]>
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 | ||
75 | static NotifierWithReturnList migration_state_notifiers[] = { | |
76 | NOTIFIER_ELEM_INIT(migration_state_notifiers, MIG_MODE_NORMAL), | |
77 | NOTIFIER_ELEM_INIT(migration_state_notifiers, MIG_MODE_CPR_REBOOT), | |
78 | }; | |
99a0db9b | 79 | |
da6f1790 JQ |
80 | /* Messages sent on the return path from destination to source */ |
81 | enum mig_rp_message_type { | |
82 | MIG_RP_MSG_INVALID = 0, /* Must be 0 */ | |
83 | MIG_RP_MSG_SHUT, /* sibling will not send any more RP messages */ | |
84 | MIG_RP_MSG_PONG, /* Response to a PING; data (seq: be32 ) */ | |
85 | ||
86 | MIG_RP_MSG_REQ_PAGES_ID, /* data (start: be64, len: be32, id: string) */ | |
87 | MIG_RP_MSG_REQ_PAGES, /* data (start: be64, len: be32) */ | |
a335debb | 88 | MIG_RP_MSG_RECV_BITMAP, /* send recved_bitmap back to source */ |
13955b89 | 89 | MIG_RP_MSG_RESUME_ACK, /* tell source that we are ready to resume */ |
1b4adb10 | 90 | MIG_RP_MSG_SWITCHOVER_ACK, /* Tell source it's OK to do switchover */ |
da6f1790 JQ |
91 | |
92 | MIG_RP_MSG_MAX | |
93 | }; | |
94 | ||
17549e84 JQ |
95 | /* When we add fault tolerance, we could have several |
96 | migrations at once. For now we don't need to add | |
97 | dynamic creation of migration */ | |
98 | ||
e5cb7e76 | 99 | static MigrationState *current_migration; |
e1b1b1bc | 100 | static MigrationIncomingState *current_incoming; |
e5cb7e76 | 101 | |
fa3673e4 | 102 | static GSList *migration_blockers[MIG_MODE__MAX]; |
3af8554b | 103 | |
8b0b29dc | 104 | static bool migration_object_check(MigrationState *ms, Error **errp); |
0331c8ca DDAG |
105 | static int migration_maybe_pause(MigrationState *s, |
106 | int *current_active_state, | |
107 | int new_state); | |
892ae715 | 108 | static void migrate_fd_cancel(MigrationState *s); |
7aa6070d | 109 | static bool close_return_path_on_source(MigrationState *s); |
63f64d77 | 110 | static void migration_completion_end(MigrationState *s); |
8b0b29dc | 111 | |
e22ffad0 PX |
112 | static void migration_downtime_start(MigrationState *s) |
113 | { | |
3e5f3bcd | 114 | trace_vmstate_downtime_checkpoint("src-downtime-start"); |
e22ffad0 PX |
115 | s->downtime_start = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); |
116 | } | |
117 | ||
118 | static void migration_downtime_end(MigrationState *s) | |
119 | { | |
120 | int64_t now = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); | |
121 | ||
122 | /* | |
123 | * If downtime already set, should mean that postcopy already set it, | |
124 | * then that should be the real downtime already. | |
125 | */ | |
126 | if (!s->downtime) { | |
127 | s->downtime = now - s->downtime_start; | |
128 | } | |
3e5f3bcd PX |
129 | |
130 | trace_vmstate_downtime_checkpoint("src-downtime-end"); | |
e22ffad0 PX |
131 | } |
132 | ||
d6f74fd1 PX |
133 | static bool migration_needs_multiple_sockets(void) |
134 | { | |
51b07548 | 135 | return migrate_multifd() || migrate_postcopy_preempt(); |
d6f74fd1 | 136 | } |
f444eeda | 137 | |
3205bebd | 138 | static bool transport_supports_multi_channels(MigrationAddress *addr) |
f444eeda | 139 | { |
3205bebd AH |
140 | if (addr->transport == MIGRATION_ADDRESS_TYPE_SOCKET) { |
141 | SocketAddress *saddr = &addr->u.socket; | |
142 | ||
decdc767 FR |
143 | if (saddr->type == SOCKET_ADDRESS_TYPE_FD) { |
144 | return migrate_mapped_ram(); | |
145 | } | |
146 | ||
f427d90b FR |
147 | return (saddr->type == SOCKET_ADDRESS_TYPE_INET || |
148 | saddr->type == SOCKET_ADDRESS_TYPE_UNIX || | |
149 | saddr->type == SOCKET_ADDRESS_TYPE_VSOCK); | |
150 | } else if (addr->transport == MIGRATION_ADDRESS_TYPE_FILE) { | |
151 | return migrate_mapped_ram(); | |
152 | } else { | |
153 | return false; | |
3205bebd | 154 | } |
f444eeda PX |
155 | } |
156 | ||
8d9e0d41 FR |
157 | static bool migration_needs_seekable_channel(void) |
158 | { | |
159 | return migrate_mapped_ram(); | |
160 | } | |
161 | ||
162 | static bool transport_supports_seeking(MigrationAddress *addr) | |
163 | { | |
164 | if (addr->transport == MIGRATION_ADDRESS_TYPE_FILE) { | |
165 | return true; | |
166 | } | |
167 | ||
168 | /* | |
169 | * At this point, the user might not yet have passed the file | |
170 | * descriptor to QEMU, so we cannot know for sure whether it | |
171 | * refers to a plain file or a socket. Let it through anyway. | |
172 | */ | |
173 | if (addr->transport == MIGRATION_ADDRESS_TYPE_SOCKET) { | |
174 | return addr->u.socket.type == SOCKET_ADDRESS_TYPE_FD; | |
175 | } | |
176 | ||
177 | return false; | |
178 | } | |
179 | ||
d6f74fd1 | 180 | static bool |
d95533e1 HG |
181 | migration_channels_and_transport_compatible(MigrationAddress *addr, |
182 | Error **errp) | |
f444eeda | 183 | { |
8d9e0d41 FR |
184 | if (migration_needs_seekable_channel() && |
185 | !transport_supports_seeking(addr)) { | |
186 | error_setg(errp, "Migration requires seekable transport (e.g. file)"); | |
187 | return false; | |
188 | } | |
189 | ||
d6f74fd1 | 190 | if (migration_needs_multiple_sockets() && |
3205bebd | 191 | !transport_supports_multi_channels(addr)) { |
d6f74fd1 PX |
192 | error_setg(errp, "Migration requires multi-channel URIs (e.g. tcp)"); |
193 | return false; | |
194 | } | |
195 | ||
196 | return true; | |
f444eeda PX |
197 | } |
198 | ||
8f8bfffc PX |
199 | static gint page_request_addr_cmp(gconstpointer ap, gconstpointer bp) |
200 | { | |
201 | uintptr_t a = (uintptr_t) ap, b = (uintptr_t) bp; | |
202 | ||
203 | return (a > b) - (a < b); | |
204 | } | |
205 | ||
9867d4dd | 206 | static int migration_stop_vm(MigrationState *s, RunState state) |
93bdf888 | 207 | { |
9867d4dd SS |
208 | int ret; |
209 | ||
210 | migration_downtime_start(s); | |
211 | ||
212 | s->vm_old_state = runstate_get(); | |
213 | global_state_store(); | |
214 | ||
215 | ret = vm_stop_force_state(state); | |
3e5f3bcd PX |
216 | |
217 | trace_vmstate_downtime_checkpoint("src-vm-stopped"); | |
9867d4dd | 218 | trace_migration_completion_vm_stop(ret); |
3e5f3bcd PX |
219 | |
220 | return ret; | |
93bdf888 PX |
221 | } |
222 | ||
e5cb7e76 PX |
223 | void migration_object_init(void) |
224 | { | |
225 | /* This can only be called once. */ | |
226 | assert(!current_migration); | |
227 | current_migration = MIGRATION_OBJ(object_new(TYPE_MIGRATION)); | |
4ffdb337 | 228 | |
e1b1b1bc PX |
229 | /* |
230 | * Init the migrate incoming object as well no matter whether | |
231 | * we'll use it or not. | |
232 | */ | |
233 | assert(!current_incoming); | |
234 | current_incoming = g_new0(MigrationIncomingState, 1); | |
235 | current_incoming->state = MIGRATION_STATUS_NONE; | |
236 | current_incoming->postcopy_remote_fds = | |
237 | g_array_new(FALSE, TRUE, sizeof(struct PostCopyFD)); | |
238 | qemu_mutex_init(¤t_incoming->rp_mutex); | |
60bb3c58 | 239 | qemu_mutex_init(¤t_incoming->postcopy_prio_thread_mutex); |
e1b1b1bc PX |
240 | qemu_event_init(¤t_incoming->main_thread_load_event, false); |
241 | qemu_sem_init(¤t_incoming->postcopy_pause_sem_dst, 0); | |
242 | qemu_sem_init(¤t_incoming->postcopy_pause_sem_fault, 0); | |
60bb3c58 | 243 | qemu_sem_init(¤t_incoming->postcopy_pause_sem_fast_load, 0); |
5655aab0 PX |
244 | qemu_sem_init(¤t_incoming->postcopy_qemufile_dst_done, 0); |
245 | ||
8f8bfffc | 246 | qemu_mutex_init(¤t_incoming->page_request_mutex); |
cf02f29e | 247 | qemu_cond_init(¤t_incoming->page_request_cond); |
8f8bfffc | 248 | current_incoming->page_requested = g_tree_new(page_request_addr_cmp); |
e1b1b1bc | 249 | |
f9734d5d | 250 | migration_object_check(current_migration, &error_fatal); |
e0d17dfd PB |
251 | |
252 | blk_mig_init(); | |
253 | ram_mig_init(); | |
254 | dirty_bitmap_mig_init(); | |
e5cb7e76 PX |
255 | } |
256 | ||
44d0d456 FR |
257 | typedef struct { |
258 | QEMUBH *bh; | |
259 | QEMUBHFunc *cb; | |
260 | void *opaque; | |
261 | } MigrationBH; | |
262 | ||
263 | static void migration_bh_dispatch_bh(void *opaque) | |
699d9476 | 264 | { |
44d0d456 FR |
265 | MigrationState *s = migrate_get_current(); |
266 | MigrationBH *migbh = opaque; | |
267 | ||
268 | /* cleanup this BH */ | |
269 | qemu_bh_delete(migbh->bh); | |
270 | migbh->bh = NULL; | |
271 | ||
272 | /* dispatch the other one */ | |
273 | migbh->cb(migbh->opaque); | |
274 | object_unref(OBJECT(s)); | |
275 | ||
276 | g_free(migbh); | |
277 | } | |
278 | ||
279 | void migration_bh_schedule(QEMUBHFunc *cb, void *opaque) | |
280 | { | |
281 | MigrationState *s = migrate_get_current(); | |
282 | MigrationBH *migbh = g_new0(MigrationBH, 1); | |
283 | QEMUBH *bh = qemu_bh_new(migration_bh_dispatch_bh, migbh); | |
284 | ||
285 | /* Store these to dispatch when the BH runs */ | |
286 | migbh->bh = bh; | |
287 | migbh->cb = cb; | |
288 | migbh->opaque = opaque; | |
289 | ||
699d9476 FR |
290 | /* |
291 | * Ref the state for bh, because it may be called when | |
292 | * there're already no other refs | |
293 | */ | |
294 | object_ref(OBJECT(s)); | |
295 | qemu_bh_schedule(bh); | |
296 | } | |
297 | ||
458fecca | 298 | void migration_cancel(const Error *error) |
c7c0e724 | 299 | { |
458fecca LV |
300 | if (error) { |
301 | migrate_set_error(current_migration, error); | |
302 | } | |
acac51ba HH |
303 | if (migrate_dirty_limit()) { |
304 | qmp_cancel_vcpu_dirty_limit(false, -1, NULL); | |
305 | } | |
c7c0e724 DH |
306 | migrate_fd_cancel(current_migration); |
307 | } | |
308 | ||
892ae715 | 309 | void migration_shutdown(void) |
1f895604 | 310 | { |
795969ab RL |
311 | /* |
312 | * When the QEMU main thread exit, the COLO thread | |
313 | * may wait a semaphore. So, we should wakeup the | |
314 | * COLO thread before migration shutdown. | |
315 | */ | |
316 | colo_shutdown(); | |
892ae715 DDAG |
317 | /* |
318 | * Cancel the current migration - that will (eventually) | |
319 | * stop the migration using this structure | |
320 | */ | |
458fecca | 321 | migration_cancel(NULL); |
1f895604 | 322 | object_unref(OBJECT(current_migration)); |
1499ab09 VSO |
323 | |
324 | /* | |
325 | * Cancel outgoing migration of dirty bitmaps. It should | |
326 | * at least unref used block nodes. | |
327 | */ | |
328 | dirty_bitmap_mig_cancel_outgoing(); | |
329 | ||
330 | /* | |
331 | * Cancel incoming migration of dirty bitmaps. Dirty bitmaps | |
332 | * are non-critical data, and their loss never considered as | |
333 | * something serious. | |
334 | */ | |
335 | dirty_bitmap_mig_cancel_incoming(); | |
1f895604 VSO |
336 | } |
337 | ||
bca7856a | 338 | /* For outgoing */ |
859bc756 | 339 | MigrationState *migrate_get_current(void) |
17549e84 | 340 | { |
e5cb7e76 PX |
341 | /* This can only be called after the object created. */ |
342 | assert(current_migration); | |
343 | return current_migration; | |
17549e84 JQ |
344 | } |
345 | ||
bca7856a DDAG |
346 | MigrationIncomingState *migration_incoming_get_current(void) |
347 | { | |
e1b1b1bc PX |
348 | assert(current_incoming); |
349 | return current_incoming; | |
bca7856a DDAG |
350 | } |
351 | ||
e031149c PX |
352 | void migration_incoming_transport_cleanup(MigrationIncomingState *mis) |
353 | { | |
354 | if (mis->socket_address_list) { | |
355 | qapi_free_SocketAddressList(mis->socket_address_list); | |
356 | mis->socket_address_list = NULL; | |
357 | } | |
358 | ||
359 | if (mis->transport_cleanup) { | |
360 | mis->transport_cleanup(mis->transport_data); | |
361 | mis->transport_data = mis->transport_cleanup = NULL; | |
362 | } | |
363 | } | |
364 | ||
bca7856a DDAG |
365 | void migration_incoming_state_destroy(void) |
366 | { | |
b4b076da JQ |
367 | struct MigrationIncomingState *mis = migration_incoming_get_current(); |
368 | ||
cde85c37 | 369 | multifd_recv_cleanup(); |
c323518a | 370 | compress_threads_load_cleanup(); |
cfc3bcf3 | 371 | |
3482655b | 372 | if (mis->to_src_file) { |
660819b1 PX |
373 | /* Tell source that we are done */ |
374 | migrate_send_rp_shut(mis, qemu_file_get_error(mis->from_src_file) != 0); | |
3482655b PX |
375 | qemu_fclose(mis->to_src_file); |
376 | mis->to_src_file = NULL; | |
377 | } | |
378 | ||
660819b1 | 379 | if (mis->from_src_file) { |
39675fff | 380 | migration_ioc_unregister_yank_from_file(mis->from_src_file); |
660819b1 PX |
381 | qemu_fclose(mis->from_src_file); |
382 | mis->from_src_file = NULL; | |
383 | } | |
00fa4fc8 DDAG |
384 | if (mis->postcopy_remote_fds) { |
385 | g_array_free(mis->postcopy_remote_fds, TRUE); | |
386 | mis->postcopy_remote_fds = NULL; | |
387 | } | |
660819b1 | 388 | |
e031149c | 389 | migration_incoming_transport_cleanup(mis); |
1783c00f DDAG |
390 | qemu_event_reset(&mis->main_thread_load_event); |
391 | ||
8f8bfffc PX |
392 | if (mis->page_requested) { |
393 | g_tree_destroy(mis->page_requested); | |
394 | mis->page_requested = NULL; | |
395 | } | |
396 | ||
36f62f11 PX |
397 | if (mis->postcopy_qemufile_dst) { |
398 | migration_ioc_unregister_yank_from_file(mis->postcopy_qemufile_dst); | |
399 | qemu_fclose(mis->postcopy_qemufile_dst); | |
400 | mis->postcopy_qemufile_dst = NULL; | |
401 | } | |
402 | ||
b5eea99e | 403 | yank_unregister_instance(MIGRATION_YANK_INSTANCE); |
bca7856a DDAG |
404 | } |
405 | ||
b05dc723 JQ |
406 | static void migrate_generate_event(int new_state) |
407 | { | |
b890902c | 408 | if (migrate_events()) { |
3ab72385 | 409 | qapi_event_send_migration(new_state); |
b05dc723 JQ |
410 | } |
411 | } | |
412 | ||
da6f1790 JQ |
413 | /* |
414 | * Send a message on the return channel back to the source | |
415 | * of the migration. | |
416 | */ | |
d6208e35 PX |
417 | static int migrate_send_rp_message(MigrationIncomingState *mis, |
418 | enum mig_rp_message_type message_type, | |
419 | uint16_t len, void *data) | |
da6f1790 | 420 | { |
d6208e35 PX |
421 | int ret = 0; |
422 | ||
da6f1790 | 423 | trace_migrate_send_rp_message((int)message_type, len); |
37396950 | 424 | QEMU_LOCK_GUARD(&mis->rp_mutex); |
d6208e35 PX |
425 | |
426 | /* | |
427 | * It's possible that the file handle got lost due to network | |
428 | * failures. | |
429 | */ | |
430 | if (!mis->to_src_file) { | |
431 | ret = -EIO; | |
37396950 | 432 | return ret; |
d6208e35 PX |
433 | } |
434 | ||
da6f1790 JQ |
435 | qemu_put_be16(mis->to_src_file, (unsigned int)message_type); |
436 | qemu_put_be16(mis->to_src_file, len); | |
437 | qemu_put_buffer(mis->to_src_file, data, len); | |
be07a0ed | 438 | return qemu_fflush(mis->to_src_file); |
da6f1790 JQ |
439 | } |
440 | ||
2e2bce16 PX |
441 | /* Request one page from the source VM at the given start address. |
442 | * rb: the RAMBlock to request the page in | |
1e2d90eb DDAG |
443 | * Start: Address offset within the RB |
444 | * Len: Length in bytes required - must be a multiple of pagesize | |
445 | */ | |
7a267fc4 PX |
446 | int migrate_send_rp_message_req_pages(MigrationIncomingState *mis, |
447 | RAMBlock *rb, ram_addr_t start) | |
1e2d90eb | 448 | { |
cb8d4c8f | 449 | uint8_t bufc[12 + 1 + 255]; /* start (8), len (4), rbname up to 256 */ |
1e2d90eb | 450 | size_t msglen = 12; /* start + len */ |
2e2bce16 | 451 | size_t len = qemu_ram_pagesize(rb); |
d6208e35 | 452 | enum mig_rp_message_type msg_type; |
2e2bce16 PX |
453 | const char *rbname; |
454 | int rbname_len; | |
1e2d90eb DDAG |
455 | |
456 | *(uint64_t *)bufc = cpu_to_be64((uint64_t)start); | |
457 | *(uint32_t *)(bufc + 8) = cpu_to_be32((uint32_t)len); | |
458 | ||
2e2bce16 PX |
459 | /* |
460 | * We maintain the last ramblock that we requested for page. Note that we | |
461 | * don't need locking because this function will only be called within the | |
462 | * postcopy ram fault thread. | |
463 | */ | |
464 | if (rb != mis->last_rb) { | |
465 | mis->last_rb = rb; | |
466 | ||
467 | rbname = qemu_ram_get_idstr(rb); | |
468 | rbname_len = strlen(rbname); | |
469 | ||
1e2d90eb DDAG |
470 | assert(rbname_len < 256); |
471 | ||
472 | bufc[msglen++] = rbname_len; | |
473 | memcpy(bufc + msglen, rbname, rbname_len); | |
474 | msglen += rbname_len; | |
d6208e35 | 475 | msg_type = MIG_RP_MSG_REQ_PAGES_ID; |
1e2d90eb | 476 | } else { |
d6208e35 | 477 | msg_type = MIG_RP_MSG_REQ_PAGES; |
1e2d90eb | 478 | } |
d6208e35 PX |
479 | |
480 | return migrate_send_rp_message(mis, msg_type, msglen, bufc); | |
1e2d90eb DDAG |
481 | } |
482 | ||
7a267fc4 | 483 | int migrate_send_rp_req_pages(MigrationIncomingState *mis, |
8f8bfffc | 484 | RAMBlock *rb, ram_addr_t start, uint64_t haddr) |
7a267fc4 | 485 | { |
7648297d | 486 | void *aligned = (void *)(uintptr_t)ROUND_DOWN(haddr, qemu_ram_pagesize(rb)); |
a2429283 | 487 | bool received = false; |
8f8bfffc PX |
488 | |
489 | WITH_QEMU_LOCK_GUARD(&mis->page_request_mutex) { | |
490 | received = ramblock_recv_bitmap_test_byte_offset(rb, start); | |
491 | if (!received && !g_tree_lookup(mis->page_requested, aligned)) { | |
492 | /* | |
493 | * The page has not been received, and it's not yet in the page | |
494 | * request list. Queue it. Set the value of element to 1, so that | |
495 | * things like g_tree_lookup() will return TRUE (1) when found. | |
496 | */ | |
497 | g_tree_insert(mis->page_requested, aligned, (gpointer)1); | |
cf02f29e | 498 | qatomic_inc(&mis->page_requested_count); |
8f8bfffc PX |
499 | trace_postcopy_page_req_add(aligned, mis->page_requested_count); |
500 | } | |
501 | } | |
502 | ||
503 | /* | |
504 | * If the page is there, skip sending the message. We don't even need the | |
505 | * lock because as long as the page arrived, it'll be there forever. | |
506 | */ | |
507 | if (received) { | |
508 | return 0; | |
509 | } | |
510 | ||
7a267fc4 PX |
511 | return migrate_send_rp_message_req_pages(mis, rb, start); |
512 | } | |
513 | ||
aad555c2 ZC |
514 | static bool migration_colo_enabled; |
515 | bool migration_incoming_colo_enabled(void) | |
516 | { | |
517 | return migration_colo_enabled; | |
518 | } | |
519 | ||
520 | void migration_incoming_disable_colo(void) | |
521 | { | |
18b1d3c9 | 522 | ram_block_discard_disable(false); |
aad555c2 ZC |
523 | migration_colo_enabled = false; |
524 | } | |
525 | ||
18b1d3c9 | 526 | int migration_incoming_enable_colo(void) |
aad555c2 | 527 | { |
51e47cf8 VSO |
528 | #ifndef CONFIG_REPLICATION |
529 | error_report("ENABLE_COLO command come in migration stream, but COLO " | |
530 | "module is not built in"); | |
531 | return -ENOTSUP; | |
532 | #endif | |
533 | ||
121ccedc VSO |
534 | if (!migrate_colo()) { |
535 | error_report("ENABLE_COLO command come in migration stream, but c-colo " | |
536 | "capability is not set"); | |
537 | return -EINVAL; | |
538 | } | |
539 | ||
18b1d3c9 DH |
540 | if (ram_block_discard_disable(true)) { |
541 | error_report("COLO: cannot disable RAM discard"); | |
542 | return -EBUSY; | |
543 | } | |
aad555c2 | 544 | migration_colo_enabled = true; |
18b1d3c9 | 545 | return 0; |
aad555c2 ZC |
546 | } |
547 | ||
9aca82ba JQ |
548 | void migrate_add_address(SocketAddress *address) |
549 | { | |
550 | MigrationIncomingState *mis = migration_incoming_get_current(); | |
9aca82ba | 551 | |
54aa3de7 EB |
552 | QAPI_LIST_PREPEND(mis->socket_address_list, |
553 | QAPI_CLONE(SocketAddress, address)); | |
9aca82ba JQ |
554 | } |
555 | ||
967f2de5 HG |
556 | bool migrate_uri_parse(const char *uri, MigrationChannel **channel, |
557 | Error **errp) | |
72a8192e | 558 | { |
5994024f | 559 | g_autoptr(MigrationChannel) val = g_new0(MigrationChannel, 1); |
72a8192e | 560 | g_autoptr(MigrationAddress) addr = g_new0(MigrationAddress, 1); |
72a8192e HG |
561 | InetSocketAddress *isock = &addr->u.rdma; |
562 | strList **tail = &addr->u.exec.args; | |
563 | ||
564 | if (strstart(uri, "exec:", NULL)) { | |
565 | addr->transport = MIGRATION_ADDRESS_TYPE_EXEC; | |
566 | #ifdef WIN32 | |
567 | QAPI_LIST_APPEND(tail, g_strdup(exec_get_cmd_path())); | |
568 | QAPI_LIST_APPEND(tail, g_strdup("/c")); | |
569 | #else | |
570 | QAPI_LIST_APPEND(tail, g_strdup("/bin/sh")); | |
571 | QAPI_LIST_APPEND(tail, g_strdup("-c")); | |
572 | #endif | |
573 | QAPI_LIST_APPEND(tail, g_strdup(uri + strlen("exec:"))); | |
574 | } else if (strstart(uri, "rdma:", NULL)) { | |
575 | if (inet_parse(isock, uri + strlen("rdma:"), errp)) { | |
576 | qapi_free_InetSocketAddress(isock); | |
577 | return false; | |
578 | } | |
579 | addr->transport = MIGRATION_ADDRESS_TYPE_RDMA; | |
580 | } else if (strstart(uri, "tcp:", NULL) || | |
581 | strstart(uri, "unix:", NULL) || | |
582 | strstart(uri, "vsock:", NULL) || | |
583 | strstart(uri, "fd:", NULL)) { | |
584 | addr->transport = MIGRATION_ADDRESS_TYPE_SOCKET; | |
41581265 | 585 | SocketAddress *saddr = socket_parse(uri, errp); |
72a8192e HG |
586 | if (!saddr) { |
587 | return false; | |
588 | } | |
589 | addr->u.socket.type = saddr->type; | |
590 | addr->u.socket.u = saddr->u; | |
41581265 ZZ |
591 | /* Don't free the objects inside; their ownership moved to "addr" */ |
592 | g_free(saddr); | |
72a8192e HG |
593 | } else if (strstart(uri, "file:", NULL)) { |
594 | addr->transport = MIGRATION_ADDRESS_TYPE_FILE; | |
595 | addr->u.file.filename = g_strdup(uri + strlen("file:")); | |
596 | if (file_parse_offset(addr->u.file.filename, &addr->u.file.offset, | |
597 | errp)) { | |
598 | return false; | |
599 | } | |
600 | } else { | |
601 | error_setg(errp, "unknown migration protocol: %s", uri); | |
602 | return false; | |
603 | } | |
604 | ||
5994024f HG |
605 | val->channel_type = MIGRATION_CHANNEL_TYPE_MAIN; |
606 | val->addr = g_steal_pointer(&addr); | |
607 | *channel = g_steal_pointer(&val); | |
72a8192e HG |
608 | return true; |
609 | } | |
610 | ||
074dbce5 HG |
611 | static void qemu_start_incoming_migration(const char *uri, bool has_channels, |
612 | MigrationChannelList *channels, | |
613 | Error **errp) | |
5bb7910a | 614 | { |
bc1d54ee | 615 | g_autoptr(MigrationChannel) channel = NULL; |
5994024f | 616 | MigrationAddress *addr = NULL; |
4111a732 | 617 | MigrationIncomingState *mis = migration_incoming_get_current(); |
34c9dd8e | 618 | |
074dbce5 HG |
619 | /* |
620 | * Having preliminary checks for uri and channel | |
621 | */ | |
0770ad43 HG |
622 | if (!uri == !channels) { |
623 | error_setg(errp, "need either 'uri' or 'channels' argument"); | |
074dbce5 | 624 | return; |
0770ad43 HG |
625 | } |
626 | ||
627 | if (channels) { | |
5994024f HG |
628 | /* To verify that Migrate channel list has only item */ |
629 | if (channels->next) { | |
630 | error_setg(errp, "Channel list has more than one entries"); | |
631 | return; | |
632 | } | |
bc1d54ee | 633 | addr = channels->value->addr; |
0770ad43 HG |
634 | } |
635 | ||
636 | if (uri) { | |
5994024f HG |
637 | /* caller uses the old URI syntax */ |
638 | if (!migrate_uri_parse(uri, &channel, errp)) { | |
639 | return; | |
640 | } | |
bc1d54ee | 641 | addr = channel->addr; |
074dbce5 | 642 | } |
d6f74fd1 | 643 | |
d95533e1 | 644 | /* transport mechanism not suitable for migration? */ |
5994024f | 645 | if (!migration_channels_and_transport_compatible(addr, errp)) { |
72a8192e HG |
646 | return; |
647 | } | |
648 | ||
4111a732 FR |
649 | migrate_set_state(&mis->state, MIGRATION_STATUS_NONE, |
650 | MIGRATION_STATUS_SETUP); | |
651 | ||
5994024f HG |
652 | if (addr->transport == MIGRATION_ADDRESS_TYPE_SOCKET) { |
653 | SocketAddress *saddr = &addr->u.socket; | |
34dfc5e4 HG |
654 | if (saddr->type == SOCKET_ADDRESS_TYPE_INET || |
655 | saddr->type == SOCKET_ADDRESS_TYPE_UNIX || | |
656 | saddr->type == SOCKET_ADDRESS_TYPE_VSOCK) { | |
657 | socket_start_incoming_migration(saddr, errp); | |
658 | } else if (saddr->type == SOCKET_ADDRESS_TYPE_FD) { | |
659 | fd_start_incoming_migration(saddr->u.fd.str, errp); | |
660 | } | |
2da776db | 661 | #ifdef CONFIG_RDMA |
5994024f | 662 | } else if (addr->transport == MIGRATION_ADDRESS_TYPE_RDMA) { |
b88a3306 JQ |
663 | if (migrate_compress()) { |
664 | error_setg(errp, "RDMA and compression can't be used together"); | |
665 | return; | |
666 | } | |
667 | if (migrate_xbzrle()) { | |
668 | error_setg(errp, "RDMA and XBZRLE can't be used together"); | |
669 | return; | |
670 | } | |
671 | if (migrate_multifd()) { | |
672 | error_setg(errp, "RDMA and multifd can't be used together"); | |
673 | return; | |
674 | } | |
5994024f | 675 | rdma_start_incoming_migration(&addr->u.rdma, errp); |
2da776db | 676 | #endif |
5994024f HG |
677 | } else if (addr->transport == MIGRATION_ADDRESS_TYPE_EXEC) { |
678 | exec_start_incoming_migration(addr->u.exec.args, errp); | |
679 | } else if (addr->transport == MIGRATION_ADDRESS_TYPE_FILE) { | |
680 | file_start_incoming_migration(&addr->u.file, errp); | |
adde220a | 681 | } else { |
312fd5f2 | 682 | error_setg(errp, "unknown migration protocol: %s", uri); |
8ca5e801 | 683 | } |
5bb7910a AL |
684 | } |
685 | ||
0aa6aefc DL |
686 | static void process_incoming_migration_bh(void *opaque) |
687 | { | |
688 | Error *local_err = NULL; | |
689 | MigrationIncomingState *mis = opaque; | |
690 | ||
3e5f3bcd PX |
691 | trace_vmstate_downtime_checkpoint("dst-precopy-bh-enter"); |
692 | ||
0f073f44 DDAG |
693 | /* If capability late_block_activate is set: |
694 | * Only fire up the block code now if we're going to restart the | |
695 | * VM, else 'cont' will do it. | |
696 | * This causes file locking to happen; so we don't want it to happen | |
697 | * unless we really are starting the VM. | |
698 | */ | |
699 | if (!migrate_late_block_activate() || | |
700 | (autostart && (!global_state_received() || | |
b4e9ddcc | 701 | runstate_is_live(global_state_get_runstate())))) { |
3b717194 | 702 | /* Make sure all file formats throw away their mutable metadata. |
0f073f44 | 703 | * If we get an error here, just don't restart the VM yet. */ |
3b717194 | 704 | bdrv_activate_all(&local_err); |
0f073f44 DDAG |
705 | if (local_err) { |
706 | error_report_err(local_err); | |
707 | local_err = NULL; | |
708 | autostart = false; | |
709 | } | |
d35ff5e6 KW |
710 | } |
711 | ||
0aa6aefc DL |
712 | /* |
713 | * This must happen after all error conditions are dealt with and | |
714 | * we're sure the VM is going to be running on this host. | |
715 | */ | |
7659505c | 716 | qemu_announce_self(&mis->announce_timer, migrate_announce_params()); |
0aa6aefc | 717 | |
3e5f3bcd PX |
718 | trace_vmstate_downtime_checkpoint("dst-precopy-bh-announced"); |
719 | ||
cde85c37 | 720 | multifd_recv_shutdown(); |
0aa6aefc | 721 | |
b35ebdf0 VSO |
722 | dirty_bitmap_mig_before_vm_start(); |
723 | ||
0aa6aefc | 724 | if (!global_state_received() || |
b4e9ddcc | 725 | runstate_is_live(global_state_get_runstate())) { |
0aa6aefc DL |
726 | if (autostart) { |
727 | vm_start(); | |
728 | } else { | |
729 | runstate_set(RUN_STATE_PAUSED); | |
730 | } | |
db009729 ZC |
731 | } else if (migration_incoming_colo_enabled()) { |
732 | migration_incoming_disable_colo(); | |
733 | vm_start(); | |
0aa6aefc DL |
734 | } else { |
735 | runstate_set(global_state_get_runstate()); | |
736 | } | |
3e5f3bcd | 737 | trace_vmstate_downtime_checkpoint("dst-precopy-bh-vm-started"); |
0aa6aefc DL |
738 | /* |
739 | * This must happen after any state changes since as soon as an external | |
740 | * observer sees this event they might start to prod at the VM assuming | |
741 | * it's ready to use. | |
742 | */ | |
743 | migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE, | |
744 | MIGRATION_STATUS_COMPLETED); | |
0aa6aefc DL |
745 | migration_incoming_state_destroy(); |
746 | } | |
747 | ||
38e8f9af MAL |
748 | static void coroutine_fn |
749 | process_incoming_migration_co(void *opaque) | |
511c0231 | 750 | { |
b4b076da | 751 | MigrationIncomingState *mis = migration_incoming_get_current(); |
e9bef235 | 752 | PostcopyState ps; |
1c12e1f5 PB |
753 | int ret; |
754 | ||
4f0fae7f | 755 | assert(mis->from_src_file); |
c323518a LS |
756 | |
757 | if (compress_threads_load_setup(mis->from_src_file)) { | |
758 | error_report("Failed to setup decompress threads"); | |
759 | goto fail; | |
760 | } | |
761 | ||
67f11b5c | 762 | mis->largest_page_size = qemu_ram_pagesize_largest(); |
093e3c42 | 763 | postcopy_state_set(POSTCOPY_INCOMING_NONE); |
4111a732 | 764 | migrate_set_state(&mis->state, MIGRATION_STATUS_SETUP, |
93d7af6f | 765 | MIGRATION_STATUS_ACTIVE); |
dd42ce24 VSO |
766 | |
767 | mis->loadvm_co = qemu_coroutine_self(); | |
4f0fae7f | 768 | ret = qemu_loadvm_state(mis->from_src_file); |
dd42ce24 | 769 | mis->loadvm_co = NULL; |
bca7856a | 770 | |
3e5f3bcd PX |
771 | trace_vmstate_downtime_checkpoint("dst-precopy-loadvm-completed"); |
772 | ||
e9bef235 DDAG |
773 | ps = postcopy_state_get(); |
774 | trace_process_incoming_migration_co_end(ret, ps); | |
775 | if (ps != POSTCOPY_INCOMING_NONE) { | |
776 | if (ps == POSTCOPY_INCOMING_ADVISE) { | |
777 | /* | |
778 | * Where a migration had postcopy enabled (and thus went to advise) | |
779 | * but managed to complete within the precopy period, we can use | |
780 | * the normal exit. | |
781 | */ | |
782 | postcopy_ram_incoming_cleanup(mis); | |
783 | } else if (ret >= 0) { | |
784 | /* | |
785 | * Postcopy was started, cleanup should happen at the end of the | |
786 | * postcopy thread. | |
787 | */ | |
788 | trace_process_incoming_migration_co_postcopy_end_main(); | |
789 | return; | |
790 | } | |
791 | /* Else if something went wrong then just fall out of the normal exit */ | |
792 | } | |
793 | ||
ecbfec6d | 794 | if (ret < 0) { |
e3b8ad5c FR |
795 | MigrationState *s = migrate_get_current(); |
796 | ||
797 | if (migrate_has_error(s)) { | |
798 | WITH_QEMU_LOCK_GUARD(&s->error_mutex) { | |
799 | error_report_err(s->error); | |
800 | } | |
801 | } | |
ecbfec6d VSO |
802 | error_report("load of migration failed: %s", strerror(-ret)); |
803 | goto fail; | |
804 | } | |
805 | ||
d0a14a2b VSO |
806 | if (colo_incoming_co() < 0) { |
807 | goto fail; | |
25d0c16f HZ |
808 | } |
809 | ||
44d0d456 | 810 | migration_bh_schedule(process_incoming_migration_bh, mis); |
6d99c2d4 FL |
811 | return; |
812 | fail: | |
6d99c2d4 FL |
813 | migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE, |
814 | MIGRATION_STATUS_FAILED); | |
815 | qemu_fclose(mis->from_src_file); | |
e5bac1f5 | 816 | |
cde85c37 | 817 | multifd_recv_cleanup(); |
c323518a | 818 | compress_threads_load_cleanup(); |
e5bac1f5 | 819 | |
6d99c2d4 | 820 | exit(EXIT_FAILURE); |
511c0231 JQ |
821 | } |
822 | ||
b673eab4 | 823 | /** |
7d6f6933 | 824 | * migration_incoming_setup: Setup incoming migration |
b673eab4 | 825 | * @f: file for main migration channel |
b673eab4 | 826 | */ |
c606f3ba | 827 | static void migration_incoming_setup(QEMUFile *f) |
82a4da79 | 828 | { |
4f0fae7f | 829 | MigrationIncomingState *mis = migration_incoming_get_current(); |
82a4da79 | 830 | |
4f0fae7f JQ |
831 | if (!mis->from_src_file) { |
832 | mis->from_src_file = f; | |
833 | } | |
06ad5135 | 834 | qemu_file_set_blocking(f, false); |
e595a01a JQ |
835 | } |
836 | ||
36c2f8be | 837 | void migration_incoming_process(void) |
e595a01a JQ |
838 | { |
839 | Coroutine *co = qemu_coroutine_create(process_incoming_migration_co, NULL); | |
0b8b8753 | 840 | qemu_coroutine_enter(co); |
82a4da79 PB |
841 | } |
842 | ||
884835fa | 843 | /* Returns true if recovered from a paused migration, otherwise false */ |
a39e9339 | 844 | static bool postcopy_try_recover(void) |
e595a01a | 845 | { |
d96c9e8d PX |
846 | MigrationIncomingState *mis = migration_incoming_get_current(); |
847 | ||
848 | if (mis->state == MIGRATION_STATUS_POSTCOPY_PAUSED) { | |
849 | /* Resumed from a paused postcopy migration */ | |
850 | ||
a39e9339 PX |
851 | /* This should be set already in migration_incoming_setup() */ |
852 | assert(mis->from_src_file); | |
d96c9e8d | 853 | /* Postcopy has standalone thread to do vm load */ |
a39e9339 | 854 | qemu_file_set_blocking(mis->from_src_file, true); |
d96c9e8d PX |
855 | |
856 | /* Re-configure the return path */ | |
a39e9339 | 857 | mis->to_src_file = qemu_file_get_return_path(mis->from_src_file); |
d96c9e8d PX |
858 | |
859 | migrate_set_state(&mis->state, MIGRATION_STATUS_POSTCOPY_PAUSED, | |
860 | MIGRATION_STATUS_POSTCOPY_RECOVER); | |
861 | ||
862 | /* | |
863 | * Here, we only wake up the main loading thread (while the | |
60bb3c58 | 864 | * rest threads will still be waiting), so that we can receive |
d96c9e8d | 865 | * commands from source now, and answer it if needed. The |
60bb3c58 | 866 | * rest threads will be woken up afterwards until we are sure |
d96c9e8d PX |
867 | * that source is ready to reply to page requests. |
868 | */ | |
869 | qemu_sem_post(&mis->postcopy_pause_sem_dst); | |
884835fa PX |
870 | return true; |
871 | } | |
872 | ||
873 | return false; | |
874 | } | |
875 | ||
b0cf3bfc | 876 | void migration_fd_process_incoming(QEMUFile *f) |
884835fa | 877 | { |
c606f3ba | 878 | migration_incoming_setup(f); |
a39e9339 | 879 | if (postcopy_try_recover()) { |
b673eab4 JQ |
880 | return; |
881 | } | |
884835fa | 882 | migration_incoming_process(); |
e595a01a JQ |
883 | } |
884 | ||
5655aab0 PX |
885 | /* |
886 | * Returns true when we want to start a new incoming migration process, | |
887 | * false otherwise. | |
888 | */ | |
889 | static bool migration_should_start_incoming(bool main_channel) | |
890 | { | |
891 | /* Multifd doesn't start unless all channels are established */ | |
51b07548 | 892 | if (migrate_multifd()) { |
5655aab0 PX |
893 | return migration_has_all_channels(); |
894 | } | |
895 | ||
896 | /* Preempt channel only starts when the main channel is created */ | |
897 | if (migrate_postcopy_preempt()) { | |
898 | return main_channel; | |
899 | } | |
900 | ||
901 | /* | |
902 | * For all the rest types of migration, we should only reach here when | |
903 | * it's the main channel that's being created, and we should always | |
904 | * proceed with this channel. | |
905 | */ | |
906 | assert(main_channel); | |
907 | return true; | |
908 | } | |
909 | ||
49ed0d24 | 910 | void migration_ioc_process_incoming(QIOChannel *ioc, Error **errp) |
4f0fae7f JQ |
911 | { |
912 | MigrationIncomingState *mis = migration_incoming_get_current(); | |
b673eab4 | 913 | Error *local_err = NULL; |
36f62f11 | 914 | QEMUFile *f; |
6720c2b3 | 915 | bool default_channel = true; |
916 | uint32_t channel_magic = 0; | |
917 | int ret = 0; | |
4f0fae7f | 918 | |
2dd7ee7a FR |
919 | if (migrate_multifd() && !migrate_mapped_ram() && |
920 | !migrate_postcopy_ram() && | |
6720c2b3 | 921 | qio_channel_has_feature(ioc, QIO_CHANNEL_FEATURE_READ_MSG_PEEK)) { |
922 | /* | |
923 | * With multiple channels, it is possible that we receive channels | |
924 | * out of order on destination side, causing incorrect mapping of | |
925 | * source channels on destination side. Check channel MAGIC to | |
926 | * decide type of channel. Please note this is best effort, postcopy | |
927 | * preempt channel does not send any magic number so avoid it for | |
928 | * postcopy live migration. Also tls live migration already does | |
929 | * tls handshake while initializing main channel so with tls this | |
930 | * issue is not possible. | |
931 | */ | |
932 | ret = migration_channel_read_peek(ioc, (void *)&channel_magic, | |
b6f4c0c7 | 933 | sizeof(channel_magic), errp); |
6720c2b3 | 934 | |
935 | if (ret != 0) { | |
6720c2b3 | 936 | return; |
937 | } | |
938 | ||
939 | default_channel = (channel_magic == cpu_to_be32(QEMU_VM_FILE_MAGIC)); | |
940 | } else { | |
941 | default_channel = !mis->from_src_file; | |
942 | } | |
943 | ||
cde85c37 | 944 | if (multifd_recv_setup(errp) != 0) { |
6720c2b3 | 945 | return; |
946 | } | |
947 | ||
948 | if (default_channel) { | |
36f62f11 | 949 | f = qemu_file_new_input(ioc); |
c606f3ba | 950 | migration_incoming_setup(f); |
a429e7f4 PX |
951 | } else { |
952 | /* Multiple connections */ | |
36f62f11 | 953 | assert(migration_needs_multiple_sockets()); |
51b07548 | 954 | if (migrate_multifd()) { |
6720c2b3 | 955 | multifd_recv_new_channel(ioc, &local_err); |
36f62f11 PX |
956 | } else { |
957 | assert(migrate_postcopy_preempt()); | |
958 | f = qemu_file_new_input(ioc); | |
6720c2b3 | 959 | postcopy_preempt_new_channel(mis, f); |
36f62f11 | 960 | } |
49ed0d24 FL |
961 | if (local_err) { |
962 | error_propagate(errp, local_err); | |
963 | return; | |
964 | } | |
4f0fae7f | 965 | } |
81e62053 | 966 | |
5655aab0 | 967 | if (migration_should_start_incoming(default_channel)) { |
a39e9339 PX |
968 | /* If it's a recovery, we're done */ |
969 | if (postcopy_try_recover()) { | |
970 | return; | |
971 | } | |
81e62053 PX |
972 | migration_incoming_process(); |
973 | } | |
4f0fae7f JQ |
974 | } |
975 | ||
428d8908 JQ |
976 | /** |
977 | * @migration_has_all_channels: We have received all channels that we need | |
978 | * | |
979 | * Returns true when we have got connections to all the channels that | |
980 | * we need for migration. | |
981 | */ | |
982 | bool migration_has_all_channels(void) | |
983 | { | |
ca273df3 | 984 | MigrationIncomingState *mis = migration_incoming_get_current(); |
62c1e0ca | 985 | |
36f62f11 PX |
986 | if (!mis->from_src_file) { |
987 | return false; | |
988 | } | |
62c1e0ca | 989 | |
51b07548 | 990 | if (migrate_multifd()) { |
36f62f11 PX |
991 | return multifd_recv_all_channels_created(); |
992 | } | |
993 | ||
994 | if (migrate_postcopy_preempt()) { | |
995 | return mis->postcopy_qemufile_dst != NULL; | |
996 | } | |
997 | ||
998 | return true; | |
428d8908 JQ |
999 | } |
1000 | ||
1b4adb10 AH |
1001 | int migrate_send_rp_switchover_ack(MigrationIncomingState *mis) |
1002 | { | |
1003 | return migrate_send_rp_message(mis, MIG_RP_MSG_SWITCHOVER_ACK, 0, NULL); | |
1004 | } | |
1005 | ||
6decec93 DDAG |
1006 | /* |
1007 | * Send a 'SHUT' message on the return channel with the given value | |
1008 | * to indicate that we've finished with the RP. Non-0 value indicates | |
1009 | * error. | |
1010 | */ | |
1011 | void migrate_send_rp_shut(MigrationIncomingState *mis, | |
1012 | uint32_t value) | |
1013 | { | |
1014 | uint32_t buf; | |
1015 | ||
1016 | buf = cpu_to_be32(value); | |
1017 | migrate_send_rp_message(mis, MIG_RP_MSG_SHUT, sizeof(buf), &buf); | |
1018 | } | |
1019 | ||
1020 | /* | |
1021 | * Send a 'PONG' message on the return channel with the given value | |
1022 | * (normally in response to a 'PING') | |
1023 | */ | |
1024 | void migrate_send_rp_pong(MigrationIncomingState *mis, | |
1025 | uint32_t value) | |
1026 | { | |
1027 | uint32_t buf; | |
1028 | ||
1029 | buf = cpu_to_be32(value); | |
1030 | migrate_send_rp_message(mis, MIG_RP_MSG_PONG, sizeof(buf), &buf); | |
1031 | } | |
1032 | ||
a335debb PX |
1033 | void migrate_send_rp_recv_bitmap(MigrationIncomingState *mis, |
1034 | char *block_name) | |
1035 | { | |
1036 | char buf[512]; | |
1037 | int len; | |
1038 | int64_t res; | |
1039 | ||
1040 | /* | |
1041 | * First, we send the header part. It contains only the len of | |
1042 | * idstr, and the idstr itself. | |
1043 | */ | |
1044 | len = strlen(block_name); | |
1045 | buf[0] = len; | |
1046 | memcpy(buf + 1, block_name, len); | |
1047 | ||
1048 | if (mis->state != MIGRATION_STATUS_POSTCOPY_RECOVER) { | |
1049 | error_report("%s: MSG_RP_RECV_BITMAP only used for recovery", | |
1050 | __func__); | |
1051 | return; | |
1052 | } | |
1053 | ||
1054 | migrate_send_rp_message(mis, MIG_RP_MSG_RECV_BITMAP, len + 1, buf); | |
1055 | ||
1056 | /* | |
1057 | * Next, we dump the received bitmap to the stream. | |
1058 | * | |
1059 | * TODO: currently we are safe since we are the only one that is | |
1060 | * using the to_src_file handle (fault thread is still paused), | |
1061 | * and it's ok even not taking the mutex. However the best way is | |
1062 | * to take the lock before sending the message header, and release | |
1063 | * the lock after sending the bitmap. | |
1064 | */ | |
1065 | qemu_mutex_lock(&mis->rp_mutex); | |
1066 | res = ramblock_recv_bitmap_send(mis->to_src_file, block_name); | |
1067 | qemu_mutex_unlock(&mis->rp_mutex); | |
1068 | ||
1069 | trace_migrate_send_rp_recv_bitmap(block_name, res); | |
1070 | } | |
1071 | ||
13955b89 PX |
1072 | void migrate_send_rp_resume_ack(MigrationIncomingState *mis, uint32_t value) |
1073 | { | |
1074 | uint32_t buf; | |
1075 | ||
1076 | buf = cpu_to_be32(value); | |
1077 | migrate_send_rp_message(mis, MIG_RP_MSG_RESUME_ACK, sizeof(buf), &buf); | |
1078 | } | |
1079 | ||
f6844b99 DDAG |
1080 | /* |
1081 | * Return true if we're already in the middle of a migration | |
1082 | * (i.e. any of the active or setup states) | |
1083 | */ | |
7dcb3c87 | 1084 | bool migration_is_setup_or_active(void) |
f6844b99 | 1085 | { |
7dcb3c87 SS |
1086 | MigrationState *s = current_migration; |
1087 | ||
1088 | switch (s->state) { | |
f6844b99 | 1089 | case MIGRATION_STATUS_ACTIVE: |
9ec055ae | 1090 | case MIGRATION_STATUS_POSTCOPY_ACTIVE: |
a688d2c1 | 1091 | case MIGRATION_STATUS_POSTCOPY_PAUSED: |
135b87b4 | 1092 | case MIGRATION_STATUS_POSTCOPY_RECOVER: |
f6844b99 | 1093 | case MIGRATION_STATUS_SETUP: |
31e06077 DDAG |
1094 | case MIGRATION_STATUS_PRE_SWITCHOVER: |
1095 | case MIGRATION_STATUS_DEVICE: | |
c7e0acd5 | 1096 | case MIGRATION_STATUS_WAIT_UNPLUG: |
19dd408a | 1097 | case MIGRATION_STATUS_COLO: |
f6844b99 DDAG |
1098 | return true; |
1099 | ||
1100 | default: | |
1101 | return false; | |
1102 | ||
1103 | } | |
1104 | } | |
1105 | ||
392d87e2 JQ |
1106 | bool migration_is_running(int state) |
1107 | { | |
1108 | switch (state) { | |
1109 | case MIGRATION_STATUS_ACTIVE: | |
1110 | case MIGRATION_STATUS_POSTCOPY_ACTIVE: | |
1111 | case MIGRATION_STATUS_POSTCOPY_PAUSED: | |
1112 | case MIGRATION_STATUS_POSTCOPY_RECOVER: | |
1113 | case MIGRATION_STATUS_SETUP: | |
1114 | case MIGRATION_STATUS_PRE_SWITCHOVER: | |
1115 | case MIGRATION_STATUS_DEVICE: | |
1116 | case MIGRATION_STATUS_WAIT_UNPLUG: | |
1117 | case MIGRATION_STATUS_CANCELLING: | |
392d87e2 JQ |
1118 | return true; |
1119 | ||
1120 | default: | |
1121 | return false; | |
1122 | ||
1123 | } | |
1124 | } | |
1125 | ||
db18dee7 PX |
1126 | static bool migrate_show_downtime(MigrationState *s) |
1127 | { | |
1128 | return (s->state == MIGRATION_STATUS_COMPLETED) || migration_in_postcopy(); | |
1129 | } | |
1130 | ||
640dfb14 WY |
1131 | static void populate_time_info(MigrationInfo *info, MigrationState *s) |
1132 | { | |
1133 | info->has_status = true; | |
1134 | info->has_setup_time = true; | |
1135 | info->setup_time = s->setup_time; | |
db18dee7 | 1136 | |
640dfb14 WY |
1137 | if (s->state == MIGRATION_STATUS_COMPLETED) { |
1138 | info->has_total_time = true; | |
1139 | info->total_time = s->total_time; | |
640dfb14 WY |
1140 | } else { |
1141 | info->has_total_time = true; | |
1142 | info->total_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME) - | |
1143 | s->start_time; | |
db18dee7 PX |
1144 | } |
1145 | ||
1146 | if (migrate_show_downtime(s)) { | |
1147 | info->has_downtime = true; | |
1148 | info->downtime = s->downtime; | |
1149 | } else { | |
640dfb14 WY |
1150 | info->has_expected_downtime = true; |
1151 | info->expected_downtime = s->expected_downtime; | |
1152 | } | |
1153 | } | |
1154 | ||
a22463a5 DDAG |
1155 | static void populate_ram_info(MigrationInfo *info, MigrationState *s) |
1156 | { | |
144fa06b JQ |
1157 | size_t page_size = qemu_target_page_size(); |
1158 | ||
a22463a5 | 1159 | info->ram = g_malloc0(sizeof(*info->ram)); |
897fd8bd | 1160 | info->ram->transferred = migration_transferred_bytes(); |
a22463a5 | 1161 | info->ram->total = ram_bytes_total(); |
aff3f660 | 1162 | info->ram->duplicate = stat64_get(&mig_stats.zero_pages); |
bedf53c1 JQ |
1163 | /* legacy value. It is not used anymore */ |
1164 | info->ram->skipped = 0; | |
aff3f660 | 1165 | info->ram->normal = stat64_get(&mig_stats.normal_pages); |
23b7576d | 1166 | info->ram->normal_bytes = info->ram->normal * page_size; |
a22463a5 | 1167 | info->ram->mbps = s->mbps; |
536b5a4e | 1168 | info->ram->dirty_sync_count = |
aff3f660 | 1169 | stat64_get(&mig_stats.dirty_sync_count); |
cf20c897 | 1170 | info->ram->dirty_sync_missed_zero_copy = |
aff3f660 | 1171 | stat64_get(&mig_stats.dirty_sync_missed_zero_copy); |
3c764f9b | 1172 | info->ram->postcopy_requests = |
aff3f660 | 1173 | stat64_get(&mig_stats.postcopy_requests); |
144fa06b | 1174 | info->ram->page_size = page_size; |
aff3f660 | 1175 | info->ram->multifd_bytes = stat64_get(&mig_stats.multifd_bytes); |
aecbfe9c | 1176 | info->ram->pages_per_second = s->pages_per_second; |
aff3f660 JQ |
1177 | info->ram->precopy_bytes = stat64_get(&mig_stats.precopy_bytes); |
1178 | info->ram->downtime_bytes = stat64_get(&mig_stats.downtime_bytes); | |
1179 | info->ram->postcopy_bytes = stat64_get(&mig_stats.postcopy_bytes); | |
a22463a5 | 1180 | |
87dca0c9 | 1181 | if (migrate_xbzrle()) { |
114f5aee JQ |
1182 | info->xbzrle_cache = g_malloc0(sizeof(*info->xbzrle_cache)); |
1183 | info->xbzrle_cache->cache_size = migrate_xbzrle_cache_size(); | |
9360447d JQ |
1184 | info->xbzrle_cache->bytes = xbzrle_counters.bytes; |
1185 | info->xbzrle_cache->pages = xbzrle_counters.pages; | |
1186 | info->xbzrle_cache->cache_miss = xbzrle_counters.cache_miss; | |
1187 | info->xbzrle_cache->cache_miss_rate = xbzrle_counters.cache_miss_rate; | |
e460a4b1 | 1188 | info->xbzrle_cache->encoding_rate = xbzrle_counters.encoding_rate; |
9360447d | 1189 | info->xbzrle_cache->overflow = xbzrle_counters.overflow; |
114f5aee JQ |
1190 | } |
1191 | ||
6f609005 | 1192 | populate_compress(info); |
76e03000 | 1193 | |
338182c8 JQ |
1194 | if (cpu_throttle_active()) { |
1195 | info->has_cpu_throttle_percentage = true; | |
1196 | info->cpu_throttle_percentage = cpu_throttle_get_percentage(); | |
1197 | } | |
1198 | ||
a22463a5 DDAG |
1199 | if (s->state != MIGRATION_STATUS_COMPLETED) { |
1200 | info->ram->remaining = ram_bytes_remaining(); | |
72f8e587 | 1201 | info->ram->dirty_pages_rate = |
aff3f660 | 1202 | stat64_get(&mig_stats.dirty_pages_rate); |
a22463a5 | 1203 | } |
15699cf5 HH |
1204 | |
1205 | if (migrate_dirty_limit() && dirtylimit_in_service()) { | |
1206 | info->has_dirty_limit_throttle_time_per_round = true; | |
1207 | info->dirty_limit_throttle_time_per_round = | |
1208 | dirtylimit_throttle_time_per_round(); | |
1209 | ||
1210 | info->has_dirty_limit_ring_full_time = true; | |
1211 | info->dirty_limit_ring_full_time = dirtylimit_ring_full_time(); | |
1212 | } | |
a22463a5 DDAG |
1213 | } |
1214 | ||
930ac04c JQ |
1215 | static void populate_disk_info(MigrationInfo *info) |
1216 | { | |
1217 | if (blk_mig_active()) { | |
930ac04c JQ |
1218 | info->disk = g_malloc0(sizeof(*info->disk)); |
1219 | info->disk->transferred = blk_mig_bytes_transferred(); | |
1220 | info->disk->remaining = blk_mig_bytes_remaining(); | |
1221 | info->disk->total = blk_mig_bytes_total(); | |
1222 | } | |
1223 | } | |
1224 | ||
65ace060 | 1225 | static void fill_source_migration_info(MigrationInfo *info) |
5bb7910a | 1226 | { |
17549e84 | 1227 | MigrationState *s = migrate_get_current(); |
552de79b | 1228 | int state = qatomic_read(&s->state); |
fa3673e4 | 1229 | GSList *cur_blocker = migration_blockers[migrate_mode()]; |
17549e84 | 1230 | |
3af8554b | 1231 | info->blocked_reasons = NULL; |
3af8554b | 1232 | |
372043f3 MA |
1233 | /* |
1234 | * There are two types of reasons a migration might be blocked; | |
1235 | * a) devices marked in VMState as non-migratable, and | |
1236 | * b) Explicit migration blockers | |
1237 | * We need to add both of them here. | |
1238 | */ | |
1239 | qemu_savevm_non_migratable_list(&info->blocked_reasons); | |
3af8554b | 1240 | |
372043f3 MA |
1241 | while (cur_blocker) { |
1242 | QAPI_LIST_PREPEND(info->blocked_reasons, | |
1243 | g_strdup(error_get_pretty(cur_blocker->data))); | |
1244 | cur_blocker = g_slist_next(cur_blocker); | |
3af8554b | 1245 | } |
372043f3 | 1246 | info->has_blocked_reasons = info->blocked_reasons != NULL; |
3af8554b | 1247 | |
552de79b | 1248 | switch (state) { |
31194731 | 1249 | case MIGRATION_STATUS_NONE: |
17549e84 | 1250 | /* no migration has happened ever */ |
65ace060 AP |
1251 | /* do not overwrite destination migration status */ |
1252 | return; | |
31194731 | 1253 | case MIGRATION_STATUS_SETUP: |
29ae8a41 | 1254 | info->has_status = true; |
ed4fbd10 | 1255 | info->has_total_time = false; |
29ae8a41 | 1256 | break; |
31194731 HZ |
1257 | case MIGRATION_STATUS_ACTIVE: |
1258 | case MIGRATION_STATUS_CANCELLING: | |
9ec055ae | 1259 | case MIGRATION_STATUS_POSTCOPY_ACTIVE: |
31e06077 DDAG |
1260 | case MIGRATION_STATUS_PRE_SWITCHOVER: |
1261 | case MIGRATION_STATUS_DEVICE: | |
a688d2c1 | 1262 | case MIGRATION_STATUS_POSTCOPY_PAUSED: |
135b87b4 | 1263 | case MIGRATION_STATUS_POSTCOPY_RECOVER: |
640dfb14 WY |
1264 | /* TODO add some postcopy stats */ |
1265 | populate_time_info(info, s); | |
a22463a5 | 1266 | populate_ram_info(info, s); |
930ac04c | 1267 | populate_disk_info(info); |
38c482b4 | 1268 | migration_populate_vfio_info(info); |
17549e84 | 1269 | break; |
0b827d5e HZ |
1270 | case MIGRATION_STATUS_COLO: |
1271 | info->has_status = true; | |
1272 | /* TODO: display COLO specific information (checkpoint info etc.) */ | |
1273 | break; | |
31194731 | 1274 | case MIGRATION_STATUS_COMPLETED: |
640dfb14 | 1275 | populate_time_info(info, s); |
a22463a5 | 1276 | populate_ram_info(info, s); |
38c482b4 | 1277 | migration_populate_vfio_info(info); |
17549e84 | 1278 | break; |
31194731 | 1279 | case MIGRATION_STATUS_FAILED: |
791e7c82 | 1280 | info->has_status = true; |
17549e84 | 1281 | break; |
31194731 | 1282 | case MIGRATION_STATUS_CANCELLED: |
791e7c82 | 1283 | info->has_status = true; |
17549e84 | 1284 | break; |
c7e0acd5 JF |
1285 | case MIGRATION_STATUS_WAIT_UNPLUG: |
1286 | info->has_status = true; | |
1287 | break; | |
5bb7910a | 1288 | } |
552de79b | 1289 | info->status = state; |
c94143e5 PX |
1290 | |
1291 | QEMU_LOCK_GUARD(&s->error_mutex); | |
1292 | if (s->error) { | |
1293 | info->error_desc = g_strdup(error_get_pretty(s->error)); | |
1294 | } | |
5bb7910a AL |
1295 | } |
1296 | ||
65ace060 AP |
1297 | static void fill_destination_migration_info(MigrationInfo *info) |
1298 | { | |
1299 | MigrationIncomingState *mis = migration_incoming_get_current(); | |
1300 | ||
9aca82ba JQ |
1301 | if (mis->socket_address_list) { |
1302 | info->has_socket_address = true; | |
1303 | info->socket_address = | |
1304 | QAPI_CLONE(SocketAddressList, mis->socket_address_list); | |
1305 | } | |
1306 | ||
65ace060 AP |
1307 | switch (mis->state) { |
1308 | case MIGRATION_STATUS_NONE: | |
1309 | return; | |
65ace060 AP |
1310 | case MIGRATION_STATUS_SETUP: |
1311 | case MIGRATION_STATUS_CANCELLING: | |
1312 | case MIGRATION_STATUS_CANCELLED: | |
1313 | case MIGRATION_STATUS_ACTIVE: | |
1314 | case MIGRATION_STATUS_POSTCOPY_ACTIVE: | |
3c9928d9 PX |
1315 | case MIGRATION_STATUS_POSTCOPY_PAUSED: |
1316 | case MIGRATION_STATUS_POSTCOPY_RECOVER: | |
65ace060 AP |
1317 | case MIGRATION_STATUS_FAILED: |
1318 | case MIGRATION_STATUS_COLO: | |
1319 | info->has_status = true; | |
1320 | break; | |
1321 | case MIGRATION_STATUS_COMPLETED: | |
1322 | info->has_status = true; | |
1323 | fill_destination_postcopy_migration_info(info); | |
1324 | break; | |
1325 | } | |
1326 | info->status = mis->state; | |
1327 | } | |
1328 | ||
1329 | MigrationInfo *qmp_query_migrate(Error **errp) | |
1330 | { | |
1331 | MigrationInfo *info = g_malloc0(sizeof(*info)); | |
1332 | ||
1333 | fill_destination_migration_info(info); | |
1334 | fill_source_migration_info(info); | |
1335 | ||
1336 | return info; | |
1337 | } | |
1338 | ||
4886a1bc DDAG |
1339 | void qmp_migrate_start_postcopy(Error **errp) |
1340 | { | |
1341 | MigrationState *s = migrate_get_current(); | |
1342 | ||
16b0fd32 | 1343 | if (!migrate_postcopy()) { |
a54d340b | 1344 | error_setg(errp, "Enable postcopy with migrate_set_capability before" |
4886a1bc DDAG |
1345 | " the start of migration"); |
1346 | return; | |
1347 | } | |
1348 | ||
1349 | if (s->state == MIGRATION_STATUS_NONE) { | |
1350 | error_setg(errp, "Postcopy must be started after migration has been" | |
1351 | " started"); | |
1352 | return; | |
1353 | } | |
1354 | /* | |
1355 | * we don't error if migration has finished since that would be racy | |
1356 | * with issuing this command. | |
1357 | */ | |
d73415a3 | 1358 | qatomic_set(&s->start_postcopy, true); |
4886a1bc DDAG |
1359 | } |
1360 | ||
065e2813 AL |
1361 | /* shared migration helpers */ |
1362 | ||
48781e5b | 1363 | void migrate_set_state(int *state, int old_state, int new_state) |
51cf4c1a | 1364 | { |
a31fedee | 1365 | assert(new_state < MIGRATION_STATUS__MAX); |
d73415a3 | 1366 | if (qatomic_cmpxchg(state, old_state, new_state) == old_state) { |
a31fedee | 1367 | trace_migrate_set_state(MigrationStatus_str(new_state)); |
b05dc723 | 1368 | migrate_generate_event(new_state); |
51cf4c1a Z |
1369 | } |
1370 | } | |
1371 | ||
fd392cfa | 1372 | static void migrate_fd_cleanup(MigrationState *s) |
065e2813 | 1373 | { |
9d9babf7 SS |
1374 | MigrationEventType type; |
1375 | ||
83174765 PX |
1376 | g_free(s->hostname); |
1377 | s->hostname = NULL; | |
e3bf5e68 DH |
1378 | json_writer_free(s->vmdesc); |
1379 | s->vmdesc = NULL; | |
83174765 | 1380 | |
0ceccd85 PX |
1381 | qemu_savevm_state_cleanup(); |
1382 | ||
22b04245 FR |
1383 | close_return_path_on_source(s); |
1384 | ||
89a02a9f | 1385 | if (s->to_dst_file) { |
62df066f | 1386 | QEMUFile *tmp; |
f986c3d2 | 1387 | |
9013dca5 | 1388 | trace_migrate_fd_cleanup(); |
195801d7 | 1389 | bql_unlock(); |
1d34e4bf DDAG |
1390 | if (s->migration_thread_running) { |
1391 | qemu_thread_join(&s->thread); | |
1392 | s->migration_thread_running = false; | |
1393 | } | |
195801d7 | 1394 | bql_lock(); |
404a7c05 | 1395 | |
cde85c37 | 1396 | multifd_send_shutdown(); |
62df066f PX |
1397 | qemu_mutex_lock(&s->qemu_file_lock); |
1398 | tmp = s->to_dst_file; | |
89a02a9f | 1399 | s->to_dst_file = NULL; |
62df066f PX |
1400 | qemu_mutex_unlock(&s->qemu_file_lock); |
1401 | /* | |
1402 | * Close the file handle without the lock to make sure the | |
1403 | * critical section won't block for long. | |
1404 | */ | |
39675fff | 1405 | migration_ioc_unregister_yank_from_file(tmp); |
62df066f | 1406 | qemu_fclose(tmp); |
065e2813 AL |
1407 | } |
1408 | ||
3a6813b6 | 1409 | assert(!migration_is_active()); |
7a2c1721 | 1410 | |
94f5a437 | 1411 | if (s->state == MIGRATION_STATUS_CANCELLING) { |
48781e5b | 1412 | migrate_set_state(&s->state, MIGRATION_STATUS_CANCELLING, |
94f5a437 | 1413 | MIGRATION_STATUS_CANCELLED); |
7a2c1721 | 1414 | } |
a3fa1d78 | 1415 | |
87db1a7d JQ |
1416 | if (s->error) { |
1417 | /* It is used on info migrate. We can't free it */ | |
1418 | error_report_err(error_copy(s->error)); | |
1419 | } | |
9d9babf7 SS |
1420 | type = migration_has_failed(s) ? MIG_EVENT_PRECOPY_FAILED : |
1421 | MIG_EVENT_PRECOPY_DONE; | |
4af667f8 | 1422 | migration_call_notifiers(s, type, NULL); |
b1a87956 | 1423 | block_cleanup_parameters(); |
b5eea99e | 1424 | yank_unregister_instance(MIGRATION_YANK_INSTANCE); |
065e2813 AL |
1425 | } |
1426 | ||
fd392cfa YK |
1427 | static void migrate_fd_cleanup_bh(void *opaque) |
1428 | { | |
44d0d456 | 1429 | migrate_fd_cleanup(opaque); |
fd392cfa YK |
1430 | } |
1431 | ||
87db1a7d JQ |
1432 | void migrate_set_error(MigrationState *s, const Error *error) |
1433 | { | |
6e8a355d | 1434 | QEMU_LOCK_GUARD(&s->error_mutex); |
87db1a7d JQ |
1435 | if (!s->error) { |
1436 | s->error = error_copy(error); | |
1437 | } | |
87db1a7d JQ |
1438 | } |
1439 | ||
2b2f6f41 PX |
1440 | bool migrate_has_error(MigrationState *s) |
1441 | { | |
1442 | /* The lock is not helpful here, but still follow the rule */ | |
1443 | QEMU_LOCK_GUARD(&s->error_mutex); | |
1444 | return qatomic_read(&s->error); | |
1445 | } | |
1446 | ||
ca7bd082 PX |
1447 | static void migrate_error_free(MigrationState *s) |
1448 | { | |
1449 | QEMU_LOCK_GUARD(&s->error_mutex); | |
1450 | if (s->error) { | |
1451 | error_free(s->error); | |
1452 | s->error = NULL; | |
1453 | } | |
1454 | } | |
1455 | ||
aaf26bd3 | 1456 | static void migrate_fd_error(MigrationState *s, const Error *error) |
065e2813 | 1457 | { |
25174055 | 1458 | trace_migrate_fd_error(error_get_pretty(error)); |
89a02a9f | 1459 | assert(s->to_dst_file == NULL); |
48781e5b HZ |
1460 | migrate_set_state(&s->state, MIGRATION_STATUS_SETUP, |
1461 | MIGRATION_STATUS_FAILED); | |
87db1a7d | 1462 | migrate_set_error(s, error); |
458cf28e JQ |
1463 | } |
1464 | ||
0edda1c4 | 1465 | static void migrate_fd_cancel(MigrationState *s) |
065e2813 | 1466 | { |
6f2b811a | 1467 | int old_state ; |
7478fb0d | 1468 | |
9013dca5 | 1469 | trace_migrate_fd_cancel(); |
065e2813 | 1470 | |
43044ac0 PX |
1471 | WITH_QEMU_LOCK_GUARD(&s->qemu_file_lock) { |
1472 | if (s->rp_state.from_dst_file) { | |
1473 | /* shutdown the rp socket, so causing the rp thread to shutdown */ | |
1474 | qemu_file_shutdown(s->rp_state.from_dst_file); | |
1475 | } | |
70b20477 DDAG |
1476 | } |
1477 | ||
6f2b811a Z |
1478 | do { |
1479 | old_state = s->state; | |
392d87e2 | 1480 | if (!migration_is_running(old_state)) { |
6f2b811a Z |
1481 | break; |
1482 | } | |
a7b36b48 DDAG |
1483 | /* If the migration is paused, kick it out of the pause */ |
1484 | if (old_state == MIGRATION_STATUS_PRE_SWITCHOVER) { | |
1485 | qemu_sem_post(&s->pause_sem); | |
1486 | } | |
48781e5b | 1487 | migrate_set_state(&s->state, old_state, MIGRATION_STATUS_CANCELLING); |
31194731 | 1488 | } while (s->state != MIGRATION_STATUS_CANCELLING); |
a26ba26e DDAG |
1489 | |
1490 | /* | |
1491 | * If we're unlucky the migration code might be stuck somewhere in a | |
1492 | * send/write while the network has failed and is waiting to timeout; | |
1493 | * if we've got shutdown(2) available then we can force it to quit. | |
a26ba26e | 1494 | */ |
7478fb0d FR |
1495 | if (s->state == MIGRATION_STATUS_CANCELLING) { |
1496 | WITH_QEMU_LOCK_GUARD(&s->qemu_file_lock) { | |
1497 | if (s->to_dst_file) { | |
1498 | qemu_file_shutdown(s->to_dst_file); | |
1499 | } | |
1500 | } | |
a26ba26e | 1501 | } |
1d2acc31 HZ |
1502 | if (s->state == MIGRATION_STATUS_CANCELLING && s->block_inactive) { |
1503 | Error *local_err = NULL; | |
1504 | ||
3b717194 | 1505 | bdrv_activate_all(&local_err); |
1d2acc31 HZ |
1506 | if (local_err) { |
1507 | error_report_err(local_err); | |
1508 | } else { | |
1509 | s->block_inactive = false; | |
1510 | } | |
1511 | } | |
065e2813 AL |
1512 | } |
1513 | ||
6835f5a1 SS |
1514 | void migration_add_notifier_mode(NotifierWithReturn *notify, |
1515 | MigrationNotifyFunc func, MigMode mode) | |
1516 | { | |
1517 | notify->notify = (NotifierWithReturnFunc)func; | |
1518 | notifier_with_return_list_add(&migration_state_notifiers[mode], notify); | |
1519 | } | |
1520 | ||
3e775730 | 1521 | void migration_add_notifier(NotifierWithReturn *notify, |
5663dd3f | 1522 | MigrationNotifyFunc func) |
99a0db9b | 1523 | { |
6835f5a1 | 1524 | migration_add_notifier_mode(notify, func, MIG_MODE_NORMAL); |
99a0db9b GH |
1525 | } |
1526 | ||
3e775730 | 1527 | void migration_remove_notifier(NotifierWithReturn *notify) |
d9cda213 SS |
1528 | { |
1529 | if (notify->notify) { | |
3e775730 | 1530 | notifier_with_return_remove(notify); |
d9cda213 SS |
1531 | notify->notify = NULL; |
1532 | } | |
1533 | } | |
1534 | ||
4af667f8 SS |
1535 | int migration_call_notifiers(MigrationState *s, MigrationEventType type, |
1536 | Error **errp) | |
99a0db9b | 1537 | { |
6835f5a1 | 1538 | MigMode mode = s->parameters.mode; |
9d9babf7 | 1539 | MigrationEvent e; |
4af667f8 | 1540 | int ret; |
9d9babf7 SS |
1541 | |
1542 | e.type = type; | |
4af667f8 SS |
1543 | ret = notifier_with_return_list_notify(&migration_state_notifiers[mode], |
1544 | &e, errp); | |
1545 | assert(!ret || type == MIG_EVENT_PRECOPY_SETUP); | |
1546 | return ret; | |
99a0db9b GH |
1547 | } |
1548 | ||
02edd2e7 | 1549 | bool migration_in_setup(MigrationState *s) |
afe2df69 | 1550 | { |
31194731 | 1551 | return s->state == MIGRATION_STATUS_SETUP; |
afe2df69 GH |
1552 | } |
1553 | ||
7073693b | 1554 | bool migration_has_finished(MigrationState *s) |
99a0db9b | 1555 | { |
31194731 | 1556 | return s->state == MIGRATION_STATUS_COMPLETED; |
99a0db9b | 1557 | } |
0edda1c4 | 1558 | |
afe2df69 GH |
1559 | bool migration_has_failed(MigrationState *s) |
1560 | { | |
31194731 HZ |
1561 | return (s->state == MIGRATION_STATUS_CANCELLED || |
1562 | s->state == MIGRATION_STATUS_FAILED); | |
afe2df69 GH |
1563 | } |
1564 | ||
5727309d | 1565 | bool migration_in_postcopy(void) |
9ec055ae | 1566 | { |
5727309d JQ |
1567 | MigrationState *s = migrate_get_current(); |
1568 | ||
3748fef9 DDAG |
1569 | switch (s->state) { |
1570 | case MIGRATION_STATUS_POSTCOPY_ACTIVE: | |
1571 | case MIGRATION_STATUS_POSTCOPY_PAUSED: | |
1572 | case MIGRATION_STATUS_POSTCOPY_RECOVER: | |
1573 | return true; | |
1574 | default: | |
1575 | return false; | |
1576 | } | |
9ec055ae DDAG |
1577 | } |
1578 | ||
f8c543e8 PX |
1579 | bool migration_postcopy_is_alive(int state) |
1580 | { | |
1581 | switch (state) { | |
1582 | case MIGRATION_STATUS_POSTCOPY_ACTIVE: | |
1583 | case MIGRATION_STATUS_POSTCOPY_RECOVER: | |
1584 | return true; | |
1585 | default: | |
1586 | return false; | |
1587 | } | |
1588 | } | |
1589 | ||
06df2e69 DH |
1590 | bool migration_in_incoming_postcopy(void) |
1591 | { | |
1592 | PostcopyState ps = postcopy_state_get(); | |
1593 | ||
1594 | return ps >= POSTCOPY_INCOMING_DISCARD && ps < POSTCOPY_INCOMING_END; | |
1595 | } | |
1596 | ||
80fe315c DH |
1597 | bool migration_incoming_postcopy_advised(void) |
1598 | { | |
1599 | PostcopyState ps = postcopy_state_get(); | |
1600 | ||
1601 | return ps >= POSTCOPY_INCOMING_ADVISE && ps < POSTCOPY_INCOMING_END; | |
1602 | } | |
1603 | ||
1a8e44a8 AG |
1604 | bool migration_in_bg_snapshot(void) |
1605 | { | |
1a8e44a8 | 1606 | return migrate_background_snapshot() && |
7dcb3c87 | 1607 | migration_is_setup_or_active(); |
1a8e44a8 AG |
1608 | } |
1609 | ||
fab35005 | 1610 | bool migration_is_idle(void) |
fe44dc91 | 1611 | { |
daff7f0b MA |
1612 | MigrationState *s = current_migration; |
1613 | ||
1614 | if (!s) { | |
1615 | return true; | |
1616 | } | |
fe44dc91 AA |
1617 | |
1618 | switch (s->state) { | |
1619 | case MIGRATION_STATUS_NONE: | |
1620 | case MIGRATION_STATUS_CANCELLED: | |
1621 | case MIGRATION_STATUS_COMPLETED: | |
1622 | case MIGRATION_STATUS_FAILED: | |
1623 | return true; | |
1624 | case MIGRATION_STATUS_SETUP: | |
1625 | case MIGRATION_STATUS_CANCELLING: | |
1626 | case MIGRATION_STATUS_ACTIVE: | |
1627 | case MIGRATION_STATUS_POSTCOPY_ACTIVE: | |
1628 | case MIGRATION_STATUS_COLO: | |
31e06077 DDAG |
1629 | case MIGRATION_STATUS_PRE_SWITCHOVER: |
1630 | case MIGRATION_STATUS_DEVICE: | |
c7e0acd5 | 1631 | case MIGRATION_STATUS_WAIT_UNPLUG: |
fe44dc91 AA |
1632 | return false; |
1633 | case MIGRATION_STATUS__MAX: | |
1634 | g_assert_not_reached(); | |
1635 | } | |
1636 | ||
1637 | return false; | |
1638 | } | |
1639 | ||
3a6813b6 | 1640 | bool migration_is_active(void) |
8f8d528e | 1641 | { |
3a6813b6 SS |
1642 | MigrationState *s = current_migration; |
1643 | ||
8f8d528e WY |
1644 | return (s->state == MIGRATION_STATUS_ACTIVE || |
1645 | s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE); | |
1646 | } | |
1647 | ||
9867d4dd SS |
1648 | bool migrate_mode_is_cpr(MigrationState *s) |
1649 | { | |
1650 | return s->parameters.mode == MIG_MODE_CPR_REBOOT; | |
1651 | } | |
1652 | ||
08fc4cb5 | 1653 | int migrate_init(MigrationState *s, Error **errp) |
0edda1c4 | 1654 | { |
08fc4cb5 AH |
1655 | int ret; |
1656 | ||
1657 | ret = qemu_savevm_state_prepare(errp); | |
1658 | if (ret) { | |
1659 | return ret; | |
1660 | } | |
1661 | ||
389775d1 DDAG |
1662 | /* |
1663 | * Reinitialise all migration state, except | |
1664 | * parameters/capabilities that the user set, and | |
1665 | * locks. | |
1666 | */ | |
89a02a9f | 1667 | s->to_dst_file = NULL; |
389775d1 | 1668 | s->state = MIGRATION_STATUS_NONE; |
389775d1 | 1669 | s->rp_state.from_dst_file = NULL; |
389775d1 | 1670 | s->mbps = 0.0; |
aecbfe9c | 1671 | s->pages_per_second = 0.0; |
389775d1 DDAG |
1672 | s->downtime = 0; |
1673 | s->expected_downtime = 0; | |
389775d1 | 1674 | s->setup_time = 0; |
389775d1 DDAG |
1675 | s->start_postcopy = false; |
1676 | s->migration_thread_running = false; | |
d59ce6f3 DB |
1677 | error_free(s->error); |
1678 | s->error = NULL; | |
2aae1eb8 | 1679 | s->vmdesc = NULL; |
389775d1 | 1680 | |
48781e5b | 1681 | migrate_set_state(&s->state, MIGRATION_STATUS_NONE, MIGRATION_STATUS_SETUP); |
0edda1c4 | 1682 | |
4af246a3 PX |
1683 | s->start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); |
1684 | s->total_time = 0; | |
f4584076 | 1685 | s->vm_old_state = -1; |
b15df1ae PX |
1686 | s->iteration_initial_bytes = 0; |
1687 | s->threshold_size = 0; | |
1b4adb10 | 1688 | s->switchover_acked = false; |
27fd25b0 | 1689 | s->rdma_migration = false; |
f543aa22 | 1690 | /* |
809f188a | 1691 | * set mig_stats memory to zero for a new migration |
f543aa22 AH |
1692 | */ |
1693 | memset(&mig_stats, 0, sizeof(mig_stats)); | |
f543aa22 | 1694 | migration_reset_vfio_bytes_transferred(); |
08fc4cb5 AH |
1695 | |
1696 | return 0; | |
0edda1c4 | 1697 | } |
cab30143 | 1698 | |
fa3673e4 | 1699 | static bool is_busy(Error **reasonp, Error **errp) |
fa2756b7 | 1700 | { |
fa3673e4 SS |
1701 | ERRP_GUARD(); |
1702 | ||
4c170330 PX |
1703 | /* Snapshots are similar to migrations, so check RUN_STATE_SAVE_VM too. */ |
1704 | if (runstate_check(RUN_STATE_SAVE_VM) || !migration_is_idle()) { | |
c8a7fc51 | 1705 | error_propagate_prepend(errp, *reasonp, |
4c170330 PX |
1706 | "disallowing migration blocker " |
1707 | "(migration/snapshot in progress) for: "); | |
c8a7fc51 | 1708 | *reasonp = NULL; |
fa3673e4 | 1709 | return true; |
fe44dc91 | 1710 | } |
fa3673e4 | 1711 | return false; |
fa2756b7 AL |
1712 | } |
1713 | ||
fa3673e4 | 1714 | static bool is_only_migratable(Error **reasonp, Error **errp, int modes) |
60fd6801 | 1715 | { |
fa3673e4 SS |
1716 | ERRP_GUARD(); |
1717 | ||
1718 | if (only_migratable && (modes & BIT(MIG_MODE_NORMAL))) { | |
c8a7fc51 | 1719 | error_propagate_prepend(errp, *reasonp, |
60fd6801 PX |
1720 | "disallowing migration blocker " |
1721 | "(--only-migratable) for: "); | |
c8a7fc51 | 1722 | *reasonp = NULL; |
fa3673e4 SS |
1723 | return true; |
1724 | } | |
1725 | return false; | |
1726 | } | |
1727 | ||
1728 | static int get_modes(MigMode mode, va_list ap) | |
1729 | { | |
1730 | int modes = 0; | |
1731 | ||
1732 | while (mode != -1 && mode != MIG_MODE_ALL) { | |
1733 | assert(mode >= MIG_MODE_NORMAL && mode < MIG_MODE__MAX); | |
1734 | modes |= BIT(mode); | |
1735 | mode = va_arg(ap, MigMode); | |
1736 | } | |
1737 | if (mode == MIG_MODE_ALL) { | |
1738 | modes = BIT(MIG_MODE__MAX) - 1; | |
1739 | } | |
1740 | return modes; | |
1741 | } | |
1742 | ||
1743 | static int add_blockers(Error **reasonp, Error **errp, int modes) | |
1744 | { | |
1745 | for (MigMode mode = 0; mode < MIG_MODE__MAX; mode++) { | |
1746 | if (modes & BIT(mode)) { | |
1747 | migration_blockers[mode] = g_slist_prepend(migration_blockers[mode], | |
1748 | *reasonp); | |
1749 | } | |
1750 | } | |
1751 | return 0; | |
1752 | } | |
1753 | ||
1754 | int migrate_add_blocker(Error **reasonp, Error **errp) | |
1755 | { | |
1756 | return migrate_add_blocker_modes(reasonp, errp, MIG_MODE_ALL); | |
1757 | } | |
1758 | ||
1759 | int migrate_add_blocker_normal(Error **reasonp, Error **errp) | |
1760 | { | |
1761 | return migrate_add_blocker_modes(reasonp, errp, MIG_MODE_NORMAL, -1); | |
1762 | } | |
1763 | ||
1764 | int migrate_add_blocker_modes(Error **reasonp, Error **errp, MigMode mode, ...) | |
1765 | { | |
1766 | int modes; | |
1767 | va_list ap; | |
1768 | ||
1769 | va_start(ap, mode); | |
1770 | modes = get_modes(mode, ap); | |
1771 | va_end(ap); | |
1772 | ||
1773 | if (is_only_migratable(reasonp, errp, modes)) { | |
60fd6801 | 1774 | return -EACCES; |
fa3673e4 SS |
1775 | } else if (is_busy(reasonp, errp)) { |
1776 | return -EBUSY; | |
60fd6801 | 1777 | } |
fa3673e4 SS |
1778 | return add_blockers(reasonp, errp, modes); |
1779 | } | |
60fd6801 | 1780 | |
fa3673e4 SS |
1781 | int migrate_add_blocker_internal(Error **reasonp, Error **errp) |
1782 | { | |
1783 | int modes = BIT(MIG_MODE__MAX) - 1; | |
1784 | ||
1785 | if (is_busy(reasonp, errp)) { | |
1786 | return -EBUSY; | |
1787 | } | |
1788 | return add_blockers(reasonp, errp, modes); | |
60fd6801 PX |
1789 | } |
1790 | ||
c8a7fc51 | 1791 | void migrate_del_blocker(Error **reasonp) |
fa2756b7 | 1792 | { |
c8a7fc51 | 1793 | if (*reasonp) { |
fa3673e4 SS |
1794 | for (MigMode mode = 0; mode < MIG_MODE__MAX; mode++) { |
1795 | migration_blockers[mode] = g_slist_remove(migration_blockers[mode], | |
1796 | *reasonp); | |
1797 | } | |
c8a7fc51 SS |
1798 | error_free(*reasonp); |
1799 | *reasonp = NULL; | |
1800 | } | |
fa2756b7 AL |
1801 | } |
1802 | ||
074dbce5 HG |
1803 | void qmp_migrate_incoming(const char *uri, bool has_channels, |
1804 | MigrationChannelList *channels, Error **errp) | |
bf1ae1f4 DDAG |
1805 | { |
1806 | Error *local_err = NULL; | |
4debb5f5 | 1807 | static bool once = true; |
bf1ae1f4 | 1808 | |
4debb5f5 DDAG |
1809 | if (!once) { |
1810 | error_setg(errp, "The incoming migration has already been started"); | |
603d5a42 | 1811 | return; |
4debb5f5 | 1812 | } |
e69d50d6 PB |
1813 | if (!runstate_check(RUN_STATE_INMIGRATE)) { |
1814 | error_setg(errp, "'-incoming' was not specified on the command line"); | |
1815 | return; | |
1816 | } | |
bf1ae1f4 | 1817 | |
cc48c587 PX |
1818 | if (!yank_register_instance(MIGRATION_YANK_INSTANCE, errp)) { |
1819 | return; | |
1820 | } | |
1821 | ||
074dbce5 | 1822 | qemu_start_incoming_migration(uri, has_channels, channels, &local_err); |
bf1ae1f4 DDAG |
1823 | |
1824 | if (local_err) { | |
cc48c587 | 1825 | yank_unregister_instance(MIGRATION_YANK_INSTANCE); |
bf1ae1f4 DDAG |
1826 | error_propagate(errp, local_err); |
1827 | return; | |
1828 | } | |
1829 | ||
4debb5f5 | 1830 | once = false; |
bf1ae1f4 DDAG |
1831 | } |
1832 | ||
02affd41 PX |
1833 | void qmp_migrate_recover(const char *uri, Error **errp) |
1834 | { | |
1835 | MigrationIncomingState *mis = migration_incoming_get_current(); | |
1836 | ||
b7f9afd4 PX |
1837 | /* |
1838 | * Don't even bother to use ERRP_GUARD() as it _must_ always be set by | |
1839 | * callers (no one should ignore a recover failure); if there is, it's a | |
1840 | * programming error. | |
1841 | */ | |
1842 | assert(errp); | |
1843 | ||
02affd41 PX |
1844 | if (mis->state != MIGRATION_STATUS_POSTCOPY_PAUSED) { |
1845 | error_setg(errp, "Migrate recover can only be run " | |
1846 | "when postcopy is paused."); | |
1847 | return; | |
1848 | } | |
1849 | ||
08401c04 PX |
1850 | /* If there's an existing transport, release it */ |
1851 | migration_incoming_transport_cleanup(mis); | |
02affd41 PX |
1852 | |
1853 | /* | |
1854 | * Note that this call will never start a real migration; it will | |
1855 | * only re-setup the migration stream and poke existing migration | |
1856 | * to continue using that newly established channel. | |
1857 | */ | |
074dbce5 | 1858 | qemu_start_incoming_migration(uri, false, NULL, errp); |
02affd41 PX |
1859 | } |
1860 | ||
bfbf89c2 PX |
1861 | void qmp_migrate_pause(Error **errp) |
1862 | { | |
1863 | MigrationState *ms = migrate_get_current(); | |
1864 | MigrationIncomingState *mis = migration_incoming_get_current(); | |
7478fb0d | 1865 | int ret = 0; |
bfbf89c2 | 1866 | |
f8c543e8 | 1867 | if (migration_postcopy_is_alive(ms->state)) { |
bfbf89c2 | 1868 | /* Source side, during postcopy */ |
f8c543e8 PX |
1869 | Error *error = NULL; |
1870 | ||
1871 | /* Tell the core migration that we're pausing */ | |
1872 | error_setg(&error, "Postcopy migration is paused by the user"); | |
1873 | migrate_set_error(ms, error); | |
1874 | error_free(error); | |
1875 | ||
bfbf89c2 | 1876 | qemu_mutex_lock(&ms->qemu_file_lock); |
7478fb0d FR |
1877 | if (ms->to_dst_file) { |
1878 | ret = qemu_file_shutdown(ms->to_dst_file); | |
1879 | } | |
bfbf89c2 PX |
1880 | qemu_mutex_unlock(&ms->qemu_file_lock); |
1881 | if (ret) { | |
1882 | error_setg(errp, "Failed to pause source migration"); | |
1883 | } | |
f8c543e8 PX |
1884 | |
1885 | /* | |
1886 | * Kick the migration thread out of any waiting windows (on behalf | |
1887 | * of the rp thread). | |
1888 | */ | |
1889 | migration_rp_kick(ms); | |
1890 | ||
bfbf89c2 PX |
1891 | return; |
1892 | } | |
1893 | ||
f8c543e8 | 1894 | if (migration_postcopy_is_alive(mis->state)) { |
bfbf89c2 PX |
1895 | ret = qemu_file_shutdown(mis->from_src_file); |
1896 | if (ret) { | |
1897 | error_setg(errp, "Failed to pause destination migration"); | |
1898 | } | |
1899 | return; | |
1900 | } | |
1901 | ||
1902 | error_setg(errp, "migrate-pause is currently only supported " | |
f8c543e8 | 1903 | "during postcopy-active or postcopy-recover state"); |
bfbf89c2 PX |
1904 | } |
1905 | ||
24f3902b GK |
1906 | bool migration_is_blocked(Error **errp) |
1907 | { | |
fa3673e4 SS |
1908 | GSList *blockers = migration_blockers[migrate_mode()]; |
1909 | ||
24f3902b GK |
1910 | if (qemu_savevm_state_blocked(errp)) { |
1911 | return true; | |
1912 | } | |
1913 | ||
fa3673e4 SS |
1914 | if (blockers) { |
1915 | error_propagate(errp, error_copy(blockers->data)); | |
24f3902b GK |
1916 | return true; |
1917 | } | |
1918 | ||
1919 | return false; | |
1920 | } | |
1921 | ||
d3e35b8f PX |
1922 | /* Returns true if continue to migrate, or false if error detected */ |
1923 | static bool migrate_prepare(MigrationState *s, bool blk, bool blk_inc, | |
1924 | bool resume, Error **errp) | |
cab30143 | 1925 | { |
40101f32 JQ |
1926 | if (blk_inc) { |
1927 | warn_report("parameter 'inc' is deprecated;" | |
1928 | " use blockdev-mirror with NBD instead"); | |
1929 | } | |
1930 | ||
8846b5bf JQ |
1931 | if (blk) { |
1932 | warn_report("parameter 'blk' is deprecated;" | |
1933 | " use blockdev-mirror with NBD instead"); | |
1934 | } | |
1935 | ||
d3e35b8f PX |
1936 | if (resume) { |
1937 | if (s->state != MIGRATION_STATUS_POSTCOPY_PAUSED) { | |
1938 | error_setg(errp, "Cannot resume if there is no " | |
1939 | "paused migration"); | |
1940 | return false; | |
1941 | } | |
97ca211c PX |
1942 | |
1943 | /* | |
1944 | * Postcopy recovery won't work well with release-ram | |
1945 | * capability since release-ram will drop the page buffer as | |
1946 | * long as the page is put into the send buffer. So if there | |
1947 | * is a network failure happened, any page buffers that have | |
1948 | * not yet reached the destination VM but have already been | |
1949 | * sent from the source VM will be lost forever. Let's refuse | |
1950 | * the client from resuming such a postcopy migration. | |
1951 | * Luckily release-ram was designed to only be used when src | |
1952 | * and destination VMs are on the same host, so it should be | |
1953 | * fine. | |
1954 | */ | |
1955 | if (migrate_release_ram()) { | |
1956 | error_setg(errp, "Postcopy recovery cannot work " | |
1957 | "when release-ram capability is set"); | |
1958 | return false; | |
1959 | } | |
1960 | ||
d3e35b8f PX |
1961 | /* This is a resume, skip init status */ |
1962 | return true; | |
1963 | } | |
cab30143 | 1964 | |
392d87e2 | 1965 | if (migration_is_running(s->state)) { |
c6bd8c70 | 1966 | error_setg(errp, QERR_MIGRATION_ACTIVE); |
d3e35b8f | 1967 | return false; |
cab30143 | 1968 | } |
d3e35b8f | 1969 | |
ca99993a DDAG |
1970 | if (runstate_check(RUN_STATE_INMIGRATE)) { |
1971 | error_setg(errp, "Guest is waiting for an incoming migration"); | |
d3e35b8f | 1972 | return false; |
ca99993a DDAG |
1973 | } |
1974 | ||
36d0fe65 T |
1975 | if (runstate_check(RUN_STATE_POSTMIGRATE)) { |
1976 | error_setg(errp, "Can't migrate the vm that was paused due to " | |
1977 | "previous migration"); | |
1978 | return false; | |
1979 | } | |
1980 | ||
06152b89 WR |
1981 | if (kvm_hwpoisoned_mem()) { |
1982 | error_setg(errp, "Can't migrate this vm with hardware poisoned memory, " | |
1983 | "please reboot the vm and try again"); | |
1984 | return false; | |
1985 | } | |
1986 | ||
24f3902b | 1987 | if (migration_is_blocked(errp)) { |
d3e35b8f | 1988 | return false; |
fa2756b7 AL |
1989 | } |
1990 | ||
4ed49feb FR |
1991 | if (migrate_mapped_ram()) { |
1992 | if (migrate_tls()) { | |
1993 | error_setg(errp, "Cannot use TLS with mapped-ram"); | |
1994 | return false; | |
1995 | } | |
f427d90b FR |
1996 | |
1997 | if (migrate_multifd_compression()) { | |
1998 | error_setg(errp, "Cannot use compression with mapped-ram"); | |
1999 | return false; | |
2000 | } | |
4ed49feb FR |
2001 | } |
2002 | ||
cbdafc1b SS |
2003 | if (migrate_mode_is_cpr(s)) { |
2004 | const char *conflict = NULL; | |
2005 | ||
2006 | if (migrate_postcopy()) { | |
2007 | conflict = "postcopy"; | |
2008 | } else if (migrate_background_snapshot()) { | |
2009 | conflict = "background snapshot"; | |
2010 | } else if (migrate_colo()) { | |
2011 | conflict = "COLO"; | |
2012 | } | |
2013 | ||
2014 | if (conflict) { | |
2015 | error_setg(errp, "Cannot use %s with CPR", conflict); | |
2016 | return false; | |
2017 | } | |
2018 | } | |
2019 | ||
d3e35b8f | 2020 | if (blk || blk_inc) { |
5e804644 | 2021 | if (migrate_colo()) { |
3ba02445 RL |
2022 | error_setg(errp, "No disk migration is required in COLO mode"); |
2023 | return false; | |
2024 | } | |
6f8be708 | 2025 | if (migrate_block() || migrate_block_incremental()) { |
2833c59b JQ |
2026 | error_setg(errp, "Command options are incompatible with " |
2027 | "current migration capabilities"); | |
d3e35b8f | 2028 | return false; |
2833c59b | 2029 | } |
b6f4c0c7 | 2030 | if (!migrate_cap_set(MIGRATION_CAPABILITY_BLOCK, true, errp)) { |
d3e35b8f | 2031 | return false; |
2833c59b JQ |
2032 | } |
2033 | s->must_remove_block_options = true; | |
2034 | } | |
2035 | ||
d3e35b8f | 2036 | if (blk_inc) { |
87c22901 | 2037 | migrate_set_block_incremental(true); |
2833c59b JQ |
2038 | } |
2039 | ||
08fc4cb5 AH |
2040 | if (migrate_init(s, errp)) { |
2041 | return false; | |
2042 | } | |
cab30143 | 2043 | |
d3e35b8f PX |
2044 | return true; |
2045 | } | |
2046 | ||
074dbce5 HG |
2047 | void qmp_migrate(const char *uri, bool has_channels, |
2048 | MigrationChannelList *channels, bool has_blk, bool blk, | |
d3e35b8f PX |
2049 | bool has_inc, bool inc, bool has_detach, bool detach, |
2050 | bool has_resume, bool resume, Error **errp) | |
2051 | { | |
8c69ae9e | 2052 | bool resume_requested; |
d3e35b8f PX |
2053 | Error *local_err = NULL; |
2054 | MigrationState *s = migrate_get_current(); | |
bc1d54ee | 2055 | g_autoptr(MigrationChannel) channel = NULL; |
5994024f | 2056 | MigrationAddress *addr = NULL; |
d3e35b8f | 2057 | |
074dbce5 HG |
2058 | /* |
2059 | * Having preliminary checks for uri and channel | |
2060 | */ | |
0770ad43 HG |
2061 | if (!uri == !channels) { |
2062 | error_setg(errp, "need either 'uri' or 'channels' argument"); | |
074dbce5 | 2063 | return; |
0770ad43 HG |
2064 | } |
2065 | ||
2066 | if (channels) { | |
5994024f HG |
2067 | /* To verify that Migrate channel list has only item */ |
2068 | if (channels->next) { | |
2069 | error_setg(errp, "Channel list has more than one entries"); | |
2070 | return; | |
2071 | } | |
bc1d54ee | 2072 | addr = channels->value->addr; |
0770ad43 HG |
2073 | } |
2074 | ||
2075 | if (uri) { | |
5994024f HG |
2076 | /* caller uses the old URI syntax */ |
2077 | if (!migrate_uri_parse(uri, &channel, errp)) { | |
2078 | return; | |
2079 | } | |
bc1d54ee | 2080 | addr = channel->addr; |
074dbce5 | 2081 | } |
d6f74fd1 | 2082 | |
d95533e1 | 2083 | /* transport mechanism not suitable for migration? */ |
5994024f | 2084 | if (!migration_channels_and_transport_compatible(addr, errp)) { |
72a8192e HG |
2085 | return; |
2086 | } | |
2087 | ||
8c69ae9e | 2088 | resume_requested = has_resume && resume; |
d3e35b8f | 2089 | if (!migrate_prepare(s, has_blk && blk, has_inc && inc, |
8c69ae9e | 2090 | resume_requested, errp)) { |
d3e35b8f PX |
2091 | /* Error detected, put into errp */ |
2092 | return; | |
2093 | } | |
2094 | ||
8c69ae9e | 2095 | if (!resume_requested) { |
b5eea99e LS |
2096 | if (!yank_register_instance(MIGRATION_YANK_INSTANCE, errp)) { |
2097 | return; | |
2098 | } | |
2099 | } | |
2100 | ||
5994024f HG |
2101 | if (addr->transport == MIGRATION_ADDRESS_TYPE_SOCKET) { |
2102 | SocketAddress *saddr = &addr->u.socket; | |
34dfc5e4 HG |
2103 | if (saddr->type == SOCKET_ADDRESS_TYPE_INET || |
2104 | saddr->type == SOCKET_ADDRESS_TYPE_UNIX || | |
2105 | saddr->type == SOCKET_ADDRESS_TYPE_VSOCK) { | |
2106 | socket_start_outgoing_migration(s, saddr, &local_err); | |
2107 | } else if (saddr->type == SOCKET_ADDRESS_TYPE_FD) { | |
2108 | fd_start_outgoing_migration(s, saddr->u.fd.str, &local_err); | |
2109 | } | |
2da776db | 2110 | #ifdef CONFIG_RDMA |
5994024f HG |
2111 | } else if (addr->transport == MIGRATION_ADDRESS_TYPE_RDMA) { |
2112 | rdma_start_outgoing_migration(s, &addr->u.rdma, &local_err); | |
2da776db | 2113 | #endif |
5994024f HG |
2114 | } else if (addr->transport == MIGRATION_ADDRESS_TYPE_EXEC) { |
2115 | exec_start_outgoing_migration(s, addr->u.exec.args, &local_err); | |
2116 | } else if (addr->transport == MIGRATION_ADDRESS_TYPE_FILE) { | |
2117 | file_start_outgoing_migration(s, &addr->u.file, &local_err); | |
99a0db9b | 2118 | } else { |
908927db | 2119 | error_setg(&local_err, QERR_INVALID_PARAMETER_VALUE, "uri", |
c6bd8c70 | 2120 | "a valid migration protocol"); |
48781e5b HZ |
2121 | migrate_set_state(&s->state, MIGRATION_STATUS_SETUP, |
2122 | MIGRATION_STATUS_FAILED); | |
b1a87956 | 2123 | block_cleanup_parameters(); |
cab30143 JQ |
2124 | } |
2125 | ||
f37afb5a | 2126 | if (local_err) { |
8c69ae9e | 2127 | if (!resume_requested) { |
b5eea99e LS |
2128 | yank_unregister_instance(MIGRATION_YANK_INSTANCE); |
2129 | } | |
d59ce6f3 | 2130 | migrate_fd_error(s, local_err); |
f37afb5a | 2131 | error_propagate(errp, local_err); |
e1c37d0e | 2132 | return; |
1299c631 | 2133 | } |
cab30143 JQ |
2134 | } |
2135 | ||
6cdedb07 | 2136 | void qmp_migrate_cancel(Error **errp) |
cab30143 | 2137 | { |
458fecca | 2138 | migration_cancel(NULL); |
cab30143 JQ |
2139 | } |
2140 | ||
89cfc02c DDAG |
2141 | void qmp_migrate_continue(MigrationStatus state, Error **errp) |
2142 | { | |
2143 | MigrationState *s = migrate_get_current(); | |
2144 | if (s->state != state) { | |
2145 | error_setg(errp, "Migration not in expected state: %s", | |
2146 | MigrationStatus_str(s->state)); | |
2147 | return; | |
2148 | } | |
2149 | qemu_sem_post(&s->pause_sem); | |
2150 | } | |
2151 | ||
f8c543e8 | 2152 | int migration_rp_wait(MigrationState *s) |
5e79a4bf | 2153 | { |
f8c543e8 PX |
2154 | /* If migration has failure already, ignore the wait */ |
2155 | if (migrate_has_error(s)) { | |
2156 | return -1; | |
2157 | } | |
2158 | ||
5e79a4bf | 2159 | qemu_sem_wait(&s->rp_state.rp_sem); |
f8c543e8 PX |
2160 | |
2161 | /* After wait, double check that there's no failure */ | |
2162 | if (migrate_has_error(s)) { | |
2163 | return -1; | |
2164 | } | |
2165 | ||
2166 | return 0; | |
5e79a4bf PX |
2167 | } |
2168 | ||
2169 | void migration_rp_kick(MigrationState *s) | |
2170 | { | |
2171 | qemu_sem_post(&s->rp_state.rp_sem); | |
2172 | } | |
2173 | ||
70b20477 DDAG |
2174 | static struct rp_cmd_args { |
2175 | ssize_t len; /* -1 = variable */ | |
2176 | const char *name; | |
2177 | } rp_cmd_args[] = { | |
2178 | [MIG_RP_MSG_INVALID] = { .len = -1, .name = "INVALID" }, | |
2179 | [MIG_RP_MSG_SHUT] = { .len = 4, .name = "SHUT" }, | |
2180 | [MIG_RP_MSG_PONG] = { .len = 4, .name = "PONG" }, | |
1e2d90eb DDAG |
2181 | [MIG_RP_MSG_REQ_PAGES] = { .len = 12, .name = "REQ_PAGES" }, |
2182 | [MIG_RP_MSG_REQ_PAGES_ID] = { .len = -1, .name = "REQ_PAGES_ID" }, | |
a335debb | 2183 | [MIG_RP_MSG_RECV_BITMAP] = { .len = -1, .name = "RECV_BITMAP" }, |
13955b89 | 2184 | [MIG_RP_MSG_RESUME_ACK] = { .len = 4, .name = "RESUME_ACK" }, |
1b4adb10 | 2185 | [MIG_RP_MSG_SWITCHOVER_ACK] = { .len = 0, .name = "SWITCHOVER_ACK" }, |
70b20477 DDAG |
2186 | [MIG_RP_MSG_MAX] = { .len = -1, .name = "MAX" }, |
2187 | }; | |
2188 | ||
1e2d90eb DDAG |
2189 | /* |
2190 | * Process a request for pages received on the return path, | |
2191 | * We're allowed to send more than requested (e.g. to round to our page size) | |
2192 | * and we don't need to send pages that have already been sent. | |
2193 | */ | |
7aa6070d PX |
2194 | static void |
2195 | migrate_handle_rp_req_pages(MigrationState *ms, const char* rbname, | |
2196 | ram_addr_t start, size_t len, Error **errp) | |
1e2d90eb | 2197 | { |
8e3b0cbb | 2198 | long our_host_ps = qemu_real_host_page_size(); |
6c595cde | 2199 | |
1e2d90eb | 2200 | trace_migrate_handle_rp_req_pages(rbname, start, len); |
6c595cde DDAG |
2201 | |
2202 | /* | |
2203 | * Since we currently insist on matching page sizes, just sanity check | |
2204 | * we're being asked for whole host pages. | |
2205 | */ | |
7648297d DH |
2206 | if (!QEMU_IS_ALIGNED(start, our_host_ps) || |
2207 | !QEMU_IS_ALIGNED(len, our_host_ps)) { | |
7aa6070d PX |
2208 | error_setg(errp, "MIG_RP_MSG_REQ_PAGES: Misaligned page request, start:" |
2209 | RAM_ADDR_FMT " len: %zd", start, len); | |
6c595cde DDAG |
2210 | return; |
2211 | } | |
2212 | ||
7aa6070d | 2213 | ram_save_queue_pages(rbname, start, len, errp); |
1e2d90eb DDAG |
2214 | } |
2215 | ||
88577f32 PX |
2216 | static bool migrate_handle_rp_recv_bitmap(MigrationState *s, char *block_name, |
2217 | Error **errp) | |
a335debb PX |
2218 | { |
2219 | RAMBlock *block = qemu_ram_block_by_name(block_name); | |
2220 | ||
2221 | if (!block) { | |
7aa6070d PX |
2222 | error_setg(errp, "MIG_RP_MSG_RECV_BITMAP has invalid block name '%s'", |
2223 | block_name); | |
88577f32 | 2224 | return false; |
a335debb PX |
2225 | } |
2226 | ||
2227 | /* Fetch the received bitmap and refresh the dirty bitmap */ | |
7aa6070d | 2228 | return ram_dirty_bitmap_reload(s, block, errp); |
a335debb PX |
2229 | } |
2230 | ||
88577f32 PX |
2231 | static bool migrate_handle_rp_resume_ack(MigrationState *s, |
2232 | uint32_t value, Error **errp) | |
13955b89 PX |
2233 | { |
2234 | trace_source_return_path_thread_resume_ack(value); | |
2235 | ||
2236 | if (value != MIGRATION_RESUME_ACK_VALUE) { | |
7aa6070d | 2237 | error_setg(errp, "illegal resume_ack value %"PRIu32, value); |
88577f32 | 2238 | return false; |
13955b89 PX |
2239 | } |
2240 | ||
2241 | /* Now both sides are active. */ | |
2242 | migrate_set_state(&s->state, MIGRATION_STATUS_POSTCOPY_RECOVER, | |
2243 | MIGRATION_STATUS_POSTCOPY_ACTIVE); | |
2244 | ||
94190696 | 2245 | /* Notify send thread that time to continue send pages */ |
5e79a4bf | 2246 | migration_rp_kick(s); |
13955b89 | 2247 | |
88577f32 | 2248 | return true; |
13955b89 PX |
2249 | } |
2250 | ||
93589827 PX |
2251 | /* |
2252 | * Release ms->rp_state.from_dst_file (and postcopy_qemufile_src if | |
2253 | * existed) in a safe way. | |
2254 | */ | |
2255 | static void migration_release_dst_files(MigrationState *ms) | |
43044ac0 PX |
2256 | { |
2257 | QEMUFile *file; | |
2258 | ||
2259 | WITH_QEMU_LOCK_GUARD(&ms->qemu_file_lock) { | |
2260 | /* | |
2261 | * Reset the from_dst_file pointer first before releasing it, as we | |
2262 | * can't block within lock section | |
2263 | */ | |
2264 | file = ms->rp_state.from_dst_file; | |
2265 | ms->rp_state.from_dst_file = NULL; | |
2266 | } | |
2267 | ||
93589827 PX |
2268 | /* |
2269 | * Do the same to postcopy fast path socket too if there is. No | |
2270 | * locking needed because this qemufile should only be managed by | |
2271 | * return path thread. | |
2272 | */ | |
2273 | if (ms->postcopy_qemufile_src) { | |
2274 | migration_ioc_unregister_yank_from_file(ms->postcopy_qemufile_src); | |
2275 | qemu_file_shutdown(ms->postcopy_qemufile_src); | |
2276 | qemu_fclose(ms->postcopy_qemufile_src); | |
2277 | ms->postcopy_qemufile_src = NULL; | |
2278 | } | |
2279 | ||
43044ac0 PX |
2280 | qemu_fclose(file); |
2281 | } | |
2282 | ||
70b20477 DDAG |
2283 | /* |
2284 | * Handles messages sent on the return path towards the source VM | |
2285 | * | |
2286 | */ | |
2287 | static void *source_return_path_thread(void *opaque) | |
2288 | { | |
2289 | MigrationState *ms = opaque; | |
2290 | QEMUFile *rp = ms->rp_state.from_dst_file; | |
2291 | uint16_t header_len, header_type; | |
568b01ca | 2292 | uint8_t buf[512]; |
70b20477 | 2293 | uint32_t tmp32, sibling_error; |
1e2d90eb DDAG |
2294 | ram_addr_t start = 0; /* =0 to silence warning */ |
2295 | size_t len = 0, expected_len; | |
7aa6070d | 2296 | Error *err = NULL; |
70b20477 DDAG |
2297 | int res; |
2298 | ||
2299 | trace_source_return_path_thread_entry(); | |
74637e6f | 2300 | rcu_register_thread(); |
14b1742e | 2301 | |
7dcb3c87 | 2302 | while (migration_is_setup_or_active()) { |
70b20477 | 2303 | trace_source_return_path_thread_loop_top(); |
7aa6070d | 2304 | |
70b20477 DDAG |
2305 | header_type = qemu_get_be16(rp); |
2306 | header_len = qemu_get_be16(rp); | |
2307 | ||
7a9ddfbf | 2308 | if (qemu_file_get_error(rp)) { |
7aa6070d | 2309 | qemu_file_get_error_obj(rp, &err); |
7a9ddfbf PX |
2310 | goto out; |
2311 | } | |
2312 | ||
70b20477 DDAG |
2313 | if (header_type >= MIG_RP_MSG_MAX || |
2314 | header_type == MIG_RP_MSG_INVALID) { | |
7aa6070d PX |
2315 | error_setg(&err, "Received invalid message 0x%04x length 0x%04x", |
2316 | header_type, header_len); | |
70b20477 DDAG |
2317 | goto out; |
2318 | } | |
2319 | ||
2320 | if ((rp_cmd_args[header_type].len != -1 && | |
2321 | header_len != rp_cmd_args[header_type].len) || | |
568b01ca | 2322 | header_len > sizeof(buf)) { |
7aa6070d PX |
2323 | error_setg(&err, "Received '%s' message (0x%04x) with" |
2324 | "incorrect length %d expecting %zu", | |
2325 | rp_cmd_args[header_type].name, header_type, header_len, | |
2326 | (size_t)rp_cmd_args[header_type].len); | |
70b20477 DDAG |
2327 | goto out; |
2328 | } | |
2329 | ||
2330 | /* We know we've got a valid header by this point */ | |
2331 | res = qemu_get_buffer(rp, buf, header_len); | |
2332 | if (res != header_len) { | |
7aa6070d PX |
2333 | error_setg(&err, "Failed reading data for message 0x%04x" |
2334 | " read %d expected %d", | |
2335 | header_type, res, header_len); | |
70b20477 DDAG |
2336 | goto out; |
2337 | } | |
2338 | ||
2339 | /* OK, we have the message and the data */ | |
2340 | switch (header_type) { | |
2341 | case MIG_RP_MSG_SHUT: | |
4d885131 | 2342 | sibling_error = ldl_be_p(buf); |
70b20477 DDAG |
2343 | trace_source_return_path_thread_shut(sibling_error); |
2344 | if (sibling_error) { | |
7aa6070d | 2345 | error_setg(&err, "Sibling indicated error %d", sibling_error); |
70b20477 DDAG |
2346 | } |
2347 | /* | |
2348 | * We'll let the main thread deal with closing the RP | |
2349 | * we could do a shutdown(2) on it, but we're the only user | |
2350 | * anyway, so there's nothing gained. | |
2351 | */ | |
2352 | goto out; | |
2353 | ||
2354 | case MIG_RP_MSG_PONG: | |
4d885131 | 2355 | tmp32 = ldl_be_p(buf); |
70b20477 | 2356 | trace_source_return_path_thread_pong(tmp32); |
b28fb582 | 2357 | qemu_sem_post(&ms->rp_state.rp_pong_acks); |
70b20477 DDAG |
2358 | break; |
2359 | ||
1e2d90eb | 2360 | case MIG_RP_MSG_REQ_PAGES: |
4d885131 PM |
2361 | start = ldq_be_p(buf); |
2362 | len = ldl_be_p(buf + 8); | |
7aa6070d PX |
2363 | migrate_handle_rp_req_pages(ms, NULL, start, len, &err); |
2364 | if (err) { | |
2365 | goto out; | |
2366 | } | |
1e2d90eb DDAG |
2367 | break; |
2368 | ||
2369 | case MIG_RP_MSG_REQ_PAGES_ID: | |
2370 | expected_len = 12 + 1; /* header + termination */ | |
2371 | ||
2372 | if (header_len >= expected_len) { | |
4d885131 PM |
2373 | start = ldq_be_p(buf); |
2374 | len = ldl_be_p(buf + 8); | |
1e2d90eb DDAG |
2375 | /* Now we expect an idstr */ |
2376 | tmp32 = buf[12]; /* Length of the following idstr */ | |
2377 | buf[13 + tmp32] = '\0'; | |
2378 | expected_len += tmp32; | |
2379 | } | |
2380 | if (header_len != expected_len) { | |
7aa6070d PX |
2381 | error_setg(&err, "Req_Page_id with length %d expecting %zd", |
2382 | header_len, expected_len); | |
2383 | goto out; | |
2384 | } | |
2385 | migrate_handle_rp_req_pages(ms, (char *)&buf[13], start, len, | |
2386 | &err); | |
2387 | if (err) { | |
1e2d90eb DDAG |
2388 | goto out; |
2389 | } | |
1e2d90eb DDAG |
2390 | break; |
2391 | ||
a335debb PX |
2392 | case MIG_RP_MSG_RECV_BITMAP: |
2393 | if (header_len < 1) { | |
7aa6070d | 2394 | error_setg(&err, "MIG_RP_MSG_RECV_BITMAP missing block name"); |
a335debb PX |
2395 | goto out; |
2396 | } | |
2397 | /* Format: len (1B) + idstr (<255B). This ends the idstr. */ | |
2398 | buf[buf[0] + 1] = '\0'; | |
88577f32 | 2399 | if (!migrate_handle_rp_recv_bitmap(ms, (char *)(buf + 1), &err)) { |
a335debb PX |
2400 | goto out; |
2401 | } | |
2402 | break; | |
2403 | ||
13955b89 PX |
2404 | case MIG_RP_MSG_RESUME_ACK: |
2405 | tmp32 = ldl_be_p(buf); | |
88577f32 | 2406 | if (!migrate_handle_rp_resume_ack(ms, tmp32, &err)) { |
13955b89 PX |
2407 | goto out; |
2408 | } | |
2409 | break; | |
2410 | ||
1b4adb10 AH |
2411 | case MIG_RP_MSG_SWITCHOVER_ACK: |
2412 | ms->switchover_acked = true; | |
2413 | trace_source_return_path_thread_switchover_acked(); | |
2414 | break; | |
2415 | ||
70b20477 DDAG |
2416 | default: |
2417 | break; | |
2418 | } | |
2419 | } | |
14b1742e PX |
2420 | |
2421 | out: | |
7aa6070d PX |
2422 | if (err) { |
2423 | migrate_set_error(ms, err); | |
2424 | error_free(err); | |
70b20477 | 2425 | trace_source_return_path_thread_bad_end(); |
70b20477 DDAG |
2426 | } |
2427 | ||
f8c543e8 PX |
2428 | if (ms->state == MIGRATION_STATUS_POSTCOPY_RECOVER) { |
2429 | /* | |
2430 | * this will be extremely unlikely: that we got yet another network | |
2431 | * issue during recovering of the 1st network failure.. during this | |
2432 | * period the main migration thread can be waiting on rp_sem for | |
2433 | * this thread to sync with the other side. | |
2434 | * | |
2435 | * When this happens, explicitly kick the migration thread out of | |
2436 | * RECOVER stage and back to PAUSED, so the admin can try | |
2437 | * everything again. | |
2438 | */ | |
2439 | migration_rp_kick(ms); | |
2440 | } | |
2441 | ||
70b20477 | 2442 | trace_source_return_path_thread_end(); |
74637e6f | 2443 | rcu_unregister_thread(); |
7aa6070d | 2444 | |
70b20477 DDAG |
2445 | return NULL; |
2446 | } | |
2447 | ||
ef796ee9 | 2448 | static int open_return_path_on_source(MigrationState *ms) |
70b20477 | 2449 | { |
89a02a9f | 2450 | ms->rp_state.from_dst_file = qemu_file_get_return_path(ms->to_dst_file); |
70b20477 DDAG |
2451 | if (!ms->rp_state.from_dst_file) { |
2452 | return -1; | |
2453 | } | |
2454 | ||
2455 | trace_open_return_path_on_source(); | |
d3e35b8f | 2456 | |
70b20477 DDAG |
2457 | qemu_thread_create(&ms->rp_state.rp_thread, "return path", |
2458 | source_return_path_thread, ms, QEMU_THREAD_JOINABLE); | |
53021ea1 | 2459 | ms->rp_state.rp_thread_created = true; |
70b20477 DDAG |
2460 | |
2461 | trace_open_return_path_on_source_continue(); | |
2462 | ||
2463 | return 0; | |
2464 | } | |
2465 | ||
7aa6070d PX |
2466 | /* Return true if error detected, or false otherwise */ |
2467 | static bool close_return_path_on_source(MigrationState *ms) | |
70b20477 | 2468 | { |
d50f5dc0 | 2469 | if (!ms->rp_state.rp_thread_created) { |
7aa6070d | 2470 | return false; |
d50f5dc0 FR |
2471 | } |
2472 | ||
2473 | trace_migration_return_path_end_before(); | |
2474 | ||
70b20477 | 2475 | /* |
639decf5 FR |
2476 | * If this is a normal exit then the destination will send a SHUT |
2477 | * and the rp_thread will exit, however if there's an error we | |
2478 | * need to cause it to exit. shutdown(2), if we have it, will | |
2479 | * cause it to unblock if it's stuck waiting for the destination. | |
70b20477 | 2480 | */ |
639decf5 | 2481 | WITH_QEMU_LOCK_GUARD(&ms->qemu_file_lock) { |
9425ef3f | 2482 | if (migrate_has_error(ms) && ms->rp_state.from_dst_file) { |
639decf5 FR |
2483 | qemu_file_shutdown(ms->rp_state.from_dst_file); |
2484 | } | |
70b20477 | 2485 | } |
639decf5 | 2486 | |
70b20477 | 2487 | qemu_thread_join(&ms->rp_state.rp_thread); |
53021ea1 | 2488 | ms->rp_state.rp_thread_created = false; |
36e9aab3 | 2489 | migration_release_dst_files(ms); |
7aa6070d | 2490 | trace_migration_return_path_end_after(); |
36e9aab3 | 2491 | |
7aa6070d PX |
2492 | /* Return path will persist the error in MigrationState when quit */ |
2493 | return migrate_has_error(ms); | |
70b20477 DDAG |
2494 | } |
2495 | ||
5655aab0 PX |
2496 | static inline void |
2497 | migration_wait_main_channel(MigrationState *ms) | |
2498 | { | |
2499 | /* Wait until one PONG message received */ | |
2500 | qemu_sem_wait(&ms->rp_state.rp_pong_acks); | |
2501 | } | |
2502 | ||
1d34e4bf DDAG |
2503 | /* |
2504 | * Switch from normal iteration to postcopy | |
2505 | * Returns non-0 on error | |
2506 | */ | |
908927db | 2507 | static int postcopy_start(MigrationState *ms, Error **errp) |
1d34e4bf DDAG |
2508 | { |
2509 | int ret; | |
61b67d47 DB |
2510 | QIOChannelBuffer *bioc; |
2511 | QEMUFile *fb; | |
52033349 | 2512 | uint64_t bandwidth = migrate_max_postcopy_bandwidth(); |
ef8d6488 | 2513 | bool restart_block = false; |
0331c8ca | 2514 | int cur_state = MIGRATION_STATUS_ACTIVE; |
d0edb8a1 | 2515 | |
5655aab0 PX |
2516 | if (migrate_postcopy_preempt()) { |
2517 | migration_wait_main_channel(ms); | |
2518 | if (postcopy_preempt_establish_channel(ms)) { | |
2519 | migrate_set_state(&ms->state, ms->state, MIGRATION_STATUS_FAILED); | |
2520 | return -1; | |
2521 | } | |
d0edb8a1 PX |
2522 | } |
2523 | ||
0331c8ca DDAG |
2524 | if (!migrate_pause_before_switchover()) { |
2525 | migrate_set_state(&ms->state, MIGRATION_STATUS_ACTIVE, | |
2526 | MIGRATION_STATUS_POSTCOPY_ACTIVE); | |
2527 | } | |
1d34e4bf DDAG |
2528 | |
2529 | trace_postcopy_start(); | |
195801d7 | 2530 | bql_lock(); |
1d34e4bf DDAG |
2531 | trace_postcopy_start_set_run(); |
2532 | ||
9867d4dd | 2533 | ret = migration_stop_vm(ms, RUN_STATE_FINISH_MIGRATE); |
76b1c7fe KW |
2534 | if (ret < 0) { |
2535 | goto fail; | |
2536 | } | |
1d34e4bf | 2537 | |
0331c8ca DDAG |
2538 | ret = migration_maybe_pause(ms, &cur_state, |
2539 | MIGRATION_STATUS_POSTCOPY_ACTIVE); | |
2540 | if (ret < 0) { | |
2541 | goto fail; | |
2542 | } | |
2543 | ||
76b1c7fe | 2544 | ret = bdrv_inactivate_all(); |
1d34e4bf DDAG |
2545 | if (ret < 0) { |
2546 | goto fail; | |
2547 | } | |
ef8d6488 | 2548 | restart_block = true; |
1d34e4bf | 2549 | |
1c0d249d DDAG |
2550 | /* |
2551 | * Cause any non-postcopiable, but iterative devices to | |
2552 | * send out their final data. | |
2553 | */ | |
a1fbe750 | 2554 | qemu_savevm_state_complete_precopy(ms->to_dst_file, true, false); |
1c0d249d | 2555 | |
1d34e4bf DDAG |
2556 | /* |
2557 | * in Finish migrate and with the io-lock held everything should | |
2558 | * be quiet, but we've potentially still got dirty pages and we | |
2559 | * need to tell the destination to throw any pages it's already received | |
2560 | * that are dirty | |
2561 | */ | |
58110f0a | 2562 | if (migrate_postcopy_ram()) { |
739fcc1b | 2563 | ram_postcopy_send_discard_bitmap(ms); |
1d34e4bf DDAG |
2564 | } |
2565 | ||
2566 | /* | |
2567 | * send rest of state - note things that are doing postcopy | |
2568 | * will notice we're in POSTCOPY_ACTIVE and not actually | |
2569 | * wrap their state up here | |
2570 | */ | |
e1fde0e0 | 2571 | migration_rate_set(bandwidth); |
58110f0a VSO |
2572 | if (migrate_postcopy_ram()) { |
2573 | /* Ping just for debugging, helps line traces up */ | |
2574 | qemu_savevm_send_ping(ms->to_dst_file, 2); | |
2575 | } | |
1d34e4bf DDAG |
2576 | |
2577 | /* | |
2578 | * While loading the device state we may trigger page transfer | |
2579 | * requests and the fd must be free to process those, and thus | |
2580 | * the destination must read the whole device state off the fd before | |
2581 | * it starts processing it. Unfortunately the ad-hoc migration format | |
2582 | * doesn't allow the destination to know the size to read without fully | |
2583 | * parsing it through each devices load-state code (especially the open | |
2584 | * coded devices that use get/put). | |
2585 | * So we wrap the device state up in a package with a length at the start; | |
2586 | * to do this we use a qemu_buf to hold the whole of the device state. | |
2587 | */ | |
61b67d47 | 2588 | bioc = qio_channel_buffer_new(4096); |
6f01f136 | 2589 | qio_channel_set_name(QIO_CHANNEL(bioc), "migration-postcopy-buffer"); |
77ef2dc1 | 2590 | fb = qemu_file_new_output(QIO_CHANNEL(bioc)); |
61b67d47 | 2591 | object_unref(OBJECT(bioc)); |
1d34e4bf | 2592 | |
c76201ab DDAG |
2593 | /* |
2594 | * Make sure the receiver can get incoming pages before we send the rest | |
2595 | * of the state | |
2596 | */ | |
2597 | qemu_savevm_send_postcopy_listen(fb); | |
2598 | ||
a1fbe750 | 2599 | qemu_savevm_state_complete_precopy(fb, false, false); |
58110f0a VSO |
2600 | if (migrate_postcopy_ram()) { |
2601 | qemu_savevm_send_ping(fb, 3); | |
2602 | } | |
1d34e4bf DDAG |
2603 | |
2604 | qemu_savevm_send_postcopy_run(fb); | |
2605 | ||
2606 | /* <><> end of stuff going into the package */ | |
1d34e4bf | 2607 | |
ef8d6488 DDAG |
2608 | /* Last point of recovery; as soon as we send the package the destination |
2609 | * can open devices and potentially start running. | |
2610 | * Lets just check again we've not got any errors. | |
2611 | */ | |
2612 | ret = qemu_file_get_error(ms->to_dst_file); | |
2613 | if (ret) { | |
908927db | 2614 | error_setg(errp, "postcopy_start: Migration stream errored (pre package)"); |
ef8d6488 DDAG |
2615 | goto fail_closefb; |
2616 | } | |
2617 | ||
2618 | restart_block = false; | |
2619 | ||
1d34e4bf | 2620 | /* Now send that blob */ |
61b67d47 | 2621 | if (qemu_savevm_send_packaged(ms->to_dst_file, bioc->data, bioc->usage)) { |
1d34e4bf DDAG |
2622 | goto fail_closefb; |
2623 | } | |
2624 | qemu_fclose(fb); | |
b82fc321 DDAG |
2625 | |
2626 | /* Send a notify to give a chance for anything that needs to happen | |
2627 | * at the transition to postcopy and after the device state; in particular | |
2628 | * spice needs to trigger a transition now | |
2629 | */ | |
4af667f8 | 2630 | migration_call_notifiers(ms, MIG_EVENT_PRECOPY_DONE, NULL); |
b82fc321 | 2631 | |
e22ffad0 | 2632 | migration_downtime_end(ms); |
1d34e4bf | 2633 | |
195801d7 | 2634 | bql_unlock(); |
1d34e4bf | 2635 | |
58110f0a VSO |
2636 | if (migrate_postcopy_ram()) { |
2637 | /* | |
2638 | * Although this ping is just for debug, it could potentially be | |
2639 | * used for getting a better measurement of downtime at the source. | |
2640 | */ | |
2641 | qemu_savevm_send_ping(ms->to_dst_file, 4); | |
2642 | } | |
1d34e4bf | 2643 | |
ced1c616 PB |
2644 | if (migrate_release_ram()) { |
2645 | ram_postcopy_migrated_memory_release(ms); | |
2646 | } | |
2647 | ||
89a02a9f | 2648 | ret = qemu_file_get_error(ms->to_dst_file); |
1d34e4bf | 2649 | if (ret) { |
4af667f8 SS |
2650 | error_setg_errno(errp, -ret, "postcopy_start: Migration stream error"); |
2651 | bql_lock(); | |
2652 | goto fail; | |
1d34e4bf | 2653 | } |
c01b16ed PX |
2654 | trace_postcopy_preempt_enabled(migrate_postcopy_preempt()); |
2655 | ||
1d34e4bf DDAG |
2656 | return ret; |
2657 | ||
2658 | fail_closefb: | |
2659 | qemu_fclose(fb); | |
2660 | fail: | |
48781e5b | 2661 | migrate_set_state(&ms->state, MIGRATION_STATUS_POSTCOPY_ACTIVE, |
1d34e4bf | 2662 | MIGRATION_STATUS_FAILED); |
ef8d6488 DDAG |
2663 | if (restart_block) { |
2664 | /* A failure happened early enough that we know the destination hasn't | |
2665 | * accessed block devices, so we're safe to recover. | |
2666 | */ | |
2667 | Error *local_err = NULL; | |
2668 | ||
3b717194 | 2669 | bdrv_activate_all(&local_err); |
ef8d6488 DDAG |
2670 | if (local_err) { |
2671 | error_report_err(local_err); | |
2672 | } | |
2673 | } | |
4af667f8 | 2674 | migration_call_notifiers(ms, MIG_EVENT_PRECOPY_FAILED, NULL); |
195801d7 | 2675 | bql_unlock(); |
1d34e4bf DDAG |
2676 | return -1; |
2677 | } | |
2678 | ||
e91d8951 DDAG |
2679 | /** |
2680 | * migration_maybe_pause: Pause if required to by | |
a4a411fb | 2681 | * migrate_pause_before_switchover called with the BQL locked |
e91d8951 DDAG |
2682 | * Returns: 0 on success |
2683 | */ | |
0331c8ca DDAG |
2684 | static int migration_maybe_pause(MigrationState *s, |
2685 | int *current_active_state, | |
2686 | int new_state) | |
e91d8951 DDAG |
2687 | { |
2688 | if (!migrate_pause_before_switchover()) { | |
2689 | return 0; | |
2690 | } | |
2691 | ||
2692 | /* Since leaving this state is not atomic with posting the semaphore | |
2693 | * it's possible that someone could have issued multiple migrate_continue | |
2694 | * and the semaphore is incorrectly positive at this point; | |
2695 | * the docs say it's undefined to reinit a semaphore that's already | |
2696 | * init'd, so use timedwait to eat up any existing posts. | |
2697 | */ | |
2698 | while (qemu_sem_timedwait(&s->pause_sem, 1) == 0) { | |
2699 | /* This block intentionally left blank */ | |
2700 | } | |
2701 | ||
8958338b ZF |
2702 | /* |
2703 | * If the migration is cancelled when it is in the completion phase, | |
2704 | * the migration state is set to MIGRATION_STATUS_CANCELLING. | |
2705 | * So we don't need to wait a semaphore, otherwise we would always | |
2706 | * wait for the 'pause_sem' semaphore. | |
2707 | */ | |
2708 | if (s->state != MIGRATION_STATUS_CANCELLING) { | |
195801d7 | 2709 | bql_unlock(); |
8958338b ZF |
2710 | migrate_set_state(&s->state, *current_active_state, |
2711 | MIGRATION_STATUS_PRE_SWITCHOVER); | |
2712 | qemu_sem_wait(&s->pause_sem); | |
2713 | migrate_set_state(&s->state, MIGRATION_STATUS_PRE_SWITCHOVER, | |
2714 | new_state); | |
2715 | *current_active_state = new_state; | |
195801d7 | 2716 | bql_lock(); |
8958338b | 2717 | } |
e91d8951 | 2718 | |
0331c8ca | 2719 | return s->state == new_state ? 0 : -EINVAL; |
e91d8951 DDAG |
2720 | } |
2721 | ||
d7f5a043 WW |
2722 | static int migration_completion_precopy(MigrationState *s, |
2723 | int *current_active_state) | |
09f6c85e DDAG |
2724 | { |
2725 | int ret; | |
2726 | ||
195801d7 | 2727 | bql_lock(); |
c33f1829 | 2728 | |
9867d4dd SS |
2729 | if (!migrate_mode_is_cpr(s)) { |
2730 | ret = migration_stop_vm(s, RUN_STATE_FINISH_MIGRATE); | |
2731 | if (ret < 0) { | |
2732 | goto out_unlock; | |
2733 | } | |
d7f5a043 | 2734 | } |
c33f1829 | 2735 | |
d7f5a043 WW |
2736 | ret = migration_maybe_pause(s, current_active_state, |
2737 | MIGRATION_STATUS_DEVICE); | |
2738 | if (ret < 0) { | |
2739 | goto out_unlock; | |
2740 | } | |
09f6c85e | 2741 | |
d7f5a043 WW |
2742 | /* |
2743 | * Inactivate disks except in COLO, and track that we have done so in order | |
2744 | * to remember to reactivate them if migration fails or is cancelled. | |
2745 | */ | |
2746 | s->block_inactive = !migrate_colo(); | |
2747 | migration_rate_set(RATE_LIMIT_DISABLED); | |
2748 | ret = qemu_savevm_state_complete_precopy(s->to_dst_file, false, | |
2749 | s->block_inactive); | |
2750 | out_unlock: | |
195801d7 | 2751 | bql_unlock(); |
d7f5a043 WW |
2752 | return ret; |
2753 | } | |
b10ac0c4 | 2754 | |
d7f5a043 WW |
2755 | static void migration_completion_postcopy(MigrationState *s) |
2756 | { | |
2757 | trace_migration_completion_postcopy_end(); | |
2758 | ||
195801d7 | 2759 | bql_lock(); |
d7f5a043 | 2760 | qemu_savevm_state_complete_postcopy(s->to_dst_file); |
195801d7 | 2761 | bql_unlock(); |
d7f5a043 WW |
2762 | |
2763 | /* | |
2764 | * Shutdown the postcopy fast path thread. This is only needed when dest | |
2765 | * QEMU binary is old (7.1/7.2). QEMU 8.0+ doesn't need this. | |
2766 | */ | |
2767 | if (migrate_postcopy_preempt() && s->preempt_pre_7_2) { | |
2768 | postcopy_preempt_shutdown_file(s); | |
2769 | } | |
68b88468 | 2770 | |
d7f5a043 WW |
2771 | trace_migration_completion_postcopy_end_after_complete(); |
2772 | } | |
2773 | ||
2774 | static void migration_completion_failed(MigrationState *s, | |
2775 | int current_active_state) | |
2776 | { | |
2777 | if (s->block_inactive && (s->state == MIGRATION_STATUS_ACTIVE || | |
2778 | s->state == MIGRATION_STATUS_DEVICE)) { | |
6621883f | 2779 | /* |
d7f5a043 WW |
2780 | * If not doing postcopy, vm_start() will be called: let's |
2781 | * regain control on images. | |
6621883f | 2782 | */ |
d7f5a043 WW |
2783 | Error *local_err = NULL; |
2784 | ||
195801d7 | 2785 | bql_lock(); |
d7f5a043 WW |
2786 | bdrv_activate_all(&local_err); |
2787 | if (local_err) { | |
2788 | error_report_err(local_err); | |
2789 | } else { | |
2790 | s->block_inactive = false; | |
36f62f11 | 2791 | } |
195801d7 | 2792 | bql_unlock(); |
d7f5a043 WW |
2793 | } |
2794 | ||
2795 | migrate_set_state(&s->state, current_active_state, | |
2796 | MIGRATION_STATUS_FAILED); | |
2797 | } | |
36f62f11 | 2798 | |
d7f5a043 WW |
2799 | /** |
2800 | * migration_completion: Used by migration_thread when there's not much left. | |
2801 | * The caller 'breaks' the loop when this returns. | |
2802 | * | |
2803 | * @s: Current migration state | |
2804 | */ | |
2805 | static void migration_completion(MigrationState *s) | |
2806 | { | |
2807 | int ret = 0; | |
2808 | int current_active_state = s->state; | |
2809 | ||
2810 | if (s->state == MIGRATION_STATUS_ACTIVE) { | |
2811 | ret = migration_completion_precopy(s, ¤t_active_state); | |
2812 | } else if (s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) { | |
2813 | migration_completion_postcopy(s); | |
444252b9 | 2814 | } else { |
d7f5a043 WW |
2815 | ret = -1; |
2816 | } | |
2817 | ||
2818 | if (ret < 0) { | |
6ba11211 | 2819 | goto fail; |
09f6c85e | 2820 | } |
09f6c85e | 2821 | |
d7f5a043 | 2822 | if (close_return_path_on_source(s)) { |
d50f5dc0 | 2823 | goto fail; |
09f6c85e DDAG |
2824 | } |
2825 | ||
89a02a9f | 2826 | if (qemu_file_get_error(s->to_dst_file)) { |
09f6c85e | 2827 | trace_migration_completion_file_err(); |
6dab4c93 | 2828 | goto fail; |
09f6c85e DDAG |
2829 | } |
2830 | ||
5e804644 | 2831 | if (migrate_colo() && s->state == MIGRATION_STATUS_ACTIVE) { |
eeeb48ee ZC |
2832 | /* COLO does not support postcopy */ |
2833 | migrate_set_state(&s->state, MIGRATION_STATUS_ACTIVE, | |
2834 | MIGRATION_STATUS_COLO); | |
2835 | } else { | |
63f64d77 | 2836 | migration_completion_end(s); |
0b827d5e HZ |
2837 | } |
2838 | ||
09f6c85e DDAG |
2839 | return; |
2840 | ||
6dab4c93 | 2841 | fail: |
d7f5a043 | 2842 | migration_completion_failed(s, current_active_state); |
09f6c85e DDAG |
2843 | } |
2844 | ||
8518278a AG |
2845 | /** |
2846 | * bg_migration_completion: Used by bg_migration_thread when after all the | |
2847 | * RAM has been saved. The caller 'breaks' the loop when this returns. | |
2848 | * | |
2849 | * @s: Current migration state | |
2850 | */ | |
2851 | static void bg_migration_completion(MigrationState *s) | |
2852 | { | |
2853 | int current_active_state = s->state; | |
2854 | ||
8518278a AG |
2855 | if (s->state == MIGRATION_STATUS_ACTIVE) { |
2856 | /* | |
2857 | * By this moment we have RAM content saved into the migration stream. | |
2858 | * The next step is to flush the non-RAM content (device state) | |
2859 | * right after the ram content. The device state has been stored into | |
2860 | * the temporary buffer before RAM saving started. | |
2861 | */ | |
2862 | qemu_put_buffer(s->to_dst_file, s->bioc->data, s->bioc->usage); | |
2863 | qemu_fflush(s->to_dst_file); | |
2864 | } else if (s->state == MIGRATION_STATUS_CANCELLING) { | |
2865 | goto fail; | |
2866 | } | |
2867 | ||
2868 | if (qemu_file_get_error(s->to_dst_file)) { | |
2869 | trace_migration_completion_file_err(); | |
2870 | goto fail; | |
2871 | } | |
2872 | ||
63f64d77 | 2873 | migration_completion_end(s); |
8518278a AG |
2874 | return; |
2875 | ||
2876 | fail: | |
2877 | migrate_set_state(&s->state, current_active_state, | |
2878 | MIGRATION_STATUS_FAILED); | |
2879 | } | |
2880 | ||
b23c2ade PX |
2881 | typedef enum MigThrError { |
2882 | /* No error detected */ | |
2883 | MIG_THR_ERR_NONE = 0, | |
2884 | /* Detected error, but resumed successfully */ | |
2885 | MIG_THR_ERR_RECOVERED = 1, | |
2886 | /* Detected fatal error, need to exit */ | |
2887 | MIG_THR_ERR_FATAL = 2, | |
2888 | } MigThrError; | |
2889 | ||
94190696 PX |
2890 | static int postcopy_resume_handshake(MigrationState *s) |
2891 | { | |
2892 | qemu_savevm_send_postcopy_resume(s->to_dst_file); | |
2893 | ||
2894 | while (s->state == MIGRATION_STATUS_POSTCOPY_RECOVER) { | |
f8c543e8 PX |
2895 | if (migration_rp_wait(s)) { |
2896 | return -1; | |
2897 | } | |
94190696 PX |
2898 | } |
2899 | ||
2900 | if (s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) { | |
2901 | return 0; | |
2902 | } | |
2903 | ||
2904 | return -1; | |
2905 | } | |
2906 | ||
135b87b4 PX |
2907 | /* Return zero if success, or <0 for error */ |
2908 | static int postcopy_do_resume(MigrationState *s) | |
2909 | { | |
d1b8eadb PX |
2910 | int ret; |
2911 | ||
2912 | /* | |
2913 | * Call all the resume_prepare() hooks, so that modules can be | |
2914 | * ready for the migration resume. | |
2915 | */ | |
2916 | ret = qemu_savevm_state_resume_prepare(s); | |
2917 | if (ret) { | |
2918 | error_report("%s: resume_prepare() failure detected: %d", | |
2919 | __func__, ret); | |
2920 | return ret; | |
2921 | } | |
2922 | ||
5655aab0 PX |
2923 | /* |
2924 | * If preempt is enabled, re-establish the preempt channel. Note that | |
2925 | * we do it after resume prepare to make sure the main channel will be | |
2926 | * created before the preempt channel. E.g. with weak network, the | |
2927 | * dest QEMU may get messed up with the preempt and main channels on | |
2928 | * the order of connection setup. This guarantees the correct order. | |
2929 | */ | |
2930 | ret = postcopy_preempt_establish_channel(s); | |
2931 | if (ret) { | |
2932 | error_report("%s: postcopy_preempt_establish_channel(): %d", | |
2933 | __func__, ret); | |
2934 | return ret; | |
2935 | } | |
2936 | ||
d1b8eadb | 2937 | /* |
94190696 PX |
2938 | * Last handshake with destination on the resume (destination will |
2939 | * switch to postcopy-active afterwards) | |
d1b8eadb | 2940 | */ |
94190696 PX |
2941 | ret = postcopy_resume_handshake(s); |
2942 | if (ret) { | |
2943 | error_report("%s: handshake failed: %d", __func__, ret); | |
2944 | return ret; | |
2945 | } | |
d1b8eadb | 2946 | |
135b87b4 PX |
2947 | return 0; |
2948 | } | |
2949 | ||
b23c2ade PX |
2950 | /* |
2951 | * We don't return until we are in a safe state to continue current | |
2952 | * postcopy migration. Returns MIG_THR_ERR_RECOVERED if recovered, or | |
2953 | * MIG_THR_ERR_FATAL if unrecovery failure happened. | |
2954 | */ | |
2955 | static MigThrError postcopy_pause(MigrationState *s) | |
2956 | { | |
2957 | assert(s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE); | |
b23c2ade | 2958 | |
135b87b4 | 2959 | while (true) { |
62df066f PX |
2960 | QEMUFile *file; |
2961 | ||
22b04245 FR |
2962 | /* |
2963 | * We're already pausing, so ignore any errors on the return | |
2964 | * path and just wait for the thread to finish. It will be | |
2965 | * re-created when we resume. | |
2966 | */ | |
2967 | close_return_path_on_source(s); | |
2968 | ||
39675fff PX |
2969 | /* |
2970 | * Current channel is possibly broken. Release it. Note that this is | |
2971 | * guaranteed even without lock because to_dst_file should only be | |
2972 | * modified by the migration thread. That also guarantees that the | |
2973 | * unregister of yank is safe too without the lock. It should be safe | |
2974 | * even to be within the qemu_file_lock, but we didn't do that to avoid | |
2975 | * taking more mutex (yank_lock) within qemu_file_lock. TL;DR: we make | |
2976 | * the qemu_file_lock critical section as small as possible. | |
2977 | */ | |
135b87b4 | 2978 | assert(s->to_dst_file); |
39675fff | 2979 | migration_ioc_unregister_yank_from_file(s->to_dst_file); |
62df066f PX |
2980 | qemu_mutex_lock(&s->qemu_file_lock); |
2981 | file = s->to_dst_file; | |
135b87b4 | 2982 | s->to_dst_file = NULL; |
62df066f PX |
2983 | qemu_mutex_unlock(&s->qemu_file_lock); |
2984 | ||
2985 | qemu_file_shutdown(file); | |
2986 | qemu_fclose(file); | |
b23c2ade | 2987 | |
d246ea50 PX |
2988 | migrate_set_state(&s->state, s->state, |
2989 | MIGRATION_STATUS_POSTCOPY_PAUSED); | |
2990 | ||
135b87b4 PX |
2991 | error_report("Detected IO failure for postcopy. " |
2992 | "Migration paused."); | |
b23c2ade | 2993 | |
135b87b4 PX |
2994 | /* |
2995 | * We wait until things fixed up. Then someone will setup the | |
2996 | * status back for us. | |
2997 | */ | |
2998 | while (s->state == MIGRATION_STATUS_POSTCOPY_PAUSED) { | |
2999 | qemu_sem_wait(&s->postcopy_pause_sem); | |
3000 | } | |
3001 | ||
3002 | if (s->state == MIGRATION_STATUS_POSTCOPY_RECOVER) { | |
3003 | /* Woken up by a recover procedure. Give it a shot */ | |
b23c2ade | 3004 | |
135b87b4 PX |
3005 | /* Do the resume logic */ |
3006 | if (postcopy_do_resume(s) == 0) { | |
3007 | /* Let's continue! */ | |
3008 | trace_postcopy_pause_continued(); | |
3009 | return MIG_THR_ERR_RECOVERED; | |
3010 | } else { | |
3011 | /* | |
3012 | * Something wrong happened during the recovery, let's | |
3013 | * pause again. Pause is always better than throwing | |
3014 | * data away. | |
3015 | */ | |
3016 | continue; | |
3017 | } | |
3018 | } else { | |
3019 | /* This is not right... Time to quit. */ | |
3020 | return MIG_THR_ERR_FATAL; | |
3021 | } | |
3022 | } | |
b23c2ade PX |
3023 | } |
3024 | ||
3025 | static MigThrError migration_detect_error(MigrationState *s) | |
3026 | { | |
3027 | int ret; | |
c3c5eae6 | 3028 | int state = s->state; |
3d661c8a | 3029 | Error *local_error = NULL; |
c3c5eae6 DDAG |
3030 | |
3031 | if (state == MIGRATION_STATUS_CANCELLING || | |
3032 | state == MIGRATION_STATUS_CANCELLED) { | |
3033 | /* End the migration, but don't set the state to failed */ | |
3034 | return MIG_THR_ERR_FATAL; | |
3035 | } | |
b23c2ade | 3036 | |
60bb3c58 PX |
3037 | /* |
3038 | * Try to detect any file errors. Note that postcopy_qemufile_src will | |
3039 | * be NULL when postcopy preempt is not enabled. | |
3040 | */ | |
3041 | ret = qemu_file_get_error_obj_any(s->to_dst_file, | |
3042 | s->postcopy_qemufile_src, | |
3043 | &local_error); | |
b23c2ade PX |
3044 | if (!ret) { |
3045 | /* Everything is fine */ | |
3d661c8a | 3046 | assert(!local_error); |
b23c2ade PX |
3047 | return MIG_THR_ERR_NONE; |
3048 | } | |
3049 | ||
3d661c8a YK |
3050 | if (local_error) { |
3051 | migrate_set_error(s, local_error); | |
3052 | error_free(local_error); | |
3053 | } | |
3054 | ||
d5c8f2af | 3055 | if (state == MIGRATION_STATUS_POSTCOPY_ACTIVE && ret) { |
b23c2ade PX |
3056 | /* |
3057 | * For postcopy, we allow the network to be down for a | |
3058 | * while. After that, it can be continued by a | |
3059 | * recovery phase. | |
3060 | */ | |
3061 | return postcopy_pause(s); | |
3062 | } else { | |
3063 | /* | |
3064 | * For precopy (or postcopy with error outside IO), we fail | |
3065 | * with no time. | |
3066 | */ | |
c3c5eae6 | 3067 | migrate_set_state(&s->state, state, MIGRATION_STATUS_FAILED); |
b23c2ade PX |
3068 | trace_migration_thread_file_err(); |
3069 | ||
3070 | /* Time to stop the migration, now. */ | |
3071 | return MIG_THR_ERR_FATAL; | |
3072 | } | |
3073 | } | |
3074 | ||
63f64d77 | 3075 | static void migration_completion_end(MigrationState *s) |
cf011f08 | 3076 | { |
f57e5a6c | 3077 | uint64_t bytes = migration_transferred_bytes(); |
cf011f08 | 3078 | int64_t end_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); |
6cde6fbe | 3079 | int64_t transfer_time; |
cf011f08 | 3080 | |
63f64d77 FR |
3081 | /* |
3082 | * Take the BQL here so that query-migrate on the QMP thread sees: | |
3083 | * - atomic update of s->total_time and s->mbps; | |
3084 | * - correct ordering of s->mbps update vs. s->state; | |
3085 | */ | |
3086 | bql_lock(); | |
e22ffad0 | 3087 | migration_downtime_end(s); |
cf011f08 | 3088 | s->total_time = end_time - s->start_time; |
6cde6fbe JQ |
3089 | transfer_time = s->total_time - s->setup_time; |
3090 | if (transfer_time) { | |
3091 | s->mbps = ((double) bytes * 8.0) / transfer_time / 1000; | |
cf011f08 | 3092 | } |
63f64d77 FR |
3093 | |
3094 | migrate_set_state(&s->state, s->state, | |
3095 | MIGRATION_STATUS_COMPLETED); | |
3096 | bql_unlock(); | |
cf011f08 PX |
3097 | } |
3098 | ||
87f3bd87 IR |
3099 | static void update_iteration_initial_status(MigrationState *s) |
3100 | { | |
3101 | /* | |
3102 | * Update these three fields at the same time to avoid mismatch info lead | |
3103 | * wrong speed calculation. | |
3104 | */ | |
3105 | s->iteration_start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); | |
f57e5a6c | 3106 | s->iteration_initial_bytes = migration_transferred_bytes(); |
87f3bd87 IR |
3107 | s->iteration_initial_pages = ram_get_total_transferred_pages(); |
3108 | } | |
3109 | ||
b15df1ae PX |
3110 | static void migration_update_counters(MigrationState *s, |
3111 | int64_t current_time) | |
3112 | { | |
aecbfe9c | 3113 | uint64_t transferred, transferred_pages, time_spent; |
0c8f0efd | 3114 | uint64_t current_bytes; /* bytes transferred since the beginning */ |
8b239597 PX |
3115 | uint64_t switchover_bw; |
3116 | /* Expected bandwidth when switching over to destination QEMU */ | |
3117 | double expected_bw_per_ms; | |
b15df1ae PX |
3118 | double bandwidth; |
3119 | ||
3120 | if (current_time < s->iteration_start_time + BUFFER_DELAY) { | |
3121 | return; | |
3122 | } | |
3123 | ||
8b239597 | 3124 | switchover_bw = migrate_avail_switchover_bandwidth(); |
f57e5a6c | 3125 | current_bytes = migration_transferred_bytes(); |
0c8f0efd | 3126 | transferred = current_bytes - s->iteration_initial_bytes; |
b15df1ae PX |
3127 | time_spent = current_time - s->iteration_start_time; |
3128 | bandwidth = (double)transferred / time_spent; | |
8b239597 PX |
3129 | |
3130 | if (switchover_bw) { | |
3131 | /* | |
3132 | * If the user specified a switchover bandwidth, let's trust the | |
3133 | * user so that can be more accurate than what we estimated. | |
3134 | */ | |
3135 | expected_bw_per_ms = switchover_bw / 1000; | |
3136 | } else { | |
3137 | /* If the user doesn't specify bandwidth, we use the estimated */ | |
3138 | expected_bw_per_ms = bandwidth; | |
3139 | } | |
3140 | ||
3141 | s->threshold_size = expected_bw_per_ms * migrate_downtime_limit(); | |
b15df1ae PX |
3142 | |
3143 | s->mbps = (((double) transferred * 8.0) / | |
3144 | ((double) time_spent / 1000.0)) / 1000.0 / 1000.0; | |
3145 | ||
aecbfe9c XG |
3146 | transferred_pages = ram_get_total_transferred_pages() - |
3147 | s->iteration_initial_pages; | |
3148 | s->pages_per_second = (double) transferred_pages / | |
3149 | (((double) time_spent / 1000.0)); | |
3150 | ||
b15df1ae PX |
3151 | /* |
3152 | * if we haven't sent anything, we don't want to | |
3153 | * recalculate. 10000 is a small enough number for our purposes | |
3154 | */ | |
aff3f660 | 3155 | if (stat64_get(&mig_stats.dirty_pages_rate) && |
72f8e587 | 3156 | transferred > 10000) { |
73208a33 | 3157 | s->expected_downtime = |
8b239597 | 3158 | stat64_get(&mig_stats.dirty_bytes_last_sync) / expected_bw_per_ms; |
b15df1ae PX |
3159 | } |
3160 | ||
0743f41f | 3161 | migration_rate_reset(); |
b15df1ae | 3162 | |
87f3bd87 | 3163 | update_iteration_initial_status(s); |
b15df1ae PX |
3164 | |
3165 | trace_migrate_transferred(transferred, time_spent, | |
8b239597 PX |
3166 | /* Both in unit bytes/ms */ |
3167 | bandwidth, switchover_bw / 1000, | |
3168 | s->threshold_size); | |
b15df1ae PX |
3169 | } |
3170 | ||
1b4adb10 AH |
3171 | static bool migration_can_switchover(MigrationState *s) |
3172 | { | |
3173 | if (!migrate_switchover_ack()) { | |
3174 | return true; | |
3175 | } | |
3176 | ||
3177 | /* No reason to wait for switchover ACK if VM is stopped */ | |
3178 | if (!runstate_is_running()) { | |
3179 | return true; | |
3180 | } | |
3181 | ||
3182 | return s->switchover_acked; | |
3183 | } | |
3184 | ||
2ad87305 PX |
3185 | /* Migration thread iteration status */ |
3186 | typedef enum { | |
3187 | MIG_ITERATE_RESUME, /* Resume current iteration */ | |
3188 | MIG_ITERATE_SKIP, /* Skip current iteration */ | |
3189 | MIG_ITERATE_BREAK, /* Break the loop */ | |
3190 | } MigIterateState; | |
3191 | ||
3192 | /* | |
3193 | * Return true if continue to the next iteration directly, false | |
3194 | * otherwise. | |
3195 | */ | |
3196 | static MigIterateState migration_iteration_run(MigrationState *s) | |
3197 | { | |
24beea4e | 3198 | uint64_t must_precopy, can_postcopy; |
908927db | 3199 | Error *local_err = NULL; |
2ad87305 | 3200 | bool in_postcopy = s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE; |
1b4adb10 | 3201 | bool can_switchover = migration_can_switchover(s); |
2ad87305 | 3202 | |
24beea4e JQ |
3203 | qemu_savevm_state_pending_estimate(&must_precopy, &can_postcopy); |
3204 | uint64_t pending_size = must_precopy + can_postcopy; | |
2ad87305 | 3205 | |
24beea4e | 3206 | trace_migrate_pending_estimate(pending_size, must_precopy, can_postcopy); |
c8df4a7a | 3207 | |
24beea4e JQ |
3208 | if (must_precopy <= s->threshold_size) { |
3209 | qemu_savevm_state_pending_exact(&must_precopy, &can_postcopy); | |
3210 | pending_size = must_precopy + can_postcopy; | |
3211 | trace_migrate_pending_exact(pending_size, must_precopy, can_postcopy); | |
c8df4a7a | 3212 | } |
2ad87305 | 3213 | |
1b4adb10 | 3214 | if ((!pending_size || pending_size < s->threshold_size) && can_switchover) { |
2ad87305 PX |
3215 | trace_migration_thread_low_pending(pending_size); |
3216 | migration_completion(s); | |
3217 | return MIG_ITERATE_BREAK; | |
3218 | } | |
3219 | ||
d9df9292 | 3220 | /* Still a significant amount to transfer */ |
1b4adb10 | 3221 | if (!in_postcopy && must_precopy <= s->threshold_size && can_switchover && |
d9df9292 | 3222 | qatomic_read(&s->start_postcopy)) { |
908927db TG |
3223 | if (postcopy_start(s, &local_err)) { |
3224 | migrate_set_error(s, local_err); | |
3225 | error_report_err(local_err); | |
d9df9292 JQ |
3226 | } |
3227 | return MIG_ITERATE_SKIP; | |
3228 | } | |
3229 | ||
3230 | /* Just another iteration step */ | |
3231 | qemu_savevm_state_iterate(s->to_dst_file, in_postcopy); | |
2ad87305 PX |
3232 | return MIG_ITERATE_RESUME; |
3233 | } | |
3234 | ||
199aa6d4 PX |
3235 | static void migration_iteration_finish(MigrationState *s) |
3236 | { | |
3237 | /* If we enabled cpu throttling for auto-converge, turn it off. */ | |
3238 | cpu_throttle_stop(); | |
3239 | ||
195801d7 | 3240 | bql_lock(); |
199aa6d4 PX |
3241 | switch (s->state) { |
3242 | case MIGRATION_STATUS_COMPLETED: | |
199aa6d4 PX |
3243 | runstate_set(RUN_STATE_POSTMIGRATE); |
3244 | break; | |
751fe4c6 | 3245 | case MIGRATION_STATUS_COLO: |
d70178a8 | 3246 | assert(migrate_colo()); |
199aa6d4 | 3247 | migrate_start_colo_process(s); |
f4584076 | 3248 | s->vm_old_state = RUN_STATE_RUNNING; |
199aa6d4 PX |
3249 | /* Fallthrough */ |
3250 | case MIGRATION_STATUS_FAILED: | |
3251 | case MIGRATION_STATUS_CANCELLED: | |
57225e5f | 3252 | case MIGRATION_STATUS_CANCELLING: |
b4e9ddcc | 3253 | if (runstate_is_live(s->vm_old_state)) { |
aa505f8e RL |
3254 | if (!runstate_check(RUN_STATE_SHUTDOWN)) { |
3255 | vm_start(); | |
3256 | } | |
199aa6d4 PX |
3257 | } else { |
3258 | if (runstate_check(RUN_STATE_FINISH_MIGRATE)) { | |
a4c6275a | 3259 | runstate_set(s->vm_old_state); |
199aa6d4 PX |
3260 | } |
3261 | } | |
3262 | break; | |
3263 | ||
3264 | default: | |
3265 | /* Should not reach here, but if so, forgive the VM. */ | |
3266 | error_report("%s: Unknown ending state %d", __func__, s->state); | |
3267 | break; | |
3268 | } | |
44d0d456 FR |
3269 | |
3270 | migration_bh_schedule(migrate_fd_cleanup_bh, s); | |
195801d7 | 3271 | bql_unlock(); |
199aa6d4 PX |
3272 | } |
3273 | ||
8518278a AG |
3274 | static void bg_migration_iteration_finish(MigrationState *s) |
3275 | { | |
3a8b81f2 FE |
3276 | /* |
3277 | * Stop tracking RAM writes - un-protect memory, un-register UFFD | |
3278 | * memory ranges, flush kernel wait queues and wake up threads | |
3279 | * waiting for write fault to be resolved. | |
3280 | */ | |
3281 | ram_write_tracking_stop(); | |
3282 | ||
195801d7 | 3283 | bql_lock(); |
8518278a AG |
3284 | switch (s->state) { |
3285 | case MIGRATION_STATUS_COMPLETED: | |
8518278a AG |
3286 | case MIGRATION_STATUS_ACTIVE: |
3287 | case MIGRATION_STATUS_FAILED: | |
3288 | case MIGRATION_STATUS_CANCELLED: | |
3289 | case MIGRATION_STATUS_CANCELLING: | |
3290 | break; | |
3291 | ||
3292 | default: | |
3293 | /* Should not reach here, but if so, forgive the VM. */ | |
3294 | error_report("%s: Unknown ending state %d", __func__, s->state); | |
3295 | break; | |
3296 | } | |
3297 | ||
44d0d456 | 3298 | migration_bh_schedule(migrate_fd_cleanup_bh, s); |
195801d7 | 3299 | bql_unlock(); |
8518278a AG |
3300 | } |
3301 | ||
3302 | /* | |
3303 | * Return true if continue to the next iteration directly, false | |
3304 | * otherwise. | |
3305 | */ | |
3306 | static MigIterateState bg_migration_iteration_run(MigrationState *s) | |
3307 | { | |
3308 | int res; | |
3309 | ||
3310 | res = qemu_savevm_state_iterate(s->to_dst_file, false); | |
3311 | if (res > 0) { | |
3312 | bg_migration_completion(s); | |
3313 | return MIG_ITERATE_BREAK; | |
3314 | } | |
3315 | ||
3316 | return MIG_ITERATE_RESUME; | |
3317 | } | |
3318 | ||
ad767bed DDAG |
3319 | void migration_make_urgent_request(void) |
3320 | { | |
3321 | qemu_sem_post(&migrate_get_current()->rate_limit_sem); | |
3322 | } | |
3323 | ||
3324 | void migration_consume_urgent_request(void) | |
3325 | { | |
3326 | qemu_sem_wait(&migrate_get_current()->rate_limit_sem); | |
3327 | } | |
3328 | ||
97e1e067 DDAG |
3329 | /* Returns true if the rate limiting was broken by an urgent request */ |
3330 | bool migration_rate_limit(void) | |
3331 | { | |
3332 | int64_t now = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); | |
3333 | MigrationState *s = migrate_get_current(); | |
3334 | ||
3335 | bool urgent = false; | |
3336 | migration_update_counters(s, now); | |
e1fde0e0 | 3337 | if (migration_rate_exceeded(s->to_dst_file)) { |
77386127 LS |
3338 | |
3339 | if (qemu_file_get_error(s->to_dst_file)) { | |
3340 | return false; | |
3341 | } | |
97e1e067 DDAG |
3342 | /* |
3343 | * Wait for a delay to do rate limiting OR | |
3344 | * something urgent to post the semaphore. | |
3345 | */ | |
3346 | int ms = s->iteration_start_time + BUFFER_DELAY - now; | |
3347 | trace_migration_rate_limit_pre(ms); | |
3348 | if (qemu_sem_timedwait(&s->rate_limit_sem, ms) == 0) { | |
3349 | /* | |
3350 | * We were woken by one or more urgent things but | |
3351 | * the timedwait will have consumed one of them. | |
3352 | * The service routine for the urgent wake will dec | |
3353 | * the semaphore itself for each item it consumes, | |
3354 | * so add this one we just eat back. | |
3355 | */ | |
3356 | qemu_sem_post(&s->rate_limit_sem); | |
3357 | urgent = true; | |
3358 | } | |
3359 | trace_migration_rate_limit_post(urgent); | |
3360 | } | |
3361 | return urgent; | |
3362 | } | |
3363 | ||
fde93d99 LV |
3364 | /* |
3365 | * if failover devices are present, wait they are completely | |
3366 | * unplugged | |
3367 | */ | |
3368 | ||
3369 | static void qemu_savevm_wait_unplug(MigrationState *s, int old_state, | |
3370 | int new_state) | |
3371 | { | |
3372 | if (qemu_savevm_state_guest_unplug_pending()) { | |
3373 | migrate_set_state(&s->state, old_state, MIGRATION_STATUS_WAIT_UNPLUG); | |
3374 | ||
3375 | while (s->state == MIGRATION_STATUS_WAIT_UNPLUG && | |
3376 | qemu_savevm_state_guest_unplug_pending()) { | |
3377 | qemu_sem_timedwait(&s->wait_unplug_sem, 250); | |
3378 | } | |
944bc528 LV |
3379 | if (s->state != MIGRATION_STATUS_WAIT_UNPLUG) { |
3380 | int timeout = 120; /* 30 seconds */ | |
3381 | /* | |
3382 | * migration has been canceled | |
3383 | * but as we have started an unplug we must wait the end | |
3384 | * to be able to plug back the card | |
3385 | */ | |
3386 | while (timeout-- && qemu_savevm_state_guest_unplug_pending()) { | |
3387 | qemu_sem_timedwait(&s->wait_unplug_sem, 250); | |
3388 | } | |
1b529d90 LV |
3389 | if (qemu_savevm_state_guest_unplug_pending() && |
3390 | !qtest_enabled()) { | |
a51dcef0 LV |
3391 | warn_report("migration: partially unplugged device on " |
3392 | "failure"); | |
3393 | } | |
944bc528 | 3394 | } |
fde93d99 LV |
3395 | |
3396 | migrate_set_state(&s->state, MIGRATION_STATUS_WAIT_UNPLUG, new_state); | |
3397 | } else { | |
3398 | migrate_set_state(&s->state, old_state, new_state); | |
3399 | } | |
3400 | } | |
3401 | ||
70b20477 DDAG |
3402 | /* |
3403 | * Master migration thread on the source VM. | |
3404 | * It drives the migration and pumps the data down the outgoing channel. | |
3405 | */ | |
5f496a1b | 3406 | static void *migration_thread(void *opaque) |
0d82d0e8 | 3407 | { |
9848a404 | 3408 | MigrationState *s = opaque; |
1b1f4ab6 | 3409 | MigrationThread *thread = NULL; |
bc72ad67 | 3410 | int64_t setup_start = qemu_clock_get_ms(QEMU_CLOCK_HOST); |
b23c2ade | 3411 | MigThrError thr_error; |
ad767bed | 3412 | bool urgent = false; |
76f5933a | 3413 | |
788fa680 | 3414 | thread = migration_threads_add("live_migration", qemu_get_thread_id()); |
1b1f4ab6 | 3415 | |
ab28bd23 PB |
3416 | rcu_register_thread(); |
3417 | ||
892ae715 | 3418 | object_ref(OBJECT(s)); |
87f3bd87 | 3419 | update_iteration_initial_status(s); |
b15df1ae | 3420 | |
dd904bc1 FR |
3421 | if (!multifd_send_setup()) { |
3422 | goto out; | |
3423 | } | |
3424 | ||
195801d7 | 3425 | bql_lock(); |
89a02a9f | 3426 | qemu_savevm_state_header(s->to_dst_file); |
195801d7 | 3427 | bql_unlock(); |
1d34e4bf | 3428 | |
62a02658 PX |
3429 | /* |
3430 | * If we opened the return path, we need to make sure dst has it | |
3431 | * opened as well. | |
3432 | */ | |
43044ac0 | 3433 | if (s->rp_state.rp_thread_created) { |
1d34e4bf | 3434 | /* Now tell the dest that it should open its end so it can reply */ |
89a02a9f | 3435 | qemu_savevm_send_open_return_path(s->to_dst_file); |
1d34e4bf DDAG |
3436 | |
3437 | /* And do a ping that will make stuff easier to debug */ | |
89a02a9f | 3438 | qemu_savevm_send_ping(s->to_dst_file, 1); |
0425dc97 | 3439 | } |
1d34e4bf | 3440 | |
58110f0a | 3441 | if (migrate_postcopy()) { |
1d34e4bf DDAG |
3442 | /* |
3443 | * Tell the destination that we *might* want to do postcopy later; | |
3444 | * if the other end can't do postcopy it should fail now, nice and | |
3445 | * early. | |
3446 | */ | |
89a02a9f | 3447 | qemu_savevm_send_postcopy_advise(s->to_dst_file); |
1d34e4bf DDAG |
3448 | } |
3449 | ||
5e804644 | 3450 | if (migrate_colo()) { |
aad555c2 ZC |
3451 | /* Notify migration destination that we enable COLO */ |
3452 | qemu_savevm_send_colo_enable(s->to_dst_file); | |
3453 | } | |
3454 | ||
195801d7 | 3455 | bql_lock(); |
9907e842 | 3456 | qemu_savevm_state_setup(s->to_dst_file); |
195801d7 | 3457 | bql_unlock(); |
0d82d0e8 | 3458 | |
fde93d99 LV |
3459 | qemu_savevm_wait_unplug(s, MIGRATION_STATUS_SETUP, |
3460 | MIGRATION_STATUS_ACTIVE); | |
c7e0acd5 | 3461 | |
bc72ad67 | 3462 | s->setup_time = qemu_clock_get_ms(QEMU_CLOCK_HOST) - setup_start; |
29ae8a41 | 3463 | |
9ec055ae DDAG |
3464 | trace_migration_thread_setup_complete(); |
3465 | ||
3a6813b6 | 3466 | while (migration_is_active()) { |
e1fde0e0 | 3467 | if (urgent || !migration_rate_exceeded(s->to_dst_file)) { |
2ad87305 PX |
3468 | MigIterateState iter_state = migration_iteration_run(s); |
3469 | if (iter_state == MIG_ITERATE_SKIP) { | |
3470 | continue; | |
3471 | } else if (iter_state == MIG_ITERATE_BREAK) { | |
09f6c85e | 3472 | break; |
c369f40d JQ |
3473 | } |
3474 | } | |
f4410a5d | 3475 | |
b23c2ade PX |
3476 | /* |
3477 | * Try to detect any kind of failures, and see whether we | |
3478 | * should stop the migration now. | |
3479 | */ | |
3480 | thr_error = migration_detect_error(s); | |
3481 | if (thr_error == MIG_THR_ERR_FATAL) { | |
3482 | /* Stop migration */ | |
fd45ee2c | 3483 | break; |
b23c2ade PX |
3484 | } else if (thr_error == MIG_THR_ERR_RECOVERED) { |
3485 | /* | |
3486 | * Just recovered from a e.g. network failure, reset all | |
3487 | * the local variables. This is important to avoid | |
3488 | * breaking transferred_bytes and bandwidth calculation | |
3489 | */ | |
87f3bd87 | 3490 | update_iteration_initial_status(s); |
fd45ee2c | 3491 | } |
b15df1ae | 3492 | |
97e1e067 | 3493 | urgent = migration_rate_limit(); |
a3fa1d78 PB |
3494 | } |
3495 | ||
dd904bc1 | 3496 | out: |
1d34e4bf | 3497 | trace_migration_thread_after_loop(); |
199aa6d4 | 3498 | migration_iteration_finish(s); |
892ae715 | 3499 | object_unref(OBJECT(s)); |
ab28bd23 | 3500 | rcu_unregister_thread(); |
788fa680 | 3501 | migration_threads_remove(thread); |
0d82d0e8 JQ |
3502 | return NULL; |
3503 | } | |
3504 | ||
8518278a AG |
3505 | static void bg_migration_vm_start_bh(void *opaque) |
3506 | { | |
3507 | MigrationState *s = opaque; | |
3508 | ||
49a50206 | 3509 | vm_resume(s->vm_old_state); |
e22ffad0 | 3510 | migration_downtime_end(s); |
8518278a AG |
3511 | } |
3512 | ||
3513 | /** | |
3514 | * Background snapshot thread, based on live migration code. | |
3515 | * This is an alternative implementation of live migration mechanism | |
3516 | * introduced specifically to support background snapshots. | |
3517 | * | |
3518 | * It takes advantage of userfault_fd write protection mechanism introduced | |
3519 | * in v5.7 kernel. Compared to existing dirty page logging migration much | |
3520 | * lesser stream traffic is produced resulting in smaller snapshot images, | |
3521 | * simply cause of no page duplicates can get into the stream. | |
3522 | * | |
3523 | * Another key point is that generated vmstate stream reflects machine state | |
3524 | * 'frozen' at the beginning of snapshot creation compared to dirty page logging | |
3525 | * mechanism, which effectively results in that saved snapshot is the state of VM | |
3526 | * at the end of the process. | |
3527 | */ | |
3528 | static void *bg_migration_thread(void *opaque) | |
3529 | { | |
3530 | MigrationState *s = opaque; | |
3531 | int64_t setup_start; | |
3532 | MigThrError thr_error; | |
3533 | QEMUFile *fb; | |
3534 | bool early_fail = true; | |
3535 | ||
3536 | rcu_register_thread(); | |
3537 | object_ref(OBJECT(s)); | |
3538 | ||
e1fde0e0 | 3539 | migration_rate_set(RATE_LIMIT_DISABLED); |
8518278a AG |
3540 | |
3541 | setup_start = qemu_clock_get_ms(QEMU_CLOCK_HOST); | |
3542 | /* | |
3543 | * We want to save vmstate for the moment when migration has been | |
3544 | * initiated but also we want to save RAM content while VM is running. | |
3545 | * The RAM content should appear first in the vmstate. So, we first | |
3546 | * stash the non-RAM part of the vmstate to the temporary buffer, | |
3547 | * then write RAM part of the vmstate to the migration stream | |
3548 | * with vCPUs running and, finally, write stashed non-RAM part of | |
3549 | * the vmstate from the buffer to the migration stream. | |
3550 | */ | |
ecb23efe | 3551 | s->bioc = qio_channel_buffer_new(512 * 1024); |
8518278a | 3552 | qio_channel_set_name(QIO_CHANNEL(s->bioc), "vmstate-buffer"); |
77ef2dc1 | 3553 | fb = qemu_file_new_output(QIO_CHANNEL(s->bioc)); |
8518278a AG |
3554 | object_unref(OBJECT(s->bioc)); |
3555 | ||
3556 | update_iteration_initial_status(s); | |
3557 | ||
eeccb99c AG |
3558 | /* |
3559 | * Prepare for tracking memory writes with UFFD-WP - populate | |
3560 | * RAM pages before protecting. | |
3561 | */ | |
3562 | #ifdef __linux__ | |
3563 | ram_write_tracking_prepare(); | |
3564 | #endif | |
3565 | ||
195801d7 | 3566 | bql_lock(); |
8518278a AG |
3567 | qemu_savevm_state_header(s->to_dst_file); |
3568 | qemu_savevm_state_setup(s->to_dst_file); | |
195801d7 | 3569 | bql_unlock(); |
8518278a | 3570 | |
fde93d99 LV |
3571 | qemu_savevm_wait_unplug(s, MIGRATION_STATUS_SETUP, |
3572 | MIGRATION_STATUS_ACTIVE); | |
8518278a | 3573 | |
8518278a AG |
3574 | s->setup_time = qemu_clock_get_ms(QEMU_CLOCK_HOST) - setup_start; |
3575 | ||
3576 | trace_migration_thread_setup_complete(); | |
8518278a | 3577 | |
195801d7 | 3578 | bql_lock(); |
8518278a | 3579 | |
9867d4dd | 3580 | if (migration_stop_vm(s, RUN_STATE_PAUSED)) { |
8518278a AG |
3581 | goto fail; |
3582 | } | |
3583 | /* | |
3584 | * Put vCPUs in sync with shadow context structures, then | |
3585 | * save their state to channel-buffer along with devices. | |
3586 | */ | |
3587 | cpu_synchronize_all_states(); | |
3588 | if (qemu_savevm_state_complete_precopy_non_iterable(fb, false, false)) { | |
3589 | goto fail; | |
3590 | } | |
ecb23efe AG |
3591 | /* |
3592 | * Since we are going to get non-iterable state data directly | |
3593 | * from s->bioc->data, explicit flush is needed here. | |
3594 | */ | |
3595 | qemu_fflush(fb); | |
3596 | ||
8518278a AG |
3597 | /* Now initialize UFFD context and start tracking RAM writes */ |
3598 | if (ram_write_tracking_start()) { | |
3599 | goto fail; | |
3600 | } | |
3601 | early_fail = false; | |
3602 | ||
3603 | /* | |
3604 | * Start VM from BH handler to avoid write-fault lock here. | |
3605 | * UFFD-WP protection for the whole RAM is already enabled so | |
3606 | * calling VM state change notifiers from vm_start() would initiate | |
3607 | * writes to virtio VQs memory which is in write-protected region. | |
3608 | */ | |
44d0d456 | 3609 | migration_bh_schedule(bg_migration_vm_start_bh, s); |
195801d7 | 3610 | bql_unlock(); |
8518278a | 3611 | |
3a6813b6 | 3612 | while (migration_is_active()) { |
8518278a AG |
3613 | MigIterateState iter_state = bg_migration_iteration_run(s); |
3614 | if (iter_state == MIG_ITERATE_SKIP) { | |
3615 | continue; | |
3616 | } else if (iter_state == MIG_ITERATE_BREAK) { | |
3617 | break; | |
3618 | } | |
3619 | ||
3620 | /* | |
3621 | * Try to detect any kind of failures, and see whether we | |
3622 | * should stop the migration now. | |
3623 | */ | |
3624 | thr_error = migration_detect_error(s); | |
3625 | if (thr_error == MIG_THR_ERR_FATAL) { | |
3626 | /* Stop migration */ | |
3627 | break; | |
3628 | } | |
3629 | ||
3630 | migration_update_counters(s, qemu_clock_get_ms(QEMU_CLOCK_REALTIME)); | |
3631 | } | |
3632 | ||
3633 | trace_migration_thread_after_loop(); | |
3634 | ||
3635 | fail: | |
3636 | if (early_fail) { | |
3637 | migrate_set_state(&s->state, MIGRATION_STATUS_ACTIVE, | |
3638 | MIGRATION_STATUS_FAILED); | |
195801d7 | 3639 | bql_unlock(); |
8518278a AG |
3640 | } |
3641 | ||
3642 | bg_migration_iteration_finish(s); | |
3643 | ||
3644 | qemu_fclose(fb); | |
3645 | object_unref(OBJECT(s)); | |
3646 | rcu_unregister_thread(); | |
3647 | ||
3648 | return NULL; | |
3649 | } | |
3650 | ||
cce8040b | 3651 | void migrate_fd_connect(MigrationState *s, Error *error_in) |
0d82d0e8 | 3652 | { |
00f4b572 | 3653 | Error *local_err = NULL; |
52033349 | 3654 | uint64_t rate_limit; |
d3e35b8f | 3655 | bool resume = s->state == MIGRATION_STATUS_POSTCOPY_PAUSED; |
9867d4dd | 3656 | int ret; |
d3e35b8f | 3657 | |
ca7bd082 PX |
3658 | /* |
3659 | * If there's a previous error, free it and prepare for another one. | |
3660 | * Meanwhile if migration completes successfully, there won't have an error | |
3661 | * dumped when calling migrate_fd_cleanup(). | |
3662 | */ | |
3663 | migrate_error_free(s); | |
3664 | ||
f5da8ba4 | 3665 | s->expected_downtime = migrate_downtime_limit(); |
cce8040b DDAG |
3666 | if (error_in) { |
3667 | migrate_fd_error(s, error_in); | |
ca30f24d PX |
3668 | if (resume) { |
3669 | /* | |
3670 | * Don't do cleanup for resume if channel is invalid, but only dump | |
3671 | * the error. We wait for another channel connect from the user. | |
3672 | * The error_report still gives HMP user a hint on what failed. | |
3673 | * It's normally done in migrate_fd_cleanup(), but call it here | |
3674 | * explicitly. | |
3675 | */ | |
3676 | error_report_err(error_copy(s->error)); | |
3677 | } else { | |
3678 | migrate_fd_cleanup(s); | |
3679 | } | |
cce8040b DDAG |
3680 | return; |
3681 | } | |
0d82d0e8 | 3682 | |
d3e35b8f PX |
3683 | if (resume) { |
3684 | /* This is a resumed migration */ | |
9d3ebbe2 | 3685 | rate_limit = migrate_max_postcopy_bandwidth(); |
d3e35b8f PX |
3686 | } else { |
3687 | /* This is a fresh new migration */ | |
9d3ebbe2 | 3688 | rate_limit = migrate_max_bandwidth(); |
442773ce | 3689 | |
d3e35b8f | 3690 | /* Notify before starting migration thread */ |
4af667f8 SS |
3691 | if (migration_call_notifiers(s, MIG_EVENT_PRECOPY_SETUP, &local_err)) { |
3692 | goto fail; | |
3693 | } | |
d3e35b8f PX |
3694 | } |
3695 | ||
e1fde0e0 | 3696 | migration_rate_set(rate_limit); |
d3e35b8f | 3697 | qemu_file_set_blocking(s->to_dst_file, true); |
9287ac27 | 3698 | |
1d34e4bf | 3699 | /* |
c788ada8 PX |
3700 | * Open the return path. For postcopy, it is used exclusively. For |
3701 | * precopy, only if user specified "return-path" capability would | |
3702 | * QEMU uses the return path. | |
1d34e4bf | 3703 | */ |
38ad1110 | 3704 | if (migrate_postcopy_ram() || migrate_return_path()) { |
ef796ee9 | 3705 | if (open_return_path_on_source(s)) { |
908927db | 3706 | error_setg(&local_err, "Unable to open return-path for postcopy"); |
bf78a046 | 3707 | goto fail; |
1d34e4bf DDAG |
3708 | } |
3709 | } | |
3710 | ||
06064a67 PX |
3711 | /* |
3712 | * This needs to be done before resuming a postcopy. Note: for newer | |
3713 | * QEMUs we will delay the channel creation until postcopy_start(), to | |
3714 | * avoid disorder of channel creations. | |
3715 | */ | |
3716 | if (migrate_postcopy_preempt() && s->preempt_pre_7_2) { | |
3717 | postcopy_preempt_setup(s); | |
3718 | } | |
3719 | ||
d3e35b8f | 3720 | if (resume) { |
135b87b4 PX |
3721 | /* Wakeup the main migration thread to do the recovery */ |
3722 | migrate_set_state(&s->state, MIGRATION_STATUS_POSTCOPY_PAUSED, | |
3723 | MIGRATION_STATUS_POSTCOPY_RECOVER); | |
3724 | qemu_sem_post(&s->postcopy_pause_sem); | |
d3e35b8f PX |
3725 | return; |
3726 | } | |
3727 | ||
9867d4dd SS |
3728 | if (migrate_mode_is_cpr(s)) { |
3729 | ret = migration_stop_vm(s, RUN_STATE_FINISH_MIGRATE); | |
3730 | if (ret < 0) { | |
3731 | error_setg(&local_err, "migration_stop_vm failed, error %d", -ret); | |
3732 | goto fail; | |
3733 | } | |
3734 | } | |
3735 | ||
8518278a AG |
3736 | if (migrate_background_snapshot()) { |
3737 | qemu_thread_create(&s->thread, "bg_snapshot", | |
3738 | bg_migration_thread, s, QEMU_THREAD_JOINABLE); | |
3739 | } else { | |
3740 | qemu_thread_create(&s->thread, "live_migration", | |
3741 | migration_thread, s, QEMU_THREAD_JOINABLE); | |
3742 | } | |
1d34e4bf | 3743 | s->migration_thread_running = true; |
bf78a046 SS |
3744 | return; |
3745 | ||
3746 | fail: | |
3747 | migrate_set_error(s, local_err); | |
3748 | migrate_set_state(&s->state, s->state, MIGRATION_STATUS_FAILED); | |
3749 | error_report_err(local_err); | |
3750 | migrate_fd_cleanup(s); | |
0d82d0e8 | 3751 | } |
093e3c42 | 3752 | |
e5cb7e76 PX |
3753 | static void migration_class_init(ObjectClass *klass, void *data) |
3754 | { | |
3755 | DeviceClass *dc = DEVICE_CLASS(klass); | |
3756 | ||
3757 | dc->user_creatable = false; | |
4f67d30b | 3758 | device_class_set_props(dc, migration_properties); |
e5cb7e76 PX |
3759 | } |
3760 | ||
b91bf5e4 MAL |
3761 | static void migration_instance_finalize(Object *obj) |
3762 | { | |
3763 | MigrationState *ms = MIGRATION_OBJ(obj); | |
b91bf5e4 | 3764 | |
87db1a7d | 3765 | qemu_mutex_destroy(&ms->error_mutex); |
62df066f | 3766 | qemu_mutex_destroy(&ms->qemu_file_lock); |
c7e0acd5 | 3767 | qemu_sem_destroy(&ms->wait_unplug_sem); |
ad767bed | 3768 | qemu_sem_destroy(&ms->rate_limit_sem); |
e91d8951 | 3769 | qemu_sem_destroy(&ms->pause_sem); |
b23c2ade | 3770 | qemu_sem_destroy(&ms->postcopy_pause_sem); |
edd090c7 | 3771 | qemu_sem_destroy(&ms->rp_state.rp_sem); |
b28fb582 | 3772 | qemu_sem_destroy(&ms->rp_state.rp_pong_acks); |
d0edb8a1 | 3773 | qemu_sem_destroy(&ms->postcopy_qemufile_src_sem); |
ab105cc1 | 3774 | error_free(ms->error); |
b91bf5e4 MAL |
3775 | } |
3776 | ||
e5cb7e76 PX |
3777 | static void migration_instance_init(Object *obj) |
3778 | { | |
3779 | MigrationState *ms = MIGRATION_OBJ(obj); | |
3780 | ||
3781 | ms->state = MIGRATION_STATUS_NONE; | |
e5cb7e76 | 3782 | ms->mbps = -1; |
aecbfe9c | 3783 | ms->pages_per_second = -1; |
e91d8951 | 3784 | qemu_sem_init(&ms->pause_sem, 0); |
87db1a7d | 3785 | qemu_mutex_init(&ms->error_mutex); |
8b0b29dc | 3786 | |
61a174e2 | 3787 | migrate_params_init(&ms->parameters); |
b23c2ade PX |
3788 | |
3789 | qemu_sem_init(&ms->postcopy_pause_sem, 0); | |
edd090c7 | 3790 | qemu_sem_init(&ms->rp_state.rp_sem, 0); |
b28fb582 | 3791 | qemu_sem_init(&ms->rp_state.rp_pong_acks, 0); |
ad767bed | 3792 | qemu_sem_init(&ms->rate_limit_sem, 0); |
c7e0acd5 | 3793 | qemu_sem_init(&ms->wait_unplug_sem, 0); |
d0edb8a1 | 3794 | qemu_sem_init(&ms->postcopy_qemufile_src_sem, 0); |
62df066f | 3795 | qemu_mutex_init(&ms->qemu_file_lock); |
8b0b29dc PX |
3796 | } |
3797 | ||
3798 | /* | |
3799 | * Return true if check pass, false otherwise. Error will be put | |
3800 | * inside errp if provided. | |
3801 | */ | |
3802 | static bool migration_object_check(MigrationState *ms, Error **errp) | |
3803 | { | |
6b19a7d9 | 3804 | /* Assuming all off */ |
b02c7fc9 | 3805 | bool old_caps[MIGRATION_CAPABILITY__MAX] = { 0 }; |
6b19a7d9 | 3806 | |
8b0b29dc PX |
3807 | if (!migrate_params_check(&ms->parameters, errp)) { |
3808 | return false; | |
3809 | } | |
3810 | ||
b02c7fc9 | 3811 | return migrate_caps_check(old_caps, ms->capabilities, errp); |
e5cb7e76 PX |
3812 | } |
3813 | ||
3814 | static const TypeInfo migration_type = { | |
3815 | .name = TYPE_MIGRATION, | |
01f6e14c | 3816 | /* |
c8d3ff38 | 3817 | * NOTE: TYPE_MIGRATION is not really a device, as the object is |
2194abd6 | 3818 | * not created using qdev_new(), it is not attached to the qdev |
c8d3ff38 PX |
3819 | * device tree, and it is never realized. |
3820 | * | |
3821 | * TODO: Make this TYPE_OBJECT once QOM provides something like | |
3822 | * TYPE_DEVICE's "-global" properties. | |
01f6e14c | 3823 | */ |
e5cb7e76 PX |
3824 | .parent = TYPE_DEVICE, |
3825 | .class_init = migration_class_init, | |
3826 | .class_size = sizeof(MigrationClass), | |
3827 | .instance_size = sizeof(MigrationState), | |
3828 | .instance_init = migration_instance_init, | |
b91bf5e4 | 3829 | .instance_finalize = migration_instance_finalize, |
e5cb7e76 PX |
3830 | }; |
3831 | ||
3832 | static void register_migration_types(void) | |
3833 | { | |
3834 | type_register_static(&migration_type); | |
3835 | } | |
3836 | ||
3837 | type_init(register_migration_types); |