]>
Commit | Line | Data |
---|---|---|
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 | } SaveState; | |
269 | ||
270 | static SaveState savevm_state = { | |
271 | .handlers = QTAILQ_HEAD_INITIALIZER(savevm_state.handlers), | |
272 | .global_section_id = 0, | |
273 | .skip_configuration = false, | |
274 | }; | |
275 | ||
276 | void savevm_skip_configuration(void) | |
277 | { | |
278 | savevm_state.skip_configuration = true; | |
279 | } | |
280 | ||
281 | ||
282 | static 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 | ||
291 | static 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) { | |
297 | error_report("Machine type received is '%.*s' and local is '%s'", | |
298 | (int) state->len, state->name, current_name); | |
299 | return -EINVAL; | |
300 | } | |
301 | return 0; | |
302 | } | |
303 | ||
304 | static 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 | }, | |
314 | }; | |
315 | ||
316 | static void dump_vmstate_vmsd(FILE *out_file, | |
317 | const VMStateDescription *vmsd, int indent, | |
318 | bool is_subsection); | |
319 | ||
320 | static 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 | ||
338 | static void dump_vmstate_vmss(FILE *out_file, | |
339 | const VMStateDescription **subsection, | |
340 | int indent) | |
341 | { | |
342 | if (*subsection != NULL) { | |
343 | dump_vmstate_vmsd(out_file, *subsection, indent, true); | |
344 | } | |
345 | } | |
346 | ||
347 | static 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) { | |
384 | const VMStateDescription **subsection = vmsd->subsections; | |
385 | bool first; | |
386 | ||
387 | fprintf(out_file, ",\n%*s\"Subsections\": [\n", indent, ""); | |
388 | first = true; | |
389 | while (*subsection != NULL) { | |
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 | ||
402 | static 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 | ||
413 | void 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 | ||
454 | static int calculate_new_instance_id(const char *idstr) | |
455 | { | |
456 | SaveStateEntry *se; | |
457 | int instance_id = 0; | |
458 | ||
459 | QTAILQ_FOREACH(se, &savevm_state.handlers, entry) { | |
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 | ||
468 | static int calculate_compat_instance_id(const char *idstr) | |
469 | { | |
470 | SaveStateEntry *se; | |
471 | int instance_id = 0; | |
472 | ||
473 | QTAILQ_FOREACH(se, &savevm_state.handlers, entry) { | |
474 | if (!se->compat) { | |
475 | continue; | |
476 | } | |
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 | ||
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. */ | |
490 | int register_savevm_live(DeviceState *dev, | |
491 | const char *idstr, | |
492 | int instance_id, | |
493 | int version_id, | |
494 | SaveVMHandlers *ops, | |
495 | void *opaque) | |
496 | { | |
497 | SaveStateEntry *se; | |
498 | ||
499 | se = g_new0(SaveStateEntry, 1); | |
500 | se->version_id = version_id; | |
501 | se->section_id = savevm_state.global_section_id++; | |
502 | se->ops = ops; | |
503 | se->opaque = opaque; | |
504 | se->vmsd = NULL; | |
505 | /* if this is a live_savem then set is_ram */ | |
506 | if (ops->save_live_setup != NULL) { | |
507 | se->is_ram = 1; | |
508 | } | |
509 | ||
510 | if (dev) { | |
511 | char *id = qdev_get_dev_path(dev); | |
512 | if (id) { | |
513 | pstrcpy(se->idstr, sizeof(se->idstr), id); | |
514 | pstrcat(se->idstr, sizeof(se->idstr), "/"); | |
515 | g_free(id); | |
516 | ||
517 | se->compat = g_new0(CompatEntry, 1); | |
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 | ||
526 | if (instance_id == -1) { | |
527 | se->instance_id = calculate_new_instance_id(se->idstr); | |
528 | } else { | |
529 | se->instance_id = instance_id; | |
530 | } | |
531 | assert(!se->compat || se->instance_id == 0); | |
532 | /* add at the end of list */ | |
533 | QTAILQ_INSERT_TAIL(&savevm_state.handlers, se, entry); | |
534 | return 0; | |
535 | } | |
536 | ||
537 | int register_savevm(DeviceState *dev, | |
538 | const char *idstr, | |
539 | int instance_id, | |
540 | int version_id, | |
541 | SaveStateHandler *save_state, | |
542 | LoadStateHandler *load_state, | |
543 | void *opaque) | |
544 | { | |
545 | SaveVMHandlers *ops = g_new0(SaveVMHandlers, 1); | |
546 | ops->save_state = save_state; | |
547 | ops->load_state = load_state; | |
548 | return register_savevm_live(dev, idstr, instance_id, version_id, | |
549 | ops, opaque); | |
550 | } | |
551 | ||
552 | void unregister_savevm(DeviceState *dev, const char *idstr, void *opaque) | |
553 | { | |
554 | SaveStateEntry *se, *new_se; | |
555 | char id[256] = ""; | |
556 | ||
557 | if (dev) { | |
558 | char *path = qdev_get_dev_path(dev); | |
559 | if (path) { | |
560 | pstrcpy(id, sizeof(id), path); | |
561 | pstrcat(id, sizeof(id), "/"); | |
562 | g_free(path); | |
563 | } | |
564 | } | |
565 | pstrcat(id, sizeof(id), idstr); | |
566 | ||
567 | QTAILQ_FOREACH_SAFE(se, &savevm_state.handlers, entry, new_se) { | |
568 | if (strcmp(se->idstr, id) == 0 && se->opaque == opaque) { | |
569 | QTAILQ_REMOVE(&savevm_state.handlers, se, entry); | |
570 | g_free(se->compat); | |
571 | g_free(se->ops); | |
572 | g_free(se); | |
573 | } | |
574 | } | |
575 | } | |
576 | ||
577 | int vmstate_register_with_alias_id(DeviceState *dev, int instance_id, | |
578 | const VMStateDescription *vmsd, | |
579 | void *opaque, int alias_id, | |
580 | int required_for_version) | |
581 | { | |
582 | SaveStateEntry *se; | |
583 | ||
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 | ||
587 | se = g_new0(SaveStateEntry, 1); | |
588 | se->version_id = vmsd->version_id; | |
589 | se->section_id = savevm_state.global_section_id++; | |
590 | se->opaque = opaque; | |
591 | se->vmsd = vmsd; | |
592 | se->alias_id = alias_id; | |
593 | ||
594 | if (dev) { | |
595 | char *id = qdev_get_dev_path(dev); | |
596 | if (id) { | |
597 | pstrcpy(se->idstr, sizeof(se->idstr), id); | |
598 | pstrcat(se->idstr, sizeof(se->idstr), "/"); | |
599 | g_free(id); | |
600 | ||
601 | se->compat = g_new0(CompatEntry, 1); | |
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 | ||
610 | if (instance_id == -1) { | |
611 | se->instance_id = calculate_new_instance_id(se->idstr); | |
612 | } else { | |
613 | se->instance_id = instance_id; | |
614 | } | |
615 | assert(!se->compat || se->instance_id == 0); | |
616 | /* add at the end of list */ | |
617 | QTAILQ_INSERT_TAIL(&savevm_state.handlers, se, entry); | |
618 | return 0; | |
619 | } | |
620 | ||
621 | void vmstate_unregister(DeviceState *dev, const VMStateDescription *vmsd, | |
622 | void *opaque) | |
623 | { | |
624 | SaveStateEntry *se, *new_se; | |
625 | ||
626 | QTAILQ_FOREACH_SAFE(se, &savevm_state.handlers, entry, new_se) { | |
627 | if (se->vmsd == vmsd && se->opaque == opaque) { | |
628 | QTAILQ_REMOVE(&savevm_state.handlers, se, entry); | |
629 | g_free(se->compat); | |
630 | g_free(se); | |
631 | } | |
632 | } | |
633 | } | |
634 | ||
635 | static int vmstate_load(QEMUFile *f, SaveStateEntry *se, int version_id) | |
636 | { | |
637 | trace_vmstate_load(se->idstr, se->vmsd ? se->vmsd->name : "(old)"); | |
638 | if (!se->vmsd) { /* Old style */ | |
639 | return se->ops->load_state(f, se->opaque, version_id); | |
640 | } | |
641 | return vmstate_load_state(f, se->vmsd, se->opaque, version_id); | |
642 | } | |
643 | ||
644 | static 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 | ||
664 | static void vmstate_save(QEMUFile *f, SaveStateEntry *se, QJSON *vmdesc) | |
665 | { | |
666 | trace_vmstate_save(se->idstr, se->vmsd ? se->vmsd->name : "(old)"); | |
667 | if (!se->vmsd) { | |
668 | vmstate_save_old_style(f, se, vmdesc); | |
669 | return; | |
670 | } | |
671 | vmstate_save_state(f, se->vmsd, se->opaque, vmdesc); | |
672 | } | |
673 | ||
674 | void savevm_skip_section_footers(void) | |
675 | { | |
676 | skip_section_footers = true; | |
677 | } | |
678 | ||
679 | /* | |
680 | * Write the header for device section (QEMU_VM_SECTION START/END/PART/FULL) | |
681 | */ | |
682 | static 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 | ||
700 | /* | |
701 | * Write a footer onto device sections that catches cases misformatted device | |
702 | * sections. | |
703 | */ | |
704 | static 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 | ||
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 | */ | |
721 | void 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 | ||
734 | void 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 | ||
743 | void 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 | ||
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 | */ | |
758 | int qemu_savevm_send_packaged(QEMUFile *f, const uint8_t *buf, size_t len) | |
759 | { | |
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 | ||
773 | qemu_put_buffer(f, buf, len); | |
774 | ||
775 | return 0; | |
776 | } | |
777 | ||
778 | /* Send prior to any postcopy transfer */ | |
779 | void 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 | */ | |
806 | void 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++) { | |
826 | cpu_to_be64w((uint64_t *)(buf + tmplen), start_list[t]); | |
827 | tmplen += 8; | |
828 | cpu_to_be64w((uint64_t *)(buf + tmplen), length_list[t]); | |
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. */ | |
836 | void 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 */ | |
843 | void 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 | ||
849 | bool qemu_savevm_state_blocked(Error **errp) | |
850 | { | |
851 | SaveStateEntry *se; | |
852 | ||
853 | QTAILQ_FOREACH(se, &savevm_state.handlers, entry) { | |
854 | if (se->vmsd && se->vmsd->unmigratable) { | |
855 | error_setg(errp, "State blocked by non-migratable device '%s'", | |
856 | se->idstr); | |
857 | return true; | |
858 | } | |
859 | } | |
860 | return false; | |
861 | } | |
862 | ||
863 | static bool enforce_config_section(void) | |
864 | { | |
865 | MachineState *machine = MACHINE(qdev_get_machine()); | |
866 | return machine->enforce_config_section; | |
867 | } | |
868 | ||
869 | void 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); | |
874 | ||
875 | if (!savevm_state.skip_configuration || enforce_config_section()) { | |
876 | qemu_put_byte(f, QEMU_VM_CONFIGURATION); | |
877 | vmstate_save_state(f, &vmstate_configuration, &savevm_state, 0); | |
878 | } | |
879 | ||
880 | } | |
881 | ||
882 | void qemu_savevm_state_begin(QEMUFile *f, | |
883 | const MigrationParams *params) | |
884 | { | |
885 | SaveStateEntry *se; | |
886 | int ret; | |
887 | ||
888 | trace_savevm_state_begin(); | |
889 | QTAILQ_FOREACH(se, &savevm_state.handlers, entry) { | |
890 | if (!se->ops || !se->ops->set_params) { | |
891 | continue; | |
892 | } | |
893 | se->ops->set_params(params, se->opaque); | |
894 | } | |
895 | ||
896 | QTAILQ_FOREACH(se, &savevm_state.handlers, entry) { | |
897 | if (!se->ops || !se->ops->save_live_setup) { | |
898 | continue; | |
899 | } | |
900 | if (se->ops && se->ops->is_active) { | |
901 | if (!se->ops->is_active(se->opaque)) { | |
902 | continue; | |
903 | } | |
904 | } | |
905 | save_section_header(f, se, QEMU_VM_SECTION_START); | |
906 | ||
907 | ret = se->ops->save_live_setup(f, se->opaque); | |
908 | save_section_footer(f, se); | |
909 | if (ret < 0) { | |
910 | qemu_file_set_error(f, ret); | |
911 | break; | |
912 | } | |
913 | } | |
914 | } | |
915 | ||
916 | /* | |
917 | * this function has three return values: | |
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 | */ | |
922 | int qemu_savevm_state_iterate(QEMUFile *f, bool postcopy) | |
923 | { | |
924 | SaveStateEntry *se; | |
925 | int ret = 1; | |
926 | ||
927 | trace_savevm_state_iterate(); | |
928 | QTAILQ_FOREACH(se, &savevm_state.handlers, entry) { | |
929 | if (!se->ops || !se->ops->save_live_iterate) { | |
930 | continue; | |
931 | } | |
932 | if (se->ops && se->ops->is_active) { | |
933 | if (!se->ops->is_active(se->opaque)) { | |
934 | continue; | |
935 | } | |
936 | } | |
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 | } | |
946 | if (qemu_file_rate_limit(f)) { | |
947 | return 0; | |
948 | } | |
949 | trace_savevm_section_start(se->idstr, se->section_id); | |
950 | ||
951 | save_section_header(f, se, QEMU_VM_SECTION_PART); | |
952 | ||
953 | ret = se->ops->save_live_iterate(f, se->opaque); | |
954 | trace_savevm_section_end(se->idstr, se->section_id, ret); | |
955 | save_section_footer(f, se); | |
956 | ||
957 | if (ret < 0) { | |
958 | qemu_file_set_error(f, ret); | |
959 | } | |
960 | if (ret <= 0) { | |
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 | } | |
967 | } | |
968 | return ret; | |
969 | } | |
970 | ||
971 | static bool should_send_vmdesc(void) | |
972 | { | |
973 | MachineState *machine = MACHINE(qdev_get_machine()); | |
974 | bool in_postcopy = migration_in_postcopy(migrate_get_current()); | |
975 | return !machine->suppress_vmdesc && !in_postcopy; | |
976 | } | |
977 | ||
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 | */ | |
985 | void 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 | ||
1017 | void qemu_savevm_state_complete_precopy(QEMUFile *f, bool iterable_only) | |
1018 | { | |
1019 | QJSON *vmdesc; | |
1020 | int vmdesc_len; | |
1021 | SaveStateEntry *se; | |
1022 | int ret; | |
1023 | bool in_postcopy = migration_in_postcopy(migrate_get_current()); | |
1024 | ||
1025 | trace_savevm_state_complete_precopy(); | |
1026 | ||
1027 | cpu_synchronize_all_states(); | |
1028 | ||
1029 | QTAILQ_FOREACH(se, &savevm_state.handlers, entry) { | |
1030 | if (!se->ops || | |
1031 | (in_postcopy && se->ops->save_live_complete_postcopy) || | |
1032 | (in_postcopy && !iterable_only) || | |
1033 | !se->ops->save_live_complete_precopy) { | |
1034 | continue; | |
1035 | } | |
1036 | ||
1037 | if (se->ops && se->ops->is_active) { | |
1038 | if (!se->ops->is_active(se->opaque)) { | |
1039 | continue; | |
1040 | } | |
1041 | } | |
1042 | trace_savevm_section_start(se->idstr, se->section_id); | |
1043 | ||
1044 | save_section_header(f, se, QEMU_VM_SECTION_END); | |
1045 | ||
1046 | ret = se->ops->save_live_complete_precopy(f, se->opaque); | |
1047 | trace_savevm_section_end(se->idstr, se->section_id, ret); | |
1048 | save_section_footer(f, se); | |
1049 | if (ret < 0) { | |
1050 | qemu_file_set_error(f, ret); | |
1051 | return; | |
1052 | } | |
1053 | } | |
1054 | ||
1055 | if (iterable_only) { | |
1056 | return; | |
1057 | } | |
1058 | ||
1059 | vmdesc = qjson_new(); | |
1060 | json_prop_int(vmdesc, "page_size", TARGET_PAGE_SIZE); | |
1061 | json_start_array(vmdesc, "devices"); | |
1062 | QTAILQ_FOREACH(se, &savevm_state.handlers, entry) { | |
1063 | ||
1064 | if ((!se->ops || !se->ops->save_state) && !se->vmsd) { | |
1065 | continue; | |
1066 | } | |
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 | ||
1072 | trace_savevm_section_start(se->idstr, se->section_id); | |
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 | ||
1078 | save_section_header(f, se, QEMU_VM_SECTION_FULL); | |
1079 | vmstate_save(f, se, vmdesc); | |
1080 | trace_savevm_section_end(se->idstr, se->section_id, 0); | |
1081 | save_section_footer(f, se); | |
1082 | ||
1083 | json_end_object(vmdesc); | |
1084 | } | |
1085 | ||
1086 | if (!in_postcopy) { | |
1087 | /* Postcopy stream will still be going */ | |
1088 | qemu_put_byte(f, QEMU_VM_EOF); | |
1089 | } | |
1090 | ||
1091 | json_end_array(vmdesc); | |
1092 | qjson_finish(vmdesc); | |
1093 | vmdesc_len = strlen(qjson_get_str(vmdesc)); | |
1094 | ||
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 | } | |
1100 | qjson_destroy(vmdesc); | |
1101 | ||
1102 | qemu_fflush(f); | |
1103 | } | |
1104 | ||
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 | */ | |
1109 | void qemu_savevm_state_pending(QEMUFile *f, uint64_t max_size, | |
1110 | uint64_t *res_non_postcopiable, | |
1111 | uint64_t *res_postcopiable) | |
1112 | { | |
1113 | SaveStateEntry *se; | |
1114 | ||
1115 | *res_non_postcopiable = 0; | |
1116 | *res_postcopiable = 0; | |
1117 | ||
1118 | ||
1119 | QTAILQ_FOREACH(se, &savevm_state.handlers, entry) { | |
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 | } | |
1128 | se->ops->save_live_pending(f, se->opaque, max_size, | |
1129 | res_non_postcopiable, res_postcopiable); | |
1130 | } | |
1131 | } | |
1132 | ||
1133 | void qemu_savevm_state_cleanup(void) | |
1134 | { | |
1135 | SaveStateEntry *se; | |
1136 | ||
1137 | trace_savevm_state_cleanup(); | |
1138 | QTAILQ_FOREACH(se, &savevm_state.handlers, entry) { | |
1139 | if (se->ops && se->ops->cleanup) { | |
1140 | se->ops->cleanup(se->opaque); | |
1141 | } | |
1142 | } | |
1143 | } | |
1144 | ||
1145 | static int qemu_savevm_state(QEMUFile *f, Error **errp) | |
1146 | { | |
1147 | int ret; | |
1148 | MigrationParams params = { | |
1149 | .blk = 0, | |
1150 | .shared = 0 | |
1151 | }; | |
1152 | MigrationState *ms = migrate_init(¶ms); | |
1153 | ms->to_dst_file = f; | |
1154 | ||
1155 | if (migration_is_blocked(errp)) { | |
1156 | return -EINVAL; | |
1157 | } | |
1158 | ||
1159 | qemu_mutex_unlock_iothread(); | |
1160 | qemu_savevm_state_header(f); | |
1161 | qemu_savevm_state_begin(f, ¶ms); | |
1162 | qemu_mutex_lock_iothread(); | |
1163 | ||
1164 | while (qemu_file_get_error(f) == 0) { | |
1165 | if (qemu_savevm_state_iterate(f, false) > 0) { | |
1166 | break; | |
1167 | } | |
1168 | } | |
1169 | ||
1170 | ret = qemu_file_get_error(f); | |
1171 | if (ret == 0) { | |
1172 | qemu_savevm_state_complete_precopy(f, false); | |
1173 | ret = qemu_file_get_error(f); | |
1174 | } | |
1175 | qemu_savevm_state_cleanup(); | |
1176 | if (ret != 0) { | |
1177 | error_setg_errno(errp, -ret, "Error while writing VM state"); | |
1178 | } | |
1179 | return ret; | |
1180 | } | |
1181 | ||
1182 | static int qemu_save_device_state(QEMUFile *f) | |
1183 | { | |
1184 | SaveStateEntry *se; | |
1185 | ||
1186 | qemu_put_be32(f, QEMU_VM_FILE_MAGIC); | |
1187 | qemu_put_be32(f, QEMU_VM_FILE_VERSION); | |
1188 | ||
1189 | cpu_synchronize_all_states(); | |
1190 | ||
1191 | QTAILQ_FOREACH(se, &savevm_state.handlers, entry) { | |
1192 | if (se->is_ram) { | |
1193 | continue; | |
1194 | } | |
1195 | if ((!se->ops || !se->ops->save_state) && !se->vmsd) { | |
1196 | continue; | |
1197 | } | |
1198 | if (se->vmsd && !vmstate_save_needed(se->vmsd, se->opaque)) { | |
1199 | continue; | |
1200 | } | |
1201 | ||
1202 | save_section_header(f, se, QEMU_VM_SECTION_FULL); | |
1203 | ||
1204 | vmstate_save(f, se, NULL); | |
1205 | ||
1206 | save_section_footer(f, se); | |
1207 | } | |
1208 | ||
1209 | qemu_put_byte(f, QEMU_VM_EOF); | |
1210 | ||
1211 | return qemu_file_get_error(f); | |
1212 | } | |
1213 | ||
1214 | static SaveStateEntry *find_se(const char *idstr, int instance_id) | |
1215 | { | |
1216 | SaveStateEntry *se; | |
1217 | ||
1218 | QTAILQ_FOREACH(se, &savevm_state.handlers, entry) { | |
1219 | if (!strcmp(se->idstr, idstr) && | |
1220 | (instance_id == se->instance_id || | |
1221 | instance_id == se->alias_id)) | |
1222 | return se; | |
1223 | /* Migrating from an older version? */ | |
1224 | if (strstr(se->idstr, idstr) && se->compat) { | |
1225 | if (!strcmp(se->compat->idstr, idstr) && | |
1226 | (instance_id == se->compat->instance_id || | |
1227 | instance_id == se->alias_id)) | |
1228 | return se; | |
1229 | } | |
1230 | } | |
1231 | return NULL; | |
1232 | } | |
1233 | ||
1234 | enum LoadVMExitCodes { | |
1235 | /* Allow a command to quit all layers of nested loadvm loops */ | |
1236 | LOADVM_QUIT = 1, | |
1237 | }; | |
1238 | ||
1239 | static int qemu_loadvm_state_main(QEMUFile *f, MigrationIncomingState *mis); | |
1240 | ||
1241 | /* ------ incoming postcopy messages ------ */ | |
1242 | /* 'advise' arrives before any transfers just to tell us that a postcopy | |
1243 | * *might* happen - it might be skipped if precopy transferred everything | |
1244 | * quickly. | |
1245 | */ | |
1246 | static int loadvm_postcopy_handle_advise(MigrationIncomingState *mis) | |
1247 | { | |
1248 | PostcopyState ps = postcopy_state_set(POSTCOPY_INCOMING_ADVISE); | |
1249 | uint64_t remote_hps, remote_tps; | |
1250 | ||
1251 | trace_loadvm_postcopy_handle_advise(); | |
1252 | if (ps != POSTCOPY_INCOMING_NONE) { | |
1253 | error_report("CMD_POSTCOPY_ADVISE in wrong postcopy state (%d)", ps); | |
1254 | return -1; | |
1255 | } | |
1256 | ||
1257 | if (!postcopy_ram_supported_by_host()) { | |
1258 | return -1; | |
1259 | } | |
1260 | ||
1261 | remote_hps = qemu_get_be64(mis->from_src_file); | |
1262 | if (remote_hps != getpagesize()) { | |
1263 | /* | |
1264 | * Some combinations of mismatch are probably possible but it gets | |
1265 | * a bit more complicated. In particular we need to place whole | |
1266 | * host pages on the dest at once, and we need to ensure that we | |
1267 | * handle dirtying to make sure we never end up sending part of | |
1268 | * a hostpage on it's own. | |
1269 | */ | |
1270 | error_report("Postcopy needs matching host page sizes (s=%d d=%d)", | |
1271 | (int)remote_hps, getpagesize()); | |
1272 | return -1; | |
1273 | } | |
1274 | ||
1275 | remote_tps = qemu_get_be64(mis->from_src_file); | |
1276 | if (remote_tps != (1ul << qemu_target_page_bits())) { | |
1277 | /* | |
1278 | * Again, some differences could be dealt with, but for now keep it | |
1279 | * simple. | |
1280 | */ | |
1281 | error_report("Postcopy needs matching target page sizes (s=%d d=%d)", | |
1282 | (int)remote_tps, 1 << qemu_target_page_bits()); | |
1283 | return -1; | |
1284 | } | |
1285 | ||
1286 | if (ram_postcopy_incoming_init(mis)) { | |
1287 | return -1; | |
1288 | } | |
1289 | ||
1290 | postcopy_state_set(POSTCOPY_INCOMING_ADVISE); | |
1291 | ||
1292 | return 0; | |
1293 | } | |
1294 | ||
1295 | /* After postcopy we will be told to throw some pages away since they're | |
1296 | * dirty and will have to be demand fetched. Must happen before CPU is | |
1297 | * started. | |
1298 | * There can be 0..many of these messages, each encoding multiple pages. | |
1299 | */ | |
1300 | static int loadvm_postcopy_ram_handle_discard(MigrationIncomingState *mis, | |
1301 | uint16_t len) | |
1302 | { | |
1303 | int tmp; | |
1304 | char ramid[256]; | |
1305 | PostcopyState ps = postcopy_state_get(); | |
1306 | ||
1307 | trace_loadvm_postcopy_ram_handle_discard(); | |
1308 | ||
1309 | switch (ps) { | |
1310 | case POSTCOPY_INCOMING_ADVISE: | |
1311 | /* 1st discard */ | |
1312 | tmp = postcopy_ram_prepare_discard(mis); | |
1313 | if (tmp) { | |
1314 | return tmp; | |
1315 | } | |
1316 | break; | |
1317 | ||
1318 | case POSTCOPY_INCOMING_DISCARD: | |
1319 | /* Expected state */ | |
1320 | break; | |
1321 | ||
1322 | default: | |
1323 | error_report("CMD_POSTCOPY_RAM_DISCARD in wrong postcopy state (%d)", | |
1324 | ps); | |
1325 | return -1; | |
1326 | } | |
1327 | /* We're expecting a | |
1328 | * Version (0) | |
1329 | * a RAM ID string (length byte, name, 0 term) | |
1330 | * then at least 1 16 byte chunk | |
1331 | */ | |
1332 | if (len < (1 + 1 + 1 + 1 + 2 * 8)) { | |
1333 | error_report("CMD_POSTCOPY_RAM_DISCARD invalid length (%d)", len); | |
1334 | return -1; | |
1335 | } | |
1336 | ||
1337 | tmp = qemu_get_byte(mis->from_src_file); | |
1338 | if (tmp != postcopy_ram_discard_version) { | |
1339 | error_report("CMD_POSTCOPY_RAM_DISCARD invalid version (%d)", tmp); | |
1340 | return -1; | |
1341 | } | |
1342 | ||
1343 | if (!qemu_get_counted_string(mis->from_src_file, ramid)) { | |
1344 | error_report("CMD_POSTCOPY_RAM_DISCARD Failed to read RAMBlock ID"); | |
1345 | return -1; | |
1346 | } | |
1347 | tmp = qemu_get_byte(mis->from_src_file); | |
1348 | if (tmp != 0) { | |
1349 | error_report("CMD_POSTCOPY_RAM_DISCARD missing nil (%d)", tmp); | |
1350 | return -1; | |
1351 | } | |
1352 | ||
1353 | len -= 3 + strlen(ramid); | |
1354 | if (len % 16) { | |
1355 | error_report("CMD_POSTCOPY_RAM_DISCARD invalid length (%d)", len); | |
1356 | return -1; | |
1357 | } | |
1358 | trace_loadvm_postcopy_ram_handle_discard_header(ramid, len); | |
1359 | while (len) { | |
1360 | uint64_t start_addr, block_length; | |
1361 | start_addr = qemu_get_be64(mis->from_src_file); | |
1362 | block_length = qemu_get_be64(mis->from_src_file); | |
1363 | ||
1364 | len -= 16; | |
1365 | int ret = ram_discard_range(mis, ramid, start_addr, | |
1366 | block_length); | |
1367 | if (ret) { | |
1368 | return ret; | |
1369 | } | |
1370 | } | |
1371 | trace_loadvm_postcopy_ram_handle_discard_end(); | |
1372 | ||
1373 | return 0; | |
1374 | } | |
1375 | ||
1376 | /* | |
1377 | * Triggered by a postcopy_listen command; this thread takes over reading | |
1378 | * the input stream, leaving the main thread free to carry on loading the rest | |
1379 | * of the device state (from RAM). | |
1380 | * (TODO:This could do with being in a postcopy file - but there again it's | |
1381 | * just another input loop, not that postcopy specific) | |
1382 | */ | |
1383 | static void *postcopy_ram_listen_thread(void *opaque) | |
1384 | { | |
1385 | QEMUFile *f = opaque; | |
1386 | MigrationIncomingState *mis = migration_incoming_get_current(); | |
1387 | int load_res; | |
1388 | ||
1389 | migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE, | |
1390 | MIGRATION_STATUS_POSTCOPY_ACTIVE); | |
1391 | qemu_sem_post(&mis->listen_thread_sem); | |
1392 | trace_postcopy_ram_listen_thread_start(); | |
1393 | ||
1394 | /* | |
1395 | * Because we're a thread and not a coroutine we can't yield | |
1396 | * in qemu_file, and thus we must be blocking now. | |
1397 | */ | |
1398 | qemu_file_set_blocking(f, true); | |
1399 | load_res = qemu_loadvm_state_main(f, mis); | |
1400 | /* And non-blocking again so we don't block in any cleanup */ | |
1401 | qemu_file_set_blocking(f, false); | |
1402 | ||
1403 | trace_postcopy_ram_listen_thread_exit(); | |
1404 | if (load_res < 0) { | |
1405 | error_report("%s: loadvm failed: %d", __func__, load_res); | |
1406 | qemu_file_set_error(f, load_res); | |
1407 | migrate_set_state(&mis->state, MIGRATION_STATUS_POSTCOPY_ACTIVE, | |
1408 | MIGRATION_STATUS_FAILED); | |
1409 | } else { | |
1410 | /* | |
1411 | * This looks good, but it's possible that the device loading in the | |
1412 | * main thread hasn't finished yet, and so we might not be in 'RUN' | |
1413 | * state yet; wait for the end of the main thread. | |
1414 | */ | |
1415 | qemu_event_wait(&mis->main_thread_load_event); | |
1416 | } | |
1417 | postcopy_ram_incoming_cleanup(mis); | |
1418 | ||
1419 | if (load_res < 0) { | |
1420 | /* | |
1421 | * If something went wrong then we have a bad state so exit; | |
1422 | * depending how far we got it might be possible at this point | |
1423 | * to leave the guest running and fire MCEs for pages that never | |
1424 | * arrived as a desperate recovery step. | |
1425 | */ | |
1426 | exit(EXIT_FAILURE); | |
1427 | } | |
1428 | ||
1429 | migrate_set_state(&mis->state, MIGRATION_STATUS_POSTCOPY_ACTIVE, | |
1430 | MIGRATION_STATUS_COMPLETED); | |
1431 | /* | |
1432 | * If everything has worked fine, then the main thread has waited | |
1433 | * for us to start, and we're the last use of the mis. | |
1434 | * (If something broke then qemu will have to exit anyway since it's | |
1435 | * got a bad migration state). | |
1436 | */ | |
1437 | migration_incoming_state_destroy(); | |
1438 | ||
1439 | ||
1440 | return NULL; | |
1441 | } | |
1442 | ||
1443 | /* After this message we must be able to immediately receive postcopy data */ | |
1444 | static int loadvm_postcopy_handle_listen(MigrationIncomingState *mis) | |
1445 | { | |
1446 | PostcopyState ps = postcopy_state_set(POSTCOPY_INCOMING_LISTENING); | |
1447 | trace_loadvm_postcopy_handle_listen(); | |
1448 | if (ps != POSTCOPY_INCOMING_ADVISE && ps != POSTCOPY_INCOMING_DISCARD) { | |
1449 | error_report("CMD_POSTCOPY_LISTEN in wrong postcopy state (%d)", ps); | |
1450 | return -1; | |
1451 | } | |
1452 | if (ps == POSTCOPY_INCOMING_ADVISE) { | |
1453 | /* | |
1454 | * A rare case, we entered listen without having to do any discards, | |
1455 | * so do the setup that's normally done at the time of the 1st discard. | |
1456 | */ | |
1457 | postcopy_ram_prepare_discard(mis); | |
1458 | } | |
1459 | ||
1460 | /* | |
1461 | * Sensitise RAM - can now generate requests for blocks that don't exist | |
1462 | * However, at this point the CPU shouldn't be running, and the IO | |
1463 | * shouldn't be doing anything yet so don't actually expect requests | |
1464 | */ | |
1465 | if (postcopy_ram_enable_notify(mis)) { | |
1466 | return -1; | |
1467 | } | |
1468 | ||
1469 | if (mis->have_listen_thread) { | |
1470 | error_report("CMD_POSTCOPY_RAM_LISTEN already has a listen thread"); | |
1471 | return -1; | |
1472 | } | |
1473 | ||
1474 | mis->have_listen_thread = true; | |
1475 | /* Start up the listening thread and wait for it to signal ready */ | |
1476 | qemu_sem_init(&mis->listen_thread_sem, 0); | |
1477 | qemu_thread_create(&mis->listen_thread, "postcopy/listen", | |
1478 | postcopy_ram_listen_thread, mis->from_src_file, | |
1479 | QEMU_THREAD_DETACHED); | |
1480 | qemu_sem_wait(&mis->listen_thread_sem); | |
1481 | qemu_sem_destroy(&mis->listen_thread_sem); | |
1482 | ||
1483 | return 0; | |
1484 | } | |
1485 | ||
1486 | ||
1487 | typedef struct { | |
1488 | QEMUBH *bh; | |
1489 | } HandleRunBhData; | |
1490 | ||
1491 | static void loadvm_postcopy_handle_run_bh(void *opaque) | |
1492 | { | |
1493 | Error *local_err = NULL; | |
1494 | HandleRunBhData *data = opaque; | |
1495 | ||
1496 | /* TODO we should move all of this lot into postcopy_ram.c or a shared code | |
1497 | * in migration.c | |
1498 | */ | |
1499 | cpu_synchronize_all_post_init(); | |
1500 | ||
1501 | qemu_announce_self(); | |
1502 | ||
1503 | /* Make sure all file formats flush their mutable metadata */ | |
1504 | bdrv_invalidate_cache_all(&local_err); | |
1505 | if (local_err) { | |
1506 | error_report_err(local_err); | |
1507 | } | |
1508 | ||
1509 | trace_loadvm_postcopy_handle_run_cpu_sync(); | |
1510 | cpu_synchronize_all_post_init(); | |
1511 | ||
1512 | trace_loadvm_postcopy_handle_run_vmstart(); | |
1513 | ||
1514 | if (autostart) { | |
1515 | /* Hold onto your hats, starting the CPU */ | |
1516 | vm_start(); | |
1517 | } else { | |
1518 | /* leave it paused and let management decide when to start the CPU */ | |
1519 | runstate_set(RUN_STATE_PAUSED); | |
1520 | } | |
1521 | ||
1522 | qemu_bh_delete(data->bh); | |
1523 | g_free(data); | |
1524 | } | |
1525 | ||
1526 | /* After all discards we can start running and asking for pages */ | |
1527 | static int loadvm_postcopy_handle_run(MigrationIncomingState *mis) | |
1528 | { | |
1529 | PostcopyState ps = postcopy_state_set(POSTCOPY_INCOMING_RUNNING); | |
1530 | HandleRunBhData *data; | |
1531 | ||
1532 | trace_loadvm_postcopy_handle_run(); | |
1533 | if (ps != POSTCOPY_INCOMING_LISTENING) { | |
1534 | error_report("CMD_POSTCOPY_RUN in wrong postcopy state (%d)", ps); | |
1535 | return -1; | |
1536 | } | |
1537 | ||
1538 | data = g_new(HandleRunBhData, 1); | |
1539 | data->bh = qemu_bh_new(loadvm_postcopy_handle_run_bh, data); | |
1540 | qemu_bh_schedule(data->bh); | |
1541 | ||
1542 | /* We need to finish reading the stream from the package | |
1543 | * and also stop reading anything more from the stream that loaded the | |
1544 | * package (since it's now being read by the listener thread). | |
1545 | * LOADVM_QUIT will quit all the layers of nested loadvm loops. | |
1546 | */ | |
1547 | return LOADVM_QUIT; | |
1548 | } | |
1549 | ||
1550 | /** | |
1551 | * Immediately following this command is a blob of data containing an embedded | |
1552 | * chunk of migration stream; read it and load it. | |
1553 | * | |
1554 | * @mis: Incoming state | |
1555 | * @length: Length of packaged data to read | |
1556 | * | |
1557 | * Returns: Negative values on error | |
1558 | * | |
1559 | */ | |
1560 | static int loadvm_handle_cmd_packaged(MigrationIncomingState *mis) | |
1561 | { | |
1562 | int ret; | |
1563 | size_t length; | |
1564 | QIOChannelBuffer *bioc; | |
1565 | ||
1566 | length = qemu_get_be32(mis->from_src_file); | |
1567 | trace_loadvm_handle_cmd_packaged(length); | |
1568 | ||
1569 | if (length > MAX_VM_CMD_PACKAGED_SIZE) { | |
1570 | error_report("Unreasonably large packaged state: %zu", length); | |
1571 | return -1; | |
1572 | } | |
1573 | ||
1574 | bioc = qio_channel_buffer_new(length); | |
1575 | ret = qemu_get_buffer(mis->from_src_file, | |
1576 | bioc->data, | |
1577 | length); | |
1578 | if (ret != length) { | |
1579 | object_unref(OBJECT(bioc)); | |
1580 | error_report("CMD_PACKAGED: Buffer receive fail ret=%d length=%zu", | |
1581 | ret, length); | |
1582 | return (ret < 0) ? ret : -EAGAIN; | |
1583 | } | |
1584 | bioc->usage += length; | |
1585 | trace_loadvm_handle_cmd_packaged_received(ret); | |
1586 | ||
1587 | QEMUFile *packf = qemu_fopen_channel_input(QIO_CHANNEL(bioc)); | |
1588 | ||
1589 | ret = qemu_loadvm_state_main(packf, mis); | |
1590 | trace_loadvm_handle_cmd_packaged_main(ret); | |
1591 | qemu_fclose(packf); | |
1592 | object_unref(OBJECT(bioc)); | |
1593 | ||
1594 | return ret; | |
1595 | } | |
1596 | ||
1597 | /* | |
1598 | * Process an incoming 'QEMU_VM_COMMAND' | |
1599 | * 0 just a normal return | |
1600 | * LOADVM_QUIT All good, but exit the loop | |
1601 | * <0 Error | |
1602 | */ | |
1603 | static int loadvm_process_command(QEMUFile *f) | |
1604 | { | |
1605 | MigrationIncomingState *mis = migration_incoming_get_current(); | |
1606 | uint16_t cmd; | |
1607 | uint16_t len; | |
1608 | uint32_t tmp32; | |
1609 | ||
1610 | cmd = qemu_get_be16(f); | |
1611 | len = qemu_get_be16(f); | |
1612 | ||
1613 | trace_loadvm_process_command(cmd, len); | |
1614 | if (cmd >= MIG_CMD_MAX || cmd == MIG_CMD_INVALID) { | |
1615 | error_report("MIG_CMD 0x%x unknown (len 0x%x)", cmd, len); | |
1616 | return -EINVAL; | |
1617 | } | |
1618 | ||
1619 | if (mig_cmd_args[cmd].len != -1 && mig_cmd_args[cmd].len != len) { | |
1620 | error_report("%s received with bad length - expecting %zu, got %d", | |
1621 | mig_cmd_args[cmd].name, | |
1622 | (size_t)mig_cmd_args[cmd].len, len); | |
1623 | return -ERANGE; | |
1624 | } | |
1625 | ||
1626 | switch (cmd) { | |
1627 | case MIG_CMD_OPEN_RETURN_PATH: | |
1628 | if (mis->to_src_file) { | |
1629 | error_report("CMD_OPEN_RETURN_PATH called when RP already open"); | |
1630 | /* Not really a problem, so don't give up */ | |
1631 | return 0; | |
1632 | } | |
1633 | mis->to_src_file = qemu_file_get_return_path(f); | |
1634 | if (!mis->to_src_file) { | |
1635 | error_report("CMD_OPEN_RETURN_PATH failed"); | |
1636 | return -1; | |
1637 | } | |
1638 | break; | |
1639 | ||
1640 | case MIG_CMD_PING: | |
1641 | tmp32 = qemu_get_be32(f); | |
1642 | trace_loadvm_process_command_ping(tmp32); | |
1643 | if (!mis->to_src_file) { | |
1644 | error_report("CMD_PING (0x%x) received with no return path", | |
1645 | tmp32); | |
1646 | return -1; | |
1647 | } | |
1648 | migrate_send_rp_pong(mis, tmp32); | |
1649 | break; | |
1650 | ||
1651 | case MIG_CMD_PACKAGED: | |
1652 | return loadvm_handle_cmd_packaged(mis); | |
1653 | ||
1654 | case MIG_CMD_POSTCOPY_ADVISE: | |
1655 | return loadvm_postcopy_handle_advise(mis); | |
1656 | ||
1657 | case MIG_CMD_POSTCOPY_LISTEN: | |
1658 | return loadvm_postcopy_handle_listen(mis); | |
1659 | ||
1660 | case MIG_CMD_POSTCOPY_RUN: | |
1661 | return loadvm_postcopy_handle_run(mis); | |
1662 | ||
1663 | case MIG_CMD_POSTCOPY_RAM_DISCARD: | |
1664 | return loadvm_postcopy_ram_handle_discard(mis, len); | |
1665 | } | |
1666 | ||
1667 | return 0; | |
1668 | } | |
1669 | ||
1670 | struct LoadStateEntry { | |
1671 | QLIST_ENTRY(LoadStateEntry) entry; | |
1672 | SaveStateEntry *se; | |
1673 | int section_id; | |
1674 | int version_id; | |
1675 | }; | |
1676 | ||
1677 | /* | |
1678 | * Read a footer off the wire and check that it matches the expected section | |
1679 | * | |
1680 | * Returns: true if the footer was good | |
1681 | * false if there is a problem (and calls error_report to say why) | |
1682 | */ | |
1683 | static bool check_section_footer(QEMUFile *f, LoadStateEntry *le) | |
1684 | { | |
1685 | uint8_t read_mark; | |
1686 | uint32_t read_section_id; | |
1687 | ||
1688 | if (skip_section_footers) { | |
1689 | /* No footer to check */ | |
1690 | return true; | |
1691 | } | |
1692 | ||
1693 | read_mark = qemu_get_byte(f); | |
1694 | ||
1695 | if (read_mark != QEMU_VM_SECTION_FOOTER) { | |
1696 | error_report("Missing section footer for %s", le->se->idstr); | |
1697 | return false; | |
1698 | } | |
1699 | ||
1700 | read_section_id = qemu_get_be32(f); | |
1701 | if (read_section_id != le->section_id) { | |
1702 | error_report("Mismatched section id in footer for %s -" | |
1703 | " read 0x%x expected 0x%x", | |
1704 | le->se->idstr, read_section_id, le->section_id); | |
1705 | return false; | |
1706 | } | |
1707 | ||
1708 | /* All good */ | |
1709 | return true; | |
1710 | } | |
1711 | ||
1712 | void loadvm_free_handlers(MigrationIncomingState *mis) | |
1713 | { | |
1714 | LoadStateEntry *le, *new_le; | |
1715 | ||
1716 | QLIST_FOREACH_SAFE(le, &mis->loadvm_handlers, entry, new_le) { | |
1717 | QLIST_REMOVE(le, entry); | |
1718 | g_free(le); | |
1719 | } | |
1720 | } | |
1721 | ||
1722 | static int | |
1723 | qemu_loadvm_section_start_full(QEMUFile *f, MigrationIncomingState *mis) | |
1724 | { | |
1725 | uint32_t instance_id, version_id, section_id; | |
1726 | SaveStateEntry *se; | |
1727 | LoadStateEntry *le; | |
1728 | char idstr[256]; | |
1729 | int ret; | |
1730 | ||
1731 | /* Read section start */ | |
1732 | section_id = qemu_get_be32(f); | |
1733 | if (!qemu_get_counted_string(f, idstr)) { | |
1734 | error_report("Unable to read ID string for section %u", | |
1735 | section_id); | |
1736 | return -EINVAL; | |
1737 | } | |
1738 | instance_id = qemu_get_be32(f); | |
1739 | version_id = qemu_get_be32(f); | |
1740 | ||
1741 | trace_qemu_loadvm_state_section_startfull(section_id, idstr, | |
1742 | instance_id, version_id); | |
1743 | /* Find savevm section */ | |
1744 | se = find_se(idstr, instance_id); | |
1745 | if (se == NULL) { | |
1746 | error_report("Unknown savevm section or instance '%s' %d", | |
1747 | idstr, instance_id); | |
1748 | return -EINVAL; | |
1749 | } | |
1750 | ||
1751 | /* Validate version */ | |
1752 | if (version_id > se->version_id) { | |
1753 | error_report("savevm: unsupported version %d for '%s' v%d", | |
1754 | version_id, idstr, se->version_id); | |
1755 | return -EINVAL; | |
1756 | } | |
1757 | ||
1758 | /* Validate if it is a device's state */ | |
1759 | if (xen_enabled() && se->is_ram) { | |
1760 | error_report("loadvm: %s RAM loading not allowed on Xen", idstr); | |
1761 | return -EINVAL; | |
1762 | } | |
1763 | ||
1764 | /* Add entry */ | |
1765 | le = g_malloc0(sizeof(*le)); | |
1766 | ||
1767 | le->se = se; | |
1768 | le->section_id = section_id; | |
1769 | le->version_id = version_id; | |
1770 | QLIST_INSERT_HEAD(&mis->loadvm_handlers, le, entry); | |
1771 | ||
1772 | ret = vmstate_load(f, le->se, le->version_id); | |
1773 | if (ret < 0) { | |
1774 | error_report("error while loading state for instance 0x%x of" | |
1775 | " device '%s'", instance_id, idstr); | |
1776 | return ret; | |
1777 | } | |
1778 | if (!check_section_footer(f, le)) { | |
1779 | return -EINVAL; | |
1780 | } | |
1781 | ||
1782 | return 0; | |
1783 | } | |
1784 | ||
1785 | static int | |
1786 | qemu_loadvm_section_part_end(QEMUFile *f, MigrationIncomingState *mis) | |
1787 | { | |
1788 | uint32_t section_id; | |
1789 | LoadStateEntry *le; | |
1790 | int ret; | |
1791 | ||
1792 | section_id = qemu_get_be32(f); | |
1793 | ||
1794 | trace_qemu_loadvm_state_section_partend(section_id); | |
1795 | QLIST_FOREACH(le, &mis->loadvm_handlers, entry) { | |
1796 | if (le->section_id == section_id) { | |
1797 | break; | |
1798 | } | |
1799 | } | |
1800 | if (le == NULL) { | |
1801 | error_report("Unknown savevm section %d", section_id); | |
1802 | return -EINVAL; | |
1803 | } | |
1804 | ||
1805 | ret = vmstate_load(f, le->se, le->version_id); | |
1806 | if (ret < 0) { | |
1807 | error_report("error while loading state section id %d(%s)", | |
1808 | section_id, le->se->idstr); | |
1809 | return ret; | |
1810 | } | |
1811 | if (!check_section_footer(f, le)) { | |
1812 | return -EINVAL; | |
1813 | } | |
1814 | ||
1815 | return 0; | |
1816 | } | |
1817 | ||
1818 | static int qemu_loadvm_state_main(QEMUFile *f, MigrationIncomingState *mis) | |
1819 | { | |
1820 | uint8_t section_type; | |
1821 | int ret; | |
1822 | ||
1823 | while ((section_type = qemu_get_byte(f)) != QEMU_VM_EOF) { | |
1824 | ||
1825 | trace_qemu_loadvm_state_section(section_type); | |
1826 | switch (section_type) { | |
1827 | case QEMU_VM_SECTION_START: | |
1828 | case QEMU_VM_SECTION_FULL: | |
1829 | ret = qemu_loadvm_section_start_full(f, mis); | |
1830 | if (ret < 0) { | |
1831 | return ret; | |
1832 | } | |
1833 | break; | |
1834 | case QEMU_VM_SECTION_PART: | |
1835 | case QEMU_VM_SECTION_END: | |
1836 | ret = qemu_loadvm_section_part_end(f, mis); | |
1837 | if (ret < 0) { | |
1838 | return ret; | |
1839 | } | |
1840 | break; | |
1841 | case QEMU_VM_COMMAND: | |
1842 | ret = loadvm_process_command(f); | |
1843 | trace_qemu_loadvm_state_section_command(ret); | |
1844 | if ((ret < 0) || (ret & LOADVM_QUIT)) { | |
1845 | return ret; | |
1846 | } | |
1847 | break; | |
1848 | default: | |
1849 | error_report("Unknown savevm section type %d", section_type); | |
1850 | return -EINVAL; | |
1851 | } | |
1852 | } | |
1853 | ||
1854 | return 0; | |
1855 | } | |
1856 | ||
1857 | int qemu_loadvm_state(QEMUFile *f) | |
1858 | { | |
1859 | MigrationIncomingState *mis = migration_incoming_get_current(); | |
1860 | Error *local_err = NULL; | |
1861 | unsigned int v; | |
1862 | int ret; | |
1863 | ||
1864 | if (qemu_savevm_state_blocked(&local_err)) { | |
1865 | error_report_err(local_err); | |
1866 | return -EINVAL; | |
1867 | } | |
1868 | ||
1869 | v = qemu_get_be32(f); | |
1870 | if (v != QEMU_VM_FILE_MAGIC) { | |
1871 | error_report("Not a migration stream"); | |
1872 | return -EINVAL; | |
1873 | } | |
1874 | ||
1875 | v = qemu_get_be32(f); | |
1876 | if (v == QEMU_VM_FILE_VERSION_COMPAT) { | |
1877 | error_report("SaveVM v2 format is obsolete and don't work anymore"); | |
1878 | return -ENOTSUP; | |
1879 | } | |
1880 | if (v != QEMU_VM_FILE_VERSION) { | |
1881 | error_report("Unsupported migration stream version"); | |
1882 | return -ENOTSUP; | |
1883 | } | |
1884 | ||
1885 | if (!savevm_state.skip_configuration || enforce_config_section()) { | |
1886 | if (qemu_get_byte(f) != QEMU_VM_CONFIGURATION) { | |
1887 | error_report("Configuration section missing"); | |
1888 | return -EINVAL; | |
1889 | } | |
1890 | ret = vmstate_load_state(f, &vmstate_configuration, &savevm_state, 0); | |
1891 | ||
1892 | if (ret) { | |
1893 | return ret; | |
1894 | } | |
1895 | } | |
1896 | ||
1897 | ret = qemu_loadvm_state_main(f, mis); | |
1898 | qemu_event_set(&mis->main_thread_load_event); | |
1899 | ||
1900 | trace_qemu_loadvm_state_post_main(ret); | |
1901 | ||
1902 | if (mis->have_listen_thread) { | |
1903 | /* Listen thread still going, can't clean up yet */ | |
1904 | return ret; | |
1905 | } | |
1906 | ||
1907 | if (ret == 0) { | |
1908 | ret = qemu_file_get_error(f); | |
1909 | } | |
1910 | ||
1911 | /* | |
1912 | * Try to read in the VMDESC section as well, so that dumping tools that | |
1913 | * intercept our migration stream have the chance to see it. | |
1914 | */ | |
1915 | ||
1916 | /* We've got to be careful; if we don't read the data and just shut the fd | |
1917 | * then the sender can error if we close while it's still sending. | |
1918 | * We also mustn't read data that isn't there; some transports (RDMA) | |
1919 | * will stall waiting for that data when the source has already closed. | |
1920 | */ | |
1921 | if (ret == 0 && should_send_vmdesc()) { | |
1922 | uint8_t *buf; | |
1923 | uint32_t size; | |
1924 | uint8_t section_type = qemu_get_byte(f); | |
1925 | ||
1926 | if (section_type != QEMU_VM_VMDESCRIPTION) { | |
1927 | error_report("Expected vmdescription section, but got %d", | |
1928 | section_type); | |
1929 | /* | |
1930 | * It doesn't seem worth failing at this point since | |
1931 | * we apparently have an otherwise valid VM state | |
1932 | */ | |
1933 | } else { | |
1934 | buf = g_malloc(0x1000); | |
1935 | size = qemu_get_be32(f); | |
1936 | ||
1937 | while (size > 0) { | |
1938 | uint32_t read_chunk = MIN(size, 0x1000); | |
1939 | qemu_get_buffer(f, buf, read_chunk); | |
1940 | size -= read_chunk; | |
1941 | } | |
1942 | g_free(buf); | |
1943 | } | |
1944 | } | |
1945 | ||
1946 | cpu_synchronize_all_post_init(); | |
1947 | ||
1948 | return ret; | |
1949 | } | |
1950 | ||
1951 | void hmp_savevm(Monitor *mon, const QDict *qdict) | |
1952 | { | |
1953 | BlockDriverState *bs, *bs1; | |
1954 | QEMUSnapshotInfo sn1, *sn = &sn1, old_sn1, *old_sn = &old_sn1; | |
1955 | int ret; | |
1956 | QEMUFile *f; | |
1957 | int saved_vm_running; | |
1958 | uint64_t vm_state_size; | |
1959 | qemu_timeval tv; | |
1960 | struct tm tm; | |
1961 | const char *name = qdict_get_try_str(qdict, "name"); | |
1962 | Error *local_err = NULL; | |
1963 | AioContext *aio_context; | |
1964 | ||
1965 | if (!bdrv_all_can_snapshot(&bs)) { | |
1966 | monitor_printf(mon, "Device '%s' is writable but does not " | |
1967 | "support snapshots.\n", bdrv_get_device_name(bs)); | |
1968 | return; | |
1969 | } | |
1970 | ||
1971 | /* Delete old snapshots of the same name */ | |
1972 | if (name && bdrv_all_delete_snapshot(name, &bs1, &local_err) < 0) { | |
1973 | error_reportf_err(local_err, | |
1974 | "Error while deleting snapshot on device '%s': ", | |
1975 | bdrv_get_device_name(bs1)); | |
1976 | return; | |
1977 | } | |
1978 | ||
1979 | bs = bdrv_all_find_vmstate_bs(); | |
1980 | if (bs == NULL) { | |
1981 | monitor_printf(mon, "No block device can accept snapshots\n"); | |
1982 | return; | |
1983 | } | |
1984 | aio_context = bdrv_get_aio_context(bs); | |
1985 | ||
1986 | saved_vm_running = runstate_is_running(); | |
1987 | ||
1988 | ret = global_state_store(); | |
1989 | if (ret) { | |
1990 | monitor_printf(mon, "Error saving global state\n"); | |
1991 | return; | |
1992 | } | |
1993 | vm_stop(RUN_STATE_SAVE_VM); | |
1994 | ||
1995 | aio_context_acquire(aio_context); | |
1996 | ||
1997 | memset(sn, 0, sizeof(*sn)); | |
1998 | ||
1999 | /* fill auxiliary fields */ | |
2000 | qemu_gettimeofday(&tv); | |
2001 | sn->date_sec = tv.tv_sec; | |
2002 | sn->date_nsec = tv.tv_usec * 1000; | |
2003 | sn->vm_clock_nsec = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL); | |
2004 | ||
2005 | if (name) { | |
2006 | ret = bdrv_snapshot_find(bs, old_sn, name); | |
2007 | if (ret >= 0) { | |
2008 | pstrcpy(sn->name, sizeof(sn->name), old_sn->name); | |
2009 | pstrcpy(sn->id_str, sizeof(sn->id_str), old_sn->id_str); | |
2010 | } else { | |
2011 | pstrcpy(sn->name, sizeof(sn->name), name); | |
2012 | } | |
2013 | } else { | |
2014 | /* cast below needed for OpenBSD where tv_sec is still 'long' */ | |
2015 | localtime_r((const time_t *)&tv.tv_sec, &tm); | |
2016 | strftime(sn->name, sizeof(sn->name), "vm-%Y%m%d%H%M%S", &tm); | |
2017 | } | |
2018 | ||
2019 | /* save the VM state */ | |
2020 | f = qemu_fopen_bdrv(bs, 1); | |
2021 | if (!f) { | |
2022 | monitor_printf(mon, "Could not open VM state file\n"); | |
2023 | goto the_end; | |
2024 | } | |
2025 | ret = qemu_savevm_state(f, &local_err); | |
2026 | vm_state_size = qemu_ftell(f); | |
2027 | qemu_fclose(f); | |
2028 | if (ret < 0) { | |
2029 | error_report_err(local_err); | |
2030 | goto the_end; | |
2031 | } | |
2032 | ||
2033 | ret = bdrv_all_create_snapshot(sn, bs, vm_state_size, &bs); | |
2034 | if (ret < 0) { | |
2035 | monitor_printf(mon, "Error while creating snapshot on '%s'\n", | |
2036 | bdrv_get_device_name(bs)); | |
2037 | } | |
2038 | ||
2039 | the_end: | |
2040 | aio_context_release(aio_context); | |
2041 | if (saved_vm_running) { | |
2042 | vm_start(); | |
2043 | } | |
2044 | } | |
2045 | ||
2046 | void qmp_xen_save_devices_state(const char *filename, Error **errp) | |
2047 | { | |
2048 | QEMUFile *f; | |
2049 | QIOChannelFile *ioc; | |
2050 | int saved_vm_running; | |
2051 | int ret; | |
2052 | ||
2053 | saved_vm_running = runstate_is_running(); | |
2054 | vm_stop(RUN_STATE_SAVE_VM); | |
2055 | global_state_store_running(); | |
2056 | ||
2057 | ioc = qio_channel_file_new_path(filename, O_WRONLY | O_CREAT, 0660, errp); | |
2058 | if (!ioc) { | |
2059 | goto the_end; | |
2060 | } | |
2061 | f = qemu_fopen_channel_output(QIO_CHANNEL(ioc)); | |
2062 | ret = qemu_save_device_state(f); | |
2063 | qemu_fclose(f); | |
2064 | if (ret < 0) { | |
2065 | error_setg(errp, QERR_IO_ERROR); | |
2066 | } | |
2067 | ||
2068 | the_end: | |
2069 | if (saved_vm_running) { | |
2070 | vm_start(); | |
2071 | } | |
2072 | } | |
2073 | ||
2074 | void qmp_xen_load_devices_state(const char *filename, Error **errp) | |
2075 | { | |
2076 | QEMUFile *f; | |
2077 | QIOChannelFile *ioc; | |
2078 | int ret; | |
2079 | ||
2080 | /* Guest must be paused before loading the device state; the RAM state | |
2081 | * will already have been loaded by xc | |
2082 | */ | |
2083 | if (runstate_is_running()) { | |
2084 | error_setg(errp, "Cannot update device state while vm is running"); | |
2085 | return; | |
2086 | } | |
2087 | vm_stop(RUN_STATE_RESTORE_VM); | |
2088 | ||
2089 | ioc = qio_channel_file_new_path(filename, O_RDONLY | O_BINARY, 0, errp); | |
2090 | if (!ioc) { | |
2091 | return; | |
2092 | } | |
2093 | f = qemu_fopen_channel_input(QIO_CHANNEL(ioc)); | |
2094 | ||
2095 | migration_incoming_state_new(f); | |
2096 | ret = qemu_loadvm_state(f); | |
2097 | qemu_fclose(f); | |
2098 | if (ret < 0) { | |
2099 | error_setg(errp, QERR_IO_ERROR); | |
2100 | } | |
2101 | migration_incoming_state_destroy(); | |
2102 | } | |
2103 | ||
2104 | int load_vmstate(const char *name) | |
2105 | { | |
2106 | BlockDriverState *bs, *bs_vm_state; | |
2107 | QEMUSnapshotInfo sn; | |
2108 | QEMUFile *f; | |
2109 | int ret; | |
2110 | AioContext *aio_context; | |
2111 | ||
2112 | if (!bdrv_all_can_snapshot(&bs)) { | |
2113 | error_report("Device '%s' is writable but does not support snapshots.", | |
2114 | bdrv_get_device_name(bs)); | |
2115 | return -ENOTSUP; | |
2116 | } | |
2117 | ret = bdrv_all_find_snapshot(name, &bs); | |
2118 | if (ret < 0) { | |
2119 | error_report("Device '%s' does not have the requested snapshot '%s'", | |
2120 | bdrv_get_device_name(bs), name); | |
2121 | return ret; | |
2122 | } | |
2123 | ||
2124 | bs_vm_state = bdrv_all_find_vmstate_bs(); | |
2125 | if (!bs_vm_state) { | |
2126 | error_report("No block device supports snapshots"); | |
2127 | return -ENOTSUP; | |
2128 | } | |
2129 | aio_context = bdrv_get_aio_context(bs_vm_state); | |
2130 | ||
2131 | /* Don't even try to load empty VM states */ | |
2132 | aio_context_acquire(aio_context); | |
2133 | ret = bdrv_snapshot_find(bs_vm_state, &sn, name); | |
2134 | aio_context_release(aio_context); | |
2135 | if (ret < 0) { | |
2136 | return ret; | |
2137 | } else if (sn.vm_state_size == 0) { | |
2138 | error_report("This is a disk-only snapshot. Revert to it offline " | |
2139 | "using qemu-img."); | |
2140 | return -EINVAL; | |
2141 | } | |
2142 | ||
2143 | /* Flush all IO requests so they don't interfere with the new state. */ | |
2144 | bdrv_drain_all(); | |
2145 | ||
2146 | ret = bdrv_all_goto_snapshot(name, &bs); | |
2147 | if (ret < 0) { | |
2148 | error_report("Error %d while activating snapshot '%s' on '%s'", | |
2149 | ret, name, bdrv_get_device_name(bs)); | |
2150 | return ret; | |
2151 | } | |
2152 | ||
2153 | /* restore the VM state */ | |
2154 | f = qemu_fopen_bdrv(bs_vm_state, 0); | |
2155 | if (!f) { | |
2156 | error_report("Could not open VM state file"); | |
2157 | return -EINVAL; | |
2158 | } | |
2159 | ||
2160 | qemu_system_reset(VMRESET_SILENT); | |
2161 | migration_incoming_state_new(f); | |
2162 | ||
2163 | aio_context_acquire(aio_context); | |
2164 | ret = qemu_loadvm_state(f); | |
2165 | qemu_fclose(f); | |
2166 | aio_context_release(aio_context); | |
2167 | ||
2168 | migration_incoming_state_destroy(); | |
2169 | if (ret < 0) { | |
2170 | error_report("Error %d while loading VM state", ret); | |
2171 | return ret; | |
2172 | } | |
2173 | ||
2174 | return 0; | |
2175 | } | |
2176 | ||
2177 | void hmp_delvm(Monitor *mon, const QDict *qdict) | |
2178 | { | |
2179 | BlockDriverState *bs; | |
2180 | Error *err; | |
2181 | const char *name = qdict_get_str(qdict, "name"); | |
2182 | ||
2183 | if (bdrv_all_delete_snapshot(name, &bs, &err) < 0) { | |
2184 | error_reportf_err(err, | |
2185 | "Error while deleting snapshot on device '%s': ", | |
2186 | bdrv_get_device_name(bs)); | |
2187 | } | |
2188 | } | |
2189 | ||
2190 | void hmp_info_snapshots(Monitor *mon, const QDict *qdict) | |
2191 | { | |
2192 | BlockDriverState *bs, *bs1; | |
2193 | QEMUSnapshotInfo *sn_tab, *sn; | |
2194 | int nb_sns, i; | |
2195 | int total; | |
2196 | int *available_snapshots; | |
2197 | AioContext *aio_context; | |
2198 | ||
2199 | bs = bdrv_all_find_vmstate_bs(); | |
2200 | if (!bs) { | |
2201 | monitor_printf(mon, "No available block device supports snapshots\n"); | |
2202 | return; | |
2203 | } | |
2204 | aio_context = bdrv_get_aio_context(bs); | |
2205 | ||
2206 | aio_context_acquire(aio_context); | |
2207 | nb_sns = bdrv_snapshot_list(bs, &sn_tab); | |
2208 | aio_context_release(aio_context); | |
2209 | ||
2210 | if (nb_sns < 0) { | |
2211 | monitor_printf(mon, "bdrv_snapshot_list: error %d\n", nb_sns); | |
2212 | return; | |
2213 | } | |
2214 | ||
2215 | if (nb_sns == 0) { | |
2216 | monitor_printf(mon, "There is no snapshot available.\n"); | |
2217 | return; | |
2218 | } | |
2219 | ||
2220 | available_snapshots = g_new0(int, nb_sns); | |
2221 | total = 0; | |
2222 | for (i = 0; i < nb_sns; i++) { | |
2223 | if (bdrv_all_find_snapshot(sn_tab[i].id_str, &bs1) == 0) { | |
2224 | available_snapshots[total] = i; | |
2225 | total++; | |
2226 | } | |
2227 | } | |
2228 | ||
2229 | if (total > 0) { | |
2230 | bdrv_snapshot_dump((fprintf_function)monitor_printf, mon, NULL); | |
2231 | monitor_printf(mon, "\n"); | |
2232 | for (i = 0; i < total; i++) { | |
2233 | sn = &sn_tab[available_snapshots[i]]; | |
2234 | bdrv_snapshot_dump((fprintf_function)monitor_printf, mon, sn); | |
2235 | monitor_printf(mon, "\n"); | |
2236 | } | |
2237 | } else { | |
2238 | monitor_printf(mon, "There is no suitable snapshot available\n"); | |
2239 | } | |
2240 | ||
2241 | g_free(sn_tab); | |
2242 | g_free(available_snapshots); | |
2243 | ||
2244 | } | |
2245 | ||
2246 | void vmstate_register_ram(MemoryRegion *mr, DeviceState *dev) | |
2247 | { | |
2248 | qemu_ram_set_idstr(mr->ram_block, | |
2249 | memory_region_name(mr), dev); | |
2250 | } | |
2251 | ||
2252 | void vmstate_unregister_ram(MemoryRegion *mr, DeviceState *dev) | |
2253 | { | |
2254 | qemu_ram_unset_idstr(mr->ram_block); | |
2255 | } | |
2256 | ||
2257 | void vmstate_register_ram_global(MemoryRegion *mr) | |
2258 | { | |
2259 | vmstate_register_ram(mr, NULL); | |
2260 | } |