hw/arm/virt-acpi-build: don't save VirtGuestInfo on AcpiBuildState
[qemu.git] / migration / savevm.c
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 "cpu.h"
31 #include "hw/boards.h"
32 #include "hw/hw.h"
33 #include "hw/qdev.h"
34 #include "hw/xen/xen.h"
35 #include "net/net.h"
36 #include "monitor/monitor.h"
37 #include "sysemu/sysemu.h"
38 #include "qemu/timer.h"
39 #include "audio/audio.h"
40 #include "migration/migration.h"
41 #include "migration/postcopy-ram.h"
42 #include "qapi/qmp/qerror.h"
43 #include "qemu/error-report.h"
44 #include "qemu/sockets.h"
45 #include "qemu/queue.h"
46 #include "sysemu/cpus.h"
47 #include "exec/memory.h"
48 #include "qmp-commands.h"
49 #include "trace.h"
50 #include "qemu/bitops.h"
51 #include "qemu/iov.h"
52 #include "block/snapshot.h"
53 #include "block/qapi.h"
54 #include "qemu/cutils.h"
55 #include "io/channel-buffer.h"
56 #include "io/channel-file.h"
57
58 #ifndef ETH_P_RARP
59 #define ETH_P_RARP 0x8035
60 #endif
61 #define ARP_HTYPE_ETH 0x0001
62 #define ARP_PTYPE_IP 0x0800
63 #define ARP_OP_REQUEST_REV 0x3
64
65 const unsigned int postcopy_ram_discard_version = 0;
66
67 static bool skip_section_footers;
68
69 static struct mig_cmd_args {
70 ssize_t len; /* -1 = variable */
71 const char *name;
72 } mig_cmd_args[] = {
73 [MIG_CMD_INVALID] = { .len = -1, .name = "INVALID" },
74 [MIG_CMD_OPEN_RETURN_PATH] = { .len = 0, .name = "OPEN_RETURN_PATH" },
75 [MIG_CMD_PING] = { .len = sizeof(uint32_t), .name = "PING" },
76 [MIG_CMD_POSTCOPY_ADVISE] = { .len = 16, .name = "POSTCOPY_ADVISE" },
77 [MIG_CMD_POSTCOPY_LISTEN] = { .len = 0, .name = "POSTCOPY_LISTEN" },
78 [MIG_CMD_POSTCOPY_RUN] = { .len = 0, .name = "POSTCOPY_RUN" },
79 [MIG_CMD_POSTCOPY_RAM_DISCARD] = {
80 .len = -1, .name = "POSTCOPY_RAM_DISCARD" },
81 [MIG_CMD_PACKAGED] = { .len = 4, .name = "PACKAGED" },
82 [MIG_CMD_MAX] = { .len = -1, .name = "MAX" },
83 };
84
85 static int announce_self_create(uint8_t *buf,
86 uint8_t *mac_addr)
87 {
88 /* Ethernet header. */
89 memset(buf, 0xff, 6); /* destination MAC addr */
90 memcpy(buf + 6, mac_addr, 6); /* source MAC addr */
91 *(uint16_t *)(buf + 12) = htons(ETH_P_RARP); /* ethertype */
92
93 /* RARP header. */
94 *(uint16_t *)(buf + 14) = htons(ARP_HTYPE_ETH); /* hardware addr space */
95 *(uint16_t *)(buf + 16) = htons(ARP_PTYPE_IP); /* protocol addr space */
96 *(buf + 18) = 6; /* hardware addr length (ethernet) */
97 *(buf + 19) = 4; /* protocol addr length (IPv4) */
98 *(uint16_t *)(buf + 20) = htons(ARP_OP_REQUEST_REV); /* opcode */
99 memcpy(buf + 22, mac_addr, 6); /* source hw addr */
100 memset(buf + 28, 0x00, 4); /* source protocol addr */
101 memcpy(buf + 32, mac_addr, 6); /* target hw addr */
102 memset(buf + 38, 0x00, 4); /* target protocol addr */
103
104 /* Padding to get up to 60 bytes (ethernet min packet size, minus FCS). */
105 memset(buf + 42, 0x00, 18);
106
107 return 60; /* len (FCS will be added by hardware) */
108 }
109
110 static void qemu_announce_self_iter(NICState *nic, void *opaque)
111 {
112 uint8_t buf[60];
113 int len;
114
115 trace_qemu_announce_self_iter(qemu_ether_ntoa(&nic->conf->macaddr));
116 len = announce_self_create(buf, nic->conf->macaddr.a);
117
118 qemu_send_packet_raw(qemu_get_queue(nic), buf, len);
119 }
120
121
122 static void qemu_announce_self_once(void *opaque)
123 {
124 static int count = SELF_ANNOUNCE_ROUNDS;
125 QEMUTimer *timer = *(QEMUTimer **)opaque;
126
127 qemu_foreach_nic(qemu_announce_self_iter, NULL);
128
129 if (--count) {
130 /* delay 50ms, 150ms, 250ms, ... */
131 timer_mod(timer, qemu_clock_get_ms(QEMU_CLOCK_REALTIME) +
132 self_announce_delay(count));
133 } else {
134 timer_del(timer);
135 timer_free(timer);
136 }
137 }
138
139 void qemu_announce_self(void)
140 {
141 static QEMUTimer *timer;
142 timer = timer_new_ms(QEMU_CLOCK_REALTIME, qemu_announce_self_once, &timer);
143 qemu_announce_self_once(&timer);
144 }
145
146 /***********************************************************/
147 /* savevm/loadvm support */
148
149 static ssize_t block_writev_buffer(void *opaque, struct iovec *iov, int iovcnt,
150 int64_t pos)
151 {
152 int ret;
153 QEMUIOVector qiov;
154
155 qemu_iovec_init_external(&qiov, iov, iovcnt);
156 ret = bdrv_writev_vmstate(opaque, &qiov, pos);
157 if (ret < 0) {
158 return ret;
159 }
160
161 return qiov.size;
162 }
163
164 static ssize_t block_get_buffer(void *opaque, uint8_t *buf, int64_t pos,
165 size_t size)
166 {
167 return bdrv_load_vmstate(opaque, buf, pos, size);
168 }
169
170 static int bdrv_fclose(void *opaque)
171 {
172 return bdrv_flush(opaque);
173 }
174
175 static const QEMUFileOps bdrv_read_ops = {
176 .get_buffer = block_get_buffer,
177 .close = bdrv_fclose
178 };
179
180 static const QEMUFileOps bdrv_write_ops = {
181 .writev_buffer = block_writev_buffer,
182 .close = bdrv_fclose
183 };
184
185 static QEMUFile *qemu_fopen_bdrv(BlockDriverState *bs, int is_writable)
186 {
187 if (is_writable) {
188 return qemu_fopen_ops(bs, &bdrv_write_ops);
189 }
190 return qemu_fopen_ops(bs, &bdrv_read_ops);
191 }
192
193
194 /* QEMUFile timer support.
195 * Not in qemu-file.c to not add qemu-timer.c as dependency to qemu-file.c
196 */
197
198 void timer_put(QEMUFile *f, QEMUTimer *ts)
199 {
200 uint64_t expire_time;
201
202 expire_time = timer_expire_time_ns(ts);
203 qemu_put_be64(f, expire_time);
204 }
205
206 void timer_get(QEMUFile *f, QEMUTimer *ts)
207 {
208 uint64_t expire_time;
209
210 expire_time = qemu_get_be64(f);
211 if (expire_time != -1) {
212 timer_mod_ns(ts, expire_time);
213 } else {
214 timer_del(ts);
215 }
216 }
217
218
219 /* VMState timer support.
220 * Not in vmstate.c to not add qemu-timer.c as dependency to vmstate.c
221 */
222
223 static int get_timer(QEMUFile *f, void *pv, size_t size)
224 {
225 QEMUTimer *v = pv;
226 timer_get(f, v);
227 return 0;
228 }
229
230 static void put_timer(QEMUFile *f, void *pv, size_t size)
231 {
232 QEMUTimer *v = pv;
233 timer_put(f, v);
234 }
235
236 const VMStateInfo vmstate_info_timer = {
237 .name = "timer",
238 .get = get_timer,
239 .put = put_timer,
240 };
241
242
243 typedef struct CompatEntry {
244 char idstr[256];
245 int instance_id;
246 } CompatEntry;
247
248 typedef struct SaveStateEntry {
249 QTAILQ_ENTRY(SaveStateEntry) entry;
250 char idstr[256];
251 int instance_id;
252 int alias_id;
253 int version_id;
254 int section_id;
255 SaveVMHandlers *ops;
256 const VMStateDescription *vmsd;
257 void *opaque;
258 CompatEntry *compat;
259 int is_ram;
260 } SaveStateEntry;
261
262 typedef struct SaveState {
263 QTAILQ_HEAD(, SaveStateEntry) handlers;
264 int global_section_id;
265 bool skip_configuration;
266 uint32_t len;
267 const char *name;
268 uint32_t target_page_bits;
269 } SaveState;
270
271 static SaveState savevm_state = {
272 .handlers = QTAILQ_HEAD_INITIALIZER(savevm_state.handlers),
273 .global_section_id = 0,
274 .skip_configuration = false,
275 };
276
277 void savevm_skip_configuration(void)
278 {
279 savevm_state.skip_configuration = true;
280 }
281
282
283 static void configuration_pre_save(void *opaque)
284 {
285 SaveState *state = opaque;
286 const char *current_name = MACHINE_GET_CLASS(current_machine)->name;
287
288 state->len = strlen(current_name);
289 state->name = current_name;
290 state->target_page_bits = TARGET_PAGE_BITS;
291 }
292
293 static int configuration_pre_load(void *opaque)
294 {
295 SaveState *state = opaque;
296
297 /* If there is no target-page-bits subsection it means the source
298 * predates the variable-target-page-bits support and is using the
299 * minimum possible value for this CPU.
300 */
301 state->target_page_bits = TARGET_PAGE_BITS_MIN;
302 return 0;
303 }
304
305 static int configuration_post_load(void *opaque, int version_id)
306 {
307 SaveState *state = opaque;
308 const char *current_name = MACHINE_GET_CLASS(current_machine)->name;
309
310 if (strncmp(state->name, current_name, state->len) != 0) {
311 error_report("Machine type received is '%.*s' and local is '%s'",
312 (int) state->len, state->name, current_name);
313 return -EINVAL;
314 }
315
316 if (state->target_page_bits != TARGET_PAGE_BITS) {
317 error_report("Received TARGET_PAGE_BITS is %d but local is %d",
318 state->target_page_bits, TARGET_PAGE_BITS);
319 return -EINVAL;
320 }
321
322 return 0;
323 }
324
325 /* The target-page-bits subsection is present only if the
326 * target page size is not the same as the default (ie the
327 * minimum page size for a variable-page-size guest CPU).
328 * If it is present then it contains the actual target page
329 * bits for the machine, and migration will fail if the
330 * two ends don't agree about it.
331 */
332 static bool vmstate_target_page_bits_needed(void *opaque)
333 {
334 return TARGET_PAGE_BITS > TARGET_PAGE_BITS_MIN;
335 }
336
337 static const VMStateDescription vmstate_target_page_bits = {
338 .name = "configuration/target-page-bits",
339 .version_id = 1,
340 .minimum_version_id = 1,
341 .needed = vmstate_target_page_bits_needed,
342 .fields = (VMStateField[]) {
343 VMSTATE_UINT32(target_page_bits, SaveState),
344 VMSTATE_END_OF_LIST()
345 }
346 };
347
348 static const VMStateDescription vmstate_configuration = {
349 .name = "configuration",
350 .version_id = 1,
351 .pre_load = configuration_pre_load,
352 .post_load = configuration_post_load,
353 .pre_save = configuration_pre_save,
354 .fields = (VMStateField[]) {
355 VMSTATE_UINT32(len, SaveState),
356 VMSTATE_VBUFFER_ALLOC_UINT32(name, SaveState, 0, NULL, 0, len),
357 VMSTATE_END_OF_LIST()
358 },
359 .subsections = (const VMStateDescription*[]) {
360 &vmstate_target_page_bits,
361 NULL
362 }
363 };
364
365 static void dump_vmstate_vmsd(FILE *out_file,
366 const VMStateDescription *vmsd, int indent,
367 bool is_subsection);
368
369 static void dump_vmstate_vmsf(FILE *out_file, const VMStateField *field,
370 int indent)
371 {
372 fprintf(out_file, "%*s{\n", indent, "");
373 indent += 2;
374 fprintf(out_file, "%*s\"field\": \"%s\",\n", indent, "", field->name);
375 fprintf(out_file, "%*s\"version_id\": %d,\n", indent, "",
376 field->version_id);
377 fprintf(out_file, "%*s\"field_exists\": %s,\n", indent, "",
378 field->field_exists ? "true" : "false");
379 fprintf(out_file, "%*s\"size\": %zu", indent, "", field->size);
380 if (field->vmsd != NULL) {
381 fprintf(out_file, ",\n");
382 dump_vmstate_vmsd(out_file, field->vmsd, indent, false);
383 }
384 fprintf(out_file, "\n%*s}", indent - 2, "");
385 }
386
387 static void dump_vmstate_vmss(FILE *out_file,
388 const VMStateDescription **subsection,
389 int indent)
390 {
391 if (*subsection != NULL) {
392 dump_vmstate_vmsd(out_file, *subsection, indent, true);
393 }
394 }
395
396 static void dump_vmstate_vmsd(FILE *out_file,
397 const VMStateDescription *vmsd, int indent,
398 bool is_subsection)
399 {
400 if (is_subsection) {
401 fprintf(out_file, "%*s{\n", indent, "");
402 } else {
403 fprintf(out_file, "%*s\"%s\": {\n", indent, "", "Description");
404 }
405 indent += 2;
406 fprintf(out_file, "%*s\"name\": \"%s\",\n", indent, "", vmsd->name);
407 fprintf(out_file, "%*s\"version_id\": %d,\n", indent, "",
408 vmsd->version_id);
409 fprintf(out_file, "%*s\"minimum_version_id\": %d", indent, "",
410 vmsd->minimum_version_id);
411 if (vmsd->fields != NULL) {
412 const VMStateField *field = vmsd->fields;
413 bool first;
414
415 fprintf(out_file, ",\n%*s\"Fields\": [\n", indent, "");
416 first = true;
417 while (field->name != NULL) {
418 if (field->flags & VMS_MUST_EXIST) {
419 /* Ignore VMSTATE_VALIDATE bits; these don't get migrated */
420 field++;
421 continue;
422 }
423 if (!first) {
424 fprintf(out_file, ",\n");
425 }
426 dump_vmstate_vmsf(out_file, field, indent + 2);
427 field++;
428 first = false;
429 }
430 fprintf(out_file, "\n%*s]", indent, "");
431 }
432 if (vmsd->subsections != NULL) {
433 const VMStateDescription **subsection = vmsd->subsections;
434 bool first;
435
436 fprintf(out_file, ",\n%*s\"Subsections\": [\n", indent, "");
437 first = true;
438 while (*subsection != NULL) {
439 if (!first) {
440 fprintf(out_file, ",\n");
441 }
442 dump_vmstate_vmss(out_file, subsection, indent + 2);
443 subsection++;
444 first = false;
445 }
446 fprintf(out_file, "\n%*s]", indent, "");
447 }
448 fprintf(out_file, "\n%*s}", indent - 2, "");
449 }
450
451 static void dump_machine_type(FILE *out_file)
452 {
453 MachineClass *mc;
454
455 mc = MACHINE_GET_CLASS(current_machine);
456
457 fprintf(out_file, " \"vmschkmachine\": {\n");
458 fprintf(out_file, " \"Name\": \"%s\"\n", mc->name);
459 fprintf(out_file, " },\n");
460 }
461
462 void dump_vmstate_json_to_file(FILE *out_file)
463 {
464 GSList *list, *elt;
465 bool first;
466
467 fprintf(out_file, "{\n");
468 dump_machine_type(out_file);
469
470 first = true;
471 list = object_class_get_list(TYPE_DEVICE, true);
472 for (elt = list; elt; elt = elt->next) {
473 DeviceClass *dc = OBJECT_CLASS_CHECK(DeviceClass, elt->data,
474 TYPE_DEVICE);
475 const char *name;
476 int indent = 2;
477
478 if (!dc->vmsd) {
479 continue;
480 }
481
482 if (!first) {
483 fprintf(out_file, ",\n");
484 }
485 name = object_class_get_name(OBJECT_CLASS(dc));
486 fprintf(out_file, "%*s\"%s\": {\n", indent, "", name);
487 indent += 2;
488 fprintf(out_file, "%*s\"Name\": \"%s\",\n", indent, "", name);
489 fprintf(out_file, "%*s\"version_id\": %d,\n", indent, "",
490 dc->vmsd->version_id);
491 fprintf(out_file, "%*s\"minimum_version_id\": %d,\n", indent, "",
492 dc->vmsd->minimum_version_id);
493
494 dump_vmstate_vmsd(out_file, dc->vmsd, indent, false);
495
496 fprintf(out_file, "\n%*s}", indent - 2, "");
497 first = false;
498 }
499 fprintf(out_file, "\n}\n");
500 fclose(out_file);
501 }
502
503 static int calculate_new_instance_id(const char *idstr)
504 {
505 SaveStateEntry *se;
506 int instance_id = 0;
507
508 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
509 if (strcmp(idstr, se->idstr) == 0
510 && instance_id <= se->instance_id) {
511 instance_id = se->instance_id + 1;
512 }
513 }
514 return instance_id;
515 }
516
517 static int calculate_compat_instance_id(const char *idstr)
518 {
519 SaveStateEntry *se;
520 int instance_id = 0;
521
522 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
523 if (!se->compat) {
524 continue;
525 }
526
527 if (strcmp(idstr, se->compat->idstr) == 0
528 && instance_id <= se->compat->instance_id) {
529 instance_id = se->compat->instance_id + 1;
530 }
531 }
532 return instance_id;
533 }
534
535 /* TODO: Individual devices generally have very little idea about the rest
536 of the system, so instance_id should be removed/replaced.
537 Meanwhile pass -1 as instance_id if you do not already have a clearly
538 distinguishing id for all instances of your device class. */
539 int register_savevm_live(DeviceState *dev,
540 const char *idstr,
541 int instance_id,
542 int version_id,
543 SaveVMHandlers *ops,
544 void *opaque)
545 {
546 SaveStateEntry *se;
547
548 se = g_new0(SaveStateEntry, 1);
549 se->version_id = version_id;
550 se->section_id = savevm_state.global_section_id++;
551 se->ops = ops;
552 se->opaque = opaque;
553 se->vmsd = NULL;
554 /* if this is a live_savem then set is_ram */
555 if (ops->save_live_setup != NULL) {
556 se->is_ram = 1;
557 }
558
559 if (dev) {
560 char *id = qdev_get_dev_path(dev);
561 if (id) {
562 pstrcpy(se->idstr, sizeof(se->idstr), id);
563 pstrcat(se->idstr, sizeof(se->idstr), "/");
564 g_free(id);
565
566 se->compat = g_new0(CompatEntry, 1);
567 pstrcpy(se->compat->idstr, sizeof(se->compat->idstr), idstr);
568 se->compat->instance_id = instance_id == -1 ?
569 calculate_compat_instance_id(idstr) : instance_id;
570 instance_id = -1;
571 }
572 }
573 pstrcat(se->idstr, sizeof(se->idstr), idstr);
574
575 if (instance_id == -1) {
576 se->instance_id = calculate_new_instance_id(se->idstr);
577 } else {
578 se->instance_id = instance_id;
579 }
580 assert(!se->compat || se->instance_id == 0);
581 /* add at the end of list */
582 QTAILQ_INSERT_TAIL(&savevm_state.handlers, se, entry);
583 return 0;
584 }
585
586 int register_savevm(DeviceState *dev,
587 const char *idstr,
588 int instance_id,
589 int version_id,
590 SaveStateHandler *save_state,
591 LoadStateHandler *load_state,
592 void *opaque)
593 {
594 SaveVMHandlers *ops = g_new0(SaveVMHandlers, 1);
595 ops->save_state = save_state;
596 ops->load_state = load_state;
597 return register_savevm_live(dev, idstr, instance_id, version_id,
598 ops, opaque);
599 }
600
601 void unregister_savevm(DeviceState *dev, const char *idstr, void *opaque)
602 {
603 SaveStateEntry *se, *new_se;
604 char id[256] = "";
605
606 if (dev) {
607 char *path = qdev_get_dev_path(dev);
608 if (path) {
609 pstrcpy(id, sizeof(id), path);
610 pstrcat(id, sizeof(id), "/");
611 g_free(path);
612 }
613 }
614 pstrcat(id, sizeof(id), idstr);
615
616 QTAILQ_FOREACH_SAFE(se, &savevm_state.handlers, entry, new_se) {
617 if (strcmp(se->idstr, id) == 0 && se->opaque == opaque) {
618 QTAILQ_REMOVE(&savevm_state.handlers, se, entry);
619 g_free(se->compat);
620 g_free(se->ops);
621 g_free(se);
622 }
623 }
624 }
625
626 int vmstate_register_with_alias_id(DeviceState *dev, int instance_id,
627 const VMStateDescription *vmsd,
628 void *opaque, int alias_id,
629 int required_for_version)
630 {
631 SaveStateEntry *se;
632
633 /* If this triggers, alias support can be dropped for the vmsd. */
634 assert(alias_id == -1 || required_for_version >= vmsd->minimum_version_id);
635
636 se = g_new0(SaveStateEntry, 1);
637 se->version_id = vmsd->version_id;
638 se->section_id = savevm_state.global_section_id++;
639 se->opaque = opaque;
640 se->vmsd = vmsd;
641 se->alias_id = alias_id;
642
643 if (dev) {
644 char *id = qdev_get_dev_path(dev);
645 if (id) {
646 pstrcpy(se->idstr, sizeof(se->idstr), id);
647 pstrcat(se->idstr, sizeof(se->idstr), "/");
648 g_free(id);
649
650 se->compat = g_new0(CompatEntry, 1);
651 pstrcpy(se->compat->idstr, sizeof(se->compat->idstr), vmsd->name);
652 se->compat->instance_id = instance_id == -1 ?
653 calculate_compat_instance_id(vmsd->name) : instance_id;
654 instance_id = -1;
655 }
656 }
657 pstrcat(se->idstr, sizeof(se->idstr), vmsd->name);
658
659 if (instance_id == -1) {
660 se->instance_id = calculate_new_instance_id(se->idstr);
661 } else {
662 se->instance_id = instance_id;
663 }
664 assert(!se->compat || se->instance_id == 0);
665 /* add at the end of list */
666 QTAILQ_INSERT_TAIL(&savevm_state.handlers, se, entry);
667 return 0;
668 }
669
670 void vmstate_unregister(DeviceState *dev, const VMStateDescription *vmsd,
671 void *opaque)
672 {
673 SaveStateEntry *se, *new_se;
674
675 QTAILQ_FOREACH_SAFE(se, &savevm_state.handlers, entry, new_se) {
676 if (se->vmsd == vmsd && se->opaque == opaque) {
677 QTAILQ_REMOVE(&savevm_state.handlers, se, entry);
678 g_free(se->compat);
679 g_free(se);
680 }
681 }
682 }
683
684 static int vmstate_load(QEMUFile *f, SaveStateEntry *se, int version_id)
685 {
686 trace_vmstate_load(se->idstr, se->vmsd ? se->vmsd->name : "(old)");
687 if (!se->vmsd) { /* Old style */
688 return se->ops->load_state(f, se->opaque, version_id);
689 }
690 return vmstate_load_state(f, se->vmsd, se->opaque, version_id);
691 }
692
693 static void vmstate_save_old_style(QEMUFile *f, SaveStateEntry *se, QJSON *vmdesc)
694 {
695 int64_t old_offset, size;
696
697 old_offset = qemu_ftell_fast(f);
698 se->ops->save_state(f, se->opaque);
699 size = qemu_ftell_fast(f) - old_offset;
700
701 if (vmdesc) {
702 json_prop_int(vmdesc, "size", size);
703 json_start_array(vmdesc, "fields");
704 json_start_object(vmdesc, NULL);
705 json_prop_str(vmdesc, "name", "data");
706 json_prop_int(vmdesc, "size", size);
707 json_prop_str(vmdesc, "type", "buffer");
708 json_end_object(vmdesc);
709 json_end_array(vmdesc);
710 }
711 }
712
713 static void vmstate_save(QEMUFile *f, SaveStateEntry *se, QJSON *vmdesc)
714 {
715 trace_vmstate_save(se->idstr, se->vmsd ? se->vmsd->name : "(old)");
716 if (!se->vmsd) {
717 vmstate_save_old_style(f, se, vmdesc);
718 return;
719 }
720 vmstate_save_state(f, se->vmsd, se->opaque, vmdesc);
721 }
722
723 void savevm_skip_section_footers(void)
724 {
725 skip_section_footers = true;
726 }
727
728 /*
729 * Write the header for device section (QEMU_VM_SECTION START/END/PART/FULL)
730 */
731 static void save_section_header(QEMUFile *f, SaveStateEntry *se,
732 uint8_t section_type)
733 {
734 qemu_put_byte(f, section_type);
735 qemu_put_be32(f, se->section_id);
736
737 if (section_type == QEMU_VM_SECTION_FULL ||
738 section_type == QEMU_VM_SECTION_START) {
739 /* ID string */
740 size_t len = strlen(se->idstr);
741 qemu_put_byte(f, len);
742 qemu_put_buffer(f, (uint8_t *)se->idstr, len);
743
744 qemu_put_be32(f, se->instance_id);
745 qemu_put_be32(f, se->version_id);
746 }
747 }
748
749 /*
750 * Write a footer onto device sections that catches cases misformatted device
751 * sections.
752 */
753 static void save_section_footer(QEMUFile *f, SaveStateEntry *se)
754 {
755 if (!skip_section_footers) {
756 qemu_put_byte(f, QEMU_VM_SECTION_FOOTER);
757 qemu_put_be32(f, se->section_id);
758 }
759 }
760
761 /**
762 * qemu_savevm_command_send: Send a 'QEMU_VM_COMMAND' type element with the
763 * command and associated data.
764 *
765 * @f: File to send command on
766 * @command: Command type to send
767 * @len: Length of associated data
768 * @data: Data associated with command.
769 */
770 void qemu_savevm_command_send(QEMUFile *f,
771 enum qemu_vm_cmd command,
772 uint16_t len,
773 uint8_t *data)
774 {
775 trace_savevm_command_send(command, len);
776 qemu_put_byte(f, QEMU_VM_COMMAND);
777 qemu_put_be16(f, (uint16_t)command);
778 qemu_put_be16(f, len);
779 qemu_put_buffer(f, data, len);
780 qemu_fflush(f);
781 }
782
783 void qemu_savevm_send_ping(QEMUFile *f, uint32_t value)
784 {
785 uint32_t buf;
786
787 trace_savevm_send_ping(value);
788 buf = cpu_to_be32(value);
789 qemu_savevm_command_send(f, MIG_CMD_PING, sizeof(value), (uint8_t *)&buf);
790 }
791
792 void qemu_savevm_send_open_return_path(QEMUFile *f)
793 {
794 trace_savevm_send_open_return_path();
795 qemu_savevm_command_send(f, MIG_CMD_OPEN_RETURN_PATH, 0, NULL);
796 }
797
798 /* We have a buffer of data to send; we don't want that all to be loaded
799 * by the command itself, so the command contains just the length of the
800 * extra buffer that we then send straight after it.
801 * TODO: Must be a better way to organise that
802 *
803 * Returns:
804 * 0 on success
805 * -ve on error
806 */
807 int qemu_savevm_send_packaged(QEMUFile *f, const uint8_t *buf, size_t len)
808 {
809 uint32_t tmp;
810
811 if (len > MAX_VM_CMD_PACKAGED_SIZE) {
812 error_report("%s: Unreasonably large packaged state: %zu",
813 __func__, len);
814 return -1;
815 }
816
817 tmp = cpu_to_be32(len);
818
819 trace_qemu_savevm_send_packaged();
820 qemu_savevm_command_send(f, MIG_CMD_PACKAGED, 4, (uint8_t *)&tmp);
821
822 qemu_put_buffer(f, buf, len);
823
824 return 0;
825 }
826
827 /* Send prior to any postcopy transfer */
828 void qemu_savevm_send_postcopy_advise(QEMUFile *f)
829 {
830 uint64_t tmp[2];
831 tmp[0] = cpu_to_be64(getpagesize());
832 tmp[1] = cpu_to_be64(1ul << qemu_target_page_bits());
833
834 trace_qemu_savevm_send_postcopy_advise();
835 qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_ADVISE, 16, (uint8_t *)tmp);
836 }
837
838 /* Sent prior to starting the destination running in postcopy, discard pages
839 * that have already been sent but redirtied on the source.
840 * CMD_POSTCOPY_RAM_DISCARD consist of:
841 * byte version (0)
842 * byte Length of name field (not including 0)
843 * n x byte RAM block name
844 * byte 0 terminator (just for safety)
845 * n x Byte ranges within the named RAMBlock
846 * be64 Start of the range
847 * be64 Length
848 *
849 * name: RAMBlock name that these entries are part of
850 * len: Number of page entries
851 * start_list: 'len' addresses
852 * length_list: 'len' addresses
853 *
854 */
855 void qemu_savevm_send_postcopy_ram_discard(QEMUFile *f, const char *name,
856 uint16_t len,
857 uint64_t *start_list,
858 uint64_t *length_list)
859 {
860 uint8_t *buf;
861 uint16_t tmplen;
862 uint16_t t;
863 size_t name_len = strlen(name);
864
865 trace_qemu_savevm_send_postcopy_ram_discard(name, len);
866 assert(name_len < 256);
867 buf = g_malloc0(1 + 1 + name_len + 1 + (8 + 8) * len);
868 buf[0] = postcopy_ram_discard_version;
869 buf[1] = name_len;
870 memcpy(buf + 2, name, name_len);
871 tmplen = 2 + name_len;
872 buf[tmplen++] = '\0';
873
874 for (t = 0; t < len; t++) {
875 stq_be_p(buf + tmplen, start_list[t]);
876 tmplen += 8;
877 stq_be_p(buf + tmplen, length_list[t]);
878 tmplen += 8;
879 }
880 qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_RAM_DISCARD, tmplen, buf);
881 g_free(buf);
882 }
883
884 /* Get the destination into a state where it can receive postcopy data. */
885 void qemu_savevm_send_postcopy_listen(QEMUFile *f)
886 {
887 trace_savevm_send_postcopy_listen();
888 qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_LISTEN, 0, NULL);
889 }
890
891 /* Kick the destination into running */
892 void qemu_savevm_send_postcopy_run(QEMUFile *f)
893 {
894 trace_savevm_send_postcopy_run();
895 qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_RUN, 0, NULL);
896 }
897
898 bool qemu_savevm_state_blocked(Error **errp)
899 {
900 SaveStateEntry *se;
901
902 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
903 if (se->vmsd && se->vmsd->unmigratable) {
904 error_setg(errp, "State blocked by non-migratable device '%s'",
905 se->idstr);
906 return true;
907 }
908 }
909 return false;
910 }
911
912 static bool enforce_config_section(void)
913 {
914 MachineState *machine = MACHINE(qdev_get_machine());
915 return machine->enforce_config_section;
916 }
917
918 void qemu_savevm_state_header(QEMUFile *f)
919 {
920 trace_savevm_state_header();
921 qemu_put_be32(f, QEMU_VM_FILE_MAGIC);
922 qemu_put_be32(f, QEMU_VM_FILE_VERSION);
923
924 if (!savevm_state.skip_configuration || enforce_config_section()) {
925 qemu_put_byte(f, QEMU_VM_CONFIGURATION);
926 vmstate_save_state(f, &vmstate_configuration, &savevm_state, 0);
927 }
928
929 }
930
931 void qemu_savevm_state_begin(QEMUFile *f,
932 const MigrationParams *params)
933 {
934 SaveStateEntry *se;
935 int ret;
936
937 trace_savevm_state_begin();
938 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
939 if (!se->ops || !se->ops->set_params) {
940 continue;
941 }
942 se->ops->set_params(params, se->opaque);
943 }
944
945 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
946 if (!se->ops || !se->ops->save_live_setup) {
947 continue;
948 }
949 if (se->ops && se->ops->is_active) {
950 if (!se->ops->is_active(se->opaque)) {
951 continue;
952 }
953 }
954 save_section_header(f, se, QEMU_VM_SECTION_START);
955
956 ret = se->ops->save_live_setup(f, se->opaque);
957 save_section_footer(f, se);
958 if (ret < 0) {
959 qemu_file_set_error(f, ret);
960 break;
961 }
962 }
963 }
964
965 /*
966 * this function has three return values:
967 * negative: there was one error, and we have -errno.
968 * 0 : We haven't finished, caller have to go again
969 * 1 : We have finished, we can go to complete phase
970 */
971 int qemu_savevm_state_iterate(QEMUFile *f, bool postcopy)
972 {
973 SaveStateEntry *se;
974 int ret = 1;
975
976 trace_savevm_state_iterate();
977 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
978 if (!se->ops || !se->ops->save_live_iterate) {
979 continue;
980 }
981 if (se->ops && se->ops->is_active) {
982 if (!se->ops->is_active(se->opaque)) {
983 continue;
984 }
985 }
986 /*
987 * In the postcopy phase, any device that doesn't know how to
988 * do postcopy should have saved it's state in the _complete
989 * call that's already run, it might get confused if we call
990 * iterate afterwards.
991 */
992 if (postcopy && !se->ops->save_live_complete_postcopy) {
993 continue;
994 }
995 if (qemu_file_rate_limit(f)) {
996 return 0;
997 }
998 trace_savevm_section_start(se->idstr, se->section_id);
999
1000 save_section_header(f, se, QEMU_VM_SECTION_PART);
1001
1002 ret = se->ops->save_live_iterate(f, se->opaque);
1003 trace_savevm_section_end(se->idstr, se->section_id, ret);
1004 save_section_footer(f, se);
1005
1006 if (ret < 0) {
1007 qemu_file_set_error(f, ret);
1008 }
1009 if (ret <= 0) {
1010 /* Do not proceed to the next vmstate before this one reported
1011 completion of the current stage. This serializes the migration
1012 and reduces the probability that a faster changing state is
1013 synchronized over and over again. */
1014 break;
1015 }
1016 }
1017 return ret;
1018 }
1019
1020 static bool should_send_vmdesc(void)
1021 {
1022 MachineState *machine = MACHINE(qdev_get_machine());
1023 bool in_postcopy = migration_in_postcopy(migrate_get_current());
1024 return !machine->suppress_vmdesc && !in_postcopy;
1025 }
1026
1027 /*
1028 * Calls the save_live_complete_postcopy methods
1029 * causing the last few pages to be sent immediately and doing any associated
1030 * cleanup.
1031 * Note postcopy also calls qemu_savevm_state_complete_precopy to complete
1032 * all the other devices, but that happens at the point we switch to postcopy.
1033 */
1034 void qemu_savevm_state_complete_postcopy(QEMUFile *f)
1035 {
1036 SaveStateEntry *se;
1037 int ret;
1038
1039 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1040 if (!se->ops || !se->ops->save_live_complete_postcopy) {
1041 continue;
1042 }
1043 if (se->ops && se->ops->is_active) {
1044 if (!se->ops->is_active(se->opaque)) {
1045 continue;
1046 }
1047 }
1048 trace_savevm_section_start(se->idstr, se->section_id);
1049 /* Section type */
1050 qemu_put_byte(f, QEMU_VM_SECTION_END);
1051 qemu_put_be32(f, se->section_id);
1052
1053 ret = se->ops->save_live_complete_postcopy(f, se->opaque);
1054 trace_savevm_section_end(se->idstr, se->section_id, ret);
1055 save_section_footer(f, se);
1056 if (ret < 0) {
1057 qemu_file_set_error(f, ret);
1058 return;
1059 }
1060 }
1061
1062 qemu_put_byte(f, QEMU_VM_EOF);
1063 qemu_fflush(f);
1064 }
1065
1066 void qemu_savevm_state_complete_precopy(QEMUFile *f, bool iterable_only)
1067 {
1068 QJSON *vmdesc;
1069 int vmdesc_len;
1070 SaveStateEntry *se;
1071 int ret;
1072 bool in_postcopy = migration_in_postcopy(migrate_get_current());
1073
1074 trace_savevm_state_complete_precopy();
1075
1076 cpu_synchronize_all_states();
1077
1078 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1079 if (!se->ops ||
1080 (in_postcopy && se->ops->save_live_complete_postcopy) ||
1081 (in_postcopy && !iterable_only) ||
1082 !se->ops->save_live_complete_precopy) {
1083 continue;
1084 }
1085
1086 if (se->ops && se->ops->is_active) {
1087 if (!se->ops->is_active(se->opaque)) {
1088 continue;
1089 }
1090 }
1091 trace_savevm_section_start(se->idstr, se->section_id);
1092
1093 save_section_header(f, se, QEMU_VM_SECTION_END);
1094
1095 ret = se->ops->save_live_complete_precopy(f, se->opaque);
1096 trace_savevm_section_end(se->idstr, se->section_id, ret);
1097 save_section_footer(f, se);
1098 if (ret < 0) {
1099 qemu_file_set_error(f, ret);
1100 return;
1101 }
1102 }
1103
1104 if (iterable_only) {
1105 return;
1106 }
1107
1108 vmdesc = qjson_new();
1109 json_prop_int(vmdesc, "page_size", TARGET_PAGE_SIZE);
1110 json_start_array(vmdesc, "devices");
1111 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1112
1113 if ((!se->ops || !se->ops->save_state) && !se->vmsd) {
1114 continue;
1115 }
1116 if (se->vmsd && !vmstate_save_needed(se->vmsd, se->opaque)) {
1117 trace_savevm_section_skip(se->idstr, se->section_id);
1118 continue;
1119 }
1120
1121 trace_savevm_section_start(se->idstr, se->section_id);
1122
1123 json_start_object(vmdesc, NULL);
1124 json_prop_str(vmdesc, "name", se->idstr);
1125 json_prop_int(vmdesc, "instance_id", se->instance_id);
1126
1127 save_section_header(f, se, QEMU_VM_SECTION_FULL);
1128 vmstate_save(f, se, vmdesc);
1129 trace_savevm_section_end(se->idstr, se->section_id, 0);
1130 save_section_footer(f, se);
1131
1132 json_end_object(vmdesc);
1133 }
1134
1135 if (!in_postcopy) {
1136 /* Postcopy stream will still be going */
1137 qemu_put_byte(f, QEMU_VM_EOF);
1138 }
1139
1140 json_end_array(vmdesc);
1141 qjson_finish(vmdesc);
1142 vmdesc_len = strlen(qjson_get_str(vmdesc));
1143
1144 if (should_send_vmdesc()) {
1145 qemu_put_byte(f, QEMU_VM_VMDESCRIPTION);
1146 qemu_put_be32(f, vmdesc_len);
1147 qemu_put_buffer(f, (uint8_t *)qjson_get_str(vmdesc), vmdesc_len);
1148 }
1149 qjson_destroy(vmdesc);
1150
1151 qemu_fflush(f);
1152 }
1153
1154 /* Give an estimate of the amount left to be transferred,
1155 * the result is split into the amount for units that can and
1156 * for units that can't do postcopy.
1157 */
1158 void qemu_savevm_state_pending(QEMUFile *f, uint64_t max_size,
1159 uint64_t *res_non_postcopiable,
1160 uint64_t *res_postcopiable)
1161 {
1162 SaveStateEntry *se;
1163
1164 *res_non_postcopiable = 0;
1165 *res_postcopiable = 0;
1166
1167
1168 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1169 if (!se->ops || !se->ops->save_live_pending) {
1170 continue;
1171 }
1172 if (se->ops && se->ops->is_active) {
1173 if (!se->ops->is_active(se->opaque)) {
1174 continue;
1175 }
1176 }
1177 se->ops->save_live_pending(f, se->opaque, max_size,
1178 res_non_postcopiable, res_postcopiable);
1179 }
1180 }
1181
1182 void qemu_savevm_state_cleanup(void)
1183 {
1184 SaveStateEntry *se;
1185
1186 trace_savevm_state_cleanup();
1187 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1188 if (se->ops && se->ops->cleanup) {
1189 se->ops->cleanup(se->opaque);
1190 }
1191 }
1192 }
1193
1194 static int qemu_savevm_state(QEMUFile *f, Error **errp)
1195 {
1196 int ret;
1197 MigrationParams params = {
1198 .blk = 0,
1199 .shared = 0
1200 };
1201 MigrationState *ms = migrate_init(&params);
1202 MigrationStatus status;
1203 ms->to_dst_file = f;
1204
1205 if (migration_is_blocked(errp)) {
1206 ret = -EINVAL;
1207 goto done;
1208 }
1209
1210 qemu_mutex_unlock_iothread();
1211 qemu_savevm_state_header(f);
1212 qemu_savevm_state_begin(f, &params);
1213 qemu_mutex_lock_iothread();
1214
1215 while (qemu_file_get_error(f) == 0) {
1216 if (qemu_savevm_state_iterate(f, false) > 0) {
1217 break;
1218 }
1219 }
1220
1221 ret = qemu_file_get_error(f);
1222 if (ret == 0) {
1223 qemu_savevm_state_complete_precopy(f, false);
1224 ret = qemu_file_get_error(f);
1225 }
1226 qemu_savevm_state_cleanup();
1227 if (ret != 0) {
1228 error_setg_errno(errp, -ret, "Error while writing VM state");
1229 }
1230
1231 done:
1232 if (ret != 0) {
1233 status = MIGRATION_STATUS_FAILED;
1234 } else {
1235 status = MIGRATION_STATUS_COMPLETED;
1236 }
1237 migrate_set_state(&ms->state, MIGRATION_STATUS_SETUP, status);
1238 return ret;
1239 }
1240
1241 static int qemu_save_device_state(QEMUFile *f)
1242 {
1243 SaveStateEntry *se;
1244
1245 qemu_put_be32(f, QEMU_VM_FILE_MAGIC);
1246 qemu_put_be32(f, QEMU_VM_FILE_VERSION);
1247
1248 cpu_synchronize_all_states();
1249
1250 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1251 if (se->is_ram) {
1252 continue;
1253 }
1254 if ((!se->ops || !se->ops->save_state) && !se->vmsd) {
1255 continue;
1256 }
1257 if (se->vmsd && !vmstate_save_needed(se->vmsd, se->opaque)) {
1258 continue;
1259 }
1260
1261 save_section_header(f, se, QEMU_VM_SECTION_FULL);
1262
1263 vmstate_save(f, se, NULL);
1264
1265 save_section_footer(f, se);
1266 }
1267
1268 qemu_put_byte(f, QEMU_VM_EOF);
1269
1270 return qemu_file_get_error(f);
1271 }
1272
1273 static SaveStateEntry *find_se(const char *idstr, int instance_id)
1274 {
1275 SaveStateEntry *se;
1276
1277 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1278 if (!strcmp(se->idstr, idstr) &&
1279 (instance_id == se->instance_id ||
1280 instance_id == se->alias_id))
1281 return se;
1282 /* Migrating from an older version? */
1283 if (strstr(se->idstr, idstr) && se->compat) {
1284 if (!strcmp(se->compat->idstr, idstr) &&
1285 (instance_id == se->compat->instance_id ||
1286 instance_id == se->alias_id))
1287 return se;
1288 }
1289 }
1290 return NULL;
1291 }
1292
1293 enum LoadVMExitCodes {
1294 /* Allow a command to quit all layers of nested loadvm loops */
1295 LOADVM_QUIT = 1,
1296 };
1297
1298 static int qemu_loadvm_state_main(QEMUFile *f, MigrationIncomingState *mis);
1299
1300 /* ------ incoming postcopy messages ------ */
1301 /* 'advise' arrives before any transfers just to tell us that a postcopy
1302 * *might* happen - it might be skipped if precopy transferred everything
1303 * quickly.
1304 */
1305 static int loadvm_postcopy_handle_advise(MigrationIncomingState *mis)
1306 {
1307 PostcopyState ps = postcopy_state_set(POSTCOPY_INCOMING_ADVISE);
1308 uint64_t remote_hps, remote_tps;
1309
1310 trace_loadvm_postcopy_handle_advise();
1311 if (ps != POSTCOPY_INCOMING_NONE) {
1312 error_report("CMD_POSTCOPY_ADVISE in wrong postcopy state (%d)", ps);
1313 return -1;
1314 }
1315
1316 if (!postcopy_ram_supported_by_host()) {
1317 return -1;
1318 }
1319
1320 remote_hps = qemu_get_be64(mis->from_src_file);
1321 if (remote_hps != getpagesize()) {
1322 /*
1323 * Some combinations of mismatch are probably possible but it gets
1324 * a bit more complicated. In particular we need to place whole
1325 * host pages on the dest at once, and we need to ensure that we
1326 * handle dirtying to make sure we never end up sending part of
1327 * a hostpage on it's own.
1328 */
1329 error_report("Postcopy needs matching host page sizes (s=%d d=%d)",
1330 (int)remote_hps, getpagesize());
1331 return -1;
1332 }
1333
1334 remote_tps = qemu_get_be64(mis->from_src_file);
1335 if (remote_tps != (1ul << qemu_target_page_bits())) {
1336 /*
1337 * Again, some differences could be dealt with, but for now keep it
1338 * simple.
1339 */
1340 error_report("Postcopy needs matching target page sizes (s=%d d=%d)",
1341 (int)remote_tps, 1 << qemu_target_page_bits());
1342 return -1;
1343 }
1344
1345 if (ram_postcopy_incoming_init(mis)) {
1346 return -1;
1347 }
1348
1349 postcopy_state_set(POSTCOPY_INCOMING_ADVISE);
1350
1351 return 0;
1352 }
1353
1354 /* After postcopy we will be told to throw some pages away since they're
1355 * dirty and will have to be demand fetched. Must happen before CPU is
1356 * started.
1357 * There can be 0..many of these messages, each encoding multiple pages.
1358 */
1359 static int loadvm_postcopy_ram_handle_discard(MigrationIncomingState *mis,
1360 uint16_t len)
1361 {
1362 int tmp;
1363 char ramid[256];
1364 PostcopyState ps = postcopy_state_get();
1365
1366 trace_loadvm_postcopy_ram_handle_discard();
1367
1368 switch (ps) {
1369 case POSTCOPY_INCOMING_ADVISE:
1370 /* 1st discard */
1371 tmp = postcopy_ram_prepare_discard(mis);
1372 if (tmp) {
1373 return tmp;
1374 }
1375 break;
1376
1377 case POSTCOPY_INCOMING_DISCARD:
1378 /* Expected state */
1379 break;
1380
1381 default:
1382 error_report("CMD_POSTCOPY_RAM_DISCARD in wrong postcopy state (%d)",
1383 ps);
1384 return -1;
1385 }
1386 /* We're expecting a
1387 * Version (0)
1388 * a RAM ID string (length byte, name, 0 term)
1389 * then at least 1 16 byte chunk
1390 */
1391 if (len < (1 + 1 + 1 + 1 + 2 * 8)) {
1392 error_report("CMD_POSTCOPY_RAM_DISCARD invalid length (%d)", len);
1393 return -1;
1394 }
1395
1396 tmp = qemu_get_byte(mis->from_src_file);
1397 if (tmp != postcopy_ram_discard_version) {
1398 error_report("CMD_POSTCOPY_RAM_DISCARD invalid version (%d)", tmp);
1399 return -1;
1400 }
1401
1402 if (!qemu_get_counted_string(mis->from_src_file, ramid)) {
1403 error_report("CMD_POSTCOPY_RAM_DISCARD Failed to read RAMBlock ID");
1404 return -1;
1405 }
1406 tmp = qemu_get_byte(mis->from_src_file);
1407 if (tmp != 0) {
1408 error_report("CMD_POSTCOPY_RAM_DISCARD missing nil (%d)", tmp);
1409 return -1;
1410 }
1411
1412 len -= 3 + strlen(ramid);
1413 if (len % 16) {
1414 error_report("CMD_POSTCOPY_RAM_DISCARD invalid length (%d)", len);
1415 return -1;
1416 }
1417 trace_loadvm_postcopy_ram_handle_discard_header(ramid, len);
1418 while (len) {
1419 uint64_t start_addr, block_length;
1420 start_addr = qemu_get_be64(mis->from_src_file);
1421 block_length = qemu_get_be64(mis->from_src_file);
1422
1423 len -= 16;
1424 int ret = ram_discard_range(mis, ramid, start_addr,
1425 block_length);
1426 if (ret) {
1427 return ret;
1428 }
1429 }
1430 trace_loadvm_postcopy_ram_handle_discard_end();
1431
1432 return 0;
1433 }
1434
1435 /*
1436 * Triggered by a postcopy_listen command; this thread takes over reading
1437 * the input stream, leaving the main thread free to carry on loading the rest
1438 * of the device state (from RAM).
1439 * (TODO:This could do with being in a postcopy file - but there again it's
1440 * just another input loop, not that postcopy specific)
1441 */
1442 static void *postcopy_ram_listen_thread(void *opaque)
1443 {
1444 QEMUFile *f = opaque;
1445 MigrationIncomingState *mis = migration_incoming_get_current();
1446 int load_res;
1447
1448 migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE,
1449 MIGRATION_STATUS_POSTCOPY_ACTIVE);
1450 qemu_sem_post(&mis->listen_thread_sem);
1451 trace_postcopy_ram_listen_thread_start();
1452
1453 /*
1454 * Because we're a thread and not a coroutine we can't yield
1455 * in qemu_file, and thus we must be blocking now.
1456 */
1457 qemu_file_set_blocking(f, true);
1458 load_res = qemu_loadvm_state_main(f, mis);
1459 /* And non-blocking again so we don't block in any cleanup */
1460 qemu_file_set_blocking(f, false);
1461
1462 trace_postcopy_ram_listen_thread_exit();
1463 if (load_res < 0) {
1464 error_report("%s: loadvm failed: %d", __func__, load_res);
1465 qemu_file_set_error(f, load_res);
1466 migrate_set_state(&mis->state, MIGRATION_STATUS_POSTCOPY_ACTIVE,
1467 MIGRATION_STATUS_FAILED);
1468 } else {
1469 /*
1470 * This looks good, but it's possible that the device loading in the
1471 * main thread hasn't finished yet, and so we might not be in 'RUN'
1472 * state yet; wait for the end of the main thread.
1473 */
1474 qemu_event_wait(&mis->main_thread_load_event);
1475 }
1476 postcopy_ram_incoming_cleanup(mis);
1477
1478 if (load_res < 0) {
1479 /*
1480 * If something went wrong then we have a bad state so exit;
1481 * depending how far we got it might be possible at this point
1482 * to leave the guest running and fire MCEs for pages that never
1483 * arrived as a desperate recovery step.
1484 */
1485 exit(EXIT_FAILURE);
1486 }
1487
1488 migrate_set_state(&mis->state, MIGRATION_STATUS_POSTCOPY_ACTIVE,
1489 MIGRATION_STATUS_COMPLETED);
1490 /*
1491 * If everything has worked fine, then the main thread has waited
1492 * for us to start, and we're the last use of the mis.
1493 * (If something broke then qemu will have to exit anyway since it's
1494 * got a bad migration state).
1495 */
1496 migration_incoming_state_destroy();
1497
1498
1499 return NULL;
1500 }
1501
1502 /* After this message we must be able to immediately receive postcopy data */
1503 static int loadvm_postcopy_handle_listen(MigrationIncomingState *mis)
1504 {
1505 PostcopyState ps = postcopy_state_set(POSTCOPY_INCOMING_LISTENING);
1506 trace_loadvm_postcopy_handle_listen();
1507 if (ps != POSTCOPY_INCOMING_ADVISE && ps != POSTCOPY_INCOMING_DISCARD) {
1508 error_report("CMD_POSTCOPY_LISTEN in wrong postcopy state (%d)", ps);
1509 return -1;
1510 }
1511 if (ps == POSTCOPY_INCOMING_ADVISE) {
1512 /*
1513 * A rare case, we entered listen without having to do any discards,
1514 * so do the setup that's normally done at the time of the 1st discard.
1515 */
1516 postcopy_ram_prepare_discard(mis);
1517 }
1518
1519 /*
1520 * Sensitise RAM - can now generate requests for blocks that don't exist
1521 * However, at this point the CPU shouldn't be running, and the IO
1522 * shouldn't be doing anything yet so don't actually expect requests
1523 */
1524 if (postcopy_ram_enable_notify(mis)) {
1525 return -1;
1526 }
1527
1528 if (mis->have_listen_thread) {
1529 error_report("CMD_POSTCOPY_RAM_LISTEN already has a listen thread");
1530 return -1;
1531 }
1532
1533 mis->have_listen_thread = true;
1534 /* Start up the listening thread and wait for it to signal ready */
1535 qemu_sem_init(&mis->listen_thread_sem, 0);
1536 qemu_thread_create(&mis->listen_thread, "postcopy/listen",
1537 postcopy_ram_listen_thread, mis->from_src_file,
1538 QEMU_THREAD_DETACHED);
1539 qemu_sem_wait(&mis->listen_thread_sem);
1540 qemu_sem_destroy(&mis->listen_thread_sem);
1541
1542 return 0;
1543 }
1544
1545
1546 typedef struct {
1547 QEMUBH *bh;
1548 } HandleRunBhData;
1549
1550 static void loadvm_postcopy_handle_run_bh(void *opaque)
1551 {
1552 Error *local_err = NULL;
1553 HandleRunBhData *data = opaque;
1554
1555 /* TODO we should move all of this lot into postcopy_ram.c or a shared code
1556 * in migration.c
1557 */
1558 cpu_synchronize_all_post_init();
1559
1560 qemu_announce_self();
1561
1562 /* Make sure all file formats flush their mutable metadata */
1563 bdrv_invalidate_cache_all(&local_err);
1564 if (local_err) {
1565 error_report_err(local_err);
1566 }
1567
1568 trace_loadvm_postcopy_handle_run_cpu_sync();
1569 cpu_synchronize_all_post_init();
1570
1571 trace_loadvm_postcopy_handle_run_vmstart();
1572
1573 if (autostart) {
1574 /* Hold onto your hats, starting the CPU */
1575 vm_start();
1576 } else {
1577 /* leave it paused and let management decide when to start the CPU */
1578 runstate_set(RUN_STATE_PAUSED);
1579 }
1580
1581 qemu_bh_delete(data->bh);
1582 g_free(data);
1583 }
1584
1585 /* After all discards we can start running and asking for pages */
1586 static int loadvm_postcopy_handle_run(MigrationIncomingState *mis)
1587 {
1588 PostcopyState ps = postcopy_state_set(POSTCOPY_INCOMING_RUNNING);
1589 HandleRunBhData *data;
1590
1591 trace_loadvm_postcopy_handle_run();
1592 if (ps != POSTCOPY_INCOMING_LISTENING) {
1593 error_report("CMD_POSTCOPY_RUN in wrong postcopy state (%d)", ps);
1594 return -1;
1595 }
1596
1597 data = g_new(HandleRunBhData, 1);
1598 data->bh = qemu_bh_new(loadvm_postcopy_handle_run_bh, data);
1599 qemu_bh_schedule(data->bh);
1600
1601 /* We need to finish reading the stream from the package
1602 * and also stop reading anything more from the stream that loaded the
1603 * package (since it's now being read by the listener thread).
1604 * LOADVM_QUIT will quit all the layers of nested loadvm loops.
1605 */
1606 return LOADVM_QUIT;
1607 }
1608
1609 /**
1610 * Immediately following this command is a blob of data containing an embedded
1611 * chunk of migration stream; read it and load it.
1612 *
1613 * @mis: Incoming state
1614 * @length: Length of packaged data to read
1615 *
1616 * Returns: Negative values on error
1617 *
1618 */
1619 static int loadvm_handle_cmd_packaged(MigrationIncomingState *mis)
1620 {
1621 int ret;
1622 size_t length;
1623 QIOChannelBuffer *bioc;
1624
1625 length = qemu_get_be32(mis->from_src_file);
1626 trace_loadvm_handle_cmd_packaged(length);
1627
1628 if (length > MAX_VM_CMD_PACKAGED_SIZE) {
1629 error_report("Unreasonably large packaged state: %zu", length);
1630 return -1;
1631 }
1632
1633 bioc = qio_channel_buffer_new(length);
1634 qio_channel_set_name(QIO_CHANNEL(bioc), "migration-loadvm-buffer");
1635 ret = qemu_get_buffer(mis->from_src_file,
1636 bioc->data,
1637 length);
1638 if (ret != length) {
1639 object_unref(OBJECT(bioc));
1640 error_report("CMD_PACKAGED: Buffer receive fail ret=%d length=%zu",
1641 ret, length);
1642 return (ret < 0) ? ret : -EAGAIN;
1643 }
1644 bioc->usage += length;
1645 trace_loadvm_handle_cmd_packaged_received(ret);
1646
1647 QEMUFile *packf = qemu_fopen_channel_input(QIO_CHANNEL(bioc));
1648
1649 ret = qemu_loadvm_state_main(packf, mis);
1650 trace_loadvm_handle_cmd_packaged_main(ret);
1651 qemu_fclose(packf);
1652 object_unref(OBJECT(bioc));
1653
1654 return ret;
1655 }
1656
1657 /*
1658 * Process an incoming 'QEMU_VM_COMMAND'
1659 * 0 just a normal return
1660 * LOADVM_QUIT All good, but exit the loop
1661 * <0 Error
1662 */
1663 static int loadvm_process_command(QEMUFile *f)
1664 {
1665 MigrationIncomingState *mis = migration_incoming_get_current();
1666 uint16_t cmd;
1667 uint16_t len;
1668 uint32_t tmp32;
1669
1670 cmd = qemu_get_be16(f);
1671 len = qemu_get_be16(f);
1672
1673 trace_loadvm_process_command(cmd, len);
1674 if (cmd >= MIG_CMD_MAX || cmd == MIG_CMD_INVALID) {
1675 error_report("MIG_CMD 0x%x unknown (len 0x%x)", cmd, len);
1676 return -EINVAL;
1677 }
1678
1679 if (mig_cmd_args[cmd].len != -1 && mig_cmd_args[cmd].len != len) {
1680 error_report("%s received with bad length - expecting %zu, got %d",
1681 mig_cmd_args[cmd].name,
1682 (size_t)mig_cmd_args[cmd].len, len);
1683 return -ERANGE;
1684 }
1685
1686 switch (cmd) {
1687 case MIG_CMD_OPEN_RETURN_PATH:
1688 if (mis->to_src_file) {
1689 error_report("CMD_OPEN_RETURN_PATH called when RP already open");
1690 /* Not really a problem, so don't give up */
1691 return 0;
1692 }
1693 mis->to_src_file = qemu_file_get_return_path(f);
1694 if (!mis->to_src_file) {
1695 error_report("CMD_OPEN_RETURN_PATH failed");
1696 return -1;
1697 }
1698 break;
1699
1700 case MIG_CMD_PING:
1701 tmp32 = qemu_get_be32(f);
1702 trace_loadvm_process_command_ping(tmp32);
1703 if (!mis->to_src_file) {
1704 error_report("CMD_PING (0x%x) received with no return path",
1705 tmp32);
1706 return -1;
1707 }
1708 migrate_send_rp_pong(mis, tmp32);
1709 break;
1710
1711 case MIG_CMD_PACKAGED:
1712 return loadvm_handle_cmd_packaged(mis);
1713
1714 case MIG_CMD_POSTCOPY_ADVISE:
1715 return loadvm_postcopy_handle_advise(mis);
1716
1717 case MIG_CMD_POSTCOPY_LISTEN:
1718 return loadvm_postcopy_handle_listen(mis);
1719
1720 case MIG_CMD_POSTCOPY_RUN:
1721 return loadvm_postcopy_handle_run(mis);
1722
1723 case MIG_CMD_POSTCOPY_RAM_DISCARD:
1724 return loadvm_postcopy_ram_handle_discard(mis, len);
1725 }
1726
1727 return 0;
1728 }
1729
1730 struct LoadStateEntry {
1731 QLIST_ENTRY(LoadStateEntry) entry;
1732 SaveStateEntry *se;
1733 int section_id;
1734 int version_id;
1735 };
1736
1737 /*
1738 * Read a footer off the wire and check that it matches the expected section
1739 *
1740 * Returns: true if the footer was good
1741 * false if there is a problem (and calls error_report to say why)
1742 */
1743 static bool check_section_footer(QEMUFile *f, LoadStateEntry *le)
1744 {
1745 uint8_t read_mark;
1746 uint32_t read_section_id;
1747
1748 if (skip_section_footers) {
1749 /* No footer to check */
1750 return true;
1751 }
1752
1753 read_mark = qemu_get_byte(f);
1754
1755 if (read_mark != QEMU_VM_SECTION_FOOTER) {
1756 error_report("Missing section footer for %s", le->se->idstr);
1757 return false;
1758 }
1759
1760 read_section_id = qemu_get_be32(f);
1761 if (read_section_id != le->section_id) {
1762 error_report("Mismatched section id in footer for %s -"
1763 " read 0x%x expected 0x%x",
1764 le->se->idstr, read_section_id, le->section_id);
1765 return false;
1766 }
1767
1768 /* All good */
1769 return true;
1770 }
1771
1772 void loadvm_free_handlers(MigrationIncomingState *mis)
1773 {
1774 LoadStateEntry *le, *new_le;
1775
1776 QLIST_FOREACH_SAFE(le, &mis->loadvm_handlers, entry, new_le) {
1777 QLIST_REMOVE(le, entry);
1778 g_free(le);
1779 }
1780 }
1781
1782 static int
1783 qemu_loadvm_section_start_full(QEMUFile *f, MigrationIncomingState *mis)
1784 {
1785 uint32_t instance_id, version_id, section_id;
1786 SaveStateEntry *se;
1787 LoadStateEntry *le;
1788 char idstr[256];
1789 int ret;
1790
1791 /* Read section start */
1792 section_id = qemu_get_be32(f);
1793 if (!qemu_get_counted_string(f, idstr)) {
1794 error_report("Unable to read ID string for section %u",
1795 section_id);
1796 return -EINVAL;
1797 }
1798 instance_id = qemu_get_be32(f);
1799 version_id = qemu_get_be32(f);
1800
1801 trace_qemu_loadvm_state_section_startfull(section_id, idstr,
1802 instance_id, version_id);
1803 /* Find savevm section */
1804 se = find_se(idstr, instance_id);
1805 if (se == NULL) {
1806 error_report("Unknown savevm section or instance '%s' %d",
1807 idstr, instance_id);
1808 return -EINVAL;
1809 }
1810
1811 /* Validate version */
1812 if (version_id > se->version_id) {
1813 error_report("savevm: unsupported version %d for '%s' v%d",
1814 version_id, idstr, se->version_id);
1815 return -EINVAL;
1816 }
1817
1818 /* Validate if it is a device's state */
1819 if (xen_enabled() && se->is_ram) {
1820 error_report("loadvm: %s RAM loading not allowed on Xen", idstr);
1821 return -EINVAL;
1822 }
1823
1824 /* Add entry */
1825 le = g_malloc0(sizeof(*le));
1826
1827 le->se = se;
1828 le->section_id = section_id;
1829 le->version_id = version_id;
1830 QLIST_INSERT_HEAD(&mis->loadvm_handlers, le, entry);
1831
1832 ret = vmstate_load(f, le->se, le->version_id);
1833 if (ret < 0) {
1834 error_report("error while loading state for instance 0x%x of"
1835 " device '%s'", instance_id, idstr);
1836 return ret;
1837 }
1838 if (!check_section_footer(f, le)) {
1839 return -EINVAL;
1840 }
1841
1842 return 0;
1843 }
1844
1845 static int
1846 qemu_loadvm_section_part_end(QEMUFile *f, MigrationIncomingState *mis)
1847 {
1848 uint32_t section_id;
1849 LoadStateEntry *le;
1850 int ret;
1851
1852 section_id = qemu_get_be32(f);
1853
1854 trace_qemu_loadvm_state_section_partend(section_id);
1855 QLIST_FOREACH(le, &mis->loadvm_handlers, entry) {
1856 if (le->section_id == section_id) {
1857 break;
1858 }
1859 }
1860 if (le == NULL) {
1861 error_report("Unknown savevm section %d", section_id);
1862 return -EINVAL;
1863 }
1864
1865 ret = vmstate_load(f, le->se, le->version_id);
1866 if (ret < 0) {
1867 error_report("error while loading state section id %d(%s)",
1868 section_id, le->se->idstr);
1869 return ret;
1870 }
1871 if (!check_section_footer(f, le)) {
1872 return -EINVAL;
1873 }
1874
1875 return 0;
1876 }
1877
1878 static int qemu_loadvm_state_main(QEMUFile *f, MigrationIncomingState *mis)
1879 {
1880 uint8_t section_type;
1881 int ret = 0;
1882
1883 while ((section_type = qemu_get_byte(f)) != QEMU_VM_EOF) {
1884 ret = 0;
1885 trace_qemu_loadvm_state_section(section_type);
1886 switch (section_type) {
1887 case QEMU_VM_SECTION_START:
1888 case QEMU_VM_SECTION_FULL:
1889 ret = qemu_loadvm_section_start_full(f, mis);
1890 if (ret < 0) {
1891 goto out;
1892 }
1893 break;
1894 case QEMU_VM_SECTION_PART:
1895 case QEMU_VM_SECTION_END:
1896 ret = qemu_loadvm_section_part_end(f, mis);
1897 if (ret < 0) {
1898 goto out;
1899 }
1900 break;
1901 case QEMU_VM_COMMAND:
1902 ret = loadvm_process_command(f);
1903 trace_qemu_loadvm_state_section_command(ret);
1904 if ((ret < 0) || (ret & LOADVM_QUIT)) {
1905 goto out;
1906 }
1907 break;
1908 default:
1909 error_report("Unknown savevm section type %d", section_type);
1910 ret = -EINVAL;
1911 goto out;
1912 }
1913 }
1914
1915 out:
1916 if (ret < 0) {
1917 qemu_file_set_error(f, ret);
1918 }
1919 return ret;
1920 }
1921
1922 int qemu_loadvm_state(QEMUFile *f)
1923 {
1924 MigrationIncomingState *mis = migration_incoming_get_current();
1925 Error *local_err = NULL;
1926 unsigned int v;
1927 int ret;
1928
1929 if (qemu_savevm_state_blocked(&local_err)) {
1930 error_report_err(local_err);
1931 return -EINVAL;
1932 }
1933
1934 v = qemu_get_be32(f);
1935 if (v != QEMU_VM_FILE_MAGIC) {
1936 error_report("Not a migration stream");
1937 return -EINVAL;
1938 }
1939
1940 v = qemu_get_be32(f);
1941 if (v == QEMU_VM_FILE_VERSION_COMPAT) {
1942 error_report("SaveVM v2 format is obsolete and don't work anymore");
1943 return -ENOTSUP;
1944 }
1945 if (v != QEMU_VM_FILE_VERSION) {
1946 error_report("Unsupported migration stream version");
1947 return -ENOTSUP;
1948 }
1949
1950 if (!savevm_state.skip_configuration || enforce_config_section()) {
1951 if (qemu_get_byte(f) != QEMU_VM_CONFIGURATION) {
1952 error_report("Configuration section missing");
1953 return -EINVAL;
1954 }
1955 ret = vmstate_load_state(f, &vmstate_configuration, &savevm_state, 0);
1956
1957 if (ret) {
1958 return ret;
1959 }
1960 }
1961
1962 ret = qemu_loadvm_state_main(f, mis);
1963 qemu_event_set(&mis->main_thread_load_event);
1964
1965 trace_qemu_loadvm_state_post_main(ret);
1966
1967 if (mis->have_listen_thread) {
1968 /* Listen thread still going, can't clean up yet */
1969 return ret;
1970 }
1971
1972 if (ret == 0) {
1973 ret = qemu_file_get_error(f);
1974 }
1975
1976 /*
1977 * Try to read in the VMDESC section as well, so that dumping tools that
1978 * intercept our migration stream have the chance to see it.
1979 */
1980
1981 /* We've got to be careful; if we don't read the data and just shut the fd
1982 * then the sender can error if we close while it's still sending.
1983 * We also mustn't read data that isn't there; some transports (RDMA)
1984 * will stall waiting for that data when the source has already closed.
1985 */
1986 if (ret == 0 && should_send_vmdesc()) {
1987 uint8_t *buf;
1988 uint32_t size;
1989 uint8_t section_type = qemu_get_byte(f);
1990
1991 if (section_type != QEMU_VM_VMDESCRIPTION) {
1992 error_report("Expected vmdescription section, but got %d",
1993 section_type);
1994 /*
1995 * It doesn't seem worth failing at this point since
1996 * we apparently have an otherwise valid VM state
1997 */
1998 } else {
1999 buf = g_malloc(0x1000);
2000 size = qemu_get_be32(f);
2001
2002 while (size > 0) {
2003 uint32_t read_chunk = MIN(size, 0x1000);
2004 qemu_get_buffer(f, buf, read_chunk);
2005 size -= read_chunk;
2006 }
2007 g_free(buf);
2008 }
2009 }
2010
2011 cpu_synchronize_all_post_init();
2012
2013 return ret;
2014 }
2015
2016 void hmp_savevm(Monitor *mon, const QDict *qdict)
2017 {
2018 BlockDriverState *bs, *bs1;
2019 QEMUSnapshotInfo sn1, *sn = &sn1, old_sn1, *old_sn = &old_sn1;
2020 int ret;
2021 QEMUFile *f;
2022 int saved_vm_running;
2023 uint64_t vm_state_size;
2024 qemu_timeval tv;
2025 struct tm tm;
2026 const char *name = qdict_get_try_str(qdict, "name");
2027 Error *local_err = NULL;
2028 AioContext *aio_context;
2029
2030 if (!bdrv_all_can_snapshot(&bs)) {
2031 monitor_printf(mon, "Device '%s' is writable but does not "
2032 "support snapshots.\n", bdrv_get_device_name(bs));
2033 return;
2034 }
2035
2036 /* Delete old snapshots of the same name */
2037 if (name && bdrv_all_delete_snapshot(name, &bs1, &local_err) < 0) {
2038 error_reportf_err(local_err,
2039 "Error while deleting snapshot on device '%s': ",
2040 bdrv_get_device_name(bs1));
2041 return;
2042 }
2043
2044 bs = bdrv_all_find_vmstate_bs();
2045 if (bs == NULL) {
2046 monitor_printf(mon, "No block device can accept snapshots\n");
2047 return;
2048 }
2049 aio_context = bdrv_get_aio_context(bs);
2050
2051 saved_vm_running = runstate_is_running();
2052
2053 ret = global_state_store();
2054 if (ret) {
2055 monitor_printf(mon, "Error saving global state\n");
2056 return;
2057 }
2058 vm_stop(RUN_STATE_SAVE_VM);
2059
2060 aio_context_acquire(aio_context);
2061
2062 memset(sn, 0, sizeof(*sn));
2063
2064 /* fill auxiliary fields */
2065 qemu_gettimeofday(&tv);
2066 sn->date_sec = tv.tv_sec;
2067 sn->date_nsec = tv.tv_usec * 1000;
2068 sn->vm_clock_nsec = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL);
2069
2070 if (name) {
2071 ret = bdrv_snapshot_find(bs, old_sn, name);
2072 if (ret >= 0) {
2073 pstrcpy(sn->name, sizeof(sn->name), old_sn->name);
2074 pstrcpy(sn->id_str, sizeof(sn->id_str), old_sn->id_str);
2075 } else {
2076 pstrcpy(sn->name, sizeof(sn->name), name);
2077 }
2078 } else {
2079 /* cast below needed for OpenBSD where tv_sec is still 'long' */
2080 localtime_r((const time_t *)&tv.tv_sec, &tm);
2081 strftime(sn->name, sizeof(sn->name), "vm-%Y%m%d%H%M%S", &tm);
2082 }
2083
2084 /* save the VM state */
2085 f = qemu_fopen_bdrv(bs, 1);
2086 if (!f) {
2087 monitor_printf(mon, "Could not open VM state file\n");
2088 goto the_end;
2089 }
2090 ret = qemu_savevm_state(f, &local_err);
2091 vm_state_size = qemu_ftell(f);
2092 qemu_fclose(f);
2093 if (ret < 0) {
2094 error_report_err(local_err);
2095 goto the_end;
2096 }
2097
2098 ret = bdrv_all_create_snapshot(sn, bs, vm_state_size, &bs);
2099 if (ret < 0) {
2100 monitor_printf(mon, "Error while creating snapshot on '%s'\n",
2101 bdrv_get_device_name(bs));
2102 }
2103
2104 the_end:
2105 aio_context_release(aio_context);
2106 if (saved_vm_running) {
2107 vm_start();
2108 }
2109 }
2110
2111 void qmp_xen_save_devices_state(const char *filename, Error **errp)
2112 {
2113 QEMUFile *f;
2114 QIOChannelFile *ioc;
2115 int saved_vm_running;
2116 int ret;
2117
2118 saved_vm_running = runstate_is_running();
2119 vm_stop(RUN_STATE_SAVE_VM);
2120 global_state_store_running();
2121
2122 ioc = qio_channel_file_new_path(filename, O_WRONLY | O_CREAT, 0660, errp);
2123 if (!ioc) {
2124 goto the_end;
2125 }
2126 qio_channel_set_name(QIO_CHANNEL(ioc), "migration-xen-save-state");
2127 f = qemu_fopen_channel_output(QIO_CHANNEL(ioc));
2128 ret = qemu_save_device_state(f);
2129 qemu_fclose(f);
2130 if (ret < 0) {
2131 error_setg(errp, QERR_IO_ERROR);
2132 }
2133
2134 the_end:
2135 if (saved_vm_running) {
2136 vm_start();
2137 }
2138 }
2139
2140 void qmp_xen_load_devices_state(const char *filename, Error **errp)
2141 {
2142 QEMUFile *f;
2143 QIOChannelFile *ioc;
2144 int ret;
2145
2146 /* Guest must be paused before loading the device state; the RAM state
2147 * will already have been loaded by xc
2148 */
2149 if (runstate_is_running()) {
2150 error_setg(errp, "Cannot update device state while vm is running");
2151 return;
2152 }
2153 vm_stop(RUN_STATE_RESTORE_VM);
2154
2155 ioc = qio_channel_file_new_path(filename, O_RDONLY | O_BINARY, 0, errp);
2156 if (!ioc) {
2157 return;
2158 }
2159 qio_channel_set_name(QIO_CHANNEL(ioc), "migration-xen-load-state");
2160 f = qemu_fopen_channel_input(QIO_CHANNEL(ioc));
2161
2162 migration_incoming_state_new(f);
2163 ret = qemu_loadvm_state(f);
2164 qemu_fclose(f);
2165 if (ret < 0) {
2166 error_setg(errp, QERR_IO_ERROR);
2167 }
2168 migration_incoming_state_destroy();
2169 }
2170
2171 int load_vmstate(const char *name)
2172 {
2173 BlockDriverState *bs, *bs_vm_state;
2174 QEMUSnapshotInfo sn;
2175 QEMUFile *f;
2176 int ret;
2177 AioContext *aio_context;
2178
2179 if (!bdrv_all_can_snapshot(&bs)) {
2180 error_report("Device '%s' is writable but does not support snapshots.",
2181 bdrv_get_device_name(bs));
2182 return -ENOTSUP;
2183 }
2184 ret = bdrv_all_find_snapshot(name, &bs);
2185 if (ret < 0) {
2186 error_report("Device '%s' does not have the requested snapshot '%s'",
2187 bdrv_get_device_name(bs), name);
2188 return ret;
2189 }
2190
2191 bs_vm_state = bdrv_all_find_vmstate_bs();
2192 if (!bs_vm_state) {
2193 error_report("No block device supports snapshots");
2194 return -ENOTSUP;
2195 }
2196 aio_context = bdrv_get_aio_context(bs_vm_state);
2197
2198 /* Don't even try to load empty VM states */
2199 aio_context_acquire(aio_context);
2200 ret = bdrv_snapshot_find(bs_vm_state, &sn, name);
2201 aio_context_release(aio_context);
2202 if (ret < 0) {
2203 return ret;
2204 } else if (sn.vm_state_size == 0) {
2205 error_report("This is a disk-only snapshot. Revert to it offline "
2206 "using qemu-img.");
2207 return -EINVAL;
2208 }
2209
2210 /* Flush all IO requests so they don't interfere with the new state. */
2211 bdrv_drain_all();
2212
2213 ret = bdrv_all_goto_snapshot(name, &bs);
2214 if (ret < 0) {
2215 error_report("Error %d while activating snapshot '%s' on '%s'",
2216 ret, name, bdrv_get_device_name(bs));
2217 return ret;
2218 }
2219
2220 /* restore the VM state */
2221 f = qemu_fopen_bdrv(bs_vm_state, 0);
2222 if (!f) {
2223 error_report("Could not open VM state file");
2224 return -EINVAL;
2225 }
2226
2227 qemu_system_reset(VMRESET_SILENT);
2228 migration_incoming_state_new(f);
2229
2230 aio_context_acquire(aio_context);
2231 ret = qemu_loadvm_state(f);
2232 qemu_fclose(f);
2233 aio_context_release(aio_context);
2234
2235 migration_incoming_state_destroy();
2236 if (ret < 0) {
2237 error_report("Error %d while loading VM state", ret);
2238 return ret;
2239 }
2240
2241 return 0;
2242 }
2243
2244 void hmp_delvm(Monitor *mon, const QDict *qdict)
2245 {
2246 BlockDriverState *bs;
2247 Error *err;
2248 const char *name = qdict_get_str(qdict, "name");
2249
2250 if (bdrv_all_delete_snapshot(name, &bs, &err) < 0) {
2251 error_reportf_err(err,
2252 "Error while deleting snapshot on device '%s': ",
2253 bdrv_get_device_name(bs));
2254 }
2255 }
2256
2257 void hmp_info_snapshots(Monitor *mon, const QDict *qdict)
2258 {
2259 BlockDriverState *bs, *bs1;
2260 BdrvNextIterator it1;
2261 QEMUSnapshotInfo *sn_tab, *sn;
2262 bool no_snapshot = true;
2263 int nb_sns, i;
2264 int total;
2265 int *global_snapshots;
2266 AioContext *aio_context;
2267
2268 typedef struct SnapshotEntry {
2269 QEMUSnapshotInfo sn;
2270 QTAILQ_ENTRY(SnapshotEntry) next;
2271 } SnapshotEntry;
2272
2273 typedef struct ImageEntry {
2274 const char *imagename;
2275 QTAILQ_ENTRY(ImageEntry) next;
2276 QTAILQ_HEAD(, SnapshotEntry) snapshots;
2277 } ImageEntry;
2278
2279 QTAILQ_HEAD(, ImageEntry) image_list =
2280 QTAILQ_HEAD_INITIALIZER(image_list);
2281
2282 ImageEntry *image_entry, *next_ie;
2283 SnapshotEntry *snapshot_entry;
2284
2285 bs = bdrv_all_find_vmstate_bs();
2286 if (!bs) {
2287 monitor_printf(mon, "No available block device supports snapshots\n");
2288 return;
2289 }
2290 aio_context = bdrv_get_aio_context(bs);
2291
2292 aio_context_acquire(aio_context);
2293 nb_sns = bdrv_snapshot_list(bs, &sn_tab);
2294 aio_context_release(aio_context);
2295
2296 if (nb_sns < 0) {
2297 monitor_printf(mon, "bdrv_snapshot_list: error %d\n", nb_sns);
2298 return;
2299 }
2300
2301 for (bs1 = bdrv_first(&it1); bs1; bs1 = bdrv_next(&it1)) {
2302 int bs1_nb_sns = 0;
2303 ImageEntry *ie;
2304 SnapshotEntry *se;
2305 AioContext *ctx = bdrv_get_aio_context(bs1);
2306
2307 aio_context_acquire(ctx);
2308 if (bdrv_can_snapshot(bs1)) {
2309 sn = NULL;
2310 bs1_nb_sns = bdrv_snapshot_list(bs1, &sn);
2311 if (bs1_nb_sns > 0) {
2312 no_snapshot = false;
2313 ie = g_new0(ImageEntry, 1);
2314 ie->imagename = bdrv_get_device_name(bs1);
2315 QTAILQ_INIT(&ie->snapshots);
2316 QTAILQ_INSERT_TAIL(&image_list, ie, next);
2317 for (i = 0; i < bs1_nb_sns; i++) {
2318 se = g_new0(SnapshotEntry, 1);
2319 se->sn = sn[i];
2320 QTAILQ_INSERT_TAIL(&ie->snapshots, se, next);
2321 }
2322 }
2323 g_free(sn);
2324 }
2325 aio_context_release(ctx);
2326 }
2327
2328 if (no_snapshot) {
2329 monitor_printf(mon, "There is no snapshot available.\n");
2330 return;
2331 }
2332
2333 global_snapshots = g_new0(int, nb_sns);
2334 total = 0;
2335 for (i = 0; i < nb_sns; i++) {
2336 SnapshotEntry *next_sn;
2337 if (bdrv_all_find_snapshot(sn_tab[i].name, &bs1) == 0) {
2338 global_snapshots[total] = i;
2339 total++;
2340 QTAILQ_FOREACH(image_entry, &image_list, next) {
2341 QTAILQ_FOREACH_SAFE(snapshot_entry, &image_entry->snapshots,
2342 next, next_sn) {
2343 if (!strcmp(sn_tab[i].name, snapshot_entry->sn.name)) {
2344 QTAILQ_REMOVE(&image_entry->snapshots, snapshot_entry,
2345 next);
2346 g_free(snapshot_entry);
2347 }
2348 }
2349 }
2350 }
2351 }
2352
2353 monitor_printf(mon, "List of snapshots present on all disks:\n");
2354
2355 if (total > 0) {
2356 bdrv_snapshot_dump((fprintf_function)monitor_printf, mon, NULL);
2357 monitor_printf(mon, "\n");
2358 for (i = 0; i < total; i++) {
2359 sn = &sn_tab[global_snapshots[i]];
2360 /* The ID is not guaranteed to be the same on all images, so
2361 * overwrite it.
2362 */
2363 pstrcpy(sn->id_str, sizeof(sn->id_str), "--");
2364 bdrv_snapshot_dump((fprintf_function)monitor_printf, mon, sn);
2365 monitor_printf(mon, "\n");
2366 }
2367 } else {
2368 monitor_printf(mon, "None\n");
2369 }
2370
2371 QTAILQ_FOREACH(image_entry, &image_list, next) {
2372 if (QTAILQ_EMPTY(&image_entry->snapshots)) {
2373 continue;
2374 }
2375 monitor_printf(mon,
2376 "\nList of partial (non-loadable) snapshots on '%s':\n",
2377 image_entry->imagename);
2378 bdrv_snapshot_dump((fprintf_function)monitor_printf, mon, NULL);
2379 monitor_printf(mon, "\n");
2380 QTAILQ_FOREACH(snapshot_entry, &image_entry->snapshots, next) {
2381 bdrv_snapshot_dump((fprintf_function)monitor_printf, mon,
2382 &snapshot_entry->sn);
2383 monitor_printf(mon, "\n");
2384 }
2385 }
2386
2387 QTAILQ_FOREACH_SAFE(image_entry, &image_list, next, next_ie) {
2388 SnapshotEntry *next_sn;
2389 QTAILQ_FOREACH_SAFE(snapshot_entry, &image_entry->snapshots, next,
2390 next_sn) {
2391 g_free(snapshot_entry);
2392 }
2393 g_free(image_entry);
2394 }
2395 g_free(sn_tab);
2396 g_free(global_snapshots);
2397
2398 }
2399
2400 void vmstate_register_ram(MemoryRegion *mr, DeviceState *dev)
2401 {
2402 qemu_ram_set_idstr(mr->ram_block,
2403 memory_region_name(mr), dev);
2404 }
2405
2406 void vmstate_unregister_ram(MemoryRegion *mr, DeviceState *dev)
2407 {
2408 qemu_ram_unset_idstr(mr->ram_block);
2409 }
2410
2411 void vmstate_register_ram_global(MemoryRegion *mr)
2412 {
2413 vmstate_register_ram(mr, NULL);
2414 }