]> git.proxmox.com Git - mirror_qemu.git/blame - migration/savevm.c
cpu: Support a target CPU having a variable page size
[mirror_qemu.git] / migration / savevm.c
CommitLineData
a672b469
AL
1/*
2 * QEMU System Emulator
3 *
4 * Copyright (c) 2003-2008 Fabrice Bellard
76cc7b58
JQ
5 * Copyright (c) 2009-2015 Red Hat Inc
6 *
7 * Authors:
8 * Juan Quintela <quintela@redhat.com>
a672b469
AL
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 */
a672b469 28
1393a485 29#include "qemu/osdep.h"
33c11879 30#include "cpu.h"
abfd9ce3 31#include "hw/boards.h"
511d2b14 32#include "hw/hw.h"
7685ee6a 33#include "hw/qdev.h"
88c16567 34#include "hw/xen/xen.h"
1422e32d 35#include "net/net.h"
83c9089e 36#include "monitor/monitor.h"
9c17d615 37#include "sysemu/sysemu.h"
1de7afc9 38#include "qemu/timer.h"
511d2b14 39#include "audio/audio.h"
caf71f86 40#include "migration/migration.h"
eb59db53 41#include "migration/postcopy-ram.h"
cc7a8ea7 42#include "qapi/qmp/qerror.h"
d49b6836 43#include "qemu/error-report.h"
1de7afc9
PB
44#include "qemu/sockets.h"
45#include "qemu/queue.h"
9c17d615 46#include "sysemu/cpus.h"
022c62cb 47#include "exec/memory.h"
a7ae8355 48#include "qmp-commands.h"
517a13c9 49#include "trace.h"
093e3c42 50#include "qemu/bitops.h"
28085f7b 51#include "qemu/iov.h"
de08c606 52#include "block/snapshot.h"
f364ec65 53#include "block/qapi.h"
f348b6d1 54#include "qemu/cutils.h"
61b67d47 55#include "io/channel-buffer.h"
8925839f 56#include "io/channel-file.h"
a672b469 57
18995b98 58#ifndef ETH_P_RARP
f8778a77 59#define ETH_P_RARP 0x8035
18995b98
N
60#endif
61#define ARP_HTYPE_ETH 0x0001
62#define ARP_PTYPE_IP 0x0800
63#define ARP_OP_REQUEST_REV 0x3
64
093e3c42
DDAG
65const unsigned int postcopy_ram_discard_version = 0;
66
37fb569c
DDAG
67static bool skip_section_footers;
68
c76ca188
DDAG
69static 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" },
2e37701e
DDAG
74 [MIG_CMD_OPEN_RETURN_PATH] = { .len = 0, .name = "OPEN_RETURN_PATH" },
75 [MIG_CMD_PING] = { .len = sizeof(uint32_t), .name = "PING" },
093e3c42
DDAG
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" },
11cf1d98 81 [MIG_CMD_PACKAGED] = { .len = 4, .name = "PACKAGED" },
c76ca188
DDAG
82 [MIG_CMD_MAX] = { .len = -1, .name = "MAX" },
83};
84
18995b98 85static int announce_self_create(uint8_t *buf,
5cecf414 86 uint8_t *mac_addr)
a672b469 87{
18995b98
N
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) */
a672b469
AL
108}
109
f401ca22 110static void qemu_announce_self_iter(NICState *nic, void *opaque)
a672b469 111{
18995b98 112 uint8_t buf[60];
f401ca22
MM
113 int len;
114
9013dca5 115 trace_qemu_announce_self_iter(qemu_ether_ntoa(&nic->conf->macaddr));
f401ca22
MM
116 len = announce_self_create(buf, nic->conf->macaddr.a);
117
b356f76d 118 qemu_send_packet_raw(qemu_get_queue(nic), buf, len);
f401ca22
MM
119}
120
121
122static void qemu_announce_self_once(void *opaque)
123{
ed8b330b
GN
124 static int count = SELF_ANNOUNCE_ROUNDS;
125 QEMUTimer *timer = *(QEMUTimer **)opaque;
a672b469 126
f401ca22
MM
127 qemu_foreach_nic(qemu_announce_self_iter, NULL);
128
18995b98
N
129 if (--count) {
130 /* delay 50ms, 150ms, 250ms, ... */
bc72ad67 131 timer_mod(timer, qemu_clock_get_ms(QEMU_CLOCK_REALTIME) +
508e1180 132 self_announce_delay(count));
ed8b330b 133 } else {
5cecf414
EH
134 timer_del(timer);
135 timer_free(timer);
ed8b330b
GN
136 }
137}
138
139void qemu_announce_self(void)
140{
5cecf414
EH
141 static QEMUTimer *timer;
142 timer = timer_new_ms(QEMU_CLOCK_REALTIME, qemu_announce_self_once, &timer);
143 qemu_announce_self_once(&timer);
a672b469
AL
144}
145
146/***********************************************************/
147/* savevm/loadvm support */
148
05fcc848
KW
149static 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
a202a4c0
DDAG
164static ssize_t block_get_buffer(void *opaque, uint8_t *buf, int64_t pos,
165 size_t size)
a672b469 166{
45566e9c 167 return bdrv_load_vmstate(opaque, buf, pos, size);
a672b469
AL
168}
169
170static int bdrv_fclose(void *opaque)
171{
ad492c92 172 return bdrv_flush(opaque);
a672b469
AL
173}
174
9229bf3c
PB
175static const QEMUFileOps bdrv_read_ops = {
176 .get_buffer = block_get_buffer,
177 .close = bdrv_fclose
178};
179
180static const QEMUFileOps bdrv_write_ops = {
05fcc848
KW
181 .writev_buffer = block_writev_buffer,
182 .close = bdrv_fclose
9229bf3c
PB
183};
184
45566e9c 185static QEMUFile *qemu_fopen_bdrv(BlockDriverState *bs, int is_writable)
a672b469 186{
38ff78d3 187 if (is_writable) {
9229bf3c 188 return qemu_fopen_ops(bs, &bdrv_write_ops);
38ff78d3 189 }
9229bf3c 190 return qemu_fopen_ops(bs, &bdrv_read_ops);
a672b469
AL
191}
192
2ff68d07 193
bb1a6d8c
EH
194/* QEMUFile timer support.
195 * Not in qemu-file.c to not add qemu-timer.c as dependency to qemu-file.c
196 */
2ff68d07 197
40daca54 198void timer_put(QEMUFile *f, QEMUTimer *ts)
2ff68d07
PB
199{
200 uint64_t expire_time;
201
e93379b0 202 expire_time = timer_expire_time_ns(ts);
2ff68d07
PB
203 qemu_put_be64(f, expire_time);
204}
205
40daca54 206void timer_get(QEMUFile *f, QEMUTimer *ts)
2ff68d07
PB
207{
208 uint64_t expire_time;
209
210 expire_time = qemu_get_be64(f);
211 if (expire_time != -1) {
bc72ad67 212 timer_mod_ns(ts, expire_time);
2ff68d07 213 } else {
bc72ad67 214 timer_del(ts);
2ff68d07
PB
215 }
216}
217
218
bb1a6d8c
EH
219/* VMState timer support.
220 * Not in vmstate.c to not add qemu-timer.c as dependency to vmstate.c
221 */
dde0463b
JQ
222
223static int get_timer(QEMUFile *f, void *pv, size_t size)
224{
225 QEMUTimer *v = pv;
40daca54 226 timer_get(f, v);
dde0463b
JQ
227 return 0;
228}
229
84e2e3eb 230static void put_timer(QEMUFile *f, void *pv, size_t size)
dde0463b 231{
84e2e3eb 232 QEMUTimer *v = pv;
40daca54 233 timer_put(f, v);
dde0463b
JQ
234}
235
236const VMStateInfo vmstate_info_timer = {
237 .name = "timer",
238 .get = get_timer,
239 .put = put_timer,
240};
241
08e99e29 242
7685ee6a
AW
243typedef struct CompatEntry {
244 char idstr[256];
245 int instance_id;
246} CompatEntry;
247
a672b469 248typedef struct SaveStateEntry {
72cf2d4f 249 QTAILQ_ENTRY(SaveStateEntry) entry;
a672b469
AL
250 char idstr[256];
251 int instance_id;
4d2ffa08 252 int alias_id;
a672b469
AL
253 int version_id;
254 int section_id;
22ea40f4 255 SaveVMHandlers *ops;
9ed7d6ae 256 const VMStateDescription *vmsd;
a672b469 257 void *opaque;
7685ee6a 258 CompatEntry *compat;
a7ae8355 259 int is_ram;
a672b469
AL
260} SaveStateEntry;
261
0163a2e0
JQ
262typedef struct SaveState {
263 QTAILQ_HEAD(, SaveStateEntry) handlers;
264 int global_section_id;
61964c23
JQ
265 bool skip_configuration;
266 uint32_t len;
267 const char *name;
0163a2e0
JQ
268} SaveState;
269
270static SaveState savevm_state = {
271 .handlers = QTAILQ_HEAD_INITIALIZER(savevm_state.handlers),
272 .global_section_id = 0,
61964c23
JQ
273 .skip_configuration = false,
274};
275
276void savevm_skip_configuration(void)
277{
278 savevm_state.skip_configuration = true;
279}
280
281
282static void configuration_pre_save(void *opaque)
283{
284 SaveState *state = opaque;
285 const char *current_name = MACHINE_GET_CLASS(current_machine)->name;
286
287 state->len = strlen(current_name);
288 state->name = current_name;
289}
290
291static int configuration_post_load(void *opaque, int version_id)
292{
293 SaveState *state = opaque;
294 const char *current_name = MACHINE_GET_CLASS(current_machine)->name;
295
296 if (strncmp(state->name, current_name, state->len) != 0) {
15d61692
GK
297 error_report("Machine type received is '%.*s' and local is '%s'",
298 (int) state->len, state->name, current_name);
61964c23
JQ
299 return -EINVAL;
300 }
301 return 0;
302}
303
304static const VMStateDescription vmstate_configuration = {
305 .name = "configuration",
306 .version_id = 1,
307 .post_load = configuration_post_load,
308 .pre_save = configuration_pre_save,
309 .fields = (VMStateField[]) {
310 VMSTATE_UINT32(len, SaveState),
311 VMSTATE_VBUFFER_ALLOC_UINT32(name, SaveState, 0, NULL, 0, len),
312 VMSTATE_END_OF_LIST()
313 },
0163a2e0 314};
a672b469 315
abfd9ce3
AS
316static void dump_vmstate_vmsd(FILE *out_file,
317 const VMStateDescription *vmsd, int indent,
318 bool is_subsection);
319
320static void dump_vmstate_vmsf(FILE *out_file, const VMStateField *field,
321 int indent)
322{
323 fprintf(out_file, "%*s{\n", indent, "");
324 indent += 2;
325 fprintf(out_file, "%*s\"field\": \"%s\",\n", indent, "", field->name);
326 fprintf(out_file, "%*s\"version_id\": %d,\n", indent, "",
327 field->version_id);
328 fprintf(out_file, "%*s\"field_exists\": %s,\n", indent, "",
329 field->field_exists ? "true" : "false");
330 fprintf(out_file, "%*s\"size\": %zu", indent, "", field->size);
331 if (field->vmsd != NULL) {
332 fprintf(out_file, ",\n");
333 dump_vmstate_vmsd(out_file, field->vmsd, indent, false);
334 }
335 fprintf(out_file, "\n%*s}", indent - 2, "");
336}
337
338static void dump_vmstate_vmss(FILE *out_file,
5cd8cada 339 const VMStateDescription **subsection,
abfd9ce3
AS
340 int indent)
341{
5cd8cada
JQ
342 if (*subsection != NULL) {
343 dump_vmstate_vmsd(out_file, *subsection, indent, true);
abfd9ce3
AS
344 }
345}
346
347static void dump_vmstate_vmsd(FILE *out_file,
348 const VMStateDescription *vmsd, int indent,
349 bool is_subsection)
350{
351 if (is_subsection) {
352 fprintf(out_file, "%*s{\n", indent, "");
353 } else {
354 fprintf(out_file, "%*s\"%s\": {\n", indent, "", "Description");
355 }
356 indent += 2;
357 fprintf(out_file, "%*s\"name\": \"%s\",\n", indent, "", vmsd->name);
358 fprintf(out_file, "%*s\"version_id\": %d,\n", indent, "",
359 vmsd->version_id);
360 fprintf(out_file, "%*s\"minimum_version_id\": %d", indent, "",
361 vmsd->minimum_version_id);
362 if (vmsd->fields != NULL) {
363 const VMStateField *field = vmsd->fields;
364 bool first;
365
366 fprintf(out_file, ",\n%*s\"Fields\": [\n", indent, "");
367 first = true;
368 while (field->name != NULL) {
369 if (field->flags & VMS_MUST_EXIST) {
370 /* Ignore VMSTATE_VALIDATE bits; these don't get migrated */
371 field++;
372 continue;
373 }
374 if (!first) {
375 fprintf(out_file, ",\n");
376 }
377 dump_vmstate_vmsf(out_file, field, indent + 2);
378 field++;
379 first = false;
380 }
381 fprintf(out_file, "\n%*s]", indent, "");
382 }
383 if (vmsd->subsections != NULL) {
5cd8cada 384 const VMStateDescription **subsection = vmsd->subsections;
abfd9ce3
AS
385 bool first;
386
387 fprintf(out_file, ",\n%*s\"Subsections\": [\n", indent, "");
388 first = true;
5cd8cada 389 while (*subsection != NULL) {
abfd9ce3
AS
390 if (!first) {
391 fprintf(out_file, ",\n");
392 }
393 dump_vmstate_vmss(out_file, subsection, indent + 2);
394 subsection++;
395 first = false;
396 }
397 fprintf(out_file, "\n%*s]", indent, "");
398 }
399 fprintf(out_file, "\n%*s}", indent - 2, "");
400}
401
402static void dump_machine_type(FILE *out_file)
403{
404 MachineClass *mc;
405
406 mc = MACHINE_GET_CLASS(current_machine);
407
408 fprintf(out_file, " \"vmschkmachine\": {\n");
409 fprintf(out_file, " \"Name\": \"%s\"\n", mc->name);
410 fprintf(out_file, " },\n");
411}
412
413void dump_vmstate_json_to_file(FILE *out_file)
414{
415 GSList *list, *elt;
416 bool first;
417
418 fprintf(out_file, "{\n");
419 dump_machine_type(out_file);
420
421 first = true;
422 list = object_class_get_list(TYPE_DEVICE, true);
423 for (elt = list; elt; elt = elt->next) {
424 DeviceClass *dc = OBJECT_CLASS_CHECK(DeviceClass, elt->data,
425 TYPE_DEVICE);
426 const char *name;
427 int indent = 2;
428
429 if (!dc->vmsd) {
430 continue;
431 }
432
433 if (!first) {
434 fprintf(out_file, ",\n");
435 }
436 name = object_class_get_name(OBJECT_CLASS(dc));
437 fprintf(out_file, "%*s\"%s\": {\n", indent, "", name);
438 indent += 2;
439 fprintf(out_file, "%*s\"Name\": \"%s\",\n", indent, "", name);
440 fprintf(out_file, "%*s\"version_id\": %d,\n", indent, "",
441 dc->vmsd->version_id);
442 fprintf(out_file, "%*s\"minimum_version_id\": %d,\n", indent, "",
443 dc->vmsd->minimum_version_id);
444
445 dump_vmstate_vmsd(out_file, dc->vmsd, indent, false);
446
447 fprintf(out_file, "\n%*s}", indent - 2, "");
448 first = false;
449 }
450 fprintf(out_file, "\n}\n");
451 fclose(out_file);
452}
453
8718e999
JQ
454static int calculate_new_instance_id(const char *idstr)
455{
456 SaveStateEntry *se;
457 int instance_id = 0;
458
0163a2e0 459 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
8718e999
JQ
460 if (strcmp(idstr, se->idstr) == 0
461 && instance_id <= se->instance_id) {
462 instance_id = se->instance_id + 1;
463 }
464 }
465 return instance_id;
466}
467
7685ee6a
AW
468static int calculate_compat_instance_id(const char *idstr)
469{
470 SaveStateEntry *se;
471 int instance_id = 0;
472
0163a2e0 473 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
38ff78d3 474 if (!se->compat) {
7685ee6a 475 continue;
38ff78d3 476 }
7685ee6a
AW
477
478 if (strcmp(idstr, se->compat->idstr) == 0
479 && instance_id <= se->compat->instance_id) {
480 instance_id = se->compat->instance_id + 1;
481 }
482 }
483 return instance_id;
484}
485
a672b469
AL
486/* TODO: Individual devices generally have very little idea about the rest
487 of the system, so instance_id should be removed/replaced.
488 Meanwhile pass -1 as instance_id if you do not already have a clearly
489 distinguishing id for all instances of your device class. */
0be71e32
AW
490int register_savevm_live(DeviceState *dev,
491 const char *idstr,
a672b469
AL
492 int instance_id,
493 int version_id,
7908c78d 494 SaveVMHandlers *ops,
a672b469
AL
495 void *opaque)
496{
8718e999 497 SaveStateEntry *se;
a672b469 498
97f3ad35 499 se = g_new0(SaveStateEntry, 1);
a672b469 500 se->version_id = version_id;
0163a2e0 501 se->section_id = savevm_state.global_section_id++;
7908c78d 502 se->ops = ops;
a672b469 503 se->opaque = opaque;
9ed7d6ae 504 se->vmsd = NULL;
a7ae8355 505 /* if this is a live_savem then set is_ram */
16310a3c 506 if (ops->save_live_setup != NULL) {
a7ae8355
SS
507 se->is_ram = 1;
508 }
a672b469 509
09e5ab63
AL
510 if (dev) {
511 char *id = qdev_get_dev_path(dev);
7685ee6a
AW
512 if (id) {
513 pstrcpy(se->idstr, sizeof(se->idstr), id);
514 pstrcat(se->idstr, sizeof(se->idstr), "/");
7267c094 515 g_free(id);
7685ee6a 516
97f3ad35 517 se->compat = g_new0(CompatEntry, 1);
7685ee6a
AW
518 pstrcpy(se->compat->idstr, sizeof(se->compat->idstr), idstr);
519 se->compat->instance_id = instance_id == -1 ?
520 calculate_compat_instance_id(idstr) : instance_id;
521 instance_id = -1;
522 }
523 }
524 pstrcat(se->idstr, sizeof(se->idstr), idstr);
525
8718e999 526 if (instance_id == -1) {
7685ee6a 527 se->instance_id = calculate_new_instance_id(se->idstr);
8718e999
JQ
528 } else {
529 se->instance_id = instance_id;
a672b469 530 }
7685ee6a 531 assert(!se->compat || se->instance_id == 0);
8718e999 532 /* add at the end of list */
0163a2e0 533 QTAILQ_INSERT_TAIL(&savevm_state.handlers, se, entry);
a672b469
AL
534 return 0;
535}
536
0be71e32
AW
537int register_savevm(DeviceState *dev,
538 const char *idstr,
a672b469
AL
539 int instance_id,
540 int version_id,
541 SaveStateHandler *save_state,
542 LoadStateHandler *load_state,
543 void *opaque)
544{
97f3ad35 545 SaveVMHandlers *ops = g_new0(SaveVMHandlers, 1);
7908c78d
JQ
546 ops->save_state = save_state;
547 ops->load_state = load_state;
0be71e32 548 return register_savevm_live(dev, idstr, instance_id, version_id,
7908c78d 549 ops, opaque);
a672b469
AL
550}
551
0be71e32 552void unregister_savevm(DeviceState *dev, const char *idstr, void *opaque)
41bd13af 553{
8718e999 554 SaveStateEntry *se, *new_se;
7685ee6a
AW
555 char id[256] = "";
556
09e5ab63
AL
557 if (dev) {
558 char *path = qdev_get_dev_path(dev);
7685ee6a
AW
559 if (path) {
560 pstrcpy(id, sizeof(id), path);
561 pstrcat(id, sizeof(id), "/");
7267c094 562 g_free(path);
7685ee6a
AW
563 }
564 }
565 pstrcat(id, sizeof(id), idstr);
41bd13af 566
0163a2e0 567 QTAILQ_FOREACH_SAFE(se, &savevm_state.handlers, entry, new_se) {
7685ee6a 568 if (strcmp(se->idstr, id) == 0 && se->opaque == opaque) {
0163a2e0 569 QTAILQ_REMOVE(&savevm_state.handlers, se, entry);
ef1e1e07 570 g_free(se->compat);
22ea40f4 571 g_free(se->ops);
7267c094 572 g_free(se);
41bd13af 573 }
41bd13af
AL
574 }
575}
576
0be71e32 577int vmstate_register_with_alias_id(DeviceState *dev, int instance_id,
4d2ffa08
JK
578 const VMStateDescription *vmsd,
579 void *opaque, int alias_id,
580 int required_for_version)
9ed7d6ae 581{
8718e999 582 SaveStateEntry *se;
9ed7d6ae 583
4d2ffa08
JK
584 /* If this triggers, alias support can be dropped for the vmsd. */
585 assert(alias_id == -1 || required_for_version >= vmsd->minimum_version_id);
586
97f3ad35 587 se = g_new0(SaveStateEntry, 1);
9ed7d6ae 588 se->version_id = vmsd->version_id;
0163a2e0 589 se->section_id = savevm_state.global_section_id++;
9ed7d6ae
JQ
590 se->opaque = opaque;
591 se->vmsd = vmsd;
4d2ffa08 592 se->alias_id = alias_id;
9ed7d6ae 593
09e5ab63
AL
594 if (dev) {
595 char *id = qdev_get_dev_path(dev);
7685ee6a
AW
596 if (id) {
597 pstrcpy(se->idstr, sizeof(se->idstr), id);
598 pstrcat(se->idstr, sizeof(se->idstr), "/");
7267c094 599 g_free(id);
7685ee6a 600
97f3ad35 601 se->compat = g_new0(CompatEntry, 1);
7685ee6a
AW
602 pstrcpy(se->compat->idstr, sizeof(se->compat->idstr), vmsd->name);
603 se->compat->instance_id = instance_id == -1 ?
604 calculate_compat_instance_id(vmsd->name) : instance_id;
605 instance_id = -1;
606 }
607 }
608 pstrcat(se->idstr, sizeof(se->idstr), vmsd->name);
609
8718e999 610 if (instance_id == -1) {
7685ee6a 611 se->instance_id = calculate_new_instance_id(se->idstr);
8718e999
JQ
612 } else {
613 se->instance_id = instance_id;
9ed7d6ae 614 }
7685ee6a 615 assert(!se->compat || se->instance_id == 0);
8718e999 616 /* add at the end of list */
0163a2e0 617 QTAILQ_INSERT_TAIL(&savevm_state.handlers, se, entry);
9ed7d6ae
JQ
618 return 0;
619}
620
0be71e32
AW
621void vmstate_unregister(DeviceState *dev, const VMStateDescription *vmsd,
622 void *opaque)
9ed7d6ae 623{
1eb7538b
JQ
624 SaveStateEntry *se, *new_se;
625
0163a2e0 626 QTAILQ_FOREACH_SAFE(se, &savevm_state.handlers, entry, new_se) {
1eb7538b 627 if (se->vmsd == vmsd && se->opaque == opaque) {
0163a2e0 628 QTAILQ_REMOVE(&savevm_state.handlers, se, entry);
ef1e1e07 629 g_free(se->compat);
7267c094 630 g_free(se);
1eb7538b
JQ
631 }
632 }
9ed7d6ae
JQ
633}
634
4082be4d
JQ
635static int vmstate_load(QEMUFile *f, SaveStateEntry *se, int version_id)
636{
9013dca5 637 trace_vmstate_load(se->idstr, se->vmsd ? se->vmsd->name : "(old)");
9ed7d6ae 638 if (!se->vmsd) { /* Old style */
22ea40f4 639 return se->ops->load_state(f, se->opaque, version_id);
9ed7d6ae
JQ
640 }
641 return vmstate_load_state(f, se->vmsd, se->opaque, version_id);
4082be4d
JQ
642}
643
8118f095
AG
644static void vmstate_save_old_style(QEMUFile *f, SaveStateEntry *se, QJSON *vmdesc)
645{
646 int64_t old_offset, size;
647
648 old_offset = qemu_ftell_fast(f);
649 se->ops->save_state(f, se->opaque);
650 size = qemu_ftell_fast(f) - old_offset;
651
652 if (vmdesc) {
653 json_prop_int(vmdesc, "size", size);
654 json_start_array(vmdesc, "fields");
655 json_start_object(vmdesc, NULL);
656 json_prop_str(vmdesc, "name", "data");
657 json_prop_int(vmdesc, "size", size);
658 json_prop_str(vmdesc, "type", "buffer");
659 json_end_object(vmdesc);
660 json_end_array(vmdesc);
661 }
662}
663
664static void vmstate_save(QEMUFile *f, SaveStateEntry *se, QJSON *vmdesc)
4082be4d 665{
9013dca5 666 trace_vmstate_save(se->idstr, se->vmsd ? se->vmsd->name : "(old)");
8118f095
AG
667 if (!se->vmsd) {
668 vmstate_save_old_style(f, se, vmdesc);
dc912121 669 return;
9ed7d6ae 670 }
8118f095 671 vmstate_save_state(f, se->vmsd, se->opaque, vmdesc);
4082be4d
JQ
672}
673
37fb569c
DDAG
674void savevm_skip_section_footers(void)
675{
676 skip_section_footers = true;
677}
678
ce39bfc9
DDAG
679/*
680 * Write the header for device section (QEMU_VM_SECTION START/END/PART/FULL)
681 */
682static void save_section_header(QEMUFile *f, SaveStateEntry *se,
683 uint8_t section_type)
684{
685 qemu_put_byte(f, section_type);
686 qemu_put_be32(f, se->section_id);
687
688 if (section_type == QEMU_VM_SECTION_FULL ||
689 section_type == QEMU_VM_SECTION_START) {
690 /* ID string */
691 size_t len = strlen(se->idstr);
692 qemu_put_byte(f, len);
693 qemu_put_buffer(f, (uint8_t *)se->idstr, len);
694
695 qemu_put_be32(f, se->instance_id);
696 qemu_put_be32(f, se->version_id);
697 }
698}
699
f68945d4
DDAG
700/*
701 * Write a footer onto device sections that catches cases misformatted device
702 * sections.
703 */
704static void save_section_footer(QEMUFile *f, SaveStateEntry *se)
705{
706 if (!skip_section_footers) {
707 qemu_put_byte(f, QEMU_VM_SECTION_FOOTER);
708 qemu_put_be32(f, se->section_id);
709 }
710}
711
c76ca188
DDAG
712/**
713 * qemu_savevm_command_send: Send a 'QEMU_VM_COMMAND' type element with the
714 * command and associated data.
715 *
716 * @f: File to send command on
717 * @command: Command type to send
718 * @len: Length of associated data
719 * @data: Data associated with command.
720 */
721void qemu_savevm_command_send(QEMUFile *f,
722 enum qemu_vm_cmd command,
723 uint16_t len,
724 uint8_t *data)
725{
726 trace_savevm_command_send(command, len);
727 qemu_put_byte(f, QEMU_VM_COMMAND);
728 qemu_put_be16(f, (uint16_t)command);
729 qemu_put_be16(f, len);
730 qemu_put_buffer(f, data, len);
731 qemu_fflush(f);
732}
733
2e37701e
DDAG
734void qemu_savevm_send_ping(QEMUFile *f, uint32_t value)
735{
736 uint32_t buf;
737
738 trace_savevm_send_ping(value);
739 buf = cpu_to_be32(value);
740 qemu_savevm_command_send(f, MIG_CMD_PING, sizeof(value), (uint8_t *)&buf);
741}
742
743void qemu_savevm_send_open_return_path(QEMUFile *f)
744{
745 trace_savevm_send_open_return_path();
746 qemu_savevm_command_send(f, MIG_CMD_OPEN_RETURN_PATH, 0, NULL);
747}
748
11cf1d98
DDAG
749/* We have a buffer of data to send; we don't want that all to be loaded
750 * by the command itself, so the command contains just the length of the
751 * extra buffer that we then send straight after it.
752 * TODO: Must be a better way to organise that
753 *
754 * Returns:
755 * 0 on success
756 * -ve on error
757 */
61b67d47 758int qemu_savevm_send_packaged(QEMUFile *f, const uint8_t *buf, size_t len)
11cf1d98 759{
11cf1d98
DDAG
760 uint32_t tmp;
761
762 if (len > MAX_VM_CMD_PACKAGED_SIZE) {
763 error_report("%s: Unreasonably large packaged state: %zu",
764 __func__, len);
765 return -1;
766 }
767
768 tmp = cpu_to_be32(len);
769
770 trace_qemu_savevm_send_packaged();
771 qemu_savevm_command_send(f, MIG_CMD_PACKAGED, 4, (uint8_t *)&tmp);
772
61b67d47 773 qemu_put_buffer(f, buf, len);
11cf1d98
DDAG
774
775 return 0;
776}
777
093e3c42
DDAG
778/* Send prior to any postcopy transfer */
779void qemu_savevm_send_postcopy_advise(QEMUFile *f)
780{
781 uint64_t tmp[2];
782 tmp[0] = cpu_to_be64(getpagesize());
783 tmp[1] = cpu_to_be64(1ul << qemu_target_page_bits());
784
785 trace_qemu_savevm_send_postcopy_advise();
786 qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_ADVISE, 16, (uint8_t *)tmp);
787}
788
789/* Sent prior to starting the destination running in postcopy, discard pages
790 * that have already been sent but redirtied on the source.
791 * CMD_POSTCOPY_RAM_DISCARD consist of:
792 * byte version (0)
793 * byte Length of name field (not including 0)
794 * n x byte RAM block name
795 * byte 0 terminator (just for safety)
796 * n x Byte ranges within the named RAMBlock
797 * be64 Start of the range
798 * be64 Length
799 *
800 * name: RAMBlock name that these entries are part of
801 * len: Number of page entries
802 * start_list: 'len' addresses
803 * length_list: 'len' addresses
804 *
805 */
806void qemu_savevm_send_postcopy_ram_discard(QEMUFile *f, const char *name,
807 uint16_t len,
808 uint64_t *start_list,
809 uint64_t *length_list)
810{
811 uint8_t *buf;
812 uint16_t tmplen;
813 uint16_t t;
814 size_t name_len = strlen(name);
815
816 trace_qemu_savevm_send_postcopy_ram_discard(name, len);
817 assert(name_len < 256);
818 buf = g_malloc0(1 + 1 + name_len + 1 + (8 + 8) * len);
819 buf[0] = postcopy_ram_discard_version;
820 buf[1] = name_len;
821 memcpy(buf + 2, name, name_len);
822 tmplen = 2 + name_len;
823 buf[tmplen++] = '\0';
824
825 for (t = 0; t < len; t++) {
4d885131 826 stq_be_p(buf + tmplen, start_list[t]);
093e3c42 827 tmplen += 8;
4d885131 828 stq_be_p(buf + tmplen, length_list[t]);
093e3c42
DDAG
829 tmplen += 8;
830 }
831 qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_RAM_DISCARD, tmplen, buf);
832 g_free(buf);
833}
834
835/* Get the destination into a state where it can receive postcopy data. */
836void qemu_savevm_send_postcopy_listen(QEMUFile *f)
837{
838 trace_savevm_send_postcopy_listen();
839 qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_LISTEN, 0, NULL);
840}
841
842/* Kick the destination into running */
843void qemu_savevm_send_postcopy_run(QEMUFile *f)
844{
845 trace_savevm_send_postcopy_run();
846 qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_RUN, 0, NULL);
847}
848
e1c37d0e 849bool qemu_savevm_state_blocked(Error **errp)
dc912121
AW
850{
851 SaveStateEntry *se;
852
0163a2e0 853 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
7d854c47 854 if (se->vmsd && se->vmsd->unmigratable) {
f231b88d
CR
855 error_setg(errp, "State blocked by non-migratable device '%s'",
856 se->idstr);
dc912121
AW
857 return true;
858 }
859 }
860 return false;
861}
862
902c053d
GK
863static bool enforce_config_section(void)
864{
865 MachineState *machine = MACHINE(qdev_get_machine());
866 return machine->enforce_config_section;
867}
868
f796baa1
DDAG
869void qemu_savevm_state_header(QEMUFile *f)
870{
871 trace_savevm_state_header();
872 qemu_put_be32(f, QEMU_VM_FILE_MAGIC);
873 qemu_put_be32(f, QEMU_VM_FILE_VERSION);
172dfd4f 874
902c053d 875 if (!savevm_state.skip_configuration || enforce_config_section()) {
172dfd4f
DDAG
876 qemu_put_byte(f, QEMU_VM_CONFIGURATION);
877 vmstate_save_state(f, &vmstate_configuration, &savevm_state, 0);
878 }
879
f796baa1
DDAG
880}
881
47c8c17a
PB
882void qemu_savevm_state_begin(QEMUFile *f,
883 const MigrationParams *params)
a672b469
AL
884{
885 SaveStateEntry *se;
39346385 886 int ret;
a672b469 887
9013dca5 888 trace_savevm_state_begin();
0163a2e0 889 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
22ea40f4 890 if (!se->ops || !se->ops->set_params) {
c163b5ca 891 continue;
6607ae23 892 }
22ea40f4 893 se->ops->set_params(params, se->opaque);
c163b5ca 894 }
38ff78d3 895
0163a2e0 896 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
d1315aac 897 if (!se->ops || !se->ops->save_live_setup) {
a672b469 898 continue;
22ea40f4 899 }
6bd68781
JQ
900 if (se->ops && se->ops->is_active) {
901 if (!se->ops->is_active(se->opaque)) {
902 continue;
903 }
904 }
ce39bfc9 905 save_section_header(f, se, QEMU_VM_SECTION_START);
a672b469 906
d1315aac 907 ret = se->ops->save_live_setup(f, se->opaque);
f68945d4 908 save_section_footer(f, se);
2975725f 909 if (ret < 0) {
47c8c17a
PB
910 qemu_file_set_error(f, ret);
911 break;
2975725f 912 }
a672b469 913 }
a672b469
AL
914}
915
39346385 916/*
07f35073 917 * this function has three return values:
39346385
JQ
918 * negative: there was one error, and we have -errno.
919 * 0 : We haven't finished, caller have to go again
920 * 1 : We have finished, we can go to complete phase
921 */
35ecd943 922int qemu_savevm_state_iterate(QEMUFile *f, bool postcopy)
a672b469
AL
923{
924 SaveStateEntry *se;
925 int ret = 1;
926
9013dca5 927 trace_savevm_state_iterate();
0163a2e0 928 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
16310a3c 929 if (!se->ops || !se->ops->save_live_iterate) {
a672b469 930 continue;
22ea40f4 931 }
6bd68781
JQ
932 if (se->ops && se->ops->is_active) {
933 if (!se->ops->is_active(se->opaque)) {
934 continue;
935 }
936 }
35ecd943
DDAG
937 /*
938 * In the postcopy phase, any device that doesn't know how to
939 * do postcopy should have saved it's state in the _complete
940 * call that's already run, it might get confused if we call
941 * iterate afterwards.
942 */
943 if (postcopy && !se->ops->save_live_complete_postcopy) {
944 continue;
945 }
aac844ed
JQ
946 if (qemu_file_rate_limit(f)) {
947 return 0;
948 }
464400f6 949 trace_savevm_section_start(se->idstr, se->section_id);
ce39bfc9
DDAG
950
951 save_section_header(f, se, QEMU_VM_SECTION_PART);
a672b469 952
16310a3c 953 ret = se->ops->save_live_iterate(f, se->opaque);
a5df2a02 954 trace_savevm_section_end(se->idstr, se->section_id, ret);
f68945d4 955 save_section_footer(f, se);
517a13c9 956
47c8c17a
PB
957 if (ret < 0) {
958 qemu_file_set_error(f, ret);
959 }
2975725f 960 if (ret <= 0) {
90697be8
JK
961 /* Do not proceed to the next vmstate before this one reported
962 completion of the current stage. This serializes the migration
963 and reduces the probability that a faster changing state is
964 synchronized over and over again. */
965 break;
966 }
a672b469 967 }
39346385 968 return ret;
a672b469
AL
969}
970
9850c604
AG
971static bool should_send_vmdesc(void)
972{
973 MachineState *machine = MACHINE(qdev_get_machine());
8421b205
DDAG
974 bool in_postcopy = migration_in_postcopy(migrate_get_current());
975 return !machine->suppress_vmdesc && !in_postcopy;
9850c604
AG
976}
977
763c906b
DDAG
978/*
979 * Calls the save_live_complete_postcopy methods
980 * causing the last few pages to be sent immediately and doing any associated
981 * cleanup.
982 * Note postcopy also calls qemu_savevm_state_complete_precopy to complete
983 * all the other devices, but that happens at the point we switch to postcopy.
984 */
985void qemu_savevm_state_complete_postcopy(QEMUFile *f)
986{
987 SaveStateEntry *se;
988 int ret;
989
990 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
991 if (!se->ops || !se->ops->save_live_complete_postcopy) {
992 continue;
993 }
994 if (se->ops && se->ops->is_active) {
995 if (!se->ops->is_active(se->opaque)) {
996 continue;
997 }
998 }
999 trace_savevm_section_start(se->idstr, se->section_id);
1000 /* Section type */
1001 qemu_put_byte(f, QEMU_VM_SECTION_END);
1002 qemu_put_be32(f, se->section_id);
1003
1004 ret = se->ops->save_live_complete_postcopy(f, se->opaque);
1005 trace_savevm_section_end(se->idstr, se->section_id, ret);
1006 save_section_footer(f, se);
1007 if (ret < 0) {
1008 qemu_file_set_error(f, ret);
1009 return;
1010 }
1011 }
1012
1013 qemu_put_byte(f, QEMU_VM_EOF);
1014 qemu_fflush(f);
1015}
1016
1c0d249d 1017void qemu_savevm_state_complete_precopy(QEMUFile *f, bool iterable_only)
a672b469 1018{
8118f095
AG
1019 QJSON *vmdesc;
1020 int vmdesc_len;
a672b469 1021 SaveStateEntry *se;
2975725f 1022 int ret;
763c906b 1023 bool in_postcopy = migration_in_postcopy(migrate_get_current());
a672b469 1024
a3e06c3d 1025 trace_savevm_state_complete_precopy();
9013dca5 1026
ea375f9a
JK
1027 cpu_synchronize_all_states();
1028
0163a2e0 1029 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
763c906b
DDAG
1030 if (!se->ops ||
1031 (in_postcopy && se->ops->save_live_complete_postcopy) ||
1c0d249d 1032 (in_postcopy && !iterable_only) ||
763c906b 1033 !se->ops->save_live_complete_precopy) {
a672b469 1034 continue;
22ea40f4 1035 }
1c0d249d 1036
6bd68781
JQ
1037 if (se->ops && se->ops->is_active) {
1038 if (!se->ops->is_active(se->opaque)) {
1039 continue;
1040 }
1041 }
464400f6 1042 trace_savevm_section_start(se->idstr, se->section_id);
ce39bfc9
DDAG
1043
1044 save_section_header(f, se, QEMU_VM_SECTION_END);
a672b469 1045
a3e06c3d 1046 ret = se->ops->save_live_complete_precopy(f, se->opaque);
a5df2a02 1047 trace_savevm_section_end(se->idstr, se->section_id, ret);
f68945d4 1048 save_section_footer(f, se);
2975725f 1049 if (ret < 0) {
47c8c17a
PB
1050 qemu_file_set_error(f, ret);
1051 return;
2975725f 1052 }
a672b469
AL
1053 }
1054
1c0d249d
DDAG
1055 if (iterable_only) {
1056 return;
1057 }
1058
8118f095
AG
1059 vmdesc = qjson_new();
1060 json_prop_int(vmdesc, "page_size", TARGET_PAGE_SIZE);
1061 json_start_array(vmdesc, "devices");
0163a2e0 1062 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
a672b469 1063
22ea40f4 1064 if ((!se->ops || !se->ops->save_state) && !se->vmsd) {
5cecf414 1065 continue;
22ea40f4 1066 }
df896152
JQ
1067 if (se->vmsd && !vmstate_save_needed(se->vmsd, se->opaque)) {
1068 trace_savevm_section_skip(se->idstr, se->section_id);
1069 continue;
1070 }
1071
464400f6 1072 trace_savevm_section_start(se->idstr, se->section_id);
8118f095
AG
1073
1074 json_start_object(vmdesc, NULL);
1075 json_prop_str(vmdesc, "name", se->idstr);
1076 json_prop_int(vmdesc, "instance_id", se->instance_id);
1077
ce39bfc9 1078 save_section_header(f, se, QEMU_VM_SECTION_FULL);
8118f095 1079 vmstate_save(f, se, vmdesc);
a5df2a02 1080 trace_savevm_section_end(se->idstr, se->section_id, 0);
f68945d4 1081 save_section_footer(f, se);
bdf46d64
WY
1082
1083 json_end_object(vmdesc);
a672b469
AL
1084 }
1085
763c906b
DDAG
1086 if (!in_postcopy) {
1087 /* Postcopy stream will still be going */
1088 qemu_put_byte(f, QEMU_VM_EOF);
1089 }
8118f095
AG
1090
1091 json_end_array(vmdesc);
1092 qjson_finish(vmdesc);
1093 vmdesc_len = strlen(qjson_get_str(vmdesc));
1094
9850c604
AG
1095 if (should_send_vmdesc()) {
1096 qemu_put_byte(f, QEMU_VM_VMDESCRIPTION);
1097 qemu_put_be32(f, vmdesc_len);
1098 qemu_put_buffer(f, (uint8_t *)qjson_get_str(vmdesc), vmdesc_len);
1099 }
b72fe9e6 1100 qjson_destroy(vmdesc);
8118f095 1101
edaae611 1102 qemu_fflush(f);
a672b469
AL
1103}
1104
c31b098f
DDAG
1105/* Give an estimate of the amount left to be transferred,
1106 * the result is split into the amount for units that can and
1107 * for units that can't do postcopy.
1108 */
1109void qemu_savevm_state_pending(QEMUFile *f, uint64_t max_size,
1110 uint64_t *res_non_postcopiable,
1111 uint64_t *res_postcopiable)
e4ed1541
JQ
1112{
1113 SaveStateEntry *se;
c31b098f
DDAG
1114
1115 *res_non_postcopiable = 0;
1116 *res_postcopiable = 0;
1117
e4ed1541 1118
0163a2e0 1119 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
e4ed1541
JQ
1120 if (!se->ops || !se->ops->save_live_pending) {
1121 continue;
1122 }
1123 if (se->ops && se->ops->is_active) {
1124 if (!se->ops->is_active(se->opaque)) {
1125 continue;
1126 }
1127 }
c31b098f
DDAG
1128 se->ops->save_live_pending(f, se->opaque, max_size,
1129 res_non_postcopiable, res_postcopiable);
e4ed1541 1130 }
e4ed1541
JQ
1131}
1132
ea7415fa 1133void qemu_savevm_state_cleanup(void)
4ec7fcc7
JK
1134{
1135 SaveStateEntry *se;
1136
ea7415fa 1137 trace_savevm_state_cleanup();
0163a2e0 1138 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
d1a8548c
LL
1139 if (se->ops && se->ops->cleanup) {
1140 se->ops->cleanup(se->opaque);
4ec7fcc7
JK
1141 }
1142 }
1143}
1144
5d80448c 1145static int qemu_savevm_state(QEMUFile *f, Error **errp)
a672b469 1146{
a672b469 1147 int ret;
6607ae23
IY
1148 MigrationParams params = {
1149 .blk = 0,
1150 .shared = 0
1151 };
aefeb18b 1152 MigrationState *ms = migrate_init(&params);
6dcf6668 1153 MigrationStatus status;
89a02a9f 1154 ms->to_dst_file = f;
a672b469 1155
24f3902b 1156 if (migration_is_blocked(errp)) {
6dcf6668
DL
1157 ret = -EINVAL;
1158 goto done;
dc912121
AW
1159 }
1160
9b095037 1161 qemu_mutex_unlock_iothread();
f796baa1 1162 qemu_savevm_state_header(f);
47c8c17a 1163 qemu_savevm_state_begin(f, &params);
9b095037
PB
1164 qemu_mutex_lock_iothread();
1165
47c8c17a 1166 while (qemu_file_get_error(f) == 0) {
35ecd943 1167 if (qemu_savevm_state_iterate(f, false) > 0) {
47c8c17a
PB
1168 break;
1169 }
1170 }
a672b469 1171
47c8c17a 1172 ret = qemu_file_get_error(f);
39346385 1173 if (ret == 0) {
1c0d249d 1174 qemu_savevm_state_complete_precopy(f, false);
624b9cc2 1175 ret = qemu_file_get_error(f);
39346385 1176 }
15b3b8ea 1177 qemu_savevm_state_cleanup();
04943eba 1178 if (ret != 0) {
5d80448c 1179 error_setg_errno(errp, -ret, "Error while writing VM state");
04943eba 1180 }
6dcf6668
DL
1181
1182done:
1183 if (ret != 0) {
1184 status = MIGRATION_STATUS_FAILED;
1185 } else {
1186 status = MIGRATION_STATUS_COMPLETED;
1187 }
1188 migrate_set_state(&ms->state, MIGRATION_STATUS_SETUP, status);
a672b469
AL
1189 return ret;
1190}
1191
a7ae8355
SS
1192static int qemu_save_device_state(QEMUFile *f)
1193{
1194 SaveStateEntry *se;
1195
1196 qemu_put_be32(f, QEMU_VM_FILE_MAGIC);
1197 qemu_put_be32(f, QEMU_VM_FILE_VERSION);
1198
1199 cpu_synchronize_all_states();
1200
0163a2e0 1201 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
a7ae8355
SS
1202 if (se->is_ram) {
1203 continue;
1204 }
22ea40f4 1205 if ((!se->ops || !se->ops->save_state) && !se->vmsd) {
a7ae8355
SS
1206 continue;
1207 }
df896152
JQ
1208 if (se->vmsd && !vmstate_save_needed(se->vmsd, se->opaque)) {
1209 continue;
1210 }
a7ae8355 1211
ce39bfc9 1212 save_section_header(f, se, QEMU_VM_SECTION_FULL);
a7ae8355 1213
8118f095 1214 vmstate_save(f, se, NULL);
f68945d4
DDAG
1215
1216 save_section_footer(f, se);
a7ae8355
SS
1217 }
1218
1219 qemu_put_byte(f, QEMU_VM_EOF);
1220
1221 return qemu_file_get_error(f);
1222}
1223
a672b469
AL
1224static SaveStateEntry *find_se(const char *idstr, int instance_id)
1225{
1226 SaveStateEntry *se;
1227
0163a2e0 1228 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
a672b469 1229 if (!strcmp(se->idstr, idstr) &&
4d2ffa08
JK
1230 (instance_id == se->instance_id ||
1231 instance_id == se->alias_id))
a672b469 1232 return se;
7685ee6a
AW
1233 /* Migrating from an older version? */
1234 if (strstr(se->idstr, idstr) && se->compat) {
1235 if (!strcmp(se->compat->idstr, idstr) &&
1236 (instance_id == se->compat->instance_id ||
1237 instance_id == se->alias_id))
1238 return se;
1239 }
a672b469
AL
1240 }
1241 return NULL;
1242}
1243
7b89bf27
DDAG
1244enum LoadVMExitCodes {
1245 /* Allow a command to quit all layers of nested loadvm loops */
1246 LOADVM_QUIT = 1,
1247};
1248
1249static int qemu_loadvm_state_main(QEMUFile *f, MigrationIncomingState *mis);
093e3c42
DDAG
1250
1251/* ------ incoming postcopy messages ------ */
1252/* 'advise' arrives before any transfers just to tell us that a postcopy
1253 * *might* happen - it might be skipped if precopy transferred everything
1254 * quickly.
1255 */
1256static int loadvm_postcopy_handle_advise(MigrationIncomingState *mis)
1257{
1258 PostcopyState ps = postcopy_state_set(POSTCOPY_INCOMING_ADVISE);
1259 uint64_t remote_hps, remote_tps;
1260
1261 trace_loadvm_postcopy_handle_advise();
1262 if (ps != POSTCOPY_INCOMING_NONE) {
1263 error_report("CMD_POSTCOPY_ADVISE in wrong postcopy state (%d)", ps);
1264 return -1;
1265 }
1266
eb59db53
DDAG
1267 if (!postcopy_ram_supported_by_host()) {
1268 return -1;
1269 }
1270
093e3c42
DDAG
1271 remote_hps = qemu_get_be64(mis->from_src_file);
1272 if (remote_hps != getpagesize()) {
1273 /*
1274 * Some combinations of mismatch are probably possible but it gets
1275 * a bit more complicated. In particular we need to place whole
1276 * host pages on the dest at once, and we need to ensure that we
1277 * handle dirtying to make sure we never end up sending part of
1278 * a hostpage on it's own.
1279 */
1280 error_report("Postcopy needs matching host page sizes (s=%d d=%d)",
1281 (int)remote_hps, getpagesize());
1282 return -1;
1283 }
1284
1285 remote_tps = qemu_get_be64(mis->from_src_file);
1286 if (remote_tps != (1ul << qemu_target_page_bits())) {
1287 /*
1288 * Again, some differences could be dealt with, but for now keep it
1289 * simple.
1290 */
1291 error_report("Postcopy needs matching target page sizes (s=%d d=%d)",
1292 (int)remote_tps, 1 << qemu_target_page_bits());
1293 return -1;
1294 }
1295
1caddf8a
DDAG
1296 if (ram_postcopy_incoming_init(mis)) {
1297 return -1;
1298 }
1299
1300 postcopy_state_set(POSTCOPY_INCOMING_ADVISE);
1301
093e3c42
DDAG
1302 return 0;
1303}
1304
1305/* After postcopy we will be told to throw some pages away since they're
1306 * dirty and will have to be demand fetched. Must happen before CPU is
1307 * started.
1308 * There can be 0..many of these messages, each encoding multiple pages.
1309 */
1310static int loadvm_postcopy_ram_handle_discard(MigrationIncomingState *mis,
1311 uint16_t len)
1312{
1313 int tmp;
1314 char ramid[256];
1315 PostcopyState ps = postcopy_state_get();
1316
1317 trace_loadvm_postcopy_ram_handle_discard();
1318
1319 switch (ps) {
1320 case POSTCOPY_INCOMING_ADVISE:
1321 /* 1st discard */
f9527107 1322 tmp = postcopy_ram_prepare_discard(mis);
093e3c42
DDAG
1323 if (tmp) {
1324 return tmp;
1325 }
1326 break;
1327
1328 case POSTCOPY_INCOMING_DISCARD:
1329 /* Expected state */
1330 break;
1331
1332 default:
1333 error_report("CMD_POSTCOPY_RAM_DISCARD in wrong postcopy state (%d)",
1334 ps);
1335 return -1;
1336 }
1337 /* We're expecting a
1338 * Version (0)
1339 * a RAM ID string (length byte, name, 0 term)
1340 * then at least 1 16 byte chunk
1341 */
1342 if (len < (1 + 1 + 1 + 1 + 2 * 8)) {
1343 error_report("CMD_POSTCOPY_RAM_DISCARD invalid length (%d)", len);
1344 return -1;
1345 }
1346
1347 tmp = qemu_get_byte(mis->from_src_file);
1348 if (tmp != postcopy_ram_discard_version) {
1349 error_report("CMD_POSTCOPY_RAM_DISCARD invalid version (%d)", tmp);
1350 return -1;
1351 }
1352
1353 if (!qemu_get_counted_string(mis->from_src_file, ramid)) {
1354 error_report("CMD_POSTCOPY_RAM_DISCARD Failed to read RAMBlock ID");
1355 return -1;
1356 }
1357 tmp = qemu_get_byte(mis->from_src_file);
1358 if (tmp != 0) {
1359 error_report("CMD_POSTCOPY_RAM_DISCARD missing nil (%d)", tmp);
1360 return -1;
1361 }
1362
1363 len -= 3 + strlen(ramid);
1364 if (len % 16) {
1365 error_report("CMD_POSTCOPY_RAM_DISCARD invalid length (%d)", len);
1366 return -1;
1367 }
1368 trace_loadvm_postcopy_ram_handle_discard_header(ramid, len);
1369 while (len) {
093e3c42
DDAG
1370 uint64_t start_addr, block_length;
1371 start_addr = qemu_get_be64(mis->from_src_file);
1372 block_length = qemu_get_be64(mis->from_src_file);
1373
1374 len -= 16;
1375 int ret = ram_discard_range(mis, ramid, start_addr,
1376 block_length);
1377 if (ret) {
1378 return ret;
1379 }
093e3c42
DDAG
1380 }
1381 trace_loadvm_postcopy_ram_handle_discard_end();
1382
1383 return 0;
1384}
1385
c76201ab
DDAG
1386/*
1387 * Triggered by a postcopy_listen command; this thread takes over reading
1388 * the input stream, leaving the main thread free to carry on loading the rest
1389 * of the device state (from RAM).
1390 * (TODO:This could do with being in a postcopy file - but there again it's
1391 * just another input loop, not that postcopy specific)
1392 */
1393static void *postcopy_ram_listen_thread(void *opaque)
1394{
1395 QEMUFile *f = opaque;
1396 MigrationIncomingState *mis = migration_incoming_get_current();
1397 int load_res;
1398
6ba996bb
DDAG
1399 migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE,
1400 MIGRATION_STATUS_POSTCOPY_ACTIVE);
c76201ab
DDAG
1401 qemu_sem_post(&mis->listen_thread_sem);
1402 trace_postcopy_ram_listen_thread_start();
1403
1404 /*
1405 * Because we're a thread and not a coroutine we can't yield
1406 * in qemu_file, and thus we must be blocking now.
1407 */
1408 qemu_file_set_blocking(f, true);
1409 load_res = qemu_loadvm_state_main(f, mis);
1410 /* And non-blocking again so we don't block in any cleanup */
1411 qemu_file_set_blocking(f, false);
1412
1413 trace_postcopy_ram_listen_thread_exit();
1414 if (load_res < 0) {
1415 error_report("%s: loadvm failed: %d", __func__, load_res);
1416 qemu_file_set_error(f, load_res);
6ba996bb
DDAG
1417 migrate_set_state(&mis->state, MIGRATION_STATUS_POSTCOPY_ACTIVE,
1418 MIGRATION_STATUS_FAILED);
c76201ab
DDAG
1419 } else {
1420 /*
1421 * This looks good, but it's possible that the device loading in the
1422 * main thread hasn't finished yet, and so we might not be in 'RUN'
1423 * state yet; wait for the end of the main thread.
1424 */
1425 qemu_event_wait(&mis->main_thread_load_event);
1426 }
1427 postcopy_ram_incoming_cleanup(mis);
c76201ab
DDAG
1428
1429 if (load_res < 0) {
1430 /*
1431 * If something went wrong then we have a bad state so exit;
1432 * depending how far we got it might be possible at this point
1433 * to leave the guest running and fire MCEs for pages that never
1434 * arrived as a desperate recovery step.
1435 */
1436 exit(EXIT_FAILURE);
1437 }
1438
6ba996bb
DDAG
1439 migrate_set_state(&mis->state, MIGRATION_STATUS_POSTCOPY_ACTIVE,
1440 MIGRATION_STATUS_COMPLETED);
1441 /*
1442 * If everything has worked fine, then the main thread has waited
1443 * for us to start, and we're the last use of the mis.
1444 * (If something broke then qemu will have to exit anyway since it's
1445 * got a bad migration state).
1446 */
1447 migration_incoming_state_destroy();
1448
1449
c76201ab
DDAG
1450 return NULL;
1451}
1452
093e3c42
DDAG
1453/* After this message we must be able to immediately receive postcopy data */
1454static int loadvm_postcopy_handle_listen(MigrationIncomingState *mis)
1455{
1456 PostcopyState ps = postcopy_state_set(POSTCOPY_INCOMING_LISTENING);
1457 trace_loadvm_postcopy_handle_listen();
1458 if (ps != POSTCOPY_INCOMING_ADVISE && ps != POSTCOPY_INCOMING_DISCARD) {
1459 error_report("CMD_POSTCOPY_LISTEN in wrong postcopy state (%d)", ps);
1460 return -1;
1461 }
f9527107
DDAG
1462 if (ps == POSTCOPY_INCOMING_ADVISE) {
1463 /*
1464 * A rare case, we entered listen without having to do any discards,
1465 * so do the setup that's normally done at the time of the 1st discard.
1466 */
1467 postcopy_ram_prepare_discard(mis);
1468 }
093e3c42 1469
f0a227ad
DDAG
1470 /*
1471 * Sensitise RAM - can now generate requests for blocks that don't exist
1472 * However, at this point the CPU shouldn't be running, and the IO
1473 * shouldn't be doing anything yet so don't actually expect requests
1474 */
1475 if (postcopy_ram_enable_notify(mis)) {
1476 return -1;
1477 }
1478
c76201ab
DDAG
1479 if (mis->have_listen_thread) {
1480 error_report("CMD_POSTCOPY_RAM_LISTEN already has a listen thread");
1481 return -1;
1482 }
1483
1484 mis->have_listen_thread = true;
1485 /* Start up the listening thread and wait for it to signal ready */
1486 qemu_sem_init(&mis->listen_thread_sem, 0);
1487 qemu_thread_create(&mis->listen_thread, "postcopy/listen",
1488 postcopy_ram_listen_thread, mis->from_src_file,
a587a3fe 1489 QEMU_THREAD_DETACHED);
c76201ab
DDAG
1490 qemu_sem_wait(&mis->listen_thread_sem);
1491 qemu_sem_destroy(&mis->listen_thread_sem);
1492
093e3c42
DDAG
1493 return 0;
1494}
1495
86469922
DL
1496
1497typedef struct {
1498 QEMUBH *bh;
1499} HandleRunBhData;
1500
ea6a55bc 1501static void loadvm_postcopy_handle_run_bh(void *opaque)
093e3c42 1502{
27c6825b 1503 Error *local_err = NULL;
86469922 1504 HandleRunBhData *data = opaque;
093e3c42 1505
27c6825b
DDAG
1506 /* TODO we should move all of this lot into postcopy_ram.c or a shared code
1507 * in migration.c
1508 */
1509 cpu_synchronize_all_post_init();
1510
1511 qemu_announce_self();
1512
1513 /* Make sure all file formats flush their mutable metadata */
1514 bdrv_invalidate_cache_all(&local_err);
1515 if (local_err) {
1516 error_report_err(local_err);
27c6825b
DDAG
1517 }
1518
1519 trace_loadvm_postcopy_handle_run_cpu_sync();
1520 cpu_synchronize_all_post_init();
1521
1522 trace_loadvm_postcopy_handle_run_vmstart();
1523
093e3c42
DDAG
1524 if (autostart) {
1525 /* Hold onto your hats, starting the CPU */
1526 vm_start();
1527 } else {
1528 /* leave it paused and let management decide when to start the CPU */
1529 runstate_set(RUN_STATE_PAUSED);
1530 }
1531
86469922
DL
1532 qemu_bh_delete(data->bh);
1533 g_free(data);
ea6a55bc
DL
1534}
1535
1536/* After all discards we can start running and asking for pages */
1537static int loadvm_postcopy_handle_run(MigrationIncomingState *mis)
1538{
1539 PostcopyState ps = postcopy_state_set(POSTCOPY_INCOMING_RUNNING);
86469922 1540 HandleRunBhData *data;
ea6a55bc
DL
1541
1542 trace_loadvm_postcopy_handle_run();
1543 if (ps != POSTCOPY_INCOMING_LISTENING) {
1544 error_report("CMD_POSTCOPY_RUN in wrong postcopy state (%d)", ps);
1545 return -1;
1546 }
1547
86469922
DL
1548 data = g_new(HandleRunBhData, 1);
1549 data->bh = qemu_bh_new(loadvm_postcopy_handle_run_bh, data);
1550 qemu_bh_schedule(data->bh);
ea6a55bc 1551
27c6825b
DDAG
1552 /* We need to finish reading the stream from the package
1553 * and also stop reading anything more from the stream that loaded the
1554 * package (since it's now being read by the listener thread).
1555 * LOADVM_QUIT will quit all the layers of nested loadvm loops.
1556 */
1557 return LOADVM_QUIT;
093e3c42
DDAG
1558}
1559
c76ca188 1560/**
11cf1d98
DDAG
1561 * Immediately following this command is a blob of data containing an embedded
1562 * chunk of migration stream; read it and load it.
1563 *
1564 * @mis: Incoming state
1565 * @length: Length of packaged data to read
c76ca188 1566 *
11cf1d98
DDAG
1567 * Returns: Negative values on error
1568 *
1569 */
1570static int loadvm_handle_cmd_packaged(MigrationIncomingState *mis)
1571{
1572 int ret;
61b67d47
DB
1573 size_t length;
1574 QIOChannelBuffer *bioc;
11cf1d98
DDAG
1575
1576 length = qemu_get_be32(mis->from_src_file);
1577 trace_loadvm_handle_cmd_packaged(length);
1578
1579 if (length > MAX_VM_CMD_PACKAGED_SIZE) {
61b67d47 1580 error_report("Unreasonably large packaged state: %zu", length);
11cf1d98
DDAG
1581 return -1;
1582 }
61b67d47
DB
1583
1584 bioc = qio_channel_buffer_new(length);
1585 ret = qemu_get_buffer(mis->from_src_file,
1586 bioc->data,
1587 length);
11cf1d98 1588 if (ret != length) {
61b67d47
DB
1589 object_unref(OBJECT(bioc));
1590 error_report("CMD_PACKAGED: Buffer receive fail ret=%d length=%zu",
9af9e0fe 1591 ret, length);
11cf1d98
DDAG
1592 return (ret < 0) ? ret : -EAGAIN;
1593 }
61b67d47 1594 bioc->usage += length;
11cf1d98
DDAG
1595 trace_loadvm_handle_cmd_packaged_received(ret);
1596
61b67d47 1597 QEMUFile *packf = qemu_fopen_channel_input(QIO_CHANNEL(bioc));
11cf1d98
DDAG
1598
1599 ret = qemu_loadvm_state_main(packf, mis);
1600 trace_loadvm_handle_cmd_packaged_main(ret);
1601 qemu_fclose(packf);
61b67d47 1602 object_unref(OBJECT(bioc));
11cf1d98
DDAG
1603
1604 return ret;
1605}
1606
1607/*
1608 * Process an incoming 'QEMU_VM_COMMAND'
1609 * 0 just a normal return
1610 * LOADVM_QUIT All good, but exit the loop
1611 * <0 Error
c76ca188
DDAG
1612 */
1613static int loadvm_process_command(QEMUFile *f)
1614{
2e37701e 1615 MigrationIncomingState *mis = migration_incoming_get_current();
c76ca188
DDAG
1616 uint16_t cmd;
1617 uint16_t len;
2e37701e 1618 uint32_t tmp32;
c76ca188
DDAG
1619
1620 cmd = qemu_get_be16(f);
1621 len = qemu_get_be16(f);
1622
1623 trace_loadvm_process_command(cmd, len);
1624 if (cmd >= MIG_CMD_MAX || cmd == MIG_CMD_INVALID) {
1625 error_report("MIG_CMD 0x%x unknown (len 0x%x)", cmd, len);
1626 return -EINVAL;
1627 }
1628
1629 if (mig_cmd_args[cmd].len != -1 && mig_cmd_args[cmd].len != len) {
1630 error_report("%s received with bad length - expecting %zu, got %d",
1631 mig_cmd_args[cmd].name,
1632 (size_t)mig_cmd_args[cmd].len, len);
1633 return -ERANGE;
1634 }
1635
1636 switch (cmd) {
2e37701e
DDAG
1637 case MIG_CMD_OPEN_RETURN_PATH:
1638 if (mis->to_src_file) {
1639 error_report("CMD_OPEN_RETURN_PATH called when RP already open");
1640 /* Not really a problem, so don't give up */
1641 return 0;
1642 }
1643 mis->to_src_file = qemu_file_get_return_path(f);
1644 if (!mis->to_src_file) {
1645 error_report("CMD_OPEN_RETURN_PATH failed");
1646 return -1;
1647 }
1648 break;
1649
1650 case MIG_CMD_PING:
1651 tmp32 = qemu_get_be32(f);
1652 trace_loadvm_process_command_ping(tmp32);
1653 if (!mis->to_src_file) {
1654 error_report("CMD_PING (0x%x) received with no return path",
1655 tmp32);
1656 return -1;
1657 }
6decec93 1658 migrate_send_rp_pong(mis, tmp32);
2e37701e 1659 break;
093e3c42 1660
11cf1d98
DDAG
1661 case MIG_CMD_PACKAGED:
1662 return loadvm_handle_cmd_packaged(mis);
1663
093e3c42
DDAG
1664 case MIG_CMD_POSTCOPY_ADVISE:
1665 return loadvm_postcopy_handle_advise(mis);
1666
1667 case MIG_CMD_POSTCOPY_LISTEN:
1668 return loadvm_postcopy_handle_listen(mis);
1669
1670 case MIG_CMD_POSTCOPY_RUN:
1671 return loadvm_postcopy_handle_run(mis);
1672
1673 case MIG_CMD_POSTCOPY_RAM_DISCARD:
1674 return loadvm_postcopy_ram_handle_discard(mis, len);
c76ca188
DDAG
1675 }
1676
1677 return 0;
1678}
1679
1a8f46f8 1680struct LoadStateEntry {
72cf2d4f 1681 QLIST_ENTRY(LoadStateEntry) entry;
a672b469
AL
1682 SaveStateEntry *se;
1683 int section_id;
1684 int version_id;
1a8f46f8 1685};
a672b469 1686
59f39a47
DDAG
1687/*
1688 * Read a footer off the wire and check that it matches the expected section
1689 *
1690 * Returns: true if the footer was good
1691 * false if there is a problem (and calls error_report to say why)
1692 */
1693static bool check_section_footer(QEMUFile *f, LoadStateEntry *le)
1694{
1695 uint8_t read_mark;
1696 uint32_t read_section_id;
1697
1698 if (skip_section_footers) {
1699 /* No footer to check */
1700 return true;
1701 }
1702
1703 read_mark = qemu_get_byte(f);
1704
1705 if (read_mark != QEMU_VM_SECTION_FOOTER) {
1706 error_report("Missing section footer for %s", le->se->idstr);
1707 return false;
1708 }
1709
1710 read_section_id = qemu_get_be32(f);
1711 if (read_section_id != le->section_id) {
1712 error_report("Mismatched section id in footer for %s -"
1713 " read 0x%x expected 0x%x",
1714 le->se->idstr, read_section_id, le->section_id);
1715 return false;
1716 }
1717
1718 /* All good */
1719 return true;
1720}
1721
1a8f46f8 1722void loadvm_free_handlers(MigrationIncomingState *mis)
a672b469 1723{
f4dbb8dd 1724 LoadStateEntry *le, *new_le;
1a8f46f8
DDAG
1725
1726 QLIST_FOREACH_SAFE(le, &mis->loadvm_handlers, entry, new_le) {
1727 QLIST_REMOVE(le, entry);
1728 g_free(le);
1729 }
1730}
1731
fb3520a8
HZ
1732static int
1733qemu_loadvm_section_start_full(QEMUFile *f, MigrationIncomingState *mis)
1734{
1735 uint32_t instance_id, version_id, section_id;
1736 SaveStateEntry *se;
1737 LoadStateEntry *le;
1738 char idstr[256];
1739 int ret;
1740
1741 /* Read section start */
1742 section_id = qemu_get_be32(f);
1743 if (!qemu_get_counted_string(f, idstr)) {
1744 error_report("Unable to read ID string for section %u",
1745 section_id);
1746 return -EINVAL;
1747 }
1748 instance_id = qemu_get_be32(f);
1749 version_id = qemu_get_be32(f);
1750
1751 trace_qemu_loadvm_state_section_startfull(section_id, idstr,
1752 instance_id, version_id);
1753 /* Find savevm section */
1754 se = find_se(idstr, instance_id);
1755 if (se == NULL) {
1756 error_report("Unknown savevm section or instance '%s' %d",
1757 idstr, instance_id);
1758 return -EINVAL;
1759 }
1760
1761 /* Validate version */
1762 if (version_id > se->version_id) {
1763 error_report("savevm: unsupported version %d for '%s' v%d",
1764 version_id, idstr, se->version_id);
1765 return -EINVAL;
1766 }
1767
88c16567
WC
1768 /* Validate if it is a device's state */
1769 if (xen_enabled() && se->is_ram) {
1770 error_report("loadvm: %s RAM loading not allowed on Xen", idstr);
1771 return -EINVAL;
1772 }
1773
fb3520a8
HZ
1774 /* Add entry */
1775 le = g_malloc0(sizeof(*le));
1776
1777 le->se = se;
1778 le->section_id = section_id;
1779 le->version_id = version_id;
1780 QLIST_INSERT_HEAD(&mis->loadvm_handlers, le, entry);
1781
1782 ret = vmstate_load(f, le->se, le->version_id);
1783 if (ret < 0) {
1784 error_report("error while loading state for instance 0x%x of"
1785 " device '%s'", instance_id, idstr);
1786 return ret;
1787 }
1788 if (!check_section_footer(f, le)) {
1789 return -EINVAL;
1790 }
1791
1792 return 0;
1793}
1794
1795static int
1796qemu_loadvm_section_part_end(QEMUFile *f, MigrationIncomingState *mis)
1797{
1798 uint32_t section_id;
1799 LoadStateEntry *le;
1800 int ret;
1801
1802 section_id = qemu_get_be32(f);
1803
1804 trace_qemu_loadvm_state_section_partend(section_id);
1805 QLIST_FOREACH(le, &mis->loadvm_handlers, entry) {
1806 if (le->section_id == section_id) {
1807 break;
1808 }
1809 }
1810 if (le == NULL) {
1811 error_report("Unknown savevm section %d", section_id);
1812 return -EINVAL;
1813 }
1814
1815 ret = vmstate_load(f, le->se, le->version_id);
1816 if (ret < 0) {
1817 error_report("error while loading state section id %d(%s)",
1818 section_id, le->se->idstr);
1819 return ret;
1820 }
1821 if (!check_section_footer(f, le)) {
1822 return -EINVAL;
1823 }
1824
1825 return 0;
1826}
1827
7b89bf27 1828static int qemu_loadvm_state_main(QEMUFile *f, MigrationIncomingState *mis)
1a8f46f8 1829{
a672b469 1830 uint8_t section_type;
ccb783c3 1831 int ret = 0;
61964c23 1832
a672b469 1833 while ((section_type = qemu_get_byte(f)) != QEMU_VM_EOF) {
ccb783c3 1834 ret = 0;
a5df2a02 1835 trace_qemu_loadvm_state_section(section_type);
a672b469
AL
1836 switch (section_type) {
1837 case QEMU_VM_SECTION_START:
1838 case QEMU_VM_SECTION_FULL:
fb3520a8 1839 ret = qemu_loadvm_section_start_full(f, mis);
b5a22e4a 1840 if (ret < 0) {
ccb783c3 1841 goto out;
b5a22e4a 1842 }
a672b469
AL
1843 break;
1844 case QEMU_VM_SECTION_PART:
1845 case QEMU_VM_SECTION_END:
fb3520a8 1846 ret = qemu_loadvm_section_part_end(f, mis);
b5a22e4a 1847 if (ret < 0) {
ccb783c3 1848 goto out;
b5a22e4a 1849 }
a672b469 1850 break;
c76ca188
DDAG
1851 case QEMU_VM_COMMAND:
1852 ret = loadvm_process_command(f);
7b89bf27
DDAG
1853 trace_qemu_loadvm_state_section_command(ret);
1854 if ((ret < 0) || (ret & LOADVM_QUIT)) {
ccb783c3 1855 goto out;
c76ca188
DDAG
1856 }
1857 break;
a672b469 1858 default:
6a64b644 1859 error_report("Unknown savevm section type %d", section_type);
ccb783c3
DDAG
1860 ret = -EINVAL;
1861 goto out;
a672b469
AL
1862 }
1863 }
1864
ccb783c3
DDAG
1865out:
1866 if (ret < 0) {
1867 qemu_file_set_error(f, ret);
1868 }
1869 return ret;
7b89bf27
DDAG
1870}
1871
1872int qemu_loadvm_state(QEMUFile *f)
1873{
1874 MigrationIncomingState *mis = migration_incoming_get_current();
1875 Error *local_err = NULL;
1876 unsigned int v;
1877 int ret;
1878
1879 if (qemu_savevm_state_blocked(&local_err)) {
1880 error_report_err(local_err);
1881 return -EINVAL;
1882 }
1883
1884 v = qemu_get_be32(f);
1885 if (v != QEMU_VM_FILE_MAGIC) {
1886 error_report("Not a migration stream");
1887 return -EINVAL;
1888 }
1889
1890 v = qemu_get_be32(f);
1891 if (v == QEMU_VM_FILE_VERSION_COMPAT) {
1892 error_report("SaveVM v2 format is obsolete and don't work anymore");
1893 return -ENOTSUP;
1894 }
1895 if (v != QEMU_VM_FILE_VERSION) {
1896 error_report("Unsupported migration stream version");
1897 return -ENOTSUP;
1898 }
1899
902c053d 1900 if (!savevm_state.skip_configuration || enforce_config_section()) {
7b89bf27
DDAG
1901 if (qemu_get_byte(f) != QEMU_VM_CONFIGURATION) {
1902 error_report("Configuration section missing");
1903 return -EINVAL;
1904 }
1905 ret = vmstate_load_state(f, &vmstate_configuration, &savevm_state, 0);
1906
1907 if (ret) {
1908 return ret;
1909 }
1910 }
1911
1912 ret = qemu_loadvm_state_main(f, mis);
1913 qemu_event_set(&mis->main_thread_load_event);
1914
1915 trace_qemu_loadvm_state_post_main(ret);
1916
c76201ab
DDAG
1917 if (mis->have_listen_thread) {
1918 /* Listen thread still going, can't clean up yet */
1919 return ret;
1920 }
1921
7b89bf27
DDAG
1922 if (ret == 0) {
1923 ret = qemu_file_get_error(f);
1924 }
1925cebc
AG
1925
1926 /*
1927 * Try to read in the VMDESC section as well, so that dumping tools that
1928 * intercept our migration stream have the chance to see it.
1929 */
1aca9a5f
DDAG
1930
1931 /* We've got to be careful; if we don't read the data and just shut the fd
1932 * then the sender can error if we close while it's still sending.
1933 * We also mustn't read data that isn't there; some transports (RDMA)
1934 * will stall waiting for that data when the source has already closed.
1935 */
7b89bf27 1936 if (ret == 0 && should_send_vmdesc()) {
1aca9a5f
DDAG
1937 uint8_t *buf;
1938 uint32_t size;
7b89bf27 1939 uint8_t section_type = qemu_get_byte(f);
1aca9a5f
DDAG
1940
1941 if (section_type != QEMU_VM_VMDESCRIPTION) {
1942 error_report("Expected vmdescription section, but got %d",
1943 section_type);
1944 /*
1945 * It doesn't seem worth failing at this point since
1946 * we apparently have an otherwise valid VM state
1947 */
1948 } else {
1949 buf = g_malloc(0x1000);
1950 size = qemu_get_be32(f);
1951
1952 while (size > 0) {
1953 uint32_t read_chunk = MIN(size, 0x1000);
1954 qemu_get_buffer(f, buf, read_chunk);
1955 size -= read_chunk;
1956 }
1957 g_free(buf);
1925cebc 1958 }
1925cebc
AG
1959 }
1960
ea375f9a
JK
1961 cpu_synchronize_all_post_init();
1962
a672b469
AL
1963 return ret;
1964}
1965
3e5a50d6 1966void hmp_savevm(Monitor *mon, const QDict *qdict)
a672b469
AL
1967{
1968 BlockDriverState *bs, *bs1;
1969 QEMUSnapshotInfo sn1, *sn = &sn1, old_sn1, *old_sn = &old_sn1;
cb499fb2 1970 int ret;
a672b469
AL
1971 QEMUFile *f;
1972 int saved_vm_running;
c2c9a466 1973 uint64_t vm_state_size;
68b891ec 1974 qemu_timeval tv;
7d631a11 1975 struct tm tm;
d54908a5 1976 const char *name = qdict_get_try_str(qdict, "name");
5d80448c 1977 Error *local_err = NULL;
79b3c12a 1978 AioContext *aio_context;
a672b469 1979
e9ff957a
DL
1980 if (!bdrv_all_can_snapshot(&bs)) {
1981 monitor_printf(mon, "Device '%s' is writable but does not "
1982 "support snapshots.\n", bdrv_get_device_name(bs));
1983 return;
feeee5ac
MDCF
1984 }
1985
0b461605
DL
1986 /* Delete old snapshots of the same name */
1987 if (name && bdrv_all_delete_snapshot(name, &bs1, &local_err) < 0) {
d410fe14
MA
1988 error_reportf_err(local_err,
1989 "Error while deleting snapshot on device '%s': ",
1990 bdrv_get_device_name(bs1));
0b461605
DL
1991 return;
1992 }
1993
7cb14481
DL
1994 bs = bdrv_all_find_vmstate_bs();
1995 if (bs == NULL) {
376253ec 1996 monitor_printf(mon, "No block device can accept snapshots\n");
a672b469
AL
1997 return;
1998 }
79b3c12a 1999 aio_context = bdrv_get_aio_context(bs);
a672b469 2000
1354869c 2001 saved_vm_running = runstate_is_running();
560d027b
JQ
2002
2003 ret = global_state_store();
2004 if (ret) {
2005 monitor_printf(mon, "Error saving global state\n");
2006 return;
2007 }
0461d5a6 2008 vm_stop(RUN_STATE_SAVE_VM);
a672b469 2009
79b3c12a
DL
2010 aio_context_acquire(aio_context);
2011
cb499fb2 2012 memset(sn, 0, sizeof(*sn));
a672b469
AL
2013
2014 /* fill auxiliary fields */
68b891ec 2015 qemu_gettimeofday(&tv);
a672b469
AL
2016 sn->date_sec = tv.tv_sec;
2017 sn->date_nsec = tv.tv_usec * 1000;
bc72ad67 2018 sn->vm_clock_nsec = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL);
a672b469 2019
7d631a11
MDCF
2020 if (name) {
2021 ret = bdrv_snapshot_find(bs, old_sn, name);
2022 if (ret >= 0) {
2023 pstrcpy(sn->name, sizeof(sn->name), old_sn->name);
2024 pstrcpy(sn->id_str, sizeof(sn->id_str), old_sn->id_str);
2025 } else {
2026 pstrcpy(sn->name, sizeof(sn->name), name);
2027 }
2028 } else {
d7d9b528
BS
2029 /* cast below needed for OpenBSD where tv_sec is still 'long' */
2030 localtime_r((const time_t *)&tv.tv_sec, &tm);
7d631a11 2031 strftime(sn->name, sizeof(sn->name), "vm-%Y%m%d%H%M%S", &tm);
7d631a11
MDCF
2032 }
2033
a672b469 2034 /* save the VM state */
45566e9c 2035 f = qemu_fopen_bdrv(bs, 1);
a672b469 2036 if (!f) {
376253ec 2037 monitor_printf(mon, "Could not open VM state file\n");
a672b469
AL
2038 goto the_end;
2039 }
5d80448c 2040 ret = qemu_savevm_state(f, &local_err);
2d22b18f 2041 vm_state_size = qemu_ftell(f);
a672b469
AL
2042 qemu_fclose(f);
2043 if (ret < 0) {
193227f9 2044 error_report_err(local_err);
a672b469
AL
2045 goto the_end;
2046 }
2047
a9085f9b
DL
2048 ret = bdrv_all_create_snapshot(sn, bs, vm_state_size, &bs);
2049 if (ret < 0) {
2050 monitor_printf(mon, "Error while creating snapshot on '%s'\n",
2051 bdrv_get_device_name(bs));
a672b469
AL
2052 }
2053
2054 the_end:
79b3c12a 2055 aio_context_release(aio_context);
38ff78d3 2056 if (saved_vm_running) {
a672b469 2057 vm_start();
38ff78d3 2058 }
a672b469
AL
2059}
2060
a7ae8355
SS
2061void qmp_xen_save_devices_state(const char *filename, Error **errp)
2062{
2063 QEMUFile *f;
8925839f 2064 QIOChannelFile *ioc;
a7ae8355
SS
2065 int saved_vm_running;
2066 int ret;
2067
2068 saved_vm_running = runstate_is_running();
2069 vm_stop(RUN_STATE_SAVE_VM);
c69adea4 2070 global_state_store_running();
a7ae8355 2071
8925839f
DB
2072 ioc = qio_channel_file_new_path(filename, O_WRONLY | O_CREAT, 0660, errp);
2073 if (!ioc) {
a7ae8355
SS
2074 goto the_end;
2075 }
8925839f 2076 f = qemu_fopen_channel_output(QIO_CHANNEL(ioc));
a7ae8355
SS
2077 ret = qemu_save_device_state(f);
2078 qemu_fclose(f);
2079 if (ret < 0) {
c6bd8c70 2080 error_setg(errp, QERR_IO_ERROR);
a7ae8355
SS
2081 }
2082
2083 the_end:
38ff78d3 2084 if (saved_vm_running) {
a7ae8355 2085 vm_start();
38ff78d3 2086 }
a7ae8355
SS
2087}
2088
88c16567
WC
2089void qmp_xen_load_devices_state(const char *filename, Error **errp)
2090{
2091 QEMUFile *f;
2092 QIOChannelFile *ioc;
2093 int ret;
2094
2095 /* Guest must be paused before loading the device state; the RAM state
2096 * will already have been loaded by xc
2097 */
2098 if (runstate_is_running()) {
2099 error_setg(errp, "Cannot update device state while vm is running");
2100 return;
2101 }
2102 vm_stop(RUN_STATE_RESTORE_VM);
2103
2104 ioc = qio_channel_file_new_path(filename, O_RDONLY | O_BINARY, 0, errp);
2105 if (!ioc) {
2106 return;
2107 }
2108 f = qemu_fopen_channel_input(QIO_CHANNEL(ioc));
2109
2110 migration_incoming_state_new(f);
2111 ret = qemu_loadvm_state(f);
2112 qemu_fclose(f);
2113 if (ret < 0) {
2114 error_setg(errp, QERR_IO_ERROR);
2115 }
2116 migration_incoming_state_destroy();
2117}
2118
03cd4655 2119int load_vmstate(const char *name)
a672b469 2120{
f0aa7a8b 2121 BlockDriverState *bs, *bs_vm_state;
2d22b18f 2122 QEMUSnapshotInfo sn;
a672b469 2123 QEMUFile *f;
751c6a17 2124 int ret;
79b3c12a 2125 AioContext *aio_context;
a672b469 2126
849f96e2
DL
2127 if (!bdrv_all_can_snapshot(&bs)) {
2128 error_report("Device '%s' is writable but does not support snapshots.",
2129 bdrv_get_device_name(bs));
2130 return -ENOTSUP;
2131 }
723ccda1
DL
2132 ret = bdrv_all_find_snapshot(name, &bs);
2133 if (ret < 0) {
2134 error_report("Device '%s' does not have the requested snapshot '%s'",
2135 bdrv_get_device_name(bs), name);
2136 return ret;
2137 }
849f96e2 2138
7cb14481 2139 bs_vm_state = bdrv_all_find_vmstate_bs();
f0aa7a8b
MDCF
2140 if (!bs_vm_state) {
2141 error_report("No block device supports snapshots");
2142 return -ENOTSUP;
2143 }
79b3c12a 2144 aio_context = bdrv_get_aio_context(bs_vm_state);
f0aa7a8b
MDCF
2145
2146 /* Don't even try to load empty VM states */
79b3c12a 2147 aio_context_acquire(aio_context);
f0aa7a8b 2148 ret = bdrv_snapshot_find(bs_vm_state, &sn, name);
79b3c12a 2149 aio_context_release(aio_context);
f0aa7a8b
MDCF
2150 if (ret < 0) {
2151 return ret;
2152 } else if (sn.vm_state_size == 0) {
e11480db
KW
2153 error_report("This is a disk-only snapshot. Revert to it offline "
2154 "using qemu-img.");
f0aa7a8b
MDCF
2155 return -EINVAL;
2156 }
2157
a672b469 2158 /* Flush all IO requests so they don't interfere with the new state. */
922453bc 2159 bdrv_drain_all();
a672b469 2160
4c1cdbaa
DL
2161 ret = bdrv_all_goto_snapshot(name, &bs);
2162 if (ret < 0) {
2163 error_report("Error %d while activating snapshot '%s' on '%s'",
2164 ret, name, bdrv_get_device_name(bs));
2165 return ret;
a672b469
AL
2166 }
2167
a672b469 2168 /* restore the VM state */
f0aa7a8b 2169 f = qemu_fopen_bdrv(bs_vm_state, 0);
a672b469 2170 if (!f) {
1ecda02b 2171 error_report("Could not open VM state file");
05f2401e 2172 return -EINVAL;
a672b469 2173 }
f0aa7a8b 2174
5a8a49d7 2175 qemu_system_reset(VMRESET_SILENT);
bca7856a 2176 migration_incoming_state_new(f);
f0aa7a8b 2177
79b3c12a
DL
2178 aio_context_acquire(aio_context);
2179 ret = qemu_loadvm_state(f);
a672b469 2180 qemu_fclose(f);
79b3c12a
DL
2181 aio_context_release(aio_context);
2182
bca7856a 2183 migration_incoming_state_destroy();
a672b469 2184 if (ret < 0) {
1ecda02b 2185 error_report("Error %d while loading VM state", ret);
05f2401e 2186 return ret;
a672b469 2187 }
f0aa7a8b 2188
05f2401e 2189 return 0;
7b630349
JQ
2190}
2191
3e5a50d6 2192void hmp_delvm(Monitor *mon, const QDict *qdict)
a672b469 2193{
af957387 2194 BlockDriverState *bs;
ba2b2288 2195 Error *err;
d54908a5 2196 const char *name = qdict_get_str(qdict, "name");
a672b469 2197
9b00ea37 2198 if (bdrv_all_delete_snapshot(name, &bs, &err) < 0) {
d410fe14
MA
2199 error_reportf_err(err,
2200 "Error while deleting snapshot on device '%s': ",
2201 bdrv_get_device_name(bs));
a672b469
AL
2202 }
2203}
2204
1ce6be24 2205void hmp_info_snapshots(Monitor *mon, const QDict *qdict)
a672b469
AL
2206{
2207 BlockDriverState *bs, *bs1;
0c204cc8 2208 BdrvNextIterator it1;
723ccda1 2209 QEMUSnapshotInfo *sn_tab, *sn;
0c204cc8 2210 bool no_snapshot = true;
723ccda1 2211 int nb_sns, i;
f9209915 2212 int total;
0c204cc8 2213 int *global_snapshots;
79b3c12a 2214 AioContext *aio_context;
a672b469 2215
0c204cc8
LM
2216 typedef struct SnapshotEntry {
2217 QEMUSnapshotInfo sn;
2218 QTAILQ_ENTRY(SnapshotEntry) next;
2219 } SnapshotEntry;
2220
2221 typedef struct ImageEntry {
2222 const char *imagename;
2223 QTAILQ_ENTRY(ImageEntry) next;
2224 QTAILQ_HEAD(, SnapshotEntry) snapshots;
2225 } ImageEntry;
2226
2227 QTAILQ_HEAD(, ImageEntry) image_list =
2228 QTAILQ_HEAD_INITIALIZER(image_list);
2229
2230 ImageEntry *image_entry, *next_ie;
2231 SnapshotEntry *snapshot_entry;
2232
7cb14481 2233 bs = bdrv_all_find_vmstate_bs();
a672b469 2234 if (!bs) {
376253ec 2235 monitor_printf(mon, "No available block device supports snapshots\n");
a672b469
AL
2236 return;
2237 }
79b3c12a 2238 aio_context = bdrv_get_aio_context(bs);
a672b469 2239
79b3c12a 2240 aio_context_acquire(aio_context);
a672b469 2241 nb_sns = bdrv_snapshot_list(bs, &sn_tab);
79b3c12a
DL
2242 aio_context_release(aio_context);
2243
a672b469 2244 if (nb_sns < 0) {
376253ec 2245 monitor_printf(mon, "bdrv_snapshot_list: error %d\n", nb_sns);
a672b469
AL
2246 return;
2247 }
f9209915 2248
0c204cc8
LM
2249 for (bs1 = bdrv_first(&it1); bs1; bs1 = bdrv_next(&it1)) {
2250 int bs1_nb_sns = 0;
2251 ImageEntry *ie;
2252 SnapshotEntry *se;
2253 AioContext *ctx = bdrv_get_aio_context(bs1);
2254
2255 aio_context_acquire(ctx);
2256 if (bdrv_can_snapshot(bs1)) {
2257 sn = NULL;
2258 bs1_nb_sns = bdrv_snapshot_list(bs1, &sn);
2259 if (bs1_nb_sns > 0) {
2260 no_snapshot = false;
2261 ie = g_new0(ImageEntry, 1);
2262 ie->imagename = bdrv_get_device_name(bs1);
2263 QTAILQ_INIT(&ie->snapshots);
2264 QTAILQ_INSERT_TAIL(&image_list, ie, next);
2265 for (i = 0; i < bs1_nb_sns; i++) {
2266 se = g_new0(SnapshotEntry, 1);
2267 se->sn = sn[i];
2268 QTAILQ_INSERT_TAIL(&ie->snapshots, se, next);
2269 }
2270 }
2271 g_free(sn);
2272 }
2273 aio_context_release(ctx);
2274 }
2275
2276 if (no_snapshot) {
f9209915
MDCF
2277 monitor_printf(mon, "There is no snapshot available.\n");
2278 return;
2279 }
2280
0c204cc8 2281 global_snapshots = g_new0(int, nb_sns);
f9209915
MDCF
2282 total = 0;
2283 for (i = 0; i < nb_sns; i++) {
0c204cc8 2284 SnapshotEntry *next_sn;
3a1ee711 2285 if (bdrv_all_find_snapshot(sn_tab[i].name, &bs1) == 0) {
0c204cc8 2286 global_snapshots[total] = i;
f9209915 2287 total++;
0c204cc8
LM
2288 QTAILQ_FOREACH(image_entry, &image_list, next) {
2289 QTAILQ_FOREACH_SAFE(snapshot_entry, &image_entry->snapshots,
2290 next, next_sn) {
2291 if (!strcmp(sn_tab[i].name, snapshot_entry->sn.name)) {
2292 QTAILQ_REMOVE(&image_entry->snapshots, snapshot_entry,
2293 next);
2294 g_free(snapshot_entry);
2295 }
2296 }
2297 }
f9209915 2298 }
a672b469 2299 }
f9209915 2300
0c204cc8
LM
2301 monitor_printf(mon, "List of snapshots present on all disks:\n");
2302
f9209915 2303 if (total > 0) {
5b917044
WX
2304 bdrv_snapshot_dump((fprintf_function)monitor_printf, mon, NULL);
2305 monitor_printf(mon, "\n");
f9209915 2306 for (i = 0; i < total; i++) {
0c204cc8 2307 sn = &sn_tab[global_snapshots[i]];
3a1ee711
LM
2308 /* The ID is not guaranteed to be the same on all images, so
2309 * overwrite it.
2310 */
2311 pstrcpy(sn->id_str, sizeof(sn->id_str), "--");
5b917044
WX
2312 bdrv_snapshot_dump((fprintf_function)monitor_printf, mon, sn);
2313 monitor_printf(mon, "\n");
f9209915
MDCF
2314 }
2315 } else {
0c204cc8 2316 monitor_printf(mon, "None\n");
f9209915
MDCF
2317 }
2318
0c204cc8
LM
2319 QTAILQ_FOREACH(image_entry, &image_list, next) {
2320 if (QTAILQ_EMPTY(&image_entry->snapshots)) {
2321 continue;
2322 }
2323 monitor_printf(mon,
2324 "\nList of partial (non-loadable) snapshots on '%s':\n",
2325 image_entry->imagename);
2326 bdrv_snapshot_dump((fprintf_function)monitor_printf, mon, NULL);
2327 monitor_printf(mon, "\n");
2328 QTAILQ_FOREACH(snapshot_entry, &image_entry->snapshots, next) {
2329 bdrv_snapshot_dump((fprintf_function)monitor_printf, mon,
2330 &snapshot_entry->sn);
2331 monitor_printf(mon, "\n");
2332 }
2333 }
2334
2335 QTAILQ_FOREACH_SAFE(image_entry, &image_list, next, next_ie) {
2336 SnapshotEntry *next_sn;
2337 QTAILQ_FOREACH_SAFE(snapshot_entry, &image_entry->snapshots, next,
2338 next_sn) {
2339 g_free(snapshot_entry);
2340 }
2341 g_free(image_entry);
2342 }
7267c094 2343 g_free(sn_tab);
0c204cc8 2344 g_free(global_snapshots);
f9209915 2345
a672b469 2346}
c5705a77
AK
2347
2348void vmstate_register_ram(MemoryRegion *mr, DeviceState *dev)
2349{
fa53a0e5 2350 qemu_ram_set_idstr(mr->ram_block,
c5705a77
AK
2351 memory_region_name(mr), dev);
2352}
2353
2354void vmstate_unregister_ram(MemoryRegion *mr, DeviceState *dev)
2355{
fa53a0e5 2356 qemu_ram_unset_idstr(mr->ram_block);
c5705a77
AK
2357}
2358
2359void vmstate_register_ram_global(MemoryRegion *mr)
2360{
2361 vmstate_register_ram(mr, NULL);
2362}