]>
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" |
db725815 | 19 | #include "qemu/main-loop.h" |
795c40b8 | 20 | #include "migration/blocker.h" |
f4dbe1bf | 21 | #include "exec.h" |
7fcac4a2 | 22 | #include "fd.h" |
2a9e2e59 | 23 | #include "file.h" |
61e8b148 | 24 | #include "socket.h" |
54d31236 | 25 | #include "sysemu/runstate.h" |
46517dd4 | 26 | #include "sysemu/sysemu.h" |
b0c3cf94 | 27 | #include "sysemu/cpu-throttle.h" |
e1a3ecee | 28 | #include "rdma.h" |
7b1e1a22 | 29 | #include "ram.h" |
c323518a | 30 | #include "ram-compress.h" |
84a899de | 31 | #include "migration/global_state.h" |
c4b63b7c | 32 | #include "migration/misc.h" |
6666c96a | 33 | #include "migration.h" |
947701cc | 34 | #include "migration-stats.h" |
20a519a0 | 35 | #include "savevm.h" |
08a0aee1 | 36 | #include "qemu-file.h" |
6720c2b3 | 37 | #include "channel.h" |
987772d9 | 38 | #include "migration/vmstate.h" |
737e150e | 39 | #include "block/block.h" |
e688df6b | 40 | #include "qapi/error.h" |
9aca82ba | 41 | #include "qapi/clone-visitor.h" |
31e4c354 | 42 | #include "qapi/qapi-visit-migration.h" |
9aca82ba | 43 | #include "qapi/qapi-visit-sockets.h" |
9af23989 MA |
44 | #include "qapi/qapi-commands-migration.h" |
45 | #include "qapi/qapi-events-migration.h" | |
cc7a8ea7 | 46 | #include "qapi/qmp/qerror.h" |
15280c36 | 47 | #include "qapi/qmp/qnull.h" |
ab28bd23 | 48 | #include "qemu/rcu.h" |
2c9e6fec | 49 | #include "block.h" |
be07b0ac | 50 | #include "postcopy-ram.h" |
766bd176 | 51 | #include "qemu/thread.h" |
c09e5bb1 | 52 | #include "trace.h" |
51180423 | 53 | #include "exec/target_page.h" |
61b67d47 | 54 | #include "io/channel-buffer.h" |
85a8578e | 55 | #include "io/channel-tls.h" |
35a6ed4f | 56 | #include "migration/colo.h" |
4ffdb337 | 57 | #include "hw/boards.h" |
9d18af93 | 58 | #include "monitor/monitor.h" |
50510ea2 | 59 | #include "net/announce.h" |
c7e0acd5 | 60 | #include "qemu/queue.h" |
d32ca5ad | 61 | #include "multifd.h" |
1b1f4ab6 | 62 | #include "threadinfo.h" |
b5eea99e | 63 | #include "qemu/yank.h" |
6e8c25b4 | 64 | #include "sysemu/cpus.h" |
39675fff | 65 | #include "yank_functions.h" |
1b529d90 | 66 | #include "sysemu/qtest.h" |
1f0776f1 | 67 | #include "options.h" |
15699cf5 | 68 | #include "sysemu/dirtylimit.h" |
72a8192e | 69 | #include "qemu/sockets.h" |
06152b89 | 70 | #include "sysemu/kvm.h" |
065e2813 | 71 | |
6835f5a1 SS |
72 | #define NOTIFIER_ELEM_INIT(array, elem) \ |
73 | [elem] = NOTIFIER_WITH_RETURN_LIST_INITIALIZER((array)[elem]) | |
74 | ||
dbea1c89 VSO |
75 | #define INMIGRATE_DEFAULT_EXIT_ON_ERROR true |
76 | ||
6835f5a1 SS |
77 | static NotifierWithReturnList migration_state_notifiers[] = { |
78 | NOTIFIER_ELEM_INIT(migration_state_notifiers, MIG_MODE_NORMAL), | |
79 | NOTIFIER_ELEM_INIT(migration_state_notifiers, MIG_MODE_CPR_REBOOT), | |
80 | }; | |
99a0db9b | 81 | |
da6f1790 JQ |
82 | /* Messages sent on the return path from destination to source */ |
83 | enum mig_rp_message_type { | |
84 | MIG_RP_MSG_INVALID = 0, /* Must be 0 */ | |
85 | MIG_RP_MSG_SHUT, /* sibling will not send any more RP messages */ | |
86 | MIG_RP_MSG_PONG, /* Response to a PING; data (seq: be32 ) */ | |
87 | ||
88 | MIG_RP_MSG_REQ_PAGES_ID, /* data (start: be64, len: be32, id: string) */ | |
89 | MIG_RP_MSG_REQ_PAGES, /* data (start: be64, len: be32) */ | |
a335debb | 90 | MIG_RP_MSG_RECV_BITMAP, /* send recved_bitmap back to source */ |
13955b89 | 91 | MIG_RP_MSG_RESUME_ACK, /* tell source that we are ready to resume */ |
1b4adb10 | 92 | MIG_RP_MSG_SWITCHOVER_ACK, /* Tell source it's OK to do switchover */ |
da6f1790 JQ |
93 | |
94 | MIG_RP_MSG_MAX | |
95 | }; | |
96 | ||
17549e84 JQ |
97 | /* When we add fault tolerance, we could have several |
98 | migrations at once. For now we don't need to add | |
99 | dynamic creation of migration */ | |
100 | ||
e5cb7e76 | 101 | static MigrationState *current_migration; |
e1b1b1bc | 102 | static MigrationIncomingState *current_incoming; |
e5cb7e76 | 103 | |
fa3673e4 | 104 | static GSList *migration_blockers[MIG_MODE__MAX]; |
3af8554b | 105 | |
8b0b29dc | 106 | static bool migration_object_check(MigrationState *ms, Error **errp); |
0331c8ca DDAG |
107 | static int migration_maybe_pause(MigrationState *s, |
108 | int *current_active_state, | |
109 | int new_state); | |
892ae715 | 110 | static void migrate_fd_cancel(MigrationState *s); |
7aa6070d | 111 | static bool close_return_path_on_source(MigrationState *s); |
63f64d77 | 112 | static void migration_completion_end(MigrationState *s); |
8b0b29dc | 113 | |
e22ffad0 PX |
114 | static void migration_downtime_start(MigrationState *s) |
115 | { | |
3e5f3bcd | 116 | trace_vmstate_downtime_checkpoint("src-downtime-start"); |
e22ffad0 PX |
117 | s->downtime_start = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); |
118 | } | |
119 | ||
120 | static void migration_downtime_end(MigrationState *s) | |
121 | { | |
122 | int64_t now = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); | |
123 | ||
124 | /* | |
125 | * If downtime already set, should mean that postcopy already set it, | |
126 | * then that should be the real downtime already. | |
127 | */ | |
128 | if (!s->downtime) { | |
129 | s->downtime = now - s->downtime_start; | |
130 | } | |
3e5f3bcd PX |
131 | |
132 | trace_vmstate_downtime_checkpoint("src-downtime-end"); | |
e22ffad0 PX |
133 | } |
134 | ||
d6f74fd1 PX |
135 | static bool migration_needs_multiple_sockets(void) |
136 | { | |
51b07548 | 137 | return migrate_multifd() || migrate_postcopy_preempt(); |
d6f74fd1 | 138 | } |
f444eeda | 139 | |
3205bebd | 140 | static bool transport_supports_multi_channels(MigrationAddress *addr) |
f444eeda | 141 | { |
3205bebd AH |
142 | if (addr->transport == MIGRATION_ADDRESS_TYPE_SOCKET) { |
143 | SocketAddress *saddr = &addr->u.socket; | |
144 | ||
f427d90b FR |
145 | return (saddr->type == SOCKET_ADDRESS_TYPE_INET || |
146 | saddr->type == SOCKET_ADDRESS_TYPE_UNIX || | |
147 | saddr->type == SOCKET_ADDRESS_TYPE_VSOCK); | |
148 | } else if (addr->transport == MIGRATION_ADDRESS_TYPE_FILE) { | |
149 | return migrate_mapped_ram(); | |
150 | } else { | |
151 | return false; | |
3205bebd | 152 | } |
f444eeda PX |
153 | } |
154 | ||
8d9e0d41 FR |
155 | static bool migration_needs_seekable_channel(void) |
156 | { | |
157 | return migrate_mapped_ram(); | |
158 | } | |
159 | ||
160 | static bool transport_supports_seeking(MigrationAddress *addr) | |
161 | { | |
162 | if (addr->transport == MIGRATION_ADDRESS_TYPE_FILE) { | |
163 | return true; | |
164 | } | |
165 | ||
8d9e0d41 FR |
166 | return false; |
167 | } | |
168 | ||
d6f74fd1 | 169 | static bool |
d95533e1 HG |
170 | migration_channels_and_transport_compatible(MigrationAddress *addr, |
171 | Error **errp) | |
f444eeda | 172 | { |
8d9e0d41 FR |
173 | if (migration_needs_seekable_channel() && |
174 | !transport_supports_seeking(addr)) { | |
175 | error_setg(errp, "Migration requires seekable transport (e.g. file)"); | |
176 | return false; | |
177 | } | |
178 | ||
d6f74fd1 | 179 | if (migration_needs_multiple_sockets() && |
3205bebd | 180 | !transport_supports_multi_channels(addr)) { |
d6f74fd1 PX |
181 | error_setg(errp, "Migration requires multi-channel URIs (e.g. tcp)"); |
182 | return false; | |
183 | } | |
184 | ||
185 | return true; | |
f444eeda PX |
186 | } |
187 | ||
8f8bfffc PX |
188 | static gint page_request_addr_cmp(gconstpointer ap, gconstpointer bp) |
189 | { | |
190 | uintptr_t a = (uintptr_t) ap, b = (uintptr_t) bp; | |
191 | ||
192 | return (a > b) - (a < b); | |
193 | } | |
194 | ||
9867d4dd | 195 | static int migration_stop_vm(MigrationState *s, RunState state) |
93bdf888 | 196 | { |
9867d4dd SS |
197 | int ret; |
198 | ||
199 | migration_downtime_start(s); | |
200 | ||
201 | s->vm_old_state = runstate_get(); | |
202 | global_state_store(); | |
203 | ||
204 | ret = vm_stop_force_state(state); | |
3e5f3bcd PX |
205 | |
206 | trace_vmstate_downtime_checkpoint("src-vm-stopped"); | |
9867d4dd | 207 | trace_migration_completion_vm_stop(ret); |
3e5f3bcd PX |
208 | |
209 | return ret; | |
93bdf888 PX |
210 | } |
211 | ||
e5cb7e76 PX |
212 | void migration_object_init(void) |
213 | { | |
214 | /* This can only be called once. */ | |
215 | assert(!current_migration); | |
216 | current_migration = MIGRATION_OBJ(object_new(TYPE_MIGRATION)); | |
4ffdb337 | 217 | |
e1b1b1bc PX |
218 | /* |
219 | * Init the migrate incoming object as well no matter whether | |
220 | * we'll use it or not. | |
221 | */ | |
222 | assert(!current_incoming); | |
223 | current_incoming = g_new0(MigrationIncomingState, 1); | |
224 | current_incoming->state = MIGRATION_STATUS_NONE; | |
225 | current_incoming->postcopy_remote_fds = | |
226 | g_array_new(FALSE, TRUE, sizeof(struct PostCopyFD)); | |
227 | qemu_mutex_init(¤t_incoming->rp_mutex); | |
60bb3c58 | 228 | qemu_mutex_init(¤t_incoming->postcopy_prio_thread_mutex); |
e1b1b1bc PX |
229 | qemu_event_init(¤t_incoming->main_thread_load_event, false); |
230 | qemu_sem_init(¤t_incoming->postcopy_pause_sem_dst, 0); | |
231 | qemu_sem_init(¤t_incoming->postcopy_pause_sem_fault, 0); | |
60bb3c58 | 232 | qemu_sem_init(¤t_incoming->postcopy_pause_sem_fast_load, 0); |
5655aab0 PX |
233 | qemu_sem_init(¤t_incoming->postcopy_qemufile_dst_done, 0); |
234 | ||
8f8bfffc | 235 | qemu_mutex_init(¤t_incoming->page_request_mutex); |
cf02f29e | 236 | qemu_cond_init(¤t_incoming->page_request_cond); |
8f8bfffc | 237 | current_incoming->page_requested = g_tree_new(page_request_addr_cmp); |
e1b1b1bc | 238 | |
dbea1c89 VSO |
239 | current_incoming->exit_on_error = INMIGRATE_DEFAULT_EXIT_ON_ERROR; |
240 | ||
f9734d5d | 241 | migration_object_check(current_migration, &error_fatal); |
e0d17dfd PB |
242 | |
243 | blk_mig_init(); | |
244 | ram_mig_init(); | |
245 | dirty_bitmap_mig_init(); | |
e5cb7e76 PX |
246 | } |
247 | ||
44d0d456 FR |
248 | typedef struct { |
249 | QEMUBH *bh; | |
250 | QEMUBHFunc *cb; | |
251 | void *opaque; | |
252 | } MigrationBH; | |
253 | ||
254 | static void migration_bh_dispatch_bh(void *opaque) | |
699d9476 | 255 | { |
44d0d456 FR |
256 | MigrationState *s = migrate_get_current(); |
257 | MigrationBH *migbh = opaque; | |
258 | ||
259 | /* cleanup this BH */ | |
260 | qemu_bh_delete(migbh->bh); | |
261 | migbh->bh = NULL; | |
262 | ||
263 | /* dispatch the other one */ | |
264 | migbh->cb(migbh->opaque); | |
265 | object_unref(OBJECT(s)); | |
266 | ||
267 | g_free(migbh); | |
268 | } | |
269 | ||
270 | void migration_bh_schedule(QEMUBHFunc *cb, void *opaque) | |
271 | { | |
272 | MigrationState *s = migrate_get_current(); | |
273 | MigrationBH *migbh = g_new0(MigrationBH, 1); | |
274 | QEMUBH *bh = qemu_bh_new(migration_bh_dispatch_bh, migbh); | |
275 | ||
276 | /* Store these to dispatch when the BH runs */ | |
277 | migbh->bh = bh; | |
278 | migbh->cb = cb; | |
279 | migbh->opaque = opaque; | |
280 | ||
699d9476 FR |
281 | /* |
282 | * Ref the state for bh, because it may be called when | |
283 | * there're already no other refs | |
284 | */ | |
285 | object_ref(OBJECT(s)); | |
286 | qemu_bh_schedule(bh); | |
287 | } | |
288 | ||
458fecca | 289 | void migration_cancel(const Error *error) |
c7c0e724 | 290 | { |
458fecca LV |
291 | if (error) { |
292 | migrate_set_error(current_migration, error); | |
293 | } | |
acac51ba HH |
294 | if (migrate_dirty_limit()) { |
295 | qmp_cancel_vcpu_dirty_limit(false, -1, NULL); | |
296 | } | |
c7c0e724 DH |
297 | migrate_fd_cancel(current_migration); |
298 | } | |
299 | ||
892ae715 | 300 | void migration_shutdown(void) |
1f895604 | 301 | { |
795969ab RL |
302 | /* |
303 | * When the QEMU main thread exit, the COLO thread | |
304 | * may wait a semaphore. So, we should wakeup the | |
305 | * COLO thread before migration shutdown. | |
306 | */ | |
307 | colo_shutdown(); | |
892ae715 DDAG |
308 | /* |
309 | * Cancel the current migration - that will (eventually) | |
310 | * stop the migration using this structure | |
311 | */ | |
458fecca | 312 | migration_cancel(NULL); |
1f895604 | 313 | object_unref(OBJECT(current_migration)); |
1499ab09 VSO |
314 | |
315 | /* | |
316 | * Cancel outgoing migration of dirty bitmaps. It should | |
317 | * at least unref used block nodes. | |
318 | */ | |
319 | dirty_bitmap_mig_cancel_outgoing(); | |
320 | ||
321 | /* | |
322 | * Cancel incoming migration of dirty bitmaps. Dirty bitmaps | |
323 | * are non-critical data, and their loss never considered as | |
324 | * something serious. | |
325 | */ | |
326 | dirty_bitmap_mig_cancel_incoming(); | |
1f895604 VSO |
327 | } |
328 | ||
bca7856a | 329 | /* For outgoing */ |
859bc756 | 330 | MigrationState *migrate_get_current(void) |
17549e84 | 331 | { |
e5cb7e76 PX |
332 | /* This can only be called after the object created. */ |
333 | assert(current_migration); | |
334 | return current_migration; | |
17549e84 JQ |
335 | } |
336 | ||
bca7856a DDAG |
337 | MigrationIncomingState *migration_incoming_get_current(void) |
338 | { | |
e1b1b1bc PX |
339 | assert(current_incoming); |
340 | return current_incoming; | |
bca7856a DDAG |
341 | } |
342 | ||
e031149c PX |
343 | void migration_incoming_transport_cleanup(MigrationIncomingState *mis) |
344 | { | |
345 | if (mis->socket_address_list) { | |
346 | qapi_free_SocketAddressList(mis->socket_address_list); | |
347 | mis->socket_address_list = NULL; | |
348 | } | |
349 | ||
350 | if (mis->transport_cleanup) { | |
351 | mis->transport_cleanup(mis->transport_data); | |
352 | mis->transport_data = mis->transport_cleanup = NULL; | |
353 | } | |
354 | } | |
355 | ||
bca7856a DDAG |
356 | void migration_incoming_state_destroy(void) |
357 | { | |
b4b076da JQ |
358 | struct MigrationIncomingState *mis = migration_incoming_get_current(); |
359 | ||
cde85c37 | 360 | multifd_recv_cleanup(); |
c323518a | 361 | compress_threads_load_cleanup(); |
cfc3bcf3 | 362 | |
3482655b | 363 | if (mis->to_src_file) { |
660819b1 PX |
364 | /* Tell source that we are done */ |
365 | migrate_send_rp_shut(mis, qemu_file_get_error(mis->from_src_file) != 0); | |
3482655b PX |
366 | qemu_fclose(mis->to_src_file); |
367 | mis->to_src_file = NULL; | |
368 | } | |
369 | ||
660819b1 | 370 | if (mis->from_src_file) { |
39675fff | 371 | migration_ioc_unregister_yank_from_file(mis->from_src_file); |
660819b1 PX |
372 | qemu_fclose(mis->from_src_file); |
373 | mis->from_src_file = NULL; | |
374 | } | |
00fa4fc8 DDAG |
375 | if (mis->postcopy_remote_fds) { |
376 | g_array_free(mis->postcopy_remote_fds, TRUE); | |
377 | mis->postcopy_remote_fds = NULL; | |
378 | } | |
660819b1 | 379 | |
e031149c | 380 | migration_incoming_transport_cleanup(mis); |
1783c00f DDAG |
381 | qemu_event_reset(&mis->main_thread_load_event); |
382 | ||
8f8bfffc PX |
383 | if (mis->page_requested) { |
384 | g_tree_destroy(mis->page_requested); | |
385 | mis->page_requested = NULL; | |
386 | } | |
387 | ||
36f62f11 PX |
388 | if (mis->postcopy_qemufile_dst) { |
389 | migration_ioc_unregister_yank_from_file(mis->postcopy_qemufile_dst); | |
390 | qemu_fclose(mis->postcopy_qemufile_dst); | |
391 | mis->postcopy_qemufile_dst = NULL; | |
392 | } | |
393 | ||
b5eea99e | 394 | yank_unregister_instance(MIGRATION_YANK_INSTANCE); |
bca7856a DDAG |
395 | } |
396 | ||
b05dc723 JQ |
397 | static void migrate_generate_event(int new_state) |
398 | { | |
b890902c | 399 | if (migrate_events()) { |
3ab72385 | 400 | qapi_event_send_migration(new_state); |
b05dc723 JQ |
401 | } |
402 | } | |
403 | ||
da6f1790 JQ |
404 | /* |
405 | * Send a message on the return channel back to the source | |
406 | * of the migration. | |
407 | */ | |
d6208e35 PX |
408 | static int migrate_send_rp_message(MigrationIncomingState *mis, |
409 | enum mig_rp_message_type message_type, | |
410 | uint16_t len, void *data) | |
da6f1790 | 411 | { |
d6208e35 PX |
412 | int ret = 0; |
413 | ||
da6f1790 | 414 | trace_migrate_send_rp_message((int)message_type, len); |
37396950 | 415 | QEMU_LOCK_GUARD(&mis->rp_mutex); |
d6208e35 PX |
416 | |
417 | /* | |
418 | * It's possible that the file handle got lost due to network | |
419 | * failures. | |
420 | */ | |
421 | if (!mis->to_src_file) { | |
422 | ret = -EIO; | |
37396950 | 423 | return ret; |
d6208e35 PX |
424 | } |
425 | ||
da6f1790 JQ |
426 | qemu_put_be16(mis->to_src_file, (unsigned int)message_type); |
427 | qemu_put_be16(mis->to_src_file, len); | |
428 | qemu_put_buffer(mis->to_src_file, data, len); | |
be07a0ed | 429 | return qemu_fflush(mis->to_src_file); |
da6f1790 JQ |
430 | } |
431 | ||
2e2bce16 PX |
432 | /* Request one page from the source VM at the given start address. |
433 | * rb: the RAMBlock to request the page in | |
1e2d90eb DDAG |
434 | * Start: Address offset within the RB |
435 | * Len: Length in bytes required - must be a multiple of pagesize | |
436 | */ | |
7a267fc4 PX |
437 | int migrate_send_rp_message_req_pages(MigrationIncomingState *mis, |
438 | RAMBlock *rb, ram_addr_t start) | |
1e2d90eb | 439 | { |
cb8d4c8f | 440 | uint8_t bufc[12 + 1 + 255]; /* start (8), len (4), rbname up to 256 */ |
1e2d90eb | 441 | size_t msglen = 12; /* start + len */ |
2e2bce16 | 442 | size_t len = qemu_ram_pagesize(rb); |
d6208e35 | 443 | enum mig_rp_message_type msg_type; |
2e2bce16 PX |
444 | const char *rbname; |
445 | int rbname_len; | |
1e2d90eb DDAG |
446 | |
447 | *(uint64_t *)bufc = cpu_to_be64((uint64_t)start); | |
448 | *(uint32_t *)(bufc + 8) = cpu_to_be32((uint32_t)len); | |
449 | ||
2e2bce16 PX |
450 | /* |
451 | * We maintain the last ramblock that we requested for page. Note that we | |
452 | * don't need locking because this function will only be called within the | |
453 | * postcopy ram fault thread. | |
454 | */ | |
455 | if (rb != mis->last_rb) { | |
456 | mis->last_rb = rb; | |
457 | ||
458 | rbname = qemu_ram_get_idstr(rb); | |
459 | rbname_len = strlen(rbname); | |
460 | ||
1e2d90eb DDAG |
461 | assert(rbname_len < 256); |
462 | ||
463 | bufc[msglen++] = rbname_len; | |
464 | memcpy(bufc + msglen, rbname, rbname_len); | |
465 | msglen += rbname_len; | |
d6208e35 | 466 | msg_type = MIG_RP_MSG_REQ_PAGES_ID; |
1e2d90eb | 467 | } else { |
d6208e35 | 468 | msg_type = MIG_RP_MSG_REQ_PAGES; |
1e2d90eb | 469 | } |
d6208e35 PX |
470 | |
471 | return migrate_send_rp_message(mis, msg_type, msglen, bufc); | |
1e2d90eb DDAG |
472 | } |
473 | ||
7a267fc4 | 474 | int migrate_send_rp_req_pages(MigrationIncomingState *mis, |
8f8bfffc | 475 | RAMBlock *rb, ram_addr_t start, uint64_t haddr) |
7a267fc4 | 476 | { |
7648297d | 477 | void *aligned = (void *)(uintptr_t)ROUND_DOWN(haddr, qemu_ram_pagesize(rb)); |
a2429283 | 478 | bool received = false; |
8f8bfffc PX |
479 | |
480 | WITH_QEMU_LOCK_GUARD(&mis->page_request_mutex) { | |
481 | received = ramblock_recv_bitmap_test_byte_offset(rb, start); | |
482 | if (!received && !g_tree_lookup(mis->page_requested, aligned)) { | |
483 | /* | |
484 | * The page has not been received, and it's not yet in the page | |
485 | * request list. Queue it. Set the value of element to 1, so that | |
486 | * things like g_tree_lookup() will return TRUE (1) when found. | |
487 | */ | |
488 | g_tree_insert(mis->page_requested, aligned, (gpointer)1); | |
cf02f29e | 489 | qatomic_inc(&mis->page_requested_count); |
8f8bfffc PX |
490 | trace_postcopy_page_req_add(aligned, mis->page_requested_count); |
491 | } | |
492 | } | |
493 | ||
494 | /* | |
495 | * If the page is there, skip sending the message. We don't even need the | |
496 | * lock because as long as the page arrived, it'll be there forever. | |
497 | */ | |
498 | if (received) { | |
499 | return 0; | |
500 | } | |
501 | ||
7a267fc4 PX |
502 | return migrate_send_rp_message_req_pages(mis, rb, start); |
503 | } | |
504 | ||
aad555c2 ZC |
505 | static bool migration_colo_enabled; |
506 | bool migration_incoming_colo_enabled(void) | |
507 | { | |
508 | return migration_colo_enabled; | |
509 | } | |
510 | ||
511 | void migration_incoming_disable_colo(void) | |
512 | { | |
18b1d3c9 | 513 | ram_block_discard_disable(false); |
aad555c2 ZC |
514 | migration_colo_enabled = false; |
515 | } | |
516 | ||
18b1d3c9 | 517 | int migration_incoming_enable_colo(void) |
aad555c2 | 518 | { |
51e47cf8 VSO |
519 | #ifndef CONFIG_REPLICATION |
520 | error_report("ENABLE_COLO command come in migration stream, but COLO " | |
521 | "module is not built in"); | |
522 | return -ENOTSUP; | |
523 | #endif | |
524 | ||
121ccedc VSO |
525 | if (!migrate_colo()) { |
526 | error_report("ENABLE_COLO command come in migration stream, but c-colo " | |
527 | "capability is not set"); | |
528 | return -EINVAL; | |
529 | } | |
530 | ||
18b1d3c9 DH |
531 | if (ram_block_discard_disable(true)) { |
532 | error_report("COLO: cannot disable RAM discard"); | |
533 | return -EBUSY; | |
534 | } | |
aad555c2 | 535 | migration_colo_enabled = true; |
18b1d3c9 | 536 | return 0; |
aad555c2 ZC |
537 | } |
538 | ||
9aca82ba JQ |
539 | void migrate_add_address(SocketAddress *address) |
540 | { | |
541 | MigrationIncomingState *mis = migration_incoming_get_current(); | |
9aca82ba | 542 | |
54aa3de7 EB |
543 | QAPI_LIST_PREPEND(mis->socket_address_list, |
544 | QAPI_CLONE(SocketAddress, address)); | |
9aca82ba JQ |
545 | } |
546 | ||
967f2de5 HG |
547 | bool migrate_uri_parse(const char *uri, MigrationChannel **channel, |
548 | Error **errp) | |
72a8192e | 549 | { |
5994024f | 550 | g_autoptr(MigrationChannel) val = g_new0(MigrationChannel, 1); |
72a8192e | 551 | g_autoptr(MigrationAddress) addr = g_new0(MigrationAddress, 1); |
72a8192e HG |
552 | InetSocketAddress *isock = &addr->u.rdma; |
553 | strList **tail = &addr->u.exec.args; | |
554 | ||
555 | if (strstart(uri, "exec:", NULL)) { | |
556 | addr->transport = MIGRATION_ADDRESS_TYPE_EXEC; | |
557 | #ifdef WIN32 | |
558 | QAPI_LIST_APPEND(tail, g_strdup(exec_get_cmd_path())); | |
559 | QAPI_LIST_APPEND(tail, g_strdup("/c")); | |
560 | #else | |
561 | QAPI_LIST_APPEND(tail, g_strdup("/bin/sh")); | |
562 | QAPI_LIST_APPEND(tail, g_strdup("-c")); | |
563 | #endif | |
564 | QAPI_LIST_APPEND(tail, g_strdup(uri + strlen("exec:"))); | |
565 | } else if (strstart(uri, "rdma:", NULL)) { | |
566 | if (inet_parse(isock, uri + strlen("rdma:"), errp)) { | |
567 | qapi_free_InetSocketAddress(isock); | |
568 | return false; | |
569 | } | |
570 | addr->transport = MIGRATION_ADDRESS_TYPE_RDMA; | |
571 | } else if (strstart(uri, "tcp:", NULL) || | |
572 | strstart(uri, "unix:", NULL) || | |
573 | strstart(uri, "vsock:", NULL) || | |
574 | strstart(uri, "fd:", NULL)) { | |
575 | addr->transport = MIGRATION_ADDRESS_TYPE_SOCKET; | |
41581265 | 576 | SocketAddress *saddr = socket_parse(uri, errp); |
72a8192e HG |
577 | if (!saddr) { |
578 | return false; | |
579 | } | |
580 | addr->u.socket.type = saddr->type; | |
581 | addr->u.socket.u = saddr->u; | |
41581265 ZZ |
582 | /* Don't free the objects inside; their ownership moved to "addr" */ |
583 | g_free(saddr); | |
72a8192e HG |
584 | } else if (strstart(uri, "file:", NULL)) { |
585 | addr->transport = MIGRATION_ADDRESS_TYPE_FILE; | |
586 | addr->u.file.filename = g_strdup(uri + strlen("file:")); | |
587 | if (file_parse_offset(addr->u.file.filename, &addr->u.file.offset, | |
588 | errp)) { | |
589 | return false; | |
590 | } | |
591 | } else { | |
592 | error_setg(errp, "unknown migration protocol: %s", uri); | |
593 | return false; | |
594 | } | |
595 | ||
5994024f HG |
596 | val->channel_type = MIGRATION_CHANNEL_TYPE_MAIN; |
597 | val->addr = g_steal_pointer(&addr); | |
598 | *channel = g_steal_pointer(&val); | |
72a8192e HG |
599 | return true; |
600 | } | |
601 | ||
074dbce5 HG |
602 | static void qemu_start_incoming_migration(const char *uri, bool has_channels, |
603 | MigrationChannelList *channels, | |
604 | Error **errp) | |
5bb7910a | 605 | { |
bc1d54ee | 606 | g_autoptr(MigrationChannel) channel = NULL; |
5994024f | 607 | MigrationAddress *addr = NULL; |
4111a732 | 608 | MigrationIncomingState *mis = migration_incoming_get_current(); |
34c9dd8e | 609 | |
074dbce5 HG |
610 | /* |
611 | * Having preliminary checks for uri and channel | |
612 | */ | |
0770ad43 HG |
613 | if (!uri == !channels) { |
614 | error_setg(errp, "need either 'uri' or 'channels' argument"); | |
074dbce5 | 615 | return; |
0770ad43 HG |
616 | } |
617 | ||
618 | if (channels) { | |
5994024f HG |
619 | /* To verify that Migrate channel list has only item */ |
620 | if (channels->next) { | |
621 | error_setg(errp, "Channel list has more than one entries"); | |
622 | return; | |
623 | } | |
bc1d54ee | 624 | addr = channels->value->addr; |
0770ad43 HG |
625 | } |
626 | ||
627 | if (uri) { | |
5994024f HG |
628 | /* caller uses the old URI syntax */ |
629 | if (!migrate_uri_parse(uri, &channel, errp)) { | |
630 | return; | |
631 | } | |
bc1d54ee | 632 | addr = channel->addr; |
074dbce5 | 633 | } |
d6f74fd1 | 634 | |
d95533e1 | 635 | /* transport mechanism not suitable for migration? */ |
5994024f | 636 | if (!migration_channels_and_transport_compatible(addr, errp)) { |
72a8192e HG |
637 | return; |
638 | } | |
639 | ||
4111a732 FR |
640 | migrate_set_state(&mis->state, MIGRATION_STATUS_NONE, |
641 | MIGRATION_STATUS_SETUP); | |
642 | ||
5994024f HG |
643 | if (addr->transport == MIGRATION_ADDRESS_TYPE_SOCKET) { |
644 | SocketAddress *saddr = &addr->u.socket; | |
34dfc5e4 HG |
645 | if (saddr->type == SOCKET_ADDRESS_TYPE_INET || |
646 | saddr->type == SOCKET_ADDRESS_TYPE_UNIX || | |
647 | saddr->type == SOCKET_ADDRESS_TYPE_VSOCK) { | |
648 | socket_start_incoming_migration(saddr, errp); | |
649 | } else if (saddr->type == SOCKET_ADDRESS_TYPE_FD) { | |
650 | fd_start_incoming_migration(saddr->u.fd.str, errp); | |
651 | } | |
2da776db | 652 | #ifdef CONFIG_RDMA |
5994024f | 653 | } else if (addr->transport == MIGRATION_ADDRESS_TYPE_RDMA) { |
b88a3306 JQ |
654 | if (migrate_compress()) { |
655 | error_setg(errp, "RDMA and compression can't be used together"); | |
656 | return; | |
657 | } | |
658 | if (migrate_xbzrle()) { | |
659 | error_setg(errp, "RDMA and XBZRLE can't be used together"); | |
660 | return; | |
661 | } | |
662 | if (migrate_multifd()) { | |
663 | error_setg(errp, "RDMA and multifd can't be used together"); | |
664 | return; | |
665 | } | |
5994024f | 666 | rdma_start_incoming_migration(&addr->u.rdma, errp); |
2da776db | 667 | #endif |
5994024f HG |
668 | } else if (addr->transport == MIGRATION_ADDRESS_TYPE_EXEC) { |
669 | exec_start_incoming_migration(addr->u.exec.args, errp); | |
670 | } else if (addr->transport == MIGRATION_ADDRESS_TYPE_FILE) { | |
671 | file_start_incoming_migration(&addr->u.file, errp); | |
adde220a | 672 | } else { |
312fd5f2 | 673 | error_setg(errp, "unknown migration protocol: %s", uri); |
8ca5e801 | 674 | } |
5bb7910a AL |
675 | } |
676 | ||
0aa6aefc DL |
677 | static void process_incoming_migration_bh(void *opaque) |
678 | { | |
679 | Error *local_err = NULL; | |
680 | MigrationIncomingState *mis = opaque; | |
681 | ||
3e5f3bcd PX |
682 | trace_vmstate_downtime_checkpoint("dst-precopy-bh-enter"); |
683 | ||
0f073f44 DDAG |
684 | /* If capability late_block_activate is set: |
685 | * Only fire up the block code now if we're going to restart the | |
686 | * VM, else 'cont' will do it. | |
687 | * This causes file locking to happen; so we don't want it to happen | |
688 | * unless we really are starting the VM. | |
689 | */ | |
690 | if (!migrate_late_block_activate() || | |
691 | (autostart && (!global_state_received() || | |
b4e9ddcc | 692 | runstate_is_live(global_state_get_runstate())))) { |
3b717194 | 693 | /* Make sure all file formats throw away their mutable metadata. |
0f073f44 | 694 | * If we get an error here, just don't restart the VM yet. */ |
3b717194 | 695 | bdrv_activate_all(&local_err); |
0f073f44 DDAG |
696 | if (local_err) { |
697 | error_report_err(local_err); | |
698 | local_err = NULL; | |
699 | autostart = false; | |
700 | } | |
d35ff5e6 KW |
701 | } |
702 | ||
0aa6aefc DL |
703 | /* |
704 | * This must happen after all error conditions are dealt with and | |
705 | * we're sure the VM is going to be running on this host. | |
706 | */ | |
7659505c | 707 | qemu_announce_self(&mis->announce_timer, migrate_announce_params()); |
0aa6aefc | 708 | |
3e5f3bcd PX |
709 | trace_vmstate_downtime_checkpoint("dst-precopy-bh-announced"); |
710 | ||
cde85c37 | 711 | multifd_recv_shutdown(); |
0aa6aefc | 712 | |
b35ebdf0 VSO |
713 | dirty_bitmap_mig_before_vm_start(); |
714 | ||
0aa6aefc | 715 | if (!global_state_received() || |
b4e9ddcc | 716 | runstate_is_live(global_state_get_runstate())) { |
0aa6aefc DL |
717 | if (autostart) { |
718 | vm_start(); | |
719 | } else { | |
720 | runstate_set(RUN_STATE_PAUSED); | |
721 | } | |
db009729 ZC |
722 | } else if (migration_incoming_colo_enabled()) { |
723 | migration_incoming_disable_colo(); | |
724 | vm_start(); | |
0aa6aefc DL |
725 | } else { |
726 | runstate_set(global_state_get_runstate()); | |
727 | } | |
3e5f3bcd | 728 | trace_vmstate_downtime_checkpoint("dst-precopy-bh-vm-started"); |
0aa6aefc DL |
729 | /* |
730 | * This must happen after any state changes since as soon as an external | |
731 | * observer sees this event they might start to prod at the VM assuming | |
732 | * it's ready to use. | |
733 | */ | |
734 | migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE, | |
735 | MIGRATION_STATUS_COMPLETED); | |
0aa6aefc DL |
736 | migration_incoming_state_destroy(); |
737 | } | |
738 | ||
38e8f9af MAL |
739 | static void coroutine_fn |
740 | process_incoming_migration_co(void *opaque) | |
511c0231 | 741 | { |
f84eaa9f | 742 | MigrationState *s = migrate_get_current(); |
b4b076da | 743 | MigrationIncomingState *mis = migration_incoming_get_current(); |
e9bef235 | 744 | PostcopyState ps; |
1c12e1f5 | 745 | int ret; |
f84eaa9f | 746 | Error *local_err = NULL; |
1c12e1f5 | 747 | |
4f0fae7f | 748 | assert(mis->from_src_file); |
c323518a LS |
749 | |
750 | if (compress_threads_load_setup(mis->from_src_file)) { | |
f84eaa9f | 751 | error_setg(&local_err, "Failed to setup decompress threads"); |
c323518a LS |
752 | goto fail; |
753 | } | |
754 | ||
67f11b5c | 755 | mis->largest_page_size = qemu_ram_pagesize_largest(); |
093e3c42 | 756 | postcopy_state_set(POSTCOPY_INCOMING_NONE); |
4111a732 | 757 | migrate_set_state(&mis->state, MIGRATION_STATUS_SETUP, |
93d7af6f | 758 | MIGRATION_STATUS_ACTIVE); |
dd42ce24 VSO |
759 | |
760 | mis->loadvm_co = qemu_coroutine_self(); | |
4f0fae7f | 761 | ret = qemu_loadvm_state(mis->from_src_file); |
dd42ce24 | 762 | mis->loadvm_co = NULL; |
bca7856a | 763 | |
3e5f3bcd PX |
764 | trace_vmstate_downtime_checkpoint("dst-precopy-loadvm-completed"); |
765 | ||
e9bef235 DDAG |
766 | ps = postcopy_state_get(); |
767 | trace_process_incoming_migration_co_end(ret, ps); | |
768 | if (ps != POSTCOPY_INCOMING_NONE) { | |
769 | if (ps == POSTCOPY_INCOMING_ADVISE) { | |
770 | /* | |
771 | * Where a migration had postcopy enabled (and thus went to advise) | |
772 | * but managed to complete within the precopy period, we can use | |
773 | * the normal exit. | |
774 | */ | |
775 | postcopy_ram_incoming_cleanup(mis); | |
776 | } else if (ret >= 0) { | |
777 | /* | |
778 | * Postcopy was started, cleanup should happen at the end of the | |
779 | * postcopy thread. | |
780 | */ | |
781 | trace_process_incoming_migration_co_postcopy_end_main(); | |
782 | return; | |
783 | } | |
784 | /* Else if something went wrong then just fall out of the normal exit */ | |
785 | } | |
786 | ||
ecbfec6d | 787 | if (ret < 0) { |
f84eaa9f | 788 | error_setg(&local_err, "load of migration failed: %s", strerror(-ret)); |
ecbfec6d VSO |
789 | goto fail; |
790 | } | |
791 | ||
d0a14a2b | 792 | if (colo_incoming_co() < 0) { |
f84eaa9f | 793 | error_setg(&local_err, "colo incoming failed"); |
d0a14a2b | 794 | goto fail; |
25d0c16f HZ |
795 | } |
796 | ||
44d0d456 | 797 | migration_bh_schedule(process_incoming_migration_bh, mis); |
6d99c2d4 FL |
798 | return; |
799 | fail: | |
6d99c2d4 FL |
800 | migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE, |
801 | MIGRATION_STATUS_FAILED); | |
f84eaa9f VSO |
802 | migrate_set_error(s, local_err); |
803 | error_free(local_err); | |
804 | ||
246f54e0 | 805 | migration_incoming_state_destroy(); |
e5bac1f5 | 806 | |
dbea1c89 VSO |
807 | if (mis->exit_on_error) { |
808 | WITH_QEMU_LOCK_GUARD(&s->error_mutex) { | |
809 | error_report_err(s->error); | |
810 | s->error = NULL; | |
811 | } | |
f84eaa9f | 812 | |
dbea1c89 VSO |
813 | exit(EXIT_FAILURE); |
814 | } | |
511c0231 JQ |
815 | } |
816 | ||
b673eab4 | 817 | /** |
7d6f6933 | 818 | * migration_incoming_setup: Setup incoming migration |
b673eab4 | 819 | * @f: file for main migration channel |
b673eab4 | 820 | */ |
c606f3ba | 821 | static void migration_incoming_setup(QEMUFile *f) |
82a4da79 | 822 | { |
4f0fae7f | 823 | MigrationIncomingState *mis = migration_incoming_get_current(); |
82a4da79 | 824 | |
4f0fae7f JQ |
825 | if (!mis->from_src_file) { |
826 | mis->from_src_file = f; | |
827 | } | |
06ad5135 | 828 | qemu_file_set_blocking(f, false); |
e595a01a JQ |
829 | } |
830 | ||
36c2f8be | 831 | void migration_incoming_process(void) |
e595a01a JQ |
832 | { |
833 | Coroutine *co = qemu_coroutine_create(process_incoming_migration_co, NULL); | |
0b8b8753 | 834 | qemu_coroutine_enter(co); |
82a4da79 PB |
835 | } |
836 | ||
884835fa | 837 | /* Returns true if recovered from a paused migration, otherwise false */ |
a39e9339 | 838 | static bool postcopy_try_recover(void) |
e595a01a | 839 | { |
d96c9e8d PX |
840 | MigrationIncomingState *mis = migration_incoming_get_current(); |
841 | ||
842 | if (mis->state == MIGRATION_STATUS_POSTCOPY_PAUSED) { | |
843 | /* Resumed from a paused postcopy migration */ | |
844 | ||
a39e9339 PX |
845 | /* This should be set already in migration_incoming_setup() */ |
846 | assert(mis->from_src_file); | |
d96c9e8d | 847 | /* Postcopy has standalone thread to do vm load */ |
a39e9339 | 848 | qemu_file_set_blocking(mis->from_src_file, true); |
d96c9e8d PX |
849 | |
850 | /* Re-configure the return path */ | |
a39e9339 | 851 | mis->to_src_file = qemu_file_get_return_path(mis->from_src_file); |
d96c9e8d PX |
852 | |
853 | migrate_set_state(&mis->state, MIGRATION_STATUS_POSTCOPY_PAUSED, | |
854 | MIGRATION_STATUS_POSTCOPY_RECOVER); | |
855 | ||
856 | /* | |
857 | * Here, we only wake up the main loading thread (while the | |
60bb3c58 | 858 | * rest threads will still be waiting), so that we can receive |
d96c9e8d | 859 | * commands from source now, and answer it if needed. The |
60bb3c58 | 860 | * rest threads will be woken up afterwards until we are sure |
d96c9e8d PX |
861 | * that source is ready to reply to page requests. |
862 | */ | |
863 | qemu_sem_post(&mis->postcopy_pause_sem_dst); | |
884835fa PX |
864 | return true; |
865 | } | |
866 | ||
867 | return false; | |
868 | } | |
869 | ||
b0cf3bfc | 870 | void migration_fd_process_incoming(QEMUFile *f) |
884835fa | 871 | { |
c606f3ba | 872 | migration_incoming_setup(f); |
a39e9339 | 873 | if (postcopy_try_recover()) { |
b673eab4 JQ |
874 | return; |
875 | } | |
884835fa | 876 | migration_incoming_process(); |
e595a01a JQ |
877 | } |
878 | ||
5655aab0 PX |
879 | /* |
880 | * Returns true when we want to start a new incoming migration process, | |
881 | * false otherwise. | |
882 | */ | |
883 | static bool migration_should_start_incoming(bool main_channel) | |
884 | { | |
885 | /* Multifd doesn't start unless all channels are established */ | |
51b07548 | 886 | if (migrate_multifd()) { |
5655aab0 PX |
887 | return migration_has_all_channels(); |
888 | } | |
889 | ||
890 | /* Preempt channel only starts when the main channel is created */ | |
891 | if (migrate_postcopy_preempt()) { | |
892 | return main_channel; | |
893 | } | |
894 | ||
895 | /* | |
896 | * For all the rest types of migration, we should only reach here when | |
897 | * it's the main channel that's being created, and we should always | |
898 | * proceed with this channel. | |
899 | */ | |
900 | assert(main_channel); | |
901 | return true; | |
902 | } | |
903 | ||
49ed0d24 | 904 | void migration_ioc_process_incoming(QIOChannel *ioc, Error **errp) |
4f0fae7f JQ |
905 | { |
906 | MigrationIncomingState *mis = migration_incoming_get_current(); | |
b673eab4 | 907 | Error *local_err = NULL; |
36f62f11 | 908 | QEMUFile *f; |
6720c2b3 | 909 | bool default_channel = true; |
910 | uint32_t channel_magic = 0; | |
911 | int ret = 0; | |
4f0fae7f | 912 | |
2dd7ee7a FR |
913 | if (migrate_multifd() && !migrate_mapped_ram() && |
914 | !migrate_postcopy_ram() && | |
6720c2b3 | 915 | qio_channel_has_feature(ioc, QIO_CHANNEL_FEATURE_READ_MSG_PEEK)) { |
916 | /* | |
917 | * With multiple channels, it is possible that we receive channels | |
918 | * out of order on destination side, causing incorrect mapping of | |
919 | * source channels on destination side. Check channel MAGIC to | |
920 | * decide type of channel. Please note this is best effort, postcopy | |
921 | * preempt channel does not send any magic number so avoid it for | |
922 | * postcopy live migration. Also tls live migration already does | |
923 | * tls handshake while initializing main channel so with tls this | |
924 | * issue is not possible. | |
925 | */ | |
926 | ret = migration_channel_read_peek(ioc, (void *)&channel_magic, | |
b6f4c0c7 | 927 | sizeof(channel_magic), errp); |
6720c2b3 | 928 | |
929 | if (ret != 0) { | |
6720c2b3 | 930 | return; |
931 | } | |
932 | ||
933 | default_channel = (channel_magic == cpu_to_be32(QEMU_VM_FILE_MAGIC)); | |
934 | } else { | |
935 | default_channel = !mis->from_src_file; | |
936 | } | |
937 | ||
cde85c37 | 938 | if (multifd_recv_setup(errp) != 0) { |
6720c2b3 | 939 | return; |
940 | } | |
941 | ||
942 | if (default_channel) { | |
36f62f11 | 943 | f = qemu_file_new_input(ioc); |
c606f3ba | 944 | migration_incoming_setup(f); |
a429e7f4 PX |
945 | } else { |
946 | /* Multiple connections */ | |
36f62f11 | 947 | assert(migration_needs_multiple_sockets()); |
51b07548 | 948 | if (migrate_multifd()) { |
6720c2b3 | 949 | multifd_recv_new_channel(ioc, &local_err); |
36f62f11 PX |
950 | } else { |
951 | assert(migrate_postcopy_preempt()); | |
952 | f = qemu_file_new_input(ioc); | |
6720c2b3 | 953 | postcopy_preempt_new_channel(mis, f); |
36f62f11 | 954 | } |
49ed0d24 FL |
955 | if (local_err) { |
956 | error_propagate(errp, local_err); | |
957 | return; | |
958 | } | |
4f0fae7f | 959 | } |
81e62053 | 960 | |
5655aab0 | 961 | if (migration_should_start_incoming(default_channel)) { |
a39e9339 PX |
962 | /* If it's a recovery, we're done */ |
963 | if (postcopy_try_recover()) { | |
964 | return; | |
965 | } | |
81e62053 PX |
966 | migration_incoming_process(); |
967 | } | |
4f0fae7f JQ |
968 | } |
969 | ||
428d8908 JQ |
970 | /** |
971 | * @migration_has_all_channels: We have received all channels that we need | |
972 | * | |
973 | * Returns true when we have got connections to all the channels that | |
974 | * we need for migration. | |
975 | */ | |
976 | bool migration_has_all_channels(void) | |
977 | { | |
ca273df3 | 978 | MigrationIncomingState *mis = migration_incoming_get_current(); |
62c1e0ca | 979 | |
36f62f11 PX |
980 | if (!mis->from_src_file) { |
981 | return false; | |
982 | } | |
62c1e0ca | 983 | |
51b07548 | 984 | if (migrate_multifd()) { |
36f62f11 PX |
985 | return multifd_recv_all_channels_created(); |
986 | } | |
987 | ||
988 | if (migrate_postcopy_preempt()) { | |
989 | return mis->postcopy_qemufile_dst != NULL; | |
990 | } | |
991 | ||
992 | return true; | |
428d8908 JQ |
993 | } |
994 | ||
1b4adb10 AH |
995 | int migrate_send_rp_switchover_ack(MigrationIncomingState *mis) |
996 | { | |
997 | return migrate_send_rp_message(mis, MIG_RP_MSG_SWITCHOVER_ACK, 0, NULL); | |
998 | } | |
999 | ||
6decec93 DDAG |
1000 | /* |
1001 | * Send a 'SHUT' message on the return channel with the given value | |
1002 | * to indicate that we've finished with the RP. Non-0 value indicates | |
1003 | * error. | |
1004 | */ | |
1005 | void migrate_send_rp_shut(MigrationIncomingState *mis, | |
1006 | uint32_t value) | |
1007 | { | |
1008 | uint32_t buf; | |
1009 | ||
1010 | buf = cpu_to_be32(value); | |
1011 | migrate_send_rp_message(mis, MIG_RP_MSG_SHUT, sizeof(buf), &buf); | |
1012 | } | |
1013 | ||
1014 | /* | |
1015 | * Send a 'PONG' message on the return channel with the given value | |
1016 | * (normally in response to a 'PING') | |
1017 | */ | |
1018 | void migrate_send_rp_pong(MigrationIncomingState *mis, | |
1019 | uint32_t value) | |
1020 | { | |
1021 | uint32_t buf; | |
1022 | ||
1023 | buf = cpu_to_be32(value); | |
1024 | migrate_send_rp_message(mis, MIG_RP_MSG_PONG, sizeof(buf), &buf); | |
1025 | } | |
1026 | ||
a335debb PX |
1027 | void migrate_send_rp_recv_bitmap(MigrationIncomingState *mis, |
1028 | char *block_name) | |
1029 | { | |
1030 | char buf[512]; | |
1031 | int len; | |
1032 | int64_t res; | |
1033 | ||
1034 | /* | |
1035 | * First, we send the header part. It contains only the len of | |
1036 | * idstr, and the idstr itself. | |
1037 | */ | |
1038 | len = strlen(block_name); | |
1039 | buf[0] = len; | |
1040 | memcpy(buf + 1, block_name, len); | |
1041 | ||
1042 | if (mis->state != MIGRATION_STATUS_POSTCOPY_RECOVER) { | |
1043 | error_report("%s: MSG_RP_RECV_BITMAP only used for recovery", | |
1044 | __func__); | |
1045 | return; | |
1046 | } | |
1047 | ||
1048 | migrate_send_rp_message(mis, MIG_RP_MSG_RECV_BITMAP, len + 1, buf); | |
1049 | ||
1050 | /* | |
1051 | * Next, we dump the received bitmap to the stream. | |
1052 | * | |
1053 | * TODO: currently we are safe since we are the only one that is | |
1054 | * using the to_src_file handle (fault thread is still paused), | |
1055 | * and it's ok even not taking the mutex. However the best way is | |
1056 | * to take the lock before sending the message header, and release | |
1057 | * the lock after sending the bitmap. | |
1058 | */ | |
1059 | qemu_mutex_lock(&mis->rp_mutex); | |
1060 | res = ramblock_recv_bitmap_send(mis->to_src_file, block_name); | |
1061 | qemu_mutex_unlock(&mis->rp_mutex); | |
1062 | ||
1063 | trace_migrate_send_rp_recv_bitmap(block_name, res); | |
1064 | } | |
1065 | ||
13955b89 PX |
1066 | void migrate_send_rp_resume_ack(MigrationIncomingState *mis, uint32_t value) |
1067 | { | |
1068 | uint32_t buf; | |
1069 | ||
1070 | buf = cpu_to_be32(value); | |
1071 | migrate_send_rp_message(mis, MIG_RP_MSG_RESUME_ACK, sizeof(buf), &buf); | |
1072 | } | |
1073 | ||
f6844b99 DDAG |
1074 | /* |
1075 | * Return true if we're already in the middle of a migration | |
1076 | * (i.e. any of the active or setup states) | |
1077 | */ | |
7dcb3c87 | 1078 | bool migration_is_setup_or_active(void) |
f6844b99 | 1079 | { |
7dcb3c87 SS |
1080 | MigrationState *s = current_migration; |
1081 | ||
1082 | switch (s->state) { | |
f6844b99 | 1083 | case MIGRATION_STATUS_ACTIVE: |
9ec055ae | 1084 | case MIGRATION_STATUS_POSTCOPY_ACTIVE: |
a688d2c1 | 1085 | case MIGRATION_STATUS_POSTCOPY_PAUSED: |
135b87b4 | 1086 | case MIGRATION_STATUS_POSTCOPY_RECOVER: |
f6844b99 | 1087 | case MIGRATION_STATUS_SETUP: |
31e06077 DDAG |
1088 | case MIGRATION_STATUS_PRE_SWITCHOVER: |
1089 | case MIGRATION_STATUS_DEVICE: | |
c7e0acd5 | 1090 | case MIGRATION_STATUS_WAIT_UNPLUG: |
19dd408a | 1091 | case MIGRATION_STATUS_COLO: |
f6844b99 DDAG |
1092 | return true; |
1093 | ||
1094 | default: | |
1095 | return false; | |
1096 | ||
1097 | } | |
1098 | } | |
1099 | ||
aeaafb1e | 1100 | bool migration_is_running(void) |
392d87e2 | 1101 | { |
aeaafb1e SS |
1102 | MigrationState *s = current_migration; |
1103 | ||
1104 | switch (s->state) { | |
392d87e2 JQ |
1105 | case MIGRATION_STATUS_ACTIVE: |
1106 | case MIGRATION_STATUS_POSTCOPY_ACTIVE: | |
1107 | case MIGRATION_STATUS_POSTCOPY_PAUSED: | |
1108 | case MIGRATION_STATUS_POSTCOPY_RECOVER: | |
1109 | case MIGRATION_STATUS_SETUP: | |
1110 | case MIGRATION_STATUS_PRE_SWITCHOVER: | |
1111 | case MIGRATION_STATUS_DEVICE: | |
1112 | case MIGRATION_STATUS_WAIT_UNPLUG: | |
1113 | case MIGRATION_STATUS_CANCELLING: | |
392d87e2 JQ |
1114 | return true; |
1115 | ||
1116 | default: | |
1117 | return false; | |
1118 | ||
1119 | } | |
1120 | } | |
1121 | ||
db18dee7 PX |
1122 | static bool migrate_show_downtime(MigrationState *s) |
1123 | { | |
1124 | return (s->state == MIGRATION_STATUS_COMPLETED) || migration_in_postcopy(); | |
1125 | } | |
1126 | ||
640dfb14 WY |
1127 | static void populate_time_info(MigrationInfo *info, MigrationState *s) |
1128 | { | |
1129 | info->has_status = true; | |
1130 | info->has_setup_time = true; | |
1131 | info->setup_time = s->setup_time; | |
db18dee7 | 1132 | |
640dfb14 WY |
1133 | if (s->state == MIGRATION_STATUS_COMPLETED) { |
1134 | info->has_total_time = true; | |
1135 | info->total_time = s->total_time; | |
640dfb14 WY |
1136 | } else { |
1137 | info->has_total_time = true; | |
1138 | info->total_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME) - | |
1139 | s->start_time; | |
db18dee7 PX |
1140 | } |
1141 | ||
1142 | if (migrate_show_downtime(s)) { | |
1143 | info->has_downtime = true; | |
1144 | info->downtime = s->downtime; | |
1145 | } else { | |
640dfb14 WY |
1146 | info->has_expected_downtime = true; |
1147 | info->expected_downtime = s->expected_downtime; | |
1148 | } | |
1149 | } | |
1150 | ||
a22463a5 DDAG |
1151 | static void populate_ram_info(MigrationInfo *info, MigrationState *s) |
1152 | { | |
144fa06b JQ |
1153 | size_t page_size = qemu_target_page_size(); |
1154 | ||
a22463a5 | 1155 | info->ram = g_malloc0(sizeof(*info->ram)); |
897fd8bd | 1156 | info->ram->transferred = migration_transferred_bytes(); |
a22463a5 | 1157 | info->ram->total = ram_bytes_total(); |
aff3f660 | 1158 | info->ram->duplicate = stat64_get(&mig_stats.zero_pages); |
bedf53c1 JQ |
1159 | /* legacy value. It is not used anymore */ |
1160 | info->ram->skipped = 0; | |
aff3f660 | 1161 | info->ram->normal = stat64_get(&mig_stats.normal_pages); |
23b7576d | 1162 | info->ram->normal_bytes = info->ram->normal * page_size; |
a22463a5 | 1163 | info->ram->mbps = s->mbps; |
536b5a4e | 1164 | info->ram->dirty_sync_count = |
aff3f660 | 1165 | stat64_get(&mig_stats.dirty_sync_count); |
cf20c897 | 1166 | info->ram->dirty_sync_missed_zero_copy = |
aff3f660 | 1167 | stat64_get(&mig_stats.dirty_sync_missed_zero_copy); |
3c764f9b | 1168 | info->ram->postcopy_requests = |
aff3f660 | 1169 | stat64_get(&mig_stats.postcopy_requests); |
144fa06b | 1170 | info->ram->page_size = page_size; |
aff3f660 | 1171 | info->ram->multifd_bytes = stat64_get(&mig_stats.multifd_bytes); |
aecbfe9c | 1172 | info->ram->pages_per_second = s->pages_per_second; |
aff3f660 JQ |
1173 | info->ram->precopy_bytes = stat64_get(&mig_stats.precopy_bytes); |
1174 | info->ram->downtime_bytes = stat64_get(&mig_stats.downtime_bytes); | |
1175 | info->ram->postcopy_bytes = stat64_get(&mig_stats.postcopy_bytes); | |
a22463a5 | 1176 | |
87dca0c9 | 1177 | if (migrate_xbzrle()) { |
114f5aee JQ |
1178 | info->xbzrle_cache = g_malloc0(sizeof(*info->xbzrle_cache)); |
1179 | info->xbzrle_cache->cache_size = migrate_xbzrle_cache_size(); | |
9360447d JQ |
1180 | info->xbzrle_cache->bytes = xbzrle_counters.bytes; |
1181 | info->xbzrle_cache->pages = xbzrle_counters.pages; | |
1182 | info->xbzrle_cache->cache_miss = xbzrle_counters.cache_miss; | |
1183 | info->xbzrle_cache->cache_miss_rate = xbzrle_counters.cache_miss_rate; | |
e460a4b1 | 1184 | info->xbzrle_cache->encoding_rate = xbzrle_counters.encoding_rate; |
9360447d | 1185 | info->xbzrle_cache->overflow = xbzrle_counters.overflow; |
114f5aee JQ |
1186 | } |
1187 | ||
6f609005 | 1188 | populate_compress(info); |
76e03000 | 1189 | |
338182c8 JQ |
1190 | if (cpu_throttle_active()) { |
1191 | info->has_cpu_throttle_percentage = true; | |
1192 | info->cpu_throttle_percentage = cpu_throttle_get_percentage(); | |
1193 | } | |
1194 | ||
a22463a5 DDAG |
1195 | if (s->state != MIGRATION_STATUS_COMPLETED) { |
1196 | info->ram->remaining = ram_bytes_remaining(); | |
72f8e587 | 1197 | info->ram->dirty_pages_rate = |
aff3f660 | 1198 | stat64_get(&mig_stats.dirty_pages_rate); |
a22463a5 | 1199 | } |
15699cf5 HH |
1200 | |
1201 | if (migrate_dirty_limit() && dirtylimit_in_service()) { | |
1202 | info->has_dirty_limit_throttle_time_per_round = true; | |
1203 | info->dirty_limit_throttle_time_per_round = | |
1204 | dirtylimit_throttle_time_per_round(); | |
1205 | ||
1206 | info->has_dirty_limit_ring_full_time = true; | |
1207 | info->dirty_limit_ring_full_time = dirtylimit_ring_full_time(); | |
1208 | } | |
a22463a5 DDAG |
1209 | } |
1210 | ||
930ac04c JQ |
1211 | static void populate_disk_info(MigrationInfo *info) |
1212 | { | |
1213 | if (blk_mig_active()) { | |
930ac04c JQ |
1214 | info->disk = g_malloc0(sizeof(*info->disk)); |
1215 | info->disk->transferred = blk_mig_bytes_transferred(); | |
1216 | info->disk->remaining = blk_mig_bytes_remaining(); | |
1217 | info->disk->total = blk_mig_bytes_total(); | |
1218 | } | |
1219 | } | |
1220 | ||
65ace060 | 1221 | static void fill_source_migration_info(MigrationInfo *info) |
5bb7910a | 1222 | { |
17549e84 | 1223 | MigrationState *s = migrate_get_current(); |
552de79b | 1224 | int state = qatomic_read(&s->state); |
fa3673e4 | 1225 | GSList *cur_blocker = migration_blockers[migrate_mode()]; |
17549e84 | 1226 | |
3af8554b | 1227 | info->blocked_reasons = NULL; |
3af8554b | 1228 | |
372043f3 MA |
1229 | /* |
1230 | * There are two types of reasons a migration might be blocked; | |
1231 | * a) devices marked in VMState as non-migratable, and | |
1232 | * b) Explicit migration blockers | |
1233 | * We need to add both of them here. | |
1234 | */ | |
1235 | qemu_savevm_non_migratable_list(&info->blocked_reasons); | |
3af8554b | 1236 | |
372043f3 MA |
1237 | while (cur_blocker) { |
1238 | QAPI_LIST_PREPEND(info->blocked_reasons, | |
1239 | g_strdup(error_get_pretty(cur_blocker->data))); | |
1240 | cur_blocker = g_slist_next(cur_blocker); | |
3af8554b | 1241 | } |
372043f3 | 1242 | info->has_blocked_reasons = info->blocked_reasons != NULL; |
3af8554b | 1243 | |
552de79b | 1244 | switch (state) { |
31194731 | 1245 | case MIGRATION_STATUS_NONE: |
17549e84 | 1246 | /* no migration has happened ever */ |
65ace060 AP |
1247 | /* do not overwrite destination migration status */ |
1248 | return; | |
31194731 | 1249 | case MIGRATION_STATUS_SETUP: |
29ae8a41 | 1250 | info->has_status = true; |
ed4fbd10 | 1251 | info->has_total_time = false; |
29ae8a41 | 1252 | break; |
31194731 HZ |
1253 | case MIGRATION_STATUS_ACTIVE: |
1254 | case MIGRATION_STATUS_CANCELLING: | |
9ec055ae | 1255 | case MIGRATION_STATUS_POSTCOPY_ACTIVE: |
31e06077 DDAG |
1256 | case MIGRATION_STATUS_PRE_SWITCHOVER: |
1257 | case MIGRATION_STATUS_DEVICE: | |
a688d2c1 | 1258 | case MIGRATION_STATUS_POSTCOPY_PAUSED: |
135b87b4 | 1259 | case MIGRATION_STATUS_POSTCOPY_RECOVER: |
640dfb14 WY |
1260 | /* TODO add some postcopy stats */ |
1261 | populate_time_info(info, s); | |
a22463a5 | 1262 | populate_ram_info(info, s); |
930ac04c | 1263 | populate_disk_info(info); |
38c482b4 | 1264 | migration_populate_vfio_info(info); |
17549e84 | 1265 | break; |
0b827d5e HZ |
1266 | case MIGRATION_STATUS_COLO: |
1267 | info->has_status = true; | |
1268 | /* TODO: display COLO specific information (checkpoint info etc.) */ | |
1269 | break; | |
31194731 | 1270 | case MIGRATION_STATUS_COMPLETED: |
640dfb14 | 1271 | populate_time_info(info, s); |
a22463a5 | 1272 | populate_ram_info(info, s); |
38c482b4 | 1273 | migration_populate_vfio_info(info); |
17549e84 | 1274 | break; |
31194731 | 1275 | case MIGRATION_STATUS_FAILED: |
791e7c82 | 1276 | info->has_status = true; |
17549e84 | 1277 | break; |
31194731 | 1278 | case MIGRATION_STATUS_CANCELLED: |
791e7c82 | 1279 | info->has_status = true; |
17549e84 | 1280 | break; |
c7e0acd5 JF |
1281 | case MIGRATION_STATUS_WAIT_UNPLUG: |
1282 | info->has_status = true; | |
1283 | break; | |
5bb7910a | 1284 | } |
552de79b | 1285 | info->status = state; |
c94143e5 PX |
1286 | |
1287 | QEMU_LOCK_GUARD(&s->error_mutex); | |
1288 | if (s->error) { | |
1289 | info->error_desc = g_strdup(error_get_pretty(s->error)); | |
1290 | } | |
5bb7910a AL |
1291 | } |
1292 | ||
65ace060 AP |
1293 | static void fill_destination_migration_info(MigrationInfo *info) |
1294 | { | |
1295 | MigrationIncomingState *mis = migration_incoming_get_current(); | |
1296 | ||
9aca82ba JQ |
1297 | if (mis->socket_address_list) { |
1298 | info->has_socket_address = true; | |
1299 | info->socket_address = | |
1300 | QAPI_CLONE(SocketAddressList, mis->socket_address_list); | |
1301 | } | |
1302 | ||
65ace060 AP |
1303 | switch (mis->state) { |
1304 | case MIGRATION_STATUS_NONE: | |
1305 | return; | |
65ace060 AP |
1306 | case MIGRATION_STATUS_SETUP: |
1307 | case MIGRATION_STATUS_CANCELLING: | |
1308 | case MIGRATION_STATUS_CANCELLED: | |
1309 | case MIGRATION_STATUS_ACTIVE: | |
1310 | case MIGRATION_STATUS_POSTCOPY_ACTIVE: | |
3c9928d9 PX |
1311 | case MIGRATION_STATUS_POSTCOPY_PAUSED: |
1312 | case MIGRATION_STATUS_POSTCOPY_RECOVER: | |
65ace060 AP |
1313 | case MIGRATION_STATUS_FAILED: |
1314 | case MIGRATION_STATUS_COLO: | |
1315 | info->has_status = true; | |
1316 | break; | |
1317 | case MIGRATION_STATUS_COMPLETED: | |
1318 | info->has_status = true; | |
1319 | fill_destination_postcopy_migration_info(info); | |
1320 | break; | |
1321 | } | |
1322 | info->status = mis->state; | |
dbea1c89 VSO |
1323 | |
1324 | if (!info->error_desc) { | |
1325 | MigrationState *s = migrate_get_current(); | |
1326 | QEMU_LOCK_GUARD(&s->error_mutex); | |
1327 | ||
1328 | if (s->error) { | |
1329 | info->error_desc = g_strdup(error_get_pretty(s->error)); | |
1330 | } | |
1331 | } | |
65ace060 AP |
1332 | } |
1333 | ||
1334 | MigrationInfo *qmp_query_migrate(Error **errp) | |
1335 | { | |
1336 | MigrationInfo *info = g_malloc0(sizeof(*info)); | |
1337 | ||
1338 | fill_destination_migration_info(info); | |
1339 | fill_source_migration_info(info); | |
1340 | ||
1341 | return info; | |
1342 | } | |
1343 | ||
4886a1bc DDAG |
1344 | void qmp_migrate_start_postcopy(Error **errp) |
1345 | { | |
1346 | MigrationState *s = migrate_get_current(); | |
1347 | ||
16b0fd32 | 1348 | if (!migrate_postcopy()) { |
a54d340b | 1349 | error_setg(errp, "Enable postcopy with migrate_set_capability before" |
4886a1bc DDAG |
1350 | " the start of migration"); |
1351 | return; | |
1352 | } | |
1353 | ||
1354 | if (s->state == MIGRATION_STATUS_NONE) { | |
1355 | error_setg(errp, "Postcopy must be started after migration has been" | |
1356 | " started"); | |
1357 | return; | |
1358 | } | |
1359 | /* | |
1360 | * we don't error if migration has finished since that would be racy | |
1361 | * with issuing this command. | |
1362 | */ | |
d73415a3 | 1363 | qatomic_set(&s->start_postcopy, true); |
4886a1bc DDAG |
1364 | } |
1365 | ||
065e2813 AL |
1366 | /* shared migration helpers */ |
1367 | ||
48781e5b | 1368 | void migrate_set_state(int *state, int old_state, int new_state) |
51cf4c1a | 1369 | { |
a31fedee | 1370 | assert(new_state < MIGRATION_STATUS__MAX); |
d73415a3 | 1371 | if (qatomic_cmpxchg(state, old_state, new_state) == old_state) { |
a31fedee | 1372 | trace_migrate_set_state(MigrationStatus_str(new_state)); |
b05dc723 | 1373 | migrate_generate_event(new_state); |
51cf4c1a Z |
1374 | } |
1375 | } | |
1376 | ||
fd392cfa | 1377 | static void migrate_fd_cleanup(MigrationState *s) |
065e2813 | 1378 | { |
9d9babf7 SS |
1379 | MigrationEventType type; |
1380 | ||
83174765 PX |
1381 | g_free(s->hostname); |
1382 | s->hostname = NULL; | |
e3bf5e68 DH |
1383 | json_writer_free(s->vmdesc); |
1384 | s->vmdesc = NULL; | |
83174765 | 1385 | |
0ceccd85 PX |
1386 | qemu_savevm_state_cleanup(); |
1387 | ||
22b04245 FR |
1388 | close_return_path_on_source(s); |
1389 | ||
89a02a9f | 1390 | if (s->to_dst_file) { |
62df066f | 1391 | QEMUFile *tmp; |
f986c3d2 | 1392 | |
9013dca5 | 1393 | trace_migrate_fd_cleanup(); |
195801d7 | 1394 | bql_unlock(); |
1d34e4bf DDAG |
1395 | if (s->migration_thread_running) { |
1396 | qemu_thread_join(&s->thread); | |
1397 | s->migration_thread_running = false; | |
1398 | } | |
195801d7 | 1399 | bql_lock(); |
404a7c05 | 1400 | |
cde85c37 | 1401 | multifd_send_shutdown(); |
62df066f PX |
1402 | qemu_mutex_lock(&s->qemu_file_lock); |
1403 | tmp = s->to_dst_file; | |
89a02a9f | 1404 | s->to_dst_file = NULL; |
62df066f PX |
1405 | qemu_mutex_unlock(&s->qemu_file_lock); |
1406 | /* | |
1407 | * Close the file handle without the lock to make sure the | |
1408 | * critical section won't block for long. | |
1409 | */ | |
39675fff | 1410 | migration_ioc_unregister_yank_from_file(tmp); |
62df066f | 1411 | qemu_fclose(tmp); |
065e2813 AL |
1412 | } |
1413 | ||
3a6813b6 | 1414 | assert(!migration_is_active()); |
7a2c1721 | 1415 | |
94f5a437 | 1416 | if (s->state == MIGRATION_STATUS_CANCELLING) { |
48781e5b | 1417 | migrate_set_state(&s->state, MIGRATION_STATUS_CANCELLING, |
94f5a437 | 1418 | MIGRATION_STATUS_CANCELLED); |
7a2c1721 | 1419 | } |
a3fa1d78 | 1420 | |
87db1a7d JQ |
1421 | if (s->error) { |
1422 | /* It is used on info migrate. We can't free it */ | |
1423 | error_report_err(error_copy(s->error)); | |
1424 | } | |
9d9babf7 SS |
1425 | type = migration_has_failed(s) ? MIG_EVENT_PRECOPY_FAILED : |
1426 | MIG_EVENT_PRECOPY_DONE; | |
4af667f8 | 1427 | migration_call_notifiers(s, type, NULL); |
b1a87956 | 1428 | block_cleanup_parameters(); |
b5eea99e | 1429 | yank_unregister_instance(MIGRATION_YANK_INSTANCE); |
065e2813 AL |
1430 | } |
1431 | ||
fd392cfa YK |
1432 | static void migrate_fd_cleanup_bh(void *opaque) |
1433 | { | |
44d0d456 | 1434 | migrate_fd_cleanup(opaque); |
fd392cfa YK |
1435 | } |
1436 | ||
87db1a7d JQ |
1437 | void migrate_set_error(MigrationState *s, const Error *error) |
1438 | { | |
6e8a355d | 1439 | QEMU_LOCK_GUARD(&s->error_mutex); |
d4a17b8f VSO |
1440 | |
1441 | trace_migrate_error(error_get_pretty(error)); | |
1442 | ||
87db1a7d JQ |
1443 | if (!s->error) { |
1444 | s->error = error_copy(error); | |
1445 | } | |
87db1a7d JQ |
1446 | } |
1447 | ||
2b2f6f41 PX |
1448 | bool migrate_has_error(MigrationState *s) |
1449 | { | |
1450 | /* The lock is not helpful here, but still follow the rule */ | |
1451 | QEMU_LOCK_GUARD(&s->error_mutex); | |
1452 | return qatomic_read(&s->error); | |
1453 | } | |
1454 | ||
ca7bd082 PX |
1455 | static void migrate_error_free(MigrationState *s) |
1456 | { | |
1457 | QEMU_LOCK_GUARD(&s->error_mutex); | |
1458 | if (s->error) { | |
1459 | error_free(s->error); | |
1460 | s->error = NULL; | |
1461 | } | |
1462 | } | |
1463 | ||
aaf26bd3 | 1464 | static void migrate_fd_error(MigrationState *s, const Error *error) |
065e2813 | 1465 | { |
89a02a9f | 1466 | assert(s->to_dst_file == NULL); |
48781e5b HZ |
1467 | migrate_set_state(&s->state, MIGRATION_STATUS_SETUP, |
1468 | MIGRATION_STATUS_FAILED); | |
87db1a7d | 1469 | migrate_set_error(s, error); |
458cf28e JQ |
1470 | } |
1471 | ||
0edda1c4 | 1472 | static void migrate_fd_cancel(MigrationState *s) |
065e2813 | 1473 | { |
6f2b811a | 1474 | int old_state ; |
7478fb0d | 1475 | |
9013dca5 | 1476 | trace_migrate_fd_cancel(); |
065e2813 | 1477 | |
43044ac0 PX |
1478 | WITH_QEMU_LOCK_GUARD(&s->qemu_file_lock) { |
1479 | if (s->rp_state.from_dst_file) { | |
1480 | /* shutdown the rp socket, so causing the rp thread to shutdown */ | |
1481 | qemu_file_shutdown(s->rp_state.from_dst_file); | |
1482 | } | |
70b20477 DDAG |
1483 | } |
1484 | ||
6f2b811a Z |
1485 | do { |
1486 | old_state = s->state; | |
aeaafb1e | 1487 | if (!migration_is_running()) { |
6f2b811a Z |
1488 | break; |
1489 | } | |
a7b36b48 DDAG |
1490 | /* If the migration is paused, kick it out of the pause */ |
1491 | if (old_state == MIGRATION_STATUS_PRE_SWITCHOVER) { | |
1492 | qemu_sem_post(&s->pause_sem); | |
1493 | } | |
48781e5b | 1494 | migrate_set_state(&s->state, old_state, MIGRATION_STATUS_CANCELLING); |
31194731 | 1495 | } while (s->state != MIGRATION_STATUS_CANCELLING); |
a26ba26e DDAG |
1496 | |
1497 | /* | |
1498 | * If we're unlucky the migration code might be stuck somewhere in a | |
1499 | * send/write while the network has failed and is waiting to timeout; | |
1500 | * if we've got shutdown(2) available then we can force it to quit. | |
a26ba26e | 1501 | */ |
7478fb0d FR |
1502 | if (s->state == MIGRATION_STATUS_CANCELLING) { |
1503 | WITH_QEMU_LOCK_GUARD(&s->qemu_file_lock) { | |
1504 | if (s->to_dst_file) { | |
1505 | qemu_file_shutdown(s->to_dst_file); | |
1506 | } | |
1507 | } | |
a26ba26e | 1508 | } |
1d2acc31 HZ |
1509 | if (s->state == MIGRATION_STATUS_CANCELLING && s->block_inactive) { |
1510 | Error *local_err = NULL; | |
1511 | ||
3b717194 | 1512 | bdrv_activate_all(&local_err); |
1d2acc31 HZ |
1513 | if (local_err) { |
1514 | error_report_err(local_err); | |
1515 | } else { | |
1516 | s->block_inactive = false; | |
1517 | } | |
1518 | } | |
065e2813 AL |
1519 | } |
1520 | ||
6835f5a1 SS |
1521 | void migration_add_notifier_mode(NotifierWithReturn *notify, |
1522 | MigrationNotifyFunc func, MigMode mode) | |
1523 | { | |
1524 | notify->notify = (NotifierWithReturnFunc)func; | |
1525 | notifier_with_return_list_add(&migration_state_notifiers[mode], notify); | |
1526 | } | |
1527 | ||
3e775730 | 1528 | void migration_add_notifier(NotifierWithReturn *notify, |
5663dd3f | 1529 | MigrationNotifyFunc func) |
99a0db9b | 1530 | { |
6835f5a1 | 1531 | migration_add_notifier_mode(notify, func, MIG_MODE_NORMAL); |
99a0db9b GH |
1532 | } |
1533 | ||
3e775730 | 1534 | void migration_remove_notifier(NotifierWithReturn *notify) |
d9cda213 SS |
1535 | { |
1536 | if (notify->notify) { | |
3e775730 | 1537 | notifier_with_return_remove(notify); |
d9cda213 SS |
1538 | notify->notify = NULL; |
1539 | } | |
1540 | } | |
1541 | ||
4af667f8 SS |
1542 | int migration_call_notifiers(MigrationState *s, MigrationEventType type, |
1543 | Error **errp) | |
99a0db9b | 1544 | { |
6835f5a1 | 1545 | MigMode mode = s->parameters.mode; |
9d9babf7 | 1546 | MigrationEvent e; |
4af667f8 | 1547 | int ret; |
9d9babf7 SS |
1548 | |
1549 | e.type = type; | |
4af667f8 SS |
1550 | ret = notifier_with_return_list_notify(&migration_state_notifiers[mode], |
1551 | &e, errp); | |
1552 | assert(!ret || type == MIG_EVENT_PRECOPY_SETUP); | |
1553 | return ret; | |
99a0db9b GH |
1554 | } |
1555 | ||
afe2df69 GH |
1556 | bool migration_has_failed(MigrationState *s) |
1557 | { | |
31194731 HZ |
1558 | return (s->state == MIGRATION_STATUS_CANCELLED || |
1559 | s->state == MIGRATION_STATUS_FAILED); | |
afe2df69 GH |
1560 | } |
1561 | ||
5727309d | 1562 | bool migration_in_postcopy(void) |
9ec055ae | 1563 | { |
5727309d JQ |
1564 | MigrationState *s = migrate_get_current(); |
1565 | ||
3748fef9 DDAG |
1566 | switch (s->state) { |
1567 | case MIGRATION_STATUS_POSTCOPY_ACTIVE: | |
1568 | case MIGRATION_STATUS_POSTCOPY_PAUSED: | |
1569 | case MIGRATION_STATUS_POSTCOPY_RECOVER: | |
1570 | return true; | |
1571 | default: | |
1572 | return false; | |
1573 | } | |
9ec055ae DDAG |
1574 | } |
1575 | ||
f8c543e8 PX |
1576 | bool migration_postcopy_is_alive(int state) |
1577 | { | |
1578 | switch (state) { | |
1579 | case MIGRATION_STATUS_POSTCOPY_ACTIVE: | |
1580 | case MIGRATION_STATUS_POSTCOPY_RECOVER: | |
1581 | return true; | |
1582 | default: | |
1583 | return false; | |
1584 | } | |
1585 | } | |
1586 | ||
06df2e69 DH |
1587 | bool migration_in_incoming_postcopy(void) |
1588 | { | |
1589 | PostcopyState ps = postcopy_state_get(); | |
1590 | ||
1591 | return ps >= POSTCOPY_INCOMING_DISCARD && ps < POSTCOPY_INCOMING_END; | |
1592 | } | |
1593 | ||
80fe315c DH |
1594 | bool migration_incoming_postcopy_advised(void) |
1595 | { | |
1596 | PostcopyState ps = postcopy_state_get(); | |
1597 | ||
1598 | return ps >= POSTCOPY_INCOMING_ADVISE && ps < POSTCOPY_INCOMING_END; | |
1599 | } | |
1600 | ||
1a8e44a8 AG |
1601 | bool migration_in_bg_snapshot(void) |
1602 | { | |
1a8e44a8 | 1603 | return migrate_background_snapshot() && |
7dcb3c87 | 1604 | migration_is_setup_or_active(); |
1a8e44a8 AG |
1605 | } |
1606 | ||
fab35005 | 1607 | bool migration_is_idle(void) |
fe44dc91 | 1608 | { |
daff7f0b MA |
1609 | MigrationState *s = current_migration; |
1610 | ||
1611 | if (!s) { | |
1612 | return true; | |
1613 | } | |
fe44dc91 AA |
1614 | |
1615 | switch (s->state) { | |
1616 | case MIGRATION_STATUS_NONE: | |
1617 | case MIGRATION_STATUS_CANCELLED: | |
1618 | case MIGRATION_STATUS_COMPLETED: | |
1619 | case MIGRATION_STATUS_FAILED: | |
1620 | return true; | |
1621 | case MIGRATION_STATUS_SETUP: | |
1622 | case MIGRATION_STATUS_CANCELLING: | |
1623 | case MIGRATION_STATUS_ACTIVE: | |
1624 | case MIGRATION_STATUS_POSTCOPY_ACTIVE: | |
1625 | case MIGRATION_STATUS_COLO: | |
31e06077 DDAG |
1626 | case MIGRATION_STATUS_PRE_SWITCHOVER: |
1627 | case MIGRATION_STATUS_DEVICE: | |
c7e0acd5 | 1628 | case MIGRATION_STATUS_WAIT_UNPLUG: |
fe44dc91 AA |
1629 | return false; |
1630 | case MIGRATION_STATUS__MAX: | |
1631 | g_assert_not_reached(); | |
1632 | } | |
1633 | ||
1634 | return false; | |
1635 | } | |
1636 | ||
3a6813b6 | 1637 | bool migration_is_active(void) |
8f8d528e | 1638 | { |
3a6813b6 SS |
1639 | MigrationState *s = current_migration; |
1640 | ||
8f8d528e WY |
1641 | return (s->state == MIGRATION_STATUS_ACTIVE || |
1642 | s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE); | |
1643 | } | |
1644 | ||
9bb630c6 SS |
1645 | bool migration_is_device(void) |
1646 | { | |
1647 | MigrationState *s = current_migration; | |
1648 | ||
1649 | return s->state == MIGRATION_STATUS_DEVICE; | |
1650 | } | |
1651 | ||
6e785639 SS |
1652 | bool migration_thread_is_self(void) |
1653 | { | |
1654 | MigrationState *s = current_migration; | |
1655 | ||
1656 | return qemu_thread_is_self(&s->thread); | |
1657 | } | |
1658 | ||
9867d4dd SS |
1659 | bool migrate_mode_is_cpr(MigrationState *s) |
1660 | { | |
1661 | return s->parameters.mode == MIG_MODE_CPR_REBOOT; | |
1662 | } | |
1663 | ||
08fc4cb5 | 1664 | int migrate_init(MigrationState *s, Error **errp) |
0edda1c4 | 1665 | { |
08fc4cb5 AH |
1666 | int ret; |
1667 | ||
1668 | ret = qemu_savevm_state_prepare(errp); | |
1669 | if (ret) { | |
1670 | return ret; | |
1671 | } | |
1672 | ||
389775d1 DDAG |
1673 | /* |
1674 | * Reinitialise all migration state, except | |
1675 | * parameters/capabilities that the user set, and | |
1676 | * locks. | |
1677 | */ | |
89a02a9f | 1678 | s->to_dst_file = NULL; |
389775d1 | 1679 | s->state = MIGRATION_STATUS_NONE; |
389775d1 | 1680 | s->rp_state.from_dst_file = NULL; |
389775d1 | 1681 | s->mbps = 0.0; |
aecbfe9c | 1682 | s->pages_per_second = 0.0; |
389775d1 DDAG |
1683 | s->downtime = 0; |
1684 | s->expected_downtime = 0; | |
389775d1 | 1685 | s->setup_time = 0; |
389775d1 DDAG |
1686 | s->start_postcopy = false; |
1687 | s->migration_thread_running = false; | |
d59ce6f3 DB |
1688 | error_free(s->error); |
1689 | s->error = NULL; | |
2aae1eb8 | 1690 | s->vmdesc = NULL; |
389775d1 | 1691 | |
48781e5b | 1692 | migrate_set_state(&s->state, MIGRATION_STATUS_NONE, MIGRATION_STATUS_SETUP); |
0edda1c4 | 1693 | |
4af246a3 PX |
1694 | s->start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); |
1695 | s->total_time = 0; | |
f4584076 | 1696 | s->vm_old_state = -1; |
b15df1ae PX |
1697 | s->iteration_initial_bytes = 0; |
1698 | s->threshold_size = 0; | |
1b4adb10 | 1699 | s->switchover_acked = false; |
27fd25b0 | 1700 | s->rdma_migration = false; |
f543aa22 | 1701 | /* |
809f188a | 1702 | * set mig_stats memory to zero for a new migration |
f543aa22 AH |
1703 | */ |
1704 | memset(&mig_stats, 0, sizeof(mig_stats)); | |
f543aa22 | 1705 | migration_reset_vfio_bytes_transferred(); |
08fc4cb5 AH |
1706 | |
1707 | return 0; | |
0edda1c4 | 1708 | } |
cab30143 | 1709 | |
fa3673e4 | 1710 | static bool is_busy(Error **reasonp, Error **errp) |
fa2756b7 | 1711 | { |
fa3673e4 SS |
1712 | ERRP_GUARD(); |
1713 | ||
4c170330 PX |
1714 | /* Snapshots are similar to migrations, so check RUN_STATE_SAVE_VM too. */ |
1715 | if (runstate_check(RUN_STATE_SAVE_VM) || !migration_is_idle()) { | |
c8a7fc51 | 1716 | error_propagate_prepend(errp, *reasonp, |
4c170330 PX |
1717 | "disallowing migration blocker " |
1718 | "(migration/snapshot in progress) for: "); | |
c8a7fc51 | 1719 | *reasonp = NULL; |
fa3673e4 | 1720 | return true; |
fe44dc91 | 1721 | } |
fa3673e4 | 1722 | return false; |
fa2756b7 AL |
1723 | } |
1724 | ||
fa3673e4 | 1725 | static bool is_only_migratable(Error **reasonp, Error **errp, int modes) |
60fd6801 | 1726 | { |
fa3673e4 SS |
1727 | ERRP_GUARD(); |
1728 | ||
1729 | if (only_migratable && (modes & BIT(MIG_MODE_NORMAL))) { | |
c8a7fc51 | 1730 | error_propagate_prepend(errp, *reasonp, |
60fd6801 PX |
1731 | "disallowing migration blocker " |
1732 | "(--only-migratable) for: "); | |
c8a7fc51 | 1733 | *reasonp = NULL; |
fa3673e4 SS |
1734 | return true; |
1735 | } | |
1736 | return false; | |
1737 | } | |
1738 | ||
1739 | static int get_modes(MigMode mode, va_list ap) | |
1740 | { | |
1741 | int modes = 0; | |
1742 | ||
1743 | while (mode != -1 && mode != MIG_MODE_ALL) { | |
1744 | assert(mode >= MIG_MODE_NORMAL && mode < MIG_MODE__MAX); | |
1745 | modes |= BIT(mode); | |
1746 | mode = va_arg(ap, MigMode); | |
1747 | } | |
1748 | if (mode == MIG_MODE_ALL) { | |
1749 | modes = BIT(MIG_MODE__MAX) - 1; | |
1750 | } | |
1751 | return modes; | |
1752 | } | |
1753 | ||
1754 | static int add_blockers(Error **reasonp, Error **errp, int modes) | |
1755 | { | |
1756 | for (MigMode mode = 0; mode < MIG_MODE__MAX; mode++) { | |
1757 | if (modes & BIT(mode)) { | |
1758 | migration_blockers[mode] = g_slist_prepend(migration_blockers[mode], | |
1759 | *reasonp); | |
1760 | } | |
1761 | } | |
1762 | return 0; | |
1763 | } | |
1764 | ||
1765 | int migrate_add_blocker(Error **reasonp, Error **errp) | |
1766 | { | |
1767 | return migrate_add_blocker_modes(reasonp, errp, MIG_MODE_ALL); | |
1768 | } | |
1769 | ||
1770 | int migrate_add_blocker_normal(Error **reasonp, Error **errp) | |
1771 | { | |
1772 | return migrate_add_blocker_modes(reasonp, errp, MIG_MODE_NORMAL, -1); | |
1773 | } | |
1774 | ||
1775 | int migrate_add_blocker_modes(Error **reasonp, Error **errp, MigMode mode, ...) | |
1776 | { | |
1777 | int modes; | |
1778 | va_list ap; | |
1779 | ||
1780 | va_start(ap, mode); | |
1781 | modes = get_modes(mode, ap); | |
1782 | va_end(ap); | |
1783 | ||
1784 | if (is_only_migratable(reasonp, errp, modes)) { | |
60fd6801 | 1785 | return -EACCES; |
fa3673e4 SS |
1786 | } else if (is_busy(reasonp, errp)) { |
1787 | return -EBUSY; | |
60fd6801 | 1788 | } |
fa3673e4 SS |
1789 | return add_blockers(reasonp, errp, modes); |
1790 | } | |
60fd6801 | 1791 | |
fa3673e4 SS |
1792 | int migrate_add_blocker_internal(Error **reasonp, Error **errp) |
1793 | { | |
1794 | int modes = BIT(MIG_MODE__MAX) - 1; | |
1795 | ||
1796 | if (is_busy(reasonp, errp)) { | |
1797 | return -EBUSY; | |
1798 | } | |
1799 | return add_blockers(reasonp, errp, modes); | |
60fd6801 PX |
1800 | } |
1801 | ||
c8a7fc51 | 1802 | void migrate_del_blocker(Error **reasonp) |
fa2756b7 | 1803 | { |
c8a7fc51 | 1804 | if (*reasonp) { |
fa3673e4 SS |
1805 | for (MigMode mode = 0; mode < MIG_MODE__MAX; mode++) { |
1806 | migration_blockers[mode] = g_slist_remove(migration_blockers[mode], | |
1807 | *reasonp); | |
1808 | } | |
c8a7fc51 SS |
1809 | error_free(*reasonp); |
1810 | *reasonp = NULL; | |
1811 | } | |
fa2756b7 AL |
1812 | } |
1813 | ||
074dbce5 | 1814 | void qmp_migrate_incoming(const char *uri, bool has_channels, |
dbea1c89 VSO |
1815 | MigrationChannelList *channels, |
1816 | bool has_exit_on_error, bool exit_on_error, | |
1817 | Error **errp) | |
bf1ae1f4 DDAG |
1818 | { |
1819 | Error *local_err = NULL; | |
4debb5f5 | 1820 | static bool once = true; |
dbea1c89 | 1821 | MigrationIncomingState *mis = migration_incoming_get_current(); |
bf1ae1f4 | 1822 | |
4debb5f5 DDAG |
1823 | if (!once) { |
1824 | error_setg(errp, "The incoming migration has already been started"); | |
603d5a42 | 1825 | return; |
4debb5f5 | 1826 | } |
e69d50d6 PB |
1827 | if (!runstate_check(RUN_STATE_INMIGRATE)) { |
1828 | error_setg(errp, "'-incoming' was not specified on the command line"); | |
1829 | return; | |
1830 | } | |
bf1ae1f4 | 1831 | |
cc48c587 PX |
1832 | if (!yank_register_instance(MIGRATION_YANK_INSTANCE, errp)) { |
1833 | return; | |
1834 | } | |
1835 | ||
dbea1c89 VSO |
1836 | mis->exit_on_error = |
1837 | has_exit_on_error ? exit_on_error : INMIGRATE_DEFAULT_EXIT_ON_ERROR; | |
1838 | ||
074dbce5 | 1839 | qemu_start_incoming_migration(uri, has_channels, channels, &local_err); |
bf1ae1f4 DDAG |
1840 | |
1841 | if (local_err) { | |
cc48c587 | 1842 | yank_unregister_instance(MIGRATION_YANK_INSTANCE); |
bf1ae1f4 DDAG |
1843 | error_propagate(errp, local_err); |
1844 | return; | |
1845 | } | |
1846 | ||
4debb5f5 | 1847 | once = false; |
bf1ae1f4 DDAG |
1848 | } |
1849 | ||
02affd41 PX |
1850 | void qmp_migrate_recover(const char *uri, Error **errp) |
1851 | { | |
1852 | MigrationIncomingState *mis = migration_incoming_get_current(); | |
1853 | ||
b7f9afd4 PX |
1854 | /* |
1855 | * Don't even bother to use ERRP_GUARD() as it _must_ always be set by | |
1856 | * callers (no one should ignore a recover failure); if there is, it's a | |
1857 | * programming error. | |
1858 | */ | |
1859 | assert(errp); | |
1860 | ||
02affd41 PX |
1861 | if (mis->state != MIGRATION_STATUS_POSTCOPY_PAUSED) { |
1862 | error_setg(errp, "Migrate recover can only be run " | |
1863 | "when postcopy is paused."); | |
1864 | return; | |
1865 | } | |
1866 | ||
08401c04 PX |
1867 | /* If there's an existing transport, release it */ |
1868 | migration_incoming_transport_cleanup(mis); | |
02affd41 PX |
1869 | |
1870 | /* | |
1871 | * Note that this call will never start a real migration; it will | |
1872 | * only re-setup the migration stream and poke existing migration | |
1873 | * to continue using that newly established channel. | |
1874 | */ | |
074dbce5 | 1875 | qemu_start_incoming_migration(uri, false, NULL, errp); |
02affd41 PX |
1876 | } |
1877 | ||
bfbf89c2 PX |
1878 | void qmp_migrate_pause(Error **errp) |
1879 | { | |
1880 | MigrationState *ms = migrate_get_current(); | |
1881 | MigrationIncomingState *mis = migration_incoming_get_current(); | |
7478fb0d | 1882 | int ret = 0; |
bfbf89c2 | 1883 | |
f8c543e8 | 1884 | if (migration_postcopy_is_alive(ms->state)) { |
bfbf89c2 | 1885 | /* Source side, during postcopy */ |
f8c543e8 PX |
1886 | Error *error = NULL; |
1887 | ||
1888 | /* Tell the core migration that we're pausing */ | |
1889 | error_setg(&error, "Postcopy migration is paused by the user"); | |
1890 | migrate_set_error(ms, error); | |
1891 | error_free(error); | |
1892 | ||
bfbf89c2 | 1893 | qemu_mutex_lock(&ms->qemu_file_lock); |
7478fb0d FR |
1894 | if (ms->to_dst_file) { |
1895 | ret = qemu_file_shutdown(ms->to_dst_file); | |
1896 | } | |
bfbf89c2 PX |
1897 | qemu_mutex_unlock(&ms->qemu_file_lock); |
1898 | if (ret) { | |
1899 | error_setg(errp, "Failed to pause source migration"); | |
1900 | } | |
f8c543e8 PX |
1901 | |
1902 | /* | |
1903 | * Kick the migration thread out of any waiting windows (on behalf | |
1904 | * of the rp thread). | |
1905 | */ | |
1906 | migration_rp_kick(ms); | |
1907 | ||
bfbf89c2 PX |
1908 | return; |
1909 | } | |
1910 | ||
f8c543e8 | 1911 | if (migration_postcopy_is_alive(mis->state)) { |
bfbf89c2 PX |
1912 | ret = qemu_file_shutdown(mis->from_src_file); |
1913 | if (ret) { | |
1914 | error_setg(errp, "Failed to pause destination migration"); | |
1915 | } | |
1916 | return; | |
1917 | } | |
1918 | ||
1919 | error_setg(errp, "migrate-pause is currently only supported " | |
f8c543e8 | 1920 | "during postcopy-active or postcopy-recover state"); |
bfbf89c2 PX |
1921 | } |
1922 | ||
24f3902b GK |
1923 | bool migration_is_blocked(Error **errp) |
1924 | { | |
fa3673e4 SS |
1925 | GSList *blockers = migration_blockers[migrate_mode()]; |
1926 | ||
24f3902b GK |
1927 | if (qemu_savevm_state_blocked(errp)) { |
1928 | return true; | |
1929 | } | |
1930 | ||
fa3673e4 SS |
1931 | if (blockers) { |
1932 | error_propagate(errp, error_copy(blockers->data)); | |
24f3902b GK |
1933 | return true; |
1934 | } | |
1935 | ||
1936 | return false; | |
1937 | } | |
1938 | ||
d3e35b8f PX |
1939 | /* Returns true if continue to migrate, or false if error detected */ |
1940 | static bool migrate_prepare(MigrationState *s, bool blk, bool blk_inc, | |
1941 | bool resume, Error **errp) | |
cab30143 | 1942 | { |
40101f32 JQ |
1943 | if (blk_inc) { |
1944 | warn_report("parameter 'inc' is deprecated;" | |
1945 | " use blockdev-mirror with NBD instead"); | |
1946 | } | |
1947 | ||
8846b5bf JQ |
1948 | if (blk) { |
1949 | warn_report("parameter 'blk' is deprecated;" | |
1950 | " use blockdev-mirror with NBD instead"); | |
1951 | } | |
1952 | ||
d3e35b8f PX |
1953 | if (resume) { |
1954 | if (s->state != MIGRATION_STATUS_POSTCOPY_PAUSED) { | |
1955 | error_setg(errp, "Cannot resume if there is no " | |
1956 | "paused migration"); | |
1957 | return false; | |
1958 | } | |
97ca211c PX |
1959 | |
1960 | /* | |
1961 | * Postcopy recovery won't work well with release-ram | |
1962 | * capability since release-ram will drop the page buffer as | |
1963 | * long as the page is put into the send buffer. So if there | |
1964 | * is a network failure happened, any page buffers that have | |
1965 | * not yet reached the destination VM but have already been | |
1966 | * sent from the source VM will be lost forever. Let's refuse | |
1967 | * the client from resuming such a postcopy migration. | |
1968 | * Luckily release-ram was designed to only be used when src | |
1969 | * and destination VMs are on the same host, so it should be | |
1970 | * fine. | |
1971 | */ | |
1972 | if (migrate_release_ram()) { | |
1973 | error_setg(errp, "Postcopy recovery cannot work " | |
1974 | "when release-ram capability is set"); | |
1975 | return false; | |
1976 | } | |
1977 | ||
d3e35b8f PX |
1978 | /* This is a resume, skip init status */ |
1979 | return true; | |
1980 | } | |
cab30143 | 1981 | |
aeaafb1e | 1982 | if (migration_is_running()) { |
00580786 | 1983 | error_setg(errp, "There's a migration process in progress"); |
d3e35b8f | 1984 | return false; |
cab30143 | 1985 | } |
d3e35b8f | 1986 | |
ca99993a DDAG |
1987 | if (runstate_check(RUN_STATE_INMIGRATE)) { |
1988 | error_setg(errp, "Guest is waiting for an incoming migration"); | |
d3e35b8f | 1989 | return false; |
ca99993a DDAG |
1990 | } |
1991 | ||
36d0fe65 T |
1992 | if (runstate_check(RUN_STATE_POSTMIGRATE)) { |
1993 | error_setg(errp, "Can't migrate the vm that was paused due to " | |
1994 | "previous migration"); | |
1995 | return false; | |
1996 | } | |
1997 | ||
06152b89 WR |
1998 | if (kvm_hwpoisoned_mem()) { |
1999 | error_setg(errp, "Can't migrate this vm with hardware poisoned memory, " | |
2000 | "please reboot the vm and try again"); | |
2001 | return false; | |
2002 | } | |
2003 | ||
24f3902b | 2004 | if (migration_is_blocked(errp)) { |
d3e35b8f | 2005 | return false; |
fa2756b7 AL |
2006 | } |
2007 | ||
4ed49feb FR |
2008 | if (migrate_mapped_ram()) { |
2009 | if (migrate_tls()) { | |
2010 | error_setg(errp, "Cannot use TLS with mapped-ram"); | |
2011 | return false; | |
2012 | } | |
f427d90b FR |
2013 | |
2014 | if (migrate_multifd_compression()) { | |
2015 | error_setg(errp, "Cannot use compression with mapped-ram"); | |
2016 | return false; | |
2017 | } | |
4ed49feb FR |
2018 | } |
2019 | ||
cbdafc1b SS |
2020 | if (migrate_mode_is_cpr(s)) { |
2021 | const char *conflict = NULL; | |
2022 | ||
2023 | if (migrate_postcopy()) { | |
2024 | conflict = "postcopy"; | |
2025 | } else if (migrate_background_snapshot()) { | |
2026 | conflict = "background snapshot"; | |
2027 | } else if (migrate_colo()) { | |
2028 | conflict = "COLO"; | |
2029 | } | |
2030 | ||
2031 | if (conflict) { | |
2032 | error_setg(errp, "Cannot use %s with CPR", conflict); | |
2033 | return false; | |
2034 | } | |
2035 | } | |
2036 | ||
d3e35b8f | 2037 | if (blk || blk_inc) { |
5e804644 | 2038 | if (migrate_colo()) { |
3ba02445 RL |
2039 | error_setg(errp, "No disk migration is required in COLO mode"); |
2040 | return false; | |
2041 | } | |
6f8be708 | 2042 | if (migrate_block() || migrate_block_incremental()) { |
2833c59b JQ |
2043 | error_setg(errp, "Command options are incompatible with " |
2044 | "current migration capabilities"); | |
d3e35b8f | 2045 | return false; |
2833c59b | 2046 | } |
b6f4c0c7 | 2047 | if (!migrate_cap_set(MIGRATION_CAPABILITY_BLOCK, true, errp)) { |
d3e35b8f | 2048 | return false; |
2833c59b JQ |
2049 | } |
2050 | s->must_remove_block_options = true; | |
2051 | } | |
2052 | ||
d3e35b8f | 2053 | if (blk_inc) { |
87c22901 | 2054 | migrate_set_block_incremental(true); |
2833c59b JQ |
2055 | } |
2056 | ||
08fc4cb5 AH |
2057 | if (migrate_init(s, errp)) { |
2058 | return false; | |
2059 | } | |
cab30143 | 2060 | |
d3e35b8f PX |
2061 | return true; |
2062 | } | |
2063 | ||
074dbce5 HG |
2064 | void qmp_migrate(const char *uri, bool has_channels, |
2065 | MigrationChannelList *channels, bool has_blk, bool blk, | |
d3e35b8f PX |
2066 | bool has_inc, bool inc, bool has_detach, bool detach, |
2067 | bool has_resume, bool resume, Error **errp) | |
2068 | { | |
8c69ae9e | 2069 | bool resume_requested; |
d3e35b8f PX |
2070 | Error *local_err = NULL; |
2071 | MigrationState *s = migrate_get_current(); | |
bc1d54ee | 2072 | g_autoptr(MigrationChannel) channel = NULL; |
5994024f | 2073 | MigrationAddress *addr = NULL; |
d3e35b8f | 2074 | |
074dbce5 HG |
2075 | /* |
2076 | * Having preliminary checks for uri and channel | |
2077 | */ | |
0770ad43 HG |
2078 | if (!uri == !channels) { |
2079 | error_setg(errp, "need either 'uri' or 'channels' argument"); | |
074dbce5 | 2080 | return; |
0770ad43 HG |
2081 | } |
2082 | ||
2083 | if (channels) { | |
5994024f HG |
2084 | /* To verify that Migrate channel list has only item */ |
2085 | if (channels->next) { | |
2086 | error_setg(errp, "Channel list has more than one entries"); | |
2087 | return; | |
2088 | } | |
bc1d54ee | 2089 | addr = channels->value->addr; |
0770ad43 HG |
2090 | } |
2091 | ||
2092 | if (uri) { | |
5994024f HG |
2093 | /* caller uses the old URI syntax */ |
2094 | if (!migrate_uri_parse(uri, &channel, errp)) { | |
2095 | return; | |
2096 | } | |
bc1d54ee | 2097 | addr = channel->addr; |
074dbce5 | 2098 | } |
d6f74fd1 | 2099 | |
d95533e1 | 2100 | /* transport mechanism not suitable for migration? */ |
5994024f | 2101 | if (!migration_channels_and_transport_compatible(addr, errp)) { |
72a8192e HG |
2102 | return; |
2103 | } | |
2104 | ||
8c69ae9e | 2105 | resume_requested = has_resume && resume; |
d3e35b8f | 2106 | if (!migrate_prepare(s, has_blk && blk, has_inc && inc, |
8c69ae9e | 2107 | resume_requested, errp)) { |
d3e35b8f PX |
2108 | /* Error detected, put into errp */ |
2109 | return; | |
2110 | } | |
2111 | ||
8c69ae9e | 2112 | if (!resume_requested) { |
b5eea99e LS |
2113 | if (!yank_register_instance(MIGRATION_YANK_INSTANCE, errp)) { |
2114 | return; | |
2115 | } | |
2116 | } | |
2117 | ||
5994024f HG |
2118 | if (addr->transport == MIGRATION_ADDRESS_TYPE_SOCKET) { |
2119 | SocketAddress *saddr = &addr->u.socket; | |
34dfc5e4 HG |
2120 | if (saddr->type == SOCKET_ADDRESS_TYPE_INET || |
2121 | saddr->type == SOCKET_ADDRESS_TYPE_UNIX || | |
2122 | saddr->type == SOCKET_ADDRESS_TYPE_VSOCK) { | |
2123 | socket_start_outgoing_migration(s, saddr, &local_err); | |
2124 | } else if (saddr->type == SOCKET_ADDRESS_TYPE_FD) { | |
2125 | fd_start_outgoing_migration(s, saddr->u.fd.str, &local_err); | |
2126 | } | |
2da776db | 2127 | #ifdef CONFIG_RDMA |
5994024f HG |
2128 | } else if (addr->transport == MIGRATION_ADDRESS_TYPE_RDMA) { |
2129 | rdma_start_outgoing_migration(s, &addr->u.rdma, &local_err); | |
2da776db | 2130 | #endif |
5994024f HG |
2131 | } else if (addr->transport == MIGRATION_ADDRESS_TYPE_EXEC) { |
2132 | exec_start_outgoing_migration(s, addr->u.exec.args, &local_err); | |
2133 | } else if (addr->transport == MIGRATION_ADDRESS_TYPE_FILE) { | |
2134 | file_start_outgoing_migration(s, &addr->u.file, &local_err); | |
99a0db9b | 2135 | } else { |
908927db | 2136 | error_setg(&local_err, QERR_INVALID_PARAMETER_VALUE, "uri", |
c6bd8c70 | 2137 | "a valid migration protocol"); |
48781e5b HZ |
2138 | migrate_set_state(&s->state, MIGRATION_STATUS_SETUP, |
2139 | MIGRATION_STATUS_FAILED); | |
b1a87956 | 2140 | block_cleanup_parameters(); |
cab30143 JQ |
2141 | } |
2142 | ||
f37afb5a | 2143 | if (local_err) { |
8c69ae9e | 2144 | if (!resume_requested) { |
b5eea99e LS |
2145 | yank_unregister_instance(MIGRATION_YANK_INSTANCE); |
2146 | } | |
d59ce6f3 | 2147 | migrate_fd_error(s, local_err); |
f37afb5a | 2148 | error_propagate(errp, local_err); |
e1c37d0e | 2149 | return; |
1299c631 | 2150 | } |
cab30143 JQ |
2151 | } |
2152 | ||
6cdedb07 | 2153 | void qmp_migrate_cancel(Error **errp) |
cab30143 | 2154 | { |
458fecca | 2155 | migration_cancel(NULL); |
cab30143 JQ |
2156 | } |
2157 | ||
89cfc02c DDAG |
2158 | void qmp_migrate_continue(MigrationStatus state, Error **errp) |
2159 | { | |
2160 | MigrationState *s = migrate_get_current(); | |
2161 | if (s->state != state) { | |
2162 | error_setg(errp, "Migration not in expected state: %s", | |
2163 | MigrationStatus_str(s->state)); | |
2164 | return; | |
2165 | } | |
2166 | qemu_sem_post(&s->pause_sem); | |
2167 | } | |
2168 | ||
f8c543e8 | 2169 | int migration_rp_wait(MigrationState *s) |
5e79a4bf | 2170 | { |
f8c543e8 PX |
2171 | /* If migration has failure already, ignore the wait */ |
2172 | if (migrate_has_error(s)) { | |
2173 | return -1; | |
2174 | } | |
2175 | ||
5e79a4bf | 2176 | qemu_sem_wait(&s->rp_state.rp_sem); |
f8c543e8 PX |
2177 | |
2178 | /* After wait, double check that there's no failure */ | |
2179 | if (migrate_has_error(s)) { | |
2180 | return -1; | |
2181 | } | |
2182 | ||
2183 | return 0; | |
5e79a4bf PX |
2184 | } |
2185 | ||
2186 | void migration_rp_kick(MigrationState *s) | |
2187 | { | |
2188 | qemu_sem_post(&s->rp_state.rp_sem); | |
2189 | } | |
2190 | ||
70b20477 DDAG |
2191 | static struct rp_cmd_args { |
2192 | ssize_t len; /* -1 = variable */ | |
2193 | const char *name; | |
2194 | } rp_cmd_args[] = { | |
2195 | [MIG_RP_MSG_INVALID] = { .len = -1, .name = "INVALID" }, | |
2196 | [MIG_RP_MSG_SHUT] = { .len = 4, .name = "SHUT" }, | |
2197 | [MIG_RP_MSG_PONG] = { .len = 4, .name = "PONG" }, | |
1e2d90eb DDAG |
2198 | [MIG_RP_MSG_REQ_PAGES] = { .len = 12, .name = "REQ_PAGES" }, |
2199 | [MIG_RP_MSG_REQ_PAGES_ID] = { .len = -1, .name = "REQ_PAGES_ID" }, | |
a335debb | 2200 | [MIG_RP_MSG_RECV_BITMAP] = { .len = -1, .name = "RECV_BITMAP" }, |
13955b89 | 2201 | [MIG_RP_MSG_RESUME_ACK] = { .len = 4, .name = "RESUME_ACK" }, |
1b4adb10 | 2202 | [MIG_RP_MSG_SWITCHOVER_ACK] = { .len = 0, .name = "SWITCHOVER_ACK" }, |
70b20477 DDAG |
2203 | [MIG_RP_MSG_MAX] = { .len = -1, .name = "MAX" }, |
2204 | }; | |
2205 | ||
1e2d90eb DDAG |
2206 | /* |
2207 | * Process a request for pages received on the return path, | |
2208 | * We're allowed to send more than requested (e.g. to round to our page size) | |
2209 | * and we don't need to send pages that have already been sent. | |
2210 | */ | |
7aa6070d PX |
2211 | static void |
2212 | migrate_handle_rp_req_pages(MigrationState *ms, const char* rbname, | |
2213 | ram_addr_t start, size_t len, Error **errp) | |
1e2d90eb | 2214 | { |
8e3b0cbb | 2215 | long our_host_ps = qemu_real_host_page_size(); |
6c595cde | 2216 | |
1e2d90eb | 2217 | trace_migrate_handle_rp_req_pages(rbname, start, len); |
6c595cde DDAG |
2218 | |
2219 | /* | |
2220 | * Since we currently insist on matching page sizes, just sanity check | |
2221 | * we're being asked for whole host pages. | |
2222 | */ | |
7648297d DH |
2223 | if (!QEMU_IS_ALIGNED(start, our_host_ps) || |
2224 | !QEMU_IS_ALIGNED(len, our_host_ps)) { | |
7aa6070d PX |
2225 | error_setg(errp, "MIG_RP_MSG_REQ_PAGES: Misaligned page request, start:" |
2226 | RAM_ADDR_FMT " len: %zd", start, len); | |
6c595cde DDAG |
2227 | return; |
2228 | } | |
2229 | ||
7aa6070d | 2230 | ram_save_queue_pages(rbname, start, len, errp); |
1e2d90eb DDAG |
2231 | } |
2232 | ||
88577f32 PX |
2233 | static bool migrate_handle_rp_recv_bitmap(MigrationState *s, char *block_name, |
2234 | Error **errp) | |
a335debb PX |
2235 | { |
2236 | RAMBlock *block = qemu_ram_block_by_name(block_name); | |
2237 | ||
2238 | if (!block) { | |
7aa6070d PX |
2239 | error_setg(errp, "MIG_RP_MSG_RECV_BITMAP has invalid block name '%s'", |
2240 | block_name); | |
88577f32 | 2241 | return false; |
a335debb PX |
2242 | } |
2243 | ||
2244 | /* Fetch the received bitmap and refresh the dirty bitmap */ | |
7aa6070d | 2245 | return ram_dirty_bitmap_reload(s, block, errp); |
a335debb PX |
2246 | } |
2247 | ||
88577f32 PX |
2248 | static bool migrate_handle_rp_resume_ack(MigrationState *s, |
2249 | uint32_t value, Error **errp) | |
13955b89 PX |
2250 | { |
2251 | trace_source_return_path_thread_resume_ack(value); | |
2252 | ||
2253 | if (value != MIGRATION_RESUME_ACK_VALUE) { | |
7aa6070d | 2254 | error_setg(errp, "illegal resume_ack value %"PRIu32, value); |
88577f32 | 2255 | return false; |
13955b89 PX |
2256 | } |
2257 | ||
2258 | /* Now both sides are active. */ | |
2259 | migrate_set_state(&s->state, MIGRATION_STATUS_POSTCOPY_RECOVER, | |
2260 | MIGRATION_STATUS_POSTCOPY_ACTIVE); | |
2261 | ||
94190696 | 2262 | /* Notify send thread that time to continue send pages */ |
5e79a4bf | 2263 | migration_rp_kick(s); |
13955b89 | 2264 | |
88577f32 | 2265 | return true; |
13955b89 PX |
2266 | } |
2267 | ||
93589827 PX |
2268 | /* |
2269 | * Release ms->rp_state.from_dst_file (and postcopy_qemufile_src if | |
2270 | * existed) in a safe way. | |
2271 | */ | |
2272 | static void migration_release_dst_files(MigrationState *ms) | |
43044ac0 PX |
2273 | { |
2274 | QEMUFile *file; | |
2275 | ||
2276 | WITH_QEMU_LOCK_GUARD(&ms->qemu_file_lock) { | |
2277 | /* | |
2278 | * Reset the from_dst_file pointer first before releasing it, as we | |
2279 | * can't block within lock section | |
2280 | */ | |
2281 | file = ms->rp_state.from_dst_file; | |
2282 | ms->rp_state.from_dst_file = NULL; | |
2283 | } | |
2284 | ||
93589827 PX |
2285 | /* |
2286 | * Do the same to postcopy fast path socket too if there is. No | |
2287 | * locking needed because this qemufile should only be managed by | |
2288 | * return path thread. | |
2289 | */ | |
2290 | if (ms->postcopy_qemufile_src) { | |
2291 | migration_ioc_unregister_yank_from_file(ms->postcopy_qemufile_src); | |
2292 | qemu_file_shutdown(ms->postcopy_qemufile_src); | |
2293 | qemu_fclose(ms->postcopy_qemufile_src); | |
2294 | ms->postcopy_qemufile_src = NULL; | |
2295 | } | |
2296 | ||
43044ac0 PX |
2297 | qemu_fclose(file); |
2298 | } | |
2299 | ||
70b20477 DDAG |
2300 | /* |
2301 | * Handles messages sent on the return path towards the source VM | |
2302 | * | |
2303 | */ | |
2304 | static void *source_return_path_thread(void *opaque) | |
2305 | { | |
2306 | MigrationState *ms = opaque; | |
2307 | QEMUFile *rp = ms->rp_state.from_dst_file; | |
2308 | uint16_t header_len, header_type; | |
568b01ca | 2309 | uint8_t buf[512]; |
70b20477 | 2310 | uint32_t tmp32, sibling_error; |
1e2d90eb DDAG |
2311 | ram_addr_t start = 0; /* =0 to silence warning */ |
2312 | size_t len = 0, expected_len; | |
7aa6070d | 2313 | Error *err = NULL; |
70b20477 DDAG |
2314 | int res; |
2315 | ||
2316 | trace_source_return_path_thread_entry(); | |
74637e6f | 2317 | rcu_register_thread(); |
14b1742e | 2318 | |
7dcb3c87 | 2319 | while (migration_is_setup_or_active()) { |
70b20477 | 2320 | trace_source_return_path_thread_loop_top(); |
7aa6070d | 2321 | |
70b20477 DDAG |
2322 | header_type = qemu_get_be16(rp); |
2323 | header_len = qemu_get_be16(rp); | |
2324 | ||
7a9ddfbf | 2325 | if (qemu_file_get_error(rp)) { |
7aa6070d | 2326 | qemu_file_get_error_obj(rp, &err); |
7a9ddfbf PX |
2327 | goto out; |
2328 | } | |
2329 | ||
70b20477 DDAG |
2330 | if (header_type >= MIG_RP_MSG_MAX || |
2331 | header_type == MIG_RP_MSG_INVALID) { | |
7aa6070d PX |
2332 | error_setg(&err, "Received invalid message 0x%04x length 0x%04x", |
2333 | header_type, header_len); | |
70b20477 DDAG |
2334 | goto out; |
2335 | } | |
2336 | ||
2337 | if ((rp_cmd_args[header_type].len != -1 && | |
2338 | header_len != rp_cmd_args[header_type].len) || | |
568b01ca | 2339 | header_len > sizeof(buf)) { |
7aa6070d PX |
2340 | error_setg(&err, "Received '%s' message (0x%04x) with" |
2341 | "incorrect length %d expecting %zu", | |
2342 | rp_cmd_args[header_type].name, header_type, header_len, | |
2343 | (size_t)rp_cmd_args[header_type].len); | |
70b20477 DDAG |
2344 | goto out; |
2345 | } | |
2346 | ||
2347 | /* We know we've got a valid header by this point */ | |
2348 | res = qemu_get_buffer(rp, buf, header_len); | |
2349 | if (res != header_len) { | |
7aa6070d PX |
2350 | error_setg(&err, "Failed reading data for message 0x%04x" |
2351 | " read %d expected %d", | |
2352 | header_type, res, header_len); | |
70b20477 DDAG |
2353 | goto out; |
2354 | } | |
2355 | ||
2356 | /* OK, we have the message and the data */ | |
2357 | switch (header_type) { | |
2358 | case MIG_RP_MSG_SHUT: | |
4d885131 | 2359 | sibling_error = ldl_be_p(buf); |
70b20477 DDAG |
2360 | trace_source_return_path_thread_shut(sibling_error); |
2361 | if (sibling_error) { | |
7aa6070d | 2362 | error_setg(&err, "Sibling indicated error %d", sibling_error); |
70b20477 DDAG |
2363 | } |
2364 | /* | |
2365 | * We'll let the main thread deal with closing the RP | |
2366 | * we could do a shutdown(2) on it, but we're the only user | |
2367 | * anyway, so there's nothing gained. | |
2368 | */ | |
2369 | goto out; | |
2370 | ||
2371 | case MIG_RP_MSG_PONG: | |
4d885131 | 2372 | tmp32 = ldl_be_p(buf); |
70b20477 | 2373 | trace_source_return_path_thread_pong(tmp32); |
b28fb582 | 2374 | qemu_sem_post(&ms->rp_state.rp_pong_acks); |
70b20477 DDAG |
2375 | break; |
2376 | ||
1e2d90eb | 2377 | case MIG_RP_MSG_REQ_PAGES: |
4d885131 PM |
2378 | start = ldq_be_p(buf); |
2379 | len = ldl_be_p(buf + 8); | |
7aa6070d PX |
2380 | migrate_handle_rp_req_pages(ms, NULL, start, len, &err); |
2381 | if (err) { | |
2382 | goto out; | |
2383 | } | |
1e2d90eb DDAG |
2384 | break; |
2385 | ||
2386 | case MIG_RP_MSG_REQ_PAGES_ID: | |
2387 | expected_len = 12 + 1; /* header + termination */ | |
2388 | ||
2389 | if (header_len >= expected_len) { | |
4d885131 PM |
2390 | start = ldq_be_p(buf); |
2391 | len = ldl_be_p(buf + 8); | |
1e2d90eb DDAG |
2392 | /* Now we expect an idstr */ |
2393 | tmp32 = buf[12]; /* Length of the following idstr */ | |
2394 | buf[13 + tmp32] = '\0'; | |
2395 | expected_len += tmp32; | |
2396 | } | |
2397 | if (header_len != expected_len) { | |
7aa6070d PX |
2398 | error_setg(&err, "Req_Page_id with length %d expecting %zd", |
2399 | header_len, expected_len); | |
2400 | goto out; | |
2401 | } | |
2402 | migrate_handle_rp_req_pages(ms, (char *)&buf[13], start, len, | |
2403 | &err); | |
2404 | if (err) { | |
1e2d90eb DDAG |
2405 | goto out; |
2406 | } | |
1e2d90eb DDAG |
2407 | break; |
2408 | ||
a335debb PX |
2409 | case MIG_RP_MSG_RECV_BITMAP: |
2410 | if (header_len < 1) { | |
7aa6070d | 2411 | error_setg(&err, "MIG_RP_MSG_RECV_BITMAP missing block name"); |
a335debb PX |
2412 | goto out; |
2413 | } | |
2414 | /* Format: len (1B) + idstr (<255B). This ends the idstr. */ | |
2415 | buf[buf[0] + 1] = '\0'; | |
88577f32 | 2416 | if (!migrate_handle_rp_recv_bitmap(ms, (char *)(buf + 1), &err)) { |
a335debb PX |
2417 | goto out; |
2418 | } | |
2419 | break; | |
2420 | ||
13955b89 PX |
2421 | case MIG_RP_MSG_RESUME_ACK: |
2422 | tmp32 = ldl_be_p(buf); | |
88577f32 | 2423 | if (!migrate_handle_rp_resume_ack(ms, tmp32, &err)) { |
13955b89 PX |
2424 | goto out; |
2425 | } | |
2426 | break; | |
2427 | ||
1b4adb10 AH |
2428 | case MIG_RP_MSG_SWITCHOVER_ACK: |
2429 | ms->switchover_acked = true; | |
2430 | trace_source_return_path_thread_switchover_acked(); | |
2431 | break; | |
2432 | ||
70b20477 DDAG |
2433 | default: |
2434 | break; | |
2435 | } | |
2436 | } | |
14b1742e PX |
2437 | |
2438 | out: | |
7aa6070d PX |
2439 | if (err) { |
2440 | migrate_set_error(ms, err); | |
2441 | error_free(err); | |
70b20477 | 2442 | trace_source_return_path_thread_bad_end(); |
70b20477 DDAG |
2443 | } |
2444 | ||
f8c543e8 PX |
2445 | if (ms->state == MIGRATION_STATUS_POSTCOPY_RECOVER) { |
2446 | /* | |
2447 | * this will be extremely unlikely: that we got yet another network | |
2448 | * issue during recovering of the 1st network failure.. during this | |
2449 | * period the main migration thread can be waiting on rp_sem for | |
2450 | * this thread to sync with the other side. | |
2451 | * | |
2452 | * When this happens, explicitly kick the migration thread out of | |
2453 | * RECOVER stage and back to PAUSED, so the admin can try | |
2454 | * everything again. | |
2455 | */ | |
2456 | migration_rp_kick(ms); | |
2457 | } | |
2458 | ||
70b20477 | 2459 | trace_source_return_path_thread_end(); |
74637e6f | 2460 | rcu_unregister_thread(); |
7aa6070d | 2461 | |
70b20477 DDAG |
2462 | return NULL; |
2463 | } | |
2464 | ||
ef796ee9 | 2465 | static int open_return_path_on_source(MigrationState *ms) |
70b20477 | 2466 | { |
89a02a9f | 2467 | ms->rp_state.from_dst_file = qemu_file_get_return_path(ms->to_dst_file); |
70b20477 DDAG |
2468 | if (!ms->rp_state.from_dst_file) { |
2469 | return -1; | |
2470 | } | |
2471 | ||
2472 | trace_open_return_path_on_source(); | |
d3e35b8f | 2473 | |
70b20477 DDAG |
2474 | qemu_thread_create(&ms->rp_state.rp_thread, "return path", |
2475 | source_return_path_thread, ms, QEMU_THREAD_JOINABLE); | |
53021ea1 | 2476 | ms->rp_state.rp_thread_created = true; |
70b20477 DDAG |
2477 | |
2478 | trace_open_return_path_on_source_continue(); | |
2479 | ||
2480 | return 0; | |
2481 | } | |
2482 | ||
7aa6070d PX |
2483 | /* Return true if error detected, or false otherwise */ |
2484 | static bool close_return_path_on_source(MigrationState *ms) | |
70b20477 | 2485 | { |
d50f5dc0 | 2486 | if (!ms->rp_state.rp_thread_created) { |
7aa6070d | 2487 | return false; |
d50f5dc0 FR |
2488 | } |
2489 | ||
2490 | trace_migration_return_path_end_before(); | |
2491 | ||
70b20477 | 2492 | /* |
639decf5 FR |
2493 | * If this is a normal exit then the destination will send a SHUT |
2494 | * and the rp_thread will exit, however if there's an error we | |
2495 | * need to cause it to exit. shutdown(2), if we have it, will | |
2496 | * cause it to unblock if it's stuck waiting for the destination. | |
70b20477 | 2497 | */ |
639decf5 | 2498 | WITH_QEMU_LOCK_GUARD(&ms->qemu_file_lock) { |
9425ef3f | 2499 | if (migrate_has_error(ms) && ms->rp_state.from_dst_file) { |
639decf5 FR |
2500 | qemu_file_shutdown(ms->rp_state.from_dst_file); |
2501 | } | |
70b20477 | 2502 | } |
639decf5 | 2503 | |
70b20477 | 2504 | qemu_thread_join(&ms->rp_state.rp_thread); |
53021ea1 | 2505 | ms->rp_state.rp_thread_created = false; |
36e9aab3 | 2506 | migration_release_dst_files(ms); |
7aa6070d | 2507 | trace_migration_return_path_end_after(); |
36e9aab3 | 2508 | |
7aa6070d PX |
2509 | /* Return path will persist the error in MigrationState when quit */ |
2510 | return migrate_has_error(ms); | |
70b20477 DDAG |
2511 | } |
2512 | ||
5655aab0 PX |
2513 | static inline void |
2514 | migration_wait_main_channel(MigrationState *ms) | |
2515 | { | |
2516 | /* Wait until one PONG message received */ | |
2517 | qemu_sem_wait(&ms->rp_state.rp_pong_acks); | |
2518 | } | |
2519 | ||
1d34e4bf DDAG |
2520 | /* |
2521 | * Switch from normal iteration to postcopy | |
2522 | * Returns non-0 on error | |
2523 | */ | |
908927db | 2524 | static int postcopy_start(MigrationState *ms, Error **errp) |
1d34e4bf DDAG |
2525 | { |
2526 | int ret; | |
61b67d47 DB |
2527 | QIOChannelBuffer *bioc; |
2528 | QEMUFile *fb; | |
52033349 | 2529 | uint64_t bandwidth = migrate_max_postcopy_bandwidth(); |
ef8d6488 | 2530 | bool restart_block = false; |
0331c8ca | 2531 | int cur_state = MIGRATION_STATUS_ACTIVE; |
d0edb8a1 | 2532 | |
5655aab0 PX |
2533 | if (migrate_postcopy_preempt()) { |
2534 | migration_wait_main_channel(ms); | |
2535 | if (postcopy_preempt_establish_channel(ms)) { | |
2536 | migrate_set_state(&ms->state, ms->state, MIGRATION_STATUS_FAILED); | |
d0ad271a AH |
2537 | error_setg(errp, "%s: Failed to establish preempt channel", |
2538 | __func__); | |
5655aab0 PX |
2539 | return -1; |
2540 | } | |
d0edb8a1 PX |
2541 | } |
2542 | ||
0331c8ca DDAG |
2543 | if (!migrate_pause_before_switchover()) { |
2544 | migrate_set_state(&ms->state, MIGRATION_STATUS_ACTIVE, | |
2545 | MIGRATION_STATUS_POSTCOPY_ACTIVE); | |
2546 | } | |
1d34e4bf DDAG |
2547 | |
2548 | trace_postcopy_start(); | |
195801d7 | 2549 | bql_lock(); |
1d34e4bf DDAG |
2550 | trace_postcopy_start_set_run(); |
2551 | ||
9867d4dd | 2552 | ret = migration_stop_vm(ms, RUN_STATE_FINISH_MIGRATE); |
76b1c7fe | 2553 | if (ret < 0) { |
d0ad271a | 2554 | error_setg_errno(errp, -ret, "%s: Failed to stop the VM", __func__); |
76b1c7fe KW |
2555 | goto fail; |
2556 | } | |
1d34e4bf | 2557 | |
0331c8ca DDAG |
2558 | ret = migration_maybe_pause(ms, &cur_state, |
2559 | MIGRATION_STATUS_POSTCOPY_ACTIVE); | |
2560 | if (ret < 0) { | |
d0ad271a AH |
2561 | error_setg_errno(errp, -ret, "%s: Failed in migration_maybe_pause()", |
2562 | __func__); | |
0331c8ca DDAG |
2563 | goto fail; |
2564 | } | |
2565 | ||
76b1c7fe | 2566 | ret = bdrv_inactivate_all(); |
1d34e4bf | 2567 | if (ret < 0) { |
d0ad271a AH |
2568 | error_setg_errno(errp, -ret, "%s: Failed in bdrv_inactivate_all()", |
2569 | __func__); | |
1d34e4bf DDAG |
2570 | goto fail; |
2571 | } | |
ef8d6488 | 2572 | restart_block = true; |
1d34e4bf | 2573 | |
1c0d249d DDAG |
2574 | /* |
2575 | * Cause any non-postcopiable, but iterative devices to | |
2576 | * send out their final data. | |
2577 | */ | |
a1fbe750 | 2578 | qemu_savevm_state_complete_precopy(ms->to_dst_file, true, false); |
1c0d249d | 2579 | |
1d34e4bf DDAG |
2580 | /* |
2581 | * in Finish migrate and with the io-lock held everything should | |
2582 | * be quiet, but we've potentially still got dirty pages and we | |
2583 | * need to tell the destination to throw any pages it's already received | |
2584 | * that are dirty | |
2585 | */ | |
58110f0a | 2586 | if (migrate_postcopy_ram()) { |
739fcc1b | 2587 | ram_postcopy_send_discard_bitmap(ms); |
1d34e4bf DDAG |
2588 | } |
2589 | ||
2590 | /* | |
2591 | * send rest of state - note things that are doing postcopy | |
2592 | * will notice we're in POSTCOPY_ACTIVE and not actually | |
2593 | * wrap their state up here | |
2594 | */ | |
e1fde0e0 | 2595 | migration_rate_set(bandwidth); |
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"); |
77ef2dc1 | 2614 | fb = qemu_file_new_output(QIO_CHANNEL(bioc)); |
61b67d47 | 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) { | |
908927db | 2638 | error_setg(errp, "postcopy_start: Migration stream errored (pre package)"); |
ef8d6488 DDAG |
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)) { |
d0ad271a | 2646 | error_setg(errp, "%s: Failed to send packaged data", __func__); |
1d34e4bf DDAG |
2647 | goto fail_closefb; |
2648 | } | |
2649 | qemu_fclose(fb); | |
b82fc321 DDAG |
2650 | |
2651 | /* Send a notify to give a chance for anything that needs to happen | |
2652 | * at the transition to postcopy and after the device state; in particular | |
2653 | * spice needs to trigger a transition now | |
2654 | */ | |
4af667f8 | 2655 | migration_call_notifiers(ms, MIG_EVENT_PRECOPY_DONE, NULL); |
b82fc321 | 2656 | |
e22ffad0 | 2657 | migration_downtime_end(ms); |
1d34e4bf | 2658 | |
195801d7 | 2659 | bql_unlock(); |
1d34e4bf | 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 | 2674 | if (ret) { |
4af667f8 SS |
2675 | error_setg_errno(errp, -ret, "postcopy_start: Migration stream error"); |
2676 | bql_lock(); | |
2677 | goto fail; | |
1d34e4bf | 2678 | } |
c01b16ed PX |
2679 | trace_postcopy_preempt_enabled(migrate_postcopy_preempt()); |
2680 | ||
1d34e4bf DDAG |
2681 | return ret; |
2682 | ||
2683 | fail_closefb: | |
2684 | qemu_fclose(fb); | |
2685 | fail: | |
48781e5b | 2686 | migrate_set_state(&ms->state, MIGRATION_STATUS_POSTCOPY_ACTIVE, |
1d34e4bf | 2687 | MIGRATION_STATUS_FAILED); |
ef8d6488 DDAG |
2688 | if (restart_block) { |
2689 | /* A failure happened early enough that we know the destination hasn't | |
2690 | * accessed block devices, so we're safe to recover. | |
2691 | */ | |
2692 | Error *local_err = NULL; | |
2693 | ||
3b717194 | 2694 | bdrv_activate_all(&local_err); |
ef8d6488 DDAG |
2695 | if (local_err) { |
2696 | error_report_err(local_err); | |
2697 | } | |
2698 | } | |
4af667f8 | 2699 | migration_call_notifiers(ms, MIG_EVENT_PRECOPY_FAILED, NULL); |
195801d7 | 2700 | bql_unlock(); |
1d34e4bf DDAG |
2701 | return -1; |
2702 | } | |
2703 | ||
e91d8951 DDAG |
2704 | /** |
2705 | * migration_maybe_pause: Pause if required to by | |
a4a411fb | 2706 | * migrate_pause_before_switchover called with the BQL locked |
e91d8951 DDAG |
2707 | * Returns: 0 on success |
2708 | */ | |
0331c8ca DDAG |
2709 | static int migration_maybe_pause(MigrationState *s, |
2710 | int *current_active_state, | |
2711 | int new_state) | |
e91d8951 DDAG |
2712 | { |
2713 | if (!migrate_pause_before_switchover()) { | |
2714 | return 0; | |
2715 | } | |
2716 | ||
2717 | /* Since leaving this state is not atomic with posting the semaphore | |
2718 | * it's possible that someone could have issued multiple migrate_continue | |
2719 | * and the semaphore is incorrectly positive at this point; | |
2720 | * the docs say it's undefined to reinit a semaphore that's already | |
2721 | * init'd, so use timedwait to eat up any existing posts. | |
2722 | */ | |
2723 | while (qemu_sem_timedwait(&s->pause_sem, 1) == 0) { | |
2724 | /* This block intentionally left blank */ | |
2725 | } | |
2726 | ||
8958338b ZF |
2727 | /* |
2728 | * If the migration is cancelled when it is in the completion phase, | |
2729 | * the migration state is set to MIGRATION_STATUS_CANCELLING. | |
2730 | * So we don't need to wait a semaphore, otherwise we would always | |
2731 | * wait for the 'pause_sem' semaphore. | |
2732 | */ | |
2733 | if (s->state != MIGRATION_STATUS_CANCELLING) { | |
195801d7 | 2734 | bql_unlock(); |
8958338b ZF |
2735 | migrate_set_state(&s->state, *current_active_state, |
2736 | MIGRATION_STATUS_PRE_SWITCHOVER); | |
2737 | qemu_sem_wait(&s->pause_sem); | |
2738 | migrate_set_state(&s->state, MIGRATION_STATUS_PRE_SWITCHOVER, | |
2739 | new_state); | |
2740 | *current_active_state = new_state; | |
195801d7 | 2741 | bql_lock(); |
8958338b | 2742 | } |
e91d8951 | 2743 | |
0331c8ca | 2744 | return s->state == new_state ? 0 : -EINVAL; |
e91d8951 DDAG |
2745 | } |
2746 | ||
d7f5a043 WW |
2747 | static int migration_completion_precopy(MigrationState *s, |
2748 | int *current_active_state) | |
09f6c85e DDAG |
2749 | { |
2750 | int ret; | |
2751 | ||
195801d7 | 2752 | bql_lock(); |
c33f1829 | 2753 | |
9867d4dd SS |
2754 | if (!migrate_mode_is_cpr(s)) { |
2755 | ret = migration_stop_vm(s, RUN_STATE_FINISH_MIGRATE); | |
2756 | if (ret < 0) { | |
2757 | goto out_unlock; | |
2758 | } | |
d7f5a043 | 2759 | } |
c33f1829 | 2760 | |
d7f5a043 WW |
2761 | ret = migration_maybe_pause(s, current_active_state, |
2762 | MIGRATION_STATUS_DEVICE); | |
2763 | if (ret < 0) { | |
2764 | goto out_unlock; | |
2765 | } | |
09f6c85e | 2766 | |
d7f5a043 WW |
2767 | /* |
2768 | * Inactivate disks except in COLO, and track that we have done so in order | |
2769 | * to remember to reactivate them if migration fails or is cancelled. | |
2770 | */ | |
2771 | s->block_inactive = !migrate_colo(); | |
2772 | migration_rate_set(RATE_LIMIT_DISABLED); | |
2773 | ret = qemu_savevm_state_complete_precopy(s->to_dst_file, false, | |
2774 | s->block_inactive); | |
2775 | out_unlock: | |
195801d7 | 2776 | bql_unlock(); |
d7f5a043 WW |
2777 | return ret; |
2778 | } | |
b10ac0c4 | 2779 | |
d7f5a043 WW |
2780 | static void migration_completion_postcopy(MigrationState *s) |
2781 | { | |
2782 | trace_migration_completion_postcopy_end(); | |
2783 | ||
195801d7 | 2784 | bql_lock(); |
d7f5a043 | 2785 | qemu_savevm_state_complete_postcopy(s->to_dst_file); |
195801d7 | 2786 | bql_unlock(); |
d7f5a043 WW |
2787 | |
2788 | /* | |
2789 | * Shutdown the postcopy fast path thread. This is only needed when dest | |
2790 | * QEMU binary is old (7.1/7.2). QEMU 8.0+ doesn't need this. | |
2791 | */ | |
2792 | if (migrate_postcopy_preempt() && s->preempt_pre_7_2) { | |
2793 | postcopy_preempt_shutdown_file(s); | |
2794 | } | |
68b88468 | 2795 | |
d7f5a043 WW |
2796 | trace_migration_completion_postcopy_end_after_complete(); |
2797 | } | |
2798 | ||
2799 | static void migration_completion_failed(MigrationState *s, | |
2800 | int current_active_state) | |
2801 | { | |
2802 | if (s->block_inactive && (s->state == MIGRATION_STATUS_ACTIVE || | |
2803 | s->state == MIGRATION_STATUS_DEVICE)) { | |
6621883f | 2804 | /* |
d7f5a043 WW |
2805 | * If not doing postcopy, vm_start() will be called: let's |
2806 | * regain control on images. | |
6621883f | 2807 | */ |
d7f5a043 WW |
2808 | Error *local_err = NULL; |
2809 | ||
195801d7 | 2810 | bql_lock(); |
d7f5a043 WW |
2811 | bdrv_activate_all(&local_err); |
2812 | if (local_err) { | |
2813 | error_report_err(local_err); | |
2814 | } else { | |
2815 | s->block_inactive = false; | |
36f62f11 | 2816 | } |
195801d7 | 2817 | bql_unlock(); |
d7f5a043 WW |
2818 | } |
2819 | ||
2820 | migrate_set_state(&s->state, current_active_state, | |
2821 | MIGRATION_STATUS_FAILED); | |
2822 | } | |
36f62f11 | 2823 | |
d7f5a043 WW |
2824 | /** |
2825 | * migration_completion: Used by migration_thread when there's not much left. | |
2826 | * The caller 'breaks' the loop when this returns. | |
2827 | * | |
2828 | * @s: Current migration state | |
2829 | */ | |
2830 | static void migration_completion(MigrationState *s) | |
2831 | { | |
2832 | int ret = 0; | |
2833 | int current_active_state = s->state; | |
30158d88 | 2834 | Error *local_err = NULL; |
d7f5a043 WW |
2835 | |
2836 | if (s->state == MIGRATION_STATUS_ACTIVE) { | |
2837 | ret = migration_completion_precopy(s, ¤t_active_state); | |
2838 | } else if (s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) { | |
2839 | migration_completion_postcopy(s); | |
444252b9 | 2840 | } else { |
d7f5a043 WW |
2841 | ret = -1; |
2842 | } | |
2843 | ||
2844 | if (ret < 0) { | |
6ba11211 | 2845 | goto fail; |
09f6c85e | 2846 | } |
09f6c85e | 2847 | |
d7f5a043 | 2848 | if (close_return_path_on_source(s)) { |
d50f5dc0 | 2849 | goto fail; |
09f6c85e DDAG |
2850 | } |
2851 | ||
89a02a9f | 2852 | if (qemu_file_get_error(s->to_dst_file)) { |
09f6c85e | 2853 | trace_migration_completion_file_err(); |
6dab4c93 | 2854 | goto fail; |
09f6c85e DDAG |
2855 | } |
2856 | ||
5e804644 | 2857 | if (migrate_colo() && s->state == MIGRATION_STATUS_ACTIVE) { |
eeeb48ee ZC |
2858 | /* COLO does not support postcopy */ |
2859 | migrate_set_state(&s->state, MIGRATION_STATUS_ACTIVE, | |
2860 | MIGRATION_STATUS_COLO); | |
2861 | } else { | |
63f64d77 | 2862 | migration_completion_end(s); |
0b827d5e HZ |
2863 | } |
2864 | ||
09f6c85e DDAG |
2865 | return; |
2866 | ||
6dab4c93 | 2867 | fail: |
30158d88 AH |
2868 | if (qemu_file_get_error_obj(s->to_dst_file, &local_err)) { |
2869 | migrate_set_error(s, local_err); | |
2870 | error_free(local_err); | |
2871 | } else if (ret) { | |
2872 | error_setg_errno(&local_err, -ret, "Error in migration completion"); | |
2873 | migrate_set_error(s, local_err); | |
2874 | error_free(local_err); | |
2875 | } | |
2876 | ||
d7f5a043 | 2877 | migration_completion_failed(s, current_active_state); |
09f6c85e DDAG |
2878 | } |
2879 | ||
8518278a AG |
2880 | /** |
2881 | * bg_migration_completion: Used by bg_migration_thread when after all the | |
2882 | * RAM has been saved. The caller 'breaks' the loop when this returns. | |
2883 | * | |
2884 | * @s: Current migration state | |
2885 | */ | |
2886 | static void bg_migration_completion(MigrationState *s) | |
2887 | { | |
2888 | int current_active_state = s->state; | |
2889 | ||
8518278a AG |
2890 | if (s->state == MIGRATION_STATUS_ACTIVE) { |
2891 | /* | |
2892 | * By this moment we have RAM content saved into the migration stream. | |
2893 | * The next step is to flush the non-RAM content (device state) | |
2894 | * right after the ram content. The device state has been stored into | |
2895 | * the temporary buffer before RAM saving started. | |
2896 | */ | |
2897 | qemu_put_buffer(s->to_dst_file, s->bioc->data, s->bioc->usage); | |
2898 | qemu_fflush(s->to_dst_file); | |
2899 | } else if (s->state == MIGRATION_STATUS_CANCELLING) { | |
2900 | goto fail; | |
2901 | } | |
2902 | ||
2903 | if (qemu_file_get_error(s->to_dst_file)) { | |
2904 | trace_migration_completion_file_err(); | |
2905 | goto fail; | |
2906 | } | |
2907 | ||
63f64d77 | 2908 | migration_completion_end(s); |
8518278a AG |
2909 | return; |
2910 | ||
2911 | fail: | |
2912 | migrate_set_state(&s->state, current_active_state, | |
2913 | MIGRATION_STATUS_FAILED); | |
2914 | } | |
2915 | ||
b23c2ade PX |
2916 | typedef enum MigThrError { |
2917 | /* No error detected */ | |
2918 | MIG_THR_ERR_NONE = 0, | |
2919 | /* Detected error, but resumed successfully */ | |
2920 | MIG_THR_ERR_RECOVERED = 1, | |
2921 | /* Detected fatal error, need to exit */ | |
2922 | MIG_THR_ERR_FATAL = 2, | |
2923 | } MigThrError; | |
2924 | ||
94190696 PX |
2925 | static int postcopy_resume_handshake(MigrationState *s) |
2926 | { | |
2927 | qemu_savevm_send_postcopy_resume(s->to_dst_file); | |
2928 | ||
2929 | while (s->state == MIGRATION_STATUS_POSTCOPY_RECOVER) { | |
f8c543e8 PX |
2930 | if (migration_rp_wait(s)) { |
2931 | return -1; | |
2932 | } | |
94190696 PX |
2933 | } |
2934 | ||
2935 | if (s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) { | |
2936 | return 0; | |
2937 | } | |
2938 | ||
2939 | return -1; | |
2940 | } | |
2941 | ||
135b87b4 PX |
2942 | /* Return zero if success, or <0 for error */ |
2943 | static int postcopy_do_resume(MigrationState *s) | |
2944 | { | |
d1b8eadb PX |
2945 | int ret; |
2946 | ||
2947 | /* | |
2948 | * Call all the resume_prepare() hooks, so that modules can be | |
2949 | * ready for the migration resume. | |
2950 | */ | |
2951 | ret = qemu_savevm_state_resume_prepare(s); | |
2952 | if (ret) { | |
2953 | error_report("%s: resume_prepare() failure detected: %d", | |
2954 | __func__, ret); | |
2955 | return ret; | |
2956 | } | |
2957 | ||
5655aab0 PX |
2958 | /* |
2959 | * If preempt is enabled, re-establish the preempt channel. Note that | |
2960 | * we do it after resume prepare to make sure the main channel will be | |
2961 | * created before the preempt channel. E.g. with weak network, the | |
2962 | * dest QEMU may get messed up with the preempt and main channels on | |
2963 | * the order of connection setup. This guarantees the correct order. | |
2964 | */ | |
2965 | ret = postcopy_preempt_establish_channel(s); | |
2966 | if (ret) { | |
2967 | error_report("%s: postcopy_preempt_establish_channel(): %d", | |
2968 | __func__, ret); | |
2969 | return ret; | |
2970 | } | |
2971 | ||
d1b8eadb | 2972 | /* |
94190696 PX |
2973 | * Last handshake with destination on the resume (destination will |
2974 | * switch to postcopy-active afterwards) | |
d1b8eadb | 2975 | */ |
94190696 PX |
2976 | ret = postcopy_resume_handshake(s); |
2977 | if (ret) { | |
2978 | error_report("%s: handshake failed: %d", __func__, ret); | |
2979 | return ret; | |
2980 | } | |
d1b8eadb | 2981 | |
135b87b4 PX |
2982 | return 0; |
2983 | } | |
2984 | ||
b23c2ade PX |
2985 | /* |
2986 | * We don't return until we are in a safe state to continue current | |
2987 | * postcopy migration. Returns MIG_THR_ERR_RECOVERED if recovered, or | |
2988 | * MIG_THR_ERR_FATAL if unrecovery failure happened. | |
2989 | */ | |
2990 | static MigThrError postcopy_pause(MigrationState *s) | |
2991 | { | |
2992 | assert(s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE); | |
b23c2ade | 2993 | |
135b87b4 | 2994 | while (true) { |
62df066f PX |
2995 | QEMUFile *file; |
2996 | ||
22b04245 FR |
2997 | /* |
2998 | * We're already pausing, so ignore any errors on the return | |
2999 | * path and just wait for the thread to finish. It will be | |
3000 | * re-created when we resume. | |
3001 | */ | |
3002 | close_return_path_on_source(s); | |
3003 | ||
39675fff PX |
3004 | /* |
3005 | * Current channel is possibly broken. Release it. Note that this is | |
3006 | * guaranteed even without lock because to_dst_file should only be | |
3007 | * modified by the migration thread. That also guarantees that the | |
3008 | * unregister of yank is safe too without the lock. It should be safe | |
3009 | * even to be within the qemu_file_lock, but we didn't do that to avoid | |
3010 | * taking more mutex (yank_lock) within qemu_file_lock. TL;DR: we make | |
3011 | * the qemu_file_lock critical section as small as possible. | |
3012 | */ | |
135b87b4 | 3013 | assert(s->to_dst_file); |
39675fff | 3014 | migration_ioc_unregister_yank_from_file(s->to_dst_file); |
62df066f PX |
3015 | qemu_mutex_lock(&s->qemu_file_lock); |
3016 | file = s->to_dst_file; | |
135b87b4 | 3017 | s->to_dst_file = NULL; |
62df066f PX |
3018 | qemu_mutex_unlock(&s->qemu_file_lock); |
3019 | ||
3020 | qemu_file_shutdown(file); | |
3021 | qemu_fclose(file); | |
b23c2ade | 3022 | |
d246ea50 PX |
3023 | migrate_set_state(&s->state, s->state, |
3024 | MIGRATION_STATUS_POSTCOPY_PAUSED); | |
3025 | ||
135b87b4 PX |
3026 | error_report("Detected IO failure for postcopy. " |
3027 | "Migration paused."); | |
b23c2ade | 3028 | |
135b87b4 PX |
3029 | /* |
3030 | * We wait until things fixed up. Then someone will setup the | |
3031 | * status back for us. | |
3032 | */ | |
3033 | while (s->state == MIGRATION_STATUS_POSTCOPY_PAUSED) { | |
3034 | qemu_sem_wait(&s->postcopy_pause_sem); | |
3035 | } | |
3036 | ||
3037 | if (s->state == MIGRATION_STATUS_POSTCOPY_RECOVER) { | |
3038 | /* Woken up by a recover procedure. Give it a shot */ | |
b23c2ade | 3039 | |
135b87b4 PX |
3040 | /* Do the resume logic */ |
3041 | if (postcopy_do_resume(s) == 0) { | |
3042 | /* Let's continue! */ | |
3043 | trace_postcopy_pause_continued(); | |
3044 | return MIG_THR_ERR_RECOVERED; | |
3045 | } else { | |
3046 | /* | |
3047 | * Something wrong happened during the recovery, let's | |
3048 | * pause again. Pause is always better than throwing | |
3049 | * data away. | |
3050 | */ | |
3051 | continue; | |
3052 | } | |
3053 | } else { | |
3054 | /* This is not right... Time to quit. */ | |
3055 | return MIG_THR_ERR_FATAL; | |
3056 | } | |
3057 | } | |
b23c2ade PX |
3058 | } |
3059 | ||
20c64c8a SS |
3060 | void migration_file_set_error(int err) |
3061 | { | |
3062 | MigrationState *s = current_migration; | |
3063 | ||
3064 | WITH_QEMU_LOCK_GUARD(&s->qemu_file_lock) { | |
3065 | if (s->to_dst_file) { | |
3066 | qemu_file_set_error(s->to_dst_file, err); | |
3067 | } | |
3068 | } | |
3069 | } | |
3070 | ||
b23c2ade PX |
3071 | static MigThrError migration_detect_error(MigrationState *s) |
3072 | { | |
3073 | int ret; | |
c3c5eae6 | 3074 | int state = s->state; |
3d661c8a | 3075 | Error *local_error = NULL; |
c3c5eae6 DDAG |
3076 | |
3077 | if (state == MIGRATION_STATUS_CANCELLING || | |
3078 | state == MIGRATION_STATUS_CANCELLED) { | |
3079 | /* End the migration, but don't set the state to failed */ | |
3080 | return MIG_THR_ERR_FATAL; | |
3081 | } | |
b23c2ade | 3082 | |
60bb3c58 PX |
3083 | /* |
3084 | * Try to detect any file errors. Note that postcopy_qemufile_src will | |
3085 | * be NULL when postcopy preempt is not enabled. | |
3086 | */ | |
3087 | ret = qemu_file_get_error_obj_any(s->to_dst_file, | |
3088 | s->postcopy_qemufile_src, | |
3089 | &local_error); | |
b23c2ade PX |
3090 | if (!ret) { |
3091 | /* Everything is fine */ | |
3d661c8a | 3092 | assert(!local_error); |
b23c2ade PX |
3093 | return MIG_THR_ERR_NONE; |
3094 | } | |
3095 | ||
3d661c8a YK |
3096 | if (local_error) { |
3097 | migrate_set_error(s, local_error); | |
3098 | error_free(local_error); | |
3099 | } | |
3100 | ||
d5c8f2af | 3101 | if (state == MIGRATION_STATUS_POSTCOPY_ACTIVE && ret) { |
b23c2ade PX |
3102 | /* |
3103 | * For postcopy, we allow the network to be down for a | |
3104 | * while. After that, it can be continued by a | |
3105 | * recovery phase. | |
3106 | */ | |
3107 | return postcopy_pause(s); | |
3108 | } else { | |
3109 | /* | |
3110 | * For precopy (or postcopy with error outside IO), we fail | |
3111 | * with no time. | |
3112 | */ | |
c3c5eae6 | 3113 | migrate_set_state(&s->state, state, MIGRATION_STATUS_FAILED); |
b23c2ade PX |
3114 | trace_migration_thread_file_err(); |
3115 | ||
3116 | /* Time to stop the migration, now. */ | |
3117 | return MIG_THR_ERR_FATAL; | |
3118 | } | |
3119 | } | |
3120 | ||
63f64d77 | 3121 | static void migration_completion_end(MigrationState *s) |
cf011f08 | 3122 | { |
f57e5a6c | 3123 | uint64_t bytes = migration_transferred_bytes(); |
cf011f08 | 3124 | int64_t end_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); |
6cde6fbe | 3125 | int64_t transfer_time; |
cf011f08 | 3126 | |
63f64d77 FR |
3127 | /* |
3128 | * Take the BQL here so that query-migrate on the QMP thread sees: | |
3129 | * - atomic update of s->total_time and s->mbps; | |
3130 | * - correct ordering of s->mbps update vs. s->state; | |
3131 | */ | |
3132 | bql_lock(); | |
e22ffad0 | 3133 | migration_downtime_end(s); |
cf011f08 | 3134 | s->total_time = end_time - s->start_time; |
6cde6fbe JQ |
3135 | transfer_time = s->total_time - s->setup_time; |
3136 | if (transfer_time) { | |
3137 | s->mbps = ((double) bytes * 8.0) / transfer_time / 1000; | |
cf011f08 | 3138 | } |
63f64d77 FR |
3139 | |
3140 | migrate_set_state(&s->state, s->state, | |
3141 | MIGRATION_STATUS_COMPLETED); | |
3142 | bql_unlock(); | |
cf011f08 PX |
3143 | } |
3144 | ||
87f3bd87 IR |
3145 | static void update_iteration_initial_status(MigrationState *s) |
3146 | { | |
3147 | /* | |
3148 | * Update these three fields at the same time to avoid mismatch info lead | |
3149 | * wrong speed calculation. | |
3150 | */ | |
3151 | s->iteration_start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); | |
f57e5a6c | 3152 | s->iteration_initial_bytes = migration_transferred_bytes(); |
87f3bd87 IR |
3153 | s->iteration_initial_pages = ram_get_total_transferred_pages(); |
3154 | } | |
3155 | ||
b15df1ae PX |
3156 | static void migration_update_counters(MigrationState *s, |
3157 | int64_t current_time) | |
3158 | { | |
aecbfe9c | 3159 | uint64_t transferred, transferred_pages, time_spent; |
0c8f0efd | 3160 | uint64_t current_bytes; /* bytes transferred since the beginning */ |
8b239597 PX |
3161 | uint64_t switchover_bw; |
3162 | /* Expected bandwidth when switching over to destination QEMU */ | |
3163 | double expected_bw_per_ms; | |
b15df1ae PX |
3164 | double bandwidth; |
3165 | ||
3166 | if (current_time < s->iteration_start_time + BUFFER_DELAY) { | |
3167 | return; | |
3168 | } | |
3169 | ||
8b239597 | 3170 | switchover_bw = migrate_avail_switchover_bandwidth(); |
f57e5a6c | 3171 | current_bytes = migration_transferred_bytes(); |
0c8f0efd | 3172 | transferred = current_bytes - s->iteration_initial_bytes; |
b15df1ae PX |
3173 | time_spent = current_time - s->iteration_start_time; |
3174 | bandwidth = (double)transferred / time_spent; | |
8b239597 PX |
3175 | |
3176 | if (switchover_bw) { | |
3177 | /* | |
3178 | * If the user specified a switchover bandwidth, let's trust the | |
3179 | * user so that can be more accurate than what we estimated. | |
3180 | */ | |
3181 | expected_bw_per_ms = switchover_bw / 1000; | |
3182 | } else { | |
3183 | /* If the user doesn't specify bandwidth, we use the estimated */ | |
3184 | expected_bw_per_ms = bandwidth; | |
3185 | } | |
3186 | ||
3187 | s->threshold_size = expected_bw_per_ms * migrate_downtime_limit(); | |
b15df1ae PX |
3188 | |
3189 | s->mbps = (((double) transferred * 8.0) / | |
3190 | ((double) time_spent / 1000.0)) / 1000.0 / 1000.0; | |
3191 | ||
aecbfe9c XG |
3192 | transferred_pages = ram_get_total_transferred_pages() - |
3193 | s->iteration_initial_pages; | |
3194 | s->pages_per_second = (double) transferred_pages / | |
3195 | (((double) time_spent / 1000.0)); | |
3196 | ||
b15df1ae PX |
3197 | /* |
3198 | * if we haven't sent anything, we don't want to | |
3199 | * recalculate. 10000 is a small enough number for our purposes | |
3200 | */ | |
aff3f660 | 3201 | if (stat64_get(&mig_stats.dirty_pages_rate) && |
72f8e587 | 3202 | transferred > 10000) { |
73208a33 | 3203 | s->expected_downtime = |
8b239597 | 3204 | stat64_get(&mig_stats.dirty_bytes_last_sync) / expected_bw_per_ms; |
b15df1ae PX |
3205 | } |
3206 | ||
0743f41f | 3207 | migration_rate_reset(); |
b15df1ae | 3208 | |
87f3bd87 | 3209 | update_iteration_initial_status(s); |
b15df1ae PX |
3210 | |
3211 | trace_migrate_transferred(transferred, time_spent, | |
8b239597 PX |
3212 | /* Both in unit bytes/ms */ |
3213 | bandwidth, switchover_bw / 1000, | |
3214 | s->threshold_size); | |
b15df1ae PX |
3215 | } |
3216 | ||
1b4adb10 AH |
3217 | static bool migration_can_switchover(MigrationState *s) |
3218 | { | |
3219 | if (!migrate_switchover_ack()) { | |
3220 | return true; | |
3221 | } | |
3222 | ||
3223 | /* No reason to wait for switchover ACK if VM is stopped */ | |
3224 | if (!runstate_is_running()) { | |
3225 | return true; | |
3226 | } | |
3227 | ||
3228 | return s->switchover_acked; | |
3229 | } | |
3230 | ||
2ad87305 PX |
3231 | /* Migration thread iteration status */ |
3232 | typedef enum { | |
3233 | MIG_ITERATE_RESUME, /* Resume current iteration */ | |
3234 | MIG_ITERATE_SKIP, /* Skip current iteration */ | |
3235 | MIG_ITERATE_BREAK, /* Break the loop */ | |
3236 | } MigIterateState; | |
3237 | ||
3238 | /* | |
3239 | * Return true if continue to the next iteration directly, false | |
3240 | * otherwise. | |
3241 | */ | |
3242 | static MigIterateState migration_iteration_run(MigrationState *s) | |
3243 | { | |
910c1647 | 3244 | uint64_t must_precopy, can_postcopy, pending_size; |
908927db | 3245 | Error *local_err = NULL; |
2ad87305 | 3246 | bool in_postcopy = s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE; |
1b4adb10 | 3247 | bool can_switchover = migration_can_switchover(s); |
2ad87305 | 3248 | |
24beea4e | 3249 | qemu_savevm_state_pending_estimate(&must_precopy, &can_postcopy); |
910c1647 | 3250 | pending_size = must_precopy + can_postcopy; |
24beea4e | 3251 | trace_migrate_pending_estimate(pending_size, must_precopy, can_postcopy); |
c8df4a7a | 3252 | |
910c1647 | 3253 | if (pending_size < s->threshold_size) { |
24beea4e JQ |
3254 | qemu_savevm_state_pending_exact(&must_precopy, &can_postcopy); |
3255 | pending_size = must_precopy + can_postcopy; | |
3256 | trace_migrate_pending_exact(pending_size, must_precopy, can_postcopy); | |
c8df4a7a | 3257 | } |
2ad87305 | 3258 | |
1b4adb10 | 3259 | if ((!pending_size || pending_size < s->threshold_size) && can_switchover) { |
2ad87305 PX |
3260 | trace_migration_thread_low_pending(pending_size); |
3261 | migration_completion(s); | |
3262 | return MIG_ITERATE_BREAK; | |
3263 | } | |
3264 | ||
d9df9292 | 3265 | /* Still a significant amount to transfer */ |
1b4adb10 | 3266 | if (!in_postcopy && must_precopy <= s->threshold_size && can_switchover && |
d9df9292 | 3267 | qatomic_read(&s->start_postcopy)) { |
908927db TG |
3268 | if (postcopy_start(s, &local_err)) { |
3269 | migrate_set_error(s, local_err); | |
3270 | error_report_err(local_err); | |
d9df9292 JQ |
3271 | } |
3272 | return MIG_ITERATE_SKIP; | |
3273 | } | |
3274 | ||
3275 | /* Just another iteration step */ | |
3276 | qemu_savevm_state_iterate(s->to_dst_file, in_postcopy); | |
2ad87305 PX |
3277 | return MIG_ITERATE_RESUME; |
3278 | } | |
3279 | ||
199aa6d4 PX |
3280 | static void migration_iteration_finish(MigrationState *s) |
3281 | { | |
3282 | /* If we enabled cpu throttling for auto-converge, turn it off. */ | |
3283 | cpu_throttle_stop(); | |
3284 | ||
195801d7 | 3285 | bql_lock(); |
199aa6d4 PX |
3286 | switch (s->state) { |
3287 | case MIGRATION_STATUS_COMPLETED: | |
199aa6d4 PX |
3288 | runstate_set(RUN_STATE_POSTMIGRATE); |
3289 | break; | |
751fe4c6 | 3290 | case MIGRATION_STATUS_COLO: |
d70178a8 | 3291 | assert(migrate_colo()); |
199aa6d4 | 3292 | migrate_start_colo_process(s); |
f4584076 | 3293 | s->vm_old_state = RUN_STATE_RUNNING; |
199aa6d4 PX |
3294 | /* Fallthrough */ |
3295 | case MIGRATION_STATUS_FAILED: | |
3296 | case MIGRATION_STATUS_CANCELLED: | |
57225e5f | 3297 | case MIGRATION_STATUS_CANCELLING: |
b4e9ddcc | 3298 | if (runstate_is_live(s->vm_old_state)) { |
aa505f8e RL |
3299 | if (!runstate_check(RUN_STATE_SHUTDOWN)) { |
3300 | vm_start(); | |
3301 | } | |
199aa6d4 PX |
3302 | } else { |
3303 | if (runstate_check(RUN_STATE_FINISH_MIGRATE)) { | |
a4c6275a | 3304 | runstate_set(s->vm_old_state); |
199aa6d4 PX |
3305 | } |
3306 | } | |
3307 | break; | |
3308 | ||
3309 | default: | |
3310 | /* Should not reach here, but if so, forgive the VM. */ | |
3311 | error_report("%s: Unknown ending state %d", __func__, s->state); | |
3312 | break; | |
3313 | } | |
44d0d456 FR |
3314 | |
3315 | migration_bh_schedule(migrate_fd_cleanup_bh, s); | |
195801d7 | 3316 | bql_unlock(); |
199aa6d4 PX |
3317 | } |
3318 | ||
8518278a AG |
3319 | static void bg_migration_iteration_finish(MigrationState *s) |
3320 | { | |
3a8b81f2 FE |
3321 | /* |
3322 | * Stop tracking RAM writes - un-protect memory, un-register UFFD | |
3323 | * memory ranges, flush kernel wait queues and wake up threads | |
3324 | * waiting for write fault to be resolved. | |
3325 | */ | |
3326 | ram_write_tracking_stop(); | |
3327 | ||
195801d7 | 3328 | bql_lock(); |
8518278a AG |
3329 | switch (s->state) { |
3330 | case MIGRATION_STATUS_COMPLETED: | |
8518278a AG |
3331 | case MIGRATION_STATUS_ACTIVE: |
3332 | case MIGRATION_STATUS_FAILED: | |
3333 | case MIGRATION_STATUS_CANCELLED: | |
3334 | case MIGRATION_STATUS_CANCELLING: | |
3335 | break; | |
3336 | ||
3337 | default: | |
3338 | /* Should not reach here, but if so, forgive the VM. */ | |
3339 | error_report("%s: Unknown ending state %d", __func__, s->state); | |
3340 | break; | |
3341 | } | |
3342 | ||
44d0d456 | 3343 | migration_bh_schedule(migrate_fd_cleanup_bh, s); |
195801d7 | 3344 | bql_unlock(); |
8518278a AG |
3345 | } |
3346 | ||
3347 | /* | |
3348 | * Return true if continue to the next iteration directly, false | |
3349 | * otherwise. | |
3350 | */ | |
3351 | static MigIterateState bg_migration_iteration_run(MigrationState *s) | |
3352 | { | |
3353 | int res; | |
3354 | ||
3355 | res = qemu_savevm_state_iterate(s->to_dst_file, false); | |
3356 | if (res > 0) { | |
3357 | bg_migration_completion(s); | |
3358 | return MIG_ITERATE_BREAK; | |
3359 | } | |
3360 | ||
3361 | return MIG_ITERATE_RESUME; | |
3362 | } | |
3363 | ||
ad767bed DDAG |
3364 | void migration_make_urgent_request(void) |
3365 | { | |
3366 | qemu_sem_post(&migrate_get_current()->rate_limit_sem); | |
3367 | } | |
3368 | ||
3369 | void migration_consume_urgent_request(void) | |
3370 | { | |
3371 | qemu_sem_wait(&migrate_get_current()->rate_limit_sem); | |
3372 | } | |
3373 | ||
97e1e067 DDAG |
3374 | /* Returns true if the rate limiting was broken by an urgent request */ |
3375 | bool migration_rate_limit(void) | |
3376 | { | |
3377 | int64_t now = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); | |
3378 | MigrationState *s = migrate_get_current(); | |
3379 | ||
3380 | bool urgent = false; | |
3381 | migration_update_counters(s, now); | |
e1fde0e0 | 3382 | if (migration_rate_exceeded(s->to_dst_file)) { |
77386127 LS |
3383 | |
3384 | if (qemu_file_get_error(s->to_dst_file)) { | |
3385 | return false; | |
3386 | } | |
97e1e067 DDAG |
3387 | /* |
3388 | * Wait for a delay to do rate limiting OR | |
3389 | * something urgent to post the semaphore. | |
3390 | */ | |
3391 | int ms = s->iteration_start_time + BUFFER_DELAY - now; | |
3392 | trace_migration_rate_limit_pre(ms); | |
3393 | if (qemu_sem_timedwait(&s->rate_limit_sem, ms) == 0) { | |
3394 | /* | |
3395 | * We were woken by one or more urgent things but | |
3396 | * the timedwait will have consumed one of them. | |
3397 | * The service routine for the urgent wake will dec | |
3398 | * the semaphore itself for each item it consumes, | |
3399 | * so add this one we just eat back. | |
3400 | */ | |
3401 | qemu_sem_post(&s->rate_limit_sem); | |
3402 | urgent = true; | |
3403 | } | |
3404 | trace_migration_rate_limit_post(urgent); | |
3405 | } | |
3406 | return urgent; | |
3407 | } | |
3408 | ||
fde93d99 LV |
3409 | /* |
3410 | * if failover devices are present, wait they are completely | |
3411 | * unplugged | |
3412 | */ | |
3413 | ||
3414 | static void qemu_savevm_wait_unplug(MigrationState *s, int old_state, | |
3415 | int new_state) | |
3416 | { | |
3417 | if (qemu_savevm_state_guest_unplug_pending()) { | |
3418 | migrate_set_state(&s->state, old_state, MIGRATION_STATUS_WAIT_UNPLUG); | |
3419 | ||
3420 | while (s->state == MIGRATION_STATUS_WAIT_UNPLUG && | |
3421 | qemu_savevm_state_guest_unplug_pending()) { | |
3422 | qemu_sem_timedwait(&s->wait_unplug_sem, 250); | |
3423 | } | |
944bc528 LV |
3424 | if (s->state != MIGRATION_STATUS_WAIT_UNPLUG) { |
3425 | int timeout = 120; /* 30 seconds */ | |
3426 | /* | |
3427 | * migration has been canceled | |
3428 | * but as we have started an unplug we must wait the end | |
3429 | * to be able to plug back the card | |
3430 | */ | |
3431 | while (timeout-- && qemu_savevm_state_guest_unplug_pending()) { | |
3432 | qemu_sem_timedwait(&s->wait_unplug_sem, 250); | |
3433 | } | |
1b529d90 LV |
3434 | if (qemu_savevm_state_guest_unplug_pending() && |
3435 | !qtest_enabled()) { | |
a51dcef0 LV |
3436 | warn_report("migration: partially unplugged device on " |
3437 | "failure"); | |
3438 | } | |
944bc528 | 3439 | } |
fde93d99 LV |
3440 | |
3441 | migrate_set_state(&s->state, MIGRATION_STATUS_WAIT_UNPLUG, new_state); | |
3442 | } else { | |
3443 | migrate_set_state(&s->state, old_state, new_state); | |
3444 | } | |
3445 | } | |
3446 | ||
70b20477 DDAG |
3447 | /* |
3448 | * Master migration thread on the source VM. | |
3449 | * It drives the migration and pumps the data down the outgoing channel. | |
3450 | */ | |
5f496a1b | 3451 | static void *migration_thread(void *opaque) |
0d82d0e8 | 3452 | { |
9848a404 | 3453 | MigrationState *s = opaque; |
1b1f4ab6 | 3454 | MigrationThread *thread = NULL; |
bc72ad67 | 3455 | int64_t setup_start = qemu_clock_get_ms(QEMU_CLOCK_HOST); |
b23c2ade | 3456 | MigThrError thr_error; |
ad767bed | 3457 | bool urgent = false; |
057a2009 CLG |
3458 | Error *local_err = NULL; |
3459 | int ret; | |
76f5933a | 3460 | |
788fa680 | 3461 | thread = migration_threads_add("live_migration", qemu_get_thread_id()); |
1b1f4ab6 | 3462 | |
ab28bd23 PB |
3463 | rcu_register_thread(); |
3464 | ||
892ae715 | 3465 | object_ref(OBJECT(s)); |
87f3bd87 | 3466 | update_iteration_initial_status(s); |
b15df1ae | 3467 | |
dd904bc1 FR |
3468 | if (!multifd_send_setup()) { |
3469 | goto out; | |
3470 | } | |
3471 | ||
195801d7 | 3472 | bql_lock(); |
89a02a9f | 3473 | qemu_savevm_state_header(s->to_dst_file); |
195801d7 | 3474 | bql_unlock(); |
1d34e4bf | 3475 | |
62a02658 PX |
3476 | /* |
3477 | * If we opened the return path, we need to make sure dst has it | |
3478 | * opened as well. | |
3479 | */ | |
43044ac0 | 3480 | if (s->rp_state.rp_thread_created) { |
1d34e4bf | 3481 | /* Now tell the dest that it should open its end so it can reply */ |
89a02a9f | 3482 | qemu_savevm_send_open_return_path(s->to_dst_file); |
1d34e4bf DDAG |
3483 | |
3484 | /* And do a ping that will make stuff easier to debug */ | |
89a02a9f | 3485 | qemu_savevm_send_ping(s->to_dst_file, 1); |
0425dc97 | 3486 | } |
1d34e4bf | 3487 | |
58110f0a | 3488 | if (migrate_postcopy()) { |
1d34e4bf DDAG |
3489 | /* |
3490 | * Tell the destination that we *might* want to do postcopy later; | |
3491 | * if the other end can't do postcopy it should fail now, nice and | |
3492 | * early. | |
3493 | */ | |
89a02a9f | 3494 | qemu_savevm_send_postcopy_advise(s->to_dst_file); |
1d34e4bf DDAG |
3495 | } |
3496 | ||
5e804644 | 3497 | if (migrate_colo()) { |
aad555c2 ZC |
3498 | /* Notify migration destination that we enable COLO */ |
3499 | qemu_savevm_send_colo_enable(s->to_dst_file); | |
3500 | } | |
3501 | ||
195801d7 | 3502 | bql_lock(); |
057a2009 | 3503 | ret = qemu_savevm_state_setup(s->to_dst_file, &local_err); |
195801d7 | 3504 | bql_unlock(); |
0d82d0e8 | 3505 | |
fde93d99 LV |
3506 | qemu_savevm_wait_unplug(s, MIGRATION_STATUS_SETUP, |
3507 | MIGRATION_STATUS_ACTIVE); | |
c7e0acd5 | 3508 | |
057a2009 CLG |
3509 | /* |
3510 | * Handle SETUP failures after waiting for virtio-net-failover | |
3511 | * devices to unplug. This to preserve migration state transitions. | |
3512 | */ | |
3513 | if (ret) { | |
3514 | migrate_set_error(s, local_err); | |
3515 | error_free(local_err); | |
3516 | migrate_set_state(&s->state, MIGRATION_STATUS_ACTIVE, | |
3517 | MIGRATION_STATUS_FAILED); | |
3518 | goto out; | |
3519 | } | |
3520 | ||
bc72ad67 | 3521 | s->setup_time = qemu_clock_get_ms(QEMU_CLOCK_HOST) - setup_start; |
29ae8a41 | 3522 | |
9ec055ae DDAG |
3523 | trace_migration_thread_setup_complete(); |
3524 | ||
3a6813b6 | 3525 | while (migration_is_active()) { |
e1fde0e0 | 3526 | if (urgent || !migration_rate_exceeded(s->to_dst_file)) { |
2ad87305 PX |
3527 | MigIterateState iter_state = migration_iteration_run(s); |
3528 | if (iter_state == MIG_ITERATE_SKIP) { | |
3529 | continue; | |
3530 | } else if (iter_state == MIG_ITERATE_BREAK) { | |
09f6c85e | 3531 | break; |
c369f40d JQ |
3532 | } |
3533 | } | |
f4410a5d | 3534 | |
b23c2ade PX |
3535 | /* |
3536 | * Try to detect any kind of failures, and see whether we | |
3537 | * should stop the migration now. | |
3538 | */ | |
3539 | thr_error = migration_detect_error(s); | |
3540 | if (thr_error == MIG_THR_ERR_FATAL) { | |
3541 | /* Stop migration */ | |
fd45ee2c | 3542 | break; |
b23c2ade PX |
3543 | } else if (thr_error == MIG_THR_ERR_RECOVERED) { |
3544 | /* | |
3545 | * Just recovered from a e.g. network failure, reset all | |
3546 | * the local variables. This is important to avoid | |
3547 | * breaking transferred_bytes and bandwidth calculation | |
3548 | */ | |
87f3bd87 | 3549 | update_iteration_initial_status(s); |
fd45ee2c | 3550 | } |
b15df1ae | 3551 | |
97e1e067 | 3552 | urgent = migration_rate_limit(); |
a3fa1d78 PB |
3553 | } |
3554 | ||
dd904bc1 | 3555 | out: |
1d34e4bf | 3556 | trace_migration_thread_after_loop(); |
199aa6d4 | 3557 | migration_iteration_finish(s); |
892ae715 | 3558 | object_unref(OBJECT(s)); |
ab28bd23 | 3559 | rcu_unregister_thread(); |
788fa680 | 3560 | migration_threads_remove(thread); |
0d82d0e8 JQ |
3561 | return NULL; |
3562 | } | |
3563 | ||
8518278a AG |
3564 | static void bg_migration_vm_start_bh(void *opaque) |
3565 | { | |
3566 | MigrationState *s = opaque; | |
3567 | ||
49a50206 | 3568 | vm_resume(s->vm_old_state); |
e22ffad0 | 3569 | migration_downtime_end(s); |
8518278a AG |
3570 | } |
3571 | ||
3572 | /** | |
3573 | * Background snapshot thread, based on live migration code. | |
3574 | * This is an alternative implementation of live migration mechanism | |
3575 | * introduced specifically to support background snapshots. | |
3576 | * | |
3577 | * It takes advantage of userfault_fd write protection mechanism introduced | |
3578 | * in v5.7 kernel. Compared to existing dirty page logging migration much | |
3579 | * lesser stream traffic is produced resulting in smaller snapshot images, | |
3580 | * simply cause of no page duplicates can get into the stream. | |
3581 | * | |
3582 | * Another key point is that generated vmstate stream reflects machine state | |
3583 | * 'frozen' at the beginning of snapshot creation compared to dirty page logging | |
3584 | * mechanism, which effectively results in that saved snapshot is the state of VM | |
3585 | * at the end of the process. | |
3586 | */ | |
3587 | static void *bg_migration_thread(void *opaque) | |
3588 | { | |
3589 | MigrationState *s = opaque; | |
3590 | int64_t setup_start; | |
3591 | MigThrError thr_error; | |
3592 | QEMUFile *fb; | |
3593 | bool early_fail = true; | |
057a2009 CLG |
3594 | Error *local_err = NULL; |
3595 | int ret; | |
8518278a AG |
3596 | |
3597 | rcu_register_thread(); | |
3598 | object_ref(OBJECT(s)); | |
3599 | ||
e1fde0e0 | 3600 | migration_rate_set(RATE_LIMIT_DISABLED); |
8518278a AG |
3601 | |
3602 | setup_start = qemu_clock_get_ms(QEMU_CLOCK_HOST); | |
3603 | /* | |
3604 | * We want to save vmstate for the moment when migration has been | |
3605 | * initiated but also we want to save RAM content while VM is running. | |
3606 | * The RAM content should appear first in the vmstate. So, we first | |
3607 | * stash the non-RAM part of the vmstate to the temporary buffer, | |
3608 | * then write RAM part of the vmstate to the migration stream | |
3609 | * with vCPUs running and, finally, write stashed non-RAM part of | |
3610 | * the vmstate from the buffer to the migration stream. | |
3611 | */ | |
ecb23efe | 3612 | s->bioc = qio_channel_buffer_new(512 * 1024); |
8518278a | 3613 | qio_channel_set_name(QIO_CHANNEL(s->bioc), "vmstate-buffer"); |
77ef2dc1 | 3614 | fb = qemu_file_new_output(QIO_CHANNEL(s->bioc)); |
8518278a AG |
3615 | object_unref(OBJECT(s->bioc)); |
3616 | ||
3617 | update_iteration_initial_status(s); | |
3618 | ||
eeccb99c AG |
3619 | /* |
3620 | * Prepare for tracking memory writes with UFFD-WP - populate | |
3621 | * RAM pages before protecting. | |
3622 | */ | |
3623 | #ifdef __linux__ | |
3624 | ram_write_tracking_prepare(); | |
3625 | #endif | |
3626 | ||
195801d7 | 3627 | bql_lock(); |
8518278a | 3628 | qemu_savevm_state_header(s->to_dst_file); |
057a2009 | 3629 | ret = qemu_savevm_state_setup(s->to_dst_file, &local_err); |
195801d7 | 3630 | bql_unlock(); |
8518278a | 3631 | |
fde93d99 LV |
3632 | qemu_savevm_wait_unplug(s, MIGRATION_STATUS_SETUP, |
3633 | MIGRATION_STATUS_ACTIVE); | |
8518278a | 3634 | |
057a2009 CLG |
3635 | /* |
3636 | * Handle SETUP failures after waiting for virtio-net-failover | |
3637 | * devices to unplug. This to preserve migration state transitions. | |
3638 | */ | |
3639 | if (ret) { | |
3640 | migrate_set_error(s, local_err); | |
3641 | error_free(local_err); | |
3642 | migrate_set_state(&s->state, MIGRATION_STATUS_ACTIVE, | |
3643 | MIGRATION_STATUS_FAILED); | |
3644 | goto fail_setup; | |
3645 | } | |
3646 | ||
8518278a AG |
3647 | s->setup_time = qemu_clock_get_ms(QEMU_CLOCK_HOST) - setup_start; |
3648 | ||
3649 | trace_migration_thread_setup_complete(); | |
8518278a | 3650 | |
195801d7 | 3651 | bql_lock(); |
8518278a | 3652 | |
9867d4dd | 3653 | if (migration_stop_vm(s, RUN_STATE_PAUSED)) { |
8518278a AG |
3654 | goto fail; |
3655 | } | |
3656 | /* | |
3657 | * Put vCPUs in sync with shadow context structures, then | |
3658 | * save their state to channel-buffer along with devices. | |
3659 | */ | |
3660 | cpu_synchronize_all_states(); | |
3661 | if (qemu_savevm_state_complete_precopy_non_iterable(fb, false, false)) { | |
3662 | goto fail; | |
3663 | } | |
ecb23efe AG |
3664 | /* |
3665 | * Since we are going to get non-iterable state data directly | |
3666 | * from s->bioc->data, explicit flush is needed here. | |
3667 | */ | |
3668 | qemu_fflush(fb); | |
3669 | ||
8518278a AG |
3670 | /* Now initialize UFFD context and start tracking RAM writes */ |
3671 | if (ram_write_tracking_start()) { | |
3672 | goto fail; | |
3673 | } | |
3674 | early_fail = false; | |
3675 | ||
3676 | /* | |
3677 | * Start VM from BH handler to avoid write-fault lock here. | |
3678 | * UFFD-WP protection for the whole RAM is already enabled so | |
3679 | * calling VM state change notifiers from vm_start() would initiate | |
3680 | * writes to virtio VQs memory which is in write-protected region. | |
3681 | */ | |
44d0d456 | 3682 | migration_bh_schedule(bg_migration_vm_start_bh, s); |
195801d7 | 3683 | bql_unlock(); |
8518278a | 3684 | |
3a6813b6 | 3685 | while (migration_is_active()) { |
8518278a AG |
3686 | MigIterateState iter_state = bg_migration_iteration_run(s); |
3687 | if (iter_state == MIG_ITERATE_SKIP) { | |
3688 | continue; | |
3689 | } else if (iter_state == MIG_ITERATE_BREAK) { | |
3690 | break; | |
3691 | } | |
3692 | ||
3693 | /* | |
3694 | * Try to detect any kind of failures, and see whether we | |
3695 | * should stop the migration now. | |
3696 | */ | |
3697 | thr_error = migration_detect_error(s); | |
3698 | if (thr_error == MIG_THR_ERR_FATAL) { | |
3699 | /* Stop migration */ | |
3700 | break; | |
3701 | } | |
3702 | ||
3703 | migration_update_counters(s, qemu_clock_get_ms(QEMU_CLOCK_REALTIME)); | |
3704 | } | |
3705 | ||
3706 | trace_migration_thread_after_loop(); | |
3707 | ||
3708 | fail: | |
3709 | if (early_fail) { | |
3710 | migrate_set_state(&s->state, MIGRATION_STATUS_ACTIVE, | |
3711 | MIGRATION_STATUS_FAILED); | |
195801d7 | 3712 | bql_unlock(); |
8518278a AG |
3713 | } |
3714 | ||
057a2009 | 3715 | fail_setup: |
8518278a AG |
3716 | bg_migration_iteration_finish(s); |
3717 | ||
3718 | qemu_fclose(fb); | |
3719 | object_unref(OBJECT(s)); | |
3720 | rcu_unregister_thread(); | |
3721 | ||
3722 | return NULL; | |
3723 | } | |
3724 | ||
cce8040b | 3725 | void migrate_fd_connect(MigrationState *s, Error *error_in) |
0d82d0e8 | 3726 | { |
00f4b572 | 3727 | Error *local_err = NULL; |
52033349 | 3728 | uint64_t rate_limit; |
d3e35b8f | 3729 | bool resume = s->state == MIGRATION_STATUS_POSTCOPY_PAUSED; |
9867d4dd | 3730 | int ret; |
d3e35b8f | 3731 | |
ca7bd082 PX |
3732 | /* |
3733 | * If there's a previous error, free it and prepare for another one. | |
3734 | * Meanwhile if migration completes successfully, there won't have an error | |
3735 | * dumped when calling migrate_fd_cleanup(). | |
3736 | */ | |
3737 | migrate_error_free(s); | |
3738 | ||
f5da8ba4 | 3739 | s->expected_downtime = migrate_downtime_limit(); |
cce8040b DDAG |
3740 | if (error_in) { |
3741 | migrate_fd_error(s, error_in); | |
ca30f24d PX |
3742 | if (resume) { |
3743 | /* | |
3744 | * Don't do cleanup for resume if channel is invalid, but only dump | |
3745 | * the error. We wait for another channel connect from the user. | |
3746 | * The error_report still gives HMP user a hint on what failed. | |
3747 | * It's normally done in migrate_fd_cleanup(), but call it here | |
3748 | * explicitly. | |
3749 | */ | |
3750 | error_report_err(error_copy(s->error)); | |
3751 | } else { | |
3752 | migrate_fd_cleanup(s); | |
3753 | } | |
cce8040b DDAG |
3754 | return; |
3755 | } | |
0d82d0e8 | 3756 | |
d3e35b8f PX |
3757 | if (resume) { |
3758 | /* This is a resumed migration */ | |
9d3ebbe2 | 3759 | rate_limit = migrate_max_postcopy_bandwidth(); |
d3e35b8f PX |
3760 | } else { |
3761 | /* This is a fresh new migration */ | |
9d3ebbe2 | 3762 | rate_limit = migrate_max_bandwidth(); |
442773ce | 3763 | |
d3e35b8f | 3764 | /* Notify before starting migration thread */ |
4af667f8 SS |
3765 | if (migration_call_notifiers(s, MIG_EVENT_PRECOPY_SETUP, &local_err)) { |
3766 | goto fail; | |
3767 | } | |
d3e35b8f PX |
3768 | } |
3769 | ||
e1fde0e0 | 3770 | migration_rate_set(rate_limit); |
d3e35b8f | 3771 | qemu_file_set_blocking(s->to_dst_file, true); |
9287ac27 | 3772 | |
1d34e4bf | 3773 | /* |
c788ada8 PX |
3774 | * Open the return path. For postcopy, it is used exclusively. For |
3775 | * precopy, only if user specified "return-path" capability would | |
3776 | * QEMU uses the return path. | |
1d34e4bf | 3777 | */ |
38ad1110 | 3778 | if (migrate_postcopy_ram() || migrate_return_path()) { |
ef796ee9 | 3779 | if (open_return_path_on_source(s)) { |
908927db | 3780 | error_setg(&local_err, "Unable to open return-path for postcopy"); |
bf78a046 | 3781 | goto fail; |
1d34e4bf DDAG |
3782 | } |
3783 | } | |
3784 | ||
06064a67 PX |
3785 | /* |
3786 | * This needs to be done before resuming a postcopy. Note: for newer | |
3787 | * QEMUs we will delay the channel creation until postcopy_start(), to | |
3788 | * avoid disorder of channel creations. | |
3789 | */ | |
3790 | if (migrate_postcopy_preempt() && s->preempt_pre_7_2) { | |
3791 | postcopy_preempt_setup(s); | |
3792 | } | |
3793 | ||
d3e35b8f | 3794 | if (resume) { |
135b87b4 PX |
3795 | /* Wakeup the main migration thread to do the recovery */ |
3796 | migrate_set_state(&s->state, MIGRATION_STATUS_POSTCOPY_PAUSED, | |
3797 | MIGRATION_STATUS_POSTCOPY_RECOVER); | |
3798 | qemu_sem_post(&s->postcopy_pause_sem); | |
d3e35b8f PX |
3799 | return; |
3800 | } | |
3801 | ||
9867d4dd SS |
3802 | if (migrate_mode_is_cpr(s)) { |
3803 | ret = migration_stop_vm(s, RUN_STATE_FINISH_MIGRATE); | |
3804 | if (ret < 0) { | |
3805 | error_setg(&local_err, "migration_stop_vm failed, error %d", -ret); | |
3806 | goto fail; | |
3807 | } | |
3808 | } | |
3809 | ||
8518278a AG |
3810 | if (migrate_background_snapshot()) { |
3811 | qemu_thread_create(&s->thread, "bg_snapshot", | |
3812 | bg_migration_thread, s, QEMU_THREAD_JOINABLE); | |
3813 | } else { | |
3814 | qemu_thread_create(&s->thread, "live_migration", | |
3815 | migration_thread, s, QEMU_THREAD_JOINABLE); | |
3816 | } | |
1d34e4bf | 3817 | s->migration_thread_running = true; |
bf78a046 SS |
3818 | return; |
3819 | ||
3820 | fail: | |
3821 | migrate_set_error(s, local_err); | |
3822 | migrate_set_state(&s->state, s->state, MIGRATION_STATUS_FAILED); | |
3823 | error_report_err(local_err); | |
3824 | migrate_fd_cleanup(s); | |
0d82d0e8 | 3825 | } |
093e3c42 | 3826 | |
e5cb7e76 PX |
3827 | static void migration_class_init(ObjectClass *klass, void *data) |
3828 | { | |
3829 | DeviceClass *dc = DEVICE_CLASS(klass); | |
3830 | ||
3831 | dc->user_creatable = false; | |
4f67d30b | 3832 | device_class_set_props(dc, migration_properties); |
e5cb7e76 PX |
3833 | } |
3834 | ||
b91bf5e4 MAL |
3835 | static void migration_instance_finalize(Object *obj) |
3836 | { | |
3837 | MigrationState *ms = MIGRATION_OBJ(obj); | |
b91bf5e4 | 3838 | |
87db1a7d | 3839 | qemu_mutex_destroy(&ms->error_mutex); |
62df066f | 3840 | qemu_mutex_destroy(&ms->qemu_file_lock); |
c7e0acd5 | 3841 | qemu_sem_destroy(&ms->wait_unplug_sem); |
ad767bed | 3842 | qemu_sem_destroy(&ms->rate_limit_sem); |
e91d8951 | 3843 | qemu_sem_destroy(&ms->pause_sem); |
b23c2ade | 3844 | qemu_sem_destroy(&ms->postcopy_pause_sem); |
edd090c7 | 3845 | qemu_sem_destroy(&ms->rp_state.rp_sem); |
b28fb582 | 3846 | qemu_sem_destroy(&ms->rp_state.rp_pong_acks); |
d0edb8a1 | 3847 | qemu_sem_destroy(&ms->postcopy_qemufile_src_sem); |
ab105cc1 | 3848 | error_free(ms->error); |
b91bf5e4 MAL |
3849 | } |
3850 | ||
e5cb7e76 PX |
3851 | static void migration_instance_init(Object *obj) |
3852 | { | |
3853 | MigrationState *ms = MIGRATION_OBJ(obj); | |
3854 | ||
3855 | ms->state = MIGRATION_STATUS_NONE; | |
e5cb7e76 | 3856 | ms->mbps = -1; |
aecbfe9c | 3857 | ms->pages_per_second = -1; |
e91d8951 | 3858 | qemu_sem_init(&ms->pause_sem, 0); |
87db1a7d | 3859 | qemu_mutex_init(&ms->error_mutex); |
8b0b29dc | 3860 | |
61a174e2 | 3861 | migrate_params_init(&ms->parameters); |
b23c2ade PX |
3862 | |
3863 | qemu_sem_init(&ms->postcopy_pause_sem, 0); | |
edd090c7 | 3864 | qemu_sem_init(&ms->rp_state.rp_sem, 0); |
b28fb582 | 3865 | qemu_sem_init(&ms->rp_state.rp_pong_acks, 0); |
ad767bed | 3866 | qemu_sem_init(&ms->rate_limit_sem, 0); |
c7e0acd5 | 3867 | qemu_sem_init(&ms->wait_unplug_sem, 0); |
d0edb8a1 | 3868 | qemu_sem_init(&ms->postcopy_qemufile_src_sem, 0); |
62df066f | 3869 | qemu_mutex_init(&ms->qemu_file_lock); |
8b0b29dc PX |
3870 | } |
3871 | ||
3872 | /* | |
3873 | * Return true if check pass, false otherwise. Error will be put | |
3874 | * inside errp if provided. | |
3875 | */ | |
3876 | static bool migration_object_check(MigrationState *ms, Error **errp) | |
3877 | { | |
6b19a7d9 | 3878 | /* Assuming all off */ |
b02c7fc9 | 3879 | bool old_caps[MIGRATION_CAPABILITY__MAX] = { 0 }; |
6b19a7d9 | 3880 | |
8b0b29dc PX |
3881 | if (!migrate_params_check(&ms->parameters, errp)) { |
3882 | return false; | |
3883 | } | |
3884 | ||
b02c7fc9 | 3885 | return migrate_caps_check(old_caps, ms->capabilities, errp); |
e5cb7e76 PX |
3886 | } |
3887 | ||
3888 | static const TypeInfo migration_type = { | |
3889 | .name = TYPE_MIGRATION, | |
01f6e14c | 3890 | /* |
c8d3ff38 | 3891 | * NOTE: TYPE_MIGRATION is not really a device, as the object is |
2194abd6 | 3892 | * not created using qdev_new(), it is not attached to the qdev |
c8d3ff38 PX |
3893 | * device tree, and it is never realized. |
3894 | * | |
3895 | * TODO: Make this TYPE_OBJECT once QOM provides something like | |
3896 | * TYPE_DEVICE's "-global" properties. | |
01f6e14c | 3897 | */ |
e5cb7e76 PX |
3898 | .parent = TYPE_DEVICE, |
3899 | .class_init = migration_class_init, | |
3900 | .class_size = sizeof(MigrationClass), | |
3901 | .instance_size = sizeof(MigrationState), | |
3902 | .instance_init = migration_instance_init, | |
b91bf5e4 | 3903 | .instance_finalize = migration_instance_finalize, |
e5cb7e76 PX |
3904 | }; |
3905 | ||
3906 | static void register_migration_types(void) | |
3907 | { | |
3908 | type_register_static(&migration_type); | |
3909 | } | |
3910 | ||
3911 | type_init(register_migration_types); |