xref: /openbmc/qemu/migration/savevm.c (revision 6c35ed68)
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 "hw/xen/xen.h"
32 #include "net/net.h"
33 #include "migration.h"
34 #include "migration/snapshot.h"
35 #include "migration/vmstate.h"
36 #include "migration/misc.h"
37 #include "migration/register.h"
38 #include "migration/global_state.h"
39 #include "ram.h"
40 #include "qemu-file-channel.h"
41 #include "qemu-file.h"
42 #include "savevm.h"
43 #include "postcopy-ram.h"
44 #include "qapi/error.h"
45 #include "qapi/qapi-commands-migration.h"
46 #include "qapi/qapi-commands-misc.h"
47 #include "qapi/qmp/qerror.h"
48 #include "qemu/error-report.h"
49 #include "sysemu/cpus.h"
50 #include "exec/memory.h"
51 #include "exec/target_page.h"
52 #include "trace.h"
53 #include "qemu/iov.h"
54 #include "qemu/main-loop.h"
55 #include "block/snapshot.h"
56 #include "qemu/cutils.h"
57 #include "io/channel-buffer.h"
58 #include "io/channel-file.h"
59 #include "sysemu/replay.h"
60 #include "sysemu/runstate.h"
61 #include "sysemu/sysemu.h"
62 #include "qjson.h"
63 #include "migration/colo.h"
64 #include "qemu/bitmap.h"
65 #include "net/announce.h"
66 
67 const unsigned int postcopy_ram_discard_version = 0;
68 
69 /* Subcommands for QEMU_VM_COMMAND */
70 enum qemu_vm_cmd {
71     MIG_CMD_INVALID = 0,   /* Must be 0 */
72     MIG_CMD_OPEN_RETURN_PATH,  /* Tell the dest to open the Return path */
73     MIG_CMD_PING,              /* Request a PONG on the RP */
74 
75     MIG_CMD_POSTCOPY_ADVISE,       /* Prior to any page transfers, just
76                                       warn we might want to do PC */
77     MIG_CMD_POSTCOPY_LISTEN,       /* Start listening for incoming
78                                       pages as it's running. */
79     MIG_CMD_POSTCOPY_RUN,          /* Start execution */
80 
81     MIG_CMD_POSTCOPY_RAM_DISCARD,  /* A list of pages to discard that
82                                       were previously sent during
83                                       precopy but are dirty. */
84     MIG_CMD_PACKAGED,          /* Send a wrapped stream within this stream */
85     MIG_CMD_ENABLE_COLO,       /* Enable COLO */
86     MIG_CMD_POSTCOPY_RESUME,   /* resume postcopy on dest */
87     MIG_CMD_RECV_BITMAP,       /* Request for recved bitmap on dst */
88     MIG_CMD_MAX
89 };
90 
91 #define MAX_VM_CMD_PACKAGED_SIZE UINT32_MAX
92 static struct mig_cmd_args {
93     ssize_t     len; /* -1 = variable */
94     const char *name;
95 } mig_cmd_args[] = {
96     [MIG_CMD_INVALID]          = { .len = -1, .name = "INVALID" },
97     [MIG_CMD_OPEN_RETURN_PATH] = { .len =  0, .name = "OPEN_RETURN_PATH" },
98     [MIG_CMD_PING]             = { .len = sizeof(uint32_t), .name = "PING" },
99     [MIG_CMD_POSTCOPY_ADVISE]  = { .len = -1, .name = "POSTCOPY_ADVISE" },
100     [MIG_CMD_POSTCOPY_LISTEN]  = { .len =  0, .name = "POSTCOPY_LISTEN" },
101     [MIG_CMD_POSTCOPY_RUN]     = { .len =  0, .name = "POSTCOPY_RUN" },
102     [MIG_CMD_POSTCOPY_RAM_DISCARD] = {
103                                    .len = -1, .name = "POSTCOPY_RAM_DISCARD" },
104     [MIG_CMD_POSTCOPY_RESUME]  = { .len =  0, .name = "POSTCOPY_RESUME" },
105     [MIG_CMD_PACKAGED]         = { .len =  4, .name = "PACKAGED" },
106     [MIG_CMD_RECV_BITMAP]      = { .len = -1, .name = "RECV_BITMAP" },
107     [MIG_CMD_MAX]              = { .len = -1, .name = "MAX" },
108 };
109 
110 /* Note for MIG_CMD_POSTCOPY_ADVISE:
111  * The format of arguments is depending on postcopy mode:
112  * - postcopy RAM only
113  *   uint64_t host page size
114  *   uint64_t taget page size
115  *
116  * - postcopy RAM and postcopy dirty bitmaps
117  *   format is the same as for postcopy RAM only
118  *
119  * - postcopy dirty bitmaps only
120  *   Nothing. Command length field is 0.
121  *
122  * Be careful: adding a new postcopy entity with some other parameters should
123  * not break format self-description ability. Good way is to introduce some
124  * generic extendable format with an exception for two old entities.
125  */
126 
127 /***********************************************************/
128 /* savevm/loadvm support */
129 
130 static ssize_t block_writev_buffer(void *opaque, struct iovec *iov, int iovcnt,
131                                    int64_t pos, Error **errp)
132 {
133     int ret;
134     QEMUIOVector qiov;
135 
136     qemu_iovec_init_external(&qiov, iov, iovcnt);
137     ret = bdrv_writev_vmstate(opaque, &qiov, pos);
138     if (ret < 0) {
139         return ret;
140     }
141 
142     return qiov.size;
143 }
144 
145 static ssize_t block_get_buffer(void *opaque, uint8_t *buf, int64_t pos,
146                                 size_t size, Error **errp)
147 {
148     return bdrv_load_vmstate(opaque, buf, pos, size);
149 }
150 
151 static int bdrv_fclose(void *opaque, Error **errp)
152 {
153     return bdrv_flush(opaque);
154 }
155 
156 static const QEMUFileOps bdrv_read_ops = {
157     .get_buffer = block_get_buffer,
158     .close =      bdrv_fclose
159 };
160 
161 static const QEMUFileOps bdrv_write_ops = {
162     .writev_buffer  = block_writev_buffer,
163     .close          = bdrv_fclose
164 };
165 
166 static QEMUFile *qemu_fopen_bdrv(BlockDriverState *bs, int is_writable)
167 {
168     if (is_writable) {
169         return qemu_fopen_ops(bs, &bdrv_write_ops);
170     }
171     return qemu_fopen_ops(bs, &bdrv_read_ops);
172 }
173 
174 
175 /* QEMUFile timer support.
176  * Not in qemu-file.c to not add qemu-timer.c as dependency to qemu-file.c
177  */
178 
179 void timer_put(QEMUFile *f, QEMUTimer *ts)
180 {
181     uint64_t expire_time;
182 
183     expire_time = timer_expire_time_ns(ts);
184     qemu_put_be64(f, expire_time);
185 }
186 
187 void timer_get(QEMUFile *f, QEMUTimer *ts)
188 {
189     uint64_t expire_time;
190 
191     expire_time = qemu_get_be64(f);
192     if (expire_time != -1) {
193         timer_mod_ns(ts, expire_time);
194     } else {
195         timer_del(ts);
196     }
197 }
198 
199 
200 /* VMState timer support.
201  * Not in vmstate.c to not add qemu-timer.c as dependency to vmstate.c
202  */
203 
204 static int get_timer(QEMUFile *f, void *pv, size_t size,
205                      const VMStateField *field)
206 {
207     QEMUTimer *v = pv;
208     timer_get(f, v);
209     return 0;
210 }
211 
212 static int put_timer(QEMUFile *f, void *pv, size_t size,
213                      const VMStateField *field, QJSON *vmdesc)
214 {
215     QEMUTimer *v = pv;
216     timer_put(f, v);
217 
218     return 0;
219 }
220 
221 const VMStateInfo vmstate_info_timer = {
222     .name = "timer",
223     .get  = get_timer,
224     .put  = put_timer,
225 };
226 
227 
228 typedef struct CompatEntry {
229     char idstr[256];
230     int instance_id;
231 } CompatEntry;
232 
233 typedef struct SaveStateEntry {
234     QTAILQ_ENTRY(SaveStateEntry) entry;
235     char idstr[256];
236     int instance_id;
237     int alias_id;
238     int version_id;
239     /* version id read from the stream */
240     int load_version_id;
241     int section_id;
242     /* section id read from the stream */
243     int load_section_id;
244     const SaveVMHandlers *ops;
245     const VMStateDescription *vmsd;
246     void *opaque;
247     CompatEntry *compat;
248     int is_ram;
249 } SaveStateEntry;
250 
251 typedef struct SaveState {
252     QTAILQ_HEAD(, SaveStateEntry) handlers;
253     int global_section_id;
254     uint32_t len;
255     const char *name;
256     uint32_t target_page_bits;
257     uint32_t caps_count;
258     MigrationCapability *capabilities;
259     QemuUUID uuid;
260 } SaveState;
261 
262 static SaveState savevm_state = {
263     .handlers = QTAILQ_HEAD_INITIALIZER(savevm_state.handlers),
264     .global_section_id = 0,
265 };
266 
267 static bool should_validate_capability(int capability)
268 {
269     assert(capability >= 0 && capability < MIGRATION_CAPABILITY__MAX);
270     /* Validate only new capabilities to keep compatibility. */
271     switch (capability) {
272     case MIGRATION_CAPABILITY_X_IGNORE_SHARED:
273         return true;
274     default:
275         return false;
276     }
277 }
278 
279 static uint32_t get_validatable_capabilities_count(void)
280 {
281     MigrationState *s = migrate_get_current();
282     uint32_t result = 0;
283     int i;
284     for (i = 0; i < MIGRATION_CAPABILITY__MAX; i++) {
285         if (should_validate_capability(i) && s->enabled_capabilities[i]) {
286             result++;
287         }
288     }
289     return result;
290 }
291 
292 static int configuration_pre_save(void *opaque)
293 {
294     SaveState *state = opaque;
295     const char *current_name = MACHINE_GET_CLASS(current_machine)->name;
296     MigrationState *s = migrate_get_current();
297     int i, j;
298 
299     state->len = strlen(current_name);
300     state->name = current_name;
301     state->target_page_bits = qemu_target_page_bits();
302 
303     state->caps_count = get_validatable_capabilities_count();
304     state->capabilities = g_renew(MigrationCapability, state->capabilities,
305                                   state->caps_count);
306     for (i = j = 0; i < MIGRATION_CAPABILITY__MAX; i++) {
307         if (should_validate_capability(i) && s->enabled_capabilities[i]) {
308             state->capabilities[j++] = i;
309         }
310     }
311     state->uuid = qemu_uuid;
312 
313     return 0;
314 }
315 
316 static int configuration_pre_load(void *opaque)
317 {
318     SaveState *state = opaque;
319 
320     /* If there is no target-page-bits subsection it means the source
321      * predates the variable-target-page-bits support and is using the
322      * minimum possible value for this CPU.
323      */
324     state->target_page_bits = qemu_target_page_bits_min();
325     return 0;
326 }
327 
328 static bool configuration_validate_capabilities(SaveState *state)
329 {
330     bool ret = true;
331     MigrationState *s = migrate_get_current();
332     unsigned long *source_caps_bm;
333     int i;
334 
335     source_caps_bm = bitmap_new(MIGRATION_CAPABILITY__MAX);
336     for (i = 0; i < state->caps_count; i++) {
337         MigrationCapability capability = state->capabilities[i];
338         set_bit(capability, source_caps_bm);
339     }
340 
341     for (i = 0; i < MIGRATION_CAPABILITY__MAX; i++) {
342         bool source_state, target_state;
343         if (!should_validate_capability(i)) {
344             continue;
345         }
346         source_state = test_bit(i, source_caps_bm);
347         target_state = s->enabled_capabilities[i];
348         if (source_state != target_state) {
349             error_report("Capability %s is %s, but received capability is %s",
350                          MigrationCapability_str(i),
351                          target_state ? "on" : "off",
352                          source_state ? "on" : "off");
353             ret = false;
354             /* Don't break here to report all failed capabilities */
355         }
356     }
357 
358     g_free(source_caps_bm);
359     return ret;
360 }
361 
362 static int configuration_post_load(void *opaque, int version_id)
363 {
364     SaveState *state = opaque;
365     const char *current_name = MACHINE_GET_CLASS(current_machine)->name;
366 
367     if (strncmp(state->name, current_name, state->len) != 0) {
368         error_report("Machine type received is '%.*s' and local is '%s'",
369                      (int) state->len, state->name, current_name);
370         return -EINVAL;
371     }
372 
373     if (state->target_page_bits != qemu_target_page_bits()) {
374         error_report("Received TARGET_PAGE_BITS is %d but local is %d",
375                      state->target_page_bits, qemu_target_page_bits());
376         return -EINVAL;
377     }
378 
379     if (!configuration_validate_capabilities(state)) {
380         return -EINVAL;
381     }
382 
383     return 0;
384 }
385 
386 static int get_capability(QEMUFile *f, void *pv, size_t size,
387                           const VMStateField *field)
388 {
389     MigrationCapability *capability = pv;
390     char capability_str[UINT8_MAX + 1];
391     uint8_t len;
392     int i;
393 
394     len = qemu_get_byte(f);
395     qemu_get_buffer(f, (uint8_t *)capability_str, len);
396     capability_str[len] = '\0';
397     for (i = 0; i < MIGRATION_CAPABILITY__MAX; i++) {
398         if (!strcmp(MigrationCapability_str(i), capability_str)) {
399             *capability = i;
400             return 0;
401         }
402     }
403     error_report("Received unknown capability %s", capability_str);
404     return -EINVAL;
405 }
406 
407 static int put_capability(QEMUFile *f, void *pv, size_t size,
408                           const VMStateField *field, QJSON *vmdesc)
409 {
410     MigrationCapability *capability = pv;
411     const char *capability_str = MigrationCapability_str(*capability);
412     size_t len = strlen(capability_str);
413     assert(len <= UINT8_MAX);
414 
415     qemu_put_byte(f, len);
416     qemu_put_buffer(f, (uint8_t *)capability_str, len);
417     return 0;
418 }
419 
420 static const VMStateInfo vmstate_info_capability = {
421     .name = "capability",
422     .get  = get_capability,
423     .put  = put_capability,
424 };
425 
426 /* The target-page-bits subsection is present only if the
427  * target page size is not the same as the default (ie the
428  * minimum page size for a variable-page-size guest CPU).
429  * If it is present then it contains the actual target page
430  * bits for the machine, and migration will fail if the
431  * two ends don't agree about it.
432  */
433 static bool vmstate_target_page_bits_needed(void *opaque)
434 {
435     return qemu_target_page_bits()
436         > qemu_target_page_bits_min();
437 }
438 
439 static const VMStateDescription vmstate_target_page_bits = {
440     .name = "configuration/target-page-bits",
441     .version_id = 1,
442     .minimum_version_id = 1,
443     .needed = vmstate_target_page_bits_needed,
444     .fields = (VMStateField[]) {
445         VMSTATE_UINT32(target_page_bits, SaveState),
446         VMSTATE_END_OF_LIST()
447     }
448 };
449 
450 static bool vmstate_capabilites_needed(void *opaque)
451 {
452     return get_validatable_capabilities_count() > 0;
453 }
454 
455 static const VMStateDescription vmstate_capabilites = {
456     .name = "configuration/capabilities",
457     .version_id = 1,
458     .minimum_version_id = 1,
459     .needed = vmstate_capabilites_needed,
460     .fields = (VMStateField[]) {
461         VMSTATE_UINT32_V(caps_count, SaveState, 1),
462         VMSTATE_VARRAY_UINT32_ALLOC(capabilities, SaveState, caps_count, 1,
463                                     vmstate_info_capability,
464                                     MigrationCapability),
465         VMSTATE_END_OF_LIST()
466     }
467 };
468 
469 static bool vmstate_uuid_needed(void *opaque)
470 {
471     return qemu_uuid_set && migrate_validate_uuid();
472 }
473 
474 static int vmstate_uuid_post_load(void *opaque, int version_id)
475 {
476     SaveState *state = opaque;
477     char uuid_src[UUID_FMT_LEN + 1];
478     char uuid_dst[UUID_FMT_LEN + 1];
479 
480     if (!qemu_uuid_set) {
481         /*
482          * It's warning because user might not know UUID in some cases,
483          * e.g. load an old snapshot
484          */
485         qemu_uuid_unparse(&state->uuid, uuid_src);
486         warn_report("UUID is received %s, but local uuid isn't set",
487                      uuid_src);
488         return 0;
489     }
490     if (!qemu_uuid_is_equal(&state->uuid, &qemu_uuid)) {
491         qemu_uuid_unparse(&state->uuid, uuid_src);
492         qemu_uuid_unparse(&qemu_uuid, uuid_dst);
493         error_report("UUID received is %s and local is %s", uuid_src, uuid_dst);
494         return -EINVAL;
495     }
496     return 0;
497 }
498 
499 static const VMStateDescription vmstate_uuid = {
500     .name = "configuration/uuid",
501     .version_id = 1,
502     .minimum_version_id = 1,
503     .needed = vmstate_uuid_needed,
504     .post_load = vmstate_uuid_post_load,
505     .fields = (VMStateField[]) {
506         VMSTATE_UINT8_ARRAY_V(uuid.data, SaveState, sizeof(QemuUUID), 1),
507         VMSTATE_END_OF_LIST()
508     }
509 };
510 
511 static const VMStateDescription vmstate_configuration = {
512     .name = "configuration",
513     .version_id = 1,
514     .pre_load = configuration_pre_load,
515     .post_load = configuration_post_load,
516     .pre_save = configuration_pre_save,
517     .fields = (VMStateField[]) {
518         VMSTATE_UINT32(len, SaveState),
519         VMSTATE_VBUFFER_ALLOC_UINT32(name, SaveState, 0, NULL, len),
520         VMSTATE_END_OF_LIST()
521     },
522     .subsections = (const VMStateDescription*[]) {
523         &vmstate_target_page_bits,
524         &vmstate_capabilites,
525         &vmstate_uuid,
526         NULL
527     }
528 };
529 
530 static void dump_vmstate_vmsd(FILE *out_file,
531                               const VMStateDescription *vmsd, int indent,
532                               bool is_subsection);
533 
534 static void dump_vmstate_vmsf(FILE *out_file, const VMStateField *field,
535                               int indent)
536 {
537     fprintf(out_file, "%*s{\n", indent, "");
538     indent += 2;
539     fprintf(out_file, "%*s\"field\": \"%s\",\n", indent, "", field->name);
540     fprintf(out_file, "%*s\"version_id\": %d,\n", indent, "",
541             field->version_id);
542     fprintf(out_file, "%*s\"field_exists\": %s,\n", indent, "",
543             field->field_exists ? "true" : "false");
544     fprintf(out_file, "%*s\"size\": %zu", indent, "", field->size);
545     if (field->vmsd != NULL) {
546         fprintf(out_file, ",\n");
547         dump_vmstate_vmsd(out_file, field->vmsd, indent, false);
548     }
549     fprintf(out_file, "\n%*s}", indent - 2, "");
550 }
551 
552 static void dump_vmstate_vmss(FILE *out_file,
553                               const VMStateDescription **subsection,
554                               int indent)
555 {
556     if (*subsection != NULL) {
557         dump_vmstate_vmsd(out_file, *subsection, indent, true);
558     }
559 }
560 
561 static void dump_vmstate_vmsd(FILE *out_file,
562                               const VMStateDescription *vmsd, int indent,
563                               bool is_subsection)
564 {
565     if (is_subsection) {
566         fprintf(out_file, "%*s{\n", indent, "");
567     } else {
568         fprintf(out_file, "%*s\"%s\": {\n", indent, "", "Description");
569     }
570     indent += 2;
571     fprintf(out_file, "%*s\"name\": \"%s\",\n", indent, "", vmsd->name);
572     fprintf(out_file, "%*s\"version_id\": %d,\n", indent, "",
573             vmsd->version_id);
574     fprintf(out_file, "%*s\"minimum_version_id\": %d", indent, "",
575             vmsd->minimum_version_id);
576     if (vmsd->fields != NULL) {
577         const VMStateField *field = vmsd->fields;
578         bool first;
579 
580         fprintf(out_file, ",\n%*s\"Fields\": [\n", indent, "");
581         first = true;
582         while (field->name != NULL) {
583             if (field->flags & VMS_MUST_EXIST) {
584                 /* Ignore VMSTATE_VALIDATE bits; these don't get migrated */
585                 field++;
586                 continue;
587             }
588             if (!first) {
589                 fprintf(out_file, ",\n");
590             }
591             dump_vmstate_vmsf(out_file, field, indent + 2);
592             field++;
593             first = false;
594         }
595         fprintf(out_file, "\n%*s]", indent, "");
596     }
597     if (vmsd->subsections != NULL) {
598         const VMStateDescription **subsection = vmsd->subsections;
599         bool first;
600 
601         fprintf(out_file, ",\n%*s\"Subsections\": [\n", indent, "");
602         first = true;
603         while (*subsection != NULL) {
604             if (!first) {
605                 fprintf(out_file, ",\n");
606             }
607             dump_vmstate_vmss(out_file, subsection, indent + 2);
608             subsection++;
609             first = false;
610         }
611         fprintf(out_file, "\n%*s]", indent, "");
612     }
613     fprintf(out_file, "\n%*s}", indent - 2, "");
614 }
615 
616 static void dump_machine_type(FILE *out_file)
617 {
618     MachineClass *mc;
619 
620     mc = MACHINE_GET_CLASS(current_machine);
621 
622     fprintf(out_file, "  \"vmschkmachine\": {\n");
623     fprintf(out_file, "    \"Name\": \"%s\"\n", mc->name);
624     fprintf(out_file, "  },\n");
625 }
626 
627 void dump_vmstate_json_to_file(FILE *out_file)
628 {
629     GSList *list, *elt;
630     bool first;
631 
632     fprintf(out_file, "{\n");
633     dump_machine_type(out_file);
634 
635     first = true;
636     list = object_class_get_list(TYPE_DEVICE, true);
637     for (elt = list; elt; elt = elt->next) {
638         DeviceClass *dc = OBJECT_CLASS_CHECK(DeviceClass, elt->data,
639                                              TYPE_DEVICE);
640         const char *name;
641         int indent = 2;
642 
643         if (!dc->vmsd) {
644             continue;
645         }
646 
647         if (!first) {
648             fprintf(out_file, ",\n");
649         }
650         name = object_class_get_name(OBJECT_CLASS(dc));
651         fprintf(out_file, "%*s\"%s\": {\n", indent, "", name);
652         indent += 2;
653         fprintf(out_file, "%*s\"Name\": \"%s\",\n", indent, "", name);
654         fprintf(out_file, "%*s\"version_id\": %d,\n", indent, "",
655                 dc->vmsd->version_id);
656         fprintf(out_file, "%*s\"minimum_version_id\": %d,\n", indent, "",
657                 dc->vmsd->minimum_version_id);
658 
659         dump_vmstate_vmsd(out_file, dc->vmsd, indent, false);
660 
661         fprintf(out_file, "\n%*s}", indent - 2, "");
662         first = false;
663     }
664     fprintf(out_file, "\n}\n");
665     fclose(out_file);
666 }
667 
668 static int calculate_new_instance_id(const char *idstr)
669 {
670     SaveStateEntry *se;
671     int instance_id = 0;
672 
673     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
674         if (strcmp(idstr, se->idstr) == 0
675             && instance_id <= se->instance_id) {
676             instance_id = se->instance_id + 1;
677         }
678     }
679     return instance_id;
680 }
681 
682 static int calculate_compat_instance_id(const char *idstr)
683 {
684     SaveStateEntry *se;
685     int instance_id = 0;
686 
687     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
688         if (!se->compat) {
689             continue;
690         }
691 
692         if (strcmp(idstr, se->compat->idstr) == 0
693             && instance_id <= se->compat->instance_id) {
694             instance_id = se->compat->instance_id + 1;
695         }
696     }
697     return instance_id;
698 }
699 
700 static inline MigrationPriority save_state_priority(SaveStateEntry *se)
701 {
702     if (se->vmsd) {
703         return se->vmsd->priority;
704     }
705     return MIG_PRI_DEFAULT;
706 }
707 
708 static void savevm_state_handler_insert(SaveStateEntry *nse)
709 {
710     MigrationPriority priority = save_state_priority(nse);
711     SaveStateEntry *se;
712 
713     assert(priority <= MIG_PRI_MAX);
714 
715     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
716         if (save_state_priority(se) < priority) {
717             break;
718         }
719     }
720 
721     if (se) {
722         QTAILQ_INSERT_BEFORE(se, nse, entry);
723     } else {
724         QTAILQ_INSERT_TAIL(&savevm_state.handlers, nse, entry);
725     }
726 }
727 
728 /* TODO: Individual devices generally have very little idea about the rest
729    of the system, so instance_id should be removed/replaced.
730    Meanwhile pass -1 as instance_id if you do not already have a clearly
731    distinguishing id for all instances of your device class. */
732 int register_savevm_live(const char *idstr,
733                          int instance_id,
734                          int version_id,
735                          const SaveVMHandlers *ops,
736                          void *opaque)
737 {
738     SaveStateEntry *se;
739 
740     se = g_new0(SaveStateEntry, 1);
741     se->version_id = version_id;
742     se->section_id = savevm_state.global_section_id++;
743     se->ops = ops;
744     se->opaque = opaque;
745     se->vmsd = NULL;
746     /* if this is a live_savem then set is_ram */
747     if (ops->save_setup != NULL) {
748         se->is_ram = 1;
749     }
750 
751     pstrcat(se->idstr, sizeof(se->idstr), idstr);
752 
753     if (instance_id == -1) {
754         se->instance_id = calculate_new_instance_id(se->idstr);
755     } else {
756         se->instance_id = instance_id;
757     }
758     assert(!se->compat || se->instance_id == 0);
759     savevm_state_handler_insert(se);
760     return 0;
761 }
762 
763 void unregister_savevm(DeviceState *dev, const char *idstr, void *opaque)
764 {
765     SaveStateEntry *se, *new_se;
766     char id[256] = "";
767 
768     if (dev) {
769         char *path = qdev_get_dev_path(dev);
770         if (path) {
771             pstrcpy(id, sizeof(id), path);
772             pstrcat(id, sizeof(id), "/");
773             g_free(path);
774         }
775     }
776     pstrcat(id, sizeof(id), idstr);
777 
778     QTAILQ_FOREACH_SAFE(se, &savevm_state.handlers, entry, new_se) {
779         if (strcmp(se->idstr, id) == 0 && se->opaque == opaque) {
780             QTAILQ_REMOVE(&savevm_state.handlers, se, entry);
781             g_free(se->compat);
782             g_free(se);
783         }
784     }
785 }
786 
787 int vmstate_register_with_alias_id(DeviceState *dev, int instance_id,
788                                    const VMStateDescription *vmsd,
789                                    void *opaque, int alias_id,
790                                    int required_for_version,
791                                    Error **errp)
792 {
793     SaveStateEntry *se;
794 
795     /* If this triggers, alias support can be dropped for the vmsd. */
796     assert(alias_id == -1 || required_for_version >= vmsd->minimum_version_id);
797 
798     se = g_new0(SaveStateEntry, 1);
799     se->version_id = vmsd->version_id;
800     se->section_id = savevm_state.global_section_id++;
801     se->opaque = opaque;
802     se->vmsd = vmsd;
803     se->alias_id = alias_id;
804 
805     if (dev) {
806         char *id = qdev_get_dev_path(dev);
807         if (id) {
808             if (snprintf(se->idstr, sizeof(se->idstr), "%s/", id) >=
809                 sizeof(se->idstr)) {
810                 error_setg(errp, "Path too long for VMState (%s)", id);
811                 g_free(id);
812                 g_free(se);
813 
814                 return -1;
815             }
816             g_free(id);
817 
818             se->compat = g_new0(CompatEntry, 1);
819             pstrcpy(se->compat->idstr, sizeof(se->compat->idstr), vmsd->name);
820             se->compat->instance_id = instance_id == -1 ?
821                          calculate_compat_instance_id(vmsd->name) : instance_id;
822             instance_id = -1;
823         }
824     }
825     pstrcat(se->idstr, sizeof(se->idstr), vmsd->name);
826 
827     if (instance_id == -1) {
828         se->instance_id = calculate_new_instance_id(se->idstr);
829     } else {
830         se->instance_id = instance_id;
831     }
832     assert(!se->compat || se->instance_id == 0);
833     savevm_state_handler_insert(se);
834     return 0;
835 }
836 
837 void vmstate_unregister(DeviceState *dev, const VMStateDescription *vmsd,
838                         void *opaque)
839 {
840     SaveStateEntry *se, *new_se;
841 
842     QTAILQ_FOREACH_SAFE(se, &savevm_state.handlers, entry, new_se) {
843         if (se->vmsd == vmsd && se->opaque == opaque) {
844             QTAILQ_REMOVE(&savevm_state.handlers, se, entry);
845             g_free(se->compat);
846             g_free(se);
847         }
848     }
849 }
850 
851 static int vmstate_load(QEMUFile *f, SaveStateEntry *se)
852 {
853     trace_vmstate_load(se->idstr, se->vmsd ? se->vmsd->name : "(old)");
854     if (!se->vmsd) {         /* Old style */
855         return se->ops->load_state(f, se->opaque, se->load_version_id);
856     }
857     return vmstate_load_state(f, se->vmsd, se->opaque, se->load_version_id);
858 }
859 
860 static void vmstate_save_old_style(QEMUFile *f, SaveStateEntry *se, QJSON *vmdesc)
861 {
862     int64_t old_offset, size;
863 
864     old_offset = qemu_ftell_fast(f);
865     se->ops->save_state(f, se->opaque);
866     size = qemu_ftell_fast(f) - old_offset;
867 
868     if (vmdesc) {
869         json_prop_int(vmdesc, "size", size);
870         json_start_array(vmdesc, "fields");
871         json_start_object(vmdesc, NULL);
872         json_prop_str(vmdesc, "name", "data");
873         json_prop_int(vmdesc, "size", size);
874         json_prop_str(vmdesc, "type", "buffer");
875         json_end_object(vmdesc);
876         json_end_array(vmdesc);
877     }
878 }
879 
880 static int vmstate_save(QEMUFile *f, SaveStateEntry *se, QJSON *vmdesc)
881 {
882     trace_vmstate_save(se->idstr, se->vmsd ? se->vmsd->name : "(old)");
883     if (!se->vmsd) {
884         vmstate_save_old_style(f, se, vmdesc);
885         return 0;
886     }
887     return vmstate_save_state(f, se->vmsd, se->opaque, vmdesc);
888 }
889 
890 /*
891  * Write the header for device section (QEMU_VM_SECTION START/END/PART/FULL)
892  */
893 static void save_section_header(QEMUFile *f, SaveStateEntry *se,
894                                 uint8_t section_type)
895 {
896     qemu_put_byte(f, section_type);
897     qemu_put_be32(f, se->section_id);
898 
899     if (section_type == QEMU_VM_SECTION_FULL ||
900         section_type == QEMU_VM_SECTION_START) {
901         /* ID string */
902         size_t len = strlen(se->idstr);
903         qemu_put_byte(f, len);
904         qemu_put_buffer(f, (uint8_t *)se->idstr, len);
905 
906         qemu_put_be32(f, se->instance_id);
907         qemu_put_be32(f, se->version_id);
908     }
909 }
910 
911 /*
912  * Write a footer onto device sections that catches cases misformatted device
913  * sections.
914  */
915 static void save_section_footer(QEMUFile *f, SaveStateEntry *se)
916 {
917     if (migrate_get_current()->send_section_footer) {
918         qemu_put_byte(f, QEMU_VM_SECTION_FOOTER);
919         qemu_put_be32(f, se->section_id);
920     }
921 }
922 
923 /**
924  * qemu_savevm_command_send: Send a 'QEMU_VM_COMMAND' type element with the
925  *                           command and associated data.
926  *
927  * @f: File to send command on
928  * @command: Command type to send
929  * @len: Length of associated data
930  * @data: Data associated with command.
931  */
932 static void qemu_savevm_command_send(QEMUFile *f,
933                                      enum qemu_vm_cmd command,
934                                      uint16_t len,
935                                      uint8_t *data)
936 {
937     trace_savevm_command_send(command, len);
938     qemu_put_byte(f, QEMU_VM_COMMAND);
939     qemu_put_be16(f, (uint16_t)command);
940     qemu_put_be16(f, len);
941     qemu_put_buffer(f, data, len);
942     qemu_fflush(f);
943 }
944 
945 void qemu_savevm_send_colo_enable(QEMUFile *f)
946 {
947     trace_savevm_send_colo_enable();
948     qemu_savevm_command_send(f, MIG_CMD_ENABLE_COLO, 0, NULL);
949 }
950 
951 void qemu_savevm_send_ping(QEMUFile *f, uint32_t value)
952 {
953     uint32_t buf;
954 
955     trace_savevm_send_ping(value);
956     buf = cpu_to_be32(value);
957     qemu_savevm_command_send(f, MIG_CMD_PING, sizeof(value), (uint8_t *)&buf);
958 }
959 
960 void qemu_savevm_send_open_return_path(QEMUFile *f)
961 {
962     trace_savevm_send_open_return_path();
963     qemu_savevm_command_send(f, MIG_CMD_OPEN_RETURN_PATH, 0, NULL);
964 }
965 
966 /* We have a buffer of data to send; we don't want that all to be loaded
967  * by the command itself, so the command contains just the length of the
968  * extra buffer that we then send straight after it.
969  * TODO: Must be a better way to organise that
970  *
971  * Returns:
972  *    0 on success
973  *    -ve on error
974  */
975 int qemu_savevm_send_packaged(QEMUFile *f, const uint8_t *buf, size_t len)
976 {
977     uint32_t tmp;
978 
979     if (len > MAX_VM_CMD_PACKAGED_SIZE) {
980         error_report("%s: Unreasonably large packaged state: %zu",
981                      __func__, len);
982         return -1;
983     }
984 
985     tmp = cpu_to_be32(len);
986 
987     trace_qemu_savevm_send_packaged();
988     qemu_savevm_command_send(f, MIG_CMD_PACKAGED, 4, (uint8_t *)&tmp);
989 
990     qemu_put_buffer(f, buf, len);
991 
992     return 0;
993 }
994 
995 /* Send prior to any postcopy transfer */
996 void qemu_savevm_send_postcopy_advise(QEMUFile *f)
997 {
998     if (migrate_postcopy_ram()) {
999         uint64_t tmp[2];
1000         tmp[0] = cpu_to_be64(ram_pagesize_summary());
1001         tmp[1] = cpu_to_be64(qemu_target_page_size());
1002 
1003         trace_qemu_savevm_send_postcopy_advise();
1004         qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_ADVISE,
1005                                  16, (uint8_t *)tmp);
1006     } else {
1007         qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_ADVISE, 0, NULL);
1008     }
1009 }
1010 
1011 /* Sent prior to starting the destination running in postcopy, discard pages
1012  * that have already been sent but redirtied on the source.
1013  * CMD_POSTCOPY_RAM_DISCARD consist of:
1014  *      byte   version (0)
1015  *      byte   Length of name field (not including 0)
1016  *  n x byte   RAM block name
1017  *      byte   0 terminator (just for safety)
1018  *  n x        Byte ranges within the named RAMBlock
1019  *      be64   Start of the range
1020  *      be64   Length
1021  *
1022  *  name:  RAMBlock name that these entries are part of
1023  *  len: Number of page entries
1024  *  start_list: 'len' addresses
1025  *  length_list: 'len' addresses
1026  *
1027  */
1028 void qemu_savevm_send_postcopy_ram_discard(QEMUFile *f, const char *name,
1029                                            uint16_t len,
1030                                            uint64_t *start_list,
1031                                            uint64_t *length_list)
1032 {
1033     uint8_t *buf;
1034     uint16_t tmplen;
1035     uint16_t t;
1036     size_t name_len = strlen(name);
1037 
1038     trace_qemu_savevm_send_postcopy_ram_discard(name, len);
1039     assert(name_len < 256);
1040     buf = g_malloc0(1 + 1 + name_len + 1 + (8 + 8) * len);
1041     buf[0] = postcopy_ram_discard_version;
1042     buf[1] = name_len;
1043     memcpy(buf + 2, name, name_len);
1044     tmplen = 2 + name_len;
1045     buf[tmplen++] = '\0';
1046 
1047     for (t = 0; t < len; t++) {
1048         stq_be_p(buf + tmplen, start_list[t]);
1049         tmplen += 8;
1050         stq_be_p(buf + tmplen, length_list[t]);
1051         tmplen += 8;
1052     }
1053     qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_RAM_DISCARD, tmplen, buf);
1054     g_free(buf);
1055 }
1056 
1057 /* Get the destination into a state where it can receive postcopy data. */
1058 void qemu_savevm_send_postcopy_listen(QEMUFile *f)
1059 {
1060     trace_savevm_send_postcopy_listen();
1061     qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_LISTEN, 0, NULL);
1062 }
1063 
1064 /* Kick the destination into running */
1065 void qemu_savevm_send_postcopy_run(QEMUFile *f)
1066 {
1067     trace_savevm_send_postcopy_run();
1068     qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_RUN, 0, NULL);
1069 }
1070 
1071 void qemu_savevm_send_postcopy_resume(QEMUFile *f)
1072 {
1073     trace_savevm_send_postcopy_resume();
1074     qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_RESUME, 0, NULL);
1075 }
1076 
1077 void qemu_savevm_send_recv_bitmap(QEMUFile *f, char *block_name)
1078 {
1079     size_t len;
1080     char buf[256];
1081 
1082     trace_savevm_send_recv_bitmap(block_name);
1083 
1084     buf[0] = len = strlen(block_name);
1085     memcpy(buf + 1, block_name, len);
1086 
1087     qemu_savevm_command_send(f, MIG_CMD_RECV_BITMAP, len + 1, (uint8_t *)buf);
1088 }
1089 
1090 bool qemu_savevm_state_blocked(Error **errp)
1091 {
1092     SaveStateEntry *se;
1093 
1094     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1095         if (se->vmsd && se->vmsd->unmigratable) {
1096             error_setg(errp, "State blocked by non-migratable device '%s'",
1097                        se->idstr);
1098             return true;
1099         }
1100     }
1101     return false;
1102 }
1103 
1104 void qemu_savevm_state_header(QEMUFile *f)
1105 {
1106     trace_savevm_state_header();
1107     qemu_put_be32(f, QEMU_VM_FILE_MAGIC);
1108     qemu_put_be32(f, QEMU_VM_FILE_VERSION);
1109 
1110     if (migrate_get_current()->send_configuration) {
1111         qemu_put_byte(f, QEMU_VM_CONFIGURATION);
1112         vmstate_save_state(f, &vmstate_configuration, &savevm_state, 0);
1113     }
1114 }
1115 
1116 void qemu_savevm_state_setup(QEMUFile *f)
1117 {
1118     SaveStateEntry *se;
1119     Error *local_err = NULL;
1120     int ret;
1121 
1122     trace_savevm_state_setup();
1123     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1124         if (!se->ops || !se->ops->save_setup) {
1125             continue;
1126         }
1127         if (se->ops->is_active) {
1128             if (!se->ops->is_active(se->opaque)) {
1129                 continue;
1130             }
1131         }
1132         save_section_header(f, se, QEMU_VM_SECTION_START);
1133 
1134         ret = se->ops->save_setup(f, se->opaque);
1135         save_section_footer(f, se);
1136         if (ret < 0) {
1137             qemu_file_set_error(f, ret);
1138             break;
1139         }
1140     }
1141 
1142     if (precopy_notify(PRECOPY_NOTIFY_SETUP, &local_err)) {
1143         error_report_err(local_err);
1144     }
1145 }
1146 
1147 int qemu_savevm_state_resume_prepare(MigrationState *s)
1148 {
1149     SaveStateEntry *se;
1150     int ret;
1151 
1152     trace_savevm_state_resume_prepare();
1153 
1154     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1155         if (!se->ops || !se->ops->resume_prepare) {
1156             continue;
1157         }
1158         if (se->ops->is_active) {
1159             if (!se->ops->is_active(se->opaque)) {
1160                 continue;
1161             }
1162         }
1163         ret = se->ops->resume_prepare(s, se->opaque);
1164         if (ret < 0) {
1165             return ret;
1166         }
1167     }
1168 
1169     return 0;
1170 }
1171 
1172 /*
1173  * this function has three return values:
1174  *   negative: there was one error, and we have -errno.
1175  *   0 : We haven't finished, caller have to go again
1176  *   1 : We have finished, we can go to complete phase
1177  */
1178 int qemu_savevm_state_iterate(QEMUFile *f, bool postcopy)
1179 {
1180     SaveStateEntry *se;
1181     int ret = 1;
1182 
1183     trace_savevm_state_iterate();
1184     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1185         if (!se->ops || !se->ops->save_live_iterate) {
1186             continue;
1187         }
1188         if (se->ops->is_active &&
1189             !se->ops->is_active(se->opaque)) {
1190             continue;
1191         }
1192         if (se->ops->is_active_iterate &&
1193             !se->ops->is_active_iterate(se->opaque)) {
1194             continue;
1195         }
1196         /*
1197          * In the postcopy phase, any device that doesn't know how to
1198          * do postcopy should have saved it's state in the _complete
1199          * call that's already run, it might get confused if we call
1200          * iterate afterwards.
1201          */
1202         if (postcopy &&
1203             !(se->ops->has_postcopy && se->ops->has_postcopy(se->opaque))) {
1204             continue;
1205         }
1206         if (qemu_file_rate_limit(f)) {
1207             return 0;
1208         }
1209         trace_savevm_section_start(se->idstr, se->section_id);
1210 
1211         save_section_header(f, se, QEMU_VM_SECTION_PART);
1212 
1213         ret = se->ops->save_live_iterate(f, se->opaque);
1214         trace_savevm_section_end(se->idstr, se->section_id, ret);
1215         save_section_footer(f, se);
1216 
1217         if (ret < 0) {
1218             qemu_file_set_error(f, ret);
1219         }
1220         if (ret <= 0) {
1221             /* Do not proceed to the next vmstate before this one reported
1222                completion of the current stage. This serializes the migration
1223                and reduces the probability that a faster changing state is
1224                synchronized over and over again. */
1225             break;
1226         }
1227     }
1228     return ret;
1229 }
1230 
1231 static bool should_send_vmdesc(void)
1232 {
1233     MachineState *machine = MACHINE(qdev_get_machine());
1234     bool in_postcopy = migration_in_postcopy();
1235     return !machine->suppress_vmdesc && !in_postcopy;
1236 }
1237 
1238 /*
1239  * Calls the save_live_complete_postcopy methods
1240  * causing the last few pages to be sent immediately and doing any associated
1241  * cleanup.
1242  * Note postcopy also calls qemu_savevm_state_complete_precopy to complete
1243  * all the other devices, but that happens at the point we switch to postcopy.
1244  */
1245 void qemu_savevm_state_complete_postcopy(QEMUFile *f)
1246 {
1247     SaveStateEntry *se;
1248     int ret;
1249 
1250     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1251         if (!se->ops || !se->ops->save_live_complete_postcopy) {
1252             continue;
1253         }
1254         if (se->ops->is_active) {
1255             if (!se->ops->is_active(se->opaque)) {
1256                 continue;
1257             }
1258         }
1259         trace_savevm_section_start(se->idstr, se->section_id);
1260         /* Section type */
1261         qemu_put_byte(f, QEMU_VM_SECTION_END);
1262         qemu_put_be32(f, se->section_id);
1263 
1264         ret = se->ops->save_live_complete_postcopy(f, se->opaque);
1265         trace_savevm_section_end(se->idstr, se->section_id, ret);
1266         save_section_footer(f, se);
1267         if (ret < 0) {
1268             qemu_file_set_error(f, ret);
1269             return;
1270         }
1271     }
1272 
1273     qemu_put_byte(f, QEMU_VM_EOF);
1274     qemu_fflush(f);
1275 }
1276 
1277 static
1278 int qemu_savevm_state_complete_precopy_iterable(QEMUFile *f, bool in_postcopy)
1279 {
1280     SaveStateEntry *se;
1281     int ret;
1282 
1283     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1284         if (!se->ops ||
1285             (in_postcopy && se->ops->has_postcopy &&
1286              se->ops->has_postcopy(se->opaque)) ||
1287             !se->ops->save_live_complete_precopy) {
1288             continue;
1289         }
1290 
1291         if (se->ops->is_active) {
1292             if (!se->ops->is_active(se->opaque)) {
1293                 continue;
1294             }
1295         }
1296         trace_savevm_section_start(se->idstr, se->section_id);
1297 
1298         save_section_header(f, se, QEMU_VM_SECTION_END);
1299 
1300         ret = se->ops->save_live_complete_precopy(f, se->opaque);
1301         trace_savevm_section_end(se->idstr, se->section_id, ret);
1302         save_section_footer(f, se);
1303         if (ret < 0) {
1304             qemu_file_set_error(f, ret);
1305             return -1;
1306         }
1307     }
1308 
1309     return 0;
1310 }
1311 
1312 static
1313 int qemu_savevm_state_complete_precopy_non_iterable(QEMUFile *f,
1314                                                     bool in_postcopy,
1315                                                     bool inactivate_disks)
1316 {
1317     g_autoptr(QJSON) vmdesc = NULL;
1318     int vmdesc_len;
1319     SaveStateEntry *se;
1320     int ret;
1321 
1322     vmdesc = qjson_new();
1323     json_prop_int(vmdesc, "page_size", qemu_target_page_size());
1324     json_start_array(vmdesc, "devices");
1325     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1326 
1327         if ((!se->ops || !se->ops->save_state) && !se->vmsd) {
1328             continue;
1329         }
1330         if (se->vmsd && !vmstate_save_needed(se->vmsd, se->opaque)) {
1331             trace_savevm_section_skip(se->idstr, se->section_id);
1332             continue;
1333         }
1334 
1335         trace_savevm_section_start(se->idstr, se->section_id);
1336 
1337         json_start_object(vmdesc, NULL);
1338         json_prop_str(vmdesc, "name", se->idstr);
1339         json_prop_int(vmdesc, "instance_id", se->instance_id);
1340 
1341         save_section_header(f, se, QEMU_VM_SECTION_FULL);
1342         ret = vmstate_save(f, se, vmdesc);
1343         if (ret) {
1344             qemu_file_set_error(f, ret);
1345             return ret;
1346         }
1347         trace_savevm_section_end(se->idstr, se->section_id, 0);
1348         save_section_footer(f, se);
1349 
1350         json_end_object(vmdesc);
1351     }
1352 
1353     if (inactivate_disks) {
1354         /* Inactivate before sending QEMU_VM_EOF so that the
1355          * bdrv_invalidate_cache_all() on the other end won't fail. */
1356         ret = bdrv_inactivate_all();
1357         if (ret) {
1358             error_report("%s: bdrv_inactivate_all() failed (%d)",
1359                          __func__, ret);
1360             qemu_file_set_error(f, ret);
1361             return ret;
1362         }
1363     }
1364     if (!in_postcopy) {
1365         /* Postcopy stream will still be going */
1366         qemu_put_byte(f, QEMU_VM_EOF);
1367     }
1368 
1369     json_end_array(vmdesc);
1370     qjson_finish(vmdesc);
1371     vmdesc_len = strlen(qjson_get_str(vmdesc));
1372 
1373     if (should_send_vmdesc()) {
1374         qemu_put_byte(f, QEMU_VM_VMDESCRIPTION);
1375         qemu_put_be32(f, vmdesc_len);
1376         qemu_put_buffer(f, (uint8_t *)qjson_get_str(vmdesc), vmdesc_len);
1377     }
1378 
1379     return 0;
1380 }
1381 
1382 int qemu_savevm_state_complete_precopy(QEMUFile *f, bool iterable_only,
1383                                        bool inactivate_disks)
1384 {
1385     int ret;
1386     Error *local_err = NULL;
1387     bool in_postcopy = migration_in_postcopy();
1388 
1389     if (precopy_notify(PRECOPY_NOTIFY_COMPLETE, &local_err)) {
1390         error_report_err(local_err);
1391     }
1392 
1393     trace_savevm_state_complete_precopy();
1394 
1395     cpu_synchronize_all_states();
1396 
1397     if (!in_postcopy || iterable_only) {
1398         ret = qemu_savevm_state_complete_precopy_iterable(f, in_postcopy);
1399         if (ret) {
1400             return ret;
1401         }
1402     }
1403 
1404     if (iterable_only) {
1405         goto flush;
1406     }
1407 
1408     ret = qemu_savevm_state_complete_precopy_non_iterable(f, in_postcopy,
1409                                                           inactivate_disks);
1410     if (ret) {
1411         return ret;
1412     }
1413 
1414 flush:
1415     qemu_fflush(f);
1416     return 0;
1417 }
1418 
1419 /* Give an estimate of the amount left to be transferred,
1420  * the result is split into the amount for units that can and
1421  * for units that can't do postcopy.
1422  */
1423 void qemu_savevm_state_pending(QEMUFile *f, uint64_t threshold_size,
1424                                uint64_t *res_precopy_only,
1425                                uint64_t *res_compatible,
1426                                uint64_t *res_postcopy_only)
1427 {
1428     SaveStateEntry *se;
1429 
1430     *res_precopy_only = 0;
1431     *res_compatible = 0;
1432     *res_postcopy_only = 0;
1433 
1434 
1435     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1436         if (!se->ops || !se->ops->save_live_pending) {
1437             continue;
1438         }
1439         if (se->ops->is_active) {
1440             if (!se->ops->is_active(se->opaque)) {
1441                 continue;
1442             }
1443         }
1444         se->ops->save_live_pending(f, se->opaque, threshold_size,
1445                                    res_precopy_only, res_compatible,
1446                                    res_postcopy_only);
1447     }
1448 }
1449 
1450 void qemu_savevm_state_cleanup(void)
1451 {
1452     SaveStateEntry *se;
1453     Error *local_err = NULL;
1454 
1455     if (precopy_notify(PRECOPY_NOTIFY_CLEANUP, &local_err)) {
1456         error_report_err(local_err);
1457     }
1458 
1459     trace_savevm_state_cleanup();
1460     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1461         if (se->ops && se->ops->save_cleanup) {
1462             se->ops->save_cleanup(se->opaque);
1463         }
1464     }
1465 }
1466 
1467 static int qemu_savevm_state(QEMUFile *f, Error **errp)
1468 {
1469     int ret;
1470     MigrationState *ms = migrate_get_current();
1471     MigrationStatus status;
1472 
1473     if (migration_is_setup_or_active(ms->state) ||
1474         ms->state == MIGRATION_STATUS_CANCELLING ||
1475         ms->state == MIGRATION_STATUS_COLO) {
1476         error_setg(errp, QERR_MIGRATION_ACTIVE);
1477         return -EINVAL;
1478     }
1479 
1480     if (migrate_use_block()) {
1481         error_setg(errp, "Block migration and snapshots are incompatible");
1482         return -EINVAL;
1483     }
1484 
1485     migrate_init(ms);
1486     memset(&ram_counters, 0, sizeof(ram_counters));
1487     ms->to_dst_file = f;
1488 
1489     qemu_mutex_unlock_iothread();
1490     qemu_savevm_state_header(f);
1491     qemu_savevm_state_setup(f);
1492     qemu_mutex_lock_iothread();
1493 
1494     while (qemu_file_get_error(f) == 0) {
1495         if (qemu_savevm_state_iterate(f, false) > 0) {
1496             break;
1497         }
1498     }
1499 
1500     ret = qemu_file_get_error(f);
1501     if (ret == 0) {
1502         qemu_savevm_state_complete_precopy(f, false, false);
1503         ret = qemu_file_get_error(f);
1504     }
1505     qemu_savevm_state_cleanup();
1506     if (ret != 0) {
1507         error_setg_errno(errp, -ret, "Error while writing VM state");
1508     }
1509 
1510     if (ret != 0) {
1511         status = MIGRATION_STATUS_FAILED;
1512     } else {
1513         status = MIGRATION_STATUS_COMPLETED;
1514     }
1515     migrate_set_state(&ms->state, MIGRATION_STATUS_SETUP, status);
1516 
1517     /* f is outer parameter, it should not stay in global migration state after
1518      * this function finished */
1519     ms->to_dst_file = NULL;
1520 
1521     return ret;
1522 }
1523 
1524 void qemu_savevm_live_state(QEMUFile *f)
1525 {
1526     /* save QEMU_VM_SECTION_END section */
1527     qemu_savevm_state_complete_precopy(f, true, false);
1528     qemu_put_byte(f, QEMU_VM_EOF);
1529 }
1530 
1531 int qemu_save_device_state(QEMUFile *f)
1532 {
1533     SaveStateEntry *se;
1534 
1535     if (!migration_in_colo_state()) {
1536         qemu_put_be32(f, QEMU_VM_FILE_MAGIC);
1537         qemu_put_be32(f, QEMU_VM_FILE_VERSION);
1538     }
1539     cpu_synchronize_all_states();
1540 
1541     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1542         int ret;
1543 
1544         if (se->is_ram) {
1545             continue;
1546         }
1547         if ((!se->ops || !se->ops->save_state) && !se->vmsd) {
1548             continue;
1549         }
1550         if (se->vmsd && !vmstate_save_needed(se->vmsd, se->opaque)) {
1551             continue;
1552         }
1553 
1554         save_section_header(f, se, QEMU_VM_SECTION_FULL);
1555 
1556         ret = vmstate_save(f, se, NULL);
1557         if (ret) {
1558             return ret;
1559         }
1560 
1561         save_section_footer(f, se);
1562     }
1563 
1564     qemu_put_byte(f, QEMU_VM_EOF);
1565 
1566     return qemu_file_get_error(f);
1567 }
1568 
1569 static SaveStateEntry *find_se(const char *idstr, int instance_id)
1570 {
1571     SaveStateEntry *se;
1572 
1573     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1574         if (!strcmp(se->idstr, idstr) &&
1575             (instance_id == se->instance_id ||
1576              instance_id == se->alias_id))
1577             return se;
1578         /* Migrating from an older version? */
1579         if (strstr(se->idstr, idstr) && se->compat) {
1580             if (!strcmp(se->compat->idstr, idstr) &&
1581                 (instance_id == se->compat->instance_id ||
1582                  instance_id == se->alias_id))
1583                 return se;
1584         }
1585     }
1586     return NULL;
1587 }
1588 
1589 enum LoadVMExitCodes {
1590     /* Allow a command to quit all layers of nested loadvm loops */
1591     LOADVM_QUIT     =  1,
1592 };
1593 
1594 /* ------ incoming postcopy messages ------ */
1595 /* 'advise' arrives before any transfers just to tell us that a postcopy
1596  * *might* happen - it might be skipped if precopy transferred everything
1597  * quickly.
1598  */
1599 static int loadvm_postcopy_handle_advise(MigrationIncomingState *mis,
1600                                          uint16_t len)
1601 {
1602     PostcopyState ps = postcopy_state_set(POSTCOPY_INCOMING_ADVISE);
1603     uint64_t remote_pagesize_summary, local_pagesize_summary, remote_tps;
1604     Error *local_err = NULL;
1605 
1606     trace_loadvm_postcopy_handle_advise();
1607     if (ps != POSTCOPY_INCOMING_NONE) {
1608         error_report("CMD_POSTCOPY_ADVISE in wrong postcopy state (%d)", ps);
1609         return -1;
1610     }
1611 
1612     switch (len) {
1613     case 0:
1614         if (migrate_postcopy_ram()) {
1615             error_report("RAM postcopy is enabled but have 0 byte advise");
1616             return -EINVAL;
1617         }
1618         return 0;
1619     case 8 + 8:
1620         if (!migrate_postcopy_ram()) {
1621             error_report("RAM postcopy is disabled but have 16 byte advise");
1622             return -EINVAL;
1623         }
1624         break;
1625     default:
1626         error_report("CMD_POSTCOPY_ADVISE invalid length (%d)", len);
1627         return -EINVAL;
1628     }
1629 
1630     if (!postcopy_ram_supported_by_host(mis)) {
1631         postcopy_state_set(POSTCOPY_INCOMING_NONE);
1632         return -1;
1633     }
1634 
1635     remote_pagesize_summary = qemu_get_be64(mis->from_src_file);
1636     local_pagesize_summary = ram_pagesize_summary();
1637 
1638     if (remote_pagesize_summary != local_pagesize_summary)  {
1639         /*
1640          * This detects two potential causes of mismatch:
1641          *   a) A mismatch in host page sizes
1642          *      Some combinations of mismatch are probably possible but it gets
1643          *      a bit more complicated.  In particular we need to place whole
1644          *      host pages on the dest at once, and we need to ensure that we
1645          *      handle dirtying to make sure we never end up sending part of
1646          *      a hostpage on it's own.
1647          *   b) The use of different huge page sizes on source/destination
1648          *      a more fine grain test is performed during RAM block migration
1649          *      but this test here causes a nice early clear failure, and
1650          *      also fails when passed to an older qemu that doesn't
1651          *      do huge pages.
1652          */
1653         error_report("Postcopy needs matching RAM page sizes (s=%" PRIx64
1654                                                              " d=%" PRIx64 ")",
1655                      remote_pagesize_summary, local_pagesize_summary);
1656         return -1;
1657     }
1658 
1659     remote_tps = qemu_get_be64(mis->from_src_file);
1660     if (remote_tps != qemu_target_page_size()) {
1661         /*
1662          * Again, some differences could be dealt with, but for now keep it
1663          * simple.
1664          */
1665         error_report("Postcopy needs matching target page sizes (s=%d d=%zd)",
1666                      (int)remote_tps, qemu_target_page_size());
1667         return -1;
1668     }
1669 
1670     if (postcopy_notify(POSTCOPY_NOTIFY_INBOUND_ADVISE, &local_err)) {
1671         error_report_err(local_err);
1672         return -1;
1673     }
1674 
1675     if (ram_postcopy_incoming_init(mis)) {
1676         return -1;
1677     }
1678 
1679     return 0;
1680 }
1681 
1682 /* After postcopy we will be told to throw some pages away since they're
1683  * dirty and will have to be demand fetched.  Must happen before CPU is
1684  * started.
1685  * There can be 0..many of these messages, each encoding multiple pages.
1686  */
1687 static int loadvm_postcopy_ram_handle_discard(MigrationIncomingState *mis,
1688                                               uint16_t len)
1689 {
1690     int tmp;
1691     char ramid[256];
1692     PostcopyState ps = postcopy_state_get();
1693 
1694     trace_loadvm_postcopy_ram_handle_discard();
1695 
1696     switch (ps) {
1697     case POSTCOPY_INCOMING_ADVISE:
1698         /* 1st discard */
1699         tmp = postcopy_ram_prepare_discard(mis);
1700         if (tmp) {
1701             return tmp;
1702         }
1703         break;
1704 
1705     case POSTCOPY_INCOMING_DISCARD:
1706         /* Expected state */
1707         break;
1708 
1709     default:
1710         error_report("CMD_POSTCOPY_RAM_DISCARD in wrong postcopy state (%d)",
1711                      ps);
1712         return -1;
1713     }
1714     /* We're expecting a
1715      *    Version (0)
1716      *    a RAM ID string (length byte, name, 0 term)
1717      *    then at least 1 16 byte chunk
1718     */
1719     if (len < (1 + 1 + 1 + 1 + 2 * 8)) {
1720         error_report("CMD_POSTCOPY_RAM_DISCARD invalid length (%d)", len);
1721         return -1;
1722     }
1723 
1724     tmp = qemu_get_byte(mis->from_src_file);
1725     if (tmp != postcopy_ram_discard_version) {
1726         error_report("CMD_POSTCOPY_RAM_DISCARD invalid version (%d)", tmp);
1727         return -1;
1728     }
1729 
1730     if (!qemu_get_counted_string(mis->from_src_file, ramid)) {
1731         error_report("CMD_POSTCOPY_RAM_DISCARD Failed to read RAMBlock ID");
1732         return -1;
1733     }
1734     tmp = qemu_get_byte(mis->from_src_file);
1735     if (tmp != 0) {
1736         error_report("CMD_POSTCOPY_RAM_DISCARD missing nil (%d)", tmp);
1737         return -1;
1738     }
1739 
1740     len -= 3 + strlen(ramid);
1741     if (len % 16) {
1742         error_report("CMD_POSTCOPY_RAM_DISCARD invalid length (%d)", len);
1743         return -1;
1744     }
1745     trace_loadvm_postcopy_ram_handle_discard_header(ramid, len);
1746     while (len) {
1747         uint64_t start_addr, block_length;
1748         start_addr = qemu_get_be64(mis->from_src_file);
1749         block_length = qemu_get_be64(mis->from_src_file);
1750 
1751         len -= 16;
1752         int ret = ram_discard_range(ramid, start_addr, block_length);
1753         if (ret) {
1754             return ret;
1755         }
1756     }
1757     trace_loadvm_postcopy_ram_handle_discard_end();
1758 
1759     return 0;
1760 }
1761 
1762 /*
1763  * Triggered by a postcopy_listen command; this thread takes over reading
1764  * the input stream, leaving the main thread free to carry on loading the rest
1765  * of the device state (from RAM).
1766  * (TODO:This could do with being in a postcopy file - but there again it's
1767  * just another input loop, not that postcopy specific)
1768  */
1769 static void *postcopy_ram_listen_thread(void *opaque)
1770 {
1771     MigrationIncomingState *mis = migration_incoming_get_current();
1772     QEMUFile *f = mis->from_src_file;
1773     int load_res;
1774 
1775     migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE,
1776                                    MIGRATION_STATUS_POSTCOPY_ACTIVE);
1777     qemu_sem_post(&mis->listen_thread_sem);
1778     trace_postcopy_ram_listen_thread_start();
1779 
1780     rcu_register_thread();
1781     /*
1782      * Because we're a thread and not a coroutine we can't yield
1783      * in qemu_file, and thus we must be blocking now.
1784      */
1785     qemu_file_set_blocking(f, true);
1786     load_res = qemu_loadvm_state_main(f, mis);
1787 
1788     /*
1789      * This is tricky, but, mis->from_src_file can change after it
1790      * returns, when postcopy recovery happened. In the future, we may
1791      * want a wrapper for the QEMUFile handle.
1792      */
1793     f = mis->from_src_file;
1794 
1795     /* And non-blocking again so we don't block in any cleanup */
1796     qemu_file_set_blocking(f, false);
1797 
1798     trace_postcopy_ram_listen_thread_exit();
1799     if (load_res < 0) {
1800         error_report("%s: loadvm failed: %d", __func__, load_res);
1801         qemu_file_set_error(f, load_res);
1802         migrate_set_state(&mis->state, MIGRATION_STATUS_POSTCOPY_ACTIVE,
1803                                        MIGRATION_STATUS_FAILED);
1804     } else {
1805         /*
1806          * This looks good, but it's possible that the device loading in the
1807          * main thread hasn't finished yet, and so we might not be in 'RUN'
1808          * state yet; wait for the end of the main thread.
1809          */
1810         qemu_event_wait(&mis->main_thread_load_event);
1811     }
1812     postcopy_ram_incoming_cleanup(mis);
1813 
1814     if (load_res < 0) {
1815         /*
1816          * If something went wrong then we have a bad state so exit;
1817          * depending how far we got it might be possible at this point
1818          * to leave the guest running and fire MCEs for pages that never
1819          * arrived as a desperate recovery step.
1820          */
1821         rcu_unregister_thread();
1822         exit(EXIT_FAILURE);
1823     }
1824 
1825     migrate_set_state(&mis->state, MIGRATION_STATUS_POSTCOPY_ACTIVE,
1826                                    MIGRATION_STATUS_COMPLETED);
1827     /*
1828      * If everything has worked fine, then the main thread has waited
1829      * for us to start, and we're the last use of the mis.
1830      * (If something broke then qemu will have to exit anyway since it's
1831      * got a bad migration state).
1832      */
1833     migration_incoming_state_destroy();
1834     qemu_loadvm_state_cleanup();
1835 
1836     rcu_unregister_thread();
1837     mis->have_listen_thread = false;
1838     return NULL;
1839 }
1840 
1841 /* After this message we must be able to immediately receive postcopy data */
1842 static int loadvm_postcopy_handle_listen(MigrationIncomingState *mis)
1843 {
1844     PostcopyState ps = postcopy_state_set(POSTCOPY_INCOMING_LISTENING);
1845     trace_loadvm_postcopy_handle_listen();
1846     Error *local_err = NULL;
1847 
1848     if (ps != POSTCOPY_INCOMING_ADVISE && ps != POSTCOPY_INCOMING_DISCARD) {
1849         error_report("CMD_POSTCOPY_LISTEN in wrong postcopy state (%d)", ps);
1850         return -1;
1851     }
1852     if (ps == POSTCOPY_INCOMING_ADVISE) {
1853         /*
1854          * A rare case, we entered listen without having to do any discards,
1855          * so do the setup that's normally done at the time of the 1st discard.
1856          */
1857         if (migrate_postcopy_ram()) {
1858             postcopy_ram_prepare_discard(mis);
1859         }
1860     }
1861 
1862     /*
1863      * Sensitise RAM - can now generate requests for blocks that don't exist
1864      * However, at this point the CPU shouldn't be running, and the IO
1865      * shouldn't be doing anything yet so don't actually expect requests
1866      */
1867     if (migrate_postcopy_ram()) {
1868         if (postcopy_ram_enable_notify(mis)) {
1869             postcopy_ram_incoming_cleanup(mis);
1870             return -1;
1871         }
1872     }
1873 
1874     if (postcopy_notify(POSTCOPY_NOTIFY_INBOUND_LISTEN, &local_err)) {
1875         error_report_err(local_err);
1876         return -1;
1877     }
1878 
1879     if (mis->have_listen_thread) {
1880         error_report("CMD_POSTCOPY_RAM_LISTEN already has a listen thread");
1881         return -1;
1882     }
1883 
1884     mis->have_listen_thread = true;
1885     /* Start up the listening thread and wait for it to signal ready */
1886     qemu_sem_init(&mis->listen_thread_sem, 0);
1887     qemu_thread_create(&mis->listen_thread, "postcopy/listen",
1888                        postcopy_ram_listen_thread, NULL,
1889                        QEMU_THREAD_DETACHED);
1890     qemu_sem_wait(&mis->listen_thread_sem);
1891     qemu_sem_destroy(&mis->listen_thread_sem);
1892 
1893     return 0;
1894 }
1895 
1896 static void loadvm_postcopy_handle_run_bh(void *opaque)
1897 {
1898     Error *local_err = NULL;
1899     MigrationIncomingState *mis = opaque;
1900 
1901     /* TODO we should move all of this lot into postcopy_ram.c or a shared code
1902      * in migration.c
1903      */
1904     cpu_synchronize_all_post_init();
1905 
1906     qemu_announce_self(&mis->announce_timer, migrate_announce_params());
1907 
1908     /* Make sure all file formats flush their mutable metadata.
1909      * If we get an error here, just don't restart the VM yet. */
1910     bdrv_invalidate_cache_all(&local_err);
1911     if (local_err) {
1912         error_report_err(local_err);
1913         local_err = NULL;
1914         autostart = false;
1915     }
1916 
1917     trace_loadvm_postcopy_handle_run_cpu_sync();
1918 
1919     trace_loadvm_postcopy_handle_run_vmstart();
1920 
1921     dirty_bitmap_mig_before_vm_start();
1922 
1923     if (autostart) {
1924         /* Hold onto your hats, starting the CPU */
1925         vm_start();
1926     } else {
1927         /* leave it paused and let management decide when to start the CPU */
1928         runstate_set(RUN_STATE_PAUSED);
1929     }
1930 
1931     qemu_bh_delete(mis->bh);
1932 }
1933 
1934 /* After all discards we can start running and asking for pages */
1935 static int loadvm_postcopy_handle_run(MigrationIncomingState *mis)
1936 {
1937     PostcopyState ps = postcopy_state_set(POSTCOPY_INCOMING_RUNNING);
1938 
1939     trace_loadvm_postcopy_handle_run();
1940     if (ps != POSTCOPY_INCOMING_LISTENING) {
1941         error_report("CMD_POSTCOPY_RUN in wrong postcopy state (%d)", ps);
1942         return -1;
1943     }
1944 
1945     mis->bh = qemu_bh_new(loadvm_postcopy_handle_run_bh, mis);
1946     qemu_bh_schedule(mis->bh);
1947 
1948     /* We need to finish reading the stream from the package
1949      * and also stop reading anything more from the stream that loaded the
1950      * package (since it's now being read by the listener thread).
1951      * LOADVM_QUIT will quit all the layers of nested loadvm loops.
1952      */
1953     return LOADVM_QUIT;
1954 }
1955 
1956 static int loadvm_postcopy_handle_resume(MigrationIncomingState *mis)
1957 {
1958     if (mis->state != MIGRATION_STATUS_POSTCOPY_RECOVER) {
1959         error_report("%s: illegal resume received", __func__);
1960         /* Don't fail the load, only for this. */
1961         return 0;
1962     }
1963 
1964     /*
1965      * This means source VM is ready to resume the postcopy migration.
1966      * It's time to switch state and release the fault thread to
1967      * continue service page faults.
1968      */
1969     migrate_set_state(&mis->state, MIGRATION_STATUS_POSTCOPY_RECOVER,
1970                       MIGRATION_STATUS_POSTCOPY_ACTIVE);
1971     qemu_sem_post(&mis->postcopy_pause_sem_fault);
1972 
1973     trace_loadvm_postcopy_handle_resume();
1974 
1975     /* Tell source that "we are ready" */
1976     migrate_send_rp_resume_ack(mis, MIGRATION_RESUME_ACK_VALUE);
1977 
1978     return 0;
1979 }
1980 
1981 /**
1982  * Immediately following this command is a blob of data containing an embedded
1983  * chunk of migration stream; read it and load it.
1984  *
1985  * @mis: Incoming state
1986  * @length: Length of packaged data to read
1987  *
1988  * Returns: Negative values on error
1989  *
1990  */
1991 static int loadvm_handle_cmd_packaged(MigrationIncomingState *mis)
1992 {
1993     int ret;
1994     size_t length;
1995     QIOChannelBuffer *bioc;
1996 
1997     length = qemu_get_be32(mis->from_src_file);
1998     trace_loadvm_handle_cmd_packaged(length);
1999 
2000     if (length > MAX_VM_CMD_PACKAGED_SIZE) {
2001         error_report("Unreasonably large packaged state: %zu", length);
2002         return -1;
2003     }
2004 
2005     bioc = qio_channel_buffer_new(length);
2006     qio_channel_set_name(QIO_CHANNEL(bioc), "migration-loadvm-buffer");
2007     ret = qemu_get_buffer(mis->from_src_file,
2008                           bioc->data,
2009                           length);
2010     if (ret != length) {
2011         object_unref(OBJECT(bioc));
2012         error_report("CMD_PACKAGED: Buffer receive fail ret=%d length=%zu",
2013                      ret, length);
2014         return (ret < 0) ? ret : -EAGAIN;
2015     }
2016     bioc->usage += length;
2017     trace_loadvm_handle_cmd_packaged_received(ret);
2018 
2019     QEMUFile *packf = qemu_fopen_channel_input(QIO_CHANNEL(bioc));
2020 
2021     ret = qemu_loadvm_state_main(packf, mis);
2022     trace_loadvm_handle_cmd_packaged_main(ret);
2023     qemu_fclose(packf);
2024     object_unref(OBJECT(bioc));
2025 
2026     return ret;
2027 }
2028 
2029 /*
2030  * Handle request that source requests for recved_bitmap on
2031  * destination. Payload format:
2032  *
2033  * len (1 byte) + ramblock_name (<255 bytes)
2034  */
2035 static int loadvm_handle_recv_bitmap(MigrationIncomingState *mis,
2036                                      uint16_t len)
2037 {
2038     QEMUFile *file = mis->from_src_file;
2039     RAMBlock *rb;
2040     char block_name[256];
2041     size_t cnt;
2042 
2043     cnt = qemu_get_counted_string(file, block_name);
2044     if (!cnt) {
2045         error_report("%s: failed to read block name", __func__);
2046         return -EINVAL;
2047     }
2048 
2049     /* Validate before using the data */
2050     if (qemu_file_get_error(file)) {
2051         return qemu_file_get_error(file);
2052     }
2053 
2054     if (len != cnt + 1) {
2055         error_report("%s: invalid payload length (%d)", __func__, len);
2056         return -EINVAL;
2057     }
2058 
2059     rb = qemu_ram_block_by_name(block_name);
2060     if (!rb) {
2061         error_report("%s: block '%s' not found", __func__, block_name);
2062         return -EINVAL;
2063     }
2064 
2065     migrate_send_rp_recv_bitmap(mis, block_name);
2066 
2067     trace_loadvm_handle_recv_bitmap(block_name);
2068 
2069     return 0;
2070 }
2071 
2072 static int loadvm_process_enable_colo(MigrationIncomingState *mis)
2073 {
2074     migration_incoming_enable_colo();
2075     return colo_init_ram_cache();
2076 }
2077 
2078 /*
2079  * Process an incoming 'QEMU_VM_COMMAND'
2080  * 0           just a normal return
2081  * LOADVM_QUIT All good, but exit the loop
2082  * <0          Error
2083  */
2084 static int loadvm_process_command(QEMUFile *f)
2085 {
2086     MigrationIncomingState *mis = migration_incoming_get_current();
2087     uint16_t cmd;
2088     uint16_t len;
2089     uint32_t tmp32;
2090 
2091     cmd = qemu_get_be16(f);
2092     len = qemu_get_be16(f);
2093 
2094     /* Check validity before continue processing of cmds */
2095     if (qemu_file_get_error(f)) {
2096         return qemu_file_get_error(f);
2097     }
2098 
2099     trace_loadvm_process_command(cmd, len);
2100     if (cmd >= MIG_CMD_MAX || cmd == MIG_CMD_INVALID) {
2101         error_report("MIG_CMD 0x%x unknown (len 0x%x)", cmd, len);
2102         return -EINVAL;
2103     }
2104 
2105     if (mig_cmd_args[cmd].len != -1 && mig_cmd_args[cmd].len != len) {
2106         error_report("%s received with bad length - expecting %zu, got %d",
2107                      mig_cmd_args[cmd].name,
2108                      (size_t)mig_cmd_args[cmd].len, len);
2109         return -ERANGE;
2110     }
2111 
2112     switch (cmd) {
2113     case MIG_CMD_OPEN_RETURN_PATH:
2114         if (mis->to_src_file) {
2115             error_report("CMD_OPEN_RETURN_PATH called when RP already open");
2116             /* Not really a problem, so don't give up */
2117             return 0;
2118         }
2119         mis->to_src_file = qemu_file_get_return_path(f);
2120         if (!mis->to_src_file) {
2121             error_report("CMD_OPEN_RETURN_PATH failed");
2122             return -1;
2123         }
2124         break;
2125 
2126     case MIG_CMD_PING:
2127         tmp32 = qemu_get_be32(f);
2128         trace_loadvm_process_command_ping(tmp32);
2129         if (!mis->to_src_file) {
2130             error_report("CMD_PING (0x%x) received with no return path",
2131                          tmp32);
2132             return -1;
2133         }
2134         migrate_send_rp_pong(mis, tmp32);
2135         break;
2136 
2137     case MIG_CMD_PACKAGED:
2138         return loadvm_handle_cmd_packaged(mis);
2139 
2140     case MIG_CMD_POSTCOPY_ADVISE:
2141         return loadvm_postcopy_handle_advise(mis, len);
2142 
2143     case MIG_CMD_POSTCOPY_LISTEN:
2144         return loadvm_postcopy_handle_listen(mis);
2145 
2146     case MIG_CMD_POSTCOPY_RUN:
2147         return loadvm_postcopy_handle_run(mis);
2148 
2149     case MIG_CMD_POSTCOPY_RAM_DISCARD:
2150         return loadvm_postcopy_ram_handle_discard(mis, len);
2151 
2152     case MIG_CMD_POSTCOPY_RESUME:
2153         return loadvm_postcopy_handle_resume(mis);
2154 
2155     case MIG_CMD_RECV_BITMAP:
2156         return loadvm_handle_recv_bitmap(mis, len);
2157 
2158     case MIG_CMD_ENABLE_COLO:
2159         return loadvm_process_enable_colo(mis);
2160     }
2161 
2162     return 0;
2163 }
2164 
2165 /*
2166  * Read a footer off the wire and check that it matches the expected section
2167  *
2168  * Returns: true if the footer was good
2169  *          false if there is a problem (and calls error_report to say why)
2170  */
2171 static bool check_section_footer(QEMUFile *f, SaveStateEntry *se)
2172 {
2173     int ret;
2174     uint8_t read_mark;
2175     uint32_t read_section_id;
2176 
2177     if (!migrate_get_current()->send_section_footer) {
2178         /* No footer to check */
2179         return true;
2180     }
2181 
2182     read_mark = qemu_get_byte(f);
2183 
2184     ret = qemu_file_get_error(f);
2185     if (ret) {
2186         error_report("%s: Read section footer failed: %d",
2187                      __func__, ret);
2188         return false;
2189     }
2190 
2191     if (read_mark != QEMU_VM_SECTION_FOOTER) {
2192         error_report("Missing section footer for %s", se->idstr);
2193         return false;
2194     }
2195 
2196     read_section_id = qemu_get_be32(f);
2197     if (read_section_id != se->load_section_id) {
2198         error_report("Mismatched section id in footer for %s -"
2199                      " read 0x%x expected 0x%x",
2200                      se->idstr, read_section_id, se->load_section_id);
2201         return false;
2202     }
2203 
2204     /* All good */
2205     return true;
2206 }
2207 
2208 static int
2209 qemu_loadvm_section_start_full(QEMUFile *f, MigrationIncomingState *mis)
2210 {
2211     uint32_t instance_id, version_id, section_id;
2212     SaveStateEntry *se;
2213     char idstr[256];
2214     int ret;
2215 
2216     /* Read section start */
2217     section_id = qemu_get_be32(f);
2218     if (!qemu_get_counted_string(f, idstr)) {
2219         error_report("Unable to read ID string for section %u",
2220                      section_id);
2221         return -EINVAL;
2222     }
2223     instance_id = qemu_get_be32(f);
2224     version_id = qemu_get_be32(f);
2225 
2226     ret = qemu_file_get_error(f);
2227     if (ret) {
2228         error_report("%s: Failed to read instance/version ID: %d",
2229                      __func__, ret);
2230         return ret;
2231     }
2232 
2233     trace_qemu_loadvm_state_section_startfull(section_id, idstr,
2234             instance_id, version_id);
2235     /* Find savevm section */
2236     se = find_se(idstr, instance_id);
2237     if (se == NULL) {
2238         error_report("Unknown savevm section or instance '%s' %d. "
2239                      "Make sure that your current VM setup matches your "
2240                      "saved VM setup, including any hotplugged devices",
2241                      idstr, instance_id);
2242         return -EINVAL;
2243     }
2244 
2245     /* Validate version */
2246     if (version_id > se->version_id) {
2247         error_report("savevm: unsupported version %d for '%s' v%d",
2248                      version_id, idstr, se->version_id);
2249         return -EINVAL;
2250     }
2251     se->load_version_id = version_id;
2252     se->load_section_id = section_id;
2253 
2254     /* Validate if it is a device's state */
2255     if (xen_enabled() && se->is_ram) {
2256         error_report("loadvm: %s RAM loading not allowed on Xen", idstr);
2257         return -EINVAL;
2258     }
2259 
2260     ret = vmstate_load(f, se);
2261     if (ret < 0) {
2262         error_report("error while loading state for instance 0x%x of"
2263                      " device '%s'", instance_id, idstr);
2264         return ret;
2265     }
2266     if (!check_section_footer(f, se)) {
2267         return -EINVAL;
2268     }
2269 
2270     return 0;
2271 }
2272 
2273 static int
2274 qemu_loadvm_section_part_end(QEMUFile *f, MigrationIncomingState *mis)
2275 {
2276     uint32_t section_id;
2277     SaveStateEntry *se;
2278     int ret;
2279 
2280     section_id = qemu_get_be32(f);
2281 
2282     ret = qemu_file_get_error(f);
2283     if (ret) {
2284         error_report("%s: Failed to read section ID: %d",
2285                      __func__, ret);
2286         return ret;
2287     }
2288 
2289     trace_qemu_loadvm_state_section_partend(section_id);
2290     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
2291         if (se->load_section_id == section_id) {
2292             break;
2293         }
2294     }
2295     if (se == NULL) {
2296         error_report("Unknown savevm section %d", section_id);
2297         return -EINVAL;
2298     }
2299 
2300     ret = vmstate_load(f, se);
2301     if (ret < 0) {
2302         error_report("error while loading state section id %d(%s)",
2303                      section_id, se->idstr);
2304         return ret;
2305     }
2306     if (!check_section_footer(f, se)) {
2307         return -EINVAL;
2308     }
2309 
2310     return 0;
2311 }
2312 
2313 static int qemu_loadvm_state_header(QEMUFile *f)
2314 {
2315     unsigned int v;
2316     int ret;
2317 
2318     v = qemu_get_be32(f);
2319     if (v != QEMU_VM_FILE_MAGIC) {
2320         error_report("Not a migration stream");
2321         return -EINVAL;
2322     }
2323 
2324     v = qemu_get_be32(f);
2325     if (v == QEMU_VM_FILE_VERSION_COMPAT) {
2326         error_report("SaveVM v2 format is obsolete and don't work anymore");
2327         return -ENOTSUP;
2328     }
2329     if (v != QEMU_VM_FILE_VERSION) {
2330         error_report("Unsupported migration stream version");
2331         return -ENOTSUP;
2332     }
2333 
2334     if (migrate_get_current()->send_configuration) {
2335         if (qemu_get_byte(f) != QEMU_VM_CONFIGURATION) {
2336             error_report("Configuration section missing");
2337             qemu_loadvm_state_cleanup();
2338             return -EINVAL;
2339         }
2340         ret = vmstate_load_state(f, &vmstate_configuration, &savevm_state, 0);
2341 
2342         if (ret) {
2343             qemu_loadvm_state_cleanup();
2344             return ret;
2345         }
2346     }
2347     return 0;
2348 }
2349 
2350 static int qemu_loadvm_state_setup(QEMUFile *f)
2351 {
2352     SaveStateEntry *se;
2353     int ret;
2354 
2355     trace_loadvm_state_setup();
2356     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
2357         if (!se->ops || !se->ops->load_setup) {
2358             continue;
2359         }
2360         if (se->ops->is_active) {
2361             if (!se->ops->is_active(se->opaque)) {
2362                 continue;
2363             }
2364         }
2365 
2366         ret = se->ops->load_setup(f, se->opaque);
2367         if (ret < 0) {
2368             qemu_file_set_error(f, ret);
2369             error_report("Load state of device %s failed", se->idstr);
2370             return ret;
2371         }
2372     }
2373     return 0;
2374 }
2375 
2376 void qemu_loadvm_state_cleanup(void)
2377 {
2378     SaveStateEntry *se;
2379 
2380     trace_loadvm_state_cleanup();
2381     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
2382         if (se->ops && se->ops->load_cleanup) {
2383             se->ops->load_cleanup(se->opaque);
2384         }
2385     }
2386 }
2387 
2388 /* Return true if we should continue the migration, or false. */
2389 static bool postcopy_pause_incoming(MigrationIncomingState *mis)
2390 {
2391     trace_postcopy_pause_incoming();
2392 
2393     /* Clear the triggered bit to allow one recovery */
2394     mis->postcopy_recover_triggered = false;
2395 
2396     assert(mis->from_src_file);
2397     qemu_file_shutdown(mis->from_src_file);
2398     qemu_fclose(mis->from_src_file);
2399     mis->from_src_file = NULL;
2400 
2401     assert(mis->to_src_file);
2402     qemu_file_shutdown(mis->to_src_file);
2403     qemu_mutex_lock(&mis->rp_mutex);
2404     qemu_fclose(mis->to_src_file);
2405     mis->to_src_file = NULL;
2406     qemu_mutex_unlock(&mis->rp_mutex);
2407 
2408     migrate_set_state(&mis->state, MIGRATION_STATUS_POSTCOPY_ACTIVE,
2409                       MIGRATION_STATUS_POSTCOPY_PAUSED);
2410 
2411     /* Notify the fault thread for the invalidated file handle */
2412     postcopy_fault_thread_notify(mis);
2413 
2414     error_report("Detected IO failure for postcopy. "
2415                  "Migration paused.");
2416 
2417     while (mis->state == MIGRATION_STATUS_POSTCOPY_PAUSED) {
2418         qemu_sem_wait(&mis->postcopy_pause_sem_dst);
2419     }
2420 
2421     trace_postcopy_pause_incoming_continued();
2422 
2423     return true;
2424 }
2425 
2426 int qemu_loadvm_state_main(QEMUFile *f, MigrationIncomingState *mis)
2427 {
2428     uint8_t section_type;
2429     int ret = 0;
2430 
2431 retry:
2432     while (true) {
2433         section_type = qemu_get_byte(f);
2434 
2435         if (qemu_file_get_error(f)) {
2436             ret = qemu_file_get_error(f);
2437             break;
2438         }
2439 
2440         trace_qemu_loadvm_state_section(section_type);
2441         switch (section_type) {
2442         case QEMU_VM_SECTION_START:
2443         case QEMU_VM_SECTION_FULL:
2444             ret = qemu_loadvm_section_start_full(f, mis);
2445             if (ret < 0) {
2446                 goto out;
2447             }
2448             break;
2449         case QEMU_VM_SECTION_PART:
2450         case QEMU_VM_SECTION_END:
2451             ret = qemu_loadvm_section_part_end(f, mis);
2452             if (ret < 0) {
2453                 goto out;
2454             }
2455             break;
2456         case QEMU_VM_COMMAND:
2457             ret = loadvm_process_command(f);
2458             trace_qemu_loadvm_state_section_command(ret);
2459             if ((ret < 0) || (ret == LOADVM_QUIT)) {
2460                 goto out;
2461             }
2462             break;
2463         case QEMU_VM_EOF:
2464             /* This is the end of migration */
2465             goto out;
2466         default:
2467             error_report("Unknown savevm section type %d", section_type);
2468             ret = -EINVAL;
2469             goto out;
2470         }
2471     }
2472 
2473 out:
2474     if (ret < 0) {
2475         qemu_file_set_error(f, ret);
2476 
2477         /*
2478          * If we are during an active postcopy, then we pause instead
2479          * of bail out to at least keep the VM's dirty data.  Note
2480          * that POSTCOPY_INCOMING_LISTENING stage is still not enough,
2481          * during which we're still receiving device states and we
2482          * still haven't yet started the VM on destination.
2483          */
2484         if (postcopy_state_get() == POSTCOPY_INCOMING_RUNNING &&
2485             postcopy_pause_incoming(mis)) {
2486             /* Reset f to point to the newly created channel */
2487             f = mis->from_src_file;
2488             goto retry;
2489         }
2490     }
2491     return ret;
2492 }
2493 
2494 int qemu_loadvm_state(QEMUFile *f)
2495 {
2496     MigrationIncomingState *mis = migration_incoming_get_current();
2497     Error *local_err = NULL;
2498     int ret;
2499 
2500     if (qemu_savevm_state_blocked(&local_err)) {
2501         error_report_err(local_err);
2502         return -EINVAL;
2503     }
2504 
2505     ret = qemu_loadvm_state_header(f);
2506     if (ret) {
2507         return ret;
2508     }
2509 
2510     if (qemu_loadvm_state_setup(f) != 0) {
2511         return -EINVAL;
2512     }
2513 
2514     cpu_synchronize_all_pre_loadvm();
2515 
2516     ret = qemu_loadvm_state_main(f, mis);
2517     qemu_event_set(&mis->main_thread_load_event);
2518 
2519     trace_qemu_loadvm_state_post_main(ret);
2520 
2521     if (mis->have_listen_thread) {
2522         /* Listen thread still going, can't clean up yet */
2523         return ret;
2524     }
2525 
2526     if (ret == 0) {
2527         ret = qemu_file_get_error(f);
2528     }
2529 
2530     /*
2531      * Try to read in the VMDESC section as well, so that dumping tools that
2532      * intercept our migration stream have the chance to see it.
2533      */
2534 
2535     /* We've got to be careful; if we don't read the data and just shut the fd
2536      * then the sender can error if we close while it's still sending.
2537      * We also mustn't read data that isn't there; some transports (RDMA)
2538      * will stall waiting for that data when the source has already closed.
2539      */
2540     if (ret == 0 && should_send_vmdesc()) {
2541         uint8_t *buf;
2542         uint32_t size;
2543         uint8_t  section_type = qemu_get_byte(f);
2544 
2545         if (section_type != QEMU_VM_VMDESCRIPTION) {
2546             error_report("Expected vmdescription section, but got %d",
2547                          section_type);
2548             /*
2549              * It doesn't seem worth failing at this point since
2550              * we apparently have an otherwise valid VM state
2551              */
2552         } else {
2553             buf = g_malloc(0x1000);
2554             size = qemu_get_be32(f);
2555 
2556             while (size > 0) {
2557                 uint32_t read_chunk = MIN(size, 0x1000);
2558                 qemu_get_buffer(f, buf, read_chunk);
2559                 size -= read_chunk;
2560             }
2561             g_free(buf);
2562         }
2563     }
2564 
2565     qemu_loadvm_state_cleanup();
2566     cpu_synchronize_all_post_init();
2567 
2568     return ret;
2569 }
2570 
2571 int qemu_load_device_state(QEMUFile *f)
2572 {
2573     MigrationIncomingState *mis = migration_incoming_get_current();
2574     int ret;
2575 
2576     /* Load QEMU_VM_SECTION_FULL section */
2577     ret = qemu_loadvm_state_main(f, mis);
2578     if (ret < 0) {
2579         error_report("Failed to load device state: %d", ret);
2580         return ret;
2581     }
2582 
2583     cpu_synchronize_all_post_init();
2584     return 0;
2585 }
2586 
2587 int save_snapshot(const char *name, Error **errp)
2588 {
2589     BlockDriverState *bs, *bs1;
2590     QEMUSnapshotInfo sn1, *sn = &sn1, old_sn1, *old_sn = &old_sn1;
2591     int ret = -1;
2592     QEMUFile *f;
2593     int saved_vm_running;
2594     uint64_t vm_state_size;
2595     qemu_timeval tv;
2596     struct tm tm;
2597     AioContext *aio_context;
2598 
2599     if (migration_is_blocked(errp)) {
2600         return ret;
2601     }
2602 
2603     if (!replay_can_snapshot()) {
2604         error_setg(errp, "Record/replay does not allow making snapshot "
2605                    "right now. Try once more later.");
2606         return ret;
2607     }
2608 
2609     if (!bdrv_all_can_snapshot(&bs)) {
2610         error_setg(errp, "Device '%s' is writable but does not support "
2611                    "snapshots", bdrv_get_device_name(bs));
2612         return ret;
2613     }
2614 
2615     /* Delete old snapshots of the same name */
2616     if (name) {
2617         ret = bdrv_all_delete_snapshot(name, &bs1, errp);
2618         if (ret < 0) {
2619             error_prepend(errp, "Error while deleting snapshot on device "
2620                           "'%s': ", bdrv_get_device_name(bs1));
2621             return ret;
2622         }
2623     }
2624 
2625     bs = bdrv_all_find_vmstate_bs();
2626     if (bs == NULL) {
2627         error_setg(errp, "No block device can accept snapshots");
2628         return ret;
2629     }
2630     aio_context = bdrv_get_aio_context(bs);
2631 
2632     saved_vm_running = runstate_is_running();
2633 
2634     ret = global_state_store();
2635     if (ret) {
2636         error_setg(errp, "Error saving global state");
2637         return ret;
2638     }
2639     vm_stop(RUN_STATE_SAVE_VM);
2640 
2641     bdrv_drain_all_begin();
2642 
2643     aio_context_acquire(aio_context);
2644 
2645     memset(sn, 0, sizeof(*sn));
2646 
2647     /* fill auxiliary fields */
2648     qemu_gettimeofday(&tv);
2649     sn->date_sec = tv.tv_sec;
2650     sn->date_nsec = tv.tv_usec * 1000;
2651     sn->vm_clock_nsec = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL);
2652 
2653     if (name) {
2654         ret = bdrv_snapshot_find(bs, old_sn, name);
2655         if (ret >= 0) {
2656             pstrcpy(sn->name, sizeof(sn->name), old_sn->name);
2657             pstrcpy(sn->id_str, sizeof(sn->id_str), old_sn->id_str);
2658         } else {
2659             pstrcpy(sn->name, sizeof(sn->name), name);
2660         }
2661     } else {
2662         /* cast below needed for OpenBSD where tv_sec is still 'long' */
2663         localtime_r((const time_t *)&tv.tv_sec, &tm);
2664         strftime(sn->name, sizeof(sn->name), "vm-%Y%m%d%H%M%S", &tm);
2665     }
2666 
2667     /* save the VM state */
2668     f = qemu_fopen_bdrv(bs, 1);
2669     if (!f) {
2670         error_setg(errp, "Could not open VM state file");
2671         goto the_end;
2672     }
2673     ret = qemu_savevm_state(f, errp);
2674     vm_state_size = qemu_ftell(f);
2675     qemu_fclose(f);
2676     if (ret < 0) {
2677         goto the_end;
2678     }
2679 
2680     /* The bdrv_all_create_snapshot() call that follows acquires the AioContext
2681      * for itself.  BDRV_POLL_WHILE() does not support nested locking because
2682      * it only releases the lock once.  Therefore synchronous I/O will deadlock
2683      * unless we release the AioContext before bdrv_all_create_snapshot().
2684      */
2685     aio_context_release(aio_context);
2686     aio_context = NULL;
2687 
2688     ret = bdrv_all_create_snapshot(sn, bs, vm_state_size, &bs);
2689     if (ret < 0) {
2690         error_setg(errp, "Error while creating snapshot on '%s'",
2691                    bdrv_get_device_name(bs));
2692         goto the_end;
2693     }
2694 
2695     ret = 0;
2696 
2697  the_end:
2698     if (aio_context) {
2699         aio_context_release(aio_context);
2700     }
2701 
2702     bdrv_drain_all_end();
2703 
2704     if (saved_vm_running) {
2705         vm_start();
2706     }
2707     return ret;
2708 }
2709 
2710 void qmp_xen_save_devices_state(const char *filename, bool has_live, bool live,
2711                                 Error **errp)
2712 {
2713     QEMUFile *f;
2714     QIOChannelFile *ioc;
2715     int saved_vm_running;
2716     int ret;
2717 
2718     if (!has_live) {
2719         /* live default to true so old version of Xen tool stack can have a
2720          * successfull live migration */
2721         live = true;
2722     }
2723 
2724     saved_vm_running = runstate_is_running();
2725     vm_stop(RUN_STATE_SAVE_VM);
2726     global_state_store_running();
2727 
2728     ioc = qio_channel_file_new_path(filename, O_WRONLY | O_CREAT, 0660, errp);
2729     if (!ioc) {
2730         goto the_end;
2731     }
2732     qio_channel_set_name(QIO_CHANNEL(ioc), "migration-xen-save-state");
2733     f = qemu_fopen_channel_output(QIO_CHANNEL(ioc));
2734     object_unref(OBJECT(ioc));
2735     ret = qemu_save_device_state(f);
2736     if (ret < 0 || qemu_fclose(f) < 0) {
2737         error_setg(errp, QERR_IO_ERROR);
2738     } else {
2739         /* libxl calls the QMP command "stop" before calling
2740          * "xen-save-devices-state" and in case of migration failure, libxl
2741          * would call "cont".
2742          * So call bdrv_inactivate_all (release locks) here to let the other
2743          * side of the migration take controle of the images.
2744          */
2745         if (live && !saved_vm_running) {
2746             ret = bdrv_inactivate_all();
2747             if (ret) {
2748                 error_setg(errp, "%s: bdrv_inactivate_all() failed (%d)",
2749                            __func__, ret);
2750             }
2751         }
2752     }
2753 
2754  the_end:
2755     if (saved_vm_running) {
2756         vm_start();
2757     }
2758 }
2759 
2760 void qmp_xen_load_devices_state(const char *filename, Error **errp)
2761 {
2762     QEMUFile *f;
2763     QIOChannelFile *ioc;
2764     int ret;
2765 
2766     /* Guest must be paused before loading the device state; the RAM state
2767      * will already have been loaded by xc
2768      */
2769     if (runstate_is_running()) {
2770         error_setg(errp, "Cannot update device state while vm is running");
2771         return;
2772     }
2773     vm_stop(RUN_STATE_RESTORE_VM);
2774 
2775     ioc = qio_channel_file_new_path(filename, O_RDONLY | O_BINARY, 0, errp);
2776     if (!ioc) {
2777         return;
2778     }
2779     qio_channel_set_name(QIO_CHANNEL(ioc), "migration-xen-load-state");
2780     f = qemu_fopen_channel_input(QIO_CHANNEL(ioc));
2781     object_unref(OBJECT(ioc));
2782 
2783     ret = qemu_loadvm_state(f);
2784     qemu_fclose(f);
2785     if (ret < 0) {
2786         error_setg(errp, QERR_IO_ERROR);
2787     }
2788     migration_incoming_state_destroy();
2789 }
2790 
2791 int load_snapshot(const char *name, Error **errp)
2792 {
2793     BlockDriverState *bs, *bs_vm_state;
2794     QEMUSnapshotInfo sn;
2795     QEMUFile *f;
2796     int ret;
2797     AioContext *aio_context;
2798     MigrationIncomingState *mis = migration_incoming_get_current();
2799 
2800     if (!replay_can_snapshot()) {
2801         error_setg(errp, "Record/replay does not allow loading snapshot "
2802                    "right now. Try once more later.");
2803         return -EINVAL;
2804     }
2805 
2806     if (!bdrv_all_can_snapshot(&bs)) {
2807         error_setg(errp,
2808                    "Device '%s' is writable but does not support snapshots",
2809                    bdrv_get_device_name(bs));
2810         return -ENOTSUP;
2811     }
2812     ret = bdrv_all_find_snapshot(name, &bs);
2813     if (ret < 0) {
2814         error_setg(errp,
2815                    "Device '%s' does not have the requested snapshot '%s'",
2816                    bdrv_get_device_name(bs), name);
2817         return ret;
2818     }
2819 
2820     bs_vm_state = bdrv_all_find_vmstate_bs();
2821     if (!bs_vm_state) {
2822         error_setg(errp, "No block device supports snapshots");
2823         return -ENOTSUP;
2824     }
2825     aio_context = bdrv_get_aio_context(bs_vm_state);
2826 
2827     /* Don't even try to load empty VM states */
2828     aio_context_acquire(aio_context);
2829     ret = bdrv_snapshot_find(bs_vm_state, &sn, name);
2830     aio_context_release(aio_context);
2831     if (ret < 0) {
2832         return ret;
2833     } else if (sn.vm_state_size == 0) {
2834         error_setg(errp, "This is a disk-only snapshot. Revert to it "
2835                    " offline using qemu-img");
2836         return -EINVAL;
2837     }
2838 
2839     /* Flush all IO requests so they don't interfere with the new state.  */
2840     bdrv_drain_all_begin();
2841 
2842     ret = bdrv_all_goto_snapshot(name, &bs, errp);
2843     if (ret < 0) {
2844         error_prepend(errp, "Could not load snapshot '%s' on '%s': ",
2845                       name, bdrv_get_device_name(bs));
2846         goto err_drain;
2847     }
2848 
2849     /* restore the VM state */
2850     f = qemu_fopen_bdrv(bs_vm_state, 0);
2851     if (!f) {
2852         error_setg(errp, "Could not open VM state file");
2853         ret = -EINVAL;
2854         goto err_drain;
2855     }
2856 
2857     qemu_system_reset(SHUTDOWN_CAUSE_NONE);
2858     mis->from_src_file = f;
2859 
2860     aio_context_acquire(aio_context);
2861     ret = qemu_loadvm_state(f);
2862     migration_incoming_state_destroy();
2863     aio_context_release(aio_context);
2864 
2865     bdrv_drain_all_end();
2866 
2867     if (ret < 0) {
2868         error_setg(errp, "Error %d while loading VM state", ret);
2869         return ret;
2870     }
2871 
2872     return 0;
2873 
2874 err_drain:
2875     bdrv_drain_all_end();
2876     return ret;
2877 }
2878 
2879 void vmstate_register_ram(MemoryRegion *mr, DeviceState *dev)
2880 {
2881     qemu_ram_set_idstr(mr->ram_block,
2882                        memory_region_name(mr), dev);
2883     qemu_ram_set_migratable(mr->ram_block);
2884 }
2885 
2886 void vmstate_unregister_ram(MemoryRegion *mr, DeviceState *dev)
2887 {
2888     qemu_ram_unset_idstr(mr->ram_block);
2889     qemu_ram_unset_migratable(mr->ram_block);
2890 }
2891 
2892 void vmstate_register_ram_global(MemoryRegion *mr)
2893 {
2894     vmstate_register_ram(mr, NULL);
2895 }
2896 
2897 bool vmstate_check_only_migratable(const VMStateDescription *vmsd)
2898 {
2899     /* check needed if --only-migratable is specified */
2900     if (!only_migratable) {
2901         return true;
2902     }
2903 
2904     return !(vmsd && vmsd->unmigratable);
2905 }
2906