xref: /openbmc/qemu/migration/savevm.c (revision beb1a917)
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-stats.h"
35 #include "migration/vmstate.h"
36 #include "migration/misc.h"
37 #include "migration/register.h"
38 #include "migration/global_state.h"
39 #include "migration/channel-block.h"
40 #include "ram.h"
41 #include "qemu-file.h"
42 #include "savevm.h"
43 #include "postcopy-ram.h"
44 #include "qapi/error.h"
45 #include "qapi/qapi-commands-migration.h"
46 #include "qapi/clone-visitor.h"
47 #include "qapi/qapi-builtin-visit.h"
48 #include "qapi/qmp/qerror.h"
49 #include "qemu/error-report.h"
50 #include "sysemu/cpus.h"
51 #include "exec/memory.h"
52 #include "exec/target_page.h"
53 #include "trace.h"
54 #include "qemu/iov.h"
55 #include "qemu/job.h"
56 #include "qemu/main-loop.h"
57 #include "block/snapshot.h"
58 #include "qemu/cutils.h"
59 #include "io/channel-buffer.h"
60 #include "io/channel-file.h"
61 #include "sysemu/replay.h"
62 #include "sysemu/runstate.h"
63 #include "sysemu/sysemu.h"
64 #include "sysemu/xen.h"
65 #include "migration/colo.h"
66 #include "qemu/bitmap.h"
67 #include "net/announce.h"
68 #include "qemu/yank.h"
69 #include "yank_functions.h"
70 #include "sysemu/qtest.h"
71 #include "options.h"
72 
73 const unsigned int postcopy_ram_discard_version;
74 
75 /* Subcommands for QEMU_VM_COMMAND */
76 enum qemu_vm_cmd {
77     MIG_CMD_INVALID = 0,   /* Must be 0 */
78     MIG_CMD_OPEN_RETURN_PATH,  /* Tell the dest to open the Return path */
79     MIG_CMD_PING,              /* Request a PONG on the RP */
80 
81     MIG_CMD_POSTCOPY_ADVISE,       /* Prior to any page transfers, just
82                                       warn we might want to do PC */
83     MIG_CMD_POSTCOPY_LISTEN,       /* Start listening for incoming
84                                       pages as it's running. */
85     MIG_CMD_POSTCOPY_RUN,          /* Start execution */
86 
87     MIG_CMD_POSTCOPY_RAM_DISCARD,  /* A list of pages to discard that
88                                       were previously sent during
89                                       precopy but are dirty. */
90     MIG_CMD_PACKAGED,          /* Send a wrapped stream within this stream */
91     MIG_CMD_ENABLE_COLO,       /* Enable COLO */
92     MIG_CMD_POSTCOPY_RESUME,   /* resume postcopy on dest */
93     MIG_CMD_RECV_BITMAP,       /* Request for recved bitmap on dst */
94     MIG_CMD_MAX
95 };
96 
97 #define MAX_VM_CMD_PACKAGED_SIZE UINT32_MAX
98 static struct mig_cmd_args {
99     ssize_t     len; /* -1 = variable */
100     const char *name;
101 } mig_cmd_args[] = {
102     [MIG_CMD_INVALID]          = { .len = -1, .name = "INVALID" },
103     [MIG_CMD_OPEN_RETURN_PATH] = { .len =  0, .name = "OPEN_RETURN_PATH" },
104     [MIG_CMD_PING]             = { .len = sizeof(uint32_t), .name = "PING" },
105     [MIG_CMD_POSTCOPY_ADVISE]  = { .len = -1, .name = "POSTCOPY_ADVISE" },
106     [MIG_CMD_POSTCOPY_LISTEN]  = { .len =  0, .name = "POSTCOPY_LISTEN" },
107     [MIG_CMD_POSTCOPY_RUN]     = { .len =  0, .name = "POSTCOPY_RUN" },
108     [MIG_CMD_POSTCOPY_RAM_DISCARD] = {
109                                    .len = -1, .name = "POSTCOPY_RAM_DISCARD" },
110     [MIG_CMD_POSTCOPY_RESUME]  = { .len =  0, .name = "POSTCOPY_RESUME" },
111     [MIG_CMD_PACKAGED]         = { .len =  4, .name = "PACKAGED" },
112     [MIG_CMD_RECV_BITMAP]      = { .len = -1, .name = "RECV_BITMAP" },
113     [MIG_CMD_MAX]              = { .len = -1, .name = "MAX" },
114 };
115 
116 /* Note for MIG_CMD_POSTCOPY_ADVISE:
117  * The format of arguments is depending on postcopy mode:
118  * - postcopy RAM only
119  *   uint64_t host page size
120  *   uint64_t target page size
121  *
122  * - postcopy RAM and postcopy dirty bitmaps
123  *   format is the same as for postcopy RAM only
124  *
125  * - postcopy dirty bitmaps only
126  *   Nothing. Command length field is 0.
127  *
128  * Be careful: adding a new postcopy entity with some other parameters should
129  * not break format self-description ability. Good way is to introduce some
130  * generic extendable format with an exception for two old entities.
131  */
132 
133 /***********************************************************/
134 /* savevm/loadvm support */
135 
136 static QEMUFile *qemu_fopen_bdrv(BlockDriverState *bs, int is_writable)
137 {
138     if (is_writable) {
139         return qemu_file_new_output(QIO_CHANNEL(qio_channel_block_new(bs)));
140     } else {
141         return qemu_file_new_input(QIO_CHANNEL(qio_channel_block_new(bs)));
142     }
143 }
144 
145 
146 /* QEMUFile timer support.
147  * Not in qemu-file.c to not add qemu-timer.c as dependency to qemu-file.c
148  */
149 
150 void timer_put(QEMUFile *f, QEMUTimer *ts)
151 {
152     uint64_t expire_time;
153 
154     expire_time = timer_expire_time_ns(ts);
155     qemu_put_be64(f, expire_time);
156 }
157 
158 void timer_get(QEMUFile *f, QEMUTimer *ts)
159 {
160     uint64_t expire_time;
161 
162     expire_time = qemu_get_be64(f);
163     if (expire_time != -1) {
164         timer_mod_ns(ts, expire_time);
165     } else {
166         timer_del(ts);
167     }
168 }
169 
170 
171 /* VMState timer support.
172  * Not in vmstate.c to not add qemu-timer.c as dependency to vmstate.c
173  */
174 
175 static int get_timer(QEMUFile *f, void *pv, size_t size,
176                      const VMStateField *field)
177 {
178     QEMUTimer *v = pv;
179     timer_get(f, v);
180     return 0;
181 }
182 
183 static int put_timer(QEMUFile *f, void *pv, size_t size,
184                      const VMStateField *field, JSONWriter *vmdesc)
185 {
186     QEMUTimer *v = pv;
187     timer_put(f, v);
188 
189     return 0;
190 }
191 
192 const VMStateInfo vmstate_info_timer = {
193     .name = "timer",
194     .get  = get_timer,
195     .put  = put_timer,
196 };
197 
198 
199 typedef struct CompatEntry {
200     char idstr[256];
201     int instance_id;
202 } CompatEntry;
203 
204 typedef struct SaveStateEntry {
205     QTAILQ_ENTRY(SaveStateEntry) entry;
206     char idstr[256];
207     uint32_t instance_id;
208     int alias_id;
209     int version_id;
210     /* version id read from the stream */
211     int load_version_id;
212     int section_id;
213     /* section id read from the stream */
214     int load_section_id;
215     const SaveVMHandlers *ops;
216     const VMStateDescription *vmsd;
217     void *opaque;
218     CompatEntry *compat;
219     int is_ram;
220 } SaveStateEntry;
221 
222 typedef struct SaveState {
223     QTAILQ_HEAD(, SaveStateEntry) handlers;
224     SaveStateEntry *handler_pri_head[MIG_PRI_MAX + 1];
225     int global_section_id;
226     uint32_t len;
227     const char *name;
228     uint32_t target_page_bits;
229     uint32_t caps_count;
230     MigrationCapability *capabilities;
231     QemuUUID uuid;
232 } SaveState;
233 
234 static SaveState savevm_state = {
235     .handlers = QTAILQ_HEAD_INITIALIZER(savevm_state.handlers),
236     .handler_pri_head = { [MIG_PRI_DEFAULT ... MIG_PRI_MAX] = NULL },
237     .global_section_id = 0,
238 };
239 
240 static bool should_validate_capability(int capability)
241 {
242     assert(capability >= 0 && capability < MIGRATION_CAPABILITY__MAX);
243     /* Validate only new capabilities to keep compatibility. */
244     switch (capability) {
245     case MIGRATION_CAPABILITY_X_IGNORE_SHARED:
246         return true;
247     default:
248         return false;
249     }
250 }
251 
252 static uint32_t get_validatable_capabilities_count(void)
253 {
254     MigrationState *s = migrate_get_current();
255     uint32_t result = 0;
256     int i;
257     for (i = 0; i < MIGRATION_CAPABILITY__MAX; i++) {
258         if (should_validate_capability(i) && s->capabilities[i]) {
259             result++;
260         }
261     }
262     return result;
263 }
264 
265 static int configuration_pre_save(void *opaque)
266 {
267     SaveState *state = opaque;
268     const char *current_name = MACHINE_GET_CLASS(current_machine)->name;
269     MigrationState *s = migrate_get_current();
270     int i, j;
271 
272     state->len = strlen(current_name);
273     state->name = current_name;
274     state->target_page_bits = qemu_target_page_bits();
275 
276     state->caps_count = get_validatable_capabilities_count();
277     state->capabilities = g_renew(MigrationCapability, state->capabilities,
278                                   state->caps_count);
279     for (i = j = 0; i < MIGRATION_CAPABILITY__MAX; i++) {
280         if (should_validate_capability(i) && s->capabilities[i]) {
281             state->capabilities[j++] = i;
282         }
283     }
284     state->uuid = qemu_uuid;
285 
286     return 0;
287 }
288 
289 static int configuration_post_save(void *opaque)
290 {
291     SaveState *state = opaque;
292 
293     g_free(state->capabilities);
294     state->capabilities = NULL;
295     state->caps_count = 0;
296     return 0;
297 }
298 
299 static int configuration_pre_load(void *opaque)
300 {
301     SaveState *state = opaque;
302 
303     /* If there is no target-page-bits subsection it means the source
304      * predates the variable-target-page-bits support and is using the
305      * minimum possible value for this CPU.
306      */
307     state->target_page_bits = qemu_target_page_bits_min();
308     return 0;
309 }
310 
311 static bool configuration_validate_capabilities(SaveState *state)
312 {
313     bool ret = true;
314     MigrationState *s = migrate_get_current();
315     unsigned long *source_caps_bm;
316     int i;
317 
318     source_caps_bm = bitmap_new(MIGRATION_CAPABILITY__MAX);
319     for (i = 0; i < state->caps_count; i++) {
320         MigrationCapability capability = state->capabilities[i];
321         set_bit(capability, source_caps_bm);
322     }
323 
324     for (i = 0; i < MIGRATION_CAPABILITY__MAX; i++) {
325         bool source_state, target_state;
326         if (!should_validate_capability(i)) {
327             continue;
328         }
329         source_state = test_bit(i, source_caps_bm);
330         target_state = s->capabilities[i];
331         if (source_state != target_state) {
332             error_report("Capability %s is %s, but received capability is %s",
333                          MigrationCapability_str(i),
334                          target_state ? "on" : "off",
335                          source_state ? "on" : "off");
336             ret = false;
337             /* Don't break here to report all failed capabilities */
338         }
339     }
340 
341     g_free(source_caps_bm);
342     return ret;
343 }
344 
345 static int configuration_post_load(void *opaque, int version_id)
346 {
347     SaveState *state = opaque;
348     const char *current_name = MACHINE_GET_CLASS(current_machine)->name;
349     int ret = 0;
350 
351     if (strncmp(state->name, current_name, state->len) != 0) {
352         error_report("Machine type received is '%.*s' and local is '%s'",
353                      (int) state->len, state->name, current_name);
354         ret = -EINVAL;
355         goto out;
356     }
357 
358     if (state->target_page_bits != qemu_target_page_bits()) {
359         error_report("Received TARGET_PAGE_BITS is %d but local is %d",
360                      state->target_page_bits, qemu_target_page_bits());
361         ret = -EINVAL;
362         goto out;
363     }
364 
365     if (!configuration_validate_capabilities(state)) {
366         ret = -EINVAL;
367         goto out;
368     }
369 
370 out:
371     g_free((void *)state->name);
372     state->name = NULL;
373     state->len = 0;
374     g_free(state->capabilities);
375     state->capabilities = NULL;
376     state->caps_count = 0;
377 
378     return ret;
379 }
380 
381 static int get_capability(QEMUFile *f, void *pv, size_t size,
382                           const VMStateField *field)
383 {
384     MigrationCapability *capability = pv;
385     char capability_str[UINT8_MAX + 1];
386     uint8_t len;
387     int i;
388 
389     len = qemu_get_byte(f);
390     qemu_get_buffer(f, (uint8_t *)capability_str, len);
391     capability_str[len] = '\0';
392     for (i = 0; i < MIGRATION_CAPABILITY__MAX; i++) {
393         if (!strcmp(MigrationCapability_str(i), capability_str)) {
394             *capability = i;
395             return 0;
396         }
397     }
398     error_report("Received unknown capability %s", capability_str);
399     return -EINVAL;
400 }
401 
402 static int put_capability(QEMUFile *f, void *pv, size_t size,
403                           const VMStateField *field, JSONWriter *vmdesc)
404 {
405     MigrationCapability *capability = pv;
406     const char *capability_str = MigrationCapability_str(*capability);
407     size_t len = strlen(capability_str);
408     assert(len <= UINT8_MAX);
409 
410     qemu_put_byte(f, len);
411     qemu_put_buffer(f, (uint8_t *)capability_str, len);
412     return 0;
413 }
414 
415 static const VMStateInfo vmstate_info_capability = {
416     .name = "capability",
417     .get  = get_capability,
418     .put  = put_capability,
419 };
420 
421 /* The target-page-bits subsection is present only if the
422  * target page size is not the same as the default (ie the
423  * minimum page size for a variable-page-size guest CPU).
424  * If it is present then it contains the actual target page
425  * bits for the machine, and migration will fail if the
426  * two ends don't agree about it.
427  */
428 static bool vmstate_target_page_bits_needed(void *opaque)
429 {
430     return qemu_target_page_bits()
431         > qemu_target_page_bits_min();
432 }
433 
434 static const VMStateDescription vmstate_target_page_bits = {
435     .name = "configuration/target-page-bits",
436     .version_id = 1,
437     .minimum_version_id = 1,
438     .needed = vmstate_target_page_bits_needed,
439     .fields = (VMStateField[]) {
440         VMSTATE_UINT32(target_page_bits, SaveState),
441         VMSTATE_END_OF_LIST()
442     }
443 };
444 
445 static bool vmstate_capabilites_needed(void *opaque)
446 {
447     return get_validatable_capabilities_count() > 0;
448 }
449 
450 static const VMStateDescription vmstate_capabilites = {
451     .name = "configuration/capabilities",
452     .version_id = 1,
453     .minimum_version_id = 1,
454     .needed = vmstate_capabilites_needed,
455     .fields = (VMStateField[]) {
456         VMSTATE_UINT32_V(caps_count, SaveState, 1),
457         VMSTATE_VARRAY_UINT32_ALLOC(capabilities, SaveState, caps_count, 1,
458                                     vmstate_info_capability,
459                                     MigrationCapability),
460         VMSTATE_END_OF_LIST()
461     }
462 };
463 
464 static bool vmstate_uuid_needed(void *opaque)
465 {
466     return qemu_uuid_set && migrate_validate_uuid();
467 }
468 
469 static int vmstate_uuid_post_load(void *opaque, int version_id)
470 {
471     SaveState *state = opaque;
472     char uuid_src[UUID_FMT_LEN + 1];
473     char uuid_dst[UUID_FMT_LEN + 1];
474 
475     if (!qemu_uuid_set) {
476         /*
477          * It's warning because user might not know UUID in some cases,
478          * e.g. load an old snapshot
479          */
480         qemu_uuid_unparse(&state->uuid, uuid_src);
481         warn_report("UUID is received %s, but local uuid isn't set",
482                      uuid_src);
483         return 0;
484     }
485     if (!qemu_uuid_is_equal(&state->uuid, &qemu_uuid)) {
486         qemu_uuid_unparse(&state->uuid, uuid_src);
487         qemu_uuid_unparse(&qemu_uuid, uuid_dst);
488         error_report("UUID received is %s and local is %s", uuid_src, uuid_dst);
489         return -EINVAL;
490     }
491     return 0;
492 }
493 
494 static const VMStateDescription vmstate_uuid = {
495     .name = "configuration/uuid",
496     .version_id = 1,
497     .minimum_version_id = 1,
498     .needed = vmstate_uuid_needed,
499     .post_load = vmstate_uuid_post_load,
500     .fields = (VMStateField[]) {
501         VMSTATE_UINT8_ARRAY_V(uuid.data, SaveState, sizeof(QemuUUID), 1),
502         VMSTATE_END_OF_LIST()
503     }
504 };
505 
506 static const VMStateDescription vmstate_configuration = {
507     .name = "configuration",
508     .version_id = 1,
509     .pre_load = configuration_pre_load,
510     .post_load = configuration_post_load,
511     .pre_save = configuration_pre_save,
512     .post_save = configuration_post_save,
513     .fields = (VMStateField[]) {
514         VMSTATE_UINT32(len, SaveState),
515         VMSTATE_VBUFFER_ALLOC_UINT32(name, SaveState, 0, NULL, len),
516         VMSTATE_END_OF_LIST()
517     },
518     .subsections = (const VMStateDescription *[]) {
519         &vmstate_target_page_bits,
520         &vmstate_capabilites,
521         &vmstate_uuid,
522         NULL
523     }
524 };
525 
526 static void dump_vmstate_vmsd(FILE *out_file,
527                               const VMStateDescription *vmsd, int indent,
528                               bool is_subsection);
529 
530 static void dump_vmstate_vmsf(FILE *out_file, const VMStateField *field,
531                               int indent)
532 {
533     fprintf(out_file, "%*s{\n", indent, "");
534     indent += 2;
535     fprintf(out_file, "%*s\"field\": \"%s\",\n", indent, "", field->name);
536     fprintf(out_file, "%*s\"version_id\": %d,\n", indent, "",
537             field->version_id);
538     fprintf(out_file, "%*s\"field_exists\": %s,\n", indent, "",
539             field->field_exists ? "true" : "false");
540     if (field->flags & VMS_ARRAY) {
541         fprintf(out_file, "%*s\"num\": %d,\n", indent, "", field->num);
542     }
543     fprintf(out_file, "%*s\"size\": %zu", indent, "", field->size);
544     if (field->vmsd != NULL) {
545         fprintf(out_file, ",\n");
546         dump_vmstate_vmsd(out_file, field->vmsd, indent, false);
547     }
548     fprintf(out_file, "\n%*s}", indent - 2, "");
549 }
550 
551 static void dump_vmstate_vmss(FILE *out_file,
552                               const VMStateDescription **subsection,
553                               int indent)
554 {
555     if (*subsection != NULL) {
556         dump_vmstate_vmsd(out_file, *subsection, indent, true);
557     }
558 }
559 
560 static void dump_vmstate_vmsd(FILE *out_file,
561                               const VMStateDescription *vmsd, int indent,
562                               bool is_subsection)
563 {
564     if (is_subsection) {
565         fprintf(out_file, "%*s{\n", indent, "");
566     } else {
567         fprintf(out_file, "%*s\"%s\": {\n", indent, "", "Description");
568     }
569     indent += 2;
570     fprintf(out_file, "%*s\"name\": \"%s\",\n", indent, "", vmsd->name);
571     fprintf(out_file, "%*s\"version_id\": %d,\n", indent, "",
572             vmsd->version_id);
573     fprintf(out_file, "%*s\"minimum_version_id\": %d", indent, "",
574             vmsd->minimum_version_id);
575     if (vmsd->fields != NULL) {
576         const VMStateField *field = vmsd->fields;
577         bool first;
578 
579         fprintf(out_file, ",\n%*s\"Fields\": [\n", indent, "");
580         first = true;
581         while (field->name != NULL) {
582             if (field->flags & VMS_MUST_EXIST) {
583                 /* Ignore VMSTATE_VALIDATE bits; these don't get migrated */
584                 field++;
585                 continue;
586             }
587             if (!first) {
588                 fprintf(out_file, ",\n");
589             }
590             dump_vmstate_vmsf(out_file, field, indent + 2);
591             field++;
592             first = false;
593         }
594         assert(field->flags == VMS_END);
595         fprintf(out_file, "\n%*s]", indent, "");
596     }
597     if (vmsd->subsections != NULL) {
598         const VMStateDescription **subsection = vmsd->subsections;
599         bool first;
600 
601         fprintf(out_file, ",\n%*s\"Subsections\": [\n", indent, "");
602         first = true;
603         while (*subsection != NULL) {
604             if (!first) {
605                 fprintf(out_file, ",\n");
606             }
607             dump_vmstate_vmss(out_file, subsection, indent + 2);
608             subsection++;
609             first = false;
610         }
611         fprintf(out_file, "\n%*s]", indent, "");
612     }
613     fprintf(out_file, "\n%*s}", indent - 2, "");
614 }
615 
616 static void dump_machine_type(FILE *out_file)
617 {
618     MachineClass *mc;
619 
620     mc = MACHINE_GET_CLASS(current_machine);
621 
622     fprintf(out_file, "  \"vmschkmachine\": {\n");
623     fprintf(out_file, "    \"Name\": \"%s\"\n", mc->name);
624     fprintf(out_file, "  },\n");
625 }
626 
627 void dump_vmstate_json_to_file(FILE *out_file)
628 {
629     GSList *list, *elt;
630     bool first;
631 
632     fprintf(out_file, "{\n");
633     dump_machine_type(out_file);
634 
635     first = true;
636     list = object_class_get_list(TYPE_DEVICE, true);
637     for (elt = list; elt; elt = elt->next) {
638         DeviceClass *dc = OBJECT_CLASS_CHECK(DeviceClass, elt->data,
639                                              TYPE_DEVICE);
640         const char *name;
641         int indent = 2;
642 
643         if (!dc->vmsd) {
644             continue;
645         }
646 
647         if (!first) {
648             fprintf(out_file, ",\n");
649         }
650         name = object_class_get_name(OBJECT_CLASS(dc));
651         fprintf(out_file, "%*s\"%s\": {\n", indent, "", name);
652         indent += 2;
653         fprintf(out_file, "%*s\"Name\": \"%s\",\n", indent, "", name);
654         fprintf(out_file, "%*s\"version_id\": %d,\n", indent, "",
655                 dc->vmsd->version_id);
656         fprintf(out_file, "%*s\"minimum_version_id\": %d,\n", indent, "",
657                 dc->vmsd->minimum_version_id);
658 
659         dump_vmstate_vmsd(out_file, dc->vmsd, indent, false);
660 
661         fprintf(out_file, "\n%*s}", indent - 2, "");
662         first = false;
663     }
664     fprintf(out_file, "\n}\n");
665     fclose(out_file);
666     g_slist_free(list);
667 }
668 
669 static uint32_t calculate_new_instance_id(const char *idstr)
670 {
671     SaveStateEntry *se;
672     uint32_t instance_id = 0;
673 
674     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
675         if (strcmp(idstr, se->idstr) == 0
676             && instance_id <= se->instance_id) {
677             instance_id = se->instance_id + 1;
678         }
679     }
680     /* Make sure we never loop over without being noticed */
681     assert(instance_id != VMSTATE_INSTANCE_ID_ANY);
682     return instance_id;
683 }
684 
685 static int calculate_compat_instance_id(const char *idstr)
686 {
687     SaveStateEntry *se;
688     int instance_id = 0;
689 
690     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
691         if (!se->compat) {
692             continue;
693         }
694 
695         if (strcmp(idstr, se->compat->idstr) == 0
696             && instance_id <= se->compat->instance_id) {
697             instance_id = se->compat->instance_id + 1;
698         }
699     }
700     return instance_id;
701 }
702 
703 static inline MigrationPriority save_state_priority(SaveStateEntry *se)
704 {
705     if (se->vmsd) {
706         return se->vmsd->priority;
707     }
708     return MIG_PRI_DEFAULT;
709 }
710 
711 static void savevm_state_handler_insert(SaveStateEntry *nse)
712 {
713     MigrationPriority priority = save_state_priority(nse);
714     SaveStateEntry *se;
715     int i;
716 
717     assert(priority <= MIG_PRI_MAX);
718 
719     for (i = priority - 1; i >= 0; i--) {
720         se = savevm_state.handler_pri_head[i];
721         if (se != NULL) {
722             assert(save_state_priority(se) < priority);
723             break;
724         }
725     }
726 
727     if (i >= 0) {
728         QTAILQ_INSERT_BEFORE(se, nse, entry);
729     } else {
730         QTAILQ_INSERT_TAIL(&savevm_state.handlers, nse, entry);
731     }
732 
733     if (savevm_state.handler_pri_head[priority] == NULL) {
734         savevm_state.handler_pri_head[priority] = nse;
735     }
736 }
737 
738 static void savevm_state_handler_remove(SaveStateEntry *se)
739 {
740     SaveStateEntry *next;
741     MigrationPriority priority = save_state_priority(se);
742 
743     if (se == savevm_state.handler_pri_head[priority]) {
744         next = QTAILQ_NEXT(se, entry);
745         if (next != NULL && save_state_priority(next) == priority) {
746             savevm_state.handler_pri_head[priority] = next;
747         } else {
748             savevm_state.handler_pri_head[priority] = NULL;
749         }
750     }
751     QTAILQ_REMOVE(&savevm_state.handlers, se, entry);
752 }
753 
754 /* TODO: Individual devices generally have very little idea about the rest
755    of the system, so instance_id should be removed/replaced.
756    Meanwhile pass -1 as instance_id if you do not already have a clearly
757    distinguishing id for all instances of your device class. */
758 int register_savevm_live(const char *idstr,
759                          uint32_t instance_id,
760                          int version_id,
761                          const SaveVMHandlers *ops,
762                          void *opaque)
763 {
764     SaveStateEntry *se;
765 
766     se = g_new0(SaveStateEntry, 1);
767     se->version_id = version_id;
768     se->section_id = savevm_state.global_section_id++;
769     se->ops = ops;
770     se->opaque = opaque;
771     se->vmsd = NULL;
772     /* if this is a live_savem then set is_ram */
773     if (ops->save_setup != NULL) {
774         se->is_ram = 1;
775     }
776 
777     pstrcat(se->idstr, sizeof(se->idstr), idstr);
778 
779     if (instance_id == VMSTATE_INSTANCE_ID_ANY) {
780         se->instance_id = calculate_new_instance_id(se->idstr);
781     } else {
782         se->instance_id = instance_id;
783     }
784     assert(!se->compat || se->instance_id == 0);
785     savevm_state_handler_insert(se);
786     return 0;
787 }
788 
789 void unregister_savevm(VMStateIf *obj, const char *idstr, void *opaque)
790 {
791     SaveStateEntry *se, *new_se;
792     char id[256] = "";
793 
794     if (obj) {
795         char *oid = vmstate_if_get_id(obj);
796         if (oid) {
797             pstrcpy(id, sizeof(id), oid);
798             pstrcat(id, sizeof(id), "/");
799             g_free(oid);
800         }
801     }
802     pstrcat(id, sizeof(id), idstr);
803 
804     QTAILQ_FOREACH_SAFE(se, &savevm_state.handlers, entry, new_se) {
805         if (strcmp(se->idstr, id) == 0 && se->opaque == opaque) {
806             savevm_state_handler_remove(se);
807             g_free(se->compat);
808             g_free(se);
809         }
810     }
811 }
812 
813 /*
814  * Perform some basic checks on vmsd's at registration
815  * time.
816  */
817 static void vmstate_check(const VMStateDescription *vmsd)
818 {
819     const VMStateField *field = vmsd->fields;
820     const VMStateDescription **subsection = vmsd->subsections;
821 
822     if (field) {
823         while (field->name) {
824             if (field->flags & (VMS_STRUCT | VMS_VSTRUCT)) {
825                 /* Recurse to sub structures */
826                 vmstate_check(field->vmsd);
827             }
828             /* Carry on */
829             field++;
830         }
831         /* Check for the end of field list canary */
832         if (field->flags != VMS_END) {
833             error_report("VMSTATE not ending with VMS_END: %s", vmsd->name);
834             g_assert_not_reached();
835         }
836     }
837 
838     while (subsection && *subsection) {
839         /*
840          * The name of a subsection should start with the name of the
841          * current object.
842          */
843         assert(!strncmp(vmsd->name, (*subsection)->name, strlen(vmsd->name)));
844         vmstate_check(*subsection);
845         subsection++;
846     }
847 }
848 
849 int vmstate_register_with_alias_id(VMStateIf *obj, uint32_t instance_id,
850                                    const VMStateDescription *vmsd,
851                                    void *opaque, int alias_id,
852                                    int required_for_version,
853                                    Error **errp)
854 {
855     SaveStateEntry *se;
856 
857     /* If this triggers, alias support can be dropped for the vmsd. */
858     assert(alias_id == -1 || required_for_version >= vmsd->minimum_version_id);
859 
860     se = g_new0(SaveStateEntry, 1);
861     se->version_id = vmsd->version_id;
862     se->section_id = savevm_state.global_section_id++;
863     se->opaque = opaque;
864     se->vmsd = vmsd;
865     se->alias_id = alias_id;
866 
867     if (obj) {
868         char *id = vmstate_if_get_id(obj);
869         if (id) {
870             if (snprintf(se->idstr, sizeof(se->idstr), "%s/", id) >=
871                 sizeof(se->idstr)) {
872                 error_setg(errp, "Path too long for VMState (%s)", id);
873                 g_free(id);
874                 g_free(se);
875 
876                 return -1;
877             }
878             g_free(id);
879 
880             se->compat = g_new0(CompatEntry, 1);
881             pstrcpy(se->compat->idstr, sizeof(se->compat->idstr), vmsd->name);
882             se->compat->instance_id = instance_id == VMSTATE_INSTANCE_ID_ANY ?
883                          calculate_compat_instance_id(vmsd->name) : instance_id;
884             instance_id = VMSTATE_INSTANCE_ID_ANY;
885         }
886     }
887     pstrcat(se->idstr, sizeof(se->idstr), vmsd->name);
888 
889     if (instance_id == VMSTATE_INSTANCE_ID_ANY) {
890         se->instance_id = calculate_new_instance_id(se->idstr);
891     } else {
892         se->instance_id = instance_id;
893     }
894 
895     /* Perform a recursive sanity check during the test runs */
896     if (qtest_enabled()) {
897         vmstate_check(vmsd);
898     }
899     assert(!se->compat || se->instance_id == 0);
900     savevm_state_handler_insert(se);
901     return 0;
902 }
903 
904 void vmstate_unregister(VMStateIf *obj, const VMStateDescription *vmsd,
905                         void *opaque)
906 {
907     SaveStateEntry *se, *new_se;
908 
909     QTAILQ_FOREACH_SAFE(se, &savevm_state.handlers, entry, new_se) {
910         if (se->vmsd == vmsd && se->opaque == opaque) {
911             savevm_state_handler_remove(se);
912             g_free(se->compat);
913             g_free(se);
914         }
915     }
916 }
917 
918 static int vmstate_load(QEMUFile *f, SaveStateEntry *se)
919 {
920     trace_vmstate_load(se->idstr, se->vmsd ? se->vmsd->name : "(old)");
921     if (!se->vmsd) {         /* Old style */
922         return se->ops->load_state(f, se->opaque, se->load_version_id);
923     }
924     return vmstate_load_state(f, se->vmsd, se->opaque, se->load_version_id);
925 }
926 
927 static void vmstate_save_old_style(QEMUFile *f, SaveStateEntry *se,
928                                    JSONWriter *vmdesc)
929 {
930     uint64_t old_offset = qemu_file_transferred_noflush(f);
931     se->ops->save_state(f, se->opaque);
932     uint64_t size = qemu_file_transferred_noflush(f) - old_offset;
933 
934     if (vmdesc) {
935         json_writer_int64(vmdesc, "size", size);
936         json_writer_start_array(vmdesc, "fields");
937         json_writer_start_object(vmdesc, NULL);
938         json_writer_str(vmdesc, "name", "data");
939         json_writer_int64(vmdesc, "size", size);
940         json_writer_str(vmdesc, "type", "buffer");
941         json_writer_end_object(vmdesc);
942         json_writer_end_array(vmdesc);
943     }
944 }
945 
946 /*
947  * Write the header for device section (QEMU_VM_SECTION START/END/PART/FULL)
948  */
949 static void save_section_header(QEMUFile *f, SaveStateEntry *se,
950                                 uint8_t section_type)
951 {
952     qemu_put_byte(f, section_type);
953     qemu_put_be32(f, se->section_id);
954 
955     if (section_type == QEMU_VM_SECTION_FULL ||
956         section_type == QEMU_VM_SECTION_START) {
957         /* ID string */
958         size_t len = strlen(se->idstr);
959         qemu_put_byte(f, len);
960         qemu_put_buffer(f, (uint8_t *)se->idstr, len);
961 
962         qemu_put_be32(f, se->instance_id);
963         qemu_put_be32(f, se->version_id);
964     }
965 }
966 
967 /*
968  * Write a footer onto device sections that catches cases misformatted device
969  * sections.
970  */
971 static void save_section_footer(QEMUFile *f, SaveStateEntry *se)
972 {
973     if (migrate_get_current()->send_section_footer) {
974         qemu_put_byte(f, QEMU_VM_SECTION_FOOTER);
975         qemu_put_be32(f, se->section_id);
976     }
977 }
978 
979 static int vmstate_save(QEMUFile *f, SaveStateEntry *se, JSONWriter *vmdesc)
980 {
981     int ret;
982 
983     if ((!se->ops || !se->ops->save_state) && !se->vmsd) {
984         return 0;
985     }
986     if (se->vmsd && !vmstate_save_needed(se->vmsd, se->opaque)) {
987         trace_savevm_section_skip(se->idstr, se->section_id);
988         return 0;
989     }
990 
991     trace_savevm_section_start(se->idstr, se->section_id);
992     save_section_header(f, se, QEMU_VM_SECTION_FULL);
993     if (vmdesc) {
994         json_writer_start_object(vmdesc, NULL);
995         json_writer_str(vmdesc, "name", se->idstr);
996         json_writer_int64(vmdesc, "instance_id", se->instance_id);
997     }
998 
999     trace_vmstate_save(se->idstr, se->vmsd ? se->vmsd->name : "(old)");
1000     if (!se->vmsd) {
1001         vmstate_save_old_style(f, se, vmdesc);
1002     } else {
1003         ret = vmstate_save_state(f, se->vmsd, se->opaque, vmdesc);
1004         if (ret) {
1005             return ret;
1006         }
1007     }
1008 
1009     trace_savevm_section_end(se->idstr, se->section_id, 0);
1010     save_section_footer(f, se);
1011     if (vmdesc) {
1012         json_writer_end_object(vmdesc);
1013     }
1014     return 0;
1015 }
1016 /**
1017  * qemu_savevm_command_send: Send a 'QEMU_VM_COMMAND' type element with the
1018  *                           command and associated data.
1019  *
1020  * @f: File to send command on
1021  * @command: Command type to send
1022  * @len: Length of associated data
1023  * @data: Data associated with command.
1024  */
1025 static void qemu_savevm_command_send(QEMUFile *f,
1026                                      enum qemu_vm_cmd command,
1027                                      uint16_t len,
1028                                      uint8_t *data)
1029 {
1030     trace_savevm_command_send(command, len);
1031     qemu_put_byte(f, QEMU_VM_COMMAND);
1032     qemu_put_be16(f, (uint16_t)command);
1033     qemu_put_be16(f, len);
1034     qemu_put_buffer(f, data, len);
1035     qemu_fflush(f);
1036 }
1037 
1038 void qemu_savevm_send_colo_enable(QEMUFile *f)
1039 {
1040     trace_savevm_send_colo_enable();
1041     qemu_savevm_command_send(f, MIG_CMD_ENABLE_COLO, 0, NULL);
1042 }
1043 
1044 void qemu_savevm_send_ping(QEMUFile *f, uint32_t value)
1045 {
1046     uint32_t buf;
1047 
1048     trace_savevm_send_ping(value);
1049     buf = cpu_to_be32(value);
1050     qemu_savevm_command_send(f, MIG_CMD_PING, sizeof(value), (uint8_t *)&buf);
1051 }
1052 
1053 void qemu_savevm_send_open_return_path(QEMUFile *f)
1054 {
1055     trace_savevm_send_open_return_path();
1056     qemu_savevm_command_send(f, MIG_CMD_OPEN_RETURN_PATH, 0, NULL);
1057 }
1058 
1059 /* We have a buffer of data to send; we don't want that all to be loaded
1060  * by the command itself, so the command contains just the length of the
1061  * extra buffer that we then send straight after it.
1062  * TODO: Must be a better way to organise that
1063  *
1064  * Returns:
1065  *    0 on success
1066  *    -ve on error
1067  */
1068 int qemu_savevm_send_packaged(QEMUFile *f, const uint8_t *buf, size_t len)
1069 {
1070     uint32_t tmp;
1071 
1072     if (len > MAX_VM_CMD_PACKAGED_SIZE) {
1073         error_report("%s: Unreasonably large packaged state: %zu",
1074                      __func__, len);
1075         return -1;
1076     }
1077 
1078     tmp = cpu_to_be32(len);
1079 
1080     trace_qemu_savevm_send_packaged();
1081     qemu_savevm_command_send(f, MIG_CMD_PACKAGED, 4, (uint8_t *)&tmp);
1082 
1083     qemu_put_buffer(f, buf, len);
1084 
1085     return 0;
1086 }
1087 
1088 /* Send prior to any postcopy transfer */
1089 void qemu_savevm_send_postcopy_advise(QEMUFile *f)
1090 {
1091     if (migrate_postcopy_ram()) {
1092         uint64_t tmp[2];
1093         tmp[0] = cpu_to_be64(ram_pagesize_summary());
1094         tmp[1] = cpu_to_be64(qemu_target_page_size());
1095 
1096         trace_qemu_savevm_send_postcopy_advise();
1097         qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_ADVISE,
1098                                  16, (uint8_t *)tmp);
1099     } else {
1100         qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_ADVISE, 0, NULL);
1101     }
1102 }
1103 
1104 /* Sent prior to starting the destination running in postcopy, discard pages
1105  * that have already been sent but redirtied on the source.
1106  * CMD_POSTCOPY_RAM_DISCARD consist of:
1107  *      byte   version (0)
1108  *      byte   Length of name field (not including 0)
1109  *  n x byte   RAM block name
1110  *      byte   0 terminator (just for safety)
1111  *  n x        Byte ranges within the named RAMBlock
1112  *      be64   Start of the range
1113  *      be64   Length
1114  *
1115  *  name:  RAMBlock name that these entries are part of
1116  *  len: Number of page entries
1117  *  start_list: 'len' addresses
1118  *  length_list: 'len' addresses
1119  *
1120  */
1121 void qemu_savevm_send_postcopy_ram_discard(QEMUFile *f, const char *name,
1122                                            uint16_t len,
1123                                            uint64_t *start_list,
1124                                            uint64_t *length_list)
1125 {
1126     uint8_t *buf;
1127     uint16_t tmplen;
1128     uint16_t t;
1129     size_t name_len = strlen(name);
1130 
1131     trace_qemu_savevm_send_postcopy_ram_discard(name, len);
1132     assert(name_len < 256);
1133     buf = g_malloc0(1 + 1 + name_len + 1 + (8 + 8) * len);
1134     buf[0] = postcopy_ram_discard_version;
1135     buf[1] = name_len;
1136     memcpy(buf + 2, name, name_len);
1137     tmplen = 2 + name_len;
1138     buf[tmplen++] = '\0';
1139 
1140     for (t = 0; t < len; t++) {
1141         stq_be_p(buf + tmplen, start_list[t]);
1142         tmplen += 8;
1143         stq_be_p(buf + tmplen, length_list[t]);
1144         tmplen += 8;
1145     }
1146     qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_RAM_DISCARD, tmplen, buf);
1147     g_free(buf);
1148 }
1149 
1150 /* Get the destination into a state where it can receive postcopy data. */
1151 void qemu_savevm_send_postcopy_listen(QEMUFile *f)
1152 {
1153     trace_savevm_send_postcopy_listen();
1154     qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_LISTEN, 0, NULL);
1155 }
1156 
1157 /* Kick the destination into running */
1158 void qemu_savevm_send_postcopy_run(QEMUFile *f)
1159 {
1160     trace_savevm_send_postcopy_run();
1161     qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_RUN, 0, NULL);
1162 }
1163 
1164 void qemu_savevm_send_postcopy_resume(QEMUFile *f)
1165 {
1166     trace_savevm_send_postcopy_resume();
1167     qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_RESUME, 0, NULL);
1168 }
1169 
1170 void qemu_savevm_send_recv_bitmap(QEMUFile *f, char *block_name)
1171 {
1172     size_t len;
1173     char buf[256];
1174 
1175     trace_savevm_send_recv_bitmap(block_name);
1176 
1177     buf[0] = len = strlen(block_name);
1178     memcpy(buf + 1, block_name, len);
1179 
1180     qemu_savevm_command_send(f, MIG_CMD_RECV_BITMAP, len + 1, (uint8_t *)buf);
1181 }
1182 
1183 bool qemu_savevm_state_blocked(Error **errp)
1184 {
1185     SaveStateEntry *se;
1186 
1187     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1188         if (se->vmsd && se->vmsd->unmigratable) {
1189             error_setg(errp, "State blocked by non-migratable device '%s'",
1190                        se->idstr);
1191             return true;
1192         }
1193     }
1194     return false;
1195 }
1196 
1197 void qemu_savevm_non_migratable_list(strList **reasons)
1198 {
1199     SaveStateEntry *se;
1200 
1201     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1202         if (se->vmsd && se->vmsd->unmigratable) {
1203             QAPI_LIST_PREPEND(*reasons,
1204                               g_strdup_printf("non-migratable device: %s",
1205                                               se->idstr));
1206         }
1207     }
1208 }
1209 
1210 void qemu_savevm_state_header(QEMUFile *f)
1211 {
1212     trace_savevm_state_header();
1213     qemu_put_be32(f, QEMU_VM_FILE_MAGIC);
1214     qemu_put_be32(f, QEMU_VM_FILE_VERSION);
1215 
1216     if (migrate_get_current()->send_configuration) {
1217         qemu_put_byte(f, QEMU_VM_CONFIGURATION);
1218         vmstate_save_state(f, &vmstate_configuration, &savevm_state, 0);
1219     }
1220 }
1221 
1222 bool qemu_savevm_state_guest_unplug_pending(void)
1223 {
1224     SaveStateEntry *se;
1225 
1226     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1227         if (se->vmsd && se->vmsd->dev_unplug_pending &&
1228             se->vmsd->dev_unplug_pending(se->opaque)) {
1229             return true;
1230         }
1231     }
1232 
1233     return false;
1234 }
1235 
1236 void qemu_savevm_state_setup(QEMUFile *f)
1237 {
1238     MigrationState *ms = migrate_get_current();
1239     SaveStateEntry *se;
1240     Error *local_err = NULL;
1241     int ret;
1242 
1243     ms->vmdesc = json_writer_new(false);
1244     json_writer_start_object(ms->vmdesc, NULL);
1245     json_writer_int64(ms->vmdesc, "page_size", qemu_target_page_size());
1246     json_writer_start_array(ms->vmdesc, "devices");
1247 
1248     trace_savevm_state_setup();
1249     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1250         if (se->vmsd && se->vmsd->early_setup) {
1251             ret = vmstate_save(f, se, ms->vmdesc);
1252             if (ret) {
1253                 qemu_file_set_error(f, ret);
1254                 break;
1255             }
1256             continue;
1257         }
1258 
1259         if (!se->ops || !se->ops->save_setup) {
1260             continue;
1261         }
1262         if (se->ops->is_active) {
1263             if (!se->ops->is_active(se->opaque)) {
1264                 continue;
1265             }
1266         }
1267         save_section_header(f, se, QEMU_VM_SECTION_START);
1268 
1269         ret = se->ops->save_setup(f, se->opaque);
1270         save_section_footer(f, se);
1271         if (ret < 0) {
1272             qemu_file_set_error(f, ret);
1273             break;
1274         }
1275     }
1276 
1277     if (precopy_notify(PRECOPY_NOTIFY_SETUP, &local_err)) {
1278         error_report_err(local_err);
1279     }
1280 }
1281 
1282 int qemu_savevm_state_resume_prepare(MigrationState *s)
1283 {
1284     SaveStateEntry *se;
1285     int ret;
1286 
1287     trace_savevm_state_resume_prepare();
1288 
1289     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1290         if (!se->ops || !se->ops->resume_prepare) {
1291             continue;
1292         }
1293         if (se->ops->is_active) {
1294             if (!se->ops->is_active(se->opaque)) {
1295                 continue;
1296             }
1297         }
1298         ret = se->ops->resume_prepare(s, se->opaque);
1299         if (ret < 0) {
1300             return ret;
1301         }
1302     }
1303 
1304     return 0;
1305 }
1306 
1307 /*
1308  * this function has three return values:
1309  *   negative: there was one error, and we have -errno.
1310  *   0 : We haven't finished, caller have to go again
1311  *   1 : We have finished, we can go to complete phase
1312  */
1313 int qemu_savevm_state_iterate(QEMUFile *f, bool postcopy)
1314 {
1315     SaveStateEntry *se;
1316     int ret = 1;
1317 
1318     trace_savevm_state_iterate();
1319     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1320         if (!se->ops || !se->ops->save_live_iterate) {
1321             continue;
1322         }
1323         if (se->ops->is_active &&
1324             !se->ops->is_active(se->opaque)) {
1325             continue;
1326         }
1327         if (se->ops->is_active_iterate &&
1328             !se->ops->is_active_iterate(se->opaque)) {
1329             continue;
1330         }
1331         /*
1332          * In the postcopy phase, any device that doesn't know how to
1333          * do postcopy should have saved it's state in the _complete
1334          * call that's already run, it might get confused if we call
1335          * iterate afterwards.
1336          */
1337         if (postcopy &&
1338             !(se->ops->has_postcopy && se->ops->has_postcopy(se->opaque))) {
1339             continue;
1340         }
1341         if (migration_rate_exceeded(f)) {
1342             return 0;
1343         }
1344         trace_savevm_section_start(se->idstr, se->section_id);
1345 
1346         save_section_header(f, se, QEMU_VM_SECTION_PART);
1347 
1348         ret = se->ops->save_live_iterate(f, se->opaque);
1349         trace_savevm_section_end(se->idstr, se->section_id, ret);
1350         save_section_footer(f, se);
1351 
1352         if (ret < 0) {
1353             error_report("failed to save SaveStateEntry with id(name): "
1354                          "%d(%s): %d",
1355                          se->section_id, se->idstr, ret);
1356             qemu_file_set_error(f, ret);
1357         }
1358         if (ret <= 0) {
1359             /* Do not proceed to the next vmstate before this one reported
1360                completion of the current stage. This serializes the migration
1361                and reduces the probability that a faster changing state is
1362                synchronized over and over again. */
1363             break;
1364         }
1365     }
1366     return ret;
1367 }
1368 
1369 static bool should_send_vmdesc(void)
1370 {
1371     MachineState *machine = MACHINE(qdev_get_machine());
1372     bool in_postcopy = migration_in_postcopy();
1373     return !machine->suppress_vmdesc && !in_postcopy;
1374 }
1375 
1376 /*
1377  * Calls the save_live_complete_postcopy methods
1378  * causing the last few pages to be sent immediately and doing any associated
1379  * cleanup.
1380  * Note postcopy also calls qemu_savevm_state_complete_precopy to complete
1381  * all the other devices, but that happens at the point we switch to postcopy.
1382  */
1383 void qemu_savevm_state_complete_postcopy(QEMUFile *f)
1384 {
1385     SaveStateEntry *se;
1386     int ret;
1387 
1388     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1389         if (!se->ops || !se->ops->save_live_complete_postcopy) {
1390             continue;
1391         }
1392         if (se->ops->is_active) {
1393             if (!se->ops->is_active(se->opaque)) {
1394                 continue;
1395             }
1396         }
1397         trace_savevm_section_start(se->idstr, se->section_id);
1398         /* Section type */
1399         qemu_put_byte(f, QEMU_VM_SECTION_END);
1400         qemu_put_be32(f, se->section_id);
1401 
1402         ret = se->ops->save_live_complete_postcopy(f, se->opaque);
1403         trace_savevm_section_end(se->idstr, se->section_id, ret);
1404         save_section_footer(f, se);
1405         if (ret < 0) {
1406             qemu_file_set_error(f, ret);
1407             return;
1408         }
1409     }
1410 
1411     qemu_put_byte(f, QEMU_VM_EOF);
1412     qemu_fflush(f);
1413 }
1414 
1415 static
1416 int qemu_savevm_state_complete_precopy_iterable(QEMUFile *f, bool in_postcopy)
1417 {
1418     SaveStateEntry *se;
1419     int ret;
1420 
1421     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1422         if (!se->ops ||
1423             (in_postcopy && se->ops->has_postcopy &&
1424              se->ops->has_postcopy(se->opaque)) ||
1425             !se->ops->save_live_complete_precopy) {
1426             continue;
1427         }
1428 
1429         if (se->ops->is_active) {
1430             if (!se->ops->is_active(se->opaque)) {
1431                 continue;
1432             }
1433         }
1434         trace_savevm_section_start(se->idstr, se->section_id);
1435 
1436         save_section_header(f, se, QEMU_VM_SECTION_END);
1437 
1438         ret = se->ops->save_live_complete_precopy(f, se->opaque);
1439         trace_savevm_section_end(se->idstr, se->section_id, ret);
1440         save_section_footer(f, se);
1441         if (ret < 0) {
1442             qemu_file_set_error(f, ret);
1443             return -1;
1444         }
1445     }
1446 
1447     return 0;
1448 }
1449 
1450 int qemu_savevm_state_complete_precopy_non_iterable(QEMUFile *f,
1451                                                     bool in_postcopy,
1452                                                     bool inactivate_disks)
1453 {
1454     MigrationState *ms = migrate_get_current();
1455     JSONWriter *vmdesc = ms->vmdesc;
1456     int vmdesc_len;
1457     SaveStateEntry *se;
1458     int ret;
1459 
1460     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1461         if (se->vmsd && se->vmsd->early_setup) {
1462             /* Already saved during qemu_savevm_state_setup(). */
1463             continue;
1464         }
1465 
1466         ret = vmstate_save(f, se, vmdesc);
1467         if (ret) {
1468             qemu_file_set_error(f, ret);
1469             return ret;
1470         }
1471     }
1472 
1473     if (inactivate_disks) {
1474         /* Inactivate before sending QEMU_VM_EOF so that the
1475          * bdrv_activate_all() on the other end won't fail. */
1476         ret = bdrv_inactivate_all();
1477         if (ret) {
1478             error_report("%s: bdrv_inactivate_all() failed (%d)",
1479                          __func__, ret);
1480             qemu_file_set_error(f, ret);
1481             return ret;
1482         }
1483     }
1484     if (!in_postcopy) {
1485         /* Postcopy stream will still be going */
1486         qemu_put_byte(f, QEMU_VM_EOF);
1487     }
1488 
1489     json_writer_end_array(vmdesc);
1490     json_writer_end_object(vmdesc);
1491     vmdesc_len = strlen(json_writer_get(vmdesc));
1492 
1493     if (should_send_vmdesc()) {
1494         qemu_put_byte(f, QEMU_VM_VMDESCRIPTION);
1495         qemu_put_be32(f, vmdesc_len);
1496         qemu_put_buffer(f, (uint8_t *)json_writer_get(vmdesc), vmdesc_len);
1497     }
1498 
1499     /* Free it now to detect any inconsistencies. */
1500     json_writer_free(vmdesc);
1501     ms->vmdesc = NULL;
1502 
1503     return 0;
1504 }
1505 
1506 int qemu_savevm_state_complete_precopy(QEMUFile *f, bool iterable_only,
1507                                        bool inactivate_disks)
1508 {
1509     int ret;
1510     Error *local_err = NULL;
1511     bool in_postcopy = migration_in_postcopy();
1512 
1513     if (precopy_notify(PRECOPY_NOTIFY_COMPLETE, &local_err)) {
1514         error_report_err(local_err);
1515     }
1516 
1517     trace_savevm_state_complete_precopy();
1518 
1519     cpu_synchronize_all_states();
1520 
1521     if (!in_postcopy || iterable_only) {
1522         ret = qemu_savevm_state_complete_precopy_iterable(f, in_postcopy);
1523         if (ret) {
1524             return ret;
1525         }
1526     }
1527 
1528     if (iterable_only) {
1529         goto flush;
1530     }
1531 
1532     ret = qemu_savevm_state_complete_precopy_non_iterable(f, in_postcopy,
1533                                                           inactivate_disks);
1534     if (ret) {
1535         return ret;
1536     }
1537 
1538 flush:
1539     qemu_fflush(f);
1540     return 0;
1541 }
1542 
1543 /* Give an estimate of the amount left to be transferred,
1544  * the result is split into the amount for units that can and
1545  * for units that can't do postcopy.
1546  */
1547 void qemu_savevm_state_pending_estimate(uint64_t *must_precopy,
1548                                         uint64_t *can_postcopy)
1549 {
1550     SaveStateEntry *se;
1551 
1552     *must_precopy = 0;
1553     *can_postcopy = 0;
1554 
1555     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1556         if (!se->ops || !se->ops->state_pending_estimate) {
1557             continue;
1558         }
1559         if (se->ops->is_active) {
1560             if (!se->ops->is_active(se->opaque)) {
1561                 continue;
1562             }
1563         }
1564         se->ops->state_pending_estimate(se->opaque, must_precopy, can_postcopy);
1565     }
1566 }
1567 
1568 void qemu_savevm_state_pending_exact(uint64_t *must_precopy,
1569                                      uint64_t *can_postcopy)
1570 {
1571     SaveStateEntry *se;
1572 
1573     *must_precopy = 0;
1574     *can_postcopy = 0;
1575 
1576     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1577         if (!se->ops || !se->ops->state_pending_exact) {
1578             continue;
1579         }
1580         if (se->ops->is_active) {
1581             if (!se->ops->is_active(se->opaque)) {
1582                 continue;
1583             }
1584         }
1585         se->ops->state_pending_exact(se->opaque, must_precopy, can_postcopy);
1586     }
1587 }
1588 
1589 void qemu_savevm_state_cleanup(void)
1590 {
1591     SaveStateEntry *se;
1592     Error *local_err = NULL;
1593 
1594     if (precopy_notify(PRECOPY_NOTIFY_CLEANUP, &local_err)) {
1595         error_report_err(local_err);
1596     }
1597 
1598     trace_savevm_state_cleanup();
1599     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1600         if (se->ops && se->ops->save_cleanup) {
1601             se->ops->save_cleanup(se->opaque);
1602         }
1603     }
1604 }
1605 
1606 static int qemu_savevm_state(QEMUFile *f, Error **errp)
1607 {
1608     int ret;
1609     MigrationState *ms = migrate_get_current();
1610     MigrationStatus status;
1611 
1612     if (migration_is_running(ms->state)) {
1613         error_setg(errp, QERR_MIGRATION_ACTIVE);
1614         return -EINVAL;
1615     }
1616 
1617     if (migrate_block()) {
1618         error_setg(errp, "Block migration and snapshots are incompatible");
1619         return -EINVAL;
1620     }
1621 
1622     migrate_init(ms);
1623     memset(&mig_stats, 0, sizeof(mig_stats));
1624     memset(&compression_counters, 0, sizeof(compression_counters));
1625     reset_vfio_bytes_transferred();
1626     ms->to_dst_file = f;
1627 
1628     qemu_mutex_unlock_iothread();
1629     qemu_savevm_state_header(f);
1630     qemu_savevm_state_setup(f);
1631     qemu_mutex_lock_iothread();
1632 
1633     while (qemu_file_get_error(f) == 0) {
1634         if (qemu_savevm_state_iterate(f, false) > 0) {
1635             break;
1636         }
1637     }
1638 
1639     ret = qemu_file_get_error(f);
1640     if (ret == 0) {
1641         qemu_savevm_state_complete_precopy(f, false, false);
1642         ret = qemu_file_get_error(f);
1643     }
1644     qemu_savevm_state_cleanup();
1645     if (ret != 0) {
1646         error_setg_errno(errp, -ret, "Error while writing VM state");
1647     }
1648 
1649     if (ret != 0) {
1650         status = MIGRATION_STATUS_FAILED;
1651     } else {
1652         status = MIGRATION_STATUS_COMPLETED;
1653     }
1654     migrate_set_state(&ms->state, MIGRATION_STATUS_SETUP, status);
1655 
1656     /* f is outer parameter, it should not stay in global migration state after
1657      * this function finished */
1658     ms->to_dst_file = NULL;
1659 
1660     return ret;
1661 }
1662 
1663 void qemu_savevm_live_state(QEMUFile *f)
1664 {
1665     /* save QEMU_VM_SECTION_END section */
1666     qemu_savevm_state_complete_precopy(f, true, false);
1667     qemu_put_byte(f, QEMU_VM_EOF);
1668 }
1669 
1670 int qemu_save_device_state(QEMUFile *f)
1671 {
1672     SaveStateEntry *se;
1673 
1674     if (!migration_in_colo_state()) {
1675         qemu_put_be32(f, QEMU_VM_FILE_MAGIC);
1676         qemu_put_be32(f, QEMU_VM_FILE_VERSION);
1677     }
1678     cpu_synchronize_all_states();
1679 
1680     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1681         int ret;
1682 
1683         if (se->is_ram) {
1684             continue;
1685         }
1686         ret = vmstate_save(f, se, NULL);
1687         if (ret) {
1688             return ret;
1689         }
1690     }
1691 
1692     qemu_put_byte(f, QEMU_VM_EOF);
1693 
1694     return qemu_file_get_error(f);
1695 }
1696 
1697 static SaveStateEntry *find_se(const char *idstr, uint32_t instance_id)
1698 {
1699     SaveStateEntry *se;
1700 
1701     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1702         if (!strcmp(se->idstr, idstr) &&
1703             (instance_id == se->instance_id ||
1704              instance_id == se->alias_id))
1705             return se;
1706         /* Migrating from an older version? */
1707         if (strstr(se->idstr, idstr) && se->compat) {
1708             if (!strcmp(se->compat->idstr, idstr) &&
1709                 (instance_id == se->compat->instance_id ||
1710                  instance_id == se->alias_id))
1711                 return se;
1712         }
1713     }
1714     return NULL;
1715 }
1716 
1717 enum LoadVMExitCodes {
1718     /* Allow a command to quit all layers of nested loadvm loops */
1719     LOADVM_QUIT     =  1,
1720 };
1721 
1722 /* ------ incoming postcopy messages ------ */
1723 /* 'advise' arrives before any transfers just to tell us that a postcopy
1724  * *might* happen - it might be skipped if precopy transferred everything
1725  * quickly.
1726  */
1727 static int loadvm_postcopy_handle_advise(MigrationIncomingState *mis,
1728                                          uint16_t len)
1729 {
1730     PostcopyState ps = postcopy_state_set(POSTCOPY_INCOMING_ADVISE);
1731     uint64_t remote_pagesize_summary, local_pagesize_summary, remote_tps;
1732     size_t page_size = qemu_target_page_size();
1733     Error *local_err = NULL;
1734 
1735     trace_loadvm_postcopy_handle_advise();
1736     if (ps != POSTCOPY_INCOMING_NONE) {
1737         error_report("CMD_POSTCOPY_ADVISE in wrong postcopy state (%d)", ps);
1738         return -1;
1739     }
1740 
1741     switch (len) {
1742     case 0:
1743         if (migrate_postcopy_ram()) {
1744             error_report("RAM postcopy is enabled but have 0 byte advise");
1745             return -EINVAL;
1746         }
1747         return 0;
1748     case 8 + 8:
1749         if (!migrate_postcopy_ram()) {
1750             error_report("RAM postcopy is disabled but have 16 byte advise");
1751             return -EINVAL;
1752         }
1753         break;
1754     default:
1755         error_report("CMD_POSTCOPY_ADVISE invalid length (%d)", len);
1756         return -EINVAL;
1757     }
1758 
1759     if (!postcopy_ram_supported_by_host(mis, &local_err)) {
1760         error_report_err(local_err);
1761         postcopy_state_set(POSTCOPY_INCOMING_NONE);
1762         return -1;
1763     }
1764 
1765     remote_pagesize_summary = qemu_get_be64(mis->from_src_file);
1766     local_pagesize_summary = ram_pagesize_summary();
1767 
1768     if (remote_pagesize_summary != local_pagesize_summary)  {
1769         /*
1770          * This detects two potential causes of mismatch:
1771          *   a) A mismatch in host page sizes
1772          *      Some combinations of mismatch are probably possible but it gets
1773          *      a bit more complicated.  In particular we need to place whole
1774          *      host pages on the dest at once, and we need to ensure that we
1775          *      handle dirtying to make sure we never end up sending part of
1776          *      a hostpage on it's own.
1777          *   b) The use of different huge page sizes on source/destination
1778          *      a more fine grain test is performed during RAM block migration
1779          *      but this test here causes a nice early clear failure, and
1780          *      also fails when passed to an older qemu that doesn't
1781          *      do huge pages.
1782          */
1783         error_report("Postcopy needs matching RAM page sizes (s=%" PRIx64
1784                                                              " d=%" PRIx64 ")",
1785                      remote_pagesize_summary, local_pagesize_summary);
1786         return -1;
1787     }
1788 
1789     remote_tps = qemu_get_be64(mis->from_src_file);
1790     if (remote_tps != page_size) {
1791         /*
1792          * Again, some differences could be dealt with, but for now keep it
1793          * simple.
1794          */
1795         error_report("Postcopy needs matching target page sizes (s=%d d=%zd)",
1796                      (int)remote_tps, page_size);
1797         return -1;
1798     }
1799 
1800     if (postcopy_notify(POSTCOPY_NOTIFY_INBOUND_ADVISE, &local_err)) {
1801         error_report_err(local_err);
1802         return -1;
1803     }
1804 
1805     if (ram_postcopy_incoming_init(mis)) {
1806         return -1;
1807     }
1808 
1809     return 0;
1810 }
1811 
1812 /* After postcopy we will be told to throw some pages away since they're
1813  * dirty and will have to be demand fetched.  Must happen before CPU is
1814  * started.
1815  * There can be 0..many of these messages, each encoding multiple pages.
1816  */
1817 static int loadvm_postcopy_ram_handle_discard(MigrationIncomingState *mis,
1818                                               uint16_t len)
1819 {
1820     int tmp;
1821     char ramid[256];
1822     PostcopyState ps = postcopy_state_get();
1823 
1824     trace_loadvm_postcopy_ram_handle_discard();
1825 
1826     switch (ps) {
1827     case POSTCOPY_INCOMING_ADVISE:
1828         /* 1st discard */
1829         tmp = postcopy_ram_prepare_discard(mis);
1830         if (tmp) {
1831             return tmp;
1832         }
1833         break;
1834 
1835     case POSTCOPY_INCOMING_DISCARD:
1836         /* Expected state */
1837         break;
1838 
1839     default:
1840         error_report("CMD_POSTCOPY_RAM_DISCARD in wrong postcopy state (%d)",
1841                      ps);
1842         return -1;
1843     }
1844     /* We're expecting a
1845      *    Version (0)
1846      *    a RAM ID string (length byte, name, 0 term)
1847      *    then at least 1 16 byte chunk
1848     */
1849     if (len < (1 + 1 + 1 + 1 + 2 * 8)) {
1850         error_report("CMD_POSTCOPY_RAM_DISCARD invalid length (%d)", len);
1851         return -1;
1852     }
1853 
1854     tmp = qemu_get_byte(mis->from_src_file);
1855     if (tmp != postcopy_ram_discard_version) {
1856         error_report("CMD_POSTCOPY_RAM_DISCARD invalid version (%d)", tmp);
1857         return -1;
1858     }
1859 
1860     if (!qemu_get_counted_string(mis->from_src_file, ramid)) {
1861         error_report("CMD_POSTCOPY_RAM_DISCARD Failed to read RAMBlock ID");
1862         return -1;
1863     }
1864     tmp = qemu_get_byte(mis->from_src_file);
1865     if (tmp != 0) {
1866         error_report("CMD_POSTCOPY_RAM_DISCARD missing nil (%d)", tmp);
1867         return -1;
1868     }
1869 
1870     len -= 3 + strlen(ramid);
1871     if (len % 16) {
1872         error_report("CMD_POSTCOPY_RAM_DISCARD invalid length (%d)", len);
1873         return -1;
1874     }
1875     trace_loadvm_postcopy_ram_handle_discard_header(ramid, len);
1876     while (len) {
1877         uint64_t start_addr, block_length;
1878         start_addr = qemu_get_be64(mis->from_src_file);
1879         block_length = qemu_get_be64(mis->from_src_file);
1880 
1881         len -= 16;
1882         int ret = ram_discard_range(ramid, start_addr, block_length);
1883         if (ret) {
1884             return ret;
1885         }
1886     }
1887     trace_loadvm_postcopy_ram_handle_discard_end();
1888 
1889     return 0;
1890 }
1891 
1892 /*
1893  * Triggered by a postcopy_listen command; this thread takes over reading
1894  * the input stream, leaving the main thread free to carry on loading the rest
1895  * of the device state (from RAM).
1896  * (TODO:This could do with being in a postcopy file - but there again it's
1897  * just another input loop, not that postcopy specific)
1898  */
1899 static void *postcopy_ram_listen_thread(void *opaque)
1900 {
1901     MigrationIncomingState *mis = migration_incoming_get_current();
1902     QEMUFile *f = mis->from_src_file;
1903     int load_res;
1904     MigrationState *migr = migrate_get_current();
1905 
1906     object_ref(OBJECT(migr));
1907 
1908     migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE,
1909                                    MIGRATION_STATUS_POSTCOPY_ACTIVE);
1910     qemu_sem_post(&mis->thread_sync_sem);
1911     trace_postcopy_ram_listen_thread_start();
1912 
1913     rcu_register_thread();
1914     /*
1915      * Because we're a thread and not a coroutine we can't yield
1916      * in qemu_file, and thus we must be blocking now.
1917      */
1918     qemu_file_set_blocking(f, true);
1919     load_res = qemu_loadvm_state_main(f, mis);
1920 
1921     /*
1922      * This is tricky, but, mis->from_src_file can change after it
1923      * returns, when postcopy recovery happened. In the future, we may
1924      * want a wrapper for the QEMUFile handle.
1925      */
1926     f = mis->from_src_file;
1927 
1928     /* And non-blocking again so we don't block in any cleanup */
1929     qemu_file_set_blocking(f, false);
1930 
1931     trace_postcopy_ram_listen_thread_exit();
1932     if (load_res < 0) {
1933         qemu_file_set_error(f, load_res);
1934         dirty_bitmap_mig_cancel_incoming();
1935         if (postcopy_state_get() == POSTCOPY_INCOMING_RUNNING &&
1936             !migrate_postcopy_ram() && migrate_dirty_bitmaps())
1937         {
1938             error_report("%s: loadvm failed during postcopy: %d. All states "
1939                          "are migrated except dirty bitmaps. Some dirty "
1940                          "bitmaps may be lost, and present migrated dirty "
1941                          "bitmaps are correctly migrated and valid.",
1942                          __func__, load_res);
1943             load_res = 0; /* prevent further exit() */
1944         } else {
1945             error_report("%s: loadvm failed: %d", __func__, load_res);
1946             migrate_set_state(&mis->state, MIGRATION_STATUS_POSTCOPY_ACTIVE,
1947                                            MIGRATION_STATUS_FAILED);
1948         }
1949     }
1950     if (load_res >= 0) {
1951         /*
1952          * This looks good, but it's possible that the device loading in the
1953          * main thread hasn't finished yet, and so we might not be in 'RUN'
1954          * state yet; wait for the end of the main thread.
1955          */
1956         qemu_event_wait(&mis->main_thread_load_event);
1957     }
1958     postcopy_ram_incoming_cleanup(mis);
1959 
1960     if (load_res < 0) {
1961         /*
1962          * If something went wrong then we have a bad state so exit;
1963          * depending how far we got it might be possible at this point
1964          * to leave the guest running and fire MCEs for pages that never
1965          * arrived as a desperate recovery step.
1966          */
1967         rcu_unregister_thread();
1968         exit(EXIT_FAILURE);
1969     }
1970 
1971     migrate_set_state(&mis->state, MIGRATION_STATUS_POSTCOPY_ACTIVE,
1972                                    MIGRATION_STATUS_COMPLETED);
1973     /*
1974      * If everything has worked fine, then the main thread has waited
1975      * for us to start, and we're the last use of the mis.
1976      * (If something broke then qemu will have to exit anyway since it's
1977      * got a bad migration state).
1978      */
1979     migration_incoming_state_destroy();
1980     qemu_loadvm_state_cleanup();
1981 
1982     rcu_unregister_thread();
1983     mis->have_listen_thread = false;
1984     postcopy_state_set(POSTCOPY_INCOMING_END);
1985 
1986     object_unref(OBJECT(migr));
1987 
1988     return NULL;
1989 }
1990 
1991 /* After this message we must be able to immediately receive postcopy data */
1992 static int loadvm_postcopy_handle_listen(MigrationIncomingState *mis)
1993 {
1994     PostcopyState ps = postcopy_state_set(POSTCOPY_INCOMING_LISTENING);
1995     Error *local_err = NULL;
1996 
1997     trace_loadvm_postcopy_handle_listen("enter");
1998 
1999     if (ps != POSTCOPY_INCOMING_ADVISE && ps != POSTCOPY_INCOMING_DISCARD) {
2000         error_report("CMD_POSTCOPY_LISTEN in wrong postcopy state (%d)", ps);
2001         return -1;
2002     }
2003     if (ps == POSTCOPY_INCOMING_ADVISE) {
2004         /*
2005          * A rare case, we entered listen without having to do any discards,
2006          * so do the setup that's normally done at the time of the 1st discard.
2007          */
2008         if (migrate_postcopy_ram()) {
2009             postcopy_ram_prepare_discard(mis);
2010         }
2011     }
2012 
2013     trace_loadvm_postcopy_handle_listen("after discard");
2014 
2015     /*
2016      * Sensitise RAM - can now generate requests for blocks that don't exist
2017      * However, at this point the CPU shouldn't be running, and the IO
2018      * shouldn't be doing anything yet so don't actually expect requests
2019      */
2020     if (migrate_postcopy_ram()) {
2021         if (postcopy_ram_incoming_setup(mis)) {
2022             postcopy_ram_incoming_cleanup(mis);
2023             return -1;
2024         }
2025     }
2026 
2027     trace_loadvm_postcopy_handle_listen("after uffd");
2028 
2029     if (postcopy_notify(POSTCOPY_NOTIFY_INBOUND_LISTEN, &local_err)) {
2030         error_report_err(local_err);
2031         return -1;
2032     }
2033 
2034     mis->have_listen_thread = true;
2035     postcopy_thread_create(mis, &mis->listen_thread, "postcopy/listen",
2036                            postcopy_ram_listen_thread, QEMU_THREAD_DETACHED);
2037     trace_loadvm_postcopy_handle_listen("return");
2038 
2039     return 0;
2040 }
2041 
2042 static void loadvm_postcopy_handle_run_bh(void *opaque)
2043 {
2044     Error *local_err = NULL;
2045     MigrationIncomingState *mis = opaque;
2046 
2047     trace_loadvm_postcopy_handle_run_bh("enter");
2048 
2049     /* TODO we should move all of this lot into postcopy_ram.c or a shared code
2050      * in migration.c
2051      */
2052     cpu_synchronize_all_post_init();
2053 
2054     trace_loadvm_postcopy_handle_run_bh("after cpu sync");
2055 
2056     qemu_announce_self(&mis->announce_timer, migrate_announce_params());
2057 
2058     trace_loadvm_postcopy_handle_run_bh("after announce");
2059 
2060     /* Make sure all file formats throw away their mutable metadata.
2061      * If we get an error here, just don't restart the VM yet. */
2062     bdrv_activate_all(&local_err);
2063     if (local_err) {
2064         error_report_err(local_err);
2065         local_err = NULL;
2066         autostart = false;
2067     }
2068 
2069     trace_loadvm_postcopy_handle_run_bh("after invalidate cache");
2070 
2071     dirty_bitmap_mig_before_vm_start();
2072 
2073     if (autostart) {
2074         /* Hold onto your hats, starting the CPU */
2075         vm_start();
2076     } else {
2077         /* leave it paused and let management decide when to start the CPU */
2078         runstate_set(RUN_STATE_PAUSED);
2079     }
2080 
2081     qemu_bh_delete(mis->bh);
2082 
2083     trace_loadvm_postcopy_handle_run_bh("return");
2084 }
2085 
2086 /* After all discards we can start running and asking for pages */
2087 static int loadvm_postcopy_handle_run(MigrationIncomingState *mis)
2088 {
2089     PostcopyState ps = postcopy_state_get();
2090 
2091     trace_loadvm_postcopy_handle_run();
2092     if (ps != POSTCOPY_INCOMING_LISTENING) {
2093         error_report("CMD_POSTCOPY_RUN in wrong postcopy state (%d)", ps);
2094         return -1;
2095     }
2096 
2097     postcopy_state_set(POSTCOPY_INCOMING_RUNNING);
2098     mis->bh = qemu_bh_new(loadvm_postcopy_handle_run_bh, mis);
2099     qemu_bh_schedule(mis->bh);
2100 
2101     /* We need to finish reading the stream from the package
2102      * and also stop reading anything more from the stream that loaded the
2103      * package (since it's now being read by the listener thread).
2104      * LOADVM_QUIT will quit all the layers of nested loadvm loops.
2105      */
2106     return LOADVM_QUIT;
2107 }
2108 
2109 /* We must be with page_request_mutex held */
2110 static gboolean postcopy_sync_page_req(gpointer key, gpointer value,
2111                                        gpointer data)
2112 {
2113     MigrationIncomingState *mis = data;
2114     void *host_addr = (void *) key;
2115     ram_addr_t rb_offset;
2116     RAMBlock *rb;
2117     int ret;
2118 
2119     rb = qemu_ram_block_from_host(host_addr, true, &rb_offset);
2120     if (!rb) {
2121         /*
2122          * This should _never_ happen.  However be nice for a migrating VM to
2123          * not crash/assert.  Post an error (note: intended to not use *_once
2124          * because we do want to see all the illegal addresses; and this can
2125          * never be triggered by the guest so we're safe) and move on next.
2126          */
2127         error_report("%s: illegal host addr %p", __func__, host_addr);
2128         /* Try the next entry */
2129         return FALSE;
2130     }
2131 
2132     ret = migrate_send_rp_message_req_pages(mis, rb, rb_offset);
2133     if (ret) {
2134         /* Please refer to above comment. */
2135         error_report("%s: send rp message failed for addr %p",
2136                      __func__, host_addr);
2137         return FALSE;
2138     }
2139 
2140     trace_postcopy_page_req_sync(host_addr);
2141 
2142     return FALSE;
2143 }
2144 
2145 static void migrate_send_rp_req_pages_pending(MigrationIncomingState *mis)
2146 {
2147     WITH_QEMU_LOCK_GUARD(&mis->page_request_mutex) {
2148         g_tree_foreach(mis->page_requested, postcopy_sync_page_req, mis);
2149     }
2150 }
2151 
2152 static int loadvm_postcopy_handle_resume(MigrationIncomingState *mis)
2153 {
2154     if (mis->state != MIGRATION_STATUS_POSTCOPY_RECOVER) {
2155         error_report("%s: illegal resume received", __func__);
2156         /* Don't fail the load, only for this. */
2157         return 0;
2158     }
2159 
2160     /*
2161      * Reset the last_rb before we resend any page req to source again, since
2162      * the source should have it reset already.
2163      */
2164     mis->last_rb = NULL;
2165 
2166     /*
2167      * This means source VM is ready to resume the postcopy migration.
2168      */
2169     migrate_set_state(&mis->state, MIGRATION_STATUS_POSTCOPY_RECOVER,
2170                       MIGRATION_STATUS_POSTCOPY_ACTIVE);
2171 
2172     trace_loadvm_postcopy_handle_resume();
2173 
2174     /* Tell source that "we are ready" */
2175     migrate_send_rp_resume_ack(mis, MIGRATION_RESUME_ACK_VALUE);
2176 
2177     /*
2178      * After a postcopy recovery, the source should have lost the postcopy
2179      * queue, or potentially the requested pages could have been lost during
2180      * the network down phase.  Let's re-sync with the source VM by re-sending
2181      * all the pending pages that we eagerly need, so these threads won't get
2182      * blocked too long due to the recovery.
2183      *
2184      * Without this procedure, the faulted destination VM threads (waiting for
2185      * page requests right before the postcopy is interrupted) can keep hanging
2186      * until the pages are sent by the source during the background copying of
2187      * pages, or another thread faulted on the same address accidentally.
2188      */
2189     migrate_send_rp_req_pages_pending(mis);
2190 
2191     /*
2192      * It's time to switch state and release the fault thread to continue
2193      * service page faults.  Note that this should be explicitly after the
2194      * above call to migrate_send_rp_req_pages_pending().  In short:
2195      * migrate_send_rp_message_req_pages() is not thread safe, yet.
2196      */
2197     qemu_sem_post(&mis->postcopy_pause_sem_fault);
2198 
2199     if (migrate_postcopy_preempt()) {
2200         /*
2201          * The preempt channel will be created in async manner, now let's
2202          * wait for it and make sure it's created.
2203          */
2204         qemu_sem_wait(&mis->postcopy_qemufile_dst_done);
2205         assert(mis->postcopy_qemufile_dst);
2206         /* Kick the fast ram load thread too */
2207         qemu_sem_post(&mis->postcopy_pause_sem_fast_load);
2208     }
2209 
2210     return 0;
2211 }
2212 
2213 /**
2214  * Immediately following this command is a blob of data containing an embedded
2215  * chunk of migration stream; read it and load it.
2216  *
2217  * @mis: Incoming state
2218  * @length: Length of packaged data to read
2219  *
2220  * Returns: Negative values on error
2221  *
2222  */
2223 static int loadvm_handle_cmd_packaged(MigrationIncomingState *mis)
2224 {
2225     int ret;
2226     size_t length;
2227     QIOChannelBuffer *bioc;
2228 
2229     length = qemu_get_be32(mis->from_src_file);
2230     trace_loadvm_handle_cmd_packaged(length);
2231 
2232     if (length > MAX_VM_CMD_PACKAGED_SIZE) {
2233         error_report("Unreasonably large packaged state: %zu", length);
2234         return -1;
2235     }
2236 
2237     bioc = qio_channel_buffer_new(length);
2238     qio_channel_set_name(QIO_CHANNEL(bioc), "migration-loadvm-buffer");
2239     ret = qemu_get_buffer(mis->from_src_file,
2240                           bioc->data,
2241                           length);
2242     if (ret != length) {
2243         object_unref(OBJECT(bioc));
2244         error_report("CMD_PACKAGED: Buffer receive fail ret=%d length=%zu",
2245                      ret, length);
2246         return (ret < 0) ? ret : -EAGAIN;
2247     }
2248     bioc->usage += length;
2249     trace_loadvm_handle_cmd_packaged_received(ret);
2250 
2251     QEMUFile *packf = qemu_file_new_input(QIO_CHANNEL(bioc));
2252 
2253     ret = qemu_loadvm_state_main(packf, mis);
2254     trace_loadvm_handle_cmd_packaged_main(ret);
2255     qemu_fclose(packf);
2256     object_unref(OBJECT(bioc));
2257 
2258     return ret;
2259 }
2260 
2261 /*
2262  * Handle request that source requests for recved_bitmap on
2263  * destination. Payload format:
2264  *
2265  * len (1 byte) + ramblock_name (<255 bytes)
2266  */
2267 static int loadvm_handle_recv_bitmap(MigrationIncomingState *mis,
2268                                      uint16_t len)
2269 {
2270     QEMUFile *file = mis->from_src_file;
2271     RAMBlock *rb;
2272     char block_name[256];
2273     size_t cnt;
2274 
2275     cnt = qemu_get_counted_string(file, block_name);
2276     if (!cnt) {
2277         error_report("%s: failed to read block name", __func__);
2278         return -EINVAL;
2279     }
2280 
2281     /* Validate before using the data */
2282     if (qemu_file_get_error(file)) {
2283         return qemu_file_get_error(file);
2284     }
2285 
2286     if (len != cnt + 1) {
2287         error_report("%s: invalid payload length (%d)", __func__, len);
2288         return -EINVAL;
2289     }
2290 
2291     rb = qemu_ram_block_by_name(block_name);
2292     if (!rb) {
2293         error_report("%s: block '%s' not found", __func__, block_name);
2294         return -EINVAL;
2295     }
2296 
2297     migrate_send_rp_recv_bitmap(mis, block_name);
2298 
2299     trace_loadvm_handle_recv_bitmap(block_name);
2300 
2301     return 0;
2302 }
2303 
2304 static int loadvm_process_enable_colo(MigrationIncomingState *mis)
2305 {
2306     int ret = migration_incoming_enable_colo();
2307 
2308     if (!ret) {
2309         ret = colo_init_ram_cache();
2310         if (ret) {
2311             migration_incoming_disable_colo();
2312         }
2313     }
2314     return ret;
2315 }
2316 
2317 /*
2318  * Process an incoming 'QEMU_VM_COMMAND'
2319  * 0           just a normal return
2320  * LOADVM_QUIT All good, but exit the loop
2321  * <0          Error
2322  */
2323 static int loadvm_process_command(QEMUFile *f)
2324 {
2325     MigrationIncomingState *mis = migration_incoming_get_current();
2326     uint16_t cmd;
2327     uint16_t len;
2328     uint32_t tmp32;
2329 
2330     cmd = qemu_get_be16(f);
2331     len = qemu_get_be16(f);
2332 
2333     /* Check validity before continue processing of cmds */
2334     if (qemu_file_get_error(f)) {
2335         return qemu_file_get_error(f);
2336     }
2337 
2338     if (cmd >= MIG_CMD_MAX || cmd == MIG_CMD_INVALID) {
2339         error_report("MIG_CMD 0x%x unknown (len 0x%x)", cmd, len);
2340         return -EINVAL;
2341     }
2342 
2343     trace_loadvm_process_command(mig_cmd_args[cmd].name, len);
2344 
2345     if (mig_cmd_args[cmd].len != -1 && mig_cmd_args[cmd].len != len) {
2346         error_report("%s received with bad length - expecting %zu, got %d",
2347                      mig_cmd_args[cmd].name,
2348                      (size_t)mig_cmd_args[cmd].len, len);
2349         return -ERANGE;
2350     }
2351 
2352     switch (cmd) {
2353     case MIG_CMD_OPEN_RETURN_PATH:
2354         if (mis->to_src_file) {
2355             error_report("CMD_OPEN_RETURN_PATH called when RP already open");
2356             /* Not really a problem, so don't give up */
2357             return 0;
2358         }
2359         mis->to_src_file = qemu_file_get_return_path(f);
2360         if (!mis->to_src_file) {
2361             error_report("CMD_OPEN_RETURN_PATH failed");
2362             return -1;
2363         }
2364 
2365         /*
2366          * Switchover ack is enabled but no device uses it, so send an ACK to
2367          * source that it's OK to switchover. Do it here, after return path has
2368          * been created.
2369          */
2370         if (migrate_switchover_ack() && !mis->switchover_ack_pending_num) {
2371             int ret = migrate_send_rp_switchover_ack(mis);
2372             if (ret) {
2373                 error_report(
2374                     "Could not send switchover ack RP MSG, err %d (%s)", ret,
2375                     strerror(-ret));
2376                 return ret;
2377             }
2378         }
2379         break;
2380 
2381     case MIG_CMD_PING:
2382         tmp32 = qemu_get_be32(f);
2383         trace_loadvm_process_command_ping(tmp32);
2384         if (!mis->to_src_file) {
2385             error_report("CMD_PING (0x%x) received with no return path",
2386                          tmp32);
2387             return -1;
2388         }
2389         migrate_send_rp_pong(mis, tmp32);
2390         break;
2391 
2392     case MIG_CMD_PACKAGED:
2393         return loadvm_handle_cmd_packaged(mis);
2394 
2395     case MIG_CMD_POSTCOPY_ADVISE:
2396         return loadvm_postcopy_handle_advise(mis, len);
2397 
2398     case MIG_CMD_POSTCOPY_LISTEN:
2399         return loadvm_postcopy_handle_listen(mis);
2400 
2401     case MIG_CMD_POSTCOPY_RUN:
2402         return loadvm_postcopy_handle_run(mis);
2403 
2404     case MIG_CMD_POSTCOPY_RAM_DISCARD:
2405         return loadvm_postcopy_ram_handle_discard(mis, len);
2406 
2407     case MIG_CMD_POSTCOPY_RESUME:
2408         return loadvm_postcopy_handle_resume(mis);
2409 
2410     case MIG_CMD_RECV_BITMAP:
2411         return loadvm_handle_recv_bitmap(mis, len);
2412 
2413     case MIG_CMD_ENABLE_COLO:
2414         return loadvm_process_enable_colo(mis);
2415     }
2416 
2417     return 0;
2418 }
2419 
2420 /*
2421  * Read a footer off the wire and check that it matches the expected section
2422  *
2423  * Returns: true if the footer was good
2424  *          false if there is a problem (and calls error_report to say why)
2425  */
2426 static bool check_section_footer(QEMUFile *f, SaveStateEntry *se)
2427 {
2428     int ret;
2429     uint8_t read_mark;
2430     uint32_t read_section_id;
2431 
2432     if (!migrate_get_current()->send_section_footer) {
2433         /* No footer to check */
2434         return true;
2435     }
2436 
2437     read_mark = qemu_get_byte(f);
2438 
2439     ret = qemu_file_get_error(f);
2440     if (ret) {
2441         error_report("%s: Read section footer failed: %d",
2442                      __func__, ret);
2443         return false;
2444     }
2445 
2446     if (read_mark != QEMU_VM_SECTION_FOOTER) {
2447         error_report("Missing section footer for %s", se->idstr);
2448         return false;
2449     }
2450 
2451     read_section_id = qemu_get_be32(f);
2452     if (read_section_id != se->load_section_id) {
2453         error_report("Mismatched section id in footer for %s -"
2454                      " read 0x%x expected 0x%x",
2455                      se->idstr, read_section_id, se->load_section_id);
2456         return false;
2457     }
2458 
2459     /* All good */
2460     return true;
2461 }
2462 
2463 static int
2464 qemu_loadvm_section_start_full(QEMUFile *f, MigrationIncomingState *mis)
2465 {
2466     uint32_t instance_id, version_id, section_id;
2467     SaveStateEntry *se;
2468     char idstr[256];
2469     int ret;
2470 
2471     /* Read section start */
2472     section_id = qemu_get_be32(f);
2473     if (!qemu_get_counted_string(f, idstr)) {
2474         error_report("Unable to read ID string for section %u",
2475                      section_id);
2476         return -EINVAL;
2477     }
2478     instance_id = qemu_get_be32(f);
2479     version_id = qemu_get_be32(f);
2480 
2481     ret = qemu_file_get_error(f);
2482     if (ret) {
2483         error_report("%s: Failed to read instance/version ID: %d",
2484                      __func__, ret);
2485         return ret;
2486     }
2487 
2488     trace_qemu_loadvm_state_section_startfull(section_id, idstr,
2489             instance_id, version_id);
2490     /* Find savevm section */
2491     se = find_se(idstr, instance_id);
2492     if (se == NULL) {
2493         error_report("Unknown savevm section or instance '%s' %"PRIu32". "
2494                      "Make sure that your current VM setup matches your "
2495                      "saved VM setup, including any hotplugged devices",
2496                      idstr, instance_id);
2497         return -EINVAL;
2498     }
2499 
2500     /* Validate version */
2501     if (version_id > se->version_id) {
2502         error_report("savevm: unsupported version %d for '%s' v%d",
2503                      version_id, idstr, se->version_id);
2504         return -EINVAL;
2505     }
2506     se->load_version_id = version_id;
2507     se->load_section_id = section_id;
2508 
2509     /* Validate if it is a device's state */
2510     if (xen_enabled() && se->is_ram) {
2511         error_report("loadvm: %s RAM loading not allowed on Xen", idstr);
2512         return -EINVAL;
2513     }
2514 
2515     ret = vmstate_load(f, se);
2516     if (ret < 0) {
2517         error_report("error while loading state for instance 0x%"PRIx32" of"
2518                      " device '%s'", instance_id, idstr);
2519         return ret;
2520     }
2521     if (!check_section_footer(f, se)) {
2522         return -EINVAL;
2523     }
2524 
2525     return 0;
2526 }
2527 
2528 static int
2529 qemu_loadvm_section_part_end(QEMUFile *f, MigrationIncomingState *mis)
2530 {
2531     uint32_t section_id;
2532     SaveStateEntry *se;
2533     int ret;
2534 
2535     section_id = qemu_get_be32(f);
2536 
2537     ret = qemu_file_get_error(f);
2538     if (ret) {
2539         error_report("%s: Failed to read section ID: %d",
2540                      __func__, ret);
2541         return ret;
2542     }
2543 
2544     trace_qemu_loadvm_state_section_partend(section_id);
2545     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
2546         if (se->load_section_id == section_id) {
2547             break;
2548         }
2549     }
2550     if (se == NULL) {
2551         error_report("Unknown savevm section %d", section_id);
2552         return -EINVAL;
2553     }
2554 
2555     ret = vmstate_load(f, se);
2556     if (ret < 0) {
2557         error_report("error while loading state section id %d(%s)",
2558                      section_id, se->idstr);
2559         return ret;
2560     }
2561     if (!check_section_footer(f, se)) {
2562         return -EINVAL;
2563     }
2564 
2565     return 0;
2566 }
2567 
2568 static int qemu_loadvm_state_header(QEMUFile *f)
2569 {
2570     unsigned int v;
2571     int ret;
2572 
2573     v = qemu_get_be32(f);
2574     if (v != QEMU_VM_FILE_MAGIC) {
2575         error_report("Not a migration stream");
2576         return -EINVAL;
2577     }
2578 
2579     v = qemu_get_be32(f);
2580     if (v == QEMU_VM_FILE_VERSION_COMPAT) {
2581         error_report("SaveVM v2 format is obsolete and don't work anymore");
2582         return -ENOTSUP;
2583     }
2584     if (v != QEMU_VM_FILE_VERSION) {
2585         error_report("Unsupported migration stream version");
2586         return -ENOTSUP;
2587     }
2588 
2589     if (migrate_get_current()->send_configuration) {
2590         if (qemu_get_byte(f) != QEMU_VM_CONFIGURATION) {
2591             error_report("Configuration section missing");
2592             qemu_loadvm_state_cleanup();
2593             return -EINVAL;
2594         }
2595         ret = vmstate_load_state(f, &vmstate_configuration, &savevm_state, 0);
2596 
2597         if (ret) {
2598             qemu_loadvm_state_cleanup();
2599             return ret;
2600         }
2601     }
2602     return 0;
2603 }
2604 
2605 static void qemu_loadvm_state_switchover_ack_needed(MigrationIncomingState *mis)
2606 {
2607     SaveStateEntry *se;
2608 
2609     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
2610         if (!se->ops || !se->ops->switchover_ack_needed) {
2611             continue;
2612         }
2613 
2614         if (se->ops->switchover_ack_needed(se->opaque)) {
2615             mis->switchover_ack_pending_num++;
2616         }
2617     }
2618 
2619     trace_loadvm_state_switchover_ack_needed(mis->switchover_ack_pending_num);
2620 }
2621 
2622 static int qemu_loadvm_state_setup(QEMUFile *f)
2623 {
2624     SaveStateEntry *se;
2625     int ret;
2626 
2627     trace_loadvm_state_setup();
2628     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
2629         if (!se->ops || !se->ops->load_setup) {
2630             continue;
2631         }
2632         if (se->ops->is_active) {
2633             if (!se->ops->is_active(se->opaque)) {
2634                 continue;
2635             }
2636         }
2637 
2638         ret = se->ops->load_setup(f, se->opaque);
2639         if (ret < 0) {
2640             qemu_file_set_error(f, ret);
2641             error_report("Load state of device %s failed", se->idstr);
2642             return ret;
2643         }
2644     }
2645     return 0;
2646 }
2647 
2648 void qemu_loadvm_state_cleanup(void)
2649 {
2650     SaveStateEntry *se;
2651 
2652     trace_loadvm_state_cleanup();
2653     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
2654         if (se->ops && se->ops->load_cleanup) {
2655             se->ops->load_cleanup(se->opaque);
2656         }
2657     }
2658 }
2659 
2660 /* Return true if we should continue the migration, or false. */
2661 static bool postcopy_pause_incoming(MigrationIncomingState *mis)
2662 {
2663     int i;
2664 
2665     trace_postcopy_pause_incoming();
2666 
2667     assert(migrate_postcopy_ram());
2668 
2669     /*
2670      * Unregister yank with either from/to src would work, since ioc behind it
2671      * is the same
2672      */
2673     migration_ioc_unregister_yank_from_file(mis->from_src_file);
2674 
2675     assert(mis->from_src_file);
2676     qemu_file_shutdown(mis->from_src_file);
2677     qemu_fclose(mis->from_src_file);
2678     mis->from_src_file = NULL;
2679 
2680     assert(mis->to_src_file);
2681     qemu_file_shutdown(mis->to_src_file);
2682     qemu_mutex_lock(&mis->rp_mutex);
2683     qemu_fclose(mis->to_src_file);
2684     mis->to_src_file = NULL;
2685     qemu_mutex_unlock(&mis->rp_mutex);
2686 
2687     /*
2688      * NOTE: this must happen before reset the PostcopyTmpPages below,
2689      * otherwise it's racy to reset those fields when the fast load thread
2690      * can be accessing it in parallel.
2691      */
2692     if (mis->postcopy_qemufile_dst) {
2693         qemu_file_shutdown(mis->postcopy_qemufile_dst);
2694         /* Take the mutex to make sure the fast ram load thread halted */
2695         qemu_mutex_lock(&mis->postcopy_prio_thread_mutex);
2696         migration_ioc_unregister_yank_from_file(mis->postcopy_qemufile_dst);
2697         qemu_fclose(mis->postcopy_qemufile_dst);
2698         mis->postcopy_qemufile_dst = NULL;
2699         qemu_mutex_unlock(&mis->postcopy_prio_thread_mutex);
2700     }
2701 
2702     migrate_set_state(&mis->state, MIGRATION_STATUS_POSTCOPY_ACTIVE,
2703                       MIGRATION_STATUS_POSTCOPY_PAUSED);
2704 
2705     /* Notify the fault thread for the invalidated file handle */
2706     postcopy_fault_thread_notify(mis);
2707 
2708     /*
2709      * If network is interrupted, any temp page we received will be useless
2710      * because we didn't mark them as "received" in receivedmap.  After a
2711      * proper recovery later (which will sync src dirty bitmap with receivedmap
2712      * on dest) these cached small pages will be resent again.
2713      */
2714     for (i = 0; i < mis->postcopy_channels; i++) {
2715         postcopy_temp_page_reset(&mis->postcopy_tmp_pages[i]);
2716     }
2717 
2718     error_report("Detected IO failure for postcopy. "
2719                  "Migration paused.");
2720 
2721     while (mis->state == MIGRATION_STATUS_POSTCOPY_PAUSED) {
2722         qemu_sem_wait(&mis->postcopy_pause_sem_dst);
2723     }
2724 
2725     trace_postcopy_pause_incoming_continued();
2726 
2727     return true;
2728 }
2729 
2730 int qemu_loadvm_state_main(QEMUFile *f, MigrationIncomingState *mis)
2731 {
2732     uint8_t section_type;
2733     int ret = 0;
2734 
2735 retry:
2736     while (true) {
2737         section_type = qemu_get_byte(f);
2738 
2739         ret = qemu_file_get_error_obj_any(f, mis->postcopy_qemufile_dst, NULL);
2740         if (ret) {
2741             break;
2742         }
2743 
2744         trace_qemu_loadvm_state_section(section_type);
2745         switch (section_type) {
2746         case QEMU_VM_SECTION_START:
2747         case QEMU_VM_SECTION_FULL:
2748             ret = qemu_loadvm_section_start_full(f, mis);
2749             if (ret < 0) {
2750                 goto out;
2751             }
2752             break;
2753         case QEMU_VM_SECTION_PART:
2754         case QEMU_VM_SECTION_END:
2755             ret = qemu_loadvm_section_part_end(f, mis);
2756             if (ret < 0) {
2757                 goto out;
2758             }
2759             break;
2760         case QEMU_VM_COMMAND:
2761             ret = loadvm_process_command(f);
2762             trace_qemu_loadvm_state_section_command(ret);
2763             if ((ret < 0) || (ret == LOADVM_QUIT)) {
2764                 goto out;
2765             }
2766             break;
2767         case QEMU_VM_EOF:
2768             /* This is the end of migration */
2769             goto out;
2770         default:
2771             error_report("Unknown savevm section type %d", section_type);
2772             ret = -EINVAL;
2773             goto out;
2774         }
2775     }
2776 
2777 out:
2778     if (ret < 0) {
2779         qemu_file_set_error(f, ret);
2780 
2781         /* Cancel bitmaps incoming regardless of recovery */
2782         dirty_bitmap_mig_cancel_incoming();
2783 
2784         /*
2785          * If we are during an active postcopy, then we pause instead
2786          * of bail out to at least keep the VM's dirty data.  Note
2787          * that POSTCOPY_INCOMING_LISTENING stage is still not enough,
2788          * during which we're still receiving device states and we
2789          * still haven't yet started the VM on destination.
2790          *
2791          * Only RAM postcopy supports recovery. Still, if RAM postcopy is
2792          * enabled, canceled bitmaps postcopy will not affect RAM postcopy
2793          * recovering.
2794          */
2795         if (postcopy_state_get() == POSTCOPY_INCOMING_RUNNING &&
2796             migrate_postcopy_ram() && postcopy_pause_incoming(mis)) {
2797             /* Reset f to point to the newly created channel */
2798             f = mis->from_src_file;
2799             goto retry;
2800         }
2801     }
2802     return ret;
2803 }
2804 
2805 int qemu_loadvm_state(QEMUFile *f)
2806 {
2807     MigrationIncomingState *mis = migration_incoming_get_current();
2808     Error *local_err = NULL;
2809     int ret;
2810 
2811     if (qemu_savevm_state_blocked(&local_err)) {
2812         error_report_err(local_err);
2813         return -EINVAL;
2814     }
2815 
2816     ret = qemu_loadvm_state_header(f);
2817     if (ret) {
2818         return ret;
2819     }
2820 
2821     if (qemu_loadvm_state_setup(f) != 0) {
2822         return -EINVAL;
2823     }
2824 
2825     if (migrate_switchover_ack()) {
2826         qemu_loadvm_state_switchover_ack_needed(mis);
2827     }
2828 
2829     cpu_synchronize_all_pre_loadvm();
2830 
2831     ret = qemu_loadvm_state_main(f, mis);
2832     qemu_event_set(&mis->main_thread_load_event);
2833 
2834     trace_qemu_loadvm_state_post_main(ret);
2835 
2836     if (mis->have_listen_thread) {
2837         /* Listen thread still going, can't clean up yet */
2838         return ret;
2839     }
2840 
2841     if (ret == 0) {
2842         ret = qemu_file_get_error(f);
2843     }
2844 
2845     /*
2846      * Try to read in the VMDESC section as well, so that dumping tools that
2847      * intercept our migration stream have the chance to see it.
2848      */
2849 
2850     /* We've got to be careful; if we don't read the data and just shut the fd
2851      * then the sender can error if we close while it's still sending.
2852      * We also mustn't read data that isn't there; some transports (RDMA)
2853      * will stall waiting for that data when the source has already closed.
2854      */
2855     if (ret == 0 && should_send_vmdesc()) {
2856         uint8_t *buf;
2857         uint32_t size;
2858         uint8_t  section_type = qemu_get_byte(f);
2859 
2860         if (section_type != QEMU_VM_VMDESCRIPTION) {
2861             error_report("Expected vmdescription section, but got %d",
2862                          section_type);
2863             /*
2864              * It doesn't seem worth failing at this point since
2865              * we apparently have an otherwise valid VM state
2866              */
2867         } else {
2868             buf = g_malloc(0x1000);
2869             size = qemu_get_be32(f);
2870 
2871             while (size > 0) {
2872                 uint32_t read_chunk = MIN(size, 0x1000);
2873                 qemu_get_buffer(f, buf, read_chunk);
2874                 size -= read_chunk;
2875             }
2876             g_free(buf);
2877         }
2878     }
2879 
2880     qemu_loadvm_state_cleanup();
2881     cpu_synchronize_all_post_init();
2882 
2883     return ret;
2884 }
2885 
2886 int qemu_load_device_state(QEMUFile *f)
2887 {
2888     MigrationIncomingState *mis = migration_incoming_get_current();
2889     int ret;
2890 
2891     /* Load QEMU_VM_SECTION_FULL section */
2892     ret = qemu_loadvm_state_main(f, mis);
2893     if (ret < 0) {
2894         error_report("Failed to load device state: %d", ret);
2895         return ret;
2896     }
2897 
2898     cpu_synchronize_all_post_init();
2899     return 0;
2900 }
2901 
2902 int qemu_loadvm_approve_switchover(void)
2903 {
2904     MigrationIncomingState *mis = migration_incoming_get_current();
2905 
2906     if (!mis->switchover_ack_pending_num) {
2907         return -EINVAL;
2908     }
2909 
2910     mis->switchover_ack_pending_num--;
2911     trace_loadvm_approve_switchover(mis->switchover_ack_pending_num);
2912 
2913     if (mis->switchover_ack_pending_num) {
2914         return 0;
2915     }
2916 
2917     return migrate_send_rp_switchover_ack(mis);
2918 }
2919 
2920 bool save_snapshot(const char *name, bool overwrite, const char *vmstate,
2921                   bool has_devices, strList *devices, Error **errp)
2922 {
2923     BlockDriverState *bs;
2924     QEMUSnapshotInfo sn1, *sn = &sn1;
2925     int ret = -1, ret2;
2926     QEMUFile *f;
2927     int saved_vm_running;
2928     uint64_t vm_state_size;
2929     g_autoptr(GDateTime) now = g_date_time_new_now_local();
2930     AioContext *aio_context;
2931 
2932     GLOBAL_STATE_CODE();
2933 
2934     if (migration_is_blocked(errp)) {
2935         return false;
2936     }
2937 
2938     if (!replay_can_snapshot()) {
2939         error_setg(errp, "Record/replay does not allow making snapshot "
2940                    "right now. Try once more later.");
2941         return false;
2942     }
2943 
2944     if (!bdrv_all_can_snapshot(has_devices, devices, errp)) {
2945         return false;
2946     }
2947 
2948     /* Delete old snapshots of the same name */
2949     if (name) {
2950         if (overwrite) {
2951             if (bdrv_all_delete_snapshot(name, has_devices,
2952                                          devices, errp) < 0) {
2953                 return false;
2954             }
2955         } else {
2956             ret2 = bdrv_all_has_snapshot(name, has_devices, devices, errp);
2957             if (ret2 < 0) {
2958                 return false;
2959             }
2960             if (ret2 == 1) {
2961                 error_setg(errp,
2962                            "Snapshot '%s' already exists in one or more devices",
2963                            name);
2964                 return false;
2965             }
2966         }
2967     }
2968 
2969     bs = bdrv_all_find_vmstate_bs(vmstate, has_devices, devices, errp);
2970     if (bs == NULL) {
2971         return false;
2972     }
2973     aio_context = bdrv_get_aio_context(bs);
2974 
2975     saved_vm_running = runstate_is_running();
2976 
2977     global_state_store();
2978     vm_stop(RUN_STATE_SAVE_VM);
2979 
2980     bdrv_drain_all_begin();
2981 
2982     aio_context_acquire(aio_context);
2983 
2984     memset(sn, 0, sizeof(*sn));
2985 
2986     /* fill auxiliary fields */
2987     sn->date_sec = g_date_time_to_unix(now);
2988     sn->date_nsec = g_date_time_get_microsecond(now) * 1000;
2989     sn->vm_clock_nsec = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL);
2990     if (replay_mode != REPLAY_MODE_NONE) {
2991         sn->icount = replay_get_current_icount();
2992     } else {
2993         sn->icount = -1ULL;
2994     }
2995 
2996     if (name) {
2997         pstrcpy(sn->name, sizeof(sn->name), name);
2998     } else {
2999         g_autofree char *autoname = g_date_time_format(now,  "vm-%Y%m%d%H%M%S");
3000         pstrcpy(sn->name, sizeof(sn->name), autoname);
3001     }
3002 
3003     /* save the VM state */
3004     f = qemu_fopen_bdrv(bs, 1);
3005     if (!f) {
3006         error_setg(errp, "Could not open VM state file");
3007         goto the_end;
3008     }
3009     ret = qemu_savevm_state(f, errp);
3010     vm_state_size = qemu_file_transferred_noflush(f);
3011     ret2 = qemu_fclose(f);
3012     if (ret < 0) {
3013         goto the_end;
3014     }
3015     if (ret2 < 0) {
3016         ret = ret2;
3017         goto the_end;
3018     }
3019 
3020     /* The bdrv_all_create_snapshot() call that follows acquires the AioContext
3021      * for itself.  BDRV_POLL_WHILE() does not support nested locking because
3022      * it only releases the lock once.  Therefore synchronous I/O will deadlock
3023      * unless we release the AioContext before bdrv_all_create_snapshot().
3024      */
3025     aio_context_release(aio_context);
3026     aio_context = NULL;
3027 
3028     ret = bdrv_all_create_snapshot(sn, bs, vm_state_size,
3029                                    has_devices, devices, errp);
3030     if (ret < 0) {
3031         bdrv_all_delete_snapshot(sn->name, has_devices, devices, NULL);
3032         goto the_end;
3033     }
3034 
3035     ret = 0;
3036 
3037  the_end:
3038     if (aio_context) {
3039         aio_context_release(aio_context);
3040     }
3041 
3042     bdrv_drain_all_end();
3043 
3044     if (saved_vm_running) {
3045         vm_start();
3046     }
3047     return ret == 0;
3048 }
3049 
3050 void qmp_xen_save_devices_state(const char *filename, bool has_live, bool live,
3051                                 Error **errp)
3052 {
3053     QEMUFile *f;
3054     QIOChannelFile *ioc;
3055     int saved_vm_running;
3056     int ret;
3057 
3058     if (!has_live) {
3059         /* live default to true so old version of Xen tool stack can have a
3060          * successful live migration */
3061         live = true;
3062     }
3063 
3064     saved_vm_running = runstate_is_running();
3065     vm_stop(RUN_STATE_SAVE_VM);
3066     global_state_store_running();
3067 
3068     ioc = qio_channel_file_new_path(filename, O_WRONLY | O_CREAT | O_TRUNC,
3069                                     0660, errp);
3070     if (!ioc) {
3071         goto the_end;
3072     }
3073     qio_channel_set_name(QIO_CHANNEL(ioc), "migration-xen-save-state");
3074     f = qemu_file_new_output(QIO_CHANNEL(ioc));
3075     object_unref(OBJECT(ioc));
3076     ret = qemu_save_device_state(f);
3077     if (ret < 0 || qemu_fclose(f) < 0) {
3078         error_setg(errp, QERR_IO_ERROR);
3079     } else {
3080         /* libxl calls the QMP command "stop" before calling
3081          * "xen-save-devices-state" and in case of migration failure, libxl
3082          * would call "cont".
3083          * So call bdrv_inactivate_all (release locks) here to let the other
3084          * side of the migration take control of the images.
3085          */
3086         if (live && !saved_vm_running) {
3087             ret = bdrv_inactivate_all();
3088             if (ret) {
3089                 error_setg(errp, "%s: bdrv_inactivate_all() failed (%d)",
3090                            __func__, ret);
3091             }
3092         }
3093     }
3094 
3095  the_end:
3096     if (saved_vm_running) {
3097         vm_start();
3098     }
3099 }
3100 
3101 void qmp_xen_load_devices_state(const char *filename, Error **errp)
3102 {
3103     QEMUFile *f;
3104     QIOChannelFile *ioc;
3105     int ret;
3106 
3107     /* Guest must be paused before loading the device state; the RAM state
3108      * will already have been loaded by xc
3109      */
3110     if (runstate_is_running()) {
3111         error_setg(errp, "Cannot update device state while vm is running");
3112         return;
3113     }
3114     vm_stop(RUN_STATE_RESTORE_VM);
3115 
3116     ioc = qio_channel_file_new_path(filename, O_RDONLY | O_BINARY, 0, errp);
3117     if (!ioc) {
3118         return;
3119     }
3120     qio_channel_set_name(QIO_CHANNEL(ioc), "migration-xen-load-state");
3121     f = qemu_file_new_input(QIO_CHANNEL(ioc));
3122     object_unref(OBJECT(ioc));
3123 
3124     ret = qemu_loadvm_state(f);
3125     qemu_fclose(f);
3126     if (ret < 0) {
3127         error_setg(errp, QERR_IO_ERROR);
3128     }
3129     migration_incoming_state_destroy();
3130 }
3131 
3132 bool load_snapshot(const char *name, const char *vmstate,
3133                    bool has_devices, strList *devices, Error **errp)
3134 {
3135     BlockDriverState *bs_vm_state;
3136     QEMUSnapshotInfo sn;
3137     QEMUFile *f;
3138     int ret;
3139     AioContext *aio_context;
3140     MigrationIncomingState *mis = migration_incoming_get_current();
3141 
3142     if (!bdrv_all_can_snapshot(has_devices, devices, errp)) {
3143         return false;
3144     }
3145     ret = bdrv_all_has_snapshot(name, has_devices, devices, errp);
3146     if (ret < 0) {
3147         return false;
3148     }
3149     if (ret == 0) {
3150         error_setg(errp, "Snapshot '%s' does not exist in one or more devices",
3151                    name);
3152         return false;
3153     }
3154 
3155     bs_vm_state = bdrv_all_find_vmstate_bs(vmstate, has_devices, devices, errp);
3156     if (!bs_vm_state) {
3157         return false;
3158     }
3159     aio_context = bdrv_get_aio_context(bs_vm_state);
3160 
3161     /* Don't even try to load empty VM states */
3162     aio_context_acquire(aio_context);
3163     ret = bdrv_snapshot_find(bs_vm_state, &sn, name);
3164     aio_context_release(aio_context);
3165     if (ret < 0) {
3166         return false;
3167     } else if (sn.vm_state_size == 0) {
3168         error_setg(errp, "This is a disk-only snapshot. Revert to it "
3169                    " offline using qemu-img");
3170         return false;
3171     }
3172 
3173     /*
3174      * Flush the record/replay queue. Now the VM state is going
3175      * to change. Therefore we don't need to preserve its consistency
3176      */
3177     replay_flush_events();
3178 
3179     /* Flush all IO requests so they don't interfere with the new state.  */
3180     bdrv_drain_all_begin();
3181 
3182     ret = bdrv_all_goto_snapshot(name, has_devices, devices, errp);
3183     if (ret < 0) {
3184         goto err_drain;
3185     }
3186 
3187     /* restore the VM state */
3188     f = qemu_fopen_bdrv(bs_vm_state, 0);
3189     if (!f) {
3190         error_setg(errp, "Could not open VM state file");
3191         goto err_drain;
3192     }
3193 
3194     qemu_system_reset(SHUTDOWN_CAUSE_SNAPSHOT_LOAD);
3195     mis->from_src_file = f;
3196 
3197     if (!yank_register_instance(MIGRATION_YANK_INSTANCE, errp)) {
3198         ret = -EINVAL;
3199         goto err_drain;
3200     }
3201     aio_context_acquire(aio_context);
3202     ret = qemu_loadvm_state(f);
3203     migration_incoming_state_destroy();
3204     aio_context_release(aio_context);
3205 
3206     bdrv_drain_all_end();
3207 
3208     if (ret < 0) {
3209         error_setg(errp, "Error %d while loading VM state", ret);
3210         return false;
3211     }
3212 
3213     return true;
3214 
3215 err_drain:
3216     bdrv_drain_all_end();
3217     return false;
3218 }
3219 
3220 bool delete_snapshot(const char *name, bool has_devices,
3221                      strList *devices, Error **errp)
3222 {
3223     if (!bdrv_all_can_snapshot(has_devices, devices, errp)) {
3224         return false;
3225     }
3226 
3227     if (bdrv_all_delete_snapshot(name, has_devices, devices, errp) < 0) {
3228         return false;
3229     }
3230 
3231     return true;
3232 }
3233 
3234 void vmstate_register_ram(MemoryRegion *mr, DeviceState *dev)
3235 {
3236     qemu_ram_set_idstr(mr->ram_block,
3237                        memory_region_name(mr), dev);
3238     qemu_ram_set_migratable(mr->ram_block);
3239 }
3240 
3241 void vmstate_unregister_ram(MemoryRegion *mr, DeviceState *dev)
3242 {
3243     qemu_ram_unset_idstr(mr->ram_block);
3244     qemu_ram_unset_migratable(mr->ram_block);
3245 }
3246 
3247 void vmstate_register_ram_global(MemoryRegion *mr)
3248 {
3249     vmstate_register_ram(mr, NULL);
3250 }
3251 
3252 bool vmstate_check_only_migratable(const VMStateDescription *vmsd)
3253 {
3254     /* check needed if --only-migratable is specified */
3255     if (!only_migratable) {
3256         return true;
3257     }
3258 
3259     return !(vmsd && vmsd->unmigratable);
3260 }
3261 
3262 typedef struct SnapshotJob {
3263     Job common;
3264     char *tag;
3265     char *vmstate;
3266     strList *devices;
3267     Coroutine *co;
3268     Error **errp;
3269     bool ret;
3270 } SnapshotJob;
3271 
3272 static void qmp_snapshot_job_free(SnapshotJob *s)
3273 {
3274     g_free(s->tag);
3275     g_free(s->vmstate);
3276     qapi_free_strList(s->devices);
3277 }
3278 
3279 
3280 static void snapshot_load_job_bh(void *opaque)
3281 {
3282     Job *job = opaque;
3283     SnapshotJob *s = container_of(job, SnapshotJob, common);
3284     int orig_vm_running;
3285 
3286     job_progress_set_remaining(&s->common, 1);
3287 
3288     orig_vm_running = runstate_is_running();
3289     vm_stop(RUN_STATE_RESTORE_VM);
3290 
3291     s->ret = load_snapshot(s->tag, s->vmstate, true, s->devices, s->errp);
3292     if (s->ret && orig_vm_running) {
3293         vm_start();
3294     }
3295 
3296     job_progress_update(&s->common, 1);
3297 
3298     qmp_snapshot_job_free(s);
3299     aio_co_wake(s->co);
3300 }
3301 
3302 static void snapshot_save_job_bh(void *opaque)
3303 {
3304     Job *job = opaque;
3305     SnapshotJob *s = container_of(job, SnapshotJob, common);
3306 
3307     job_progress_set_remaining(&s->common, 1);
3308     s->ret = save_snapshot(s->tag, false, s->vmstate,
3309                            true, s->devices, s->errp);
3310     job_progress_update(&s->common, 1);
3311 
3312     qmp_snapshot_job_free(s);
3313     aio_co_wake(s->co);
3314 }
3315 
3316 static void snapshot_delete_job_bh(void *opaque)
3317 {
3318     Job *job = opaque;
3319     SnapshotJob *s = container_of(job, SnapshotJob, common);
3320 
3321     job_progress_set_remaining(&s->common, 1);
3322     s->ret = delete_snapshot(s->tag, true, s->devices, s->errp);
3323     job_progress_update(&s->common, 1);
3324 
3325     qmp_snapshot_job_free(s);
3326     aio_co_wake(s->co);
3327 }
3328 
3329 static int coroutine_fn snapshot_save_job_run(Job *job, Error **errp)
3330 {
3331     SnapshotJob *s = container_of(job, SnapshotJob, common);
3332     s->errp = errp;
3333     s->co = qemu_coroutine_self();
3334     aio_bh_schedule_oneshot(qemu_get_aio_context(),
3335                             snapshot_save_job_bh, job);
3336     qemu_coroutine_yield();
3337     return s->ret ? 0 : -1;
3338 }
3339 
3340 static int coroutine_fn snapshot_load_job_run(Job *job, Error **errp)
3341 {
3342     SnapshotJob *s = container_of(job, SnapshotJob, common);
3343     s->errp = errp;
3344     s->co = qemu_coroutine_self();
3345     aio_bh_schedule_oneshot(qemu_get_aio_context(),
3346                             snapshot_load_job_bh, job);
3347     qemu_coroutine_yield();
3348     return s->ret ? 0 : -1;
3349 }
3350 
3351 static int coroutine_fn snapshot_delete_job_run(Job *job, Error **errp)
3352 {
3353     SnapshotJob *s = container_of(job, SnapshotJob, common);
3354     s->errp = errp;
3355     s->co = qemu_coroutine_self();
3356     aio_bh_schedule_oneshot(qemu_get_aio_context(),
3357                             snapshot_delete_job_bh, job);
3358     qemu_coroutine_yield();
3359     return s->ret ? 0 : -1;
3360 }
3361 
3362 
3363 static const JobDriver snapshot_load_job_driver = {
3364     .instance_size = sizeof(SnapshotJob),
3365     .job_type      = JOB_TYPE_SNAPSHOT_LOAD,
3366     .run           = snapshot_load_job_run,
3367 };
3368 
3369 static const JobDriver snapshot_save_job_driver = {
3370     .instance_size = sizeof(SnapshotJob),
3371     .job_type      = JOB_TYPE_SNAPSHOT_SAVE,
3372     .run           = snapshot_save_job_run,
3373 };
3374 
3375 static const JobDriver snapshot_delete_job_driver = {
3376     .instance_size = sizeof(SnapshotJob),
3377     .job_type      = JOB_TYPE_SNAPSHOT_DELETE,
3378     .run           = snapshot_delete_job_run,
3379 };
3380 
3381 
3382 void qmp_snapshot_save(const char *job_id,
3383                        const char *tag,
3384                        const char *vmstate,
3385                        strList *devices,
3386                        Error **errp)
3387 {
3388     SnapshotJob *s;
3389 
3390     s = job_create(job_id, &snapshot_save_job_driver, NULL,
3391                    qemu_get_aio_context(), JOB_MANUAL_DISMISS,
3392                    NULL, NULL, errp);
3393     if (!s) {
3394         return;
3395     }
3396 
3397     s->tag = g_strdup(tag);
3398     s->vmstate = g_strdup(vmstate);
3399     s->devices = QAPI_CLONE(strList, devices);
3400 
3401     job_start(&s->common);
3402 }
3403 
3404 void qmp_snapshot_load(const char *job_id,
3405                        const char *tag,
3406                        const char *vmstate,
3407                        strList *devices,
3408                        Error **errp)
3409 {
3410     SnapshotJob *s;
3411 
3412     s = job_create(job_id, &snapshot_load_job_driver, NULL,
3413                    qemu_get_aio_context(), JOB_MANUAL_DISMISS,
3414                    NULL, NULL, errp);
3415     if (!s) {
3416         return;
3417     }
3418 
3419     s->tag = g_strdup(tag);
3420     s->vmstate = g_strdup(vmstate);
3421     s->devices = QAPI_CLONE(strList, devices);
3422 
3423     job_start(&s->common);
3424 }
3425 
3426 void qmp_snapshot_delete(const char *job_id,
3427                          const char *tag,
3428                          strList *devices,
3429                          Error **errp)
3430 {
3431     SnapshotJob *s;
3432 
3433     s = job_create(job_id, &snapshot_delete_job_driver, NULL,
3434                    qemu_get_aio_context(), JOB_MANUAL_DISMISS,
3435                    NULL, NULL, errp);
3436     if (!s) {
3437         return;
3438     }
3439 
3440     s->tag = g_strdup(tag);
3441     s->devices = QAPI_CLONE(strList, devices);
3442 
3443     job_start(&s->common);
3444 }
3445