xref: /openbmc/qemu/migration/savevm.c (revision c11b0583)
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 "config-host.h"
30 #include "qemu-common.h"
31 #include "hw/boards.h"
32 #include "hw/hw.h"
33 #include "hw/qdev.h"
34 #include "net/net.h"
35 #include "monitor/monitor.h"
36 #include "sysemu/sysemu.h"
37 #include "qemu/timer.h"
38 #include "audio/audio.h"
39 #include "migration/migration.h"
40 #include "qapi/qmp/qerror.h"
41 #include "qemu/error-report.h"
42 #include "qemu/sockets.h"
43 #include "qemu/queue.h"
44 #include "sysemu/cpus.h"
45 #include "exec/memory.h"
46 #include "qmp-commands.h"
47 #include "trace.h"
48 #include "qemu/iov.h"
49 #include "block/snapshot.h"
50 #include "block/qapi.h"
51 
52 
53 #ifndef ETH_P_RARP
54 #define ETH_P_RARP 0x8035
55 #endif
56 #define ARP_HTYPE_ETH 0x0001
57 #define ARP_PTYPE_IP 0x0800
58 #define ARP_OP_REQUEST_REV 0x3
59 
60 static bool skip_section_footers;
61 
62 static int announce_self_create(uint8_t *buf,
63                                 uint8_t *mac_addr)
64 {
65     /* Ethernet header. */
66     memset(buf, 0xff, 6);         /* destination MAC addr */
67     memcpy(buf + 6, mac_addr, 6); /* source MAC addr */
68     *(uint16_t *)(buf + 12) = htons(ETH_P_RARP); /* ethertype */
69 
70     /* RARP header. */
71     *(uint16_t *)(buf + 14) = htons(ARP_HTYPE_ETH); /* hardware addr space */
72     *(uint16_t *)(buf + 16) = htons(ARP_PTYPE_IP); /* protocol addr space */
73     *(buf + 18) = 6; /* hardware addr length (ethernet) */
74     *(buf + 19) = 4; /* protocol addr length (IPv4) */
75     *(uint16_t *)(buf + 20) = htons(ARP_OP_REQUEST_REV); /* opcode */
76     memcpy(buf + 22, mac_addr, 6); /* source hw addr */
77     memset(buf + 28, 0x00, 4);     /* source protocol addr */
78     memcpy(buf + 32, mac_addr, 6); /* target hw addr */
79     memset(buf + 38, 0x00, 4);     /* target protocol addr */
80 
81     /* Padding to get up to 60 bytes (ethernet min packet size, minus FCS). */
82     memset(buf + 42, 0x00, 18);
83 
84     return 60; /* len (FCS will be added by hardware) */
85 }
86 
87 static void qemu_announce_self_iter(NICState *nic, void *opaque)
88 {
89     uint8_t buf[60];
90     int len;
91 
92     trace_qemu_announce_self_iter(qemu_ether_ntoa(&nic->conf->macaddr));
93     len = announce_self_create(buf, nic->conf->macaddr.a);
94 
95     qemu_send_packet_raw(qemu_get_queue(nic), buf, len);
96 }
97 
98 
99 static void qemu_announce_self_once(void *opaque)
100 {
101     static int count = SELF_ANNOUNCE_ROUNDS;
102     QEMUTimer *timer = *(QEMUTimer **)opaque;
103 
104     qemu_foreach_nic(qemu_announce_self_iter, NULL);
105 
106     if (--count) {
107         /* delay 50ms, 150ms, 250ms, ... */
108         timer_mod(timer, qemu_clock_get_ms(QEMU_CLOCK_REALTIME) +
109                   self_announce_delay(count));
110     } else {
111             timer_del(timer);
112             timer_free(timer);
113     }
114 }
115 
116 void qemu_announce_self(void)
117 {
118     static QEMUTimer *timer;
119     timer = timer_new_ms(QEMU_CLOCK_REALTIME, qemu_announce_self_once, &timer);
120     qemu_announce_self_once(&timer);
121 }
122 
123 /***********************************************************/
124 /* savevm/loadvm support */
125 
126 static ssize_t block_writev_buffer(void *opaque, struct iovec *iov, int iovcnt,
127                                    int64_t pos)
128 {
129     int ret;
130     QEMUIOVector qiov;
131 
132     qemu_iovec_init_external(&qiov, iov, iovcnt);
133     ret = bdrv_writev_vmstate(opaque, &qiov, pos);
134     if (ret < 0) {
135         return ret;
136     }
137 
138     return qiov.size;
139 }
140 
141 static int block_put_buffer(void *opaque, const uint8_t *buf,
142                            int64_t pos, int size)
143 {
144     bdrv_save_vmstate(opaque, buf, pos, size);
145     return size;
146 }
147 
148 static int block_get_buffer(void *opaque, uint8_t *buf, int64_t pos, int size)
149 {
150     return bdrv_load_vmstate(opaque, buf, pos, size);
151 }
152 
153 static int bdrv_fclose(void *opaque)
154 {
155     return bdrv_flush(opaque);
156 }
157 
158 static const QEMUFileOps bdrv_read_ops = {
159     .get_buffer = block_get_buffer,
160     .close =      bdrv_fclose
161 };
162 
163 static const QEMUFileOps bdrv_write_ops = {
164     .put_buffer     = block_put_buffer,
165     .writev_buffer  = block_writev_buffer,
166     .close          = bdrv_fclose
167 };
168 
169 static QEMUFile *qemu_fopen_bdrv(BlockDriverState *bs, int is_writable)
170 {
171     if (is_writable) {
172         return qemu_fopen_ops(bs, &bdrv_write_ops);
173     }
174     return qemu_fopen_ops(bs, &bdrv_read_ops);
175 }
176 
177 
178 /* QEMUFile timer support.
179  * Not in qemu-file.c to not add qemu-timer.c as dependency to qemu-file.c
180  */
181 
182 void timer_put(QEMUFile *f, QEMUTimer *ts)
183 {
184     uint64_t expire_time;
185 
186     expire_time = timer_expire_time_ns(ts);
187     qemu_put_be64(f, expire_time);
188 }
189 
190 void timer_get(QEMUFile *f, QEMUTimer *ts)
191 {
192     uint64_t expire_time;
193 
194     expire_time = qemu_get_be64(f);
195     if (expire_time != -1) {
196         timer_mod_ns(ts, expire_time);
197     } else {
198         timer_del(ts);
199     }
200 }
201 
202 
203 /* VMState timer support.
204  * Not in vmstate.c to not add qemu-timer.c as dependency to vmstate.c
205  */
206 
207 static int get_timer(QEMUFile *f, void *pv, size_t size)
208 {
209     QEMUTimer *v = pv;
210     timer_get(f, v);
211     return 0;
212 }
213 
214 static void put_timer(QEMUFile *f, void *pv, size_t size)
215 {
216     QEMUTimer *v = pv;
217     timer_put(f, v);
218 }
219 
220 const VMStateInfo vmstate_info_timer = {
221     .name = "timer",
222     .get  = get_timer,
223     .put  = put_timer,
224 };
225 
226 
227 typedef struct CompatEntry {
228     char idstr[256];
229     int instance_id;
230 } CompatEntry;
231 
232 typedef struct SaveStateEntry {
233     QTAILQ_ENTRY(SaveStateEntry) entry;
234     char idstr[256];
235     int instance_id;
236     int alias_id;
237     int version_id;
238     int section_id;
239     SaveVMHandlers *ops;
240     const VMStateDescription *vmsd;
241     void *opaque;
242     CompatEntry *compat;
243     int is_ram;
244 } SaveStateEntry;
245 
246 typedef struct SaveState {
247     QTAILQ_HEAD(, SaveStateEntry) handlers;
248     int global_section_id;
249     bool skip_configuration;
250     uint32_t len;
251     const char *name;
252 } SaveState;
253 
254 static SaveState savevm_state = {
255     .handlers = QTAILQ_HEAD_INITIALIZER(savevm_state.handlers),
256     .global_section_id = 0,
257     .skip_configuration = false,
258 };
259 
260 void savevm_skip_configuration(void)
261 {
262     savevm_state.skip_configuration = true;
263 }
264 
265 
266 static void configuration_pre_save(void *opaque)
267 {
268     SaveState *state = opaque;
269     const char *current_name = MACHINE_GET_CLASS(current_machine)->name;
270 
271     state->len = strlen(current_name);
272     state->name = current_name;
273 }
274 
275 static int configuration_post_load(void *opaque, int version_id)
276 {
277     SaveState *state = opaque;
278     const char *current_name = MACHINE_GET_CLASS(current_machine)->name;
279 
280     if (strncmp(state->name, current_name, state->len) != 0) {
281         error_report("Machine type received is '%s' and local is '%s'",
282                      state->name, current_name);
283         return -EINVAL;
284     }
285     return 0;
286 }
287 
288 static const VMStateDescription vmstate_configuration = {
289     .name = "configuration",
290     .version_id = 1,
291     .post_load = configuration_post_load,
292     .pre_save = configuration_pre_save,
293     .fields = (VMStateField[]) {
294         VMSTATE_UINT32(len, SaveState),
295         VMSTATE_VBUFFER_ALLOC_UINT32(name, SaveState, 0, NULL, 0, len),
296         VMSTATE_END_OF_LIST()
297     },
298 };
299 
300 static void dump_vmstate_vmsd(FILE *out_file,
301                               const VMStateDescription *vmsd, int indent,
302                               bool is_subsection);
303 
304 static void dump_vmstate_vmsf(FILE *out_file, const VMStateField *field,
305                               int indent)
306 {
307     fprintf(out_file, "%*s{\n", indent, "");
308     indent += 2;
309     fprintf(out_file, "%*s\"field\": \"%s\",\n", indent, "", field->name);
310     fprintf(out_file, "%*s\"version_id\": %d,\n", indent, "",
311             field->version_id);
312     fprintf(out_file, "%*s\"field_exists\": %s,\n", indent, "",
313             field->field_exists ? "true" : "false");
314     fprintf(out_file, "%*s\"size\": %zu", indent, "", field->size);
315     if (field->vmsd != NULL) {
316         fprintf(out_file, ",\n");
317         dump_vmstate_vmsd(out_file, field->vmsd, indent, false);
318     }
319     fprintf(out_file, "\n%*s}", indent - 2, "");
320 }
321 
322 static void dump_vmstate_vmss(FILE *out_file,
323                               const VMStateDescription **subsection,
324                               int indent)
325 {
326     if (*subsection != NULL) {
327         dump_vmstate_vmsd(out_file, *subsection, indent, true);
328     }
329 }
330 
331 static void dump_vmstate_vmsd(FILE *out_file,
332                               const VMStateDescription *vmsd, int indent,
333                               bool is_subsection)
334 {
335     if (is_subsection) {
336         fprintf(out_file, "%*s{\n", indent, "");
337     } else {
338         fprintf(out_file, "%*s\"%s\": {\n", indent, "", "Description");
339     }
340     indent += 2;
341     fprintf(out_file, "%*s\"name\": \"%s\",\n", indent, "", vmsd->name);
342     fprintf(out_file, "%*s\"version_id\": %d,\n", indent, "",
343             vmsd->version_id);
344     fprintf(out_file, "%*s\"minimum_version_id\": %d", indent, "",
345             vmsd->minimum_version_id);
346     if (vmsd->fields != NULL) {
347         const VMStateField *field = vmsd->fields;
348         bool first;
349 
350         fprintf(out_file, ",\n%*s\"Fields\": [\n", indent, "");
351         first = true;
352         while (field->name != NULL) {
353             if (field->flags & VMS_MUST_EXIST) {
354                 /* Ignore VMSTATE_VALIDATE bits; these don't get migrated */
355                 field++;
356                 continue;
357             }
358             if (!first) {
359                 fprintf(out_file, ",\n");
360             }
361             dump_vmstate_vmsf(out_file, field, indent + 2);
362             field++;
363             first = false;
364         }
365         fprintf(out_file, "\n%*s]", indent, "");
366     }
367     if (vmsd->subsections != NULL) {
368         const VMStateDescription **subsection = vmsd->subsections;
369         bool first;
370 
371         fprintf(out_file, ",\n%*s\"Subsections\": [\n", indent, "");
372         first = true;
373         while (*subsection != NULL) {
374             if (!first) {
375                 fprintf(out_file, ",\n");
376             }
377             dump_vmstate_vmss(out_file, subsection, indent + 2);
378             subsection++;
379             first = false;
380         }
381         fprintf(out_file, "\n%*s]", indent, "");
382     }
383     fprintf(out_file, "\n%*s}", indent - 2, "");
384 }
385 
386 static void dump_machine_type(FILE *out_file)
387 {
388     MachineClass *mc;
389 
390     mc = MACHINE_GET_CLASS(current_machine);
391 
392     fprintf(out_file, "  \"vmschkmachine\": {\n");
393     fprintf(out_file, "    \"Name\": \"%s\"\n", mc->name);
394     fprintf(out_file, "  },\n");
395 }
396 
397 void dump_vmstate_json_to_file(FILE *out_file)
398 {
399     GSList *list, *elt;
400     bool first;
401 
402     fprintf(out_file, "{\n");
403     dump_machine_type(out_file);
404 
405     first = true;
406     list = object_class_get_list(TYPE_DEVICE, true);
407     for (elt = list; elt; elt = elt->next) {
408         DeviceClass *dc = OBJECT_CLASS_CHECK(DeviceClass, elt->data,
409                                              TYPE_DEVICE);
410         const char *name;
411         int indent = 2;
412 
413         if (!dc->vmsd) {
414             continue;
415         }
416 
417         if (!first) {
418             fprintf(out_file, ",\n");
419         }
420         name = object_class_get_name(OBJECT_CLASS(dc));
421         fprintf(out_file, "%*s\"%s\": {\n", indent, "", name);
422         indent += 2;
423         fprintf(out_file, "%*s\"Name\": \"%s\",\n", indent, "", name);
424         fprintf(out_file, "%*s\"version_id\": %d,\n", indent, "",
425                 dc->vmsd->version_id);
426         fprintf(out_file, "%*s\"minimum_version_id\": %d,\n", indent, "",
427                 dc->vmsd->minimum_version_id);
428 
429         dump_vmstate_vmsd(out_file, dc->vmsd, indent, false);
430 
431         fprintf(out_file, "\n%*s}", indent - 2, "");
432         first = false;
433     }
434     fprintf(out_file, "\n}\n");
435     fclose(out_file);
436 }
437 
438 static int calculate_new_instance_id(const char *idstr)
439 {
440     SaveStateEntry *se;
441     int instance_id = 0;
442 
443     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
444         if (strcmp(idstr, se->idstr) == 0
445             && instance_id <= se->instance_id) {
446             instance_id = se->instance_id + 1;
447         }
448     }
449     return instance_id;
450 }
451 
452 static int calculate_compat_instance_id(const char *idstr)
453 {
454     SaveStateEntry *se;
455     int instance_id = 0;
456 
457     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
458         if (!se->compat) {
459             continue;
460         }
461 
462         if (strcmp(idstr, se->compat->idstr) == 0
463             && instance_id <= se->compat->instance_id) {
464             instance_id = se->compat->instance_id + 1;
465         }
466     }
467     return instance_id;
468 }
469 
470 /* TODO: Individual devices generally have very little idea about the rest
471    of the system, so instance_id should be removed/replaced.
472    Meanwhile pass -1 as instance_id if you do not already have a clearly
473    distinguishing id for all instances of your device class. */
474 int register_savevm_live(DeviceState *dev,
475                          const char *idstr,
476                          int instance_id,
477                          int version_id,
478                          SaveVMHandlers *ops,
479                          void *opaque)
480 {
481     SaveStateEntry *se;
482 
483     se = g_malloc0(sizeof(SaveStateEntry));
484     se->version_id = version_id;
485     se->section_id = savevm_state.global_section_id++;
486     se->ops = ops;
487     se->opaque = opaque;
488     se->vmsd = NULL;
489     /* if this is a live_savem then set is_ram */
490     if (ops->save_live_setup != NULL) {
491         se->is_ram = 1;
492     }
493 
494     if (dev) {
495         char *id = qdev_get_dev_path(dev);
496         if (id) {
497             pstrcpy(se->idstr, sizeof(se->idstr), id);
498             pstrcat(se->idstr, sizeof(se->idstr), "/");
499             g_free(id);
500 
501             se->compat = g_malloc0(sizeof(CompatEntry));
502             pstrcpy(se->compat->idstr, sizeof(se->compat->idstr), idstr);
503             se->compat->instance_id = instance_id == -1 ?
504                          calculate_compat_instance_id(idstr) : instance_id;
505             instance_id = -1;
506         }
507     }
508     pstrcat(se->idstr, sizeof(se->idstr), idstr);
509 
510     if (instance_id == -1) {
511         se->instance_id = calculate_new_instance_id(se->idstr);
512     } else {
513         se->instance_id = instance_id;
514     }
515     assert(!se->compat || se->instance_id == 0);
516     /* add at the end of list */
517     QTAILQ_INSERT_TAIL(&savevm_state.handlers, se, entry);
518     return 0;
519 }
520 
521 int register_savevm(DeviceState *dev,
522                     const char *idstr,
523                     int instance_id,
524                     int version_id,
525                     SaveStateHandler *save_state,
526                     LoadStateHandler *load_state,
527                     void *opaque)
528 {
529     SaveVMHandlers *ops = g_malloc0(sizeof(SaveVMHandlers));
530     ops->save_state = save_state;
531     ops->load_state = load_state;
532     return register_savevm_live(dev, idstr, instance_id, version_id,
533                                 ops, opaque);
534 }
535 
536 void unregister_savevm(DeviceState *dev, const char *idstr, void *opaque)
537 {
538     SaveStateEntry *se, *new_se;
539     char id[256] = "";
540 
541     if (dev) {
542         char *path = qdev_get_dev_path(dev);
543         if (path) {
544             pstrcpy(id, sizeof(id), path);
545             pstrcat(id, sizeof(id), "/");
546             g_free(path);
547         }
548     }
549     pstrcat(id, sizeof(id), idstr);
550 
551     QTAILQ_FOREACH_SAFE(se, &savevm_state.handlers, entry, new_se) {
552         if (strcmp(se->idstr, id) == 0 && se->opaque == opaque) {
553             QTAILQ_REMOVE(&savevm_state.handlers, se, entry);
554             g_free(se->compat);
555             g_free(se->ops);
556             g_free(se);
557         }
558     }
559 }
560 
561 int vmstate_register_with_alias_id(DeviceState *dev, int instance_id,
562                                    const VMStateDescription *vmsd,
563                                    void *opaque, int alias_id,
564                                    int required_for_version)
565 {
566     SaveStateEntry *se;
567 
568     /* If this triggers, alias support can be dropped for the vmsd. */
569     assert(alias_id == -1 || required_for_version >= vmsd->minimum_version_id);
570 
571     se = g_malloc0(sizeof(SaveStateEntry));
572     se->version_id = vmsd->version_id;
573     se->section_id = savevm_state.global_section_id++;
574     se->opaque = opaque;
575     se->vmsd = vmsd;
576     se->alias_id = alias_id;
577 
578     if (dev) {
579         char *id = qdev_get_dev_path(dev);
580         if (id) {
581             pstrcpy(se->idstr, sizeof(se->idstr), id);
582             pstrcat(se->idstr, sizeof(se->idstr), "/");
583             g_free(id);
584 
585             se->compat = g_malloc0(sizeof(CompatEntry));
586             pstrcpy(se->compat->idstr, sizeof(se->compat->idstr), vmsd->name);
587             se->compat->instance_id = instance_id == -1 ?
588                          calculate_compat_instance_id(vmsd->name) : instance_id;
589             instance_id = -1;
590         }
591     }
592     pstrcat(se->idstr, sizeof(se->idstr), vmsd->name);
593 
594     if (instance_id == -1) {
595         se->instance_id = calculate_new_instance_id(se->idstr);
596     } else {
597         se->instance_id = instance_id;
598     }
599     assert(!se->compat || se->instance_id == 0);
600     /* add at the end of list */
601     QTAILQ_INSERT_TAIL(&savevm_state.handlers, se, entry);
602     return 0;
603 }
604 
605 void vmstate_unregister(DeviceState *dev, const VMStateDescription *vmsd,
606                         void *opaque)
607 {
608     SaveStateEntry *se, *new_se;
609 
610     QTAILQ_FOREACH_SAFE(se, &savevm_state.handlers, entry, new_se) {
611         if (se->vmsd == vmsd && se->opaque == opaque) {
612             QTAILQ_REMOVE(&savevm_state.handlers, se, entry);
613             g_free(se->compat);
614             g_free(se);
615         }
616     }
617 }
618 
619 static int vmstate_load(QEMUFile *f, SaveStateEntry *se, int version_id)
620 {
621     trace_vmstate_load(se->idstr, se->vmsd ? se->vmsd->name : "(old)");
622     if (!se->vmsd) {         /* Old style */
623         return se->ops->load_state(f, se->opaque, version_id);
624     }
625     return vmstate_load_state(f, se->vmsd, se->opaque, version_id);
626 }
627 
628 static void vmstate_save_old_style(QEMUFile *f, SaveStateEntry *se, QJSON *vmdesc)
629 {
630     int64_t old_offset, size;
631 
632     old_offset = qemu_ftell_fast(f);
633     se->ops->save_state(f, se->opaque);
634     size = qemu_ftell_fast(f) - old_offset;
635 
636     if (vmdesc) {
637         json_prop_int(vmdesc, "size", size);
638         json_start_array(vmdesc, "fields");
639         json_start_object(vmdesc, NULL);
640         json_prop_str(vmdesc, "name", "data");
641         json_prop_int(vmdesc, "size", size);
642         json_prop_str(vmdesc, "type", "buffer");
643         json_end_object(vmdesc);
644         json_end_array(vmdesc);
645     }
646 }
647 
648 static void vmstate_save(QEMUFile *f, SaveStateEntry *se, QJSON *vmdesc)
649 {
650     trace_vmstate_save(se->idstr, se->vmsd ? se->vmsd->name : "(old)");
651     if (!se->vmsd) {
652         vmstate_save_old_style(f, se, vmdesc);
653         return;
654     }
655     vmstate_save_state(f, se->vmsd, se->opaque, vmdesc);
656 }
657 
658 void savevm_skip_section_footers(void)
659 {
660     skip_section_footers = true;
661 }
662 
663 /*
664  * Write the header for device section (QEMU_VM_SECTION START/END/PART/FULL)
665  */
666 static void save_section_header(QEMUFile *f, SaveStateEntry *se,
667                                 uint8_t section_type)
668 {
669     qemu_put_byte(f, section_type);
670     qemu_put_be32(f, se->section_id);
671 
672     if (section_type == QEMU_VM_SECTION_FULL ||
673         section_type == QEMU_VM_SECTION_START) {
674         /* ID string */
675         size_t len = strlen(se->idstr);
676         qemu_put_byte(f, len);
677         qemu_put_buffer(f, (uint8_t *)se->idstr, len);
678 
679         qemu_put_be32(f, se->instance_id);
680         qemu_put_be32(f, se->version_id);
681     }
682 }
683 
684 /*
685  * Write a footer onto device sections that catches cases misformatted device
686  * sections.
687  */
688 static void save_section_footer(QEMUFile *f, SaveStateEntry *se)
689 {
690     if (!skip_section_footers) {
691         qemu_put_byte(f, QEMU_VM_SECTION_FOOTER);
692         qemu_put_be32(f, se->section_id);
693     }
694 }
695 
696 bool qemu_savevm_state_blocked(Error **errp)
697 {
698     SaveStateEntry *se;
699 
700     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
701         if (se->vmsd && se->vmsd->unmigratable) {
702             error_setg(errp, "State blocked by non-migratable device '%s'",
703                        se->idstr);
704             return true;
705         }
706     }
707     return false;
708 }
709 
710 void qemu_savevm_state_header(QEMUFile *f)
711 {
712     trace_savevm_state_header();
713     qemu_put_be32(f, QEMU_VM_FILE_MAGIC);
714     qemu_put_be32(f, QEMU_VM_FILE_VERSION);
715 }
716 
717 void qemu_savevm_state_begin(QEMUFile *f,
718                              const MigrationParams *params)
719 {
720     SaveStateEntry *se;
721     int ret;
722 
723     trace_savevm_state_begin();
724     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
725         if (!se->ops || !se->ops->set_params) {
726             continue;
727         }
728         se->ops->set_params(params, se->opaque);
729     }
730 
731     if (!savevm_state.skip_configuration) {
732         qemu_put_byte(f, QEMU_VM_CONFIGURATION);
733         vmstate_save_state(f, &vmstate_configuration, &savevm_state, 0);
734     }
735 
736     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
737         if (!se->ops || !se->ops->save_live_setup) {
738             continue;
739         }
740         if (se->ops && se->ops->is_active) {
741             if (!se->ops->is_active(se->opaque)) {
742                 continue;
743             }
744         }
745         save_section_header(f, se, QEMU_VM_SECTION_START);
746 
747         ret = se->ops->save_live_setup(f, se->opaque);
748         save_section_footer(f, se);
749         if (ret < 0) {
750             qemu_file_set_error(f, ret);
751             break;
752         }
753     }
754 }
755 
756 /*
757  * this function has three return values:
758  *   negative: there was one error, and we have -errno.
759  *   0 : We haven't finished, caller have to go again
760  *   1 : We have finished, we can go to complete phase
761  */
762 int qemu_savevm_state_iterate(QEMUFile *f)
763 {
764     SaveStateEntry *se;
765     int ret = 1;
766 
767     trace_savevm_state_iterate();
768     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
769         if (!se->ops || !se->ops->save_live_iterate) {
770             continue;
771         }
772         if (se->ops && se->ops->is_active) {
773             if (!se->ops->is_active(se->opaque)) {
774                 continue;
775             }
776         }
777         if (qemu_file_rate_limit(f)) {
778             return 0;
779         }
780         trace_savevm_section_start(se->idstr, se->section_id);
781 
782         save_section_header(f, se, QEMU_VM_SECTION_PART);
783 
784         ret = se->ops->save_live_iterate(f, se->opaque);
785         trace_savevm_section_end(se->idstr, se->section_id, ret);
786         save_section_footer(f, se);
787 
788         if (ret < 0) {
789             qemu_file_set_error(f, ret);
790         }
791         if (ret <= 0) {
792             /* Do not proceed to the next vmstate before this one reported
793                completion of the current stage. This serializes the migration
794                and reduces the probability that a faster changing state is
795                synchronized over and over again. */
796             break;
797         }
798     }
799     return ret;
800 }
801 
802 static bool should_send_vmdesc(void)
803 {
804     MachineState *machine = MACHINE(qdev_get_machine());
805     return !machine->suppress_vmdesc;
806 }
807 
808 void qemu_savevm_state_complete(QEMUFile *f)
809 {
810     QJSON *vmdesc;
811     int vmdesc_len;
812     SaveStateEntry *se;
813     int ret;
814 
815     trace_savevm_state_complete();
816 
817     cpu_synchronize_all_states();
818 
819     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
820         if (!se->ops || !se->ops->save_live_complete) {
821             continue;
822         }
823         if (se->ops && se->ops->is_active) {
824             if (!se->ops->is_active(se->opaque)) {
825                 continue;
826             }
827         }
828         trace_savevm_section_start(se->idstr, se->section_id);
829 
830         save_section_header(f, se, QEMU_VM_SECTION_END);
831 
832         ret = se->ops->save_live_complete(f, se->opaque);
833         trace_savevm_section_end(se->idstr, se->section_id, ret);
834         save_section_footer(f, se);
835         if (ret < 0) {
836             qemu_file_set_error(f, ret);
837             return;
838         }
839     }
840 
841     vmdesc = qjson_new();
842     json_prop_int(vmdesc, "page_size", TARGET_PAGE_SIZE);
843     json_start_array(vmdesc, "devices");
844     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
845 
846         if ((!se->ops || !se->ops->save_state) && !se->vmsd) {
847             continue;
848         }
849         if (se->vmsd && !vmstate_save_needed(se->vmsd, se->opaque)) {
850             trace_savevm_section_skip(se->idstr, se->section_id);
851             continue;
852         }
853 
854         trace_savevm_section_start(se->idstr, se->section_id);
855 
856         json_start_object(vmdesc, NULL);
857         json_prop_str(vmdesc, "name", se->idstr);
858         json_prop_int(vmdesc, "instance_id", se->instance_id);
859 
860         save_section_header(f, se, QEMU_VM_SECTION_FULL);
861 
862         vmstate_save(f, se, vmdesc);
863 
864         json_end_object(vmdesc);
865         trace_savevm_section_end(se->idstr, se->section_id, 0);
866         save_section_footer(f, se);
867     }
868 
869     qemu_put_byte(f, QEMU_VM_EOF);
870 
871     json_end_array(vmdesc);
872     qjson_finish(vmdesc);
873     vmdesc_len = strlen(qjson_get_str(vmdesc));
874 
875     if (should_send_vmdesc()) {
876         qemu_put_byte(f, QEMU_VM_VMDESCRIPTION);
877         qemu_put_be32(f, vmdesc_len);
878         qemu_put_buffer(f, (uint8_t *)qjson_get_str(vmdesc), vmdesc_len);
879     }
880     object_unref(OBJECT(vmdesc));
881 
882     qemu_fflush(f);
883 }
884 
885 uint64_t qemu_savevm_state_pending(QEMUFile *f, uint64_t max_size)
886 {
887     SaveStateEntry *se;
888     uint64_t ret = 0;
889 
890     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
891         if (!se->ops || !se->ops->save_live_pending) {
892             continue;
893         }
894         if (se->ops && se->ops->is_active) {
895             if (!se->ops->is_active(se->opaque)) {
896                 continue;
897             }
898         }
899         ret += se->ops->save_live_pending(f, se->opaque, max_size);
900     }
901     return ret;
902 }
903 
904 void qemu_savevm_state_cancel(void)
905 {
906     SaveStateEntry *se;
907 
908     trace_savevm_state_cancel();
909     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
910         if (se->ops && se->ops->cancel) {
911             se->ops->cancel(se->opaque);
912         }
913     }
914 }
915 
916 static int qemu_savevm_state(QEMUFile *f, Error **errp)
917 {
918     int ret;
919     MigrationParams params = {
920         .blk = 0,
921         .shared = 0
922     };
923 
924     if (qemu_savevm_state_blocked(errp)) {
925         return -EINVAL;
926     }
927 
928     qemu_mutex_unlock_iothread();
929     qemu_savevm_state_header(f);
930     qemu_savevm_state_begin(f, &params);
931     qemu_mutex_lock_iothread();
932 
933     while (qemu_file_get_error(f) == 0) {
934         if (qemu_savevm_state_iterate(f) > 0) {
935             break;
936         }
937     }
938 
939     ret = qemu_file_get_error(f);
940     if (ret == 0) {
941         qemu_savevm_state_complete(f);
942         ret = qemu_file_get_error(f);
943     }
944     if (ret != 0) {
945         qemu_savevm_state_cancel();
946         error_setg_errno(errp, -ret, "Error while writing VM state");
947     }
948     return ret;
949 }
950 
951 static int qemu_save_device_state(QEMUFile *f)
952 {
953     SaveStateEntry *se;
954 
955     qemu_put_be32(f, QEMU_VM_FILE_MAGIC);
956     qemu_put_be32(f, QEMU_VM_FILE_VERSION);
957 
958     cpu_synchronize_all_states();
959 
960     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
961         if (se->is_ram) {
962             continue;
963         }
964         if ((!se->ops || !se->ops->save_state) && !se->vmsd) {
965             continue;
966         }
967         if (se->vmsd && !vmstate_save_needed(se->vmsd, se->opaque)) {
968             continue;
969         }
970 
971         save_section_header(f, se, QEMU_VM_SECTION_FULL);
972 
973         vmstate_save(f, se, NULL);
974 
975         save_section_footer(f, se);
976     }
977 
978     qemu_put_byte(f, QEMU_VM_EOF);
979 
980     return qemu_file_get_error(f);
981 }
982 
983 static SaveStateEntry *find_se(const char *idstr, int instance_id)
984 {
985     SaveStateEntry *se;
986 
987     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
988         if (!strcmp(se->idstr, idstr) &&
989             (instance_id == se->instance_id ||
990              instance_id == se->alias_id))
991             return se;
992         /* Migrating from an older version? */
993         if (strstr(se->idstr, idstr) && se->compat) {
994             if (!strcmp(se->compat->idstr, idstr) &&
995                 (instance_id == se->compat->instance_id ||
996                  instance_id == se->alias_id))
997                 return se;
998         }
999     }
1000     return NULL;
1001 }
1002 
1003 struct LoadStateEntry {
1004     QLIST_ENTRY(LoadStateEntry) entry;
1005     SaveStateEntry *se;
1006     int section_id;
1007     int version_id;
1008 };
1009 
1010 /*
1011  * Read a footer off the wire and check that it matches the expected section
1012  *
1013  * Returns: true if the footer was good
1014  *          false if there is a problem (and calls error_report to say why)
1015  */
1016 static bool check_section_footer(QEMUFile *f, LoadStateEntry *le)
1017 {
1018     uint8_t read_mark;
1019     uint32_t read_section_id;
1020 
1021     if (skip_section_footers) {
1022         /* No footer to check */
1023         return true;
1024     }
1025 
1026     read_mark = qemu_get_byte(f);
1027 
1028     if (read_mark != QEMU_VM_SECTION_FOOTER) {
1029         error_report("Missing section footer for %s", le->se->idstr);
1030         return false;
1031     }
1032 
1033     read_section_id = qemu_get_be32(f);
1034     if (read_section_id != le->section_id) {
1035         error_report("Mismatched section id in footer for %s -"
1036                      " read 0x%x expected 0x%x",
1037                      le->se->idstr, read_section_id, le->section_id);
1038         return false;
1039     }
1040 
1041     /* All good */
1042     return true;
1043 }
1044 
1045 void loadvm_free_handlers(MigrationIncomingState *mis)
1046 {
1047     LoadStateEntry *le, *new_le;
1048 
1049     QLIST_FOREACH_SAFE(le, &mis->loadvm_handlers, entry, new_le) {
1050         QLIST_REMOVE(le, entry);
1051         g_free(le);
1052     }
1053 }
1054 
1055 int qemu_loadvm_state(QEMUFile *f)
1056 {
1057     MigrationIncomingState *mis = migration_incoming_get_current();
1058     Error *local_err = NULL;
1059     uint8_t section_type;
1060     unsigned int v;
1061     int ret;
1062     int file_error_after_eof = -1;
1063 
1064     if (qemu_savevm_state_blocked(&local_err)) {
1065         error_report_err(local_err);
1066         return -EINVAL;
1067     }
1068 
1069     v = qemu_get_be32(f);
1070     if (v != QEMU_VM_FILE_MAGIC) {
1071         error_report("Not a migration stream");
1072         return -EINVAL;
1073     }
1074 
1075     v = qemu_get_be32(f);
1076     if (v == QEMU_VM_FILE_VERSION_COMPAT) {
1077         error_report("SaveVM v2 format is obsolete and don't work anymore");
1078         return -ENOTSUP;
1079     }
1080     if (v != QEMU_VM_FILE_VERSION) {
1081         error_report("Unsupported migration stream version");
1082         return -ENOTSUP;
1083     }
1084 
1085     if (!savevm_state.skip_configuration) {
1086         if (qemu_get_byte(f) != QEMU_VM_CONFIGURATION) {
1087             error_report("Configuration section missing");
1088             return -EINVAL;
1089         }
1090         ret = vmstate_load_state(f, &vmstate_configuration, &savevm_state, 0);
1091 
1092         if (ret) {
1093             return ret;
1094         }
1095     }
1096 
1097     while ((section_type = qemu_get_byte(f)) != QEMU_VM_EOF) {
1098         uint32_t instance_id, version_id, section_id;
1099         SaveStateEntry *se;
1100         LoadStateEntry *le;
1101         char idstr[256];
1102 
1103         trace_qemu_loadvm_state_section(section_type);
1104         switch (section_type) {
1105         case QEMU_VM_SECTION_START:
1106         case QEMU_VM_SECTION_FULL:
1107             /* Read section start */
1108             section_id = qemu_get_be32(f);
1109             if (!qemu_get_counted_string(f, idstr)) {
1110                 error_report("Unable to read ID string for section %u",
1111                             section_id);
1112                 return -EINVAL;
1113             }
1114             instance_id = qemu_get_be32(f);
1115             version_id = qemu_get_be32(f);
1116 
1117             trace_qemu_loadvm_state_section_startfull(section_id, idstr,
1118                                                       instance_id, version_id);
1119             /* Find savevm section */
1120             se = find_se(idstr, instance_id);
1121             if (se == NULL) {
1122                 error_report("Unknown savevm section or instance '%s' %d",
1123                              idstr, instance_id);
1124                 ret = -EINVAL;
1125                 goto out;
1126             }
1127 
1128             /* Validate version */
1129             if (version_id > se->version_id) {
1130                 error_report("savevm: unsupported version %d for '%s' v%d",
1131                              version_id, idstr, se->version_id);
1132                 ret = -EINVAL;
1133                 goto out;
1134             }
1135 
1136             /* Add entry */
1137             le = g_malloc0(sizeof(*le));
1138 
1139             le->se = se;
1140             le->section_id = section_id;
1141             le->version_id = version_id;
1142             QLIST_INSERT_HEAD(&mis->loadvm_handlers, le, entry);
1143 
1144             ret = vmstate_load(f, le->se, le->version_id);
1145             if (ret < 0) {
1146                 error_report("error while loading state for instance 0x%x of"
1147                              " device '%s'", instance_id, idstr);
1148                 goto out;
1149             }
1150             if (!check_section_footer(f, le)) {
1151                 ret = -EINVAL;
1152                 goto out;
1153             }
1154             break;
1155         case QEMU_VM_SECTION_PART:
1156         case QEMU_VM_SECTION_END:
1157             section_id = qemu_get_be32(f);
1158 
1159             trace_qemu_loadvm_state_section_partend(section_id);
1160             QLIST_FOREACH(le, &mis->loadvm_handlers, entry) {
1161                 if (le->section_id == section_id) {
1162                     break;
1163                 }
1164             }
1165             if (le == NULL) {
1166                 error_report("Unknown savevm section %d", section_id);
1167                 ret = -EINVAL;
1168                 goto out;
1169             }
1170 
1171             ret = vmstate_load(f, le->se, le->version_id);
1172             if (ret < 0) {
1173                 error_report("error while loading state section id %d(%s)",
1174                              section_id, le->se->idstr);
1175                 goto out;
1176             }
1177             if (!check_section_footer(f, le)) {
1178                 ret = -EINVAL;
1179                 goto out;
1180             }
1181             break;
1182         default:
1183             error_report("Unknown savevm section type %d", section_type);
1184             ret = -EINVAL;
1185             goto out;
1186         }
1187     }
1188 
1189     file_error_after_eof = qemu_file_get_error(f);
1190 
1191     /*
1192      * Try to read in the VMDESC section as well, so that dumping tools that
1193      * intercept our migration stream have the chance to see it.
1194      */
1195 
1196     /* We've got to be careful; if we don't read the data and just shut the fd
1197      * then the sender can error if we close while it's still sending.
1198      * We also mustn't read data that isn't there; some transports (RDMA)
1199      * will stall waiting for that data when the source has already closed.
1200      */
1201     if (should_send_vmdesc()) {
1202         uint8_t *buf;
1203         uint32_t size;
1204         section_type = qemu_get_byte(f);
1205 
1206         if (section_type != QEMU_VM_VMDESCRIPTION) {
1207             error_report("Expected vmdescription section, but got %d",
1208                          section_type);
1209             /*
1210              * It doesn't seem worth failing at this point since
1211              * we apparently have an otherwise valid VM state
1212              */
1213         } else {
1214             buf = g_malloc(0x1000);
1215             size = qemu_get_be32(f);
1216 
1217             while (size > 0) {
1218                 uint32_t read_chunk = MIN(size, 0x1000);
1219                 qemu_get_buffer(f, buf, read_chunk);
1220                 size -= read_chunk;
1221             }
1222             g_free(buf);
1223         }
1224     }
1225 
1226     cpu_synchronize_all_post_init();
1227 
1228     ret = 0;
1229 
1230 out:
1231     if (ret == 0) {
1232         /* We may not have a VMDESC section, so ignore relative errors */
1233         ret = file_error_after_eof;
1234     }
1235 
1236     return ret;
1237 }
1238 
1239 static BlockDriverState *find_vmstate_bs(void)
1240 {
1241     BlockDriverState *bs = NULL;
1242     while ((bs = bdrv_next(bs))) {
1243         if (bdrv_can_snapshot(bs)) {
1244             return bs;
1245         }
1246     }
1247     return NULL;
1248 }
1249 
1250 /*
1251  * Deletes snapshots of a given name in all opened images.
1252  */
1253 static int del_existing_snapshots(Monitor *mon, const char *name)
1254 {
1255     BlockDriverState *bs;
1256     QEMUSnapshotInfo sn1, *snapshot = &sn1;
1257     Error *err = NULL;
1258 
1259     bs = NULL;
1260     while ((bs = bdrv_next(bs))) {
1261         if (bdrv_can_snapshot(bs) &&
1262             bdrv_snapshot_find(bs, snapshot, name) >= 0) {
1263             bdrv_snapshot_delete_by_id_or_name(bs, name, &err);
1264             if (err) {
1265                 monitor_printf(mon,
1266                                "Error while deleting snapshot on device '%s':"
1267                                " %s\n",
1268                                bdrv_get_device_name(bs),
1269                                error_get_pretty(err));
1270                 error_free(err);
1271                 return -1;
1272             }
1273         }
1274     }
1275 
1276     return 0;
1277 }
1278 
1279 void hmp_savevm(Monitor *mon, const QDict *qdict)
1280 {
1281     BlockDriverState *bs, *bs1;
1282     QEMUSnapshotInfo sn1, *sn = &sn1, old_sn1, *old_sn = &old_sn1;
1283     int ret;
1284     QEMUFile *f;
1285     int saved_vm_running;
1286     uint64_t vm_state_size;
1287     qemu_timeval tv;
1288     struct tm tm;
1289     const char *name = qdict_get_try_str(qdict, "name");
1290     Error *local_err = NULL;
1291 
1292     /* Verify if there is a device that doesn't support snapshots and is writable */
1293     bs = NULL;
1294     while ((bs = bdrv_next(bs))) {
1295 
1296         if (!bdrv_is_inserted(bs) || bdrv_is_read_only(bs)) {
1297             continue;
1298         }
1299 
1300         if (!bdrv_can_snapshot(bs)) {
1301             monitor_printf(mon, "Device '%s' is writable but does not support snapshots.\n",
1302                                bdrv_get_device_name(bs));
1303             return;
1304         }
1305     }
1306 
1307     bs = find_vmstate_bs();
1308     if (!bs) {
1309         monitor_printf(mon, "No block device can accept snapshots\n");
1310         return;
1311     }
1312 
1313     saved_vm_running = runstate_is_running();
1314 
1315     ret = global_state_store();
1316     if (ret) {
1317         monitor_printf(mon, "Error saving global state\n");
1318         return;
1319     }
1320     vm_stop(RUN_STATE_SAVE_VM);
1321 
1322     memset(sn, 0, sizeof(*sn));
1323 
1324     /* fill auxiliary fields */
1325     qemu_gettimeofday(&tv);
1326     sn->date_sec = tv.tv_sec;
1327     sn->date_nsec = tv.tv_usec * 1000;
1328     sn->vm_clock_nsec = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL);
1329 
1330     if (name) {
1331         ret = bdrv_snapshot_find(bs, old_sn, name);
1332         if (ret >= 0) {
1333             pstrcpy(sn->name, sizeof(sn->name), old_sn->name);
1334             pstrcpy(sn->id_str, sizeof(sn->id_str), old_sn->id_str);
1335         } else {
1336             pstrcpy(sn->name, sizeof(sn->name), name);
1337         }
1338     } else {
1339         /* cast below needed for OpenBSD where tv_sec is still 'long' */
1340         localtime_r((const time_t *)&tv.tv_sec, &tm);
1341         strftime(sn->name, sizeof(sn->name), "vm-%Y%m%d%H%M%S", &tm);
1342     }
1343 
1344     /* Delete old snapshots of the same name */
1345     if (name && del_existing_snapshots(mon, name) < 0) {
1346         goto the_end;
1347     }
1348 
1349     /* save the VM state */
1350     f = qemu_fopen_bdrv(bs, 1);
1351     if (!f) {
1352         monitor_printf(mon, "Could not open VM state file\n");
1353         goto the_end;
1354     }
1355     ret = qemu_savevm_state(f, &local_err);
1356     vm_state_size = qemu_ftell(f);
1357     qemu_fclose(f);
1358     if (ret < 0) {
1359         monitor_printf(mon, "%s\n", error_get_pretty(local_err));
1360         error_free(local_err);
1361         goto the_end;
1362     }
1363 
1364     /* create the snapshots */
1365 
1366     bs1 = NULL;
1367     while ((bs1 = bdrv_next(bs1))) {
1368         if (bdrv_can_snapshot(bs1)) {
1369             /* Write VM state size only to the image that contains the state */
1370             sn->vm_state_size = (bs == bs1 ? vm_state_size : 0);
1371             ret = bdrv_snapshot_create(bs1, sn);
1372             if (ret < 0) {
1373                 monitor_printf(mon, "Error while creating snapshot on '%s'\n",
1374                                bdrv_get_device_name(bs1));
1375             }
1376         }
1377     }
1378 
1379  the_end:
1380     if (saved_vm_running) {
1381         vm_start();
1382     }
1383 }
1384 
1385 void qmp_xen_save_devices_state(const char *filename, Error **errp)
1386 {
1387     QEMUFile *f;
1388     int saved_vm_running;
1389     int ret;
1390 
1391     saved_vm_running = runstate_is_running();
1392     vm_stop(RUN_STATE_SAVE_VM);
1393     global_state_store_running();
1394 
1395     f = qemu_fopen(filename, "wb");
1396     if (!f) {
1397         error_setg_file_open(errp, errno, filename);
1398         goto the_end;
1399     }
1400     ret = qemu_save_device_state(f);
1401     qemu_fclose(f);
1402     if (ret < 0) {
1403         error_setg(errp, QERR_IO_ERROR);
1404     }
1405 
1406  the_end:
1407     if (saved_vm_running) {
1408         vm_start();
1409     }
1410 }
1411 
1412 int load_vmstate(const char *name)
1413 {
1414     BlockDriverState *bs, *bs_vm_state;
1415     QEMUSnapshotInfo sn;
1416     QEMUFile *f;
1417     int ret;
1418 
1419     bs_vm_state = find_vmstate_bs();
1420     if (!bs_vm_state) {
1421         error_report("No block device supports snapshots");
1422         return -ENOTSUP;
1423     }
1424 
1425     /* Don't even try to load empty VM states */
1426     ret = bdrv_snapshot_find(bs_vm_state, &sn, name);
1427     if (ret < 0) {
1428         return ret;
1429     } else if (sn.vm_state_size == 0) {
1430         error_report("This is a disk-only snapshot. Revert to it offline "
1431             "using qemu-img.");
1432         return -EINVAL;
1433     }
1434 
1435     /* Verify if there is any device that doesn't support snapshots and is
1436     writable and check if the requested snapshot is available too. */
1437     bs = NULL;
1438     while ((bs = bdrv_next(bs))) {
1439 
1440         if (!bdrv_is_inserted(bs) || bdrv_is_read_only(bs)) {
1441             continue;
1442         }
1443 
1444         if (!bdrv_can_snapshot(bs)) {
1445             error_report("Device '%s' is writable but does not support snapshots.",
1446                                bdrv_get_device_name(bs));
1447             return -ENOTSUP;
1448         }
1449 
1450         ret = bdrv_snapshot_find(bs, &sn, name);
1451         if (ret < 0) {
1452             error_report("Device '%s' does not have the requested snapshot '%s'",
1453                            bdrv_get_device_name(bs), name);
1454             return ret;
1455         }
1456     }
1457 
1458     /* Flush all IO requests so they don't interfere with the new state.  */
1459     bdrv_drain_all();
1460 
1461     bs = NULL;
1462     while ((bs = bdrv_next(bs))) {
1463         if (bdrv_can_snapshot(bs)) {
1464             ret = bdrv_snapshot_goto(bs, name);
1465             if (ret < 0) {
1466                 error_report("Error %d while activating snapshot '%s' on '%s'",
1467                              ret, name, bdrv_get_device_name(bs));
1468                 return ret;
1469             }
1470         }
1471     }
1472 
1473     /* restore the VM state */
1474     f = qemu_fopen_bdrv(bs_vm_state, 0);
1475     if (!f) {
1476         error_report("Could not open VM state file");
1477         return -EINVAL;
1478     }
1479 
1480     qemu_system_reset(VMRESET_SILENT);
1481     migration_incoming_state_new(f);
1482     ret = qemu_loadvm_state(f);
1483 
1484     qemu_fclose(f);
1485     migration_incoming_state_destroy();
1486     if (ret < 0) {
1487         error_report("Error %d while loading VM state", ret);
1488         return ret;
1489     }
1490 
1491     return 0;
1492 }
1493 
1494 void hmp_delvm(Monitor *mon, const QDict *qdict)
1495 {
1496     BlockDriverState *bs;
1497     Error *err;
1498     const char *name = qdict_get_str(qdict, "name");
1499 
1500     if (!find_vmstate_bs()) {
1501         monitor_printf(mon, "No block device supports snapshots\n");
1502         return;
1503     }
1504 
1505     bs = NULL;
1506     while ((bs = bdrv_next(bs))) {
1507         if (bdrv_can_snapshot(bs)) {
1508             err = NULL;
1509             bdrv_snapshot_delete_by_id_or_name(bs, name, &err);
1510             if (err) {
1511                 monitor_printf(mon,
1512                                "Error while deleting snapshot on device '%s':"
1513                                " %s\n",
1514                                bdrv_get_device_name(bs),
1515                                error_get_pretty(err));
1516                 error_free(err);
1517             }
1518         }
1519     }
1520 }
1521 
1522 void hmp_info_snapshots(Monitor *mon, const QDict *qdict)
1523 {
1524     BlockDriverState *bs, *bs1;
1525     QEMUSnapshotInfo *sn_tab, *sn, s, *sn_info = &s;
1526     int nb_sns, i, ret, available;
1527     int total;
1528     int *available_snapshots;
1529 
1530     bs = find_vmstate_bs();
1531     if (!bs) {
1532         monitor_printf(mon, "No available block device supports snapshots\n");
1533         return;
1534     }
1535 
1536     nb_sns = bdrv_snapshot_list(bs, &sn_tab);
1537     if (nb_sns < 0) {
1538         monitor_printf(mon, "bdrv_snapshot_list: error %d\n", nb_sns);
1539         return;
1540     }
1541 
1542     if (nb_sns == 0) {
1543         monitor_printf(mon, "There is no snapshot available.\n");
1544         return;
1545     }
1546 
1547     available_snapshots = g_malloc0(sizeof(int) * nb_sns);
1548     total = 0;
1549     for (i = 0; i < nb_sns; i++) {
1550         sn = &sn_tab[i];
1551         available = 1;
1552         bs1 = NULL;
1553 
1554         while ((bs1 = bdrv_next(bs1))) {
1555             if (bdrv_can_snapshot(bs1) && bs1 != bs) {
1556                 ret = bdrv_snapshot_find(bs1, sn_info, sn->id_str);
1557                 if (ret < 0) {
1558                     available = 0;
1559                     break;
1560                 }
1561             }
1562         }
1563 
1564         if (available) {
1565             available_snapshots[total] = i;
1566             total++;
1567         }
1568     }
1569 
1570     if (total > 0) {
1571         bdrv_snapshot_dump((fprintf_function)monitor_printf, mon, NULL);
1572         monitor_printf(mon, "\n");
1573         for (i = 0; i < total; i++) {
1574             sn = &sn_tab[available_snapshots[i]];
1575             bdrv_snapshot_dump((fprintf_function)monitor_printf, mon, sn);
1576             monitor_printf(mon, "\n");
1577         }
1578     } else {
1579         monitor_printf(mon, "There is no suitable snapshot available\n");
1580     }
1581 
1582     g_free(sn_tab);
1583     g_free(available_snapshots);
1584 
1585 }
1586 
1587 void vmstate_register_ram(MemoryRegion *mr, DeviceState *dev)
1588 {
1589     qemu_ram_set_idstr(memory_region_get_ram_addr(mr) & TARGET_PAGE_MASK,
1590                        memory_region_name(mr), dev);
1591 }
1592 
1593 void vmstate_unregister_ram(MemoryRegion *mr, DeviceState *dev)
1594 {
1595     qemu_ram_unset_idstr(memory_region_get_ram_addr(mr) & TARGET_PAGE_MASK);
1596 }
1597 
1598 void vmstate_register_ram_global(MemoryRegion *mr)
1599 {
1600     vmstate_register_ram(mr, NULL);
1601 }
1602