xref: /openbmc/qemu/migration/savevm.c (revision f664b88247487c4cb020d016bef0f3b1daf9f4e5)
1 /*
2  * QEMU System Emulator
3  *
4  * Copyright (c) 2003-2008 Fabrice Bellard
5  * Copyright (c) 2009-2015 Red Hat Inc
6  *
7  * Authors:
8  *  Juan Quintela <quintela@redhat.com>
9  *
10  * Permission is hereby granted, free of charge, to any person obtaining a copy
11  * of this software and associated documentation files (the "Software"), to deal
12  * in the Software without restriction, including without limitation the rights
13  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
14  * copies of the Software, and to permit persons to whom the Software is
15  * furnished to do so, subject to the following conditions:
16  *
17  * The above copyright notice and this permission notice shall be included in
18  * all copies or substantial portions of the Software.
19  *
20  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
21  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
22  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
23  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
24  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
25  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
26  * THE SOFTWARE.
27  */
28 
29 #include "qemu/osdep.h"
30 #include "hw/boards.h"
31 #include "hw/hw.h"
32 #include "hw/qdev.h"
33 #include "hw/xen/xen.h"
34 #include "net/net.h"
35 #include "sysemu/sysemu.h"
36 #include "qemu/timer.h"
37 #include "migration/migration.h"
38 #include "qemu-file-channel.h"
39 #include "savevm.h"
40 #include "postcopy-ram.h"
41 #include "qapi/qmp/qerror.h"
42 #include "qemu/error-report.h"
43 #include "qemu/queue.h"
44 #include "sysemu/cpus.h"
45 #include "exec/memory.h"
46 #include "exec/target_page.h"
47 #include "qmp-commands.h"
48 #include "trace.h"
49 #include "qemu/bitops.h"
50 #include "qemu/iov.h"
51 #include "block/snapshot.h"
52 #include "qemu/cutils.h"
53 #include "io/channel-buffer.h"
54 #include "io/channel-file.h"
55 
56 #ifndef ETH_P_RARP
57 #define ETH_P_RARP 0x8035
58 #endif
59 #define ARP_HTYPE_ETH 0x0001
60 #define ARP_PTYPE_IP 0x0800
61 #define ARP_OP_REQUEST_REV 0x3
62 
63 const unsigned int postcopy_ram_discard_version = 0;
64 
65 static bool skip_section_footers;
66 
67 /* Subcommands for QEMU_VM_COMMAND */
68 enum qemu_vm_cmd {
69     MIG_CMD_INVALID = 0,   /* Must be 0 */
70     MIG_CMD_OPEN_RETURN_PATH,  /* Tell the dest to open the Return path */
71     MIG_CMD_PING,              /* Request a PONG on the RP */
72 
73     MIG_CMD_POSTCOPY_ADVISE,       /* Prior to any page transfers, just
74                                       warn we might want to do PC */
75     MIG_CMD_POSTCOPY_LISTEN,       /* Start listening for incoming
76                                       pages as it's running. */
77     MIG_CMD_POSTCOPY_RUN,          /* Start execution */
78 
79     MIG_CMD_POSTCOPY_RAM_DISCARD,  /* A list of pages to discard that
80                                       were previously sent during
81                                       precopy but are dirty. */
82     MIG_CMD_PACKAGED,          /* Send a wrapped stream within this stream */
83     MIG_CMD_MAX
84 };
85 
86 #define MAX_VM_CMD_PACKAGED_SIZE (1ul << 24)
87 static struct mig_cmd_args {
88     ssize_t     len; /* -1 = variable */
89     const char *name;
90 } mig_cmd_args[] = {
91     [MIG_CMD_INVALID]          = { .len = -1, .name = "INVALID" },
92     [MIG_CMD_OPEN_RETURN_PATH] = { .len =  0, .name = "OPEN_RETURN_PATH" },
93     [MIG_CMD_PING]             = { .len = sizeof(uint32_t), .name = "PING" },
94     [MIG_CMD_POSTCOPY_ADVISE]  = { .len = 16, .name = "POSTCOPY_ADVISE" },
95     [MIG_CMD_POSTCOPY_LISTEN]  = { .len =  0, .name = "POSTCOPY_LISTEN" },
96     [MIG_CMD_POSTCOPY_RUN]     = { .len =  0, .name = "POSTCOPY_RUN" },
97     [MIG_CMD_POSTCOPY_RAM_DISCARD] = {
98                                    .len = -1, .name = "POSTCOPY_RAM_DISCARD" },
99     [MIG_CMD_PACKAGED]         = { .len =  4, .name = "PACKAGED" },
100     [MIG_CMD_MAX]              = { .len = -1, .name = "MAX" },
101 };
102 
103 static int announce_self_create(uint8_t *buf,
104                                 uint8_t *mac_addr)
105 {
106     /* Ethernet header. */
107     memset(buf, 0xff, 6);         /* destination MAC addr */
108     memcpy(buf + 6, mac_addr, 6); /* source MAC addr */
109     *(uint16_t *)(buf + 12) = htons(ETH_P_RARP); /* ethertype */
110 
111     /* RARP header. */
112     *(uint16_t *)(buf + 14) = htons(ARP_HTYPE_ETH); /* hardware addr space */
113     *(uint16_t *)(buf + 16) = htons(ARP_PTYPE_IP); /* protocol addr space */
114     *(buf + 18) = 6; /* hardware addr length (ethernet) */
115     *(buf + 19) = 4; /* protocol addr length (IPv4) */
116     *(uint16_t *)(buf + 20) = htons(ARP_OP_REQUEST_REV); /* opcode */
117     memcpy(buf + 22, mac_addr, 6); /* source hw addr */
118     memset(buf + 28, 0x00, 4);     /* source protocol addr */
119     memcpy(buf + 32, mac_addr, 6); /* target hw addr */
120     memset(buf + 38, 0x00, 4);     /* target protocol addr */
121 
122     /* Padding to get up to 60 bytes (ethernet min packet size, minus FCS). */
123     memset(buf + 42, 0x00, 18);
124 
125     return 60; /* len (FCS will be added by hardware) */
126 }
127 
128 static void qemu_announce_self_iter(NICState *nic, void *opaque)
129 {
130     uint8_t buf[60];
131     int len;
132 
133     trace_qemu_announce_self_iter(qemu_ether_ntoa(&nic->conf->macaddr));
134     len = announce_self_create(buf, nic->conf->macaddr.a);
135 
136     qemu_send_packet_raw(qemu_get_queue(nic), buf, len);
137 }
138 
139 
140 static void qemu_announce_self_once(void *opaque)
141 {
142     static int count = SELF_ANNOUNCE_ROUNDS;
143     QEMUTimer *timer = *(QEMUTimer **)opaque;
144 
145     qemu_foreach_nic(qemu_announce_self_iter, NULL);
146 
147     if (--count) {
148         /* delay 50ms, 150ms, 250ms, ... */
149         timer_mod(timer, qemu_clock_get_ms(QEMU_CLOCK_REALTIME) +
150                   self_announce_delay(count));
151     } else {
152             timer_del(timer);
153             timer_free(timer);
154     }
155 }
156 
157 void qemu_announce_self(void)
158 {
159     static QEMUTimer *timer;
160     timer = timer_new_ms(QEMU_CLOCK_REALTIME, qemu_announce_self_once, &timer);
161     qemu_announce_self_once(&timer);
162 }
163 
164 /***********************************************************/
165 /* savevm/loadvm support */
166 
167 static ssize_t block_writev_buffer(void *opaque, struct iovec *iov, int iovcnt,
168                                    int64_t pos)
169 {
170     int ret;
171     QEMUIOVector qiov;
172 
173     qemu_iovec_init_external(&qiov, iov, iovcnt);
174     ret = bdrv_writev_vmstate(opaque, &qiov, pos);
175     if (ret < 0) {
176         return ret;
177     }
178 
179     return qiov.size;
180 }
181 
182 static ssize_t block_get_buffer(void *opaque, uint8_t *buf, int64_t pos,
183                                 size_t size)
184 {
185     return bdrv_load_vmstate(opaque, buf, pos, size);
186 }
187 
188 static int bdrv_fclose(void *opaque)
189 {
190     return bdrv_flush(opaque);
191 }
192 
193 static const QEMUFileOps bdrv_read_ops = {
194     .get_buffer = block_get_buffer,
195     .close =      bdrv_fclose
196 };
197 
198 static const QEMUFileOps bdrv_write_ops = {
199     .writev_buffer  = block_writev_buffer,
200     .close          = bdrv_fclose
201 };
202 
203 static QEMUFile *qemu_fopen_bdrv(BlockDriverState *bs, int is_writable)
204 {
205     if (is_writable) {
206         return qemu_fopen_ops(bs, &bdrv_write_ops);
207     }
208     return qemu_fopen_ops(bs, &bdrv_read_ops);
209 }
210 
211 
212 /* QEMUFile timer support.
213  * Not in qemu-file.c to not add qemu-timer.c as dependency to qemu-file.c
214  */
215 
216 void timer_put(QEMUFile *f, QEMUTimer *ts)
217 {
218     uint64_t expire_time;
219 
220     expire_time = timer_expire_time_ns(ts);
221     qemu_put_be64(f, expire_time);
222 }
223 
224 void timer_get(QEMUFile *f, QEMUTimer *ts)
225 {
226     uint64_t expire_time;
227 
228     expire_time = qemu_get_be64(f);
229     if (expire_time != -1) {
230         timer_mod_ns(ts, expire_time);
231     } else {
232         timer_del(ts);
233     }
234 }
235 
236 
237 /* VMState timer support.
238  * Not in vmstate.c to not add qemu-timer.c as dependency to vmstate.c
239  */
240 
241 static int get_timer(QEMUFile *f, void *pv, size_t size, VMStateField *field)
242 {
243     QEMUTimer *v = pv;
244     timer_get(f, v);
245     return 0;
246 }
247 
248 static int put_timer(QEMUFile *f, void *pv, size_t size, VMStateField *field,
249                      QJSON *vmdesc)
250 {
251     QEMUTimer *v = pv;
252     timer_put(f, v);
253 
254     return 0;
255 }
256 
257 const VMStateInfo vmstate_info_timer = {
258     .name = "timer",
259     .get  = get_timer,
260     .put  = put_timer,
261 };
262 
263 
264 typedef struct CompatEntry {
265     char idstr[256];
266     int instance_id;
267 } CompatEntry;
268 
269 typedef struct SaveStateEntry {
270     QTAILQ_ENTRY(SaveStateEntry) entry;
271     char idstr[256];
272     int instance_id;
273     int alias_id;
274     int version_id;
275     int section_id;
276     SaveVMHandlers *ops;
277     const VMStateDescription *vmsd;
278     void *opaque;
279     CompatEntry *compat;
280     int is_ram;
281 } SaveStateEntry;
282 
283 typedef struct SaveState {
284     QTAILQ_HEAD(, SaveStateEntry) handlers;
285     int global_section_id;
286     bool skip_configuration;
287     uint32_t len;
288     const char *name;
289     uint32_t target_page_bits;
290 } SaveState;
291 
292 static SaveState savevm_state = {
293     .handlers = QTAILQ_HEAD_INITIALIZER(savevm_state.handlers),
294     .global_section_id = 0,
295     .skip_configuration = false,
296 };
297 
298 void savevm_skip_configuration(void)
299 {
300     savevm_state.skip_configuration = true;
301 }
302 
303 
304 static void configuration_pre_save(void *opaque)
305 {
306     SaveState *state = opaque;
307     const char *current_name = MACHINE_GET_CLASS(current_machine)->name;
308 
309     state->len = strlen(current_name);
310     state->name = current_name;
311     state->target_page_bits = qemu_target_page_bits();
312 }
313 
314 static int configuration_pre_load(void *opaque)
315 {
316     SaveState *state = opaque;
317 
318     /* If there is no target-page-bits subsection it means the source
319      * predates the variable-target-page-bits support and is using the
320      * minimum possible value for this CPU.
321      */
322     state->target_page_bits = qemu_target_page_bits_min();
323     return 0;
324 }
325 
326 static int configuration_post_load(void *opaque, int version_id)
327 {
328     SaveState *state = opaque;
329     const char *current_name = MACHINE_GET_CLASS(current_machine)->name;
330 
331     if (strncmp(state->name, current_name, state->len) != 0) {
332         error_report("Machine type received is '%.*s' and local is '%s'",
333                      (int) state->len, state->name, current_name);
334         return -EINVAL;
335     }
336 
337     if (state->target_page_bits != qemu_target_page_bits()) {
338         error_report("Received TARGET_PAGE_BITS is %d but local is %d",
339                      state->target_page_bits, qemu_target_page_bits());
340         return -EINVAL;
341     }
342 
343     return 0;
344 }
345 
346 /* The target-page-bits subsection is present only if the
347  * target page size is not the same as the default (ie the
348  * minimum page size for a variable-page-size guest CPU).
349  * If it is present then it contains the actual target page
350  * bits for the machine, and migration will fail if the
351  * two ends don't agree about it.
352  */
353 static bool vmstate_target_page_bits_needed(void *opaque)
354 {
355     return qemu_target_page_bits()
356         > qemu_target_page_bits_min();
357 }
358 
359 static const VMStateDescription vmstate_target_page_bits = {
360     .name = "configuration/target-page-bits",
361     .version_id = 1,
362     .minimum_version_id = 1,
363     .needed = vmstate_target_page_bits_needed,
364     .fields = (VMStateField[]) {
365         VMSTATE_UINT32(target_page_bits, SaveState),
366         VMSTATE_END_OF_LIST()
367     }
368 };
369 
370 static const VMStateDescription vmstate_configuration = {
371     .name = "configuration",
372     .version_id = 1,
373     .pre_load = configuration_pre_load,
374     .post_load = configuration_post_load,
375     .pre_save = configuration_pre_save,
376     .fields = (VMStateField[]) {
377         VMSTATE_UINT32(len, SaveState),
378         VMSTATE_VBUFFER_ALLOC_UINT32(name, SaveState, 0, NULL, len),
379         VMSTATE_END_OF_LIST()
380     },
381     .subsections = (const VMStateDescription*[]) {
382         &vmstate_target_page_bits,
383         NULL
384     }
385 };
386 
387 static void dump_vmstate_vmsd(FILE *out_file,
388                               const VMStateDescription *vmsd, int indent,
389                               bool is_subsection);
390 
391 static void dump_vmstate_vmsf(FILE *out_file, const VMStateField *field,
392                               int indent)
393 {
394     fprintf(out_file, "%*s{\n", indent, "");
395     indent += 2;
396     fprintf(out_file, "%*s\"field\": \"%s\",\n", indent, "", field->name);
397     fprintf(out_file, "%*s\"version_id\": %d,\n", indent, "",
398             field->version_id);
399     fprintf(out_file, "%*s\"field_exists\": %s,\n", indent, "",
400             field->field_exists ? "true" : "false");
401     fprintf(out_file, "%*s\"size\": %zu", indent, "", field->size);
402     if (field->vmsd != NULL) {
403         fprintf(out_file, ",\n");
404         dump_vmstate_vmsd(out_file, field->vmsd, indent, false);
405     }
406     fprintf(out_file, "\n%*s}", indent - 2, "");
407 }
408 
409 static void dump_vmstate_vmss(FILE *out_file,
410                               const VMStateDescription **subsection,
411                               int indent)
412 {
413     if (*subsection != NULL) {
414         dump_vmstate_vmsd(out_file, *subsection, indent, true);
415     }
416 }
417 
418 static void dump_vmstate_vmsd(FILE *out_file,
419                               const VMStateDescription *vmsd, int indent,
420                               bool is_subsection)
421 {
422     if (is_subsection) {
423         fprintf(out_file, "%*s{\n", indent, "");
424     } else {
425         fprintf(out_file, "%*s\"%s\": {\n", indent, "", "Description");
426     }
427     indent += 2;
428     fprintf(out_file, "%*s\"name\": \"%s\",\n", indent, "", vmsd->name);
429     fprintf(out_file, "%*s\"version_id\": %d,\n", indent, "",
430             vmsd->version_id);
431     fprintf(out_file, "%*s\"minimum_version_id\": %d", indent, "",
432             vmsd->minimum_version_id);
433     if (vmsd->fields != NULL) {
434         const VMStateField *field = vmsd->fields;
435         bool first;
436 
437         fprintf(out_file, ",\n%*s\"Fields\": [\n", indent, "");
438         first = true;
439         while (field->name != NULL) {
440             if (field->flags & VMS_MUST_EXIST) {
441                 /* Ignore VMSTATE_VALIDATE bits; these don't get migrated */
442                 field++;
443                 continue;
444             }
445             if (!first) {
446                 fprintf(out_file, ",\n");
447             }
448             dump_vmstate_vmsf(out_file, field, indent + 2);
449             field++;
450             first = false;
451         }
452         fprintf(out_file, "\n%*s]", indent, "");
453     }
454     if (vmsd->subsections != NULL) {
455         const VMStateDescription **subsection = vmsd->subsections;
456         bool first;
457 
458         fprintf(out_file, ",\n%*s\"Subsections\": [\n", indent, "");
459         first = true;
460         while (*subsection != NULL) {
461             if (!first) {
462                 fprintf(out_file, ",\n");
463             }
464             dump_vmstate_vmss(out_file, subsection, indent + 2);
465             subsection++;
466             first = false;
467         }
468         fprintf(out_file, "\n%*s]", indent, "");
469     }
470     fprintf(out_file, "\n%*s}", indent - 2, "");
471 }
472 
473 static void dump_machine_type(FILE *out_file)
474 {
475     MachineClass *mc;
476 
477     mc = MACHINE_GET_CLASS(current_machine);
478 
479     fprintf(out_file, "  \"vmschkmachine\": {\n");
480     fprintf(out_file, "    \"Name\": \"%s\"\n", mc->name);
481     fprintf(out_file, "  },\n");
482 }
483 
484 void dump_vmstate_json_to_file(FILE *out_file)
485 {
486     GSList *list, *elt;
487     bool first;
488 
489     fprintf(out_file, "{\n");
490     dump_machine_type(out_file);
491 
492     first = true;
493     list = object_class_get_list(TYPE_DEVICE, true);
494     for (elt = list; elt; elt = elt->next) {
495         DeviceClass *dc = OBJECT_CLASS_CHECK(DeviceClass, elt->data,
496                                              TYPE_DEVICE);
497         const char *name;
498         int indent = 2;
499 
500         if (!dc->vmsd) {
501             continue;
502         }
503 
504         if (!first) {
505             fprintf(out_file, ",\n");
506         }
507         name = object_class_get_name(OBJECT_CLASS(dc));
508         fprintf(out_file, "%*s\"%s\": {\n", indent, "", name);
509         indent += 2;
510         fprintf(out_file, "%*s\"Name\": \"%s\",\n", indent, "", name);
511         fprintf(out_file, "%*s\"version_id\": %d,\n", indent, "",
512                 dc->vmsd->version_id);
513         fprintf(out_file, "%*s\"minimum_version_id\": %d,\n", indent, "",
514                 dc->vmsd->minimum_version_id);
515 
516         dump_vmstate_vmsd(out_file, dc->vmsd, indent, false);
517 
518         fprintf(out_file, "\n%*s}", indent - 2, "");
519         first = false;
520     }
521     fprintf(out_file, "\n}\n");
522     fclose(out_file);
523 }
524 
525 static int calculate_new_instance_id(const char *idstr)
526 {
527     SaveStateEntry *se;
528     int instance_id = 0;
529 
530     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
531         if (strcmp(idstr, se->idstr) == 0
532             && instance_id <= se->instance_id) {
533             instance_id = se->instance_id + 1;
534         }
535     }
536     return instance_id;
537 }
538 
539 static int calculate_compat_instance_id(const char *idstr)
540 {
541     SaveStateEntry *se;
542     int instance_id = 0;
543 
544     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
545         if (!se->compat) {
546             continue;
547         }
548 
549         if (strcmp(idstr, se->compat->idstr) == 0
550             && instance_id <= se->compat->instance_id) {
551             instance_id = se->compat->instance_id + 1;
552         }
553     }
554     return instance_id;
555 }
556 
557 static inline MigrationPriority save_state_priority(SaveStateEntry *se)
558 {
559     if (se->vmsd) {
560         return se->vmsd->priority;
561     }
562     return MIG_PRI_DEFAULT;
563 }
564 
565 static void savevm_state_handler_insert(SaveStateEntry *nse)
566 {
567     MigrationPriority priority = save_state_priority(nse);
568     SaveStateEntry *se;
569 
570     assert(priority <= MIG_PRI_MAX);
571 
572     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
573         if (save_state_priority(se) < priority) {
574             break;
575         }
576     }
577 
578     if (se) {
579         QTAILQ_INSERT_BEFORE(se, nse, entry);
580     } else {
581         QTAILQ_INSERT_TAIL(&savevm_state.handlers, nse, entry);
582     }
583 }
584 
585 /* TODO: Individual devices generally have very little idea about the rest
586    of the system, so instance_id should be removed/replaced.
587    Meanwhile pass -1 as instance_id if you do not already have a clearly
588    distinguishing id for all instances of your device class. */
589 int register_savevm_live(DeviceState *dev,
590                          const char *idstr,
591                          int instance_id,
592                          int version_id,
593                          SaveVMHandlers *ops,
594                          void *opaque)
595 {
596     SaveStateEntry *se;
597 
598     se = g_new0(SaveStateEntry, 1);
599     se->version_id = version_id;
600     se->section_id = savevm_state.global_section_id++;
601     se->ops = ops;
602     se->opaque = opaque;
603     se->vmsd = NULL;
604     /* if this is a live_savem then set is_ram */
605     if (ops->save_live_setup != NULL) {
606         se->is_ram = 1;
607     }
608 
609     if (dev) {
610         char *id = qdev_get_dev_path(dev);
611         if (id) {
612             if (snprintf(se->idstr, sizeof(se->idstr), "%s/", id) >=
613                 sizeof(se->idstr)) {
614                 error_report("Path too long for VMState (%s)", id);
615                 g_free(id);
616                 g_free(se);
617 
618                 return -1;
619             }
620             g_free(id);
621 
622             se->compat = g_new0(CompatEntry, 1);
623             pstrcpy(se->compat->idstr, sizeof(se->compat->idstr), idstr);
624             se->compat->instance_id = instance_id == -1 ?
625                          calculate_compat_instance_id(idstr) : instance_id;
626             instance_id = -1;
627         }
628     }
629     pstrcat(se->idstr, sizeof(se->idstr), idstr);
630 
631     if (instance_id == -1) {
632         se->instance_id = calculate_new_instance_id(se->idstr);
633     } else {
634         se->instance_id = instance_id;
635     }
636     assert(!se->compat || se->instance_id == 0);
637     savevm_state_handler_insert(se);
638     return 0;
639 }
640 
641 int register_savevm(DeviceState *dev,
642                     const char *idstr,
643                     int instance_id,
644                     int version_id,
645                     SaveStateHandler *save_state,
646                     LoadStateHandler *load_state,
647                     void *opaque)
648 {
649     SaveVMHandlers *ops = g_new0(SaveVMHandlers, 1);
650     ops->save_state = save_state;
651     ops->load_state = load_state;
652     return register_savevm_live(dev, idstr, instance_id, version_id,
653                                 ops, opaque);
654 }
655 
656 void unregister_savevm(DeviceState *dev, const char *idstr, void *opaque)
657 {
658     SaveStateEntry *se, *new_se;
659     char id[256] = "";
660 
661     if (dev) {
662         char *path = qdev_get_dev_path(dev);
663         if (path) {
664             pstrcpy(id, sizeof(id), path);
665             pstrcat(id, sizeof(id), "/");
666             g_free(path);
667         }
668     }
669     pstrcat(id, sizeof(id), idstr);
670 
671     QTAILQ_FOREACH_SAFE(se, &savevm_state.handlers, entry, new_se) {
672         if (strcmp(se->idstr, id) == 0 && se->opaque == opaque) {
673             QTAILQ_REMOVE(&savevm_state.handlers, se, entry);
674             g_free(se->compat);
675             g_free(se->ops);
676             g_free(se);
677         }
678     }
679 }
680 
681 int vmstate_register_with_alias_id(DeviceState *dev, int instance_id,
682                                    const VMStateDescription *vmsd,
683                                    void *opaque, int alias_id,
684                                    int required_for_version,
685                                    Error **errp)
686 {
687     SaveStateEntry *se;
688 
689     /* If this triggers, alias support can be dropped for the vmsd. */
690     assert(alias_id == -1 || required_for_version >= vmsd->minimum_version_id);
691 
692     se = g_new0(SaveStateEntry, 1);
693     se->version_id = vmsd->version_id;
694     se->section_id = savevm_state.global_section_id++;
695     se->opaque = opaque;
696     se->vmsd = vmsd;
697     se->alias_id = alias_id;
698 
699     if (dev) {
700         char *id = qdev_get_dev_path(dev);
701         if (id) {
702             if (snprintf(se->idstr, sizeof(se->idstr), "%s/", id) >=
703                 sizeof(se->idstr)) {
704                 error_setg(errp, "Path too long for VMState (%s)", id);
705                 g_free(id);
706                 g_free(se);
707 
708                 return -1;
709             }
710             g_free(id);
711 
712             se->compat = g_new0(CompatEntry, 1);
713             pstrcpy(se->compat->idstr, sizeof(se->compat->idstr), vmsd->name);
714             se->compat->instance_id = instance_id == -1 ?
715                          calculate_compat_instance_id(vmsd->name) : instance_id;
716             instance_id = -1;
717         }
718     }
719     pstrcat(se->idstr, sizeof(se->idstr), vmsd->name);
720 
721     if (instance_id == -1) {
722         se->instance_id = calculate_new_instance_id(se->idstr);
723     } else {
724         se->instance_id = instance_id;
725     }
726     assert(!se->compat || se->instance_id == 0);
727     savevm_state_handler_insert(se);
728     return 0;
729 }
730 
731 void vmstate_unregister(DeviceState *dev, const VMStateDescription *vmsd,
732                         void *opaque)
733 {
734     SaveStateEntry *se, *new_se;
735 
736     QTAILQ_FOREACH_SAFE(se, &savevm_state.handlers, entry, new_se) {
737         if (se->vmsd == vmsd && se->opaque == opaque) {
738             QTAILQ_REMOVE(&savevm_state.handlers, se, entry);
739             g_free(se->compat);
740             g_free(se);
741         }
742     }
743 }
744 
745 static int vmstate_load(QEMUFile *f, SaveStateEntry *se, int version_id)
746 {
747     trace_vmstate_load(se->idstr, se->vmsd ? se->vmsd->name : "(old)");
748     if (!se->vmsd) {         /* Old style */
749         return se->ops->load_state(f, se->opaque, version_id);
750     }
751     return vmstate_load_state(f, se->vmsd, se->opaque, version_id);
752 }
753 
754 static void vmstate_save_old_style(QEMUFile *f, SaveStateEntry *se, QJSON *vmdesc)
755 {
756     int64_t old_offset, size;
757 
758     old_offset = qemu_ftell_fast(f);
759     se->ops->save_state(f, se->opaque);
760     size = qemu_ftell_fast(f) - old_offset;
761 
762     if (vmdesc) {
763         json_prop_int(vmdesc, "size", size);
764         json_start_array(vmdesc, "fields");
765         json_start_object(vmdesc, NULL);
766         json_prop_str(vmdesc, "name", "data");
767         json_prop_int(vmdesc, "size", size);
768         json_prop_str(vmdesc, "type", "buffer");
769         json_end_object(vmdesc);
770         json_end_array(vmdesc);
771     }
772 }
773 
774 static void vmstate_save(QEMUFile *f, SaveStateEntry *se, QJSON *vmdesc)
775 {
776     trace_vmstate_save(se->idstr, se->vmsd ? se->vmsd->name : "(old)");
777     if (!se->vmsd) {
778         vmstate_save_old_style(f, se, vmdesc);
779         return;
780     }
781     vmstate_save_state(f, se->vmsd, se->opaque, vmdesc);
782 }
783 
784 void savevm_skip_section_footers(void)
785 {
786     skip_section_footers = true;
787 }
788 
789 /*
790  * Write the header for device section (QEMU_VM_SECTION START/END/PART/FULL)
791  */
792 static void save_section_header(QEMUFile *f, SaveStateEntry *se,
793                                 uint8_t section_type)
794 {
795     qemu_put_byte(f, section_type);
796     qemu_put_be32(f, se->section_id);
797 
798     if (section_type == QEMU_VM_SECTION_FULL ||
799         section_type == QEMU_VM_SECTION_START) {
800         /* ID string */
801         size_t len = strlen(se->idstr);
802         qemu_put_byte(f, len);
803         qemu_put_buffer(f, (uint8_t *)se->idstr, len);
804 
805         qemu_put_be32(f, se->instance_id);
806         qemu_put_be32(f, se->version_id);
807     }
808 }
809 
810 /*
811  * Write a footer onto device sections that catches cases misformatted device
812  * sections.
813  */
814 static void save_section_footer(QEMUFile *f, SaveStateEntry *se)
815 {
816     if (!skip_section_footers) {
817         qemu_put_byte(f, QEMU_VM_SECTION_FOOTER);
818         qemu_put_be32(f, se->section_id);
819     }
820 }
821 
822 /**
823  * qemu_savevm_command_send: Send a 'QEMU_VM_COMMAND' type element with the
824  *                           command and associated data.
825  *
826  * @f: File to send command on
827  * @command: Command type to send
828  * @len: Length of associated data
829  * @data: Data associated with command.
830  */
831 static void qemu_savevm_command_send(QEMUFile *f,
832                                      enum qemu_vm_cmd command,
833                                      uint16_t len,
834                                      uint8_t *data)
835 {
836     trace_savevm_command_send(command, len);
837     qemu_put_byte(f, QEMU_VM_COMMAND);
838     qemu_put_be16(f, (uint16_t)command);
839     qemu_put_be16(f, len);
840     qemu_put_buffer(f, data, len);
841     qemu_fflush(f);
842 }
843 
844 void qemu_savevm_send_ping(QEMUFile *f, uint32_t value)
845 {
846     uint32_t buf;
847 
848     trace_savevm_send_ping(value);
849     buf = cpu_to_be32(value);
850     qemu_savevm_command_send(f, MIG_CMD_PING, sizeof(value), (uint8_t *)&buf);
851 }
852 
853 void qemu_savevm_send_open_return_path(QEMUFile *f)
854 {
855     trace_savevm_send_open_return_path();
856     qemu_savevm_command_send(f, MIG_CMD_OPEN_RETURN_PATH, 0, NULL);
857 }
858 
859 /* We have a buffer of data to send; we don't want that all to be loaded
860  * by the command itself, so the command contains just the length of the
861  * extra buffer that we then send straight after it.
862  * TODO: Must be a better way to organise that
863  *
864  * Returns:
865  *    0 on success
866  *    -ve on error
867  */
868 int qemu_savevm_send_packaged(QEMUFile *f, const uint8_t *buf, size_t len)
869 {
870     uint32_t tmp;
871 
872     if (len > MAX_VM_CMD_PACKAGED_SIZE) {
873         error_report("%s: Unreasonably large packaged state: %zu",
874                      __func__, len);
875         return -1;
876     }
877 
878     tmp = cpu_to_be32(len);
879 
880     trace_qemu_savevm_send_packaged();
881     qemu_savevm_command_send(f, MIG_CMD_PACKAGED, 4, (uint8_t *)&tmp);
882 
883     qemu_put_buffer(f, buf, len);
884 
885     return 0;
886 }
887 
888 /* Send prior to any postcopy transfer */
889 void qemu_savevm_send_postcopy_advise(QEMUFile *f)
890 {
891     uint64_t tmp[2];
892     tmp[0] = cpu_to_be64(ram_pagesize_summary());
893     tmp[1] = cpu_to_be64(qemu_target_page_size());
894 
895     trace_qemu_savevm_send_postcopy_advise();
896     qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_ADVISE, 16, (uint8_t *)tmp);
897 }
898 
899 /* Sent prior to starting the destination running in postcopy, discard pages
900  * that have already been sent but redirtied on the source.
901  * CMD_POSTCOPY_RAM_DISCARD consist of:
902  *      byte   version (0)
903  *      byte   Length of name field (not including 0)
904  *  n x byte   RAM block name
905  *      byte   0 terminator (just for safety)
906  *  n x        Byte ranges within the named RAMBlock
907  *      be64   Start of the range
908  *      be64   Length
909  *
910  *  name:  RAMBlock name that these entries are part of
911  *  len: Number of page entries
912  *  start_list: 'len' addresses
913  *  length_list: 'len' addresses
914  *
915  */
916 void qemu_savevm_send_postcopy_ram_discard(QEMUFile *f, const char *name,
917                                            uint16_t len,
918                                            uint64_t *start_list,
919                                            uint64_t *length_list)
920 {
921     uint8_t *buf;
922     uint16_t tmplen;
923     uint16_t t;
924     size_t name_len = strlen(name);
925 
926     trace_qemu_savevm_send_postcopy_ram_discard(name, len);
927     assert(name_len < 256);
928     buf = g_malloc0(1 + 1 + name_len + 1 + (8 + 8) * len);
929     buf[0] = postcopy_ram_discard_version;
930     buf[1] = name_len;
931     memcpy(buf + 2, name, name_len);
932     tmplen = 2 + name_len;
933     buf[tmplen++] = '\0';
934 
935     for (t = 0; t < len; t++) {
936         stq_be_p(buf + tmplen, start_list[t]);
937         tmplen += 8;
938         stq_be_p(buf + tmplen, length_list[t]);
939         tmplen += 8;
940     }
941     qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_RAM_DISCARD, tmplen, buf);
942     g_free(buf);
943 }
944 
945 /* Get the destination into a state where it can receive postcopy data. */
946 void qemu_savevm_send_postcopy_listen(QEMUFile *f)
947 {
948     trace_savevm_send_postcopy_listen();
949     qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_LISTEN, 0, NULL);
950 }
951 
952 /* Kick the destination into running */
953 void qemu_savevm_send_postcopy_run(QEMUFile *f)
954 {
955     trace_savevm_send_postcopy_run();
956     qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_RUN, 0, NULL);
957 }
958 
959 bool qemu_savevm_state_blocked(Error **errp)
960 {
961     SaveStateEntry *se;
962 
963     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
964         if (se->vmsd && se->vmsd->unmigratable) {
965             error_setg(errp, "State blocked by non-migratable device '%s'",
966                        se->idstr);
967             return true;
968         }
969     }
970     return false;
971 }
972 
973 static bool enforce_config_section(void)
974 {
975     MachineState *machine = MACHINE(qdev_get_machine());
976     return machine->enforce_config_section;
977 }
978 
979 void qemu_savevm_state_header(QEMUFile *f)
980 {
981     trace_savevm_state_header();
982     qemu_put_be32(f, QEMU_VM_FILE_MAGIC);
983     qemu_put_be32(f, QEMU_VM_FILE_VERSION);
984 
985     if (!savevm_state.skip_configuration || enforce_config_section()) {
986         qemu_put_byte(f, QEMU_VM_CONFIGURATION);
987         vmstate_save_state(f, &vmstate_configuration, &savevm_state, 0);
988     }
989 
990 }
991 
992 void qemu_savevm_state_begin(QEMUFile *f)
993 {
994     SaveStateEntry *se;
995     int ret;
996 
997     trace_savevm_state_begin();
998     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
999         if (!se->ops || !se->ops->save_live_setup) {
1000             continue;
1001         }
1002         if (se->ops && se->ops->is_active) {
1003             if (!se->ops->is_active(se->opaque)) {
1004                 continue;
1005             }
1006         }
1007         save_section_header(f, se, QEMU_VM_SECTION_START);
1008 
1009         ret = se->ops->save_live_setup(f, se->opaque);
1010         save_section_footer(f, se);
1011         if (ret < 0) {
1012             qemu_file_set_error(f, ret);
1013             break;
1014         }
1015     }
1016 }
1017 
1018 /*
1019  * this function has three return values:
1020  *   negative: there was one error, and we have -errno.
1021  *   0 : We haven't finished, caller have to go again
1022  *   1 : We have finished, we can go to complete phase
1023  */
1024 int qemu_savevm_state_iterate(QEMUFile *f, bool postcopy)
1025 {
1026     SaveStateEntry *se;
1027     int ret = 1;
1028 
1029     trace_savevm_state_iterate();
1030     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1031         if (!se->ops || !se->ops->save_live_iterate) {
1032             continue;
1033         }
1034         if (se->ops && se->ops->is_active) {
1035             if (!se->ops->is_active(se->opaque)) {
1036                 continue;
1037             }
1038         }
1039         /*
1040          * In the postcopy phase, any device that doesn't know how to
1041          * do postcopy should have saved it's state in the _complete
1042          * call that's already run, it might get confused if we call
1043          * iterate afterwards.
1044          */
1045         if (postcopy && !se->ops->save_live_complete_postcopy) {
1046             continue;
1047         }
1048         if (qemu_file_rate_limit(f)) {
1049             return 0;
1050         }
1051         trace_savevm_section_start(se->idstr, se->section_id);
1052 
1053         save_section_header(f, se, QEMU_VM_SECTION_PART);
1054 
1055         ret = se->ops->save_live_iterate(f, se->opaque);
1056         trace_savevm_section_end(se->idstr, se->section_id, ret);
1057         save_section_footer(f, se);
1058 
1059         if (ret < 0) {
1060             qemu_file_set_error(f, ret);
1061         }
1062         if (ret <= 0) {
1063             /* Do not proceed to the next vmstate before this one reported
1064                completion of the current stage. This serializes the migration
1065                and reduces the probability that a faster changing state is
1066                synchronized over and over again. */
1067             break;
1068         }
1069     }
1070     return ret;
1071 }
1072 
1073 static bool should_send_vmdesc(void)
1074 {
1075     MachineState *machine = MACHINE(qdev_get_machine());
1076     bool in_postcopy = migration_in_postcopy();
1077     return !machine->suppress_vmdesc && !in_postcopy;
1078 }
1079 
1080 /*
1081  * Calls the save_live_complete_postcopy methods
1082  * causing the last few pages to be sent immediately and doing any associated
1083  * cleanup.
1084  * Note postcopy also calls qemu_savevm_state_complete_precopy to complete
1085  * all the other devices, but that happens at the point we switch to postcopy.
1086  */
1087 void qemu_savevm_state_complete_postcopy(QEMUFile *f)
1088 {
1089     SaveStateEntry *se;
1090     int ret;
1091 
1092     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1093         if (!se->ops || !se->ops->save_live_complete_postcopy) {
1094             continue;
1095         }
1096         if (se->ops && se->ops->is_active) {
1097             if (!se->ops->is_active(se->opaque)) {
1098                 continue;
1099             }
1100         }
1101         trace_savevm_section_start(se->idstr, se->section_id);
1102         /* Section type */
1103         qemu_put_byte(f, QEMU_VM_SECTION_END);
1104         qemu_put_be32(f, se->section_id);
1105 
1106         ret = se->ops->save_live_complete_postcopy(f, se->opaque);
1107         trace_savevm_section_end(se->idstr, se->section_id, ret);
1108         save_section_footer(f, se);
1109         if (ret < 0) {
1110             qemu_file_set_error(f, ret);
1111             return;
1112         }
1113     }
1114 
1115     qemu_put_byte(f, QEMU_VM_EOF);
1116     qemu_fflush(f);
1117 }
1118 
1119 void qemu_savevm_state_complete_precopy(QEMUFile *f, bool iterable_only)
1120 {
1121     QJSON *vmdesc;
1122     int vmdesc_len;
1123     SaveStateEntry *se;
1124     int ret;
1125     bool in_postcopy = migration_in_postcopy();
1126 
1127     trace_savevm_state_complete_precopy();
1128 
1129     cpu_synchronize_all_states();
1130 
1131     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1132         if (!se->ops ||
1133             (in_postcopy && se->ops->save_live_complete_postcopy) ||
1134             (in_postcopy && !iterable_only) ||
1135             !se->ops->save_live_complete_precopy) {
1136             continue;
1137         }
1138 
1139         if (se->ops && se->ops->is_active) {
1140             if (!se->ops->is_active(se->opaque)) {
1141                 continue;
1142             }
1143         }
1144         trace_savevm_section_start(se->idstr, se->section_id);
1145 
1146         save_section_header(f, se, QEMU_VM_SECTION_END);
1147 
1148         ret = se->ops->save_live_complete_precopy(f, se->opaque);
1149         trace_savevm_section_end(se->idstr, se->section_id, ret);
1150         save_section_footer(f, se);
1151         if (ret < 0) {
1152             qemu_file_set_error(f, ret);
1153             return;
1154         }
1155     }
1156 
1157     if (iterable_only) {
1158         return;
1159     }
1160 
1161     vmdesc = qjson_new();
1162     json_prop_int(vmdesc, "page_size", qemu_target_page_size());
1163     json_start_array(vmdesc, "devices");
1164     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1165 
1166         if ((!se->ops || !se->ops->save_state) && !se->vmsd) {
1167             continue;
1168         }
1169         if (se->vmsd && !vmstate_save_needed(se->vmsd, se->opaque)) {
1170             trace_savevm_section_skip(se->idstr, se->section_id);
1171             continue;
1172         }
1173 
1174         trace_savevm_section_start(se->idstr, se->section_id);
1175 
1176         json_start_object(vmdesc, NULL);
1177         json_prop_str(vmdesc, "name", se->idstr);
1178         json_prop_int(vmdesc, "instance_id", se->instance_id);
1179 
1180         save_section_header(f, se, QEMU_VM_SECTION_FULL);
1181         vmstate_save(f, se, vmdesc);
1182         trace_savevm_section_end(se->idstr, se->section_id, 0);
1183         save_section_footer(f, se);
1184 
1185         json_end_object(vmdesc);
1186     }
1187 
1188     if (!in_postcopy) {
1189         /* Postcopy stream will still be going */
1190         qemu_put_byte(f, QEMU_VM_EOF);
1191     }
1192 
1193     json_end_array(vmdesc);
1194     qjson_finish(vmdesc);
1195     vmdesc_len = strlen(qjson_get_str(vmdesc));
1196 
1197     if (should_send_vmdesc()) {
1198         qemu_put_byte(f, QEMU_VM_VMDESCRIPTION);
1199         qemu_put_be32(f, vmdesc_len);
1200         qemu_put_buffer(f, (uint8_t *)qjson_get_str(vmdesc), vmdesc_len);
1201     }
1202     qjson_destroy(vmdesc);
1203 
1204     qemu_fflush(f);
1205 }
1206 
1207 /* Give an estimate of the amount left to be transferred,
1208  * the result is split into the amount for units that can and
1209  * for units that can't do postcopy.
1210  */
1211 void qemu_savevm_state_pending(QEMUFile *f, uint64_t threshold_size,
1212                                uint64_t *res_non_postcopiable,
1213                                uint64_t *res_postcopiable)
1214 {
1215     SaveStateEntry *se;
1216 
1217     *res_non_postcopiable = 0;
1218     *res_postcopiable = 0;
1219 
1220 
1221     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1222         if (!se->ops || !se->ops->save_live_pending) {
1223             continue;
1224         }
1225         if (se->ops && se->ops->is_active) {
1226             if (!se->ops->is_active(se->opaque)) {
1227                 continue;
1228             }
1229         }
1230         se->ops->save_live_pending(f, se->opaque, threshold_size,
1231                                    res_non_postcopiable, res_postcopiable);
1232     }
1233 }
1234 
1235 void qemu_savevm_state_cleanup(void)
1236 {
1237     SaveStateEntry *se;
1238 
1239     trace_savevm_state_cleanup();
1240     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1241         if (se->ops && se->ops->cleanup) {
1242             se->ops->cleanup(se->opaque);
1243         }
1244     }
1245 }
1246 
1247 static int qemu_savevm_state(QEMUFile *f, Error **errp)
1248 {
1249     int ret;
1250     MigrationState *ms = migrate_init();
1251     MigrationStatus status;
1252     ms->to_dst_file = f;
1253 
1254     if (migration_is_blocked(errp)) {
1255         ret = -EINVAL;
1256         goto done;
1257     }
1258 
1259     if (migrate_use_block()) {
1260         error_setg(errp, "Block migration and snapshots are incompatible");
1261         ret = -EINVAL;
1262         goto done;
1263     }
1264 
1265     qemu_mutex_unlock_iothread();
1266     qemu_savevm_state_header(f);
1267     qemu_savevm_state_begin(f);
1268     qemu_mutex_lock_iothread();
1269 
1270     while (qemu_file_get_error(f) == 0) {
1271         if (qemu_savevm_state_iterate(f, false) > 0) {
1272             break;
1273         }
1274     }
1275 
1276     ret = qemu_file_get_error(f);
1277     if (ret == 0) {
1278         qemu_savevm_state_complete_precopy(f, false);
1279         ret = qemu_file_get_error(f);
1280     }
1281     qemu_savevm_state_cleanup();
1282     if (ret != 0) {
1283         error_setg_errno(errp, -ret, "Error while writing VM state");
1284     }
1285 
1286 done:
1287     if (ret != 0) {
1288         status = MIGRATION_STATUS_FAILED;
1289     } else {
1290         status = MIGRATION_STATUS_COMPLETED;
1291     }
1292     migrate_set_state(&ms->state, MIGRATION_STATUS_SETUP, status);
1293 
1294     /* f is outer parameter, it should not stay in global migration state after
1295      * this function finished */
1296     ms->to_dst_file = NULL;
1297 
1298     return ret;
1299 }
1300 
1301 static int qemu_save_device_state(QEMUFile *f)
1302 {
1303     SaveStateEntry *se;
1304 
1305     qemu_put_be32(f, QEMU_VM_FILE_MAGIC);
1306     qemu_put_be32(f, QEMU_VM_FILE_VERSION);
1307 
1308     cpu_synchronize_all_states();
1309 
1310     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1311         if (se->is_ram) {
1312             continue;
1313         }
1314         if ((!se->ops || !se->ops->save_state) && !se->vmsd) {
1315             continue;
1316         }
1317         if (se->vmsd && !vmstate_save_needed(se->vmsd, se->opaque)) {
1318             continue;
1319         }
1320 
1321         save_section_header(f, se, QEMU_VM_SECTION_FULL);
1322 
1323         vmstate_save(f, se, NULL);
1324 
1325         save_section_footer(f, se);
1326     }
1327 
1328     qemu_put_byte(f, QEMU_VM_EOF);
1329 
1330     return qemu_file_get_error(f);
1331 }
1332 
1333 static SaveStateEntry *find_se(const char *idstr, int instance_id)
1334 {
1335     SaveStateEntry *se;
1336 
1337     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1338         if (!strcmp(se->idstr, idstr) &&
1339             (instance_id == se->instance_id ||
1340              instance_id == se->alias_id))
1341             return se;
1342         /* Migrating from an older version? */
1343         if (strstr(se->idstr, idstr) && se->compat) {
1344             if (!strcmp(se->compat->idstr, idstr) &&
1345                 (instance_id == se->compat->instance_id ||
1346                  instance_id == se->alias_id))
1347                 return se;
1348         }
1349     }
1350     return NULL;
1351 }
1352 
1353 enum LoadVMExitCodes {
1354     /* Allow a command to quit all layers of nested loadvm loops */
1355     LOADVM_QUIT     =  1,
1356 };
1357 
1358 static int qemu_loadvm_state_main(QEMUFile *f, MigrationIncomingState *mis);
1359 
1360 /* ------ incoming postcopy messages ------ */
1361 /* 'advise' arrives before any transfers just to tell us that a postcopy
1362  * *might* happen - it might be skipped if precopy transferred everything
1363  * quickly.
1364  */
1365 static int loadvm_postcopy_handle_advise(MigrationIncomingState *mis)
1366 {
1367     PostcopyState ps = postcopy_state_set(POSTCOPY_INCOMING_ADVISE);
1368     uint64_t remote_pagesize_summary, local_pagesize_summary, remote_tps;
1369 
1370     trace_loadvm_postcopy_handle_advise();
1371     if (ps != POSTCOPY_INCOMING_NONE) {
1372         error_report("CMD_POSTCOPY_ADVISE in wrong postcopy state (%d)", ps);
1373         return -1;
1374     }
1375 
1376     if (!postcopy_ram_supported_by_host()) {
1377         postcopy_state_set(POSTCOPY_INCOMING_NONE);
1378         return -1;
1379     }
1380 
1381     remote_pagesize_summary = qemu_get_be64(mis->from_src_file);
1382     local_pagesize_summary = ram_pagesize_summary();
1383 
1384     if (remote_pagesize_summary != local_pagesize_summary)  {
1385         /*
1386          * This detects two potential causes of mismatch:
1387          *   a) A mismatch in host page sizes
1388          *      Some combinations of mismatch are probably possible but it gets
1389          *      a bit more complicated.  In particular we need to place whole
1390          *      host pages on the dest at once, and we need to ensure that we
1391          *      handle dirtying to make sure we never end up sending part of
1392          *      a hostpage on it's own.
1393          *   b) The use of different huge page sizes on source/destination
1394          *      a more fine grain test is performed during RAM block migration
1395          *      but this test here causes a nice early clear failure, and
1396          *      also fails when passed to an older qemu that doesn't
1397          *      do huge pages.
1398          */
1399         error_report("Postcopy needs matching RAM page sizes (s=%" PRIx64
1400                                                              " d=%" PRIx64 ")",
1401                      remote_pagesize_summary, local_pagesize_summary);
1402         return -1;
1403     }
1404 
1405     remote_tps = qemu_get_be64(mis->from_src_file);
1406     if (remote_tps != qemu_target_page_size()) {
1407         /*
1408          * Again, some differences could be dealt with, but for now keep it
1409          * simple.
1410          */
1411         error_report("Postcopy needs matching target page sizes (s=%d d=%zd)",
1412                      (int)remote_tps, qemu_target_page_size());
1413         return -1;
1414     }
1415 
1416     if (ram_postcopy_incoming_init(mis)) {
1417         return -1;
1418     }
1419 
1420     postcopy_state_set(POSTCOPY_INCOMING_ADVISE);
1421 
1422     return 0;
1423 }
1424 
1425 /* After postcopy we will be told to throw some pages away since they're
1426  * dirty and will have to be demand fetched.  Must happen before CPU is
1427  * started.
1428  * There can be 0..many of these messages, each encoding multiple pages.
1429  */
1430 static int loadvm_postcopy_ram_handle_discard(MigrationIncomingState *mis,
1431                                               uint16_t len)
1432 {
1433     int tmp;
1434     char ramid[256];
1435     PostcopyState ps = postcopy_state_get();
1436 
1437     trace_loadvm_postcopy_ram_handle_discard();
1438 
1439     switch (ps) {
1440     case POSTCOPY_INCOMING_ADVISE:
1441         /* 1st discard */
1442         tmp = postcopy_ram_prepare_discard(mis);
1443         if (tmp) {
1444             return tmp;
1445         }
1446         break;
1447 
1448     case POSTCOPY_INCOMING_DISCARD:
1449         /* Expected state */
1450         break;
1451 
1452     default:
1453         error_report("CMD_POSTCOPY_RAM_DISCARD in wrong postcopy state (%d)",
1454                      ps);
1455         return -1;
1456     }
1457     /* We're expecting a
1458      *    Version (0)
1459      *    a RAM ID string (length byte, name, 0 term)
1460      *    then at least 1 16 byte chunk
1461     */
1462     if (len < (1 + 1 + 1 + 1 + 2 * 8)) {
1463         error_report("CMD_POSTCOPY_RAM_DISCARD invalid length (%d)", len);
1464         return -1;
1465     }
1466 
1467     tmp = qemu_get_byte(mis->from_src_file);
1468     if (tmp != postcopy_ram_discard_version) {
1469         error_report("CMD_POSTCOPY_RAM_DISCARD invalid version (%d)", tmp);
1470         return -1;
1471     }
1472 
1473     if (!qemu_get_counted_string(mis->from_src_file, ramid)) {
1474         error_report("CMD_POSTCOPY_RAM_DISCARD Failed to read RAMBlock ID");
1475         return -1;
1476     }
1477     tmp = qemu_get_byte(mis->from_src_file);
1478     if (tmp != 0) {
1479         error_report("CMD_POSTCOPY_RAM_DISCARD missing nil (%d)", tmp);
1480         return -1;
1481     }
1482 
1483     len -= 3 + strlen(ramid);
1484     if (len % 16) {
1485         error_report("CMD_POSTCOPY_RAM_DISCARD invalid length (%d)", len);
1486         return -1;
1487     }
1488     trace_loadvm_postcopy_ram_handle_discard_header(ramid, len);
1489     while (len) {
1490         uint64_t start_addr, block_length;
1491         start_addr = qemu_get_be64(mis->from_src_file);
1492         block_length = qemu_get_be64(mis->from_src_file);
1493 
1494         len -= 16;
1495         int ret = ram_discard_range(ramid, start_addr, block_length);
1496         if (ret) {
1497             return ret;
1498         }
1499     }
1500     trace_loadvm_postcopy_ram_handle_discard_end();
1501 
1502     return 0;
1503 }
1504 
1505 /*
1506  * Triggered by a postcopy_listen command; this thread takes over reading
1507  * the input stream, leaving the main thread free to carry on loading the rest
1508  * of the device state (from RAM).
1509  * (TODO:This could do with being in a postcopy file - but there again it's
1510  * just another input loop, not that postcopy specific)
1511  */
1512 static void *postcopy_ram_listen_thread(void *opaque)
1513 {
1514     QEMUFile *f = opaque;
1515     MigrationIncomingState *mis = migration_incoming_get_current();
1516     int load_res;
1517 
1518     migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE,
1519                                    MIGRATION_STATUS_POSTCOPY_ACTIVE);
1520     qemu_sem_post(&mis->listen_thread_sem);
1521     trace_postcopy_ram_listen_thread_start();
1522 
1523     /*
1524      * Because we're a thread and not a coroutine we can't yield
1525      * in qemu_file, and thus we must be blocking now.
1526      */
1527     qemu_file_set_blocking(f, true);
1528     load_res = qemu_loadvm_state_main(f, mis);
1529     /* And non-blocking again so we don't block in any cleanup */
1530     qemu_file_set_blocking(f, false);
1531 
1532     trace_postcopy_ram_listen_thread_exit();
1533     if (load_res < 0) {
1534         error_report("%s: loadvm failed: %d", __func__, load_res);
1535         qemu_file_set_error(f, load_res);
1536         migrate_set_state(&mis->state, MIGRATION_STATUS_POSTCOPY_ACTIVE,
1537                                        MIGRATION_STATUS_FAILED);
1538     } else {
1539         /*
1540          * This looks good, but it's possible that the device loading in the
1541          * main thread hasn't finished yet, and so we might not be in 'RUN'
1542          * state yet; wait for the end of the main thread.
1543          */
1544         qemu_event_wait(&mis->main_thread_load_event);
1545     }
1546     postcopy_ram_incoming_cleanup(mis);
1547 
1548     if (load_res < 0) {
1549         /*
1550          * If something went wrong then we have a bad state so exit;
1551          * depending how far we got it might be possible at this point
1552          * to leave the guest running and fire MCEs for pages that never
1553          * arrived as a desperate recovery step.
1554          */
1555         exit(EXIT_FAILURE);
1556     }
1557 
1558     migrate_set_state(&mis->state, MIGRATION_STATUS_POSTCOPY_ACTIVE,
1559                                    MIGRATION_STATUS_COMPLETED);
1560     /*
1561      * If everything has worked fine, then the main thread has waited
1562      * for us to start, and we're the last use of the mis.
1563      * (If something broke then qemu will have to exit anyway since it's
1564      * got a bad migration state).
1565      */
1566     migration_incoming_state_destroy();
1567 
1568 
1569     return NULL;
1570 }
1571 
1572 /* After this message we must be able to immediately receive postcopy data */
1573 static int loadvm_postcopy_handle_listen(MigrationIncomingState *mis)
1574 {
1575     PostcopyState ps = postcopy_state_set(POSTCOPY_INCOMING_LISTENING);
1576     trace_loadvm_postcopy_handle_listen();
1577     if (ps != POSTCOPY_INCOMING_ADVISE && ps != POSTCOPY_INCOMING_DISCARD) {
1578         error_report("CMD_POSTCOPY_LISTEN in wrong postcopy state (%d)", ps);
1579         return -1;
1580     }
1581     if (ps == POSTCOPY_INCOMING_ADVISE) {
1582         /*
1583          * A rare case, we entered listen without having to do any discards,
1584          * so do the setup that's normally done at the time of the 1st discard.
1585          */
1586         postcopy_ram_prepare_discard(mis);
1587     }
1588 
1589     /*
1590      * Sensitise RAM - can now generate requests for blocks that don't exist
1591      * However, at this point the CPU shouldn't be running, and the IO
1592      * shouldn't be doing anything yet so don't actually expect requests
1593      */
1594     if (postcopy_ram_enable_notify(mis)) {
1595         return -1;
1596     }
1597 
1598     if (mis->have_listen_thread) {
1599         error_report("CMD_POSTCOPY_RAM_LISTEN already has a listen thread");
1600         return -1;
1601     }
1602 
1603     mis->have_listen_thread = true;
1604     /* Start up the listening thread and wait for it to signal ready */
1605     qemu_sem_init(&mis->listen_thread_sem, 0);
1606     qemu_thread_create(&mis->listen_thread, "postcopy/listen",
1607                        postcopy_ram_listen_thread, mis->from_src_file,
1608                        QEMU_THREAD_DETACHED);
1609     qemu_sem_wait(&mis->listen_thread_sem);
1610     qemu_sem_destroy(&mis->listen_thread_sem);
1611 
1612     return 0;
1613 }
1614 
1615 
1616 typedef struct {
1617     QEMUBH *bh;
1618 } HandleRunBhData;
1619 
1620 static void loadvm_postcopy_handle_run_bh(void *opaque)
1621 {
1622     Error *local_err = NULL;
1623     HandleRunBhData *data = opaque;
1624 
1625     /* TODO we should move all of this lot into postcopy_ram.c or a shared code
1626      * in migration.c
1627      */
1628     cpu_synchronize_all_post_init();
1629 
1630     qemu_announce_self();
1631 
1632     /* Make sure all file formats flush their mutable metadata.
1633      * If we get an error here, just don't restart the VM yet. */
1634     bdrv_invalidate_cache_all(&local_err);
1635     if (local_err) {
1636         error_report_err(local_err);
1637         local_err = NULL;
1638         autostart = false;
1639     }
1640 
1641     trace_loadvm_postcopy_handle_run_cpu_sync();
1642     cpu_synchronize_all_post_init();
1643 
1644     trace_loadvm_postcopy_handle_run_vmstart();
1645 
1646     if (autostart) {
1647         /* Hold onto your hats, starting the CPU */
1648         vm_start();
1649     } else {
1650         /* leave it paused and let management decide when to start the CPU */
1651         runstate_set(RUN_STATE_PAUSED);
1652     }
1653 
1654     qemu_bh_delete(data->bh);
1655     g_free(data);
1656 }
1657 
1658 /* After all discards we can start running and asking for pages */
1659 static int loadvm_postcopy_handle_run(MigrationIncomingState *mis)
1660 {
1661     PostcopyState ps = postcopy_state_set(POSTCOPY_INCOMING_RUNNING);
1662     HandleRunBhData *data;
1663 
1664     trace_loadvm_postcopy_handle_run();
1665     if (ps != POSTCOPY_INCOMING_LISTENING) {
1666         error_report("CMD_POSTCOPY_RUN in wrong postcopy state (%d)", ps);
1667         return -1;
1668     }
1669 
1670     data = g_new(HandleRunBhData, 1);
1671     data->bh = qemu_bh_new(loadvm_postcopy_handle_run_bh, data);
1672     qemu_bh_schedule(data->bh);
1673 
1674     /* We need to finish reading the stream from the package
1675      * and also stop reading anything more from the stream that loaded the
1676      * package (since it's now being read by the listener thread).
1677      * LOADVM_QUIT will quit all the layers of nested loadvm loops.
1678      */
1679     return LOADVM_QUIT;
1680 }
1681 
1682 /**
1683  * Immediately following this command is a blob of data containing an embedded
1684  * chunk of migration stream; read it and load it.
1685  *
1686  * @mis: Incoming state
1687  * @length: Length of packaged data to read
1688  *
1689  * Returns: Negative values on error
1690  *
1691  */
1692 static int loadvm_handle_cmd_packaged(MigrationIncomingState *mis)
1693 {
1694     int ret;
1695     size_t length;
1696     QIOChannelBuffer *bioc;
1697 
1698     length = qemu_get_be32(mis->from_src_file);
1699     trace_loadvm_handle_cmd_packaged(length);
1700 
1701     if (length > MAX_VM_CMD_PACKAGED_SIZE) {
1702         error_report("Unreasonably large packaged state: %zu", length);
1703         return -1;
1704     }
1705 
1706     bioc = qio_channel_buffer_new(length);
1707     qio_channel_set_name(QIO_CHANNEL(bioc), "migration-loadvm-buffer");
1708     ret = qemu_get_buffer(mis->from_src_file,
1709                           bioc->data,
1710                           length);
1711     if (ret != length) {
1712         object_unref(OBJECT(bioc));
1713         error_report("CMD_PACKAGED: Buffer receive fail ret=%d length=%zu",
1714                      ret, length);
1715         return (ret < 0) ? ret : -EAGAIN;
1716     }
1717     bioc->usage += length;
1718     trace_loadvm_handle_cmd_packaged_received(ret);
1719 
1720     QEMUFile *packf = qemu_fopen_channel_input(QIO_CHANNEL(bioc));
1721 
1722     ret = qemu_loadvm_state_main(packf, mis);
1723     trace_loadvm_handle_cmd_packaged_main(ret);
1724     qemu_fclose(packf);
1725     object_unref(OBJECT(bioc));
1726 
1727     return ret;
1728 }
1729 
1730 /*
1731  * Process an incoming 'QEMU_VM_COMMAND'
1732  * 0           just a normal return
1733  * LOADVM_QUIT All good, but exit the loop
1734  * <0          Error
1735  */
1736 static int loadvm_process_command(QEMUFile *f)
1737 {
1738     MigrationIncomingState *mis = migration_incoming_get_current();
1739     uint16_t cmd;
1740     uint16_t len;
1741     uint32_t tmp32;
1742 
1743     cmd = qemu_get_be16(f);
1744     len = qemu_get_be16(f);
1745 
1746     trace_loadvm_process_command(cmd, len);
1747     if (cmd >= MIG_CMD_MAX || cmd == MIG_CMD_INVALID) {
1748         error_report("MIG_CMD 0x%x unknown (len 0x%x)", cmd, len);
1749         return -EINVAL;
1750     }
1751 
1752     if (mig_cmd_args[cmd].len != -1 && mig_cmd_args[cmd].len != len) {
1753         error_report("%s received with bad length - expecting %zu, got %d",
1754                      mig_cmd_args[cmd].name,
1755                      (size_t)mig_cmd_args[cmd].len, len);
1756         return -ERANGE;
1757     }
1758 
1759     switch (cmd) {
1760     case MIG_CMD_OPEN_RETURN_PATH:
1761         if (mis->to_src_file) {
1762             error_report("CMD_OPEN_RETURN_PATH called when RP already open");
1763             /* Not really a problem, so don't give up */
1764             return 0;
1765         }
1766         mis->to_src_file = qemu_file_get_return_path(f);
1767         if (!mis->to_src_file) {
1768             error_report("CMD_OPEN_RETURN_PATH failed");
1769             return -1;
1770         }
1771         break;
1772 
1773     case MIG_CMD_PING:
1774         tmp32 = qemu_get_be32(f);
1775         trace_loadvm_process_command_ping(tmp32);
1776         if (!mis->to_src_file) {
1777             error_report("CMD_PING (0x%x) received with no return path",
1778                          tmp32);
1779             return -1;
1780         }
1781         migrate_send_rp_pong(mis, tmp32);
1782         break;
1783 
1784     case MIG_CMD_PACKAGED:
1785         return loadvm_handle_cmd_packaged(mis);
1786 
1787     case MIG_CMD_POSTCOPY_ADVISE:
1788         return loadvm_postcopy_handle_advise(mis);
1789 
1790     case MIG_CMD_POSTCOPY_LISTEN:
1791         return loadvm_postcopy_handle_listen(mis);
1792 
1793     case MIG_CMD_POSTCOPY_RUN:
1794         return loadvm_postcopy_handle_run(mis);
1795 
1796     case MIG_CMD_POSTCOPY_RAM_DISCARD:
1797         return loadvm_postcopy_ram_handle_discard(mis, len);
1798     }
1799 
1800     return 0;
1801 }
1802 
1803 struct LoadStateEntry {
1804     QLIST_ENTRY(LoadStateEntry) entry;
1805     SaveStateEntry *se;
1806     int section_id;
1807     int version_id;
1808 };
1809 
1810 /*
1811  * Read a footer off the wire and check that it matches the expected section
1812  *
1813  * Returns: true if the footer was good
1814  *          false if there is a problem (and calls error_report to say why)
1815  */
1816 static bool check_section_footer(QEMUFile *f, LoadStateEntry *le)
1817 {
1818     uint8_t read_mark;
1819     uint32_t read_section_id;
1820 
1821     if (skip_section_footers) {
1822         /* No footer to check */
1823         return true;
1824     }
1825 
1826     read_mark = qemu_get_byte(f);
1827 
1828     if (read_mark != QEMU_VM_SECTION_FOOTER) {
1829         error_report("Missing section footer for %s", le->se->idstr);
1830         return false;
1831     }
1832 
1833     read_section_id = qemu_get_be32(f);
1834     if (read_section_id != le->section_id) {
1835         error_report("Mismatched section id in footer for %s -"
1836                      " read 0x%x expected 0x%x",
1837                      le->se->idstr, read_section_id, le->section_id);
1838         return false;
1839     }
1840 
1841     /* All good */
1842     return true;
1843 }
1844 
1845 void loadvm_free_handlers(MigrationIncomingState *mis)
1846 {
1847     LoadStateEntry *le, *new_le;
1848 
1849     QLIST_FOREACH_SAFE(le, &mis->loadvm_handlers, entry, new_le) {
1850         QLIST_REMOVE(le, entry);
1851         g_free(le);
1852     }
1853 }
1854 
1855 static int
1856 qemu_loadvm_section_start_full(QEMUFile *f, MigrationIncomingState *mis)
1857 {
1858     uint32_t instance_id, version_id, section_id;
1859     SaveStateEntry *se;
1860     LoadStateEntry *le;
1861     char idstr[256];
1862     int ret;
1863 
1864     /* Read section start */
1865     section_id = qemu_get_be32(f);
1866     if (!qemu_get_counted_string(f, idstr)) {
1867         error_report("Unable to read ID string for section %u",
1868                      section_id);
1869         return -EINVAL;
1870     }
1871     instance_id = qemu_get_be32(f);
1872     version_id = qemu_get_be32(f);
1873 
1874     trace_qemu_loadvm_state_section_startfull(section_id, idstr,
1875             instance_id, version_id);
1876     /* Find savevm section */
1877     se = find_se(idstr, instance_id);
1878     if (se == NULL) {
1879         error_report("Unknown savevm section or instance '%s' %d",
1880                      idstr, instance_id);
1881         return -EINVAL;
1882     }
1883 
1884     /* Validate version */
1885     if (version_id > se->version_id) {
1886         error_report("savevm: unsupported version %d for '%s' v%d",
1887                      version_id, idstr, se->version_id);
1888         return -EINVAL;
1889     }
1890 
1891     /* Validate if it is a device's state */
1892     if (xen_enabled() && se->is_ram) {
1893         error_report("loadvm: %s RAM loading not allowed on Xen", idstr);
1894         return -EINVAL;
1895     }
1896 
1897     /* Add entry */
1898     le = g_malloc0(sizeof(*le));
1899 
1900     le->se = se;
1901     le->section_id = section_id;
1902     le->version_id = version_id;
1903     QLIST_INSERT_HEAD(&mis->loadvm_handlers, le, entry);
1904 
1905     ret = vmstate_load(f, le->se, le->version_id);
1906     if (ret < 0) {
1907         error_report("error while loading state for instance 0x%x of"
1908                      " device '%s'", instance_id, idstr);
1909         return ret;
1910     }
1911     if (!check_section_footer(f, le)) {
1912         return -EINVAL;
1913     }
1914 
1915     return 0;
1916 }
1917 
1918 static int
1919 qemu_loadvm_section_part_end(QEMUFile *f, MigrationIncomingState *mis)
1920 {
1921     uint32_t section_id;
1922     LoadStateEntry *le;
1923     int ret;
1924 
1925     section_id = qemu_get_be32(f);
1926 
1927     trace_qemu_loadvm_state_section_partend(section_id);
1928     QLIST_FOREACH(le, &mis->loadvm_handlers, entry) {
1929         if (le->section_id == section_id) {
1930             break;
1931         }
1932     }
1933     if (le == NULL) {
1934         error_report("Unknown savevm section %d", section_id);
1935         return -EINVAL;
1936     }
1937 
1938     ret = vmstate_load(f, le->se, le->version_id);
1939     if (ret < 0) {
1940         error_report("error while loading state section id %d(%s)",
1941                      section_id, le->se->idstr);
1942         return ret;
1943     }
1944     if (!check_section_footer(f, le)) {
1945         return -EINVAL;
1946     }
1947 
1948     return 0;
1949 }
1950 
1951 static int qemu_loadvm_state_main(QEMUFile *f, MigrationIncomingState *mis)
1952 {
1953     uint8_t section_type;
1954     int ret = 0;
1955 
1956     while ((section_type = qemu_get_byte(f)) != QEMU_VM_EOF) {
1957         ret = 0;
1958         trace_qemu_loadvm_state_section(section_type);
1959         switch (section_type) {
1960         case QEMU_VM_SECTION_START:
1961         case QEMU_VM_SECTION_FULL:
1962             ret = qemu_loadvm_section_start_full(f, mis);
1963             if (ret < 0) {
1964                 goto out;
1965             }
1966             break;
1967         case QEMU_VM_SECTION_PART:
1968         case QEMU_VM_SECTION_END:
1969             ret = qemu_loadvm_section_part_end(f, mis);
1970             if (ret < 0) {
1971                 goto out;
1972             }
1973             break;
1974         case QEMU_VM_COMMAND:
1975             ret = loadvm_process_command(f);
1976             trace_qemu_loadvm_state_section_command(ret);
1977             if ((ret < 0) || (ret & LOADVM_QUIT)) {
1978                 goto out;
1979             }
1980             break;
1981         default:
1982             error_report("Unknown savevm section type %d", section_type);
1983             ret = -EINVAL;
1984             goto out;
1985         }
1986     }
1987 
1988 out:
1989     if (ret < 0) {
1990         qemu_file_set_error(f, ret);
1991     }
1992     return ret;
1993 }
1994 
1995 int qemu_loadvm_state(QEMUFile *f)
1996 {
1997     MigrationIncomingState *mis = migration_incoming_get_current();
1998     Error *local_err = NULL;
1999     unsigned int v;
2000     int ret;
2001 
2002     if (qemu_savevm_state_blocked(&local_err)) {
2003         error_report_err(local_err);
2004         return -EINVAL;
2005     }
2006 
2007     v = qemu_get_be32(f);
2008     if (v != QEMU_VM_FILE_MAGIC) {
2009         error_report("Not a migration stream");
2010         return -EINVAL;
2011     }
2012 
2013     v = qemu_get_be32(f);
2014     if (v == QEMU_VM_FILE_VERSION_COMPAT) {
2015         error_report("SaveVM v2 format is obsolete and don't work anymore");
2016         return -ENOTSUP;
2017     }
2018     if (v != QEMU_VM_FILE_VERSION) {
2019         error_report("Unsupported migration stream version");
2020         return -ENOTSUP;
2021     }
2022 
2023     if (!savevm_state.skip_configuration || enforce_config_section()) {
2024         if (qemu_get_byte(f) != QEMU_VM_CONFIGURATION) {
2025             error_report("Configuration section missing");
2026             return -EINVAL;
2027         }
2028         ret = vmstate_load_state(f, &vmstate_configuration, &savevm_state, 0);
2029 
2030         if (ret) {
2031             return ret;
2032         }
2033     }
2034 
2035     ret = qemu_loadvm_state_main(f, mis);
2036     qemu_event_set(&mis->main_thread_load_event);
2037 
2038     trace_qemu_loadvm_state_post_main(ret);
2039 
2040     if (mis->have_listen_thread) {
2041         /* Listen thread still going, can't clean up yet */
2042         return ret;
2043     }
2044 
2045     if (ret == 0) {
2046         ret = qemu_file_get_error(f);
2047     }
2048 
2049     /*
2050      * Try to read in the VMDESC section as well, so that dumping tools that
2051      * intercept our migration stream have the chance to see it.
2052      */
2053 
2054     /* We've got to be careful; if we don't read the data and just shut the fd
2055      * then the sender can error if we close while it's still sending.
2056      * We also mustn't read data that isn't there; some transports (RDMA)
2057      * will stall waiting for that data when the source has already closed.
2058      */
2059     if (ret == 0 && should_send_vmdesc()) {
2060         uint8_t *buf;
2061         uint32_t size;
2062         uint8_t  section_type = qemu_get_byte(f);
2063 
2064         if (section_type != QEMU_VM_VMDESCRIPTION) {
2065             error_report("Expected vmdescription section, but got %d",
2066                          section_type);
2067             /*
2068              * It doesn't seem worth failing at this point since
2069              * we apparently have an otherwise valid VM state
2070              */
2071         } else {
2072             buf = g_malloc(0x1000);
2073             size = qemu_get_be32(f);
2074 
2075             while (size > 0) {
2076                 uint32_t read_chunk = MIN(size, 0x1000);
2077                 qemu_get_buffer(f, buf, read_chunk);
2078                 size -= read_chunk;
2079             }
2080             g_free(buf);
2081         }
2082     }
2083 
2084     cpu_synchronize_all_post_init();
2085 
2086     return ret;
2087 }
2088 
2089 int save_vmstate(const char *name, Error **errp)
2090 {
2091     BlockDriverState *bs, *bs1;
2092     QEMUSnapshotInfo sn1, *sn = &sn1, old_sn1, *old_sn = &old_sn1;
2093     int ret = -1;
2094     QEMUFile *f;
2095     int saved_vm_running;
2096     uint64_t vm_state_size;
2097     qemu_timeval tv;
2098     struct tm tm;
2099     AioContext *aio_context;
2100 
2101     if (!bdrv_all_can_snapshot(&bs)) {
2102         error_setg(errp, "Device '%s' is writable but does not support "
2103                    "snapshots", bdrv_get_device_name(bs));
2104         return ret;
2105     }
2106 
2107     /* Delete old snapshots of the same name */
2108     if (name) {
2109         ret = bdrv_all_delete_snapshot(name, &bs1, errp);
2110         if (ret < 0) {
2111             error_prepend(errp, "Error while deleting snapshot on device "
2112                           "'%s': ", bdrv_get_device_name(bs1));
2113             return ret;
2114         }
2115     }
2116 
2117     bs = bdrv_all_find_vmstate_bs();
2118     if (bs == NULL) {
2119         error_setg(errp, "No block device can accept snapshots");
2120         return ret;
2121     }
2122     aio_context = bdrv_get_aio_context(bs);
2123 
2124     saved_vm_running = runstate_is_running();
2125 
2126     ret = global_state_store();
2127     if (ret) {
2128         error_setg(errp, "Error saving global state");
2129         return ret;
2130     }
2131     vm_stop(RUN_STATE_SAVE_VM);
2132 
2133     aio_context_acquire(aio_context);
2134 
2135     memset(sn, 0, sizeof(*sn));
2136 
2137     /* fill auxiliary fields */
2138     qemu_gettimeofday(&tv);
2139     sn->date_sec = tv.tv_sec;
2140     sn->date_nsec = tv.tv_usec * 1000;
2141     sn->vm_clock_nsec = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL);
2142 
2143     if (name) {
2144         ret = bdrv_snapshot_find(bs, old_sn, name);
2145         if (ret >= 0) {
2146             pstrcpy(sn->name, sizeof(sn->name), old_sn->name);
2147             pstrcpy(sn->id_str, sizeof(sn->id_str), old_sn->id_str);
2148         } else {
2149             pstrcpy(sn->name, sizeof(sn->name), name);
2150         }
2151     } else {
2152         /* cast below needed for OpenBSD where tv_sec is still 'long' */
2153         localtime_r((const time_t *)&tv.tv_sec, &tm);
2154         strftime(sn->name, sizeof(sn->name), "vm-%Y%m%d%H%M%S", &tm);
2155     }
2156 
2157     /* save the VM state */
2158     f = qemu_fopen_bdrv(bs, 1);
2159     if (!f) {
2160         error_setg(errp, "Could not open VM state file");
2161         goto the_end;
2162     }
2163     ret = qemu_savevm_state(f, errp);
2164     vm_state_size = qemu_ftell(f);
2165     qemu_fclose(f);
2166     if (ret < 0) {
2167         goto the_end;
2168     }
2169 
2170     ret = bdrv_all_create_snapshot(sn, bs, vm_state_size, &bs);
2171     if (ret < 0) {
2172         error_setg(errp, "Error while creating snapshot on '%s'",
2173                    bdrv_get_device_name(bs));
2174         goto the_end;
2175     }
2176 
2177     ret = 0;
2178 
2179  the_end:
2180     aio_context_release(aio_context);
2181     if (saved_vm_running) {
2182         vm_start();
2183     }
2184     return ret;
2185 }
2186 
2187 void qmp_xen_save_devices_state(const char *filename, Error **errp)
2188 {
2189     QEMUFile *f;
2190     QIOChannelFile *ioc;
2191     int saved_vm_running;
2192     int ret;
2193 
2194     saved_vm_running = runstate_is_running();
2195     vm_stop(RUN_STATE_SAVE_VM);
2196     global_state_store_running();
2197 
2198     ioc = qio_channel_file_new_path(filename, O_WRONLY | O_CREAT, 0660, errp);
2199     if (!ioc) {
2200         goto the_end;
2201     }
2202     qio_channel_set_name(QIO_CHANNEL(ioc), "migration-xen-save-state");
2203     f = qemu_fopen_channel_output(QIO_CHANNEL(ioc));
2204     ret = qemu_save_device_state(f);
2205     qemu_fclose(f);
2206     if (ret < 0) {
2207         error_setg(errp, QERR_IO_ERROR);
2208     }
2209 
2210  the_end:
2211     if (saved_vm_running) {
2212         vm_start();
2213     }
2214 }
2215 
2216 void qmp_xen_load_devices_state(const char *filename, Error **errp)
2217 {
2218     QEMUFile *f;
2219     QIOChannelFile *ioc;
2220     int ret;
2221 
2222     /* Guest must be paused before loading the device state; the RAM state
2223      * will already have been loaded by xc
2224      */
2225     if (runstate_is_running()) {
2226         error_setg(errp, "Cannot update device state while vm is running");
2227         return;
2228     }
2229     vm_stop(RUN_STATE_RESTORE_VM);
2230 
2231     ioc = qio_channel_file_new_path(filename, O_RDONLY | O_BINARY, 0, errp);
2232     if (!ioc) {
2233         return;
2234     }
2235     qio_channel_set_name(QIO_CHANNEL(ioc), "migration-xen-load-state");
2236     f = qemu_fopen_channel_input(QIO_CHANNEL(ioc));
2237 
2238     ret = qemu_loadvm_state(f);
2239     qemu_fclose(f);
2240     if (ret < 0) {
2241         error_setg(errp, QERR_IO_ERROR);
2242     }
2243     migration_incoming_state_destroy();
2244 }
2245 
2246 int load_vmstate(const char *name, Error **errp)
2247 {
2248     BlockDriverState *bs, *bs_vm_state;
2249     QEMUSnapshotInfo sn;
2250     QEMUFile *f;
2251     int ret;
2252     AioContext *aio_context;
2253     MigrationIncomingState *mis = migration_incoming_get_current();
2254 
2255     if (!bdrv_all_can_snapshot(&bs)) {
2256         error_setg(errp,
2257                    "Device '%s' is writable but does not support snapshots",
2258                    bdrv_get_device_name(bs));
2259         return -ENOTSUP;
2260     }
2261     ret = bdrv_all_find_snapshot(name, &bs);
2262     if (ret < 0) {
2263         error_setg(errp,
2264                    "Device '%s' does not have the requested snapshot '%s'",
2265                    bdrv_get_device_name(bs), name);
2266         return ret;
2267     }
2268 
2269     bs_vm_state = bdrv_all_find_vmstate_bs();
2270     if (!bs_vm_state) {
2271         error_setg(errp, "No block device supports snapshots");
2272         return -ENOTSUP;
2273     }
2274     aio_context = bdrv_get_aio_context(bs_vm_state);
2275 
2276     /* Don't even try to load empty VM states */
2277     aio_context_acquire(aio_context);
2278     ret = bdrv_snapshot_find(bs_vm_state, &sn, name);
2279     aio_context_release(aio_context);
2280     if (ret < 0) {
2281         return ret;
2282     } else if (sn.vm_state_size == 0) {
2283         error_setg(errp, "This is a disk-only snapshot. Revert to it "
2284                    " offline using qemu-img");
2285         return -EINVAL;
2286     }
2287 
2288     /* Flush all IO requests so they don't interfere with the new state.  */
2289     bdrv_drain_all();
2290 
2291     ret = bdrv_all_goto_snapshot(name, &bs);
2292     if (ret < 0) {
2293         error_setg(errp, "Error %d while activating snapshot '%s' on '%s'",
2294                      ret, name, bdrv_get_device_name(bs));
2295         return ret;
2296     }
2297 
2298     /* restore the VM state */
2299     f = qemu_fopen_bdrv(bs_vm_state, 0);
2300     if (!f) {
2301         error_setg(errp, "Could not open VM state file");
2302         return -EINVAL;
2303     }
2304 
2305     qemu_system_reset(SHUTDOWN_CAUSE_NONE);
2306     mis->from_src_file = f;
2307 
2308     aio_context_acquire(aio_context);
2309     ret = qemu_loadvm_state(f);
2310     qemu_fclose(f);
2311     aio_context_release(aio_context);
2312 
2313     migration_incoming_state_destroy();
2314     if (ret < 0) {
2315         error_setg(errp, "Error %d while loading VM state", ret);
2316         return ret;
2317     }
2318 
2319     return 0;
2320 }
2321 
2322 void vmstate_register_ram(MemoryRegion *mr, DeviceState *dev)
2323 {
2324     qemu_ram_set_idstr(mr->ram_block,
2325                        memory_region_name(mr), dev);
2326 }
2327 
2328 void vmstate_unregister_ram(MemoryRegion *mr, DeviceState *dev)
2329 {
2330     qemu_ram_unset_idstr(mr->ram_block);
2331 }
2332 
2333 void vmstate_register_ram_global(MemoryRegion *mr)
2334 {
2335     vmstate_register_ram(mr, NULL);
2336 }
2337 
2338 bool vmstate_check_only_migratable(const VMStateDescription *vmsd)
2339 {
2340     /* check needed if --only-migratable is specified */
2341     if (!only_migratable) {
2342         return true;
2343     }
2344 
2345     return !(vmsd && vmsd->unmigratable);
2346 }
2347