* Ignore VMSTATE_VALIDATE bits; these don't get migrated */
578                 field++;
579                 continue;
580             }
581             if (!first) {
582                 fprintf(out_file, ",\n");
583             }
584             dump_vmstate_vmsf(out_file, field, indent + 2);
585             field++;
586             first = false;
587         }
588         fprintf(out_file, "\n%*s]", indent, "");
589     }
590     if (vmsd->subsections != NULL) {
591         const VMStateDescription **subsection = vmsd->subsections;
592         bool first;
593 
594         fprintf(out_file, ",\n%*s\"Subsections\": [\n", indent, "");
595         first = true;
596         while (*subsection != NULL) {
597             if (!first) {
598                 fprintf(out_file, ",\n");
599             }
600             dump_vmstate_vmss(out_file, subsection, indent + 2);
601             subsection++;
602             first = false;
603         }
604         fprintf(out_file, "\n%*s]", indent, "");
605     }
606     fprintf(out_file, "\n%*s}", indent - 2, "");
607 }
608 
609 static void dump_machine_type(FILE *out_file)
610 {
611     MachineClass *mc;
612 
613     mc = MACHINE_GET_CLASS(current_machine);
614 
615     fprintf(out_file, "  \"vmschkmachine\": {\n");
616     fprintf(out_file, "    \"Name\": \"%s\"\n", mc->name);
617     fprintf(out_file, "  },\n");
618 }
619 
620 void dump_vmstate_json_to_file(FILE *out_file)
621 {
622     GSList *list, *elt;
623     bool first;
624 
625     fprintf(out_file, "{\n");
626     dump_machine_type(out_file);
627 
628     first = true;
629     list = object_class_get_list(TYPE_DEVICE, true);
630     for (elt = list; elt; elt = elt->next) {
631         DeviceClass *dc = OBJECT_CLASS_CHECK(DeviceClass, elt->data,
632                                              TYPE_DEVICE);
633         const char *name;
634         int indent = 2;
635 
636         if (!dc->vmsd) {
637             continue;
638         }
639 
640         if (!first) {
641             fprintf(out_file, ",\n");
642         }
643         name = object_class_get_name(OBJECT_CLASS(dc));
644         fprintf(out_file, "%*s\"%s\": {\n", indent, "", name);
645         indent += 2;
646         fprintf(out_file, "%*s\"Name\": \"%s\",\n", indent, "", name);
647         fprintf(out_file, "%*s\"version_id\": %d,\n", indent, "",
648                 dc->vmsd->version_id);
649         fprintf(out_file, "%*s\"minimum_version_id\": %d,\n", indent, "",
650                 dc->vmsd->minimum_version_id);
651 
652         dump_vmstate_vmsd(out_file, dc->vmsd, indent, false);
653 
654         fprintf(out_file, "\n%*s}", indent - 2, "");
655         first = false;
656     }
657     fprintf(out_file, "\n}\n");
658     fclose(out_file);
659     g_slist_free(list);
660 }
661 
662 static uint32_t calculate_new_instance_id(const char *idstr)
663 {
664     SaveStateEntry *se;
665     uint32_t instance_id = 0;
666 
667     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
668         if (strcmp(idstr, se->idstr) == 0
669             && instance_id <= se->instance_id) {
670             instance_id = se->instance_id + 1;
671         }
672     }
673     /* Make sure we never loop over without being noticed */
674     assert(instance_id != VMSTATE_INSTANCE_ID_ANY);
675     return instance_id;
676 }
677 
678 static int calculate_compat_instance_id(const char *idstr)
679 {
680     SaveStateEntry *se;
681     int instance_id = 0;
682 
683     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
684         if (!se->compat) {
685             continue;
686         }
687 
688         if (strcmp(idstr, se->compat->idstr) == 0
689             && instance_id <= se->compat->instance_id) {
690             instance_id = se->compat->instance_id + 1;
691         }
692     }
693     return instance_id;
694 }
695 
696 static inline MigrationPriority save_state_priority(SaveStateEntry *se)
697 {
698     if (se->vmsd) {
699         return se->vmsd->priority;
700     }
701     return MIG_PRI_DEFAULT;
702 }
703 
704 static void savevm_state_handler_insert(SaveStateEntry *nse)
705 {
706     MigrationPriority priority = save_state_priority(nse);
707     SaveStateEntry *se;
708     int i;
709 
710     assert(priority <= MIG_PRI_MAX);
711 
712     for (i = priority - 1; i >= 0; i--) {
713         se = savevm_state.handler_pri_head[i];
714         if (se != NULL) {
715             assert(save_state_priority(se) < priority);
716             break;
717         }
718     }
719 
720     if (i >= 0) {
721         QTAILQ_INSERT_BEFORE(se, nse, entry);
722     } else {
723         QTAILQ_INSERT_TAIL(&savevm_state.handlers, nse, entry);
724     }
725 
726     if (savevm_state.handler_pri_head[priority] == NULL) {
727         savevm_state.handler_pri_head[priority] = nse;
728     }
729 }
730 
731 static void savevm_state_handler_remove(SaveStateEntry *se)
732 {
733     SaveStateEntry *next;
734     MigrationPriority priority = save_state_priority(se);
735 
736     if (se == savevm_state.handler_pri_head[priority]) {
737         next = QTAILQ_NEXT(se, entry);
738         if (next != NULL && save_state_priority(next) == priority) {
739             savevm_state.handler_pri_head[priority] = next;
740         } else {
741             savevm_state.handler_pri_head[priority] = NULL;
742         }
743     }
744     QTAILQ_REMOVE(&savevm_state.handlers, se, entry);
745 }
746 
747 /* TODO: Individual devices generally have very little idea about the rest
748    of the system, so instance_id should be removed/replaced.
749    Meanwhile pass -1 as instance_id if you do not already have a clearly
750    distinguishing id for all instances of your device class. */
751 int register_savevm_live(const char *idstr,
752                          uint32_t instance_id,
753                          int version_id,
754                          const SaveVMHandlers *ops,
755                          void *opaque)
756 {
757     SaveStateEntry *se;
758 
759     se = g_new0(SaveStateEntry, 1);
760     se->version_id = version_id;
761     se->section_id = savevm_state.global_section_id++;
762     se->ops = ops;
763     se->opaque = opaque;
764     se->vmsd = NULL;
765     /* if this is a live_savem then set is_ram */
766     if (ops->save_setup != NULL) {
767         se->is_ram = 1;
768     }
769 
770     pstrcat(se->idstr, sizeof(se->idstr), idstr);
771 
772     if (instance_id == VMSTATE_INSTANCE_ID_ANY) {
773         se->instance_id = calculate_new_instance_id(se->idstr);
774     } else {
775         se->instance_id = instance_id;
776     }
777     assert(!se->compat || se->instance_id == 0);
778     savevm_state_handler_insert(se);
779     return 0;
780 }
781 
782 void unregister_savevm(VMStateIf *obj, const char *idstr, void *opaque)
783 {
784     SaveStateEntry *se, *new_se;
785     char id[256] = "";
786 
787     if (obj) {
788         char *oid = vmstate_if_get_id(obj);
789         if (oid) {
790             pstrcpy(id, sizeof(id), oid);
791             pstrcat(id, sizeof(id), "/");
792             g_free(oid);
793         }
794     }
795     pstrcat(id, sizeof(id), idstr);
796 
797     QTAILQ_FOREACH_SAFE(se, &savevm_state.handlers, entry, new_se) {
798         if (strcmp(se->idstr, id) == 0 && se->opaque == opaque) {
799             savevm_state_handler_remove(se);
800             g_free(se->compat);
801             g_free(se);
802         }
803     }
804 }
805 
806 int vmstate_register_with_alias_id(VMStateIf *obj, uint32_t instance_id,
807                                    const VMStateDescription *vmsd,
808                                    void *opaque, int alias_id,
809                                    int required_for_version,
810                                    Error **errp)
811 {
812     SaveStateEntry *se;
813 
814     /* If this triggers, alias support can be dropped for the vmsd. */
815     assert(alias_id == -1 || required_for_version >= vmsd->minimum_version_id);
816 
817     se = g_new0(SaveStateEntry, 1);
818     se->version_id = vmsd->version_id;
819     se->section_id = savevm_state.global_section_id++;
820     se->opaque = opaque;
821     se->vmsd = vmsd;
822     se->alias_id = alias_id;
823 
824     if (obj) {
825         char *id = vmstate_if_get_id(obj);
826         if (id) {
827             if (snprintf(se->idstr, sizeof(se->idstr), "%s/", id) >=
828                 sizeof(se->idstr)) {
829                 error_setg(errp, "Path too long for VMState (%s)", id);
830                 g_free(id);
831                 g_free(se);
832 
833                 return -1;
834             }
835             g_free(id);
836 
837             se->compat = g_new0(CompatEntry, 1);
838             pstrcpy(se->compat->idstr, sizeof(se->compat->idstr), vmsd->name);
839             se->compat->instance_id = instance_id == VMSTATE_INSTANCE_ID_ANY ?
840                          calculate_compat_instance_id(vmsd->name) : instance_id;
841             instance_id = VMSTATE_INSTANCE_ID_ANY;
842         }
843     }
844     pstrcat(se->idstr, sizeof(se->idstr), vmsd->name);
845 
846     if (instance_id == VMSTATE_INSTANCE_ID_ANY) {
847         se->instance_id = calculate_new_instance_id(se->idstr);
848     } else {
849         se->instance_id = instance_id;
850     }
851     assert(!se->compat || se->instance_id == 0);
852     savevm_state_handler_insert(se);
853     return 0;
854 }
855 
856 void vmstate_unregister(VMStateIf *obj, const VMStateDescription *vmsd,
857                         void *opaque)
858 {
859     SaveStateEntry *se, *new_se;
860 
861     QTAILQ_FOREACH_SAFE(se, &savevm_state.handlers, entry, new_se) {
862         if (se->vmsd == vmsd && se->opaque == opaque) {
863             savevm_state_handler_remove(se);
864             g_free(se->compat);
865             g_free(se);
866         }
867     }
868 }
869 
870 static int vmstate_load(QEMUFile *f, SaveStateEntry *se)
871 {
872     trace_vmstate_load(se->idstr, se->vmsd ? se->vmsd->name : "(old)");
873     if (!se->vmsd) {         /* Old style */
874         return se->ops->load_state(f, se->opaque, se->load_version_id);
875     }
876     return vmstate_load_state(f, se->vmsd, se->opaque, se->load_version_id);
877 }
878 
879 static void vmstate_save_old_style(QEMUFile *f, SaveStateEntry *se,
880                                    JSONWriter *vmdesc)
881 {
882     int64_t old_offset, size;
883 
884     old_offset = qemu_file_total_transferred_fast(f);
885     se->ops->save_state(f, se->opaque);
886     size = qemu_file_total_transferred_fast(f) - old_offset;
887 
888     if (vmdesc) {
889         json_writer_int64(vmdesc, "size", size);
890         json_writer_start_array(vmdesc, "fields");
891         json_writer_start_object(vmdesc, NULL);
892         json_writer_str(vmdesc, "name", "data");
893         json_writer_int64(vmdesc, "size", size);
894         json_writer_str(vmdesc, "type", "buffer");
895         json_writer_end_object(vmdesc);
896         json_writer_end_array(vmdesc);
897     }
898 }
899 
900 static int vmstate_save(QEMUFile *f, SaveStateEntry *se,
901                         JSONWriter *vmdesc)
902 {
903     trace_vmstate_save(se->idstr, se->vmsd ? se->vmsd->name : "(old)");
904     if (!se->vmsd) {
905         vmstate_save_old_style(f, se, vmdesc);
906         return 0;
907     }
908     return vmstate_save_state(f, se->vmsd, se->opaque, vmdesc);
909 }
910 
911 /*
912  * Write the header for device section (QEMU_VM_SECTION START/END/PART/FULL)
913  */
914 static void save_section_header(QEMUFile *f, SaveStateEntry *se,
915                                 uint8_t section_type)
916 {
917     qemu_put_byte(f, section_type);
918     qemu_put_be32(f, se->section_id);
919 
920     if (section_type == QEMU_VM_SECTION_FULL ||
921         section_type == QEMU_VM_SECTION_START) {
922         /* ID string */
923         size_t len = strlen(se->idstr);
924         qemu_put_byte(f, len);
925         qemu_put_buffer(f, (uint8_t *)se->idstr, len);
926 
927         qemu_put_be32(f, se->instance_id);
928         qemu_put_be32(f, se->version_id);
929     }
930 }
931 
932 /*
933  * Write a footer onto device sections that catches cases misformatted device
934  * sections.
935  */
936 static void save_section_footer(QEMUFile *f, SaveStateEntry *se)
937 {
938     if (migrate_get_current()->send_section_footer) {
939         qemu_put_byte(f, QEMU_VM_SECTION_FOOTER);
940         qemu_put_be32(f, se->section_id);
941     }
942 }
943 
944 /**
945  * qemu_savevm_command_send: Send a 'QEMU_VM_COMMAND' type element with the
946  *                           command and associated data.
947  *
948  * @f: File to send command on
949  * @command: Command type to send
950  * @len: Length of associated data
951  * @data: Data associated with command.
952  */
953 static void qemu_savevm_command_send(QEMUFile *f,
954                                      enum qemu_vm_cmd command,
955                                      uint16_t len,
956                                      uint8_t *data)
957 {
958     trace_savevm_command_send(command, len);
959     qemu_put_byte(f, QEMU_VM_COMMAND);
960     qemu_put_be16(f, (uint16_t)command);
961     qemu_put_be16(f, len);
962     qemu_put_buffer(f, data, len);
963     qemu_fflush(f);
964 }
965 
966 void qemu_savevm_send_colo_enable(QEMUFile *f)
967 {
968     trace_savevm_send_colo_enable();
969     qemu_savevm_command_send(f, MIG_CMD_ENABLE_COLO, 0, NULL);
970 }
971 
972 void qemu_savevm_send_ping(QEMUFile *f, uint32_t value)
973 {
974     uint32_t buf;
975 
976     trace_savevm_send_ping(value);
977     buf = cpu_to_be32(value);
978     qemu_savevm_command_send(f, MIG_CMD_PING, sizeof(value), (uint8_t *)&buf);
979 }
980 
981 void qemu_savevm_send_open_return_path(QEMUFile *f)
982 {
983     trace_savevm_send_open_return_path();
984     qemu_savevm_command_send(f, MIG_CMD_OPEN_RETURN_PATH, 0, NULL);
985 }
986 
987 /* We have a buffer of data to send; we don't want that all to be loaded
988  * by the command itself, so the command contains just the length of the
989  * extra buffer that we then send straight after it.
990  * TODO: Must be a better way to organise that
991  *
992  * Returns:
993  *    0 on success
994  *    -ve on error
995  */
996 int qemu_savevm_send_packaged(QEMUFile *f, const uint8_t *buf, size_t len)
997 {
998     uint32_t tmp;
999 
1000     if (len > MAX_VM_CMD_PACKAGED_SIZE) {
1001         error_report("%s: Unreasonably large packaged state: %zu",
1002                      __func__, len);
1003         return -1;
1004     }
1005 
1006     tmp = cpu_to_be32(len);
1007 
1008     trace_qemu_savevm_send_packaged();
1009     qemu_savevm_command_send(f, MIG_CMD_PACKAGED, 4, (uint8_t *)&tmp);
1010 
1011     qemu_put_buffer(f, buf, len);
1012 
1013     return 0;
1014 }
1015 
1016 /* Send prior to any postcopy transfer */
1017 void qemu_savevm_send_postcopy_advise(QEMUFile *f)
1018 {
1019     if (migrate_postcopy_ram()) {
1020         uint64_t tmp[2];
1021         tmp[0] = cpu_to_be64(ram_pagesize_summary());
1022         tmp[1] = cpu_to_be64(qemu_target_page_size());
1023 
1024         trace_qemu_savevm_send_postcopy_advise();
1025         qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_ADVISE,
1026                                  16, (uint8_t *)tmp);
1027     } else {
1028         qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_ADVISE, 0, NULL);
1029     }
1030 }
1031 
1032 /* Sent prior to starting the destination running in postcopy, discard pages
1033  * that have already been sent but redirtied on the source.
1034  * CMD_POSTCOPY_RAM_DISCARD consist of:
1035  *      byte   version (0)
1036  *      byte   Length of name field (not including 0)
1037  *  n x byte   RAM block name
1038  *      byte   0 terminator (just for safety)
1039  *  n x        Byte ranges within the named RAMBlock
1040  *      be64   Start of the range
1041  *      be64   Length
1042  *
1043  *  name:  RAMBlock name that these entries are part of
1044  *  len: Number of page entries
1045  *  start_list: 'len' addresses
1046  *  length_list: 'len' addresses
1047  *
1048  */
1049 void qemu_savevm_send_postcopy_ram_discard(QEMUFile *f, const char *name,
1050                                            uint16_t len,
1051                                            uint64_t *start_list,
1052                                            uint64_t *length_list)
1053 {
1054     uint8_t *buf;
1055     uint16_t tmplen;
1056     uint16_t t;
1057     size_t name_len = strlen(name);
1058 
1059     trace_qemu_savevm_send_postcopy_ram_discard(name, len);
1060     assert(name_len < 256);
1061     buf = g_malloc0(1 + 1 + name_len + 1 + (8 + 8) * len);
1062     buf[0] = postcopy_ram_discard_version;
1063     buf[1] = name_len;
1064     memcpy(buf + 2, name, name_len);
1065     tmplen = 2 + name_len;
1066     buf[tmplen++] = '\0';
1067 
1068     for (t = 0; t < len; t++) {
1069         stq_be_p(buf + tmplen, start_list[t]);
1070         tmplen += 8;
1071         stq_be_p(buf + tmplen, length_list[t]);
1072         tmplen += 8;
1073     }
1074     qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_RAM_DISCARD, tmplen, buf);
1075     g_free(buf);
1076 }
1077 
1078 /* Get the destination into a state where it can receive postcopy data. */
1079 void qemu_savevm_send_postcopy_listen(QEMUFile *f)
1080 {
1081     trace_savevm_send_postcopy_listen();
1082     qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_LISTEN, 0, NULL);
1083 }
1084 
1085 /* Kick the destination into running */
1086 void qemu_savevm_send_postcopy_run(QEMUFile *f)
1087 {
1088     trace_savevm_send_postcopy_run();
1089     qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_RUN, 0, NULL);
1090 }
1091 
1092 void qemu_savevm_send_postcopy_resume(QEMUFile *f)
1093 {
1094     trace_savevm_send_postcopy_resume();
1095     qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_RESUME, 0, NULL);
1096 }
1097 
1098 void qemu_savevm_send_recv_bitmap(QEMUFile *f, char *block_name)
1099 {
1100     size_t len;
1101     char buf[256];
1102 
1103     trace_savevm_send_recv_bitmap(block_name);
1104 
1105     buf[0] = len = strlen(block_name);
1106     memcpy(buf + 1, block_name, len);
1107 
1108     qemu_savevm_command_send(f, MIG_CMD_RECV_BITMAP, len + 1, (uint8_t *)buf);
1109 }
1110 
1111 bool qemu_savevm_state_blocked(Error **errp)
1112 {
1113     SaveStateEntry *se;
1114 
1115     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1116         if (se->vmsd && se->vmsd->unmigratable) {
1117             error_setg(errp, "State blocked by non-migratable device '%s'",
1118                        se->idstr);
1119             return true;
1120         }
1121     }
1122     return false;
1123 }
1124 
1125 void qemu_savevm_non_migratable_list(strList **reasons)
1126 {
1127     SaveStateEntry *se;
1128 
1129     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1130         if (se->vmsd && se->vmsd->unmigratable) {
1131             QAPI_LIST_PREPEND(*reasons,
1132                               g_strdup_printf("non-migratable device: %s",
1133                                               se->idstr));
1134         }
1135     }
1136 }
1137 
1138 void qemu_savevm_state_header(QEMUFile *f)
1139 {
1140     trace_savevm_state_header();
1141     qemu_put_be32(f, QEMU_VM_FILE_MAGIC);
1142     qemu_put_be32(f, QEMU_VM_FILE_VERSION);
1143 
1144     if (migrate_get_current()->send_configuration) {
1145         qemu_put_byte(f, QEMU_VM_CONFIGURATION);
1146         vmstate_save_state(f, &vmstate_configuration, &savevm_state, 0);
1147     }
1148 }
1149 
1150 bool qemu_savevm_state_guest_unplug_pending(void)
1151 {
1152     SaveStateEntry *se;
1153 
1154     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1155         if (se->vmsd && se->vmsd->dev_unplug_pending &&
1156             se->vmsd->dev_unplug_pending(se->opaque)) {
1157             return true;
1158         }
1159     }
1160 
1161     return false;
1162 }
1163 
1164 void qemu_savevm_state_setup(QEMUFile *f)
1165 {
1166     SaveStateEntry *se;
1167     Error *local_err = NULL;
1168     int ret;
1169 
1170     trace_savevm_state_setup();
1171     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1172         if (!se->ops || !se->ops->save_setup) {
1173             continue;
1174         }
1175         if (se->ops->is_active) {
1176             if (!se->ops->is_active(se->opaque)) {
1177                 continue;
1178             }
1179         }
1180         save_section_header(f, se, QEMU_VM_SECTION_START);
1181 
1182         ret = se->ops->save_setup(f, se->opaque);
1183         save_section_footer(f, se);
1184         if (ret < 0) {
1185             qemu_file_set_error(f, ret);
1186             break;
1187         }
1188     }
1189 
1190     if (precopy_notify(PRECOPY_NOTIFY_SETUP, &local_err)) {
1191         error_report_err(local_err);
1192     }
1193 }
1194 
1195 int qemu_savevm_state_resume_prepare(MigrationState *s)
1196 {
1197     SaveStateEntry *se;
1198     int ret;
1199 
1200     trace_savevm_state_resume_prepare();
1201 
1202     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1203         if (!se->ops || !se->ops->resume_prepare) {
1204             continue;
1205         }
1206         if (se->ops->is_active) {
1207             if (!se->ops->is_active(se->opaque)) {
1208                 continue;
1209             }
1210         }
1211         ret = se->ops->resume_prepare(s, se->opaque);
1212         if (ret < 0) {
1213             return ret;
1214         }
1215     }
1216 
1217     return 0;
1218 }
1219 
1220 /*
1221  * this function has three return values:
1222  *   negative: there was one error, and we have -errno.
1223  *   0 : We haven't finished, caller have to go again
1224  *   1 : We have finished, we can go to complete phase
1225  */
1226 int qemu_savevm_state_iterate(QEMUFile *f, bool postcopy)
1227 {
1228     SaveStateEntry *se;
1229     int ret = 1;
1230 
1231     trace_savevm_state_iterate();
1232     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1233         if (!se->ops || !se->ops->save_live_iterate) {
1234             continue;
1235         }
1236         if (se->ops->is_active &&
1237             !se->ops->is_active(se->opaque)) {
1238             continue;
1239         }
1240         if (se->ops->is_active_iterate &&
1241             !se->ops->is_active_iterate(se->opaque)) {
1242             continue;
1243         }
1244         /*
1245          * In the postcopy phase, any device that doesn't know how to
1246          * do postcopy should have saved it's state in the _complete
1247          * call that's already run, it might get confused if we call
1248          * iterate afterwards.
1249          */
1250         if (postcopy &&
1251             !(se->ops->has_postcopy && se->ops->has_postcopy(se->opaque))) {
1252             continue;
1253         }
1254         if (qemu_file_rate_limit(f)) {
1255             return 0;
1256         }
1257         trace_savevm_section_start(se->idstr, se->section_id);
1258 
1259         save_section_header(f, se, QEMU_VM_SECTION_PART);
1260 
1261         ret = se->ops->save_live_iterate(f, se->opaque);
1262         trace_savevm_section_end(se->idstr, se->section_id, ret);
1263         save_section_footer(f, se);
1264 
1265         if (ret < 0) {
1266             error_report("failed to save SaveStateEntry with id(name): "
1267                          "%d(%s): %d",
1268                          se->section_id, se->idstr, ret);
1269             qemu_file_set_error(f, ret);
1270         }
1271         if (ret <= 0) {
1272             /* Do not proceed to the next vmstate before this one reported
1273                completion of the current stage. This serializes the migration
1274                and reduces the probability that a faster changing state is
1275                synchronized over and over again. */
1276             break;
1277         }
1278     }
1279     return ret;
1280 }
1281 
1282 static bool should_send_vmdesc(void)
1283 {
1284     MachineState *machine = MACHINE(qdev_get_machine());
1285     bool in_postcopy = migration_in_postcopy();
1286     return !machine->suppress_vmdesc && !in_postcopy;
1287 }
1288 
1289 /*
1290  * Calls the save_live_complete_postcopy methods
1291  * causing the last few pages to be sent immediately and doing any associated
1292  * cleanup.
1293  * Note postcopy also calls qemu_savevm_state_complete_precopy to complete
1294  * all the other devices, but that happens at the point we switch to postcopy.
1295  */
1296 void qemu_savevm_state_complete_postcopy(QEMUFile *f)
1297 {
1298     SaveStateEntry *se;
1299     int ret;
1300 
1301     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1302         if (!se->ops || !se->ops->save_live_complete_postcopy) {
1303             continue;
1304         }
1305         if (se->ops->is_active) {
1306             if (!se->ops->is_active(se->opaque)) {
1307                 continue;
1308             }
1309         }
1310         trace_savevm_section_start(se->idstr, se->section_id);
1311         /* Section type */
1312         qemu_put_byte(f, QEMU_VM_SECTION_END);
1313         qemu_put_be32(f, se->section_id);
1314 
1315         ret = se->ops->save_live_complete_postcopy(f, se->opaque);
1316         trace_savevm_section_end(se->idstr, se->section_id, ret);
1317         save_section_footer(f, se);
1318         if (ret < 0) {
1319             qemu_file_set_error(f, ret);
1320             return;
1321         }
1322     }
1323 
1324     qemu_put_byte(f, QEMU_VM_EOF);
1325     qemu_fflush(f);
1326 }
1327 
1328 static
1329 int qemu_savevm_state_complete_precopy_iterable(QEMUFile *f, bool in_postcopy)
1330 {
1331     SaveStateEntry *se;
1332     int ret;
1333 
1334     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1335         if (!se->ops ||
1336             (in_postcopy && se->ops->has_postcopy &&
1337              se->ops->has_postcopy(se->opaque)) ||
1338             !se->ops->save_live_complete_precopy) {
1339             continue;
1340         }
1341 
1342         if (se->ops->is_active) {
1343             if (!se->ops->is_active(se->opaque)) {
1344                 continue;
1345             }
1346         }
1347         trace_savevm_section_start(se->idstr, se->section_id);
1348 
1349         save_section_header(f, se, QEMU_VM_SECTION_END);
1350 
1351         ret = se->ops->save_live_complete_precopy(f, se->opaque);
1352         trace_savevm_section_end(se->idstr, se->section_id, ret);
1353         save_section_footer(f, se);
1354         if (ret < 0) {
1355             qemu_file_set_error(f, ret);
1356             return -1;
1357         }
1358     }
1359 
1360     return 0;
1361 }
1362 
1363 int qemu_savevm_state_complete_precopy_non_iterable(QEMUFile *f,
1364                                                     bool in_postcopy,
1365                                                     bool inactivate_disks)
1366 {
1367     g_autoptr(JSONWriter) vmdesc = NULL;
1368     int vmdesc_len;
1369     SaveStateEntry *se;
1370     int ret;
1371 
1372     vmdesc = json_writer_new(false);
1373     json_writer_start_object(vmdesc, NULL);
1374     json_writer_int64(vmdesc, "page_size", qemu_target_page_size());
1375     json_writer_start_array(vmdesc, "devices");
1376     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1377 
1378         if ((!se->ops || !se->ops->save_state) && !se->vmsd) {
1379             continue;
1380         }
1381         if (se->vmsd && !vmstate_save_needed(se->vmsd, se->opaque)) {
1382             trace_savevm_section_skip(se->idstr, se->section_id);
1383             continue;
1384         }
1385 
1386         trace_savevm_section_start(se->idstr, se->section_id);
1387 
1388         json_writer_start_object(vmdesc, NULL);
1389         json_writer_str(vmdesc, "name", se->idstr);
1390         json_writer_int64(vmdesc, "instance_id", se->instance_id);
1391 
1392         save_section_header(f, se, QEMU_VM_SECTION_FULL);
1393         ret = vmstate_save(f, se, vmdesc);
1394         if (ret) {
1395             qemu_file_set_error(f, ret);
1396             return ret;
1397         }
1398         trace_savevm_section_end(se->idstr, se->section_id, 0);
1399         save_section_footer(f, se);
1400 
1401         json_writer_end_object(vmdesc);
1402     }
1403 
1404     if (inactivate_disks) {
1405         /* Inactivate before sending QEMU_VM_EOF so that the
1406          * bdrv_activate_all() on the other end won't fail. */
1407         ret = bdrv_inactivate_all();
1408         if (ret) {
1409             error_report("%s: bdrv_inactivate_all() failed (%d)",
1410                          __func__, ret);
1411             qemu_file_set_error(f, ret);
1412             return ret;
1413         }
1414     }
1415     if (!in_postcopy) {
1416         /* Postcopy stream will still be going */
1417         qemu_put_byte(f, QEMU_VM_EOF);
1418     }
1419 
1420     json_writer_end_array(vmdesc);
1421     json_writer_end_object(vmdesc);
1422     vmdesc_len = strlen(json_writer_get(vmdesc));
1423 
1424     if (should_send_vmdesc()) {
1425         qemu_put_byte(f, QEMU_VM_VMDESCRIPTION);
1426         qemu_put_be32(f, vmdesc_len);
1427         qemu_put_buffer(f, (uint8_t *)json_writer_get(vmdesc), vmdesc_len);
1428     }
1429 
1430     return 0;
1431 }
1432 
1433 int qemu_savevm_state_complete_precopy(QEMUFile *f, bool iterable_only,
1434                                        bool inactivate_disks)
1435 {
1436     int ret;
1437     Error *local_err = NULL;
1438     bool in_postcopy = migration_in_postcopy();
1439 
1440     if (precopy_notify(PRECOPY_NOTIFY_COMPLETE, &local_err)) {
1441         error_report_err(local_err);
1442     }
1443 
1444     trace_savevm_state_complete_precopy();
1445 
1446     cpu_synchronize_all_states();
1447 
1448     if (!in_postcopy || iterable_only) {
1449         ret = qemu_savevm_state_complete_precopy_iterable(f, in_postcopy);
1450         if (ret) {
1451             return ret;
1452         }
1453     }
1454 
1455     if (iterable_only) {
1456         goto flush;
1457     }
1458 
1459     ret = qemu_savevm_state_complete_precopy_non_iterable(f, in_postcopy,
1460                                                           inactivate_disks);
1461     if (ret) {
1462         return ret;
1463     }
1464 
1465 flush:
1466     qemu_fflush(f);
1467     return 0;
1468 }
1469 
1470 /* Give an estimate of the amount left to be transferred,
1471  * the result is split into the amount for units that can and
1472  * for units that can't do postcopy.
1473  */
1474 void qemu_savevm_state_pending(QEMUFile *f, uint64_t threshold_size,
1475                                uint64_t *res_precopy_only,
1476                                uint64_t *res_compatible,
1477                                uint64_t *res_postcopy_only)
1478 {
1479     SaveStateEntry *se;
1480 
1481     *res_precopy_only = 0;
1482     *res_compatible = 0;
1483     *res_postcopy_only = 0;
1484 
1485 
1486     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1487         if (!se->ops || !se->ops->save_live_pending) {
1488             continue;
1489         }
1490         if (se->ops->is_active) {
1491             if (!se->ops->is_active(se->opaque)) {
1492                 continue;
1493             }
1494         }
1495         se->ops->save_live_pending(f, se->opaque, threshold_size,
1496                                    res_precopy_only, res_compatible,
1497                                    res_postcopy_only);
1498     }
1499 }
1500 
1501 void qemu_savevm_state_cleanup(void)
1502 {
1503     SaveStateEntry *se;
1504     Error *local_err = NULL;
1505 
1506     if (precopy_notify(PRECOPY_NOTIFY_CLEANUP, &local_err)) {
1507         error_report_err(local_err);
1508     }
1509 
1510     trace_savevm_state_cleanup();
1511     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1512         if (se->ops && se->ops->save_cleanup) {
1513             se->ops->save_cleanup(se->opaque);
1514         }
1515     }
1516 }
1517 
1518 static int qemu_savevm_state(QEMUFile *f, Error **errp)
1519 {
1520     int ret;
1521     MigrationState *ms = migrate_get_current();
1522     MigrationStatus status;
1523 
1524     if (migration_is_running(ms->state)) {
1525         error_setg(errp, QERR_MIGRATION_ACTIVE);
1526         return -EINVAL;
1527     }
1528 
1529     if (migrate_use_block()) {
1530         error_setg(errp, "Block migration and snapshots are incompatible");
1531         return -EINVAL;
1532     }
1533 
1534     migrate_init(ms);
1535     memset(&ram_counters, 0, sizeof(ram_counters));
1536     memset(&compression_counters, 0, sizeof(compression_counters));
1537     ms->to_dst_file = f;
1538 
1539     qemu_mutex_unlock_iothread();
1540     qemu_savevm_state_header(f);
1541     qemu_savevm_state_setup(f);
1542     qemu_mutex_lock_iothread();
1543 
1544     while (qemu_file_get_error(f) == 0) {
1545         if (qemu_savevm_state_iterate(f, false) > 0) {
1546             break;
1547         }
1548     }
1549 
1550     ret = qemu_file_get_error(f);
1551     if (ret == 0) {
1552         qemu_savevm_state_complete_precopy(f, false, false);
1553         ret = qemu_file_get_error(f);
1554     }
1555     qemu_savevm_state_cleanup();
1556     if (ret != 0) {
1557         error_setg_errno(errp, -ret, "Error while writing VM state");
1558     }
1559 
1560     if (ret != 0) {
1561         status = MIGRATION_STATUS_FAILED;
1562     } else {
1563         status = MIGRATION_STATUS_COMPLETED;
1564     }
1565     migrate_set_state(&ms->state, MIGRATION_STATUS_SETUP, status);
1566 
1567     /* f is outer parameter, it should not stay in global migration state after
1568      * this function finished */
1569     ms->to_dst_file = NULL;
1570 
1571     return ret;
1572 }
1573 
1574 void qemu_savevm_live_state(QEMUFile *f)
1575 {
1576     /* save QEMU_VM_SECTION_END section */
1577     qemu_savevm_state_complete_precopy(f, true, false);
1578     qemu_put_byte(f, QEMU_VM_EOF);
1579 }
1580 
1581 int qemu_save_device_state(QEMUFile *f)
1582 {
1583     SaveStateEntry *se;
1584 
1585     if (!migration_in_colo_state()) {
1586         qemu_put_be32(f, QEMU_VM_FILE_MAGIC);
1587         qemu_put_be32(f, QEMU_VM_FILE_VERSION);
1588     }
1589     cpu_synchronize_all_states();
1590 
1591     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1592         int ret;
1593 
1594         if (se->is_ram) {
1595             continue;
1596         }
1597         if ((!se->ops || !se->ops->save_state) && !se->vmsd) {
1598             continue;
1599         }
1600         if (se->vmsd && !vmstate_save_needed(se->vmsd, se->opaque)) {
1601             continue;
1602         }
1603 
1604         save_section_header(f, se, QEMU_VM_SECTION_FULL);
1605 
1606         ret = vmstate_save(f, se, NULL);
1607         if (ret) {
1608             return ret;
1609         }
1610 
1611         save_section_footer(f, se);
1612     }
1613 
1614     qemu_put_byte(f, QEMU_VM_EOF);
1615 
1616     return qemu_file_get_error(f);
1617 }
1618 
1619 static SaveStateEntry *find_se(const char *idstr, uint32_t instance_id)
1620 {
1621     SaveStateEntry *se;
1622 
1623     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1624         if (!strcmp(se->idstr, idstr) &&
1625             (instance_id == se->instance_id ||
1626              instance_id == se->alias_id))
1627             return se;
1628         /* Migrating from an older version? */
1629         if (strstr(se->idstr, idstr) && se->compat) {
1630             if (!strcmp(se->compat->idstr, idstr) &&
1631                 (instance_id == se->compat->instance_id ||
1632                  instance_id == se->alias_id))
1633                 return se;
1634         }
1635     }
1636     return NULL;
1637 }
1638 
1639 enum LoadVMExitCodes {
1640     /* Allow a command to quit all layers of nested loadvm loops */
1641     LOADVM_QUIT     =  1,
1642 };
1643 
1644 /* ------ incoming postcopy messages ------ */
1645 /* 'advise' arrives before any transfers just to tell us that a postcopy
1646  * *might* happen - it might be skipped if precopy transferred everything
1647  * quickly.
1648  */
1649 static int loadvm_postcopy_handle_advise(MigrationIncomingState *mis,
1650                                          uint16_t len)
1651 {
1652     PostcopyState ps = postcopy_state_set(POSTCOPY_INCOMING_ADVISE);
1653     uint64_t remote_pagesize_summary, local_pagesize_summary, remote_tps;
1654     size_t page_size = qemu_target_page_size();
1655     Error *local_err = NULL;
1656 
1657     trace_loadvm_postcopy_handle_advise();
1658     if (ps != POSTCOPY_INCOMING_NONE) {
1659         error_report("CMD_POSTCOPY_ADVISE in wrong postcopy state (%d)", ps);
1660         return -1;
1661     }
1662 
1663     switch (len) {
1664     case 0:
1665         if (migrate_postcopy_ram()) {
1666             error_report("RAM postcopy is enabled but have 0 byte advise");
1667             return -EINVAL;
1668         }
1669         return 0;
1670     case 8 + 8:
1671         if (!migrate_postcopy_ram()) {
1672             error_report("RAM postcopy is disabled but have 16 byte advise");
1673             return -EINVAL;
1674         }
1675         break;
1676     default:
1677         error_report("CMD_POSTCOPY_ADVISE invalid length (%d)", len);
1678         return -EINVAL;
1679     }
1680 
1681     if (!postcopy_ram_supported_by_host(mis)) {
1682         postcopy_state_set(POSTCOPY_INCOMING_NONE);
1683         return -1;
1684     }
1685 
1686     remote_pagesize_summary = qemu_get_be64(mis->from_src_file);
1687     local_pagesize_summary = ram_pagesize_summary();
1688 
1689     if (remote_pagesize_summary != local_pagesize_summary)  {
1690         /*
1691          * This detects two potential causes of mismatch:
1692          *   a) A mismatch in host page sizes
1693          *      Some combinations of mismatch are probably possible but it gets
1694          *      a bit more complicated.  In particular we need to place whole
1695          *      host pages on the dest at once, and we need to ensure that we
1696          *      handle dirtying to make sure we never end up sending part of
1697          *      a hostpage on it's own.
1698          *   b) The use of different huge page sizes on source/destination
1699          *      a more fine grain test is performed during RAM block migration
1700          *      but this test here causes a nice early clear failure, and
1701          *      also fails when passed to an older qemu that doesn't
1702          *      do huge pages.
1703          */
1704         error_report("Postcopy needs matching RAM page sizes (s=%" PRIx64
1705                                                              " d=%" PRIx64 ")",
1706                      remote_pagesize_summary, local_pagesize_summary);
1707         return -1;
1708     }
1709 
1710     remote_tps = qemu_get_be64(mis->from_src_file);
1711     if (remote_tps != page_size) {
1712         /*
1713          * Again, some differences could be dealt with, but for now keep it
1714          * simple.
1715          */
1716         error_report("Postcopy needs matching target page sizes (s=%d d=%zd)",
1717                      (int)remote_tps, page_size);
1718         return -1;
1719     }
1720 
1721     if (postcopy_notify(POSTCOPY_NOTIFY_INBOUND_ADVISE, &local_err)) {
1722         error_report_err(local_err);
1723         return -1;
1724     }
1725 
1726     if (ram_postcopy_incoming_init(mis)) {
1727         return -1;
1728     }
1729 
1730     return 0;
1731 }
1732 
1733 /* After postcopy we will be told to throw some pages away since they're
1734  * dirty and will have to be demand fetched.  Must happen before CPU is
1735  * started.
1736  * There can be 0..many of these messages, each encoding multiple pages.
1737  */
1738 static int loadvm_postcopy_ram_handle_discard(MigrationIncomingState *mis,
1739                                               uint16_t len)
1740 {
1741     int tmp;
1742     char ramid[256];
1743     PostcopyState ps = postcopy_state_get();
1744 
1745     trace_loadvm_postcopy_ram_handle_discard();
1746 
1747     switch (ps) {
1748     case POSTCOPY_INCOMING_ADVISE:
1749         /* 1st discard */
1750         tmp = postcopy_ram_prepare_discard(mis);
1751         if (tmp) {
1752             return tmp;
1753         }
1754         break;
1755 
1756     case POSTCOPY_INCOMING_DISCARD:
1757         /* Expected state */
1758         break;
1759 
1760     default:
1761         error_report("CMD_POSTCOPY_RAM_DISCARD in wrong postcopy state (%d)",
1762                      ps);
1763         return -1;
1764     }
1765     /* We're expecting a
1766      *    Version (0)
1767      *    a RAM ID string (length byte, name, 0 term)
1768      *    then at least 1 16 byte chunk
1769     */
1770     if (len < (1 + 1 + 1 + 1 + 2 * 8)) {
1771         error_report("CMD_POSTCOPY_RAM_DISCARD invalid length (%d)", len);
1772         return -1;
1773     }
1774 
1775     tmp = qemu_get_byte(mis->from_src_file);
1776     if (tmp != postcopy_ram_discard_version) {
1777         error_report("CMD_POSTCOPY_RAM_DISCARD invalid version (%d)", tmp);
1778         return -1;
1779     }
1780 
1781     if (!qemu_get_counted_string(mis->from_src_file, ramid)) {
1782         error_report("CMD_POSTCOPY_RAM_DISCARD Failed to read RAMBlock ID");
1783         return -1;
1784     }
1785     tmp = qemu_get_byte(mis->from_src_file);
1786     if (tmp != 0) {
1787         error_report("CMD_POSTCOPY_RAM_DISCARD missing nil (%d)", tmp);
1788         return -1;
1789     }
1790 
1791     len -= 3 + strlen(ramid);
1792     if (len % 16) {
1793         error_report("CMD_POSTCOPY_RAM_DISCARD invalid length (%d)", len);
1794         return -1;
1795     }
1796     trace_loadvm_postcopy_ram_handle_discard_header(ramid, len);
1797     while (len) {
1798         uint64_t start_addr, block_length;
1799         start_addr = qemu_get_be64(mis->from_src_file);
1800         block_length = qemu_get_be64(mis->from_src_file);
1801 
1802         len -= 16;
1803         int ret = ram_discard_range(ramid, start_addr, block_length);
1804         if (ret) {
1805             return ret;
1806         }
1807     }
1808     trace_loadvm_postcopy_ram_handle_discard_end();
1809 
1810     return 0;
1811 }
1812 
1813 /*
1814  * Triggered by a postcopy_listen command; this thread takes over reading
1815  * the input stream, leaving the main thread free to carry on loading the rest
1816  * of the device state (from RAM).
1817  * (TODO:This could do with being in a postcopy file - but there again it's
1818  * just another input loop, not that postcopy specific)
1819  */
1820 static void *postcopy_ram_listen_thread(void *opaque)
1821 {
1822     MigrationIncomingState *mis = migration_incoming_get_current();
1823     QEMUFile *f = mis->from_src_file;
1824     int load_res;
1825     MigrationState *migr = migrate_get_current();
1826 
1827     object_ref(OBJECT(migr));
1828 
1829     migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE,
1830                                    MIGRATION_STATUS_POSTCOPY_ACTIVE);
1831     qemu_sem_post(&mis->thread_sync_sem);
1832     trace_postcopy_ram_listen_thread_start();
1833 
1834     rcu_register_thread();
1835     /*
1836      * Because we're a thread and not a coroutine we can't yield
1837      * in qemu_file, and thus we must be blocking now.
1838      */
1839     qemu_file_set_blocking(f, true);
1840     load_res = qemu_loadvm_state_main(f, mis);
1841 
1842     /*
1843      * This is tricky, but, mis->from_src_file can change after it
1844      * returns, when postcopy recovery happened. In the future, we may
1845      * want a wrapper for the QEMUFile handle.
1846      */
1847     f = mis->from_src_file;
1848 
1849     /* And non-blocking again so we don't block in any cleanup */
1850     qemu_file_set_blocking(f, false);
1851 
1852     trace_postcopy_ram_listen_thread_exit();
1853     if (load_res < 0) {
1854         qemu_file_set_error(f, load_res);
1855         dirty_bitmap_mig_cancel_incoming();
1856         if (postcopy_state_get() == POSTCOPY_INCOMING_RUNNING &&
1857             !migrate_postcopy_ram() && migrate_dirty_bitmaps())
1858         {
1859             error_report("%s: loadvm failed during postcopy: %d. All states "
1860                          "are migrated except dirty bitmaps. Some dirty "
1861                          "bitmaps may be lost, and present migrated dirty "
1862                          "bitmaps are correctly migrated and valid.",
1863                          __func__, load_res);
1864             load_res = 0; /* prevent further exit() */
1865         } else {
1866             error_report("%s: loadvm failed: %d", __func__, load_res);
1867             migrate_set_state(&mis->state, MIGRATION_STATUS_POSTCOPY_ACTIVE,
1868                                            MIGRATION_STATUS_FAILED);
1869         }
1870     }
1871     if (load_res >= 0) {
1872         /*
1873          * This looks good, but it's possible that the device loading in the
1874          * main thread hasn't finished yet, and so we might not be in 'RUN'
1875          * state yet; wait for the end of the main thread.
1876          */
1877         qemu_event_wait(&mis->main_thread_load_event);
1878     }
1879     postcopy_ram_incoming_cleanup(mis);
1880 
1881     if (load_res < 0) {
1882         /*
1883          * If something went wrong then we have a bad state so exit;
1884          * depending how far we got it might be possible at this point
1885          * to leave the guest running and fire MCEs for pages that never
1886          * arrived as a desperate recovery step.
1887          */
1888         rcu_unregister_thread();
1889         exit(EXIT_FAILURE);
1890     }
1891 
1892     migrate_set_state(&mis->state, MIGRATION_STATUS_POSTCOPY_ACTIVE,
1893                                    MIGRATION_STATUS_COMPLETED);
1894     /*
1895      * If everything has worked fine, then the main thread has waited
1896      * for us to start, and we're the last use of the mis.
1897      * (If something broke then qemu will have to exit anyway since it's
1898      * got a bad migration state).
1899      */
1900     migration_incoming_state_destroy();
1901     qemu_loadvm_state_cleanup();
1902 
1903     rcu_unregister_thread();
1904     mis->have_listen_thread = false;
1905     postcopy_state_set(POSTCOPY_INCOMING_END);
1906 
1907     object_unref(OBJECT(migr));
1908 
1909     return NULL;
1910 }
1911 
1912 /* After this message we must be able to immediately receive postcopy data */
1913 static int loadvm_postcopy_handle_listen(MigrationIncomingState *mis)
1914 {
1915     PostcopyState ps = postcopy_state_set(POSTCOPY_INCOMING_LISTENING);
1916     Error *local_err = NULL;
1917 
1918     trace_loadvm_postcopy_handle_listen("enter");
1919 
1920     if (ps != POSTCOPY_INCOMING_ADVISE && ps != POSTCOPY_INCOMING_DISCARD) {
1921         error_report("CMD_POSTCOPY_LISTEN in wrong postcopy state (%d)", ps);
1922         return -1;
1923     }
1924     if (ps == POSTCOPY_INCOMING_ADVISE) {
1925         /*
1926          * A rare case, we entered listen without having to do any discards,
1927          * so do the setup that's normally done at the time of the 1st discard.
1928          */
1929         if (migrate_postcopy_ram()) {
1930             postcopy_ram_prepare_discard(mis);
1931         }
1932     }
1933 
1934     trace_loadvm_postcopy_handle_listen("after discard");
1935 
1936     /*
1937      * Sensitise RAM - can now generate requests for blocks that don't exist
1938      * However, at this point the CPU shouldn't be running, and the IO
1939      * shouldn't be doing anything yet so don't actually expect requests
1940      */
1941     if (migrate_postcopy_ram()) {
1942         if (postcopy_ram_incoming_setup(mis)) {
1943             postcopy_ram_incoming_cleanup(mis);
1944             return -1;
1945         }
1946     }
1947 
1948     trace_loadvm_postcopy_handle_listen("after uffd");
1949 
1950     if (postcopy_notify(POSTCOPY_NOTIFY_INBOUND_LISTEN, &local_err)) {
1951         error_report_err(local_err);
1952         return -1;
1953     }
1954 
1955     mis->have_listen_thread = true;
1956     postcopy_thread_create(mis, &mis->listen_thread, "postcopy/listen",
1957                            postcopy_ram_listen_thread, QEMU_THREAD_DETACHED);
1958     trace_loadvm_postcopy_handle_listen("return");
1959 
1960     return 0;
1961 }
1962 
1963 static void loadvm_postcopy_handle_run_bh(void *opaque)
1964 {
1965     Error *local_err = NULL;
1966     MigrationIncomingState *mis = opaque;
1967 
1968     trace_loadvm_postcopy_handle_run_bh("enter");
1969 
1970     /* TODO we should move all of this lot into postcopy_ram.c or a shared code
1971      * in migration.c
1972      */
1973     cpu_synchronize_all_post_init();
1974 
1975     trace_loadvm_postcopy_handle_run_bh("after cpu sync");
1976 
1977     qemu_announce_self(&mis->announce_timer, migrate_announce_params());
1978 
1979     trace_loadvm_postcopy_handle_run_bh("after announce");
1980 
1981     /* Make sure all file formats throw away their mutable metadata.
1982      * If we get an error here, just don't restart the VM yet. */
1983     bdrv_activate_all(&local_err);
1984     if (local_err) {
1985         error_report_err(local_err);
1986         local_err = NULL;
1987         autostart = false;
1988     }
1989 
1990     trace_loadvm_postcopy_handle_run_bh("after invalidate cache");
1991 
1992     dirty_bitmap_mig_before_vm_start();
1993 
1994     if (autostart) {
1995         /* Hold onto your hats, starting the CPU */
1996         vm_start();
1997     } else {
1998         /* leave it paused and let management decide when to start the CPU */
1999         runstate_set(RUN_STATE_PAUSED);
2000     }
2001 
2002     qemu_bh_delete(mis->bh);
2003 
2004     trace_loadvm_postcopy_handle_run_bh("return");
2005 }
2006 
2007 /* After all discards we can start running and asking for pages */
2008 static int loadvm_postcopy_handle_run(MigrationIncomingState *mis)
2009 {
2010     PostcopyState ps = postcopy_state_get();
2011 
2012     trace_loadvm_postcopy_handle_run();
2013     if (ps != POSTCOPY_INCOMING_LISTENING) {
2014         error_report("CMD_POSTCOPY_RUN in wrong postcopy state (%d)", ps);
2015         return -1;
2016     }
2017 
2018     postcopy_state_set(POSTCOPY_INCOMING_RUNNING);
2019     mis->bh = qemu_bh_new(loadvm_postcopy_handle_run_bh, mis);
2020     qemu_bh_schedule(mis->bh);
2021 
2022     /* We need to finish reading the stream from the package
2023      * and also stop reading anything more from the stream that loaded the
2024      * package (since it's now being read by the listener thread).
2025      * LOADVM_QUIT will quit all the layers of nested loadvm loops.
2026      */
2027     return LOADVM_QUIT;
2028 }
2029 
2030 /* We must be with page_request_mutex held */
2031 static gboolean postcopy_sync_page_req(gpointer key, gpointer value,
2032                                        gpointer data)
2033 {
2034     MigrationIncomingState *mis = data;
2035     void *host_addr = (void *) key;
2036     ram_addr_t rb_offset;
2037     RAMBlock *rb;
2038     int ret;
2039 
2040     rb = qemu_ram_block_from_host(host_addr, true, &rb_offset);
2041     if (!rb) {
2042         /*
2043          * This should _never_ happen.  However be nice for a migrating VM to
2044          * not crash/assert.  Post an error (note: intended to not use *_once
2045          * because we do want to see all the illegal addresses; and this can
2046          * never be triggered by the guest so we're safe) and move on next.
2047          */
2048         error_report("%s: illegal host addr %p", __func__, host_addr);
2049         /* Try the next entry */
2050         return FALSE;
2051     }
2052 
2053     ret = migrate_send_rp_message_req_pages(mis, rb, rb_offset);
2054     if (ret) {
2055         /* Please refer to above comment. */
2056         error_report("%s: send rp message failed for addr %p",
2057                      __func__, host_addr);
2058         return FALSE;
2059     }
2060 
2061     trace_postcopy_page_req_sync(host_addr);
2062 
2063     return FALSE;
2064 }
2065 
2066 static void migrate_send_rp_req_pages_pending(MigrationIncomingState *mis)
2067 {
2068     WITH_QEMU_LOCK_GUARD(&mis->page_request_mutex) {
2069         g_tree_foreach(mis->page_requested, postcopy_sync_page_req, mis);
2070     }
2071 }
2072 
2073 static int loadvm_postcopy_handle_resume(MigrationIncomingState *mis)
2074 {
2075     if (mis->state != MIGRATION_STATUS_POSTCOPY_RECOVER) {
2076         error_report("%s: illegal resume received", __func__);
2077         /* Don't fail the load, only for this. */
2078         return 0;
2079     }
2080 
2081     /*
2082      * Reset the last_rb before we resend any page req to source again, since
2083      * the source should have it reset already.
2084      */
2085     mis->last_rb = NULL;
2086 
2087     /*
2088      * This means source VM is ready to resume the postcopy migration.
2089      */
2090     migrate_set_state(&mis->state, MIGRATION_STATUS_POSTCOPY_RECOVER,
2091                       MIGRATION_STATUS_POSTCOPY_ACTIVE);
2092 
2093     trace_loadvm_postcopy_handle_resume();
2094 
2095     /* Tell source that "we are ready" */
2096     migrate_send_rp_resume_ack(mis, MIGRATION_RESUME_ACK_VALUE);
2097 
2098     /*
2099      * After a postcopy recovery, the source should have lost the postcopy
2100      * queue, or potentially the requested pages could have been lost during
2101      * the network down phase.  Let's re-sync with the source VM by re-sending
2102      * all the pending pages that we eagerly need, so these threads won't get
2103      * blocked too long due to the recovery.
2104      *
2105      * Without this procedure, the faulted destination VM threads (waiting for
2106      * page requests right before the postcopy is interrupted) can keep hanging
2107      * until the pages are sent by the source during the background copying of
2108      * pages, or another thread faulted on the same address accidentally.
2109      */
2110     migrate_send_rp_req_pages_pending(mis);
2111 
2112     /*
2113      * It's time to switch state and release the fault thread to continue
2114      * service page faults.  Note that this should be explicitly after the
2115      * above call to migrate_send_rp_req_pages_pending().  In short:
2116      * migrate_send_rp_message_req_pages() is not thread safe, yet.
2117      */
2118     qemu_sem_post(&mis->postcopy_pause_sem_fault);
2119 
2120     return 0;
2121 }
2122 
2123 /**
2124  * Immediately following this command is a blob of data containing an embedded
2125  * chunk of migration stream; read it and load it.
2126  *
2127  * @mis: Incoming state
2128  * @length: Length of packaged data to read
2129  *
2130  * Returns: Negative values on error
2131  *
2132  */
2133 static int loadvm_handle_cmd_packaged(MigrationIncomingState *mis)
2134 {
2135     int ret;
2136     size_t length;
2137     QIOChannelBuffer *bioc;
2138 
2139     length = qemu_get_be32(mis->from_src_file);
2140     trace_loadvm_handle_cmd_packaged(length);
2141 
2142     if (length > MAX_VM_CMD_PACKAGED_SIZE) {
2143         error_report("Unreasonably large packaged state: %zu", length);
2144         return -1;
2145     }
2146 
2147     bioc = qio_channel_buffer_new(length);
2148     qio_channel_set_name(QIO_CHANNEL(bioc), "migration-loadvm-buffer");
2149     ret = qemu_get_buffer(mis->from_src_file,
2150                           bioc->data,
2151                           length);
2152     if (ret != length) {
2153         object_unref(OBJECT(bioc));
2154         error_report("CMD_PACKAGED: Buffer receive fail ret=%d length=%zu",
2155                      ret, length);
2156         return (ret < 0) ? ret : -EAGAIN;
2157     }
2158     bioc->usage += length;
2159     trace_loadvm_handle_cmd_packaged_received(ret);
2160 
2161     QEMUFile *packf = qemu_file_new_input(QIO_CHANNEL(bioc));
2162 
2163     ret = qemu_loadvm_state_main(packf, mis);
2164     trace_loadvm_handle_cmd_packaged_main(ret);
2165     qemu_fclose(packf);
2166     object_unref(OBJECT(bioc));
2167 
2168     return ret;
2169 }
2170 
2171 /*
2172  * Handle request that source requests for recved_bitmap on
2173  * destination. Payload format:
2174  *
2175  * len (1 byte) + ramblock_name (<255 bytes)
2176  */
2177 static int loadvm_handle_recv_bitmap(MigrationIncomingState *mis,
2178                                      uint16_t len)
2179 {
2180     QEMUFile *file = mis->from_src_file;
2181     RAMBlock *rb;
2182     char block_name[256];
2183     size_t cnt;
2184 
2185     cnt = qemu_get_counted_string(file, block_name);
2186     if (!cnt) {
2187         error_report("%s: failed to read block name", __func__);
2188         return -EINVAL;
2189     }
2190 
2191     /* Validate before using the data */
2192     if (qemu_file_get_error(file)) {
2193         return qemu_file_get_error(file);
2194     }
2195 
2196     if (len != cnt + 1) {
2197         error_report("%s: invalid payload length (%d)", __func__, len);
2198         return -EINVAL;
2199     }
2200 
2201     rb = qemu_ram_block_by_name(block_name);
2202     if (!rb) {
2203         error_report("%s: block '%s' not found", __func__, block_name);
2204         return -EINVAL;
2205     }
2206 
2207     migrate_send_rp_recv_bitmap(mis, block_name);
2208 
2209     trace_loadvm_handle_recv_bitmap(block_name);
2210 
2211     return 0;
2212 }
2213 
2214 static int loadvm_process_enable_colo(MigrationIncomingState *mis)
2215 {
2216     int ret = migration_incoming_enable_colo();
2217 
2218     if (!ret) {
2219         ret = colo_init_ram_cache();
2220         if (ret) {
2221             migration_incoming_disable_colo();
2222         }
2223     }
2224     return ret;
2225 }
2226 
2227 /*
2228  * Process an incoming 'QEMU_VM_COMMAND'
2229  * 0           just a normal return
2230  * LOADVM_QUIT All good, but exit the loop
2231  * <0          Error
2232  */
2233 static int loadvm_process_command(QEMUFile *f)
2234 {
2235     MigrationIncomingState *mis = migration_incoming_get_current();
2236     uint16_t cmd;
2237     uint16_t len;
2238     uint32_t tmp32;
2239 
2240     cmd = qemu_get_be16(f);
2241     len = qemu_get_be16(f);
2242 
2243     /* Check validity before continue processing of cmds */
2244     if (qemu_file_get_error(f)) {
2245         return qemu_file_get_error(f);
2246     }
2247 
2248     if (cmd >= MIG_CMD_MAX || cmd == MIG_CMD_INVALID) {
2249         error_report("MIG_CMD 0x%x unknown (len 0x%x)", cmd, len);
2250         return -EINVAL;
2251     }
2252 
2253     trace_loadvm_process_command(mig_cmd_args[cmd].name, len);
2254 
2255     if (mig_cmd_args[cmd].len != -1 && mig_cmd_args[cmd].len != len) {
2256         error_report("%s received with bad length - expecting %zu, got %d",
2257                      mig_cmd_args[cmd].name,
2258                      (size_t)mig_cmd_args[cmd].len, len);
2259         return -ERANGE;
2260     }
2261 
2262     switch (cmd) {
2263     case MIG_CMD_OPEN_RETURN_PATH:
2264         if (mis->to_src_file) {
2265             error_report("CMD_OPEN_RETURN_PATH called when RP already open");
2266             /* Not really a problem, so don't give up */
2267             return 0;
2268         }
2269         mis->to_src_file = qemu_file_get_return_path(f);
2270         if (!mis->to_src_file) {
2271             error_report("CMD_OPEN_RETURN_PATH failed");
2272             return -1;
2273         }
2274         break;
2275 
2276     case MIG_CMD_PING:
2277         tmp32 = qemu_get_be32(f);
2278         trace_loadvm_process_command_ping(tmp32);
2279         if (!mis->to_src_file) {
2280             error_report("CMD_PING (0x%x) received with no return path",
2281                          tmp32);
2282             return -1;
2283         }
2284         migrate_send_rp_pong(mis, tmp32);
2285         break;
2286 
2287     case MIG_CMD_PACKAGED:
2288         return loadvm_handle_cmd_packaged(mis);
2289 
2290     case MIG_CMD_POSTCOPY_ADVISE:
2291         return loadvm_postcopy_handle_advise(mis, len);
2292 
2293     case MIG_CMD_POSTCOPY_LISTEN:
2294         return loadvm_postcopy_handle_listen(mis);
2295 
2296     case MIG_CMD_POSTCOPY_RUN:
2297         return loadvm_postcopy_handle_run(mis);
2298 
2299     case MIG_CMD_POSTCOPY_RAM_DISCARD:
2300         return loadvm_postcopy_ram_handle_discard(mis, len);
2301 
2302     case MIG_CMD_POSTCOPY_RESUME:
2303         return loadvm_postcopy_handle_resume(mis);
2304 
2305     case MIG_CMD_RECV_BITMAP:
2306         return loadvm_handle_recv_bitmap(mis, len);
2307 
2308     case MIG_CMD_ENABLE_COLO:
2309         return loadvm_process_enable_colo(mis);
2310     }
2311 
2312     return 0;
2313 }
2314 
2315 /*
2316  * Read a footer off the wire and check that it matches the expected section
2317  *
2318  * Returns: true if the footer was good
2319  *          false if there is a problem (and calls error_report to say why)
2320  */
2321 static bool check_section_footer(QEMUFile *f, SaveStateEntry *se)
2322 {
2323     int ret;
2324     uint8_t read_mark;
2325     uint32_t read_section_id;
2326 
2327     if (!migrate_get_current()->send_section_footer) {
2328         /* No footer to check */
2329         return true;
2330     }
2331 
2332     read_mark = qemu_get_byte(f);
2333 
2334     ret = qemu_file_get_error(f);
2335     if (ret) {
2336         error_report("%s: Read section footer failed: %d",
2337                      __func__, ret);
2338         return false;
2339     }
2340 
2341     if (read_mark != QEMU_VM_SECTION_FOOTER) {
2342         error_report("Missing section footer for %s", se->idstr);
2343         return false;
2344     }
2345 
2346     read_section_id = qemu_get_be32(f);
2347     if (read_section_id != se->load_section_id) {
2348         error_report("Mismatched section id in footer for %s -"
2349                      " read 0x%x expected 0x%x",
2350                      se->idstr, read_section_id, se->load_section_id);
2351         return false;
2352     }
2353 
2354     /* All good */
2355     return true;
2356 }
2357 
2358 static int
2359 qemu_loadvm_section_start_full(QEMUFile *f, MigrationIncomingState *mis)
2360 {
2361     uint32_t instance_id, version_id, section_id;
2362     SaveStateEntry *se;
2363     char idstr[256];
2364     int ret;
2365 
2366     /* Read section start */
2367     section_id = qemu_get_be32(f);
2368     if (!qemu_get_counted_string(f, idstr)) {
2369         error_report("Unable to read ID string for section %u",
2370                      section_id);
2371         return -EINVAL;
2372     }
2373     instance_id = qemu_get_be32(f);
2374     version_id = qemu_get_be32(f);
2375 
2376     ret = qemu_file_get_error(f);
2377     if (ret) {
2378         error_report("%s: Failed to read instance/version ID: %d",
2379                      __func__, ret);
2380         return ret;
2381     }
2382 
2383     trace_qemu_loadvm_state_section_startfull(section_id, idstr,
2384             instance_id, version_id);
2385     /* Find savevm section */
2386     se = find_se(idstr, instance_id);
2387     if (se == NULL) {
2388         error_report("Unknown savevm section or instance '%s' %"PRIu32". "
2389                      "Make sure that your current VM setup matches your "
2390                      "saved VM setup, including any hotplugged devices",
2391                      idstr, instance_id);
2392         return -EINVAL;
2393     }
2394 
2395     /* Validate version */
2396     if (version_id > se->version_id) {
2397         error_report("savevm: unsupported version %d for '%s' v%d",
2398                      version_id, idstr, se->version_id);
2399         return -EINVAL;
2400     }
2401     se->load_version_id = version_id;
2402     se->load_section_id = section_id;
2403 
2404     /* Validate if it is a device's state */
2405     if (xen_enabled() && se->is_ram) {
2406         error_report("loadvm: %s RAM loading not allowed on Xen", idstr);
2407         return -EINVAL;
2408     }
2409 
2410     ret = vmstate_load(f, se);
2411     if (ret < 0) {
2412         error_report("error while loading state for instance 0x%"PRIx32" of"
2413                      " device '%s'", instance_id, idstr);
2414         return ret;
2415     }
2416     if (!check_section_footer(f, se)) {
2417         return -EINVAL;
2418     }
2419 
2420     return 0;
2421 }
2422 
2423 static int
2424 qemu_loadvm_section_part_end(QEMUFile *f, MigrationIncomingState *mis)
2425 {
2426     uint32_t section_id;
2427     SaveStateEntry *se;
2428     int ret;
2429 
2430     section_id = qemu_get_be32(f);
2431 
2432     ret = qemu_file_get_error(f);
2433     if (ret) {
2434         error_report("%s: Failed to read section ID: %d",
2435                      __func__, ret);
2436         return ret;
2437     }
2438 
2439     trace_qemu_loadvm_state_section_partend(section_id);
2440     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
2441         if (se->load_section_id == section_id) {
2442             break;
2443         }
2444     }
2445     if (se == NULL) {
2446         error_report("Unknown savevm section %d", section_id);
2447         return -EINVAL;
2448     }
2449 
2450     ret = vmstate_load(f, se);
2451     if (ret < 0) {
2452         error_report("error while loading state section id %d(%s)",
2453                      section_id, se->idstr);
2454         return ret;
2455     }
2456     if (!check_section_footer(f, se)) {
2457         return -EINVAL;
2458     }
2459 
2460     return 0;
2461 }
2462 
2463 static int qemu_loadvm_state_header(QEMUFile *f)
2464 {
2465     unsigned int v;
2466     int ret;
2467 
2468     v = qemu_get_be32(f);
2469     if (v != QEMU_VM_FILE_MAGIC) {
2470         error_report("Not a migration stream");
2471         return -EINVAL;
2472     }
2473 
2474     v = qemu_get_be32(f);
2475     if (v == QEMU_VM_FILE_VERSION_COMPAT) {
2476         error_report("SaveVM v2 format is obsolete and don't work anymore");
2477         return -ENOTSUP;
2478     }
2479     if (v != QEMU_VM_FILE_VERSION) {
2480         error_report("Unsupported migration stream version");
2481         return -ENOTSUP;
2482     }
2483 
2484     if (migrate_get_current()->send_configuration) {
2485         if (qemu_get_byte(f) != QEMU_VM_CONFIGURATION) {
2486             error_report("Configuration section missing");
2487             qemu_loadvm_state_cleanup();
2488             return -EINVAL;
2489         }
2490         ret = vmstate_load_state(f, &vmstate_configuration, &savevm_state, 0);
2491 
2492         if (ret) {
2493             qemu_loadvm_state_cleanup();
2494             return ret;
2495         }
2496     }
2497     return 0;
2498 }
2499 
2500 static int qemu_loadvm_state_setup(QEMUFile *f)
2501 {
2502     SaveStateEntry *se;
2503     int ret;
2504 
2505     trace_loadvm_state_setup();
2506     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
2507         if (!se->ops || !se->ops->load_setup) {
2508             continue;
2509         }
2510         if (se->ops->is_active) {
2511             if (!se->ops->is_active(se->opaque)) {
2512                 continue;
2513             }
2514         }
2515 
2516         ret = se->ops->load_setup(f, se->opaque);
2517         if (ret < 0) {
2518             qemu_file_set_error(f, ret);
2519             error_report("Load state of device %s failed", se->idstr);
2520             return ret;
2521         }
2522     }
2523     return 0;
2524 }
2525 
2526 void qemu_loadvm_state_cleanup(void)
2527 {
2528     SaveStateEntry *se;
2529 
2530     trace_loadvm_state_cleanup();
2531     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
2532         if (se->ops && se->ops->load_cleanup) {
2533             se->ops->load_cleanup(se->opaque);
2534         }
2535     }
2536 }
2537 
2538 /* Return true if we should continue the migration, or false. */
2539 static bool postcopy_pause_incoming(MigrationIncomingState *mis)
2540 {
2541     int i;
2542 
2543     /*
2544      * If network is interrupted, any temp page we received will be useless
2545      * because we didn't mark them as "received" in receivedmap.  After a
2546      * proper recovery later (which will sync src dirty bitmap with receivedmap
2547      * on dest) these cached small pages will be resent again.
2548      */
2549     for (i = 0; i < mis->postcopy_channels; i++) {
2550         postcopy_temp_page_reset(&mis->postcopy_tmp_pages[i]);
2551     }
2552 
2553     trace_postcopy_pause_incoming();
2554 
2555     assert(migrate_postcopy_ram());
2556 
2557     /*
2558      * Unregister yank with either from/to src would work, since ioc behind it
2559      * is the same
2560      */
2561     migration_ioc_unregister_yank_from_file(mis->from_src_file);
2562 
2563     assert(mis->from_src_file);
2564     qemu_file_shutdown(mis->from_src_file);
2565     qemu_fclose(mis->from_src_file);
2566     mis->from_src_file = NULL;
2567 
2568     assert(mis->to_src_file);
2569     qemu_file_shutdown(mis->to_src_file);
2570     qemu_mutex_lock(&mis->rp_mutex);
2571     qemu_fclose(mis->to_src_file);
2572     mis->to_src_file = NULL;
2573     qemu_mutex_unlock(&mis->rp_mutex);
2574 
2575     migrate_set_state(&mis->state, MIGRATION_STATUS_POSTCOPY_ACTIVE,
2576                       MIGRATION_STATUS_POSTCOPY_PAUSED);
2577 
2578     /* Notify the fault thread for the invalidated file handle */
2579     postcopy_fault_thread_notify(mis);
2580 
2581     error_report("Detected IO failure for postcopy. "
2582                  "Migration paused.");
2583 
2584     while (mis->state == MIGRATION_STATUS_POSTCOPY_PAUSED) {
2585         qemu_sem_wait(&mis->postcopy_pause_sem_dst);
2586     }
2587 
2588     trace_postcopy_pause_incoming_continued();
2589 
2590     return true;
2591 }
2592 
2593 int qemu_loadvm_state_main(QEMUFile *f, MigrationIncomingState *mis)
2594 {
2595     uint8_t section_type;
2596     int ret = 0;
2597 
2598 retry:
2599     while (true) {
2600         section_type = qemu_get_byte(f);
2601 
2602         if (qemu_file_get_error(f)) {
2603             ret = qemu_file_get_error(f);
2604             break;
2605         }
2606 
2607         trace_qemu_loadvm_state_section(section_type);
2608         switch (section_type) {
2609         case QEMU_VM_SECTION_START:
2610         case QEMU_VM_SECTION_FULL:
2611             ret = qemu_loadvm_section_start_full(f, mis);
2612             if (ret < 0) {
2613                 goto out;
2614             }
2615             break;
2616         case QEMU_VM_SECTION_PART:
2617         case QEMU_VM_SECTION_END:
2618             ret = qemu_loadvm_section_part_end(f, mis);
2619             if (ret < 0) {
2620                 goto out;
2621             }
2622             break;
2623         case QEMU_VM_COMMAND:
2624             ret = loadvm_process_command(f);
2625             trace_qemu_loadvm_state_section_command(ret);
2626             if ((ret < 0) || (ret == LOADVM_QUIT)) {
2627                 goto out;
2628             }
2629             break;
2630         case QEMU_VM_EOF:
2631             /* This is the end of migration */
2632             goto out;
2633         default:
2634             error_report("Unknown savevm section type %d", section_type);
2635             ret = -EINVAL;
2636             goto out;
2637         }
2638     }
2639 
2640 out:
2641     if (ret < 0) {
2642         qemu_file_set_error(f, ret);
2643 
2644         /* Cancel bitmaps incoming regardless of recovery */
2645         dirty_bitmap_mig_cancel_incoming();
2646 
2647         /*
2648          * If we are during an active postcopy, then we pause instead
2649          * of bail out to at least keep the VM's dirty data.  Note
2650          * that POSTCOPY_INCOMING_LISTENING stage is still not enough,
2651          * during which we're still receiving device states and we
2652          * still haven't yet started the VM on destination.
2653          *
2654          * Only RAM postcopy supports recovery. Still, if RAM postcopy is
2655          * enabled, canceled bitmaps postcopy will not affect RAM postcopy
2656          * recovering.
2657          */
2658         if (postcopy_state_get() == POSTCOPY_INCOMING_RUNNING &&
2659             migrate_postcopy_ram() && postcopy_pause_incoming(mis)) {
2660             /* Reset f to point to the newly created channel */
2661             f = mis->from_src_file;
2662             goto retry;
2663         }
2664     }
2665     return ret;
2666 }
2667 
2668 int qemu_loadvm_state(QEMUFile *f)
2669 {
2670     MigrationIncomingState *mis = migration_incoming_get_current();
2671     Error *local_err = NULL;
2672     int ret;
2673 
2674     if (qemu_savevm_state_blocked(&local_err)) {
2675         error_report_err(local_err);
2676         return -EINVAL;
2677     }
2678 
2679     ret = qemu_loadvm_state_header(f);
2680     if (ret) {
2681         return ret;
2682     }
2683 
2684     if (qemu_loadvm_state_setup(f) != 0) {
2685         return -EINVAL;
2686     }
2687 
2688     cpu_synchronize_all_pre_loadvm();
2689 
2690     ret = qemu_loadvm_state_main(f, mis);
2691     qemu_event_set(&mis->main_thread_load_event);
2692 
2693     trace_qemu_loadvm_state_post_main(ret);
2694 
2695     if (mis->have_listen_thread) {
2696         /* Listen thread still going, can't clean up yet */
2697         return ret;
2698     }
2699 
2700     if (ret == 0) {
2701         ret = qemu_file_get_error(f);
2702     }
2703 
2704     /*
2705      * Try to read in the VMDESC section as well, so that dumping tools that
2706      * intercept our migration stream have the chance to see it.
2707      */
2708 
2709     /* We've got to be careful; if we don't read the data and just shut the fd
2710      * then the sender can error if we close while it's still sending.
2711      * We also mustn't read data that isn't there; some transports (RDMA)
2712      * will stall waiting for that data when the source has already closed.
2713      */
2714     if (ret == 0 && should_send_vmdesc()) {
2715         uint8_t *buf;
2716         uint32_t size;
2717         uint8_t  section_type = qemu_get_byte(f);
2718 
2719         if (section_type != QEMU_VM_VMDESCRIPTION) {
2720             error_report("Expected vmdescription section, but got %d",
2721                          section_type);
2722             /*
2723              * It doesn't seem worth failing at this point since
2724              * we apparently have an otherwise valid VM state
2725              */
2726         } else {
2727             buf = g_malloc(0x1000);
2728             size = qemu_get_be32(f);
2729 
2730             while (size > 0) {
2731                 uint32_t read_chunk = MIN(size, 0x1000);
2732                 qemu_get_buffer(f, buf, read_chunk);
2733                 size -= read_chunk;
2734             }
2735             g_free(buf);
2736         }
2737     }
2738 
2739     qemu_loadvm_state_cleanup();
2740     cpu_synchronize_all_post_init();
2741 
2742     return ret;
2743 }
2744 
2745 int qemu_load_device_state(QEMUFile *f)
2746 {
2747     MigrationIncomingState *mis = migration_incoming_get_current();
2748     int ret;
2749 
2750     /* Load QEMU_VM_SECTION_FULL section */
2751     ret = qemu_loadvm_state_main(f, mis);
2752     if (ret < 0) {
2753         error_report("Failed to load device state: %d", ret);
2754         return ret;
2755     }
2756 
2757     cpu_synchronize_all_post_init();
2758     return 0;
2759 }
2760 
2761 bool save_snapshot(const char *name, bool overwrite, const char *vmstate,
2762                   bool has_devices, strList *devices, Error **errp)
2763 {
2764     BlockDriverState *bs;
2765     QEMUSnapshotInfo sn1, *sn = &sn1;
2766     int ret = -1, ret2;
2767     QEMUFile *f;
2768     int saved_vm_running;
2769     uint64_t vm_state_size;
2770     g_autoptr(GDateTime) now = g_date_time_new_now_local();
2771     AioContext *aio_context;
2772 
2773     GLOBAL_STATE_CODE();
2774 
2775     if (migration_is_blocked(errp)) {
2776         return false;
2777     }
2778 
2779     if (!replay_can_snapshot()) {
2780         error_setg(errp, "Record/replay does not allow making snapshot "
2781                    "right now. Try once more later.");
2782         return false;
2783     }
2784 
2785     if (!bdrv_all_can_snapshot(has_devices, devices, errp)) {
2786         return false;
2787     }
2788 
2789     /* Delete old snapshots of the same name */
2790     if (name) {
2791         if (overwrite) {
2792             if (bdrv_all_delete_snapshot(name, has_devices,
2793                                          devices, errp) < 0) {
2794                 return false;
2795             }
2796         } else {
2797             ret2 = bdrv_all_has_snapshot(name, has_devices, devices, errp);
2798             if (ret2 < 0) {
2799                 return false;
2800             }
2801             if (ret2 == 1) {
2802                 error_setg(errp,
2803                            "Snapshot '%s' already exists in one or more devices",
2804                            name);
2805                 return false;
2806             }
2807         }
2808     }
2809 
2810     bs = bdrv_all_find_vmstate_bs(vmstate, has_devices, devices, errp);
2811     if (bs == NULL) {
2812         return false;
2813     }
2814     aio_context = bdrv_get_aio_context(bs);
2815 
2816     saved_vm_running = runstate_is_running();
2817 
2818     ret = global_state_store();
2819     if (ret) {
2820         error_setg(errp, "Error saving global state");
2821         return false;
2822     }
2823     vm_stop(RUN_STATE_SAVE_VM);
2824 
2825     bdrv_drain_all_begin();
2826 
2827     aio_context_acquire(aio_context);
2828 
2829     memset(sn, 0, sizeof(*sn));
2830 
2831     /* fill auxiliary fields */
2832     sn->date_sec = g_date_time_to_unix(now);
2833     sn->date_nsec = g_date_time_get_microsecond(now) * 1000;
2834     sn->vm_clock_nsec = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL);
2835     if (replay_mode != REPLAY_MODE_NONE) {
2836         sn->icount = replay_get_current_icount();
2837     } else {
2838         sn->icount = -1ULL;
2839     }
2840 
2841     if (name) {
2842         pstrcpy(sn->name, sizeof(sn->name), name);
2843     } else {
2844         g_autofree char *autoname = g_date_time_format(now,  "vm-%Y%m%d%H%M%S");
2845         pstrcpy(sn->name, sizeof(sn->name), autoname);
2846     }
2847 
2848     /* save the VM state */
2849     f = qemu_fopen_bdrv(bs, 1);
2850     if (!f) {
2851         error_setg(errp, "Could not open VM state file");
2852         goto the_end;
2853     }
2854     ret = qemu_savevm_state(f, errp);
2855     vm_state_size = qemu_file_total_transferred(f);
2856     ret2 = qemu_fclose(f);
2857     if (ret < 0) {
2858         goto the_end;
2859     }
2860     if (ret2 < 0) {
2861         ret = ret2;
2862         goto the_end;
2863     }
2864 
2865     /* The bdrv_all_create_snapshot() call that follows acquires the AioContext
2866      * for itself.  BDRV_POLL_WHILE() does not support nested locking because
2867      * it only releases the lock once.  Therefore synchronous I/O will deadlock
2868      * unless we release the AioContext before bdrv_all_create_snapshot().
2869      */
2870     aio_context_release(aio_context);
2871     aio_context = NULL;
2872 
2873     ret = bdrv_all_create_snapshot(sn, bs, vm_state_size,
2874                                    has_devices, devices, errp);
2875     if (ret < 0) {
2876         bdrv_all_delete_snapshot(sn->name, has_devices, devices, NULL);
2877         goto the_end;
2878     }
2879 
2880     ret = 0;
2881 
2882  the_end:
2883     if (aio_context) {
2884         aio_context_release(aio_context);
2885     }
2886 
2887     bdrv_drain_all_end();
2888 
2889     if (saved_vm_running) {
2890         vm_start();
2891     }
2892     return ret == 0;
2893 }
2894 
2895 void qmp_xen_save_devices_state(const char *filename, bool has_live, bool live,
2896                                 Error **errp)
2897 {
2898     QEMUFile *f;
2899     QIOChannelFile *ioc;
2900     int saved_vm_running;
2901     int ret;
2902 
2903     if (!has_live) {
2904         /* live default to true so old version of Xen tool stack can have a
2905          * successful live migration */
2906         live = true;
2907     }
2908 
2909     saved_vm_running = runstate_is_running();
2910     vm_stop(RUN_STATE_SAVE_VM);
2911     global_state_store_running();
2912 
2913     ioc = qio_channel_file_new_path(filename, O_WRONLY | O_CREAT | O_TRUNC,
2914                                     0660, errp);
2915     if (!ioc) {
2916         goto the_end;
2917     }
2918     qio_channel_set_name(QIO_CHANNEL(ioc), "migration-xen-save-state");
2919     f = qemu_file_new_output(QIO_CHANNEL(ioc));
2920     object_unref(OBJECT(ioc));
2921     ret = qemu_save_device_state(f);
2922     if (ret < 0 || qemu_fclose(f) < 0) {
2923         error_setg(errp, QERR_IO_ERROR);
2924     } else {
2925         /* libxl calls the QMP command "stop" before calling
2926          * "xen-save-devices-state" and in case of migration failure, libxl
2927          * would call "cont".
2928          * So call bdrv_inactivate_all (release locks) here to let the other
2929          * side of the migration take control of the images.
2930          */
2931         if (live && !saved_vm_running) {
2932             ret = bdrv_inactivate_all();
2933             if (ret) {
2934                 error_setg(errp, "%s: bdrv_inactivate_all() failed (%d)",
2935                            __func__, ret);
2936             }
2937         }
2938     }
2939 
2940  the_end:
2941     if (saved_vm_running) {
2942         vm_start();
2943     }
2944 }
2945 
2946 void qmp_xen_load_devices_state(const char *filename, Error **errp)
2947 {
2948     QEMUFile *f;
2949     QIOChannelFile *ioc;
2950     int ret;
2951 
2952     /* Guest must be paused before loading the device state; the RAM state
2953      * will already have been loaded by xc
2954      */
2955     if (runstate_is_running()) {
2956         error_setg(errp, "Cannot update device state while vm is running");
2957         return;
2958     }
2959     vm_stop(RUN_STATE_RESTORE_VM);
2960 
2961     ioc = qio_channel_file_new_path(filename, O_RDONLY | O_BINARY, 0, errp);
2962     if (!ioc) {
2963         return;
2964     }
2965     qio_channel_set_name(QIO_CHANNEL(ioc), "migration-xen-load-state");
2966     f = qemu_file_new_input(QIO_CHANNEL(ioc));
2967     object_unref(OBJECT(ioc));
2968 
2969     ret = qemu_loadvm_state(f);
2970     qemu_fclose(f);
2971     if (ret < 0) {
2972         error_setg(errp, QERR_IO_ERROR);
2973     }
2974     migration_incoming_state_destroy();
2975 }
2976 
2977 bool load_snapshot(const char *name, const char *vmstate,
2978                    bool has_devices, strList *devices, Error **errp)
2979 {
2980     BlockDriverState *bs_vm_state;
2981     QEMUSnapshotInfo sn;
2982     QEMUFile *f;
2983     int ret;
2984     AioContext *aio_context;
2985     MigrationIncomingState *mis = migration_incoming_get_current();
2986 
2987     if (!bdrv_all_can_snapshot(has_devices, devices, errp)) {
2988         return false;
2989     }
2990     ret = bdrv_all_has_snapshot(name, has_devices, devices, errp);
2991     if (ret < 0) {
2992         return false;
2993     }
2994     if (ret == 0) {
2995         error_setg(errp, "Snapshot '%s' does not exist in one or more devices",
2996                    name);
2997         return false;
2998     }
2999 
3000     bs_vm_state = bdrv_all_find_vmstate_bs(vmstate, has_devices, devices, errp);
3001     if (!bs_vm_state) {
3002         return false;
3003     }
3004     aio_context = bdrv_get_aio_context(bs_vm_state);
3005 
3006     /* Don't even try to load empty VM states */
3007     aio_context_acquire(aio_context);
3008     ret = bdrv_snapshot_find(bs_vm_state, &sn, name);
3009     aio_context_release(aio_context);
3010     if (ret < 0) {
3011         return false;
3012     } else if (sn.vm_state_size == 0) {
3013         error_setg(errp, "This is a disk-only snapshot. Revert to it "
3014                    " offline using qemu-img");
3015         return false;
3016     }
3017 
3018     /*
3019      * Flush the record/replay queue. Now the VM state is going
3020      * to change. Therefore we don't need to preserve its consistency
3021      */
3022     replay_flush_events();
3023 
3024     /* Flush all IO requests so they don't interfere with the new state.  */
3025     bdrv_drain_all_begin();
3026 
3027     ret = bdrv_all_goto_snapshot(name, has_devices, devices, errp);
3028     if (ret < 0) {
3029         goto err_drain;
3030     }
3031 
3032     /* restore the VM state */
3033     f = qemu_fopen_bdrv(bs_vm_state, 0);
3034     if (!f) {
3035         error_setg(errp, "Could not open VM state file");
3036         goto err_drain;
3037     }
3038 
3039     qemu_system_reset(SHUTDOWN_CAUSE_NONE);
3040     mis->from_src_file = f;
3041 
3042     if (!yank_register_instance(MIGRATION_YANK_INSTANCE, errp)) {
3043         ret = -EINVAL;
3044         goto err_drain;
3045     }
3046     aio_context_acquire(aio_context);
3047     ret = qemu_loadvm_state(f);
3048     migration_incoming_state_destroy();
3049     aio_context_release(aio_context);
3050 
3051     bdrv_drain_all_end();
3052 
3053     if (ret < 0) {
3054         error_setg(errp, "Error %d while loading VM state", ret);
3055         return false;
3056     }
3057 
3058     return true;
3059 
3060 err_drain:
3061     bdrv_drain_all_end();
3062     return false;
3063 }
3064 
3065 bool delete_snapshot(const char *name, bool has_devices,
3066                      strList *devices, Error **errp)
3067 {
3068     if (!bdrv_all_can_snapshot(has_devices, devices, errp)) {
3069         return false;
3070     }
3071 
3072     if (bdrv_all_delete_snapshot(name, has_devices, devices, errp) < 0) {
3073         return false;
3074     }
3075 
3076     return true;
3077 }
3078 
3079 void vmstate_register_ram(MemoryRegion *mr, DeviceState *dev)
3080 {
3081     qemu_ram_set_idstr(mr->ram_block,
3082                        memory_region_name(mr), dev);
3083     qemu_ram_set_migratable(mr->ram_block);
3084 }
3085 
3086 void vmstate_unregister_ram(MemoryRegion *mr, DeviceState *dev)
3087 {
3088     qemu_ram_unset_idstr(mr->ram_block);
3089     qemu_ram_unset_migratable(mr->ram_block);
3090 }
3091 
3092 void vmstate_register_ram_global(MemoryRegion *mr)
3093 {
3094     vmstate_register_ram(mr, NULL);
3095 }
3096 
3097 bool vmstate_check_only_migratable(const VMStateDescription *vmsd)
3098 {
3099     /* check needed if --only-migratable is specified */
3100     if (!only_migratable) {
3101         return true;
3102     }
3103 
3104     return !(vmsd && vmsd->unmigratable);
3105 }
3106 
3107 typedef struct SnapshotJob {
3108     Job common;
3109     char *tag;
3110     char *vmstate;
3111     strList *devices;
3112     Coroutine *co;
3113     Error **errp;
3114     bool ret;
3115 } SnapshotJob;
3116 
3117 static void qmp_snapshot_job_free(SnapshotJob *s)
3118 {
3119     g_free(s->tag);
3120     g_free(s->vmstate);
3121     qapi_free_strList(s->devices);
3122 }
3123 
3124 
3125 static void snapshot_load_job_bh(void *opaque)
3126 {
3127     Job *job = opaque;
3128     SnapshotJob *s = container_of(job, SnapshotJob, common);
3129     int orig_vm_running;
3130 
3131     job_progress_set_remaining(&s->common, 1);
3132 
3133     orig_vm_running = runstate_is_running();
3134     vm_stop(RUN_STATE_RESTORE_VM);
3135 
3136     s->ret = load_snapshot(s->tag, s->vmstate, true, s->devices, s->errp);
3137     if (s->ret && orig_vm_running) {
3138         vm_start();
3139     }
3140 
3141     job_progress_update(&s->common, 1);
3142 
3143     qmp_snapshot_job_free(s);
3144     aio_co_wake(s->co);
3145 }
3146 
3147 static void snapshot_save_job_bh(void *opaque)
3148 {
3149     Job *job = opaque;
3150     SnapshotJob *s = container_of(job, SnapshotJob, common);
3151 
3152     job_progress_set_remaining(&s->common, 1);
3153     s->ret = save_snapshot(s->tag, false, s->vmstate,
3154                            true, s->devices, s->errp);
3155     job_progress_update(&s->common, 1);
3156 
3157     qmp_snapshot_job_free(s);
3158     aio_co_wake(s->co);
3159 }
3160 
3161 static void snapshot_delete_job_bh(void *opaque)
3162 {
3163     Job *job = opaque;
3164     SnapshotJob *s = container_of(job, SnapshotJob, common);
3165 
3166     job_progress_set_remaining(&s->common, 1);
3167     s->ret = delete_snapshot(s->tag, true, s->devices, s->errp);
3168     job_progress_update(&s->common, 1);
3169 
3170     qmp_snapshot_job_free(s);
3171     aio_co_wake(s->co);
3172 }
3173 
3174 static int coroutine_fn snapshot_save_job_run(Job *job, Error **errp)
3175 {
3176     SnapshotJob *s = container_of(job, SnapshotJob, common);
3177     s->errp = errp;
3178     s->co = qemu_coroutine_self();
3179     aio_bh_schedule_oneshot(qemu_get_aio_context(),
3180                             snapshot_save_job_bh, job);
3181     qemu_coroutine_yield();
3182     return s->ret ? 0 : -1;
3183 }
3184 
3185 static int coroutine_fn snapshot_load_job_run(Job *job, Error **errp)
3186 {
3187     SnapshotJob *s = container_of(job, SnapshotJob, common);
3188     s->errp = errp;
3189     s->co = qemu_coroutine_self();
3190     aio_bh_schedule_oneshot(qemu_get_aio_context(),
3191                             snapshot_load_job_bh, job);
3192     qemu_coroutine_yield();
3193     return s->ret ? 0 : -1;
3194 }
3195 
3196 static int coroutine_fn snapshot_delete_job_run(Job *job, Error **errp)
3197 {
3198     SnapshotJob *s = container_of(job, SnapshotJob, common);
3199     s->errp = errp;
3200     s->co = qemu_coroutine_self();
3201     aio_bh_schedule_oneshot(qemu_get_aio_context(),
3202                             snapshot_delete_job_bh, job);
3203     qemu_coroutine_yield();
3204     return s->ret ? 0 : -1;
3205 }
3206 
3207 
3208 static const JobDriver snapshot_load_job_driver = {
3209     .instance_size = sizeof(SnapshotJob),
3210     .job_type      = JOB_TYPE_SNAPSHOT_LOAD,
3211     .run           = snapshot_load_job_run,
3212 };
3213 
3214 static const JobDriver snapshot_save_job_driver = {
3215     .instance_size = sizeof(SnapshotJob),
3216     .job_type      = JOB_TYPE_SNAPSHOT_SAVE,
3217     .run           = snapshot_save_job_run,
3218 };
3219 
3220 static const JobDriver snapshot_delete_job_driver = {
3221     .instance_size = sizeof(SnapshotJob),
3222     .job_type      = JOB_TYPE_SNAPSHOT_DELETE,
3223     .run           = snapshot_delete_job_run,
3224 };
3225 
3226 
3227 void qmp_snapshot_save(const char *job_id,
3228                        const char *tag,
3229                        const char *vmstate,
3230                        strList *devices,
3231                        Error **errp)
3232 {
3233     SnapshotJob *s;
3234 
3235     s = job_create(job_id, &snapshot_save_job_driver, NULL,
3236                    qemu_get_aio_context(), JOB_MANUAL_DISMISS,
3237                    NULL, NULL, errp);
3238     if (!s) {
3239         return;
3240     }
3241 
3242     s->tag = g_strdup(tag);
3243     s->vmstate = g_strdup(vmstate);
3244     s->devices = QAPI_CLONE(strList, devices);
3245 
3246     job_start(&s->common);
3247 }
3248 
3249 void qmp_snapshot_load(const char *job_id,
3250                        const char *tag,
3251                        const char *vmstate,
3252                        strList *devices,
3253                        Error **errp)
3254 {
3255     SnapshotJob *s;
3256 
3257     s = job_create(job_id, &snapshot_load_job_driver, NULL,
3258                    qemu_get_aio_context(), JOB_MANUAL_DISMISS,
3259                    NULL, NULL, errp);
3260     if (!s) {
3261         return;
3262     }
3263 
3264     s->tag = g_strdup(tag);
3265     s->vmstate = g_strdup(vmstate);
3266     s->devices = QAPI_CLONE(strList, devices);
3267 
3268     job_start(&s->common);
3269 }
3270 
3271 void qmp_snapshot_delete(const char *job_id,
3272                          const char *tag,
3273                          strList *devices,
3274                          Error **errp)
3275 {
3276     SnapshotJob *s;
3277 
3278     s = job_create(job_id, &snapshot_delete_job_driver, NULL,
3279                    qemu_get_aio_context(), JOB_MANUAL_DISMISS,
3280                    NULL, NULL, errp);
3281     if (!s) {
3282         return;
3283     }
3284 
3285     s->tag = g_strdup(tag);
3286     s->devices = QAPI_CLONE(strList, devices);
3287 
3288     job_start(&s->common);
3289 }
3290