xref: /openbmc/qemu/migration/savevm.c (revision 650d103d3ea959212f826acb9d3fe80cf30e347b)
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 "block/snapshot.h"
55 #include "qemu/cutils.h"
56 #include "io/channel-buffer.h"
57 #include "io/channel-file.h"
58 #include "sysemu/replay.h"
59 #include "qjson.h"
60 #include "migration/colo.h"
61 #include "qemu/bitmap.h"
62 #include "net/announce.h"
63 
64 const unsigned int postcopy_ram_discard_version = 0;
65 
66 /* Subcommands for QEMU_VM_COMMAND */
67 enum qemu_vm_cmd {
68     MIG_CMD_INVALID = 0,   /* Must be 0 */
69     MIG_CMD_OPEN_RETURN_PATH,  /* Tell the dest to open the Return path */
70     MIG_CMD_PING,              /* Request a PONG on the RP */
71 
72     MIG_CMD_POSTCOPY_ADVISE,       /* Prior to any page transfers, just
73                                       warn we might want to do PC */
74     MIG_CMD_POSTCOPY_LISTEN,       /* Start listening for incoming
75                                       pages as it's running. */
76     MIG_CMD_POSTCOPY_RUN,          /* Start execution */
77 
78     MIG_CMD_POSTCOPY_RAM_DISCARD,  /* A list of pages to discard that
79                                       were previously sent during
80                                       precopy but are dirty. */
81     MIG_CMD_PACKAGED,          /* Send a wrapped stream within this stream */
82     MIG_CMD_ENABLE_COLO,       /* Enable COLO */
83     MIG_CMD_POSTCOPY_RESUME,   /* resume postcopy on dest */
84     MIG_CMD_RECV_BITMAP,       /* Request for recved bitmap on dst */
85     MIG_CMD_MAX
86 };
87 
88 #define MAX_VM_CMD_PACKAGED_SIZE UINT32_MAX
89 static struct mig_cmd_args {
90     ssize_t     len; /* -1 = variable */
91     const char *name;
92 } mig_cmd_args[] = {
93     [MIG_CMD_INVALID]          = { .len = -1, .name = "INVALID" },
94     [MIG_CMD_OPEN_RETURN_PATH] = { .len =  0, .name = "OPEN_RETURN_PATH" },
95     [MIG_CMD_PING]             = { .len = sizeof(uint32_t), .name = "PING" },
96     [MIG_CMD_POSTCOPY_ADVISE]  = { .len = -1, .name = "POSTCOPY_ADVISE" },
97     [MIG_CMD_POSTCOPY_LISTEN]  = { .len =  0, .name = "POSTCOPY_LISTEN" },
98     [MIG_CMD_POSTCOPY_RUN]     = { .len =  0, .name = "POSTCOPY_RUN" },
99     [MIG_CMD_POSTCOPY_RAM_DISCARD] = {
100                                    .len = -1, .name = "POSTCOPY_RAM_DISCARD" },
101     [MIG_CMD_POSTCOPY_RESUME]  = { .len =  0, .name = "POSTCOPY_RESUME" },
102     [MIG_CMD_PACKAGED]         = { .len =  4, .name = "PACKAGED" },
103     [MIG_CMD_RECV_BITMAP]      = { .len = -1, .name = "RECV_BITMAP" },
104     [MIG_CMD_MAX]              = { .len = -1, .name = "MAX" },
105 };
106 
107 /* Note for MIG_CMD_POSTCOPY_ADVISE:
108  * The format of arguments is depending on postcopy mode:
109  * - postcopy RAM only
110  *   uint64_t host page size
111  *   uint64_t taget page size
112  *
113  * - postcopy RAM and postcopy dirty bitmaps
114  *   format is the same as for postcopy RAM only
115  *
116  * - postcopy dirty bitmaps only
117  *   Nothing. Command length field is 0.
118  *
119  * Be careful: adding a new postcopy entity with some other parameters should
120  * not break format self-description ability. Good way is to introduce some
121  * generic extendable format with an exception for two old entities.
122  */
123 
124 /***********************************************************/
125 /* savevm/loadvm support */
126 
127 static ssize_t block_writev_buffer(void *opaque, struct iovec *iov, int iovcnt,
128                                    int64_t pos)
129 {
130     int ret;
131     QEMUIOVector qiov;
132 
133     qemu_iovec_init_external(&qiov, iov, iovcnt);
134     ret = bdrv_writev_vmstate(opaque, &qiov, pos);
135     if (ret < 0) {
136         return ret;
137     }
138 
139     return qiov.size;
140 }
141 
142 static ssize_t block_get_buffer(void *opaque, uint8_t *buf, int64_t pos,
143                                 size_t size)
144 {
145     return bdrv_load_vmstate(opaque, buf, pos, size);
146 }
147 
148 static int bdrv_fclose(void *opaque)
149 {
150     return bdrv_flush(opaque);
151 }
152 
153 static const QEMUFileOps bdrv_read_ops = {
154     .get_buffer = block_get_buffer,
155     .close =      bdrv_fclose
156 };
157 
158 static const QEMUFileOps bdrv_write_ops = {
159     .writev_buffer  = block_writev_buffer,
160     .close          = bdrv_fclose
161 };
162 
163 static QEMUFile *qemu_fopen_bdrv(BlockDriverState *bs, int is_writable)
164 {
165     if (is_writable) {
166         return qemu_fopen_ops(bs, &bdrv_write_ops);
167     }
168     return qemu_fopen_ops(bs, &bdrv_read_ops);
169 }
170 
171 
172 /* QEMUFile timer support.
173  * Not in qemu-file.c to not add qemu-timer.c as dependency to qemu-file.c
174  */
175 
176 void timer_put(QEMUFile *f, QEMUTimer *ts)
177 {
178     uint64_t expire_time;
179 
180     expire_time = timer_expire_time_ns(ts);
181     qemu_put_be64(f, expire_time);
182 }
183 
184 void timer_get(QEMUFile *f, QEMUTimer *ts)
185 {
186     uint64_t expire_time;
187 
188     expire_time = qemu_get_be64(f);
189     if (expire_time != -1) {
190         timer_mod_ns(ts, expire_time);
191     } else {
192         timer_del(ts);
193     }
194 }
195 
196 
197 /* VMState timer support.
198  * Not in vmstate.c to not add qemu-timer.c as dependency to vmstate.c
199  */
200 
201 static int get_timer(QEMUFile *f, void *pv, size_t size,
202                      const VMStateField *field)
203 {
204     QEMUTimer *v = pv;
205     timer_get(f, v);
206     return 0;
207 }
208 
209 static int put_timer(QEMUFile *f, void *pv, size_t size,
210                      const VMStateField *field, QJSON *vmdesc)
211 {
212     QEMUTimer *v = pv;
213     timer_put(f, v);
214 
215     return 0;
216 }
217 
218 const VMStateInfo vmstate_info_timer = {
219     .name = "timer",
220     .get  = get_timer,
221     .put  = put_timer,
222 };
223 
224 
225 typedef struct CompatEntry {
226     char idstr[256];
227     int instance_id;
228 } CompatEntry;
229 
230 typedef struct SaveStateEntry {
231     QTAILQ_ENTRY(SaveStateEntry) entry;
232     char idstr[256];
233     int instance_id;
234     int alias_id;
235     int version_id;
236     /* version id read from the stream */
237     int load_version_id;
238     int section_id;
239     /* section id read from the stream */
240     int load_section_id;
241     const SaveVMHandlers *ops;
242     const VMStateDescription *vmsd;
243     void *opaque;
244     CompatEntry *compat;
245     int is_ram;
246 } SaveStateEntry;
247 
248 typedef struct SaveState {
249     QTAILQ_HEAD(, SaveStateEntry) handlers;
250     int global_section_id;
251     uint32_t len;
252     const char *name;
253     uint32_t target_page_bits;
254     uint32_t caps_count;
255     MigrationCapability *capabilities;
256 } SaveState;
257 
258 static SaveState savevm_state = {
259     .handlers = QTAILQ_HEAD_INITIALIZER(savevm_state.handlers),
260     .global_section_id = 0,
261 };
262 
263 static bool should_validate_capability(int capability)
264 {
265     assert(capability >= 0 && capability < MIGRATION_CAPABILITY__MAX);
266     /* Validate only new capabilities to keep compatibility. */
267     switch (capability) {
268     case MIGRATION_CAPABILITY_X_IGNORE_SHARED:
269         return true;
270     default:
271         return false;
272     }
273 }
274 
275 static uint32_t get_validatable_capabilities_count(void)
276 {
277     MigrationState *s = migrate_get_current();
278     uint32_t result = 0;
279     int i;
280     for (i = 0; i < MIGRATION_CAPABILITY__MAX; i++) {
281         if (should_validate_capability(i) && s->enabled_capabilities[i]) {
282             result++;
283         }
284     }
285     return result;
286 }
287 
288 static int configuration_pre_save(void *opaque)
289 {
290     SaveState *state = opaque;
291     const char *current_name = MACHINE_GET_CLASS(current_machine)->name;
292     MigrationState *s = migrate_get_current();
293     int i, j;
294 
295     state->len = strlen(current_name);
296     state->name = current_name;
297     state->target_page_bits = qemu_target_page_bits();
298 
299     state->caps_count = get_validatable_capabilities_count();
300     state->capabilities = g_renew(MigrationCapability, state->capabilities,
301                                   state->caps_count);
302     for (i = j = 0; i < MIGRATION_CAPABILITY__MAX; i++) {
303         if (should_validate_capability(i) && s->enabled_capabilities[i]) {
304             state->capabilities[j++] = i;
305         }
306     }
307 
308     return 0;
309 }
310 
311 static int configuration_pre_load(void *opaque)
312 {
313     SaveState *state = opaque;
314 
315     /* If there is no target-page-bits subsection it means the source
316      * predates the variable-target-page-bits support and is using the
317      * minimum possible value for this CPU.
318      */
319     state->target_page_bits = qemu_target_page_bits_min();
320     return 0;
321 }
322 
323 static bool configuration_validate_capabilities(SaveState *state)
324 {
325     bool ret = true;
326     MigrationState *s = migrate_get_current();
327     unsigned long *source_caps_bm;
328     int i;
329 
330     source_caps_bm = bitmap_new(MIGRATION_CAPABILITY__MAX);
331     for (i = 0; i < state->caps_count; i++) {
332         MigrationCapability capability = state->capabilities[i];
333         set_bit(capability, source_caps_bm);
334     }
335 
336     for (i = 0; i < MIGRATION_CAPABILITY__MAX; i++) {
337         bool source_state, target_state;
338         if (!should_validate_capability(i)) {
339             continue;
340         }
341         source_state = test_bit(i, source_caps_bm);
342         target_state = s->enabled_capabilities[i];
343         if (source_state != target_state) {
344             error_report("Capability %s is %s, but received capability is %s",
345                          MigrationCapability_str(i),
346                          target_state ? "on" : "off",
347                          source_state ? "on" : "off");
348             ret = false;
349             /* Don't break here to report all failed capabilities */
350         }
351     }
352 
353     g_free(source_caps_bm);
354     return ret;
355 }
356 
357 static int configuration_post_load(void *opaque, int version_id)
358 {
359     SaveState *state = opaque;
360     const char *current_name = MACHINE_GET_CLASS(current_machine)->name;
361 
362     if (strncmp(state->name, current_name, state->len) != 0) {
363         error_report("Machine type received is '%.*s' and local is '%s'",
364                      (int) state->len, state->name, current_name);
365         return -EINVAL;
366     }
367 
368     if (state->target_page_bits != qemu_target_page_bits()) {
369         error_report("Received TARGET_PAGE_BITS is %d but local is %d",
370                      state->target_page_bits, qemu_target_page_bits());
371         return -EINVAL;
372     }
373 
374     if (!configuration_validate_capabilities(state)) {
375         return -EINVAL;
376     }
377 
378     return 0;
379 }
380 
381 static int get_capability(QEMUFile *f, void *pv, size_t size,
382                           const VMStateField *field)
383 {
384     MigrationCapability *capability = pv;
385     char capability_str[UINT8_MAX + 1];
386     uint8_t len;
387     int i;
388 
389     len = qemu_get_byte(f);
390     qemu_get_buffer(f, (uint8_t *)capability_str, len);
391     capability_str[len] = '\0';
392     for (i = 0; i < MIGRATION_CAPABILITY__MAX; i++) {
393         if (!strcmp(MigrationCapability_str(i), capability_str)) {
394             *capability = i;
395             return 0;
396         }
397     }
398     error_report("Received unknown capability %s", capability_str);
399     return -EINVAL;
400 }
401 
402 static int put_capability(QEMUFile *f, void *pv, size_t size,
403                           const VMStateField *field, QJSON *vmdesc)
404 {
405     MigrationCapability *capability = pv;
406     const char *capability_str = MigrationCapability_str(*capability);
407     size_t len = strlen(capability_str);
408     assert(len <= UINT8_MAX);
409 
410     qemu_put_byte(f, len);
411     qemu_put_buffer(f, (uint8_t *)capability_str, len);
412     return 0;
413 }
414 
415 static const VMStateInfo vmstate_info_capability = {
416     .name = "capability",
417     .get  = get_capability,
418     .put  = put_capability,
419 };
420 
421 /* The target-page-bits subsection is present only if the
422  * target page size is not the same as the default (ie the
423  * minimum page size for a variable-page-size guest CPU).
424  * If it is present then it contains the actual target page
425  * bits for the machine, and migration will fail if the
426  * two ends don't agree about it.
427  */
428 static bool vmstate_target_page_bits_needed(void *opaque)
429 {
430     return qemu_target_page_bits()
431         > qemu_target_page_bits_min();
432 }
433 
434 static const VMStateDescription vmstate_target_page_bits = {
435     .name = "configuration/target-page-bits",
436     .version_id = 1,
437     .minimum_version_id = 1,
438     .needed = vmstate_target_page_bits_needed,
439     .fields = (VMStateField[]) {
440         VMSTATE_UINT32(target_page_bits, SaveState),
441         VMSTATE_END_OF_LIST()
442     }
443 };
444 
445 static bool vmstate_capabilites_needed(void *opaque)
446 {
447     return get_validatable_capabilities_count() > 0;
448 }
449 
450 static const VMStateDescription vmstate_capabilites = {
451     .name = "configuration/capabilities",
452     .version_id = 1,
453     .minimum_version_id = 1,
454     .needed = vmstate_capabilites_needed,
455     .fields = (VMStateField[]) {
456         VMSTATE_UINT32_V(caps_count, SaveState, 1),
457         VMSTATE_VARRAY_UINT32_ALLOC(capabilities, SaveState, caps_count, 1,
458                                     vmstate_info_capability,
459                                     MigrationCapability),
460         VMSTATE_END_OF_LIST()
461     }
462 };
463 
464 static const VMStateDescription vmstate_configuration = {
465     .name = "configuration",
466     .version_id = 1,
467     .pre_load = configuration_pre_load,
468     .post_load = configuration_post_load,
469     .pre_save = configuration_pre_save,
470     .fields = (VMStateField[]) {
471         VMSTATE_UINT32(len, SaveState),
472         VMSTATE_VBUFFER_ALLOC_UINT32(name, SaveState, 0, NULL, len),
473         VMSTATE_END_OF_LIST()
474     },
475     .subsections = (const VMStateDescription*[]) {
476         &vmstate_target_page_bits,
477         &vmstate_capabilites,
478         NULL
479     }
480 };
481 
482 static void dump_vmstate_vmsd(FILE *out_file,
483                               const VMStateDescription *vmsd, int indent,
484                               bool is_subsection);
485 
486 static void dump_vmstate_vmsf(FILE *out_file, const VMStateField *field,
487                               int indent)
488 {
489     fprintf(out_file, "%*s{\n", indent, "");
490     indent += 2;
491     fprintf(out_file, "%*s\"field\": \"%s\",\n", indent, "", field->name);
492     fprintf(out_file, "%*s\"version_id\": %d,\n", indent, "",
493             field->version_id);
494     fprintf(out_file, "%*s\"field_exists\": %s,\n", indent, "",
495             field->field_exists ? "true" : "false");
496     fprintf(out_file, "%*s\"size\": %zu", indent, "", field->size);
497     if (field->vmsd != NULL) {
498         fprintf(out_file, ",\n");
499         dump_vmstate_vmsd(out_file, field->vmsd, indent, false);
500     }
501     fprintf(out_file, "\n%*s}", indent - 2, "");
502 }
503 
504 static void dump_vmstate_vmss(FILE *out_file,
505                               const VMStateDescription **subsection,
506                               int indent)
507 {
508     if (*subsection != NULL) {
509         dump_vmstate_vmsd(out_file, *subsection, indent, true);
510     }
511 }
512 
513 static void dump_vmstate_vmsd(FILE *out_file,
514                               const VMStateDescription *vmsd, int indent,
515                               bool is_subsection)
516 {
517     if (is_subsection) {
518         fprintf(out_file, "%*s{\n", indent, "");
519     } else {
520         fprintf(out_file, "%*s\"%s\": {\n", indent, "", "Description");
521     }
522     indent += 2;
523     fprintf(out_file, "%*s\"name\": \"%s\",\n", indent, "", vmsd->name);
524     fprintf(out_file, "%*s\"version_id\": %d,\n", indent, "",
525             vmsd->version_id);
526     fprintf(out_file, "%*s\"minimum_version_id\": %d", indent, "",
527             vmsd->minimum_version_id);
528     if (vmsd->fields != NULL) {
529         const VMStateField *field = vmsd->fields;
530         bool first;
531 
532         fprintf(out_file, ",\n%*s\"Fields\": [\n", indent, "");
533         first = true;
534         while (field->name != NULL) {
535             if (field->flags & VMS_MUST_EXIST) {
536                 /* Ignore VMSTATE_VALIDATE bits; these don't get migrated */
537                 field++;
538                 continue;
539             }
540             if (!first) {
541                 fprintf(out_file, ",\n");
542             }
543             dump_vmstate_vmsf(out_file, field, indent + 2);
544             field++;
545             first = false;
546         }
547         fprintf(out_file, "\n%*s]", indent, "");
548     }
549     if (vmsd->subsections != NULL) {
550         const VMStateDescription **subsection = vmsd->subsections;
551         bool first;
552 
553         fprintf(out_file, ",\n%*s\"Subsections\": [\n", indent, "");
554         first = true;
555         while (*subsection != NULL) {
556             if (!first) {
557                 fprintf(out_file, ",\n");
558             }
559             dump_vmstate_vmss(out_file, subsection, indent + 2);
560             subsection++;
561             first = false;
562         }
563         fprintf(out_file, "\n%*s]", indent, "");
564     }
565     fprintf(out_file, "\n%*s}", indent - 2, "");
566 }
567 
568 static void dump_machine_type(FILE *out_file)
569 {
570     MachineClass *mc;
571 
572     mc = MACHINE_GET_CLASS(current_machine);
573 
574     fprintf(out_file, "  \"vmschkmachine\": {\n");
575     fprintf(out_file, "    \"Name\": \"%s\"\n", mc->name);
576     fprintf(out_file, "  },\n");
577 }
578 
579 void dump_vmstate_json_to_file(FILE *out_file)
580 {
581     GSList *list, *elt;
582     bool first;
583 
584     fprintf(out_file, "{\n");
585     dump_machine_type(out_file);
586 
587     first = true;
588     list = object_class_get_list(TYPE_DEVICE, true);
589     for (elt = list; elt; elt = elt->next) {
590         DeviceClass *dc = OBJECT_CLASS_CHECK(DeviceClass, elt->data,
591                                              TYPE_DEVICE);
592         const char *name;
593         int indent = 2;
594 
595         if (!dc->vmsd) {
596             continue;
597         }
598 
599         if (!first) {
600             fprintf(out_file, ",\n");
601         }
602         name = object_class_get_name(OBJECT_CLASS(dc));
603         fprintf(out_file, "%*s\"%s\": {\n", indent, "", name);
604         indent += 2;
605         fprintf(out_file, "%*s\"Name\": \"%s\",\n", indent, "", name);
606         fprintf(out_file, "%*s\"version_id\": %d,\n", indent, "",
607                 dc->vmsd->version_id);
608         fprintf(out_file, "%*s\"minimum_version_id\": %d,\n", indent, "",
609                 dc->vmsd->minimum_version_id);
610 
611         dump_vmstate_vmsd(out_file, dc->vmsd, indent, false);
612 
613         fprintf(out_file, "\n%*s}", indent - 2, "");
614         first = false;
615     }
616     fprintf(out_file, "\n}\n");
617     fclose(out_file);
618 }
619 
620 static int calculate_new_instance_id(const char *idstr)
621 {
622     SaveStateEntry *se;
623     int instance_id = 0;
624 
625     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
626         if (strcmp(idstr, se->idstr) == 0
627             && instance_id <= se->instance_id) {
628             instance_id = se->instance_id + 1;
629         }
630     }
631     return instance_id;
632 }
633 
634 static int calculate_compat_instance_id(const char *idstr)
635 {
636     SaveStateEntry *se;
637     int instance_id = 0;
638 
639     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
640         if (!se->compat) {
641             continue;
642         }
643 
644         if (strcmp(idstr, se->compat->idstr) == 0
645             && instance_id <= se->compat->instance_id) {
646             instance_id = se->compat->instance_id + 1;
647         }
648     }
649     return instance_id;
650 }
651 
652 static inline MigrationPriority save_state_priority(SaveStateEntry *se)
653 {
654     if (se->vmsd) {
655         return se->vmsd->priority;
656     }
657     return MIG_PRI_DEFAULT;
658 }
659 
660 static void savevm_state_handler_insert(SaveStateEntry *nse)
661 {
662     MigrationPriority priority = save_state_priority(nse);
663     SaveStateEntry *se;
664 
665     assert(priority <= MIG_PRI_MAX);
666 
667     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
668         if (save_state_priority(se) < priority) {
669             break;
670         }
671     }
672 
673     if (se) {
674         QTAILQ_INSERT_BEFORE(se, nse, entry);
675     } else {
676         QTAILQ_INSERT_TAIL(&savevm_state.handlers, nse, entry);
677     }
678 }
679 
680 /* TODO: Individual devices generally have very little idea about the rest
681    of the system, so instance_id should be removed/replaced.
682    Meanwhile pass -1 as instance_id if you do not already have a clearly
683    distinguishing id for all instances of your device class. */
684 int register_savevm_live(DeviceState *dev,
685                          const char *idstr,
686                          int instance_id,
687                          int version_id,
688                          const SaveVMHandlers *ops,
689                          void *opaque)
690 {
691     SaveStateEntry *se;
692 
693     se = g_new0(SaveStateEntry, 1);
694     se->version_id = version_id;
695     se->section_id = savevm_state.global_section_id++;
696     se->ops = ops;
697     se->opaque = opaque;
698     se->vmsd = NULL;
699     /* if this is a live_savem then set is_ram */
700     if (ops->save_setup != NULL) {
701         se->is_ram = 1;
702     }
703 
704     if (dev) {
705         char *id = qdev_get_dev_path(dev);
706         if (id) {
707             if (snprintf(se->idstr, sizeof(se->idstr), "%s/", id) >=
708                 sizeof(se->idstr)) {
709                 error_report("Path too long for VMState (%s)", id);
710                 g_free(id);
711                 g_free(se);
712 
713                 return -1;
714             }
715             g_free(id);
716 
717             se->compat = g_new0(CompatEntry, 1);
718             pstrcpy(se->compat->idstr, sizeof(se->compat->idstr), idstr);
719             se->compat->instance_id = instance_id == -1 ?
720                          calculate_compat_instance_id(idstr) : instance_id;
721             instance_id = -1;
722         }
723     }
724     pstrcat(se->idstr, sizeof(se->idstr), idstr);
725 
726     if (instance_id == -1) {
727         se->instance_id = calculate_new_instance_id(se->idstr);
728     } else {
729         se->instance_id = instance_id;
730     }
731     assert(!se->compat || se->instance_id == 0);
732     savevm_state_handler_insert(se);
733     return 0;
734 }
735 
736 void unregister_savevm(DeviceState *dev, const char *idstr, void *opaque)
737 {
738     SaveStateEntry *se, *new_se;
739     char id[256] = "";
740 
741     if (dev) {
742         char *path = qdev_get_dev_path(dev);
743         if (path) {
744             pstrcpy(id, sizeof(id), path);
745             pstrcat(id, sizeof(id), "/");
746             g_free(path);
747         }
748     }
749     pstrcat(id, sizeof(id), idstr);
750 
751     QTAILQ_FOREACH_SAFE(se, &savevm_state.handlers, entry, new_se) {
752         if (strcmp(se->idstr, id) == 0 && se->opaque == opaque) {
753             QTAILQ_REMOVE(&savevm_state.handlers, se, entry);
754             g_free(se->compat);
755             g_free(se);
756         }
757     }
758 }
759 
760 int vmstate_register_with_alias_id(DeviceState *dev, int instance_id,
761                                    const VMStateDescription *vmsd,
762                                    void *opaque, int alias_id,
763                                    int required_for_version,
764                                    Error **errp)
765 {
766     SaveStateEntry *se;
767 
768     /* If this triggers, alias support can be dropped for the vmsd. */
769     assert(alias_id == -1 || required_for_version >= vmsd->minimum_version_id);
770 
771     se = g_new0(SaveStateEntry, 1);
772     se->version_id = vmsd->version_id;
773     se->section_id = savevm_state.global_section_id++;
774     se->opaque = opaque;
775     se->vmsd = vmsd;
776     se->alias_id = alias_id;
777 
778     if (dev) {
779         char *id = qdev_get_dev_path(dev);
780         if (id) {
781             if (snprintf(se->idstr, sizeof(se->idstr), "%s/", id) >=
782                 sizeof(se->idstr)) {
783                 error_setg(errp, "Path too long for VMState (%s)", id);
784                 g_free(id);
785                 g_free(se);
786 
787                 return -1;
788             }
789             g_free(id);
790 
791             se->compat = g_new0(CompatEntry, 1);
792             pstrcpy(se->compat->idstr, sizeof(se->compat->idstr), vmsd->name);
793             se->compat->instance_id = instance_id == -1 ?
794                          calculate_compat_instance_id(vmsd->name) : instance_id;
795             instance_id = -1;
796         }
797     }
798     pstrcat(se->idstr, sizeof(se->idstr), vmsd->name);
799 
800     if (instance_id == -1) {
801         se->instance_id = calculate_new_instance_id(se->idstr);
802     } else {
803         se->instance_id = instance_id;
804     }
805     assert(!se->compat || se->instance_id == 0);
806     savevm_state_handler_insert(se);
807     return 0;
808 }
809 
810 void vmstate_unregister(DeviceState *dev, const VMStateDescription *vmsd,
811                         void *opaque)
812 {
813     SaveStateEntry *se, *new_se;
814 
815     QTAILQ_FOREACH_SAFE(se, &savevm_state.handlers, entry, new_se) {
816         if (se->vmsd == vmsd && se->opaque == opaque) {
817             QTAILQ_REMOVE(&savevm_state.handlers, se, entry);
818             g_free(se->compat);
819             g_free(se);
820         }
821     }
822 }
823 
824 static int vmstate_load(QEMUFile *f, SaveStateEntry *se)
825 {
826     trace_vmstate_load(se->idstr, se->vmsd ? se->vmsd->name : "(old)");
827     if (!se->vmsd) {         /* Old style */
828         return se->ops->load_state(f, se->opaque, se->load_version_id);
829     }
830     return vmstate_load_state(f, se->vmsd, se->opaque, se->load_version_id);
831 }
832 
833 static void vmstate_save_old_style(QEMUFile *f, SaveStateEntry *se, QJSON *vmdesc)
834 {
835     int64_t old_offset, size;
836 
837     old_offset = qemu_ftell_fast(f);
838     se->ops->save_state(f, se->opaque);
839     size = qemu_ftell_fast(f) - old_offset;
840 
841     if (vmdesc) {
842         json_prop_int(vmdesc, "size", size);
843         json_start_array(vmdesc, "fields");
844         json_start_object(vmdesc, NULL);
845         json_prop_str(vmdesc, "name", "data");
846         json_prop_int(vmdesc, "size", size);
847         json_prop_str(vmdesc, "type", "buffer");
848         json_end_object(vmdesc);
849         json_end_array(vmdesc);
850     }
851 }
852 
853 static int vmstate_save(QEMUFile *f, SaveStateEntry *se, QJSON *vmdesc)
854 {
855     trace_vmstate_save(se->idstr, se->vmsd ? se->vmsd->name : "(old)");
856     if (!se->vmsd) {
857         vmstate_save_old_style(f, se, vmdesc);
858         return 0;
859     }
860     return vmstate_save_state(f, se->vmsd, se->opaque, vmdesc);
861 }
862 
863 /*
864  * Write the header for device section (QEMU_VM_SECTION START/END/PART/FULL)
865  */
866 static void save_section_header(QEMUFile *f, SaveStateEntry *se,
867                                 uint8_t section_type)
868 {
869     qemu_put_byte(f, section_type);
870     qemu_put_be32(f, se->section_id);
871 
872     if (section_type == QEMU_VM_SECTION_FULL ||
873         section_type == QEMU_VM_SECTION_START) {
874         /* ID string */
875         size_t len = strlen(se->idstr);
876         qemu_put_byte(f, len);
877         qemu_put_buffer(f, (uint8_t *)se->idstr, len);
878 
879         qemu_put_be32(f, se->instance_id);
880         qemu_put_be32(f, se->version_id);
881     }
882 }
883 
884 /*
885  * Write a footer onto device sections that catches cases misformatted device
886  * sections.
887  */
888 static void save_section_footer(QEMUFile *f, SaveStateEntry *se)
889 {
890     if (migrate_get_current()->send_section_footer) {
891         qemu_put_byte(f, QEMU_VM_SECTION_FOOTER);
892         qemu_put_be32(f, se->section_id);
893     }
894 }
895 
896 /**
897  * qemu_savevm_command_send: Send a 'QEMU_VM_COMMAND' type element with the
898  *                           command and associated data.
899  *
900  * @f: File to send command on
901  * @command: Command type to send
902  * @len: Length of associated data
903  * @data: Data associated with command.
904  */
905 static void qemu_savevm_command_send(QEMUFile *f,
906                                      enum qemu_vm_cmd command,
907                                      uint16_t len,
908                                      uint8_t *data)
909 {
910     trace_savevm_command_send(command, len);
911     qemu_put_byte(f, QEMU_VM_COMMAND);
912     qemu_put_be16(f, (uint16_t)command);
913     qemu_put_be16(f, len);
914     qemu_put_buffer(f, data, len);
915     qemu_fflush(f);
916 }
917 
918 void qemu_savevm_send_colo_enable(QEMUFile *f)
919 {
920     trace_savevm_send_colo_enable();
921     qemu_savevm_command_send(f, MIG_CMD_ENABLE_COLO, 0, NULL);
922 }
923 
924 void qemu_savevm_send_ping(QEMUFile *f, uint32_t value)
925 {
926     uint32_t buf;
927 
928     trace_savevm_send_ping(value);
929     buf = cpu_to_be32(value);
930     qemu_savevm_command_send(f, MIG_CMD_PING, sizeof(value), (uint8_t *)&buf);
931 }
932 
933 void qemu_savevm_send_open_return_path(QEMUFile *f)
934 {
935     trace_savevm_send_open_return_path();
936     qemu_savevm_command_send(f, MIG_CMD_OPEN_RETURN_PATH, 0, NULL);
937 }
938 
939 /* We have a buffer of data to send; we don't want that all to be loaded
940  * by the command itself, so the command contains just the length of the
941  * extra buffer that we then send straight after it.
942  * TODO: Must be a better way to organise that
943  *
944  * Returns:
945  *    0 on success
946  *    -ve on error
947  */
948 int qemu_savevm_send_packaged(QEMUFile *f, const uint8_t *buf, size_t len)
949 {
950     uint32_t tmp;
951 
952     if (len > MAX_VM_CMD_PACKAGED_SIZE) {
953         error_report("%s: Unreasonably large packaged state: %zu",
954                      __func__, len);
955         return -1;
956     }
957 
958     tmp = cpu_to_be32(len);
959 
960     trace_qemu_savevm_send_packaged();
961     qemu_savevm_command_send(f, MIG_CMD_PACKAGED, 4, (uint8_t *)&tmp);
962 
963     qemu_put_buffer(f, buf, len);
964 
965     return 0;
966 }
967 
968 /* Send prior to any postcopy transfer */
969 void qemu_savevm_send_postcopy_advise(QEMUFile *f)
970 {
971     if (migrate_postcopy_ram()) {
972         uint64_t tmp[2];
973         tmp[0] = cpu_to_be64(ram_pagesize_summary());
974         tmp[1] = cpu_to_be64(qemu_target_page_size());
975 
976         trace_qemu_savevm_send_postcopy_advise();
977         qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_ADVISE,
978                                  16, (uint8_t *)tmp);
979     } else {
980         qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_ADVISE, 0, NULL);
981     }
982 }
983 
984 /* Sent prior to starting the destination running in postcopy, discard pages
985  * that have already been sent but redirtied on the source.
986  * CMD_POSTCOPY_RAM_DISCARD consist of:
987  *      byte   version (0)
988  *      byte   Length of name field (not including 0)
989  *  n x byte   RAM block name
990  *      byte   0 terminator (just for safety)
991  *  n x        Byte ranges within the named RAMBlock
992  *      be64   Start of the range
993  *      be64   Length
994  *
995  *  name:  RAMBlock name that these entries are part of
996  *  len: Number of page entries
997  *  start_list: 'len' addresses
998  *  length_list: 'len' addresses
999  *
1000  */
1001 void qemu_savevm_send_postcopy_ram_discard(QEMUFile *f, const char *name,
1002                                            uint16_t len,
1003                                            uint64_t *start_list,
1004                                            uint64_t *length_list)
1005 {
1006     uint8_t *buf;
1007     uint16_t tmplen;
1008     uint16_t t;
1009     size_t name_len = strlen(name);
1010 
1011     trace_qemu_savevm_send_postcopy_ram_discard(name, len);
1012     assert(name_len < 256);
1013     buf = g_malloc0(1 + 1 + name_len + 1 + (8 + 8) * len);
1014     buf[0] = postcopy_ram_discard_version;
1015     buf[1] = name_len;
1016     memcpy(buf + 2, name, name_len);
1017     tmplen = 2 + name_len;
1018     buf[tmplen++] = '\0';
1019 
1020     for (t = 0; t < len; t++) {
1021         stq_be_p(buf + tmplen, start_list[t]);
1022         tmplen += 8;
1023         stq_be_p(buf + tmplen, length_list[t]);
1024         tmplen += 8;
1025     }
1026     qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_RAM_DISCARD, tmplen, buf);
1027     g_free(buf);
1028 }
1029 
1030 /* Get the destination into a state where it can receive postcopy data. */
1031 void qemu_savevm_send_postcopy_listen(QEMUFile *f)
1032 {
1033     trace_savevm_send_postcopy_listen();
1034     qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_LISTEN, 0, NULL);
1035 }
1036 
1037 /* Kick the destination into running */
1038 void qemu_savevm_send_postcopy_run(QEMUFile *f)
1039 {
1040     trace_savevm_send_postcopy_run();
1041     qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_RUN, 0, NULL);
1042 }
1043 
1044 void qemu_savevm_send_postcopy_resume(QEMUFile *f)
1045 {
1046     trace_savevm_send_postcopy_resume();
1047     qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_RESUME, 0, NULL);
1048 }
1049 
1050 void qemu_savevm_send_recv_bitmap(QEMUFile *f, char *block_name)
1051 {
1052     size_t len;
1053     char buf[256];
1054 
1055     trace_savevm_send_recv_bitmap(block_name);
1056 
1057     buf[0] = len = strlen(block_name);
1058     memcpy(buf + 1, block_name, len);
1059 
1060     qemu_savevm_command_send(f, MIG_CMD_RECV_BITMAP, len + 1, (uint8_t *)buf);
1061 }
1062 
1063 bool qemu_savevm_state_blocked(Error **errp)
1064 {
1065     SaveStateEntry *se;
1066 
1067     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1068         if (se->vmsd && se->vmsd->unmigratable) {
1069             error_setg(errp, "State blocked by non-migratable device '%s'",
1070                        se->idstr);
1071             return true;
1072         }
1073     }
1074     return false;
1075 }
1076 
1077 void qemu_savevm_state_header(QEMUFile *f)
1078 {
1079     trace_savevm_state_header();
1080     qemu_put_be32(f, QEMU_VM_FILE_MAGIC);
1081     qemu_put_be32(f, QEMU_VM_FILE_VERSION);
1082 
1083     if (migrate_get_current()->send_configuration) {
1084         qemu_put_byte(f, QEMU_VM_CONFIGURATION);
1085         vmstate_save_state(f, &vmstate_configuration, &savevm_state, 0);
1086     }
1087 }
1088 
1089 void qemu_savevm_state_setup(QEMUFile *f)
1090 {
1091     SaveStateEntry *se;
1092     Error *local_err = NULL;
1093     int ret;
1094 
1095     trace_savevm_state_setup();
1096     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1097         if (!se->ops || !se->ops->save_setup) {
1098             continue;
1099         }
1100         if (se->ops && se->ops->is_active) {
1101             if (!se->ops->is_active(se->opaque)) {
1102                 continue;
1103             }
1104         }
1105         save_section_header(f, se, QEMU_VM_SECTION_START);
1106 
1107         ret = se->ops->save_setup(f, se->opaque);
1108         save_section_footer(f, se);
1109         if (ret < 0) {
1110             qemu_file_set_error(f, ret);
1111             break;
1112         }
1113     }
1114 
1115     if (precopy_notify(PRECOPY_NOTIFY_SETUP, &local_err)) {
1116         error_report_err(local_err);
1117     }
1118 }
1119 
1120 int qemu_savevm_state_resume_prepare(MigrationState *s)
1121 {
1122     SaveStateEntry *se;
1123     int ret;
1124 
1125     trace_savevm_state_resume_prepare();
1126 
1127     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1128         if (!se->ops || !se->ops->resume_prepare) {
1129             continue;
1130         }
1131         if (se->ops && se->ops->is_active) {
1132             if (!se->ops->is_active(se->opaque)) {
1133                 continue;
1134             }
1135         }
1136         ret = se->ops->resume_prepare(s, se->opaque);
1137         if (ret < 0) {
1138             return ret;
1139         }
1140     }
1141 
1142     return 0;
1143 }
1144 
1145 /*
1146  * this function has three return values:
1147  *   negative: there was one error, and we have -errno.
1148  *   0 : We haven't finished, caller have to go again
1149  *   1 : We have finished, we can go to complete phase
1150  */
1151 int qemu_savevm_state_iterate(QEMUFile *f, bool postcopy)
1152 {
1153     SaveStateEntry *se;
1154     int ret = 1;
1155 
1156     trace_savevm_state_iterate();
1157     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1158         if (!se->ops || !se->ops->save_live_iterate) {
1159             continue;
1160         }
1161         if (se->ops->is_active &&
1162             !se->ops->is_active(se->opaque)) {
1163             continue;
1164         }
1165         if (se->ops->is_active_iterate &&
1166             !se->ops->is_active_iterate(se->opaque)) {
1167             continue;
1168         }
1169         /*
1170          * In the postcopy phase, any device that doesn't know how to
1171          * do postcopy should have saved it's state in the _complete
1172          * call that's already run, it might get confused if we call
1173          * iterate afterwards.
1174          */
1175         if (postcopy &&
1176             !(se->ops->has_postcopy && se->ops->has_postcopy(se->opaque))) {
1177             continue;
1178         }
1179         if (qemu_file_rate_limit(f)) {
1180             return 0;
1181         }
1182         trace_savevm_section_start(se->idstr, se->section_id);
1183 
1184         save_section_header(f, se, QEMU_VM_SECTION_PART);
1185 
1186         ret = se->ops->save_live_iterate(f, se->opaque);
1187         trace_savevm_section_end(se->idstr, se->section_id, ret);
1188         save_section_footer(f, se);
1189 
1190         if (ret < 0) {
1191             qemu_file_set_error(f, ret);
1192         }
1193         if (ret <= 0) {
1194             /* Do not proceed to the next vmstate before this one reported
1195                completion of the current stage. This serializes the migration
1196                and reduces the probability that a faster changing state is
1197                synchronized over and over again. */
1198             break;
1199         }
1200     }
1201     return ret;
1202 }
1203 
1204 static bool should_send_vmdesc(void)
1205 {
1206     MachineState *machine = MACHINE(qdev_get_machine());
1207     bool in_postcopy = migration_in_postcopy();
1208     return !machine->suppress_vmdesc && !in_postcopy;
1209 }
1210 
1211 /*
1212  * Calls the save_live_complete_postcopy methods
1213  * causing the last few pages to be sent immediately and doing any associated
1214  * cleanup.
1215  * Note postcopy also calls qemu_savevm_state_complete_precopy to complete
1216  * all the other devices, but that happens at the point we switch to postcopy.
1217  */
1218 void qemu_savevm_state_complete_postcopy(QEMUFile *f)
1219 {
1220     SaveStateEntry *se;
1221     int ret;
1222 
1223     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1224         if (!se->ops || !se->ops->save_live_complete_postcopy) {
1225             continue;
1226         }
1227         if (se->ops && se->ops->is_active) {
1228             if (!se->ops->is_active(se->opaque)) {
1229                 continue;
1230             }
1231         }
1232         trace_savevm_section_start(se->idstr, se->section_id);
1233         /* Section type */
1234         qemu_put_byte(f, QEMU_VM_SECTION_END);
1235         qemu_put_be32(f, se->section_id);
1236 
1237         ret = se->ops->save_live_complete_postcopy(f, se->opaque);
1238         trace_savevm_section_end(se->idstr, se->section_id, ret);
1239         save_section_footer(f, se);
1240         if (ret < 0) {
1241             qemu_file_set_error(f, ret);
1242             return;
1243         }
1244     }
1245 
1246     qemu_put_byte(f, QEMU_VM_EOF);
1247     qemu_fflush(f);
1248 }
1249 
1250 int qemu_savevm_state_complete_precopy(QEMUFile *f, bool iterable_only,
1251                                        bool inactivate_disks)
1252 {
1253     QJSON *vmdesc;
1254     int vmdesc_len;
1255     SaveStateEntry *se;
1256     int ret;
1257     bool in_postcopy = migration_in_postcopy();
1258     Error *local_err = NULL;
1259 
1260     if (precopy_notify(PRECOPY_NOTIFY_COMPLETE, &local_err)) {
1261         error_report_err(local_err);
1262     }
1263 
1264     trace_savevm_state_complete_precopy();
1265 
1266     cpu_synchronize_all_states();
1267 
1268     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1269         if (!se->ops ||
1270             (in_postcopy && se->ops->has_postcopy &&
1271              se->ops->has_postcopy(se->opaque)) ||
1272             (in_postcopy && !iterable_only) ||
1273             !se->ops->save_live_complete_precopy) {
1274             continue;
1275         }
1276 
1277         if (se->ops && se->ops->is_active) {
1278             if (!se->ops->is_active(se->opaque)) {
1279                 continue;
1280             }
1281         }
1282         trace_savevm_section_start(se->idstr, se->section_id);
1283 
1284         save_section_header(f, se, QEMU_VM_SECTION_END);
1285 
1286         ret = se->ops->save_live_complete_precopy(f, se->opaque);
1287         trace_savevm_section_end(se->idstr, se->section_id, ret);
1288         save_section_footer(f, se);
1289         if (ret < 0) {
1290             qemu_file_set_error(f, ret);
1291             return -1;
1292         }
1293     }
1294 
1295     if (iterable_only) {
1296         return 0;
1297     }
1298 
1299     vmdesc = qjson_new();
1300     json_prop_int(vmdesc, "page_size", qemu_target_page_size());
1301     json_start_array(vmdesc, "devices");
1302     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1303 
1304         if ((!se->ops || !se->ops->save_state) && !se->vmsd) {
1305             continue;
1306         }
1307         if (se->vmsd && !vmstate_save_needed(se->vmsd, se->opaque)) {
1308             trace_savevm_section_skip(se->idstr, se->section_id);
1309             continue;
1310         }
1311 
1312         trace_savevm_section_start(se->idstr, se->section_id);
1313 
1314         json_start_object(vmdesc, NULL);
1315         json_prop_str(vmdesc, "name", se->idstr);
1316         json_prop_int(vmdesc, "instance_id", se->instance_id);
1317 
1318         save_section_header(f, se, QEMU_VM_SECTION_FULL);
1319         ret = vmstate_save(f, se, vmdesc);
1320         if (ret) {
1321             qemu_file_set_error(f, ret);
1322             return ret;
1323         }
1324         trace_savevm_section_end(se->idstr, se->section_id, 0);
1325         save_section_footer(f, se);
1326 
1327         json_end_object(vmdesc);
1328     }
1329 
1330     if (inactivate_disks) {
1331         /* Inactivate before sending QEMU_VM_EOF so that the
1332          * bdrv_invalidate_cache_all() on the other end won't fail. */
1333         ret = bdrv_inactivate_all();
1334         if (ret) {
1335             error_report("%s: bdrv_inactivate_all() failed (%d)",
1336                          __func__, ret);
1337             qemu_file_set_error(f, ret);
1338             return ret;
1339         }
1340     }
1341     if (!in_postcopy) {
1342         /* Postcopy stream will still be going */
1343         qemu_put_byte(f, QEMU_VM_EOF);
1344     }
1345 
1346     json_end_array(vmdesc);
1347     qjson_finish(vmdesc);
1348     vmdesc_len = strlen(qjson_get_str(vmdesc));
1349 
1350     if (should_send_vmdesc()) {
1351         qemu_put_byte(f, QEMU_VM_VMDESCRIPTION);
1352         qemu_put_be32(f, vmdesc_len);
1353         qemu_put_buffer(f, (uint8_t *)qjson_get_str(vmdesc), vmdesc_len);
1354     }
1355     qjson_destroy(vmdesc);
1356 
1357     qemu_fflush(f);
1358     return 0;
1359 }
1360 
1361 /* Give an estimate of the amount left to be transferred,
1362  * the result is split into the amount for units that can and
1363  * for units that can't do postcopy.
1364  */
1365 void qemu_savevm_state_pending(QEMUFile *f, uint64_t threshold_size,
1366                                uint64_t *res_precopy_only,
1367                                uint64_t *res_compatible,
1368                                uint64_t *res_postcopy_only)
1369 {
1370     SaveStateEntry *se;
1371 
1372     *res_precopy_only = 0;
1373     *res_compatible = 0;
1374     *res_postcopy_only = 0;
1375 
1376 
1377     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1378         if (!se->ops || !se->ops->save_live_pending) {
1379             continue;
1380         }
1381         if (se->ops && se->ops->is_active) {
1382             if (!se->ops->is_active(se->opaque)) {
1383                 continue;
1384             }
1385         }
1386         se->ops->save_live_pending(f, se->opaque, threshold_size,
1387                                    res_precopy_only, res_compatible,
1388                                    res_postcopy_only);
1389     }
1390 }
1391 
1392 void qemu_savevm_state_cleanup(void)
1393 {
1394     SaveStateEntry *se;
1395     Error *local_err = NULL;
1396 
1397     if (precopy_notify(PRECOPY_NOTIFY_CLEANUP, &local_err)) {
1398         error_report_err(local_err);
1399     }
1400 
1401     trace_savevm_state_cleanup();
1402     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1403         if (se->ops && se->ops->save_cleanup) {
1404             se->ops->save_cleanup(se->opaque);
1405         }
1406     }
1407 }
1408 
1409 static int qemu_savevm_state(QEMUFile *f, Error **errp)
1410 {
1411     int ret;
1412     MigrationState *ms = migrate_get_current();
1413     MigrationStatus status;
1414 
1415     if (migration_is_setup_or_active(ms->state) ||
1416         ms->state == MIGRATION_STATUS_CANCELLING ||
1417         ms->state == MIGRATION_STATUS_COLO) {
1418         error_setg(errp, QERR_MIGRATION_ACTIVE);
1419         return -EINVAL;
1420     }
1421 
1422     if (migrate_use_block()) {
1423         error_setg(errp, "Block migration and snapshots are incompatible");
1424         return -EINVAL;
1425     }
1426 
1427     migrate_init(ms);
1428     ms->to_dst_file = f;
1429 
1430     qemu_mutex_unlock_iothread();
1431     qemu_savevm_state_header(f);
1432     qemu_savevm_state_setup(f);
1433     qemu_mutex_lock_iothread();
1434 
1435     while (qemu_file_get_error(f) == 0) {
1436         if (qemu_savevm_state_iterate(f, false) > 0) {
1437             break;
1438         }
1439     }
1440 
1441     ret = qemu_file_get_error(f);
1442     if (ret == 0) {
1443         qemu_savevm_state_complete_precopy(f, false, false);
1444         ret = qemu_file_get_error(f);
1445     }
1446     qemu_savevm_state_cleanup();
1447     if (ret != 0) {
1448         error_setg_errno(errp, -ret, "Error while writing VM state");
1449     }
1450 
1451     if (ret != 0) {
1452         status = MIGRATION_STATUS_FAILED;
1453     } else {
1454         status = MIGRATION_STATUS_COMPLETED;
1455     }
1456     migrate_set_state(&ms->state, MIGRATION_STATUS_SETUP, status);
1457 
1458     /* f is outer parameter, it should not stay in global migration state after
1459      * this function finished */
1460     ms->to_dst_file = NULL;
1461 
1462     return ret;
1463 }
1464 
1465 void qemu_savevm_live_state(QEMUFile *f)
1466 {
1467     /* save QEMU_VM_SECTION_END section */
1468     qemu_savevm_state_complete_precopy(f, true, false);
1469     qemu_put_byte(f, QEMU_VM_EOF);
1470 }
1471 
1472 int qemu_save_device_state(QEMUFile *f)
1473 {
1474     SaveStateEntry *se;
1475 
1476     if (!migration_in_colo_state()) {
1477         qemu_put_be32(f, QEMU_VM_FILE_MAGIC);
1478         qemu_put_be32(f, QEMU_VM_FILE_VERSION);
1479     }
1480     cpu_synchronize_all_states();
1481 
1482     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1483         int ret;
1484 
1485         if (se->is_ram) {
1486             continue;
1487         }
1488         if ((!se->ops || !se->ops->save_state) && !se->vmsd) {
1489             continue;
1490         }
1491         if (se->vmsd && !vmstate_save_needed(se->vmsd, se->opaque)) {
1492             continue;
1493         }
1494 
1495         save_section_header(f, se, QEMU_VM_SECTION_FULL);
1496 
1497         ret = vmstate_save(f, se, NULL);
1498         if (ret) {
1499             return ret;
1500         }
1501 
1502         save_section_footer(f, se);
1503     }
1504 
1505     qemu_put_byte(f, QEMU_VM_EOF);
1506 
1507     return qemu_file_get_error(f);
1508 }
1509 
1510 static SaveStateEntry *find_se(const char *idstr, int instance_id)
1511 {
1512     SaveStateEntry *se;
1513 
1514     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1515         if (!strcmp(se->idstr, idstr) &&
1516             (instance_id == se->instance_id ||
1517              instance_id == se->alias_id))
1518             return se;
1519         /* Migrating from an older version? */
1520         if (strstr(se->idstr, idstr) && se->compat) {
1521             if (!strcmp(se->compat->idstr, idstr) &&
1522                 (instance_id == se->compat->instance_id ||
1523                  instance_id == se->alias_id))
1524                 return se;
1525         }
1526     }
1527     return NULL;
1528 }
1529 
1530 enum LoadVMExitCodes {
1531     /* Allow a command to quit all layers of nested loadvm loops */
1532     LOADVM_QUIT     =  1,
1533 };
1534 
1535 /* ------ incoming postcopy messages ------ */
1536 /* 'advise' arrives before any transfers just to tell us that a postcopy
1537  * *might* happen - it might be skipped if precopy transferred everything
1538  * quickly.
1539  */
1540 static int loadvm_postcopy_handle_advise(MigrationIncomingState *mis,
1541                                          uint16_t len)
1542 {
1543     PostcopyState ps = postcopy_state_set(POSTCOPY_INCOMING_ADVISE);
1544     uint64_t remote_pagesize_summary, local_pagesize_summary, remote_tps;
1545     Error *local_err = NULL;
1546 
1547     trace_loadvm_postcopy_handle_advise();
1548     if (ps != POSTCOPY_INCOMING_NONE) {
1549         error_report("CMD_POSTCOPY_ADVISE in wrong postcopy state (%d)", ps);
1550         return -1;
1551     }
1552 
1553     switch (len) {
1554     case 0:
1555         if (migrate_postcopy_ram()) {
1556             error_report("RAM postcopy is enabled but have 0 byte advise");
1557             return -EINVAL;
1558         }
1559         return 0;
1560     case 8 + 8:
1561         if (!migrate_postcopy_ram()) {
1562             error_report("RAM postcopy is disabled but have 16 byte advise");
1563             return -EINVAL;
1564         }
1565         break;
1566     default:
1567         error_report("CMD_POSTCOPY_ADVISE invalid length (%d)", len);
1568         return -EINVAL;
1569     }
1570 
1571     if (!postcopy_ram_supported_by_host(mis)) {
1572         postcopy_state_set(POSTCOPY_INCOMING_NONE);
1573         return -1;
1574     }
1575 
1576     remote_pagesize_summary = qemu_get_be64(mis->from_src_file);
1577     local_pagesize_summary = ram_pagesize_summary();
1578 
1579     if (remote_pagesize_summary != local_pagesize_summary)  {
1580         /*
1581          * This detects two potential causes of mismatch:
1582          *   a) A mismatch in host page sizes
1583          *      Some combinations of mismatch are probably possible but it gets
1584          *      a bit more complicated.  In particular we need to place whole
1585          *      host pages on the dest at once, and we need to ensure that we
1586          *      handle dirtying to make sure we never end up sending part of
1587          *      a hostpage on it's own.
1588          *   b) The use of different huge page sizes on source/destination
1589          *      a more fine grain test is performed during RAM block migration
1590          *      but this test here causes a nice early clear failure, and
1591          *      also fails when passed to an older qemu that doesn't
1592          *      do huge pages.
1593          */
1594         error_report("Postcopy needs matching RAM page sizes (s=%" PRIx64
1595                                                              " d=%" PRIx64 ")",
1596                      remote_pagesize_summary, local_pagesize_summary);
1597         return -1;
1598     }
1599 
1600     remote_tps = qemu_get_be64(mis->from_src_file);
1601     if (remote_tps != qemu_target_page_size()) {
1602         /*
1603          * Again, some differences could be dealt with, but for now keep it
1604          * simple.
1605          */
1606         error_report("Postcopy needs matching target page sizes (s=%d d=%zd)",
1607                      (int)remote_tps, qemu_target_page_size());
1608         return -1;
1609     }
1610 
1611     if (postcopy_notify(POSTCOPY_NOTIFY_INBOUND_ADVISE, &local_err)) {
1612         error_report_err(local_err);
1613         return -1;
1614     }
1615 
1616     if (ram_postcopy_incoming_init(mis)) {
1617         return -1;
1618     }
1619 
1620     postcopy_state_set(POSTCOPY_INCOMING_ADVISE);
1621 
1622     return 0;
1623 }
1624 
1625 /* After postcopy we will be told to throw some pages away since they're
1626  * dirty and will have to be demand fetched.  Must happen before CPU is
1627  * started.
1628  * There can be 0..many of these messages, each encoding multiple pages.
1629  */
1630 static int loadvm_postcopy_ram_handle_discard(MigrationIncomingState *mis,
1631                                               uint16_t len)
1632 {
1633     int tmp;
1634     char ramid[256];
1635     PostcopyState ps = postcopy_state_get();
1636 
1637     trace_loadvm_postcopy_ram_handle_discard();
1638 
1639     switch (ps) {
1640     case POSTCOPY_INCOMING_ADVISE:
1641         /* 1st discard */
1642         tmp = postcopy_ram_prepare_discard(mis);
1643         if (tmp) {
1644             return tmp;
1645         }
1646         break;
1647 
1648     case POSTCOPY_INCOMING_DISCARD:
1649         /* Expected state */
1650         break;
1651 
1652     default:
1653         error_report("CMD_POSTCOPY_RAM_DISCARD in wrong postcopy state (%d)",
1654                      ps);
1655         return -1;
1656     }
1657     /* We're expecting a
1658      *    Version (0)
1659      *    a RAM ID string (length byte, name, 0 term)
1660      *    then at least 1 16 byte chunk
1661     */
1662     if (len < (1 + 1 + 1 + 1 + 2 * 8)) {
1663         error_report("CMD_POSTCOPY_RAM_DISCARD invalid length (%d)", len);
1664         return -1;
1665     }
1666 
1667     tmp = qemu_get_byte(mis->from_src_file);
1668     if (tmp != postcopy_ram_discard_version) {
1669         error_report("CMD_POSTCOPY_RAM_DISCARD invalid version (%d)", tmp);
1670         return -1;
1671     }
1672 
1673     if (!qemu_get_counted_string(mis->from_src_file, ramid)) {
1674         error_report("CMD_POSTCOPY_RAM_DISCARD Failed to read RAMBlock ID");
1675         return -1;
1676     }
1677     tmp = qemu_get_byte(mis->from_src_file);
1678     if (tmp != 0) {
1679         error_report("CMD_POSTCOPY_RAM_DISCARD missing nil (%d)", tmp);
1680         return -1;
1681     }
1682 
1683     len -= 3 + strlen(ramid);
1684     if (len % 16) {
1685         error_report("CMD_POSTCOPY_RAM_DISCARD invalid length (%d)", len);
1686         return -1;
1687     }
1688     trace_loadvm_postcopy_ram_handle_discard_header(ramid, len);
1689     while (len) {
1690         uint64_t start_addr, block_length;
1691         start_addr = qemu_get_be64(mis->from_src_file);
1692         block_length = qemu_get_be64(mis->from_src_file);
1693 
1694         len -= 16;
1695         int ret = ram_discard_range(ramid, start_addr, block_length);
1696         if (ret) {
1697             return ret;
1698         }
1699     }
1700     trace_loadvm_postcopy_ram_handle_discard_end();
1701 
1702     return 0;
1703 }
1704 
1705 /*
1706  * Triggered by a postcopy_listen command; this thread takes over reading
1707  * the input stream, leaving the main thread free to carry on loading the rest
1708  * of the device state (from RAM).
1709  * (TODO:This could do with being in a postcopy file - but there again it's
1710  * just another input loop, not that postcopy specific)
1711  */
1712 static void *postcopy_ram_listen_thread(void *opaque)
1713 {
1714     MigrationIncomingState *mis = migration_incoming_get_current();
1715     QEMUFile *f = mis->from_src_file;
1716     int load_res;
1717 
1718     migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE,
1719                                    MIGRATION_STATUS_POSTCOPY_ACTIVE);
1720     qemu_sem_post(&mis->listen_thread_sem);
1721     trace_postcopy_ram_listen_thread_start();
1722 
1723     rcu_register_thread();
1724     /*
1725      * Because we're a thread and not a coroutine we can't yield
1726      * in qemu_file, and thus we must be blocking now.
1727      */
1728     qemu_file_set_blocking(f, true);
1729     load_res = qemu_loadvm_state_main(f, mis);
1730 
1731     /*
1732      * This is tricky, but, mis->from_src_file can change after it
1733      * returns, when postcopy recovery happened. In the future, we may
1734      * want a wrapper for the QEMUFile handle.
1735      */
1736     f = mis->from_src_file;
1737 
1738     /* And non-blocking again so we don't block in any cleanup */
1739     qemu_file_set_blocking(f, false);
1740 
1741     trace_postcopy_ram_listen_thread_exit();
1742     if (load_res < 0) {
1743         error_report("%s: loadvm failed: %d", __func__, load_res);
1744         qemu_file_set_error(f, load_res);
1745         migrate_set_state(&mis->state, MIGRATION_STATUS_POSTCOPY_ACTIVE,
1746                                        MIGRATION_STATUS_FAILED);
1747     } else {
1748         /*
1749          * This looks good, but it's possible that the device loading in the
1750          * main thread hasn't finished yet, and so we might not be in 'RUN'
1751          * state yet; wait for the end of the main thread.
1752          */
1753         qemu_event_wait(&mis->main_thread_load_event);
1754     }
1755     postcopy_ram_incoming_cleanup(mis);
1756 
1757     if (load_res < 0) {
1758         /*
1759          * If something went wrong then we have a bad state so exit;
1760          * depending how far we got it might be possible at this point
1761          * to leave the guest running and fire MCEs for pages that never
1762          * arrived as a desperate recovery step.
1763          */
1764         rcu_unregister_thread();
1765         exit(EXIT_FAILURE);
1766     }
1767 
1768     migrate_set_state(&mis->state, MIGRATION_STATUS_POSTCOPY_ACTIVE,
1769                                    MIGRATION_STATUS_COMPLETED);
1770     /*
1771      * If everything has worked fine, then the main thread has waited
1772      * for us to start, and we're the last use of the mis.
1773      * (If something broke then qemu will have to exit anyway since it's
1774      * got a bad migration state).
1775      */
1776     migration_incoming_state_destroy();
1777     qemu_loadvm_state_cleanup();
1778 
1779     rcu_unregister_thread();
1780     mis->have_listen_thread = false;
1781     return NULL;
1782 }
1783 
1784 /* After this message we must be able to immediately receive postcopy data */
1785 static int loadvm_postcopy_handle_listen(MigrationIncomingState *mis)
1786 {
1787     PostcopyState ps = postcopy_state_set(POSTCOPY_INCOMING_LISTENING);
1788     trace_loadvm_postcopy_handle_listen();
1789     Error *local_err = NULL;
1790 
1791     if (ps != POSTCOPY_INCOMING_ADVISE && ps != POSTCOPY_INCOMING_DISCARD) {
1792         error_report("CMD_POSTCOPY_LISTEN in wrong postcopy state (%d)", ps);
1793         return -1;
1794     }
1795     if (ps == POSTCOPY_INCOMING_ADVISE) {
1796         /*
1797          * A rare case, we entered listen without having to do any discards,
1798          * so do the setup that's normally done at the time of the 1st discard.
1799          */
1800         if (migrate_postcopy_ram()) {
1801             postcopy_ram_prepare_discard(mis);
1802         }
1803     }
1804 
1805     /*
1806      * Sensitise RAM - can now generate requests for blocks that don't exist
1807      * However, at this point the CPU shouldn't be running, and the IO
1808      * shouldn't be doing anything yet so don't actually expect requests
1809      */
1810     if (migrate_postcopy_ram()) {
1811         if (postcopy_ram_enable_notify(mis)) {
1812             postcopy_ram_incoming_cleanup(mis);
1813             return -1;
1814         }
1815     }
1816 
1817     if (postcopy_notify(POSTCOPY_NOTIFY_INBOUND_LISTEN, &local_err)) {
1818         error_report_err(local_err);
1819         return -1;
1820     }
1821 
1822     if (mis->have_listen_thread) {
1823         error_report("CMD_POSTCOPY_RAM_LISTEN already has a listen thread");
1824         return -1;
1825     }
1826 
1827     mis->have_listen_thread = true;
1828     /* Start up the listening thread and wait for it to signal ready */
1829     qemu_sem_init(&mis->listen_thread_sem, 0);
1830     qemu_thread_create(&mis->listen_thread, "postcopy/listen",
1831                        postcopy_ram_listen_thread, NULL,
1832                        QEMU_THREAD_DETACHED);
1833     qemu_sem_wait(&mis->listen_thread_sem);
1834     qemu_sem_destroy(&mis->listen_thread_sem);
1835 
1836     return 0;
1837 }
1838 
1839 
1840 typedef struct {
1841     QEMUBH *bh;
1842 } HandleRunBhData;
1843 
1844 static void loadvm_postcopy_handle_run_bh(void *opaque)
1845 {
1846     Error *local_err = NULL;
1847     HandleRunBhData *data = opaque;
1848     MigrationIncomingState *mis = migration_incoming_get_current();
1849 
1850     /* TODO we should move all of this lot into postcopy_ram.c or a shared code
1851      * in migration.c
1852      */
1853     cpu_synchronize_all_post_init();
1854 
1855     qemu_announce_self(&mis->announce_timer, migrate_announce_params());
1856 
1857     /* Make sure all file formats flush their mutable metadata.
1858      * If we get an error here, just don't restart the VM yet. */
1859     bdrv_invalidate_cache_all(&local_err);
1860     if (local_err) {
1861         error_report_err(local_err);
1862         local_err = NULL;
1863         autostart = false;
1864     }
1865 
1866     trace_loadvm_postcopy_handle_run_cpu_sync();
1867 
1868     trace_loadvm_postcopy_handle_run_vmstart();
1869 
1870     dirty_bitmap_mig_before_vm_start();
1871 
1872     if (autostart) {
1873         /* Hold onto your hats, starting the CPU */
1874         vm_start();
1875     } else {
1876         /* leave it paused and let management decide when to start the CPU */
1877         runstate_set(RUN_STATE_PAUSED);
1878     }
1879 
1880     qemu_bh_delete(data->bh);
1881     g_free(data);
1882 }
1883 
1884 /* After all discards we can start running and asking for pages */
1885 static int loadvm_postcopy_handle_run(MigrationIncomingState *mis)
1886 {
1887     PostcopyState ps = postcopy_state_set(POSTCOPY_INCOMING_RUNNING);
1888     HandleRunBhData *data;
1889 
1890     trace_loadvm_postcopy_handle_run();
1891     if (ps != POSTCOPY_INCOMING_LISTENING) {
1892         error_report("CMD_POSTCOPY_RUN in wrong postcopy state (%d)", ps);
1893         return -1;
1894     }
1895 
1896     data = g_new(HandleRunBhData, 1);
1897     data->bh = qemu_bh_new(loadvm_postcopy_handle_run_bh, data);
1898     qemu_bh_schedule(data->bh);
1899 
1900     /* We need to finish reading the stream from the package
1901      * and also stop reading anything more from the stream that loaded the
1902      * package (since it's now being read by the listener thread).
1903      * LOADVM_QUIT will quit all the layers of nested loadvm loops.
1904      */
1905     return LOADVM_QUIT;
1906 }
1907 
1908 static int loadvm_postcopy_handle_resume(MigrationIncomingState *mis)
1909 {
1910     if (mis->state != MIGRATION_STATUS_POSTCOPY_RECOVER) {
1911         error_report("%s: illegal resume received", __func__);
1912         /* Don't fail the load, only for this. */
1913         return 0;
1914     }
1915 
1916     /*
1917      * This means source VM is ready to resume the postcopy migration.
1918      * It's time to switch state and release the fault thread to
1919      * continue service page faults.
1920      */
1921     migrate_set_state(&mis->state, MIGRATION_STATUS_POSTCOPY_RECOVER,
1922                       MIGRATION_STATUS_POSTCOPY_ACTIVE);
1923     qemu_sem_post(&mis->postcopy_pause_sem_fault);
1924 
1925     trace_loadvm_postcopy_handle_resume();
1926 
1927     /* Tell source that "we are ready" */
1928     migrate_send_rp_resume_ack(mis, MIGRATION_RESUME_ACK_VALUE);
1929 
1930     return 0;
1931 }
1932 
1933 /**
1934  * Immediately following this command is a blob of data containing an embedded
1935  * chunk of migration stream; read it and load it.
1936  *
1937  * @mis: Incoming state
1938  * @length: Length of packaged data to read
1939  *
1940  * Returns: Negative values on error
1941  *
1942  */
1943 static int loadvm_handle_cmd_packaged(MigrationIncomingState *mis)
1944 {
1945     int ret;
1946     size_t length;
1947     QIOChannelBuffer *bioc;
1948 
1949     length = qemu_get_be32(mis->from_src_file);
1950     trace_loadvm_handle_cmd_packaged(length);
1951 
1952     if (length > MAX_VM_CMD_PACKAGED_SIZE) {
1953         error_report("Unreasonably large packaged state: %zu", length);
1954         return -1;
1955     }
1956 
1957     bioc = qio_channel_buffer_new(length);
1958     qio_channel_set_name(QIO_CHANNEL(bioc), "migration-loadvm-buffer");
1959     ret = qemu_get_buffer(mis->from_src_file,
1960                           bioc->data,
1961                           length);
1962     if (ret != length) {
1963         object_unref(OBJECT(bioc));
1964         error_report("CMD_PACKAGED: Buffer receive fail ret=%d length=%zu",
1965                      ret, length);
1966         return (ret < 0) ? ret : -EAGAIN;
1967     }
1968     bioc->usage += length;
1969     trace_loadvm_handle_cmd_packaged_received(ret);
1970 
1971     QEMUFile *packf = qemu_fopen_channel_input(QIO_CHANNEL(bioc));
1972 
1973     ret = qemu_loadvm_state_main(packf, mis);
1974     trace_loadvm_handle_cmd_packaged_main(ret);
1975     qemu_fclose(packf);
1976     object_unref(OBJECT(bioc));
1977 
1978     return ret;
1979 }
1980 
1981 /*
1982  * Handle request that source requests for recved_bitmap on
1983  * destination. Payload format:
1984  *
1985  * len (1 byte) + ramblock_name (<255 bytes)
1986  */
1987 static int loadvm_handle_recv_bitmap(MigrationIncomingState *mis,
1988                                      uint16_t len)
1989 {
1990     QEMUFile *file = mis->from_src_file;
1991     RAMBlock *rb;
1992     char block_name[256];
1993     size_t cnt;
1994 
1995     cnt = qemu_get_counted_string(file, block_name);
1996     if (!cnt) {
1997         error_report("%s: failed to read block name", __func__);
1998         return -EINVAL;
1999     }
2000 
2001     /* Validate before using the data */
2002     if (qemu_file_get_error(file)) {
2003         return qemu_file_get_error(file);
2004     }
2005 
2006     if (len != cnt + 1) {
2007         error_report("%s: invalid payload length (%d)", __func__, len);
2008         return -EINVAL;
2009     }
2010 
2011     rb = qemu_ram_block_by_name(block_name);
2012     if (!rb) {
2013         error_report("%s: block '%s' not found", __func__, block_name);
2014         return -EINVAL;
2015     }
2016 
2017     migrate_send_rp_recv_bitmap(mis, block_name);
2018 
2019     trace_loadvm_handle_recv_bitmap(block_name);
2020 
2021     return 0;
2022 }
2023 
2024 static int loadvm_process_enable_colo(MigrationIncomingState *mis)
2025 {
2026     migration_incoming_enable_colo();
2027     return colo_init_ram_cache();
2028 }
2029 
2030 /*
2031  * Process an incoming 'QEMU_VM_COMMAND'
2032  * 0           just a normal return
2033  * LOADVM_QUIT All good, but exit the loop
2034  * <0          Error
2035  */
2036 static int loadvm_process_command(QEMUFile *f)
2037 {
2038     MigrationIncomingState *mis = migration_incoming_get_current();
2039     uint16_t cmd;
2040     uint16_t len;
2041     uint32_t tmp32;
2042 
2043     cmd = qemu_get_be16(f);
2044     len = qemu_get_be16(f);
2045 
2046     /* Check validity before continue processing of cmds */
2047     if (qemu_file_get_error(f)) {
2048         return qemu_file_get_error(f);
2049     }
2050 
2051     trace_loadvm_process_command(cmd, len);
2052     if (cmd >= MIG_CMD_MAX || cmd == MIG_CMD_INVALID) {
2053         error_report("MIG_CMD 0x%x unknown (len 0x%x)", cmd, len);
2054         return -EINVAL;
2055     }
2056 
2057     if (mig_cmd_args[cmd].len != -1 && mig_cmd_args[cmd].len != len) {
2058         error_report("%s received with bad length - expecting %zu, got %d",
2059                      mig_cmd_args[cmd].name,
2060                      (size_t)mig_cmd_args[cmd].len, len);
2061         return -ERANGE;
2062     }
2063 
2064     switch (cmd) {
2065     case MIG_CMD_OPEN_RETURN_PATH:
2066         if (mis->to_src_file) {
2067             error_report("CMD_OPEN_RETURN_PATH called when RP already open");
2068             /* Not really a problem, so don't give up */
2069             return 0;
2070         }
2071         mis->to_src_file = qemu_file_get_return_path(f);
2072         if (!mis->to_src_file) {
2073             error_report("CMD_OPEN_RETURN_PATH failed");
2074             return -1;
2075         }
2076         break;
2077 
2078     case MIG_CMD_PING:
2079         tmp32 = qemu_get_be32(f);
2080         trace_loadvm_process_command_ping(tmp32);
2081         if (!mis->to_src_file) {
2082             error_report("CMD_PING (0x%x) received with no return path",
2083                          tmp32);
2084             return -1;
2085         }
2086         migrate_send_rp_pong(mis, tmp32);
2087         break;
2088 
2089     case MIG_CMD_PACKAGED:
2090         return loadvm_handle_cmd_packaged(mis);
2091 
2092     case MIG_CMD_POSTCOPY_ADVISE:
2093         return loadvm_postcopy_handle_advise(mis, len);
2094 
2095     case MIG_CMD_POSTCOPY_LISTEN:
2096         return loadvm_postcopy_handle_listen(mis);
2097 
2098     case MIG_CMD_POSTCOPY_RUN:
2099         return loadvm_postcopy_handle_run(mis);
2100 
2101     case MIG_CMD_POSTCOPY_RAM_DISCARD:
2102         return loadvm_postcopy_ram_handle_discard(mis, len);
2103 
2104     case MIG_CMD_POSTCOPY_RESUME:
2105         return loadvm_postcopy_handle_resume(mis);
2106 
2107     case MIG_CMD_RECV_BITMAP:
2108         return loadvm_handle_recv_bitmap(mis, len);
2109 
2110     case MIG_CMD_ENABLE_COLO:
2111         return loadvm_process_enable_colo(mis);
2112     }
2113 
2114     return 0;
2115 }
2116 
2117 /*
2118  * Read a footer off the wire and check that it matches the expected section
2119  *
2120  * Returns: true if the footer was good
2121  *          false if there is a problem (and calls error_report to say why)
2122  */
2123 static bool check_section_footer(QEMUFile *f, SaveStateEntry *se)
2124 {
2125     int ret;
2126     uint8_t read_mark;
2127     uint32_t read_section_id;
2128 
2129     if (!migrate_get_current()->send_section_footer) {
2130         /* No footer to check */
2131         return true;
2132     }
2133 
2134     read_mark = qemu_get_byte(f);
2135 
2136     ret = qemu_file_get_error(f);
2137     if (ret) {
2138         error_report("%s: Read section footer failed: %d",
2139                      __func__, ret);
2140         return false;
2141     }
2142 
2143     if (read_mark != QEMU_VM_SECTION_FOOTER) {
2144         error_report("Missing section footer for %s", se->idstr);
2145         return false;
2146     }
2147 
2148     read_section_id = qemu_get_be32(f);
2149     if (read_section_id != se->load_section_id) {
2150         error_report("Mismatched section id in footer for %s -"
2151                      " read 0x%x expected 0x%x",
2152                      se->idstr, read_section_id, se->load_section_id);
2153         return false;
2154     }
2155 
2156     /* All good */
2157     return true;
2158 }
2159 
2160 static int
2161 qemu_loadvm_section_start_full(QEMUFile *f, MigrationIncomingState *mis)
2162 {
2163     uint32_t instance_id, version_id, section_id;
2164     SaveStateEntry *se;
2165     char idstr[256];
2166     int ret;
2167 
2168     /* Read section start */
2169     section_id = qemu_get_be32(f);
2170     if (!qemu_get_counted_string(f, idstr)) {
2171         error_report("Unable to read ID string for section %u",
2172                      section_id);
2173         return -EINVAL;
2174     }
2175     instance_id = qemu_get_be32(f);
2176     version_id = qemu_get_be32(f);
2177 
2178     ret = qemu_file_get_error(f);
2179     if (ret) {
2180         error_report("%s: Failed to read instance/version ID: %d",
2181                      __func__, ret);
2182         return ret;
2183     }
2184 
2185     trace_qemu_loadvm_state_section_startfull(section_id, idstr,
2186             instance_id, version_id);
2187     /* Find savevm section */
2188     se = find_se(idstr, instance_id);
2189     if (se == NULL) {
2190         error_report("Unknown savevm section or instance '%s' %d. "
2191                      "Make sure that your current VM setup matches your "
2192                      "saved VM setup, including any hotplugged devices",
2193                      idstr, instance_id);
2194         return -EINVAL;
2195     }
2196 
2197     /* Validate version */
2198     if (version_id > se->version_id) {
2199         error_report("savevm: unsupported version %d for '%s' v%d",
2200                      version_id, idstr, se->version_id);
2201         return -EINVAL;
2202     }
2203     se->load_version_id = version_id;
2204     se->load_section_id = section_id;
2205 
2206     /* Validate if it is a device's state */
2207     if (xen_enabled() && se->is_ram) {
2208         error_report("loadvm: %s RAM loading not allowed on Xen", idstr);
2209         return -EINVAL;
2210     }
2211 
2212     ret = vmstate_load(f, se);
2213     if (ret < 0) {
2214         error_report("error while loading state for instance 0x%x of"
2215                      " device '%s'", instance_id, idstr);
2216         return ret;
2217     }
2218     if (!check_section_footer(f, se)) {
2219         return -EINVAL;
2220     }
2221 
2222     return 0;
2223 }
2224 
2225 static int
2226 qemu_loadvm_section_part_end(QEMUFile *f, MigrationIncomingState *mis)
2227 {
2228     uint32_t section_id;
2229     SaveStateEntry *se;
2230     int ret;
2231 
2232     section_id = qemu_get_be32(f);
2233 
2234     ret = qemu_file_get_error(f);
2235     if (ret) {
2236         error_report("%s: Failed to read section ID: %d",
2237                      __func__, ret);
2238         return ret;
2239     }
2240 
2241     trace_qemu_loadvm_state_section_partend(section_id);
2242     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
2243         if (se->load_section_id == section_id) {
2244             break;
2245         }
2246     }
2247     if (se == NULL) {
2248         error_report("Unknown savevm section %d", section_id);
2249         return -EINVAL;
2250     }
2251 
2252     ret = vmstate_load(f, se);
2253     if (ret < 0) {
2254         error_report("error while loading state section id %d(%s)",
2255                      section_id, se->idstr);
2256         return ret;
2257     }
2258     if (!check_section_footer(f, se)) {
2259         return -EINVAL;
2260     }
2261 
2262     return 0;
2263 }
2264 
2265 static int qemu_loadvm_state_header(QEMUFile *f)
2266 {
2267     unsigned int v;
2268     int ret;
2269 
2270     v = qemu_get_be32(f);
2271     if (v != QEMU_VM_FILE_MAGIC) {
2272         error_report("Not a migration stream");
2273         return -EINVAL;
2274     }
2275 
2276     v = qemu_get_be32(f);
2277     if (v == QEMU_VM_FILE_VERSION_COMPAT) {
2278         error_report("SaveVM v2 format is obsolete and don't work anymore");
2279         return -ENOTSUP;
2280     }
2281     if (v != QEMU_VM_FILE_VERSION) {
2282         error_report("Unsupported migration stream version");
2283         return -ENOTSUP;
2284     }
2285 
2286     if (migrate_get_current()->send_configuration) {
2287         if (qemu_get_byte(f) != QEMU_VM_CONFIGURATION) {
2288             error_report("Configuration section missing");
2289             qemu_loadvm_state_cleanup();
2290             return -EINVAL;
2291         }
2292         ret = vmstate_load_state(f, &vmstate_configuration, &savevm_state, 0);
2293 
2294         if (ret) {
2295             qemu_loadvm_state_cleanup();
2296             return ret;
2297         }
2298     }
2299     return 0;
2300 }
2301 
2302 static int qemu_loadvm_state_setup(QEMUFile *f)
2303 {
2304     SaveStateEntry *se;
2305     int ret;
2306 
2307     trace_loadvm_state_setup();
2308     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
2309         if (!se->ops || !se->ops->load_setup) {
2310             continue;
2311         }
2312         if (se->ops && se->ops->is_active) {
2313             if (!se->ops->is_active(se->opaque)) {
2314                 continue;
2315             }
2316         }
2317 
2318         ret = se->ops->load_setup(f, se->opaque);
2319         if (ret < 0) {
2320             qemu_file_set_error(f, ret);
2321             error_report("Load state of device %s failed", se->idstr);
2322             return ret;
2323         }
2324     }
2325     return 0;
2326 }
2327 
2328 void qemu_loadvm_state_cleanup(void)
2329 {
2330     SaveStateEntry *se;
2331 
2332     trace_loadvm_state_cleanup();
2333     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
2334         if (se->ops && se->ops->load_cleanup) {
2335             se->ops->load_cleanup(se->opaque);
2336         }
2337     }
2338 }
2339 
2340 /* Return true if we should continue the migration, or false. */
2341 static bool postcopy_pause_incoming(MigrationIncomingState *mis)
2342 {
2343     trace_postcopy_pause_incoming();
2344 
2345     /* Clear the triggered bit to allow one recovery */
2346     mis->postcopy_recover_triggered = false;
2347 
2348     assert(mis->from_src_file);
2349     qemu_file_shutdown(mis->from_src_file);
2350     qemu_fclose(mis->from_src_file);
2351     mis->from_src_file = NULL;
2352 
2353     assert(mis->to_src_file);
2354     qemu_file_shutdown(mis->to_src_file);
2355     qemu_mutex_lock(&mis->rp_mutex);
2356     qemu_fclose(mis->to_src_file);
2357     mis->to_src_file = NULL;
2358     qemu_mutex_unlock(&mis->rp_mutex);
2359 
2360     migrate_set_state(&mis->state, MIGRATION_STATUS_POSTCOPY_ACTIVE,
2361                       MIGRATION_STATUS_POSTCOPY_PAUSED);
2362 
2363     /* Notify the fault thread for the invalidated file handle */
2364     postcopy_fault_thread_notify(mis);
2365 
2366     error_report("Detected IO failure for postcopy. "
2367                  "Migration paused.");
2368 
2369     while (mis->state == MIGRATION_STATUS_POSTCOPY_PAUSED) {
2370         qemu_sem_wait(&mis->postcopy_pause_sem_dst);
2371     }
2372 
2373     trace_postcopy_pause_incoming_continued();
2374 
2375     return true;
2376 }
2377 
2378 int qemu_loadvm_state_main(QEMUFile *f, MigrationIncomingState *mis)
2379 {
2380     uint8_t section_type;
2381     int ret = 0;
2382 
2383 retry:
2384     while (true) {
2385         section_type = qemu_get_byte(f);
2386 
2387         if (qemu_file_get_error(f)) {
2388             ret = qemu_file_get_error(f);
2389             break;
2390         }
2391 
2392         trace_qemu_loadvm_state_section(section_type);
2393         switch (section_type) {
2394         case QEMU_VM_SECTION_START:
2395         case QEMU_VM_SECTION_FULL:
2396             ret = qemu_loadvm_section_start_full(f, mis);
2397             if (ret < 0) {
2398                 goto out;
2399             }
2400             break;
2401         case QEMU_VM_SECTION_PART:
2402         case QEMU_VM_SECTION_END:
2403             ret = qemu_loadvm_section_part_end(f, mis);
2404             if (ret < 0) {
2405                 goto out;
2406             }
2407             break;
2408         case QEMU_VM_COMMAND:
2409             ret = loadvm_process_command(f);
2410             trace_qemu_loadvm_state_section_command(ret);
2411             if ((ret < 0) || (ret & LOADVM_QUIT)) {
2412                 goto out;
2413             }
2414             break;
2415         case QEMU_VM_EOF:
2416             /* This is the end of migration */
2417             goto out;
2418         default:
2419             error_report("Unknown savevm section type %d", section_type);
2420             ret = -EINVAL;
2421             goto out;
2422         }
2423     }
2424 
2425 out:
2426     if (ret < 0) {
2427         qemu_file_set_error(f, ret);
2428 
2429         /*
2430          * If we are during an active postcopy, then we pause instead
2431          * of bail out to at least keep the VM's dirty data.  Note
2432          * that POSTCOPY_INCOMING_LISTENING stage is still not enough,
2433          * during which we're still receiving device states and we
2434          * still haven't yet started the VM on destination.
2435          */
2436         if (postcopy_state_get() == POSTCOPY_INCOMING_RUNNING &&
2437             postcopy_pause_incoming(mis)) {
2438             /* Reset f to point to the newly created channel */
2439             f = mis->from_src_file;
2440             goto retry;
2441         }
2442     }
2443     return ret;
2444 }
2445 
2446 int qemu_loadvm_state(QEMUFile *f)
2447 {
2448     MigrationIncomingState *mis = migration_incoming_get_current();
2449     Error *local_err = NULL;
2450     int ret;
2451 
2452     if (qemu_savevm_state_blocked(&local_err)) {
2453         error_report_err(local_err);
2454         return -EINVAL;
2455     }
2456 
2457     ret = qemu_loadvm_state_header(f);
2458     if (ret) {
2459         return ret;
2460     }
2461 
2462     if (qemu_loadvm_state_setup(f) != 0) {
2463         return -EINVAL;
2464     }
2465 
2466     cpu_synchronize_all_pre_loadvm();
2467 
2468     ret = qemu_loadvm_state_main(f, mis);
2469     qemu_event_set(&mis->main_thread_load_event);
2470 
2471     trace_qemu_loadvm_state_post_main(ret);
2472 
2473     if (mis->have_listen_thread) {
2474         /* Listen thread still going, can't clean up yet */
2475         return ret;
2476     }
2477 
2478     if (ret == 0) {
2479         ret = qemu_file_get_error(f);
2480     }
2481 
2482     /*
2483      * Try to read in the VMDESC section as well, so that dumping tools that
2484      * intercept our migration stream have the chance to see it.
2485      */
2486 
2487     /* We've got to be careful; if we don't read the data and just shut the fd
2488      * then the sender can error if we close while it's still sending.
2489      * We also mustn't read data that isn't there; some transports (RDMA)
2490      * will stall waiting for that data when the source has already closed.
2491      */
2492     if (ret == 0 && should_send_vmdesc()) {
2493         uint8_t *buf;
2494         uint32_t size;
2495         uint8_t  section_type = qemu_get_byte(f);
2496 
2497         if (section_type != QEMU_VM_VMDESCRIPTION) {
2498             error_report("Expected vmdescription section, but got %d",
2499                          section_type);
2500             /*
2501              * It doesn't seem worth failing at this point since
2502              * we apparently have an otherwise valid VM state
2503              */
2504         } else {
2505             buf = g_malloc(0x1000);
2506             size = qemu_get_be32(f);
2507 
2508             while (size > 0) {
2509                 uint32_t read_chunk = MIN(size, 0x1000);
2510                 qemu_get_buffer(f, buf, read_chunk);
2511                 size -= read_chunk;
2512             }
2513             g_free(buf);
2514         }
2515     }
2516 
2517     qemu_loadvm_state_cleanup();
2518     cpu_synchronize_all_post_init();
2519 
2520     return ret;
2521 }
2522 
2523 int qemu_load_device_state(QEMUFile *f)
2524 {
2525     MigrationIncomingState *mis = migration_incoming_get_current();
2526     int ret;
2527 
2528     /* Load QEMU_VM_SECTION_FULL section */
2529     ret = qemu_loadvm_state_main(f, mis);
2530     if (ret < 0) {
2531         error_report("Failed to load device state: %d", ret);
2532         return ret;
2533     }
2534 
2535     cpu_synchronize_all_post_init();
2536     return 0;
2537 }
2538 
2539 int save_snapshot(const char *name, Error **errp)
2540 {
2541     BlockDriverState *bs, *bs1;
2542     QEMUSnapshotInfo sn1, *sn = &sn1, old_sn1, *old_sn = &old_sn1;
2543     int ret = -1;
2544     QEMUFile *f;
2545     int saved_vm_running;
2546     uint64_t vm_state_size;
2547     qemu_timeval tv;
2548     struct tm tm;
2549     AioContext *aio_context;
2550 
2551     if (migration_is_blocked(errp)) {
2552         return ret;
2553     }
2554 
2555     if (!replay_can_snapshot()) {
2556         error_setg(errp, "Record/replay does not allow making snapshot "
2557                    "right now. Try once more later.");
2558         return ret;
2559     }
2560 
2561     if (!bdrv_all_can_snapshot(&bs)) {
2562         error_setg(errp, "Device '%s' is writable but does not support "
2563                    "snapshots", bdrv_get_device_name(bs));
2564         return ret;
2565     }
2566 
2567     /* Delete old snapshots of the same name */
2568     if (name) {
2569         ret = bdrv_all_delete_snapshot(name, &bs1, errp);
2570         if (ret < 0) {
2571             error_prepend(errp, "Error while deleting snapshot on device "
2572                           "'%s': ", bdrv_get_device_name(bs1));
2573             return ret;
2574         }
2575     }
2576 
2577     bs = bdrv_all_find_vmstate_bs();
2578     if (bs == NULL) {
2579         error_setg(errp, "No block device can accept snapshots");
2580         return ret;
2581     }
2582     aio_context = bdrv_get_aio_context(bs);
2583 
2584     saved_vm_running = runstate_is_running();
2585 
2586     ret = global_state_store();
2587     if (ret) {
2588         error_setg(errp, "Error saving global state");
2589         return ret;
2590     }
2591     vm_stop(RUN_STATE_SAVE_VM);
2592 
2593     bdrv_drain_all_begin();
2594 
2595     aio_context_acquire(aio_context);
2596 
2597     memset(sn, 0, sizeof(*sn));
2598 
2599     /* fill auxiliary fields */
2600     qemu_gettimeofday(&tv);
2601     sn->date_sec = tv.tv_sec;
2602     sn->date_nsec = tv.tv_usec * 1000;
2603     sn->vm_clock_nsec = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL);
2604 
2605     if (name) {
2606         ret = bdrv_snapshot_find(bs, old_sn, name);
2607         if (ret >= 0) {
2608             pstrcpy(sn->name, sizeof(sn->name), old_sn->name);
2609             pstrcpy(sn->id_str, sizeof(sn->id_str), old_sn->id_str);
2610         } else {
2611             pstrcpy(sn->name, sizeof(sn->name), name);
2612         }
2613     } else {
2614         /* cast below needed for OpenBSD where tv_sec is still 'long' */
2615         localtime_r((const time_t *)&tv.tv_sec, &tm);
2616         strftime(sn->name, sizeof(sn->name), "vm-%Y%m%d%H%M%S", &tm);
2617     }
2618 
2619     /* save the VM state */
2620     f = qemu_fopen_bdrv(bs, 1);
2621     if (!f) {
2622         error_setg(errp, "Could not open VM state file");
2623         goto the_end;
2624     }
2625     ret = qemu_savevm_state(f, errp);
2626     vm_state_size = qemu_ftell(f);
2627     qemu_fclose(f);
2628     if (ret < 0) {
2629         goto the_end;
2630     }
2631 
2632     /* The bdrv_all_create_snapshot() call that follows acquires the AioContext
2633      * for itself.  BDRV_POLL_WHILE() does not support nested locking because
2634      * it only releases the lock once.  Therefore synchronous I/O will deadlock
2635      * unless we release the AioContext before bdrv_all_create_snapshot().
2636      */
2637     aio_context_release(aio_context);
2638     aio_context = NULL;
2639 
2640     ret = bdrv_all_create_snapshot(sn, bs, vm_state_size, &bs);
2641     if (ret < 0) {
2642         error_setg(errp, "Error while creating snapshot on '%s'",
2643                    bdrv_get_device_name(bs));
2644         goto the_end;
2645     }
2646 
2647     ret = 0;
2648 
2649  the_end:
2650     if (aio_context) {
2651         aio_context_release(aio_context);
2652     }
2653 
2654     bdrv_drain_all_end();
2655 
2656     if (saved_vm_running) {
2657         vm_start();
2658     }
2659     return ret;
2660 }
2661 
2662 void qmp_xen_save_devices_state(const char *filename, bool has_live, bool live,
2663                                 Error **errp)
2664 {
2665     QEMUFile *f;
2666     QIOChannelFile *ioc;
2667     int saved_vm_running;
2668     int ret;
2669 
2670     if (!has_live) {
2671         /* live default to true so old version of Xen tool stack can have a
2672          * successfull live migration */
2673         live = true;
2674     }
2675 
2676     saved_vm_running = runstate_is_running();
2677     vm_stop(RUN_STATE_SAVE_VM);
2678     global_state_store_running();
2679 
2680     ioc = qio_channel_file_new_path(filename, O_WRONLY | O_CREAT, 0660, errp);
2681     if (!ioc) {
2682         goto the_end;
2683     }
2684     qio_channel_set_name(QIO_CHANNEL(ioc), "migration-xen-save-state");
2685     f = qemu_fopen_channel_output(QIO_CHANNEL(ioc));
2686     object_unref(OBJECT(ioc));
2687     ret = qemu_save_device_state(f);
2688     if (ret < 0 || qemu_fclose(f) < 0) {
2689         error_setg(errp, QERR_IO_ERROR);
2690     } else {
2691         /* libxl calls the QMP command "stop" before calling
2692          * "xen-save-devices-state" and in case of migration failure, libxl
2693          * would call "cont".
2694          * So call bdrv_inactivate_all (release locks) here to let the other
2695          * side of the migration take controle of the images.
2696          */
2697         if (live && !saved_vm_running) {
2698             ret = bdrv_inactivate_all();
2699             if (ret) {
2700                 error_setg(errp, "%s: bdrv_inactivate_all() failed (%d)",
2701                            __func__, ret);
2702             }
2703         }
2704     }
2705 
2706  the_end:
2707     if (saved_vm_running) {
2708         vm_start();
2709     }
2710 }
2711 
2712 void qmp_xen_load_devices_state(const char *filename, Error **errp)
2713 {
2714     QEMUFile *f;
2715     QIOChannelFile *ioc;
2716     int ret;
2717 
2718     /* Guest must be paused before loading the device state; the RAM state
2719      * will already have been loaded by xc
2720      */
2721     if (runstate_is_running()) {
2722         error_setg(errp, "Cannot update device state while vm is running");
2723         return;
2724     }
2725     vm_stop(RUN_STATE_RESTORE_VM);
2726 
2727     ioc = qio_channel_file_new_path(filename, O_RDONLY | O_BINARY, 0, errp);
2728     if (!ioc) {
2729         return;
2730     }
2731     qio_channel_set_name(QIO_CHANNEL(ioc), "migration-xen-load-state");
2732     f = qemu_fopen_channel_input(QIO_CHANNEL(ioc));
2733     object_unref(OBJECT(ioc));
2734 
2735     ret = qemu_loadvm_state(f);
2736     qemu_fclose(f);
2737     if (ret < 0) {
2738         error_setg(errp, QERR_IO_ERROR);
2739     }
2740     migration_incoming_state_destroy();
2741 }
2742 
2743 int load_snapshot(const char *name, Error **errp)
2744 {
2745     BlockDriverState *bs, *bs_vm_state;
2746     QEMUSnapshotInfo sn;
2747     QEMUFile *f;
2748     int ret;
2749     AioContext *aio_context;
2750     MigrationIncomingState *mis = migration_incoming_get_current();
2751 
2752     if (!replay_can_snapshot()) {
2753         error_setg(errp, "Record/replay does not allow loading snapshot "
2754                    "right now. Try once more later.");
2755         return -EINVAL;
2756     }
2757 
2758     if (!bdrv_all_can_snapshot(&bs)) {
2759         error_setg(errp,
2760                    "Device '%s' is writable but does not support snapshots",
2761                    bdrv_get_device_name(bs));
2762         return -ENOTSUP;
2763     }
2764     ret = bdrv_all_find_snapshot(name, &bs);
2765     if (ret < 0) {
2766         error_setg(errp,
2767                    "Device '%s' does not have the requested snapshot '%s'",
2768                    bdrv_get_device_name(bs), name);
2769         return ret;
2770     }
2771 
2772     bs_vm_state = bdrv_all_find_vmstate_bs();
2773     if (!bs_vm_state) {
2774         error_setg(errp, "No block device supports snapshots");
2775         return -ENOTSUP;
2776     }
2777     aio_context = bdrv_get_aio_context(bs_vm_state);
2778 
2779     /* Don't even try to load empty VM states */
2780     aio_context_acquire(aio_context);
2781     ret = bdrv_snapshot_find(bs_vm_state, &sn, name);
2782     aio_context_release(aio_context);
2783     if (ret < 0) {
2784         return ret;
2785     } else if (sn.vm_state_size == 0) {
2786         error_setg(errp, "This is a disk-only snapshot. Revert to it "
2787                    " offline using qemu-img");
2788         return -EINVAL;
2789     }
2790 
2791     /* Flush all IO requests so they don't interfere with the new state.  */
2792     bdrv_drain_all_begin();
2793 
2794     ret = bdrv_all_goto_snapshot(name, &bs, errp);
2795     if (ret < 0) {
2796         error_prepend(errp, "Could not load snapshot '%s' on '%s': ",
2797                       name, bdrv_get_device_name(bs));
2798         goto err_drain;
2799     }
2800 
2801     /* restore the VM state */
2802     f = qemu_fopen_bdrv(bs_vm_state, 0);
2803     if (!f) {
2804         error_setg(errp, "Could not open VM state file");
2805         ret = -EINVAL;
2806         goto err_drain;
2807     }
2808 
2809     qemu_system_reset(SHUTDOWN_CAUSE_NONE);
2810     mis->from_src_file = f;
2811 
2812     aio_context_acquire(aio_context);
2813     ret = qemu_loadvm_state(f);
2814     migration_incoming_state_destroy();
2815     aio_context_release(aio_context);
2816 
2817     bdrv_drain_all_end();
2818 
2819     if (ret < 0) {
2820         error_setg(errp, "Error %d while loading VM state", ret);
2821         return ret;
2822     }
2823 
2824     return 0;
2825 
2826 err_drain:
2827     bdrv_drain_all_end();
2828     return ret;
2829 }
2830 
2831 void vmstate_register_ram(MemoryRegion *mr, DeviceState *dev)
2832 {
2833     qemu_ram_set_idstr(mr->ram_block,
2834                        memory_region_name(mr), dev);
2835     qemu_ram_set_migratable(mr->ram_block);
2836 }
2837 
2838 void vmstate_unregister_ram(MemoryRegion *mr, DeviceState *dev)
2839 {
2840     qemu_ram_unset_idstr(mr->ram_block);
2841     qemu_ram_unset_migratable(mr->ram_block);
2842 }
2843 
2844 void vmstate_register_ram_global(MemoryRegion *mr)
2845 {
2846     vmstate_register_ram(mr, NULL);
2847 }
2848 
2849 bool vmstate_check_only_migratable(const VMStateDescription *vmsd)
2850 {
2851     /* check needed if --only-migratable is specified */
2852     if (!only_migratable) {
2853         return true;
2854     }
2855 
2856     return !(vmsd && vmsd->unmigratable);
2857 }
2858