xref: /openbmc/qemu/migration/savevm.c (revision 4c4465ff)
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 "net/net.h"
32 #include "migration.h"
33 #include "migration/snapshot.h"
34 #include "migration/vmstate.h"
35 #include "migration/misc.h"
36 #include "migration/register.h"
37 #include "migration/global_state.h"
38 #include "ram.h"
39 #include "qemu-file-channel.h"
40 #include "qemu-file.h"
41 #include "savevm.h"
42 #include "postcopy-ram.h"
43 #include "qapi/error.h"
44 #include "qapi/qapi-commands-migration.h"
45 #include "qapi/qmp/json-writer.h"
46 #include "qapi/qmp/qerror.h"
47 #include "qemu/error-report.h"
48 #include "sysemu/cpus.h"
49 #include "exec/memory.h"
50 #include "exec/target_page.h"
51 #include "trace.h"
52 #include "qemu/iov.h"
53 #include "qemu/main-loop.h"
54 #include "block/snapshot.h"
55 #include "qemu/cutils.h"
56 #include "io/channel-buffer.h"
57 #include "io/channel-file.h"
58 #include "sysemu/replay.h"
59 #include "sysemu/runstate.h"
60 #include "sysemu/sysemu.h"
61 #include "sysemu/xen.h"
62 #include "migration/colo.h"
63 #include "qemu/bitmap.h"
64 #include "net/announce.h"
65 
66 const unsigned int postcopy_ram_discard_version;
67 
68 /* Subcommands for QEMU_VM_COMMAND */
69 enum qemu_vm_cmd {
70     MIG_CMD_INVALID = 0,   /* Must be 0 */
71     MIG_CMD_OPEN_RETURN_PATH,  /* Tell the dest to open the Return path */
72     MIG_CMD_PING,              /* Request a PONG on the RP */
73 
74     MIG_CMD_POSTCOPY_ADVISE,       /* Prior to any page transfers, just
75                                       warn we might want to do PC */
76     MIG_CMD_POSTCOPY_LISTEN,       /* Start listening for incoming
77                                       pages as it's running. */
78     MIG_CMD_POSTCOPY_RUN,          /* Start execution */
79 
80     MIG_CMD_POSTCOPY_RAM_DISCARD,  /* A list of pages to discard that
81                                       were previously sent during
82                                       precopy but are dirty. */
83     MIG_CMD_PACKAGED,          /* Send a wrapped stream within this stream */
84     MIG_CMD_ENABLE_COLO,       /* Enable COLO */
85     MIG_CMD_POSTCOPY_RESUME,   /* resume postcopy on dest */
86     MIG_CMD_RECV_BITMAP,       /* Request for recved bitmap on dst */
87     MIG_CMD_MAX
88 };
89 
90 #define MAX_VM_CMD_PACKAGED_SIZE UINT32_MAX
91 static struct mig_cmd_args {
92     ssize_t     len; /* -1 = variable */
93     const char *name;
94 } mig_cmd_args[] = {
95     [MIG_CMD_INVALID]          = { .len = -1, .name = "INVALID" },
96     [MIG_CMD_OPEN_RETURN_PATH] = { .len =  0, .name = "OPEN_RETURN_PATH" },
97     [MIG_CMD_PING]             = { .len = sizeof(uint32_t), .name = "PING" },
98     [MIG_CMD_POSTCOPY_ADVISE]  = { .len = -1, .name = "POSTCOPY_ADVISE" },
99     [MIG_CMD_POSTCOPY_LISTEN]  = { .len =  0, .name = "POSTCOPY_LISTEN" },
100     [MIG_CMD_POSTCOPY_RUN]     = { .len =  0, .name = "POSTCOPY_RUN" },
101     [MIG_CMD_POSTCOPY_RAM_DISCARD] = {
102                                    .len = -1, .name = "POSTCOPY_RAM_DISCARD" },
103     [MIG_CMD_POSTCOPY_RESUME]  = { .len =  0, .name = "POSTCOPY_RESUME" },
104     [MIG_CMD_PACKAGED]         = { .len =  4, .name = "PACKAGED" },
105     [MIG_CMD_RECV_BITMAP]      = { .len = -1, .name = "RECV_BITMAP" },
106     [MIG_CMD_MAX]              = { .len = -1, .name = "MAX" },
107 };
108 
109 /* Note for MIG_CMD_POSTCOPY_ADVISE:
110  * The format of arguments is depending on postcopy mode:
111  * - postcopy RAM only
112  *   uint64_t host page size
113  *   uint64_t taget page size
114  *
115  * - postcopy RAM and postcopy dirty bitmaps
116  *   format is the same as for postcopy RAM only
117  *
118  * - postcopy dirty bitmaps only
119  *   Nothing. Command length field is 0.
120  *
121  * Be careful: adding a new postcopy entity with some other parameters should
122  * not break format self-description ability. Good way is to introduce some
123  * generic extendable format with an exception for two old entities.
124  */
125 
126 /***********************************************************/
127 /* savevm/loadvm support */
128 
129 static ssize_t block_writev_buffer(void *opaque, struct iovec *iov, int iovcnt,
130                                    int64_t pos, Error **errp)
131 {
132     int ret;
133     QEMUIOVector qiov;
134 
135     qemu_iovec_init_external(&qiov, iov, iovcnt);
136     ret = bdrv_writev_vmstate(opaque, &qiov, pos);
137     if (ret < 0) {
138         return ret;
139     }
140 
141     return qiov.size;
142 }
143 
144 static ssize_t block_get_buffer(void *opaque, uint8_t *buf, int64_t pos,
145                                 size_t size, Error **errp)
146 {
147     return bdrv_load_vmstate(opaque, buf, pos, size);
148 }
149 
150 static int bdrv_fclose(void *opaque, Error **errp)
151 {
152     return bdrv_flush(opaque);
153 }
154 
155 static const QEMUFileOps bdrv_read_ops = {
156     .get_buffer = block_get_buffer,
157     .close =      bdrv_fclose
158 };
159 
160 static const QEMUFileOps bdrv_write_ops = {
161     .writev_buffer  = block_writev_buffer,
162     .close          = bdrv_fclose
163 };
164 
165 static QEMUFile *qemu_fopen_bdrv(BlockDriverState *bs, int is_writable)
166 {
167     if (is_writable) {
168         return qemu_fopen_ops(bs, &bdrv_write_ops);
169     }
170     return qemu_fopen_ops(bs, &bdrv_read_ops);
171 }
172 
173 
174 /* QEMUFile timer support.
175  * Not in qemu-file.c to not add qemu-timer.c as dependency to qemu-file.c
176  */
177 
178 void timer_put(QEMUFile *f, QEMUTimer *ts)
179 {
180     uint64_t expire_time;
181 
182     expire_time = timer_expire_time_ns(ts);
183     qemu_put_be64(f, expire_time);
184 }
185 
186 void timer_get(QEMUFile *f, QEMUTimer *ts)
187 {
188     uint64_t expire_time;
189 
190     expire_time = qemu_get_be64(f);
191     if (expire_time != -1) {
192         timer_mod_ns(ts, expire_time);
193     } else {
194         timer_del(ts);
195     }
196 }
197 
198 
199 /* VMState timer support.
200  * Not in vmstate.c to not add qemu-timer.c as dependency to vmstate.c
201  */
202 
203 static int get_timer(QEMUFile *f, void *pv, size_t size,
204                      const VMStateField *field)
205 {
206     QEMUTimer *v = pv;
207     timer_get(f, v);
208     return 0;
209 }
210 
211 static int put_timer(QEMUFile *f, void *pv, size_t size,
212                      const VMStateField *field, JSONWriter *vmdesc)
213 {
214     QEMUTimer *v = pv;
215     timer_put(f, v);
216 
217     return 0;
218 }
219 
220 const VMStateInfo vmstate_info_timer = {
221     .name = "timer",
222     .get  = get_timer,
223     .put  = put_timer,
224 };
225 
226 
227 typedef struct CompatEntry {
228     char idstr[256];
229     int instance_id;
230 } CompatEntry;
231 
232 typedef struct SaveStateEntry {
233     QTAILQ_ENTRY(SaveStateEntry) entry;
234     char idstr[256];
235     uint32_t instance_id;
236     int alias_id;
237     int version_id;
238     /* version id read from the stream */
239     int load_version_id;
240     int section_id;
241     /* section id read from the stream */
242     int load_section_id;
243     const SaveVMHandlers *ops;
244     const VMStateDescription *vmsd;
245     void *opaque;
246     CompatEntry *compat;
247     int is_ram;
248 } SaveStateEntry;
249 
250 typedef struct SaveState {
251     QTAILQ_HEAD(, SaveStateEntry) handlers;
252     SaveStateEntry *handler_pri_head[MIG_PRI_MAX + 1];
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     QemuUUID uuid;
260 } SaveState;
261 
262 static SaveState savevm_state = {
263     .handlers = QTAILQ_HEAD_INITIALIZER(savevm_state.handlers),
264     .handler_pri_head = { [MIG_PRI_DEFAULT ... MIG_PRI_MAX] = NULL },
265     .global_section_id = 0,
266 };
267 
268 static bool should_validate_capability(int capability)
269 {
270     assert(capability >= 0 && capability < MIGRATION_CAPABILITY__MAX);
271     /* Validate only new capabilities to keep compatibility. */
272     switch (capability) {
273     case MIGRATION_CAPABILITY_X_IGNORE_SHARED:
274         return true;
275     default:
276         return false;
277     }
278 }
279 
280 static uint32_t get_validatable_capabilities_count(void)
281 {
282     MigrationState *s = migrate_get_current();
283     uint32_t result = 0;
284     int i;
285     for (i = 0; i < MIGRATION_CAPABILITY__MAX; i++) {
286         if (should_validate_capability(i) && s->enabled_capabilities[i]) {
287             result++;
288         }
289     }
290     return result;
291 }
292 
293 static int configuration_pre_save(void *opaque)
294 {
295     SaveState *state = opaque;
296     const char *current_name = MACHINE_GET_CLASS(current_machine)->name;
297     MigrationState *s = migrate_get_current();
298     int i, j;
299 
300     state->len = strlen(current_name);
301     state->name = current_name;
302     state->target_page_bits = qemu_target_page_bits();
303 
304     state->caps_count = get_validatable_capabilities_count();
305     state->capabilities = g_renew(MigrationCapability, state->capabilities,
306                                   state->caps_count);
307     for (i = j = 0; i < MIGRATION_CAPABILITY__MAX; i++) {
308         if (should_validate_capability(i) && s->enabled_capabilities[i]) {
309             state->capabilities[j++] = i;
310         }
311     }
312     state->uuid = qemu_uuid;
313 
314     return 0;
315 }
316 
317 static int configuration_pre_load(void *opaque)
318 {
319     SaveState *state = opaque;
320 
321     /* If there is no target-page-bits subsection it means the source
322      * predates the variable-target-page-bits support and is using the
323      * minimum possible value for this CPU.
324      */
325     state->target_page_bits = qemu_target_page_bits_min();
326     return 0;
327 }
328 
329 static bool configuration_validate_capabilities(SaveState *state)
330 {
331     bool ret = true;
332     MigrationState *s = migrate_get_current();
333     unsigned long *source_caps_bm;
334     int i;
335 
336     source_caps_bm = bitmap_new(MIGRATION_CAPABILITY__MAX);
337     for (i = 0; i < state->caps_count; i++) {
338         MigrationCapability capability = state->capabilities[i];
339         set_bit(capability, source_caps_bm);
340     }
341 
342     for (i = 0; i < MIGRATION_CAPABILITY__MAX; i++) {
343         bool source_state, target_state;
344         if (!should_validate_capability(i)) {
345             continue;
346         }
347         source_state = test_bit(i, source_caps_bm);
348         target_state = s->enabled_capabilities[i];
349         if (source_state != target_state) {
350             error_report("Capability %s is %s, but received capability is %s",
351                          MigrationCapability_str(i),
352                          target_state ? "on" : "off",
353                          source_state ? "on" : "off");
354             ret = false;
355             /* Don't break here to report all failed capabilities */
356         }
357     }
358 
359     g_free(source_caps_bm);
360     return ret;
361 }
362 
363 static int configuration_post_load(void *opaque, int version_id)
364 {
365     SaveState *state = opaque;
366     const char *current_name = MACHINE_GET_CLASS(current_machine)->name;
367 
368     if (strncmp(state->name, current_name, state->len) != 0) {
369         error_report("Machine type received is '%.*s' and local is '%s'",
370                      (int) state->len, state->name, current_name);
371         return -EINVAL;
372     }
373 
374     if (state->target_page_bits != qemu_target_page_bits()) {
375         error_report("Received TARGET_PAGE_BITS is %d but local is %d",
376                      state->target_page_bits, qemu_target_page_bits());
377         return -EINVAL;
378     }
379 
380     if (!configuration_validate_capabilities(state)) {
381         return -EINVAL;
382     }
383 
384     return 0;
385 }
386 
387 static int get_capability(QEMUFile *f, void *pv, size_t size,
388                           const VMStateField *field)
389 {
390     MigrationCapability *capability = pv;
391     char capability_str[UINT8_MAX + 1];
392     uint8_t len;
393     int i;
394 
395     len = qemu_get_byte(f);
396     qemu_get_buffer(f, (uint8_t *)capability_str, len);
397     capability_str[len] = '\0';
398     for (i = 0; i < MIGRATION_CAPABILITY__MAX; i++) {
399         if (!strcmp(MigrationCapability_str(i), capability_str)) {
400             *capability = i;
401             return 0;
402         }
403     }
404     error_report("Received unknown capability %s", capability_str);
405     return -EINVAL;
406 }
407 
408 static int put_capability(QEMUFile *f, void *pv, size_t size,
409                           const VMStateField *field, JSONWriter *vmdesc)
410 {
411     MigrationCapability *capability = pv;
412     const char *capability_str = MigrationCapability_str(*capability);
413     size_t len = strlen(capability_str);
414     assert(len <= UINT8_MAX);
415 
416     qemu_put_byte(f, len);
417     qemu_put_buffer(f, (uint8_t *)capability_str, len);
418     return 0;
419 }
420 
421 static const VMStateInfo vmstate_info_capability = {
422     .name = "capability",
423     .get  = get_capability,
424     .put  = put_capability,
425 };
426 
427 /* The target-page-bits subsection is present only if the
428  * target page size is not the same as the default (ie the
429  * minimum page size for a variable-page-size guest CPU).
430  * If it is present then it contains the actual target page
431  * bits for the machine, and migration will fail if the
432  * two ends don't agree about it.
433  */
434 static bool vmstate_target_page_bits_needed(void *opaque)
435 {
436     return qemu_target_page_bits()
437         > qemu_target_page_bits_min();
438 }
439 
440 static const VMStateDescription vmstate_target_page_bits = {
441     .name = "configuration/target-page-bits",
442     .version_id = 1,
443     .minimum_version_id = 1,
444     .needed = vmstate_target_page_bits_needed,
445     .fields = (VMStateField[]) {
446         VMSTATE_UINT32(target_page_bits, SaveState),
447         VMSTATE_END_OF_LIST()
448     }
449 };
450 
451 static bool vmstate_capabilites_needed(void *opaque)
452 {
453     return get_validatable_capabilities_count() > 0;
454 }
455 
456 static const VMStateDescription vmstate_capabilites = {
457     .name = "configuration/capabilities",
458     .version_id = 1,
459     .minimum_version_id = 1,
460     .needed = vmstate_capabilites_needed,
461     .fields = (VMStateField[]) {
462         VMSTATE_UINT32_V(caps_count, SaveState, 1),
463         VMSTATE_VARRAY_UINT32_ALLOC(capabilities, SaveState, caps_count, 1,
464                                     vmstate_info_capability,
465                                     MigrationCapability),
466         VMSTATE_END_OF_LIST()
467     }
468 };
469 
470 static bool vmstate_uuid_needed(void *opaque)
471 {
472     return qemu_uuid_set && migrate_validate_uuid();
473 }
474 
475 static int vmstate_uuid_post_load(void *opaque, int version_id)
476 {
477     SaveState *state = opaque;
478     char uuid_src[UUID_FMT_LEN + 1];
479     char uuid_dst[UUID_FMT_LEN + 1];
480 
481     if (!qemu_uuid_set) {
482         /*
483          * It's warning because user might not know UUID in some cases,
484          * e.g. load an old snapshot
485          */
486         qemu_uuid_unparse(&state->uuid, uuid_src);
487         warn_report("UUID is received %s, but local uuid isn't set",
488                      uuid_src);
489         return 0;
490     }
491     if (!qemu_uuid_is_equal(&state->uuid, &qemu_uuid)) {
492         qemu_uuid_unparse(&state->uuid, uuid_src);
493         qemu_uuid_unparse(&qemu_uuid, uuid_dst);
494         error_report("UUID received is %s and local is %s", uuid_src, uuid_dst);
495         return -EINVAL;
496     }
497     return 0;
498 }
499 
500 static const VMStateDescription vmstate_uuid = {
501     .name = "configuration/uuid",
502     .version_id = 1,
503     .minimum_version_id = 1,
504     .needed = vmstate_uuid_needed,
505     .post_load = vmstate_uuid_post_load,
506     .fields = (VMStateField[]) {
507         VMSTATE_UINT8_ARRAY_V(uuid.data, SaveState, sizeof(QemuUUID), 1),
508         VMSTATE_END_OF_LIST()
509     }
510 };
511 
512 static const VMStateDescription vmstate_configuration = {
513     .name = "configuration",
514     .version_id = 1,
515     .pre_load = configuration_pre_load,
516     .post_load = configuration_post_load,
517     .pre_save = configuration_pre_save,
518     .fields = (VMStateField[]) {
519         VMSTATE_UINT32(len, SaveState),
520         VMSTATE_VBUFFER_ALLOC_UINT32(name, SaveState, 0, NULL, len),
521         VMSTATE_END_OF_LIST()
522     },
523     .subsections = (const VMStateDescription *[]) {
524         &vmstate_target_page_bits,
525         &vmstate_capabilites,
526         &vmstate_uuid,
527         NULL
528     }
529 };
530 
531 static void dump_vmstate_vmsd(FILE *out_file,
532                               const VMStateDescription *vmsd, int indent,
533                               bool is_subsection);
534 
535 static void dump_vmstate_vmsf(FILE *out_file, const VMStateField *field,
536                               int indent)
537 {
538     fprintf(out_file, "%*s{\n", indent, "");
539     indent += 2;
540     fprintf(out_file, "%*s\"field\": \"%s\",\n", indent, "", field->name);
541     fprintf(out_file, "%*s\"version_id\": %d,\n", indent, "",
542             field->version_id);
543     fprintf(out_file, "%*s\"field_exists\": %s,\n", indent, "",
544             field->field_exists ? "true" : "false");
545     fprintf(out_file, "%*s\"size\": %zu", indent, "", field->size);
546     if (field->vmsd != NULL) {
547         fprintf(out_file, ",\n");
548         dump_vmstate_vmsd(out_file, field->vmsd, indent, false);
549     }
550     fprintf(out_file, "\n%*s}", indent - 2, "");
551 }
552 
553 static void dump_vmstate_vmss(FILE *out_file,
554                               const VMStateDescription **subsection,
555                               int indent)
556 {
557     if (*subsection != NULL) {
558         dump_vmstate_vmsd(out_file, *subsection, indent, true);
559     }
560 }
561 
562 static void dump_vmstate_vmsd(FILE *out_file,
563                               const VMStateDescription *vmsd, int indent,
564                               bool is_subsection)
565 {
566     if (is_subsection) {
567         fprintf(out_file, "%*s{\n", indent, "");
568     } else {
569         fprintf(out_file, "%*s\"%s\": {\n", indent, "", "Description");
570     }
571     indent += 2;
572     fprintf(out_file, "%*s\"name\": \"%s\",\n", indent, "", vmsd->name);
573     fprintf(out_file, "%*s\"version_id\": %d,\n", indent, "",
574             vmsd->version_id);
575     fprintf(out_file, "%*s\"minimum_version_id\": %d", indent, "",
576             vmsd->minimum_version_id);
577     if (vmsd->fields != NULL) {
578         const VMStateField *field = vmsd->fields;
579         bool first;
580 
581         fprintf(out_file, ",\n%*s\"Fields\": [\n", indent, "");
582         first = true;
583         while (field->name != NULL) {
584             if (field->flags & VMS_MUST_EXIST) {
585                 /* Ignore VMSTATE_VALIDATE bits; these don't get migrated */
586                 field++;
587                 continue;
588             }
589             if (!first) {
590                 fprintf(out_file, ",\n");
591             }
592             dump_vmstate_vmsf(out_file, field, indent + 2);
593             field++;
594             first = false;
595         }
596         fprintf(out_file, "\n%*s]", indent, "");
597     }
598     if (vmsd->subsections != NULL) {
599         const VMStateDescription **subsection = vmsd->subsections;
600         bool first;
601 
602         fprintf(out_file, ",\n%*s\"Subsections\": [\n", indent, "");
603         first = true;
604         while (*subsection != NULL) {
605             if (!first) {
606                 fprintf(out_file, ",\n");
607             }
608             dump_vmstate_vmss(out_file, subsection, indent + 2);
609             subsection++;
610             first = false;
611         }
612         fprintf(out_file, "\n%*s]", indent, "");
613     }
614     fprintf(out_file, "\n%*s}", indent - 2, "");
615 }
616 
617 static void dump_machine_type(FILE *out_file)
618 {
619     MachineClass *mc;
620 
621     mc = MACHINE_GET_CLASS(current_machine);
622 
623     fprintf(out_file, "  \"vmschkmachine\": {\n");
624     fprintf(out_file, "    \"Name\": \"%s\"\n", mc->name);
625     fprintf(out_file, "  },\n");
626 }
627 
628 void dump_vmstate_json_to_file(FILE *out_file)
629 {
630     GSList *list, *elt;
631     bool first;
632 
633     fprintf(out_file, "{\n");
634     dump_machine_type(out_file);
635 
636     first = true;
637     list = object_class_get_list(TYPE_DEVICE, true);
638     for (elt = list; elt; elt = elt->next) {
639         DeviceClass *dc = OBJECT_CLASS_CHECK(DeviceClass, elt->data,
640                                              TYPE_DEVICE);
641         const char *name;
642         int indent = 2;
643 
644         if (!dc->vmsd) {
645             continue;
646         }
647 
648         if (!first) {
649             fprintf(out_file, ",\n");
650         }
651         name = object_class_get_name(OBJECT_CLASS(dc));
652         fprintf(out_file, "%*s\"%s\": {\n", indent, "", name);
653         indent += 2;
654         fprintf(out_file, "%*s\"Name\": \"%s\",\n", indent, "", name);
655         fprintf(out_file, "%*s\"version_id\": %d,\n", indent, "",
656                 dc->vmsd->version_id);
657         fprintf(out_file, "%*s\"minimum_version_id\": %d,\n", indent, "",
658                 dc->vmsd->minimum_version_id);
659 
660         dump_vmstate_vmsd(out_file, dc->vmsd, indent, false);
661 
662         fprintf(out_file, "\n%*s}", indent - 2, "");
663         first = false;
664     }
665     fprintf(out_file, "\n}\n");
666     fclose(out_file);
667     g_slist_free(list);
668 }
669 
670 static uint32_t calculate_new_instance_id(const char *idstr)
671 {
672     SaveStateEntry *se;
673     uint32_t instance_id = 0;
674 
675     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
676         if (strcmp(idstr, se->idstr) == 0
677             && instance_id <= se->instance_id) {
678             instance_id = se->instance_id + 1;
679         }
680     }
681     /* Make sure we never loop over without being noticed */
682     assert(instance_id != VMSTATE_INSTANCE_ID_ANY);
683     return instance_id;
684 }
685 
686 static int calculate_compat_instance_id(const char *idstr)
687 {
688     SaveStateEntry *se;
689     int instance_id = 0;
690 
691     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
692         if (!se->compat) {
693             continue;
694         }
695 
696         if (strcmp(idstr, se->compat->idstr) == 0
697             && instance_id <= se->compat->instance_id) {
698             instance_id = se->compat->instance_id + 1;
699         }
700     }
701     return instance_id;
702 }
703 
704 static inline MigrationPriority save_state_priority(SaveStateEntry *se)
705 {
706     if (se->vmsd) {
707         return se->vmsd->priority;
708     }
709     return MIG_PRI_DEFAULT;
710 }
711 
712 static void savevm_state_handler_insert(SaveStateEntry *nse)
713 {
714     MigrationPriority priority = save_state_priority(nse);
715     SaveStateEntry *se;
716     int i;
717 
718     assert(priority <= MIG_PRI_MAX);
719 
720     for (i = priority - 1; i >= 0; i--) {
721         se = savevm_state.handler_pri_head[i];
722         if (se != NULL) {
723             assert(save_state_priority(se) < priority);
724             break;
725         }
726     }
727 
728     if (i >= 0) {
729         QTAILQ_INSERT_BEFORE(se, nse, entry);
730     } else {
731         QTAILQ_INSERT_TAIL(&savevm_state.handlers, nse, entry);
732     }
733 
734     if (savevm_state.handler_pri_head[priority] == NULL) {
735         savevm_state.handler_pri_head[priority] = nse;
736     }
737 }
738 
739 static void savevm_state_handler_remove(SaveStateEntry *se)
740 {
741     SaveStateEntry *next;
742     MigrationPriority priority = save_state_priority(se);
743 
744     if (se == savevm_state.handler_pri_head[priority]) {
745         next = QTAILQ_NEXT(se, entry);
746         if (next != NULL && save_state_priority(next) == priority) {
747             savevm_state.handler_pri_head[priority] = next;
748         } else {
749             savevm_state.handler_pri_head[priority] = NULL;
750         }
751     }
752     QTAILQ_REMOVE(&savevm_state.handlers, se, entry);
753 }
754 
755 /* TODO: Individual devices generally have very little idea about the rest
756    of the system, so instance_id should be removed/replaced.
757    Meanwhile pass -1 as instance_id if you do not already have a clearly
758    distinguishing id for all instances of your device class. */
759 int register_savevm_live(const char *idstr,
760                          uint32_t instance_id,
761                          int version_id,
762                          const SaveVMHandlers *ops,
763                          void *opaque)
764 {
765     SaveStateEntry *se;
766 
767     se = g_new0(SaveStateEntry, 1);
768     se->version_id = version_id;
769     se->section_id = savevm_state.global_section_id++;
770     se->ops = ops;
771     se->opaque = opaque;
772     se->vmsd = NULL;
773     /* if this is a live_savem then set is_ram */
774     if (ops->save_setup != NULL) {
775         se->is_ram = 1;
776     }
777 
778     pstrcat(se->idstr, sizeof(se->idstr), idstr);
779 
780     if (instance_id == VMSTATE_INSTANCE_ID_ANY) {
781         se->instance_id = calculate_new_instance_id(se->idstr);
782     } else {
783         se->instance_id = instance_id;
784     }
785     assert(!se->compat || se->instance_id == 0);
786     savevm_state_handler_insert(se);
787     return 0;
788 }
789 
790 void unregister_savevm(VMStateIf *obj, const char *idstr, void *opaque)
791 {
792     SaveStateEntry *se, *new_se;
793     char id[256] = "";
794 
795     if (obj) {
796         char *oid = vmstate_if_get_id(obj);
797         if (oid) {
798             pstrcpy(id, sizeof(id), oid);
799             pstrcat(id, sizeof(id), "/");
800             g_free(oid);
801         }
802     }
803     pstrcat(id, sizeof(id), idstr);
804 
805     QTAILQ_FOREACH_SAFE(se, &savevm_state.handlers, entry, new_se) {
806         if (strcmp(se->idstr, id) == 0 && se->opaque == opaque) {
807             savevm_state_handler_remove(se);
808             g_free(se->compat);
809             g_free(se);
810         }
811     }
812 }
813 
814 int vmstate_register_with_alias_id(VMStateIf *obj, uint32_t instance_id,
815                                    const VMStateDescription *vmsd,
816                                    void *opaque, int alias_id,
817                                    int required_for_version,
818                                    Error **errp)
819 {
820     SaveStateEntry *se;
821 
822     /* If this triggers, alias support can be dropped for the vmsd. */
823     assert(alias_id == -1 || required_for_version >= vmsd->minimum_version_id);
824 
825     se = g_new0(SaveStateEntry, 1);
826     se->version_id = vmsd->version_id;
827     se->section_id = savevm_state.global_section_id++;
828     se->opaque = opaque;
829     se->vmsd = vmsd;
830     se->alias_id = alias_id;
831 
832     if (obj) {
833         char *id = vmstate_if_get_id(obj);
834         if (id) {
835             if (snprintf(se->idstr, sizeof(se->idstr), "%s/", id) >=
836                 sizeof(se->idstr)) {
837                 error_setg(errp, "Path too long for VMState (%s)", id);
838                 g_free(id);
839                 g_free(se);
840 
841                 return -1;
842             }
843             g_free(id);
844 
845             se->compat = g_new0(CompatEntry, 1);
846             pstrcpy(se->compat->idstr, sizeof(se->compat->idstr), vmsd->name);
847             se->compat->instance_id = instance_id == VMSTATE_INSTANCE_ID_ANY ?
848                          calculate_compat_instance_id(vmsd->name) : instance_id;
849             instance_id = VMSTATE_INSTANCE_ID_ANY;
850         }
851     }
852     pstrcat(se->idstr, sizeof(se->idstr), vmsd->name);
853 
854     if (instance_id == VMSTATE_INSTANCE_ID_ANY) {
855         se->instance_id = calculate_new_instance_id(se->idstr);
856     } else {
857         se->instance_id = instance_id;
858     }
859     assert(!se->compat || se->instance_id == 0);
860     savevm_state_handler_insert(se);
861     return 0;
862 }
863 
864 void vmstate_unregister(VMStateIf *obj, const VMStateDescription *vmsd,
865                         void *opaque)
866 {
867     SaveStateEntry *se, *new_se;
868 
869     QTAILQ_FOREACH_SAFE(se, &savevm_state.handlers, entry, new_se) {
870         if (se->vmsd == vmsd && se->opaque == opaque) {
871             savevm_state_handler_remove(se);
872             g_free(se->compat);
873             g_free(se);
874         }
875     }
876 }
877 
878 static int vmstate_load(QEMUFile *f, SaveStateEntry *se)
879 {
880     trace_vmstate_load(se->idstr, se->vmsd ? se->vmsd->name : "(old)");
881     if (!se->vmsd) {         /* Old style */
882         return se->ops->load_state(f, se->opaque, se->load_version_id);
883     }
884     return vmstate_load_state(f, se->vmsd, se->opaque, se->load_version_id);
885 }
886 
887 static void vmstate_save_old_style(QEMUFile *f, SaveStateEntry *se,
888                                    JSONWriter *vmdesc)
889 {
890     int64_t old_offset, size;
891 
892     old_offset = qemu_ftell_fast(f);
893     se->ops->save_state(f, se->opaque);
894     size = qemu_ftell_fast(f) - old_offset;
895 
896     if (vmdesc) {
897         json_writer_int64(vmdesc, "size", size);
898         json_writer_start_array(vmdesc, "fields");
899         json_writer_start_object(vmdesc, NULL);
900         json_writer_str(vmdesc, "name", "data");
901         json_writer_int64(vmdesc, "size", size);
902         json_writer_str(vmdesc, "type", "buffer");
903         json_writer_end_object(vmdesc);
904         json_writer_end_array(vmdesc);
905     }
906 }
907 
908 static int vmstate_save(QEMUFile *f, SaveStateEntry *se,
909                         JSONWriter *vmdesc)
910 {
911     trace_vmstate_save(se->idstr, se->vmsd ? se->vmsd->name : "(old)");
912     if (!se->vmsd) {
913         vmstate_save_old_style(f, se, vmdesc);
914         return 0;
915     }
916     return vmstate_save_state(f, se->vmsd, se->opaque, vmdesc);
917 }
918 
919 /*
920  * Write the header for device section (QEMU_VM_SECTION START/END/PART/FULL)
921  */
922 static void save_section_header(QEMUFile *f, SaveStateEntry *se,
923                                 uint8_t section_type)
924 {
925     qemu_put_byte(f, section_type);
926     qemu_put_be32(f, se->section_id);
927 
928     if (section_type == QEMU_VM_SECTION_FULL ||
929         section_type == QEMU_VM_SECTION_START) {
930         /* ID string */
931         size_t len = strlen(se->idstr);
932         qemu_put_byte(f, len);
933         qemu_put_buffer(f, (uint8_t *)se->idstr, len);
934 
935         qemu_put_be32(f, se->instance_id);
936         qemu_put_be32(f, se->version_id);
937     }
938 }
939 
940 /*
941  * Write a footer onto device sections that catches cases misformatted device
942  * sections.
943  */
944 static void save_section_footer(QEMUFile *f, SaveStateEntry *se)
945 {
946     if (migrate_get_current()->send_section_footer) {
947         qemu_put_byte(f, QEMU_VM_SECTION_FOOTER);
948         qemu_put_be32(f, se->section_id);
949     }
950 }
951 
952 /**
953  * qemu_savevm_command_send: Send a 'QEMU_VM_COMMAND' type element with the
954  *                           command and associated data.
955  *
956  * @f: File to send command on
957  * @command: Command type to send
958  * @len: Length of associated data
959  * @data: Data associated with command.
960  */
961 static void qemu_savevm_command_send(QEMUFile *f,
962                                      enum qemu_vm_cmd command,
963                                      uint16_t len,
964                                      uint8_t *data)
965 {
966     trace_savevm_command_send(command, len);
967     qemu_put_byte(f, QEMU_VM_COMMAND);
968     qemu_put_be16(f, (uint16_t)command);
969     qemu_put_be16(f, len);
970     qemu_put_buffer(f, data, len);
971     qemu_fflush(f);
972 }
973 
974 void qemu_savevm_send_colo_enable(QEMUFile *f)
975 {
976     trace_savevm_send_colo_enable();
977     qemu_savevm_command_send(f, MIG_CMD_ENABLE_COLO, 0, NULL);
978 }
979 
980 void qemu_savevm_send_ping(QEMUFile *f, uint32_t value)
981 {
982     uint32_t buf;
983 
984     trace_savevm_send_ping(value);
985     buf = cpu_to_be32(value);
986     qemu_savevm_command_send(f, MIG_CMD_PING, sizeof(value), (uint8_t *)&buf);
987 }
988 
989 void qemu_savevm_send_open_return_path(QEMUFile *f)
990 {
991     trace_savevm_send_open_return_path();
992     qemu_savevm_command_send(f, MIG_CMD_OPEN_RETURN_PATH, 0, NULL);
993 }
994 
995 /* We have a buffer of data to send; we don't want that all to be loaded
996  * by the command itself, so the command contains just the length of the
997  * extra buffer that we then send straight after it.
998  * TODO: Must be a better way to organise that
999  *
1000  * Returns:
1001  *    0 on success
1002  *    -ve on error
1003  */
1004 int qemu_savevm_send_packaged(QEMUFile *f, const uint8_t *buf, size_t len)
1005 {
1006     uint32_t tmp;
1007 
1008     if (len > MAX_VM_CMD_PACKAGED_SIZE) {
1009         error_report("%s: Unreasonably large packaged state: %zu",
1010                      __func__, len);
1011         return -1;
1012     }
1013 
1014     tmp = cpu_to_be32(len);
1015 
1016     trace_qemu_savevm_send_packaged();
1017     qemu_savevm_command_send(f, MIG_CMD_PACKAGED, 4, (uint8_t *)&tmp);
1018 
1019     qemu_put_buffer(f, buf, len);
1020 
1021     return 0;
1022 }
1023 
1024 /* Send prior to any postcopy transfer */
1025 void qemu_savevm_send_postcopy_advise(QEMUFile *f)
1026 {
1027     if (migrate_postcopy_ram()) {
1028         uint64_t tmp[2];
1029         tmp[0] = cpu_to_be64(ram_pagesize_summary());
1030         tmp[1] = cpu_to_be64(qemu_target_page_size());
1031 
1032         trace_qemu_savevm_send_postcopy_advise();
1033         qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_ADVISE,
1034                                  16, (uint8_t *)tmp);
1035     } else {
1036         qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_ADVISE, 0, NULL);
1037     }
1038 }
1039 
1040 /* Sent prior to starting the destination running in postcopy, discard pages
1041  * that have already been sent but redirtied on the source.
1042  * CMD_POSTCOPY_RAM_DISCARD consist of:
1043  *      byte   version (0)
1044  *      byte   Length of name field (not including 0)
1045  *  n x byte   RAM block name
1046  *      byte   0 terminator (just for safety)
1047  *  n x        Byte ranges within the named RAMBlock
1048  *      be64   Start of the range
1049  *      be64   Length
1050  *
1051  *  name:  RAMBlock name that these entries are part of
1052  *  len: Number of page entries
1053  *  start_list: 'len' addresses
1054  *  length_list: 'len' addresses
1055  *
1056  */
1057 void qemu_savevm_send_postcopy_ram_discard(QEMUFile *f, const char *name,
1058                                            uint16_t len,
1059                                            uint64_t *start_list,
1060                                            uint64_t *length_list)
1061 {
1062     uint8_t *buf;
1063     uint16_t tmplen;
1064     uint16_t t;
1065     size_t name_len = strlen(name);
1066 
1067     trace_qemu_savevm_send_postcopy_ram_discard(name, len);
1068     assert(name_len < 256);
1069     buf = g_malloc0(1 + 1 + name_len + 1 + (8 + 8) * len);
1070     buf[0] = postcopy_ram_discard_version;
1071     buf[1] = name_len;
1072     memcpy(buf + 2, name, name_len);
1073     tmplen = 2 + name_len;
1074     buf[tmplen++] = '\0';
1075 
1076     for (t = 0; t < len; t++) {
1077         stq_be_p(buf + tmplen, start_list[t]);
1078         tmplen += 8;
1079         stq_be_p(buf + tmplen, length_list[t]);
1080         tmplen += 8;
1081     }
1082     qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_RAM_DISCARD, tmplen, buf);
1083     g_free(buf);
1084 }
1085 
1086 /* Get the destination into a state where it can receive postcopy data. */
1087 void qemu_savevm_send_postcopy_listen(QEMUFile *f)
1088 {
1089     trace_savevm_send_postcopy_listen();
1090     qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_LISTEN, 0, NULL);
1091 }
1092 
1093 /* Kick the destination into running */
1094 void qemu_savevm_send_postcopy_run(QEMUFile *f)
1095 {
1096     trace_savevm_send_postcopy_run();
1097     qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_RUN, 0, NULL);
1098 }
1099 
1100 void qemu_savevm_send_postcopy_resume(QEMUFile *f)
1101 {
1102     trace_savevm_send_postcopy_resume();
1103     qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_RESUME, 0, NULL);
1104 }
1105 
1106 void qemu_savevm_send_recv_bitmap(QEMUFile *f, char *block_name)
1107 {
1108     size_t len;
1109     char buf[256];
1110 
1111     trace_savevm_send_recv_bitmap(block_name);
1112 
1113     buf[0] = len = strlen(block_name);
1114     memcpy(buf + 1, block_name, len);
1115 
1116     qemu_savevm_command_send(f, MIG_CMD_RECV_BITMAP, len + 1, (uint8_t *)buf);
1117 }
1118 
1119 bool qemu_savevm_state_blocked(Error **errp)
1120 {
1121     SaveStateEntry *se;
1122 
1123     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1124         if (se->vmsd && se->vmsd->unmigratable) {
1125             error_setg(errp, "State blocked by non-migratable device '%s'",
1126                        se->idstr);
1127             return true;
1128         }
1129     }
1130     return false;
1131 }
1132 
1133 void qemu_savevm_state_header(QEMUFile *f)
1134 {
1135     trace_savevm_state_header();
1136     qemu_put_be32(f, QEMU_VM_FILE_MAGIC);
1137     qemu_put_be32(f, QEMU_VM_FILE_VERSION);
1138 
1139     if (migrate_get_current()->send_configuration) {
1140         qemu_put_byte(f, QEMU_VM_CONFIGURATION);
1141         vmstate_save_state(f, &vmstate_configuration, &savevm_state, 0);
1142     }
1143 }
1144 
1145 bool qemu_savevm_state_guest_unplug_pending(void)
1146 {
1147     SaveStateEntry *se;
1148 
1149     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1150         if (se->vmsd && se->vmsd->dev_unplug_pending &&
1151             se->vmsd->dev_unplug_pending(se->opaque)) {
1152             return true;
1153         }
1154     }
1155 
1156     return false;
1157 }
1158 
1159 void qemu_savevm_state_setup(QEMUFile *f)
1160 {
1161     SaveStateEntry *se;
1162     Error *local_err = NULL;
1163     int ret;
1164 
1165     trace_savevm_state_setup();
1166     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1167         if (!se->ops || !se->ops->save_setup) {
1168             continue;
1169         }
1170         if (se->ops->is_active) {
1171             if (!se->ops->is_active(se->opaque)) {
1172                 continue;
1173             }
1174         }
1175         save_section_header(f, se, QEMU_VM_SECTION_START);
1176 
1177         ret = se->ops->save_setup(f, se->opaque);
1178         save_section_footer(f, se);
1179         if (ret < 0) {
1180             qemu_file_set_error(f, ret);
1181             break;
1182         }
1183     }
1184 
1185     if (precopy_notify(PRECOPY_NOTIFY_SETUP, &local_err)) {
1186         error_report_err(local_err);
1187     }
1188 }
1189 
1190 int qemu_savevm_state_resume_prepare(MigrationState *s)
1191 {
1192     SaveStateEntry *se;
1193     int ret;
1194 
1195     trace_savevm_state_resume_prepare();
1196 
1197     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1198         if (!se->ops || !se->ops->resume_prepare) {
1199             continue;
1200         }
1201         if (se->ops->is_active) {
1202             if (!se->ops->is_active(se->opaque)) {
1203                 continue;
1204             }
1205         }
1206         ret = se->ops->resume_prepare(s, se->opaque);
1207         if (ret < 0) {
1208             return ret;
1209         }
1210     }
1211 
1212     return 0;
1213 }
1214 
1215 /*
1216  * this function has three return values:
1217  *   negative: there was one error, and we have -errno.
1218  *   0 : We haven't finished, caller have to go again
1219  *   1 : We have finished, we can go to complete phase
1220  */
1221 int qemu_savevm_state_iterate(QEMUFile *f, bool postcopy)
1222 {
1223     SaveStateEntry *se;
1224     int ret = 1;
1225 
1226     trace_savevm_state_iterate();
1227     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1228         if (!se->ops || !se->ops->save_live_iterate) {
1229             continue;
1230         }
1231         if (se->ops->is_active &&
1232             !se->ops->is_active(se->opaque)) {
1233             continue;
1234         }
1235         if (se->ops->is_active_iterate &&
1236             !se->ops->is_active_iterate(se->opaque)) {
1237             continue;
1238         }
1239         /*
1240          * In the postcopy phase, any device that doesn't know how to
1241          * do postcopy should have saved it's state in the _complete
1242          * call that's already run, it might get confused if we call
1243          * iterate afterwards.
1244          */
1245         if (postcopy &&
1246             !(se->ops->has_postcopy && se->ops->has_postcopy(se->opaque))) {
1247             continue;
1248         }
1249         if (qemu_file_rate_limit(f)) {
1250             return 0;
1251         }
1252         trace_savevm_section_start(se->idstr, se->section_id);
1253 
1254         save_section_header(f, se, QEMU_VM_SECTION_PART);
1255 
1256         ret = se->ops->save_live_iterate(f, se->opaque);
1257         trace_savevm_section_end(se->idstr, se->section_id, ret);
1258         save_section_footer(f, se);
1259 
1260         if (ret < 0) {
1261             error_report("failed to save SaveStateEntry with id(name): %d(%s)",
1262                          se->section_id, se->idstr);
1263             qemu_file_set_error(f, ret);
1264         }
1265         if (ret <= 0) {
1266             /* Do not proceed to the next vmstate before this one reported
1267                completion of the current stage. This serializes the migration
1268                and reduces the probability that a faster changing state is
1269                synchronized over and over again. */
1270             break;
1271         }
1272     }
1273     return ret;
1274 }
1275 
1276 static bool should_send_vmdesc(void)
1277 {
1278     MachineState *machine = MACHINE(qdev_get_machine());
1279     bool in_postcopy = migration_in_postcopy();
1280     return !machine->suppress_vmdesc && !in_postcopy;
1281 }
1282 
1283 /*
1284  * Calls the save_live_complete_postcopy methods
1285  * causing the last few pages to be sent immediately and doing any associated
1286  * cleanup.
1287  * Note postcopy also calls qemu_savevm_state_complete_precopy to complete
1288  * all the other devices, but that happens at the point we switch to postcopy.
1289  */
1290 void qemu_savevm_state_complete_postcopy(QEMUFile *f)
1291 {
1292     SaveStateEntry *se;
1293     int ret;
1294 
1295     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1296         if (!se->ops || !se->ops->save_live_complete_postcopy) {
1297             continue;
1298         }
1299         if (se->ops->is_active) {
1300             if (!se->ops->is_active(se->opaque)) {
1301                 continue;
1302             }
1303         }
1304         trace_savevm_section_start(se->idstr, se->section_id);
1305         /* Section type */
1306         qemu_put_byte(f, QEMU_VM_SECTION_END);
1307         qemu_put_be32(f, se->section_id);
1308 
1309         ret = se->ops->save_live_complete_postcopy(f, se->opaque);
1310         trace_savevm_section_end(se->idstr, se->section_id, ret);
1311         save_section_footer(f, se);
1312         if (ret < 0) {
1313             qemu_file_set_error(f, ret);
1314             return;
1315         }
1316     }
1317 
1318     qemu_put_byte(f, QEMU_VM_EOF);
1319     qemu_fflush(f);
1320 }
1321 
1322 static
1323 int qemu_savevm_state_complete_precopy_iterable(QEMUFile *f, bool in_postcopy)
1324 {
1325     SaveStateEntry *se;
1326     int ret;
1327 
1328     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1329         if (!se->ops ||
1330             (in_postcopy && se->ops->has_postcopy &&
1331              se->ops->has_postcopy(se->opaque)) ||
1332             !se->ops->save_live_complete_precopy) {
1333             continue;
1334         }
1335 
1336         if (se->ops->is_active) {
1337             if (!se->ops->is_active(se->opaque)) {
1338                 continue;
1339             }
1340         }
1341         trace_savevm_section_start(se->idstr, se->section_id);
1342 
1343         save_section_header(f, se, QEMU_VM_SECTION_END);
1344 
1345         ret = se->ops->save_live_complete_precopy(f, se->opaque);
1346         trace_savevm_section_end(se->idstr, se->section_id, ret);
1347         save_section_footer(f, se);
1348         if (ret < 0) {
1349             qemu_file_set_error(f, ret);
1350             return -1;
1351         }
1352     }
1353 
1354     return 0;
1355 }
1356 
1357 static
1358 int qemu_savevm_state_complete_precopy_non_iterable(QEMUFile *f,
1359                                                     bool in_postcopy,
1360                                                     bool inactivate_disks)
1361 {
1362     g_autoptr(JSONWriter) vmdesc = NULL;
1363     int vmdesc_len;
1364     SaveStateEntry *se;
1365     int ret;
1366 
1367     vmdesc = json_writer_new(false);
1368     json_writer_start_object(vmdesc, NULL);
1369     json_writer_int64(vmdesc, "page_size", qemu_target_page_size());
1370     json_writer_start_array(vmdesc, "devices");
1371     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1372 
1373         if ((!se->ops || !se->ops->save_state) && !se->vmsd) {
1374             continue;
1375         }
1376         if (se->vmsd && !vmstate_save_needed(se->vmsd, se->opaque)) {
1377             trace_savevm_section_skip(se->idstr, se->section_id);
1378             continue;
1379         }
1380 
1381         trace_savevm_section_start(se->idstr, se->section_id);
1382 
1383         json_writer_start_object(vmdesc, NULL);
1384         json_writer_str(vmdesc, "name", se->idstr);
1385         json_writer_int64(vmdesc, "instance_id", se->instance_id);
1386 
1387         save_section_header(f, se, QEMU_VM_SECTION_FULL);
1388         ret = vmstate_save(f, se, vmdesc);
1389         if (ret) {
1390             qemu_file_set_error(f, ret);
1391             return ret;
1392         }
1393         trace_savevm_section_end(se->idstr, se->section_id, 0);
1394         save_section_footer(f, se);
1395 
1396         json_writer_end_object(vmdesc);
1397     }
1398 
1399     if (inactivate_disks) {
1400         /* Inactivate before sending QEMU_VM_EOF so that the
1401          * bdrv_invalidate_cache_all() on the other end won't fail. */
1402         ret = bdrv_inactivate_all();
1403         if (ret) {
1404             error_report("%s: bdrv_inactivate_all() failed (%d)",
1405                          __func__, ret);
1406             qemu_file_set_error(f, ret);
1407             return ret;
1408         }
1409     }
1410     if (!in_postcopy) {
1411         /* Postcopy stream will still be going */
1412         qemu_put_byte(f, QEMU_VM_EOF);
1413     }
1414 
1415     json_writer_end_array(vmdesc);
1416     json_writer_end_object(vmdesc);
1417     vmdesc_len = strlen(json_writer_get(vmdesc));
1418 
1419     if (should_send_vmdesc()) {
1420         qemu_put_byte(f, QEMU_VM_VMDESCRIPTION);
1421         qemu_put_be32(f, vmdesc_len);
1422         qemu_put_buffer(f, (uint8_t *)json_writer_get(vmdesc), vmdesc_len);
1423     }
1424 
1425     return 0;
1426 }
1427 
1428 int qemu_savevm_state_complete_precopy(QEMUFile *f, bool iterable_only,
1429                                        bool inactivate_disks)
1430 {
1431     int ret;
1432     Error *local_err = NULL;
1433     bool in_postcopy = migration_in_postcopy();
1434 
1435     if (precopy_notify(PRECOPY_NOTIFY_COMPLETE, &local_err)) {
1436         error_report_err(local_err);
1437     }
1438 
1439     trace_savevm_state_complete_precopy();
1440 
1441     cpu_synchronize_all_states();
1442 
1443     if (!in_postcopy || iterable_only) {
1444         ret = qemu_savevm_state_complete_precopy_iterable(f, in_postcopy);
1445         if (ret) {
1446             return ret;
1447         }
1448     }
1449 
1450     if (iterable_only) {
1451         goto flush;
1452     }
1453 
1454     ret = qemu_savevm_state_complete_precopy_non_iterable(f, in_postcopy,
1455                                                           inactivate_disks);
1456     if (ret) {
1457         return ret;
1458     }
1459 
1460 flush:
1461     qemu_fflush(f);
1462     return 0;
1463 }
1464 
1465 /* Give an estimate of the amount left to be transferred,
1466  * the result is split into the amount for units that can and
1467  * for units that can't do postcopy.
1468  */
1469 void qemu_savevm_state_pending(QEMUFile *f, uint64_t threshold_size,
1470                                uint64_t *res_precopy_only,
1471                                uint64_t *res_compatible,
1472                                uint64_t *res_postcopy_only)
1473 {
1474     SaveStateEntry *se;
1475 
1476     *res_precopy_only = 0;
1477     *res_compatible = 0;
1478     *res_postcopy_only = 0;
1479 
1480 
1481     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1482         if (!se->ops || !se->ops->save_live_pending) {
1483             continue;
1484         }
1485         if (se->ops->is_active) {
1486             if (!se->ops->is_active(se->opaque)) {
1487                 continue;
1488             }
1489         }
1490         se->ops->save_live_pending(f, se->opaque, threshold_size,
1491                                    res_precopy_only, res_compatible,
1492                                    res_postcopy_only);
1493     }
1494 }
1495 
1496 void qemu_savevm_state_cleanup(void)
1497 {
1498     SaveStateEntry *se;
1499     Error *local_err = NULL;
1500 
1501     if (precopy_notify(PRECOPY_NOTIFY_CLEANUP, &local_err)) {
1502         error_report_err(local_err);
1503     }
1504 
1505     trace_savevm_state_cleanup();
1506     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1507         if (se->ops && se->ops->save_cleanup) {
1508             se->ops->save_cleanup(se->opaque);
1509         }
1510     }
1511 }
1512 
1513 static int qemu_savevm_state(QEMUFile *f, Error **errp)
1514 {
1515     int ret;
1516     MigrationState *ms = migrate_get_current();
1517     MigrationStatus status;
1518 
1519     if (migration_is_running(ms->state)) {
1520         error_setg(errp, QERR_MIGRATION_ACTIVE);
1521         return -EINVAL;
1522     }
1523 
1524     if (migrate_use_block()) {
1525         error_setg(errp, "Block migration and snapshots are incompatible");
1526         return -EINVAL;
1527     }
1528 
1529     migrate_init(ms);
1530     memset(&ram_counters, 0, sizeof(ram_counters));
1531     ms->to_dst_file = f;
1532 
1533     qemu_mutex_unlock_iothread();
1534     qemu_savevm_state_header(f);
1535     qemu_savevm_state_setup(f);
1536     qemu_mutex_lock_iothread();
1537 
1538     while (qemu_file_get_error(f) == 0) {
1539         if (qemu_savevm_state_iterate(f, false) > 0) {
1540             break;
1541         }
1542     }
1543 
1544     ret = qemu_file_get_error(f);
1545     if (ret == 0) {
1546         qemu_savevm_state_complete_precopy(f, false, false);
1547         ret = qemu_file_get_error(f);
1548     }
1549     qemu_savevm_state_cleanup();
1550     if (ret != 0) {
1551         error_setg_errno(errp, -ret, "Error while writing VM state");
1552     }
1553 
1554     if (ret != 0) {
1555         status = MIGRATION_STATUS_FAILED;
1556     } else {
1557         status = MIGRATION_STATUS_COMPLETED;
1558     }
1559     migrate_set_state(&ms->state, MIGRATION_STATUS_SETUP, status);
1560 
1561     /* f is outer parameter, it should not stay in global migration state after
1562      * this function finished */
1563     ms->to_dst_file = NULL;
1564 
1565     return ret;
1566 }
1567 
1568 void qemu_savevm_live_state(QEMUFile *f)
1569 {
1570     /* save QEMU_VM_SECTION_END section */
1571     qemu_savevm_state_complete_precopy(f, true, false);
1572     qemu_put_byte(f, QEMU_VM_EOF);
1573 }
1574 
1575 int qemu_save_device_state(QEMUFile *f)
1576 {
1577     SaveStateEntry *se;
1578 
1579     if (!migration_in_colo_state()) {
1580         qemu_put_be32(f, QEMU_VM_FILE_MAGIC);
1581         qemu_put_be32(f, QEMU_VM_FILE_VERSION);
1582     }
1583     cpu_synchronize_all_states();
1584 
1585     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1586         int ret;
1587 
1588         if (se->is_ram) {
1589             continue;
1590         }
1591         if ((!se->ops || !se->ops->save_state) && !se->vmsd) {
1592             continue;
1593         }
1594         if (se->vmsd && !vmstate_save_needed(se->vmsd, se->opaque)) {
1595             continue;
1596         }
1597 
1598         save_section_header(f, se, QEMU_VM_SECTION_FULL);
1599 
1600         ret = vmstate_save(f, se, NULL);
1601         if (ret) {
1602             return ret;
1603         }
1604 
1605         save_section_footer(f, se);
1606     }
1607 
1608     qemu_put_byte(f, QEMU_VM_EOF);
1609 
1610     return qemu_file_get_error(f);
1611 }
1612 
1613 static SaveStateEntry *find_se(const char *idstr, uint32_t instance_id)
1614 {
1615     SaveStateEntry *se;
1616 
1617     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1618         if (!strcmp(se->idstr, idstr) &&
1619             (instance_id == se->instance_id ||
1620              instance_id == se->alias_id))
1621             return se;
1622         /* Migrating from an older version? */
1623         if (strstr(se->idstr, idstr) && se->compat) {
1624             if (!strcmp(se->compat->idstr, idstr) &&
1625                 (instance_id == se->compat->instance_id ||
1626                  instance_id == se->alias_id))
1627                 return se;
1628         }
1629     }
1630     return NULL;
1631 }
1632 
1633 enum LoadVMExitCodes {
1634     /* Allow a command to quit all layers of nested loadvm loops */
1635     LOADVM_QUIT     =  1,
1636 };
1637 
1638 /* ------ incoming postcopy messages ------ */
1639 /* 'advise' arrives before any transfers just to tell us that a postcopy
1640  * *might* happen - it might be skipped if precopy transferred everything
1641  * quickly.
1642  */
1643 static int loadvm_postcopy_handle_advise(MigrationIncomingState *mis,
1644                                          uint16_t len)
1645 {
1646     PostcopyState ps = postcopy_state_set(POSTCOPY_INCOMING_ADVISE);
1647     uint64_t remote_pagesize_summary, local_pagesize_summary, remote_tps;
1648     Error *local_err = NULL;
1649 
1650     trace_loadvm_postcopy_handle_advise();
1651     if (ps != POSTCOPY_INCOMING_NONE) {
1652         error_report("CMD_POSTCOPY_ADVISE in wrong postcopy state (%d)", ps);
1653         return -1;
1654     }
1655 
1656     switch (len) {
1657     case 0:
1658         if (migrate_postcopy_ram()) {
1659             error_report("RAM postcopy is enabled but have 0 byte advise");
1660             return -EINVAL;
1661         }
1662         return 0;
1663     case 8 + 8:
1664         if (!migrate_postcopy_ram()) {
1665             error_report("RAM postcopy is disabled but have 16 byte advise");
1666             return -EINVAL;
1667         }
1668         break;
1669     default:
1670         error_report("CMD_POSTCOPY_ADVISE invalid length (%d)", len);
1671         return -EINVAL;
1672     }
1673 
1674     if (!postcopy_ram_supported_by_host(mis)) {
1675         postcopy_state_set(POSTCOPY_INCOMING_NONE);
1676         return -1;
1677     }
1678 
1679     remote_pagesize_summary = qemu_get_be64(mis->from_src_file);
1680     local_pagesize_summary = ram_pagesize_summary();
1681 
1682     if (remote_pagesize_summary != local_pagesize_summary)  {
1683         /*
1684          * This detects two potential causes of mismatch:
1685          *   a) A mismatch in host page sizes
1686          *      Some combinations of mismatch are probably possible but it gets
1687          *      a bit more complicated.  In particular we need to place whole
1688          *      host pages on the dest at once, and we need to ensure that we
1689          *      handle dirtying to make sure we never end up sending part of
1690          *      a hostpage on it's own.
1691          *   b) The use of different huge page sizes on source/destination
1692          *      a more fine grain test is performed during RAM block migration
1693          *      but this test here causes a nice early clear failure, and
1694          *      also fails when passed to an older qemu that doesn't
1695          *      do huge pages.
1696          */
1697         error_report("Postcopy needs matching RAM page sizes (s=%" PRIx64
1698                                                              " d=%" PRIx64 ")",
1699                      remote_pagesize_summary, local_pagesize_summary);
1700         return -1;
1701     }
1702 
1703     remote_tps = qemu_get_be64(mis->from_src_file);
1704     if (remote_tps != qemu_target_page_size()) {
1705         /*
1706          * Again, some differences could be dealt with, but for now keep it
1707          * simple.
1708          */
1709         error_report("Postcopy needs matching target page sizes (s=%d d=%zd)",
1710                      (int)remote_tps, qemu_target_page_size());
1711         return -1;
1712     }
1713 
1714     if (postcopy_notify(POSTCOPY_NOTIFY_INBOUND_ADVISE, &local_err)) {
1715         error_report_err(local_err);
1716         return -1;
1717     }
1718 
1719     if (ram_postcopy_incoming_init(mis)) {
1720         return -1;
1721     }
1722 
1723     return 0;
1724 }
1725 
1726 /* After postcopy we will be told to throw some pages away since they're
1727  * dirty and will have to be demand fetched.  Must happen before CPU is
1728  * started.
1729  * There can be 0..many of these messages, each encoding multiple pages.
1730  */
1731 static int loadvm_postcopy_ram_handle_discard(MigrationIncomingState *mis,
1732                                               uint16_t len)
1733 {
1734     int tmp;
1735     char ramid[256];
1736     PostcopyState ps = postcopy_state_get();
1737 
1738     trace_loadvm_postcopy_ram_handle_discard();
1739 
1740     switch (ps) {
1741     case POSTCOPY_INCOMING_ADVISE:
1742         /* 1st discard */
1743         tmp = postcopy_ram_prepare_discard(mis);
1744         if (tmp) {
1745             return tmp;
1746         }
1747         break;
1748 
1749     case POSTCOPY_INCOMING_DISCARD:
1750         /* Expected state */
1751         break;
1752 
1753     default:
1754         error_report("CMD_POSTCOPY_RAM_DISCARD in wrong postcopy state (%d)",
1755                      ps);
1756         return -1;
1757     }
1758     /* We're expecting a
1759      *    Version (0)
1760      *    a RAM ID string (length byte, name, 0 term)
1761      *    then at least 1 16 byte chunk
1762     */
1763     if (len < (1 + 1 + 1 + 1 + 2 * 8)) {
1764         error_report("CMD_POSTCOPY_RAM_DISCARD invalid length (%d)", len);
1765         return -1;
1766     }
1767 
1768     tmp = qemu_get_byte(mis->from_src_file);
1769     if (tmp != postcopy_ram_discard_version) {
1770         error_report("CMD_POSTCOPY_RAM_DISCARD invalid version (%d)", tmp);
1771         return -1;
1772     }
1773 
1774     if (!qemu_get_counted_string(mis->from_src_file, ramid)) {
1775         error_report("CMD_POSTCOPY_RAM_DISCARD Failed to read RAMBlock ID");
1776         return -1;
1777     }
1778     tmp = qemu_get_byte(mis->from_src_file);
1779     if (tmp != 0) {
1780         error_report("CMD_POSTCOPY_RAM_DISCARD missing nil (%d)", tmp);
1781         return -1;
1782     }
1783 
1784     len -= 3 + strlen(ramid);
1785     if (len % 16) {
1786         error_report("CMD_POSTCOPY_RAM_DISCARD invalid length (%d)", len);
1787         return -1;
1788     }
1789     trace_loadvm_postcopy_ram_handle_discard_header(ramid, len);
1790     while (len) {
1791         uint64_t start_addr, block_length;
1792         start_addr = qemu_get_be64(mis->from_src_file);
1793         block_length = qemu_get_be64(mis->from_src_file);
1794 
1795         len -= 16;
1796         int ret = ram_discard_range(ramid, start_addr, block_length);
1797         if (ret) {
1798             return ret;
1799         }
1800     }
1801     trace_loadvm_postcopy_ram_handle_discard_end();
1802 
1803     return 0;
1804 }
1805 
1806 /*
1807  * Triggered by a postcopy_listen command; this thread takes over reading
1808  * the input stream, leaving the main thread free to carry on loading the rest
1809  * of the device state (from RAM).
1810  * (TODO:This could do with being in a postcopy file - but there again it's
1811  * just another input loop, not that postcopy specific)
1812  */
1813 static void *postcopy_ram_listen_thread(void *opaque)
1814 {
1815     MigrationIncomingState *mis = migration_incoming_get_current();
1816     QEMUFile *f = mis->from_src_file;
1817     int load_res;
1818     MigrationState *migr = migrate_get_current();
1819 
1820     object_ref(OBJECT(migr));
1821 
1822     migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE,
1823                                    MIGRATION_STATUS_POSTCOPY_ACTIVE);
1824     qemu_sem_post(&mis->listen_thread_sem);
1825     trace_postcopy_ram_listen_thread_start();
1826 
1827     rcu_register_thread();
1828     /*
1829      * Because we're a thread and not a coroutine we can't yield
1830      * in qemu_file, and thus we must be blocking now.
1831      */
1832     qemu_file_set_blocking(f, true);
1833     load_res = qemu_loadvm_state_main(f, mis);
1834 
1835     /*
1836      * This is tricky, but, mis->from_src_file can change after it
1837      * returns, when postcopy recovery happened. In the future, we may
1838      * want a wrapper for the QEMUFile handle.
1839      */
1840     f = mis->from_src_file;
1841 
1842     /* And non-blocking again so we don't block in any cleanup */
1843     qemu_file_set_blocking(f, false);
1844 
1845     trace_postcopy_ram_listen_thread_exit();
1846     if (load_res < 0) {
1847         qemu_file_set_error(f, load_res);
1848         dirty_bitmap_mig_cancel_incoming();
1849         if (postcopy_state_get() == POSTCOPY_INCOMING_RUNNING &&
1850             !migrate_postcopy_ram() && migrate_dirty_bitmaps())
1851         {
1852             error_report("%s: loadvm failed during postcopy: %d. All states "
1853                          "are migrated except dirty bitmaps. Some dirty "
1854                          "bitmaps may be lost, and present migrated dirty "
1855                          "bitmaps are correctly migrated and valid.",
1856                          __func__, load_res);
1857             load_res = 0; /* prevent further exit() */
1858         } else {
1859             error_report("%s: loadvm failed: %d", __func__, load_res);
1860             migrate_set_state(&mis->state, MIGRATION_STATUS_POSTCOPY_ACTIVE,
1861                                            MIGRATION_STATUS_FAILED);
1862         }
1863     }
1864     if (load_res >= 0) {
1865         /*
1866          * This looks good, but it's possible that the device loading in the
1867          * main thread hasn't finished yet, and so we might not be in 'RUN'
1868          * state yet; wait for the end of the main thread.
1869          */
1870         qemu_event_wait(&mis->main_thread_load_event);
1871     }
1872     postcopy_ram_incoming_cleanup(mis);
1873 
1874     if (load_res < 0) {
1875         /*
1876          * If something went wrong then we have a bad state so exit;
1877          * depending how far we got it might be possible at this point
1878          * to leave the guest running and fire MCEs for pages that never
1879          * arrived as a desperate recovery step.
1880          */
1881         rcu_unregister_thread();
1882         exit(EXIT_FAILURE);
1883     }
1884 
1885     migrate_set_state(&mis->state, MIGRATION_STATUS_POSTCOPY_ACTIVE,
1886                                    MIGRATION_STATUS_COMPLETED);
1887     /*
1888      * If everything has worked fine, then the main thread has waited
1889      * for us to start, and we're the last use of the mis.
1890      * (If something broke then qemu will have to exit anyway since it's
1891      * got a bad migration state).
1892      */
1893     migration_incoming_state_destroy();
1894     qemu_loadvm_state_cleanup();
1895 
1896     rcu_unregister_thread();
1897     mis->have_listen_thread = false;
1898     postcopy_state_set(POSTCOPY_INCOMING_END);
1899 
1900     object_unref(OBJECT(migr));
1901 
1902     return NULL;
1903 }
1904 
1905 /* After this message we must be able to immediately receive postcopy data */
1906 static int loadvm_postcopy_handle_listen(MigrationIncomingState *mis)
1907 {
1908     PostcopyState ps = postcopy_state_set(POSTCOPY_INCOMING_LISTENING);
1909     trace_loadvm_postcopy_handle_listen();
1910     Error *local_err = NULL;
1911 
1912     if (ps != POSTCOPY_INCOMING_ADVISE && ps != POSTCOPY_INCOMING_DISCARD) {
1913         error_report("CMD_POSTCOPY_LISTEN in wrong postcopy state (%d)", ps);
1914         return -1;
1915     }
1916     if (ps == POSTCOPY_INCOMING_ADVISE) {
1917         /*
1918          * A rare case, we entered listen without having to do any discards,
1919          * so do the setup that's normally done at the time of the 1st discard.
1920          */
1921         if (migrate_postcopy_ram()) {
1922             postcopy_ram_prepare_discard(mis);
1923         }
1924     }
1925 
1926     /*
1927      * Sensitise RAM - can now generate requests for blocks that don't exist
1928      * However, at this point the CPU shouldn't be running, and the IO
1929      * shouldn't be doing anything yet so don't actually expect requests
1930      */
1931     if (migrate_postcopy_ram()) {
1932         if (postcopy_ram_incoming_setup(mis)) {
1933             postcopy_ram_incoming_cleanup(mis);
1934             return -1;
1935         }
1936     }
1937 
1938     if (postcopy_notify(POSTCOPY_NOTIFY_INBOUND_LISTEN, &local_err)) {
1939         error_report_err(local_err);
1940         return -1;
1941     }
1942 
1943     mis->have_listen_thread = true;
1944     /* Start up the listening thread and wait for it to signal ready */
1945     qemu_sem_init(&mis->listen_thread_sem, 0);
1946     qemu_thread_create(&mis->listen_thread, "postcopy/listen",
1947                        postcopy_ram_listen_thread, NULL,
1948                        QEMU_THREAD_DETACHED);
1949     qemu_sem_wait(&mis->listen_thread_sem);
1950     qemu_sem_destroy(&mis->listen_thread_sem);
1951 
1952     return 0;
1953 }
1954 
1955 static void loadvm_postcopy_handle_run_bh(void *opaque)
1956 {
1957     Error *local_err = NULL;
1958     MigrationIncomingState *mis = opaque;
1959 
1960     /* TODO we should move all of this lot into postcopy_ram.c or a shared code
1961      * in migration.c
1962      */
1963     cpu_synchronize_all_post_init();
1964 
1965     qemu_announce_self(&mis->announce_timer, migrate_announce_params());
1966 
1967     /* Make sure all file formats flush their mutable metadata.
1968      * If we get an error here, just don't restart the VM yet. */
1969     bdrv_invalidate_cache_all(&local_err);
1970     if (local_err) {
1971         error_report_err(local_err);
1972         local_err = NULL;
1973         autostart = false;
1974     }
1975 
1976     trace_loadvm_postcopy_handle_run_cpu_sync();
1977 
1978     trace_loadvm_postcopy_handle_run_vmstart();
1979 
1980     dirty_bitmap_mig_before_vm_start();
1981 
1982     if (autostart) {
1983         /* Hold onto your hats, starting the CPU */
1984         vm_start();
1985     } else {
1986         /* leave it paused and let management decide when to start the CPU */
1987         runstate_set(RUN_STATE_PAUSED);
1988     }
1989 
1990     qemu_bh_delete(mis->bh);
1991 }
1992 
1993 /* After all discards we can start running and asking for pages */
1994 static int loadvm_postcopy_handle_run(MigrationIncomingState *mis)
1995 {
1996     PostcopyState ps = postcopy_state_get();
1997 
1998     trace_loadvm_postcopy_handle_run();
1999     if (ps != POSTCOPY_INCOMING_LISTENING) {
2000         error_report("CMD_POSTCOPY_RUN in wrong postcopy state (%d)", ps);
2001         return -1;
2002     }
2003 
2004     postcopy_state_set(POSTCOPY_INCOMING_RUNNING);
2005     mis->bh = qemu_bh_new(loadvm_postcopy_handle_run_bh, mis);
2006     qemu_bh_schedule(mis->bh);
2007 
2008     /* We need to finish reading the stream from the package
2009      * and also stop reading anything more from the stream that loaded the
2010      * package (since it's now being read by the listener thread).
2011      * LOADVM_QUIT will quit all the layers of nested loadvm loops.
2012      */
2013     return LOADVM_QUIT;
2014 }
2015 
2016 /* We must be with page_request_mutex held */
2017 static gboolean postcopy_sync_page_req(gpointer key, gpointer value,
2018                                        gpointer data)
2019 {
2020     MigrationIncomingState *mis = data;
2021     void *host_addr = (void *) key;
2022     ram_addr_t rb_offset;
2023     RAMBlock *rb;
2024     int ret;
2025 
2026     rb = qemu_ram_block_from_host(host_addr, true, &rb_offset);
2027     if (!rb) {
2028         /*
2029          * This should _never_ happen.  However be nice for a migrating VM to
2030          * not crash/assert.  Post an error (note: intended to not use *_once
2031          * because we do want to see all the illegal addresses; and this can
2032          * never be triggered by the guest so we're safe) and move on next.
2033          */
2034         error_report("%s: illegal host addr %p", __func__, host_addr);
2035         /* Try the next entry */
2036         return FALSE;
2037     }
2038 
2039     ret = migrate_send_rp_message_req_pages(mis, rb, rb_offset);
2040     if (ret) {
2041         /* Please refer to above comment. */
2042         error_report("%s: send rp message failed for addr %p",
2043                      __func__, host_addr);
2044         return FALSE;
2045     }
2046 
2047     trace_postcopy_page_req_sync(host_addr);
2048 
2049     return FALSE;
2050 }
2051 
2052 static void migrate_send_rp_req_pages_pending(MigrationIncomingState *mis)
2053 {
2054     WITH_QEMU_LOCK_GUARD(&mis->page_request_mutex) {
2055         g_tree_foreach(mis->page_requested, postcopy_sync_page_req, mis);
2056     }
2057 }
2058 
2059 static int loadvm_postcopy_handle_resume(MigrationIncomingState *mis)
2060 {
2061     if (mis->state != MIGRATION_STATUS_POSTCOPY_RECOVER) {
2062         error_report("%s: illegal resume received", __func__);
2063         /* Don't fail the load, only for this. */
2064         return 0;
2065     }
2066 
2067     /*
2068      * Reset the last_rb before we resend any page req to source again, since
2069      * the source should have it reset already.
2070      */
2071     mis->last_rb = NULL;
2072 
2073     /*
2074      * This means source VM is ready to resume the postcopy migration.
2075      */
2076     migrate_set_state(&mis->state, MIGRATION_STATUS_POSTCOPY_RECOVER,
2077                       MIGRATION_STATUS_POSTCOPY_ACTIVE);
2078 
2079     trace_loadvm_postcopy_handle_resume();
2080 
2081     /* Tell source that "we are ready" */
2082     migrate_send_rp_resume_ack(mis, MIGRATION_RESUME_ACK_VALUE);
2083 
2084     /*
2085      * After a postcopy recovery, the source should have lost the postcopy
2086      * queue, or potentially the requested pages could have been lost during
2087      * the network down phase.  Let's re-sync with the source VM by re-sending
2088      * all the pending pages that we eagerly need, so these threads won't get
2089      * blocked too long due to the recovery.
2090      *
2091      * Without this procedure, the faulted destination VM threads (waiting for
2092      * page requests right before the postcopy is interrupted) can keep hanging
2093      * until the pages are sent by the source during the background copying of
2094      * pages, or another thread faulted on the same address accidentally.
2095      */
2096     migrate_send_rp_req_pages_pending(mis);
2097 
2098     /*
2099      * It's time to switch state and release the fault thread to continue
2100      * service page faults.  Note that this should be explicitly after the
2101      * above call to migrate_send_rp_req_pages_pending().  In short:
2102      * migrate_send_rp_message_req_pages() is not thread safe, yet.
2103      */
2104     qemu_sem_post(&mis->postcopy_pause_sem_fault);
2105 
2106     return 0;
2107 }
2108 
2109 /**
2110  * Immediately following this command is a blob of data containing an embedded
2111  * chunk of migration stream; read it and load it.
2112  *
2113  * @mis: Incoming state
2114  * @length: Length of packaged data to read
2115  *
2116  * Returns: Negative values on error
2117  *
2118  */
2119 static int loadvm_handle_cmd_packaged(MigrationIncomingState *mis)
2120 {
2121     int ret;
2122     size_t length;
2123     QIOChannelBuffer *bioc;
2124 
2125     length = qemu_get_be32(mis->from_src_file);
2126     trace_loadvm_handle_cmd_packaged(length);
2127 
2128     if (length > MAX_VM_CMD_PACKAGED_SIZE) {
2129         error_report("Unreasonably large packaged state: %zu", length);
2130         return -1;
2131     }
2132 
2133     bioc = qio_channel_buffer_new(length);
2134     qio_channel_set_name(QIO_CHANNEL(bioc), "migration-loadvm-buffer");
2135     ret = qemu_get_buffer(mis->from_src_file,
2136                           bioc->data,
2137                           length);
2138     if (ret != length) {
2139         object_unref(OBJECT(bioc));
2140         error_report("CMD_PACKAGED: Buffer receive fail ret=%d length=%zu",
2141                      ret, length);
2142         return (ret < 0) ? ret : -EAGAIN;
2143     }
2144     bioc->usage += length;
2145     trace_loadvm_handle_cmd_packaged_received(ret);
2146 
2147     QEMUFile *packf = qemu_fopen_channel_input(QIO_CHANNEL(bioc));
2148 
2149     ret = qemu_loadvm_state_main(packf, mis);
2150     trace_loadvm_handle_cmd_packaged_main(ret);
2151     qemu_fclose(packf);
2152     object_unref(OBJECT(bioc));
2153 
2154     return ret;
2155 }
2156 
2157 /*
2158  * Handle request that source requests for recved_bitmap on
2159  * destination. Payload format:
2160  *
2161  * len (1 byte) + ramblock_name (<255 bytes)
2162  */
2163 static int loadvm_handle_recv_bitmap(MigrationIncomingState *mis,
2164                                      uint16_t len)
2165 {
2166     QEMUFile *file = mis->from_src_file;
2167     RAMBlock *rb;
2168     char block_name[256];
2169     size_t cnt;
2170 
2171     cnt = qemu_get_counted_string(file, block_name);
2172     if (!cnt) {
2173         error_report("%s: failed to read block name", __func__);
2174         return -EINVAL;
2175     }
2176 
2177     /* Validate before using the data */
2178     if (qemu_file_get_error(file)) {
2179         return qemu_file_get_error(file);
2180     }
2181 
2182     if (len != cnt + 1) {
2183         error_report("%s: invalid payload length (%d)", __func__, len);
2184         return -EINVAL;
2185     }
2186 
2187     rb = qemu_ram_block_by_name(block_name);
2188     if (!rb) {
2189         error_report("%s: block '%s' not found", __func__, block_name);
2190         return -EINVAL;
2191     }
2192 
2193     migrate_send_rp_recv_bitmap(mis, block_name);
2194 
2195     trace_loadvm_handle_recv_bitmap(block_name);
2196 
2197     return 0;
2198 }
2199 
2200 static int loadvm_process_enable_colo(MigrationIncomingState *mis)
2201 {
2202     int ret = migration_incoming_enable_colo();
2203 
2204     if (!ret) {
2205         ret = colo_init_ram_cache();
2206         if (ret) {
2207             migration_incoming_disable_colo();
2208         }
2209     }
2210     return ret;
2211 }
2212 
2213 /*
2214  * Process an incoming 'QEMU_VM_COMMAND'
2215  * 0           just a normal return
2216  * LOADVM_QUIT All good, but exit the loop
2217  * <0          Error
2218  */
2219 static int loadvm_process_command(QEMUFile *f)
2220 {
2221     MigrationIncomingState *mis = migration_incoming_get_current();
2222     uint16_t cmd;
2223     uint16_t len;
2224     uint32_t tmp32;
2225 
2226     cmd = qemu_get_be16(f);
2227     len = qemu_get_be16(f);
2228 
2229     /* Check validity before continue processing of cmds */
2230     if (qemu_file_get_error(f)) {
2231         return qemu_file_get_error(f);
2232     }
2233 
2234     trace_loadvm_process_command(cmd, len);
2235     if (cmd >= MIG_CMD_MAX || cmd == MIG_CMD_INVALID) {
2236         error_report("MIG_CMD 0x%x unknown (len 0x%x)", cmd, len);
2237         return -EINVAL;
2238     }
2239 
2240     if (mig_cmd_args[cmd].len != -1 && mig_cmd_args[cmd].len != len) {
2241         error_report("%s received with bad length - expecting %zu, got %d",
2242                      mig_cmd_args[cmd].name,
2243                      (size_t)mig_cmd_args[cmd].len, len);
2244         return -ERANGE;
2245     }
2246 
2247     switch (cmd) {
2248     case MIG_CMD_OPEN_RETURN_PATH:
2249         if (mis->to_src_file) {
2250             error_report("CMD_OPEN_RETURN_PATH called when RP already open");
2251             /* Not really a problem, so don't give up */
2252             return 0;
2253         }
2254         mis->to_src_file = qemu_file_get_return_path(f);
2255         if (!mis->to_src_file) {
2256             error_report("CMD_OPEN_RETURN_PATH failed");
2257             return -1;
2258         }
2259         break;
2260 
2261     case MIG_CMD_PING:
2262         tmp32 = qemu_get_be32(f);
2263         trace_loadvm_process_command_ping(tmp32);
2264         if (!mis->to_src_file) {
2265             error_report("CMD_PING (0x%x) received with no return path",
2266                          tmp32);
2267             return -1;
2268         }
2269         migrate_send_rp_pong(mis, tmp32);
2270         break;
2271 
2272     case MIG_CMD_PACKAGED:
2273         return loadvm_handle_cmd_packaged(mis);
2274 
2275     case MIG_CMD_POSTCOPY_ADVISE:
2276         return loadvm_postcopy_handle_advise(mis, len);
2277 
2278     case MIG_CMD_POSTCOPY_LISTEN:
2279         return loadvm_postcopy_handle_listen(mis);
2280 
2281     case MIG_CMD_POSTCOPY_RUN:
2282         return loadvm_postcopy_handle_run(mis);
2283 
2284     case MIG_CMD_POSTCOPY_RAM_DISCARD:
2285         return loadvm_postcopy_ram_handle_discard(mis, len);
2286 
2287     case MIG_CMD_POSTCOPY_RESUME:
2288         return loadvm_postcopy_handle_resume(mis);
2289 
2290     case MIG_CMD_RECV_BITMAP:
2291         return loadvm_handle_recv_bitmap(mis, len);
2292 
2293     case MIG_CMD_ENABLE_COLO:
2294         return loadvm_process_enable_colo(mis);
2295     }
2296 
2297     return 0;
2298 }
2299 
2300 /*
2301  * Read a footer off the wire and check that it matches the expected section
2302  *
2303  * Returns: true if the footer was good
2304  *          false if there is a problem (and calls error_report to say why)
2305  */
2306 static bool check_section_footer(QEMUFile *f, SaveStateEntry *se)
2307 {
2308     int ret;
2309     uint8_t read_mark;
2310     uint32_t read_section_id;
2311 
2312     if (!migrate_get_current()->send_section_footer) {
2313         /* No footer to check */
2314         return true;
2315     }
2316 
2317     read_mark = qemu_get_byte(f);
2318 
2319     ret = qemu_file_get_error(f);
2320     if (ret) {
2321         error_report("%s: Read section footer failed: %d",
2322                      __func__, ret);
2323         return false;
2324     }
2325 
2326     if (read_mark != QEMU_VM_SECTION_FOOTER) {
2327         error_report("Missing section footer for %s", se->idstr);
2328         return false;
2329     }
2330 
2331     read_section_id = qemu_get_be32(f);
2332     if (read_section_id != se->load_section_id) {
2333         error_report("Mismatched section id in footer for %s -"
2334                      " read 0x%x expected 0x%x",
2335                      se->idstr, read_section_id, se->load_section_id);
2336         return false;
2337     }
2338 
2339     /* All good */
2340     return true;
2341 }
2342 
2343 static int
2344 qemu_loadvm_section_start_full(QEMUFile *f, MigrationIncomingState *mis)
2345 {
2346     uint32_t instance_id, version_id, section_id;
2347     SaveStateEntry *se;
2348     char idstr[256];
2349     int ret;
2350 
2351     /* Read section start */
2352     section_id = qemu_get_be32(f);
2353     if (!qemu_get_counted_string(f, idstr)) {
2354         error_report("Unable to read ID string for section %u",
2355                      section_id);
2356         return -EINVAL;
2357     }
2358     instance_id = qemu_get_be32(f);
2359     version_id = qemu_get_be32(f);
2360 
2361     ret = qemu_file_get_error(f);
2362     if (ret) {
2363         error_report("%s: Failed to read instance/version ID: %d",
2364                      __func__, ret);
2365         return ret;
2366     }
2367 
2368     trace_qemu_loadvm_state_section_startfull(section_id, idstr,
2369             instance_id, version_id);
2370     /* Find savevm section */
2371     se = find_se(idstr, instance_id);
2372     if (se == NULL) {
2373         error_report("Unknown savevm section or instance '%s' %"PRIu32". "
2374                      "Make sure that your current VM setup matches your "
2375                      "saved VM setup, including any hotplugged devices",
2376                      idstr, instance_id);
2377         return -EINVAL;
2378     }
2379 
2380     /* Validate version */
2381     if (version_id > se->version_id) {
2382         error_report("savevm: unsupported version %d for '%s' v%d",
2383                      version_id, idstr, se->version_id);
2384         return -EINVAL;
2385     }
2386     se->load_version_id = version_id;
2387     se->load_section_id = section_id;
2388 
2389     /* Validate if it is a device's state */
2390     if (xen_enabled() && se->is_ram) {
2391         error_report("loadvm: %s RAM loading not allowed on Xen", idstr);
2392         return -EINVAL;
2393     }
2394 
2395     ret = vmstate_load(f, se);
2396     if (ret < 0) {
2397         error_report("error while loading state for instance 0x%"PRIx32" of"
2398                      " device '%s'", instance_id, idstr);
2399         return ret;
2400     }
2401     if (!check_section_footer(f, se)) {
2402         return -EINVAL;
2403     }
2404 
2405     return 0;
2406 }
2407 
2408 static int
2409 qemu_loadvm_section_part_end(QEMUFile *f, MigrationIncomingState *mis)
2410 {
2411     uint32_t section_id;
2412     SaveStateEntry *se;
2413     int ret;
2414 
2415     section_id = qemu_get_be32(f);
2416 
2417     ret = qemu_file_get_error(f);
2418     if (ret) {
2419         error_report("%s: Failed to read section ID: %d",
2420                      __func__, ret);
2421         return ret;
2422     }
2423 
2424     trace_qemu_loadvm_state_section_partend(section_id);
2425     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
2426         if (se->load_section_id == section_id) {
2427             break;
2428         }
2429     }
2430     if (se == NULL) {
2431         error_report("Unknown savevm section %d", section_id);
2432         return -EINVAL;
2433     }
2434 
2435     ret = vmstate_load(f, se);
2436     if (ret < 0) {
2437         error_report("error while loading state section id %d(%s)",
2438                      section_id, se->idstr);
2439         return ret;
2440     }
2441     if (!check_section_footer(f, se)) {
2442         return -EINVAL;
2443     }
2444 
2445     return 0;
2446 }
2447 
2448 static int qemu_loadvm_state_header(QEMUFile *f)
2449 {
2450     unsigned int v;
2451     int ret;
2452 
2453     v = qemu_get_be32(f);
2454     if (v != QEMU_VM_FILE_MAGIC) {
2455         error_report("Not a migration stream");
2456         return -EINVAL;
2457     }
2458 
2459     v = qemu_get_be32(f);
2460     if (v == QEMU_VM_FILE_VERSION_COMPAT) {
2461         error_report("SaveVM v2 format is obsolete and don't work anymore");
2462         return -ENOTSUP;
2463     }
2464     if (v != QEMU_VM_FILE_VERSION) {
2465         error_report("Unsupported migration stream version");
2466         return -ENOTSUP;
2467     }
2468 
2469     if (migrate_get_current()->send_configuration) {
2470         if (qemu_get_byte(f) != QEMU_VM_CONFIGURATION) {
2471             error_report("Configuration section missing");
2472             qemu_loadvm_state_cleanup();
2473             return -EINVAL;
2474         }
2475         ret = vmstate_load_state(f, &vmstate_configuration, &savevm_state, 0);
2476 
2477         if (ret) {
2478             qemu_loadvm_state_cleanup();
2479             return ret;
2480         }
2481     }
2482     return 0;
2483 }
2484 
2485 static int qemu_loadvm_state_setup(QEMUFile *f)
2486 {
2487     SaveStateEntry *se;
2488     int ret;
2489 
2490     trace_loadvm_state_setup();
2491     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
2492         if (!se->ops || !se->ops->load_setup) {
2493             continue;
2494         }
2495         if (se->ops->is_active) {
2496             if (!se->ops->is_active(se->opaque)) {
2497                 continue;
2498             }
2499         }
2500 
2501         ret = se->ops->load_setup(f, se->opaque);
2502         if (ret < 0) {
2503             qemu_file_set_error(f, ret);
2504             error_report("Load state of device %s failed", se->idstr);
2505             return ret;
2506         }
2507     }
2508     return 0;
2509 }
2510 
2511 void qemu_loadvm_state_cleanup(void)
2512 {
2513     SaveStateEntry *se;
2514 
2515     trace_loadvm_state_cleanup();
2516     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
2517         if (se->ops && se->ops->load_cleanup) {
2518             se->ops->load_cleanup(se->opaque);
2519         }
2520     }
2521 }
2522 
2523 /* Return true if we should continue the migration, or false. */
2524 static bool postcopy_pause_incoming(MigrationIncomingState *mis)
2525 {
2526     trace_postcopy_pause_incoming();
2527 
2528     assert(migrate_postcopy_ram());
2529 
2530     /* Clear the triggered bit to allow one recovery */
2531     mis->postcopy_recover_triggered = false;
2532 
2533     assert(mis->from_src_file);
2534     qemu_file_shutdown(mis->from_src_file);
2535     qemu_fclose(mis->from_src_file);
2536     mis->from_src_file = NULL;
2537 
2538     assert(mis->to_src_file);
2539     qemu_file_shutdown(mis->to_src_file);
2540     qemu_mutex_lock(&mis->rp_mutex);
2541     qemu_fclose(mis->to_src_file);
2542     mis->to_src_file = NULL;
2543     qemu_mutex_unlock(&mis->rp_mutex);
2544 
2545     migrate_set_state(&mis->state, MIGRATION_STATUS_POSTCOPY_ACTIVE,
2546                       MIGRATION_STATUS_POSTCOPY_PAUSED);
2547 
2548     /* Notify the fault thread for the invalidated file handle */
2549     postcopy_fault_thread_notify(mis);
2550 
2551     error_report("Detected IO failure for postcopy. "
2552                  "Migration paused.");
2553 
2554     while (mis->state == MIGRATION_STATUS_POSTCOPY_PAUSED) {
2555         qemu_sem_wait(&mis->postcopy_pause_sem_dst);
2556     }
2557 
2558     trace_postcopy_pause_incoming_continued();
2559 
2560     return true;
2561 }
2562 
2563 int qemu_loadvm_state_main(QEMUFile *f, MigrationIncomingState *mis)
2564 {
2565     uint8_t section_type;
2566     int ret = 0;
2567 
2568 retry:
2569     while (true) {
2570         section_type = qemu_get_byte(f);
2571 
2572         if (qemu_file_get_error(f)) {
2573             ret = qemu_file_get_error(f);
2574             break;
2575         }
2576 
2577         trace_qemu_loadvm_state_section(section_type);
2578         switch (section_type) {
2579         case QEMU_VM_SECTION_START:
2580         case QEMU_VM_SECTION_FULL:
2581             ret = qemu_loadvm_section_start_full(f, mis);
2582             if (ret < 0) {
2583                 goto out;
2584             }
2585             break;
2586         case QEMU_VM_SECTION_PART:
2587         case QEMU_VM_SECTION_END:
2588             ret = qemu_loadvm_section_part_end(f, mis);
2589             if (ret < 0) {
2590                 goto out;
2591             }
2592             break;
2593         case QEMU_VM_COMMAND:
2594             ret = loadvm_process_command(f);
2595             trace_qemu_loadvm_state_section_command(ret);
2596             if ((ret < 0) || (ret == LOADVM_QUIT)) {
2597                 goto out;
2598             }
2599             break;
2600         case QEMU_VM_EOF:
2601             /* This is the end of migration */
2602             goto out;
2603         default:
2604             error_report("Unknown savevm section type %d", section_type);
2605             ret = -EINVAL;
2606             goto out;
2607         }
2608     }
2609 
2610 out:
2611     if (ret < 0) {
2612         qemu_file_set_error(f, ret);
2613 
2614         /* Cancel bitmaps incoming regardless of recovery */
2615         dirty_bitmap_mig_cancel_incoming();
2616 
2617         /*
2618          * If we are during an active postcopy, then we pause instead
2619          * of bail out to at least keep the VM's dirty data.  Note
2620          * that POSTCOPY_INCOMING_LISTENING stage is still not enough,
2621          * during which we're still receiving device states and we
2622          * still haven't yet started the VM on destination.
2623          *
2624          * Only RAM postcopy supports recovery. Still, if RAM postcopy is
2625          * enabled, canceled bitmaps postcopy will not affect RAM postcopy
2626          * recovering.
2627          */
2628         if (postcopy_state_get() == POSTCOPY_INCOMING_RUNNING &&
2629             migrate_postcopy_ram() && postcopy_pause_incoming(mis)) {
2630             /* Reset f to point to the newly created channel */
2631             f = mis->from_src_file;
2632             goto retry;
2633         }
2634     }
2635     return ret;
2636 }
2637 
2638 int qemu_loadvm_state(QEMUFile *f)
2639 {
2640     MigrationIncomingState *mis = migration_incoming_get_current();
2641     Error *local_err = NULL;
2642     int ret;
2643 
2644     if (qemu_savevm_state_blocked(&local_err)) {
2645         error_report_err(local_err);
2646         return -EINVAL;
2647     }
2648 
2649     ret = qemu_loadvm_state_header(f);
2650     if (ret) {
2651         return ret;
2652     }
2653 
2654     if (qemu_loadvm_state_setup(f) != 0) {
2655         return -EINVAL;
2656     }
2657 
2658     cpu_synchronize_all_pre_loadvm();
2659 
2660     ret = qemu_loadvm_state_main(f, mis);
2661     qemu_event_set(&mis->main_thread_load_event);
2662 
2663     trace_qemu_loadvm_state_post_main(ret);
2664 
2665     if (mis->have_listen_thread) {
2666         /* Listen thread still going, can't clean up yet */
2667         return ret;
2668     }
2669 
2670     if (ret == 0) {
2671         ret = qemu_file_get_error(f);
2672     }
2673 
2674     /*
2675      * Try to read in the VMDESC section as well, so that dumping tools that
2676      * intercept our migration stream have the chance to see it.
2677      */
2678 
2679     /* We've got to be careful; if we don't read the data and just shut the fd
2680      * then the sender can error if we close while it's still sending.
2681      * We also mustn't read data that isn't there; some transports (RDMA)
2682      * will stall waiting for that data when the source has already closed.
2683      */
2684     if (ret == 0 && should_send_vmdesc()) {
2685         uint8_t *buf;
2686         uint32_t size;
2687         uint8_t  section_type = qemu_get_byte(f);
2688 
2689         if (section_type != QEMU_VM_VMDESCRIPTION) {
2690             error_report("Expected vmdescription section, but got %d",
2691                          section_type);
2692             /*
2693              * It doesn't seem worth failing at this point since
2694              * we apparently have an otherwise valid VM state
2695              */
2696         } else {
2697             buf = g_malloc(0x1000);
2698             size = qemu_get_be32(f);
2699 
2700             while (size > 0) {
2701                 uint32_t read_chunk = MIN(size, 0x1000);
2702                 qemu_get_buffer(f, buf, read_chunk);
2703                 size -= read_chunk;
2704             }
2705             g_free(buf);
2706         }
2707     }
2708 
2709     qemu_loadvm_state_cleanup();
2710     cpu_synchronize_all_post_init();
2711 
2712     return ret;
2713 }
2714 
2715 int qemu_load_device_state(QEMUFile *f)
2716 {
2717     MigrationIncomingState *mis = migration_incoming_get_current();
2718     int ret;
2719 
2720     /* Load QEMU_VM_SECTION_FULL section */
2721     ret = qemu_loadvm_state_main(f, mis);
2722     if (ret < 0) {
2723         error_report("Failed to load device state: %d", ret);
2724         return ret;
2725     }
2726 
2727     cpu_synchronize_all_post_init();
2728     return 0;
2729 }
2730 
2731 int save_snapshot(const char *name, Error **errp)
2732 {
2733     BlockDriverState *bs, *bs1;
2734     QEMUSnapshotInfo sn1, *sn = &sn1;
2735     int ret = -1, ret2;
2736     QEMUFile *f;
2737     int saved_vm_running;
2738     uint64_t vm_state_size;
2739     qemu_timeval tv;
2740     struct tm tm;
2741     AioContext *aio_context;
2742 
2743     if (migration_is_blocked(errp)) {
2744         return ret;
2745     }
2746 
2747     if (!replay_can_snapshot()) {
2748         error_setg(errp, "Record/replay does not allow making snapshot "
2749                    "right now. Try once more later.");
2750         return ret;
2751     }
2752 
2753     if (!bdrv_all_can_snapshot(&bs)) {
2754         error_setg(errp, "Device '%s' is writable but does not support "
2755                    "snapshots", bdrv_get_device_or_node_name(bs));
2756         return ret;
2757     }
2758 
2759     /* Delete old snapshots of the same name */
2760     if (name) {
2761         ret = bdrv_all_delete_snapshot(name, &bs1, errp);
2762         if (ret < 0) {
2763             error_prepend(errp, "Error while deleting snapshot on device "
2764                           "'%s': ", bdrv_get_device_or_node_name(bs1));
2765             return ret;
2766         }
2767     }
2768 
2769     bs = bdrv_all_find_vmstate_bs();
2770     if (bs == NULL) {
2771         error_setg(errp, "No block device can accept snapshots");
2772         return ret;
2773     }
2774     aio_context = bdrv_get_aio_context(bs);
2775 
2776     saved_vm_running = runstate_is_running();
2777 
2778     ret = global_state_store();
2779     if (ret) {
2780         error_setg(errp, "Error saving global state");
2781         return ret;
2782     }
2783     vm_stop(RUN_STATE_SAVE_VM);
2784 
2785     bdrv_drain_all_begin();
2786 
2787     aio_context_acquire(aio_context);
2788 
2789     memset(sn, 0, sizeof(*sn));
2790 
2791     /* fill auxiliary fields */
2792     qemu_gettimeofday(&tv);
2793     sn->date_sec = tv.tv_sec;
2794     sn->date_nsec = tv.tv_usec * 1000;
2795     sn->vm_clock_nsec = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL);
2796     if (replay_mode != REPLAY_MODE_NONE) {
2797         sn->icount = replay_get_current_icount();
2798     } else {
2799         sn->icount = -1ULL;
2800     }
2801 
2802     if (name) {
2803         pstrcpy(sn->name, sizeof(sn->name), name);
2804     } else {
2805         /* cast below needed for OpenBSD where tv_sec is still 'long' */
2806         localtime_r((const time_t *)&tv.tv_sec, &tm);
2807         strftime(sn->name, sizeof(sn->name), "vm-%Y%m%d%H%M%S", &tm);
2808     }
2809 
2810     /* save the VM state */
2811     f = qemu_fopen_bdrv(bs, 1);
2812     if (!f) {
2813         error_setg(errp, "Could not open VM state file");
2814         goto the_end;
2815     }
2816     ret = qemu_savevm_state(f, errp);
2817     vm_state_size = qemu_ftell(f);
2818     ret2 = qemu_fclose(f);
2819     if (ret < 0) {
2820         goto the_end;
2821     }
2822     if (ret2 < 0) {
2823         ret = ret2;
2824         goto the_end;
2825     }
2826 
2827     /* The bdrv_all_create_snapshot() call that follows acquires the AioContext
2828      * for itself.  BDRV_POLL_WHILE() does not support nested locking because
2829      * it only releases the lock once.  Therefore synchronous I/O will deadlock
2830      * unless we release the AioContext before bdrv_all_create_snapshot().
2831      */
2832     aio_context_release(aio_context);
2833     aio_context = NULL;
2834 
2835     ret = bdrv_all_create_snapshot(sn, bs, vm_state_size, &bs);
2836     if (ret < 0) {
2837         error_setg(errp, "Error while creating snapshot on '%s'",
2838                    bdrv_get_device_or_node_name(bs));
2839         bdrv_all_delete_snapshot(sn->name, &bs, NULL);
2840         goto the_end;
2841     }
2842 
2843     ret = 0;
2844 
2845  the_end:
2846     if (aio_context) {
2847         aio_context_release(aio_context);
2848     }
2849 
2850     bdrv_drain_all_end();
2851 
2852     if (saved_vm_running) {
2853         vm_start();
2854     }
2855     return ret;
2856 }
2857 
2858 void qmp_xen_save_devices_state(const char *filename, bool has_live, bool live,
2859                                 Error **errp)
2860 {
2861     QEMUFile *f;
2862     QIOChannelFile *ioc;
2863     int saved_vm_running;
2864     int ret;
2865 
2866     if (!has_live) {
2867         /* live default to true so old version of Xen tool stack can have a
2868          * successful live migration */
2869         live = true;
2870     }
2871 
2872     saved_vm_running = runstate_is_running();
2873     vm_stop(RUN_STATE_SAVE_VM);
2874     global_state_store_running();
2875 
2876     ioc = qio_channel_file_new_path(filename, O_WRONLY | O_CREAT | O_TRUNC,
2877                                     0660, errp);
2878     if (!ioc) {
2879         goto the_end;
2880     }
2881     qio_channel_set_name(QIO_CHANNEL(ioc), "migration-xen-save-state");
2882     f = qemu_fopen_channel_output(QIO_CHANNEL(ioc));
2883     object_unref(OBJECT(ioc));
2884     ret = qemu_save_device_state(f);
2885     if (ret < 0 || qemu_fclose(f) < 0) {
2886         error_setg(errp, QERR_IO_ERROR);
2887     } else {
2888         /* libxl calls the QMP command "stop" before calling
2889          * "xen-save-devices-state" and in case of migration failure, libxl
2890          * would call "cont".
2891          * So call bdrv_inactivate_all (release locks) here to let the other
2892          * side of the migration take control of the images.
2893          */
2894         if (live && !saved_vm_running) {
2895             ret = bdrv_inactivate_all();
2896             if (ret) {
2897                 error_setg(errp, "%s: bdrv_inactivate_all() failed (%d)",
2898                            __func__, ret);
2899             }
2900         }
2901     }
2902 
2903  the_end:
2904     if (saved_vm_running) {
2905         vm_start();
2906     }
2907 }
2908 
2909 void qmp_xen_load_devices_state(const char *filename, Error **errp)
2910 {
2911     QEMUFile *f;
2912     QIOChannelFile *ioc;
2913     int ret;
2914 
2915     /* Guest must be paused before loading the device state; the RAM state
2916      * will already have been loaded by xc
2917      */
2918     if (runstate_is_running()) {
2919         error_setg(errp, "Cannot update device state while vm is running");
2920         return;
2921     }
2922     vm_stop(RUN_STATE_RESTORE_VM);
2923 
2924     ioc = qio_channel_file_new_path(filename, O_RDONLY | O_BINARY, 0, errp);
2925     if (!ioc) {
2926         return;
2927     }
2928     qio_channel_set_name(QIO_CHANNEL(ioc), "migration-xen-load-state");
2929     f = qemu_fopen_channel_input(QIO_CHANNEL(ioc));
2930     object_unref(OBJECT(ioc));
2931 
2932     ret = qemu_loadvm_state(f);
2933     qemu_fclose(f);
2934     if (ret < 0) {
2935         error_setg(errp, QERR_IO_ERROR);
2936     }
2937     migration_incoming_state_destroy();
2938 }
2939 
2940 int load_snapshot(const char *name, Error **errp)
2941 {
2942     BlockDriverState *bs, *bs_vm_state;
2943     QEMUSnapshotInfo sn;
2944     QEMUFile *f;
2945     int ret;
2946     AioContext *aio_context;
2947     MigrationIncomingState *mis = migration_incoming_get_current();
2948 
2949     if (!bdrv_all_can_snapshot(&bs)) {
2950         error_setg(errp,
2951                    "Device '%s' is writable but does not support snapshots",
2952                    bdrv_get_device_or_node_name(bs));
2953         return -ENOTSUP;
2954     }
2955     ret = bdrv_all_find_snapshot(name, &bs);
2956     if (ret < 0) {
2957         error_setg(errp,
2958                    "Device '%s' does not have the requested snapshot '%s'",
2959                    bdrv_get_device_or_node_name(bs), name);
2960         return ret;
2961     }
2962 
2963     bs_vm_state = bdrv_all_find_vmstate_bs();
2964     if (!bs_vm_state) {
2965         error_setg(errp, "No block device supports snapshots");
2966         return -ENOTSUP;
2967     }
2968     aio_context = bdrv_get_aio_context(bs_vm_state);
2969 
2970     /* Don't even try to load empty VM states */
2971     aio_context_acquire(aio_context);
2972     ret = bdrv_snapshot_find(bs_vm_state, &sn, name);
2973     aio_context_release(aio_context);
2974     if (ret < 0) {
2975         return ret;
2976     } else if (sn.vm_state_size == 0) {
2977         error_setg(errp, "This is a disk-only snapshot. Revert to it "
2978                    " offline using qemu-img");
2979         return -EINVAL;
2980     }
2981 
2982     /*
2983      * Flush the record/replay queue. Now the VM state is going
2984      * to change. Therefore we don't need to preserve its consistency
2985      */
2986     replay_flush_events();
2987 
2988     /* Flush all IO requests so they don't interfere with the new state.  */
2989     bdrv_drain_all_begin();
2990 
2991     ret = bdrv_all_goto_snapshot(name, &bs, errp);
2992     if (ret < 0) {
2993         error_prepend(errp, "Could not load snapshot '%s' on '%s': ",
2994                       name, bdrv_get_device_or_node_name(bs));
2995         goto err_drain;
2996     }
2997 
2998     /* restore the VM state */
2999     f = qemu_fopen_bdrv(bs_vm_state, 0);
3000     if (!f) {
3001         error_setg(errp, "Could not open VM state file");
3002         ret = -EINVAL;
3003         goto err_drain;
3004     }
3005 
3006     qemu_system_reset(SHUTDOWN_CAUSE_NONE);
3007     mis->from_src_file = f;
3008 
3009     aio_context_acquire(aio_context);
3010     ret = qemu_loadvm_state(f);
3011     migration_incoming_state_destroy();
3012     aio_context_release(aio_context);
3013 
3014     bdrv_drain_all_end();
3015 
3016     if (ret < 0) {
3017         error_setg(errp, "Error %d while loading VM state", ret);
3018         return ret;
3019     }
3020 
3021     return 0;
3022 
3023 err_drain:
3024     bdrv_drain_all_end();
3025     return ret;
3026 }
3027 
3028 void vmstate_register_ram(MemoryRegion *mr, DeviceState *dev)
3029 {
3030     qemu_ram_set_idstr(mr->ram_block,
3031                        memory_region_name(mr), dev);
3032     qemu_ram_set_migratable(mr->ram_block);
3033 }
3034 
3035 void vmstate_unregister_ram(MemoryRegion *mr, DeviceState *dev)
3036 {
3037     qemu_ram_unset_idstr(mr->ram_block);
3038     qemu_ram_unset_migratable(mr->ram_block);
3039 }
3040 
3041 void vmstate_register_ram_global(MemoryRegion *mr)
3042 {
3043     vmstate_register_ram(mr, NULL);
3044 }
3045 
3046 bool vmstate_check_only_migratable(const VMStateDescription *vmsd)
3047 {
3048     /* check needed if --only-migratable is specified */
3049     if (!only_migratable) {
3050         return true;
3051     }
3052 
3053     return !(vmsd && vmsd->unmigratable);
3054 }
3055