]>
Commit | Line | Data |
---|---|---|
1 | /* | |
2 | * QEMU live migration | |
3 | * | |
4 | * Copyright IBM, Corp. 2008 | |
5 | * | |
6 | * Authors: | |
7 | * Anthony Liguori <aliguori@us.ibm.com> | |
8 | * | |
9 | * This work is licensed under the terms of the GNU GPL, version 2. See | |
10 | * the COPYING file in the top-level directory. | |
11 | * | |
12 | */ | |
13 | ||
14 | #include "qemu-common.h" | |
15 | #include "migration.h" | |
16 | #include "monitor.h" | |
17 | #include "buffered_file.h" | |
18 | #include "sysemu.h" | |
19 | #include "block.h" | |
20 | #include "qemu_socket.h" | |
21 | #include "block-migration.h" | |
22 | #include "qemu-objects.h" | |
23 | ||
24 | //#define DEBUG_MIGRATION | |
25 | ||
26 | #ifdef DEBUG_MIGRATION | |
27 | #define dprintf(fmt, ...) \ | |
28 | do { printf("migration: " fmt, ## __VA_ARGS__); } while (0) | |
29 | #else | |
30 | #define dprintf(fmt, ...) \ | |
31 | do { } while (0) | |
32 | #endif | |
33 | ||
34 | /* Migration speed throttling */ | |
35 | static uint32_t max_throttle = (32 << 20); | |
36 | ||
37 | static MigrationState *current_migration; | |
38 | ||
39 | void qemu_start_incoming_migration(const char *uri) | |
40 | { | |
41 | const char *p; | |
42 | ||
43 | if (strstart(uri, "tcp:", &p)) | |
44 | tcp_start_incoming_migration(p); | |
45 | #if !defined(WIN32) | |
46 | else if (strstart(uri, "exec:", &p)) | |
47 | exec_start_incoming_migration(p); | |
48 | else if (strstart(uri, "unix:", &p)) | |
49 | unix_start_incoming_migration(p); | |
50 | else if (strstart(uri, "fd:", &p)) | |
51 | fd_start_incoming_migration(p); | |
52 | #endif | |
53 | else | |
54 | fprintf(stderr, "unknown migration protocol: %s\n", uri); | |
55 | } | |
56 | ||
57 | void do_migrate(Monitor *mon, const QDict *qdict, QObject **ret_data) | |
58 | { | |
59 | MigrationState *s = NULL; | |
60 | const char *p; | |
61 | int detach = qdict_get_int(qdict, "detach"); | |
62 | const char *uri = qdict_get_str(qdict, "uri"); | |
63 | ||
64 | if (current_migration && | |
65 | current_migration->get_status(current_migration) == MIG_STATE_ACTIVE) { | |
66 | monitor_printf(mon, "migration already in progress\n"); | |
67 | return; | |
68 | } | |
69 | ||
70 | if (strstart(uri, "tcp:", &p)) | |
71 | s = tcp_start_outgoing_migration(mon, p, max_throttle, detach, | |
72 | (int)qdict_get_int(qdict, "blk"), | |
73 | (int)qdict_get_int(qdict, "inc")); | |
74 | #if !defined(WIN32) | |
75 | else if (strstart(uri, "exec:", &p)) | |
76 | s = exec_start_outgoing_migration(mon, p, max_throttle, detach, | |
77 | (int)qdict_get_int(qdict, "blk"), | |
78 | (int)qdict_get_int(qdict, "inc")); | |
79 | else if (strstart(uri, "unix:", &p)) | |
80 | s = unix_start_outgoing_migration(mon, p, max_throttle, detach, | |
81 | (int)qdict_get_int(qdict, "blk"), | |
82 | (int)qdict_get_int(qdict, "inc")); | |
83 | else if (strstart(uri, "fd:", &p)) | |
84 | s = fd_start_outgoing_migration(mon, p, max_throttle, detach, | |
85 | (int)qdict_get_int(qdict, "blk"), | |
86 | (int)qdict_get_int(qdict, "inc")); | |
87 | #endif | |
88 | else | |
89 | monitor_printf(mon, "unknown migration protocol: %s\n", uri); | |
90 | ||
91 | if (s == NULL) | |
92 | monitor_printf(mon, "migration failed\n"); | |
93 | else { | |
94 | if (current_migration) | |
95 | current_migration->release(current_migration); | |
96 | ||
97 | current_migration = s; | |
98 | } | |
99 | } | |
100 | ||
101 | void do_migrate_cancel(Monitor *mon, const QDict *qdict, QObject **ret_data) | |
102 | { | |
103 | MigrationState *s = current_migration; | |
104 | ||
105 | if (s) | |
106 | s->cancel(s); | |
107 | } | |
108 | ||
109 | void do_migrate_set_speed(Monitor *mon, const QDict *qdict) | |
110 | { | |
111 | double d; | |
112 | char *ptr; | |
113 | FdMigrationState *s; | |
114 | const char *value = qdict_get_str(qdict, "value"); | |
115 | ||
116 | d = strtod(value, &ptr); | |
117 | switch (*ptr) { | |
118 | case 'G': case 'g': | |
119 | d *= 1024; | |
120 | case 'M': case 'm': | |
121 | d *= 1024; | |
122 | case 'K': case 'k': | |
123 | d *= 1024; | |
124 | default: | |
125 | break; | |
126 | } | |
127 | ||
128 | max_throttle = (uint32_t)d; | |
129 | ||
130 | s = migrate_to_fms(current_migration); | |
131 | if (s && s->file) { | |
132 | qemu_file_set_rate_limit(s->file, max_throttle); | |
133 | } | |
134 | } | |
135 | ||
136 | /* amount of nanoseconds we are willing to wait for migration to be down. | |
137 | * the choice of nanoseconds is because it is the maximum resolution that | |
138 | * get_clock() can achieve. It is an internal measure. All user-visible | |
139 | * units must be in seconds */ | |
140 | static uint64_t max_downtime = 30000000; | |
141 | ||
142 | uint64_t migrate_max_downtime(void) | |
143 | { | |
144 | return max_downtime; | |
145 | } | |
146 | ||
147 | void do_migrate_set_downtime(Monitor *mon, const QDict *qdict) | |
148 | { | |
149 | char *ptr; | |
150 | double d; | |
151 | const char *value = qdict_get_str(qdict, "value"); | |
152 | ||
153 | d = strtod(value, &ptr); | |
154 | if (!strcmp(ptr,"ms")) { | |
155 | d *= 1000000; | |
156 | } else if (!strcmp(ptr,"us")) { | |
157 | d *= 1000; | |
158 | } else if (!strcmp(ptr,"ns")) { | |
159 | } else { | |
160 | /* all else considered to be seconds */ | |
161 | d *= 1000000000; | |
162 | } | |
163 | ||
164 | max_downtime = (uint64_t)d; | |
165 | } | |
166 | ||
167 | static void migrate_print_status(Monitor *mon, const char *name, | |
168 | const QDict *status_dict) | |
169 | { | |
170 | QDict *qdict; | |
171 | ||
172 | qdict = qobject_to_qdict(qdict_get(status_dict, name)); | |
173 | ||
174 | monitor_printf(mon, "transferred %s: %" PRIu64 " kbytes\n", name, | |
175 | qdict_get_int(qdict, "transferred") >> 10); | |
176 | monitor_printf(mon, "remaining %s: %" PRIu64 " kbytes\n", name, | |
177 | qdict_get_int(qdict, "remaining") >> 10); | |
178 | monitor_printf(mon, "total %s: %" PRIu64 " kbytes\n", name, | |
179 | qdict_get_int(qdict, "total") >> 10); | |
180 | } | |
181 | ||
182 | void do_info_migrate_print(Monitor *mon, const QObject *data) | |
183 | { | |
184 | QDict *qdict; | |
185 | ||
186 | qdict = qobject_to_qdict(data); | |
187 | ||
188 | monitor_printf(mon, "Migration status: %s\n", | |
189 | qdict_get_str(qdict, "status")); | |
190 | ||
191 | if (qdict_haskey(qdict, "ram")) { | |
192 | migrate_print_status(mon, "ram", qdict); | |
193 | } | |
194 | ||
195 | if (qdict_haskey(qdict, "disk")) { | |
196 | migrate_print_status(mon, "disk", qdict); | |
197 | } | |
198 | } | |
199 | ||
200 | static void migrate_put_status(QDict *qdict, const char *name, | |
201 | uint64_t trans, uint64_t rem, uint64_t total) | |
202 | { | |
203 | QObject *obj; | |
204 | ||
205 | obj = qobject_from_jsonf("{ 'transferred': %" PRId64 ", " | |
206 | "'remaining': %" PRId64 ", " | |
207 | "'total': %" PRId64 " }", trans, rem, total); | |
208 | assert(obj != NULL); | |
209 | ||
210 | qdict_put_obj(qdict, name, obj); | |
211 | } | |
212 | ||
213 | /** | |
214 | * do_info_migrate(): Migration status | |
215 | * | |
216 | * Return a QDict. If migration is active there will be another | |
217 | * QDict with RAM migration status and if block migration is active | |
218 | * another one with block migration status. | |
219 | * | |
220 | * The main QDict contains the following: | |
221 | * | |
222 | * - "status": migration status | |
223 | * - "ram": only present if "status" is "active", it is a QDict with the | |
224 | * following RAM information (in bytes): | |
225 | * - "transferred": amount transferred | |
226 | * - "remaining": amount remaining | |
227 | * - "total": total | |
228 | * - "disk": only present if "status" is "active" and it is a block migration, | |
229 | * it is a QDict with the following disk information (in bytes): | |
230 | * - "transferred": amount transferred | |
231 | * - "remaining": amount remaining | |
232 | * - "total": total | |
233 | * | |
234 | * Examples: | |
235 | * | |
236 | * 1. Migration is "completed": | |
237 | * | |
238 | * { "status": "completed" } | |
239 | * | |
240 | * 2. Migration is "active" and it is not a block migration: | |
241 | * | |
242 | * { "status": "active", | |
243 | * "ram": { "transferred": 123, "remaining": 123, "total": 246 } } | |
244 | * | |
245 | * 3. Migration is "active" and it is a block migration: | |
246 | * | |
247 | * { "status": "active", | |
248 | * "ram": { "total": 1057024, "remaining": 1053304, "transferred": 3720 }, | |
249 | * "disk": { "total": 20971520, "remaining": 20880384, "transferred": 91136 }} | |
250 | */ | |
251 | void do_info_migrate(Monitor *mon, QObject **ret_data) | |
252 | { | |
253 | QDict *qdict; | |
254 | MigrationState *s = current_migration; | |
255 | ||
256 | if (s) { | |
257 | switch (s->get_status(s)) { | |
258 | case MIG_STATE_ACTIVE: | |
259 | qdict = qdict_new(); | |
260 | qdict_put(qdict, "status", qstring_from_str("active")); | |
261 | ||
262 | migrate_put_status(qdict, "ram", ram_bytes_transferred(), | |
263 | ram_bytes_remaining(), ram_bytes_total()); | |
264 | ||
265 | if (blk_mig_active()) { | |
266 | migrate_put_status(qdict, "disk", blk_mig_bytes_transferred(), | |
267 | blk_mig_bytes_remaining(), | |
268 | blk_mig_bytes_total()); | |
269 | } | |
270 | ||
271 | *ret_data = QOBJECT(qdict); | |
272 | break; | |
273 | case MIG_STATE_COMPLETED: | |
274 | *ret_data = qobject_from_jsonf("{ 'status': 'completed' }"); | |
275 | break; | |
276 | case MIG_STATE_ERROR: | |
277 | *ret_data = qobject_from_jsonf("{ 'status': 'failed' }"); | |
278 | break; | |
279 | case MIG_STATE_CANCELLED: | |
280 | *ret_data = qobject_from_jsonf("{ 'status': 'cancelled' }"); | |
281 | break; | |
282 | } | |
283 | assert(*ret_data != NULL); | |
284 | } | |
285 | } | |
286 | ||
287 | /* shared migration helpers */ | |
288 | ||
289 | void migrate_fd_monitor_suspend(FdMigrationState *s, Monitor *mon) | |
290 | { | |
291 | s->mon = mon; | |
292 | if (monitor_suspend(mon) == 0) { | |
293 | dprintf("suspending monitor\n"); | |
294 | } else { | |
295 | monitor_printf(mon, "terminal does not allow synchronous " | |
296 | "migration, continuing detached\n"); | |
297 | } | |
298 | } | |
299 | ||
300 | void migrate_fd_error(FdMigrationState *s) | |
301 | { | |
302 | dprintf("setting error state\n"); | |
303 | s->state = MIG_STATE_ERROR; | |
304 | migrate_fd_cleanup(s); | |
305 | } | |
306 | ||
307 | void migrate_fd_cleanup(FdMigrationState *s) | |
308 | { | |
309 | qemu_set_fd_handler2(s->fd, NULL, NULL, NULL, NULL); | |
310 | ||
311 | if (s->file) { | |
312 | dprintf("closing file\n"); | |
313 | qemu_fclose(s->file); | |
314 | s->file = NULL; | |
315 | } | |
316 | ||
317 | if (s->fd != -1) | |
318 | close(s->fd); | |
319 | ||
320 | /* Don't resume monitor until we've flushed all of the buffers */ | |
321 | if (s->mon) { | |
322 | monitor_resume(s->mon); | |
323 | } | |
324 | ||
325 | s->fd = -1; | |
326 | } | |
327 | ||
328 | void migrate_fd_put_notify(void *opaque) | |
329 | { | |
330 | FdMigrationState *s = opaque; | |
331 | ||
332 | qemu_set_fd_handler2(s->fd, NULL, NULL, NULL, NULL); | |
333 | qemu_file_put_notify(s->file); | |
334 | } | |
335 | ||
336 | ssize_t migrate_fd_put_buffer(void *opaque, const void *data, size_t size) | |
337 | { | |
338 | FdMigrationState *s = opaque; | |
339 | ssize_t ret; | |
340 | ||
341 | do { | |
342 | ret = s->write(s, data, size); | |
343 | } while (ret == -1 && ((s->get_error(s)) == EINTR)); | |
344 | ||
345 | if (ret == -1) | |
346 | ret = -(s->get_error(s)); | |
347 | ||
348 | if (ret == -EAGAIN) | |
349 | qemu_set_fd_handler2(s->fd, NULL, NULL, migrate_fd_put_notify, s); | |
350 | ||
351 | return ret; | |
352 | } | |
353 | ||
354 | void migrate_fd_connect(FdMigrationState *s) | |
355 | { | |
356 | int ret; | |
357 | ||
358 | s->file = qemu_fopen_ops_buffered(s, | |
359 | s->bandwidth_limit, | |
360 | migrate_fd_put_buffer, | |
361 | migrate_fd_put_ready, | |
362 | migrate_fd_wait_for_unfreeze, | |
363 | migrate_fd_close); | |
364 | ||
365 | dprintf("beginning savevm\n"); | |
366 | ret = qemu_savevm_state_begin(s->mon, s->file, s->mig_state.blk, | |
367 | s->mig_state.shared); | |
368 | if (ret < 0) { | |
369 | dprintf("failed, %d\n", ret); | |
370 | migrate_fd_error(s); | |
371 | return; | |
372 | } | |
373 | ||
374 | migrate_fd_put_ready(s); | |
375 | } | |
376 | ||
377 | void migrate_fd_put_ready(void *opaque) | |
378 | { | |
379 | FdMigrationState *s = opaque; | |
380 | ||
381 | if (s->state != MIG_STATE_ACTIVE) { | |
382 | dprintf("put_ready returning because of non-active state\n"); | |
383 | return; | |
384 | } | |
385 | ||
386 | dprintf("iterate\n"); | |
387 | if (qemu_savevm_state_iterate(s->mon, s->file) == 1) { | |
388 | int state; | |
389 | int old_vm_running = vm_running; | |
390 | ||
391 | dprintf("done iterating\n"); | |
392 | vm_stop(0); | |
393 | ||
394 | qemu_aio_flush(); | |
395 | bdrv_flush_all(); | |
396 | if ((qemu_savevm_state_complete(s->mon, s->file)) < 0) { | |
397 | if (old_vm_running) { | |
398 | vm_start(); | |
399 | } | |
400 | state = MIG_STATE_ERROR; | |
401 | } else { | |
402 | state = MIG_STATE_COMPLETED; | |
403 | } | |
404 | migrate_fd_cleanup(s); | |
405 | s->state = state; | |
406 | } | |
407 | } | |
408 | ||
409 | int migrate_fd_get_status(MigrationState *mig_state) | |
410 | { | |
411 | FdMigrationState *s = migrate_to_fms(mig_state); | |
412 | return s->state; | |
413 | } | |
414 | ||
415 | void migrate_fd_cancel(MigrationState *mig_state) | |
416 | { | |
417 | FdMigrationState *s = migrate_to_fms(mig_state); | |
418 | ||
419 | if (s->state != MIG_STATE_ACTIVE) | |
420 | return; | |
421 | ||
422 | dprintf("cancelling migration\n"); | |
423 | ||
424 | s->state = MIG_STATE_CANCELLED; | |
425 | qemu_savevm_state_cancel(s->mon, s->file); | |
426 | ||
427 | migrate_fd_cleanup(s); | |
428 | } | |
429 | ||
430 | void migrate_fd_release(MigrationState *mig_state) | |
431 | { | |
432 | FdMigrationState *s = migrate_to_fms(mig_state); | |
433 | ||
434 | dprintf("releasing state\n"); | |
435 | ||
436 | if (s->state == MIG_STATE_ACTIVE) { | |
437 | s->state = MIG_STATE_CANCELLED; | |
438 | migrate_fd_cleanup(s); | |
439 | } | |
440 | free(s); | |
441 | } | |
442 | ||
443 | void migrate_fd_wait_for_unfreeze(void *opaque) | |
444 | { | |
445 | FdMigrationState *s = opaque; | |
446 | int ret; | |
447 | ||
448 | dprintf("wait for unfreeze\n"); | |
449 | if (s->state != MIG_STATE_ACTIVE) | |
450 | return; | |
451 | ||
452 | do { | |
453 | fd_set wfds; | |
454 | ||
455 | FD_ZERO(&wfds); | |
456 | FD_SET(s->fd, &wfds); | |
457 | ||
458 | ret = select(s->fd + 1, NULL, &wfds, NULL, NULL); | |
459 | } while (ret == -1 && (s->get_error(s)) == EINTR); | |
460 | } | |
461 | ||
462 | int migrate_fd_close(void *opaque) | |
463 | { | |
464 | FdMigrationState *s = opaque; | |
465 | ||
466 | qemu_set_fd_handler2(s->fd, NULL, NULL, NULL, NULL); | |
467 | return s->close(s); | |
468 | } |