]>
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" |
9aca82ba JQ |
34 | #include "qapi/clone-visitor.h" |
35 | #include "qapi/qapi-visit-sockets.h" | |
9af23989 MA |
36 | #include "qapi/qapi-commands-migration.h" |
37 | #include "qapi/qapi-events-migration.h" | |
cc7a8ea7 | 38 | #include "qapi/qmp/qerror.h" |
15280c36 | 39 | #include "qapi/qmp/qnull.h" |
ab28bd23 | 40 | #include "qemu/rcu.h" |
2c9e6fec | 41 | #include "block.h" |
be07b0ac | 42 | #include "postcopy-ram.h" |
766bd176 | 43 | #include "qemu/thread.h" |
c09e5bb1 | 44 | #include "trace.h" |
51180423 | 45 | #include "exec/target_page.h" |
61b67d47 | 46 | #include "io/channel-buffer.h" |
35a6ed4f | 47 | #include "migration/colo.h" |
4ffdb337 | 48 | #include "hw/boards.h" |
9d18af93 | 49 | #include "monitor/monitor.h" |
50510ea2 | 50 | #include "net/announce.h" |
065e2813 | 51 | |
dc325627 | 52 | #define MAX_THROTTLE (32 << 20) /* Migration transfer speed throttling */ |
5bb7910a | 53 | |
5b4e1eb7 JQ |
54 | /* Amount of time to allocate to each "chunk" of bandwidth-throttled |
55 | * data. */ | |
56 | #define BUFFER_DELAY 100 | |
57 | #define XFER_LIMIT_RATIO (1000 / BUFFER_DELAY) | |
58 | ||
2ff30257 AA |
59 | /* Time in milliseconds we are allowed to stop the source, |
60 | * for sending the last part */ | |
61 | #define DEFAULT_MIGRATE_SET_DOWNTIME 300 | |
62 | ||
87c9cc1c DHB |
63 | /* Maximum migrate downtime set to 2000 seconds */ |
64 | #define MAX_MIGRATE_DOWNTIME_SECONDS 2000 | |
65 | #define MAX_MIGRATE_DOWNTIME (MAX_MIGRATE_DOWNTIME_SECONDS * 1000) | |
66 | ||
8706d2d5 LL |
67 | /* Default compression thread count */ |
68 | #define DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT 8 | |
3fcb38c2 LL |
69 | /* Default decompression thread count, usually decompression is at |
70 | * least 4 times as fast as compression.*/ | |
71 | #define DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT 2 | |
8706d2d5 LL |
72 | /*0: means nocompress, 1: best speed, ... 9: best compress ratio */ |
73 | #define DEFAULT_MIGRATE_COMPRESS_LEVEL 1 | |
1626fee3 | 74 | /* Define default autoconverge cpu throttle migration parameters */ |
d85a31d1 JH |
75 | #define DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL 20 |
76 | #define DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT 10 | |
4cbc9c7f | 77 | #define DEFAULT_MIGRATE_MAX_CPU_THROTTLE 99 |
8706d2d5 | 78 | |
17ad9b35 | 79 | /* Migration XBZRLE default cache size */ |
73af8dd8 | 80 | #define DEFAULT_MIGRATE_XBZRLE_CACHE_SIZE (64 * 1024 * 1024) |
17ad9b35 | 81 | |
131b2153 ZC |
82 | /* The delay time (in ms) between two COLO checkpoints */ |
83 | #define DEFAULT_MIGRATE_X_CHECKPOINT_DELAY (200 * 100) | |
4075fb1c | 84 | #define DEFAULT_MIGRATE_MULTIFD_CHANNELS 2 |
0fb86605 | 85 | #define DEFAULT_MIGRATE_MULTIFD_PAGE_COUNT 16 |
68b53591 | 86 | |
7e555c6c DDAG |
87 | /* Background transfer rate for postcopy, 0 means unlimited, note |
88 | * that page requests can still exceed this limit. | |
89 | */ | |
90 | #define DEFAULT_MIGRATE_MAX_POSTCOPY_BANDWIDTH 0 | |
91 | ||
ee3d96ba DDAG |
92 | /* |
93 | * Parameters for self_announce_delay giving a stream of RARP/ARP | |
94 | * packets after migration. | |
95 | */ | |
96 | #define DEFAULT_MIGRATE_ANNOUNCE_INITIAL 50 | |
97 | #define DEFAULT_MIGRATE_ANNOUNCE_MAX 550 | |
98 | #define DEFAULT_MIGRATE_ANNOUNCE_ROUNDS 5 | |
99 | #define DEFAULT_MIGRATE_ANNOUNCE_STEP 100 | |
100 | ||
99a0db9b GH |
101 | static NotifierList migration_state_notifiers = |
102 | NOTIFIER_LIST_INITIALIZER(migration_state_notifiers); | |
103 | ||
adde220a DDAG |
104 | static bool deferred_incoming; |
105 | ||
da6f1790 JQ |
106 | /* Messages sent on the return path from destination to source */ |
107 | enum mig_rp_message_type { | |
108 | MIG_RP_MSG_INVALID = 0, /* Must be 0 */ | |
109 | MIG_RP_MSG_SHUT, /* sibling will not send any more RP messages */ | |
110 | MIG_RP_MSG_PONG, /* Response to a PING; data (seq: be32 ) */ | |
111 | ||
112 | MIG_RP_MSG_REQ_PAGES_ID, /* data (start: be64, len: be32, id: string) */ | |
113 | MIG_RP_MSG_REQ_PAGES, /* data (start: be64, len: be32) */ | |
a335debb | 114 | MIG_RP_MSG_RECV_BITMAP, /* send recved_bitmap back to source */ |
13955b89 | 115 | MIG_RP_MSG_RESUME_ACK, /* tell source that we are ready to resume */ |
da6f1790 JQ |
116 | |
117 | MIG_RP_MSG_MAX | |
118 | }; | |
119 | ||
17549e84 JQ |
120 | /* When we add fault tolerance, we could have several |
121 | migrations at once. For now we don't need to add | |
122 | dynamic creation of migration */ | |
123 | ||
e5cb7e76 | 124 | static MigrationState *current_migration; |
e1b1b1bc | 125 | static MigrationIncomingState *current_incoming; |
e5cb7e76 | 126 | |
8b0b29dc | 127 | static bool migration_object_check(MigrationState *ms, Error **errp); |
0331c8ca DDAG |
128 | static int migration_maybe_pause(MigrationState *s, |
129 | int *current_active_state, | |
130 | int new_state); | |
892ae715 | 131 | static void migrate_fd_cancel(MigrationState *s); |
8b0b29dc | 132 | |
e5cb7e76 PX |
133 | void migration_object_init(void) |
134 | { | |
4ffdb337 | 135 | MachineState *ms = MACHINE(qdev_get_machine()); |
8b0b29dc | 136 | Error *err = NULL; |
4ffdb337 | 137 | |
e5cb7e76 PX |
138 | /* This can only be called once. */ |
139 | assert(!current_migration); | |
140 | current_migration = MIGRATION_OBJ(object_new(TYPE_MIGRATION)); | |
4ffdb337 | 141 | |
e1b1b1bc PX |
142 | /* |
143 | * Init the migrate incoming object as well no matter whether | |
144 | * we'll use it or not. | |
145 | */ | |
146 | assert(!current_incoming); | |
147 | current_incoming = g_new0(MigrationIncomingState, 1); | |
148 | current_incoming->state = MIGRATION_STATUS_NONE; | |
149 | current_incoming->postcopy_remote_fds = | |
150 | g_array_new(FALSE, TRUE, sizeof(struct PostCopyFD)); | |
151 | qemu_mutex_init(¤t_incoming->rp_mutex); | |
152 | qemu_event_init(¤t_incoming->main_thread_load_event, false); | |
153 | qemu_sem_init(¤t_incoming->postcopy_pause_sem_dst, 0); | |
154 | qemu_sem_init(¤t_incoming->postcopy_pause_sem_fault, 0); | |
155 | ||
156 | init_dirty_bitmap_incoming_migration(); | |
157 | ||
8b0b29dc PX |
158 | if (!migration_object_check(current_migration, &err)) { |
159 | error_report_err(err); | |
160 | exit(1); | |
161 | } | |
162 | ||
4ffdb337 PX |
163 | /* |
164 | * We cannot really do this in migration_instance_init() since at | |
165 | * that time global properties are not yet applied, then this | |
166 | * value will be definitely replaced by something else. | |
167 | */ | |
168 | if (ms->enforce_config_section) { | |
169 | current_migration->send_configuration = true; | |
170 | } | |
e5cb7e76 PX |
171 | } |
172 | ||
892ae715 | 173 | void migration_shutdown(void) |
1f895604 | 174 | { |
892ae715 DDAG |
175 | /* |
176 | * Cancel the current migration - that will (eventually) | |
177 | * stop the migration using this structure | |
178 | */ | |
179 | migrate_fd_cancel(current_migration); | |
1f895604 VSO |
180 | object_unref(OBJECT(current_migration)); |
181 | } | |
182 | ||
bca7856a | 183 | /* For outgoing */ |
859bc756 | 184 | MigrationState *migrate_get_current(void) |
17549e84 | 185 | { |
e5cb7e76 PX |
186 | /* This can only be called after the object created. */ |
187 | assert(current_migration); | |
188 | return current_migration; | |
17549e84 JQ |
189 | } |
190 | ||
bca7856a DDAG |
191 | MigrationIncomingState *migration_incoming_get_current(void) |
192 | { | |
e1b1b1bc PX |
193 | assert(current_incoming); |
194 | return current_incoming; | |
bca7856a DDAG |
195 | } |
196 | ||
197 | void migration_incoming_state_destroy(void) | |
198 | { | |
b4b076da JQ |
199 | struct MigrationIncomingState *mis = migration_incoming_get_current(); |
200 | ||
3482655b | 201 | if (mis->to_src_file) { |
660819b1 PX |
202 | /* Tell source that we are done */ |
203 | migrate_send_rp_shut(mis, qemu_file_get_error(mis->from_src_file) != 0); | |
3482655b PX |
204 | qemu_fclose(mis->to_src_file); |
205 | mis->to_src_file = NULL; | |
206 | } | |
207 | ||
660819b1 PX |
208 | if (mis->from_src_file) { |
209 | qemu_fclose(mis->from_src_file); | |
210 | mis->from_src_file = NULL; | |
211 | } | |
00fa4fc8 DDAG |
212 | if (mis->postcopy_remote_fds) { |
213 | g_array_free(mis->postcopy_remote_fds, TRUE); | |
214 | mis->postcopy_remote_fds = NULL; | |
215 | } | |
660819b1 | 216 | |
5089e186 | 217 | qemu_event_reset(&mis->main_thread_load_event); |
9aca82ba JQ |
218 | |
219 | if (mis->socket_address_list) { | |
220 | qapi_free_SocketAddressList(mis->socket_address_list); | |
221 | mis->socket_address_list = NULL; | |
222 | } | |
bca7856a DDAG |
223 | } |
224 | ||
b05dc723 JQ |
225 | static void migrate_generate_event(int new_state) |
226 | { | |
227 | if (migrate_use_events()) { | |
3ab72385 | 228 | qapi_event_send_migration(new_state); |
b05dc723 JQ |
229 | } |
230 | } | |
231 | ||
0f073f44 DDAG |
232 | static bool migrate_late_block_activate(void) |
233 | { | |
234 | MigrationState *s; | |
235 | ||
236 | s = migrate_get_current(); | |
237 | ||
238 | return s->enabled_capabilities[ | |
239 | MIGRATION_CAPABILITY_LATE_BLOCK_ACTIVATE]; | |
240 | } | |
241 | ||
adde220a DDAG |
242 | /* |
243 | * Called on -incoming with a defer: uri. | |
244 | * The migration can be started later after any parameters have been | |
245 | * changed. | |
246 | */ | |
247 | static void deferred_incoming_migration(Error **errp) | |
248 | { | |
249 | if (deferred_incoming) { | |
250 | error_setg(errp, "Incoming migration already deferred"); | |
251 | } | |
252 | deferred_incoming = true; | |
253 | } | |
254 | ||
da6f1790 JQ |
255 | /* |
256 | * Send a message on the return channel back to the source | |
257 | * of the migration. | |
258 | */ | |
d6208e35 PX |
259 | static int migrate_send_rp_message(MigrationIncomingState *mis, |
260 | enum mig_rp_message_type message_type, | |
261 | uint16_t len, void *data) | |
da6f1790 | 262 | { |
d6208e35 PX |
263 | int ret = 0; |
264 | ||
da6f1790 JQ |
265 | trace_migrate_send_rp_message((int)message_type, len); |
266 | qemu_mutex_lock(&mis->rp_mutex); | |
d6208e35 PX |
267 | |
268 | /* | |
269 | * It's possible that the file handle got lost due to network | |
270 | * failures. | |
271 | */ | |
272 | if (!mis->to_src_file) { | |
273 | ret = -EIO; | |
274 | goto error; | |
275 | } | |
276 | ||
da6f1790 JQ |
277 | qemu_put_be16(mis->to_src_file, (unsigned int)message_type); |
278 | qemu_put_be16(mis->to_src_file, len); | |
279 | qemu_put_buffer(mis->to_src_file, data, len); | |
280 | qemu_fflush(mis->to_src_file); | |
d6208e35 PX |
281 | |
282 | /* It's possible that qemu file got error during sending */ | |
283 | ret = qemu_file_get_error(mis->to_src_file); | |
284 | ||
285 | error: | |
da6f1790 | 286 | qemu_mutex_unlock(&mis->rp_mutex); |
d6208e35 | 287 | return ret; |
da6f1790 JQ |
288 | } |
289 | ||
1e2d90eb DDAG |
290 | /* Request a range of pages from the source VM at the given |
291 | * start address. | |
292 | * rbname: Name of the RAMBlock to request the page in, if NULL it's the same | |
293 | * as the last request (a name must have been given previously) | |
294 | * Start: Address offset within the RB | |
295 | * Len: Length in bytes required - must be a multiple of pagesize | |
296 | */ | |
d6208e35 PX |
297 | int migrate_send_rp_req_pages(MigrationIncomingState *mis, const char *rbname, |
298 | ram_addr_t start, size_t len) | |
1e2d90eb | 299 | { |
cb8d4c8f | 300 | uint8_t bufc[12 + 1 + 255]; /* start (8), len (4), rbname up to 256 */ |
1e2d90eb | 301 | size_t msglen = 12; /* start + len */ |
d6208e35 | 302 | enum mig_rp_message_type msg_type; |
1e2d90eb DDAG |
303 | |
304 | *(uint64_t *)bufc = cpu_to_be64((uint64_t)start); | |
305 | *(uint32_t *)(bufc + 8) = cpu_to_be32((uint32_t)len); | |
306 | ||
307 | if (rbname) { | |
308 | int rbname_len = strlen(rbname); | |
309 | assert(rbname_len < 256); | |
310 | ||
311 | bufc[msglen++] = rbname_len; | |
312 | memcpy(bufc + msglen, rbname, rbname_len); | |
313 | msglen += rbname_len; | |
d6208e35 | 314 | msg_type = MIG_RP_MSG_REQ_PAGES_ID; |
1e2d90eb | 315 | } else { |
d6208e35 | 316 | msg_type = MIG_RP_MSG_REQ_PAGES; |
1e2d90eb | 317 | } |
d6208e35 PX |
318 | |
319 | return migrate_send_rp_message(mis, msg_type, msglen, bufc); | |
1e2d90eb DDAG |
320 | } |
321 | ||
aad555c2 ZC |
322 | static bool migration_colo_enabled; |
323 | bool migration_incoming_colo_enabled(void) | |
324 | { | |
325 | return migration_colo_enabled; | |
326 | } | |
327 | ||
328 | void migration_incoming_disable_colo(void) | |
329 | { | |
330 | migration_colo_enabled = false; | |
331 | } | |
332 | ||
333 | void migration_incoming_enable_colo(void) | |
334 | { | |
335 | migration_colo_enabled = true; | |
336 | } | |
337 | ||
9aca82ba JQ |
338 | void migrate_add_address(SocketAddress *address) |
339 | { | |
340 | MigrationIncomingState *mis = migration_incoming_get_current(); | |
341 | SocketAddressList *addrs; | |
342 | ||
343 | addrs = g_new0(SocketAddressList, 1); | |
344 | addrs->next = mis->socket_address_list; | |
345 | mis->socket_address_list = addrs; | |
346 | addrs->value = QAPI_CLONE(SocketAddress, address); | |
347 | } | |
348 | ||
43eaae28 | 349 | void qemu_start_incoming_migration(const char *uri, Error **errp) |
5bb7910a | 350 | { |
34c9dd8e AL |
351 | const char *p; |
352 | ||
3ab72385 | 353 | qapi_event_send_migration(MIGRATION_STATUS_SETUP); |
adde220a DDAG |
354 | if (!strcmp(uri, "defer")) { |
355 | deferred_incoming_migration(errp); | |
356 | } else if (strstart(uri, "tcp:", &p)) { | |
43eaae28 | 357 | tcp_start_incoming_migration(p, errp); |
2da776db | 358 | #ifdef CONFIG_RDMA |
adde220a | 359 | } else if (strstart(uri, "rdma:", &p)) { |
2da776db MH |
360 | rdma_start_incoming_migration(p, errp); |
361 | #endif | |
adde220a | 362 | } else if (strstart(uri, "exec:", &p)) { |
43eaae28 | 363 | exec_start_incoming_migration(p, errp); |
adde220a | 364 | } else if (strstart(uri, "unix:", &p)) { |
43eaae28 | 365 | unix_start_incoming_migration(p, errp); |
adde220a | 366 | } else if (strstart(uri, "fd:", &p)) { |
43eaae28 | 367 | fd_start_incoming_migration(p, errp); |
adde220a | 368 | } else { |
312fd5f2 | 369 | error_setg(errp, "unknown migration protocol: %s", uri); |
8ca5e801 | 370 | } |
5bb7910a AL |
371 | } |
372 | ||
0aa6aefc DL |
373 | static void process_incoming_migration_bh(void *opaque) |
374 | { | |
375 | Error *local_err = NULL; | |
376 | MigrationIncomingState *mis = opaque; | |
377 | ||
0f073f44 DDAG |
378 | /* If capability late_block_activate is set: |
379 | * Only fire up the block code now if we're going to restart the | |
380 | * VM, else 'cont' will do it. | |
381 | * This causes file locking to happen; so we don't want it to happen | |
382 | * unless we really are starting the VM. | |
383 | */ | |
384 | if (!migrate_late_block_activate() || | |
385 | (autostart && (!global_state_received() || | |
386 | global_state_get_runstate() == RUN_STATE_RUNNING))) { | |
387 | /* Make sure all file formats flush their mutable metadata. | |
388 | * If we get an error here, just don't restart the VM yet. */ | |
389 | bdrv_invalidate_cache_all(&local_err); | |
390 | if (local_err) { | |
391 | error_report_err(local_err); | |
392 | local_err = NULL; | |
393 | autostart = false; | |
394 | } | |
d35ff5e6 KW |
395 | } |
396 | ||
0aa6aefc DL |
397 | /* |
398 | * This must happen after all error conditions are dealt with and | |
399 | * we're sure the VM is going to be running on this host. | |
400 | */ | |
7659505c | 401 | qemu_announce_self(&mis->announce_timer, migrate_announce_params()); |
0aa6aefc | 402 | |
f986c3d2 JQ |
403 | if (multifd_load_cleanup(&local_err) != 0) { |
404 | error_report_err(local_err); | |
405 | autostart = false; | |
406 | } | |
0aa6aefc DL |
407 | /* If global state section was not received or we are in running |
408 | state, we need to obey autostart. Any other state is set with | |
409 | runstate_set. */ | |
410 | ||
b35ebdf0 VSO |
411 | dirty_bitmap_mig_before_vm_start(); |
412 | ||
0aa6aefc DL |
413 | if (!global_state_received() || |
414 | global_state_get_runstate() == RUN_STATE_RUNNING) { | |
415 | if (autostart) { | |
416 | vm_start(); | |
417 | } else { | |
418 | runstate_set(RUN_STATE_PAUSED); | |
419 | } | |
db009729 ZC |
420 | } else if (migration_incoming_colo_enabled()) { |
421 | migration_incoming_disable_colo(); | |
422 | vm_start(); | |
0aa6aefc DL |
423 | } else { |
424 | runstate_set(global_state_get_runstate()); | |
425 | } | |
0aa6aefc DL |
426 | /* |
427 | * This must happen after any state changes since as soon as an external | |
428 | * observer sees this event they might start to prod at the VM assuming | |
429 | * it's ready to use. | |
430 | */ | |
431 | migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE, | |
432 | MIGRATION_STATUS_COMPLETED); | |
433 | qemu_bh_delete(mis->bh); | |
434 | migration_incoming_state_destroy(); | |
435 | } | |
436 | ||
82a4da79 | 437 | static void process_incoming_migration_co(void *opaque) |
511c0231 | 438 | { |
b4b076da | 439 | MigrationIncomingState *mis = migration_incoming_get_current(); |
e9bef235 | 440 | PostcopyState ps; |
1c12e1f5 | 441 | int ret; |
8e48ac95 | 442 | Error *local_err = NULL; |
1c12e1f5 | 443 | |
4f0fae7f | 444 | assert(mis->from_src_file); |
92370989 | 445 | mis->migration_incoming_co = qemu_coroutine_self(); |
67f11b5c | 446 | mis->largest_page_size = qemu_ram_pagesize_largest(); |
093e3c42 | 447 | postcopy_state_set(POSTCOPY_INCOMING_NONE); |
93d7af6f HZ |
448 | migrate_set_state(&mis->state, MIGRATION_STATUS_NONE, |
449 | MIGRATION_STATUS_ACTIVE); | |
4f0fae7f | 450 | ret = qemu_loadvm_state(mis->from_src_file); |
bca7856a | 451 | |
e9bef235 DDAG |
452 | ps = postcopy_state_get(); |
453 | trace_process_incoming_migration_co_end(ret, ps); | |
454 | if (ps != POSTCOPY_INCOMING_NONE) { | |
455 | if (ps == POSTCOPY_INCOMING_ADVISE) { | |
456 | /* | |
457 | * Where a migration had postcopy enabled (and thus went to advise) | |
458 | * but managed to complete within the precopy period, we can use | |
459 | * the normal exit. | |
460 | */ | |
461 | postcopy_ram_incoming_cleanup(mis); | |
462 | } else if (ret >= 0) { | |
463 | /* | |
464 | * Postcopy was started, cleanup should happen at the end of the | |
465 | * postcopy thread. | |
466 | */ | |
467 | trace_process_incoming_migration_co_postcopy_end_main(); | |
468 | return; | |
469 | } | |
470 | /* Else if something went wrong then just fall out of the normal exit */ | |
471 | } | |
472 | ||
25d0c16f | 473 | /* we get COLO info, and know if we are in COLO mode */ |
aad555c2 | 474 | if (!ret && migration_incoming_colo_enabled()) { |
8e48ac95 ZC |
475 | /* Make sure all file formats flush their mutable metadata */ |
476 | bdrv_invalidate_cache_all(&local_err); | |
477 | if (local_err) { | |
8e48ac95 | 478 | error_report_err(local_err); |
6d99c2d4 | 479 | goto fail; |
8e48ac95 ZC |
480 | } |
481 | ||
13af18f2 ZC |
482 | if (colo_init_ram_cache() < 0) { |
483 | error_report("Init ram cache failed"); | |
6d99c2d4 | 484 | goto fail; |
13af18f2 ZC |
485 | } |
486 | ||
25d0c16f HZ |
487 | qemu_thread_create(&mis->colo_incoming_thread, "COLO incoming", |
488 | colo_process_incoming_thread, mis, QEMU_THREAD_JOINABLE); | |
489 | mis->have_colo_incoming_thread = true; | |
490 | qemu_coroutine_yield(); | |
491 | ||
492 | /* Wait checkpoint incoming thread exit before free resource */ | |
493 | qemu_thread_join(&mis->colo_incoming_thread); | |
13af18f2 ZC |
494 | /* We hold the global iothread lock, so it is safe here */ |
495 | colo_release_ram_cache(); | |
25d0c16f HZ |
496 | } |
497 | ||
1c12e1f5 | 498 | if (ret < 0) { |
db80face | 499 | error_report("load of migration failed: %s", strerror(-ret)); |
6d99c2d4 | 500 | goto fail; |
511c0231 | 501 | } |
0aa6aefc DL |
502 | mis->bh = qemu_bh_new(process_incoming_migration_bh, mis); |
503 | qemu_bh_schedule(mis->bh); | |
92370989 | 504 | mis->migration_incoming_co = NULL; |
6d99c2d4 FL |
505 | return; |
506 | fail: | |
507 | local_err = NULL; | |
508 | migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE, | |
509 | MIGRATION_STATUS_FAILED); | |
510 | qemu_fclose(mis->from_src_file); | |
511 | if (multifd_load_cleanup(&local_err) != 0) { | |
512 | error_report_err(local_err); | |
513 | } | |
514 | exit(EXIT_FAILURE); | |
511c0231 JQ |
515 | } |
516 | ||
e595a01a | 517 | static void migration_incoming_setup(QEMUFile *f) |
82a4da79 | 518 | { |
4f0fae7f | 519 | MigrationIncomingState *mis = migration_incoming_get_current(); |
82a4da79 | 520 | |
f986c3d2 JQ |
521 | if (multifd_load_setup() != 0) { |
522 | /* We haven't been able to create multifd threads | |
523 | nothing better to do */ | |
524 | exit(EXIT_FAILURE); | |
525 | } | |
526 | ||
4f0fae7f JQ |
527 | if (!mis->from_src_file) { |
528 | mis->from_src_file = f; | |
529 | } | |
06ad5135 | 530 | qemu_file_set_blocking(f, false); |
e595a01a JQ |
531 | } |
532 | ||
36c2f8be | 533 | void migration_incoming_process(void) |
e595a01a JQ |
534 | { |
535 | Coroutine *co = qemu_coroutine_create(process_incoming_migration_co, NULL); | |
0b8b8753 | 536 | qemu_coroutine_enter(co); |
82a4da79 PB |
537 | } |
538 | ||
884835fa PX |
539 | /* Returns true if recovered from a paused migration, otherwise false */ |
540 | static bool postcopy_try_recover(QEMUFile *f) | |
e595a01a | 541 | { |
d96c9e8d PX |
542 | MigrationIncomingState *mis = migration_incoming_get_current(); |
543 | ||
544 | if (mis->state == MIGRATION_STATUS_POSTCOPY_PAUSED) { | |
545 | /* Resumed from a paused postcopy migration */ | |
546 | ||
547 | mis->from_src_file = f; | |
548 | /* Postcopy has standalone thread to do vm load */ | |
549 | qemu_file_set_blocking(f, true); | |
550 | ||
551 | /* Re-configure the return path */ | |
552 | mis->to_src_file = qemu_file_get_return_path(f); | |
553 | ||
554 | migrate_set_state(&mis->state, MIGRATION_STATUS_POSTCOPY_PAUSED, | |
555 | MIGRATION_STATUS_POSTCOPY_RECOVER); | |
556 | ||
557 | /* | |
558 | * Here, we only wake up the main loading thread (while the | |
559 | * fault thread will still be waiting), so that we can receive | |
560 | * commands from source now, and answer it if needed. The | |
561 | * fault thread will be woken up afterwards until we are sure | |
562 | * that source is ready to reply to page requests. | |
563 | */ | |
564 | qemu_sem_post(&mis->postcopy_pause_sem_dst); | |
884835fa PX |
565 | return true; |
566 | } | |
567 | ||
568 | return false; | |
569 | } | |
570 | ||
571 | void migration_fd_process_incoming(QEMUFile *f) | |
572 | { | |
573 | if (postcopy_try_recover(f)) { | |
574 | return; | |
d96c9e8d | 575 | } |
884835fa PX |
576 | |
577 | migration_incoming_setup(f); | |
578 | migration_incoming_process(); | |
e595a01a JQ |
579 | } |
580 | ||
49ed0d24 | 581 | void migration_ioc_process_incoming(QIOChannel *ioc, Error **errp) |
4f0fae7f JQ |
582 | { |
583 | MigrationIncomingState *mis = migration_incoming_get_current(); | |
a429e7f4 | 584 | bool start_migration; |
4f0fae7f JQ |
585 | |
586 | if (!mis->from_src_file) { | |
a429e7f4 | 587 | /* The first connection (multifd may have multiple) */ |
4f0fae7f | 588 | QEMUFile *f = qemu_fopen_channel_input(ioc); |
a429e7f4 PX |
589 | |
590 | /* If it's a recovery, we're done */ | |
884835fa PX |
591 | if (postcopy_try_recover(f)) { |
592 | return; | |
593 | } | |
a429e7f4 | 594 | |
36c2f8be | 595 | migration_incoming_setup(f); |
a429e7f4 PX |
596 | |
597 | /* | |
598 | * Common migration only needs one channel, so we can start | |
599 | * right now. Multifd needs more than one channel, we wait. | |
600 | */ | |
601 | start_migration = !migrate_use_multifd(); | |
602 | } else { | |
49ed0d24 | 603 | Error *local_err = NULL; |
a429e7f4 PX |
604 | /* Multiple connections */ |
605 | assert(migrate_use_multifd()); | |
49ed0d24 FL |
606 | start_migration = multifd_recv_new_channel(ioc, &local_err); |
607 | if (local_err) { | |
608 | error_propagate(errp, local_err); | |
609 | return; | |
610 | } | |
4f0fae7f | 611 | } |
81e62053 | 612 | |
a429e7f4 | 613 | if (start_migration) { |
81e62053 PX |
614 | migration_incoming_process(); |
615 | } | |
4f0fae7f JQ |
616 | } |
617 | ||
428d8908 JQ |
618 | /** |
619 | * @migration_has_all_channels: We have received all channels that we need | |
620 | * | |
621 | * Returns true when we have got connections to all the channels that | |
622 | * we need for migration. | |
623 | */ | |
624 | bool migration_has_all_channels(void) | |
625 | { | |
ca273df3 | 626 | MigrationIncomingState *mis = migration_incoming_get_current(); |
62c1e0ca JQ |
627 | bool all_channels; |
628 | ||
629 | all_channels = multifd_recv_all_channels_created(); | |
630 | ||
ca273df3 | 631 | return all_channels && mis->from_src_file != NULL; |
428d8908 JQ |
632 | } |
633 | ||
6decec93 DDAG |
634 | /* |
635 | * Send a 'SHUT' message on the return channel with the given value | |
636 | * to indicate that we've finished with the RP. Non-0 value indicates | |
637 | * error. | |
638 | */ | |
639 | void migrate_send_rp_shut(MigrationIncomingState *mis, | |
640 | uint32_t value) | |
641 | { | |
642 | uint32_t buf; | |
643 | ||
644 | buf = cpu_to_be32(value); | |
645 | migrate_send_rp_message(mis, MIG_RP_MSG_SHUT, sizeof(buf), &buf); | |
646 | } | |
647 | ||
648 | /* | |
649 | * Send a 'PONG' message on the return channel with the given value | |
650 | * (normally in response to a 'PING') | |
651 | */ | |
652 | void migrate_send_rp_pong(MigrationIncomingState *mis, | |
653 | uint32_t value) | |
654 | { | |
655 | uint32_t buf; | |
656 | ||
657 | buf = cpu_to_be32(value); | |
658 | migrate_send_rp_message(mis, MIG_RP_MSG_PONG, sizeof(buf), &buf); | |
659 | } | |
660 | ||
a335debb PX |
661 | void migrate_send_rp_recv_bitmap(MigrationIncomingState *mis, |
662 | char *block_name) | |
663 | { | |
664 | char buf[512]; | |
665 | int len; | |
666 | int64_t res; | |
667 | ||
668 | /* | |
669 | * First, we send the header part. It contains only the len of | |
670 | * idstr, and the idstr itself. | |
671 | */ | |
672 | len = strlen(block_name); | |
673 | buf[0] = len; | |
674 | memcpy(buf + 1, block_name, len); | |
675 | ||
676 | if (mis->state != MIGRATION_STATUS_POSTCOPY_RECOVER) { | |
677 | error_report("%s: MSG_RP_RECV_BITMAP only used for recovery", | |
678 | __func__); | |
679 | return; | |
680 | } | |
681 | ||
682 | migrate_send_rp_message(mis, MIG_RP_MSG_RECV_BITMAP, len + 1, buf); | |
683 | ||
684 | /* | |
685 | * Next, we dump the received bitmap to the stream. | |
686 | * | |
687 | * TODO: currently we are safe since we are the only one that is | |
688 | * using the to_src_file handle (fault thread is still paused), | |
689 | * and it's ok even not taking the mutex. However the best way is | |
690 | * to take the lock before sending the message header, and release | |
691 | * the lock after sending the bitmap. | |
692 | */ | |
693 | qemu_mutex_lock(&mis->rp_mutex); | |
694 | res = ramblock_recv_bitmap_send(mis->to_src_file, block_name); | |
695 | qemu_mutex_unlock(&mis->rp_mutex); | |
696 | ||
697 | trace_migrate_send_rp_recv_bitmap(block_name, res); | |
698 | } | |
699 | ||
13955b89 PX |
700 | void migrate_send_rp_resume_ack(MigrationIncomingState *mis, uint32_t value) |
701 | { | |
702 | uint32_t buf; | |
703 | ||
704 | buf = cpu_to_be32(value); | |
705 | migrate_send_rp_message(mis, MIG_RP_MSG_RESUME_ACK, sizeof(buf), &buf); | |
706 | } | |
707 | ||
bbf6da32 OW |
708 | MigrationCapabilityStatusList *qmp_query_migrate_capabilities(Error **errp) |
709 | { | |
710 | MigrationCapabilityStatusList *head = NULL; | |
711 | MigrationCapabilityStatusList *caps; | |
712 | MigrationState *s = migrate_get_current(); | |
713 | int i; | |
714 | ||
387eedeb | 715 | caps = NULL; /* silence compiler warning */ |
7fb1cf16 | 716 | for (i = 0; i < MIGRATION_CAPABILITY__MAX; i++) { |
ed1701c6 DDAG |
717 | #ifndef CONFIG_LIVE_BLOCK_MIGRATION |
718 | if (i == MIGRATION_CAPABILITY_BLOCK) { | |
719 | continue; | |
720 | } | |
721 | #endif | |
bbf6da32 OW |
722 | if (head == NULL) { |
723 | head = g_malloc0(sizeof(*caps)); | |
724 | caps = head; | |
725 | } else { | |
726 | caps->next = g_malloc0(sizeof(*caps)); | |
727 | caps = caps->next; | |
728 | } | |
729 | caps->value = | |
730 | g_malloc(sizeof(*caps->value)); | |
731 | caps->value->capability = i; | |
732 | caps->value->state = s->enabled_capabilities[i]; | |
733 | } | |
734 | ||
735 | return head; | |
736 | } | |
737 | ||
85de8323 LL |
738 | MigrationParameters *qmp_query_migrate_parameters(Error **errp) |
739 | { | |
740 | MigrationParameters *params; | |
741 | MigrationState *s = migrate_get_current(); | |
742 | ||
e87fae4c | 743 | /* TODO use QAPI_CLONE() instead of duplicating it inline */ |
85de8323 | 744 | params = g_malloc0(sizeof(*params)); |
de63ab61 | 745 | params->has_compress_level = true; |
2594f56d | 746 | params->compress_level = s->parameters.compress_level; |
de63ab61 | 747 | params->has_compress_threads = true; |
2594f56d | 748 | params->compress_threads = s->parameters.compress_threads; |
1d58872a XG |
749 | params->has_compress_wait_thread = true; |
750 | params->compress_wait_thread = s->parameters.compress_wait_thread; | |
de63ab61 | 751 | params->has_decompress_threads = true; |
2594f56d | 752 | params->decompress_threads = s->parameters.decompress_threads; |
de63ab61 | 753 | params->has_cpu_throttle_initial = true; |
2594f56d | 754 | params->cpu_throttle_initial = s->parameters.cpu_throttle_initial; |
de63ab61 | 755 | params->has_cpu_throttle_increment = true; |
2594f56d | 756 | params->cpu_throttle_increment = s->parameters.cpu_throttle_increment; |
8cc99dcd | 757 | params->has_tls_creds = true; |
69ef1f36 | 758 | params->tls_creds = g_strdup(s->parameters.tls_creds); |
8cc99dcd | 759 | params->has_tls_hostname = true; |
69ef1f36 | 760 | params->tls_hostname = g_strdup(s->parameters.tls_hostname); |
2ff30257 AA |
761 | params->has_max_bandwidth = true; |
762 | params->max_bandwidth = s->parameters.max_bandwidth; | |
763 | params->has_downtime_limit = true; | |
764 | params->downtime_limit = s->parameters.downtime_limit; | |
fe39a4d4 | 765 | params->has_x_checkpoint_delay = true; |
68b53591 | 766 | params->x_checkpoint_delay = s->parameters.x_checkpoint_delay; |
2833c59b JQ |
767 | params->has_block_incremental = true; |
768 | params->block_incremental = s->parameters.block_incremental; | |
4075fb1c JQ |
769 | params->has_x_multifd_channels = true; |
770 | params->x_multifd_channels = s->parameters.x_multifd_channels; | |
0fb86605 JQ |
771 | params->has_x_multifd_page_count = true; |
772 | params->x_multifd_page_count = s->parameters.x_multifd_page_count; | |
73af8dd8 JQ |
773 | params->has_xbzrle_cache_size = true; |
774 | params->xbzrle_cache_size = s->parameters.xbzrle_cache_size; | |
7e555c6c DDAG |
775 | params->has_max_postcopy_bandwidth = true; |
776 | params->max_postcopy_bandwidth = s->parameters.max_postcopy_bandwidth; | |
4cbc9c7f LQ |
777 | params->has_max_cpu_throttle = true; |
778 | params->max_cpu_throttle = s->parameters.max_cpu_throttle; | |
ee3d96ba DDAG |
779 | params->has_announce_initial = true; |
780 | params->announce_initial = s->parameters.announce_initial; | |
781 | params->has_announce_max = true; | |
782 | params->announce_max = s->parameters.announce_max; | |
783 | params->has_announce_rounds = true; | |
784 | params->announce_rounds = s->parameters.announce_rounds; | |
785 | params->has_announce_step = true; | |
786 | params->announce_step = s->parameters.announce_step; | |
85de8323 LL |
787 | |
788 | return params; | |
789 | } | |
790 | ||
ee3d96ba DDAG |
791 | AnnounceParameters *migrate_announce_params(void) |
792 | { | |
793 | static AnnounceParameters ap; | |
794 | ||
795 | MigrationState *s = migrate_get_current(); | |
796 | ||
797 | ap.initial = s->parameters.announce_initial; | |
798 | ap.max = s->parameters.announce_max; | |
799 | ap.rounds = s->parameters.announce_rounds; | |
800 | ap.step = s->parameters.announce_step; | |
801 | ||
802 | return ≈ | |
803 | } | |
804 | ||
f6844b99 DDAG |
805 | /* |
806 | * Return true if we're already in the middle of a migration | |
807 | * (i.e. any of the active or setup states) | |
808 | */ | |
3d63da16 | 809 | bool migration_is_setup_or_active(int state) |
f6844b99 DDAG |
810 | { |
811 | switch (state) { | |
812 | case MIGRATION_STATUS_ACTIVE: | |
9ec055ae | 813 | case MIGRATION_STATUS_POSTCOPY_ACTIVE: |
a688d2c1 | 814 | case MIGRATION_STATUS_POSTCOPY_PAUSED: |
135b87b4 | 815 | case MIGRATION_STATUS_POSTCOPY_RECOVER: |
f6844b99 | 816 | case MIGRATION_STATUS_SETUP: |
31e06077 DDAG |
817 | case MIGRATION_STATUS_PRE_SWITCHOVER: |
818 | case MIGRATION_STATUS_DEVICE: | |
f6844b99 DDAG |
819 | return true; |
820 | ||
821 | default: | |
822 | return false; | |
823 | ||
824 | } | |
825 | } | |
826 | ||
a22463a5 DDAG |
827 | static void populate_ram_info(MigrationInfo *info, MigrationState *s) |
828 | { | |
829 | info->has_ram = true; | |
830 | info->ram = g_malloc0(sizeof(*info->ram)); | |
9360447d | 831 | info->ram->transferred = ram_counters.transferred; |
a22463a5 | 832 | info->ram->total = ram_bytes_total(); |
9360447d | 833 | info->ram->duplicate = ram_counters.duplicate; |
bedf53c1 JQ |
834 | /* legacy value. It is not used anymore */ |
835 | info->ram->skipped = 0; | |
9360447d JQ |
836 | info->ram->normal = ram_counters.normal; |
837 | info->ram->normal_bytes = ram_counters.normal * | |
20afaed9 | 838 | qemu_target_page_size(); |
a22463a5 | 839 | info->ram->mbps = s->mbps; |
9360447d JQ |
840 | info->ram->dirty_sync_count = ram_counters.dirty_sync_count; |
841 | info->ram->postcopy_requests = ram_counters.postcopy_requests; | |
030ce1f8 | 842 | info->ram->page_size = qemu_target_page_size(); |
a61c45bd | 843 | info->ram->multifd_bytes = ram_counters.multifd_bytes; |
aecbfe9c | 844 | info->ram->pages_per_second = s->pages_per_second; |
a22463a5 | 845 | |
114f5aee JQ |
846 | if (migrate_use_xbzrle()) { |
847 | info->has_xbzrle_cache = true; | |
848 | info->xbzrle_cache = g_malloc0(sizeof(*info->xbzrle_cache)); | |
849 | info->xbzrle_cache->cache_size = migrate_xbzrle_cache_size(); | |
9360447d JQ |
850 | info->xbzrle_cache->bytes = xbzrle_counters.bytes; |
851 | info->xbzrle_cache->pages = xbzrle_counters.pages; | |
852 | info->xbzrle_cache->cache_miss = xbzrle_counters.cache_miss; | |
853 | info->xbzrle_cache->cache_miss_rate = xbzrle_counters.cache_miss_rate; | |
854 | info->xbzrle_cache->overflow = xbzrle_counters.overflow; | |
114f5aee JQ |
855 | } |
856 | ||
76e03000 XG |
857 | if (migrate_use_compression()) { |
858 | info->has_compression = true; | |
859 | info->compression = g_malloc0(sizeof(*info->compression)); | |
860 | info->compression->pages = compression_counters.pages; | |
861 | info->compression->busy = compression_counters.busy; | |
862 | info->compression->busy_rate = compression_counters.busy_rate; | |
863 | info->compression->compressed_size = | |
864 | compression_counters.compressed_size; | |
865 | info->compression->compression_rate = | |
866 | compression_counters.compression_rate; | |
867 | } | |
868 | ||
338182c8 JQ |
869 | if (cpu_throttle_active()) { |
870 | info->has_cpu_throttle_percentage = true; | |
871 | info->cpu_throttle_percentage = cpu_throttle_get_percentage(); | |
872 | } | |
873 | ||
a22463a5 DDAG |
874 | if (s->state != MIGRATION_STATUS_COMPLETED) { |
875 | info->ram->remaining = ram_bytes_remaining(); | |
9360447d | 876 | info->ram->dirty_pages_rate = ram_counters.dirty_pages_rate; |
a22463a5 DDAG |
877 | } |
878 | } | |
879 | ||
930ac04c JQ |
880 | static void populate_disk_info(MigrationInfo *info) |
881 | { | |
882 | if (blk_mig_active()) { | |
883 | info->has_disk = true; | |
884 | info->disk = g_malloc0(sizeof(*info->disk)); | |
885 | info->disk->transferred = blk_mig_bytes_transferred(); | |
886 | info->disk->remaining = blk_mig_bytes_remaining(); | |
887 | info->disk->total = blk_mig_bytes_total(); | |
888 | } | |
889 | } | |
890 | ||
65ace060 | 891 | static void fill_source_migration_info(MigrationInfo *info) |
5bb7910a | 892 | { |
17549e84 JQ |
893 | MigrationState *s = migrate_get_current(); |
894 | ||
895 | switch (s->state) { | |
31194731 | 896 | case MIGRATION_STATUS_NONE: |
17549e84 | 897 | /* no migration has happened ever */ |
65ace060 AP |
898 | /* do not overwrite destination migration status */ |
899 | return; | |
17549e84 | 900 | break; |
31194731 | 901 | case MIGRATION_STATUS_SETUP: |
29ae8a41 | 902 | info->has_status = true; |
ed4fbd10 | 903 | info->has_total_time = false; |
29ae8a41 | 904 | break; |
31194731 HZ |
905 | case MIGRATION_STATUS_ACTIVE: |
906 | case MIGRATION_STATUS_CANCELLING: | |
9ec055ae | 907 | case MIGRATION_STATUS_POSTCOPY_ACTIVE: |
31e06077 DDAG |
908 | case MIGRATION_STATUS_PRE_SWITCHOVER: |
909 | case MIGRATION_STATUS_DEVICE: | |
a688d2c1 | 910 | case MIGRATION_STATUS_POSTCOPY_PAUSED: |
135b87b4 | 911 | case MIGRATION_STATUS_POSTCOPY_RECOVER: |
c8f9f4f4 | 912 | /* TODO add some postcopy stats */ |
9ec055ae DDAG |
913 | info->has_status = true; |
914 | info->has_total_time = true; | |
915 | info->total_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME) | |
4af246a3 | 916 | - s->start_time; |
9ec055ae DDAG |
917 | info->has_expected_downtime = true; |
918 | info->expected_downtime = s->expected_downtime; | |
919 | info->has_setup_time = true; | |
920 | info->setup_time = s->setup_time; | |
921 | ||
a22463a5 | 922 | populate_ram_info(info, s); |
930ac04c | 923 | populate_disk_info(info); |
17549e84 | 924 | break; |
0b827d5e HZ |
925 | case MIGRATION_STATUS_COLO: |
926 | info->has_status = true; | |
927 | /* TODO: display COLO specific information (checkpoint info etc.) */ | |
928 | break; | |
31194731 | 929 | case MIGRATION_STATUS_COMPLETED: |
791e7c82 | 930 | info->has_status = true; |
00c14997 | 931 | info->has_total_time = true; |
7aa939af | 932 | info->total_time = s->total_time; |
9c5a9fcf JQ |
933 | info->has_downtime = true; |
934 | info->downtime = s->downtime; | |
ed4fbd10 MH |
935 | info->has_setup_time = true; |
936 | info->setup_time = s->setup_time; | |
d5f8a570 | 937 | |
a22463a5 | 938 | populate_ram_info(info, s); |
17549e84 | 939 | break; |
31194731 | 940 | case MIGRATION_STATUS_FAILED: |
791e7c82 | 941 | info->has_status = true; |
d59ce6f3 DB |
942 | if (s->error) { |
943 | info->has_error_desc = true; | |
944 | info->error_desc = g_strdup(error_get_pretty(s->error)); | |
945 | } | |
17549e84 | 946 | break; |
31194731 | 947 | case MIGRATION_STATUS_CANCELLED: |
791e7c82 | 948 | info->has_status = true; |
17549e84 | 949 | break; |
5bb7910a | 950 | } |
cde63fbe | 951 | info->status = s->state; |
5bb7910a AL |
952 | } |
953 | ||
4a84214e PX |
954 | /** |
955 | * @migration_caps_check - check capability validity | |
956 | * | |
957 | * @cap_list: old capability list, array of bool | |
958 | * @params: new capabilities to be applied soon | |
959 | * @errp: set *errp if the check failed, with reason | |
960 | * | |
961 | * Returns true if check passed, otherwise false. | |
962 | */ | |
963 | static bool migrate_caps_check(bool *cap_list, | |
964 | MigrationCapabilityStatusList *params, | |
965 | Error **errp) | |
00458433 | 966 | { |
00458433 | 967 | MigrationCapabilityStatusList *cap; |
4a84214e | 968 | bool old_postcopy_cap; |
d7651f15 | 969 | MigrationIncomingState *mis = migration_incoming_get_current(); |
00458433 | 970 | |
4a84214e | 971 | old_postcopy_cap = cap_list[MIGRATION_CAPABILITY_POSTCOPY_RAM]; |
00458433 OW |
972 | |
973 | for (cap = params; cap; cap = cap->next) { | |
4a84214e PX |
974 | cap_list[cap->value->capability] = cap->value->state; |
975 | } | |
976 | ||
ed1701c6 | 977 | #ifndef CONFIG_LIVE_BLOCK_MIGRATION |
4a84214e PX |
978 | if (cap_list[MIGRATION_CAPABILITY_BLOCK]) { |
979 | error_setg(errp, "QEMU compiled without old-style (blk/-b, inc/-i) " | |
980 | "block migration"); | |
981 | error_append_hint(errp, "Use drive_mirror+NBD instead.\n"); | |
982 | return false; | |
00458433 | 983 | } |
4a84214e | 984 | #endif |
53dd370c | 985 | |
7e934f5b ZC |
986 | #ifndef CONFIG_REPLICATION |
987 | if (cap_list[MIGRATION_CAPABILITY_X_COLO]) { | |
988 | error_setg(errp, "QEMU compiled without replication module" | |
989 | " can't enable COLO"); | |
990 | error_append_hint(errp, "Please enable replication before COLO.\n"); | |
991 | return false; | |
992 | } | |
993 | #endif | |
994 | ||
4a84214e PX |
995 | if (cap_list[MIGRATION_CAPABILITY_POSTCOPY_RAM]) { |
996 | if (cap_list[MIGRATION_CAPABILITY_COMPRESS]) { | |
53dd370c DDAG |
997 | /* The decompression threads asynchronously write into RAM |
998 | * rather than use the atomic copies needed to avoid | |
999 | * userfaulting. It should be possible to fix the decompression | |
1000 | * threads for compatibility in future. | |
1001 | */ | |
4a84214e PX |
1002 | error_setg(errp, "Postcopy is not currently compatible " |
1003 | "with compression"); | |
1004 | return false; | |
53dd370c | 1005 | } |
4a84214e | 1006 | |
096631bd DDAG |
1007 | /* This check is reasonably expensive, so only when it's being |
1008 | * set the first time, also it's only the destination that needs | |
1009 | * special support. | |
1010 | */ | |
1011 | if (!old_postcopy_cap && runstate_check(RUN_STATE_INMIGRATE) && | |
d7651f15 | 1012 | !postcopy_ram_supported_by_host(mis)) { |
096631bd DDAG |
1013 | /* postcopy_ram_supported_by_host will have emitted a more |
1014 | * detailed message | |
1015 | */ | |
4a84214e PX |
1016 | error_setg(errp, "Postcopy is not supported"); |
1017 | return false; | |
096631bd | 1018 | } |
18269069 YK |
1019 | |
1020 | if (cap_list[MIGRATION_CAPABILITY_X_IGNORE_SHARED]) { | |
1021 | error_setg(errp, "Postcopy is not compatible with ignore-shared"); | |
1022 | return false; | |
1023 | } | |
53dd370c | 1024 | } |
4a84214e PX |
1025 | |
1026 | return true; | |
1027 | } | |
1028 | ||
65ace060 AP |
1029 | static void fill_destination_migration_info(MigrationInfo *info) |
1030 | { | |
1031 | MigrationIncomingState *mis = migration_incoming_get_current(); | |
1032 | ||
9aca82ba JQ |
1033 | if (mis->socket_address_list) { |
1034 | info->has_socket_address = true; | |
1035 | info->socket_address = | |
1036 | QAPI_CLONE(SocketAddressList, mis->socket_address_list); | |
1037 | } | |
1038 | ||
65ace060 AP |
1039 | switch (mis->state) { |
1040 | case MIGRATION_STATUS_NONE: | |
1041 | return; | |
1042 | break; | |
1043 | case MIGRATION_STATUS_SETUP: | |
1044 | case MIGRATION_STATUS_CANCELLING: | |
1045 | case MIGRATION_STATUS_CANCELLED: | |
1046 | case MIGRATION_STATUS_ACTIVE: | |
1047 | case MIGRATION_STATUS_POSTCOPY_ACTIVE: | |
3c9928d9 PX |
1048 | case MIGRATION_STATUS_POSTCOPY_PAUSED: |
1049 | case MIGRATION_STATUS_POSTCOPY_RECOVER: | |
65ace060 AP |
1050 | case MIGRATION_STATUS_FAILED: |
1051 | case MIGRATION_STATUS_COLO: | |
1052 | info->has_status = true; | |
1053 | break; | |
1054 | case MIGRATION_STATUS_COMPLETED: | |
1055 | info->has_status = true; | |
1056 | fill_destination_postcopy_migration_info(info); | |
1057 | break; | |
1058 | } | |
1059 | info->status = mis->state; | |
1060 | } | |
1061 | ||
1062 | MigrationInfo *qmp_query_migrate(Error **errp) | |
1063 | { | |
1064 | MigrationInfo *info = g_malloc0(sizeof(*info)); | |
1065 | ||
1066 | fill_destination_migration_info(info); | |
1067 | fill_source_migration_info(info); | |
1068 | ||
1069 | return info; | |
1070 | } | |
1071 | ||
4a84214e PX |
1072 | void qmp_migrate_set_capabilities(MigrationCapabilityStatusList *params, |
1073 | Error **errp) | |
1074 | { | |
1075 | MigrationState *s = migrate_get_current(); | |
1076 | MigrationCapabilityStatusList *cap; | |
dd0ee30c | 1077 | bool cap_list[MIGRATION_CAPABILITY__MAX]; |
4a84214e PX |
1078 | |
1079 | if (migration_is_setup_or_active(s->state)) { | |
1080 | error_setg(errp, QERR_MIGRATION_ACTIVE); | |
1081 | return; | |
1082 | } | |
1083 | ||
dd0ee30c PX |
1084 | memcpy(cap_list, s->enabled_capabilities, sizeof(cap_list)); |
1085 | if (!migrate_caps_check(cap_list, params, errp)) { | |
4a84214e PX |
1086 | return; |
1087 | } | |
1088 | ||
1089 | for (cap = params; cap; cap = cap->next) { | |
1090 | s->enabled_capabilities[cap->value->capability] = cap->value->state; | |
1091 | } | |
00458433 OW |
1092 | } |
1093 | ||
16d063bc PX |
1094 | /* |
1095 | * Check whether the parameters are valid. Error will be put into errp | |
1096 | * (if provided). Return true if valid, otherwise false. | |
1097 | */ | |
1098 | static bool migrate_params_check(MigrationParameters *params, Error **errp) | |
85de8323 | 1099 | { |
7f375e04 | 1100 | if (params->has_compress_level && |
741d4086 | 1101 | (params->compress_level > 9)) { |
c6bd8c70 MA |
1102 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "compress_level", |
1103 | "is invalid, it should be in the range of 0 to 9"); | |
16d063bc | 1104 | return false; |
85de8323 | 1105 | } |
16d063bc | 1106 | |
741d4086 | 1107 | if (params->has_compress_threads && (params->compress_threads < 1)) { |
c6bd8c70 MA |
1108 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, |
1109 | "compress_threads", | |
1110 | "is invalid, it should be in the range of 1 to 255"); | |
16d063bc | 1111 | return false; |
85de8323 | 1112 | } |
16d063bc | 1113 | |
741d4086 | 1114 | if (params->has_decompress_threads && (params->decompress_threads < 1)) { |
c6bd8c70 MA |
1115 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, |
1116 | "decompress_threads", | |
1117 | "is invalid, it should be in the range of 1 to 255"); | |
16d063bc | 1118 | return false; |
85de8323 | 1119 | } |
16d063bc | 1120 | |
7f375e04 EB |
1121 | if (params->has_cpu_throttle_initial && |
1122 | (params->cpu_throttle_initial < 1 || | |
1123 | params->cpu_throttle_initial > 99)) { | |
1626fee3 | 1124 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, |
d85a31d1 | 1125 | "cpu_throttle_initial", |
1626fee3 | 1126 | "an integer in the range of 1 to 99"); |
16d063bc | 1127 | return false; |
1626fee3 | 1128 | } |
16d063bc | 1129 | |
7f375e04 EB |
1130 | if (params->has_cpu_throttle_increment && |
1131 | (params->cpu_throttle_increment < 1 || | |
1132 | params->cpu_throttle_increment > 99)) { | |
1626fee3 | 1133 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, |
d85a31d1 | 1134 | "cpu_throttle_increment", |
1626fee3 | 1135 | "an integer in the range of 1 to 99"); |
16d063bc | 1136 | return false; |
1626fee3 | 1137 | } |
16d063bc | 1138 | |
741d4086 | 1139 | if (params->has_max_bandwidth && (params->max_bandwidth > SIZE_MAX)) { |
2ff30257 AA |
1140 | error_setg(errp, "Parameter 'max_bandwidth' expects an integer in the" |
1141 | " range of 0 to %zu bytes/second", SIZE_MAX); | |
16d063bc | 1142 | return false; |
2ff30257 | 1143 | } |
16d063bc | 1144 | |
2ff30257 | 1145 | if (params->has_downtime_limit && |
741d4086 | 1146 | (params->downtime_limit > MAX_MIGRATE_DOWNTIME)) { |
87c9cc1c DHB |
1147 | error_setg(errp, "Parameter 'downtime_limit' expects an integer in " |
1148 | "the range of 0 to %d milliseconds", | |
1149 | MAX_MIGRATE_DOWNTIME); | |
16d063bc | 1150 | return false; |
2ff30257 | 1151 | } |
16d063bc | 1152 | |
741d4086 JQ |
1153 | /* x_checkpoint_delay is now always positive */ |
1154 | ||
1155 | if (params->has_x_multifd_channels && (params->x_multifd_channels < 1)) { | |
4075fb1c JQ |
1156 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, |
1157 | "multifd_channels", | |
1158 | "is invalid, it should be in the range of 1 to 255"); | |
1159 | return false; | |
1160 | } | |
0fb86605 | 1161 | if (params->has_x_multifd_page_count && |
741d4086 JQ |
1162 | (params->x_multifd_page_count < 1 || |
1163 | params->x_multifd_page_count > 10000)) { | |
0fb86605 JQ |
1164 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, |
1165 | "multifd_page_count", | |
1166 | "is invalid, it should be in the range of 1 to 10000"); | |
1167 | return false; | |
1168 | } | |
16d063bc | 1169 | |
73af8dd8 JQ |
1170 | if (params->has_xbzrle_cache_size && |
1171 | (params->xbzrle_cache_size < qemu_target_page_size() || | |
1172 | !is_power_of_2(params->xbzrle_cache_size))) { | |
1173 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, | |
1174 | "xbzrle_cache_size", | |
1175 | "is invalid, it should be bigger than target page size" | |
1176 | " and a power of two"); | |
1177 | return false; | |
1178 | } | |
1179 | ||
4cbc9c7f LQ |
1180 | if (params->has_max_cpu_throttle && |
1181 | (params->max_cpu_throttle < params->cpu_throttle_initial || | |
1182 | params->max_cpu_throttle > 99)) { | |
1183 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, | |
1184 | "max_cpu_throttle", | |
1185 | "an integer in the range of cpu_throttle_initial to 99"); | |
1186 | return false; | |
1187 | } | |
1188 | ||
ee3d96ba DDAG |
1189 | if (params->has_announce_initial && |
1190 | params->announce_initial > 100000) { | |
1191 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, | |
1192 | "announce_initial", | |
1193 | "is invalid, it must be less than 100000 ms"); | |
1194 | return false; | |
1195 | } | |
1196 | if (params->has_announce_max && | |
1197 | params->announce_max > 100000) { | |
1198 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, | |
1199 | "announce_max", | |
1200 | "is invalid, it must be less than 100000 ms"); | |
1201 | return false; | |
1202 | } | |
1203 | if (params->has_announce_rounds && | |
1204 | params->announce_rounds > 1000) { | |
1205 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, | |
1206 | "announce_rounds", | |
1207 | "is invalid, it must be in the range of 0 to 1000"); | |
1208 | return false; | |
1209 | } | |
1210 | if (params->has_announce_step && | |
1211 | (params->announce_step < 1 || | |
1212 | params->announce_step > 10000)) { | |
1213 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, | |
1214 | "announce_step", | |
1215 | "is invalid, it must be in the range of 1 to 10000 ms"); | |
1216 | return false; | |
1217 | } | |
16d063bc PX |
1218 | return true; |
1219 | } | |
1220 | ||
1bda8b3c MA |
1221 | static void migrate_params_test_apply(MigrateSetParameters *params, |
1222 | MigrationParameters *dest) | |
1223 | { | |
1224 | *dest = migrate_get_current()->parameters; | |
1225 | ||
1226 | /* TODO use QAPI_CLONE() instead of duplicating it inline */ | |
1227 | ||
1228 | if (params->has_compress_level) { | |
1229 | dest->compress_level = params->compress_level; | |
1230 | } | |
1231 | ||
1232 | if (params->has_compress_threads) { | |
1233 | dest->compress_threads = params->compress_threads; | |
1234 | } | |
1235 | ||
1d58872a XG |
1236 | if (params->has_compress_wait_thread) { |
1237 | dest->compress_wait_thread = params->compress_wait_thread; | |
1238 | } | |
1239 | ||
1bda8b3c MA |
1240 | if (params->has_decompress_threads) { |
1241 | dest->decompress_threads = params->decompress_threads; | |
1242 | } | |
1243 | ||
1244 | if (params->has_cpu_throttle_initial) { | |
1245 | dest->cpu_throttle_initial = params->cpu_throttle_initial; | |
1246 | } | |
1247 | ||
1248 | if (params->has_cpu_throttle_increment) { | |
1249 | dest->cpu_throttle_increment = params->cpu_throttle_increment; | |
1250 | } | |
1251 | ||
1252 | if (params->has_tls_creds) { | |
01fa5598 MA |
1253 | assert(params->tls_creds->type == QTYPE_QSTRING); |
1254 | dest->tls_creds = g_strdup(params->tls_creds->u.s); | |
1bda8b3c MA |
1255 | } |
1256 | ||
1257 | if (params->has_tls_hostname) { | |
01fa5598 MA |
1258 | assert(params->tls_hostname->type == QTYPE_QSTRING); |
1259 | dest->tls_hostname = g_strdup(params->tls_hostname->u.s); | |
1bda8b3c MA |
1260 | } |
1261 | ||
1262 | if (params->has_max_bandwidth) { | |
1263 | dest->max_bandwidth = params->max_bandwidth; | |
1264 | } | |
1265 | ||
1266 | if (params->has_downtime_limit) { | |
1267 | dest->downtime_limit = params->downtime_limit; | |
1268 | } | |
1269 | ||
1270 | if (params->has_x_checkpoint_delay) { | |
1271 | dest->x_checkpoint_delay = params->x_checkpoint_delay; | |
1272 | } | |
1273 | ||
1274 | if (params->has_block_incremental) { | |
1275 | dest->block_incremental = params->block_incremental; | |
1276 | } | |
5e7577a1 JQ |
1277 | if (params->has_x_multifd_channels) { |
1278 | dest->x_multifd_channels = params->x_multifd_channels; | |
1279 | } | |
1280 | if (params->has_x_multifd_page_count) { | |
1281 | dest->x_multifd_page_count = params->x_multifd_page_count; | |
1282 | } | |
73af8dd8 JQ |
1283 | if (params->has_xbzrle_cache_size) { |
1284 | dest->xbzrle_cache_size = params->xbzrle_cache_size; | |
1285 | } | |
7e555c6c DDAG |
1286 | if (params->has_max_postcopy_bandwidth) { |
1287 | dest->max_postcopy_bandwidth = params->max_postcopy_bandwidth; | |
1288 | } | |
4cbc9c7f LQ |
1289 | if (params->has_max_cpu_throttle) { |
1290 | dest->max_cpu_throttle = params->max_cpu_throttle; | |
1291 | } | |
ee3d96ba DDAG |
1292 | if (params->has_announce_initial) { |
1293 | dest->announce_initial = params->announce_initial; | |
1294 | } | |
1295 | if (params->has_announce_max) { | |
1296 | dest->announce_max = params->announce_max; | |
1297 | } | |
1298 | if (params->has_announce_rounds) { | |
1299 | dest->announce_rounds = params->announce_rounds; | |
1300 | } | |
1301 | if (params->has_announce_step) { | |
1302 | dest->announce_step = params->announce_step; | |
1303 | } | |
1bda8b3c MA |
1304 | } |
1305 | ||
73af8dd8 | 1306 | static void migrate_params_apply(MigrateSetParameters *params, Error **errp) |
16d063bc PX |
1307 | { |
1308 | MigrationState *s = migrate_get_current(); | |
1309 | ||
e87fae4c MA |
1310 | /* TODO use QAPI_CLONE() instead of duplicating it inline */ |
1311 | ||
7f375e04 EB |
1312 | if (params->has_compress_level) { |
1313 | s->parameters.compress_level = params->compress_level; | |
85de8323 | 1314 | } |
476c72aa | 1315 | |
7f375e04 EB |
1316 | if (params->has_compress_threads) { |
1317 | s->parameters.compress_threads = params->compress_threads; | |
85de8323 | 1318 | } |
476c72aa | 1319 | |
1d58872a XG |
1320 | if (params->has_compress_wait_thread) { |
1321 | s->parameters.compress_wait_thread = params->compress_wait_thread; | |
1322 | } | |
1323 | ||
7f375e04 EB |
1324 | if (params->has_decompress_threads) { |
1325 | s->parameters.decompress_threads = params->decompress_threads; | |
85de8323 | 1326 | } |
476c72aa | 1327 | |
7f375e04 EB |
1328 | if (params->has_cpu_throttle_initial) { |
1329 | s->parameters.cpu_throttle_initial = params->cpu_throttle_initial; | |
1626fee3 | 1330 | } |
476c72aa | 1331 | |
7f375e04 EB |
1332 | if (params->has_cpu_throttle_increment) { |
1333 | s->parameters.cpu_throttle_increment = params->cpu_throttle_increment; | |
1626fee3 | 1334 | } |
476c72aa | 1335 | |
7f375e04 | 1336 | if (params->has_tls_creds) { |
69ef1f36 | 1337 | g_free(s->parameters.tls_creds); |
01fa5598 MA |
1338 | assert(params->tls_creds->type == QTYPE_QSTRING); |
1339 | s->parameters.tls_creds = g_strdup(params->tls_creds->u.s); | |
69ef1f36 | 1340 | } |
476c72aa | 1341 | |
7f375e04 | 1342 | if (params->has_tls_hostname) { |
69ef1f36 | 1343 | g_free(s->parameters.tls_hostname); |
01fa5598 MA |
1344 | assert(params->tls_hostname->type == QTYPE_QSTRING); |
1345 | s->parameters.tls_hostname = g_strdup(params->tls_hostname->u.s); | |
69ef1f36 | 1346 | } |
476c72aa | 1347 | |
2ff30257 AA |
1348 | if (params->has_max_bandwidth) { |
1349 | s->parameters.max_bandwidth = params->max_bandwidth; | |
1350 | if (s->to_dst_file) { | |
1351 | qemu_file_set_rate_limit(s->to_dst_file, | |
1352 | s->parameters.max_bandwidth / XFER_LIMIT_RATIO); | |
1353 | } | |
1354 | } | |
476c72aa | 1355 | |
2ff30257 AA |
1356 | if (params->has_downtime_limit) { |
1357 | s->parameters.downtime_limit = params->downtime_limit; | |
1358 | } | |
68b53591 HZ |
1359 | |
1360 | if (params->has_x_checkpoint_delay) { | |
1361 | s->parameters.x_checkpoint_delay = params->x_checkpoint_delay; | |
479125d5 HZ |
1362 | if (migration_in_colo_state()) { |
1363 | colo_checkpoint_notify(s); | |
1364 | } | |
68b53591 | 1365 | } |
476c72aa | 1366 | |
2833c59b JQ |
1367 | if (params->has_block_incremental) { |
1368 | s->parameters.block_incremental = params->block_incremental; | |
1369 | } | |
4075fb1c JQ |
1370 | if (params->has_x_multifd_channels) { |
1371 | s->parameters.x_multifd_channels = params->x_multifd_channels; | |
1372 | } | |
0fb86605 JQ |
1373 | if (params->has_x_multifd_page_count) { |
1374 | s->parameters.x_multifd_page_count = params->x_multifd_page_count; | |
1375 | } | |
73af8dd8 JQ |
1376 | if (params->has_xbzrle_cache_size) { |
1377 | s->parameters.xbzrle_cache_size = params->xbzrle_cache_size; | |
1378 | xbzrle_cache_resize(params->xbzrle_cache_size, errp); | |
1379 | } | |
7e555c6c DDAG |
1380 | if (params->has_max_postcopy_bandwidth) { |
1381 | s->parameters.max_postcopy_bandwidth = params->max_postcopy_bandwidth; | |
1382 | } | |
4cbc9c7f LQ |
1383 | if (params->has_max_cpu_throttle) { |
1384 | s->parameters.max_cpu_throttle = params->max_cpu_throttle; | |
1385 | } | |
ee3d96ba DDAG |
1386 | if (params->has_announce_initial) { |
1387 | s->parameters.announce_initial = params->announce_initial; | |
1388 | } | |
1389 | if (params->has_announce_max) { | |
1390 | s->parameters.announce_max = params->announce_max; | |
1391 | } | |
1392 | if (params->has_announce_rounds) { | |
1393 | s->parameters.announce_rounds = params->announce_rounds; | |
1394 | } | |
1395 | if (params->has_announce_step) { | |
1396 | s->parameters.announce_step = params->announce_step; | |
1397 | } | |
85de8323 LL |
1398 | } |
1399 | ||
1bda8b3c | 1400 | void qmp_migrate_set_parameters(MigrateSetParameters *params, Error **errp) |
476c72aa | 1401 | { |
1bda8b3c MA |
1402 | MigrationParameters tmp; |
1403 | ||
01fa5598 MA |
1404 | /* TODO Rewrite "" to null instead */ |
1405 | if (params->has_tls_creds | |
1406 | && params->tls_creds->type == QTYPE_QNULL) { | |
cb3e7f08 | 1407 | qobject_unref(params->tls_creds->u.n); |
01fa5598 MA |
1408 | params->tls_creds->type = QTYPE_QSTRING; |
1409 | params->tls_creds->u.s = strdup(""); | |
1410 | } | |
1411 | /* TODO Rewrite "" to null instead */ | |
1412 | if (params->has_tls_hostname | |
1413 | && params->tls_hostname->type == QTYPE_QNULL) { | |
cb3e7f08 | 1414 | qobject_unref(params->tls_hostname->u.n); |
01fa5598 MA |
1415 | params->tls_hostname->type = QTYPE_QSTRING; |
1416 | params->tls_hostname->u.s = strdup(""); | |
1417 | } | |
1418 | ||
1bda8b3c MA |
1419 | migrate_params_test_apply(params, &tmp); |
1420 | ||
1421 | if (!migrate_params_check(&tmp, errp)) { | |
476c72aa PX |
1422 | /* Invalid parameter */ |
1423 | return; | |
1424 | } | |
1425 | ||
73af8dd8 | 1426 | migrate_params_apply(params, errp); |
476c72aa PX |
1427 | } |
1428 | ||
2594f56d | 1429 | |
4886a1bc DDAG |
1430 | void qmp_migrate_start_postcopy(Error **errp) |
1431 | { | |
1432 | MigrationState *s = migrate_get_current(); | |
1433 | ||
16b0fd32 | 1434 | if (!migrate_postcopy()) { |
a54d340b | 1435 | error_setg(errp, "Enable postcopy with migrate_set_capability before" |
4886a1bc DDAG |
1436 | " the start of migration"); |
1437 | return; | |
1438 | } | |
1439 | ||
1440 | if (s->state == MIGRATION_STATUS_NONE) { | |
1441 | error_setg(errp, "Postcopy must be started after migration has been" | |
1442 | " started"); | |
1443 | return; | |
1444 | } | |
1445 | /* | |
1446 | * we don't error if migration has finished since that would be racy | |
1447 | * with issuing this command. | |
1448 | */ | |
1449 | atomic_set(&s->start_postcopy, true); | |
1450 | } | |
1451 | ||
065e2813 AL |
1452 | /* shared migration helpers */ |
1453 | ||
48781e5b | 1454 | void migrate_set_state(int *state, int old_state, int new_state) |
51cf4c1a | 1455 | { |
a31fedee | 1456 | assert(new_state < MIGRATION_STATUS__MAX); |
48781e5b | 1457 | if (atomic_cmpxchg(state, old_state, new_state) == old_state) { |
a31fedee | 1458 | trace_migrate_set_state(MigrationStatus_str(new_state)); |
b05dc723 | 1459 | migrate_generate_event(new_state); |
51cf4c1a Z |
1460 | } |
1461 | } | |
1462 | ||
4e4a3d3a PX |
1463 | static MigrationCapabilityStatusList *migrate_cap_add( |
1464 | MigrationCapabilityStatusList *list, | |
1465 | MigrationCapability index, | |
1466 | bool state) | |
2833c59b JQ |
1467 | { |
1468 | MigrationCapabilityStatusList *cap; | |
1469 | ||
1470 | cap = g_new0(MigrationCapabilityStatusList, 1); | |
1471 | cap->value = g_new0(MigrationCapabilityStatus, 1); | |
4e4a3d3a PX |
1472 | cap->value->capability = index; |
1473 | cap->value->state = state; | |
1474 | cap->next = list; | |
1475 | ||
1476 | return cap; | |
1477 | } | |
1478 | ||
1479 | void migrate_set_block_enabled(bool value, Error **errp) | |
1480 | { | |
1481 | MigrationCapabilityStatusList *cap; | |
1482 | ||
1483 | cap = migrate_cap_add(NULL, MIGRATION_CAPABILITY_BLOCK, value); | |
2833c59b JQ |
1484 | qmp_migrate_set_capabilities(cap, errp); |
1485 | qapi_free_MigrationCapabilityStatusList(cap); | |
1486 | } | |
1487 | ||
1488 | static void migrate_set_block_incremental(MigrationState *s, bool value) | |
1489 | { | |
1490 | s->parameters.block_incremental = value; | |
1491 | } | |
1492 | ||
1493 | static void block_cleanup_parameters(MigrationState *s) | |
1494 | { | |
1495 | if (s->must_remove_block_options) { | |
1496 | /* setting to false can never fail */ | |
1497 | migrate_set_block_enabled(false, &error_abort); | |
1498 | migrate_set_block_incremental(s, false); | |
1499 | s->must_remove_block_options = false; | |
1500 | } | |
1501 | } | |
1502 | ||
bb1fadc4 | 1503 | static void migrate_fd_cleanup(void *opaque) |
065e2813 | 1504 | { |
bb1fadc4 PB |
1505 | MigrationState *s = opaque; |
1506 | ||
1507 | qemu_bh_delete(s->cleanup_bh); | |
1508 | s->cleanup_bh = NULL; | |
1509 | ||
0ceccd85 PX |
1510 | qemu_savevm_state_cleanup(); |
1511 | ||
89a02a9f | 1512 | if (s->to_dst_file) { |
62df066f | 1513 | QEMUFile *tmp; |
f986c3d2 | 1514 | |
9013dca5 | 1515 | trace_migrate_fd_cleanup(); |
404a7c05 | 1516 | qemu_mutex_unlock_iothread(); |
1d34e4bf DDAG |
1517 | if (s->migration_thread_running) { |
1518 | qemu_thread_join(&s->thread); | |
1519 | s->migration_thread_running = false; | |
1520 | } | |
404a7c05 PB |
1521 | qemu_mutex_lock_iothread(); |
1522 | ||
1398b2e3 | 1523 | multifd_save_cleanup(); |
62df066f PX |
1524 | qemu_mutex_lock(&s->qemu_file_lock); |
1525 | tmp = s->to_dst_file; | |
89a02a9f | 1526 | s->to_dst_file = NULL; |
62df066f PX |
1527 | qemu_mutex_unlock(&s->qemu_file_lock); |
1528 | /* | |
1529 | * Close the file handle without the lock to make sure the | |
1530 | * critical section won't block for long. | |
1531 | */ | |
1532 | qemu_fclose(tmp); | |
065e2813 AL |
1533 | } |
1534 | ||
9ec055ae DDAG |
1535 | assert((s->state != MIGRATION_STATUS_ACTIVE) && |
1536 | (s->state != MIGRATION_STATUS_POSTCOPY_ACTIVE)); | |
7a2c1721 | 1537 | |
94f5a437 | 1538 | if (s->state == MIGRATION_STATUS_CANCELLING) { |
48781e5b | 1539 | migrate_set_state(&s->state, MIGRATION_STATUS_CANCELLING, |
94f5a437 | 1540 | MIGRATION_STATUS_CANCELLED); |
7a2c1721 | 1541 | } |
a3fa1d78 | 1542 | |
87db1a7d JQ |
1543 | if (s->error) { |
1544 | /* It is used on info migrate. We can't free it */ | |
1545 | error_report_err(error_copy(s->error)); | |
1546 | } | |
a3fa1d78 | 1547 | notifier_list_notify(&migration_state_notifiers, s); |
2833c59b | 1548 | block_cleanup_parameters(s); |
065e2813 AL |
1549 | } |
1550 | ||
87db1a7d JQ |
1551 | void migrate_set_error(MigrationState *s, const Error *error) |
1552 | { | |
1553 | qemu_mutex_lock(&s->error_mutex); | |
1554 | if (!s->error) { | |
1555 | s->error = error_copy(error); | |
1556 | } | |
1557 | qemu_mutex_unlock(&s->error_mutex); | |
1558 | } | |
1559 | ||
d59ce6f3 | 1560 | void migrate_fd_error(MigrationState *s, const Error *error) |
065e2813 | 1561 | { |
25174055 | 1562 | trace_migrate_fd_error(error_get_pretty(error)); |
89a02a9f | 1563 | assert(s->to_dst_file == NULL); |
48781e5b HZ |
1564 | migrate_set_state(&s->state, MIGRATION_STATUS_SETUP, |
1565 | MIGRATION_STATUS_FAILED); | |
87db1a7d | 1566 | migrate_set_error(s, error); |
458cf28e JQ |
1567 | } |
1568 | ||
0edda1c4 | 1569 | static void migrate_fd_cancel(MigrationState *s) |
065e2813 | 1570 | { |
6f2b811a | 1571 | int old_state ; |
89a02a9f | 1572 | QEMUFile *f = migrate_get_current()->to_dst_file; |
9013dca5 | 1573 | trace_migrate_fd_cancel(); |
065e2813 | 1574 | |
70b20477 DDAG |
1575 | if (s->rp_state.from_dst_file) { |
1576 | /* shutdown the rp socket, so causing the rp thread to shutdown */ | |
1577 | qemu_file_shutdown(s->rp_state.from_dst_file); | |
1578 | } | |
1579 | ||
6f2b811a Z |
1580 | do { |
1581 | old_state = s->state; | |
f6844b99 | 1582 | if (!migration_is_setup_or_active(old_state)) { |
6f2b811a Z |
1583 | break; |
1584 | } | |
a7b36b48 DDAG |
1585 | /* If the migration is paused, kick it out of the pause */ |
1586 | if (old_state == MIGRATION_STATUS_PRE_SWITCHOVER) { | |
1587 | qemu_sem_post(&s->pause_sem); | |
1588 | } | |
48781e5b | 1589 | migrate_set_state(&s->state, old_state, MIGRATION_STATUS_CANCELLING); |
31194731 | 1590 | } while (s->state != MIGRATION_STATUS_CANCELLING); |
a26ba26e DDAG |
1591 | |
1592 | /* | |
1593 | * If we're unlucky the migration code might be stuck somewhere in a | |
1594 | * send/write while the network has failed and is waiting to timeout; | |
1595 | * if we've got shutdown(2) available then we can force it to quit. | |
1596 | * The outgoing qemu file gets closed in migrate_fd_cleanup that is | |
1597 | * called in a bh, so there is no race against this cancel. | |
1598 | */ | |
31194731 | 1599 | if (s->state == MIGRATION_STATUS_CANCELLING && f) { |
a26ba26e DDAG |
1600 | qemu_file_shutdown(f); |
1601 | } | |
1d2acc31 HZ |
1602 | if (s->state == MIGRATION_STATUS_CANCELLING && s->block_inactive) { |
1603 | Error *local_err = NULL; | |
1604 | ||
1605 | bdrv_invalidate_cache_all(&local_err); | |
1606 | if (local_err) { | |
1607 | error_report_err(local_err); | |
1608 | } else { | |
1609 | s->block_inactive = false; | |
1610 | } | |
1611 | } | |
065e2813 AL |
1612 | } |
1613 | ||
99a0db9b GH |
1614 | void add_migration_state_change_notifier(Notifier *notify) |
1615 | { | |
1616 | notifier_list_add(&migration_state_notifiers, notify); | |
1617 | } | |
1618 | ||
1619 | void remove_migration_state_change_notifier(Notifier *notify) | |
1620 | { | |
31552529 | 1621 | notifier_remove(notify); |
99a0db9b GH |
1622 | } |
1623 | ||
02edd2e7 | 1624 | bool migration_in_setup(MigrationState *s) |
afe2df69 | 1625 | { |
31194731 | 1626 | return s->state == MIGRATION_STATUS_SETUP; |
afe2df69 GH |
1627 | } |
1628 | ||
7073693b | 1629 | bool migration_has_finished(MigrationState *s) |
99a0db9b | 1630 | { |
31194731 | 1631 | return s->state == MIGRATION_STATUS_COMPLETED; |
99a0db9b | 1632 | } |
0edda1c4 | 1633 | |
afe2df69 GH |
1634 | bool migration_has_failed(MigrationState *s) |
1635 | { | |
31194731 HZ |
1636 | return (s->state == MIGRATION_STATUS_CANCELLED || |
1637 | s->state == MIGRATION_STATUS_FAILED); | |
afe2df69 GH |
1638 | } |
1639 | ||
5727309d | 1640 | bool migration_in_postcopy(void) |
9ec055ae | 1641 | { |
5727309d JQ |
1642 | MigrationState *s = migrate_get_current(); |
1643 | ||
9ec055ae DDAG |
1644 | return (s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE); |
1645 | } | |
1646 | ||
b82fc321 DDAG |
1647 | bool migration_in_postcopy_after_devices(MigrationState *s) |
1648 | { | |
5727309d | 1649 | return migration_in_postcopy() && s->postcopy_after_devices; |
b82fc321 DDAG |
1650 | } |
1651 | ||
fab35005 | 1652 | bool migration_is_idle(void) |
fe44dc91 | 1653 | { |
fab35005 | 1654 | MigrationState *s = migrate_get_current(); |
fe44dc91 AA |
1655 | |
1656 | switch (s->state) { | |
1657 | case MIGRATION_STATUS_NONE: | |
1658 | case MIGRATION_STATUS_CANCELLED: | |
1659 | case MIGRATION_STATUS_COMPLETED: | |
1660 | case MIGRATION_STATUS_FAILED: | |
1661 | return true; | |
1662 | case MIGRATION_STATUS_SETUP: | |
1663 | case MIGRATION_STATUS_CANCELLING: | |
1664 | case MIGRATION_STATUS_ACTIVE: | |
1665 | case MIGRATION_STATUS_POSTCOPY_ACTIVE: | |
1666 | case MIGRATION_STATUS_COLO: | |
31e06077 DDAG |
1667 | case MIGRATION_STATUS_PRE_SWITCHOVER: |
1668 | case MIGRATION_STATUS_DEVICE: | |
fe44dc91 AA |
1669 | return false; |
1670 | case MIGRATION_STATUS__MAX: | |
1671 | g_assert_not_reached(); | |
1672 | } | |
1673 | ||
1674 | return false; | |
1675 | } | |
1676 | ||
3e0c8050 | 1677 | void migrate_init(MigrationState *s) |
0edda1c4 | 1678 | { |
389775d1 DDAG |
1679 | /* |
1680 | * Reinitialise all migration state, except | |
1681 | * parameters/capabilities that the user set, and | |
1682 | * locks. | |
1683 | */ | |
1684 | s->bytes_xfer = 0; | |
1685 | s->xfer_limit = 0; | |
1686 | s->cleanup_bh = 0; | |
89a02a9f | 1687 | s->to_dst_file = NULL; |
389775d1 | 1688 | s->state = MIGRATION_STATUS_NONE; |
389775d1 DDAG |
1689 | s->rp_state.from_dst_file = NULL; |
1690 | s->rp_state.error = false; | |
1691 | s->mbps = 0.0; | |
aecbfe9c | 1692 | s->pages_per_second = 0.0; |
389775d1 DDAG |
1693 | s->downtime = 0; |
1694 | s->expected_downtime = 0; | |
389775d1 | 1695 | s->setup_time = 0; |
389775d1 | 1696 | s->start_postcopy = false; |
b82fc321 | 1697 | s->postcopy_after_devices = false; |
389775d1 | 1698 | s->migration_thread_running = false; |
d59ce6f3 DB |
1699 | error_free(s->error); |
1700 | s->error = NULL; | |
389775d1 | 1701 | |
48781e5b | 1702 | migrate_set_state(&s->state, MIGRATION_STATUS_NONE, MIGRATION_STATUS_SETUP); |
0edda1c4 | 1703 | |
4af246a3 PX |
1704 | s->start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); |
1705 | s->total_time = 0; | |
7287cbd4 | 1706 | s->vm_was_running = false; |
b15df1ae PX |
1707 | s->iteration_initial_bytes = 0; |
1708 | s->threshold_size = 0; | |
0edda1c4 | 1709 | } |
cab30143 | 1710 | |
fa2756b7 AL |
1711 | static GSList *migration_blockers; |
1712 | ||
fe44dc91 | 1713 | int migrate_add_blocker(Error *reason, Error **errp) |
fa2756b7 | 1714 | { |
3df663e5 | 1715 | if (migrate_get_current()->only_migratable) { |
4b576648 MA |
1716 | error_propagate_prepend(errp, error_copy(reason), |
1717 | "disallowing migration blocker " | |
1718 | "(--only_migratable) for: "); | |
b67b8c3a AA |
1719 | return -EACCES; |
1720 | } | |
1721 | ||
fab35005 | 1722 | if (migration_is_idle()) { |
fe44dc91 AA |
1723 | migration_blockers = g_slist_prepend(migration_blockers, reason); |
1724 | return 0; | |
1725 | } | |
1726 | ||
4b576648 MA |
1727 | error_propagate_prepend(errp, error_copy(reason), |
1728 | "disallowing migration blocker " | |
1729 | "(migration in progress) for: "); | |
fe44dc91 | 1730 | return -EBUSY; |
fa2756b7 AL |
1731 | } |
1732 | ||
1733 | void migrate_del_blocker(Error *reason) | |
1734 | { | |
1735 | migration_blockers = g_slist_remove(migration_blockers, reason); | |
1736 | } | |
1737 | ||
bf1ae1f4 DDAG |
1738 | void qmp_migrate_incoming(const char *uri, Error **errp) |
1739 | { | |
1740 | Error *local_err = NULL; | |
4debb5f5 | 1741 | static bool once = true; |
bf1ae1f4 DDAG |
1742 | |
1743 | if (!deferred_incoming) { | |
4debb5f5 | 1744 | error_setg(errp, "For use with '-incoming defer'"); |
bf1ae1f4 DDAG |
1745 | return; |
1746 | } | |
4debb5f5 DDAG |
1747 | if (!once) { |
1748 | error_setg(errp, "The incoming migration has already been started"); | |
1749 | } | |
bf1ae1f4 DDAG |
1750 | |
1751 | qemu_start_incoming_migration(uri, &local_err); | |
1752 | ||
1753 | if (local_err) { | |
1754 | error_propagate(errp, local_err); | |
1755 | return; | |
1756 | } | |
1757 | ||
4debb5f5 | 1758 | once = false; |
bf1ae1f4 DDAG |
1759 | } |
1760 | ||
02affd41 PX |
1761 | void qmp_migrate_recover(const char *uri, Error **errp) |
1762 | { | |
1763 | MigrationIncomingState *mis = migration_incoming_get_current(); | |
1764 | ||
1765 | if (mis->state != MIGRATION_STATUS_POSTCOPY_PAUSED) { | |
1766 | error_setg(errp, "Migrate recover can only be run " | |
1767 | "when postcopy is paused."); | |
1768 | return; | |
1769 | } | |
1770 | ||
1771 | if (atomic_cmpxchg(&mis->postcopy_recover_triggered, | |
1772 | false, true) == true) { | |
1773 | error_setg(errp, "Migrate recovery is triggered already"); | |
1774 | return; | |
1775 | } | |
1776 | ||
1777 | /* | |
1778 | * Note that this call will never start a real migration; it will | |
1779 | * only re-setup the migration stream and poke existing migration | |
1780 | * to continue using that newly established channel. | |
1781 | */ | |
1782 | qemu_start_incoming_migration(uri, errp); | |
1783 | } | |
1784 | ||
bfbf89c2 PX |
1785 | void qmp_migrate_pause(Error **errp) |
1786 | { | |
1787 | MigrationState *ms = migrate_get_current(); | |
1788 | MigrationIncomingState *mis = migration_incoming_get_current(); | |
1789 | int ret; | |
1790 | ||
1791 | if (ms->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) { | |
1792 | /* Source side, during postcopy */ | |
1793 | qemu_mutex_lock(&ms->qemu_file_lock); | |
1794 | ret = qemu_file_shutdown(ms->to_dst_file); | |
1795 | qemu_mutex_unlock(&ms->qemu_file_lock); | |
1796 | if (ret) { | |
1797 | error_setg(errp, "Failed to pause source migration"); | |
1798 | } | |
1799 | return; | |
1800 | } | |
1801 | ||
1802 | if (mis->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) { | |
1803 | ret = qemu_file_shutdown(mis->from_src_file); | |
1804 | if (ret) { | |
1805 | error_setg(errp, "Failed to pause destination migration"); | |
1806 | } | |
1807 | return; | |
1808 | } | |
1809 | ||
1810 | error_setg(errp, "migrate-pause is currently only supported " | |
1811 | "during postcopy-active state"); | |
1812 | } | |
1813 | ||
24f3902b GK |
1814 | bool migration_is_blocked(Error **errp) |
1815 | { | |
1816 | if (qemu_savevm_state_blocked(errp)) { | |
1817 | return true; | |
1818 | } | |
1819 | ||
1820 | if (migration_blockers) { | |
250561e1 | 1821 | error_propagate(errp, error_copy(migration_blockers->data)); |
24f3902b GK |
1822 | return true; |
1823 | } | |
1824 | ||
1825 | return false; | |
1826 | } | |
1827 | ||
d3e35b8f PX |
1828 | /* Returns true if continue to migrate, or false if error detected */ |
1829 | static bool migrate_prepare(MigrationState *s, bool blk, bool blk_inc, | |
1830 | bool resume, Error **errp) | |
cab30143 | 1831 | { |
be7059cd | 1832 | Error *local_err = NULL; |
d3e35b8f PX |
1833 | |
1834 | if (resume) { | |
1835 | if (s->state != MIGRATION_STATUS_POSTCOPY_PAUSED) { | |
1836 | error_setg(errp, "Cannot resume if there is no " | |
1837 | "paused migration"); | |
1838 | return false; | |
1839 | } | |
97ca211c PX |
1840 | |
1841 | /* | |
1842 | * Postcopy recovery won't work well with release-ram | |
1843 | * capability since release-ram will drop the page buffer as | |
1844 | * long as the page is put into the send buffer. So if there | |
1845 | * is a network failure happened, any page buffers that have | |
1846 | * not yet reached the destination VM but have already been | |
1847 | * sent from the source VM will be lost forever. Let's refuse | |
1848 | * the client from resuming such a postcopy migration. | |
1849 | * Luckily release-ram was designed to only be used when src | |
1850 | * and destination VMs are on the same host, so it should be | |
1851 | * fine. | |
1852 | */ | |
1853 | if (migrate_release_ram()) { | |
1854 | error_setg(errp, "Postcopy recovery cannot work " | |
1855 | "when release-ram capability is set"); | |
1856 | return false; | |
1857 | } | |
1858 | ||
d3e35b8f PX |
1859 | /* This is a resume, skip init status */ |
1860 | return true; | |
1861 | } | |
cab30143 | 1862 | |
f6844b99 | 1863 | if (migration_is_setup_or_active(s->state) || |
0b827d5e HZ |
1864 | s->state == MIGRATION_STATUS_CANCELLING || |
1865 | s->state == MIGRATION_STATUS_COLO) { | |
c6bd8c70 | 1866 | error_setg(errp, QERR_MIGRATION_ACTIVE); |
d3e35b8f | 1867 | return false; |
cab30143 | 1868 | } |
d3e35b8f | 1869 | |
ca99993a DDAG |
1870 | if (runstate_check(RUN_STATE_INMIGRATE)) { |
1871 | error_setg(errp, "Guest is waiting for an incoming migration"); | |
d3e35b8f | 1872 | return false; |
ca99993a DDAG |
1873 | } |
1874 | ||
24f3902b | 1875 | if (migration_is_blocked(errp)) { |
d3e35b8f | 1876 | return false; |
fa2756b7 AL |
1877 | } |
1878 | ||
d3e35b8f | 1879 | if (blk || blk_inc) { |
2833c59b JQ |
1880 | if (migrate_use_block() || migrate_use_block_incremental()) { |
1881 | error_setg(errp, "Command options are incompatible with " | |
1882 | "current migration capabilities"); | |
d3e35b8f | 1883 | return false; |
2833c59b JQ |
1884 | } |
1885 | migrate_set_block_enabled(true, &local_err); | |
1886 | if (local_err) { | |
1887 | error_propagate(errp, local_err); | |
d3e35b8f | 1888 | return false; |
2833c59b JQ |
1889 | } |
1890 | s->must_remove_block_options = true; | |
1891 | } | |
1892 | ||
d3e35b8f | 1893 | if (blk_inc) { |
2833c59b JQ |
1894 | migrate_set_block_incremental(s, true); |
1895 | } | |
1896 | ||
3e0c8050 | 1897 | migrate_init(s); |
cab30143 | 1898 | |
d3e35b8f PX |
1899 | return true; |
1900 | } | |
1901 | ||
1902 | void qmp_migrate(const char *uri, bool has_blk, bool blk, | |
1903 | bool has_inc, bool inc, bool has_detach, bool detach, | |
1904 | bool has_resume, bool resume, Error **errp) | |
1905 | { | |
1906 | Error *local_err = NULL; | |
1907 | MigrationState *s = migrate_get_current(); | |
1908 | const char *p; | |
1909 | ||
1910 | if (!migrate_prepare(s, has_blk && blk, has_inc && inc, | |
1911 | has_resume && resume, errp)) { | |
1912 | /* Error detected, put into errp */ | |
1913 | return; | |
1914 | } | |
1915 | ||
cab30143 | 1916 | if (strstart(uri, "tcp:", &p)) { |
f37afb5a | 1917 | tcp_start_outgoing_migration(s, p, &local_err); |
2da776db | 1918 | #ifdef CONFIG_RDMA |
41310c68 | 1919 | } else if (strstart(uri, "rdma:", &p)) { |
2da776db MH |
1920 | rdma_start_outgoing_migration(s, p, &local_err); |
1921 | #endif | |
cab30143 | 1922 | } else if (strstart(uri, "exec:", &p)) { |
f37afb5a | 1923 | exec_start_outgoing_migration(s, p, &local_err); |
cab30143 | 1924 | } else if (strstart(uri, "unix:", &p)) { |
f37afb5a | 1925 | unix_start_outgoing_migration(s, p, &local_err); |
cab30143 | 1926 | } else if (strstart(uri, "fd:", &p)) { |
f37afb5a | 1927 | fd_start_outgoing_migration(s, p, &local_err); |
99a0db9b | 1928 | } else { |
c6bd8c70 MA |
1929 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "uri", |
1930 | "a valid migration protocol"); | |
48781e5b HZ |
1931 | migrate_set_state(&s->state, MIGRATION_STATUS_SETUP, |
1932 | MIGRATION_STATUS_FAILED); | |
09576e74 | 1933 | block_cleanup_parameters(s); |
e1c37d0e | 1934 | return; |
cab30143 JQ |
1935 | } |
1936 | ||
f37afb5a | 1937 | if (local_err) { |
d59ce6f3 | 1938 | migrate_fd_error(s, local_err); |
f37afb5a | 1939 | error_propagate(errp, local_err); |
e1c37d0e | 1940 | return; |
1299c631 | 1941 | } |
cab30143 JQ |
1942 | } |
1943 | ||
6cdedb07 | 1944 | void qmp_migrate_cancel(Error **errp) |
cab30143 | 1945 | { |
17549e84 | 1946 | migrate_fd_cancel(migrate_get_current()); |
cab30143 JQ |
1947 | } |
1948 | ||
89cfc02c DDAG |
1949 | void qmp_migrate_continue(MigrationStatus state, Error **errp) |
1950 | { | |
1951 | MigrationState *s = migrate_get_current(); | |
1952 | if (s->state != state) { | |
1953 | error_setg(errp, "Migration not in expected state: %s", | |
1954 | MigrationStatus_str(s->state)); | |
1955 | return; | |
1956 | } | |
1957 | qemu_sem_post(&s->pause_sem); | |
1958 | } | |
1959 | ||
9e1ba4cc OW |
1960 | void qmp_migrate_set_cache_size(int64_t value, Error **errp) |
1961 | { | |
73af8dd8 JQ |
1962 | MigrateSetParameters p = { |
1963 | .has_xbzrle_cache_size = true, | |
1964 | .xbzrle_cache_size = value, | |
1965 | }; | |
c91e681a | 1966 | |
73af8dd8 | 1967 | qmp_migrate_set_parameters(&p, errp); |
9e1ba4cc OW |
1968 | } |
1969 | ||
1970 | int64_t qmp_query_migrate_cache_size(Error **errp) | |
1971 | { | |
1972 | return migrate_xbzrle_cache_size(); | |
1973 | } | |
1974 | ||
3dc85383 | 1975 | void qmp_migrate_set_speed(int64_t value, Error **errp) |
cab30143 | 1976 | { |
1bda8b3c | 1977 | MigrateSetParameters p = { |
2ff30257 AA |
1978 | .has_max_bandwidth = true, |
1979 | .max_bandwidth = value, | |
1980 | }; | |
cab30143 | 1981 | |
2ff30257 | 1982 | qmp_migrate_set_parameters(&p, errp); |
cab30143 JQ |
1983 | } |
1984 | ||
4f0a993b | 1985 | void qmp_migrate_set_downtime(double value, Error **errp) |
cab30143 | 1986 | { |
87c9cc1c DHB |
1987 | if (value < 0 || value > MAX_MIGRATE_DOWNTIME_SECONDS) { |
1988 | error_setg(errp, "Parameter 'downtime_limit' expects an integer in " | |
1989 | "the range of 0 to %d seconds", | |
1990 | MAX_MIGRATE_DOWNTIME_SECONDS); | |
1991 | return; | |
1992 | } | |
1993 | ||
2ff30257 AA |
1994 | value *= 1000; /* Convert to milliseconds */ |
1995 | value = MAX(0, MIN(INT64_MAX, value)); | |
1996 | ||
1bda8b3c | 1997 | MigrateSetParameters p = { |
2ff30257 AA |
1998 | .has_downtime_limit = true, |
1999 | .downtime_limit = value, | |
2000 | }; | |
2001 | ||
2002 | qmp_migrate_set_parameters(&p, errp); | |
99a0db9b | 2003 | } |
17ad9b35 | 2004 | |
53f09a10 PB |
2005 | bool migrate_release_ram(void) |
2006 | { | |
2007 | MigrationState *s; | |
2008 | ||
2009 | s = migrate_get_current(); | |
2010 | ||
2011 | return s->enabled_capabilities[MIGRATION_CAPABILITY_RELEASE_RAM]; | |
2012 | } | |
2013 | ||
53dd370c DDAG |
2014 | bool migrate_postcopy_ram(void) |
2015 | { | |
2016 | MigrationState *s; | |
2017 | ||
2018 | s = migrate_get_current(); | |
2019 | ||
32c3db5b | 2020 | return s->enabled_capabilities[MIGRATION_CAPABILITY_POSTCOPY_RAM]; |
53dd370c DDAG |
2021 | } |
2022 | ||
58110f0a VSO |
2023 | bool migrate_postcopy(void) |
2024 | { | |
dd6bb914 | 2025 | return migrate_postcopy_ram() || migrate_dirty_bitmaps(); |
58110f0a VSO |
2026 | } |
2027 | ||
bde1e2ec CV |
2028 | bool migrate_auto_converge(void) |
2029 | { | |
2030 | MigrationState *s; | |
2031 | ||
2032 | s = migrate_get_current(); | |
2033 | ||
2034 | return s->enabled_capabilities[MIGRATION_CAPABILITY_AUTO_CONVERGE]; | |
2035 | } | |
2036 | ||
323004a3 PL |
2037 | bool migrate_zero_blocks(void) |
2038 | { | |
2039 | MigrationState *s; | |
2040 | ||
2041 | s = migrate_get_current(); | |
2042 | ||
2043 | return s->enabled_capabilities[MIGRATION_CAPABILITY_ZERO_BLOCKS]; | |
2044 | } | |
2045 | ||
f22f928e AP |
2046 | bool migrate_postcopy_blocktime(void) |
2047 | { | |
2048 | MigrationState *s; | |
2049 | ||
2050 | s = migrate_get_current(); | |
2051 | ||
2052 | return s->enabled_capabilities[MIGRATION_CAPABILITY_POSTCOPY_BLOCKTIME]; | |
2053 | } | |
2054 | ||
8706d2d5 LL |
2055 | bool migrate_use_compression(void) |
2056 | { | |
dde4e694 LL |
2057 | MigrationState *s; |
2058 | ||
2059 | s = migrate_get_current(); | |
2060 | ||
2061 | return s->enabled_capabilities[MIGRATION_CAPABILITY_COMPRESS]; | |
8706d2d5 LL |
2062 | } |
2063 | ||
2064 | int migrate_compress_level(void) | |
2065 | { | |
2066 | MigrationState *s; | |
2067 | ||
2068 | s = migrate_get_current(); | |
2069 | ||
2594f56d | 2070 | return s->parameters.compress_level; |
8706d2d5 LL |
2071 | } |
2072 | ||
2073 | int migrate_compress_threads(void) | |
2074 | { | |
2075 | MigrationState *s; | |
2076 | ||
2077 | s = migrate_get_current(); | |
2078 | ||
2594f56d | 2079 | return s->parameters.compress_threads; |
8706d2d5 LL |
2080 | } |
2081 | ||
1d58872a XG |
2082 | int migrate_compress_wait_thread(void) |
2083 | { | |
2084 | MigrationState *s; | |
2085 | ||
2086 | s = migrate_get_current(); | |
2087 | ||
2088 | return s->parameters.compress_wait_thread; | |
2089 | } | |
2090 | ||
3fcb38c2 LL |
2091 | int migrate_decompress_threads(void) |
2092 | { | |
2093 | MigrationState *s; | |
2094 | ||
2095 | s = migrate_get_current(); | |
2096 | ||
2594f56d | 2097 | return s->parameters.decompress_threads; |
3fcb38c2 LL |
2098 | } |
2099 | ||
55efc8c2 VSO |
2100 | bool migrate_dirty_bitmaps(void) |
2101 | { | |
2102 | MigrationState *s; | |
2103 | ||
2104 | s = migrate_get_current(); | |
2105 | ||
2106 | return s->enabled_capabilities[MIGRATION_CAPABILITY_DIRTY_BITMAPS]; | |
2107 | } | |
2108 | ||
18269069 YK |
2109 | bool migrate_ignore_shared(void) |
2110 | { | |
2111 | MigrationState *s; | |
2112 | ||
2113 | s = migrate_get_current(); | |
2114 | ||
2115 | return s->enabled_capabilities[MIGRATION_CAPABILITY_X_IGNORE_SHARED]; | |
2116 | } | |
2117 | ||
b05dc723 JQ |
2118 | bool migrate_use_events(void) |
2119 | { | |
2120 | MigrationState *s; | |
2121 | ||
2122 | s = migrate_get_current(); | |
2123 | ||
2124 | return s->enabled_capabilities[MIGRATION_CAPABILITY_EVENTS]; | |
2125 | } | |
2126 | ||
30126bbf JQ |
2127 | bool migrate_use_multifd(void) |
2128 | { | |
2129 | MigrationState *s; | |
2130 | ||
2131 | s = migrate_get_current(); | |
2132 | ||
2133 | return s->enabled_capabilities[MIGRATION_CAPABILITY_X_MULTIFD]; | |
2134 | } | |
2135 | ||
93fbd031 DDAG |
2136 | bool migrate_pause_before_switchover(void) |
2137 | { | |
2138 | MigrationState *s; | |
2139 | ||
2140 | s = migrate_get_current(); | |
2141 | ||
2142 | return s->enabled_capabilities[ | |
2143 | MIGRATION_CAPABILITY_PAUSE_BEFORE_SWITCHOVER]; | |
2144 | } | |
2145 | ||
4075fb1c JQ |
2146 | int migrate_multifd_channels(void) |
2147 | { | |
2148 | MigrationState *s; | |
2149 | ||
2150 | s = migrate_get_current(); | |
2151 | ||
2152 | return s->parameters.x_multifd_channels; | |
2153 | } | |
2154 | ||
0fb86605 JQ |
2155 | int migrate_multifd_page_count(void) |
2156 | { | |
2157 | MigrationState *s; | |
2158 | ||
2159 | s = migrate_get_current(); | |
2160 | ||
2161 | return s->parameters.x_multifd_page_count; | |
2162 | } | |
2163 | ||
17ad9b35 OW |
2164 | int migrate_use_xbzrle(void) |
2165 | { | |
2166 | MigrationState *s; | |
2167 | ||
2168 | s = migrate_get_current(); | |
2169 | ||
2170 | return s->enabled_capabilities[MIGRATION_CAPABILITY_XBZRLE]; | |
2171 | } | |
2172 | ||
2173 | int64_t migrate_xbzrle_cache_size(void) | |
2174 | { | |
2175 | MigrationState *s; | |
2176 | ||
2177 | s = migrate_get_current(); | |
2178 | ||
73af8dd8 | 2179 | return s->parameters.xbzrle_cache_size; |
17ad9b35 | 2180 | } |
0d82d0e8 | 2181 | |
7e555c6c DDAG |
2182 | static int64_t migrate_max_postcopy_bandwidth(void) |
2183 | { | |
2184 | MigrationState *s; | |
2185 | ||
2186 | s = migrate_get_current(); | |
2187 | ||
2188 | return s->parameters.max_postcopy_bandwidth; | |
2189 | } | |
2190 | ||
2833c59b JQ |
2191 | bool migrate_use_block(void) |
2192 | { | |
2193 | MigrationState *s; | |
2194 | ||
2195 | s = migrate_get_current(); | |
2196 | ||
2197 | return s->enabled_capabilities[MIGRATION_CAPABILITY_BLOCK]; | |
2198 | } | |
2199 | ||
c788ada8 PX |
2200 | bool migrate_use_return_path(void) |
2201 | { | |
2202 | MigrationState *s; | |
2203 | ||
2204 | s = migrate_get_current(); | |
2205 | ||
2206 | return s->enabled_capabilities[MIGRATION_CAPABILITY_RETURN_PATH]; | |
2207 | } | |
2208 | ||
2833c59b JQ |
2209 | bool migrate_use_block_incremental(void) |
2210 | { | |
2211 | MigrationState *s; | |
2212 | ||
2213 | s = migrate_get_current(); | |
2214 | ||
2215 | return s->parameters.block_incremental; | |
2216 | } | |
2217 | ||
70b20477 DDAG |
2218 | /* migration thread support */ |
2219 | /* | |
2220 | * Something bad happened to the RP stream, mark an error | |
2221 | * The caller shall print or trace something to indicate why | |
2222 | */ | |
2223 | static void mark_source_rp_bad(MigrationState *s) | |
2224 | { | |
2225 | s->rp_state.error = true; | |
2226 | } | |
2227 | ||
2228 | static struct rp_cmd_args { | |
2229 | ssize_t len; /* -1 = variable */ | |
2230 | const char *name; | |
2231 | } rp_cmd_args[] = { | |
2232 | [MIG_RP_MSG_INVALID] = { .len = -1, .name = "INVALID" }, | |
2233 | [MIG_RP_MSG_SHUT] = { .len = 4, .name = "SHUT" }, | |
2234 | [MIG_RP_MSG_PONG] = { .len = 4, .name = "PONG" }, | |
1e2d90eb DDAG |
2235 | [MIG_RP_MSG_REQ_PAGES] = { .len = 12, .name = "REQ_PAGES" }, |
2236 | [MIG_RP_MSG_REQ_PAGES_ID] = { .len = -1, .name = "REQ_PAGES_ID" }, | |
a335debb | 2237 | [MIG_RP_MSG_RECV_BITMAP] = { .len = -1, .name = "RECV_BITMAP" }, |
13955b89 | 2238 | [MIG_RP_MSG_RESUME_ACK] = { .len = 4, .name = "RESUME_ACK" }, |
70b20477 DDAG |
2239 | [MIG_RP_MSG_MAX] = { .len = -1, .name = "MAX" }, |
2240 | }; | |
2241 | ||
1e2d90eb DDAG |
2242 | /* |
2243 | * Process a request for pages received on the return path, | |
2244 | * We're allowed to send more than requested (e.g. to round to our page size) | |
2245 | * and we don't need to send pages that have already been sent. | |
2246 | */ | |
2247 | static void migrate_handle_rp_req_pages(MigrationState *ms, const char* rbname, | |
2248 | ram_addr_t start, size_t len) | |
2249 | { | |
6c595cde DDAG |
2250 | long our_host_ps = getpagesize(); |
2251 | ||
1e2d90eb | 2252 | trace_migrate_handle_rp_req_pages(rbname, start, len); |
6c595cde DDAG |
2253 | |
2254 | /* | |
2255 | * Since we currently insist on matching page sizes, just sanity check | |
2256 | * we're being asked for whole host pages. | |
2257 | */ | |
2258 | if (start & (our_host_ps-1) || | |
2259 | (len & (our_host_ps-1))) { | |
2260 | error_report("%s: Misaligned page request, start: " RAM_ADDR_FMT | |
2261 | " len: %zd", __func__, start, len); | |
2262 | mark_source_rp_bad(ms); | |
2263 | return; | |
2264 | } | |
2265 | ||
96506894 | 2266 | if (ram_save_queue_pages(rbname, start, len)) { |
6c595cde DDAG |
2267 | mark_source_rp_bad(ms); |
2268 | } | |
1e2d90eb DDAG |
2269 | } |
2270 | ||
14b1742e PX |
2271 | /* Return true to retry, false to quit */ |
2272 | static bool postcopy_pause_return_path_thread(MigrationState *s) | |
2273 | { | |
2274 | trace_postcopy_pause_return_path(); | |
2275 | ||
2276 | qemu_sem_wait(&s->postcopy_pause_rp_sem); | |
2277 | ||
2278 | trace_postcopy_pause_return_path_continued(); | |
2279 | ||
2280 | return true; | |
2281 | } | |
2282 | ||
a335debb PX |
2283 | static int migrate_handle_rp_recv_bitmap(MigrationState *s, char *block_name) |
2284 | { | |
2285 | RAMBlock *block = qemu_ram_block_by_name(block_name); | |
2286 | ||
2287 | if (!block) { | |
2288 | error_report("%s: invalid block name '%s'", __func__, block_name); | |
2289 | return -EINVAL; | |
2290 | } | |
2291 | ||
2292 | /* Fetch the received bitmap and refresh the dirty bitmap */ | |
2293 | return ram_dirty_bitmap_reload(s, block); | |
2294 | } | |
2295 | ||
13955b89 PX |
2296 | static int migrate_handle_rp_resume_ack(MigrationState *s, uint32_t value) |
2297 | { | |
2298 | trace_source_return_path_thread_resume_ack(value); | |
2299 | ||
2300 | if (value != MIGRATION_RESUME_ACK_VALUE) { | |
2301 | error_report("%s: illegal resume_ack value %"PRIu32, | |
2302 | __func__, value); | |
2303 | return -1; | |
2304 | } | |
2305 | ||
2306 | /* Now both sides are active. */ | |
2307 | migrate_set_state(&s->state, MIGRATION_STATUS_POSTCOPY_RECOVER, | |
2308 | MIGRATION_STATUS_POSTCOPY_ACTIVE); | |
2309 | ||
94190696 PX |
2310 | /* Notify send thread that time to continue send pages */ |
2311 | qemu_sem_post(&s->rp_state.rp_sem); | |
13955b89 PX |
2312 | |
2313 | return 0; | |
2314 | } | |
2315 | ||
70b20477 DDAG |
2316 | /* |
2317 | * Handles messages sent on the return path towards the source VM | |
2318 | * | |
2319 | */ | |
2320 | static void *source_return_path_thread(void *opaque) | |
2321 | { | |
2322 | MigrationState *ms = opaque; | |
2323 | QEMUFile *rp = ms->rp_state.from_dst_file; | |
2324 | uint16_t header_len, header_type; | |
568b01ca | 2325 | uint8_t buf[512]; |
70b20477 | 2326 | uint32_t tmp32, sibling_error; |
1e2d90eb DDAG |
2327 | ram_addr_t start = 0; /* =0 to silence warning */ |
2328 | size_t len = 0, expected_len; | |
70b20477 DDAG |
2329 | int res; |
2330 | ||
2331 | trace_source_return_path_thread_entry(); | |
74637e6f | 2332 | rcu_register_thread(); |
14b1742e PX |
2333 | |
2334 | retry: | |
70b20477 DDAG |
2335 | while (!ms->rp_state.error && !qemu_file_get_error(rp) && |
2336 | migration_is_setup_or_active(ms->state)) { | |
2337 | trace_source_return_path_thread_loop_top(); | |
2338 | header_type = qemu_get_be16(rp); | |
2339 | header_len = qemu_get_be16(rp); | |
2340 | ||
7a9ddfbf PX |
2341 | if (qemu_file_get_error(rp)) { |
2342 | mark_source_rp_bad(ms); | |
2343 | goto out; | |
2344 | } | |
2345 | ||
70b20477 DDAG |
2346 | if (header_type >= MIG_RP_MSG_MAX || |
2347 | header_type == MIG_RP_MSG_INVALID) { | |
2348 | error_report("RP: Received invalid message 0x%04x length 0x%04x", | |
2349 | header_type, header_len); | |
2350 | mark_source_rp_bad(ms); | |
2351 | goto out; | |
2352 | } | |
2353 | ||
2354 | if ((rp_cmd_args[header_type].len != -1 && | |
2355 | header_len != rp_cmd_args[header_type].len) || | |
568b01ca | 2356 | header_len > sizeof(buf)) { |
70b20477 DDAG |
2357 | error_report("RP: Received '%s' message (0x%04x) with" |
2358 | "incorrect length %d expecting %zu", | |
2359 | rp_cmd_args[header_type].name, header_type, header_len, | |
2360 | (size_t)rp_cmd_args[header_type].len); | |
2361 | mark_source_rp_bad(ms); | |
2362 | goto out; | |
2363 | } | |
2364 | ||
2365 | /* We know we've got a valid header by this point */ | |
2366 | res = qemu_get_buffer(rp, buf, header_len); | |
2367 | if (res != header_len) { | |
2368 | error_report("RP: Failed reading data for message 0x%04x" | |
2369 | " read %d expected %d", | |
2370 | header_type, res, header_len); | |
2371 | mark_source_rp_bad(ms); | |
2372 | goto out; | |
2373 | } | |
2374 | ||
2375 | /* OK, we have the message and the data */ | |
2376 | switch (header_type) { | |
2377 | case MIG_RP_MSG_SHUT: | |
4d885131 | 2378 | sibling_error = ldl_be_p(buf); |
70b20477 DDAG |
2379 | trace_source_return_path_thread_shut(sibling_error); |
2380 | if (sibling_error) { | |
2381 | error_report("RP: Sibling indicated error %d", sibling_error); | |
2382 | mark_source_rp_bad(ms); | |
2383 | } | |
2384 | /* | |
2385 | * We'll let the main thread deal with closing the RP | |
2386 | * we could do a shutdown(2) on it, but we're the only user | |
2387 | * anyway, so there's nothing gained. | |
2388 | */ | |
2389 | goto out; | |
2390 | ||
2391 | case MIG_RP_MSG_PONG: | |
4d885131 | 2392 | tmp32 = ldl_be_p(buf); |
70b20477 DDAG |
2393 | trace_source_return_path_thread_pong(tmp32); |
2394 | break; | |
2395 | ||
1e2d90eb | 2396 | case MIG_RP_MSG_REQ_PAGES: |
4d885131 PM |
2397 | start = ldq_be_p(buf); |
2398 | len = ldl_be_p(buf + 8); | |
1e2d90eb DDAG |
2399 | migrate_handle_rp_req_pages(ms, NULL, start, len); |
2400 | break; | |
2401 | ||
2402 | case MIG_RP_MSG_REQ_PAGES_ID: | |
2403 | expected_len = 12 + 1; /* header + termination */ | |
2404 | ||
2405 | if (header_len >= expected_len) { | |
4d885131 PM |
2406 | start = ldq_be_p(buf); |
2407 | len = ldl_be_p(buf + 8); | |
1e2d90eb DDAG |
2408 | /* Now we expect an idstr */ |
2409 | tmp32 = buf[12]; /* Length of the following idstr */ | |
2410 | buf[13 + tmp32] = '\0'; | |
2411 | expected_len += tmp32; | |
2412 | } | |
2413 | if (header_len != expected_len) { | |
2414 | error_report("RP: Req_Page_id with length %d expecting %zd", | |
2415 | header_len, expected_len); | |
2416 | mark_source_rp_bad(ms); | |
2417 | goto out; | |
2418 | } | |
2419 | migrate_handle_rp_req_pages(ms, (char *)&buf[13], start, len); | |
2420 | break; | |
2421 | ||
a335debb PX |
2422 | case MIG_RP_MSG_RECV_BITMAP: |
2423 | if (header_len < 1) { | |
2424 | error_report("%s: missing block name", __func__); | |
2425 | mark_source_rp_bad(ms); | |
2426 | goto out; | |
2427 | } | |
2428 | /* Format: len (1B) + idstr (<255B). This ends the idstr. */ | |
2429 | buf[buf[0] + 1] = '\0'; | |
2430 | if (migrate_handle_rp_recv_bitmap(ms, (char *)(buf + 1))) { | |
2431 | mark_source_rp_bad(ms); | |
2432 | goto out; | |
2433 | } | |
2434 | break; | |
2435 | ||
13955b89 PX |
2436 | case MIG_RP_MSG_RESUME_ACK: |
2437 | tmp32 = ldl_be_p(buf); | |
2438 | if (migrate_handle_rp_resume_ack(ms, tmp32)) { | |
2439 | mark_source_rp_bad(ms); | |
2440 | goto out; | |
2441 | } | |
2442 | break; | |
2443 | ||
70b20477 DDAG |
2444 | default: |
2445 | break; | |
2446 | } | |
2447 | } | |
14b1742e PX |
2448 | |
2449 | out: | |
2450 | res = qemu_file_get_error(rp); | |
2451 | if (res) { | |
2452 | if (res == -EIO) { | |
2453 | /* | |
2454 | * Maybe there is something we can do: it looks like a | |
2455 | * network down issue, and we pause for a recovery. | |
2456 | */ | |
2457 | if (postcopy_pause_return_path_thread(ms)) { | |
2458 | /* Reload rp, reset the rest */ | |
0284a2a8 MAL |
2459 | if (rp != ms->rp_state.from_dst_file) { |
2460 | qemu_fclose(rp); | |
2461 | rp = ms->rp_state.from_dst_file; | |
2462 | } | |
14b1742e PX |
2463 | ms->rp_state.error = false; |
2464 | goto retry; | |
2465 | } | |
2466 | } | |
2467 | ||
70b20477 DDAG |
2468 | trace_source_return_path_thread_bad_end(); |
2469 | mark_source_rp_bad(ms); | |
2470 | } | |
2471 | ||
2472 | trace_source_return_path_thread_end(); | |
70b20477 DDAG |
2473 | ms->rp_state.from_dst_file = NULL; |
2474 | qemu_fclose(rp); | |
74637e6f | 2475 | rcu_unregister_thread(); |
70b20477 DDAG |
2476 | return NULL; |
2477 | } | |
2478 | ||
d3e35b8f PX |
2479 | static int open_return_path_on_source(MigrationState *ms, |
2480 | bool create_thread) | |
70b20477 DDAG |
2481 | { |
2482 | ||
89a02a9f | 2483 | ms->rp_state.from_dst_file = qemu_file_get_return_path(ms->to_dst_file); |
70b20477 DDAG |
2484 | if (!ms->rp_state.from_dst_file) { |
2485 | return -1; | |
2486 | } | |
2487 | ||
2488 | trace_open_return_path_on_source(); | |
d3e35b8f PX |
2489 | |
2490 | if (!create_thread) { | |
2491 | /* We're done */ | |
2492 | return 0; | |
2493 | } | |
2494 | ||
70b20477 DDAG |
2495 | qemu_thread_create(&ms->rp_state.rp_thread, "return path", |
2496 | source_return_path_thread, ms, QEMU_THREAD_JOINABLE); | |
2497 | ||
2498 | trace_open_return_path_on_source_continue(); | |
2499 | ||
2500 | return 0; | |
2501 | } | |
2502 | ||
70b20477 DDAG |
2503 | /* Returns 0 if the RP was ok, otherwise there was an error on the RP */ |
2504 | static int await_return_path_close_on_source(MigrationState *ms) | |
2505 | { | |
2506 | /* | |
2507 | * If this is a normal exit then the destination will send a SHUT and the | |
2508 | * rp_thread will exit, however if there's an error we need to cause | |
2509 | * it to exit. | |
2510 | */ | |
89a02a9f | 2511 | if (qemu_file_get_error(ms->to_dst_file) && ms->rp_state.from_dst_file) { |
70b20477 DDAG |
2512 | /* |
2513 | * shutdown(2), if we have it, will cause it to unblock if it's stuck | |
2514 | * waiting for the destination. | |
2515 | */ | |
2516 | qemu_file_shutdown(ms->rp_state.from_dst_file); | |
2517 | mark_source_rp_bad(ms); | |
2518 | } | |
2519 | trace_await_return_path_close_on_source_joining(); | |
2520 | qemu_thread_join(&ms->rp_state.rp_thread); | |
2521 | trace_await_return_path_close_on_source_close(); | |
2522 | return ms->rp_state.error; | |
2523 | } | |
2524 | ||
1d34e4bf DDAG |
2525 | /* |
2526 | * Switch from normal iteration to postcopy | |
2527 | * Returns non-0 on error | |
2528 | */ | |
7287cbd4 | 2529 | static int postcopy_start(MigrationState *ms) |
1d34e4bf DDAG |
2530 | { |
2531 | int ret; | |
61b67d47 DB |
2532 | QIOChannelBuffer *bioc; |
2533 | QEMUFile *fb; | |
1d34e4bf | 2534 | int64_t time_at_stop = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); |
7e555c6c | 2535 | int64_t bandwidth = migrate_max_postcopy_bandwidth(); |
ef8d6488 | 2536 | bool restart_block = false; |
0331c8ca DDAG |
2537 | int cur_state = MIGRATION_STATUS_ACTIVE; |
2538 | if (!migrate_pause_before_switchover()) { | |
2539 | migrate_set_state(&ms->state, MIGRATION_STATUS_ACTIVE, | |
2540 | MIGRATION_STATUS_POSTCOPY_ACTIVE); | |
2541 | } | |
1d34e4bf DDAG |
2542 | |
2543 | trace_postcopy_start(); | |
2544 | qemu_mutex_lock_iothread(); | |
2545 | trace_postcopy_start_set_run(); | |
2546 | ||
fb064112 | 2547 | qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER, NULL); |
1d34e4bf DDAG |
2548 | global_state_store(); |
2549 | ret = vm_stop_force_state(RUN_STATE_FINISH_MIGRATE); | |
76b1c7fe KW |
2550 | if (ret < 0) { |
2551 | goto fail; | |
2552 | } | |
1d34e4bf | 2553 | |
0331c8ca DDAG |
2554 | ret = migration_maybe_pause(ms, &cur_state, |
2555 | MIGRATION_STATUS_POSTCOPY_ACTIVE); | |
2556 | if (ret < 0) { | |
2557 | goto fail; | |
2558 | } | |
2559 | ||
76b1c7fe | 2560 | ret = bdrv_inactivate_all(); |
1d34e4bf DDAG |
2561 | if (ret < 0) { |
2562 | goto fail; | |
2563 | } | |
ef8d6488 | 2564 | restart_block = true; |
1d34e4bf | 2565 | |
1c0d249d DDAG |
2566 | /* |
2567 | * Cause any non-postcopiable, but iterative devices to | |
2568 | * send out their final data. | |
2569 | */ | |
a1fbe750 | 2570 | qemu_savevm_state_complete_precopy(ms->to_dst_file, true, false); |
1c0d249d | 2571 | |
1d34e4bf DDAG |
2572 | /* |
2573 | * in Finish migrate and with the io-lock held everything should | |
2574 | * be quiet, but we've potentially still got dirty pages and we | |
2575 | * need to tell the destination to throw any pages it's already received | |
2576 | * that are dirty | |
2577 | */ | |
58110f0a VSO |
2578 | if (migrate_postcopy_ram()) { |
2579 | if (ram_postcopy_send_discard_bitmap(ms)) { | |
2580 | error_report("postcopy send discard bitmap failed"); | |
2581 | goto fail; | |
2582 | } | |
1d34e4bf DDAG |
2583 | } |
2584 | ||
2585 | /* | |
2586 | * send rest of state - note things that are doing postcopy | |
2587 | * will notice we're in POSTCOPY_ACTIVE and not actually | |
2588 | * wrap their state up here | |
2589 | */ | |
7e555c6c DDAG |
2590 | /* 0 max-postcopy-bandwidth means unlimited */ |
2591 | if (!bandwidth) { | |
2592 | qemu_file_set_rate_limit(ms->to_dst_file, INT64_MAX); | |
2593 | } else { | |
2594 | qemu_file_set_rate_limit(ms->to_dst_file, bandwidth / XFER_LIMIT_RATIO); | |
2595 | } | |
58110f0a VSO |
2596 | if (migrate_postcopy_ram()) { |
2597 | /* Ping just for debugging, helps line traces up */ | |
2598 | qemu_savevm_send_ping(ms->to_dst_file, 2); | |
2599 | } | |
1d34e4bf DDAG |
2600 | |
2601 | /* | |
2602 | * While loading the device state we may trigger page transfer | |
2603 | * requests and the fd must be free to process those, and thus | |
2604 | * the destination must read the whole device state off the fd before | |
2605 | * it starts processing it. Unfortunately the ad-hoc migration format | |
2606 | * doesn't allow the destination to know the size to read without fully | |
2607 | * parsing it through each devices load-state code (especially the open | |
2608 | * coded devices that use get/put). | |
2609 | * So we wrap the device state up in a package with a length at the start; | |
2610 | * to do this we use a qemu_buf to hold the whole of the device state. | |
2611 | */ | |
61b67d47 | 2612 | bioc = qio_channel_buffer_new(4096); |
6f01f136 | 2613 | qio_channel_set_name(QIO_CHANNEL(bioc), "migration-postcopy-buffer"); |
61b67d47 DB |
2614 | fb = qemu_fopen_channel_output(QIO_CHANNEL(bioc)); |
2615 | object_unref(OBJECT(bioc)); | |
1d34e4bf | 2616 | |
c76201ab DDAG |
2617 | /* |
2618 | * Make sure the receiver can get incoming pages before we send the rest | |
2619 | * of the state | |
2620 | */ | |
2621 | qemu_savevm_send_postcopy_listen(fb); | |
2622 | ||
a1fbe750 | 2623 | qemu_savevm_state_complete_precopy(fb, false, false); |
58110f0a VSO |
2624 | if (migrate_postcopy_ram()) { |
2625 | qemu_savevm_send_ping(fb, 3); | |
2626 | } | |
1d34e4bf DDAG |
2627 | |
2628 | qemu_savevm_send_postcopy_run(fb); | |
2629 | ||
2630 | /* <><> end of stuff going into the package */ | |
1d34e4bf | 2631 | |
ef8d6488 DDAG |
2632 | /* Last point of recovery; as soon as we send the package the destination |
2633 | * can open devices and potentially start running. | |
2634 | * Lets just check again we've not got any errors. | |
2635 | */ | |
2636 | ret = qemu_file_get_error(ms->to_dst_file); | |
2637 | if (ret) { | |
2638 | error_report("postcopy_start: Migration stream errored (pre package)"); | |
2639 | goto fail_closefb; | |
2640 | } | |
2641 | ||
2642 | restart_block = false; | |
2643 | ||
1d34e4bf | 2644 | /* Now send that blob */ |
61b67d47 | 2645 | if (qemu_savevm_send_packaged(ms->to_dst_file, bioc->data, bioc->usage)) { |
1d34e4bf DDAG |
2646 | goto fail_closefb; |
2647 | } | |
2648 | qemu_fclose(fb); | |
b82fc321 DDAG |
2649 | |
2650 | /* Send a notify to give a chance for anything that needs to happen | |
2651 | * at the transition to postcopy and after the device state; in particular | |
2652 | * spice needs to trigger a transition now | |
2653 | */ | |
2654 | ms->postcopy_after_devices = true; | |
2655 | notifier_list_notify(&migration_state_notifiers, ms); | |
2656 | ||
1d34e4bf DDAG |
2657 | ms->downtime = qemu_clock_get_ms(QEMU_CLOCK_REALTIME) - time_at_stop; |
2658 | ||
2659 | qemu_mutex_unlock_iothread(); | |
2660 | ||
58110f0a VSO |
2661 | if (migrate_postcopy_ram()) { |
2662 | /* | |
2663 | * Although this ping is just for debug, it could potentially be | |
2664 | * used for getting a better measurement of downtime at the source. | |
2665 | */ | |
2666 | qemu_savevm_send_ping(ms->to_dst_file, 4); | |
2667 | } | |
1d34e4bf | 2668 | |
ced1c616 PB |
2669 | if (migrate_release_ram()) { |
2670 | ram_postcopy_migrated_memory_release(ms); | |
2671 | } | |
2672 | ||
89a02a9f | 2673 | ret = qemu_file_get_error(ms->to_dst_file); |
1d34e4bf DDAG |
2674 | if (ret) { |
2675 | error_report("postcopy_start: Migration stream errored"); | |
48781e5b | 2676 | migrate_set_state(&ms->state, MIGRATION_STATUS_POSTCOPY_ACTIVE, |
1d34e4bf DDAG |
2677 | MIGRATION_STATUS_FAILED); |
2678 | } | |
2679 | ||
2680 | return ret; | |
2681 | ||
2682 | fail_closefb: | |
2683 | qemu_fclose(fb); | |
2684 | fail: | |
48781e5b | 2685 | migrate_set_state(&ms->state, MIGRATION_STATUS_POSTCOPY_ACTIVE, |
1d34e4bf | 2686 | MIGRATION_STATUS_FAILED); |
ef8d6488 DDAG |
2687 | if (restart_block) { |
2688 | /* A failure happened early enough that we know the destination hasn't | |
2689 | * accessed block devices, so we're safe to recover. | |
2690 | */ | |
2691 | Error *local_err = NULL; | |
2692 | ||
2693 | bdrv_invalidate_cache_all(&local_err); | |
2694 | if (local_err) { | |
2695 | error_report_err(local_err); | |
2696 | } | |
2697 | } | |
1d34e4bf DDAG |
2698 | qemu_mutex_unlock_iothread(); |
2699 | return -1; | |
2700 | } | |
2701 | ||
e91d8951 DDAG |
2702 | /** |
2703 | * migration_maybe_pause: Pause if required to by | |
2704 | * migrate_pause_before_switchover called with the iothread locked | |
2705 | * Returns: 0 on success | |
2706 | */ | |
0331c8ca DDAG |
2707 | static int migration_maybe_pause(MigrationState *s, |
2708 | int *current_active_state, | |
2709 | int new_state) | |
e91d8951 DDAG |
2710 | { |
2711 | if (!migrate_pause_before_switchover()) { | |
2712 | return 0; | |
2713 | } | |
2714 | ||
2715 | /* Since leaving this state is not atomic with posting the semaphore | |
2716 | * it's possible that someone could have issued multiple migrate_continue | |
2717 | * and the semaphore is incorrectly positive at this point; | |
2718 | * the docs say it's undefined to reinit a semaphore that's already | |
2719 | * init'd, so use timedwait to eat up any existing posts. | |
2720 | */ | |
2721 | while (qemu_sem_timedwait(&s->pause_sem, 1) == 0) { | |
2722 | /* This block intentionally left blank */ | |
2723 | } | |
2724 | ||
2725 | qemu_mutex_unlock_iothread(); | |
2726 | migrate_set_state(&s->state, *current_active_state, | |
2727 | MIGRATION_STATUS_PRE_SWITCHOVER); | |
2728 | qemu_sem_wait(&s->pause_sem); | |
2729 | migrate_set_state(&s->state, MIGRATION_STATUS_PRE_SWITCHOVER, | |
0331c8ca DDAG |
2730 | new_state); |
2731 | *current_active_state = new_state; | |
e91d8951 DDAG |
2732 | qemu_mutex_lock_iothread(); |
2733 | ||
0331c8ca | 2734 | return s->state == new_state ? 0 : -EINVAL; |
e91d8951 DDAG |
2735 | } |
2736 | ||
09f6c85e DDAG |
2737 | /** |
2738 | * migration_completion: Used by migration_thread when there's not much left. | |
2739 | * The caller 'breaks' the loop when this returns. | |
2740 | * | |
2741 | * @s: Current migration state | |
09f6c85e | 2742 | */ |
2ad87305 | 2743 | static void migration_completion(MigrationState *s) |
09f6c85e DDAG |
2744 | { |
2745 | int ret; | |
2ad87305 | 2746 | int current_active_state = s->state; |
09f6c85e | 2747 | |
b10ac0c4 DDAG |
2748 | if (s->state == MIGRATION_STATUS_ACTIVE) { |
2749 | qemu_mutex_lock_iothread(); | |
64909f97 | 2750 | s->downtime_start = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); |
fb064112 | 2751 | qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER, NULL); |
7287cbd4 | 2752 | s->vm_was_running = runstate_is_running(); |
b10ac0c4 DDAG |
2753 | ret = global_state_store(); |
2754 | ||
2755 | if (!ret) { | |
a1fbe750 | 2756 | bool inactivate = !migrate_colo_enabled(); |
b10ac0c4 | 2757 | ret = vm_stop_force_state(RUN_STATE_FINISH_MIGRATE); |
e91d8951 | 2758 | if (ret >= 0) { |
0331c8ca DDAG |
2759 | ret = migration_maybe_pause(s, ¤t_active_state, |
2760 | MIGRATION_STATUS_DEVICE); | |
e91d8951 | 2761 | } |
f07fa4cb KW |
2762 | if (ret >= 0) { |
2763 | qemu_file_set_rate_limit(s->to_dst_file, INT64_MAX); | |
a1fbe750 FZ |
2764 | ret = qemu_savevm_state_complete_precopy(s->to_dst_file, false, |
2765 | inactivate); | |
f07fa4cb | 2766 | } |
a1fbe750 FZ |
2767 | if (inactivate && ret >= 0) { |
2768 | s->block_inactive = true; | |
b10ac0c4 DDAG |
2769 | } |
2770 | } | |
2771 | qemu_mutex_unlock_iothread(); | |
09f6c85e | 2772 | |
b10ac0c4 DDAG |
2773 | if (ret < 0) { |
2774 | goto fail; | |
09f6c85e | 2775 | } |
b10ac0c4 DDAG |
2776 | } else if (s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) { |
2777 | trace_migration_completion_postcopy_end(); | |
2778 | ||
89a02a9f | 2779 | qemu_savevm_state_complete_postcopy(s->to_dst_file); |
b10ac0c4 | 2780 | trace_migration_completion_postcopy_end_after_complete(); |
09f6c85e | 2781 | } |
09f6c85e | 2782 | |
b10ac0c4 DDAG |
2783 | /* |
2784 | * If rp was opened we must clean up the thread before | |
2785 | * cleaning everything else up (since if there are no failures | |
2786 | * it will wait for the destination to send it's status in | |
2787 | * a SHUT command). | |
b10ac0c4 | 2788 | */ |
0425dc97 | 2789 | if (s->rp_state.from_dst_file) { |
b10ac0c4 | 2790 | int rp_error; |
0425dc97 | 2791 | trace_migration_return_path_end_before(); |
b10ac0c4 | 2792 | rp_error = await_return_path_close_on_source(s); |
0425dc97 | 2793 | trace_migration_return_path_end_after(rp_error); |
b10ac0c4 | 2794 | if (rp_error) { |
fe904ea8 | 2795 | goto fail_invalidate; |
b10ac0c4 | 2796 | } |
09f6c85e DDAG |
2797 | } |
2798 | ||
89a02a9f | 2799 | if (qemu_file_get_error(s->to_dst_file)) { |
09f6c85e | 2800 | trace_migration_completion_file_err(); |
fe904ea8 | 2801 | goto fail_invalidate; |
09f6c85e DDAG |
2802 | } |
2803 | ||
0b827d5e HZ |
2804 | if (!migrate_colo_enabled()) { |
2805 | migrate_set_state(&s->state, current_active_state, | |
2806 | MIGRATION_STATUS_COMPLETED); | |
2807 | } | |
2808 | ||
09f6c85e DDAG |
2809 | return; |
2810 | ||
fe904ea8 GK |
2811 | fail_invalidate: |
2812 | /* If not doing postcopy, vm_start() will be called: let's regain | |
2813 | * control on images. | |
2814 | */ | |
6039dd5b DDAG |
2815 | if (s->state == MIGRATION_STATUS_ACTIVE || |
2816 | s->state == MIGRATION_STATUS_DEVICE) { | |
fe904ea8 GK |
2817 | Error *local_err = NULL; |
2818 | ||
1d2acc31 | 2819 | qemu_mutex_lock_iothread(); |
fe904ea8 GK |
2820 | bdrv_invalidate_cache_all(&local_err); |
2821 | if (local_err) { | |
2822 | error_report_err(local_err); | |
1d2acc31 HZ |
2823 | } else { |
2824 | s->block_inactive = false; | |
fe904ea8 | 2825 | } |
1d2acc31 | 2826 | qemu_mutex_unlock_iothread(); |
fe904ea8 GK |
2827 | } |
2828 | ||
09f6c85e | 2829 | fail: |
48781e5b HZ |
2830 | migrate_set_state(&s->state, current_active_state, |
2831 | MIGRATION_STATUS_FAILED); | |
09f6c85e DDAG |
2832 | } |
2833 | ||
35a6ed4f HZ |
2834 | bool migrate_colo_enabled(void) |
2835 | { | |
2836 | MigrationState *s = migrate_get_current(); | |
2837 | return s->enabled_capabilities[MIGRATION_CAPABILITY_X_COLO]; | |
2838 | } | |
2839 | ||
b23c2ade PX |
2840 | typedef enum MigThrError { |
2841 | /* No error detected */ | |
2842 | MIG_THR_ERR_NONE = 0, | |
2843 | /* Detected error, but resumed successfully */ | |
2844 | MIG_THR_ERR_RECOVERED = 1, | |
2845 | /* Detected fatal error, need to exit */ | |
2846 | MIG_THR_ERR_FATAL = 2, | |
2847 | } MigThrError; | |
2848 | ||
94190696 PX |
2849 | static int postcopy_resume_handshake(MigrationState *s) |
2850 | { | |
2851 | qemu_savevm_send_postcopy_resume(s->to_dst_file); | |
2852 | ||
2853 | while (s->state == MIGRATION_STATUS_POSTCOPY_RECOVER) { | |
2854 | qemu_sem_wait(&s->rp_state.rp_sem); | |
2855 | } | |
2856 | ||
2857 | if (s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) { | |
2858 | return 0; | |
2859 | } | |
2860 | ||
2861 | return -1; | |
2862 | } | |
2863 | ||
135b87b4 PX |
2864 | /* Return zero if success, or <0 for error */ |
2865 | static int postcopy_do_resume(MigrationState *s) | |
2866 | { | |
d1b8eadb PX |
2867 | int ret; |
2868 | ||
2869 | /* | |
2870 | * Call all the resume_prepare() hooks, so that modules can be | |
2871 | * ready for the migration resume. | |
2872 | */ | |
2873 | ret = qemu_savevm_state_resume_prepare(s); | |
2874 | if (ret) { | |
2875 | error_report("%s: resume_prepare() failure detected: %d", | |
2876 | __func__, ret); | |
2877 | return ret; | |
2878 | } | |
2879 | ||
2880 | /* | |
94190696 PX |
2881 | * Last handshake with destination on the resume (destination will |
2882 | * switch to postcopy-active afterwards) | |
d1b8eadb | 2883 | */ |
94190696 PX |
2884 | ret = postcopy_resume_handshake(s); |
2885 | if (ret) { | |
2886 | error_report("%s: handshake failed: %d", __func__, ret); | |
2887 | return ret; | |
2888 | } | |
d1b8eadb | 2889 | |
135b87b4 PX |
2890 | return 0; |
2891 | } | |
2892 | ||
b23c2ade PX |
2893 | /* |
2894 | * We don't return until we are in a safe state to continue current | |
2895 | * postcopy migration. Returns MIG_THR_ERR_RECOVERED if recovered, or | |
2896 | * MIG_THR_ERR_FATAL if unrecovery failure happened. | |
2897 | */ | |
2898 | static MigThrError postcopy_pause(MigrationState *s) | |
2899 | { | |
2900 | assert(s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE); | |
b23c2ade | 2901 | |
135b87b4 | 2902 | while (true) { |
62df066f PX |
2903 | QEMUFile *file; |
2904 | ||
135b87b4 PX |
2905 | migrate_set_state(&s->state, s->state, |
2906 | MIGRATION_STATUS_POSTCOPY_PAUSED); | |
b23c2ade | 2907 | |
135b87b4 PX |
2908 | /* Current channel is possibly broken. Release it. */ |
2909 | assert(s->to_dst_file); | |
62df066f PX |
2910 | qemu_mutex_lock(&s->qemu_file_lock); |
2911 | file = s->to_dst_file; | |
135b87b4 | 2912 | s->to_dst_file = NULL; |
62df066f PX |
2913 | qemu_mutex_unlock(&s->qemu_file_lock); |
2914 | ||
2915 | qemu_file_shutdown(file); | |
2916 | qemu_fclose(file); | |
b23c2ade | 2917 | |
135b87b4 PX |
2918 | error_report("Detected IO failure for postcopy. " |
2919 | "Migration paused."); | |
b23c2ade | 2920 | |
135b87b4 PX |
2921 | /* |
2922 | * We wait until things fixed up. Then someone will setup the | |
2923 | * status back for us. | |
2924 | */ | |
2925 | while (s->state == MIGRATION_STATUS_POSTCOPY_PAUSED) { | |
2926 | qemu_sem_wait(&s->postcopy_pause_sem); | |
2927 | } | |
2928 | ||
2929 | if (s->state == MIGRATION_STATUS_POSTCOPY_RECOVER) { | |
2930 | /* Woken up by a recover procedure. Give it a shot */ | |
b23c2ade | 2931 | |
135b87b4 PX |
2932 | /* |
2933 | * Firstly, let's wake up the return path now, with a new | |
2934 | * return path channel. | |
2935 | */ | |
2936 | qemu_sem_post(&s->postcopy_pause_rp_sem); | |
2937 | ||
2938 | /* Do the resume logic */ | |
2939 | if (postcopy_do_resume(s) == 0) { | |
2940 | /* Let's continue! */ | |
2941 | trace_postcopy_pause_continued(); | |
2942 | return MIG_THR_ERR_RECOVERED; | |
2943 | } else { | |
2944 | /* | |
2945 | * Something wrong happened during the recovery, let's | |
2946 | * pause again. Pause is always better than throwing | |
2947 | * data away. | |
2948 | */ | |
2949 | continue; | |
2950 | } | |
2951 | } else { | |
2952 | /* This is not right... Time to quit. */ | |
2953 | return MIG_THR_ERR_FATAL; | |
2954 | } | |
2955 | } | |
b23c2ade PX |
2956 | } |
2957 | ||
2958 | static MigThrError migration_detect_error(MigrationState *s) | |
2959 | { | |
2960 | int ret; | |
c3c5eae6 DDAG |
2961 | int state = s->state; |
2962 | ||
2963 | if (state == MIGRATION_STATUS_CANCELLING || | |
2964 | state == MIGRATION_STATUS_CANCELLED) { | |
2965 | /* End the migration, but don't set the state to failed */ | |
2966 | return MIG_THR_ERR_FATAL; | |
2967 | } | |
b23c2ade PX |
2968 | |
2969 | /* Try to detect any file errors */ | |
2970 | ret = qemu_file_get_error(s->to_dst_file); | |
2971 | ||
2972 | if (!ret) { | |
2973 | /* Everything is fine */ | |
2974 | return MIG_THR_ERR_NONE; | |
2975 | } | |
2976 | ||
c3c5eae6 | 2977 | if (state == MIGRATION_STATUS_POSTCOPY_ACTIVE && ret == -EIO) { |
b23c2ade PX |
2978 | /* |
2979 | * For postcopy, we allow the network to be down for a | |
2980 | * while. After that, it can be continued by a | |
2981 | * recovery phase. | |
2982 | */ | |
2983 | return postcopy_pause(s); | |
2984 | } else { | |
2985 | /* | |
2986 | * For precopy (or postcopy with error outside IO), we fail | |
2987 | * with no time. | |
2988 | */ | |
c3c5eae6 | 2989 | migrate_set_state(&s->state, state, MIGRATION_STATUS_FAILED); |
b23c2ade PX |
2990 | trace_migration_thread_file_err(); |
2991 | ||
2992 | /* Time to stop the migration, now. */ | |
2993 | return MIG_THR_ERR_FATAL; | |
2994 | } | |
2995 | } | |
2996 | ||
0c8f0efd JQ |
2997 | /* How many bytes have we transferred since the beggining of the migration */ |
2998 | static uint64_t migration_total_bytes(MigrationState *s) | |
2999 | { | |
b9ee2f7d | 3000 | return qemu_ftell(s->to_dst_file) + ram_counters.multifd_bytes; |
0c8f0efd JQ |
3001 | } |
3002 | ||
cf011f08 PX |
3003 | static void migration_calculate_complete(MigrationState *s) |
3004 | { | |
0c8f0efd | 3005 | uint64_t bytes = migration_total_bytes(s); |
cf011f08 | 3006 | int64_t end_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); |
6cde6fbe | 3007 | int64_t transfer_time; |
cf011f08 PX |
3008 | |
3009 | s->total_time = end_time - s->start_time; | |
3010 | if (!s->downtime) { | |
3011 | /* | |
3012 | * It's still not set, so we are precopy migration. For | |
3013 | * postcopy, downtime is calculated during postcopy_start(). | |
3014 | */ | |
3015 | s->downtime = end_time - s->downtime_start; | |
3016 | } | |
3017 | ||
6cde6fbe JQ |
3018 | transfer_time = s->total_time - s->setup_time; |
3019 | if (transfer_time) { | |
3020 | s->mbps = ((double) bytes * 8.0) / transfer_time / 1000; | |
cf011f08 PX |
3021 | } |
3022 | } | |
3023 | ||
b15df1ae PX |
3024 | static void migration_update_counters(MigrationState *s, |
3025 | int64_t current_time) | |
3026 | { | |
aecbfe9c | 3027 | uint64_t transferred, transferred_pages, time_spent; |
0c8f0efd | 3028 | uint64_t current_bytes; /* bytes transferred since the beginning */ |
b15df1ae PX |
3029 | double bandwidth; |
3030 | ||
3031 | if (current_time < s->iteration_start_time + BUFFER_DELAY) { | |
3032 | return; | |
3033 | } | |
3034 | ||
0c8f0efd JQ |
3035 | current_bytes = migration_total_bytes(s); |
3036 | transferred = current_bytes - s->iteration_initial_bytes; | |
b15df1ae PX |
3037 | time_spent = current_time - s->iteration_start_time; |
3038 | bandwidth = (double)transferred / time_spent; | |
0781c1ed | 3039 | s->threshold_size = bandwidth * s->parameters.downtime_limit; |
b15df1ae PX |
3040 | |
3041 | s->mbps = (((double) transferred * 8.0) / | |
3042 | ((double) time_spent / 1000.0)) / 1000.0 / 1000.0; | |
3043 | ||
aecbfe9c XG |
3044 | transferred_pages = ram_get_total_transferred_pages() - |
3045 | s->iteration_initial_pages; | |
3046 | s->pages_per_second = (double) transferred_pages / | |
3047 | (((double) time_spent / 1000.0)); | |
3048 | ||
b15df1ae PX |
3049 | /* |
3050 | * if we haven't sent anything, we don't want to | |
3051 | * recalculate. 10000 is a small enough number for our purposes | |
3052 | */ | |
3053 | if (ram_counters.dirty_pages_rate && transferred > 10000) { | |
650af890 | 3054 | s->expected_downtime = ram_counters.remaining / bandwidth; |
b15df1ae PX |
3055 | } |
3056 | ||
3057 | qemu_file_reset_rate_limit(s->to_dst_file); | |
3058 | ||
3059 | s->iteration_start_time = current_time; | |
0c8f0efd | 3060 | s->iteration_initial_bytes = current_bytes; |
aecbfe9c | 3061 | s->iteration_initial_pages = ram_get_total_transferred_pages(); |
b15df1ae PX |
3062 | |
3063 | trace_migrate_transferred(transferred, time_spent, | |
0781c1ed | 3064 | bandwidth, s->threshold_size); |
b15df1ae PX |
3065 | } |
3066 | ||
2ad87305 PX |
3067 | /* Migration thread iteration status */ |
3068 | typedef enum { | |
3069 | MIG_ITERATE_RESUME, /* Resume current iteration */ | |
3070 | MIG_ITERATE_SKIP, /* Skip current iteration */ | |
3071 | MIG_ITERATE_BREAK, /* Break the loop */ | |
3072 | } MigIterateState; | |
3073 | ||
3074 | /* | |
3075 | * Return true if continue to the next iteration directly, false | |
3076 | * otherwise. | |
3077 | */ | |
3078 | static MigIterateState migration_iteration_run(MigrationState *s) | |
3079 | { | |
47995026 | 3080 | uint64_t pending_size, pend_pre, pend_compat, pend_post; |
2ad87305 PX |
3081 | bool in_postcopy = s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE; |
3082 | ||
47995026 VSO |
3083 | qemu_savevm_state_pending(s->to_dst_file, s->threshold_size, &pend_pre, |
3084 | &pend_compat, &pend_post); | |
3085 | pending_size = pend_pre + pend_compat + pend_post; | |
2ad87305 PX |
3086 | |
3087 | trace_migrate_pending(pending_size, s->threshold_size, | |
47995026 | 3088 | pend_pre, pend_compat, pend_post); |
2ad87305 PX |
3089 | |
3090 | if (pending_size && pending_size >= s->threshold_size) { | |
3091 | /* Still a significant amount to transfer */ | |
3092 | if (migrate_postcopy() && !in_postcopy && | |
47995026 | 3093 | pend_pre <= s->threshold_size && |
2ad87305 PX |
3094 | atomic_read(&s->start_postcopy)) { |
3095 | if (postcopy_start(s)) { | |
3096 | error_report("%s: postcopy failed to start", __func__); | |
3097 | } | |
3098 | return MIG_ITERATE_SKIP; | |
3099 | } | |
3100 | /* Just another iteration step */ | |
3101 | qemu_savevm_state_iterate(s->to_dst_file, | |
3102 | s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE); | |
3103 | } else { | |
3104 | trace_migration_thread_low_pending(pending_size); | |
3105 | migration_completion(s); | |
3106 | return MIG_ITERATE_BREAK; | |
3107 | } | |
3108 | ||
3109 | return MIG_ITERATE_RESUME; | |
3110 | } | |
3111 | ||
199aa6d4 PX |
3112 | static void migration_iteration_finish(MigrationState *s) |
3113 | { | |
3114 | /* If we enabled cpu throttling for auto-converge, turn it off. */ | |
3115 | cpu_throttle_stop(); | |
3116 | ||
3117 | qemu_mutex_lock_iothread(); | |
3118 | switch (s->state) { | |
3119 | case MIGRATION_STATUS_COMPLETED: | |
3120 | migration_calculate_complete(s); | |
3121 | runstate_set(RUN_STATE_POSTMIGRATE); | |
3122 | break; | |
3123 | ||
3124 | case MIGRATION_STATUS_ACTIVE: | |
3125 | /* | |
3126 | * We should really assert here, but since it's during | |
3127 | * migration, let's try to reduce the usage of assertions. | |
3128 | */ | |
3129 | if (!migrate_colo_enabled()) { | |
3130 | error_report("%s: critical error: calling COLO code without " | |
3131 | "COLO enabled", __func__); | |
3132 | } | |
3133 | migrate_start_colo_process(s); | |
3134 | /* | |
3135 | * Fixme: we will run VM in COLO no matter its old running state. | |
3136 | * After exited COLO, we will keep running. | |
3137 | */ | |
3138 | s->vm_was_running = true; | |
3139 | /* Fallthrough */ | |
3140 | case MIGRATION_STATUS_FAILED: | |
3141 | case MIGRATION_STATUS_CANCELLED: | |
57225e5f | 3142 | case MIGRATION_STATUS_CANCELLING: |
199aa6d4 PX |
3143 | if (s->vm_was_running) { |
3144 | vm_start(); | |
3145 | } else { | |
3146 | if (runstate_check(RUN_STATE_FINISH_MIGRATE)) { | |
3147 | runstate_set(RUN_STATE_POSTMIGRATE); | |
3148 | } | |
3149 | } | |
3150 | break; | |
3151 | ||
3152 | default: | |
3153 | /* Should not reach here, but if so, forgive the VM. */ | |
3154 | error_report("%s: Unknown ending state %d", __func__, s->state); | |
3155 | break; | |
3156 | } | |
3157 | qemu_bh_schedule(s->cleanup_bh); | |
3158 | qemu_mutex_unlock_iothread(); | |
3159 | } | |
3160 | ||
ad767bed DDAG |
3161 | void migration_make_urgent_request(void) |
3162 | { | |
3163 | qemu_sem_post(&migrate_get_current()->rate_limit_sem); | |
3164 | } | |
3165 | ||
3166 | void migration_consume_urgent_request(void) | |
3167 | { | |
3168 | qemu_sem_wait(&migrate_get_current()->rate_limit_sem); | |
3169 | } | |
3170 | ||
70b20477 DDAG |
3171 | /* |
3172 | * Master migration thread on the source VM. | |
3173 | * It drives the migration and pumps the data down the outgoing channel. | |
3174 | */ | |
5f496a1b | 3175 | static void *migration_thread(void *opaque) |
0d82d0e8 | 3176 | { |
9848a404 | 3177 | MigrationState *s = opaque; |
bc72ad67 | 3178 | int64_t setup_start = qemu_clock_get_ms(QEMU_CLOCK_HOST); |
b23c2ade | 3179 | MigThrError thr_error; |
ad767bed | 3180 | bool urgent = false; |
76f5933a | 3181 | |
ab28bd23 PB |
3182 | rcu_register_thread(); |
3183 | ||
892ae715 | 3184 | object_ref(OBJECT(s)); |
b15df1ae PX |
3185 | s->iteration_start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); |
3186 | ||
89a02a9f | 3187 | qemu_savevm_state_header(s->to_dst_file); |
1d34e4bf | 3188 | |
62a02658 PX |
3189 | /* |
3190 | * If we opened the return path, we need to make sure dst has it | |
3191 | * opened as well. | |
3192 | */ | |
3193 | if (s->rp_state.from_dst_file) { | |
1d34e4bf | 3194 | /* Now tell the dest that it should open its end so it can reply */ |
89a02a9f | 3195 | qemu_savevm_send_open_return_path(s->to_dst_file); |
1d34e4bf DDAG |
3196 | |
3197 | /* And do a ping that will make stuff easier to debug */ | |
89a02a9f | 3198 | qemu_savevm_send_ping(s->to_dst_file, 1); |
0425dc97 | 3199 | } |
1d34e4bf | 3200 | |
58110f0a | 3201 | if (migrate_postcopy()) { |
1d34e4bf DDAG |
3202 | /* |
3203 | * Tell the destination that we *might* want to do postcopy later; | |
3204 | * if the other end can't do postcopy it should fail now, nice and | |
3205 | * early. | |
3206 | */ | |
89a02a9f | 3207 | qemu_savevm_send_postcopy_advise(s->to_dst_file); |
1d34e4bf DDAG |
3208 | } |
3209 | ||
aad555c2 ZC |
3210 | if (migrate_colo_enabled()) { |
3211 | /* Notify migration destination that we enable COLO */ | |
3212 | qemu_savevm_send_colo_enable(s->to_dst_file); | |
3213 | } | |
3214 | ||
9907e842 | 3215 | qemu_savevm_state_setup(s->to_dst_file); |
0d82d0e8 | 3216 | |
bc72ad67 | 3217 | s->setup_time = qemu_clock_get_ms(QEMU_CLOCK_HOST) - setup_start; |
48781e5b HZ |
3218 | migrate_set_state(&s->state, MIGRATION_STATUS_SETUP, |
3219 | MIGRATION_STATUS_ACTIVE); | |
29ae8a41 | 3220 | |
9ec055ae DDAG |
3221 | trace_migration_thread_setup_complete(); |
3222 | ||
3223 | while (s->state == MIGRATION_STATUS_ACTIVE || | |
3224 | s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) { | |
a3e879cd | 3225 | int64_t current_time; |
0d82d0e8 | 3226 | |
ad767bed | 3227 | if (urgent || !qemu_file_rate_limit(s->to_dst_file)) { |
2ad87305 PX |
3228 | MigIterateState iter_state = migration_iteration_run(s); |
3229 | if (iter_state == MIG_ITERATE_SKIP) { | |
3230 | continue; | |
3231 | } else if (iter_state == MIG_ITERATE_BREAK) { | |
09f6c85e | 3232 | break; |
c369f40d JQ |
3233 | } |
3234 | } | |
f4410a5d | 3235 | |
b23c2ade PX |
3236 | /* |
3237 | * Try to detect any kind of failures, and see whether we | |
3238 | * should stop the migration now. | |
3239 | */ | |
3240 | thr_error = migration_detect_error(s); | |
3241 | if (thr_error == MIG_THR_ERR_FATAL) { | |
3242 | /* Stop migration */ | |
fd45ee2c | 3243 | break; |
b23c2ade PX |
3244 | } else if (thr_error == MIG_THR_ERR_RECOVERED) { |
3245 | /* | |
3246 | * Just recovered from a e.g. network failure, reset all | |
3247 | * the local variables. This is important to avoid | |
3248 | * breaking transferred_bytes and bandwidth calculation | |
3249 | */ | |
3250 | s->iteration_start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); | |
3251 | s->iteration_initial_bytes = 0; | |
fd45ee2c | 3252 | } |
b15df1ae | 3253 | |
bc72ad67 | 3254 | current_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); |
0d82d0e8 | 3255 | |
b15df1ae PX |
3256 | migration_update_counters(s, current_time); |
3257 | ||
ad767bed | 3258 | urgent = false; |
89a02a9f | 3259 | if (qemu_file_rate_limit(s->to_dst_file)) { |
ad767bed DDAG |
3260 | /* Wait for a delay to do rate limiting OR |
3261 | * something urgent to post the semaphore. | |
3262 | */ | |
3263 | int ms = s->iteration_start_time + BUFFER_DELAY - current_time; | |
3264 | trace_migration_thread_ratelimit_pre(ms); | |
3265 | if (qemu_sem_timedwait(&s->rate_limit_sem, ms) == 0) { | |
3266 | /* We were worken by one or more urgent things but | |
3267 | * the timedwait will have consumed one of them. | |
3268 | * The service routine for the urgent wake will dec | |
3269 | * the semaphore itself for each item it consumes, | |
3270 | * so add this one we just eat back. | |
3271 | */ | |
3272 | qemu_sem_post(&s->rate_limit_sem); | |
3273 | urgent = true; | |
3274 | } | |
3275 | trace_migration_thread_ratelimit_post(urgent); | |
0d82d0e8 | 3276 | } |
a3fa1d78 PB |
3277 | } |
3278 | ||
1d34e4bf | 3279 | trace_migration_thread_after_loop(); |
199aa6d4 | 3280 | migration_iteration_finish(s); |
892ae715 | 3281 | object_unref(OBJECT(s)); |
ab28bd23 | 3282 | rcu_unregister_thread(); |
0d82d0e8 JQ |
3283 | return NULL; |
3284 | } | |
3285 | ||
cce8040b | 3286 | void migrate_fd_connect(MigrationState *s, Error *error_in) |
0d82d0e8 | 3287 | { |
d3e35b8f PX |
3288 | int64_t rate_limit; |
3289 | bool resume = s->state == MIGRATION_STATUS_POSTCOPY_PAUSED; | |
3290 | ||
2ff30257 | 3291 | s->expected_downtime = s->parameters.downtime_limit; |
bb1fadc4 | 3292 | s->cleanup_bh = qemu_bh_new(migrate_fd_cleanup, s); |
cce8040b DDAG |
3293 | if (error_in) { |
3294 | migrate_fd_error(s, error_in); | |
3295 | migrate_fd_cleanup(s); | |
3296 | return; | |
3297 | } | |
0d82d0e8 | 3298 | |
d3e35b8f PX |
3299 | if (resume) { |
3300 | /* This is a resumed migration */ | |
3301 | rate_limit = INT64_MAX; | |
3302 | } else { | |
3303 | /* This is a fresh new migration */ | |
3304 | rate_limit = s->parameters.max_bandwidth / XFER_LIMIT_RATIO; | |
442773ce | 3305 | |
d3e35b8f PX |
3306 | /* Notify before starting migration thread */ |
3307 | notifier_list_notify(&migration_state_notifiers, s); | |
3308 | } | |
3309 | ||
3310 | qemu_file_set_rate_limit(s->to_dst_file, rate_limit); | |
3311 | qemu_file_set_blocking(s->to_dst_file, true); | |
9287ac27 | 3312 | |
1d34e4bf | 3313 | /* |
c788ada8 PX |
3314 | * Open the return path. For postcopy, it is used exclusively. For |
3315 | * precopy, only if user specified "return-path" capability would | |
3316 | * QEMU uses the return path. | |
1d34e4bf | 3317 | */ |
c788ada8 | 3318 | if (migrate_postcopy_ram() || migrate_use_return_path()) { |
d3e35b8f | 3319 | if (open_return_path_on_source(s, !resume)) { |
1d34e4bf | 3320 | error_report("Unable to open return-path for postcopy"); |
d3e35b8f | 3321 | migrate_set_state(&s->state, s->state, MIGRATION_STATUS_FAILED); |
1d34e4bf DDAG |
3322 | migrate_fd_cleanup(s); |
3323 | return; | |
3324 | } | |
3325 | } | |
3326 | ||
d3e35b8f | 3327 | if (resume) { |
135b87b4 PX |
3328 | /* Wakeup the main migration thread to do the recovery */ |
3329 | migrate_set_state(&s->state, MIGRATION_STATUS_POSTCOPY_PAUSED, | |
3330 | MIGRATION_STATUS_POSTCOPY_RECOVER); | |
3331 | qemu_sem_post(&s->postcopy_pause_sem); | |
d3e35b8f PX |
3332 | return; |
3333 | } | |
3334 | ||
f986c3d2 JQ |
3335 | if (multifd_save_setup() != 0) { |
3336 | migrate_set_state(&s->state, MIGRATION_STATUS_SETUP, | |
3337 | MIGRATION_STATUS_FAILED); | |
3338 | migrate_fd_cleanup(s); | |
3339 | return; | |
3340 | } | |
009fad7f | 3341 | qemu_thread_create(&s->thread, "live_migration", migration_thread, s, |
bb1fadc4 | 3342 | QEMU_THREAD_JOINABLE); |
1d34e4bf | 3343 | s->migration_thread_running = true; |
0d82d0e8 | 3344 | } |
093e3c42 | 3345 | |
9d18af93 PX |
3346 | void migration_global_dump(Monitor *mon) |
3347 | { | |
3348 | MigrationState *ms = migrate_get_current(); | |
3349 | ||
6f0f6428 JQ |
3350 | monitor_printf(mon, "globals:\n"); |
3351 | monitor_printf(mon, "store-global-state: %s\n", | |
3352 | ms->store_global_state ? "on" : "off"); | |
3353 | monitor_printf(mon, "only-migratable: %s\n", | |
3354 | ms->only_migratable ? "on" : "off"); | |
3355 | monitor_printf(mon, "send-configuration: %s\n", | |
3356 | ms->send_configuration ? "on" : "off"); | |
3357 | monitor_printf(mon, "send-section-footer: %s\n", | |
3358 | ms->send_section_footer ? "on" : "off"); | |
f548222c XG |
3359 | monitor_printf(mon, "decompress-error-check: %s\n", |
3360 | ms->decompress_error_check ? "on" : "off"); | |
9d18af93 PX |
3361 | } |
3362 | ||
20814758 PX |
3363 | #define DEFINE_PROP_MIG_CAP(name, x) \ |
3364 | DEFINE_PROP_BOOL(name, MigrationState, enabled_capabilities[x], false) | |
3365 | ||
5272298c PX |
3366 | static Property migration_properties[] = { |
3367 | DEFINE_PROP_BOOL("store-global-state", MigrationState, | |
3368 | store_global_state, true), | |
3df663e5 | 3369 | DEFINE_PROP_BOOL("only-migratable", MigrationState, only_migratable, false), |
71dd4c1a PX |
3370 | DEFINE_PROP_BOOL("send-configuration", MigrationState, |
3371 | send_configuration, true), | |
15c38503 PX |
3372 | DEFINE_PROP_BOOL("send-section-footer", MigrationState, |
3373 | send_section_footer, true), | |
f548222c XG |
3374 | DEFINE_PROP_BOOL("decompress-error-check", MigrationState, |
3375 | decompress_error_check, true), | |
89632faf PX |
3376 | |
3377 | /* Migration parameters */ | |
741d4086 | 3378 | DEFINE_PROP_UINT8("x-compress-level", MigrationState, |
89632faf PX |
3379 | parameters.compress_level, |
3380 | DEFAULT_MIGRATE_COMPRESS_LEVEL), | |
741d4086 | 3381 | DEFINE_PROP_UINT8("x-compress-threads", MigrationState, |
89632faf PX |
3382 | parameters.compress_threads, |
3383 | DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT), | |
1d58872a XG |
3384 | DEFINE_PROP_BOOL("x-compress-wait-thread", MigrationState, |
3385 | parameters.compress_wait_thread, true), | |
741d4086 | 3386 | DEFINE_PROP_UINT8("x-decompress-threads", MigrationState, |
89632faf PX |
3387 | parameters.decompress_threads, |
3388 | DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT), | |
741d4086 | 3389 | DEFINE_PROP_UINT8("x-cpu-throttle-initial", MigrationState, |
89632faf PX |
3390 | parameters.cpu_throttle_initial, |
3391 | DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL), | |
741d4086 | 3392 | DEFINE_PROP_UINT8("x-cpu-throttle-increment", MigrationState, |
89632faf PX |
3393 | parameters.cpu_throttle_increment, |
3394 | DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT), | |
741d4086 | 3395 | DEFINE_PROP_SIZE("x-max-bandwidth", MigrationState, |
89632faf | 3396 | parameters.max_bandwidth, MAX_THROTTLE), |
741d4086 | 3397 | DEFINE_PROP_UINT64("x-downtime-limit", MigrationState, |
89632faf PX |
3398 | parameters.downtime_limit, |
3399 | DEFAULT_MIGRATE_SET_DOWNTIME), | |
741d4086 | 3400 | DEFINE_PROP_UINT32("x-checkpoint-delay", MigrationState, |
89632faf PX |
3401 | parameters.x_checkpoint_delay, |
3402 | DEFAULT_MIGRATE_X_CHECKPOINT_DELAY), | |
741d4086 | 3403 | DEFINE_PROP_UINT8("x-multifd-channels", MigrationState, |
4075fb1c JQ |
3404 | parameters.x_multifd_channels, |
3405 | DEFAULT_MIGRATE_MULTIFD_CHANNELS), | |
741d4086 | 3406 | DEFINE_PROP_UINT32("x-multifd-page-count", MigrationState, |
0fb86605 JQ |
3407 | parameters.x_multifd_page_count, |
3408 | DEFAULT_MIGRATE_MULTIFD_PAGE_COUNT), | |
73af8dd8 JQ |
3409 | DEFINE_PROP_SIZE("xbzrle-cache-size", MigrationState, |
3410 | parameters.xbzrle_cache_size, | |
3411 | DEFAULT_MIGRATE_XBZRLE_CACHE_SIZE), | |
7e555c6c DDAG |
3412 | DEFINE_PROP_SIZE("max-postcopy-bandwidth", MigrationState, |
3413 | parameters.max_postcopy_bandwidth, | |
3414 | DEFAULT_MIGRATE_MAX_POSTCOPY_BANDWIDTH), | |
4cbc9c7f LQ |
3415 | DEFINE_PROP_UINT8("max-cpu-throttle", MigrationState, |
3416 | parameters.max_cpu_throttle, | |
3417 | DEFAULT_MIGRATE_MAX_CPU_THROTTLE), | |
ee3d96ba DDAG |
3418 | DEFINE_PROP_SIZE("announce-initial", MigrationState, |
3419 | parameters.announce_initial, | |
3420 | DEFAULT_MIGRATE_ANNOUNCE_INITIAL), | |
3421 | DEFINE_PROP_SIZE("announce-max", MigrationState, | |
3422 | parameters.announce_max, | |
3423 | DEFAULT_MIGRATE_ANNOUNCE_MAX), | |
3424 | DEFINE_PROP_SIZE("announce-rounds", MigrationState, | |
3425 | parameters.announce_rounds, | |
3426 | DEFAULT_MIGRATE_ANNOUNCE_ROUNDS), | |
3427 | DEFINE_PROP_SIZE("announce-step", MigrationState, | |
3428 | parameters.announce_step, | |
3429 | DEFAULT_MIGRATE_ANNOUNCE_STEP), | |
20814758 PX |
3430 | |
3431 | /* Migration capabilities */ | |
3432 | DEFINE_PROP_MIG_CAP("x-xbzrle", MIGRATION_CAPABILITY_XBZRLE), | |
3433 | DEFINE_PROP_MIG_CAP("x-rdma-pin-all", MIGRATION_CAPABILITY_RDMA_PIN_ALL), | |
3434 | DEFINE_PROP_MIG_CAP("x-auto-converge", MIGRATION_CAPABILITY_AUTO_CONVERGE), | |
3435 | DEFINE_PROP_MIG_CAP("x-zero-blocks", MIGRATION_CAPABILITY_ZERO_BLOCKS), | |
3436 | DEFINE_PROP_MIG_CAP("x-compress", MIGRATION_CAPABILITY_COMPRESS), | |
3437 | DEFINE_PROP_MIG_CAP("x-events", MIGRATION_CAPABILITY_EVENTS), | |
3438 | DEFINE_PROP_MIG_CAP("x-postcopy-ram", MIGRATION_CAPABILITY_POSTCOPY_RAM), | |
3439 | DEFINE_PROP_MIG_CAP("x-colo", MIGRATION_CAPABILITY_X_COLO), | |
3440 | DEFINE_PROP_MIG_CAP("x-release-ram", MIGRATION_CAPABILITY_RELEASE_RAM), | |
3441 | DEFINE_PROP_MIG_CAP("x-block", MIGRATION_CAPABILITY_BLOCK), | |
3442 | DEFINE_PROP_MIG_CAP("x-return-path", MIGRATION_CAPABILITY_RETURN_PATH), | |
30126bbf | 3443 | DEFINE_PROP_MIG_CAP("x-multifd", MIGRATION_CAPABILITY_X_MULTIFD), |
20814758 | 3444 | |
5272298c PX |
3445 | DEFINE_PROP_END_OF_LIST(), |
3446 | }; | |
3447 | ||
e5cb7e76 PX |
3448 | static void migration_class_init(ObjectClass *klass, void *data) |
3449 | { | |
3450 | DeviceClass *dc = DEVICE_CLASS(klass); | |
3451 | ||
3452 | dc->user_creatable = false; | |
5272298c | 3453 | dc->props = migration_properties; |
e5cb7e76 PX |
3454 | } |
3455 | ||
b91bf5e4 MAL |
3456 | static void migration_instance_finalize(Object *obj) |
3457 | { | |
3458 | MigrationState *ms = MIGRATION_OBJ(obj); | |
3459 | MigrationParameters *params = &ms->parameters; | |
3460 | ||
87db1a7d | 3461 | qemu_mutex_destroy(&ms->error_mutex); |
62df066f | 3462 | qemu_mutex_destroy(&ms->qemu_file_lock); |
b91bf5e4 MAL |
3463 | g_free(params->tls_hostname); |
3464 | g_free(params->tls_creds); | |
ad767bed | 3465 | qemu_sem_destroy(&ms->rate_limit_sem); |
e91d8951 | 3466 | qemu_sem_destroy(&ms->pause_sem); |
b23c2ade | 3467 | qemu_sem_destroy(&ms->postcopy_pause_sem); |
14b1742e | 3468 | qemu_sem_destroy(&ms->postcopy_pause_rp_sem); |
edd090c7 | 3469 | qemu_sem_destroy(&ms->rp_state.rp_sem); |
ab105cc1 | 3470 | error_free(ms->error); |
b91bf5e4 MAL |
3471 | } |
3472 | ||
e5cb7e76 PX |
3473 | static void migration_instance_init(Object *obj) |
3474 | { | |
3475 | MigrationState *ms = MIGRATION_OBJ(obj); | |
8b0b29dc | 3476 | MigrationParameters *params = &ms->parameters; |
e5cb7e76 PX |
3477 | |
3478 | ms->state = MIGRATION_STATUS_NONE; | |
e5cb7e76 | 3479 | ms->mbps = -1; |
aecbfe9c | 3480 | ms->pages_per_second = -1; |
e91d8951 | 3481 | qemu_sem_init(&ms->pause_sem, 0); |
87db1a7d | 3482 | qemu_mutex_init(&ms->error_mutex); |
8b0b29dc PX |
3483 | |
3484 | params->tls_hostname = g_strdup(""); | |
3485 | params->tls_creds = g_strdup(""); | |
3486 | ||
3487 | /* Set has_* up only for parameter checks */ | |
3488 | params->has_compress_level = true; | |
3489 | params->has_compress_threads = true; | |
3490 | params->has_decompress_threads = true; | |
3491 | params->has_cpu_throttle_initial = true; | |
3492 | params->has_cpu_throttle_increment = true; | |
3493 | params->has_max_bandwidth = true; | |
3494 | params->has_downtime_limit = true; | |
3495 | params->has_x_checkpoint_delay = true; | |
3496 | params->has_block_incremental = true; | |
4075fb1c | 3497 | params->has_x_multifd_channels = true; |
0fb86605 | 3498 | params->has_x_multifd_page_count = true; |
73af8dd8 | 3499 | params->has_xbzrle_cache_size = true; |
7e555c6c | 3500 | params->has_max_postcopy_bandwidth = true; |
4cbc9c7f | 3501 | params->has_max_cpu_throttle = true; |
ee3d96ba DDAG |
3502 | params->has_announce_initial = true; |
3503 | params->has_announce_max = true; | |
3504 | params->has_announce_rounds = true; | |
3505 | params->has_announce_step = true; | |
b23c2ade PX |
3506 | |
3507 | qemu_sem_init(&ms->postcopy_pause_sem, 0); | |
14b1742e | 3508 | qemu_sem_init(&ms->postcopy_pause_rp_sem, 0); |
edd090c7 | 3509 | qemu_sem_init(&ms->rp_state.rp_sem, 0); |
ad767bed | 3510 | qemu_sem_init(&ms->rate_limit_sem, 0); |
62df066f | 3511 | qemu_mutex_init(&ms->qemu_file_lock); |
8b0b29dc PX |
3512 | } |
3513 | ||
3514 | /* | |
3515 | * Return true if check pass, false otherwise. Error will be put | |
3516 | * inside errp if provided. | |
3517 | */ | |
3518 | static bool migration_object_check(MigrationState *ms, Error **errp) | |
3519 | { | |
6b19a7d9 PX |
3520 | MigrationCapabilityStatusList *head = NULL; |
3521 | /* Assuming all off */ | |
3522 | bool cap_list[MIGRATION_CAPABILITY__MAX] = { 0 }, ret; | |
3523 | int i; | |
3524 | ||
8b0b29dc PX |
3525 | if (!migrate_params_check(&ms->parameters, errp)) { |
3526 | return false; | |
3527 | } | |
3528 | ||
6b19a7d9 PX |
3529 | for (i = 0; i < MIGRATION_CAPABILITY__MAX; i++) { |
3530 | if (ms->enabled_capabilities[i]) { | |
3531 | head = migrate_cap_add(head, i, true); | |
3532 | } | |
3533 | } | |
3534 | ||
3535 | ret = migrate_caps_check(cap_list, head, errp); | |
3536 | ||
3537 | /* It works with head == NULL */ | |
3538 | qapi_free_MigrationCapabilityStatusList(head); | |
3539 | ||
3540 | return ret; | |
e5cb7e76 PX |
3541 | } |
3542 | ||
3543 | static const TypeInfo migration_type = { | |
3544 | .name = TYPE_MIGRATION, | |
01f6e14c | 3545 | /* |
c8d3ff38 PX |
3546 | * NOTE: TYPE_MIGRATION is not really a device, as the object is |
3547 | * not created using qdev_create(), it is not attached to the qdev | |
3548 | * device tree, and it is never realized. | |
3549 | * | |
3550 | * TODO: Make this TYPE_OBJECT once QOM provides something like | |
3551 | * TYPE_DEVICE's "-global" properties. | |
01f6e14c | 3552 | */ |
e5cb7e76 PX |
3553 | .parent = TYPE_DEVICE, |
3554 | .class_init = migration_class_init, | |
3555 | .class_size = sizeof(MigrationClass), | |
3556 | .instance_size = sizeof(MigrationState), | |
3557 | .instance_init = migration_instance_init, | |
b91bf5e4 | 3558 | .instance_finalize = migration_instance_finalize, |
e5cb7e76 PX |
3559 | }; |
3560 | ||
3561 | static void register_migration_types(void) | |
3562 | { | |
3563 | type_register_static(&migration_type); | |
3564 | } | |
3565 | ||
3566 | type_init(register_migration_types); |