]>
Commit | Line | Data |
---|---|---|
a672b469 AL |
1 | /* |
2 | * QEMU System Emulator | |
3 | * | |
4 | * Copyright (c) 2003-2008 Fabrice Bellard | |
76cc7b58 JQ |
5 | * Copyright (c) 2009-2015 Red Hat Inc |
6 | * | |
7 | * Authors: | |
8 | * Juan Quintela <quintela@redhat.com> | |
a672b469 AL |
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 | */ | |
a672b469 | 28 | |
1393a485 | 29 | #include "qemu/osdep.h" |
abfd9ce3 | 30 | #include "hw/boards.h" |
88c16567 | 31 | #include "hw/xen/xen.h" |
1422e32d | 32 | #include "net/net.h" |
6666c96a | 33 | #include "migration.h" |
5e22479a | 34 | #include "migration/snapshot.h" |
f8d806c9 | 35 | #include "migration/misc.h" |
f2a8f0a6 | 36 | #include "migration/register.h" |
84a899de | 37 | #include "migration/global_state.h" |
7b1e1a22 | 38 | #include "ram.h" |
40014d81 | 39 | #include "qemu-file-channel.h" |
08a0aee1 | 40 | #include "qemu-file.h" |
20a519a0 | 41 | #include "savevm.h" |
be07b0ac | 42 | #include "postcopy-ram.h" |
e688df6b | 43 | #include "qapi/error.h" |
cc7a8ea7 | 44 | #include "qapi/qmp/qerror.h" |
d49b6836 | 45 | #include "qemu/error-report.h" |
9c17d615 | 46 | #include "sysemu/cpus.h" |
022c62cb | 47 | #include "exec/memory.h" |
51180423 | 48 | #include "exec/target_page.h" |
a7ae8355 | 49 | #include "qmp-commands.h" |
517a13c9 | 50 | #include "trace.h" |
28085f7b | 51 | #include "qemu/iov.h" |
de08c606 | 52 | #include "block/snapshot.h" |
f348b6d1 | 53 | #include "qemu/cutils.h" |
61b67d47 | 54 | #include "io/channel-buffer.h" |
8925839f | 55 | #include "io/channel-file.h" |
a672b469 | 56 | |
18995b98 | 57 | #ifndef ETH_P_RARP |
f8778a77 | 58 | #define ETH_P_RARP 0x8035 |
18995b98 N |
59 | #endif |
60 | #define ARP_HTYPE_ETH 0x0001 | |
61 | #define ARP_PTYPE_IP 0x0800 | |
62 | #define ARP_OP_REQUEST_REV 0x3 | |
63 | ||
093e3c42 DDAG |
64 | const unsigned int postcopy_ram_discard_version = 0; |
65 | ||
20a519a0 JQ |
66 | /* Subcommands for QEMU_VM_COMMAND */ |
67 | enum qemu_vm_cmd { | |
68 | MIG_CMD_INVALID = 0, /* Must be 0 */ | |
69 | MIG_CMD_OPEN_RETURN_PATH, /* Tell the dest to open the Return path */ | |
70 | MIG_CMD_PING, /* Request a PONG on the RP */ | |
71 | ||
72 | MIG_CMD_POSTCOPY_ADVISE, /* Prior to any page transfers, just | |
73 | warn we might want to do PC */ | |
74 | MIG_CMD_POSTCOPY_LISTEN, /* Start listening for incoming | |
75 | pages as it's running. */ | |
76 | MIG_CMD_POSTCOPY_RUN, /* Start execution */ | |
77 | ||
78 | MIG_CMD_POSTCOPY_RAM_DISCARD, /* A list of pages to discard that | |
79 | were previously sent during | |
80 | precopy but are dirty. */ | |
81 | MIG_CMD_PACKAGED, /* Send a wrapped stream within this stream */ | |
82 | MIG_CMD_MAX | |
83 | }; | |
84 | ||
ee555cdf | 85 | #define MAX_VM_CMD_PACKAGED_SIZE UINT32_MAX |
c76ca188 DDAG |
86 | static struct mig_cmd_args { |
87 | ssize_t len; /* -1 = variable */ | |
88 | const char *name; | |
89 | } mig_cmd_args[] = { | |
90 | [MIG_CMD_INVALID] = { .len = -1, .name = "INVALID" }, | |
2e37701e DDAG |
91 | [MIG_CMD_OPEN_RETURN_PATH] = { .len = 0, .name = "OPEN_RETURN_PATH" }, |
92 | [MIG_CMD_PING] = { .len = sizeof(uint32_t), .name = "PING" }, | |
58110f0a | 93 | [MIG_CMD_POSTCOPY_ADVISE] = { .len = -1, .name = "POSTCOPY_ADVISE" }, |
093e3c42 DDAG |
94 | [MIG_CMD_POSTCOPY_LISTEN] = { .len = 0, .name = "POSTCOPY_LISTEN" }, |
95 | [MIG_CMD_POSTCOPY_RUN] = { .len = 0, .name = "POSTCOPY_RUN" }, | |
96 | [MIG_CMD_POSTCOPY_RAM_DISCARD] = { | |
97 | .len = -1, .name = "POSTCOPY_RAM_DISCARD" }, | |
11cf1d98 | 98 | [MIG_CMD_PACKAGED] = { .len = 4, .name = "PACKAGED" }, |
c76ca188 DDAG |
99 | [MIG_CMD_MAX] = { .len = -1, .name = "MAX" }, |
100 | }; | |
101 | ||
58110f0a VSO |
102 | /* Note for MIG_CMD_POSTCOPY_ADVISE: |
103 | * The format of arguments is depending on postcopy mode: | |
104 | * - postcopy RAM only | |
105 | * uint64_t host page size | |
106 | * uint64_t taget page size | |
107 | * | |
108 | * - postcopy RAM and postcopy dirty bitmaps | |
109 | * format is the same as for postcopy RAM only | |
110 | * | |
111 | * - postcopy dirty bitmaps only | |
112 | * Nothing. Command length field is 0. | |
113 | * | |
114 | * Be careful: adding a new postcopy entity with some other parameters should | |
115 | * not break format self-description ability. Good way is to introduce some | |
116 | * generic extendable format with an exception for two old entities. | |
117 | */ | |
118 | ||
18995b98 | 119 | static int announce_self_create(uint8_t *buf, |
5cecf414 | 120 | uint8_t *mac_addr) |
a672b469 | 121 | { |
18995b98 N |
122 | /* Ethernet header. */ |
123 | memset(buf, 0xff, 6); /* destination MAC addr */ | |
124 | memcpy(buf + 6, mac_addr, 6); /* source MAC addr */ | |
125 | *(uint16_t *)(buf + 12) = htons(ETH_P_RARP); /* ethertype */ | |
126 | ||
127 | /* RARP header. */ | |
128 | *(uint16_t *)(buf + 14) = htons(ARP_HTYPE_ETH); /* hardware addr space */ | |
129 | *(uint16_t *)(buf + 16) = htons(ARP_PTYPE_IP); /* protocol addr space */ | |
130 | *(buf + 18) = 6; /* hardware addr length (ethernet) */ | |
131 | *(buf + 19) = 4; /* protocol addr length (IPv4) */ | |
132 | *(uint16_t *)(buf + 20) = htons(ARP_OP_REQUEST_REV); /* opcode */ | |
133 | memcpy(buf + 22, mac_addr, 6); /* source hw addr */ | |
134 | memset(buf + 28, 0x00, 4); /* source protocol addr */ | |
135 | memcpy(buf + 32, mac_addr, 6); /* target hw addr */ | |
136 | memset(buf + 38, 0x00, 4); /* target protocol addr */ | |
137 | ||
138 | /* Padding to get up to 60 bytes (ethernet min packet size, minus FCS). */ | |
139 | memset(buf + 42, 0x00, 18); | |
140 | ||
141 | return 60; /* len (FCS will be added by hardware) */ | |
a672b469 AL |
142 | } |
143 | ||
f401ca22 | 144 | static void qemu_announce_self_iter(NICState *nic, void *opaque) |
a672b469 | 145 | { |
18995b98 | 146 | uint8_t buf[60]; |
f401ca22 MM |
147 | int len; |
148 | ||
9013dca5 | 149 | trace_qemu_announce_self_iter(qemu_ether_ntoa(&nic->conf->macaddr)); |
f401ca22 MM |
150 | len = announce_self_create(buf, nic->conf->macaddr.a); |
151 | ||
b356f76d | 152 | qemu_send_packet_raw(qemu_get_queue(nic), buf, len); |
f401ca22 MM |
153 | } |
154 | ||
155 | ||
156 | static void qemu_announce_self_once(void *opaque) | |
157 | { | |
ed8b330b GN |
158 | static int count = SELF_ANNOUNCE_ROUNDS; |
159 | QEMUTimer *timer = *(QEMUTimer **)opaque; | |
a672b469 | 160 | |
f401ca22 MM |
161 | qemu_foreach_nic(qemu_announce_self_iter, NULL); |
162 | ||
18995b98 N |
163 | if (--count) { |
164 | /* delay 50ms, 150ms, 250ms, ... */ | |
bc72ad67 | 165 | timer_mod(timer, qemu_clock_get_ms(QEMU_CLOCK_REALTIME) + |
508e1180 | 166 | self_announce_delay(count)); |
ed8b330b | 167 | } else { |
5cecf414 EH |
168 | timer_del(timer); |
169 | timer_free(timer); | |
ed8b330b GN |
170 | } |
171 | } | |
172 | ||
173 | void qemu_announce_self(void) | |
174 | { | |
5cecf414 EH |
175 | static QEMUTimer *timer; |
176 | timer = timer_new_ms(QEMU_CLOCK_REALTIME, qemu_announce_self_once, &timer); | |
177 | qemu_announce_self_once(&timer); | |
a672b469 AL |
178 | } |
179 | ||
180 | /***********************************************************/ | |
181 | /* savevm/loadvm support */ | |
182 | ||
05fcc848 KW |
183 | static ssize_t block_writev_buffer(void *opaque, struct iovec *iov, int iovcnt, |
184 | int64_t pos) | |
185 | { | |
186 | int ret; | |
187 | QEMUIOVector qiov; | |
188 | ||
189 | qemu_iovec_init_external(&qiov, iov, iovcnt); | |
190 | ret = bdrv_writev_vmstate(opaque, &qiov, pos); | |
191 | if (ret < 0) { | |
192 | return ret; | |
193 | } | |
194 | ||
195 | return qiov.size; | |
196 | } | |
197 | ||
a202a4c0 DDAG |
198 | static ssize_t block_get_buffer(void *opaque, uint8_t *buf, int64_t pos, |
199 | size_t size) | |
a672b469 | 200 | { |
45566e9c | 201 | return bdrv_load_vmstate(opaque, buf, pos, size); |
a672b469 AL |
202 | } |
203 | ||
204 | static int bdrv_fclose(void *opaque) | |
205 | { | |
ad492c92 | 206 | return bdrv_flush(opaque); |
a672b469 AL |
207 | } |
208 | ||
9229bf3c PB |
209 | static const QEMUFileOps bdrv_read_ops = { |
210 | .get_buffer = block_get_buffer, | |
211 | .close = bdrv_fclose | |
212 | }; | |
213 | ||
214 | static const QEMUFileOps bdrv_write_ops = { | |
05fcc848 KW |
215 | .writev_buffer = block_writev_buffer, |
216 | .close = bdrv_fclose | |
9229bf3c PB |
217 | }; |
218 | ||
45566e9c | 219 | static QEMUFile *qemu_fopen_bdrv(BlockDriverState *bs, int is_writable) |
a672b469 | 220 | { |
38ff78d3 | 221 | if (is_writable) { |
9229bf3c | 222 | return qemu_fopen_ops(bs, &bdrv_write_ops); |
38ff78d3 | 223 | } |
9229bf3c | 224 | return qemu_fopen_ops(bs, &bdrv_read_ops); |
a672b469 AL |
225 | } |
226 | ||
2ff68d07 | 227 | |
bb1a6d8c EH |
228 | /* QEMUFile timer support. |
229 | * Not in qemu-file.c to not add qemu-timer.c as dependency to qemu-file.c | |
230 | */ | |
2ff68d07 | 231 | |
40daca54 | 232 | void timer_put(QEMUFile *f, QEMUTimer *ts) |
2ff68d07 PB |
233 | { |
234 | uint64_t expire_time; | |
235 | ||
e93379b0 | 236 | expire_time = timer_expire_time_ns(ts); |
2ff68d07 PB |
237 | qemu_put_be64(f, expire_time); |
238 | } | |
239 | ||
40daca54 | 240 | void timer_get(QEMUFile *f, QEMUTimer *ts) |
2ff68d07 PB |
241 | { |
242 | uint64_t expire_time; | |
243 | ||
244 | expire_time = qemu_get_be64(f); | |
245 | if (expire_time != -1) { | |
bc72ad67 | 246 | timer_mod_ns(ts, expire_time); |
2ff68d07 | 247 | } else { |
bc72ad67 | 248 | timer_del(ts); |
2ff68d07 PB |
249 | } |
250 | } | |
251 | ||
252 | ||
bb1a6d8c EH |
253 | /* VMState timer support. |
254 | * Not in vmstate.c to not add qemu-timer.c as dependency to vmstate.c | |
255 | */ | |
dde0463b | 256 | |
2c21ee76 | 257 | static int get_timer(QEMUFile *f, void *pv, size_t size, VMStateField *field) |
dde0463b JQ |
258 | { |
259 | QEMUTimer *v = pv; | |
40daca54 | 260 | timer_get(f, v); |
dde0463b JQ |
261 | return 0; |
262 | } | |
263 | ||
2c21ee76 JD |
264 | static int put_timer(QEMUFile *f, void *pv, size_t size, VMStateField *field, |
265 | QJSON *vmdesc) | |
dde0463b | 266 | { |
84e2e3eb | 267 | QEMUTimer *v = pv; |
40daca54 | 268 | timer_put(f, v); |
2c21ee76 JD |
269 | |
270 | return 0; | |
dde0463b JQ |
271 | } |
272 | ||
273 | const VMStateInfo vmstate_info_timer = { | |
274 | .name = "timer", | |
275 | .get = get_timer, | |
276 | .put = put_timer, | |
277 | }; | |
278 | ||
08e99e29 | 279 | |
7685ee6a AW |
280 | typedef struct CompatEntry { |
281 | char idstr[256]; | |
282 | int instance_id; | |
283 | } CompatEntry; | |
284 | ||
a672b469 | 285 | typedef struct SaveStateEntry { |
72cf2d4f | 286 | QTAILQ_ENTRY(SaveStateEntry) entry; |
a672b469 AL |
287 | char idstr[256]; |
288 | int instance_id; | |
4d2ffa08 | 289 | int alias_id; |
a672b469 | 290 | int version_id; |
0f42f657 JQ |
291 | /* version id read from the stream */ |
292 | int load_version_id; | |
a672b469 | 293 | int section_id; |
0f42f657 JQ |
294 | /* section id read from the stream */ |
295 | int load_section_id; | |
22ea40f4 | 296 | SaveVMHandlers *ops; |
9ed7d6ae | 297 | const VMStateDescription *vmsd; |
a672b469 | 298 | void *opaque; |
7685ee6a | 299 | CompatEntry *compat; |
a7ae8355 | 300 | int is_ram; |
a672b469 AL |
301 | } SaveStateEntry; |
302 | ||
0163a2e0 JQ |
303 | typedef struct SaveState { |
304 | QTAILQ_HEAD(, SaveStateEntry) handlers; | |
305 | int global_section_id; | |
61964c23 JQ |
306 | uint32_t len; |
307 | const char *name; | |
59811a32 | 308 | uint32_t target_page_bits; |
0163a2e0 JQ |
309 | } SaveState; |
310 | ||
311 | static SaveState savevm_state = { | |
312 | .handlers = QTAILQ_HEAD_INITIALIZER(savevm_state.handlers), | |
313 | .global_section_id = 0, | |
61964c23 JQ |
314 | }; |
315 | ||
44b1ff31 | 316 | static int configuration_pre_save(void *opaque) |
61964c23 JQ |
317 | { |
318 | SaveState *state = opaque; | |
319 | const char *current_name = MACHINE_GET_CLASS(current_machine)->name; | |
320 | ||
321 | state->len = strlen(current_name); | |
322 | state->name = current_name; | |
46d702b1 | 323 | state->target_page_bits = qemu_target_page_bits(); |
44b1ff31 DDAG |
324 | |
325 | return 0; | |
59811a32 PM |
326 | } |
327 | ||
328 | static int configuration_pre_load(void *opaque) | |
329 | { | |
330 | SaveState *state = opaque; | |
331 | ||
332 | /* If there is no target-page-bits subsection it means the source | |
333 | * predates the variable-target-page-bits support and is using the | |
334 | * minimum possible value for this CPU. | |
335 | */ | |
46d702b1 | 336 | state->target_page_bits = qemu_target_page_bits_min(); |
59811a32 | 337 | return 0; |
61964c23 JQ |
338 | } |
339 | ||
340 | static int configuration_post_load(void *opaque, int version_id) | |
341 | { | |
342 | SaveState *state = opaque; | |
343 | const char *current_name = MACHINE_GET_CLASS(current_machine)->name; | |
344 | ||
345 | if (strncmp(state->name, current_name, state->len) != 0) { | |
15d61692 GK |
346 | error_report("Machine type received is '%.*s' and local is '%s'", |
347 | (int) state->len, state->name, current_name); | |
61964c23 JQ |
348 | return -EINVAL; |
349 | } | |
59811a32 | 350 | |
46d702b1 | 351 | if (state->target_page_bits != qemu_target_page_bits()) { |
59811a32 | 352 | error_report("Received TARGET_PAGE_BITS is %d but local is %d", |
46d702b1 | 353 | state->target_page_bits, qemu_target_page_bits()); |
59811a32 PM |
354 | return -EINVAL; |
355 | } | |
356 | ||
61964c23 JQ |
357 | return 0; |
358 | } | |
359 | ||
59811a32 PM |
360 | /* The target-page-bits subsection is present only if the |
361 | * target page size is not the same as the default (ie the | |
362 | * minimum page size for a variable-page-size guest CPU). | |
363 | * If it is present then it contains the actual target page | |
364 | * bits for the machine, and migration will fail if the | |
365 | * two ends don't agree about it. | |
366 | */ | |
367 | static bool vmstate_target_page_bits_needed(void *opaque) | |
368 | { | |
46d702b1 JQ |
369 | return qemu_target_page_bits() |
370 | > qemu_target_page_bits_min(); | |
59811a32 PM |
371 | } |
372 | ||
373 | static const VMStateDescription vmstate_target_page_bits = { | |
374 | .name = "configuration/target-page-bits", | |
375 | .version_id = 1, | |
376 | .minimum_version_id = 1, | |
377 | .needed = vmstate_target_page_bits_needed, | |
378 | .fields = (VMStateField[]) { | |
379 | VMSTATE_UINT32(target_page_bits, SaveState), | |
380 | VMSTATE_END_OF_LIST() | |
381 | } | |
382 | }; | |
383 | ||
61964c23 JQ |
384 | static const VMStateDescription vmstate_configuration = { |
385 | .name = "configuration", | |
386 | .version_id = 1, | |
59811a32 | 387 | .pre_load = configuration_pre_load, |
61964c23 JQ |
388 | .post_load = configuration_post_load, |
389 | .pre_save = configuration_pre_save, | |
390 | .fields = (VMStateField[]) { | |
391 | VMSTATE_UINT32(len, SaveState), | |
59046ec2 | 392 | VMSTATE_VBUFFER_ALLOC_UINT32(name, SaveState, 0, NULL, len), |
61964c23 JQ |
393 | VMSTATE_END_OF_LIST() |
394 | }, | |
59811a32 PM |
395 | .subsections = (const VMStateDescription*[]) { |
396 | &vmstate_target_page_bits, | |
397 | NULL | |
398 | } | |
0163a2e0 | 399 | }; |
a672b469 | 400 | |
abfd9ce3 AS |
401 | static void dump_vmstate_vmsd(FILE *out_file, |
402 | const VMStateDescription *vmsd, int indent, | |
403 | bool is_subsection); | |
404 | ||
405 | static void dump_vmstate_vmsf(FILE *out_file, const VMStateField *field, | |
406 | int indent) | |
407 | { | |
408 | fprintf(out_file, "%*s{\n", indent, ""); | |
409 | indent += 2; | |
410 | fprintf(out_file, "%*s\"field\": \"%s\",\n", indent, "", field->name); | |
411 | fprintf(out_file, "%*s\"version_id\": %d,\n", indent, "", | |
412 | field->version_id); | |
413 | fprintf(out_file, "%*s\"field_exists\": %s,\n", indent, "", | |
414 | field->field_exists ? "true" : "false"); | |
415 | fprintf(out_file, "%*s\"size\": %zu", indent, "", field->size); | |
416 | if (field->vmsd != NULL) { | |
417 | fprintf(out_file, ",\n"); | |
418 | dump_vmstate_vmsd(out_file, field->vmsd, indent, false); | |
419 | } | |
420 | fprintf(out_file, "\n%*s}", indent - 2, ""); | |
421 | } | |
422 | ||
423 | static void dump_vmstate_vmss(FILE *out_file, | |
5cd8cada | 424 | const VMStateDescription **subsection, |
abfd9ce3 AS |
425 | int indent) |
426 | { | |
5cd8cada JQ |
427 | if (*subsection != NULL) { |
428 | dump_vmstate_vmsd(out_file, *subsection, indent, true); | |
abfd9ce3 AS |
429 | } |
430 | } | |
431 | ||
432 | static void dump_vmstate_vmsd(FILE *out_file, | |
433 | const VMStateDescription *vmsd, int indent, | |
434 | bool is_subsection) | |
435 | { | |
436 | if (is_subsection) { | |
437 | fprintf(out_file, "%*s{\n", indent, ""); | |
438 | } else { | |
439 | fprintf(out_file, "%*s\"%s\": {\n", indent, "", "Description"); | |
440 | } | |
441 | indent += 2; | |
442 | fprintf(out_file, "%*s\"name\": \"%s\",\n", indent, "", vmsd->name); | |
443 | fprintf(out_file, "%*s\"version_id\": %d,\n", indent, "", | |
444 | vmsd->version_id); | |
445 | fprintf(out_file, "%*s\"minimum_version_id\": %d", indent, "", | |
446 | vmsd->minimum_version_id); | |
447 | if (vmsd->fields != NULL) { | |
448 | const VMStateField *field = vmsd->fields; | |
449 | bool first; | |
450 | ||
451 | fprintf(out_file, ",\n%*s\"Fields\": [\n", indent, ""); | |
452 | first = true; | |
453 | while (field->name != NULL) { | |
454 | if (field->flags & VMS_MUST_EXIST) { | |
455 | /* Ignore VMSTATE_VALIDATE bits; these don't get migrated */ | |
456 | field++; | |
457 | continue; | |
458 | } | |
459 | if (!first) { | |
460 | fprintf(out_file, ",\n"); | |
461 | } | |
462 | dump_vmstate_vmsf(out_file, field, indent + 2); | |
463 | field++; | |
464 | first = false; | |
465 | } | |
466 | fprintf(out_file, "\n%*s]", indent, ""); | |
467 | } | |
468 | if (vmsd->subsections != NULL) { | |
5cd8cada | 469 | const VMStateDescription **subsection = vmsd->subsections; |
abfd9ce3 AS |
470 | bool first; |
471 | ||
472 | fprintf(out_file, ",\n%*s\"Subsections\": [\n", indent, ""); | |
473 | first = true; | |
5cd8cada | 474 | while (*subsection != NULL) { |
abfd9ce3 AS |
475 | if (!first) { |
476 | fprintf(out_file, ",\n"); | |
477 | } | |
478 | dump_vmstate_vmss(out_file, subsection, indent + 2); | |
479 | subsection++; | |
480 | first = false; | |
481 | } | |
482 | fprintf(out_file, "\n%*s]", indent, ""); | |
483 | } | |
484 | fprintf(out_file, "\n%*s}", indent - 2, ""); | |
485 | } | |
486 | ||
487 | static void dump_machine_type(FILE *out_file) | |
488 | { | |
489 | MachineClass *mc; | |
490 | ||
491 | mc = MACHINE_GET_CLASS(current_machine); | |
492 | ||
493 | fprintf(out_file, " \"vmschkmachine\": {\n"); | |
494 | fprintf(out_file, " \"Name\": \"%s\"\n", mc->name); | |
495 | fprintf(out_file, " },\n"); | |
496 | } | |
497 | ||
498 | void dump_vmstate_json_to_file(FILE *out_file) | |
499 | { | |
500 | GSList *list, *elt; | |
501 | bool first; | |
502 | ||
503 | fprintf(out_file, "{\n"); | |
504 | dump_machine_type(out_file); | |
505 | ||
506 | first = true; | |
507 | list = object_class_get_list(TYPE_DEVICE, true); | |
508 | for (elt = list; elt; elt = elt->next) { | |
509 | DeviceClass *dc = OBJECT_CLASS_CHECK(DeviceClass, elt->data, | |
510 | TYPE_DEVICE); | |
511 | const char *name; | |
512 | int indent = 2; | |
513 | ||
514 | if (!dc->vmsd) { | |
515 | continue; | |
516 | } | |
517 | ||
518 | if (!first) { | |
519 | fprintf(out_file, ",\n"); | |
520 | } | |
521 | name = object_class_get_name(OBJECT_CLASS(dc)); | |
522 | fprintf(out_file, "%*s\"%s\": {\n", indent, "", name); | |
523 | indent += 2; | |
524 | fprintf(out_file, "%*s\"Name\": \"%s\",\n", indent, "", name); | |
525 | fprintf(out_file, "%*s\"version_id\": %d,\n", indent, "", | |
526 | dc->vmsd->version_id); | |
527 | fprintf(out_file, "%*s\"minimum_version_id\": %d,\n", indent, "", | |
528 | dc->vmsd->minimum_version_id); | |
529 | ||
530 | dump_vmstate_vmsd(out_file, dc->vmsd, indent, false); | |
531 | ||
532 | fprintf(out_file, "\n%*s}", indent - 2, ""); | |
533 | first = false; | |
534 | } | |
535 | fprintf(out_file, "\n}\n"); | |
536 | fclose(out_file); | |
537 | } | |
538 | ||
8718e999 JQ |
539 | static int calculate_new_instance_id(const char *idstr) |
540 | { | |
541 | SaveStateEntry *se; | |
542 | int instance_id = 0; | |
543 | ||
0163a2e0 | 544 | QTAILQ_FOREACH(se, &savevm_state.handlers, entry) { |
8718e999 JQ |
545 | if (strcmp(idstr, se->idstr) == 0 |
546 | && instance_id <= se->instance_id) { | |
547 | instance_id = se->instance_id + 1; | |
548 | } | |
549 | } | |
550 | return instance_id; | |
551 | } | |
552 | ||
7685ee6a AW |
553 | static int calculate_compat_instance_id(const char *idstr) |
554 | { | |
555 | SaveStateEntry *se; | |
556 | int instance_id = 0; | |
557 | ||
0163a2e0 | 558 | QTAILQ_FOREACH(se, &savevm_state.handlers, entry) { |
38ff78d3 | 559 | if (!se->compat) { |
7685ee6a | 560 | continue; |
38ff78d3 | 561 | } |
7685ee6a AW |
562 | |
563 | if (strcmp(idstr, se->compat->idstr) == 0 | |
564 | && instance_id <= se->compat->instance_id) { | |
565 | instance_id = se->compat->instance_id + 1; | |
566 | } | |
567 | } | |
568 | return instance_id; | |
569 | } | |
570 | ||
f37bc036 PX |
571 | static inline MigrationPriority save_state_priority(SaveStateEntry *se) |
572 | { | |
573 | if (se->vmsd) { | |
574 | return se->vmsd->priority; | |
575 | } | |
576 | return MIG_PRI_DEFAULT; | |
577 | } | |
578 | ||
579 | static void savevm_state_handler_insert(SaveStateEntry *nse) | |
580 | { | |
581 | MigrationPriority priority = save_state_priority(nse); | |
582 | SaveStateEntry *se; | |
583 | ||
584 | assert(priority <= MIG_PRI_MAX); | |
585 | ||
586 | QTAILQ_FOREACH(se, &savevm_state.handlers, entry) { | |
587 | if (save_state_priority(se) < priority) { | |
588 | break; | |
589 | } | |
590 | } | |
591 | ||
592 | if (se) { | |
593 | QTAILQ_INSERT_BEFORE(se, nse, entry); | |
594 | } else { | |
595 | QTAILQ_INSERT_TAIL(&savevm_state.handlers, nse, entry); | |
596 | } | |
597 | } | |
598 | ||
a672b469 AL |
599 | /* TODO: Individual devices generally have very little idea about the rest |
600 | of the system, so instance_id should be removed/replaced. | |
601 | Meanwhile pass -1 as instance_id if you do not already have a clearly | |
602 | distinguishing id for all instances of your device class. */ | |
0be71e32 AW |
603 | int register_savevm_live(DeviceState *dev, |
604 | const char *idstr, | |
a672b469 AL |
605 | int instance_id, |
606 | int version_id, | |
7908c78d | 607 | SaveVMHandlers *ops, |
a672b469 AL |
608 | void *opaque) |
609 | { | |
8718e999 | 610 | SaveStateEntry *se; |
a672b469 | 611 | |
97f3ad35 | 612 | se = g_new0(SaveStateEntry, 1); |
a672b469 | 613 | se->version_id = version_id; |
0163a2e0 | 614 | se->section_id = savevm_state.global_section_id++; |
7908c78d | 615 | se->ops = ops; |
a672b469 | 616 | se->opaque = opaque; |
9ed7d6ae | 617 | se->vmsd = NULL; |
a7ae8355 | 618 | /* if this is a live_savem then set is_ram */ |
9907e842 | 619 | if (ops->save_setup != NULL) { |
a7ae8355 SS |
620 | se->is_ram = 1; |
621 | } | |
a672b469 | 622 | |
09e5ab63 AL |
623 | if (dev) { |
624 | char *id = qdev_get_dev_path(dev); | |
7685ee6a | 625 | if (id) { |
581f08ba DDAG |
626 | if (snprintf(se->idstr, sizeof(se->idstr), "%s/", id) >= |
627 | sizeof(se->idstr)) { | |
628 | error_report("Path too long for VMState (%s)", id); | |
629 | g_free(id); | |
630 | g_free(se); | |
631 | ||
632 | return -1; | |
633 | } | |
7267c094 | 634 | g_free(id); |
7685ee6a | 635 | |
97f3ad35 | 636 | se->compat = g_new0(CompatEntry, 1); |
7685ee6a AW |
637 | pstrcpy(se->compat->idstr, sizeof(se->compat->idstr), idstr); |
638 | se->compat->instance_id = instance_id == -1 ? | |
639 | calculate_compat_instance_id(idstr) : instance_id; | |
640 | instance_id = -1; | |
641 | } | |
642 | } | |
643 | pstrcat(se->idstr, sizeof(se->idstr), idstr); | |
644 | ||
8718e999 | 645 | if (instance_id == -1) { |
7685ee6a | 646 | se->instance_id = calculate_new_instance_id(se->idstr); |
8718e999 JQ |
647 | } else { |
648 | se->instance_id = instance_id; | |
a672b469 | 649 | } |
7685ee6a | 650 | assert(!se->compat || se->instance_id == 0); |
f37bc036 | 651 | savevm_state_handler_insert(se); |
a672b469 AL |
652 | return 0; |
653 | } | |
654 | ||
0be71e32 | 655 | void unregister_savevm(DeviceState *dev, const char *idstr, void *opaque) |
41bd13af | 656 | { |
8718e999 | 657 | SaveStateEntry *se, *new_se; |
7685ee6a AW |
658 | char id[256] = ""; |
659 | ||
09e5ab63 AL |
660 | if (dev) { |
661 | char *path = qdev_get_dev_path(dev); | |
7685ee6a AW |
662 | if (path) { |
663 | pstrcpy(id, sizeof(id), path); | |
664 | pstrcat(id, sizeof(id), "/"); | |
7267c094 | 665 | g_free(path); |
7685ee6a AW |
666 | } |
667 | } | |
668 | pstrcat(id, sizeof(id), idstr); | |
41bd13af | 669 | |
0163a2e0 | 670 | QTAILQ_FOREACH_SAFE(se, &savevm_state.handlers, entry, new_se) { |
7685ee6a | 671 | if (strcmp(se->idstr, id) == 0 && se->opaque == opaque) { |
0163a2e0 | 672 | QTAILQ_REMOVE(&savevm_state.handlers, se, entry); |
ef1e1e07 | 673 | g_free(se->compat); |
7267c094 | 674 | g_free(se); |
41bd13af | 675 | } |
41bd13af AL |
676 | } |
677 | } | |
678 | ||
0be71e32 | 679 | int vmstate_register_with_alias_id(DeviceState *dev, int instance_id, |
4d2ffa08 JK |
680 | const VMStateDescription *vmsd, |
681 | void *opaque, int alias_id, | |
bc5c4f21 DDAG |
682 | int required_for_version, |
683 | Error **errp) | |
9ed7d6ae | 684 | { |
8718e999 | 685 | SaveStateEntry *se; |
9ed7d6ae | 686 | |
4d2ffa08 JK |
687 | /* If this triggers, alias support can be dropped for the vmsd. */ |
688 | assert(alias_id == -1 || required_for_version >= vmsd->minimum_version_id); | |
689 | ||
97f3ad35 | 690 | se = g_new0(SaveStateEntry, 1); |
9ed7d6ae | 691 | se->version_id = vmsd->version_id; |
0163a2e0 | 692 | se->section_id = savevm_state.global_section_id++; |
9ed7d6ae JQ |
693 | se->opaque = opaque; |
694 | se->vmsd = vmsd; | |
4d2ffa08 | 695 | se->alias_id = alias_id; |
9ed7d6ae | 696 | |
09e5ab63 AL |
697 | if (dev) { |
698 | char *id = qdev_get_dev_path(dev); | |
7685ee6a | 699 | if (id) { |
581f08ba DDAG |
700 | if (snprintf(se->idstr, sizeof(se->idstr), "%s/", id) >= |
701 | sizeof(se->idstr)) { | |
702 | error_setg(errp, "Path too long for VMState (%s)", id); | |
703 | g_free(id); | |
704 | g_free(se); | |
705 | ||
706 | return -1; | |
707 | } | |
128e4e10 | 708 | g_free(id); |
7685ee6a | 709 | |
97f3ad35 | 710 | se->compat = g_new0(CompatEntry, 1); |
7685ee6a AW |
711 | pstrcpy(se->compat->idstr, sizeof(se->compat->idstr), vmsd->name); |
712 | se->compat->instance_id = instance_id == -1 ? | |
713 | calculate_compat_instance_id(vmsd->name) : instance_id; | |
714 | instance_id = -1; | |
715 | } | |
716 | } | |
717 | pstrcat(se->idstr, sizeof(se->idstr), vmsd->name); | |
718 | ||
8718e999 | 719 | if (instance_id == -1) { |
7685ee6a | 720 | se->instance_id = calculate_new_instance_id(se->idstr); |
8718e999 JQ |
721 | } else { |
722 | se->instance_id = instance_id; | |
9ed7d6ae | 723 | } |
7685ee6a | 724 | assert(!se->compat || se->instance_id == 0); |
f37bc036 | 725 | savevm_state_handler_insert(se); |
9ed7d6ae JQ |
726 | return 0; |
727 | } | |
728 | ||
0be71e32 AW |
729 | void vmstate_unregister(DeviceState *dev, const VMStateDescription *vmsd, |
730 | void *opaque) | |
9ed7d6ae | 731 | { |
1eb7538b JQ |
732 | SaveStateEntry *se, *new_se; |
733 | ||
0163a2e0 | 734 | QTAILQ_FOREACH_SAFE(se, &savevm_state.handlers, entry, new_se) { |
1eb7538b | 735 | if (se->vmsd == vmsd && se->opaque == opaque) { |
0163a2e0 | 736 | QTAILQ_REMOVE(&savevm_state.handlers, se, entry); |
ef1e1e07 | 737 | g_free(se->compat); |
7267c094 | 738 | g_free(se); |
1eb7538b JQ |
739 | } |
740 | } | |
9ed7d6ae JQ |
741 | } |
742 | ||
3a011c26 | 743 | static int vmstate_load(QEMUFile *f, SaveStateEntry *se) |
4082be4d | 744 | { |
9013dca5 | 745 | trace_vmstate_load(se->idstr, se->vmsd ? se->vmsd->name : "(old)"); |
9ed7d6ae | 746 | if (!se->vmsd) { /* Old style */ |
3a011c26 | 747 | return se->ops->load_state(f, se->opaque, se->load_version_id); |
9ed7d6ae | 748 | } |
3a011c26 | 749 | return vmstate_load_state(f, se->vmsd, se->opaque, se->load_version_id); |
4082be4d JQ |
750 | } |
751 | ||
8118f095 AG |
752 | static void vmstate_save_old_style(QEMUFile *f, SaveStateEntry *se, QJSON *vmdesc) |
753 | { | |
754 | int64_t old_offset, size; | |
755 | ||
756 | old_offset = qemu_ftell_fast(f); | |
757 | se->ops->save_state(f, se->opaque); | |
758 | size = qemu_ftell_fast(f) - old_offset; | |
759 | ||
760 | if (vmdesc) { | |
761 | json_prop_int(vmdesc, "size", size); | |
762 | json_start_array(vmdesc, "fields"); | |
763 | json_start_object(vmdesc, NULL); | |
764 | json_prop_str(vmdesc, "name", "data"); | |
765 | json_prop_int(vmdesc, "size", size); | |
766 | json_prop_str(vmdesc, "type", "buffer"); | |
767 | json_end_object(vmdesc); | |
768 | json_end_array(vmdesc); | |
769 | } | |
770 | } | |
771 | ||
687433f6 | 772 | static int vmstate_save(QEMUFile *f, SaveStateEntry *se, QJSON *vmdesc) |
4082be4d | 773 | { |
9013dca5 | 774 | trace_vmstate_save(se->idstr, se->vmsd ? se->vmsd->name : "(old)"); |
8118f095 AG |
775 | if (!se->vmsd) { |
776 | vmstate_save_old_style(f, se, vmdesc); | |
687433f6 | 777 | return 0; |
9ed7d6ae | 778 | } |
687433f6 | 779 | return vmstate_save_state(f, se->vmsd, se->opaque, vmdesc); |
4082be4d JQ |
780 | } |
781 | ||
ce39bfc9 DDAG |
782 | /* |
783 | * Write the header for device section (QEMU_VM_SECTION START/END/PART/FULL) | |
784 | */ | |
785 | static void save_section_header(QEMUFile *f, SaveStateEntry *se, | |
786 | uint8_t section_type) | |
787 | { | |
788 | qemu_put_byte(f, section_type); | |
789 | qemu_put_be32(f, se->section_id); | |
790 | ||
791 | if (section_type == QEMU_VM_SECTION_FULL || | |
792 | section_type == QEMU_VM_SECTION_START) { | |
793 | /* ID string */ | |
794 | size_t len = strlen(se->idstr); | |
795 | qemu_put_byte(f, len); | |
796 | qemu_put_buffer(f, (uint8_t *)se->idstr, len); | |
797 | ||
798 | qemu_put_be32(f, se->instance_id); | |
799 | qemu_put_be32(f, se->version_id); | |
800 | } | |
801 | } | |
802 | ||
f68945d4 DDAG |
803 | /* |
804 | * Write a footer onto device sections that catches cases misformatted device | |
805 | * sections. | |
806 | */ | |
807 | static void save_section_footer(QEMUFile *f, SaveStateEntry *se) | |
808 | { | |
15c38503 | 809 | if (migrate_get_current()->send_section_footer) { |
f68945d4 DDAG |
810 | qemu_put_byte(f, QEMU_VM_SECTION_FOOTER); |
811 | qemu_put_be32(f, se->section_id); | |
812 | } | |
813 | } | |
814 | ||
c76ca188 DDAG |
815 | /** |
816 | * qemu_savevm_command_send: Send a 'QEMU_VM_COMMAND' type element with the | |
817 | * command and associated data. | |
818 | * | |
819 | * @f: File to send command on | |
820 | * @command: Command type to send | |
821 | * @len: Length of associated data | |
822 | * @data: Data associated with command. | |
823 | */ | |
20a519a0 JQ |
824 | static void qemu_savevm_command_send(QEMUFile *f, |
825 | enum qemu_vm_cmd command, | |
826 | uint16_t len, | |
827 | uint8_t *data) | |
c76ca188 DDAG |
828 | { |
829 | trace_savevm_command_send(command, len); | |
830 | qemu_put_byte(f, QEMU_VM_COMMAND); | |
831 | qemu_put_be16(f, (uint16_t)command); | |
832 | qemu_put_be16(f, len); | |
833 | qemu_put_buffer(f, data, len); | |
834 | qemu_fflush(f); | |
835 | } | |
836 | ||
2e37701e DDAG |
837 | void qemu_savevm_send_ping(QEMUFile *f, uint32_t value) |
838 | { | |
839 | uint32_t buf; | |
840 | ||
841 | trace_savevm_send_ping(value); | |
842 | buf = cpu_to_be32(value); | |
843 | qemu_savevm_command_send(f, MIG_CMD_PING, sizeof(value), (uint8_t *)&buf); | |
844 | } | |
845 | ||
846 | void qemu_savevm_send_open_return_path(QEMUFile *f) | |
847 | { | |
848 | trace_savevm_send_open_return_path(); | |
849 | qemu_savevm_command_send(f, MIG_CMD_OPEN_RETURN_PATH, 0, NULL); | |
850 | } | |
851 | ||
11cf1d98 DDAG |
852 | /* We have a buffer of data to send; we don't want that all to be loaded |
853 | * by the command itself, so the command contains just the length of the | |
854 | * extra buffer that we then send straight after it. | |
855 | * TODO: Must be a better way to organise that | |
856 | * | |
857 | * Returns: | |
858 | * 0 on success | |
859 | * -ve on error | |
860 | */ | |
61b67d47 | 861 | int qemu_savevm_send_packaged(QEMUFile *f, const uint8_t *buf, size_t len) |
11cf1d98 | 862 | { |
11cf1d98 DDAG |
863 | uint32_t tmp; |
864 | ||
865 | if (len > MAX_VM_CMD_PACKAGED_SIZE) { | |
866 | error_report("%s: Unreasonably large packaged state: %zu", | |
867 | __func__, len); | |
868 | return -1; | |
869 | } | |
870 | ||
871 | tmp = cpu_to_be32(len); | |
872 | ||
873 | trace_qemu_savevm_send_packaged(); | |
874 | qemu_savevm_command_send(f, MIG_CMD_PACKAGED, 4, (uint8_t *)&tmp); | |
875 | ||
61b67d47 | 876 | qemu_put_buffer(f, buf, len); |
11cf1d98 DDAG |
877 | |
878 | return 0; | |
879 | } | |
880 | ||
093e3c42 DDAG |
881 | /* Send prior to any postcopy transfer */ |
882 | void qemu_savevm_send_postcopy_advise(QEMUFile *f) | |
883 | { | |
58110f0a VSO |
884 | if (migrate_postcopy_ram()) { |
885 | uint64_t tmp[2]; | |
886 | tmp[0] = cpu_to_be64(ram_pagesize_summary()); | |
887 | tmp[1] = cpu_to_be64(qemu_target_page_size()); | |
888 | ||
889 | trace_qemu_savevm_send_postcopy_advise(); | |
890 | qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_ADVISE, | |
891 | 16, (uint8_t *)tmp); | |
892 | } else { | |
893 | qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_ADVISE, 0, NULL); | |
894 | } | |
093e3c42 DDAG |
895 | } |
896 | ||
897 | /* Sent prior to starting the destination running in postcopy, discard pages | |
898 | * that have already been sent but redirtied on the source. | |
899 | * CMD_POSTCOPY_RAM_DISCARD consist of: | |
900 | * byte version (0) | |
901 | * byte Length of name field (not including 0) | |
902 | * n x byte RAM block name | |
903 | * byte 0 terminator (just for safety) | |
904 | * n x Byte ranges within the named RAMBlock | |
905 | * be64 Start of the range | |
906 | * be64 Length | |
907 | * | |
908 | * name: RAMBlock name that these entries are part of | |
909 | * len: Number of page entries | |
910 | * start_list: 'len' addresses | |
911 | * length_list: 'len' addresses | |
912 | * | |
913 | */ | |
914 | void qemu_savevm_send_postcopy_ram_discard(QEMUFile *f, const char *name, | |
915 | uint16_t len, | |
916 | uint64_t *start_list, | |
917 | uint64_t *length_list) | |
918 | { | |
919 | uint8_t *buf; | |
920 | uint16_t tmplen; | |
921 | uint16_t t; | |
922 | size_t name_len = strlen(name); | |
923 | ||
924 | trace_qemu_savevm_send_postcopy_ram_discard(name, len); | |
925 | assert(name_len < 256); | |
926 | buf = g_malloc0(1 + 1 + name_len + 1 + (8 + 8) * len); | |
927 | buf[0] = postcopy_ram_discard_version; | |
928 | buf[1] = name_len; | |
929 | memcpy(buf + 2, name, name_len); | |
930 | tmplen = 2 + name_len; | |
931 | buf[tmplen++] = '\0'; | |
932 | ||
933 | for (t = 0; t < len; t++) { | |
4d885131 | 934 | stq_be_p(buf + tmplen, start_list[t]); |
093e3c42 | 935 | tmplen += 8; |
4d885131 | 936 | stq_be_p(buf + tmplen, length_list[t]); |
093e3c42 DDAG |
937 | tmplen += 8; |
938 | } | |
939 | qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_RAM_DISCARD, tmplen, buf); | |
940 | g_free(buf); | |
941 | } | |
942 | ||
943 | /* Get the destination into a state where it can receive postcopy data. */ | |
944 | void qemu_savevm_send_postcopy_listen(QEMUFile *f) | |
945 | { | |
946 | trace_savevm_send_postcopy_listen(); | |
947 | qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_LISTEN, 0, NULL); | |
948 | } | |
949 | ||
950 | /* Kick the destination into running */ | |
951 | void qemu_savevm_send_postcopy_run(QEMUFile *f) | |
952 | { | |
953 | trace_savevm_send_postcopy_run(); | |
954 | qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_RUN, 0, NULL); | |
955 | } | |
956 | ||
e1c37d0e | 957 | bool qemu_savevm_state_blocked(Error **errp) |
dc912121 AW |
958 | { |
959 | SaveStateEntry *se; | |
960 | ||
0163a2e0 | 961 | QTAILQ_FOREACH(se, &savevm_state.handlers, entry) { |
7d854c47 | 962 | if (se->vmsd && se->vmsd->unmigratable) { |
f231b88d CR |
963 | error_setg(errp, "State blocked by non-migratable device '%s'", |
964 | se->idstr); | |
dc912121 AW |
965 | return true; |
966 | } | |
967 | } | |
968 | return false; | |
969 | } | |
970 | ||
f796baa1 DDAG |
971 | void qemu_savevm_state_header(QEMUFile *f) |
972 | { | |
973 | trace_savevm_state_header(); | |
974 | qemu_put_be32(f, QEMU_VM_FILE_MAGIC); | |
975 | qemu_put_be32(f, QEMU_VM_FILE_VERSION); | |
172dfd4f | 976 | |
4ffdb337 | 977 | if (migrate_get_current()->send_configuration) { |
172dfd4f DDAG |
978 | qemu_put_byte(f, QEMU_VM_CONFIGURATION); |
979 | vmstate_save_state(f, &vmstate_configuration, &savevm_state, 0); | |
980 | } | |
f796baa1 DDAG |
981 | } |
982 | ||
9907e842 | 983 | void qemu_savevm_state_setup(QEMUFile *f) |
a672b469 AL |
984 | { |
985 | SaveStateEntry *se; | |
39346385 | 986 | int ret; |
a672b469 | 987 | |
9907e842 | 988 | trace_savevm_state_setup(); |
0163a2e0 | 989 | QTAILQ_FOREACH(se, &savevm_state.handlers, entry) { |
9907e842 | 990 | if (!se->ops || !se->ops->save_setup) { |
a672b469 | 991 | continue; |
22ea40f4 | 992 | } |
6bd68781 JQ |
993 | if (se->ops && se->ops->is_active) { |
994 | if (!se->ops->is_active(se->opaque)) { | |
995 | continue; | |
996 | } | |
997 | } | |
ce39bfc9 | 998 | save_section_header(f, se, QEMU_VM_SECTION_START); |
a672b469 | 999 | |
9907e842 | 1000 | ret = se->ops->save_setup(f, se->opaque); |
f68945d4 | 1001 | save_section_footer(f, se); |
2975725f | 1002 | if (ret < 0) { |
47c8c17a PB |
1003 | qemu_file_set_error(f, ret); |
1004 | break; | |
2975725f | 1005 | } |
a672b469 | 1006 | } |
a672b469 AL |
1007 | } |
1008 | ||
39346385 | 1009 | /* |
07f35073 | 1010 | * this function has three return values: |
39346385 JQ |
1011 | * negative: there was one error, and we have -errno. |
1012 | * 0 : We haven't finished, caller have to go again | |
1013 | * 1 : We have finished, we can go to complete phase | |
1014 | */ | |
35ecd943 | 1015 | int qemu_savevm_state_iterate(QEMUFile *f, bool postcopy) |
a672b469 AL |
1016 | { |
1017 | SaveStateEntry *se; | |
1018 | int ret = 1; | |
1019 | ||
9013dca5 | 1020 | trace_savevm_state_iterate(); |
0163a2e0 | 1021 | QTAILQ_FOREACH(se, &savevm_state.handlers, entry) { |
16310a3c | 1022 | if (!se->ops || !se->ops->save_live_iterate) { |
a672b469 | 1023 | continue; |
22ea40f4 | 1024 | } |
6bd68781 JQ |
1025 | if (se->ops && se->ops->is_active) { |
1026 | if (!se->ops->is_active(se->opaque)) { | |
1027 | continue; | |
1028 | } | |
1029 | } | |
35ecd943 DDAG |
1030 | /* |
1031 | * In the postcopy phase, any device that doesn't know how to | |
1032 | * do postcopy should have saved it's state in the _complete | |
1033 | * call that's already run, it might get confused if we call | |
1034 | * iterate afterwards. | |
1035 | */ | |
c6467627 VSO |
1036 | if (postcopy && |
1037 | !(se->ops->has_postcopy && se->ops->has_postcopy(se->opaque))) { | |
35ecd943 DDAG |
1038 | continue; |
1039 | } | |
aac844ed JQ |
1040 | if (qemu_file_rate_limit(f)) { |
1041 | return 0; | |
1042 | } | |
464400f6 | 1043 | trace_savevm_section_start(se->idstr, se->section_id); |
ce39bfc9 DDAG |
1044 | |
1045 | save_section_header(f, se, QEMU_VM_SECTION_PART); | |
a672b469 | 1046 | |
16310a3c | 1047 | ret = se->ops->save_live_iterate(f, se->opaque); |
a5df2a02 | 1048 | trace_savevm_section_end(se->idstr, se->section_id, ret); |
f68945d4 | 1049 | save_section_footer(f, se); |
517a13c9 | 1050 | |
47c8c17a PB |
1051 | if (ret < 0) { |
1052 | qemu_file_set_error(f, ret); | |
1053 | } | |
2975725f | 1054 | if (ret <= 0) { |
90697be8 JK |
1055 | /* Do not proceed to the next vmstate before this one reported |
1056 | completion of the current stage. This serializes the migration | |
1057 | and reduces the probability that a faster changing state is | |
1058 | synchronized over and over again. */ | |
1059 | break; | |
1060 | } | |
a672b469 | 1061 | } |
39346385 | 1062 | return ret; |
a672b469 AL |
1063 | } |
1064 | ||
9850c604 AG |
1065 | static bool should_send_vmdesc(void) |
1066 | { | |
1067 | MachineState *machine = MACHINE(qdev_get_machine()); | |
5727309d | 1068 | bool in_postcopy = migration_in_postcopy(); |
8421b205 | 1069 | return !machine->suppress_vmdesc && !in_postcopy; |
9850c604 AG |
1070 | } |
1071 | ||
763c906b DDAG |
1072 | /* |
1073 | * Calls the save_live_complete_postcopy methods | |
1074 | * causing the last few pages to be sent immediately and doing any associated | |
1075 | * cleanup. | |
1076 | * Note postcopy also calls qemu_savevm_state_complete_precopy to complete | |
1077 | * all the other devices, but that happens at the point we switch to postcopy. | |
1078 | */ | |
1079 | void qemu_savevm_state_complete_postcopy(QEMUFile *f) | |
1080 | { | |
1081 | SaveStateEntry *se; | |
1082 | int ret; | |
1083 | ||
1084 | QTAILQ_FOREACH(se, &savevm_state.handlers, entry) { | |
1085 | if (!se->ops || !se->ops->save_live_complete_postcopy) { | |
1086 | continue; | |
1087 | } | |
1088 | if (se->ops && se->ops->is_active) { | |
1089 | if (!se->ops->is_active(se->opaque)) { | |
1090 | continue; | |
1091 | } | |
1092 | } | |
1093 | trace_savevm_section_start(se->idstr, se->section_id); | |
1094 | /* Section type */ | |
1095 | qemu_put_byte(f, QEMU_VM_SECTION_END); | |
1096 | qemu_put_be32(f, se->section_id); | |
1097 | ||
1098 | ret = se->ops->save_live_complete_postcopy(f, se->opaque); | |
1099 | trace_savevm_section_end(se->idstr, se->section_id, ret); | |
1100 | save_section_footer(f, se); | |
1101 | if (ret < 0) { | |
1102 | qemu_file_set_error(f, ret); | |
1103 | return; | |
1104 | } | |
1105 | } | |
1106 | ||
1107 | qemu_put_byte(f, QEMU_VM_EOF); | |
1108 | qemu_fflush(f); | |
1109 | } | |
1110 | ||
a1fbe750 FZ |
1111 | int qemu_savevm_state_complete_precopy(QEMUFile *f, bool iterable_only, |
1112 | bool inactivate_disks) | |
a672b469 | 1113 | { |
8118f095 AG |
1114 | QJSON *vmdesc; |
1115 | int vmdesc_len; | |
a672b469 | 1116 | SaveStateEntry *se; |
2975725f | 1117 | int ret; |
5727309d | 1118 | bool in_postcopy = migration_in_postcopy(); |
a672b469 | 1119 | |
a3e06c3d | 1120 | trace_savevm_state_complete_precopy(); |
9013dca5 | 1121 | |
ea375f9a JK |
1122 | cpu_synchronize_all_states(); |
1123 | ||
0163a2e0 | 1124 | QTAILQ_FOREACH(se, &savevm_state.handlers, entry) { |
763c906b | 1125 | if (!se->ops || |
c6467627 VSO |
1126 | (in_postcopy && se->ops->has_postcopy && |
1127 | se->ops->has_postcopy(se->opaque)) || | |
1c0d249d | 1128 | (in_postcopy && !iterable_only) || |
763c906b | 1129 | !se->ops->save_live_complete_precopy) { |
a672b469 | 1130 | continue; |
22ea40f4 | 1131 | } |
1c0d249d | 1132 | |
6bd68781 JQ |
1133 | if (se->ops && se->ops->is_active) { |
1134 | if (!se->ops->is_active(se->opaque)) { | |
1135 | continue; | |
1136 | } | |
1137 | } | |
464400f6 | 1138 | trace_savevm_section_start(se->idstr, se->section_id); |
ce39bfc9 DDAG |
1139 | |
1140 | save_section_header(f, se, QEMU_VM_SECTION_END); | |
a672b469 | 1141 | |
a3e06c3d | 1142 | ret = se->ops->save_live_complete_precopy(f, se->opaque); |
a5df2a02 | 1143 | trace_savevm_section_end(se->idstr, se->section_id, ret); |
f68945d4 | 1144 | save_section_footer(f, se); |
2975725f | 1145 | if (ret < 0) { |
47c8c17a | 1146 | qemu_file_set_error(f, ret); |
a1fbe750 | 1147 | return -1; |
2975725f | 1148 | } |
a672b469 AL |
1149 | } |
1150 | ||
1c0d249d | 1151 | if (iterable_only) { |
a1fbe750 | 1152 | return 0; |
1c0d249d DDAG |
1153 | } |
1154 | ||
8118f095 | 1155 | vmdesc = qjson_new(); |
46d702b1 | 1156 | json_prop_int(vmdesc, "page_size", qemu_target_page_size()); |
8118f095 | 1157 | json_start_array(vmdesc, "devices"); |
0163a2e0 | 1158 | QTAILQ_FOREACH(se, &savevm_state.handlers, entry) { |
a672b469 | 1159 | |
22ea40f4 | 1160 | if ((!se->ops || !se->ops->save_state) && !se->vmsd) { |
5cecf414 | 1161 | continue; |
22ea40f4 | 1162 | } |
df896152 JQ |
1163 | if (se->vmsd && !vmstate_save_needed(se->vmsd, se->opaque)) { |
1164 | trace_savevm_section_skip(se->idstr, se->section_id); | |
1165 | continue; | |
1166 | } | |
1167 | ||
464400f6 | 1168 | trace_savevm_section_start(se->idstr, se->section_id); |
8118f095 AG |
1169 | |
1170 | json_start_object(vmdesc, NULL); | |
1171 | json_prop_str(vmdesc, "name", se->idstr); | |
1172 | json_prop_int(vmdesc, "instance_id", se->instance_id); | |
1173 | ||
ce39bfc9 | 1174 | save_section_header(f, se, QEMU_VM_SECTION_FULL); |
687433f6 DDAG |
1175 | ret = vmstate_save(f, se, vmdesc); |
1176 | if (ret) { | |
1177 | qemu_file_set_error(f, ret); | |
1178 | return ret; | |
1179 | } | |
a5df2a02 | 1180 | trace_savevm_section_end(se->idstr, se->section_id, 0); |
f68945d4 | 1181 | save_section_footer(f, se); |
bdf46d64 WY |
1182 | |
1183 | json_end_object(vmdesc); | |
a672b469 AL |
1184 | } |
1185 | ||
a1fbe750 FZ |
1186 | if (inactivate_disks) { |
1187 | /* Inactivate before sending QEMU_VM_EOF so that the | |
1188 | * bdrv_invalidate_cache_all() on the other end won't fail. */ | |
1189 | ret = bdrv_inactivate_all(); | |
1190 | if (ret) { | |
c232bf58 DDAG |
1191 | error_report("%s: bdrv_inactivate_all() failed (%d)", |
1192 | __func__, ret); | |
a1fbe750 FZ |
1193 | qemu_file_set_error(f, ret); |
1194 | return ret; | |
1195 | } | |
1196 | } | |
763c906b DDAG |
1197 | if (!in_postcopy) { |
1198 | /* Postcopy stream will still be going */ | |
1199 | qemu_put_byte(f, QEMU_VM_EOF); | |
1200 | } | |
8118f095 AG |
1201 | |
1202 | json_end_array(vmdesc); | |
1203 | qjson_finish(vmdesc); | |
1204 | vmdesc_len = strlen(qjson_get_str(vmdesc)); | |
1205 | ||
9850c604 AG |
1206 | if (should_send_vmdesc()) { |
1207 | qemu_put_byte(f, QEMU_VM_VMDESCRIPTION); | |
1208 | qemu_put_be32(f, vmdesc_len); | |
1209 | qemu_put_buffer(f, (uint8_t *)qjson_get_str(vmdesc), vmdesc_len); | |
1210 | } | |
b72fe9e6 | 1211 | qjson_destroy(vmdesc); |
8118f095 | 1212 | |
edaae611 | 1213 | qemu_fflush(f); |
a1fbe750 | 1214 | return 0; |
a672b469 AL |
1215 | } |
1216 | ||
c31b098f DDAG |
1217 | /* Give an estimate of the amount left to be transferred, |
1218 | * the result is split into the amount for units that can and | |
1219 | * for units that can't do postcopy. | |
1220 | */ | |
faec066a | 1221 | void qemu_savevm_state_pending(QEMUFile *f, uint64_t threshold_size, |
c31b098f DDAG |
1222 | uint64_t *res_non_postcopiable, |
1223 | uint64_t *res_postcopiable) | |
e4ed1541 JQ |
1224 | { |
1225 | SaveStateEntry *se; | |
c31b098f DDAG |
1226 | |
1227 | *res_non_postcopiable = 0; | |
1228 | *res_postcopiable = 0; | |
1229 | ||
e4ed1541 | 1230 | |
0163a2e0 | 1231 | QTAILQ_FOREACH(se, &savevm_state.handlers, entry) { |
e4ed1541 JQ |
1232 | if (!se->ops || !se->ops->save_live_pending) { |
1233 | continue; | |
1234 | } | |
1235 | if (se->ops && se->ops->is_active) { | |
1236 | if (!se->ops->is_active(se->opaque)) { | |
1237 | continue; | |
1238 | } | |
1239 | } | |
faec066a | 1240 | se->ops->save_live_pending(f, se->opaque, threshold_size, |
c31b098f | 1241 | res_non_postcopiable, res_postcopiable); |
e4ed1541 | 1242 | } |
e4ed1541 JQ |
1243 | } |
1244 | ||
ea7415fa | 1245 | void qemu_savevm_state_cleanup(void) |
4ec7fcc7 JK |
1246 | { |
1247 | SaveStateEntry *se; | |
1248 | ||
ea7415fa | 1249 | trace_savevm_state_cleanup(); |
0163a2e0 | 1250 | QTAILQ_FOREACH(se, &savevm_state.handlers, entry) { |
70f794fc JQ |
1251 | if (se->ops && se->ops->save_cleanup) { |
1252 | se->ops->save_cleanup(se->opaque); | |
4ec7fcc7 JK |
1253 | } |
1254 | } | |
1255 | } | |
1256 | ||
5d80448c | 1257 | static int qemu_savevm_state(QEMUFile *f, Error **errp) |
a672b469 | 1258 | { |
a672b469 | 1259 | int ret; |
a0762d9e | 1260 | MigrationState *ms = migrate_init(); |
6dcf6668 | 1261 | MigrationStatus status; |
89a02a9f | 1262 | ms->to_dst_file = f; |
a672b469 | 1263 | |
24f3902b | 1264 | if (migration_is_blocked(errp)) { |
6dcf6668 DL |
1265 | ret = -EINVAL; |
1266 | goto done; | |
dc912121 AW |
1267 | } |
1268 | ||
ce7c817c JQ |
1269 | if (migrate_use_block()) { |
1270 | error_setg(errp, "Block migration and snapshots are incompatible"); | |
1271 | ret = -EINVAL; | |
1272 | goto done; | |
1273 | } | |
1274 | ||
9b095037 | 1275 | qemu_mutex_unlock_iothread(); |
f796baa1 | 1276 | qemu_savevm_state_header(f); |
9907e842 | 1277 | qemu_savevm_state_setup(f); |
9b095037 PB |
1278 | qemu_mutex_lock_iothread(); |
1279 | ||
47c8c17a | 1280 | while (qemu_file_get_error(f) == 0) { |
35ecd943 | 1281 | if (qemu_savevm_state_iterate(f, false) > 0) { |
47c8c17a PB |
1282 | break; |
1283 | } | |
1284 | } | |
a672b469 | 1285 | |
47c8c17a | 1286 | ret = qemu_file_get_error(f); |
39346385 | 1287 | if (ret == 0) { |
a1fbe750 | 1288 | qemu_savevm_state_complete_precopy(f, false, false); |
624b9cc2 | 1289 | ret = qemu_file_get_error(f); |
39346385 | 1290 | } |
15b3b8ea | 1291 | qemu_savevm_state_cleanup(); |
04943eba | 1292 | if (ret != 0) { |
5d80448c | 1293 | error_setg_errno(errp, -ret, "Error while writing VM state"); |
04943eba | 1294 | } |
6dcf6668 DL |
1295 | |
1296 | done: | |
1297 | if (ret != 0) { | |
1298 | status = MIGRATION_STATUS_FAILED; | |
1299 | } else { | |
1300 | status = MIGRATION_STATUS_COMPLETED; | |
1301 | } | |
1302 | migrate_set_state(&ms->state, MIGRATION_STATUS_SETUP, status); | |
f9c8caa0 VSO |
1303 | |
1304 | /* f is outer parameter, it should not stay in global migration state after | |
1305 | * this function finished */ | |
1306 | ms->to_dst_file = NULL; | |
1307 | ||
a672b469 AL |
1308 | return ret; |
1309 | } | |
1310 | ||
a7ae8355 SS |
1311 | static int qemu_save_device_state(QEMUFile *f) |
1312 | { | |
1313 | SaveStateEntry *se; | |
1314 | ||
1315 | qemu_put_be32(f, QEMU_VM_FILE_MAGIC); | |
1316 | qemu_put_be32(f, QEMU_VM_FILE_VERSION); | |
1317 | ||
1318 | cpu_synchronize_all_states(); | |
1319 | ||
0163a2e0 | 1320 | QTAILQ_FOREACH(se, &savevm_state.handlers, entry) { |
687433f6 DDAG |
1321 | int ret; |
1322 | ||
a7ae8355 SS |
1323 | if (se->is_ram) { |
1324 | continue; | |
1325 | } | |
22ea40f4 | 1326 | if ((!se->ops || !se->ops->save_state) && !se->vmsd) { |
a7ae8355 SS |
1327 | continue; |
1328 | } | |
df896152 JQ |
1329 | if (se->vmsd && !vmstate_save_needed(se->vmsd, se->opaque)) { |
1330 | continue; | |
1331 | } | |
a7ae8355 | 1332 | |
ce39bfc9 | 1333 | save_section_header(f, se, QEMU_VM_SECTION_FULL); |
a7ae8355 | 1334 | |
687433f6 DDAG |
1335 | ret = vmstate_save(f, se, NULL); |
1336 | if (ret) { | |
1337 | return ret; | |
1338 | } | |
f68945d4 DDAG |
1339 | |
1340 | save_section_footer(f, se); | |
a7ae8355 SS |
1341 | } |
1342 | ||
1343 | qemu_put_byte(f, QEMU_VM_EOF); | |
1344 | ||
1345 | return qemu_file_get_error(f); | |
1346 | } | |
1347 | ||
a672b469 AL |
1348 | static SaveStateEntry *find_se(const char *idstr, int instance_id) |
1349 | { | |
1350 | SaveStateEntry *se; | |
1351 | ||
0163a2e0 | 1352 | QTAILQ_FOREACH(se, &savevm_state.handlers, entry) { |
a672b469 | 1353 | if (!strcmp(se->idstr, idstr) && |
4d2ffa08 JK |
1354 | (instance_id == se->instance_id || |
1355 | instance_id == se->alias_id)) | |
a672b469 | 1356 | return se; |
7685ee6a AW |
1357 | /* Migrating from an older version? */ |
1358 | if (strstr(se->idstr, idstr) && se->compat) { | |
1359 | if (!strcmp(se->compat->idstr, idstr) && | |
1360 | (instance_id == se->compat->instance_id || | |
1361 | instance_id == se->alias_id)) | |
1362 | return se; | |
1363 | } | |
a672b469 AL |
1364 | } |
1365 | return NULL; | |
1366 | } | |
1367 | ||
7b89bf27 DDAG |
1368 | enum LoadVMExitCodes { |
1369 | /* Allow a command to quit all layers of nested loadvm loops */ | |
1370 | LOADVM_QUIT = 1, | |
1371 | }; | |
1372 | ||
1373 | static int qemu_loadvm_state_main(QEMUFile *f, MigrationIncomingState *mis); | |
093e3c42 DDAG |
1374 | |
1375 | /* ------ incoming postcopy messages ------ */ | |
1376 | /* 'advise' arrives before any transfers just to tell us that a postcopy | |
1377 | * *might* happen - it might be skipped if precopy transferred everything | |
1378 | * quickly. | |
1379 | */ | |
875fcd01 GK |
1380 | static int loadvm_postcopy_handle_advise(MigrationIncomingState *mis, |
1381 | uint16_t len) | |
093e3c42 DDAG |
1382 | { |
1383 | PostcopyState ps = postcopy_state_set(POSTCOPY_INCOMING_ADVISE); | |
e8ca1db2 | 1384 | uint64_t remote_pagesize_summary, local_pagesize_summary, remote_tps; |
093e3c42 DDAG |
1385 | |
1386 | trace_loadvm_postcopy_handle_advise(); | |
1387 | if (ps != POSTCOPY_INCOMING_NONE) { | |
1388 | error_report("CMD_POSTCOPY_ADVISE in wrong postcopy state (%d)", ps); | |
1389 | return -1; | |
1390 | } | |
1391 | ||
875fcd01 GK |
1392 | switch (len) { |
1393 | case 0: | |
1394 | if (migrate_postcopy_ram()) { | |
1395 | error_report("RAM postcopy is enabled but have 0 byte advise"); | |
1396 | return -EINVAL; | |
1397 | } | |
58110f0a | 1398 | return 0; |
875fcd01 GK |
1399 | case 8 + 8: |
1400 | if (!migrate_postcopy_ram()) { | |
1401 | error_report("RAM postcopy is disabled but have 16 byte advise"); | |
1402 | return -EINVAL; | |
1403 | } | |
1404 | break; | |
1405 | default: | |
1406 | error_report("CMD_POSTCOPY_ADVISE invalid length (%d)", len); | |
1407 | return -EINVAL; | |
58110f0a VSO |
1408 | } |
1409 | ||
d7651f15 | 1410 | if (!postcopy_ram_supported_by_host(mis)) { |
328d4d85 | 1411 | postcopy_state_set(POSTCOPY_INCOMING_NONE); |
eb59db53 DDAG |
1412 | return -1; |
1413 | } | |
1414 | ||
e8ca1db2 DDAG |
1415 | remote_pagesize_summary = qemu_get_be64(mis->from_src_file); |
1416 | local_pagesize_summary = ram_pagesize_summary(); | |
1417 | ||
1418 | if (remote_pagesize_summary != local_pagesize_summary) { | |
093e3c42 | 1419 | /* |
e8ca1db2 DDAG |
1420 | * This detects two potential causes of mismatch: |
1421 | * a) A mismatch in host page sizes | |
1422 | * Some combinations of mismatch are probably possible but it gets | |
1423 | * a bit more complicated. In particular we need to place whole | |
1424 | * host pages on the dest at once, and we need to ensure that we | |
1425 | * handle dirtying to make sure we never end up sending part of | |
1426 | * a hostpage on it's own. | |
1427 | * b) The use of different huge page sizes on source/destination | |
1428 | * a more fine grain test is performed during RAM block migration | |
1429 | * but this test here causes a nice early clear failure, and | |
1430 | * also fails when passed to an older qemu that doesn't | |
1431 | * do huge pages. | |
093e3c42 | 1432 | */ |
e8ca1db2 DDAG |
1433 | error_report("Postcopy needs matching RAM page sizes (s=%" PRIx64 |
1434 | " d=%" PRIx64 ")", | |
1435 | remote_pagesize_summary, local_pagesize_summary); | |
093e3c42 DDAG |
1436 | return -1; |
1437 | } | |
1438 | ||
1439 | remote_tps = qemu_get_be64(mis->from_src_file); | |
20afaed9 | 1440 | if (remote_tps != qemu_target_page_size()) { |
093e3c42 DDAG |
1441 | /* |
1442 | * Again, some differences could be dealt with, but for now keep it | |
1443 | * simple. | |
1444 | */ | |
20afaed9 JQ |
1445 | error_report("Postcopy needs matching target page sizes (s=%d d=%zd)", |
1446 | (int)remote_tps, qemu_target_page_size()); | |
093e3c42 DDAG |
1447 | return -1; |
1448 | } | |
1449 | ||
1caddf8a DDAG |
1450 | if (ram_postcopy_incoming_init(mis)) { |
1451 | return -1; | |
1452 | } | |
1453 | ||
1454 | postcopy_state_set(POSTCOPY_INCOMING_ADVISE); | |
1455 | ||
093e3c42 DDAG |
1456 | return 0; |
1457 | } | |
1458 | ||
1459 | /* After postcopy we will be told to throw some pages away since they're | |
1460 | * dirty and will have to be demand fetched. Must happen before CPU is | |
1461 | * started. | |
1462 | * There can be 0..many of these messages, each encoding multiple pages. | |
1463 | */ | |
1464 | static int loadvm_postcopy_ram_handle_discard(MigrationIncomingState *mis, | |
1465 | uint16_t len) | |
1466 | { | |
1467 | int tmp; | |
1468 | char ramid[256]; | |
1469 | PostcopyState ps = postcopy_state_get(); | |
1470 | ||
1471 | trace_loadvm_postcopy_ram_handle_discard(); | |
1472 | ||
1473 | switch (ps) { | |
1474 | case POSTCOPY_INCOMING_ADVISE: | |
1475 | /* 1st discard */ | |
f9527107 | 1476 | tmp = postcopy_ram_prepare_discard(mis); |
093e3c42 DDAG |
1477 | if (tmp) { |
1478 | return tmp; | |
1479 | } | |
1480 | break; | |
1481 | ||
1482 | case POSTCOPY_INCOMING_DISCARD: | |
1483 | /* Expected state */ | |
1484 | break; | |
1485 | ||
1486 | default: | |
1487 | error_report("CMD_POSTCOPY_RAM_DISCARD in wrong postcopy state (%d)", | |
1488 | ps); | |
1489 | return -1; | |
1490 | } | |
1491 | /* We're expecting a | |
1492 | * Version (0) | |
1493 | * a RAM ID string (length byte, name, 0 term) | |
1494 | * then at least 1 16 byte chunk | |
1495 | */ | |
1496 | if (len < (1 + 1 + 1 + 1 + 2 * 8)) { | |
1497 | error_report("CMD_POSTCOPY_RAM_DISCARD invalid length (%d)", len); | |
1498 | return -1; | |
1499 | } | |
1500 | ||
1501 | tmp = qemu_get_byte(mis->from_src_file); | |
1502 | if (tmp != postcopy_ram_discard_version) { | |
1503 | error_report("CMD_POSTCOPY_RAM_DISCARD invalid version (%d)", tmp); | |
1504 | return -1; | |
1505 | } | |
1506 | ||
1507 | if (!qemu_get_counted_string(mis->from_src_file, ramid)) { | |
1508 | error_report("CMD_POSTCOPY_RAM_DISCARD Failed to read RAMBlock ID"); | |
1509 | return -1; | |
1510 | } | |
1511 | tmp = qemu_get_byte(mis->from_src_file); | |
1512 | if (tmp != 0) { | |
1513 | error_report("CMD_POSTCOPY_RAM_DISCARD missing nil (%d)", tmp); | |
1514 | return -1; | |
1515 | } | |
1516 | ||
1517 | len -= 3 + strlen(ramid); | |
1518 | if (len % 16) { | |
1519 | error_report("CMD_POSTCOPY_RAM_DISCARD invalid length (%d)", len); | |
1520 | return -1; | |
1521 | } | |
1522 | trace_loadvm_postcopy_ram_handle_discard_header(ramid, len); | |
1523 | while (len) { | |
093e3c42 DDAG |
1524 | uint64_t start_addr, block_length; |
1525 | start_addr = qemu_get_be64(mis->from_src_file); | |
1526 | block_length = qemu_get_be64(mis->from_src_file); | |
1527 | ||
1528 | len -= 16; | |
aaa2064c | 1529 | int ret = ram_discard_range(ramid, start_addr, block_length); |
093e3c42 DDAG |
1530 | if (ret) { |
1531 | return ret; | |
1532 | } | |
093e3c42 DDAG |
1533 | } |
1534 | trace_loadvm_postcopy_ram_handle_discard_end(); | |
1535 | ||
1536 | return 0; | |
1537 | } | |
1538 | ||
c76201ab DDAG |
1539 | /* |
1540 | * Triggered by a postcopy_listen command; this thread takes over reading | |
1541 | * the input stream, leaving the main thread free to carry on loading the rest | |
1542 | * of the device state (from RAM). | |
1543 | * (TODO:This could do with being in a postcopy file - but there again it's | |
1544 | * just another input loop, not that postcopy specific) | |
1545 | */ | |
1546 | static void *postcopy_ram_listen_thread(void *opaque) | |
1547 | { | |
1548 | QEMUFile *f = opaque; | |
1549 | MigrationIncomingState *mis = migration_incoming_get_current(); | |
1550 | int load_res; | |
1551 | ||
6ba996bb DDAG |
1552 | migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE, |
1553 | MIGRATION_STATUS_POSTCOPY_ACTIVE); | |
c76201ab DDAG |
1554 | qemu_sem_post(&mis->listen_thread_sem); |
1555 | trace_postcopy_ram_listen_thread_start(); | |
1556 | ||
1557 | /* | |
1558 | * Because we're a thread and not a coroutine we can't yield | |
1559 | * in qemu_file, and thus we must be blocking now. | |
1560 | */ | |
1561 | qemu_file_set_blocking(f, true); | |
1562 | load_res = qemu_loadvm_state_main(f, mis); | |
1563 | /* And non-blocking again so we don't block in any cleanup */ | |
1564 | qemu_file_set_blocking(f, false); | |
1565 | ||
1566 | trace_postcopy_ram_listen_thread_exit(); | |
1567 | if (load_res < 0) { | |
1568 | error_report("%s: loadvm failed: %d", __func__, load_res); | |
1569 | qemu_file_set_error(f, load_res); | |
6ba996bb DDAG |
1570 | migrate_set_state(&mis->state, MIGRATION_STATUS_POSTCOPY_ACTIVE, |
1571 | MIGRATION_STATUS_FAILED); | |
c76201ab DDAG |
1572 | } else { |
1573 | /* | |
1574 | * This looks good, but it's possible that the device loading in the | |
1575 | * main thread hasn't finished yet, and so we might not be in 'RUN' | |
1576 | * state yet; wait for the end of the main thread. | |
1577 | */ | |
1578 | qemu_event_wait(&mis->main_thread_load_event); | |
1579 | } | |
1580 | postcopy_ram_incoming_cleanup(mis); | |
c76201ab DDAG |
1581 | |
1582 | if (load_res < 0) { | |
1583 | /* | |
1584 | * If something went wrong then we have a bad state so exit; | |
1585 | * depending how far we got it might be possible at this point | |
1586 | * to leave the guest running and fire MCEs for pages that never | |
1587 | * arrived as a desperate recovery step. | |
1588 | */ | |
1589 | exit(EXIT_FAILURE); | |
1590 | } | |
1591 | ||
6ba996bb DDAG |
1592 | migrate_set_state(&mis->state, MIGRATION_STATUS_POSTCOPY_ACTIVE, |
1593 | MIGRATION_STATUS_COMPLETED); | |
1594 | /* | |
1595 | * If everything has worked fine, then the main thread has waited | |
1596 | * for us to start, and we're the last use of the mis. | |
1597 | * (If something broke then qemu will have to exit anyway since it's | |
1598 | * got a bad migration state). | |
1599 | */ | |
1600 | migration_incoming_state_destroy(); | |
acb5ea86 | 1601 | qemu_loadvm_state_cleanup(); |
6ba996bb | 1602 | |
c76201ab DDAG |
1603 | return NULL; |
1604 | } | |
1605 | ||
093e3c42 DDAG |
1606 | /* After this message we must be able to immediately receive postcopy data */ |
1607 | static int loadvm_postcopy_handle_listen(MigrationIncomingState *mis) | |
1608 | { | |
1609 | PostcopyState ps = postcopy_state_set(POSTCOPY_INCOMING_LISTENING); | |
1610 | trace_loadvm_postcopy_handle_listen(); | |
1611 | if (ps != POSTCOPY_INCOMING_ADVISE && ps != POSTCOPY_INCOMING_DISCARD) { | |
1612 | error_report("CMD_POSTCOPY_LISTEN in wrong postcopy state (%d)", ps); | |
1613 | return -1; | |
1614 | } | |
f9527107 DDAG |
1615 | if (ps == POSTCOPY_INCOMING_ADVISE) { |
1616 | /* | |
1617 | * A rare case, we entered listen without having to do any discards, | |
1618 | * so do the setup that's normally done at the time of the 1st discard. | |
1619 | */ | |
58110f0a VSO |
1620 | if (migrate_postcopy_ram()) { |
1621 | postcopy_ram_prepare_discard(mis); | |
1622 | } | |
f9527107 | 1623 | } |
093e3c42 | 1624 | |
f0a227ad DDAG |
1625 | /* |
1626 | * Sensitise RAM - can now generate requests for blocks that don't exist | |
1627 | * However, at this point the CPU shouldn't be running, and the IO | |
1628 | * shouldn't be doing anything yet so don't actually expect requests | |
1629 | */ | |
58110f0a VSO |
1630 | if (migrate_postcopy_ram()) { |
1631 | if (postcopy_ram_enable_notify(mis)) { | |
1632 | return -1; | |
1633 | } | |
f0a227ad DDAG |
1634 | } |
1635 | ||
c76201ab DDAG |
1636 | if (mis->have_listen_thread) { |
1637 | error_report("CMD_POSTCOPY_RAM_LISTEN already has a listen thread"); | |
1638 | return -1; | |
1639 | } | |
1640 | ||
1641 | mis->have_listen_thread = true; | |
1642 | /* Start up the listening thread and wait for it to signal ready */ | |
1643 | qemu_sem_init(&mis->listen_thread_sem, 0); | |
1644 | qemu_thread_create(&mis->listen_thread, "postcopy/listen", | |
1645 | postcopy_ram_listen_thread, mis->from_src_file, | |
a587a3fe | 1646 | QEMU_THREAD_DETACHED); |
c76201ab DDAG |
1647 | qemu_sem_wait(&mis->listen_thread_sem); |
1648 | qemu_sem_destroy(&mis->listen_thread_sem); | |
1649 | ||
093e3c42 DDAG |
1650 | return 0; |
1651 | } | |
1652 | ||
86469922 DL |
1653 | |
1654 | typedef struct { | |
1655 | QEMUBH *bh; | |
1656 | } HandleRunBhData; | |
1657 | ||
ea6a55bc | 1658 | static void loadvm_postcopy_handle_run_bh(void *opaque) |
093e3c42 | 1659 | { |
27c6825b | 1660 | Error *local_err = NULL; |
86469922 | 1661 | HandleRunBhData *data = opaque; |
093e3c42 | 1662 | |
27c6825b DDAG |
1663 | /* TODO we should move all of this lot into postcopy_ram.c or a shared code |
1664 | * in migration.c | |
1665 | */ | |
1666 | cpu_synchronize_all_post_init(); | |
1667 | ||
1668 | qemu_announce_self(); | |
1669 | ||
ace21a58 KW |
1670 | /* Make sure all file formats flush their mutable metadata. |
1671 | * If we get an error here, just don't restart the VM yet. */ | |
27c6825b | 1672 | bdrv_invalidate_cache_all(&local_err); |
0042fd36 | 1673 | if (local_err) { |
ace21a58 | 1674 | error_report_err(local_err); |
0042fd36 KW |
1675 | local_err = NULL; |
1676 | autostart = false; | |
1677 | } | |
1678 | ||
27c6825b DDAG |
1679 | trace_loadvm_postcopy_handle_run_cpu_sync(); |
1680 | cpu_synchronize_all_post_init(); | |
1681 | ||
1682 | trace_loadvm_postcopy_handle_run_vmstart(); | |
1683 | ||
093e3c42 DDAG |
1684 | if (autostart) { |
1685 | /* Hold onto your hats, starting the CPU */ | |
1686 | vm_start(); | |
1687 | } else { | |
1688 | /* leave it paused and let management decide when to start the CPU */ | |
1689 | runstate_set(RUN_STATE_PAUSED); | |
1690 | } | |
1691 | ||
86469922 DL |
1692 | qemu_bh_delete(data->bh); |
1693 | g_free(data); | |
ea6a55bc DL |
1694 | } |
1695 | ||
1696 | /* After all discards we can start running and asking for pages */ | |
1697 | static int loadvm_postcopy_handle_run(MigrationIncomingState *mis) | |
1698 | { | |
1699 | PostcopyState ps = postcopy_state_set(POSTCOPY_INCOMING_RUNNING); | |
86469922 | 1700 | HandleRunBhData *data; |
ea6a55bc DL |
1701 | |
1702 | trace_loadvm_postcopy_handle_run(); | |
1703 | if (ps != POSTCOPY_INCOMING_LISTENING) { | |
1704 | error_report("CMD_POSTCOPY_RUN in wrong postcopy state (%d)", ps); | |
1705 | return -1; | |
1706 | } | |
1707 | ||
86469922 DL |
1708 | data = g_new(HandleRunBhData, 1); |
1709 | data->bh = qemu_bh_new(loadvm_postcopy_handle_run_bh, data); | |
1710 | qemu_bh_schedule(data->bh); | |
ea6a55bc | 1711 | |
27c6825b DDAG |
1712 | /* We need to finish reading the stream from the package |
1713 | * and also stop reading anything more from the stream that loaded the | |
1714 | * package (since it's now being read by the listener thread). | |
1715 | * LOADVM_QUIT will quit all the layers of nested loadvm loops. | |
1716 | */ | |
1717 | return LOADVM_QUIT; | |
093e3c42 DDAG |
1718 | } |
1719 | ||
c76ca188 | 1720 | /** |
11cf1d98 DDAG |
1721 | * Immediately following this command is a blob of data containing an embedded |
1722 | * chunk of migration stream; read it and load it. | |
1723 | * | |
1724 | * @mis: Incoming state | |
1725 | * @length: Length of packaged data to read | |
c76ca188 | 1726 | * |
11cf1d98 DDAG |
1727 | * Returns: Negative values on error |
1728 | * | |
1729 | */ | |
1730 | static int loadvm_handle_cmd_packaged(MigrationIncomingState *mis) | |
1731 | { | |
1732 | int ret; | |
61b67d47 DB |
1733 | size_t length; |
1734 | QIOChannelBuffer *bioc; | |
11cf1d98 DDAG |
1735 | |
1736 | length = qemu_get_be32(mis->from_src_file); | |
1737 | trace_loadvm_handle_cmd_packaged(length); | |
1738 | ||
1739 | if (length > MAX_VM_CMD_PACKAGED_SIZE) { | |
61b67d47 | 1740 | error_report("Unreasonably large packaged state: %zu", length); |
11cf1d98 DDAG |
1741 | return -1; |
1742 | } | |
61b67d47 DB |
1743 | |
1744 | bioc = qio_channel_buffer_new(length); | |
6f01f136 | 1745 | qio_channel_set_name(QIO_CHANNEL(bioc), "migration-loadvm-buffer"); |
61b67d47 DB |
1746 | ret = qemu_get_buffer(mis->from_src_file, |
1747 | bioc->data, | |
1748 | length); | |
11cf1d98 | 1749 | if (ret != length) { |
61b67d47 DB |
1750 | object_unref(OBJECT(bioc)); |
1751 | error_report("CMD_PACKAGED: Buffer receive fail ret=%d length=%zu", | |
9af9e0fe | 1752 | ret, length); |
11cf1d98 DDAG |
1753 | return (ret < 0) ? ret : -EAGAIN; |
1754 | } | |
61b67d47 | 1755 | bioc->usage += length; |
11cf1d98 DDAG |
1756 | trace_loadvm_handle_cmd_packaged_received(ret); |
1757 | ||
61b67d47 | 1758 | QEMUFile *packf = qemu_fopen_channel_input(QIO_CHANNEL(bioc)); |
11cf1d98 DDAG |
1759 | |
1760 | ret = qemu_loadvm_state_main(packf, mis); | |
1761 | trace_loadvm_handle_cmd_packaged_main(ret); | |
1762 | qemu_fclose(packf); | |
61b67d47 | 1763 | object_unref(OBJECT(bioc)); |
11cf1d98 DDAG |
1764 | |
1765 | return ret; | |
1766 | } | |
1767 | ||
1768 | /* | |
1769 | * Process an incoming 'QEMU_VM_COMMAND' | |
1770 | * 0 just a normal return | |
1771 | * LOADVM_QUIT All good, but exit the loop | |
1772 | * <0 Error | |
c76ca188 DDAG |
1773 | */ |
1774 | static int loadvm_process_command(QEMUFile *f) | |
1775 | { | |
2e37701e | 1776 | MigrationIncomingState *mis = migration_incoming_get_current(); |
c76ca188 DDAG |
1777 | uint16_t cmd; |
1778 | uint16_t len; | |
2e37701e | 1779 | uint32_t tmp32; |
c76ca188 DDAG |
1780 | |
1781 | cmd = qemu_get_be16(f); | |
1782 | len = qemu_get_be16(f); | |
1783 | ||
1784 | trace_loadvm_process_command(cmd, len); | |
1785 | if (cmd >= MIG_CMD_MAX || cmd == MIG_CMD_INVALID) { | |
1786 | error_report("MIG_CMD 0x%x unknown (len 0x%x)", cmd, len); | |
1787 | return -EINVAL; | |
1788 | } | |
1789 | ||
1790 | if (mig_cmd_args[cmd].len != -1 && mig_cmd_args[cmd].len != len) { | |
1791 | error_report("%s received with bad length - expecting %zu, got %d", | |
1792 | mig_cmd_args[cmd].name, | |
1793 | (size_t)mig_cmd_args[cmd].len, len); | |
1794 | return -ERANGE; | |
1795 | } | |
1796 | ||
1797 | switch (cmd) { | |
2e37701e DDAG |
1798 | case MIG_CMD_OPEN_RETURN_PATH: |
1799 | if (mis->to_src_file) { | |
1800 | error_report("CMD_OPEN_RETURN_PATH called when RP already open"); | |
1801 | /* Not really a problem, so don't give up */ | |
1802 | return 0; | |
1803 | } | |
1804 | mis->to_src_file = qemu_file_get_return_path(f); | |
1805 | if (!mis->to_src_file) { | |
1806 | error_report("CMD_OPEN_RETURN_PATH failed"); | |
1807 | return -1; | |
1808 | } | |
1809 | break; | |
1810 | ||
1811 | case MIG_CMD_PING: | |
1812 | tmp32 = qemu_get_be32(f); | |
1813 | trace_loadvm_process_command_ping(tmp32); | |
1814 | if (!mis->to_src_file) { | |
1815 | error_report("CMD_PING (0x%x) received with no return path", | |
1816 | tmp32); | |
1817 | return -1; | |
1818 | } | |
6decec93 | 1819 | migrate_send_rp_pong(mis, tmp32); |
2e37701e | 1820 | break; |
093e3c42 | 1821 | |
11cf1d98 DDAG |
1822 | case MIG_CMD_PACKAGED: |
1823 | return loadvm_handle_cmd_packaged(mis); | |
1824 | ||
093e3c42 | 1825 | case MIG_CMD_POSTCOPY_ADVISE: |
875fcd01 | 1826 | return loadvm_postcopy_handle_advise(mis, len); |
093e3c42 DDAG |
1827 | |
1828 | case MIG_CMD_POSTCOPY_LISTEN: | |
1829 | return loadvm_postcopy_handle_listen(mis); | |
1830 | ||
1831 | case MIG_CMD_POSTCOPY_RUN: | |
1832 | return loadvm_postcopy_handle_run(mis); | |
1833 | ||
1834 | case MIG_CMD_POSTCOPY_RAM_DISCARD: | |
1835 | return loadvm_postcopy_ram_handle_discard(mis, len); | |
c76ca188 DDAG |
1836 | } |
1837 | ||
1838 | return 0; | |
1839 | } | |
1840 | ||
59f39a47 DDAG |
1841 | /* |
1842 | * Read a footer off the wire and check that it matches the expected section | |
1843 | * | |
1844 | * Returns: true if the footer was good | |
1845 | * false if there is a problem (and calls error_report to say why) | |
1846 | */ | |
0f42f657 | 1847 | static bool check_section_footer(QEMUFile *f, SaveStateEntry *se) |
59f39a47 DDAG |
1848 | { |
1849 | uint8_t read_mark; | |
1850 | uint32_t read_section_id; | |
1851 | ||
15c38503 | 1852 | if (!migrate_get_current()->send_section_footer) { |
59f39a47 DDAG |
1853 | /* No footer to check */ |
1854 | return true; | |
1855 | } | |
1856 | ||
1857 | read_mark = qemu_get_byte(f); | |
1858 | ||
1859 | if (read_mark != QEMU_VM_SECTION_FOOTER) { | |
0f42f657 | 1860 | error_report("Missing section footer for %s", se->idstr); |
59f39a47 DDAG |
1861 | return false; |
1862 | } | |
1863 | ||
1864 | read_section_id = qemu_get_be32(f); | |
0f42f657 | 1865 | if (read_section_id != se->load_section_id) { |
59f39a47 DDAG |
1866 | error_report("Mismatched section id in footer for %s -" |
1867 | " read 0x%x expected 0x%x", | |
0f42f657 | 1868 | se->idstr, read_section_id, se->load_section_id); |
59f39a47 DDAG |
1869 | return false; |
1870 | } | |
1871 | ||
1872 | /* All good */ | |
1873 | return true; | |
1874 | } | |
1875 | ||
fb3520a8 HZ |
1876 | static int |
1877 | qemu_loadvm_section_start_full(QEMUFile *f, MigrationIncomingState *mis) | |
1878 | { | |
1879 | uint32_t instance_id, version_id, section_id; | |
1880 | SaveStateEntry *se; | |
fb3520a8 HZ |
1881 | char idstr[256]; |
1882 | int ret; | |
1883 | ||
1884 | /* Read section start */ | |
1885 | section_id = qemu_get_be32(f); | |
1886 | if (!qemu_get_counted_string(f, idstr)) { | |
1887 | error_report("Unable to read ID string for section %u", | |
1888 | section_id); | |
1889 | return -EINVAL; | |
1890 | } | |
1891 | instance_id = qemu_get_be32(f); | |
1892 | version_id = qemu_get_be32(f); | |
1893 | ||
1894 | trace_qemu_loadvm_state_section_startfull(section_id, idstr, | |
1895 | instance_id, version_id); | |
1896 | /* Find savevm section */ | |
1897 | se = find_se(idstr, instance_id); | |
1898 | if (se == NULL) { | |
1899 | error_report("Unknown savevm section or instance '%s' %d", | |
1900 | idstr, instance_id); | |
1901 | return -EINVAL; | |
1902 | } | |
1903 | ||
1904 | /* Validate version */ | |
1905 | if (version_id > se->version_id) { | |
1906 | error_report("savevm: unsupported version %d for '%s' v%d", | |
1907 | version_id, idstr, se->version_id); | |
1908 | return -EINVAL; | |
1909 | } | |
0f42f657 JQ |
1910 | se->load_version_id = version_id; |
1911 | se->load_section_id = section_id; | |
fb3520a8 | 1912 | |
88c16567 WC |
1913 | /* Validate if it is a device's state */ |
1914 | if (xen_enabled() && se->is_ram) { | |
1915 | error_report("loadvm: %s RAM loading not allowed on Xen", idstr); | |
1916 | return -EINVAL; | |
1917 | } | |
1918 | ||
3a011c26 | 1919 | ret = vmstate_load(f, se); |
fb3520a8 HZ |
1920 | if (ret < 0) { |
1921 | error_report("error while loading state for instance 0x%x of" | |
1922 | " device '%s'", instance_id, idstr); | |
1923 | return ret; | |
1924 | } | |
0f42f657 | 1925 | if (!check_section_footer(f, se)) { |
fb3520a8 HZ |
1926 | return -EINVAL; |
1927 | } | |
1928 | ||
1929 | return 0; | |
1930 | } | |
1931 | ||
1932 | static int | |
1933 | qemu_loadvm_section_part_end(QEMUFile *f, MigrationIncomingState *mis) | |
1934 | { | |
1935 | uint32_t section_id; | |
0f42f657 | 1936 | SaveStateEntry *se; |
fb3520a8 HZ |
1937 | int ret; |
1938 | ||
1939 | section_id = qemu_get_be32(f); | |
1940 | ||
1941 | trace_qemu_loadvm_state_section_partend(section_id); | |
0f42f657 JQ |
1942 | QTAILQ_FOREACH(se, &savevm_state.handlers, entry) { |
1943 | if (se->load_section_id == section_id) { | |
fb3520a8 HZ |
1944 | break; |
1945 | } | |
1946 | } | |
0f42f657 | 1947 | if (se == NULL) { |
fb3520a8 HZ |
1948 | error_report("Unknown savevm section %d", section_id); |
1949 | return -EINVAL; | |
1950 | } | |
1951 | ||
3a011c26 | 1952 | ret = vmstate_load(f, se); |
fb3520a8 HZ |
1953 | if (ret < 0) { |
1954 | error_report("error while loading state section id %d(%s)", | |
0f42f657 | 1955 | section_id, se->idstr); |
fb3520a8 HZ |
1956 | return ret; |
1957 | } | |
0f42f657 | 1958 | if (!check_section_footer(f, se)) { |
fb3520a8 HZ |
1959 | return -EINVAL; |
1960 | } | |
1961 | ||
1962 | return 0; | |
1963 | } | |
1964 | ||
acb5ea86 JQ |
1965 | static int qemu_loadvm_state_setup(QEMUFile *f) |
1966 | { | |
1967 | SaveStateEntry *se; | |
1968 | int ret; | |
1969 | ||
1970 | trace_loadvm_state_setup(); | |
1971 | QTAILQ_FOREACH(se, &savevm_state.handlers, entry) { | |
1972 | if (!se->ops || !se->ops->load_setup) { | |
1973 | continue; | |
1974 | } | |
1975 | if (se->ops && se->ops->is_active) { | |
1976 | if (!se->ops->is_active(se->opaque)) { | |
1977 | continue; | |
1978 | } | |
1979 | } | |
1980 | ||
1981 | ret = se->ops->load_setup(f, se->opaque); | |
1982 | if (ret < 0) { | |
1983 | qemu_file_set_error(f, ret); | |
1984 | error_report("Load state of device %s failed", se->idstr); | |
1985 | return ret; | |
1986 | } | |
1987 | } | |
1988 | return 0; | |
1989 | } | |
1990 | ||
1991 | void qemu_loadvm_state_cleanup(void) | |
1992 | { | |
1993 | SaveStateEntry *se; | |
1994 | ||
1995 | trace_loadvm_state_cleanup(); | |
1996 | QTAILQ_FOREACH(se, &savevm_state.handlers, entry) { | |
1997 | if (se->ops && se->ops->load_cleanup) { | |
1998 | se->ops->load_cleanup(se->opaque); | |
1999 | } | |
2000 | } | |
2001 | } | |
2002 | ||
7b89bf27 | 2003 | static int qemu_loadvm_state_main(QEMUFile *f, MigrationIncomingState *mis) |
1a8f46f8 | 2004 | { |
a672b469 | 2005 | uint8_t section_type; |
ccb783c3 | 2006 | int ret = 0; |
61964c23 | 2007 | |
a672b469 | 2008 | while ((section_type = qemu_get_byte(f)) != QEMU_VM_EOF) { |
ccb783c3 | 2009 | ret = 0; |
a5df2a02 | 2010 | trace_qemu_loadvm_state_section(section_type); |
a672b469 AL |
2011 | switch (section_type) { |
2012 | case QEMU_VM_SECTION_START: | |
2013 | case QEMU_VM_SECTION_FULL: | |
fb3520a8 | 2014 | ret = qemu_loadvm_section_start_full(f, mis); |
b5a22e4a | 2015 | if (ret < 0) { |
ccb783c3 | 2016 | goto out; |
b5a22e4a | 2017 | } |
a672b469 AL |
2018 | break; |
2019 | case QEMU_VM_SECTION_PART: | |
2020 | case QEMU_VM_SECTION_END: | |
fb3520a8 | 2021 | ret = qemu_loadvm_section_part_end(f, mis); |
b5a22e4a | 2022 | if (ret < 0) { |
ccb783c3 | 2023 | goto out; |
b5a22e4a | 2024 | } |
a672b469 | 2025 | break; |
c76ca188 DDAG |
2026 | case QEMU_VM_COMMAND: |
2027 | ret = loadvm_process_command(f); | |
7b89bf27 DDAG |
2028 | trace_qemu_loadvm_state_section_command(ret); |
2029 | if ((ret < 0) || (ret & LOADVM_QUIT)) { | |
ccb783c3 | 2030 | goto out; |
c76ca188 DDAG |
2031 | } |
2032 | break; | |
a672b469 | 2033 | default: |
6a64b644 | 2034 | error_report("Unknown savevm section type %d", section_type); |
ccb783c3 DDAG |
2035 | ret = -EINVAL; |
2036 | goto out; | |
a672b469 AL |
2037 | } |
2038 | } | |
2039 | ||
ccb783c3 DDAG |
2040 | out: |
2041 | if (ret < 0) { | |
2042 | qemu_file_set_error(f, ret); | |
2043 | } | |
2044 | return ret; | |
7b89bf27 DDAG |
2045 | } |
2046 | ||
2047 | int qemu_loadvm_state(QEMUFile *f) | |
2048 | { | |
2049 | MigrationIncomingState *mis = migration_incoming_get_current(); | |
2050 | Error *local_err = NULL; | |
2051 | unsigned int v; | |
2052 | int ret; | |
2053 | ||
2054 | if (qemu_savevm_state_blocked(&local_err)) { | |
2055 | error_report_err(local_err); | |
2056 | return -EINVAL; | |
2057 | } | |
2058 | ||
2059 | v = qemu_get_be32(f); | |
2060 | if (v != QEMU_VM_FILE_MAGIC) { | |
2061 | error_report("Not a migration stream"); | |
2062 | return -EINVAL; | |
2063 | } | |
2064 | ||
2065 | v = qemu_get_be32(f); | |
2066 | if (v == QEMU_VM_FILE_VERSION_COMPAT) { | |
2067 | error_report("SaveVM v2 format is obsolete and don't work anymore"); | |
2068 | return -ENOTSUP; | |
2069 | } | |
2070 | if (v != QEMU_VM_FILE_VERSION) { | |
2071 | error_report("Unsupported migration stream version"); | |
2072 | return -ENOTSUP; | |
2073 | } | |
2074 | ||
acb5ea86 JQ |
2075 | if (qemu_loadvm_state_setup(f) != 0) { |
2076 | return -EINVAL; | |
2077 | } | |
2078 | ||
4ffdb337 | 2079 | if (migrate_get_current()->send_configuration) { |
7b89bf27 DDAG |
2080 | if (qemu_get_byte(f) != QEMU_VM_CONFIGURATION) { |
2081 | error_report("Configuration section missing"); | |
2082 | return -EINVAL; | |
2083 | } | |
2084 | ret = vmstate_load_state(f, &vmstate_configuration, &savevm_state, 0); | |
2085 | ||
2086 | if (ret) { | |
2087 | return ret; | |
2088 | } | |
2089 | } | |
2090 | ||
75e972da DG |
2091 | cpu_synchronize_all_pre_loadvm(); |
2092 | ||
7b89bf27 DDAG |
2093 | ret = qemu_loadvm_state_main(f, mis); |
2094 | qemu_event_set(&mis->main_thread_load_event); | |
2095 | ||
2096 | trace_qemu_loadvm_state_post_main(ret); | |
2097 | ||
c76201ab DDAG |
2098 | if (mis->have_listen_thread) { |
2099 | /* Listen thread still going, can't clean up yet */ | |
2100 | return ret; | |
2101 | } | |
2102 | ||
7b89bf27 DDAG |
2103 | if (ret == 0) { |
2104 | ret = qemu_file_get_error(f); | |
2105 | } | |
1925cebc AG |
2106 | |
2107 | /* | |
2108 | * Try to read in the VMDESC section as well, so that dumping tools that | |
2109 | * intercept our migration stream have the chance to see it. | |
2110 | */ | |
1aca9a5f DDAG |
2111 | |
2112 | /* We've got to be careful; if we don't read the data and just shut the fd | |
2113 | * then the sender can error if we close while it's still sending. | |
2114 | * We also mustn't read data that isn't there; some transports (RDMA) | |
2115 | * will stall waiting for that data when the source has already closed. | |
2116 | */ | |
7b89bf27 | 2117 | if (ret == 0 && should_send_vmdesc()) { |
1aca9a5f DDAG |
2118 | uint8_t *buf; |
2119 | uint32_t size; | |
7b89bf27 | 2120 | uint8_t section_type = qemu_get_byte(f); |
1aca9a5f DDAG |
2121 | |
2122 | if (section_type != QEMU_VM_VMDESCRIPTION) { | |
2123 | error_report("Expected vmdescription section, but got %d", | |
2124 | section_type); | |
2125 | /* | |
2126 | * It doesn't seem worth failing at this point since | |
2127 | * we apparently have an otherwise valid VM state | |
2128 | */ | |
2129 | } else { | |
2130 | buf = g_malloc(0x1000); | |
2131 | size = qemu_get_be32(f); | |
2132 | ||
2133 | while (size > 0) { | |
2134 | uint32_t read_chunk = MIN(size, 0x1000); | |
2135 | qemu_get_buffer(f, buf, read_chunk); | |
2136 | size -= read_chunk; | |
2137 | } | |
2138 | g_free(buf); | |
1925cebc | 2139 | } |
1925cebc AG |
2140 | } |
2141 | ||
acb5ea86 | 2142 | qemu_loadvm_state_cleanup(); |
ea375f9a JK |
2143 | cpu_synchronize_all_post_init(); |
2144 | ||
a672b469 AL |
2145 | return ret; |
2146 | } | |
2147 | ||
5e22479a | 2148 | int save_snapshot(const char *name, Error **errp) |
a672b469 AL |
2149 | { |
2150 | BlockDriverState *bs, *bs1; | |
2151 | QEMUSnapshotInfo sn1, *sn = &sn1, old_sn1, *old_sn = &old_sn1; | |
ac8c19ba | 2152 | int ret = -1; |
a672b469 AL |
2153 | QEMUFile *f; |
2154 | int saved_vm_running; | |
c2c9a466 | 2155 | uint64_t vm_state_size; |
68b891ec | 2156 | qemu_timeval tv; |
7d631a11 | 2157 | struct tm tm; |
79b3c12a | 2158 | AioContext *aio_context; |
a672b469 | 2159 | |
e9ff957a | 2160 | if (!bdrv_all_can_snapshot(&bs)) { |
927d6638 JQ |
2161 | error_setg(errp, "Device '%s' is writable but does not support " |
2162 | "snapshots", bdrv_get_device_name(bs)); | |
ac8c19ba | 2163 | return ret; |
feeee5ac MDCF |
2164 | } |
2165 | ||
0b461605 | 2166 | /* Delete old snapshots of the same name */ |
ac8c19ba | 2167 | if (name) { |
927d6638 | 2168 | ret = bdrv_all_delete_snapshot(name, &bs1, errp); |
ac8c19ba | 2169 | if (ret < 0) { |
927d6638 JQ |
2170 | error_prepend(errp, "Error while deleting snapshot on device " |
2171 | "'%s': ", bdrv_get_device_name(bs1)); | |
ac8c19ba PD |
2172 | return ret; |
2173 | } | |
0b461605 DL |
2174 | } |
2175 | ||
7cb14481 DL |
2176 | bs = bdrv_all_find_vmstate_bs(); |
2177 | if (bs == NULL) { | |
927d6638 | 2178 | error_setg(errp, "No block device can accept snapshots"); |
ac8c19ba | 2179 | return ret; |
a672b469 | 2180 | } |
79b3c12a | 2181 | aio_context = bdrv_get_aio_context(bs); |
a672b469 | 2182 | |
1354869c | 2183 | saved_vm_running = runstate_is_running(); |
560d027b JQ |
2184 | |
2185 | ret = global_state_store(); | |
2186 | if (ret) { | |
927d6638 | 2187 | error_setg(errp, "Error saving global state"); |
ac8c19ba | 2188 | return ret; |
560d027b | 2189 | } |
0461d5a6 | 2190 | vm_stop(RUN_STATE_SAVE_VM); |
a672b469 | 2191 | |
8649f2f9 SH |
2192 | bdrv_drain_all_begin(); |
2193 | ||
79b3c12a DL |
2194 | aio_context_acquire(aio_context); |
2195 | ||
cb499fb2 | 2196 | memset(sn, 0, sizeof(*sn)); |
a672b469 AL |
2197 | |
2198 | /* fill auxiliary fields */ | |
68b891ec | 2199 | qemu_gettimeofday(&tv); |
a672b469 AL |
2200 | sn->date_sec = tv.tv_sec; |
2201 | sn->date_nsec = tv.tv_usec * 1000; | |
bc72ad67 | 2202 | sn->vm_clock_nsec = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL); |
a672b469 | 2203 | |
7d631a11 MDCF |
2204 | if (name) { |
2205 | ret = bdrv_snapshot_find(bs, old_sn, name); | |
2206 | if (ret >= 0) { | |
2207 | pstrcpy(sn->name, sizeof(sn->name), old_sn->name); | |
2208 | pstrcpy(sn->id_str, sizeof(sn->id_str), old_sn->id_str); | |
2209 | } else { | |
2210 | pstrcpy(sn->name, sizeof(sn->name), name); | |
2211 | } | |
2212 | } else { | |
d7d9b528 BS |
2213 | /* cast below needed for OpenBSD where tv_sec is still 'long' */ |
2214 | localtime_r((const time_t *)&tv.tv_sec, &tm); | |
7d631a11 | 2215 | strftime(sn->name, sizeof(sn->name), "vm-%Y%m%d%H%M%S", &tm); |
7d631a11 MDCF |
2216 | } |
2217 | ||
a672b469 | 2218 | /* save the VM state */ |
45566e9c | 2219 | f = qemu_fopen_bdrv(bs, 1); |
a672b469 | 2220 | if (!f) { |
927d6638 | 2221 | error_setg(errp, "Could not open VM state file"); |
a672b469 AL |
2222 | goto the_end; |
2223 | } | |
927d6638 | 2224 | ret = qemu_savevm_state(f, errp); |
2d22b18f | 2225 | vm_state_size = qemu_ftell(f); |
a672b469 AL |
2226 | qemu_fclose(f); |
2227 | if (ret < 0) { | |
a672b469 AL |
2228 | goto the_end; |
2229 | } | |
2230 | ||
17e2a4a4 SH |
2231 | /* The bdrv_all_create_snapshot() call that follows acquires the AioContext |
2232 | * for itself. BDRV_POLL_WHILE() does not support nested locking because | |
2233 | * it only releases the lock once. Therefore synchronous I/O will deadlock | |
2234 | * unless we release the AioContext before bdrv_all_create_snapshot(). | |
2235 | */ | |
2236 | aio_context_release(aio_context); | |
2237 | aio_context = NULL; | |
2238 | ||
a9085f9b DL |
2239 | ret = bdrv_all_create_snapshot(sn, bs, vm_state_size, &bs); |
2240 | if (ret < 0) { | |
927d6638 JQ |
2241 | error_setg(errp, "Error while creating snapshot on '%s'", |
2242 | bdrv_get_device_name(bs)); | |
ac8c19ba | 2243 | goto the_end; |
a672b469 AL |
2244 | } |
2245 | ||
ac8c19ba PD |
2246 | ret = 0; |
2247 | ||
a672b469 | 2248 | the_end: |
17e2a4a4 SH |
2249 | if (aio_context) { |
2250 | aio_context_release(aio_context); | |
2251 | } | |
8649f2f9 SH |
2252 | |
2253 | bdrv_drain_all_end(); | |
2254 | ||
38ff78d3 | 2255 | if (saved_vm_running) { |
a672b469 | 2256 | vm_start(); |
38ff78d3 | 2257 | } |
ac8c19ba PD |
2258 | return ret; |
2259 | } | |
2260 | ||
5d6c599f AP |
2261 | void qmp_xen_save_devices_state(const char *filename, bool has_live, bool live, |
2262 | Error **errp) | |
a7ae8355 SS |
2263 | { |
2264 | QEMUFile *f; | |
8925839f | 2265 | QIOChannelFile *ioc; |
a7ae8355 SS |
2266 | int saved_vm_running; |
2267 | int ret; | |
2268 | ||
5d6c599f AP |
2269 | if (!has_live) { |
2270 | /* live default to true so old version of Xen tool stack can have a | |
2271 | * successfull live migration */ | |
2272 | live = true; | |
2273 | } | |
2274 | ||
a7ae8355 SS |
2275 | saved_vm_running = runstate_is_running(); |
2276 | vm_stop(RUN_STATE_SAVE_VM); | |
c69adea4 | 2277 | global_state_store_running(); |
a7ae8355 | 2278 | |
8925839f DB |
2279 | ioc = qio_channel_file_new_path(filename, O_WRONLY | O_CREAT, 0660, errp); |
2280 | if (!ioc) { | |
a7ae8355 SS |
2281 | goto the_end; |
2282 | } | |
6f01f136 | 2283 | qio_channel_set_name(QIO_CHANNEL(ioc), "migration-xen-save-state"); |
8925839f | 2284 | f = qemu_fopen_channel_output(QIO_CHANNEL(ioc)); |
032b79f7 | 2285 | object_unref(OBJECT(ioc)); |
a7ae8355 SS |
2286 | ret = qemu_save_device_state(f); |
2287 | qemu_fclose(f); | |
2288 | if (ret < 0) { | |
c6bd8c70 | 2289 | error_setg(errp, QERR_IO_ERROR); |
5d6c599f AP |
2290 | } else { |
2291 | /* libxl calls the QMP command "stop" before calling | |
2292 | * "xen-save-devices-state" and in case of migration failure, libxl | |
2293 | * would call "cont". | |
2294 | * So call bdrv_inactivate_all (release locks) here to let the other | |
2295 | * side of the migration take controle of the images. | |
2296 | */ | |
2297 | if (live && !saved_vm_running) { | |
2298 | ret = bdrv_inactivate_all(); | |
2299 | if (ret) { | |
2300 | error_setg(errp, "%s: bdrv_inactivate_all() failed (%d)", | |
2301 | __func__, ret); | |
2302 | } | |
2303 | } | |
a7ae8355 SS |
2304 | } |
2305 | ||
2306 | the_end: | |
38ff78d3 | 2307 | if (saved_vm_running) { |
a7ae8355 | 2308 | vm_start(); |
38ff78d3 | 2309 | } |
a7ae8355 SS |
2310 | } |
2311 | ||
88c16567 WC |
2312 | void qmp_xen_load_devices_state(const char *filename, Error **errp) |
2313 | { | |
2314 | QEMUFile *f; | |
2315 | QIOChannelFile *ioc; | |
2316 | int ret; | |
2317 | ||
2318 | /* Guest must be paused before loading the device state; the RAM state | |
2319 | * will already have been loaded by xc | |
2320 | */ | |
2321 | if (runstate_is_running()) { | |
2322 | error_setg(errp, "Cannot update device state while vm is running"); | |
2323 | return; | |
2324 | } | |
2325 | vm_stop(RUN_STATE_RESTORE_VM); | |
2326 | ||
2327 | ioc = qio_channel_file_new_path(filename, O_RDONLY | O_BINARY, 0, errp); | |
2328 | if (!ioc) { | |
2329 | return; | |
2330 | } | |
6f01f136 | 2331 | qio_channel_set_name(QIO_CHANNEL(ioc), "migration-xen-load-state"); |
88c16567 | 2332 | f = qemu_fopen_channel_input(QIO_CHANNEL(ioc)); |
032b79f7 | 2333 | object_unref(OBJECT(ioc)); |
88c16567 | 2334 | |
88c16567 WC |
2335 | ret = qemu_loadvm_state(f); |
2336 | qemu_fclose(f); | |
2337 | if (ret < 0) { | |
2338 | error_setg(errp, QERR_IO_ERROR); | |
2339 | } | |
2340 | migration_incoming_state_destroy(); | |
2341 | } | |
2342 | ||
5e22479a | 2343 | int load_snapshot(const char *name, Error **errp) |
a672b469 | 2344 | { |
f0aa7a8b | 2345 | BlockDriverState *bs, *bs_vm_state; |
2d22b18f | 2346 | QEMUSnapshotInfo sn; |
a672b469 | 2347 | QEMUFile *f; |
751c6a17 | 2348 | int ret; |
79b3c12a | 2349 | AioContext *aio_context; |
b4b076da | 2350 | MigrationIncomingState *mis = migration_incoming_get_current(); |
a672b469 | 2351 | |
849f96e2 | 2352 | if (!bdrv_all_can_snapshot(&bs)) { |
927d6638 JQ |
2353 | error_setg(errp, |
2354 | "Device '%s' is writable but does not support snapshots", | |
2355 | bdrv_get_device_name(bs)); | |
849f96e2 DL |
2356 | return -ENOTSUP; |
2357 | } | |
723ccda1 DL |
2358 | ret = bdrv_all_find_snapshot(name, &bs); |
2359 | if (ret < 0) { | |
927d6638 JQ |
2360 | error_setg(errp, |
2361 | "Device '%s' does not have the requested snapshot '%s'", | |
2362 | bdrv_get_device_name(bs), name); | |
723ccda1 DL |
2363 | return ret; |
2364 | } | |
849f96e2 | 2365 | |
7cb14481 | 2366 | bs_vm_state = bdrv_all_find_vmstate_bs(); |
f0aa7a8b | 2367 | if (!bs_vm_state) { |
927d6638 | 2368 | error_setg(errp, "No block device supports snapshots"); |
f0aa7a8b MDCF |
2369 | return -ENOTSUP; |
2370 | } | |
79b3c12a | 2371 | aio_context = bdrv_get_aio_context(bs_vm_state); |
f0aa7a8b MDCF |
2372 | |
2373 | /* Don't even try to load empty VM states */ | |
79b3c12a | 2374 | aio_context_acquire(aio_context); |
f0aa7a8b | 2375 | ret = bdrv_snapshot_find(bs_vm_state, &sn, name); |
79b3c12a | 2376 | aio_context_release(aio_context); |
f0aa7a8b MDCF |
2377 | if (ret < 0) { |
2378 | return ret; | |
2379 | } else if (sn.vm_state_size == 0) { | |
927d6638 JQ |
2380 | error_setg(errp, "This is a disk-only snapshot. Revert to it " |
2381 | " offline using qemu-img"); | |
f0aa7a8b MDCF |
2382 | return -EINVAL; |
2383 | } | |
2384 | ||
a672b469 | 2385 | /* Flush all IO requests so they don't interfere with the new state. */ |
8649f2f9 | 2386 | bdrv_drain_all_begin(); |
a672b469 | 2387 | |
2b624fe0 | 2388 | ret = bdrv_all_goto_snapshot(name, &bs, errp); |
4c1cdbaa | 2389 | if (ret < 0) { |
2b624fe0 KW |
2390 | error_prepend(errp, "Could not load snapshot '%s' on '%s': ", |
2391 | name, bdrv_get_device_name(bs)); | |
8649f2f9 | 2392 | goto err_drain; |
a672b469 AL |
2393 | } |
2394 | ||
a672b469 | 2395 | /* restore the VM state */ |
f0aa7a8b | 2396 | f = qemu_fopen_bdrv(bs_vm_state, 0); |
a672b469 | 2397 | if (!f) { |
927d6638 | 2398 | error_setg(errp, "Could not open VM state file"); |
8649f2f9 SH |
2399 | ret = -EINVAL; |
2400 | goto err_drain; | |
a672b469 | 2401 | } |
f0aa7a8b | 2402 | |
aedbe192 | 2403 | qemu_system_reset(SHUTDOWN_CAUSE_NONE); |
b4b076da | 2404 | mis->from_src_file = f; |
f0aa7a8b | 2405 | |
79b3c12a DL |
2406 | aio_context_acquire(aio_context); |
2407 | ret = qemu_loadvm_state(f); | |
1575829d | 2408 | migration_incoming_state_destroy(); |
79b3c12a DL |
2409 | aio_context_release(aio_context); |
2410 | ||
8649f2f9 SH |
2411 | bdrv_drain_all_end(); |
2412 | ||
a672b469 | 2413 | if (ret < 0) { |
927d6638 | 2414 | error_setg(errp, "Error %d while loading VM state", ret); |
05f2401e | 2415 | return ret; |
a672b469 | 2416 | } |
f0aa7a8b | 2417 | |
05f2401e | 2418 | return 0; |
8649f2f9 SH |
2419 | |
2420 | err_drain: | |
2421 | bdrv_drain_all_end(); | |
2422 | return ret; | |
7b630349 JQ |
2423 | } |
2424 | ||
c5705a77 AK |
2425 | void vmstate_register_ram(MemoryRegion *mr, DeviceState *dev) |
2426 | { | |
fa53a0e5 | 2427 | qemu_ram_set_idstr(mr->ram_block, |
c5705a77 AK |
2428 | memory_region_name(mr), dev); |
2429 | } | |
2430 | ||
2431 | void vmstate_unregister_ram(MemoryRegion *mr, DeviceState *dev) | |
2432 | { | |
fa53a0e5 | 2433 | qemu_ram_unset_idstr(mr->ram_block); |
c5705a77 AK |
2434 | } |
2435 | ||
2436 | void vmstate_register_ram_global(MemoryRegion *mr) | |
2437 | { | |
2438 | vmstate_register_ram(mr, NULL); | |
2439 | } | |
1bfe5f05 JQ |
2440 | |
2441 | bool vmstate_check_only_migratable(const VMStateDescription *vmsd) | |
2442 | { | |
2443 | /* check needed if --only-migratable is specified */ | |
3df663e5 | 2444 | if (!migrate_get_current()->only_migratable) { |
1bfe5f05 JQ |
2445 | return true; |
2446 | } | |
2447 | ||
2448 | return !(vmsd && vmsd->unmigratable); | |
2449 | } |