xref: /illumos-kvm-cmd/migration.c (revision 68396ea9)
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 int64_t max_throttle = (32 << 20);
36 
37 static MigrationState *current_migration;
38 
39 static NotifierList migration_state_notifiers =
40     NOTIFIER_LIST_INITIALIZER(migration_state_notifiers);
41 
qemu_start_incoming_migration(const char * uri)42 int qemu_start_incoming_migration(const char *uri)
43 {
44     const char *p;
45     int ret;
46 
47     if (strstart(uri, "tcp:", &p))
48         ret = tcp_start_incoming_migration(p);
49 #if !defined(WIN32)
50     else if (strstart(uri, "exec:", &p))
51         ret =  exec_start_incoming_migration(p);
52     else if (strstart(uri, "unix:", &p))
53         ret = unix_start_incoming_migration(p);
54     else if (strstart(uri, "fd:", &p))
55         ret = fd_start_incoming_migration(p);
56 #endif
57     else {
58         fprintf(stderr, "unknown migration protocol: %s\n", uri);
59         ret = -EPROTONOSUPPORT;
60     }
61     return ret;
62 }
63 
process_incoming_migration(QEMUFile * f)64 void process_incoming_migration(QEMUFile *f)
65 {
66     if (qemu_loadvm_state(f) < 0) {
67         fprintf(stderr, "load of migration failed\n");
68         exit(0);
69     }
70     qemu_announce_self();
71     DPRINTF("successfully loaded vm state\n");
72 
73     incoming_expected = false;
74 
75     if (autostart)
76         vm_start();
77 }
78 
do_migrate(Monitor * mon,const QDict * qdict,QObject ** ret_data)79 int do_migrate(Monitor *mon, const QDict *qdict, QObject **ret_data)
80 {
81     MigrationState *s = NULL;
82     const char *p;
83     int detach = qdict_get_try_bool(qdict, "detach", 0);
84     int blk = qdict_get_try_bool(qdict, "blk", 0);
85     int inc = qdict_get_try_bool(qdict, "inc", 0);
86     const char *uri = qdict_get_str(qdict, "uri");
87 
88     if (current_migration &&
89         current_migration->get_status(current_migration) == MIG_STATE_ACTIVE) {
90         monitor_printf(mon, "migration already in progress\n");
91         return -1;
92     }
93 
94     if (qemu_savevm_state_blocked(mon)) {
95         return -1;
96     }
97 
98     if (strstart(uri, "tcp:", &p)) {
99         s = tcp_start_outgoing_migration(mon, p, max_throttle, detach,
100                                          blk, inc);
101 #if !defined(WIN32)
102     } else if (strstart(uri, "exec:", &p)) {
103         s = exec_start_outgoing_migration(mon, p, max_throttle, detach,
104                                           blk, inc);
105     } else if (strstart(uri, "unix:", &p)) {
106         s = unix_start_outgoing_migration(mon, p, max_throttle, detach,
107                                           blk, inc);
108     } else if (strstart(uri, "fd:", &p)) {
109         s = fd_start_outgoing_migration(mon, p, max_throttle, detach,
110                                         blk, inc);
111 #endif
112     } else {
113         monitor_printf(mon, "unknown migration protocol: %s\n", uri);
114         return -1;
115     }
116 
117     if (s == NULL) {
118         monitor_printf(mon, "migration failed\n");
119         return -1;
120     }
121 
122     if (current_migration) {
123         current_migration->release(current_migration);
124     }
125 
126     current_migration = s;
127     notifier_list_notify(&migration_state_notifiers);
128     return 0;
129 }
130 
do_migrate_cancel(Monitor * mon,const QDict * qdict,QObject ** ret_data)131 int do_migrate_cancel(Monitor *mon, const QDict *qdict, QObject **ret_data)
132 {
133     MigrationState *s = current_migration;
134 
135     if (s)
136         s->cancel(s);
137 
138     return 0;
139 }
140 
do_migrate_set_speed(Monitor * mon,const QDict * qdict,QObject ** ret_data)141 int do_migrate_set_speed(Monitor *mon, const QDict *qdict, QObject **ret_data)
142 {
143     int64_t d;
144     FdMigrationState *s;
145 
146     d = qdict_get_int(qdict, "value");
147     if (d < 0) {
148         d = 0;
149     }
150     max_throttle = d;
151 
152     s = migrate_to_fms(current_migration);
153     if (s && s->file) {
154         qemu_file_set_rate_limit(s->file, max_throttle);
155     }
156 
157     return 0;
158 }
159 
160 /* amount of nanoseconds we are willing to wait for migration to be down.
161  * the choice of nanoseconds is because it is the maximum resolution that
162  * get_clock() can achieve. It is an internal measure. All user-visible
163  * units must be in seconds */
164 static uint64_t max_downtime = 30000000;
165 
migrate_max_downtime(void)166 uint64_t migrate_max_downtime(void)
167 {
168     return max_downtime;
169 }
170 
do_migrate_set_downtime(Monitor * mon,const QDict * qdict,QObject ** ret_data)171 int do_migrate_set_downtime(Monitor *mon, const QDict *qdict,
172                             QObject **ret_data)
173 {
174     double d;
175 
176     d = qdict_get_double(qdict, "value") * 1e9;
177     d = MAX(0, MIN(UINT64_MAX, d));
178     max_downtime = (uint64_t)d;
179 
180     return 0;
181 }
182 
migrate_print_status(Monitor * mon,const char * name,const QDict * status_dict)183 static void migrate_print_status(Monitor *mon, const char *name,
184                                  const QDict *status_dict)
185 {
186     QDict *qdict;
187 
188     qdict = qobject_to_qdict(qdict_get(status_dict, name));
189 
190     monitor_printf(mon, "transferred %s: %" PRIu64 " kbytes\n", name,
191                         qdict_get_int(qdict, "transferred") >> 10);
192     monitor_printf(mon, "remaining %s: %" PRIu64 " kbytes\n", name,
193                         qdict_get_int(qdict, "remaining") >> 10);
194     monitor_printf(mon, "total %s: %" PRIu64 " kbytes\n", name,
195                         qdict_get_int(qdict, "total") >> 10);
196 }
197 
do_info_migrate_print(Monitor * mon,const QObject * data)198 void do_info_migrate_print(Monitor *mon, const QObject *data)
199 {
200     QDict *qdict;
201 
202     qdict = qobject_to_qdict(data);
203 
204     monitor_printf(mon, "Migration status: %s\n",
205                    qdict_get_str(qdict, "status"));
206 
207     if (qdict_haskey(qdict, "ram")) {
208         migrate_print_status(mon, "ram", qdict);
209     }
210 
211     if (qdict_haskey(qdict, "disk")) {
212         migrate_print_status(mon, "disk", qdict);
213     }
214 }
215 
migrate_put_status(QDict * qdict,const char * name,uint64_t trans,uint64_t rem,uint64_t total)216 static void migrate_put_status(QDict *qdict, const char *name,
217                                uint64_t trans, uint64_t rem, uint64_t total)
218 {
219     QObject *obj;
220 
221     obj = qobject_from_jsonf("{ 'transferred': %" PRId64 ", "
222                                "'remaining': %" PRId64 ", "
223                                "'total': %" PRId64 " }", trans, rem, total);
224     qdict_put_obj(qdict, name, obj);
225 }
226 
do_info_migrate(Monitor * mon,QObject ** ret_data)227 void do_info_migrate(Monitor *mon, QObject **ret_data)
228 {
229     QDict *qdict;
230     MigrationState *s = current_migration;
231 
232     if (s) {
233         switch (s->get_status(s)) {
234         case MIG_STATE_ACTIVE:
235             qdict = qdict_new();
236             qdict_put(qdict, "status", qstring_from_str("active"));
237 
238             migrate_put_status(qdict, "ram", ram_bytes_transferred(),
239                                ram_bytes_remaining(), ram_bytes_total());
240 
241             if (blk_mig_active()) {
242                 migrate_put_status(qdict, "disk", blk_mig_bytes_transferred(),
243                                    blk_mig_bytes_remaining(),
244                                    blk_mig_bytes_total());
245             }
246 
247             *ret_data = QOBJECT(qdict);
248             break;
249         case MIG_STATE_COMPLETED:
250             *ret_data = qobject_from_jsonf("{ 'status': 'completed' }");
251             break;
252         case MIG_STATE_ERROR:
253             *ret_data = qobject_from_jsonf("{ 'status': 'failed' }");
254             break;
255         case MIG_STATE_CANCELLED:
256             *ret_data = qobject_from_jsonf("{ 'status': 'cancelled' }");
257             break;
258         }
259     }
260 }
261 
262 /* shared migration helpers */
263 
migrate_fd_monitor_suspend(FdMigrationState * s,Monitor * mon)264 void migrate_fd_monitor_suspend(FdMigrationState *s, Monitor *mon)
265 {
266     s->mon = mon;
267     if (monitor_suspend(mon) == 0) {
268         DPRINTF("suspending monitor\n");
269     } else {
270         monitor_printf(mon, "terminal does not allow synchronous "
271                        "migration, continuing detached\n");
272     }
273 }
274 
migrate_fd_error(FdMigrationState * s)275 void migrate_fd_error(FdMigrationState *s)
276 {
277     DPRINTF("setting error state\n");
278     s->state = MIG_STATE_ERROR;
279     notifier_list_notify(&migration_state_notifiers);
280     migrate_fd_cleanup(s);
281 }
282 
migrate_fd_cleanup(FdMigrationState * s)283 int migrate_fd_cleanup(FdMigrationState *s)
284 {
285     int ret = 0;
286 
287     qemu_set_fd_handler2(s->fd, NULL, NULL, NULL, NULL);
288 
289     if (s->file) {
290         DPRINTF("closing file\n");
291         if (qemu_fclose(s->file) != 0) {
292             ret = -1;
293         }
294         s->file = NULL;
295     }
296 
297     if (s->fd != -1)
298         close(s->fd);
299 
300     /* Don't resume monitor until we've flushed all of the buffers */
301     if (s->mon) {
302         monitor_resume(s->mon);
303     }
304 
305     s->fd = -1;
306 
307     return ret;
308 }
309 
migrate_fd_put_notify(void * opaque)310 void migrate_fd_put_notify(void *opaque)
311 {
312     FdMigrationState *s = opaque;
313 
314     qemu_set_fd_handler2(s->fd, NULL, NULL, NULL, NULL);
315     qemu_file_put_notify(s->file);
316 }
317 
migrate_fd_put_buffer(void * opaque,const void * data,size_t size)318 ssize_t migrate_fd_put_buffer(void *opaque, const void *data, size_t size)
319 {
320     FdMigrationState *s = opaque;
321     ssize_t ret;
322 
323     do {
324         ret = s->write(s, data, size);
325     } while (ret == -1 && ((s->get_error(s)) == EINTR));
326 
327     if (ret == -1)
328         ret = -(s->get_error(s));
329 
330     if (ret == -EAGAIN) {
331         qemu_set_fd_handler2(s->fd, NULL, NULL, migrate_fd_put_notify, s);
332     } else if (ret < 0) {
333         if (s->mon) {
334             monitor_resume(s->mon);
335         }
336         s->state = MIG_STATE_ERROR;
337         notifier_list_notify(&migration_state_notifiers);
338     }
339 
340     return ret;
341 }
342 
migrate_fd_connect(FdMigrationState * s)343 void migrate_fd_connect(FdMigrationState *s)
344 {
345     int ret;
346 
347     s->file = qemu_fopen_ops_buffered(s,
348                                       s->bandwidth_limit,
349                                       migrate_fd_put_buffer,
350                                       migrate_fd_put_ready,
351                                       migrate_fd_wait_for_unfreeze,
352                                       migrate_fd_close);
353 
354     DPRINTF("beginning savevm\n");
355     ret = qemu_savevm_state_begin(s->mon, s->file, s->mig_state.blk,
356                                   s->mig_state.shared);
357     if (ret < 0) {
358         DPRINTF("failed, %d\n", ret);
359         migrate_fd_error(s);
360         return;
361     }
362 
363     migrate_fd_put_ready(s);
364 }
365 
migrate_fd_put_ready(void * opaque)366 void migrate_fd_put_ready(void *opaque)
367 {
368     FdMigrationState *s = opaque;
369 
370     if (s->state != MIG_STATE_ACTIVE) {
371         DPRINTF("put_ready returning because of non-active state\n");
372         return;
373     }
374 
375     DPRINTF("iterate\n");
376     if (qemu_savevm_state_iterate(s->mon, s->file) == 1) {
377         int state;
378         int old_vm_running = vm_running;
379 
380         DPRINTF("done iterating\n");
381         vm_stop(0);
382 
383         if ((qemu_savevm_state_complete(s->mon, s->file)) < 0) {
384             if (old_vm_running) {
385                 vm_start();
386             }
387             state = MIG_STATE_ERROR;
388         } else {
389             state = MIG_STATE_COMPLETED;
390         }
391         if (migrate_fd_cleanup(s) < 0) {
392             if (old_vm_running) {
393                 vm_start();
394             }
395             state = MIG_STATE_ERROR;
396         }
397         s->state = state;
398         notifier_list_notify(&migration_state_notifiers);
399     }
400 }
401 
migrate_fd_get_status(MigrationState * mig_state)402 int migrate_fd_get_status(MigrationState *mig_state)
403 {
404     FdMigrationState *s = migrate_to_fms(mig_state);
405     return s->state;
406 }
407 
migrate_fd_cancel(MigrationState * mig_state)408 void migrate_fd_cancel(MigrationState *mig_state)
409 {
410     FdMigrationState *s = migrate_to_fms(mig_state);
411 
412     if (s->state != MIG_STATE_ACTIVE)
413         return;
414 
415     DPRINTF("cancelling migration\n");
416 
417     s->state = MIG_STATE_CANCELLED;
418     notifier_list_notify(&migration_state_notifiers);
419     qemu_savevm_state_cancel(s->mon, s->file);
420 
421     migrate_fd_cleanup(s);
422 }
423 
migrate_fd_release(MigrationState * mig_state)424 void migrate_fd_release(MigrationState *mig_state)
425 {
426     FdMigrationState *s = migrate_to_fms(mig_state);
427 
428     DPRINTF("releasing state\n");
429 
430     if (s->state == MIG_STATE_ACTIVE) {
431         s->state = MIG_STATE_CANCELLED;
432         notifier_list_notify(&migration_state_notifiers);
433         migrate_fd_cleanup(s);
434     }
435     qemu_free(s);
436 }
437 
migrate_fd_wait_for_unfreeze(void * opaque)438 void migrate_fd_wait_for_unfreeze(void *opaque)
439 {
440     FdMigrationState *s = opaque;
441     int ret;
442 
443     DPRINTF("wait for unfreeze\n");
444     if (s->state != MIG_STATE_ACTIVE)
445         return;
446 
447     do {
448         fd_set wfds;
449 
450         FD_ZERO(&wfds);
451         FD_SET(s->fd, &wfds);
452 
453         ret = select(s->fd + 1, NULL, &wfds, NULL, NULL);
454     } while (ret == -1 && (s->get_error(s)) == EINTR);
455 }
456 
migrate_fd_close(void * opaque)457 int migrate_fd_close(void *opaque)
458 {
459     FdMigrationState *s = opaque;
460 
461     qemu_set_fd_handler2(s->fd, NULL, NULL, NULL, NULL);
462     return s->close(s);
463 }
464 
add_migration_state_change_notifier(Notifier * notify)465 void add_migration_state_change_notifier(Notifier *notify)
466 {
467     notifier_list_add(&migration_state_notifiers, notify);
468 }
469 
remove_migration_state_change_notifier(Notifier * notify)470 void remove_migration_state_change_notifier(Notifier *notify)
471 {
472     notifier_list_remove(&migration_state_notifiers, notify);
473 }
474 
get_migration_state(void)475 int get_migration_state(void)
476 {
477     if (current_migration) {
478         return migrate_fd_get_status(current_migration);
479     } else {
480         return MIG_STATE_ERROR;
481     }
482 }
483