]>
Commit | Line | Data |
---|---|---|
5bb7910a AL |
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 | * | |
6b620ca3 PB |
12 | * Contributions after 2012-01-13 are licensed under the terms of the |
13 | * GNU GPL, version 2 or (at your option) any later version. | |
5bb7910a AL |
14 | */ |
15 | ||
1393a485 | 16 | #include "qemu/osdep.h" |
f348b6d1 | 17 | #include "qemu/cutils.h" |
d49b6836 | 18 | #include "qemu/error-report.h" |
795c40b8 | 19 | #include "migration/blocker.h" |
f4dbe1bf | 20 | #include "exec.h" |
7fcac4a2 | 21 | #include "fd.h" |
61e8b148 | 22 | #include "socket.h" |
e1a3ecee | 23 | #include "rdma.h" |
7b1e1a22 | 24 | #include "ram.h" |
84a899de | 25 | #include "migration/global_state.h" |
c4b63b7c | 26 | #include "migration/misc.h" |
6666c96a | 27 | #include "migration.h" |
20a519a0 | 28 | #include "savevm.h" |
40014d81 | 29 | #include "qemu-file-channel.h" |
08a0aee1 | 30 | #include "qemu-file.h" |
987772d9 | 31 | #include "migration/vmstate.h" |
737e150e | 32 | #include "block/block.h" |
e688df6b | 33 | #include "qapi/error.h" |
9af23989 MA |
34 | #include "qapi/qapi-commands-migration.h" |
35 | #include "qapi/qapi-events-migration.h" | |
cc7a8ea7 | 36 | #include "qapi/qmp/qerror.h" |
15280c36 | 37 | #include "qapi/qmp/qnull.h" |
ab28bd23 | 38 | #include "qemu/rcu.h" |
2c9e6fec | 39 | #include "block.h" |
be07b0ac | 40 | #include "postcopy-ram.h" |
766bd176 | 41 | #include "qemu/thread.h" |
c09e5bb1 | 42 | #include "trace.h" |
51180423 | 43 | #include "exec/target_page.h" |
61b67d47 | 44 | #include "io/channel-buffer.h" |
35a6ed4f | 45 | #include "migration/colo.h" |
4ffdb337 | 46 | #include "hw/boards.h" |
9d18af93 | 47 | #include "monitor/monitor.h" |
065e2813 | 48 | |
dc325627 | 49 | #define MAX_THROTTLE (32 << 20) /* Migration transfer speed throttling */ |
5bb7910a | 50 | |
5b4e1eb7 JQ |
51 | /* Amount of time to allocate to each "chunk" of bandwidth-throttled |
52 | * data. */ | |
53 | #define BUFFER_DELAY 100 | |
54 | #define XFER_LIMIT_RATIO (1000 / BUFFER_DELAY) | |
55 | ||
2ff30257 AA |
56 | /* Time in milliseconds we are allowed to stop the source, |
57 | * for sending the last part */ | |
58 | #define DEFAULT_MIGRATE_SET_DOWNTIME 300 | |
59 | ||
87c9cc1c DHB |
60 | /* Maximum migrate downtime set to 2000 seconds */ |
61 | #define MAX_MIGRATE_DOWNTIME_SECONDS 2000 | |
62 | #define MAX_MIGRATE_DOWNTIME (MAX_MIGRATE_DOWNTIME_SECONDS * 1000) | |
63 | ||
8706d2d5 LL |
64 | /* Default compression thread count */ |
65 | #define DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT 8 | |
3fcb38c2 LL |
66 | /* Default decompression thread count, usually decompression is at |
67 | * least 4 times as fast as compression.*/ | |
68 | #define DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT 2 | |
8706d2d5 LL |
69 | /*0: means nocompress, 1: best speed, ... 9: best compress ratio */ |
70 | #define DEFAULT_MIGRATE_COMPRESS_LEVEL 1 | |
1626fee3 | 71 | /* Define default autoconverge cpu throttle migration parameters */ |
d85a31d1 JH |
72 | #define DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL 20 |
73 | #define DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT 10 | |
8706d2d5 | 74 | |
17ad9b35 | 75 | /* Migration XBZRLE default cache size */ |
73af8dd8 | 76 | #define DEFAULT_MIGRATE_XBZRLE_CACHE_SIZE (64 * 1024 * 1024) |
17ad9b35 | 77 | |
68b53591 HZ |
78 | /* The delay time (in ms) between two COLO checkpoints |
79 | * Note: Please change this default value to 10000 when we support hybrid mode. | |
80 | */ | |
81 | #define DEFAULT_MIGRATE_X_CHECKPOINT_DELAY 200 | |
4075fb1c | 82 | #define DEFAULT_MIGRATE_MULTIFD_CHANNELS 2 |
0fb86605 | 83 | #define DEFAULT_MIGRATE_MULTIFD_PAGE_COUNT 16 |
68b53591 | 84 | |
99a0db9b GH |
85 | static NotifierList migration_state_notifiers = |
86 | NOTIFIER_LIST_INITIALIZER(migration_state_notifiers); | |
87 | ||
adde220a DDAG |
88 | static bool deferred_incoming; |
89 | ||
da6f1790 JQ |
90 | /* Messages sent on the return path from destination to source */ |
91 | enum mig_rp_message_type { | |
92 | MIG_RP_MSG_INVALID = 0, /* Must be 0 */ | |
93 | MIG_RP_MSG_SHUT, /* sibling will not send any more RP messages */ | |
94 | MIG_RP_MSG_PONG, /* Response to a PING; data (seq: be32 ) */ | |
95 | ||
96 | MIG_RP_MSG_REQ_PAGES_ID, /* data (start: be64, len: be32, id: string) */ | |
97 | MIG_RP_MSG_REQ_PAGES, /* data (start: be64, len: be32) */ | |
a335debb | 98 | MIG_RP_MSG_RECV_BITMAP, /* send recved_bitmap back to source */ |
13955b89 | 99 | MIG_RP_MSG_RESUME_ACK, /* tell source that we are ready to resume */ |
da6f1790 JQ |
100 | |
101 | MIG_RP_MSG_MAX | |
102 | }; | |
103 | ||
17549e84 JQ |
104 | /* When we add fault tolerance, we could have several |
105 | migrations at once. For now we don't need to add | |
106 | dynamic creation of migration */ | |
107 | ||
e5cb7e76 | 108 | static MigrationState *current_migration; |
e1b1b1bc | 109 | static MigrationIncomingState *current_incoming; |
e5cb7e76 | 110 | |
8b0b29dc | 111 | static bool migration_object_check(MigrationState *ms, Error **errp); |
0331c8ca DDAG |
112 | static int migration_maybe_pause(MigrationState *s, |
113 | int *current_active_state, | |
114 | int new_state); | |
8b0b29dc | 115 | |
e5cb7e76 PX |
116 | void migration_object_init(void) |
117 | { | |
4ffdb337 | 118 | MachineState *ms = MACHINE(qdev_get_machine()); |
8b0b29dc | 119 | Error *err = NULL; |
4ffdb337 | 120 | |
e5cb7e76 PX |
121 | /* This can only be called once. */ |
122 | assert(!current_migration); | |
123 | current_migration = MIGRATION_OBJ(object_new(TYPE_MIGRATION)); | |
4ffdb337 | 124 | |
e1b1b1bc PX |
125 | /* |
126 | * Init the migrate incoming object as well no matter whether | |
127 | * we'll use it or not. | |
128 | */ | |
129 | assert(!current_incoming); | |
130 | current_incoming = g_new0(MigrationIncomingState, 1); | |
131 | current_incoming->state = MIGRATION_STATUS_NONE; | |
132 | current_incoming->postcopy_remote_fds = | |
133 | g_array_new(FALSE, TRUE, sizeof(struct PostCopyFD)); | |
134 | qemu_mutex_init(¤t_incoming->rp_mutex); | |
135 | qemu_event_init(¤t_incoming->main_thread_load_event, false); | |
136 | qemu_sem_init(¤t_incoming->postcopy_pause_sem_dst, 0); | |
137 | qemu_sem_init(¤t_incoming->postcopy_pause_sem_fault, 0); | |
138 | ||
139 | init_dirty_bitmap_incoming_migration(); | |
140 | ||
8b0b29dc PX |
141 | if (!migration_object_check(current_migration, &err)) { |
142 | error_report_err(err); | |
143 | exit(1); | |
144 | } | |
145 | ||
4ffdb337 PX |
146 | /* |
147 | * We cannot really do this in migration_instance_init() since at | |
148 | * that time global properties are not yet applied, then this | |
149 | * value will be definitely replaced by something else. | |
150 | */ | |
151 | if (ms->enforce_config_section) { | |
152 | current_migration->send_configuration = true; | |
153 | } | |
e5cb7e76 PX |
154 | } |
155 | ||
1f895604 VSO |
156 | void migration_object_finalize(void) |
157 | { | |
158 | object_unref(OBJECT(current_migration)); | |
159 | } | |
160 | ||
bca7856a | 161 | /* For outgoing */ |
859bc756 | 162 | MigrationState *migrate_get_current(void) |
17549e84 | 163 | { |
e5cb7e76 PX |
164 | /* This can only be called after the object created. */ |
165 | assert(current_migration); | |
166 | return current_migration; | |
17549e84 JQ |
167 | } |
168 | ||
bca7856a DDAG |
169 | MigrationIncomingState *migration_incoming_get_current(void) |
170 | { | |
e1b1b1bc PX |
171 | assert(current_incoming); |
172 | return current_incoming; | |
bca7856a DDAG |
173 | } |
174 | ||
175 | void migration_incoming_state_destroy(void) | |
176 | { | |
b4b076da JQ |
177 | struct MigrationIncomingState *mis = migration_incoming_get_current(); |
178 | ||
3482655b | 179 | if (mis->to_src_file) { |
660819b1 PX |
180 | /* Tell source that we are done */ |
181 | migrate_send_rp_shut(mis, qemu_file_get_error(mis->from_src_file) != 0); | |
3482655b PX |
182 | qemu_fclose(mis->to_src_file); |
183 | mis->to_src_file = NULL; | |
184 | } | |
185 | ||
660819b1 PX |
186 | if (mis->from_src_file) { |
187 | qemu_fclose(mis->from_src_file); | |
188 | mis->from_src_file = NULL; | |
189 | } | |
00fa4fc8 DDAG |
190 | if (mis->postcopy_remote_fds) { |
191 | g_array_free(mis->postcopy_remote_fds, TRUE); | |
192 | mis->postcopy_remote_fds = NULL; | |
193 | } | |
660819b1 | 194 | |
5089e186 | 195 | qemu_event_reset(&mis->main_thread_load_event); |
bca7856a DDAG |
196 | } |
197 | ||
b05dc723 JQ |
198 | static void migrate_generate_event(int new_state) |
199 | { | |
200 | if (migrate_use_events()) { | |
201 | qapi_event_send_migration(new_state, &error_abort); | |
b05dc723 JQ |
202 | } |
203 | } | |
204 | ||
adde220a DDAG |
205 | /* |
206 | * Called on -incoming with a defer: uri. | |
207 | * The migration can be started later after any parameters have been | |
208 | * changed. | |
209 | */ | |
210 | static void deferred_incoming_migration(Error **errp) | |
211 | { | |
212 | if (deferred_incoming) { | |
213 | error_setg(errp, "Incoming migration already deferred"); | |
214 | } | |
215 | deferred_incoming = true; | |
216 | } | |
217 | ||
da6f1790 JQ |
218 | /* |
219 | * Send a message on the return channel back to the source | |
220 | * of the migration. | |
221 | */ | |
d6208e35 PX |
222 | static int migrate_send_rp_message(MigrationIncomingState *mis, |
223 | enum mig_rp_message_type message_type, | |
224 | uint16_t len, void *data) | |
da6f1790 | 225 | { |
d6208e35 PX |
226 | int ret = 0; |
227 | ||
da6f1790 JQ |
228 | trace_migrate_send_rp_message((int)message_type, len); |
229 | qemu_mutex_lock(&mis->rp_mutex); | |
d6208e35 PX |
230 | |
231 | /* | |
232 | * It's possible that the file handle got lost due to network | |
233 | * failures. | |
234 | */ | |
235 | if (!mis->to_src_file) { | |
236 | ret = -EIO; | |
237 | goto error; | |
238 | } | |
239 | ||
da6f1790 JQ |
240 | qemu_put_be16(mis->to_src_file, (unsigned int)message_type); |
241 | qemu_put_be16(mis->to_src_file, len); | |
242 | qemu_put_buffer(mis->to_src_file, data, len); | |
243 | qemu_fflush(mis->to_src_file); | |
d6208e35 PX |
244 | |
245 | /* It's possible that qemu file got error during sending */ | |
246 | ret = qemu_file_get_error(mis->to_src_file); | |
247 | ||
248 | error: | |
da6f1790 | 249 | qemu_mutex_unlock(&mis->rp_mutex); |
d6208e35 | 250 | return ret; |
da6f1790 JQ |
251 | } |
252 | ||
1e2d90eb DDAG |
253 | /* Request a range of pages from the source VM at the given |
254 | * start address. | |
255 | * rbname: Name of the RAMBlock to request the page in, if NULL it's the same | |
256 | * as the last request (a name must have been given previously) | |
257 | * Start: Address offset within the RB | |
258 | * Len: Length in bytes required - must be a multiple of pagesize | |
259 | */ | |
d6208e35 PX |
260 | int migrate_send_rp_req_pages(MigrationIncomingState *mis, const char *rbname, |
261 | ram_addr_t start, size_t len) | |
1e2d90eb | 262 | { |
cb8d4c8f | 263 | uint8_t bufc[12 + 1 + 255]; /* start (8), len (4), rbname up to 256 */ |
1e2d90eb | 264 | size_t msglen = 12; /* start + len */ |
d6208e35 | 265 | enum mig_rp_message_type msg_type; |
1e2d90eb DDAG |
266 | |
267 | *(uint64_t *)bufc = cpu_to_be64((uint64_t)start); | |
268 | *(uint32_t *)(bufc + 8) = cpu_to_be32((uint32_t)len); | |
269 | ||
270 | if (rbname) { | |
271 | int rbname_len = strlen(rbname); | |
272 | assert(rbname_len < 256); | |
273 | ||
274 | bufc[msglen++] = rbname_len; | |
275 | memcpy(bufc + msglen, rbname, rbname_len); | |
276 | msglen += rbname_len; | |
d6208e35 | 277 | msg_type = MIG_RP_MSG_REQ_PAGES_ID; |
1e2d90eb | 278 | } else { |
d6208e35 | 279 | msg_type = MIG_RP_MSG_REQ_PAGES; |
1e2d90eb | 280 | } |
d6208e35 PX |
281 | |
282 | return migrate_send_rp_message(mis, msg_type, msglen, bufc); | |
1e2d90eb DDAG |
283 | } |
284 | ||
43eaae28 | 285 | void qemu_start_incoming_migration(const char *uri, Error **errp) |
5bb7910a | 286 | { |
34c9dd8e AL |
287 | const char *p; |
288 | ||
7cf1fe6d | 289 | qapi_event_send_migration(MIGRATION_STATUS_SETUP, &error_abort); |
adde220a DDAG |
290 | if (!strcmp(uri, "defer")) { |
291 | deferred_incoming_migration(errp); | |
292 | } else if (strstart(uri, "tcp:", &p)) { | |
43eaae28 | 293 | tcp_start_incoming_migration(p, errp); |
2da776db | 294 | #ifdef CONFIG_RDMA |
adde220a | 295 | } else if (strstart(uri, "rdma:", &p)) { |
2da776db MH |
296 | rdma_start_incoming_migration(p, errp); |
297 | #endif | |
adde220a | 298 | } else if (strstart(uri, "exec:", &p)) { |
43eaae28 | 299 | exec_start_incoming_migration(p, errp); |
adde220a | 300 | } else if (strstart(uri, "unix:", &p)) { |
43eaae28 | 301 | unix_start_incoming_migration(p, errp); |
adde220a | 302 | } else if (strstart(uri, "fd:", &p)) { |
43eaae28 | 303 | fd_start_incoming_migration(p, errp); |
adde220a | 304 | } else { |
312fd5f2 | 305 | error_setg(errp, "unknown migration protocol: %s", uri); |
8ca5e801 | 306 | } |
5bb7910a AL |
307 | } |
308 | ||
0aa6aefc DL |
309 | static void process_incoming_migration_bh(void *opaque) |
310 | { | |
311 | Error *local_err = NULL; | |
312 | MigrationIncomingState *mis = opaque; | |
313 | ||
a18a73d7 DDAG |
314 | /* Make sure all file formats flush their mutable metadata. |
315 | * If we get an error here, just don't restart the VM yet. */ | |
316 | bdrv_invalidate_cache_all(&local_err); | |
317 | if (local_err) { | |
318 | error_report_err(local_err); | |
319 | local_err = NULL; | |
320 | autostart = false; | |
d35ff5e6 KW |
321 | } |
322 | ||
0aa6aefc DL |
323 | /* |
324 | * This must happen after all error conditions are dealt with and | |
325 | * we're sure the VM is going to be running on this host. | |
326 | */ | |
327 | qemu_announce_self(); | |
328 | ||
f986c3d2 JQ |
329 | if (multifd_load_cleanup(&local_err) != 0) { |
330 | error_report_err(local_err); | |
331 | autostart = false; | |
332 | } | |
0aa6aefc DL |
333 | /* If global state section was not received or we are in running |
334 | state, we need to obey autostart. Any other state is set with | |
335 | runstate_set. */ | |
336 | ||
b35ebdf0 VSO |
337 | dirty_bitmap_mig_before_vm_start(); |
338 | ||
0aa6aefc DL |
339 | if (!global_state_received() || |
340 | global_state_get_runstate() == RUN_STATE_RUNNING) { | |
341 | if (autostart) { | |
342 | vm_start(); | |
343 | } else { | |
344 | runstate_set(RUN_STATE_PAUSED); | |
345 | } | |
346 | } else { | |
347 | runstate_set(global_state_get_runstate()); | |
348 | } | |
0aa6aefc DL |
349 | /* |
350 | * This must happen after any state changes since as soon as an external | |
351 | * observer sees this event they might start to prod at the VM assuming | |
352 | * it's ready to use. | |
353 | */ | |
354 | migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE, | |
355 | MIGRATION_STATUS_COMPLETED); | |
356 | qemu_bh_delete(mis->bh); | |
357 | migration_incoming_state_destroy(); | |
358 | } | |
359 | ||
82a4da79 | 360 | static void process_incoming_migration_co(void *opaque) |
511c0231 | 361 | { |
b4b076da | 362 | MigrationIncomingState *mis = migration_incoming_get_current(); |
e9bef235 | 363 | PostcopyState ps; |
1c12e1f5 PB |
364 | int ret; |
365 | ||
4f0fae7f | 366 | assert(mis->from_src_file); |
67f11b5c | 367 | mis->largest_page_size = qemu_ram_pagesize_largest(); |
093e3c42 | 368 | postcopy_state_set(POSTCOPY_INCOMING_NONE); |
93d7af6f HZ |
369 | migrate_set_state(&mis->state, MIGRATION_STATUS_NONE, |
370 | MIGRATION_STATUS_ACTIVE); | |
4f0fae7f | 371 | ret = qemu_loadvm_state(mis->from_src_file); |
bca7856a | 372 | |
e9bef235 DDAG |
373 | ps = postcopy_state_get(); |
374 | trace_process_incoming_migration_co_end(ret, ps); | |
375 | if (ps != POSTCOPY_INCOMING_NONE) { | |
376 | if (ps == POSTCOPY_INCOMING_ADVISE) { | |
377 | /* | |
378 | * Where a migration had postcopy enabled (and thus went to advise) | |
379 | * but managed to complete within the precopy period, we can use | |
380 | * the normal exit. | |
381 | */ | |
382 | postcopy_ram_incoming_cleanup(mis); | |
383 | } else if (ret >= 0) { | |
384 | /* | |
385 | * Postcopy was started, cleanup should happen at the end of the | |
386 | * postcopy thread. | |
387 | */ | |
388 | trace_process_incoming_migration_co_postcopy_end_main(); | |
389 | return; | |
390 | } | |
391 | /* Else if something went wrong then just fall out of the normal exit */ | |
392 | } | |
393 | ||
25d0c16f HZ |
394 | /* we get COLO info, and know if we are in COLO mode */ |
395 | if (!ret && migration_incoming_enable_colo()) { | |
396 | mis->migration_incoming_co = qemu_coroutine_self(); | |
397 | qemu_thread_create(&mis->colo_incoming_thread, "COLO incoming", | |
398 | colo_process_incoming_thread, mis, QEMU_THREAD_JOINABLE); | |
399 | mis->have_colo_incoming_thread = true; | |
400 | qemu_coroutine_yield(); | |
401 | ||
402 | /* Wait checkpoint incoming thread exit before free resource */ | |
403 | qemu_thread_join(&mis->colo_incoming_thread); | |
404 | } | |
405 | ||
1c12e1f5 | 406 | if (ret < 0) { |
f986c3d2 JQ |
407 | Error *local_err = NULL; |
408 | ||
93d7af6f HZ |
409 | migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE, |
410 | MIGRATION_STATUS_FAILED); | |
db80face | 411 | error_report("load of migration failed: %s", strerror(-ret)); |
3a0f2cea | 412 | qemu_fclose(mis->from_src_file); |
f986c3d2 JQ |
413 | if (multifd_load_cleanup(&local_err) != 0) { |
414 | error_report_err(local_err); | |
415 | } | |
4aead692 | 416 | exit(EXIT_FAILURE); |
511c0231 | 417 | } |
0aa6aefc DL |
418 | mis->bh = qemu_bh_new(process_incoming_migration_bh, mis); |
419 | qemu_bh_schedule(mis->bh); | |
511c0231 JQ |
420 | } |
421 | ||
e595a01a | 422 | static void migration_incoming_setup(QEMUFile *f) |
82a4da79 | 423 | { |
4f0fae7f | 424 | MigrationIncomingState *mis = migration_incoming_get_current(); |
82a4da79 | 425 | |
f986c3d2 JQ |
426 | if (multifd_load_setup() != 0) { |
427 | /* We haven't been able to create multifd threads | |
428 | nothing better to do */ | |
429 | exit(EXIT_FAILURE); | |
430 | } | |
431 | ||
4f0fae7f JQ |
432 | if (!mis->from_src_file) { |
433 | mis->from_src_file = f; | |
434 | } | |
06ad5135 | 435 | qemu_file_set_blocking(f, false); |
e595a01a JQ |
436 | } |
437 | ||
36c2f8be | 438 | void migration_incoming_process(void) |
e595a01a JQ |
439 | { |
440 | Coroutine *co = qemu_coroutine_create(process_incoming_migration_co, NULL); | |
0b8b8753 | 441 | qemu_coroutine_enter(co); |
82a4da79 PB |
442 | } |
443 | ||
e595a01a JQ |
444 | void migration_fd_process_incoming(QEMUFile *f) |
445 | { | |
d96c9e8d PX |
446 | MigrationIncomingState *mis = migration_incoming_get_current(); |
447 | ||
448 | if (mis->state == MIGRATION_STATUS_POSTCOPY_PAUSED) { | |
449 | /* Resumed from a paused postcopy migration */ | |
450 | ||
451 | mis->from_src_file = f; | |
452 | /* Postcopy has standalone thread to do vm load */ | |
453 | qemu_file_set_blocking(f, true); | |
454 | ||
455 | /* Re-configure the return path */ | |
456 | mis->to_src_file = qemu_file_get_return_path(f); | |
457 | ||
458 | migrate_set_state(&mis->state, MIGRATION_STATUS_POSTCOPY_PAUSED, | |
459 | MIGRATION_STATUS_POSTCOPY_RECOVER); | |
460 | ||
461 | /* | |
462 | * Here, we only wake up the main loading thread (while the | |
463 | * fault thread will still be waiting), so that we can receive | |
464 | * commands from source now, and answer it if needed. The | |
465 | * fault thread will be woken up afterwards until we are sure | |
466 | * that source is ready to reply to page requests. | |
467 | */ | |
468 | qemu_sem_post(&mis->postcopy_pause_sem_dst); | |
469 | } else { | |
470 | /* New incoming migration */ | |
471 | migration_incoming_setup(f); | |
472 | migration_incoming_process(); | |
473 | } | |
e595a01a JQ |
474 | } |
475 | ||
4f0fae7f JQ |
476 | void migration_ioc_process_incoming(QIOChannel *ioc) |
477 | { | |
478 | MigrationIncomingState *mis = migration_incoming_get_current(); | |
479 | ||
480 | if (!mis->from_src_file) { | |
481 | QEMUFile *f = qemu_fopen_channel_input(ioc); | |
36c2f8be | 482 | migration_incoming_setup(f); |
71bb07db | 483 | return; |
4f0fae7f | 484 | } |
71bb07db | 485 | multifd_recv_new_channel(ioc); |
4f0fae7f JQ |
486 | } |
487 | ||
428d8908 JQ |
488 | /** |
489 | * @migration_has_all_channels: We have received all channels that we need | |
490 | * | |
491 | * Returns true when we have got connections to all the channels that | |
492 | * we need for migration. | |
493 | */ | |
494 | bool migration_has_all_channels(void) | |
495 | { | |
62c1e0ca JQ |
496 | bool all_channels; |
497 | ||
498 | all_channels = multifd_recv_all_channels_created(); | |
499 | ||
500 | return all_channels; | |
428d8908 JQ |
501 | } |
502 | ||
6decec93 DDAG |
503 | /* |
504 | * Send a 'SHUT' message on the return channel with the given value | |
505 | * to indicate that we've finished with the RP. Non-0 value indicates | |
506 | * error. | |
507 | */ | |
508 | void migrate_send_rp_shut(MigrationIncomingState *mis, | |
509 | uint32_t value) | |
510 | { | |
511 | uint32_t buf; | |
512 | ||
513 | buf = cpu_to_be32(value); | |
514 | migrate_send_rp_message(mis, MIG_RP_MSG_SHUT, sizeof(buf), &buf); | |
515 | } | |
516 | ||
517 | /* | |
518 | * Send a 'PONG' message on the return channel with the given value | |
519 | * (normally in response to a 'PING') | |
520 | */ | |
521 | void migrate_send_rp_pong(MigrationIncomingState *mis, | |
522 | uint32_t value) | |
523 | { | |
524 | uint32_t buf; | |
525 | ||
526 | buf = cpu_to_be32(value); | |
527 | migrate_send_rp_message(mis, MIG_RP_MSG_PONG, sizeof(buf), &buf); | |
528 | } | |
529 | ||
a335debb PX |
530 | void migrate_send_rp_recv_bitmap(MigrationIncomingState *mis, |
531 | char *block_name) | |
532 | { | |
533 | char buf[512]; | |
534 | int len; | |
535 | int64_t res; | |
536 | ||
537 | /* | |
538 | * First, we send the header part. It contains only the len of | |
539 | * idstr, and the idstr itself. | |
540 | */ | |
541 | len = strlen(block_name); | |
542 | buf[0] = len; | |
543 | memcpy(buf + 1, block_name, len); | |
544 | ||
545 | if (mis->state != MIGRATION_STATUS_POSTCOPY_RECOVER) { | |
546 | error_report("%s: MSG_RP_RECV_BITMAP only used for recovery", | |
547 | __func__); | |
548 | return; | |
549 | } | |
550 | ||
551 | migrate_send_rp_message(mis, MIG_RP_MSG_RECV_BITMAP, len + 1, buf); | |
552 | ||
553 | /* | |
554 | * Next, we dump the received bitmap to the stream. | |
555 | * | |
556 | * TODO: currently we are safe since we are the only one that is | |
557 | * using the to_src_file handle (fault thread is still paused), | |
558 | * and it's ok even not taking the mutex. However the best way is | |
559 | * to take the lock before sending the message header, and release | |
560 | * the lock after sending the bitmap. | |
561 | */ | |
562 | qemu_mutex_lock(&mis->rp_mutex); | |
563 | res = ramblock_recv_bitmap_send(mis->to_src_file, block_name); | |
564 | qemu_mutex_unlock(&mis->rp_mutex); | |
565 | ||
566 | trace_migrate_send_rp_recv_bitmap(block_name, res); | |
567 | } | |
568 | ||
13955b89 PX |
569 | void migrate_send_rp_resume_ack(MigrationIncomingState *mis, uint32_t value) |
570 | { | |
571 | uint32_t buf; | |
572 | ||
573 | buf = cpu_to_be32(value); | |
574 | migrate_send_rp_message(mis, MIG_RP_MSG_RESUME_ACK, sizeof(buf), &buf); | |
575 | } | |
576 | ||
bbf6da32 OW |
577 | MigrationCapabilityStatusList *qmp_query_migrate_capabilities(Error **errp) |
578 | { | |
579 | MigrationCapabilityStatusList *head = NULL; | |
580 | MigrationCapabilityStatusList *caps; | |
581 | MigrationState *s = migrate_get_current(); | |
582 | int i; | |
583 | ||
387eedeb | 584 | caps = NULL; /* silence compiler warning */ |
7fb1cf16 | 585 | for (i = 0; i < MIGRATION_CAPABILITY__MAX; i++) { |
ed1701c6 DDAG |
586 | #ifndef CONFIG_LIVE_BLOCK_MIGRATION |
587 | if (i == MIGRATION_CAPABILITY_BLOCK) { | |
588 | continue; | |
589 | } | |
590 | #endif | |
bbf6da32 OW |
591 | if (head == NULL) { |
592 | head = g_malloc0(sizeof(*caps)); | |
593 | caps = head; | |
594 | } else { | |
595 | caps->next = g_malloc0(sizeof(*caps)); | |
596 | caps = caps->next; | |
597 | } | |
598 | caps->value = | |
599 | g_malloc(sizeof(*caps->value)); | |
600 | caps->value->capability = i; | |
601 | caps->value->state = s->enabled_capabilities[i]; | |
602 | } | |
603 | ||
604 | return head; | |
605 | } | |
606 | ||
85de8323 LL |
607 | MigrationParameters *qmp_query_migrate_parameters(Error **errp) |
608 | { | |
609 | MigrationParameters *params; | |
610 | MigrationState *s = migrate_get_current(); | |
611 | ||
e87fae4c | 612 | /* TODO use QAPI_CLONE() instead of duplicating it inline */ |
85de8323 | 613 | params = g_malloc0(sizeof(*params)); |
de63ab61 | 614 | params->has_compress_level = true; |
2594f56d | 615 | params->compress_level = s->parameters.compress_level; |
de63ab61 | 616 | params->has_compress_threads = true; |
2594f56d | 617 | params->compress_threads = s->parameters.compress_threads; |
de63ab61 | 618 | params->has_decompress_threads = true; |
2594f56d | 619 | params->decompress_threads = s->parameters.decompress_threads; |
de63ab61 | 620 | params->has_cpu_throttle_initial = true; |
2594f56d | 621 | params->cpu_throttle_initial = s->parameters.cpu_throttle_initial; |
de63ab61 | 622 | params->has_cpu_throttle_increment = true; |
2594f56d | 623 | params->cpu_throttle_increment = s->parameters.cpu_throttle_increment; |
8cc99dcd | 624 | params->has_tls_creds = true; |
69ef1f36 | 625 | params->tls_creds = g_strdup(s->parameters.tls_creds); |
8cc99dcd | 626 | params->has_tls_hostname = true; |
69ef1f36 | 627 | params->tls_hostname = g_strdup(s->parameters.tls_hostname); |
2ff30257 AA |
628 | params->has_max_bandwidth = true; |
629 | params->max_bandwidth = s->parameters.max_bandwidth; | |
630 | params->has_downtime_limit = true; | |
631 | params->downtime_limit = s->parameters.downtime_limit; | |
fe39a4d4 | 632 | params->has_x_checkpoint_delay = true; |
68b53591 | 633 | params->x_checkpoint_delay = s->parameters.x_checkpoint_delay; |
2833c59b JQ |
634 | params->has_block_incremental = true; |
635 | params->block_incremental = s->parameters.block_incremental; | |
4075fb1c JQ |
636 | params->has_x_multifd_channels = true; |
637 | params->x_multifd_channels = s->parameters.x_multifd_channels; | |
0fb86605 JQ |
638 | params->has_x_multifd_page_count = true; |
639 | params->x_multifd_page_count = s->parameters.x_multifd_page_count; | |
73af8dd8 JQ |
640 | params->has_xbzrle_cache_size = true; |
641 | params->xbzrle_cache_size = s->parameters.xbzrle_cache_size; | |
85de8323 LL |
642 | |
643 | return params; | |
644 | } | |
645 | ||
f6844b99 DDAG |
646 | /* |
647 | * Return true if we're already in the middle of a migration | |
648 | * (i.e. any of the active or setup states) | |
649 | */ | |
650 | static bool migration_is_setup_or_active(int state) | |
651 | { | |
652 | switch (state) { | |
653 | case MIGRATION_STATUS_ACTIVE: | |
9ec055ae | 654 | case MIGRATION_STATUS_POSTCOPY_ACTIVE: |
a688d2c1 | 655 | case MIGRATION_STATUS_POSTCOPY_PAUSED: |
135b87b4 | 656 | case MIGRATION_STATUS_POSTCOPY_RECOVER: |
f6844b99 | 657 | case MIGRATION_STATUS_SETUP: |
31e06077 DDAG |
658 | case MIGRATION_STATUS_PRE_SWITCHOVER: |
659 | case MIGRATION_STATUS_DEVICE: | |
f6844b99 DDAG |
660 | return true; |
661 | ||
662 | default: | |
663 | return false; | |
664 | ||
665 | } | |
666 | } | |
667 | ||
a22463a5 DDAG |
668 | static void populate_ram_info(MigrationInfo *info, MigrationState *s) |
669 | { | |
670 | info->has_ram = true; | |
671 | info->ram = g_malloc0(sizeof(*info->ram)); | |
9360447d | 672 | info->ram->transferred = ram_counters.transferred; |
a22463a5 | 673 | info->ram->total = ram_bytes_total(); |
9360447d | 674 | info->ram->duplicate = ram_counters.duplicate; |
bedf53c1 JQ |
675 | /* legacy value. It is not used anymore */ |
676 | info->ram->skipped = 0; | |
9360447d JQ |
677 | info->ram->normal = ram_counters.normal; |
678 | info->ram->normal_bytes = ram_counters.normal * | |
20afaed9 | 679 | qemu_target_page_size(); |
a22463a5 | 680 | info->ram->mbps = s->mbps; |
9360447d JQ |
681 | info->ram->dirty_sync_count = ram_counters.dirty_sync_count; |
682 | info->ram->postcopy_requests = ram_counters.postcopy_requests; | |
030ce1f8 | 683 | info->ram->page_size = qemu_target_page_size(); |
a22463a5 | 684 | |
114f5aee JQ |
685 | if (migrate_use_xbzrle()) { |
686 | info->has_xbzrle_cache = true; | |
687 | info->xbzrle_cache = g_malloc0(sizeof(*info->xbzrle_cache)); | |
688 | info->xbzrle_cache->cache_size = migrate_xbzrle_cache_size(); | |
9360447d JQ |
689 | info->xbzrle_cache->bytes = xbzrle_counters.bytes; |
690 | info->xbzrle_cache->pages = xbzrle_counters.pages; | |
691 | info->xbzrle_cache->cache_miss = xbzrle_counters.cache_miss; | |
692 | info->xbzrle_cache->cache_miss_rate = xbzrle_counters.cache_miss_rate; | |
693 | info->xbzrle_cache->overflow = xbzrle_counters.overflow; | |
114f5aee JQ |
694 | } |
695 | ||
338182c8 JQ |
696 | if (cpu_throttle_active()) { |
697 | info->has_cpu_throttle_percentage = true; | |
698 | info->cpu_throttle_percentage = cpu_throttle_get_percentage(); | |
699 | } | |
700 | ||
a22463a5 DDAG |
701 | if (s->state != MIGRATION_STATUS_COMPLETED) { |
702 | info->ram->remaining = ram_bytes_remaining(); | |
9360447d | 703 | info->ram->dirty_pages_rate = ram_counters.dirty_pages_rate; |
a22463a5 DDAG |
704 | } |
705 | } | |
706 | ||
930ac04c JQ |
707 | static void populate_disk_info(MigrationInfo *info) |
708 | { | |
709 | if (blk_mig_active()) { | |
710 | info->has_disk = true; | |
711 | info->disk = g_malloc0(sizeof(*info->disk)); | |
712 | info->disk->transferred = blk_mig_bytes_transferred(); | |
713 | info->disk->remaining = blk_mig_bytes_remaining(); | |
714 | info->disk->total = blk_mig_bytes_total(); | |
715 | } | |
716 | } | |
717 | ||
65ace060 | 718 | static void fill_source_migration_info(MigrationInfo *info) |
5bb7910a | 719 | { |
17549e84 JQ |
720 | MigrationState *s = migrate_get_current(); |
721 | ||
722 | switch (s->state) { | |
31194731 | 723 | case MIGRATION_STATUS_NONE: |
17549e84 | 724 | /* no migration has happened ever */ |
65ace060 AP |
725 | /* do not overwrite destination migration status */ |
726 | return; | |
17549e84 | 727 | break; |
31194731 | 728 | case MIGRATION_STATUS_SETUP: |
29ae8a41 | 729 | info->has_status = true; |
ed4fbd10 | 730 | info->has_total_time = false; |
29ae8a41 | 731 | break; |
31194731 HZ |
732 | case MIGRATION_STATUS_ACTIVE: |
733 | case MIGRATION_STATUS_CANCELLING: | |
9ec055ae | 734 | case MIGRATION_STATUS_POSTCOPY_ACTIVE: |
31e06077 DDAG |
735 | case MIGRATION_STATUS_PRE_SWITCHOVER: |
736 | case MIGRATION_STATUS_DEVICE: | |
a688d2c1 | 737 | case MIGRATION_STATUS_POSTCOPY_PAUSED: |
135b87b4 | 738 | case MIGRATION_STATUS_POSTCOPY_RECOVER: |
c8f9f4f4 | 739 | /* TODO add some postcopy stats */ |
9ec055ae DDAG |
740 | info->has_status = true; |
741 | info->has_total_time = true; | |
742 | info->total_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME) | |
4af246a3 | 743 | - s->start_time; |
9ec055ae DDAG |
744 | info->has_expected_downtime = true; |
745 | info->expected_downtime = s->expected_downtime; | |
746 | info->has_setup_time = true; | |
747 | info->setup_time = s->setup_time; | |
748 | ||
a22463a5 | 749 | populate_ram_info(info, s); |
930ac04c | 750 | populate_disk_info(info); |
17549e84 | 751 | break; |
0b827d5e HZ |
752 | case MIGRATION_STATUS_COLO: |
753 | info->has_status = true; | |
754 | /* TODO: display COLO specific information (checkpoint info etc.) */ | |
755 | break; | |
31194731 | 756 | case MIGRATION_STATUS_COMPLETED: |
791e7c82 | 757 | info->has_status = true; |
00c14997 | 758 | info->has_total_time = true; |
7aa939af | 759 | info->total_time = s->total_time; |
9c5a9fcf JQ |
760 | info->has_downtime = true; |
761 | info->downtime = s->downtime; | |
ed4fbd10 MH |
762 | info->has_setup_time = true; |
763 | info->setup_time = s->setup_time; | |
d5f8a570 | 764 | |
a22463a5 | 765 | populate_ram_info(info, s); |
17549e84 | 766 | break; |
31194731 | 767 | case MIGRATION_STATUS_FAILED: |
791e7c82 | 768 | info->has_status = true; |
d59ce6f3 DB |
769 | if (s->error) { |
770 | info->has_error_desc = true; | |
771 | info->error_desc = g_strdup(error_get_pretty(s->error)); | |
772 | } | |
17549e84 | 773 | break; |
31194731 | 774 | case MIGRATION_STATUS_CANCELLED: |
791e7c82 | 775 | info->has_status = true; |
17549e84 | 776 | break; |
5bb7910a | 777 | } |
cde63fbe | 778 | info->status = s->state; |
5bb7910a AL |
779 | } |
780 | ||
4a84214e PX |
781 | /** |
782 | * @migration_caps_check - check capability validity | |
783 | * | |
784 | * @cap_list: old capability list, array of bool | |
785 | * @params: new capabilities to be applied soon | |
786 | * @errp: set *errp if the check failed, with reason | |
787 | * | |
788 | * Returns true if check passed, otherwise false. | |
789 | */ | |
790 | static bool migrate_caps_check(bool *cap_list, | |
791 | MigrationCapabilityStatusList *params, | |
792 | Error **errp) | |
00458433 | 793 | { |
00458433 | 794 | MigrationCapabilityStatusList *cap; |
4a84214e | 795 | bool old_postcopy_cap; |
d7651f15 | 796 | MigrationIncomingState *mis = migration_incoming_get_current(); |
00458433 | 797 | |
4a84214e | 798 | old_postcopy_cap = cap_list[MIGRATION_CAPABILITY_POSTCOPY_RAM]; |
00458433 OW |
799 | |
800 | for (cap = params; cap; cap = cap->next) { | |
4a84214e PX |
801 | cap_list[cap->value->capability] = cap->value->state; |
802 | } | |
803 | ||
ed1701c6 | 804 | #ifndef CONFIG_LIVE_BLOCK_MIGRATION |
4a84214e PX |
805 | if (cap_list[MIGRATION_CAPABILITY_BLOCK]) { |
806 | error_setg(errp, "QEMU compiled without old-style (blk/-b, inc/-i) " | |
807 | "block migration"); | |
808 | error_append_hint(errp, "Use drive_mirror+NBD instead.\n"); | |
809 | return false; | |
00458433 | 810 | } |
4a84214e | 811 | #endif |
53dd370c | 812 | |
4a84214e PX |
813 | if (cap_list[MIGRATION_CAPABILITY_POSTCOPY_RAM]) { |
814 | if (cap_list[MIGRATION_CAPABILITY_COMPRESS]) { | |
53dd370c DDAG |
815 | /* The decompression threads asynchronously write into RAM |
816 | * rather than use the atomic copies needed to avoid | |
817 | * userfaulting. It should be possible to fix the decompression | |
818 | * threads for compatibility in future. | |
819 | */ | |
4a84214e PX |
820 | error_setg(errp, "Postcopy is not currently compatible " |
821 | "with compression"); | |
822 | return false; | |
53dd370c | 823 | } |
4a84214e | 824 | |
096631bd DDAG |
825 | /* This check is reasonably expensive, so only when it's being |
826 | * set the first time, also it's only the destination that needs | |
827 | * special support. | |
828 | */ | |
829 | if (!old_postcopy_cap && runstate_check(RUN_STATE_INMIGRATE) && | |
d7651f15 | 830 | !postcopy_ram_supported_by_host(mis)) { |
096631bd DDAG |
831 | /* postcopy_ram_supported_by_host will have emitted a more |
832 | * detailed message | |
833 | */ | |
4a84214e PX |
834 | error_setg(errp, "Postcopy is not supported"); |
835 | return false; | |
096631bd | 836 | } |
53dd370c | 837 | } |
4a84214e PX |
838 | |
839 | return true; | |
840 | } | |
841 | ||
65ace060 AP |
842 | static void fill_destination_migration_info(MigrationInfo *info) |
843 | { | |
844 | MigrationIncomingState *mis = migration_incoming_get_current(); | |
845 | ||
846 | switch (mis->state) { | |
847 | case MIGRATION_STATUS_NONE: | |
848 | return; | |
849 | break; | |
850 | case MIGRATION_STATUS_SETUP: | |
851 | case MIGRATION_STATUS_CANCELLING: | |
852 | case MIGRATION_STATUS_CANCELLED: | |
853 | case MIGRATION_STATUS_ACTIVE: | |
854 | case MIGRATION_STATUS_POSTCOPY_ACTIVE: | |
855 | case MIGRATION_STATUS_FAILED: | |
856 | case MIGRATION_STATUS_COLO: | |
857 | info->has_status = true; | |
858 | break; | |
859 | case MIGRATION_STATUS_COMPLETED: | |
860 | info->has_status = true; | |
861 | fill_destination_postcopy_migration_info(info); | |
862 | break; | |
863 | } | |
864 | info->status = mis->state; | |
865 | } | |
866 | ||
867 | MigrationInfo *qmp_query_migrate(Error **errp) | |
868 | { | |
869 | MigrationInfo *info = g_malloc0(sizeof(*info)); | |
870 | ||
871 | fill_destination_migration_info(info); | |
872 | fill_source_migration_info(info); | |
873 | ||
874 | return info; | |
875 | } | |
876 | ||
4a84214e PX |
877 | void qmp_migrate_set_capabilities(MigrationCapabilityStatusList *params, |
878 | Error **errp) | |
879 | { | |
880 | MigrationState *s = migrate_get_current(); | |
881 | MigrationCapabilityStatusList *cap; | |
dd0ee30c | 882 | bool cap_list[MIGRATION_CAPABILITY__MAX]; |
4a84214e PX |
883 | |
884 | if (migration_is_setup_or_active(s->state)) { | |
885 | error_setg(errp, QERR_MIGRATION_ACTIVE); | |
886 | return; | |
887 | } | |
888 | ||
dd0ee30c PX |
889 | memcpy(cap_list, s->enabled_capabilities, sizeof(cap_list)); |
890 | if (!migrate_caps_check(cap_list, params, errp)) { | |
4a84214e PX |
891 | return; |
892 | } | |
893 | ||
894 | for (cap = params; cap; cap = cap->next) { | |
895 | s->enabled_capabilities[cap->value->capability] = cap->value->state; | |
896 | } | |
00458433 OW |
897 | } |
898 | ||
16d063bc PX |
899 | /* |
900 | * Check whether the parameters are valid. Error will be put into errp | |
901 | * (if provided). Return true if valid, otherwise false. | |
902 | */ | |
903 | static bool migrate_params_check(MigrationParameters *params, Error **errp) | |
85de8323 | 904 | { |
7f375e04 | 905 | if (params->has_compress_level && |
741d4086 | 906 | (params->compress_level > 9)) { |
c6bd8c70 MA |
907 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "compress_level", |
908 | "is invalid, it should be in the range of 0 to 9"); | |
16d063bc | 909 | return false; |
85de8323 | 910 | } |
16d063bc | 911 | |
741d4086 | 912 | if (params->has_compress_threads && (params->compress_threads < 1)) { |
c6bd8c70 MA |
913 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, |
914 | "compress_threads", | |
915 | "is invalid, it should be in the range of 1 to 255"); | |
16d063bc | 916 | return false; |
85de8323 | 917 | } |
16d063bc | 918 | |
741d4086 | 919 | if (params->has_decompress_threads && (params->decompress_threads < 1)) { |
c6bd8c70 MA |
920 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, |
921 | "decompress_threads", | |
922 | "is invalid, it should be in the range of 1 to 255"); | |
16d063bc | 923 | return false; |
85de8323 | 924 | } |
16d063bc | 925 | |
7f375e04 EB |
926 | if (params->has_cpu_throttle_initial && |
927 | (params->cpu_throttle_initial < 1 || | |
928 | params->cpu_throttle_initial > 99)) { | |
1626fee3 | 929 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, |
d85a31d1 | 930 | "cpu_throttle_initial", |
1626fee3 | 931 | "an integer in the range of 1 to 99"); |
16d063bc | 932 | return false; |
1626fee3 | 933 | } |
16d063bc | 934 | |
7f375e04 EB |
935 | if (params->has_cpu_throttle_increment && |
936 | (params->cpu_throttle_increment < 1 || | |
937 | params->cpu_throttle_increment > 99)) { | |
1626fee3 | 938 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, |
d85a31d1 | 939 | "cpu_throttle_increment", |
1626fee3 | 940 | "an integer in the range of 1 to 99"); |
16d063bc | 941 | return false; |
1626fee3 | 942 | } |
16d063bc | 943 | |
741d4086 | 944 | if (params->has_max_bandwidth && (params->max_bandwidth > SIZE_MAX)) { |
2ff30257 AA |
945 | error_setg(errp, "Parameter 'max_bandwidth' expects an integer in the" |
946 | " range of 0 to %zu bytes/second", SIZE_MAX); | |
16d063bc | 947 | return false; |
2ff30257 | 948 | } |
16d063bc | 949 | |
2ff30257 | 950 | if (params->has_downtime_limit && |
741d4086 | 951 | (params->downtime_limit > MAX_MIGRATE_DOWNTIME)) { |
87c9cc1c DHB |
952 | error_setg(errp, "Parameter 'downtime_limit' expects an integer in " |
953 | "the range of 0 to %d milliseconds", | |
954 | MAX_MIGRATE_DOWNTIME); | |
16d063bc | 955 | return false; |
2ff30257 | 956 | } |
16d063bc | 957 | |
741d4086 JQ |
958 | /* x_checkpoint_delay is now always positive */ |
959 | ||
960 | if (params->has_x_multifd_channels && (params->x_multifd_channels < 1)) { | |
4075fb1c JQ |
961 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, |
962 | "multifd_channels", | |
963 | "is invalid, it should be in the range of 1 to 255"); | |
964 | return false; | |
965 | } | |
0fb86605 | 966 | if (params->has_x_multifd_page_count && |
741d4086 JQ |
967 | (params->x_multifd_page_count < 1 || |
968 | params->x_multifd_page_count > 10000)) { | |
0fb86605 JQ |
969 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, |
970 | "multifd_page_count", | |
971 | "is invalid, it should be in the range of 1 to 10000"); | |
972 | return false; | |
973 | } | |
16d063bc | 974 | |
73af8dd8 JQ |
975 | if (params->has_xbzrle_cache_size && |
976 | (params->xbzrle_cache_size < qemu_target_page_size() || | |
977 | !is_power_of_2(params->xbzrle_cache_size))) { | |
978 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, | |
979 | "xbzrle_cache_size", | |
980 | "is invalid, it should be bigger than target page size" | |
981 | " and a power of two"); | |
982 | return false; | |
983 | } | |
984 | ||
16d063bc PX |
985 | return true; |
986 | } | |
987 | ||
1bda8b3c MA |
988 | static void migrate_params_test_apply(MigrateSetParameters *params, |
989 | MigrationParameters *dest) | |
990 | { | |
991 | *dest = migrate_get_current()->parameters; | |
992 | ||
993 | /* TODO use QAPI_CLONE() instead of duplicating it inline */ | |
994 | ||
995 | if (params->has_compress_level) { | |
996 | dest->compress_level = params->compress_level; | |
997 | } | |
998 | ||
999 | if (params->has_compress_threads) { | |
1000 | dest->compress_threads = params->compress_threads; | |
1001 | } | |
1002 | ||
1003 | if (params->has_decompress_threads) { | |
1004 | dest->decompress_threads = params->decompress_threads; | |
1005 | } | |
1006 | ||
1007 | if (params->has_cpu_throttle_initial) { | |
1008 | dest->cpu_throttle_initial = params->cpu_throttle_initial; | |
1009 | } | |
1010 | ||
1011 | if (params->has_cpu_throttle_increment) { | |
1012 | dest->cpu_throttle_increment = params->cpu_throttle_increment; | |
1013 | } | |
1014 | ||
1015 | if (params->has_tls_creds) { | |
01fa5598 MA |
1016 | assert(params->tls_creds->type == QTYPE_QSTRING); |
1017 | dest->tls_creds = g_strdup(params->tls_creds->u.s); | |
1bda8b3c MA |
1018 | } |
1019 | ||
1020 | if (params->has_tls_hostname) { | |
01fa5598 MA |
1021 | assert(params->tls_hostname->type == QTYPE_QSTRING); |
1022 | dest->tls_hostname = g_strdup(params->tls_hostname->u.s); | |
1bda8b3c MA |
1023 | } |
1024 | ||
1025 | if (params->has_max_bandwidth) { | |
1026 | dest->max_bandwidth = params->max_bandwidth; | |
1027 | } | |
1028 | ||
1029 | if (params->has_downtime_limit) { | |
1030 | dest->downtime_limit = params->downtime_limit; | |
1031 | } | |
1032 | ||
1033 | if (params->has_x_checkpoint_delay) { | |
1034 | dest->x_checkpoint_delay = params->x_checkpoint_delay; | |
1035 | } | |
1036 | ||
1037 | if (params->has_block_incremental) { | |
1038 | dest->block_incremental = params->block_incremental; | |
1039 | } | |
5e7577a1 JQ |
1040 | if (params->has_x_multifd_channels) { |
1041 | dest->x_multifd_channels = params->x_multifd_channels; | |
1042 | } | |
1043 | if (params->has_x_multifd_page_count) { | |
1044 | dest->x_multifd_page_count = params->x_multifd_page_count; | |
1045 | } | |
73af8dd8 JQ |
1046 | if (params->has_xbzrle_cache_size) { |
1047 | dest->xbzrle_cache_size = params->xbzrle_cache_size; | |
1048 | } | |
1bda8b3c MA |
1049 | } |
1050 | ||
73af8dd8 | 1051 | static void migrate_params_apply(MigrateSetParameters *params, Error **errp) |
16d063bc PX |
1052 | { |
1053 | MigrationState *s = migrate_get_current(); | |
1054 | ||
e87fae4c MA |
1055 | /* TODO use QAPI_CLONE() instead of duplicating it inline */ |
1056 | ||
7f375e04 EB |
1057 | if (params->has_compress_level) { |
1058 | s->parameters.compress_level = params->compress_level; | |
85de8323 | 1059 | } |
476c72aa | 1060 | |
7f375e04 EB |
1061 | if (params->has_compress_threads) { |
1062 | s->parameters.compress_threads = params->compress_threads; | |
85de8323 | 1063 | } |
476c72aa | 1064 | |
7f375e04 EB |
1065 | if (params->has_decompress_threads) { |
1066 | s->parameters.decompress_threads = params->decompress_threads; | |
85de8323 | 1067 | } |
476c72aa | 1068 | |
7f375e04 EB |
1069 | if (params->has_cpu_throttle_initial) { |
1070 | s->parameters.cpu_throttle_initial = params->cpu_throttle_initial; | |
1626fee3 | 1071 | } |
476c72aa | 1072 | |
7f375e04 EB |
1073 | if (params->has_cpu_throttle_increment) { |
1074 | s->parameters.cpu_throttle_increment = params->cpu_throttle_increment; | |
1626fee3 | 1075 | } |
476c72aa | 1076 | |
7f375e04 | 1077 | if (params->has_tls_creds) { |
69ef1f36 | 1078 | g_free(s->parameters.tls_creds); |
01fa5598 MA |
1079 | assert(params->tls_creds->type == QTYPE_QSTRING); |
1080 | s->parameters.tls_creds = g_strdup(params->tls_creds->u.s); | |
69ef1f36 | 1081 | } |
476c72aa | 1082 | |
7f375e04 | 1083 | if (params->has_tls_hostname) { |
69ef1f36 | 1084 | g_free(s->parameters.tls_hostname); |
01fa5598 MA |
1085 | assert(params->tls_hostname->type == QTYPE_QSTRING); |
1086 | s->parameters.tls_hostname = g_strdup(params->tls_hostname->u.s); | |
69ef1f36 | 1087 | } |
476c72aa | 1088 | |
2ff30257 AA |
1089 | if (params->has_max_bandwidth) { |
1090 | s->parameters.max_bandwidth = params->max_bandwidth; | |
1091 | if (s->to_dst_file) { | |
1092 | qemu_file_set_rate_limit(s->to_dst_file, | |
1093 | s->parameters.max_bandwidth / XFER_LIMIT_RATIO); | |
1094 | } | |
1095 | } | |
476c72aa | 1096 | |
2ff30257 AA |
1097 | if (params->has_downtime_limit) { |
1098 | s->parameters.downtime_limit = params->downtime_limit; | |
1099 | } | |
68b53591 HZ |
1100 | |
1101 | if (params->has_x_checkpoint_delay) { | |
1102 | s->parameters.x_checkpoint_delay = params->x_checkpoint_delay; | |
479125d5 HZ |
1103 | if (migration_in_colo_state()) { |
1104 | colo_checkpoint_notify(s); | |
1105 | } | |
68b53591 | 1106 | } |
476c72aa | 1107 | |
2833c59b JQ |
1108 | if (params->has_block_incremental) { |
1109 | s->parameters.block_incremental = params->block_incremental; | |
1110 | } | |
4075fb1c JQ |
1111 | if (params->has_x_multifd_channels) { |
1112 | s->parameters.x_multifd_channels = params->x_multifd_channels; | |
1113 | } | |
0fb86605 JQ |
1114 | if (params->has_x_multifd_page_count) { |
1115 | s->parameters.x_multifd_page_count = params->x_multifd_page_count; | |
1116 | } | |
73af8dd8 JQ |
1117 | if (params->has_xbzrle_cache_size) { |
1118 | s->parameters.xbzrle_cache_size = params->xbzrle_cache_size; | |
1119 | xbzrle_cache_resize(params->xbzrle_cache_size, errp); | |
1120 | } | |
85de8323 LL |
1121 | } |
1122 | ||
1bda8b3c | 1123 | void qmp_migrate_set_parameters(MigrateSetParameters *params, Error **errp) |
476c72aa | 1124 | { |
1bda8b3c MA |
1125 | MigrationParameters tmp; |
1126 | ||
01fa5598 MA |
1127 | /* TODO Rewrite "" to null instead */ |
1128 | if (params->has_tls_creds | |
1129 | && params->tls_creds->type == QTYPE_QNULL) { | |
cb3e7f08 | 1130 | qobject_unref(params->tls_creds->u.n); |
01fa5598 MA |
1131 | params->tls_creds->type = QTYPE_QSTRING; |
1132 | params->tls_creds->u.s = strdup(""); | |
1133 | } | |
1134 | /* TODO Rewrite "" to null instead */ | |
1135 | if (params->has_tls_hostname | |
1136 | && params->tls_hostname->type == QTYPE_QNULL) { | |
cb3e7f08 | 1137 | qobject_unref(params->tls_hostname->u.n); |
01fa5598 MA |
1138 | params->tls_hostname->type = QTYPE_QSTRING; |
1139 | params->tls_hostname->u.s = strdup(""); | |
1140 | } | |
1141 | ||
1bda8b3c MA |
1142 | migrate_params_test_apply(params, &tmp); |
1143 | ||
1144 | if (!migrate_params_check(&tmp, errp)) { | |
476c72aa PX |
1145 | /* Invalid parameter */ |
1146 | return; | |
1147 | } | |
1148 | ||
73af8dd8 | 1149 | migrate_params_apply(params, errp); |
476c72aa PX |
1150 | } |
1151 | ||
2594f56d | 1152 | |
4886a1bc DDAG |
1153 | void qmp_migrate_start_postcopy(Error **errp) |
1154 | { | |
1155 | MigrationState *s = migrate_get_current(); | |
1156 | ||
16b0fd32 | 1157 | if (!migrate_postcopy()) { |
a54d340b | 1158 | error_setg(errp, "Enable postcopy with migrate_set_capability before" |
4886a1bc DDAG |
1159 | " the start of migration"); |
1160 | return; | |
1161 | } | |
1162 | ||
1163 | if (s->state == MIGRATION_STATUS_NONE) { | |
1164 | error_setg(errp, "Postcopy must be started after migration has been" | |
1165 | " started"); | |
1166 | return; | |
1167 | } | |
1168 | /* | |
1169 | * we don't error if migration has finished since that would be racy | |
1170 | * with issuing this command. | |
1171 | */ | |
1172 | atomic_set(&s->start_postcopy, true); | |
1173 | } | |
1174 | ||
065e2813 AL |
1175 | /* shared migration helpers */ |
1176 | ||
48781e5b | 1177 | void migrate_set_state(int *state, int old_state, int new_state) |
51cf4c1a | 1178 | { |
a31fedee | 1179 | assert(new_state < MIGRATION_STATUS__MAX); |
48781e5b | 1180 | if (atomic_cmpxchg(state, old_state, new_state) == old_state) { |
a31fedee | 1181 | trace_migrate_set_state(MigrationStatus_str(new_state)); |
b05dc723 | 1182 | migrate_generate_event(new_state); |
51cf4c1a Z |
1183 | } |
1184 | } | |
1185 | ||
4e4a3d3a PX |
1186 | static MigrationCapabilityStatusList *migrate_cap_add( |
1187 | MigrationCapabilityStatusList *list, | |
1188 | MigrationCapability index, | |
1189 | bool state) | |
2833c59b JQ |
1190 | { |
1191 | MigrationCapabilityStatusList *cap; | |
1192 | ||
1193 | cap = g_new0(MigrationCapabilityStatusList, 1); | |
1194 | cap->value = g_new0(MigrationCapabilityStatus, 1); | |
4e4a3d3a PX |
1195 | cap->value->capability = index; |
1196 | cap->value->state = state; | |
1197 | cap->next = list; | |
1198 | ||
1199 | return cap; | |
1200 | } | |
1201 | ||
1202 | void migrate_set_block_enabled(bool value, Error **errp) | |
1203 | { | |
1204 | MigrationCapabilityStatusList *cap; | |
1205 | ||
1206 | cap = migrate_cap_add(NULL, MIGRATION_CAPABILITY_BLOCK, value); | |
2833c59b JQ |
1207 | qmp_migrate_set_capabilities(cap, errp); |
1208 | qapi_free_MigrationCapabilityStatusList(cap); | |
1209 | } | |
1210 | ||
1211 | static void migrate_set_block_incremental(MigrationState *s, bool value) | |
1212 | { | |
1213 | s->parameters.block_incremental = value; | |
1214 | } | |
1215 | ||
1216 | static void block_cleanup_parameters(MigrationState *s) | |
1217 | { | |
1218 | if (s->must_remove_block_options) { | |
1219 | /* setting to false can never fail */ | |
1220 | migrate_set_block_enabled(false, &error_abort); | |
1221 | migrate_set_block_incremental(s, false); | |
1222 | s->must_remove_block_options = false; | |
1223 | } | |
1224 | } | |
1225 | ||
bb1fadc4 | 1226 | static void migrate_fd_cleanup(void *opaque) |
065e2813 | 1227 | { |
bb1fadc4 PB |
1228 | MigrationState *s = opaque; |
1229 | ||
1230 | qemu_bh_delete(s->cleanup_bh); | |
1231 | s->cleanup_bh = NULL; | |
1232 | ||
0ceccd85 PX |
1233 | qemu_savevm_state_cleanup(); |
1234 | ||
89a02a9f | 1235 | if (s->to_dst_file) { |
f986c3d2 JQ |
1236 | Error *local_err = NULL; |
1237 | ||
9013dca5 | 1238 | trace_migrate_fd_cleanup(); |
404a7c05 | 1239 | qemu_mutex_unlock_iothread(); |
1d34e4bf DDAG |
1240 | if (s->migration_thread_running) { |
1241 | qemu_thread_join(&s->thread); | |
1242 | s->migration_thread_running = false; | |
1243 | } | |
404a7c05 PB |
1244 | qemu_mutex_lock_iothread(); |
1245 | ||
f986c3d2 JQ |
1246 | if (multifd_save_cleanup(&local_err) != 0) { |
1247 | error_report_err(local_err); | |
1248 | } | |
89a02a9f HZ |
1249 | qemu_fclose(s->to_dst_file); |
1250 | s->to_dst_file = NULL; | |
065e2813 AL |
1251 | } |
1252 | ||
9ec055ae DDAG |
1253 | assert((s->state != MIGRATION_STATUS_ACTIVE) && |
1254 | (s->state != MIGRATION_STATUS_POSTCOPY_ACTIVE)); | |
7a2c1721 | 1255 | |
94f5a437 | 1256 | if (s->state == MIGRATION_STATUS_CANCELLING) { |
48781e5b | 1257 | migrate_set_state(&s->state, MIGRATION_STATUS_CANCELLING, |
94f5a437 | 1258 | MIGRATION_STATUS_CANCELLED); |
7a2c1721 | 1259 | } |
a3fa1d78 | 1260 | |
87db1a7d JQ |
1261 | if (s->error) { |
1262 | /* It is used on info migrate. We can't free it */ | |
1263 | error_report_err(error_copy(s->error)); | |
1264 | } | |
a3fa1d78 | 1265 | notifier_list_notify(&migration_state_notifiers, s); |
2833c59b | 1266 | block_cleanup_parameters(s); |
065e2813 AL |
1267 | } |
1268 | ||
87db1a7d JQ |
1269 | void migrate_set_error(MigrationState *s, const Error *error) |
1270 | { | |
1271 | qemu_mutex_lock(&s->error_mutex); | |
1272 | if (!s->error) { | |
1273 | s->error = error_copy(error); | |
1274 | } | |
1275 | qemu_mutex_unlock(&s->error_mutex); | |
1276 | } | |
1277 | ||
d59ce6f3 | 1278 | void migrate_fd_error(MigrationState *s, const Error *error) |
065e2813 | 1279 | { |
25174055 | 1280 | trace_migrate_fd_error(error_get_pretty(error)); |
89a02a9f | 1281 | assert(s->to_dst_file == NULL); |
48781e5b HZ |
1282 | migrate_set_state(&s->state, MIGRATION_STATUS_SETUP, |
1283 | MIGRATION_STATUS_FAILED); | |
87db1a7d | 1284 | migrate_set_error(s, error); |
458cf28e JQ |
1285 | } |
1286 | ||
0edda1c4 | 1287 | static void migrate_fd_cancel(MigrationState *s) |
065e2813 | 1288 | { |
6f2b811a | 1289 | int old_state ; |
89a02a9f | 1290 | QEMUFile *f = migrate_get_current()->to_dst_file; |
9013dca5 | 1291 | trace_migrate_fd_cancel(); |
065e2813 | 1292 | |
70b20477 DDAG |
1293 | if (s->rp_state.from_dst_file) { |
1294 | /* shutdown the rp socket, so causing the rp thread to shutdown */ | |
1295 | qemu_file_shutdown(s->rp_state.from_dst_file); | |
1296 | } | |
1297 | ||
6f2b811a Z |
1298 | do { |
1299 | old_state = s->state; | |
f6844b99 | 1300 | if (!migration_is_setup_or_active(old_state)) { |
6f2b811a Z |
1301 | break; |
1302 | } | |
a7b36b48 DDAG |
1303 | /* If the migration is paused, kick it out of the pause */ |
1304 | if (old_state == MIGRATION_STATUS_PRE_SWITCHOVER) { | |
1305 | qemu_sem_post(&s->pause_sem); | |
1306 | } | |
48781e5b | 1307 | migrate_set_state(&s->state, old_state, MIGRATION_STATUS_CANCELLING); |
31194731 | 1308 | } while (s->state != MIGRATION_STATUS_CANCELLING); |
a26ba26e DDAG |
1309 | |
1310 | /* | |
1311 | * If we're unlucky the migration code might be stuck somewhere in a | |
1312 | * send/write while the network has failed and is waiting to timeout; | |
1313 | * if we've got shutdown(2) available then we can force it to quit. | |
1314 | * The outgoing qemu file gets closed in migrate_fd_cleanup that is | |
1315 | * called in a bh, so there is no race against this cancel. | |
1316 | */ | |
31194731 | 1317 | if (s->state == MIGRATION_STATUS_CANCELLING && f) { |
a26ba26e DDAG |
1318 | qemu_file_shutdown(f); |
1319 | } | |
1d2acc31 HZ |
1320 | if (s->state == MIGRATION_STATUS_CANCELLING && s->block_inactive) { |
1321 | Error *local_err = NULL; | |
1322 | ||
1323 | bdrv_invalidate_cache_all(&local_err); | |
1324 | if (local_err) { | |
1325 | error_report_err(local_err); | |
1326 | } else { | |
1327 | s->block_inactive = false; | |
1328 | } | |
1329 | } | |
065e2813 AL |
1330 | } |
1331 | ||
99a0db9b GH |
1332 | void add_migration_state_change_notifier(Notifier *notify) |
1333 | { | |
1334 | notifier_list_add(&migration_state_notifiers, notify); | |
1335 | } | |
1336 | ||
1337 | void remove_migration_state_change_notifier(Notifier *notify) | |
1338 | { | |
31552529 | 1339 | notifier_remove(notify); |
99a0db9b GH |
1340 | } |
1341 | ||
02edd2e7 | 1342 | bool migration_in_setup(MigrationState *s) |
afe2df69 | 1343 | { |
31194731 | 1344 | return s->state == MIGRATION_STATUS_SETUP; |
afe2df69 GH |
1345 | } |
1346 | ||
7073693b | 1347 | bool migration_has_finished(MigrationState *s) |
99a0db9b | 1348 | { |
31194731 | 1349 | return s->state == MIGRATION_STATUS_COMPLETED; |
99a0db9b | 1350 | } |
0edda1c4 | 1351 | |
afe2df69 GH |
1352 | bool migration_has_failed(MigrationState *s) |
1353 | { | |
31194731 HZ |
1354 | return (s->state == MIGRATION_STATUS_CANCELLED || |
1355 | s->state == MIGRATION_STATUS_FAILED); | |
afe2df69 GH |
1356 | } |
1357 | ||
5727309d | 1358 | bool migration_in_postcopy(void) |
9ec055ae | 1359 | { |
5727309d JQ |
1360 | MigrationState *s = migrate_get_current(); |
1361 | ||
9ec055ae DDAG |
1362 | return (s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE); |
1363 | } | |
1364 | ||
b82fc321 DDAG |
1365 | bool migration_in_postcopy_after_devices(MigrationState *s) |
1366 | { | |
5727309d | 1367 | return migration_in_postcopy() && s->postcopy_after_devices; |
b82fc321 DDAG |
1368 | } |
1369 | ||
fab35005 | 1370 | bool migration_is_idle(void) |
fe44dc91 | 1371 | { |
fab35005 | 1372 | MigrationState *s = migrate_get_current(); |
fe44dc91 AA |
1373 | |
1374 | switch (s->state) { | |
1375 | case MIGRATION_STATUS_NONE: | |
1376 | case MIGRATION_STATUS_CANCELLED: | |
1377 | case MIGRATION_STATUS_COMPLETED: | |
1378 | case MIGRATION_STATUS_FAILED: | |
1379 | return true; | |
1380 | case MIGRATION_STATUS_SETUP: | |
1381 | case MIGRATION_STATUS_CANCELLING: | |
1382 | case MIGRATION_STATUS_ACTIVE: | |
1383 | case MIGRATION_STATUS_POSTCOPY_ACTIVE: | |
1384 | case MIGRATION_STATUS_COLO: | |
31e06077 DDAG |
1385 | case MIGRATION_STATUS_PRE_SWITCHOVER: |
1386 | case MIGRATION_STATUS_DEVICE: | |
fe44dc91 AA |
1387 | return false; |
1388 | case MIGRATION_STATUS__MAX: | |
1389 | g_assert_not_reached(); | |
1390 | } | |
1391 | ||
1392 | return false; | |
1393 | } | |
1394 | ||
3e0c8050 | 1395 | void migrate_init(MigrationState *s) |
0edda1c4 | 1396 | { |
389775d1 DDAG |
1397 | /* |
1398 | * Reinitialise all migration state, except | |
1399 | * parameters/capabilities that the user set, and | |
1400 | * locks. | |
1401 | */ | |
1402 | s->bytes_xfer = 0; | |
1403 | s->xfer_limit = 0; | |
1404 | s->cleanup_bh = 0; | |
89a02a9f | 1405 | s->to_dst_file = NULL; |
389775d1 | 1406 | s->state = MIGRATION_STATUS_NONE; |
389775d1 DDAG |
1407 | s->rp_state.from_dst_file = NULL; |
1408 | s->rp_state.error = false; | |
1409 | s->mbps = 0.0; | |
1410 | s->downtime = 0; | |
1411 | s->expected_downtime = 0; | |
389775d1 | 1412 | s->setup_time = 0; |
389775d1 | 1413 | s->start_postcopy = false; |
b82fc321 | 1414 | s->postcopy_after_devices = false; |
389775d1 | 1415 | s->migration_thread_running = false; |
d59ce6f3 DB |
1416 | error_free(s->error); |
1417 | s->error = NULL; | |
389775d1 | 1418 | |
48781e5b | 1419 | migrate_set_state(&s->state, MIGRATION_STATUS_NONE, MIGRATION_STATUS_SETUP); |
0edda1c4 | 1420 | |
4af246a3 PX |
1421 | s->start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); |
1422 | s->total_time = 0; | |
7287cbd4 | 1423 | s->vm_was_running = false; |
b15df1ae PX |
1424 | s->iteration_initial_bytes = 0; |
1425 | s->threshold_size = 0; | |
0edda1c4 | 1426 | } |
cab30143 | 1427 | |
fa2756b7 AL |
1428 | static GSList *migration_blockers; |
1429 | ||
fe44dc91 | 1430 | int migrate_add_blocker(Error *reason, Error **errp) |
fa2756b7 | 1431 | { |
3df663e5 | 1432 | if (migrate_get_current()->only_migratable) { |
b67b8c3a AA |
1433 | error_propagate(errp, error_copy(reason)); |
1434 | error_prepend(errp, "disallowing migration blocker " | |
1435 | "(--only_migratable) for: "); | |
1436 | return -EACCES; | |
1437 | } | |
1438 | ||
fab35005 | 1439 | if (migration_is_idle()) { |
fe44dc91 AA |
1440 | migration_blockers = g_slist_prepend(migration_blockers, reason); |
1441 | return 0; | |
1442 | } | |
1443 | ||
1444 | error_propagate(errp, error_copy(reason)); | |
1445 | error_prepend(errp, "disallowing migration blocker (migration in " | |
1446 | "progress) for: "); | |
1447 | return -EBUSY; | |
fa2756b7 AL |
1448 | } |
1449 | ||
1450 | void migrate_del_blocker(Error *reason) | |
1451 | { | |
1452 | migration_blockers = g_slist_remove(migration_blockers, reason); | |
1453 | } | |
1454 | ||
bf1ae1f4 DDAG |
1455 | void qmp_migrate_incoming(const char *uri, Error **errp) |
1456 | { | |
1457 | Error *local_err = NULL; | |
4debb5f5 | 1458 | static bool once = true; |
bf1ae1f4 DDAG |
1459 | |
1460 | if (!deferred_incoming) { | |
4debb5f5 | 1461 | error_setg(errp, "For use with '-incoming defer'"); |
bf1ae1f4 DDAG |
1462 | return; |
1463 | } | |
4debb5f5 DDAG |
1464 | if (!once) { |
1465 | error_setg(errp, "The incoming migration has already been started"); | |
1466 | } | |
bf1ae1f4 DDAG |
1467 | |
1468 | qemu_start_incoming_migration(uri, &local_err); | |
1469 | ||
1470 | if (local_err) { | |
1471 | error_propagate(errp, local_err); | |
1472 | return; | |
1473 | } | |
1474 | ||
4debb5f5 | 1475 | once = false; |
bf1ae1f4 DDAG |
1476 | } |
1477 | ||
02affd41 PX |
1478 | void qmp_migrate_recover(const char *uri, Error **errp) |
1479 | { | |
1480 | MigrationIncomingState *mis = migration_incoming_get_current(); | |
1481 | ||
1482 | if (mis->state != MIGRATION_STATUS_POSTCOPY_PAUSED) { | |
1483 | error_setg(errp, "Migrate recover can only be run " | |
1484 | "when postcopy is paused."); | |
1485 | return; | |
1486 | } | |
1487 | ||
1488 | if (atomic_cmpxchg(&mis->postcopy_recover_triggered, | |
1489 | false, true) == true) { | |
1490 | error_setg(errp, "Migrate recovery is triggered already"); | |
1491 | return; | |
1492 | } | |
1493 | ||
1494 | /* | |
1495 | * Note that this call will never start a real migration; it will | |
1496 | * only re-setup the migration stream and poke existing migration | |
1497 | * to continue using that newly established channel. | |
1498 | */ | |
1499 | qemu_start_incoming_migration(uri, errp); | |
1500 | } | |
1501 | ||
24f3902b GK |
1502 | bool migration_is_blocked(Error **errp) |
1503 | { | |
1504 | if (qemu_savevm_state_blocked(errp)) { | |
1505 | return true; | |
1506 | } | |
1507 | ||
1508 | if (migration_blockers) { | |
250561e1 | 1509 | error_propagate(errp, error_copy(migration_blockers->data)); |
24f3902b GK |
1510 | return true; |
1511 | } | |
1512 | ||
1513 | return false; | |
1514 | } | |
1515 | ||
d3e35b8f PX |
1516 | /* Returns true if continue to migrate, or false if error detected */ |
1517 | static bool migrate_prepare(MigrationState *s, bool blk, bool blk_inc, | |
1518 | bool resume, Error **errp) | |
cab30143 | 1519 | { |
be7059cd | 1520 | Error *local_err = NULL; |
d3e35b8f PX |
1521 | |
1522 | if (resume) { | |
1523 | if (s->state != MIGRATION_STATUS_POSTCOPY_PAUSED) { | |
1524 | error_setg(errp, "Cannot resume if there is no " | |
1525 | "paused migration"); | |
1526 | return false; | |
1527 | } | |
1528 | /* This is a resume, skip init status */ | |
1529 | return true; | |
1530 | } | |
cab30143 | 1531 | |
f6844b99 | 1532 | if (migration_is_setup_or_active(s->state) || |
0b827d5e HZ |
1533 | s->state == MIGRATION_STATUS_CANCELLING || |
1534 | s->state == MIGRATION_STATUS_COLO) { | |
c6bd8c70 | 1535 | error_setg(errp, QERR_MIGRATION_ACTIVE); |
d3e35b8f | 1536 | return false; |
cab30143 | 1537 | } |
d3e35b8f | 1538 | |
ca99993a DDAG |
1539 | if (runstate_check(RUN_STATE_INMIGRATE)) { |
1540 | error_setg(errp, "Guest is waiting for an incoming migration"); | |
d3e35b8f | 1541 | return false; |
ca99993a DDAG |
1542 | } |
1543 | ||
24f3902b | 1544 | if (migration_is_blocked(errp)) { |
d3e35b8f | 1545 | return false; |
fa2756b7 AL |
1546 | } |
1547 | ||
d3e35b8f | 1548 | if (blk || blk_inc) { |
2833c59b JQ |
1549 | if (migrate_use_block() || migrate_use_block_incremental()) { |
1550 | error_setg(errp, "Command options are incompatible with " | |
1551 | "current migration capabilities"); | |
d3e35b8f | 1552 | return false; |
2833c59b JQ |
1553 | } |
1554 | migrate_set_block_enabled(true, &local_err); | |
1555 | if (local_err) { | |
1556 | error_propagate(errp, local_err); | |
d3e35b8f | 1557 | return false; |
2833c59b JQ |
1558 | } |
1559 | s->must_remove_block_options = true; | |
1560 | } | |
1561 | ||
d3e35b8f | 1562 | if (blk_inc) { |
2833c59b JQ |
1563 | migrate_set_block_incremental(s, true); |
1564 | } | |
1565 | ||
3e0c8050 | 1566 | migrate_init(s); |
cab30143 | 1567 | |
d3e35b8f PX |
1568 | return true; |
1569 | } | |
1570 | ||
1571 | void qmp_migrate(const char *uri, bool has_blk, bool blk, | |
1572 | bool has_inc, bool inc, bool has_detach, bool detach, | |
1573 | bool has_resume, bool resume, Error **errp) | |
1574 | { | |
1575 | Error *local_err = NULL; | |
1576 | MigrationState *s = migrate_get_current(); | |
1577 | const char *p; | |
1578 | ||
1579 | if (!migrate_prepare(s, has_blk && blk, has_inc && inc, | |
1580 | has_resume && resume, errp)) { | |
1581 | /* Error detected, put into errp */ | |
1582 | return; | |
1583 | } | |
1584 | ||
cab30143 | 1585 | if (strstart(uri, "tcp:", &p)) { |
f37afb5a | 1586 | tcp_start_outgoing_migration(s, p, &local_err); |
2da776db | 1587 | #ifdef CONFIG_RDMA |
41310c68 | 1588 | } else if (strstart(uri, "rdma:", &p)) { |
2da776db MH |
1589 | rdma_start_outgoing_migration(s, p, &local_err); |
1590 | #endif | |
cab30143 | 1591 | } else if (strstart(uri, "exec:", &p)) { |
f37afb5a | 1592 | exec_start_outgoing_migration(s, p, &local_err); |
cab30143 | 1593 | } else if (strstart(uri, "unix:", &p)) { |
f37afb5a | 1594 | unix_start_outgoing_migration(s, p, &local_err); |
cab30143 | 1595 | } else if (strstart(uri, "fd:", &p)) { |
f37afb5a | 1596 | fd_start_outgoing_migration(s, p, &local_err); |
99a0db9b | 1597 | } else { |
c6bd8c70 MA |
1598 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "uri", |
1599 | "a valid migration protocol"); | |
48781e5b HZ |
1600 | migrate_set_state(&s->state, MIGRATION_STATUS_SETUP, |
1601 | MIGRATION_STATUS_FAILED); | |
09576e74 | 1602 | block_cleanup_parameters(s); |
e1c37d0e | 1603 | return; |
cab30143 JQ |
1604 | } |
1605 | ||
f37afb5a | 1606 | if (local_err) { |
d59ce6f3 | 1607 | migrate_fd_error(s, local_err); |
f37afb5a | 1608 | error_propagate(errp, local_err); |
e1c37d0e | 1609 | return; |
1299c631 | 1610 | } |
cab30143 JQ |
1611 | } |
1612 | ||
6cdedb07 | 1613 | void qmp_migrate_cancel(Error **errp) |
cab30143 | 1614 | { |
17549e84 | 1615 | migrate_fd_cancel(migrate_get_current()); |
cab30143 JQ |
1616 | } |
1617 | ||
89cfc02c DDAG |
1618 | void qmp_migrate_continue(MigrationStatus state, Error **errp) |
1619 | { | |
1620 | MigrationState *s = migrate_get_current(); | |
1621 | if (s->state != state) { | |
1622 | error_setg(errp, "Migration not in expected state: %s", | |
1623 | MigrationStatus_str(s->state)); | |
1624 | return; | |
1625 | } | |
1626 | qemu_sem_post(&s->pause_sem); | |
1627 | } | |
1628 | ||
9e1ba4cc OW |
1629 | void qmp_migrate_set_cache_size(int64_t value, Error **errp) |
1630 | { | |
73af8dd8 JQ |
1631 | MigrateSetParameters p = { |
1632 | .has_xbzrle_cache_size = true, | |
1633 | .xbzrle_cache_size = value, | |
1634 | }; | |
c91e681a | 1635 | |
73af8dd8 | 1636 | qmp_migrate_set_parameters(&p, errp); |
9e1ba4cc OW |
1637 | } |
1638 | ||
1639 | int64_t qmp_query_migrate_cache_size(Error **errp) | |
1640 | { | |
1641 | return migrate_xbzrle_cache_size(); | |
1642 | } | |
1643 | ||
3dc85383 | 1644 | void qmp_migrate_set_speed(int64_t value, Error **errp) |
cab30143 | 1645 | { |
1bda8b3c | 1646 | MigrateSetParameters p = { |
2ff30257 AA |
1647 | .has_max_bandwidth = true, |
1648 | .max_bandwidth = value, | |
1649 | }; | |
cab30143 | 1650 | |
2ff30257 | 1651 | qmp_migrate_set_parameters(&p, errp); |
cab30143 JQ |
1652 | } |
1653 | ||
4f0a993b | 1654 | void qmp_migrate_set_downtime(double value, Error **errp) |
cab30143 | 1655 | { |
87c9cc1c DHB |
1656 | if (value < 0 || value > MAX_MIGRATE_DOWNTIME_SECONDS) { |
1657 | error_setg(errp, "Parameter 'downtime_limit' expects an integer in " | |
1658 | "the range of 0 to %d seconds", | |
1659 | MAX_MIGRATE_DOWNTIME_SECONDS); | |
1660 | return; | |
1661 | } | |
1662 | ||
2ff30257 AA |
1663 | value *= 1000; /* Convert to milliseconds */ |
1664 | value = MAX(0, MIN(INT64_MAX, value)); | |
1665 | ||
1bda8b3c | 1666 | MigrateSetParameters p = { |
2ff30257 AA |
1667 | .has_downtime_limit = true, |
1668 | .downtime_limit = value, | |
1669 | }; | |
1670 | ||
1671 | qmp_migrate_set_parameters(&p, errp); | |
99a0db9b | 1672 | } |
17ad9b35 | 1673 | |
53f09a10 PB |
1674 | bool migrate_release_ram(void) |
1675 | { | |
1676 | MigrationState *s; | |
1677 | ||
1678 | s = migrate_get_current(); | |
1679 | ||
1680 | return s->enabled_capabilities[MIGRATION_CAPABILITY_RELEASE_RAM]; | |
1681 | } | |
1682 | ||
53dd370c DDAG |
1683 | bool migrate_postcopy_ram(void) |
1684 | { | |
1685 | MigrationState *s; | |
1686 | ||
1687 | s = migrate_get_current(); | |
1688 | ||
32c3db5b | 1689 | return s->enabled_capabilities[MIGRATION_CAPABILITY_POSTCOPY_RAM]; |
53dd370c DDAG |
1690 | } |
1691 | ||
58110f0a VSO |
1692 | bool migrate_postcopy(void) |
1693 | { | |
dd6bb914 | 1694 | return migrate_postcopy_ram() || migrate_dirty_bitmaps(); |
58110f0a VSO |
1695 | } |
1696 | ||
bde1e2ec CV |
1697 | bool migrate_auto_converge(void) |
1698 | { | |
1699 | MigrationState *s; | |
1700 | ||
1701 | s = migrate_get_current(); | |
1702 | ||
1703 | return s->enabled_capabilities[MIGRATION_CAPABILITY_AUTO_CONVERGE]; | |
1704 | } | |
1705 | ||
323004a3 PL |
1706 | bool migrate_zero_blocks(void) |
1707 | { | |
1708 | MigrationState *s; | |
1709 | ||
1710 | s = migrate_get_current(); | |
1711 | ||
1712 | return s->enabled_capabilities[MIGRATION_CAPABILITY_ZERO_BLOCKS]; | |
1713 | } | |
1714 | ||
f22f928e AP |
1715 | bool migrate_postcopy_blocktime(void) |
1716 | { | |
1717 | MigrationState *s; | |
1718 | ||
1719 | s = migrate_get_current(); | |
1720 | ||
1721 | return s->enabled_capabilities[MIGRATION_CAPABILITY_POSTCOPY_BLOCKTIME]; | |
1722 | } | |
1723 | ||
8706d2d5 LL |
1724 | bool migrate_use_compression(void) |
1725 | { | |
dde4e694 LL |
1726 | MigrationState *s; |
1727 | ||
1728 | s = migrate_get_current(); | |
1729 | ||
1730 | return s->enabled_capabilities[MIGRATION_CAPABILITY_COMPRESS]; | |
8706d2d5 LL |
1731 | } |
1732 | ||
1733 | int migrate_compress_level(void) | |
1734 | { | |
1735 | MigrationState *s; | |
1736 | ||
1737 | s = migrate_get_current(); | |
1738 | ||
2594f56d | 1739 | return s->parameters.compress_level; |
8706d2d5 LL |
1740 | } |
1741 | ||
1742 | int migrate_compress_threads(void) | |
1743 | { | |
1744 | MigrationState *s; | |
1745 | ||
1746 | s = migrate_get_current(); | |
1747 | ||
2594f56d | 1748 | return s->parameters.compress_threads; |
8706d2d5 LL |
1749 | } |
1750 | ||
3fcb38c2 LL |
1751 | int migrate_decompress_threads(void) |
1752 | { | |
1753 | MigrationState *s; | |
1754 | ||
1755 | s = migrate_get_current(); | |
1756 | ||
2594f56d | 1757 | return s->parameters.decompress_threads; |
3fcb38c2 LL |
1758 | } |
1759 | ||
55efc8c2 VSO |
1760 | bool migrate_dirty_bitmaps(void) |
1761 | { | |
1762 | MigrationState *s; | |
1763 | ||
1764 | s = migrate_get_current(); | |
1765 | ||
1766 | return s->enabled_capabilities[MIGRATION_CAPABILITY_DIRTY_BITMAPS]; | |
1767 | } | |
1768 | ||
b05dc723 JQ |
1769 | bool migrate_use_events(void) |
1770 | { | |
1771 | MigrationState *s; | |
1772 | ||
1773 | s = migrate_get_current(); | |
1774 | ||
1775 | return s->enabled_capabilities[MIGRATION_CAPABILITY_EVENTS]; | |
1776 | } | |
1777 | ||
30126bbf JQ |
1778 | bool migrate_use_multifd(void) |
1779 | { | |
1780 | MigrationState *s; | |
1781 | ||
1782 | s = migrate_get_current(); | |
1783 | ||
1784 | return s->enabled_capabilities[MIGRATION_CAPABILITY_X_MULTIFD]; | |
1785 | } | |
1786 | ||
93fbd031 DDAG |
1787 | bool migrate_pause_before_switchover(void) |
1788 | { | |
1789 | MigrationState *s; | |
1790 | ||
1791 | s = migrate_get_current(); | |
1792 | ||
1793 | return s->enabled_capabilities[ | |
1794 | MIGRATION_CAPABILITY_PAUSE_BEFORE_SWITCHOVER]; | |
1795 | } | |
1796 | ||
4075fb1c JQ |
1797 | int migrate_multifd_channels(void) |
1798 | { | |
1799 | MigrationState *s; | |
1800 | ||
1801 | s = migrate_get_current(); | |
1802 | ||
1803 | return s->parameters.x_multifd_channels; | |
1804 | } | |
1805 | ||
0fb86605 JQ |
1806 | int migrate_multifd_page_count(void) |
1807 | { | |
1808 | MigrationState *s; | |
1809 | ||
1810 | s = migrate_get_current(); | |
1811 | ||
1812 | return s->parameters.x_multifd_page_count; | |
1813 | } | |
1814 | ||
17ad9b35 OW |
1815 | int migrate_use_xbzrle(void) |
1816 | { | |
1817 | MigrationState *s; | |
1818 | ||
1819 | s = migrate_get_current(); | |
1820 | ||
1821 | return s->enabled_capabilities[MIGRATION_CAPABILITY_XBZRLE]; | |
1822 | } | |
1823 | ||
1824 | int64_t migrate_xbzrle_cache_size(void) | |
1825 | { | |
1826 | MigrationState *s; | |
1827 | ||
1828 | s = migrate_get_current(); | |
1829 | ||
73af8dd8 | 1830 | return s->parameters.xbzrle_cache_size; |
17ad9b35 | 1831 | } |
0d82d0e8 | 1832 | |
2833c59b JQ |
1833 | bool migrate_use_block(void) |
1834 | { | |
1835 | MigrationState *s; | |
1836 | ||
1837 | s = migrate_get_current(); | |
1838 | ||
1839 | return s->enabled_capabilities[MIGRATION_CAPABILITY_BLOCK]; | |
1840 | } | |
1841 | ||
c788ada8 PX |
1842 | bool migrate_use_return_path(void) |
1843 | { | |
1844 | MigrationState *s; | |
1845 | ||
1846 | s = migrate_get_current(); | |
1847 | ||
1848 | return s->enabled_capabilities[MIGRATION_CAPABILITY_RETURN_PATH]; | |
1849 | } | |
1850 | ||
2833c59b JQ |
1851 | bool migrate_use_block_incremental(void) |
1852 | { | |
1853 | MigrationState *s; | |
1854 | ||
1855 | s = migrate_get_current(); | |
1856 | ||
1857 | return s->parameters.block_incremental; | |
1858 | } | |
1859 | ||
70b20477 DDAG |
1860 | /* migration thread support */ |
1861 | /* | |
1862 | * Something bad happened to the RP stream, mark an error | |
1863 | * The caller shall print or trace something to indicate why | |
1864 | */ | |
1865 | static void mark_source_rp_bad(MigrationState *s) | |
1866 | { | |
1867 | s->rp_state.error = true; | |
1868 | } | |
1869 | ||
1870 | static struct rp_cmd_args { | |
1871 | ssize_t len; /* -1 = variable */ | |
1872 | const char *name; | |
1873 | } rp_cmd_args[] = { | |
1874 | [MIG_RP_MSG_INVALID] = { .len = -1, .name = "INVALID" }, | |
1875 | [MIG_RP_MSG_SHUT] = { .len = 4, .name = "SHUT" }, | |
1876 | [MIG_RP_MSG_PONG] = { .len = 4, .name = "PONG" }, | |
1e2d90eb DDAG |
1877 | [MIG_RP_MSG_REQ_PAGES] = { .len = 12, .name = "REQ_PAGES" }, |
1878 | [MIG_RP_MSG_REQ_PAGES_ID] = { .len = -1, .name = "REQ_PAGES_ID" }, | |
a335debb | 1879 | [MIG_RP_MSG_RECV_BITMAP] = { .len = -1, .name = "RECV_BITMAP" }, |
13955b89 | 1880 | [MIG_RP_MSG_RESUME_ACK] = { .len = 4, .name = "RESUME_ACK" }, |
70b20477 DDAG |
1881 | [MIG_RP_MSG_MAX] = { .len = -1, .name = "MAX" }, |
1882 | }; | |
1883 | ||
1e2d90eb DDAG |
1884 | /* |
1885 | * Process a request for pages received on the return path, | |
1886 | * We're allowed to send more than requested (e.g. to round to our page size) | |
1887 | * and we don't need to send pages that have already been sent. | |
1888 | */ | |
1889 | static void migrate_handle_rp_req_pages(MigrationState *ms, const char* rbname, | |
1890 | ram_addr_t start, size_t len) | |
1891 | { | |
6c595cde DDAG |
1892 | long our_host_ps = getpagesize(); |
1893 | ||
1e2d90eb | 1894 | trace_migrate_handle_rp_req_pages(rbname, start, len); |
6c595cde DDAG |
1895 | |
1896 | /* | |
1897 | * Since we currently insist on matching page sizes, just sanity check | |
1898 | * we're being asked for whole host pages. | |
1899 | */ | |
1900 | if (start & (our_host_ps-1) || | |
1901 | (len & (our_host_ps-1))) { | |
1902 | error_report("%s: Misaligned page request, start: " RAM_ADDR_FMT | |
1903 | " len: %zd", __func__, start, len); | |
1904 | mark_source_rp_bad(ms); | |
1905 | return; | |
1906 | } | |
1907 | ||
96506894 | 1908 | if (ram_save_queue_pages(rbname, start, len)) { |
6c595cde DDAG |
1909 | mark_source_rp_bad(ms); |
1910 | } | |
1e2d90eb DDAG |
1911 | } |
1912 | ||
14b1742e PX |
1913 | /* Return true to retry, false to quit */ |
1914 | static bool postcopy_pause_return_path_thread(MigrationState *s) | |
1915 | { | |
1916 | trace_postcopy_pause_return_path(); | |
1917 | ||
1918 | qemu_sem_wait(&s->postcopy_pause_rp_sem); | |
1919 | ||
1920 | trace_postcopy_pause_return_path_continued(); | |
1921 | ||
1922 | return true; | |
1923 | } | |
1924 | ||
a335debb PX |
1925 | static int migrate_handle_rp_recv_bitmap(MigrationState *s, char *block_name) |
1926 | { | |
1927 | RAMBlock *block = qemu_ram_block_by_name(block_name); | |
1928 | ||
1929 | if (!block) { | |
1930 | error_report("%s: invalid block name '%s'", __func__, block_name); | |
1931 | return -EINVAL; | |
1932 | } | |
1933 | ||
1934 | /* Fetch the received bitmap and refresh the dirty bitmap */ | |
1935 | return ram_dirty_bitmap_reload(s, block); | |
1936 | } | |
1937 | ||
13955b89 PX |
1938 | static int migrate_handle_rp_resume_ack(MigrationState *s, uint32_t value) |
1939 | { | |
1940 | trace_source_return_path_thread_resume_ack(value); | |
1941 | ||
1942 | if (value != MIGRATION_RESUME_ACK_VALUE) { | |
1943 | error_report("%s: illegal resume_ack value %"PRIu32, | |
1944 | __func__, value); | |
1945 | return -1; | |
1946 | } | |
1947 | ||
1948 | /* Now both sides are active. */ | |
1949 | migrate_set_state(&s->state, MIGRATION_STATUS_POSTCOPY_RECOVER, | |
1950 | MIGRATION_STATUS_POSTCOPY_ACTIVE); | |
1951 | ||
94190696 PX |
1952 | /* Notify send thread that time to continue send pages */ |
1953 | qemu_sem_post(&s->rp_state.rp_sem); | |
13955b89 PX |
1954 | |
1955 | return 0; | |
1956 | } | |
1957 | ||
70b20477 DDAG |
1958 | /* |
1959 | * Handles messages sent on the return path towards the source VM | |
1960 | * | |
1961 | */ | |
1962 | static void *source_return_path_thread(void *opaque) | |
1963 | { | |
1964 | MigrationState *ms = opaque; | |
1965 | QEMUFile *rp = ms->rp_state.from_dst_file; | |
1966 | uint16_t header_len, header_type; | |
568b01ca | 1967 | uint8_t buf[512]; |
70b20477 | 1968 | uint32_t tmp32, sibling_error; |
1e2d90eb DDAG |
1969 | ram_addr_t start = 0; /* =0 to silence warning */ |
1970 | size_t len = 0, expected_len; | |
70b20477 DDAG |
1971 | int res; |
1972 | ||
1973 | trace_source_return_path_thread_entry(); | |
14b1742e PX |
1974 | |
1975 | retry: | |
70b20477 DDAG |
1976 | while (!ms->rp_state.error && !qemu_file_get_error(rp) && |
1977 | migration_is_setup_or_active(ms->state)) { | |
1978 | trace_source_return_path_thread_loop_top(); | |
1979 | header_type = qemu_get_be16(rp); | |
1980 | header_len = qemu_get_be16(rp); | |
1981 | ||
7a9ddfbf PX |
1982 | if (qemu_file_get_error(rp)) { |
1983 | mark_source_rp_bad(ms); | |
1984 | goto out; | |
1985 | } | |
1986 | ||
70b20477 DDAG |
1987 | if (header_type >= MIG_RP_MSG_MAX || |
1988 | header_type == MIG_RP_MSG_INVALID) { | |
1989 | error_report("RP: Received invalid message 0x%04x length 0x%04x", | |
1990 | header_type, header_len); | |
1991 | mark_source_rp_bad(ms); | |
1992 | goto out; | |
1993 | } | |
1994 | ||
1995 | if ((rp_cmd_args[header_type].len != -1 && | |
1996 | header_len != rp_cmd_args[header_type].len) || | |
568b01ca | 1997 | header_len > sizeof(buf)) { |
70b20477 DDAG |
1998 | error_report("RP: Received '%s' message (0x%04x) with" |
1999 | "incorrect length %d expecting %zu", | |
2000 | rp_cmd_args[header_type].name, header_type, header_len, | |
2001 | (size_t)rp_cmd_args[header_type].len); | |
2002 | mark_source_rp_bad(ms); | |
2003 | goto out; | |
2004 | } | |
2005 | ||
2006 | /* We know we've got a valid header by this point */ | |
2007 | res = qemu_get_buffer(rp, buf, header_len); | |
2008 | if (res != header_len) { | |
2009 | error_report("RP: Failed reading data for message 0x%04x" | |
2010 | " read %d expected %d", | |
2011 | header_type, res, header_len); | |
2012 | mark_source_rp_bad(ms); | |
2013 | goto out; | |
2014 | } | |
2015 | ||
2016 | /* OK, we have the message and the data */ | |
2017 | switch (header_type) { | |
2018 | case MIG_RP_MSG_SHUT: | |
4d885131 | 2019 | sibling_error = ldl_be_p(buf); |
70b20477 DDAG |
2020 | trace_source_return_path_thread_shut(sibling_error); |
2021 | if (sibling_error) { | |
2022 | error_report("RP: Sibling indicated error %d", sibling_error); | |
2023 | mark_source_rp_bad(ms); | |
2024 | } | |
2025 | /* | |
2026 | * We'll let the main thread deal with closing the RP | |
2027 | * we could do a shutdown(2) on it, but we're the only user | |
2028 | * anyway, so there's nothing gained. | |
2029 | */ | |
2030 | goto out; | |
2031 | ||
2032 | case MIG_RP_MSG_PONG: | |
4d885131 | 2033 | tmp32 = ldl_be_p(buf); |
70b20477 DDAG |
2034 | trace_source_return_path_thread_pong(tmp32); |
2035 | break; | |
2036 | ||
1e2d90eb | 2037 | case MIG_RP_MSG_REQ_PAGES: |
4d885131 PM |
2038 | start = ldq_be_p(buf); |
2039 | len = ldl_be_p(buf + 8); | |
1e2d90eb DDAG |
2040 | migrate_handle_rp_req_pages(ms, NULL, start, len); |
2041 | break; | |
2042 | ||
2043 | case MIG_RP_MSG_REQ_PAGES_ID: | |
2044 | expected_len = 12 + 1; /* header + termination */ | |
2045 | ||
2046 | if (header_len >= expected_len) { | |
4d885131 PM |
2047 | start = ldq_be_p(buf); |
2048 | len = ldl_be_p(buf + 8); | |
1e2d90eb DDAG |
2049 | /* Now we expect an idstr */ |
2050 | tmp32 = buf[12]; /* Length of the following idstr */ | |
2051 | buf[13 + tmp32] = '\0'; | |
2052 | expected_len += tmp32; | |
2053 | } | |
2054 | if (header_len != expected_len) { | |
2055 | error_report("RP: Req_Page_id with length %d expecting %zd", | |
2056 | header_len, expected_len); | |
2057 | mark_source_rp_bad(ms); | |
2058 | goto out; | |
2059 | } | |
2060 | migrate_handle_rp_req_pages(ms, (char *)&buf[13], start, len); | |
2061 | break; | |
2062 | ||
a335debb PX |
2063 | case MIG_RP_MSG_RECV_BITMAP: |
2064 | if (header_len < 1) { | |
2065 | error_report("%s: missing block name", __func__); | |
2066 | mark_source_rp_bad(ms); | |
2067 | goto out; | |
2068 | } | |
2069 | /* Format: len (1B) + idstr (<255B). This ends the idstr. */ | |
2070 | buf[buf[0] + 1] = '\0'; | |
2071 | if (migrate_handle_rp_recv_bitmap(ms, (char *)(buf + 1))) { | |
2072 | mark_source_rp_bad(ms); | |
2073 | goto out; | |
2074 | } | |
2075 | break; | |
2076 | ||
13955b89 PX |
2077 | case MIG_RP_MSG_RESUME_ACK: |
2078 | tmp32 = ldl_be_p(buf); | |
2079 | if (migrate_handle_rp_resume_ack(ms, tmp32)) { | |
2080 | mark_source_rp_bad(ms); | |
2081 | goto out; | |
2082 | } | |
2083 | break; | |
2084 | ||
70b20477 DDAG |
2085 | default: |
2086 | break; | |
2087 | } | |
2088 | } | |
14b1742e PX |
2089 | |
2090 | out: | |
2091 | res = qemu_file_get_error(rp); | |
2092 | if (res) { | |
2093 | if (res == -EIO) { | |
2094 | /* | |
2095 | * Maybe there is something we can do: it looks like a | |
2096 | * network down issue, and we pause for a recovery. | |
2097 | */ | |
2098 | if (postcopy_pause_return_path_thread(ms)) { | |
2099 | /* Reload rp, reset the rest */ | |
2100 | rp = ms->rp_state.from_dst_file; | |
2101 | ms->rp_state.error = false; | |
2102 | goto retry; | |
2103 | } | |
2104 | } | |
2105 | ||
70b20477 DDAG |
2106 | trace_source_return_path_thread_bad_end(); |
2107 | mark_source_rp_bad(ms); | |
2108 | } | |
2109 | ||
2110 | trace_source_return_path_thread_end(); | |
70b20477 DDAG |
2111 | ms->rp_state.from_dst_file = NULL; |
2112 | qemu_fclose(rp); | |
2113 | return NULL; | |
2114 | } | |
2115 | ||
d3e35b8f PX |
2116 | static int open_return_path_on_source(MigrationState *ms, |
2117 | bool create_thread) | |
70b20477 DDAG |
2118 | { |
2119 | ||
89a02a9f | 2120 | ms->rp_state.from_dst_file = qemu_file_get_return_path(ms->to_dst_file); |
70b20477 DDAG |
2121 | if (!ms->rp_state.from_dst_file) { |
2122 | return -1; | |
2123 | } | |
2124 | ||
2125 | trace_open_return_path_on_source(); | |
d3e35b8f PX |
2126 | |
2127 | if (!create_thread) { | |
2128 | /* We're done */ | |
2129 | return 0; | |
2130 | } | |
2131 | ||
70b20477 DDAG |
2132 | qemu_thread_create(&ms->rp_state.rp_thread, "return path", |
2133 | source_return_path_thread, ms, QEMU_THREAD_JOINABLE); | |
2134 | ||
2135 | trace_open_return_path_on_source_continue(); | |
2136 | ||
2137 | return 0; | |
2138 | } | |
2139 | ||
70b20477 DDAG |
2140 | /* Returns 0 if the RP was ok, otherwise there was an error on the RP */ |
2141 | static int await_return_path_close_on_source(MigrationState *ms) | |
2142 | { | |
2143 | /* | |
2144 | * If this is a normal exit then the destination will send a SHUT and the | |
2145 | * rp_thread will exit, however if there's an error we need to cause | |
2146 | * it to exit. | |
2147 | */ | |
89a02a9f | 2148 | if (qemu_file_get_error(ms->to_dst_file) && ms->rp_state.from_dst_file) { |
70b20477 DDAG |
2149 | /* |
2150 | * shutdown(2), if we have it, will cause it to unblock if it's stuck | |
2151 | * waiting for the destination. | |
2152 | */ | |
2153 | qemu_file_shutdown(ms->rp_state.from_dst_file); | |
2154 | mark_source_rp_bad(ms); | |
2155 | } | |
2156 | trace_await_return_path_close_on_source_joining(); | |
2157 | qemu_thread_join(&ms->rp_state.rp_thread); | |
2158 | trace_await_return_path_close_on_source_close(); | |
2159 | return ms->rp_state.error; | |
2160 | } | |
2161 | ||
1d34e4bf DDAG |
2162 | /* |
2163 | * Switch from normal iteration to postcopy | |
2164 | * Returns non-0 on error | |
2165 | */ | |
7287cbd4 | 2166 | static int postcopy_start(MigrationState *ms) |
1d34e4bf DDAG |
2167 | { |
2168 | int ret; | |
61b67d47 DB |
2169 | QIOChannelBuffer *bioc; |
2170 | QEMUFile *fb; | |
1d34e4bf | 2171 | int64_t time_at_stop = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); |
ef8d6488 | 2172 | bool restart_block = false; |
0331c8ca DDAG |
2173 | int cur_state = MIGRATION_STATUS_ACTIVE; |
2174 | if (!migrate_pause_before_switchover()) { | |
2175 | migrate_set_state(&ms->state, MIGRATION_STATUS_ACTIVE, | |
2176 | MIGRATION_STATUS_POSTCOPY_ACTIVE); | |
2177 | } | |
1d34e4bf DDAG |
2178 | |
2179 | trace_postcopy_start(); | |
2180 | qemu_mutex_lock_iothread(); | |
2181 | trace_postcopy_start_set_run(); | |
2182 | ||
2183 | qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER); | |
1d34e4bf DDAG |
2184 | global_state_store(); |
2185 | ret = vm_stop_force_state(RUN_STATE_FINISH_MIGRATE); | |
76b1c7fe KW |
2186 | if (ret < 0) { |
2187 | goto fail; | |
2188 | } | |
1d34e4bf | 2189 | |
0331c8ca DDAG |
2190 | ret = migration_maybe_pause(ms, &cur_state, |
2191 | MIGRATION_STATUS_POSTCOPY_ACTIVE); | |
2192 | if (ret < 0) { | |
2193 | goto fail; | |
2194 | } | |
2195 | ||
76b1c7fe | 2196 | ret = bdrv_inactivate_all(); |
1d34e4bf DDAG |
2197 | if (ret < 0) { |
2198 | goto fail; | |
2199 | } | |
ef8d6488 | 2200 | restart_block = true; |
1d34e4bf | 2201 | |
1c0d249d DDAG |
2202 | /* |
2203 | * Cause any non-postcopiable, but iterative devices to | |
2204 | * send out their final data. | |
2205 | */ | |
a1fbe750 | 2206 | qemu_savevm_state_complete_precopy(ms->to_dst_file, true, false); |
1c0d249d | 2207 | |
1d34e4bf DDAG |
2208 | /* |
2209 | * in Finish migrate and with the io-lock held everything should | |
2210 | * be quiet, but we've potentially still got dirty pages and we | |
2211 | * need to tell the destination to throw any pages it's already received | |
2212 | * that are dirty | |
2213 | */ | |
58110f0a VSO |
2214 | if (migrate_postcopy_ram()) { |
2215 | if (ram_postcopy_send_discard_bitmap(ms)) { | |
2216 | error_report("postcopy send discard bitmap failed"); | |
2217 | goto fail; | |
2218 | } | |
1d34e4bf DDAG |
2219 | } |
2220 | ||
2221 | /* | |
2222 | * send rest of state - note things that are doing postcopy | |
2223 | * will notice we're in POSTCOPY_ACTIVE and not actually | |
2224 | * wrap their state up here | |
2225 | */ | |
89a02a9f | 2226 | qemu_file_set_rate_limit(ms->to_dst_file, INT64_MAX); |
58110f0a VSO |
2227 | if (migrate_postcopy_ram()) { |
2228 | /* Ping just for debugging, helps line traces up */ | |
2229 | qemu_savevm_send_ping(ms->to_dst_file, 2); | |
2230 | } | |
1d34e4bf DDAG |
2231 | |
2232 | /* | |
2233 | * While loading the device state we may trigger page transfer | |
2234 | * requests and the fd must be free to process those, and thus | |
2235 | * the destination must read the whole device state off the fd before | |
2236 | * it starts processing it. Unfortunately the ad-hoc migration format | |
2237 | * doesn't allow the destination to know the size to read without fully | |
2238 | * parsing it through each devices load-state code (especially the open | |
2239 | * coded devices that use get/put). | |
2240 | * So we wrap the device state up in a package with a length at the start; | |
2241 | * to do this we use a qemu_buf to hold the whole of the device state. | |
2242 | */ | |
61b67d47 | 2243 | bioc = qio_channel_buffer_new(4096); |
6f01f136 | 2244 | qio_channel_set_name(QIO_CHANNEL(bioc), "migration-postcopy-buffer"); |
61b67d47 DB |
2245 | fb = qemu_fopen_channel_output(QIO_CHANNEL(bioc)); |
2246 | object_unref(OBJECT(bioc)); | |
1d34e4bf | 2247 | |
c76201ab DDAG |
2248 | /* |
2249 | * Make sure the receiver can get incoming pages before we send the rest | |
2250 | * of the state | |
2251 | */ | |
2252 | qemu_savevm_send_postcopy_listen(fb); | |
2253 | ||
a1fbe750 | 2254 | qemu_savevm_state_complete_precopy(fb, false, false); |
58110f0a VSO |
2255 | if (migrate_postcopy_ram()) { |
2256 | qemu_savevm_send_ping(fb, 3); | |
2257 | } | |
1d34e4bf DDAG |
2258 | |
2259 | qemu_savevm_send_postcopy_run(fb); | |
2260 | ||
2261 | /* <><> end of stuff going into the package */ | |
1d34e4bf | 2262 | |
ef8d6488 DDAG |
2263 | /* Last point of recovery; as soon as we send the package the destination |
2264 | * can open devices and potentially start running. | |
2265 | * Lets just check again we've not got any errors. | |
2266 | */ | |
2267 | ret = qemu_file_get_error(ms->to_dst_file); | |
2268 | if (ret) { | |
2269 | error_report("postcopy_start: Migration stream errored (pre package)"); | |
2270 | goto fail_closefb; | |
2271 | } | |
2272 | ||
2273 | restart_block = false; | |
2274 | ||
1d34e4bf | 2275 | /* Now send that blob */ |
61b67d47 | 2276 | if (qemu_savevm_send_packaged(ms->to_dst_file, bioc->data, bioc->usage)) { |
1d34e4bf DDAG |
2277 | goto fail_closefb; |
2278 | } | |
2279 | qemu_fclose(fb); | |
b82fc321 DDAG |
2280 | |
2281 | /* Send a notify to give a chance for anything that needs to happen | |
2282 | * at the transition to postcopy and after the device state; in particular | |
2283 | * spice needs to trigger a transition now | |
2284 | */ | |
2285 | ms->postcopy_after_devices = true; | |
2286 | notifier_list_notify(&migration_state_notifiers, ms); | |
2287 | ||
1d34e4bf DDAG |
2288 | ms->downtime = qemu_clock_get_ms(QEMU_CLOCK_REALTIME) - time_at_stop; |
2289 | ||
2290 | qemu_mutex_unlock_iothread(); | |
2291 | ||
58110f0a VSO |
2292 | if (migrate_postcopy_ram()) { |
2293 | /* | |
2294 | * Although this ping is just for debug, it could potentially be | |
2295 | * used for getting a better measurement of downtime at the source. | |
2296 | */ | |
2297 | qemu_savevm_send_ping(ms->to_dst_file, 4); | |
2298 | } | |
1d34e4bf | 2299 | |
ced1c616 PB |
2300 | if (migrate_release_ram()) { |
2301 | ram_postcopy_migrated_memory_release(ms); | |
2302 | } | |
2303 | ||
89a02a9f | 2304 | ret = qemu_file_get_error(ms->to_dst_file); |
1d34e4bf DDAG |
2305 | if (ret) { |
2306 | error_report("postcopy_start: Migration stream errored"); | |
48781e5b | 2307 | migrate_set_state(&ms->state, MIGRATION_STATUS_POSTCOPY_ACTIVE, |
1d34e4bf DDAG |
2308 | MIGRATION_STATUS_FAILED); |
2309 | } | |
2310 | ||
2311 | return ret; | |
2312 | ||
2313 | fail_closefb: | |
2314 | qemu_fclose(fb); | |
2315 | fail: | |
48781e5b | 2316 | migrate_set_state(&ms->state, MIGRATION_STATUS_POSTCOPY_ACTIVE, |
1d34e4bf | 2317 | MIGRATION_STATUS_FAILED); |
ef8d6488 DDAG |
2318 | if (restart_block) { |
2319 | /* A failure happened early enough that we know the destination hasn't | |
2320 | * accessed block devices, so we're safe to recover. | |
2321 | */ | |
2322 | Error *local_err = NULL; | |
2323 | ||
2324 | bdrv_invalidate_cache_all(&local_err); | |
2325 | if (local_err) { | |
2326 | error_report_err(local_err); | |
2327 | } | |
2328 | } | |
1d34e4bf DDAG |
2329 | qemu_mutex_unlock_iothread(); |
2330 | return -1; | |
2331 | } | |
2332 | ||
e91d8951 DDAG |
2333 | /** |
2334 | * migration_maybe_pause: Pause if required to by | |
2335 | * migrate_pause_before_switchover called with the iothread locked | |
2336 | * Returns: 0 on success | |
2337 | */ | |
0331c8ca DDAG |
2338 | static int migration_maybe_pause(MigrationState *s, |
2339 | int *current_active_state, | |
2340 | int new_state) | |
e91d8951 DDAG |
2341 | { |
2342 | if (!migrate_pause_before_switchover()) { | |
2343 | return 0; | |
2344 | } | |
2345 | ||
2346 | /* Since leaving this state is not atomic with posting the semaphore | |
2347 | * it's possible that someone could have issued multiple migrate_continue | |
2348 | * and the semaphore is incorrectly positive at this point; | |
2349 | * the docs say it's undefined to reinit a semaphore that's already | |
2350 | * init'd, so use timedwait to eat up any existing posts. | |
2351 | */ | |
2352 | while (qemu_sem_timedwait(&s->pause_sem, 1) == 0) { | |
2353 | /* This block intentionally left blank */ | |
2354 | } | |
2355 | ||
2356 | qemu_mutex_unlock_iothread(); | |
2357 | migrate_set_state(&s->state, *current_active_state, | |
2358 | MIGRATION_STATUS_PRE_SWITCHOVER); | |
2359 | qemu_sem_wait(&s->pause_sem); | |
2360 | migrate_set_state(&s->state, MIGRATION_STATUS_PRE_SWITCHOVER, | |
0331c8ca DDAG |
2361 | new_state); |
2362 | *current_active_state = new_state; | |
e91d8951 DDAG |
2363 | qemu_mutex_lock_iothread(); |
2364 | ||
0331c8ca | 2365 | return s->state == new_state ? 0 : -EINVAL; |
e91d8951 DDAG |
2366 | } |
2367 | ||
09f6c85e DDAG |
2368 | /** |
2369 | * migration_completion: Used by migration_thread when there's not much left. | |
2370 | * The caller 'breaks' the loop when this returns. | |
2371 | * | |
2372 | * @s: Current migration state | |
09f6c85e | 2373 | */ |
2ad87305 | 2374 | static void migration_completion(MigrationState *s) |
09f6c85e DDAG |
2375 | { |
2376 | int ret; | |
2ad87305 | 2377 | int current_active_state = s->state; |
09f6c85e | 2378 | |
b10ac0c4 DDAG |
2379 | if (s->state == MIGRATION_STATUS_ACTIVE) { |
2380 | qemu_mutex_lock_iothread(); | |
64909f97 | 2381 | s->downtime_start = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); |
b10ac0c4 | 2382 | qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER); |
7287cbd4 | 2383 | s->vm_was_running = runstate_is_running(); |
b10ac0c4 DDAG |
2384 | ret = global_state_store(); |
2385 | ||
2386 | if (!ret) { | |
a1fbe750 | 2387 | bool inactivate = !migrate_colo_enabled(); |
b10ac0c4 | 2388 | ret = vm_stop_force_state(RUN_STATE_FINISH_MIGRATE); |
e91d8951 | 2389 | if (ret >= 0) { |
0331c8ca DDAG |
2390 | ret = migration_maybe_pause(s, ¤t_active_state, |
2391 | MIGRATION_STATUS_DEVICE); | |
e91d8951 | 2392 | } |
f07fa4cb KW |
2393 | if (ret >= 0) { |
2394 | qemu_file_set_rate_limit(s->to_dst_file, INT64_MAX); | |
a1fbe750 FZ |
2395 | ret = qemu_savevm_state_complete_precopy(s->to_dst_file, false, |
2396 | inactivate); | |
f07fa4cb | 2397 | } |
a1fbe750 FZ |
2398 | if (inactivate && ret >= 0) { |
2399 | s->block_inactive = true; | |
b10ac0c4 DDAG |
2400 | } |
2401 | } | |
2402 | qemu_mutex_unlock_iothread(); | |
09f6c85e | 2403 | |
b10ac0c4 DDAG |
2404 | if (ret < 0) { |
2405 | goto fail; | |
09f6c85e | 2406 | } |
b10ac0c4 DDAG |
2407 | } else if (s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) { |
2408 | trace_migration_completion_postcopy_end(); | |
2409 | ||
89a02a9f | 2410 | qemu_savevm_state_complete_postcopy(s->to_dst_file); |
b10ac0c4 | 2411 | trace_migration_completion_postcopy_end_after_complete(); |
09f6c85e | 2412 | } |
09f6c85e | 2413 | |
b10ac0c4 DDAG |
2414 | /* |
2415 | * If rp was opened we must clean up the thread before | |
2416 | * cleaning everything else up (since if there are no failures | |
2417 | * it will wait for the destination to send it's status in | |
2418 | * a SHUT command). | |
b10ac0c4 | 2419 | */ |
0425dc97 | 2420 | if (s->rp_state.from_dst_file) { |
b10ac0c4 | 2421 | int rp_error; |
0425dc97 | 2422 | trace_migration_return_path_end_before(); |
b10ac0c4 | 2423 | rp_error = await_return_path_close_on_source(s); |
0425dc97 | 2424 | trace_migration_return_path_end_after(rp_error); |
b10ac0c4 | 2425 | if (rp_error) { |
fe904ea8 | 2426 | goto fail_invalidate; |
b10ac0c4 | 2427 | } |
09f6c85e DDAG |
2428 | } |
2429 | ||
89a02a9f | 2430 | if (qemu_file_get_error(s->to_dst_file)) { |
09f6c85e | 2431 | trace_migration_completion_file_err(); |
fe904ea8 | 2432 | goto fail_invalidate; |
09f6c85e DDAG |
2433 | } |
2434 | ||
0b827d5e HZ |
2435 | if (!migrate_colo_enabled()) { |
2436 | migrate_set_state(&s->state, current_active_state, | |
2437 | MIGRATION_STATUS_COMPLETED); | |
2438 | } | |
2439 | ||
09f6c85e DDAG |
2440 | return; |
2441 | ||
fe904ea8 GK |
2442 | fail_invalidate: |
2443 | /* If not doing postcopy, vm_start() will be called: let's regain | |
2444 | * control on images. | |
2445 | */ | |
6039dd5b DDAG |
2446 | if (s->state == MIGRATION_STATUS_ACTIVE || |
2447 | s->state == MIGRATION_STATUS_DEVICE) { | |
fe904ea8 GK |
2448 | Error *local_err = NULL; |
2449 | ||
1d2acc31 | 2450 | qemu_mutex_lock_iothread(); |
fe904ea8 GK |
2451 | bdrv_invalidate_cache_all(&local_err); |
2452 | if (local_err) { | |
2453 | error_report_err(local_err); | |
1d2acc31 HZ |
2454 | } else { |
2455 | s->block_inactive = false; | |
fe904ea8 | 2456 | } |
1d2acc31 | 2457 | qemu_mutex_unlock_iothread(); |
fe904ea8 GK |
2458 | } |
2459 | ||
09f6c85e | 2460 | fail: |
48781e5b HZ |
2461 | migrate_set_state(&s->state, current_active_state, |
2462 | MIGRATION_STATUS_FAILED); | |
09f6c85e DDAG |
2463 | } |
2464 | ||
35a6ed4f HZ |
2465 | bool migrate_colo_enabled(void) |
2466 | { | |
2467 | MigrationState *s = migrate_get_current(); | |
2468 | return s->enabled_capabilities[MIGRATION_CAPABILITY_X_COLO]; | |
2469 | } | |
2470 | ||
b23c2ade PX |
2471 | typedef enum MigThrError { |
2472 | /* No error detected */ | |
2473 | MIG_THR_ERR_NONE = 0, | |
2474 | /* Detected error, but resumed successfully */ | |
2475 | MIG_THR_ERR_RECOVERED = 1, | |
2476 | /* Detected fatal error, need to exit */ | |
2477 | MIG_THR_ERR_FATAL = 2, | |
2478 | } MigThrError; | |
2479 | ||
94190696 PX |
2480 | static int postcopy_resume_handshake(MigrationState *s) |
2481 | { | |
2482 | qemu_savevm_send_postcopy_resume(s->to_dst_file); | |
2483 | ||
2484 | while (s->state == MIGRATION_STATUS_POSTCOPY_RECOVER) { | |
2485 | qemu_sem_wait(&s->rp_state.rp_sem); | |
2486 | } | |
2487 | ||
2488 | if (s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) { | |
2489 | return 0; | |
2490 | } | |
2491 | ||
2492 | return -1; | |
2493 | } | |
2494 | ||
135b87b4 PX |
2495 | /* Return zero if success, or <0 for error */ |
2496 | static int postcopy_do_resume(MigrationState *s) | |
2497 | { | |
d1b8eadb PX |
2498 | int ret; |
2499 | ||
2500 | /* | |
2501 | * Call all the resume_prepare() hooks, so that modules can be | |
2502 | * ready for the migration resume. | |
2503 | */ | |
2504 | ret = qemu_savevm_state_resume_prepare(s); | |
2505 | if (ret) { | |
2506 | error_report("%s: resume_prepare() failure detected: %d", | |
2507 | __func__, ret); | |
2508 | return ret; | |
2509 | } | |
2510 | ||
2511 | /* | |
94190696 PX |
2512 | * Last handshake with destination on the resume (destination will |
2513 | * switch to postcopy-active afterwards) | |
d1b8eadb | 2514 | */ |
94190696 PX |
2515 | ret = postcopy_resume_handshake(s); |
2516 | if (ret) { | |
2517 | error_report("%s: handshake failed: %d", __func__, ret); | |
2518 | return ret; | |
2519 | } | |
d1b8eadb | 2520 | |
135b87b4 PX |
2521 | return 0; |
2522 | } | |
2523 | ||
b23c2ade PX |
2524 | /* |
2525 | * We don't return until we are in a safe state to continue current | |
2526 | * postcopy migration. Returns MIG_THR_ERR_RECOVERED if recovered, or | |
2527 | * MIG_THR_ERR_FATAL if unrecovery failure happened. | |
2528 | */ | |
2529 | static MigThrError postcopy_pause(MigrationState *s) | |
2530 | { | |
2531 | assert(s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE); | |
b23c2ade | 2532 | |
135b87b4 PX |
2533 | while (true) { |
2534 | migrate_set_state(&s->state, s->state, | |
2535 | MIGRATION_STATUS_POSTCOPY_PAUSED); | |
b23c2ade | 2536 | |
135b87b4 PX |
2537 | /* Current channel is possibly broken. Release it. */ |
2538 | assert(s->to_dst_file); | |
2539 | qemu_file_shutdown(s->to_dst_file); | |
2540 | qemu_fclose(s->to_dst_file); | |
2541 | s->to_dst_file = NULL; | |
b23c2ade | 2542 | |
135b87b4 PX |
2543 | error_report("Detected IO failure for postcopy. " |
2544 | "Migration paused."); | |
b23c2ade | 2545 | |
135b87b4 PX |
2546 | /* |
2547 | * We wait until things fixed up. Then someone will setup the | |
2548 | * status back for us. | |
2549 | */ | |
2550 | while (s->state == MIGRATION_STATUS_POSTCOPY_PAUSED) { | |
2551 | qemu_sem_wait(&s->postcopy_pause_sem); | |
2552 | } | |
2553 | ||
2554 | if (s->state == MIGRATION_STATUS_POSTCOPY_RECOVER) { | |
2555 | /* Woken up by a recover procedure. Give it a shot */ | |
b23c2ade | 2556 | |
135b87b4 PX |
2557 | /* |
2558 | * Firstly, let's wake up the return path now, with a new | |
2559 | * return path channel. | |
2560 | */ | |
2561 | qemu_sem_post(&s->postcopy_pause_rp_sem); | |
2562 | ||
2563 | /* Do the resume logic */ | |
2564 | if (postcopy_do_resume(s) == 0) { | |
2565 | /* Let's continue! */ | |
2566 | trace_postcopy_pause_continued(); | |
2567 | return MIG_THR_ERR_RECOVERED; | |
2568 | } else { | |
2569 | /* | |
2570 | * Something wrong happened during the recovery, let's | |
2571 | * pause again. Pause is always better than throwing | |
2572 | * data away. | |
2573 | */ | |
2574 | continue; | |
2575 | } | |
2576 | } else { | |
2577 | /* This is not right... Time to quit. */ | |
2578 | return MIG_THR_ERR_FATAL; | |
2579 | } | |
2580 | } | |
b23c2ade PX |
2581 | } |
2582 | ||
2583 | static MigThrError migration_detect_error(MigrationState *s) | |
2584 | { | |
2585 | int ret; | |
2586 | ||
2587 | /* Try to detect any file errors */ | |
2588 | ret = qemu_file_get_error(s->to_dst_file); | |
2589 | ||
2590 | if (!ret) { | |
2591 | /* Everything is fine */ | |
2592 | return MIG_THR_ERR_NONE; | |
2593 | } | |
2594 | ||
2595 | if (s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE && ret == -EIO) { | |
2596 | /* | |
2597 | * For postcopy, we allow the network to be down for a | |
2598 | * while. After that, it can be continued by a | |
2599 | * recovery phase. | |
2600 | */ | |
2601 | return postcopy_pause(s); | |
2602 | } else { | |
2603 | /* | |
2604 | * For precopy (or postcopy with error outside IO), we fail | |
2605 | * with no time. | |
2606 | */ | |
2607 | migrate_set_state(&s->state, s->state, MIGRATION_STATUS_FAILED); | |
2608 | trace_migration_thread_file_err(); | |
2609 | ||
2610 | /* Time to stop the migration, now. */ | |
2611 | return MIG_THR_ERR_FATAL; | |
2612 | } | |
2613 | } | |
2614 | ||
cf011f08 PX |
2615 | static void migration_calculate_complete(MigrationState *s) |
2616 | { | |
2617 | uint64_t bytes = qemu_ftell(s->to_dst_file); | |
2618 | int64_t end_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); | |
2619 | ||
2620 | s->total_time = end_time - s->start_time; | |
2621 | if (!s->downtime) { | |
2622 | /* | |
2623 | * It's still not set, so we are precopy migration. For | |
2624 | * postcopy, downtime is calculated during postcopy_start(). | |
2625 | */ | |
2626 | s->downtime = end_time - s->downtime_start; | |
2627 | } | |
2628 | ||
2629 | if (s->total_time) { | |
2630 | s->mbps = ((double) bytes * 8.0) / s->total_time / 1000; | |
2631 | } | |
2632 | } | |
2633 | ||
b15df1ae PX |
2634 | static void migration_update_counters(MigrationState *s, |
2635 | int64_t current_time) | |
2636 | { | |
2637 | uint64_t transferred, time_spent; | |
b15df1ae PX |
2638 | double bandwidth; |
2639 | ||
2640 | if (current_time < s->iteration_start_time + BUFFER_DELAY) { | |
2641 | return; | |
2642 | } | |
2643 | ||
2644 | transferred = qemu_ftell(s->to_dst_file) - s->iteration_initial_bytes; | |
2645 | time_spent = current_time - s->iteration_start_time; | |
2646 | bandwidth = (double)transferred / time_spent; | |
0781c1ed | 2647 | s->threshold_size = bandwidth * s->parameters.downtime_limit; |
b15df1ae PX |
2648 | |
2649 | s->mbps = (((double) transferred * 8.0) / | |
2650 | ((double) time_spent / 1000.0)) / 1000.0 / 1000.0; | |
2651 | ||
2652 | /* | |
2653 | * if we haven't sent anything, we don't want to | |
2654 | * recalculate. 10000 is a small enough number for our purposes | |
2655 | */ | |
2656 | if (ram_counters.dirty_pages_rate && transferred > 10000) { | |
2657 | s->expected_downtime = ram_counters.dirty_pages_rate * | |
2658 | qemu_target_page_size() / bandwidth; | |
2659 | } | |
2660 | ||
2661 | qemu_file_reset_rate_limit(s->to_dst_file); | |
2662 | ||
2663 | s->iteration_start_time = current_time; | |
2664 | s->iteration_initial_bytes = qemu_ftell(s->to_dst_file); | |
2665 | ||
2666 | trace_migrate_transferred(transferred, time_spent, | |
0781c1ed | 2667 | bandwidth, s->threshold_size); |
b15df1ae PX |
2668 | } |
2669 | ||
2ad87305 PX |
2670 | /* Migration thread iteration status */ |
2671 | typedef enum { | |
2672 | MIG_ITERATE_RESUME, /* Resume current iteration */ | |
2673 | MIG_ITERATE_SKIP, /* Skip current iteration */ | |
2674 | MIG_ITERATE_BREAK, /* Break the loop */ | |
2675 | } MigIterateState; | |
2676 | ||
2677 | /* | |
2678 | * Return true if continue to the next iteration directly, false | |
2679 | * otherwise. | |
2680 | */ | |
2681 | static MigIterateState migration_iteration_run(MigrationState *s) | |
2682 | { | |
47995026 | 2683 | uint64_t pending_size, pend_pre, pend_compat, pend_post; |
2ad87305 PX |
2684 | bool in_postcopy = s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE; |
2685 | ||
47995026 VSO |
2686 | qemu_savevm_state_pending(s->to_dst_file, s->threshold_size, &pend_pre, |
2687 | &pend_compat, &pend_post); | |
2688 | pending_size = pend_pre + pend_compat + pend_post; | |
2ad87305 PX |
2689 | |
2690 | trace_migrate_pending(pending_size, s->threshold_size, | |
47995026 | 2691 | pend_pre, pend_compat, pend_post); |
2ad87305 PX |
2692 | |
2693 | if (pending_size && pending_size >= s->threshold_size) { | |
2694 | /* Still a significant amount to transfer */ | |
2695 | if (migrate_postcopy() && !in_postcopy && | |
47995026 | 2696 | pend_pre <= s->threshold_size && |
2ad87305 PX |
2697 | atomic_read(&s->start_postcopy)) { |
2698 | if (postcopy_start(s)) { | |
2699 | error_report("%s: postcopy failed to start", __func__); | |
2700 | } | |
2701 | return MIG_ITERATE_SKIP; | |
2702 | } | |
2703 | /* Just another iteration step */ | |
2704 | qemu_savevm_state_iterate(s->to_dst_file, | |
2705 | s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE); | |
2706 | } else { | |
2707 | trace_migration_thread_low_pending(pending_size); | |
2708 | migration_completion(s); | |
2709 | return MIG_ITERATE_BREAK; | |
2710 | } | |
2711 | ||
2712 | return MIG_ITERATE_RESUME; | |
2713 | } | |
2714 | ||
199aa6d4 PX |
2715 | static void migration_iteration_finish(MigrationState *s) |
2716 | { | |
2717 | /* If we enabled cpu throttling for auto-converge, turn it off. */ | |
2718 | cpu_throttle_stop(); | |
2719 | ||
2720 | qemu_mutex_lock_iothread(); | |
2721 | switch (s->state) { | |
2722 | case MIGRATION_STATUS_COMPLETED: | |
2723 | migration_calculate_complete(s); | |
2724 | runstate_set(RUN_STATE_POSTMIGRATE); | |
2725 | break; | |
2726 | ||
2727 | case MIGRATION_STATUS_ACTIVE: | |
2728 | /* | |
2729 | * We should really assert here, but since it's during | |
2730 | * migration, let's try to reduce the usage of assertions. | |
2731 | */ | |
2732 | if (!migrate_colo_enabled()) { | |
2733 | error_report("%s: critical error: calling COLO code without " | |
2734 | "COLO enabled", __func__); | |
2735 | } | |
2736 | migrate_start_colo_process(s); | |
2737 | /* | |
2738 | * Fixme: we will run VM in COLO no matter its old running state. | |
2739 | * After exited COLO, we will keep running. | |
2740 | */ | |
2741 | s->vm_was_running = true; | |
2742 | /* Fallthrough */ | |
2743 | case MIGRATION_STATUS_FAILED: | |
2744 | case MIGRATION_STATUS_CANCELLED: | |
2745 | if (s->vm_was_running) { | |
2746 | vm_start(); | |
2747 | } else { | |
2748 | if (runstate_check(RUN_STATE_FINISH_MIGRATE)) { | |
2749 | runstate_set(RUN_STATE_POSTMIGRATE); | |
2750 | } | |
2751 | } | |
2752 | break; | |
2753 | ||
2754 | default: | |
2755 | /* Should not reach here, but if so, forgive the VM. */ | |
2756 | error_report("%s: Unknown ending state %d", __func__, s->state); | |
2757 | break; | |
2758 | } | |
2759 | qemu_bh_schedule(s->cleanup_bh); | |
2760 | qemu_mutex_unlock_iothread(); | |
2761 | } | |
2762 | ||
70b20477 DDAG |
2763 | /* |
2764 | * Master migration thread on the source VM. | |
2765 | * It drives the migration and pumps the data down the outgoing channel. | |
2766 | */ | |
5f496a1b | 2767 | static void *migration_thread(void *opaque) |
0d82d0e8 | 2768 | { |
9848a404 | 2769 | MigrationState *s = opaque; |
bc72ad67 | 2770 | int64_t setup_start = qemu_clock_get_ms(QEMU_CLOCK_HOST); |
b23c2ade | 2771 | MigThrError thr_error; |
76f5933a | 2772 | |
ab28bd23 PB |
2773 | rcu_register_thread(); |
2774 | ||
b15df1ae PX |
2775 | s->iteration_start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); |
2776 | ||
89a02a9f | 2777 | qemu_savevm_state_header(s->to_dst_file); |
1d34e4bf | 2778 | |
62a02658 PX |
2779 | /* |
2780 | * If we opened the return path, we need to make sure dst has it | |
2781 | * opened as well. | |
2782 | */ | |
2783 | if (s->rp_state.from_dst_file) { | |
1d34e4bf | 2784 | /* Now tell the dest that it should open its end so it can reply */ |
89a02a9f | 2785 | qemu_savevm_send_open_return_path(s->to_dst_file); |
1d34e4bf DDAG |
2786 | |
2787 | /* And do a ping that will make stuff easier to debug */ | |
89a02a9f | 2788 | qemu_savevm_send_ping(s->to_dst_file, 1); |
0425dc97 | 2789 | } |
1d34e4bf | 2790 | |
58110f0a | 2791 | if (migrate_postcopy()) { |
1d34e4bf DDAG |
2792 | /* |
2793 | * Tell the destination that we *might* want to do postcopy later; | |
2794 | * if the other end can't do postcopy it should fail now, nice and | |
2795 | * early. | |
2796 | */ | |
89a02a9f | 2797 | qemu_savevm_send_postcopy_advise(s->to_dst_file); |
1d34e4bf DDAG |
2798 | } |
2799 | ||
9907e842 | 2800 | qemu_savevm_state_setup(s->to_dst_file); |
0d82d0e8 | 2801 | |
bc72ad67 | 2802 | s->setup_time = qemu_clock_get_ms(QEMU_CLOCK_HOST) - setup_start; |
48781e5b HZ |
2803 | migrate_set_state(&s->state, MIGRATION_STATUS_SETUP, |
2804 | MIGRATION_STATUS_ACTIVE); | |
29ae8a41 | 2805 | |
9ec055ae DDAG |
2806 | trace_migration_thread_setup_complete(); |
2807 | ||
2808 | while (s->state == MIGRATION_STATUS_ACTIVE || | |
2809 | s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) { | |
a3e879cd | 2810 | int64_t current_time; |
0d82d0e8 | 2811 | |
89a02a9f | 2812 | if (!qemu_file_rate_limit(s->to_dst_file)) { |
2ad87305 PX |
2813 | MigIterateState iter_state = migration_iteration_run(s); |
2814 | if (iter_state == MIG_ITERATE_SKIP) { | |
2815 | continue; | |
2816 | } else if (iter_state == MIG_ITERATE_BREAK) { | |
09f6c85e | 2817 | break; |
c369f40d JQ |
2818 | } |
2819 | } | |
f4410a5d | 2820 | |
b23c2ade PX |
2821 | /* |
2822 | * Try to detect any kind of failures, and see whether we | |
2823 | * should stop the migration now. | |
2824 | */ | |
2825 | thr_error = migration_detect_error(s); | |
2826 | if (thr_error == MIG_THR_ERR_FATAL) { | |
2827 | /* Stop migration */ | |
fd45ee2c | 2828 | break; |
b23c2ade PX |
2829 | } else if (thr_error == MIG_THR_ERR_RECOVERED) { |
2830 | /* | |
2831 | * Just recovered from a e.g. network failure, reset all | |
2832 | * the local variables. This is important to avoid | |
2833 | * breaking transferred_bytes and bandwidth calculation | |
2834 | */ | |
2835 | s->iteration_start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); | |
2836 | s->iteration_initial_bytes = 0; | |
fd45ee2c | 2837 | } |
b15df1ae | 2838 | |
bc72ad67 | 2839 | current_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); |
0d82d0e8 | 2840 | |
b15df1ae PX |
2841 | migration_update_counters(s, current_time); |
2842 | ||
89a02a9f | 2843 | if (qemu_file_rate_limit(s->to_dst_file)) { |
0d82d0e8 | 2844 | /* usleep expects microseconds */ |
b15df1ae PX |
2845 | g_usleep((s->iteration_start_time + BUFFER_DELAY - |
2846 | current_time) * 1000); | |
0d82d0e8 | 2847 | } |
a3fa1d78 PB |
2848 | } |
2849 | ||
1d34e4bf | 2850 | trace_migration_thread_after_loop(); |
199aa6d4 | 2851 | migration_iteration_finish(s); |
ab28bd23 | 2852 | rcu_unregister_thread(); |
0d82d0e8 JQ |
2853 | return NULL; |
2854 | } | |
2855 | ||
cce8040b | 2856 | void migrate_fd_connect(MigrationState *s, Error *error_in) |
0d82d0e8 | 2857 | { |
d3e35b8f PX |
2858 | int64_t rate_limit; |
2859 | bool resume = s->state == MIGRATION_STATUS_POSTCOPY_PAUSED; | |
2860 | ||
2ff30257 | 2861 | s->expected_downtime = s->parameters.downtime_limit; |
bb1fadc4 | 2862 | s->cleanup_bh = qemu_bh_new(migrate_fd_cleanup, s); |
cce8040b DDAG |
2863 | if (error_in) { |
2864 | migrate_fd_error(s, error_in); | |
2865 | migrate_fd_cleanup(s); | |
2866 | return; | |
2867 | } | |
0d82d0e8 | 2868 | |
d3e35b8f PX |
2869 | if (resume) { |
2870 | /* This is a resumed migration */ | |
2871 | rate_limit = INT64_MAX; | |
2872 | } else { | |
2873 | /* This is a fresh new migration */ | |
2874 | rate_limit = s->parameters.max_bandwidth / XFER_LIMIT_RATIO; | |
2875 | s->expected_downtime = s->parameters.downtime_limit; | |
2876 | s->cleanup_bh = qemu_bh_new(migrate_fd_cleanup, s); | |
442773ce | 2877 | |
d3e35b8f PX |
2878 | /* Notify before starting migration thread */ |
2879 | notifier_list_notify(&migration_state_notifiers, s); | |
2880 | } | |
2881 | ||
2882 | qemu_file_set_rate_limit(s->to_dst_file, rate_limit); | |
2883 | qemu_file_set_blocking(s->to_dst_file, true); | |
9287ac27 | 2884 | |
1d34e4bf | 2885 | /* |
c788ada8 PX |
2886 | * Open the return path. For postcopy, it is used exclusively. For |
2887 | * precopy, only if user specified "return-path" capability would | |
2888 | * QEMU uses the return path. | |
1d34e4bf | 2889 | */ |
c788ada8 | 2890 | if (migrate_postcopy_ram() || migrate_use_return_path()) { |
d3e35b8f | 2891 | if (open_return_path_on_source(s, !resume)) { |
1d34e4bf | 2892 | error_report("Unable to open return-path for postcopy"); |
d3e35b8f | 2893 | migrate_set_state(&s->state, s->state, MIGRATION_STATUS_FAILED); |
1d34e4bf DDAG |
2894 | migrate_fd_cleanup(s); |
2895 | return; | |
2896 | } | |
2897 | } | |
2898 | ||
d3e35b8f | 2899 | if (resume) { |
135b87b4 PX |
2900 | /* Wakeup the main migration thread to do the recovery */ |
2901 | migrate_set_state(&s->state, MIGRATION_STATUS_POSTCOPY_PAUSED, | |
2902 | MIGRATION_STATUS_POSTCOPY_RECOVER); | |
2903 | qemu_sem_post(&s->postcopy_pause_sem); | |
d3e35b8f PX |
2904 | return; |
2905 | } | |
2906 | ||
f986c3d2 JQ |
2907 | if (multifd_save_setup() != 0) { |
2908 | migrate_set_state(&s->state, MIGRATION_STATUS_SETUP, | |
2909 | MIGRATION_STATUS_FAILED); | |
2910 | migrate_fd_cleanup(s); | |
2911 | return; | |
2912 | } | |
009fad7f | 2913 | qemu_thread_create(&s->thread, "live_migration", migration_thread, s, |
bb1fadc4 | 2914 | QEMU_THREAD_JOINABLE); |
1d34e4bf | 2915 | s->migration_thread_running = true; |
0d82d0e8 | 2916 | } |
093e3c42 | 2917 | |
9d18af93 PX |
2918 | void migration_global_dump(Monitor *mon) |
2919 | { | |
2920 | MigrationState *ms = migrate_get_current(); | |
2921 | ||
6f0f6428 JQ |
2922 | monitor_printf(mon, "globals:\n"); |
2923 | monitor_printf(mon, "store-global-state: %s\n", | |
2924 | ms->store_global_state ? "on" : "off"); | |
2925 | monitor_printf(mon, "only-migratable: %s\n", | |
2926 | ms->only_migratable ? "on" : "off"); | |
2927 | monitor_printf(mon, "send-configuration: %s\n", | |
2928 | ms->send_configuration ? "on" : "off"); | |
2929 | monitor_printf(mon, "send-section-footer: %s\n", | |
2930 | ms->send_section_footer ? "on" : "off"); | |
9d18af93 PX |
2931 | } |
2932 | ||
20814758 PX |
2933 | #define DEFINE_PROP_MIG_CAP(name, x) \ |
2934 | DEFINE_PROP_BOOL(name, MigrationState, enabled_capabilities[x], false) | |
2935 | ||
5272298c PX |
2936 | static Property migration_properties[] = { |
2937 | DEFINE_PROP_BOOL("store-global-state", MigrationState, | |
2938 | store_global_state, true), | |
3df663e5 | 2939 | DEFINE_PROP_BOOL("only-migratable", MigrationState, only_migratable, false), |
71dd4c1a PX |
2940 | DEFINE_PROP_BOOL("send-configuration", MigrationState, |
2941 | send_configuration, true), | |
15c38503 PX |
2942 | DEFINE_PROP_BOOL("send-section-footer", MigrationState, |
2943 | send_section_footer, true), | |
89632faf PX |
2944 | |
2945 | /* Migration parameters */ | |
741d4086 | 2946 | DEFINE_PROP_UINT8("x-compress-level", MigrationState, |
89632faf PX |
2947 | parameters.compress_level, |
2948 | DEFAULT_MIGRATE_COMPRESS_LEVEL), | |
741d4086 | 2949 | DEFINE_PROP_UINT8("x-compress-threads", MigrationState, |
89632faf PX |
2950 | parameters.compress_threads, |
2951 | DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT), | |
741d4086 | 2952 | DEFINE_PROP_UINT8("x-decompress-threads", MigrationState, |
89632faf PX |
2953 | parameters.decompress_threads, |
2954 | DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT), | |
741d4086 | 2955 | DEFINE_PROP_UINT8("x-cpu-throttle-initial", MigrationState, |
89632faf PX |
2956 | parameters.cpu_throttle_initial, |
2957 | DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL), | |
741d4086 | 2958 | DEFINE_PROP_UINT8("x-cpu-throttle-increment", MigrationState, |
89632faf PX |
2959 | parameters.cpu_throttle_increment, |
2960 | DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT), | |
741d4086 | 2961 | DEFINE_PROP_SIZE("x-max-bandwidth", MigrationState, |
89632faf | 2962 | parameters.max_bandwidth, MAX_THROTTLE), |
741d4086 | 2963 | DEFINE_PROP_UINT64("x-downtime-limit", MigrationState, |
89632faf PX |
2964 | parameters.downtime_limit, |
2965 | DEFAULT_MIGRATE_SET_DOWNTIME), | |
741d4086 | 2966 | DEFINE_PROP_UINT32("x-checkpoint-delay", MigrationState, |
89632faf PX |
2967 | parameters.x_checkpoint_delay, |
2968 | DEFAULT_MIGRATE_X_CHECKPOINT_DELAY), | |
741d4086 | 2969 | DEFINE_PROP_UINT8("x-multifd-channels", MigrationState, |
4075fb1c JQ |
2970 | parameters.x_multifd_channels, |
2971 | DEFAULT_MIGRATE_MULTIFD_CHANNELS), | |
741d4086 | 2972 | DEFINE_PROP_UINT32("x-multifd-page-count", MigrationState, |
0fb86605 JQ |
2973 | parameters.x_multifd_page_count, |
2974 | DEFAULT_MIGRATE_MULTIFD_PAGE_COUNT), | |
73af8dd8 JQ |
2975 | DEFINE_PROP_SIZE("xbzrle-cache-size", MigrationState, |
2976 | parameters.xbzrle_cache_size, | |
2977 | DEFAULT_MIGRATE_XBZRLE_CACHE_SIZE), | |
20814758 PX |
2978 | |
2979 | /* Migration capabilities */ | |
2980 | DEFINE_PROP_MIG_CAP("x-xbzrle", MIGRATION_CAPABILITY_XBZRLE), | |
2981 | DEFINE_PROP_MIG_CAP("x-rdma-pin-all", MIGRATION_CAPABILITY_RDMA_PIN_ALL), | |
2982 | DEFINE_PROP_MIG_CAP("x-auto-converge", MIGRATION_CAPABILITY_AUTO_CONVERGE), | |
2983 | DEFINE_PROP_MIG_CAP("x-zero-blocks", MIGRATION_CAPABILITY_ZERO_BLOCKS), | |
2984 | DEFINE_PROP_MIG_CAP("x-compress", MIGRATION_CAPABILITY_COMPRESS), | |
2985 | DEFINE_PROP_MIG_CAP("x-events", MIGRATION_CAPABILITY_EVENTS), | |
2986 | DEFINE_PROP_MIG_CAP("x-postcopy-ram", MIGRATION_CAPABILITY_POSTCOPY_RAM), | |
2987 | DEFINE_PROP_MIG_CAP("x-colo", MIGRATION_CAPABILITY_X_COLO), | |
2988 | DEFINE_PROP_MIG_CAP("x-release-ram", MIGRATION_CAPABILITY_RELEASE_RAM), | |
2989 | DEFINE_PROP_MIG_CAP("x-block", MIGRATION_CAPABILITY_BLOCK), | |
2990 | DEFINE_PROP_MIG_CAP("x-return-path", MIGRATION_CAPABILITY_RETURN_PATH), | |
30126bbf | 2991 | DEFINE_PROP_MIG_CAP("x-multifd", MIGRATION_CAPABILITY_X_MULTIFD), |
20814758 | 2992 | |
5272298c PX |
2993 | DEFINE_PROP_END_OF_LIST(), |
2994 | }; | |
2995 | ||
e5cb7e76 PX |
2996 | static void migration_class_init(ObjectClass *klass, void *data) |
2997 | { | |
2998 | DeviceClass *dc = DEVICE_CLASS(klass); | |
2999 | ||
3000 | dc->user_creatable = false; | |
5272298c | 3001 | dc->props = migration_properties; |
e5cb7e76 PX |
3002 | } |
3003 | ||
b91bf5e4 MAL |
3004 | static void migration_instance_finalize(Object *obj) |
3005 | { | |
3006 | MigrationState *ms = MIGRATION_OBJ(obj); | |
3007 | MigrationParameters *params = &ms->parameters; | |
3008 | ||
87db1a7d | 3009 | qemu_mutex_destroy(&ms->error_mutex); |
b91bf5e4 MAL |
3010 | g_free(params->tls_hostname); |
3011 | g_free(params->tls_creds); | |
e91d8951 | 3012 | qemu_sem_destroy(&ms->pause_sem); |
b23c2ade | 3013 | qemu_sem_destroy(&ms->postcopy_pause_sem); |
14b1742e | 3014 | qemu_sem_destroy(&ms->postcopy_pause_rp_sem); |
edd090c7 | 3015 | qemu_sem_destroy(&ms->rp_state.rp_sem); |
ab105cc1 | 3016 | error_free(ms->error); |
b91bf5e4 MAL |
3017 | } |
3018 | ||
e5cb7e76 PX |
3019 | static void migration_instance_init(Object *obj) |
3020 | { | |
3021 | MigrationState *ms = MIGRATION_OBJ(obj); | |
8b0b29dc | 3022 | MigrationParameters *params = &ms->parameters; |
e5cb7e76 PX |
3023 | |
3024 | ms->state = MIGRATION_STATUS_NONE; | |
e5cb7e76 | 3025 | ms->mbps = -1; |
e91d8951 | 3026 | qemu_sem_init(&ms->pause_sem, 0); |
87db1a7d | 3027 | qemu_mutex_init(&ms->error_mutex); |
8b0b29dc PX |
3028 | |
3029 | params->tls_hostname = g_strdup(""); | |
3030 | params->tls_creds = g_strdup(""); | |
3031 | ||
3032 | /* Set has_* up only for parameter checks */ | |
3033 | params->has_compress_level = true; | |
3034 | params->has_compress_threads = true; | |
3035 | params->has_decompress_threads = true; | |
3036 | params->has_cpu_throttle_initial = true; | |
3037 | params->has_cpu_throttle_increment = true; | |
3038 | params->has_max_bandwidth = true; | |
3039 | params->has_downtime_limit = true; | |
3040 | params->has_x_checkpoint_delay = true; | |
3041 | params->has_block_incremental = true; | |
4075fb1c | 3042 | params->has_x_multifd_channels = true; |
0fb86605 | 3043 | params->has_x_multifd_page_count = true; |
73af8dd8 | 3044 | params->has_xbzrle_cache_size = true; |
b23c2ade PX |
3045 | |
3046 | qemu_sem_init(&ms->postcopy_pause_sem, 0); | |
14b1742e | 3047 | qemu_sem_init(&ms->postcopy_pause_rp_sem, 0); |
edd090c7 | 3048 | qemu_sem_init(&ms->rp_state.rp_sem, 0); |
8b0b29dc PX |
3049 | } |
3050 | ||
3051 | /* | |
3052 | * Return true if check pass, false otherwise. Error will be put | |
3053 | * inside errp if provided. | |
3054 | */ | |
3055 | static bool migration_object_check(MigrationState *ms, Error **errp) | |
3056 | { | |
6b19a7d9 PX |
3057 | MigrationCapabilityStatusList *head = NULL; |
3058 | /* Assuming all off */ | |
3059 | bool cap_list[MIGRATION_CAPABILITY__MAX] = { 0 }, ret; | |
3060 | int i; | |
3061 | ||
8b0b29dc PX |
3062 | if (!migrate_params_check(&ms->parameters, errp)) { |
3063 | return false; | |
3064 | } | |
3065 | ||
6b19a7d9 PX |
3066 | for (i = 0; i < MIGRATION_CAPABILITY__MAX; i++) { |
3067 | if (ms->enabled_capabilities[i]) { | |
3068 | head = migrate_cap_add(head, i, true); | |
3069 | } | |
3070 | } | |
3071 | ||
3072 | ret = migrate_caps_check(cap_list, head, errp); | |
3073 | ||
3074 | /* It works with head == NULL */ | |
3075 | qapi_free_MigrationCapabilityStatusList(head); | |
3076 | ||
3077 | return ret; | |
e5cb7e76 PX |
3078 | } |
3079 | ||
3080 | static const TypeInfo migration_type = { | |
3081 | .name = TYPE_MIGRATION, | |
01f6e14c | 3082 | /* |
c8d3ff38 PX |
3083 | * NOTE: TYPE_MIGRATION is not really a device, as the object is |
3084 | * not created using qdev_create(), it is not attached to the qdev | |
3085 | * device tree, and it is never realized. | |
3086 | * | |
3087 | * TODO: Make this TYPE_OBJECT once QOM provides something like | |
3088 | * TYPE_DEVICE's "-global" properties. | |
01f6e14c | 3089 | */ |
e5cb7e76 PX |
3090 | .parent = TYPE_DEVICE, |
3091 | .class_init = migration_class_init, | |
3092 | .class_size = sizeof(MigrationClass), | |
3093 | .instance_size = sizeof(MigrationState), | |
3094 | .instance_init = migration_instance_init, | |
b91bf5e4 | 3095 | .instance_finalize = migration_instance_finalize, |
e5cb7e76 PX |
3096 | }; |
3097 | ||
3098 | static void register_migration_types(void) | |
3099 | { | |
3100 | type_register_static(&migration_type); | |
3101 | } | |
3102 | ||
3103 | type_init(register_migration_types); |