xref: /openbmc/qemu/migration/savevm.c (revision c51a3f5d)
1 /*
2  * QEMU System Emulator
3  *
4  * Copyright (c) 2003-2008 Fabrice Bellard
5  * Copyright (c) 2009-2015 Red Hat Inc
6  *
7  * Authors:
8  *  Juan Quintela <quintela@redhat.com>
9  *
10  * Permission is hereby granted, free of charge, to any person obtaining a copy
11  * of this software and associated documentation files (the "Software"), to deal
12  * in the Software without restriction, including without limitation the rights
13  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
14  * copies of the Software, and to permit persons to whom the Software is
15  * furnished to do so, subject to the following conditions:
16  *
17  * The above copyright notice and this permission notice shall be included in
18  * all copies or substantial portions of the Software.
19  *
20  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
21  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
22  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
23  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
24  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
25  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
26  * THE SOFTWARE.
27  */
28 
29 #include "qemu/osdep.h"
30 #include "hw/boards.h"
31 #include "net/net.h"
32 #include "migration.h"
33 #include "migration/snapshot.h"
34 #include "migration/vmstate.h"
35 #include "migration/misc.h"
36 #include "migration/register.h"
37 #include "migration/global_state.h"
38 #include "ram.h"
39 #include "qemu-file-channel.h"
40 #include "qemu-file.h"
41 #include "savevm.h"
42 #include "postcopy-ram.h"
43 #include "qapi/error.h"
44 #include "qapi/qapi-commands-migration.h"
45 #include "qapi/qapi-commands-misc.h"
46 #include "qapi/qmp/qerror.h"
47 #include "qemu/error-report.h"
48 #include "sysemu/cpus.h"
49 #include "exec/memory.h"
50 #include "exec/target_page.h"
51 #include "trace.h"
52 #include "qemu/iov.h"
53 #include "qemu/main-loop.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 "sysemu/runstate.h"
60 #include "sysemu/sysemu.h"
61 #include "sysemu/xen.h"
62 #include "qjson.h"
63 #include "migration/colo.h"
64 #include "qemu/bitmap.h"
65 #include "net/announce.h"
66 
67 const unsigned int postcopy_ram_discard_version = 0;
68 
69 /* Subcommands for QEMU_VM_COMMAND */
70 enum qemu_vm_cmd {
71     MIG_CMD_INVALID = 0,   /* Must be 0 */
72     MIG_CMD_OPEN_RETURN_PATH,  /* Tell the dest to open the Return path */
73     MIG_CMD_PING,              /* Request a PONG on the RP */
74 
75     MIG_CMD_POSTCOPY_ADVISE,       /* Prior to any page transfers, just
76                                       warn we might want to do PC */
77     MIG_CMD_POSTCOPY_LISTEN,       /* Start listening for incoming
78                                       pages as it's running. */
79     MIG_CMD_POSTCOPY_RUN,          /* Start execution */
80 
81     MIG_CMD_POSTCOPY_RAM_DISCARD,  /* A list of pages to discard that
82                                       were previously sent during
83                                       precopy but are dirty. */
84     MIG_CMD_PACKAGED,          /* Send a wrapped stream within this stream */
85     MIG_CMD_ENABLE_COLO,       /* Enable COLO */
86     MIG_CMD_POSTCOPY_RESUME,   /* resume postcopy on dest */
87     MIG_CMD_RECV_BITMAP,       /* Request for recved bitmap on dst */
88     MIG_CMD_MAX
89 };
90 
91 #define MAX_VM_CMD_PACKAGED_SIZE UINT32_MAX
92 static struct mig_cmd_args {
93     ssize_t     len; /* -1 = variable */
94     const char *name;
95 } mig_cmd_args[] = {
96     [MIG_CMD_INVALID]          = { .len = -1, .name = "INVALID" },
97     [MIG_CMD_OPEN_RETURN_PATH] = { .len =  0, .name = "OPEN_RETURN_PATH" },
98     [MIG_CMD_PING]             = { .len = sizeof(uint32_t), .name = "PING" },
99     [MIG_CMD_POSTCOPY_ADVISE]  = { .len = -1, .name = "POSTCOPY_ADVISE" },
100     [MIG_CMD_POSTCOPY_LISTEN]  = { .len =  0, .name = "POSTCOPY_LISTEN" },
101     [MIG_CMD_POSTCOPY_RUN]     = { .len =  0, .name = "POSTCOPY_RUN" },
102     [MIG_CMD_POSTCOPY_RAM_DISCARD] = {
103                                    .len = -1, .name = "POSTCOPY_RAM_DISCARD" },
104     [MIG_CMD_POSTCOPY_RESUME]  = { .len =  0, .name = "POSTCOPY_RESUME" },
105     [MIG_CMD_PACKAGED]         = { .len =  4, .name = "PACKAGED" },
106     [MIG_CMD_RECV_BITMAP]      = { .len = -1, .name = "RECV_BITMAP" },
107     [MIG_CMD_MAX]              = { .len = -1, .name = "MAX" },
108 };
109 
110 /* Note for MIG_CMD_POSTCOPY_ADVISE:
111  * The format of arguments is depending on postcopy mode:
112  * - postcopy RAM only
113  *   uint64_t host page size
114  *   uint64_t taget page size
115  *
116  * - postcopy RAM and postcopy dirty bitmaps
117  *   format is the same as for postcopy RAM only
118  *
119  * - postcopy dirty bitmaps only
120  *   Nothing. Command length field is 0.
121  *
122  * Be careful: adding a new postcopy entity with some other parameters should
123  * not break format self-description ability. Good way is to introduce some
124  * generic extendable format with an exception for two old entities.
125  */
126 
127 /***********************************************************/
128 /* savevm/loadvm support */
129 
130 static ssize_t block_writev_buffer(void *opaque, struct iovec *iov, int iovcnt,
131                                    int64_t pos, Error **errp)
132 {
133     int ret;
134     QEMUIOVector qiov;
135 
136     qemu_iovec_init_external(&qiov, iov, iovcnt);
137     ret = bdrv_writev_vmstate(opaque, &qiov, pos);
138     if (ret < 0) {
139         return ret;
140     }
141 
142     return qiov.size;
143 }
144 
145 static ssize_t block_get_buffer(void *opaque, uint8_t *buf, int64_t pos,
146                                 size_t size, Error **errp)
147 {
148     return bdrv_load_vmstate(opaque, buf, pos, size);
149 }
150 
151 static int bdrv_fclose(void *opaque, Error **errp)
152 {
153     return bdrv_flush(opaque);
154 }
155 
156 static const QEMUFileOps bdrv_read_ops = {
157     .get_buffer = block_get_buffer,
158     .close =      bdrv_fclose
159 };
160 
161 static const QEMUFileOps bdrv_write_ops = {
162     .writev_buffer  = block_writev_buffer,
163     .close          = bdrv_fclose
164 };
165 
166 static QEMUFile *qemu_fopen_bdrv(BlockDriverState *bs, int is_writable)
167 {
168     if (is_writable) {
169         return qemu_fopen_ops(bs, &bdrv_write_ops);
170     }
171     return qemu_fopen_ops(bs, &bdrv_read_ops);
172 }
173 
174 
175 /* QEMUFile timer support.
176  * Not in qemu-file.c to not add qemu-timer.c as dependency to qemu-file.c
177  */
178 
179 void timer_put(QEMUFile *f, QEMUTimer *ts)
180 {
181     uint64_t expire_time;
182 
183     expire_time = timer_expire_time_ns(ts);
184     qemu_put_be64(f, expire_time);
185 }
186 
187 void timer_get(QEMUFile *f, QEMUTimer *ts)
188 {
189     uint64_t expire_time;
190 
191     expire_time = qemu_get_be64(f);
192     if (expire_time != -1) {
193         timer_mod_ns(ts, expire_time);
194     } else {
195         timer_del(ts);
196     }
197 }
198 
199 
200 /* VMState timer support.
201  * Not in vmstate.c to not add qemu-timer.c as dependency to vmstate.c
202  */
203 
204 static int get_timer(QEMUFile *f, void *pv, size_t size,
205                      const VMStateField *field)
206 {
207     QEMUTimer *v = pv;
208     timer_get(f, v);
209     return 0;
210 }
211 
212 static int put_timer(QEMUFile *f, void *pv, size_t size,
213                      const VMStateField *field, QJSON *vmdesc)
214 {
215     QEMUTimer *v = pv;
216     timer_put(f, v);
217 
218     return 0;
219 }
220 
221 const VMStateInfo vmstate_info_timer = {
222     .name = "timer",
223     .get  = get_timer,
224     .put  = put_timer,
225 };
226 
227 
228 typedef struct CompatEntry {
229     char idstr[256];
230     int instance_id;
231 } CompatEntry;
232 
233 typedef struct SaveStateEntry {
234     QTAILQ_ENTRY(SaveStateEntry) entry;
235     char idstr[256];
236     uint32_t instance_id;
237     int alias_id;
238     int version_id;
239     /* version id read from the stream */
240     int load_version_id;
241     int section_id;
242     /* section id read from the stream */
243     int load_section_id;
244     const SaveVMHandlers *ops;
245     const VMStateDescription *vmsd;
246     void *opaque;
247     CompatEntry *compat;
248     int is_ram;
249 } SaveStateEntry;
250 
251 typedef struct SaveState {
252     QTAILQ_HEAD(, SaveStateEntry) handlers;
253     SaveStateEntry *handler_pri_head[MIG_PRI_MAX + 1];
254     int global_section_id;
255     uint32_t len;
256     const char *name;
257     uint32_t target_page_bits;
258     uint32_t caps_count;
259     MigrationCapability *capabilities;
260     QemuUUID uuid;
261 } SaveState;
262 
263 static SaveState savevm_state = {
264     .handlers = QTAILQ_HEAD_INITIALIZER(savevm_state.handlers),
265     .handler_pri_head = { [MIG_PRI_DEFAULT ... MIG_PRI_MAX] = NULL },
266     .global_section_id = 0,
267 };
268 
269 static bool should_validate_capability(int capability)
270 {
271     assert(capability >= 0 && capability < MIGRATION_CAPABILITY__MAX);
272     /* Validate only new capabilities to keep compatibility. */
273     switch (capability) {
274     case MIGRATION_CAPABILITY_X_IGNORE_SHARED:
275         return true;
276     default:
277         return false;
278     }
279 }
280 
281 static uint32_t get_validatable_capabilities_count(void)
282 {
283     MigrationState *s = migrate_get_current();
284     uint32_t result = 0;
285     int i;
286     for (i = 0; i < MIGRATION_CAPABILITY__MAX; i++) {
287         if (should_validate_capability(i) && s->enabled_capabilities[i]) {
288             result++;
289         }
290     }
291     return result;
292 }
293 
294 static int configuration_pre_save(void *opaque)
295 {
296     SaveState *state = opaque;
297     const char *current_name = MACHINE_GET_CLASS(current_machine)->name;
298     MigrationState *s = migrate_get_current();
299     int i, j;
300 
301     state->len = strlen(current_name);
302     state->name = current_name;
303     state->target_page_bits = qemu_target_page_bits();
304 
305     state->caps_count = get_validatable_capabilities_count();
306     state->capabilities = g_renew(MigrationCapability, state->capabilities,
307                                   state->caps_count);
308     for (i = j = 0; i < MIGRATION_CAPABILITY__MAX; i++) {
309         if (should_validate_capability(i) && s->enabled_capabilities[i]) {
310             state->capabilities[j++] = i;
311         }
312     }
313     state->uuid = qemu_uuid;
314 
315     return 0;
316 }
317 
318 static int configuration_pre_load(void *opaque)
319 {
320     SaveState *state = opaque;
321 
322     /* If there is no target-page-bits subsection it means the source
323      * predates the variable-target-page-bits support and is using the
324      * minimum possible value for this CPU.
325      */
326     state->target_page_bits = qemu_target_page_bits_min();
327     return 0;
328 }
329 
330 static bool configuration_validate_capabilities(SaveState *state)
331 {
332     bool ret = true;
333     MigrationState *s = migrate_get_current();
334     unsigned long *source_caps_bm;
335     int i;
336 
337     source_caps_bm = bitmap_new(MIGRATION_CAPABILITY__MAX);
338     for (i = 0; i < state->caps_count; i++) {
339         MigrationCapability capability = state->capabilities[i];
340         set_bit(capability, source_caps_bm);
341     }
342 
343     for (i = 0; i < MIGRATION_CAPABILITY__MAX; i++) {
344         bool source_state, target_state;
345         if (!should_validate_capability(i)) {
346             continue;
347         }
348         source_state = test_bit(i, source_caps_bm);
349         target_state = s->enabled_capabilities[i];
350         if (source_state != target_state) {
351             error_report("Capability %s is %s, but received capability is %s",
352                          MigrationCapability_str(i),
353                          target_state ? "on" : "off",
354                          source_state ? "on" : "off");
355             ret = false;
356             /* Don't break here to report all failed capabilities */
357         }
358     }
359 
360     g_free(source_caps_bm);
361     return ret;
362 }
363 
364 static int configuration_post_load(void *opaque, int version_id)
365 {
366     SaveState *state = opaque;
367     const char *current_name = MACHINE_GET_CLASS(current_machine)->name;
368 
369     if (strncmp(state->name, current_name, state->len) != 0) {
370         error_report("Machine type received is '%.*s' and local is '%s'",
371                      (int) state->len, state->name, current_name);
372         return -EINVAL;
373     }
374 
375     if (state->target_page_bits != qemu_target_page_bits()) {
376         error_report("Received TARGET_PAGE_BITS is %d but local is %d",
377                      state->target_page_bits, qemu_target_page_bits());
378         return -EINVAL;
379     }
380 
381     if (!configuration_validate_capabilities(state)) {
382         return -EINVAL;
383     }
384 
385     return 0;
386 }
387 
388 static int get_capability(QEMUFile *f, void *pv, size_t size,
389                           const VMStateField *field)
390 {
391     MigrationCapability *capability = pv;
392     char capability_str[UINT8_MAX + 1];
393     uint8_t len;
394     int i;
395 
396     len = qemu_get_byte(f);
397     qemu_get_buffer(f, (uint8_t *)capability_str, len);
398     capability_str[len] = '\0';
399     for (i = 0; i < MIGRATION_CAPABILITY__MAX; i++) {
400         if (!strcmp(MigrationCapability_str(i), capability_str)) {
401             *capability = i;
402             return 0;
403         }
404     }
405     error_report("Received unknown capability %s", capability_str);
406     return -EINVAL;
407 }
408 
409 static int put_capability(QEMUFile *f, void *pv, size_t size,
410                           const VMStateField *field, QJSON *vmdesc)
411 {
412     MigrationCapability *capability = pv;
413     const char *capability_str = MigrationCapability_str(*capability);
414     size_t len = strlen(capability_str);
415     assert(len <= UINT8_MAX);
416 
417     qemu_put_byte(f, len);
418     qemu_put_buffer(f, (uint8_t *)capability_str, len);
419     return 0;
420 }
421 
422 static const VMStateInfo vmstate_info_capability = {
423     .name = "capability",
424     .get  = get_capability,
425     .put  = put_capability,
426 };
427 
428 /* The target-page-bits subsection is present only if the
429  * target page size is not the same as the default (ie the
430  * minimum page size for a variable-page-size guest CPU).
431  * If it is present then it contains the actual target page
432  * bits for the machine, and migration will fail if the
433  * two ends don't agree about it.
434  */
435 static bool vmstate_target_page_bits_needed(void *opaque)
436 {
437     return qemu_target_page_bits()
438         > qemu_target_page_bits_min();
439 }
440 
441 static const VMStateDescription vmstate_target_page_bits = {
442     .name = "configuration/target-page-bits",
443     .version_id = 1,
444     .minimum_version_id = 1,
445     .needed = vmstate_target_page_bits_needed,
446     .fields = (VMStateField[]) {
447         VMSTATE_UINT32(target_page_bits, SaveState),
448         VMSTATE_END_OF_LIST()
449     }
450 };
451 
452 static bool vmstate_capabilites_needed(void *opaque)
453 {
454     return get_validatable_capabilities_count() > 0;
455 }
456 
457 static const VMStateDescription vmstate_capabilites = {
458     .name = "configuration/capabilities",
459     .version_id = 1,
460     .minimum_version_id = 1,
461     .needed = vmstate_capabilites_needed,
462     .fields = (VMStateField[]) {
463         VMSTATE_UINT32_V(caps_count, SaveState, 1),
464         VMSTATE_VARRAY_UINT32_ALLOC(capabilities, SaveState, caps_count, 1,
465                                     vmstate_info_capability,
466                                     MigrationCapability),
467         VMSTATE_END_OF_LIST()
468     }
469 };
470 
471 static bool vmstate_uuid_needed(void *opaque)
472 {
473     return qemu_uuid_set && migrate_validate_uuid();
474 }
475 
476 static int vmstate_uuid_post_load(void *opaque, int version_id)
477 {
478     SaveState *state = opaque;
479     char uuid_src[UUID_FMT_LEN + 1];
480     char uuid_dst[UUID_FMT_LEN + 1];
481 
482     if (!qemu_uuid_set) {
483         /*
484          * It's warning because user might not know UUID in some cases,
485          * e.g. load an old snapshot
486          */
487         qemu_uuid_unparse(&state->uuid, uuid_src);
488         warn_report("UUID is received %s, but local uuid isn't set",
489                      uuid_src);
490         return 0;
491     }
492     if (!qemu_uuid_is_equal(&state->uuid, &qemu_uuid)) {
493         qemu_uuid_unparse(&state->uuid, uuid_src);
494         qemu_uuid_unparse(&qemu_uuid, uuid_dst);
495         error_report("UUID received is %s and local is %s", uuid_src, uuid_dst);
496         return -EINVAL;
497     }
498     return 0;
499 }
500 
501 static const VMStateDescription vmstate_uuid = {
502     .name = "configuration/uuid",
503     .version_id = 1,
504     .minimum_version_id = 1,
505     .needed = vmstate_uuid_needed,
506     .post_load = vmstate_uuid_post_load,
507     .fields = (VMStateField[]) {
508         VMSTATE_UINT8_ARRAY_V(uuid.data, SaveState, sizeof(QemuUUID), 1),
509         VMSTATE_END_OF_LIST()
510     }
511 };
512 
513 static const VMStateDescription vmstate_configuration = {
514     .name = "configuration",
515     .version_id = 1,
516     .pre_load = configuration_pre_load,
517     .post_load = configuration_post_load,
518     .pre_save = configuration_pre_save,
519     .fields = (VMStateField[]) {
520         VMSTATE_UINT32(len, SaveState),
521         VMSTATE_VBUFFER_ALLOC_UINT32(name, SaveState, 0, NULL, len),
522         VMSTATE_END_OF_LIST()
523     },
524     .subsections = (const VMStateDescription*[]) {
525         &vmstate_target_page_bits,
526         &vmstate_capabilites,
527         &vmstate_uuid,
528         NULL
529     }
530 };
531 
532 static void dump_vmstate_vmsd(FILE *out_file,
533                               const VMStateDescription *vmsd, int indent,
534                               bool is_subsection);
535 
536 static void dump_vmstate_vmsf(FILE *out_file, const VMStateField *field,
537                               int indent)
538 {
539     fprintf(out_file, "%*s{\n", indent, "");
540     indent += 2;
541     fprintf(out_file, "%*s\"field\": \"%s\",\n", indent, "", field->name);
542     fprintf(out_file, "%*s\"version_id\": %d,\n", indent, "",
543             field->version_id);
544     fprintf(out_file, "%*s\"field_exists\": %s,\n", indent, "",
545             field->field_exists ? "true" : "false");
546     fprintf(out_file, "%*s\"size\": %zu", indent, "", field->size);
547     if (field->vmsd != NULL) {
548         fprintf(out_file, ",\n");
549         dump_vmstate_vmsd(out_file, field->vmsd, indent, false);
550     }
551     fprintf(out_file, "\n%*s}", indent - 2, "");
552 }
553 
554 static void dump_vmstate_vmss(FILE *out_file,
555                               const VMStateDescription **subsection,
556                               int indent)
557 {
558     if (*subsection != NULL) {
559         dump_vmstate_vmsd(out_file, *subsection, indent, true);
560     }
561 }
562 
563 static void dump_vmstate_vmsd(FILE *out_file,
564                               const VMStateDescription *vmsd, int indent,
565                               bool is_subsection)
566 {
567     if (is_subsection) {
568         fprintf(out_file, "%*s{\n", indent, "");
569     } else {
570         fprintf(out_file, "%*s\"%s\": {\n", indent, "", "Description");
571     }
572     indent += 2;
573     fprintf(out_file, "%*s\"name\": \"%s\",\n", indent, "", vmsd->name);
574     fprintf(out_file, "%*s\"version_id\": %d,\n", indent, "",
575             vmsd->version_id);
576     fprintf(out_file, "%*s\"minimum_version_id\": %d", indent, "",
577             vmsd->minimum_version_id);
578     if (vmsd->fields != NULL) {
579         const VMStateField *field = vmsd->fields;
580         bool first;
581 
582         fprintf(out_file, ",\n%*s\"Fields\": [\n", indent, "");
583         first = true;
584         while (field->name != NULL) {
585             if (field->flags & VMS_MUST_EXIST) {
586                 /* Ignore VMSTATE_VALIDATE bits; these don't get migrated */
587                 field++;
588                 continue;
589             }
590             if (!first) {
591                 fprintf(out_file, ",\n");
592             }
593             dump_vmstate_vmsf(out_file, field, indent + 2);
594             field++;
595             first = false;
596         }
597         fprintf(out_file, "\n%*s]", indent, "");
598     }
599     if (vmsd->subsections != NULL) {
600         const VMStateDescription **subsection = vmsd->subsections;
601         bool first;
602 
603         fprintf(out_file, ",\n%*s\"Subsections\": [\n", indent, "");
604         first = true;
605         while (*subsection != NULL) {
606             if (!first) {
607                 fprintf(out_file, ",\n");
608             }
609             dump_vmstate_vmss(out_file, subsection, indent + 2);
610             subsection++;
611             first = false;
612         }
613         fprintf(out_file, "\n%*s]", indent, "");
614     }
615     fprintf(out_file, "\n%*s}", indent - 2, "");
616 }
617 
618 static void dump_machine_type(FILE *out_file)
619 {
620     MachineClass *mc;
621 
622     mc = MACHINE_GET_CLASS(current_machine);
623 
624     fprintf(out_file, "  \"vmschkmachine\": {\n");
625     fprintf(out_file, "    \"Name\": \"%s\"\n", mc->name);
626     fprintf(out_file, "  },\n");
627 }
628 
629 void dump_vmstate_json_to_file(FILE *out_file)
630 {
631     GSList *list, *elt;
632     bool first;
633 
634     fprintf(out_file, "{\n");
635     dump_machine_type(out_file);
636 
637     first = true;
638     list = object_class_get_list(TYPE_DEVICE, true);
639     for (elt = list; elt; elt = elt->next) {
640         DeviceClass *dc = OBJECT_CLASS_CHECK(DeviceClass, elt->data,
641                                              TYPE_DEVICE);
642         const char *name;
643         int indent = 2;
644 
645         if (!dc->vmsd) {
646             continue;
647         }
648 
649         if (!first) {
650             fprintf(out_file, ",\n");
651         }
652         name = object_class_get_name(OBJECT_CLASS(dc));
653         fprintf(out_file, "%*s\"%s\": {\n", indent, "", name);
654         indent += 2;
655         fprintf(out_file, "%*s\"Name\": \"%s\",\n", indent, "", name);
656         fprintf(out_file, "%*s\"version_id\": %d,\n", indent, "",
657                 dc->vmsd->version_id);
658         fprintf(out_file, "%*s\"minimum_version_id\": %d,\n", indent, "",
659                 dc->vmsd->minimum_version_id);
660 
661         dump_vmstate_vmsd(out_file, dc->vmsd, indent, false);
662 
663         fprintf(out_file, "\n%*s}", indent - 2, "");
664         first = false;
665     }
666     fprintf(out_file, "\n}\n");
667     fclose(out_file);
668     g_slist_free(list);
669 }
670 
671 static uint32_t calculate_new_instance_id(const char *idstr)
672 {
673     SaveStateEntry *se;
674     uint32_t instance_id = 0;
675 
676     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
677         if (strcmp(idstr, se->idstr) == 0
678             && instance_id <= se->instance_id) {
679             instance_id = se->instance_id + 1;
680         }
681     }
682     /* Make sure we never loop over without being noticed */
683     assert(instance_id != VMSTATE_INSTANCE_ID_ANY);
684     return instance_id;
685 }
686 
687 static int calculate_compat_instance_id(const char *idstr)
688 {
689     SaveStateEntry *se;
690     int instance_id = 0;
691 
692     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
693         if (!se->compat) {
694             continue;
695         }
696 
697         if (strcmp(idstr, se->compat->idstr) == 0
698             && instance_id <= se->compat->instance_id) {
699             instance_id = se->compat->instance_id + 1;
700         }
701     }
702     return instance_id;
703 }
704 
705 static inline MigrationPriority save_state_priority(SaveStateEntry *se)
706 {
707     if (se->vmsd) {
708         return se->vmsd->priority;
709     }
710     return MIG_PRI_DEFAULT;
711 }
712 
713 static void savevm_state_handler_insert(SaveStateEntry *nse)
714 {
715     MigrationPriority priority = save_state_priority(nse);
716     SaveStateEntry *se;
717     int i;
718 
719     assert(priority <= MIG_PRI_MAX);
720 
721     for (i = priority - 1; i >= 0; i--) {
722         se = savevm_state.handler_pri_head[i];
723         if (se != NULL) {
724             assert(save_state_priority(se) < priority);
725             break;
726         }
727     }
728 
729     if (i >= 0) {
730         QTAILQ_INSERT_BEFORE(se, nse, entry);
731     } else {
732         QTAILQ_INSERT_TAIL(&savevm_state.handlers, nse, entry);
733     }
734 
735     if (savevm_state.handler_pri_head[priority] == NULL) {
736         savevm_state.handler_pri_head[priority] = nse;
737     }
738 }
739 
740 static void savevm_state_handler_remove(SaveStateEntry *se)
741 {
742     SaveStateEntry *next;
743     MigrationPriority priority = save_state_priority(se);
744 
745     if (se == savevm_state.handler_pri_head[priority]) {
746         next = QTAILQ_NEXT(se, entry);
747         if (next != NULL && save_state_priority(next) == priority) {
748             savevm_state.handler_pri_head[priority] = next;
749         } else {
750             savevm_state.handler_pri_head[priority] = NULL;
751         }
752     }
753     QTAILQ_REMOVE(&savevm_state.handlers, se, entry);
754 }
755 
756 /* TODO: Individual devices generally have very little idea about the rest
757    of the system, so instance_id should be removed/replaced.
758    Meanwhile pass -1 as instance_id if you do not already have a clearly
759    distinguishing id for all instances of your device class. */
760 int register_savevm_live(const char *idstr,
761                          uint32_t instance_id,
762                          int version_id,
763                          const SaveVMHandlers *ops,
764                          void *opaque)
765 {
766     SaveStateEntry *se;
767 
768     se = g_new0(SaveStateEntry, 1);
769     se->version_id = version_id;
770     se->section_id = savevm_state.global_section_id++;
771     se->ops = ops;
772     se->opaque = opaque;
773     se->vmsd = NULL;
774     /* if this is a live_savem then set is_ram */
775     if (ops->save_setup != NULL) {
776         se->is_ram = 1;
777     }
778 
779     pstrcat(se->idstr, sizeof(se->idstr), idstr);
780 
781     if (instance_id == VMSTATE_INSTANCE_ID_ANY) {
782         se->instance_id = calculate_new_instance_id(se->idstr);
783     } else {
784         se->instance_id = instance_id;
785     }
786     assert(!se->compat || se->instance_id == 0);
787     savevm_state_handler_insert(se);
788     return 0;
789 }
790 
791 void unregister_savevm(VMStateIf *obj, const char *idstr, void *opaque)
792 {
793     SaveStateEntry *se, *new_se;
794     char id[256] = "";
795 
796     if (obj) {
797         char *oid = vmstate_if_get_id(obj);
798         if (oid) {
799             pstrcpy(id, sizeof(id), oid);
800             pstrcat(id, sizeof(id), "/");
801             g_free(oid);
802         }
803     }
804     pstrcat(id, sizeof(id), idstr);
805 
806     QTAILQ_FOREACH_SAFE(se, &savevm_state.handlers, entry, new_se) {
807         if (strcmp(se->idstr, id) == 0 && se->opaque == opaque) {
808             savevm_state_handler_remove(se);
809             g_free(se->compat);
810             g_free(se);
811         }
812     }
813 }
814 
815 int vmstate_register_with_alias_id(VMStateIf *obj, uint32_t instance_id,
816                                    const VMStateDescription *vmsd,
817                                    void *opaque, int alias_id,
818                                    int required_for_version,
819                                    Error **errp)
820 {
821     SaveStateEntry *se;
822 
823     /* If this triggers, alias support can be dropped for the vmsd. */
824     assert(alias_id == -1 || required_for_version >= vmsd->minimum_version_id);
825 
826     se = g_new0(SaveStateEntry, 1);
827     se->version_id = vmsd->version_id;
828     se->section_id = savevm_state.global_section_id++;
829     se->opaque = opaque;
830     se->vmsd = vmsd;
831     se->alias_id = alias_id;
832 
833     if (obj) {
834         char *id = vmstate_if_get_id(obj);
835         if (id) {
836             if (snprintf(se->idstr, sizeof(se->idstr), "%s/", id) >=
837                 sizeof(se->idstr)) {
838                 error_setg(errp, "Path too long for VMState (%s)", id);
839                 g_free(id);
840                 g_free(se);
841 
842                 return -1;
843             }
844             g_free(id);
845 
846             se->compat = g_new0(CompatEntry, 1);
847             pstrcpy(se->compat->idstr, sizeof(se->compat->idstr), vmsd->name);
848             se->compat->instance_id = instance_id == VMSTATE_INSTANCE_ID_ANY ?
849                          calculate_compat_instance_id(vmsd->name) : instance_id;
850             instance_id = VMSTATE_INSTANCE_ID_ANY;
851         }
852     }
853     pstrcat(se->idstr, sizeof(se->idstr), vmsd->name);
854 
855     if (instance_id == VMSTATE_INSTANCE_ID_ANY) {
856         se->instance_id = calculate_new_instance_id(se->idstr);
857     } else {
858         se->instance_id = instance_id;
859     }
860     assert(!se->compat || se->instance_id == 0);
861     savevm_state_handler_insert(se);
862     return 0;
863 }
864 
865 void vmstate_unregister(VMStateIf *obj, const VMStateDescription *vmsd,
866                         void *opaque)
867 {
868     SaveStateEntry *se, *new_se;
869 
870     QTAILQ_FOREACH_SAFE(se, &savevm_state.handlers, entry, new_se) {
871         if (se->vmsd == vmsd && se->opaque == opaque) {
872             savevm_state_handler_remove(se);
873             g_free(se->compat);
874             g_free(se);
875         }
876     }
877 }
878 
879 static int vmstate_load(QEMUFile *f, SaveStateEntry *se)
880 {
881     trace_vmstate_load(se->idstr, se->vmsd ? se->vmsd->name : "(old)");
882     if (!se->vmsd) {         /* Old style */
883         return se->ops->load_state(f, se->opaque, se->load_version_id);
884     }
885     return vmstate_load_state(f, se->vmsd, se->opaque, se->load_version_id);
886 }
887 
888 static void vmstate_save_old_style(QEMUFile *f, SaveStateEntry *se, QJSON *vmdesc)
889 {
890     int64_t old_offset, size;
891 
892     old_offset = qemu_ftell_fast(f);
893     se->ops->save_state(f, se->opaque);
894     size = qemu_ftell_fast(f) - old_offset;
895 
896     if (vmdesc) {
897         json_prop_int(vmdesc, "size", size);
898         json_start_array(vmdesc, "fields");
899         json_start_object(vmdesc, NULL);
900         json_prop_str(vmdesc, "name", "data");
901         json_prop_int(vmdesc, "size", size);
902         json_prop_str(vmdesc, "type", "buffer");
903         json_end_object(vmdesc);
904         json_end_array(vmdesc);
905     }
906 }
907 
908 static int vmstate_save(QEMUFile *f, SaveStateEntry *se, QJSON *vmdesc)
909 {
910     trace_vmstate_save(se->idstr, se->vmsd ? se->vmsd->name : "(old)");
911     if (!se->vmsd) {
912         vmstate_save_old_style(f, se, vmdesc);
913         return 0;
914     }
915     return vmstate_save_state(f, se->vmsd, se->opaque, vmdesc);
916 }
917 
918 /*
919  * Write the header for device section (QEMU_VM_SECTION START/END/PART/FULL)
920  */
921 static void save_section_header(QEMUFile *f, SaveStateEntry *se,
922                                 uint8_t section_type)
923 {
924     qemu_put_byte(f, section_type);
925     qemu_put_be32(f, se->section_id);
926 
927     if (section_type == QEMU_VM_SECTION_FULL ||
928         section_type == QEMU_VM_SECTION_START) {
929         /* ID string */
930         size_t len = strlen(se->idstr);
931         qemu_put_byte(f, len);
932         qemu_put_buffer(f, (uint8_t *)se->idstr, len);
933 
934         qemu_put_be32(f, se->instance_id);
935         qemu_put_be32(f, se->version_id);
936     }
937 }
938 
939 /*
940  * Write a footer onto device sections that catches cases misformatted device
941  * sections.
942  */
943 static void save_section_footer(QEMUFile *f, SaveStateEntry *se)
944 {
945     if (migrate_get_current()->send_section_footer) {
946         qemu_put_byte(f, QEMU_VM_SECTION_FOOTER);
947         qemu_put_be32(f, se->section_id);
948     }
949 }
950 
951 /**
952  * qemu_savevm_command_send: Send a 'QEMU_VM_COMMAND' type element with the
953  *                           command and associated data.
954  *
955  * @f: File to send command on
956  * @command: Command type to send
957  * @len: Length of associated data
958  * @data: Data associated with command.
959  */
960 static void qemu_savevm_command_send(QEMUFile *f,
961                                      enum qemu_vm_cmd command,
962                                      uint16_t len,
963                                      uint8_t *data)
964 {
965     trace_savevm_command_send(command, len);
966     qemu_put_byte(f, QEMU_VM_COMMAND);
967     qemu_put_be16(f, (uint16_t)command);
968     qemu_put_be16(f, len);
969     qemu_put_buffer(f, data, len);
970     qemu_fflush(f);
971 }
972 
973 void qemu_savevm_send_colo_enable(QEMUFile *f)
974 {
975     trace_savevm_send_colo_enable();
976     qemu_savevm_command_send(f, MIG_CMD_ENABLE_COLO, 0, NULL);
977 }
978 
979 void qemu_savevm_send_ping(QEMUFile *f, uint32_t value)
980 {
981     uint32_t buf;
982 
983     trace_savevm_send_ping(value);
984     buf = cpu_to_be32(value);
985     qemu_savevm_command_send(f, MIG_CMD_PING, sizeof(value), (uint8_t *)&buf);
986 }
987 
988 void qemu_savevm_send_open_return_path(QEMUFile *f)
989 {
990     trace_savevm_send_open_return_path();
991     qemu_savevm_command_send(f, MIG_CMD_OPEN_RETURN_PATH, 0, NULL);
992 }
993 
994 /* We have a buffer of data to send; we don't want that all to be loaded
995  * by the command itself, so the command contains just the length of the
996  * extra buffer that we then send straight after it.
997  * TODO: Must be a better way to organise that
998  *
999  * Returns:
1000  *    0 on success
1001  *    -ve on error
1002  */
1003 int qemu_savevm_send_packaged(QEMUFile *f, const uint8_t *buf, size_t len)
1004 {
1005     uint32_t tmp;
1006 
1007     if (len > MAX_VM_CMD_PACKAGED_SIZE) {
1008         error_report("%s: Unreasonably large packaged state: %zu",
1009                      __func__, len);
1010         return -1;
1011     }
1012 
1013     tmp = cpu_to_be32(len);
1014 
1015     trace_qemu_savevm_send_packaged();
1016     qemu_savevm_command_send(f, MIG_CMD_PACKAGED, 4, (uint8_t *)&tmp);
1017 
1018     qemu_put_buffer(f, buf, len);
1019 
1020     return 0;
1021 }
1022 
1023 /* Send prior to any postcopy transfer */
1024 void qemu_savevm_send_postcopy_advise(QEMUFile *f)
1025 {
1026     if (migrate_postcopy_ram()) {
1027         uint64_t tmp[2];
1028         tmp[0] = cpu_to_be64(ram_pagesize_summary());
1029         tmp[1] = cpu_to_be64(qemu_target_page_size());
1030 
1031         trace_qemu_savevm_send_postcopy_advise();
1032         qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_ADVISE,
1033                                  16, (uint8_t *)tmp);
1034     } else {
1035         qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_ADVISE, 0, NULL);
1036     }
1037 }
1038 
1039 /* Sent prior to starting the destination running in postcopy, discard pages
1040  * that have already been sent but redirtied on the source.
1041  * CMD_POSTCOPY_RAM_DISCARD consist of:
1042  *      byte   version (0)
1043  *      byte   Length of name field (not including 0)
1044  *  n x byte   RAM block name
1045  *      byte   0 terminator (just for safety)
1046  *  n x        Byte ranges within the named RAMBlock
1047  *      be64   Start of the range
1048  *      be64   Length
1049  *
1050  *  name:  RAMBlock name that these entries are part of
1051  *  len: Number of page entries
1052  *  start_list: 'len' addresses
1053  *  length_list: 'len' addresses
1054  *
1055  */
1056 void qemu_savevm_send_postcopy_ram_discard(QEMUFile *f, const char *name,
1057                                            uint16_t len,
1058                                            uint64_t *start_list,
1059                                            uint64_t *length_list)
1060 {
1061     uint8_t *buf;
1062     uint16_t tmplen;
1063     uint16_t t;
1064     size_t name_len = strlen(name);
1065 
1066     trace_qemu_savevm_send_postcopy_ram_discard(name, len);
1067     assert(name_len < 256);
1068     buf = g_malloc0(1 + 1 + name_len + 1 + (8 + 8) * len);
1069     buf[0] = postcopy_ram_discard_version;
1070     buf[1] = name_len;
1071     memcpy(buf + 2, name, name_len);
1072     tmplen = 2 + name_len;
1073     buf[tmplen++] = '\0';
1074 
1075     for (t = 0; t < len; t++) {
1076         stq_be_p(buf + tmplen, start_list[t]);
1077         tmplen += 8;
1078         stq_be_p(buf + tmplen, length_list[t]);
1079         tmplen += 8;
1080     }
1081     qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_RAM_DISCARD, tmplen, buf);
1082     g_free(buf);
1083 }
1084 
1085 /* Get the destination into a state where it can receive postcopy data. */
1086 void qemu_savevm_send_postcopy_listen(QEMUFile *f)
1087 {
1088     trace_savevm_send_postcopy_listen();
1089     qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_LISTEN, 0, NULL);
1090 }
1091 
1092 /* Kick the destination into running */
1093 void qemu_savevm_send_postcopy_run(QEMUFile *f)
1094 {
1095     trace_savevm_send_postcopy_run();
1096     qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_RUN, 0, NULL);
1097 }
1098 
1099 void qemu_savevm_send_postcopy_resume(QEMUFile *f)
1100 {
1101     trace_savevm_send_postcopy_resume();
1102     qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_RESUME, 0, NULL);
1103 }
1104 
1105 void qemu_savevm_send_recv_bitmap(QEMUFile *f, char *block_name)
1106 {
1107     size_t len;
1108     char buf[256];
1109 
1110     trace_savevm_send_recv_bitmap(block_name);
1111 
1112     buf[0] = len = strlen(block_name);
1113     memcpy(buf + 1, block_name, len);
1114 
1115     qemu_savevm_command_send(f, MIG_CMD_RECV_BITMAP, len + 1, (uint8_t *)buf);
1116 }
1117 
1118 bool qemu_savevm_state_blocked(Error **errp)
1119 {
1120     SaveStateEntry *se;
1121 
1122     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1123         if (se->vmsd && se->vmsd->unmigratable) {
1124             error_setg(errp, "State blocked by non-migratable device '%s'",
1125                        se->idstr);
1126             return true;
1127         }
1128     }
1129     return false;
1130 }
1131 
1132 void qemu_savevm_state_header(QEMUFile *f)
1133 {
1134     trace_savevm_state_header();
1135     qemu_put_be32(f, QEMU_VM_FILE_MAGIC);
1136     qemu_put_be32(f, QEMU_VM_FILE_VERSION);
1137 
1138     if (migrate_get_current()->send_configuration) {
1139         qemu_put_byte(f, QEMU_VM_CONFIGURATION);
1140         vmstate_save_state(f, &vmstate_configuration, &savevm_state, 0);
1141     }
1142 }
1143 
1144 bool qemu_savevm_state_guest_unplug_pending(void)
1145 {
1146     SaveStateEntry *se;
1147 
1148     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1149         if (se->vmsd && se->vmsd->dev_unplug_pending &&
1150             se->vmsd->dev_unplug_pending(se->opaque)) {
1151             return true;
1152         }
1153     }
1154 
1155     return false;
1156 }
1157 
1158 void qemu_savevm_state_setup(QEMUFile *f)
1159 {
1160     SaveStateEntry *se;
1161     Error *local_err = NULL;
1162     int ret;
1163 
1164     trace_savevm_state_setup();
1165     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1166         if (!se->ops || !se->ops->save_setup) {
1167             continue;
1168         }
1169         if (se->ops->is_active) {
1170             if (!se->ops->is_active(se->opaque)) {
1171                 continue;
1172             }
1173         }
1174         save_section_header(f, se, QEMU_VM_SECTION_START);
1175 
1176         ret = se->ops->save_setup(f, se->opaque);
1177         save_section_footer(f, se);
1178         if (ret < 0) {
1179             qemu_file_set_error(f, ret);
1180             break;
1181         }
1182     }
1183 
1184     if (precopy_notify(PRECOPY_NOTIFY_SETUP, &local_err)) {
1185         error_report_err(local_err);
1186     }
1187 }
1188 
1189 int qemu_savevm_state_resume_prepare(MigrationState *s)
1190 {
1191     SaveStateEntry *se;
1192     int ret;
1193 
1194     trace_savevm_state_resume_prepare();
1195 
1196     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1197         if (!se->ops || !se->ops->resume_prepare) {
1198             continue;
1199         }
1200         if (se->ops->is_active) {
1201             if (!se->ops->is_active(se->opaque)) {
1202                 continue;
1203             }
1204         }
1205         ret = se->ops->resume_prepare(s, se->opaque);
1206         if (ret < 0) {
1207             return ret;
1208         }
1209     }
1210 
1211     return 0;
1212 }
1213 
1214 /*
1215  * this function has three return values:
1216  *   negative: there was one error, and we have -errno.
1217  *   0 : We haven't finished, caller have to go again
1218  *   1 : We have finished, we can go to complete phase
1219  */
1220 int qemu_savevm_state_iterate(QEMUFile *f, bool postcopy)
1221 {
1222     SaveStateEntry *se;
1223     int ret = 1;
1224 
1225     trace_savevm_state_iterate();
1226     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1227         if (!se->ops || !se->ops->save_live_iterate) {
1228             continue;
1229         }
1230         if (se->ops->is_active &&
1231             !se->ops->is_active(se->opaque)) {
1232             continue;
1233         }
1234         if (se->ops->is_active_iterate &&
1235             !se->ops->is_active_iterate(se->opaque)) {
1236             continue;
1237         }
1238         /*
1239          * In the postcopy phase, any device that doesn't know how to
1240          * do postcopy should have saved it's state in the _complete
1241          * call that's already run, it might get confused if we call
1242          * iterate afterwards.
1243          */
1244         if (postcopy &&
1245             !(se->ops->has_postcopy && se->ops->has_postcopy(se->opaque))) {
1246             continue;
1247         }
1248         if (qemu_file_rate_limit(f)) {
1249             return 0;
1250         }
1251         trace_savevm_section_start(se->idstr, se->section_id);
1252 
1253         save_section_header(f, se, QEMU_VM_SECTION_PART);
1254 
1255         ret = se->ops->save_live_iterate(f, se->opaque);
1256         trace_savevm_section_end(se->idstr, se->section_id, ret);
1257         save_section_footer(f, se);
1258 
1259         if (ret < 0) {
1260             error_report("failed to save SaveStateEntry with id(name): %d(%s)",
1261                          se->section_id, se->idstr);
1262             qemu_file_set_error(f, ret);
1263         }
1264         if (ret <= 0) {
1265             /* Do not proceed to the next vmstate before this one reported
1266                completion of the current stage. This serializes the migration
1267                and reduces the probability that a faster changing state is
1268                synchronized over and over again. */
1269             break;
1270         }
1271     }
1272     return ret;
1273 }
1274 
1275 static bool should_send_vmdesc(void)
1276 {
1277     MachineState *machine = MACHINE(qdev_get_machine());
1278     bool in_postcopy = migration_in_postcopy();
1279     return !machine->suppress_vmdesc && !in_postcopy;
1280 }
1281 
1282 /*
1283  * Calls the save_live_complete_postcopy methods
1284  * causing the last few pages to be sent immediately and doing any associated
1285  * cleanup.
1286  * Note postcopy also calls qemu_savevm_state_complete_precopy to complete
1287  * all the other devices, but that happens at the point we switch to postcopy.
1288  */
1289 void qemu_savevm_state_complete_postcopy(QEMUFile *f)
1290 {
1291     SaveStateEntry *se;
1292     int ret;
1293 
1294     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1295         if (!se->ops || !se->ops->save_live_complete_postcopy) {
1296             continue;
1297         }
1298         if (se->ops->is_active) {
1299             if (!se->ops->is_active(se->opaque)) {
1300                 continue;
1301             }
1302         }
1303         trace_savevm_section_start(se->idstr, se->section_id);
1304         /* Section type */
1305         qemu_put_byte(f, QEMU_VM_SECTION_END);
1306         qemu_put_be32(f, se->section_id);
1307 
1308         ret = se->ops->save_live_complete_postcopy(f, se->opaque);
1309         trace_savevm_section_end(se->idstr, se->section_id, ret);
1310         save_section_footer(f, se);
1311         if (ret < 0) {
1312             qemu_file_set_error(f, ret);
1313             return;
1314         }
1315     }
1316 
1317     qemu_put_byte(f, QEMU_VM_EOF);
1318     qemu_fflush(f);
1319 }
1320 
1321 static
1322 int qemu_savevm_state_complete_precopy_iterable(QEMUFile *f, bool in_postcopy)
1323 {
1324     SaveStateEntry *se;
1325     int ret;
1326 
1327     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1328         if (!se->ops ||
1329             (in_postcopy && se->ops->has_postcopy &&
1330              se->ops->has_postcopy(se->opaque)) ||
1331             !se->ops->save_live_complete_precopy) {
1332             continue;
1333         }
1334 
1335         if (se->ops->is_active) {
1336             if (!se->ops->is_active(se->opaque)) {
1337                 continue;
1338             }
1339         }
1340         trace_savevm_section_start(se->idstr, se->section_id);
1341 
1342         save_section_header(f, se, QEMU_VM_SECTION_END);
1343 
1344         ret = se->ops->save_live_complete_precopy(f, se->opaque);
1345         trace_savevm_section_end(se->idstr, se->section_id, ret);
1346         save_section_footer(f, se);
1347         if (ret < 0) {
1348             qemu_file_set_error(f, ret);
1349             return -1;
1350         }
1351     }
1352 
1353     return 0;
1354 }
1355 
1356 static
1357 int qemu_savevm_state_complete_precopy_non_iterable(QEMUFile *f,
1358                                                     bool in_postcopy,
1359                                                     bool inactivate_disks)
1360 {
1361     g_autoptr(QJSON) vmdesc = NULL;
1362     int vmdesc_len;
1363     SaveStateEntry *se;
1364     int ret;
1365 
1366     vmdesc = qjson_new();
1367     json_prop_int(vmdesc, "page_size", qemu_target_page_size());
1368     json_start_array(vmdesc, "devices");
1369     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1370 
1371         if ((!se->ops || !se->ops->save_state) && !se->vmsd) {
1372             continue;
1373         }
1374         if (se->vmsd && !vmstate_save_needed(se->vmsd, se->opaque)) {
1375             trace_savevm_section_skip(se->idstr, se->section_id);
1376             continue;
1377         }
1378 
1379         trace_savevm_section_start(se->idstr, se->section_id);
1380 
1381         json_start_object(vmdesc, NULL);
1382         json_prop_str(vmdesc, "name", se->idstr);
1383         json_prop_int(vmdesc, "instance_id", se->instance_id);
1384 
1385         save_section_header(f, se, QEMU_VM_SECTION_FULL);
1386         ret = vmstate_save(f, se, vmdesc);
1387         if (ret) {
1388             qemu_file_set_error(f, ret);
1389             return ret;
1390         }
1391         trace_savevm_section_end(se->idstr, se->section_id, 0);
1392         save_section_footer(f, se);
1393 
1394         json_end_object(vmdesc);
1395     }
1396 
1397     if (inactivate_disks) {
1398         /* Inactivate before sending QEMU_VM_EOF so that the
1399          * bdrv_invalidate_cache_all() on the other end won't fail. */
1400         ret = bdrv_inactivate_all();
1401         if (ret) {
1402             error_report("%s: bdrv_inactivate_all() failed (%d)",
1403                          __func__, ret);
1404             qemu_file_set_error(f, ret);
1405             return ret;
1406         }
1407     }
1408     if (!in_postcopy) {
1409         /* Postcopy stream will still be going */
1410         qemu_put_byte(f, QEMU_VM_EOF);
1411     }
1412 
1413     json_end_array(vmdesc);
1414     qjson_finish(vmdesc);
1415     vmdesc_len = strlen(qjson_get_str(vmdesc));
1416 
1417     if (should_send_vmdesc()) {
1418         qemu_put_byte(f, QEMU_VM_VMDESCRIPTION);
1419         qemu_put_be32(f, vmdesc_len);
1420         qemu_put_buffer(f, (uint8_t *)qjson_get_str(vmdesc), vmdesc_len);
1421     }
1422 
1423     return 0;
1424 }
1425 
1426 int qemu_savevm_state_complete_precopy(QEMUFile *f, bool iterable_only,
1427                                        bool inactivate_disks)
1428 {
1429     int ret;
1430     Error *local_err = NULL;
1431     bool in_postcopy = migration_in_postcopy();
1432 
1433     if (precopy_notify(PRECOPY_NOTIFY_COMPLETE, &local_err)) {
1434         error_report_err(local_err);
1435     }
1436 
1437     trace_savevm_state_complete_precopy();
1438 
1439     cpu_synchronize_all_states();
1440 
1441     if (!in_postcopy || iterable_only) {
1442         ret = qemu_savevm_state_complete_precopy_iterable(f, in_postcopy);
1443         if (ret) {
1444             return ret;
1445         }
1446     }
1447 
1448     if (iterable_only) {
1449         goto flush;
1450     }
1451 
1452     ret = qemu_savevm_state_complete_precopy_non_iterable(f, in_postcopy,
1453                                                           inactivate_disks);
1454     if (ret) {
1455         return ret;
1456     }
1457 
1458 flush:
1459     qemu_fflush(f);
1460     return 0;
1461 }
1462 
1463 /* Give an estimate of the amount left to be transferred,
1464  * the result is split into the amount for units that can and
1465  * for units that can't do postcopy.
1466  */
1467 void qemu_savevm_state_pending(QEMUFile *f, uint64_t threshold_size,
1468                                uint64_t *res_precopy_only,
1469                                uint64_t *res_compatible,
1470                                uint64_t *res_postcopy_only)
1471 {
1472     SaveStateEntry *se;
1473 
1474     *res_precopy_only = 0;
1475     *res_compatible = 0;
1476     *res_postcopy_only = 0;
1477 
1478 
1479     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1480         if (!se->ops || !se->ops->save_live_pending) {
1481             continue;
1482         }
1483         if (se->ops->is_active) {
1484             if (!se->ops->is_active(se->opaque)) {
1485                 continue;
1486             }
1487         }
1488         se->ops->save_live_pending(f, se->opaque, threshold_size,
1489                                    res_precopy_only, res_compatible,
1490                                    res_postcopy_only);
1491     }
1492 }
1493 
1494 void qemu_savevm_state_cleanup(void)
1495 {
1496     SaveStateEntry *se;
1497     Error *local_err = NULL;
1498 
1499     if (precopy_notify(PRECOPY_NOTIFY_CLEANUP, &local_err)) {
1500         error_report_err(local_err);
1501     }
1502 
1503     trace_savevm_state_cleanup();
1504     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1505         if (se->ops && se->ops->save_cleanup) {
1506             se->ops->save_cleanup(se->opaque);
1507         }
1508     }
1509 }
1510 
1511 static int qemu_savevm_state(QEMUFile *f, Error **errp)
1512 {
1513     int ret;
1514     MigrationState *ms = migrate_get_current();
1515     MigrationStatus status;
1516 
1517     if (migration_is_running(ms->state)) {
1518         error_setg(errp, QERR_MIGRATION_ACTIVE);
1519         return -EINVAL;
1520     }
1521 
1522     if (migrate_use_block()) {
1523         error_setg(errp, "Block migration and snapshots are incompatible");
1524         return -EINVAL;
1525     }
1526 
1527     migrate_init(ms);
1528     memset(&ram_counters, 0, sizeof(ram_counters));
1529     ms->to_dst_file = f;
1530 
1531     qemu_mutex_unlock_iothread();
1532     qemu_savevm_state_header(f);
1533     qemu_savevm_state_setup(f);
1534     qemu_mutex_lock_iothread();
1535 
1536     while (qemu_file_get_error(f) == 0) {
1537         if (qemu_savevm_state_iterate(f, false) > 0) {
1538             break;
1539         }
1540     }
1541 
1542     ret = qemu_file_get_error(f);
1543     if (ret == 0) {
1544         qemu_savevm_state_complete_precopy(f, false, false);
1545         ret = qemu_file_get_error(f);
1546     }
1547     qemu_savevm_state_cleanup();
1548     if (ret != 0) {
1549         error_setg_errno(errp, -ret, "Error while writing VM state");
1550     }
1551 
1552     if (ret != 0) {
1553         status = MIGRATION_STATUS_FAILED;
1554     } else {
1555         status = MIGRATION_STATUS_COMPLETED;
1556     }
1557     migrate_set_state(&ms->state, MIGRATION_STATUS_SETUP, status);
1558 
1559     /* f is outer parameter, it should not stay in global migration state after
1560      * this function finished */
1561     ms->to_dst_file = NULL;
1562 
1563     return ret;
1564 }
1565 
1566 void qemu_savevm_live_state(QEMUFile *f)
1567 {
1568     /* save QEMU_VM_SECTION_END section */
1569     qemu_savevm_state_complete_precopy(f, true, false);
1570     qemu_put_byte(f, QEMU_VM_EOF);
1571 }
1572 
1573 int qemu_save_device_state(QEMUFile *f)
1574 {
1575     SaveStateEntry *se;
1576 
1577     if (!migration_in_colo_state()) {
1578         qemu_put_be32(f, QEMU_VM_FILE_MAGIC);
1579         qemu_put_be32(f, QEMU_VM_FILE_VERSION);
1580     }
1581     cpu_synchronize_all_states();
1582 
1583     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1584         int ret;
1585 
1586         if (se->is_ram) {
1587             continue;
1588         }
1589         if ((!se->ops || !se->ops->save_state) && !se->vmsd) {
1590             continue;
1591         }
1592         if (se->vmsd && !vmstate_save_needed(se->vmsd, se->opaque)) {
1593             continue;
1594         }
1595 
1596         save_section_header(f, se, QEMU_VM_SECTION_FULL);
1597 
1598         ret = vmstate_save(f, se, NULL);
1599         if (ret) {
1600             return ret;
1601         }
1602 
1603         save_section_footer(f, se);
1604     }
1605 
1606     qemu_put_byte(f, QEMU_VM_EOF);
1607 
1608     return qemu_file_get_error(f);
1609 }
1610 
1611 static SaveStateEntry *find_se(const char *idstr, uint32_t instance_id)
1612 {
1613     SaveStateEntry *se;
1614 
1615     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1616         if (!strcmp(se->idstr, idstr) &&
1617             (instance_id == se->instance_id ||
1618              instance_id == se->alias_id))
1619             return se;
1620         /* Migrating from an older version? */
1621         if (strstr(se->idstr, idstr) && se->compat) {
1622             if (!strcmp(se->compat->idstr, idstr) &&
1623                 (instance_id == se->compat->instance_id ||
1624                  instance_id == se->alias_id))
1625                 return se;
1626         }
1627     }
1628     return NULL;
1629 }
1630 
1631 enum LoadVMExitCodes {
1632     /* Allow a command to quit all layers of nested loadvm loops */
1633     LOADVM_QUIT     =  1,
1634 };
1635 
1636 /* ------ incoming postcopy messages ------ */
1637 /* 'advise' arrives before any transfers just to tell us that a postcopy
1638  * *might* happen - it might be skipped if precopy transferred everything
1639  * quickly.
1640  */
1641 static int loadvm_postcopy_handle_advise(MigrationIncomingState *mis,
1642                                          uint16_t len)
1643 {
1644     PostcopyState ps = postcopy_state_set(POSTCOPY_INCOMING_ADVISE);
1645     uint64_t remote_pagesize_summary, local_pagesize_summary, remote_tps;
1646     Error *local_err = NULL;
1647 
1648     trace_loadvm_postcopy_handle_advise();
1649     if (ps != POSTCOPY_INCOMING_NONE) {
1650         error_report("CMD_POSTCOPY_ADVISE in wrong postcopy state (%d)", ps);
1651         return -1;
1652     }
1653 
1654     switch (len) {
1655     case 0:
1656         if (migrate_postcopy_ram()) {
1657             error_report("RAM postcopy is enabled but have 0 byte advise");
1658             return -EINVAL;
1659         }
1660         return 0;
1661     case 8 + 8:
1662         if (!migrate_postcopy_ram()) {
1663             error_report("RAM postcopy is disabled but have 16 byte advise");
1664             return -EINVAL;
1665         }
1666         break;
1667     default:
1668         error_report("CMD_POSTCOPY_ADVISE invalid length (%d)", len);
1669         return -EINVAL;
1670     }
1671 
1672     if (!postcopy_ram_supported_by_host(mis)) {
1673         postcopy_state_set(POSTCOPY_INCOMING_NONE);
1674         return -1;
1675     }
1676 
1677     remote_pagesize_summary = qemu_get_be64(mis->from_src_file);
1678     local_pagesize_summary = ram_pagesize_summary();
1679 
1680     if (remote_pagesize_summary != local_pagesize_summary)  {
1681         /*
1682          * This detects two potential causes of mismatch:
1683          *   a) A mismatch in host page sizes
1684          *      Some combinations of mismatch are probably possible but it gets
1685          *      a bit more complicated.  In particular we need to place whole
1686          *      host pages on the dest at once, and we need to ensure that we
1687          *      handle dirtying to make sure we never end up sending part of
1688          *      a hostpage on it's own.
1689          *   b) The use of different huge page sizes on source/destination
1690          *      a more fine grain test is performed during RAM block migration
1691          *      but this test here causes a nice early clear failure, and
1692          *      also fails when passed to an older qemu that doesn't
1693          *      do huge pages.
1694          */
1695         error_report("Postcopy needs matching RAM page sizes (s=%" PRIx64
1696                                                              " d=%" PRIx64 ")",
1697                      remote_pagesize_summary, local_pagesize_summary);
1698         return -1;
1699     }
1700 
1701     remote_tps = qemu_get_be64(mis->from_src_file);
1702     if (remote_tps != qemu_target_page_size()) {
1703         /*
1704          * Again, some differences could be dealt with, but for now keep it
1705          * simple.
1706          */
1707         error_report("Postcopy needs matching target page sizes (s=%d d=%zd)",
1708                      (int)remote_tps, qemu_target_page_size());
1709         return -1;
1710     }
1711 
1712     if (postcopy_notify(POSTCOPY_NOTIFY_INBOUND_ADVISE, &local_err)) {
1713         error_report_err(local_err);
1714         return -1;
1715     }
1716 
1717     if (ram_postcopy_incoming_init(mis)) {
1718         return -1;
1719     }
1720 
1721     return 0;
1722 }
1723 
1724 /* After postcopy we will be told to throw some pages away since they're
1725  * dirty and will have to be demand fetched.  Must happen before CPU is
1726  * started.
1727  * There can be 0..many of these messages, each encoding multiple pages.
1728  */
1729 static int loadvm_postcopy_ram_handle_discard(MigrationIncomingState *mis,
1730                                               uint16_t len)
1731 {
1732     int tmp;
1733     char ramid[256];
1734     PostcopyState ps = postcopy_state_get();
1735 
1736     trace_loadvm_postcopy_ram_handle_discard();
1737 
1738     switch (ps) {
1739     case POSTCOPY_INCOMING_ADVISE:
1740         /* 1st discard */
1741         tmp = postcopy_ram_prepare_discard(mis);
1742         if (tmp) {
1743             return tmp;
1744         }
1745         break;
1746 
1747     case POSTCOPY_INCOMING_DISCARD:
1748         /* Expected state */
1749         break;
1750 
1751     default:
1752         error_report("CMD_POSTCOPY_RAM_DISCARD in wrong postcopy state (%d)",
1753                      ps);
1754         return -1;
1755     }
1756     /* We're expecting a
1757      *    Version (0)
1758      *    a RAM ID string (length byte, name, 0 term)
1759      *    then at least 1 16 byte chunk
1760     */
1761     if (len < (1 + 1 + 1 + 1 + 2 * 8)) {
1762         error_report("CMD_POSTCOPY_RAM_DISCARD invalid length (%d)", len);
1763         return -1;
1764     }
1765 
1766     tmp = qemu_get_byte(mis->from_src_file);
1767     if (tmp != postcopy_ram_discard_version) {
1768         error_report("CMD_POSTCOPY_RAM_DISCARD invalid version (%d)", tmp);
1769         return -1;
1770     }
1771 
1772     if (!qemu_get_counted_string(mis->from_src_file, ramid)) {
1773         error_report("CMD_POSTCOPY_RAM_DISCARD Failed to read RAMBlock ID");
1774         return -1;
1775     }
1776     tmp = qemu_get_byte(mis->from_src_file);
1777     if (tmp != 0) {
1778         error_report("CMD_POSTCOPY_RAM_DISCARD missing nil (%d)", tmp);
1779         return -1;
1780     }
1781 
1782     len -= 3 + strlen(ramid);
1783     if (len % 16) {
1784         error_report("CMD_POSTCOPY_RAM_DISCARD invalid length (%d)", len);
1785         return -1;
1786     }
1787     trace_loadvm_postcopy_ram_handle_discard_header(ramid, len);
1788     while (len) {
1789         uint64_t start_addr, block_length;
1790         start_addr = qemu_get_be64(mis->from_src_file);
1791         block_length = qemu_get_be64(mis->from_src_file);
1792 
1793         len -= 16;
1794         int ret = ram_discard_range(ramid, start_addr, block_length);
1795         if (ret) {
1796             return ret;
1797         }
1798     }
1799     trace_loadvm_postcopy_ram_handle_discard_end();
1800 
1801     return 0;
1802 }
1803 
1804 /*
1805  * Triggered by a postcopy_listen command; this thread takes over reading
1806  * the input stream, leaving the main thread free to carry on loading the rest
1807  * of the device state (from RAM).
1808  * (TODO:This could do with being in a postcopy file - but there again it's
1809  * just another input loop, not that postcopy specific)
1810  */
1811 static void *postcopy_ram_listen_thread(void *opaque)
1812 {
1813     MigrationIncomingState *mis = migration_incoming_get_current();
1814     QEMUFile *f = mis->from_src_file;
1815     int load_res;
1816     MigrationState *migr = migrate_get_current();
1817 
1818     object_ref(OBJECT(migr));
1819 
1820     migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE,
1821                                    MIGRATION_STATUS_POSTCOPY_ACTIVE);
1822     qemu_sem_post(&mis->listen_thread_sem);
1823     trace_postcopy_ram_listen_thread_start();
1824 
1825     rcu_register_thread();
1826     /*
1827      * Because we're a thread and not a coroutine we can't yield
1828      * in qemu_file, and thus we must be blocking now.
1829      */
1830     qemu_file_set_blocking(f, true);
1831     load_res = qemu_loadvm_state_main(f, mis);
1832 
1833     /*
1834      * This is tricky, but, mis->from_src_file can change after it
1835      * returns, when postcopy recovery happened. In the future, we may
1836      * want a wrapper for the QEMUFile handle.
1837      */
1838     f = mis->from_src_file;
1839 
1840     /* And non-blocking again so we don't block in any cleanup */
1841     qemu_file_set_blocking(f, false);
1842 
1843     trace_postcopy_ram_listen_thread_exit();
1844     if (load_res < 0) {
1845         qemu_file_set_error(f, load_res);
1846         dirty_bitmap_mig_cancel_incoming();
1847         if (postcopy_state_get() == POSTCOPY_INCOMING_RUNNING &&
1848             !migrate_postcopy_ram() && migrate_dirty_bitmaps())
1849         {
1850             error_report("%s: loadvm failed during postcopy: %d. All states "
1851                          "are migrated except dirty bitmaps. Some dirty "
1852                          "bitmaps may be lost, and present migrated dirty "
1853                          "bitmaps are correctly migrated and valid.",
1854                          __func__, load_res);
1855             load_res = 0; /* prevent further exit() */
1856         } else {
1857             error_report("%s: loadvm failed: %d", __func__, load_res);
1858             migrate_set_state(&mis->state, MIGRATION_STATUS_POSTCOPY_ACTIVE,
1859                                            MIGRATION_STATUS_FAILED);
1860         }
1861     }
1862     if (load_res >= 0) {
1863         /*
1864          * This looks good, but it's possible that the device loading in the
1865          * main thread hasn't finished yet, and so we might not be in 'RUN'
1866          * state yet; wait for the end of the main thread.
1867          */
1868         qemu_event_wait(&mis->main_thread_load_event);
1869     }
1870     postcopy_ram_incoming_cleanup(mis);
1871 
1872     if (load_res < 0) {
1873         /*
1874          * If something went wrong then we have a bad state so exit;
1875          * depending how far we got it might be possible at this point
1876          * to leave the guest running and fire MCEs for pages that never
1877          * arrived as a desperate recovery step.
1878          */
1879         rcu_unregister_thread();
1880         exit(EXIT_FAILURE);
1881     }
1882 
1883     migrate_set_state(&mis->state, MIGRATION_STATUS_POSTCOPY_ACTIVE,
1884                                    MIGRATION_STATUS_COMPLETED);
1885     /*
1886      * If everything has worked fine, then the main thread has waited
1887      * for us to start, and we're the last use of the mis.
1888      * (If something broke then qemu will have to exit anyway since it's
1889      * got a bad migration state).
1890      */
1891     migration_incoming_state_destroy();
1892     qemu_loadvm_state_cleanup();
1893 
1894     rcu_unregister_thread();
1895     mis->have_listen_thread = false;
1896     postcopy_state_set(POSTCOPY_INCOMING_END);
1897 
1898     object_unref(OBJECT(migr));
1899 
1900     return NULL;
1901 }
1902 
1903 /* After this message we must be able to immediately receive postcopy data */
1904 static int loadvm_postcopy_handle_listen(MigrationIncomingState *mis)
1905 {
1906     PostcopyState ps = postcopy_state_set(POSTCOPY_INCOMING_LISTENING);
1907     trace_loadvm_postcopy_handle_listen();
1908     Error *local_err = NULL;
1909 
1910     if (ps != POSTCOPY_INCOMING_ADVISE && ps != POSTCOPY_INCOMING_DISCARD) {
1911         error_report("CMD_POSTCOPY_LISTEN in wrong postcopy state (%d)", ps);
1912         return -1;
1913     }
1914     if (ps == POSTCOPY_INCOMING_ADVISE) {
1915         /*
1916          * A rare case, we entered listen without having to do any discards,
1917          * so do the setup that's normally done at the time of the 1st discard.
1918          */
1919         if (migrate_postcopy_ram()) {
1920             postcopy_ram_prepare_discard(mis);
1921         }
1922     }
1923 
1924     /*
1925      * Sensitise RAM - can now generate requests for blocks that don't exist
1926      * However, at this point the CPU shouldn't be running, and the IO
1927      * shouldn't be doing anything yet so don't actually expect requests
1928      */
1929     if (migrate_postcopy_ram()) {
1930         if (postcopy_ram_incoming_setup(mis)) {
1931             postcopy_ram_incoming_cleanup(mis);
1932             return -1;
1933         }
1934     }
1935 
1936     if (postcopy_notify(POSTCOPY_NOTIFY_INBOUND_LISTEN, &local_err)) {
1937         error_report_err(local_err);
1938         return -1;
1939     }
1940 
1941     mis->have_listen_thread = true;
1942     /* Start up the listening thread and wait for it to signal ready */
1943     qemu_sem_init(&mis->listen_thread_sem, 0);
1944     qemu_thread_create(&mis->listen_thread, "postcopy/listen",
1945                        postcopy_ram_listen_thread, NULL,
1946                        QEMU_THREAD_DETACHED);
1947     qemu_sem_wait(&mis->listen_thread_sem);
1948     qemu_sem_destroy(&mis->listen_thread_sem);
1949 
1950     return 0;
1951 }
1952 
1953 static void loadvm_postcopy_handle_run_bh(void *opaque)
1954 {
1955     Error *local_err = NULL;
1956     MigrationIncomingState *mis = opaque;
1957 
1958     /* TODO we should move all of this lot into postcopy_ram.c or a shared code
1959      * in migration.c
1960      */
1961     cpu_synchronize_all_post_init();
1962 
1963     qemu_announce_self(&mis->announce_timer, migrate_announce_params());
1964 
1965     /* Make sure all file formats flush their mutable metadata.
1966      * If we get an error here, just don't restart the VM yet. */
1967     bdrv_invalidate_cache_all(&local_err);
1968     if (local_err) {
1969         error_report_err(local_err);
1970         local_err = NULL;
1971         autostart = false;
1972     }
1973 
1974     trace_loadvm_postcopy_handle_run_cpu_sync();
1975 
1976     trace_loadvm_postcopy_handle_run_vmstart();
1977 
1978     dirty_bitmap_mig_before_vm_start();
1979 
1980     if (autostart) {
1981         /* Hold onto your hats, starting the CPU */
1982         vm_start();
1983     } else {
1984         /* leave it paused and let management decide when to start the CPU */
1985         runstate_set(RUN_STATE_PAUSED);
1986     }
1987 
1988     qemu_bh_delete(mis->bh);
1989 }
1990 
1991 /* After all discards we can start running and asking for pages */
1992 static int loadvm_postcopy_handle_run(MigrationIncomingState *mis)
1993 {
1994     PostcopyState ps = postcopy_state_get();
1995 
1996     trace_loadvm_postcopy_handle_run();
1997     if (ps != POSTCOPY_INCOMING_LISTENING) {
1998         error_report("CMD_POSTCOPY_RUN in wrong postcopy state (%d)", ps);
1999         return -1;
2000     }
2001 
2002     postcopy_state_set(POSTCOPY_INCOMING_RUNNING);
2003     mis->bh = qemu_bh_new(loadvm_postcopy_handle_run_bh, mis);
2004     qemu_bh_schedule(mis->bh);
2005 
2006     /* We need to finish reading the stream from the package
2007      * and also stop reading anything more from the stream that loaded the
2008      * package (since it's now being read by the listener thread).
2009      * LOADVM_QUIT will quit all the layers of nested loadvm loops.
2010      */
2011     return LOADVM_QUIT;
2012 }
2013 
2014 static int loadvm_postcopy_handle_resume(MigrationIncomingState *mis)
2015 {
2016     if (mis->state != MIGRATION_STATUS_POSTCOPY_RECOVER) {
2017         error_report("%s: illegal resume received", __func__);
2018         /* Don't fail the load, only for this. */
2019         return 0;
2020     }
2021 
2022     /*
2023      * This means source VM is ready to resume the postcopy migration.
2024      * It's time to switch state and release the fault thread to
2025      * continue service page faults.
2026      */
2027     migrate_set_state(&mis->state, MIGRATION_STATUS_POSTCOPY_RECOVER,
2028                       MIGRATION_STATUS_POSTCOPY_ACTIVE);
2029     qemu_sem_post(&mis->postcopy_pause_sem_fault);
2030 
2031     trace_loadvm_postcopy_handle_resume();
2032 
2033     /* Tell source that "we are ready" */
2034     migrate_send_rp_resume_ack(mis, MIGRATION_RESUME_ACK_VALUE);
2035 
2036     return 0;
2037 }
2038 
2039 /**
2040  * Immediately following this command is a blob of data containing an embedded
2041  * chunk of migration stream; read it and load it.
2042  *
2043  * @mis: Incoming state
2044  * @length: Length of packaged data to read
2045  *
2046  * Returns: Negative values on error
2047  *
2048  */
2049 static int loadvm_handle_cmd_packaged(MigrationIncomingState *mis)
2050 {
2051     int ret;
2052     size_t length;
2053     QIOChannelBuffer *bioc;
2054 
2055     length = qemu_get_be32(mis->from_src_file);
2056     trace_loadvm_handle_cmd_packaged(length);
2057 
2058     if (length > MAX_VM_CMD_PACKAGED_SIZE) {
2059         error_report("Unreasonably large packaged state: %zu", length);
2060         return -1;
2061     }
2062 
2063     bioc = qio_channel_buffer_new(length);
2064     qio_channel_set_name(QIO_CHANNEL(bioc), "migration-loadvm-buffer");
2065     ret = qemu_get_buffer(mis->from_src_file,
2066                           bioc->data,
2067                           length);
2068     if (ret != length) {
2069         object_unref(OBJECT(bioc));
2070         error_report("CMD_PACKAGED: Buffer receive fail ret=%d length=%zu",
2071                      ret, length);
2072         return (ret < 0) ? ret : -EAGAIN;
2073     }
2074     bioc->usage += length;
2075     trace_loadvm_handle_cmd_packaged_received(ret);
2076 
2077     QEMUFile *packf = qemu_fopen_channel_input(QIO_CHANNEL(bioc));
2078 
2079     ret = qemu_loadvm_state_main(packf, mis);
2080     trace_loadvm_handle_cmd_packaged_main(ret);
2081     qemu_fclose(packf);
2082     object_unref(OBJECT(bioc));
2083 
2084     return ret;
2085 }
2086 
2087 /*
2088  * Handle request that source requests for recved_bitmap on
2089  * destination. Payload format:
2090  *
2091  * len (1 byte) + ramblock_name (<255 bytes)
2092  */
2093 static int loadvm_handle_recv_bitmap(MigrationIncomingState *mis,
2094                                      uint16_t len)
2095 {
2096     QEMUFile *file = mis->from_src_file;
2097     RAMBlock *rb;
2098     char block_name[256];
2099     size_t cnt;
2100 
2101     cnt = qemu_get_counted_string(file, block_name);
2102     if (!cnt) {
2103         error_report("%s: failed to read block name", __func__);
2104         return -EINVAL;
2105     }
2106 
2107     /* Validate before using the data */
2108     if (qemu_file_get_error(file)) {
2109         return qemu_file_get_error(file);
2110     }
2111 
2112     if (len != cnt + 1) {
2113         error_report("%s: invalid payload length (%d)", __func__, len);
2114         return -EINVAL;
2115     }
2116 
2117     rb = qemu_ram_block_by_name(block_name);
2118     if (!rb) {
2119         error_report("%s: block '%s' not found", __func__, block_name);
2120         return -EINVAL;
2121     }
2122 
2123     migrate_send_rp_recv_bitmap(mis, block_name);
2124 
2125     trace_loadvm_handle_recv_bitmap(block_name);
2126 
2127     return 0;
2128 }
2129 
2130 static int loadvm_process_enable_colo(MigrationIncomingState *mis)
2131 {
2132     int ret = migration_incoming_enable_colo();
2133 
2134     if (!ret) {
2135         ret = colo_init_ram_cache();
2136         if (ret) {
2137             migration_incoming_disable_colo();
2138         }
2139     }
2140     return ret;
2141 }
2142 
2143 /*
2144  * Process an incoming 'QEMU_VM_COMMAND'
2145  * 0           just a normal return
2146  * LOADVM_QUIT All good, but exit the loop
2147  * <0          Error
2148  */
2149 static int loadvm_process_command(QEMUFile *f)
2150 {
2151     MigrationIncomingState *mis = migration_incoming_get_current();
2152     uint16_t cmd;
2153     uint16_t len;
2154     uint32_t tmp32;
2155 
2156     cmd = qemu_get_be16(f);
2157     len = qemu_get_be16(f);
2158 
2159     /* Check validity before continue processing of cmds */
2160     if (qemu_file_get_error(f)) {
2161         return qemu_file_get_error(f);
2162     }
2163 
2164     trace_loadvm_process_command(cmd, len);
2165     if (cmd >= MIG_CMD_MAX || cmd == MIG_CMD_INVALID) {
2166         error_report("MIG_CMD 0x%x unknown (len 0x%x)", cmd, len);
2167         return -EINVAL;
2168     }
2169 
2170     if (mig_cmd_args[cmd].len != -1 && mig_cmd_args[cmd].len != len) {
2171         error_report("%s received with bad length - expecting %zu, got %d",
2172                      mig_cmd_args[cmd].name,
2173                      (size_t)mig_cmd_args[cmd].len, len);
2174         return -ERANGE;
2175     }
2176 
2177     switch (cmd) {
2178     case MIG_CMD_OPEN_RETURN_PATH:
2179         if (mis->to_src_file) {
2180             error_report("CMD_OPEN_RETURN_PATH called when RP already open");
2181             /* Not really a problem, so don't give up */
2182             return 0;
2183         }
2184         mis->to_src_file = qemu_file_get_return_path(f);
2185         if (!mis->to_src_file) {
2186             error_report("CMD_OPEN_RETURN_PATH failed");
2187             return -1;
2188         }
2189         break;
2190 
2191     case MIG_CMD_PING:
2192         tmp32 = qemu_get_be32(f);
2193         trace_loadvm_process_command_ping(tmp32);
2194         if (!mis->to_src_file) {
2195             error_report("CMD_PING (0x%x) received with no return path",
2196                          tmp32);
2197             return -1;
2198         }
2199         migrate_send_rp_pong(mis, tmp32);
2200         break;
2201 
2202     case MIG_CMD_PACKAGED:
2203         return loadvm_handle_cmd_packaged(mis);
2204 
2205     case MIG_CMD_POSTCOPY_ADVISE:
2206         return loadvm_postcopy_handle_advise(mis, len);
2207 
2208     case MIG_CMD_POSTCOPY_LISTEN:
2209         return loadvm_postcopy_handle_listen(mis);
2210 
2211     case MIG_CMD_POSTCOPY_RUN:
2212         return loadvm_postcopy_handle_run(mis);
2213 
2214     case MIG_CMD_POSTCOPY_RAM_DISCARD:
2215         return loadvm_postcopy_ram_handle_discard(mis, len);
2216 
2217     case MIG_CMD_POSTCOPY_RESUME:
2218         return loadvm_postcopy_handle_resume(mis);
2219 
2220     case MIG_CMD_RECV_BITMAP:
2221         return loadvm_handle_recv_bitmap(mis, len);
2222 
2223     case MIG_CMD_ENABLE_COLO:
2224         return loadvm_process_enable_colo(mis);
2225     }
2226 
2227     return 0;
2228 }
2229 
2230 /*
2231  * Read a footer off the wire and check that it matches the expected section
2232  *
2233  * Returns: true if the footer was good
2234  *          false if there is a problem (and calls error_report to say why)
2235  */
2236 static bool check_section_footer(QEMUFile *f, SaveStateEntry *se)
2237 {
2238     int ret;
2239     uint8_t read_mark;
2240     uint32_t read_section_id;
2241 
2242     if (!migrate_get_current()->send_section_footer) {
2243         /* No footer to check */
2244         return true;
2245     }
2246 
2247     read_mark = qemu_get_byte(f);
2248 
2249     ret = qemu_file_get_error(f);
2250     if (ret) {
2251         error_report("%s: Read section footer failed: %d",
2252                      __func__, ret);
2253         return false;
2254     }
2255 
2256     if (read_mark != QEMU_VM_SECTION_FOOTER) {
2257         error_report("Missing section footer for %s", se->idstr);
2258         return false;
2259     }
2260 
2261     read_section_id = qemu_get_be32(f);
2262     if (read_section_id != se->load_section_id) {
2263         error_report("Mismatched section id in footer for %s -"
2264                      " read 0x%x expected 0x%x",
2265                      se->idstr, read_section_id, se->load_section_id);
2266         return false;
2267     }
2268 
2269     /* All good */
2270     return true;
2271 }
2272 
2273 static int
2274 qemu_loadvm_section_start_full(QEMUFile *f, MigrationIncomingState *mis)
2275 {
2276     uint32_t instance_id, version_id, section_id;
2277     SaveStateEntry *se;
2278     char idstr[256];
2279     int ret;
2280 
2281     /* Read section start */
2282     section_id = qemu_get_be32(f);
2283     if (!qemu_get_counted_string(f, idstr)) {
2284         error_report("Unable to read ID string for section %u",
2285                      section_id);
2286         return -EINVAL;
2287     }
2288     instance_id = qemu_get_be32(f);
2289     version_id = qemu_get_be32(f);
2290 
2291     ret = qemu_file_get_error(f);
2292     if (ret) {
2293         error_report("%s: Failed to read instance/version ID: %d",
2294                      __func__, ret);
2295         return ret;
2296     }
2297 
2298     trace_qemu_loadvm_state_section_startfull(section_id, idstr,
2299             instance_id, version_id);
2300     /* Find savevm section */
2301     se = find_se(idstr, instance_id);
2302     if (se == NULL) {
2303         error_report("Unknown savevm section or instance '%s' %"PRIu32". "
2304                      "Make sure that your current VM setup matches your "
2305                      "saved VM setup, including any hotplugged devices",
2306                      idstr, instance_id);
2307         return -EINVAL;
2308     }
2309 
2310     /* Validate version */
2311     if (version_id > se->version_id) {
2312         error_report("savevm: unsupported version %d for '%s' v%d",
2313                      version_id, idstr, se->version_id);
2314         return -EINVAL;
2315     }
2316     se->load_version_id = version_id;
2317     se->load_section_id = section_id;
2318 
2319     /* Validate if it is a device's state */
2320     if (xen_enabled() && se->is_ram) {
2321         error_report("loadvm: %s RAM loading not allowed on Xen", idstr);
2322         return -EINVAL;
2323     }
2324 
2325     ret = vmstate_load(f, se);
2326     if (ret < 0) {
2327         error_report("error while loading state for instance 0x%"PRIx32" of"
2328                      " device '%s'", instance_id, idstr);
2329         return ret;
2330     }
2331     if (!check_section_footer(f, se)) {
2332         return -EINVAL;
2333     }
2334 
2335     return 0;
2336 }
2337 
2338 static int
2339 qemu_loadvm_section_part_end(QEMUFile *f, MigrationIncomingState *mis)
2340 {
2341     uint32_t section_id;
2342     SaveStateEntry *se;
2343     int ret;
2344 
2345     section_id = qemu_get_be32(f);
2346 
2347     ret = qemu_file_get_error(f);
2348     if (ret) {
2349         error_report("%s: Failed to read section ID: %d",
2350                      __func__, ret);
2351         return ret;
2352     }
2353 
2354     trace_qemu_loadvm_state_section_partend(section_id);
2355     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
2356         if (se->load_section_id == section_id) {
2357             break;
2358         }
2359     }
2360     if (se == NULL) {
2361         error_report("Unknown savevm section %d", section_id);
2362         return -EINVAL;
2363     }
2364 
2365     ret = vmstate_load(f, se);
2366     if (ret < 0) {
2367         error_report("error while loading state section id %d(%s)",
2368                      section_id, se->idstr);
2369         return ret;
2370     }
2371     if (!check_section_footer(f, se)) {
2372         return -EINVAL;
2373     }
2374 
2375     return 0;
2376 }
2377 
2378 static int qemu_loadvm_state_header(QEMUFile *f)
2379 {
2380     unsigned int v;
2381     int ret;
2382 
2383     v = qemu_get_be32(f);
2384     if (v != QEMU_VM_FILE_MAGIC) {
2385         error_report("Not a migration stream");
2386         return -EINVAL;
2387     }
2388 
2389     v = qemu_get_be32(f);
2390     if (v == QEMU_VM_FILE_VERSION_COMPAT) {
2391         error_report("SaveVM v2 format is obsolete and don't work anymore");
2392         return -ENOTSUP;
2393     }
2394     if (v != QEMU_VM_FILE_VERSION) {
2395         error_report("Unsupported migration stream version");
2396         return -ENOTSUP;
2397     }
2398 
2399     if (migrate_get_current()->send_configuration) {
2400         if (qemu_get_byte(f) != QEMU_VM_CONFIGURATION) {
2401             error_report("Configuration section missing");
2402             qemu_loadvm_state_cleanup();
2403             return -EINVAL;
2404         }
2405         ret = vmstate_load_state(f, &vmstate_configuration, &savevm_state, 0);
2406 
2407         if (ret) {
2408             qemu_loadvm_state_cleanup();
2409             return ret;
2410         }
2411     }
2412     return 0;
2413 }
2414 
2415 static int qemu_loadvm_state_setup(QEMUFile *f)
2416 {
2417     SaveStateEntry *se;
2418     int ret;
2419 
2420     trace_loadvm_state_setup();
2421     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
2422         if (!se->ops || !se->ops->load_setup) {
2423             continue;
2424         }
2425         if (se->ops->is_active) {
2426             if (!se->ops->is_active(se->opaque)) {
2427                 continue;
2428             }
2429         }
2430 
2431         ret = se->ops->load_setup(f, se->opaque);
2432         if (ret < 0) {
2433             qemu_file_set_error(f, ret);
2434             error_report("Load state of device %s failed", se->idstr);
2435             return ret;
2436         }
2437     }
2438     return 0;
2439 }
2440 
2441 void qemu_loadvm_state_cleanup(void)
2442 {
2443     SaveStateEntry *se;
2444 
2445     trace_loadvm_state_cleanup();
2446     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
2447         if (se->ops && se->ops->load_cleanup) {
2448             se->ops->load_cleanup(se->opaque);
2449         }
2450     }
2451 }
2452 
2453 /* Return true if we should continue the migration, or false. */
2454 static bool postcopy_pause_incoming(MigrationIncomingState *mis)
2455 {
2456     trace_postcopy_pause_incoming();
2457 
2458     assert(migrate_postcopy_ram());
2459 
2460     /* Clear the triggered bit to allow one recovery */
2461     mis->postcopy_recover_triggered = false;
2462 
2463     assert(mis->from_src_file);
2464     qemu_file_shutdown(mis->from_src_file);
2465     qemu_fclose(mis->from_src_file);
2466     mis->from_src_file = NULL;
2467 
2468     assert(mis->to_src_file);
2469     qemu_file_shutdown(mis->to_src_file);
2470     qemu_mutex_lock(&mis->rp_mutex);
2471     qemu_fclose(mis->to_src_file);
2472     mis->to_src_file = NULL;
2473     qemu_mutex_unlock(&mis->rp_mutex);
2474 
2475     migrate_set_state(&mis->state, MIGRATION_STATUS_POSTCOPY_ACTIVE,
2476                       MIGRATION_STATUS_POSTCOPY_PAUSED);
2477 
2478     /* Notify the fault thread for the invalidated file handle */
2479     postcopy_fault_thread_notify(mis);
2480 
2481     error_report("Detected IO failure for postcopy. "
2482                  "Migration paused.");
2483 
2484     while (mis->state == MIGRATION_STATUS_POSTCOPY_PAUSED) {
2485         qemu_sem_wait(&mis->postcopy_pause_sem_dst);
2486     }
2487 
2488     trace_postcopy_pause_incoming_continued();
2489 
2490     return true;
2491 }
2492 
2493 int qemu_loadvm_state_main(QEMUFile *f, MigrationIncomingState *mis)
2494 {
2495     uint8_t section_type;
2496     int ret = 0;
2497 
2498 retry:
2499     while (true) {
2500         section_type = qemu_get_byte(f);
2501 
2502         if (qemu_file_get_error(f)) {
2503             ret = qemu_file_get_error(f);
2504             break;
2505         }
2506 
2507         trace_qemu_loadvm_state_section(section_type);
2508         switch (section_type) {
2509         case QEMU_VM_SECTION_START:
2510         case QEMU_VM_SECTION_FULL:
2511             ret = qemu_loadvm_section_start_full(f, mis);
2512             if (ret < 0) {
2513                 goto out;
2514             }
2515             break;
2516         case QEMU_VM_SECTION_PART:
2517         case QEMU_VM_SECTION_END:
2518             ret = qemu_loadvm_section_part_end(f, mis);
2519             if (ret < 0) {
2520                 goto out;
2521             }
2522             break;
2523         case QEMU_VM_COMMAND:
2524             ret = loadvm_process_command(f);
2525             trace_qemu_loadvm_state_section_command(ret);
2526             if ((ret < 0) || (ret == LOADVM_QUIT)) {
2527                 goto out;
2528             }
2529             break;
2530         case QEMU_VM_EOF:
2531             /* This is the end of migration */
2532             goto out;
2533         default:
2534             error_report("Unknown savevm section type %d", section_type);
2535             ret = -EINVAL;
2536             goto out;
2537         }
2538     }
2539 
2540 out:
2541     if (ret < 0) {
2542         qemu_file_set_error(f, ret);
2543 
2544         /* Cancel bitmaps incoming regardless of recovery */
2545         dirty_bitmap_mig_cancel_incoming();
2546 
2547         /*
2548          * If we are during an active postcopy, then we pause instead
2549          * of bail out to at least keep the VM's dirty data.  Note
2550          * that POSTCOPY_INCOMING_LISTENING stage is still not enough,
2551          * during which we're still receiving device states and we
2552          * still haven't yet started the VM on destination.
2553          *
2554          * Only RAM postcopy supports recovery. Still, if RAM postcopy is
2555          * enabled, canceled bitmaps postcopy will not affect RAM postcopy
2556          * recovering.
2557          */
2558         if (postcopy_state_get() == POSTCOPY_INCOMING_RUNNING &&
2559             migrate_postcopy_ram() && postcopy_pause_incoming(mis)) {
2560             /* Reset f to point to the newly created channel */
2561             f = mis->from_src_file;
2562             goto retry;
2563         }
2564     }
2565     return ret;
2566 }
2567 
2568 int qemu_loadvm_state(QEMUFile *f)
2569 {
2570     MigrationIncomingState *mis = migration_incoming_get_current();
2571     Error *local_err = NULL;
2572     int ret;
2573 
2574     if (qemu_savevm_state_blocked(&local_err)) {
2575         error_report_err(local_err);
2576         return -EINVAL;
2577     }
2578 
2579     ret = qemu_loadvm_state_header(f);
2580     if (ret) {
2581         return ret;
2582     }
2583 
2584     if (qemu_loadvm_state_setup(f) != 0) {
2585         return -EINVAL;
2586     }
2587 
2588     cpu_synchronize_all_pre_loadvm();
2589 
2590     ret = qemu_loadvm_state_main(f, mis);
2591     qemu_event_set(&mis->main_thread_load_event);
2592 
2593     trace_qemu_loadvm_state_post_main(ret);
2594 
2595     if (mis->have_listen_thread) {
2596         /* Listen thread still going, can't clean up yet */
2597         return ret;
2598     }
2599 
2600     if (ret == 0) {
2601         ret = qemu_file_get_error(f);
2602     }
2603 
2604     /*
2605      * Try to read in the VMDESC section as well, so that dumping tools that
2606      * intercept our migration stream have the chance to see it.
2607      */
2608 
2609     /* We've got to be careful; if we don't read the data and just shut the fd
2610      * then the sender can error if we close while it's still sending.
2611      * We also mustn't read data that isn't there; some transports (RDMA)
2612      * will stall waiting for that data when the source has already closed.
2613      */
2614     if (ret == 0 && should_send_vmdesc()) {
2615         uint8_t *buf;
2616         uint32_t size;
2617         uint8_t  section_type = qemu_get_byte(f);
2618 
2619         if (section_type != QEMU_VM_VMDESCRIPTION) {
2620             error_report("Expected vmdescription section, but got %d",
2621                          section_type);
2622             /*
2623              * It doesn't seem worth failing at this point since
2624              * we apparently have an otherwise valid VM state
2625              */
2626         } else {
2627             buf = g_malloc(0x1000);
2628             size = qemu_get_be32(f);
2629 
2630             while (size > 0) {
2631                 uint32_t read_chunk = MIN(size, 0x1000);
2632                 qemu_get_buffer(f, buf, read_chunk);
2633                 size -= read_chunk;
2634             }
2635             g_free(buf);
2636         }
2637     }
2638 
2639     qemu_loadvm_state_cleanup();
2640     cpu_synchronize_all_post_init();
2641 
2642     return ret;
2643 }
2644 
2645 int qemu_load_device_state(QEMUFile *f)
2646 {
2647     MigrationIncomingState *mis = migration_incoming_get_current();
2648     int ret;
2649 
2650     /* Load QEMU_VM_SECTION_FULL section */
2651     ret = qemu_loadvm_state_main(f, mis);
2652     if (ret < 0) {
2653         error_report("Failed to load device state: %d", ret);
2654         return ret;
2655     }
2656 
2657     cpu_synchronize_all_post_init();
2658     return 0;
2659 }
2660 
2661 int save_snapshot(const char *name, Error **errp)
2662 {
2663     BlockDriverState *bs, *bs1;
2664     QEMUSnapshotInfo sn1, *sn = &sn1, old_sn1, *old_sn = &old_sn1;
2665     int ret = -1, ret2;
2666     QEMUFile *f;
2667     int saved_vm_running;
2668     uint64_t vm_state_size;
2669     qemu_timeval tv;
2670     struct tm tm;
2671     AioContext *aio_context;
2672 
2673     if (migration_is_blocked(errp)) {
2674         return ret;
2675     }
2676 
2677     if (!replay_can_snapshot()) {
2678         error_setg(errp, "Record/replay does not allow making snapshot "
2679                    "right now. Try once more later.");
2680         return ret;
2681     }
2682 
2683     if (!bdrv_all_can_snapshot(&bs)) {
2684         error_setg(errp, "Device '%s' is writable but does not support "
2685                    "snapshots", bdrv_get_device_or_node_name(bs));
2686         return ret;
2687     }
2688 
2689     /* Delete old snapshots of the same name */
2690     if (name) {
2691         ret = bdrv_all_delete_snapshot(name, &bs1, errp);
2692         if (ret < 0) {
2693             error_prepend(errp, "Error while deleting snapshot on device "
2694                           "'%s': ", bdrv_get_device_or_node_name(bs1));
2695             return ret;
2696         }
2697     }
2698 
2699     bs = bdrv_all_find_vmstate_bs();
2700     if (bs == NULL) {
2701         error_setg(errp, "No block device can accept snapshots");
2702         return ret;
2703     }
2704     aio_context = bdrv_get_aio_context(bs);
2705 
2706     saved_vm_running = runstate_is_running();
2707 
2708     ret = global_state_store();
2709     if (ret) {
2710         error_setg(errp, "Error saving global state");
2711         return ret;
2712     }
2713     vm_stop(RUN_STATE_SAVE_VM);
2714 
2715     bdrv_drain_all_begin();
2716 
2717     aio_context_acquire(aio_context);
2718 
2719     memset(sn, 0, sizeof(*sn));
2720 
2721     /* fill auxiliary fields */
2722     qemu_gettimeofday(&tv);
2723     sn->date_sec = tv.tv_sec;
2724     sn->date_nsec = tv.tv_usec * 1000;
2725     sn->vm_clock_nsec = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL);
2726 
2727     if (name) {
2728         ret = bdrv_snapshot_find(bs, old_sn, name);
2729         if (ret >= 0) {
2730             pstrcpy(sn->name, sizeof(sn->name), old_sn->name);
2731             pstrcpy(sn->id_str, sizeof(sn->id_str), old_sn->id_str);
2732         } else {
2733             pstrcpy(sn->name, sizeof(sn->name), name);
2734         }
2735     } else {
2736         /* cast below needed for OpenBSD where tv_sec is still 'long' */
2737         localtime_r((const time_t *)&tv.tv_sec, &tm);
2738         strftime(sn->name, sizeof(sn->name), "vm-%Y%m%d%H%M%S", &tm);
2739     }
2740 
2741     /* save the VM state */
2742     f = qemu_fopen_bdrv(bs, 1);
2743     if (!f) {
2744         error_setg(errp, "Could not open VM state file");
2745         goto the_end;
2746     }
2747     ret = qemu_savevm_state(f, errp);
2748     vm_state_size = qemu_ftell(f);
2749     ret2 = qemu_fclose(f);
2750     if (ret < 0) {
2751         goto the_end;
2752     }
2753     if (ret2 < 0) {
2754         ret = ret2;
2755         goto the_end;
2756     }
2757 
2758     /* The bdrv_all_create_snapshot() call that follows acquires the AioContext
2759      * for itself.  BDRV_POLL_WHILE() does not support nested locking because
2760      * it only releases the lock once.  Therefore synchronous I/O will deadlock
2761      * unless we release the AioContext before bdrv_all_create_snapshot().
2762      */
2763     aio_context_release(aio_context);
2764     aio_context = NULL;
2765 
2766     ret = bdrv_all_create_snapshot(sn, bs, vm_state_size, &bs);
2767     if (ret < 0) {
2768         error_setg(errp, "Error while creating snapshot on '%s'",
2769                    bdrv_get_device_or_node_name(bs));
2770         goto the_end;
2771     }
2772 
2773     ret = 0;
2774 
2775  the_end:
2776     if (aio_context) {
2777         aio_context_release(aio_context);
2778     }
2779 
2780     bdrv_drain_all_end();
2781 
2782     if (saved_vm_running) {
2783         vm_start();
2784     }
2785     return ret;
2786 }
2787 
2788 void qmp_xen_save_devices_state(const char *filename, bool has_live, bool live,
2789                                 Error **errp)
2790 {
2791     QEMUFile *f;
2792     QIOChannelFile *ioc;
2793     int saved_vm_running;
2794     int ret;
2795 
2796     if (!has_live) {
2797         /* live default to true so old version of Xen tool stack can have a
2798          * successfull live migration */
2799         live = true;
2800     }
2801 
2802     saved_vm_running = runstate_is_running();
2803     vm_stop(RUN_STATE_SAVE_VM);
2804     global_state_store_running();
2805 
2806     ioc = qio_channel_file_new_path(filename, O_WRONLY | O_CREAT, 0660, errp);
2807     if (!ioc) {
2808         goto the_end;
2809     }
2810     qio_channel_set_name(QIO_CHANNEL(ioc), "migration-xen-save-state");
2811     f = qemu_fopen_channel_output(QIO_CHANNEL(ioc));
2812     object_unref(OBJECT(ioc));
2813     ret = qemu_save_device_state(f);
2814     if (ret < 0 || qemu_fclose(f) < 0) {
2815         error_setg(errp, QERR_IO_ERROR);
2816     } else {
2817         /* libxl calls the QMP command "stop" before calling
2818          * "xen-save-devices-state" and in case of migration failure, libxl
2819          * would call "cont".
2820          * So call bdrv_inactivate_all (release locks) here to let the other
2821          * side of the migration take controle of the images.
2822          */
2823         if (live && !saved_vm_running) {
2824             ret = bdrv_inactivate_all();
2825             if (ret) {
2826                 error_setg(errp, "%s: bdrv_inactivate_all() failed (%d)",
2827                            __func__, ret);
2828             }
2829         }
2830     }
2831 
2832  the_end:
2833     if (saved_vm_running) {
2834         vm_start();
2835     }
2836 }
2837 
2838 void qmp_xen_load_devices_state(const char *filename, Error **errp)
2839 {
2840     QEMUFile *f;
2841     QIOChannelFile *ioc;
2842     int ret;
2843 
2844     /* Guest must be paused before loading the device state; the RAM state
2845      * will already have been loaded by xc
2846      */
2847     if (runstate_is_running()) {
2848         error_setg(errp, "Cannot update device state while vm is running");
2849         return;
2850     }
2851     vm_stop(RUN_STATE_RESTORE_VM);
2852 
2853     ioc = qio_channel_file_new_path(filename, O_RDONLY | O_BINARY, 0, errp);
2854     if (!ioc) {
2855         return;
2856     }
2857     qio_channel_set_name(QIO_CHANNEL(ioc), "migration-xen-load-state");
2858     f = qemu_fopen_channel_input(QIO_CHANNEL(ioc));
2859     object_unref(OBJECT(ioc));
2860 
2861     ret = qemu_loadvm_state(f);
2862     qemu_fclose(f);
2863     if (ret < 0) {
2864         error_setg(errp, QERR_IO_ERROR);
2865     }
2866     migration_incoming_state_destroy();
2867 }
2868 
2869 int load_snapshot(const char *name, Error **errp)
2870 {
2871     BlockDriverState *bs, *bs_vm_state;
2872     QEMUSnapshotInfo sn;
2873     QEMUFile *f;
2874     int ret;
2875     AioContext *aio_context;
2876     MigrationIncomingState *mis = migration_incoming_get_current();
2877 
2878     if (!replay_can_snapshot()) {
2879         error_setg(errp, "Record/replay does not allow loading snapshot "
2880                    "right now. Try once more later.");
2881         return -EINVAL;
2882     }
2883 
2884     if (!bdrv_all_can_snapshot(&bs)) {
2885         error_setg(errp,
2886                    "Device '%s' is writable but does not support snapshots",
2887                    bdrv_get_device_or_node_name(bs));
2888         return -ENOTSUP;
2889     }
2890     ret = bdrv_all_find_snapshot(name, &bs);
2891     if (ret < 0) {
2892         error_setg(errp,
2893                    "Device '%s' does not have the requested snapshot '%s'",
2894                    bdrv_get_device_or_node_name(bs), name);
2895         return ret;
2896     }
2897 
2898     bs_vm_state = bdrv_all_find_vmstate_bs();
2899     if (!bs_vm_state) {
2900         error_setg(errp, "No block device supports snapshots");
2901         return -ENOTSUP;
2902     }
2903     aio_context = bdrv_get_aio_context(bs_vm_state);
2904 
2905     /* Don't even try to load empty VM states */
2906     aio_context_acquire(aio_context);
2907     ret = bdrv_snapshot_find(bs_vm_state, &sn, name);
2908     aio_context_release(aio_context);
2909     if (ret < 0) {
2910         return ret;
2911     } else if (sn.vm_state_size == 0) {
2912         error_setg(errp, "This is a disk-only snapshot. Revert to it "
2913                    " offline using qemu-img");
2914         return -EINVAL;
2915     }
2916 
2917     /* Flush all IO requests so they don't interfere with the new state.  */
2918     bdrv_drain_all_begin();
2919 
2920     ret = bdrv_all_goto_snapshot(name, &bs, errp);
2921     if (ret < 0) {
2922         error_prepend(errp, "Could not load snapshot '%s' on '%s': ",
2923                       name, bdrv_get_device_or_node_name(bs));
2924         goto err_drain;
2925     }
2926 
2927     /* restore the VM state */
2928     f = qemu_fopen_bdrv(bs_vm_state, 0);
2929     if (!f) {
2930         error_setg(errp, "Could not open VM state file");
2931         ret = -EINVAL;
2932         goto err_drain;
2933     }
2934 
2935     qemu_system_reset(SHUTDOWN_CAUSE_NONE);
2936     mis->from_src_file = f;
2937 
2938     aio_context_acquire(aio_context);
2939     ret = qemu_loadvm_state(f);
2940     migration_incoming_state_destroy();
2941     aio_context_release(aio_context);
2942 
2943     bdrv_drain_all_end();
2944 
2945     if (ret < 0) {
2946         error_setg(errp, "Error %d while loading VM state", ret);
2947         return ret;
2948     }
2949 
2950     return 0;
2951 
2952 err_drain:
2953     bdrv_drain_all_end();
2954     return ret;
2955 }
2956 
2957 void vmstate_register_ram(MemoryRegion *mr, DeviceState *dev)
2958 {
2959     qemu_ram_set_idstr(mr->ram_block,
2960                        memory_region_name(mr), dev);
2961     qemu_ram_set_migratable(mr->ram_block);
2962 }
2963 
2964 void vmstate_unregister_ram(MemoryRegion *mr, DeviceState *dev)
2965 {
2966     qemu_ram_unset_idstr(mr->ram_block);
2967     qemu_ram_unset_migratable(mr->ram_block);
2968 }
2969 
2970 void vmstate_register_ram_global(MemoryRegion *mr)
2971 {
2972     vmstate_register_ram(mr, NULL);
2973 }
2974 
2975 bool vmstate_check_only_migratable(const VMStateDescription *vmsd)
2976 {
2977     /* check needed if --only-migratable is specified */
2978     if (!only_migratable) {
2979         return true;
2980     }
2981 
2982     return !(vmsd && vmsd->unmigratable);
2983 }
2984