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