]>
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" |
065e2813 | 69 | |
99a0db9b GH |
70 | static NotifierList migration_state_notifiers = |
71 | NOTIFIER_LIST_INITIALIZER(migration_state_notifiers); | |
72 | ||
da6f1790 JQ |
73 | /* Messages sent on the return path from destination to source */ |
74 | enum mig_rp_message_type { | |
75 | MIG_RP_MSG_INVALID = 0, /* Must be 0 */ | |
76 | MIG_RP_MSG_SHUT, /* sibling will not send any more RP messages */ | |
77 | MIG_RP_MSG_PONG, /* Response to a PING; data (seq: be32 ) */ | |
78 | ||
79 | MIG_RP_MSG_REQ_PAGES_ID, /* data (start: be64, len: be32, id: string) */ | |
80 | MIG_RP_MSG_REQ_PAGES, /* data (start: be64, len: be32) */ | |
a335debb | 81 | MIG_RP_MSG_RECV_BITMAP, /* send recved_bitmap back to source */ |
13955b89 | 82 | MIG_RP_MSG_RESUME_ACK, /* tell source that we are ready to resume */ |
1b4adb10 | 83 | MIG_RP_MSG_SWITCHOVER_ACK, /* Tell source it's OK to do switchover */ |
da6f1790 JQ |
84 | |
85 | MIG_RP_MSG_MAX | |
86 | }; | |
87 | ||
17549e84 JQ |
88 | /* When we add fault tolerance, we could have several |
89 | migrations at once. For now we don't need to add | |
90 | dynamic creation of migration */ | |
91 | ||
e5cb7e76 | 92 | static MigrationState *current_migration; |
e1b1b1bc | 93 | static MigrationIncomingState *current_incoming; |
e5cb7e76 | 94 | |
3af8554b DDAG |
95 | static GSList *migration_blockers; |
96 | ||
8b0b29dc | 97 | static bool migration_object_check(MigrationState *ms, Error **errp); |
0331c8ca DDAG |
98 | static int migration_maybe_pause(MigrationState *s, |
99 | int *current_active_state, | |
100 | int new_state); | |
892ae715 | 101 | static void migrate_fd_cancel(MigrationState *s); |
d7f5a043 | 102 | static int close_return_path_on_source(MigrationState *s); |
8b0b29dc | 103 | |
d6f74fd1 PX |
104 | static bool migration_needs_multiple_sockets(void) |
105 | { | |
51b07548 | 106 | return migrate_multifd() || migrate_postcopy_preempt(); |
d6f74fd1 | 107 | } |
f444eeda | 108 | |
d6f74fd1 | 109 | static bool uri_supports_multi_channels(const char *uri) |
f444eeda | 110 | { |
d6f74fd1 PX |
111 | return strstart(uri, "tcp:", NULL) || strstart(uri, "unix:", NULL) || |
112 | strstart(uri, "vsock:", NULL); | |
f444eeda PX |
113 | } |
114 | ||
d6f74fd1 PX |
115 | static bool |
116 | migration_channels_and_uri_compatible(const char *uri, Error **errp) | |
f444eeda | 117 | { |
d6f74fd1 PX |
118 | if (migration_needs_multiple_sockets() && |
119 | !uri_supports_multi_channels(uri)) { | |
120 | error_setg(errp, "Migration requires multi-channel URIs (e.g. tcp)"); | |
121 | return false; | |
122 | } | |
123 | ||
124 | return true; | |
f444eeda PX |
125 | } |
126 | ||
8f8bfffc PX |
127 | static gint page_request_addr_cmp(gconstpointer ap, gconstpointer bp) |
128 | { | |
129 | uintptr_t a = (uintptr_t) ap, b = (uintptr_t) bp; | |
130 | ||
131 | return (a > b) - (a < b); | |
132 | } | |
133 | ||
e5cb7e76 PX |
134 | void migration_object_init(void) |
135 | { | |
136 | /* This can only be called once. */ | |
137 | assert(!current_migration); | |
138 | current_migration = MIGRATION_OBJ(object_new(TYPE_MIGRATION)); | |
4ffdb337 | 139 | |
e1b1b1bc PX |
140 | /* |
141 | * Init the migrate incoming object as well no matter whether | |
142 | * we'll use it or not. | |
143 | */ | |
144 | assert(!current_incoming); | |
145 | current_incoming = g_new0(MigrationIncomingState, 1); | |
146 | current_incoming->state = MIGRATION_STATUS_NONE; | |
147 | current_incoming->postcopy_remote_fds = | |
148 | g_array_new(FALSE, TRUE, sizeof(struct PostCopyFD)); | |
149 | qemu_mutex_init(¤t_incoming->rp_mutex); | |
60bb3c58 | 150 | qemu_mutex_init(¤t_incoming->postcopy_prio_thread_mutex); |
e1b1b1bc PX |
151 | qemu_event_init(¤t_incoming->main_thread_load_event, false); |
152 | qemu_sem_init(¤t_incoming->postcopy_pause_sem_dst, 0); | |
153 | qemu_sem_init(¤t_incoming->postcopy_pause_sem_fault, 0); | |
60bb3c58 | 154 | qemu_sem_init(¤t_incoming->postcopy_pause_sem_fast_load, 0); |
5655aab0 PX |
155 | qemu_sem_init(¤t_incoming->postcopy_qemufile_dst_done, 0); |
156 | ||
8f8bfffc | 157 | qemu_mutex_init(¤t_incoming->page_request_mutex); |
cf02f29e | 158 | qemu_cond_init(¤t_incoming->page_request_cond); |
8f8bfffc | 159 | current_incoming->page_requested = g_tree_new(page_request_addr_cmp); |
e1b1b1bc | 160 | |
f9734d5d | 161 | migration_object_check(current_migration, &error_fatal); |
e0d17dfd PB |
162 | |
163 | blk_mig_init(); | |
164 | ram_mig_init(); | |
165 | dirty_bitmap_mig_init(); | |
e5cb7e76 PX |
166 | } |
167 | ||
458fecca | 168 | void migration_cancel(const Error *error) |
c7c0e724 | 169 | { |
458fecca LV |
170 | if (error) { |
171 | migrate_set_error(current_migration, error); | |
172 | } | |
acac51ba HH |
173 | if (migrate_dirty_limit()) { |
174 | qmp_cancel_vcpu_dirty_limit(false, -1, NULL); | |
175 | } | |
c7c0e724 DH |
176 | migrate_fd_cancel(current_migration); |
177 | } | |
178 | ||
892ae715 | 179 | void migration_shutdown(void) |
1f895604 | 180 | { |
795969ab RL |
181 | /* |
182 | * When the QEMU main thread exit, the COLO thread | |
183 | * may wait a semaphore. So, we should wakeup the | |
184 | * COLO thread before migration shutdown. | |
185 | */ | |
186 | colo_shutdown(); | |
892ae715 DDAG |
187 | /* |
188 | * Cancel the current migration - that will (eventually) | |
189 | * stop the migration using this structure | |
190 | */ | |
458fecca | 191 | migration_cancel(NULL); |
1f895604 | 192 | object_unref(OBJECT(current_migration)); |
1499ab09 VSO |
193 | |
194 | /* | |
195 | * Cancel outgoing migration of dirty bitmaps. It should | |
196 | * at least unref used block nodes. | |
197 | */ | |
198 | dirty_bitmap_mig_cancel_outgoing(); | |
199 | ||
200 | /* | |
201 | * Cancel incoming migration of dirty bitmaps. Dirty bitmaps | |
202 | * are non-critical data, and their loss never considered as | |
203 | * something serious. | |
204 | */ | |
205 | dirty_bitmap_mig_cancel_incoming(); | |
1f895604 VSO |
206 | } |
207 | ||
bca7856a | 208 | /* For outgoing */ |
859bc756 | 209 | MigrationState *migrate_get_current(void) |
17549e84 | 210 | { |
e5cb7e76 PX |
211 | /* This can only be called after the object created. */ |
212 | assert(current_migration); | |
213 | return current_migration; | |
17549e84 JQ |
214 | } |
215 | ||
bca7856a DDAG |
216 | MigrationIncomingState *migration_incoming_get_current(void) |
217 | { | |
e1b1b1bc PX |
218 | assert(current_incoming); |
219 | return current_incoming; | |
bca7856a DDAG |
220 | } |
221 | ||
e031149c PX |
222 | void migration_incoming_transport_cleanup(MigrationIncomingState *mis) |
223 | { | |
224 | if (mis->socket_address_list) { | |
225 | qapi_free_SocketAddressList(mis->socket_address_list); | |
226 | mis->socket_address_list = NULL; | |
227 | } | |
228 | ||
229 | if (mis->transport_cleanup) { | |
230 | mis->transport_cleanup(mis->transport_data); | |
231 | mis->transport_data = mis->transport_cleanup = NULL; | |
232 | } | |
233 | } | |
234 | ||
bca7856a DDAG |
235 | void migration_incoming_state_destroy(void) |
236 | { | |
b4b076da JQ |
237 | struct MigrationIncomingState *mis = migration_incoming_get_current(); |
238 | ||
cfc3bcf3 | 239 | multifd_load_cleanup(); |
c323518a | 240 | compress_threads_load_cleanup(); |
cfc3bcf3 | 241 | |
3482655b | 242 | if (mis->to_src_file) { |
660819b1 PX |
243 | /* Tell source that we are done */ |
244 | migrate_send_rp_shut(mis, qemu_file_get_error(mis->from_src_file) != 0); | |
3482655b PX |
245 | qemu_fclose(mis->to_src_file); |
246 | mis->to_src_file = NULL; | |
247 | } | |
248 | ||
660819b1 | 249 | if (mis->from_src_file) { |
39675fff | 250 | migration_ioc_unregister_yank_from_file(mis->from_src_file); |
660819b1 PX |
251 | qemu_fclose(mis->from_src_file); |
252 | mis->from_src_file = NULL; | |
253 | } | |
00fa4fc8 DDAG |
254 | if (mis->postcopy_remote_fds) { |
255 | g_array_free(mis->postcopy_remote_fds, TRUE); | |
256 | mis->postcopy_remote_fds = NULL; | |
257 | } | |
660819b1 | 258 | |
e031149c | 259 | migration_incoming_transport_cleanup(mis); |
1783c00f DDAG |
260 | qemu_event_reset(&mis->main_thread_load_event); |
261 | ||
8f8bfffc PX |
262 | if (mis->page_requested) { |
263 | g_tree_destroy(mis->page_requested); | |
264 | mis->page_requested = NULL; | |
265 | } | |
266 | ||
36f62f11 PX |
267 | if (mis->postcopy_qemufile_dst) { |
268 | migration_ioc_unregister_yank_from_file(mis->postcopy_qemufile_dst); | |
269 | qemu_fclose(mis->postcopy_qemufile_dst); | |
270 | mis->postcopy_qemufile_dst = NULL; | |
271 | } | |
272 | ||
b5eea99e | 273 | yank_unregister_instance(MIGRATION_YANK_INSTANCE); |
bca7856a DDAG |
274 | } |
275 | ||
b05dc723 JQ |
276 | static void migrate_generate_event(int new_state) |
277 | { | |
b890902c | 278 | if (migrate_events()) { |
3ab72385 | 279 | qapi_event_send_migration(new_state); |
b05dc723 JQ |
280 | } |
281 | } | |
282 | ||
da6f1790 JQ |
283 | /* |
284 | * Send a message on the return channel back to the source | |
285 | * of the migration. | |
286 | */ | |
d6208e35 PX |
287 | static int migrate_send_rp_message(MigrationIncomingState *mis, |
288 | enum mig_rp_message_type message_type, | |
289 | uint16_t len, void *data) | |
da6f1790 | 290 | { |
d6208e35 PX |
291 | int ret = 0; |
292 | ||
da6f1790 | 293 | trace_migrate_send_rp_message((int)message_type, len); |
37396950 | 294 | QEMU_LOCK_GUARD(&mis->rp_mutex); |
d6208e35 PX |
295 | |
296 | /* | |
297 | * It's possible that the file handle got lost due to network | |
298 | * failures. | |
299 | */ | |
300 | if (!mis->to_src_file) { | |
301 | ret = -EIO; | |
37396950 | 302 | return ret; |
d6208e35 PX |
303 | } |
304 | ||
da6f1790 JQ |
305 | qemu_put_be16(mis->to_src_file, (unsigned int)message_type); |
306 | qemu_put_be16(mis->to_src_file, len); | |
307 | qemu_put_buffer(mis->to_src_file, data, len); | |
308 | qemu_fflush(mis->to_src_file); | |
d6208e35 PX |
309 | |
310 | /* It's possible that qemu file got error during sending */ | |
311 | ret = qemu_file_get_error(mis->to_src_file); | |
312 | ||
d6208e35 | 313 | return ret; |
da6f1790 JQ |
314 | } |
315 | ||
2e2bce16 PX |
316 | /* Request one page from the source VM at the given start address. |
317 | * rb: the RAMBlock to request the page in | |
1e2d90eb DDAG |
318 | * Start: Address offset within the RB |
319 | * Len: Length in bytes required - must be a multiple of pagesize | |
320 | */ | |
7a267fc4 PX |
321 | int migrate_send_rp_message_req_pages(MigrationIncomingState *mis, |
322 | RAMBlock *rb, ram_addr_t start) | |
1e2d90eb | 323 | { |
cb8d4c8f | 324 | uint8_t bufc[12 + 1 + 255]; /* start (8), len (4), rbname up to 256 */ |
1e2d90eb | 325 | size_t msglen = 12; /* start + len */ |
2e2bce16 | 326 | size_t len = qemu_ram_pagesize(rb); |
d6208e35 | 327 | enum mig_rp_message_type msg_type; |
2e2bce16 PX |
328 | const char *rbname; |
329 | int rbname_len; | |
1e2d90eb DDAG |
330 | |
331 | *(uint64_t *)bufc = cpu_to_be64((uint64_t)start); | |
332 | *(uint32_t *)(bufc + 8) = cpu_to_be32((uint32_t)len); | |
333 | ||
2e2bce16 PX |
334 | /* |
335 | * We maintain the last ramblock that we requested for page. Note that we | |
336 | * don't need locking because this function will only be called within the | |
337 | * postcopy ram fault thread. | |
338 | */ | |
339 | if (rb != mis->last_rb) { | |
340 | mis->last_rb = rb; | |
341 | ||
342 | rbname = qemu_ram_get_idstr(rb); | |
343 | rbname_len = strlen(rbname); | |
344 | ||
1e2d90eb DDAG |
345 | assert(rbname_len < 256); |
346 | ||
347 | bufc[msglen++] = rbname_len; | |
348 | memcpy(bufc + msglen, rbname, rbname_len); | |
349 | msglen += rbname_len; | |
d6208e35 | 350 | msg_type = MIG_RP_MSG_REQ_PAGES_ID; |
1e2d90eb | 351 | } else { |
d6208e35 | 352 | msg_type = MIG_RP_MSG_REQ_PAGES; |
1e2d90eb | 353 | } |
d6208e35 PX |
354 | |
355 | return migrate_send_rp_message(mis, msg_type, msglen, bufc); | |
1e2d90eb DDAG |
356 | } |
357 | ||
7a267fc4 | 358 | int migrate_send_rp_req_pages(MigrationIncomingState *mis, |
8f8bfffc | 359 | RAMBlock *rb, ram_addr_t start, uint64_t haddr) |
7a267fc4 | 360 | { |
7648297d | 361 | void *aligned = (void *)(uintptr_t)ROUND_DOWN(haddr, qemu_ram_pagesize(rb)); |
a2429283 | 362 | bool received = false; |
8f8bfffc PX |
363 | |
364 | WITH_QEMU_LOCK_GUARD(&mis->page_request_mutex) { | |
365 | received = ramblock_recv_bitmap_test_byte_offset(rb, start); | |
366 | if (!received && !g_tree_lookup(mis->page_requested, aligned)) { | |
367 | /* | |
368 | * The page has not been received, and it's not yet in the page | |
369 | * request list. Queue it. Set the value of element to 1, so that | |
370 | * things like g_tree_lookup() will return TRUE (1) when found. | |
371 | */ | |
372 | g_tree_insert(mis->page_requested, aligned, (gpointer)1); | |
cf02f29e | 373 | qatomic_inc(&mis->page_requested_count); |
8f8bfffc PX |
374 | trace_postcopy_page_req_add(aligned, mis->page_requested_count); |
375 | } | |
376 | } | |
377 | ||
378 | /* | |
379 | * If the page is there, skip sending the message. We don't even need the | |
380 | * lock because as long as the page arrived, it'll be there forever. | |
381 | */ | |
382 | if (received) { | |
383 | return 0; | |
384 | } | |
385 | ||
7a267fc4 PX |
386 | return migrate_send_rp_message_req_pages(mis, rb, start); |
387 | } | |
388 | ||
aad555c2 ZC |
389 | static bool migration_colo_enabled; |
390 | bool migration_incoming_colo_enabled(void) | |
391 | { | |
392 | return migration_colo_enabled; | |
393 | } | |
394 | ||
395 | void migration_incoming_disable_colo(void) | |
396 | { | |
18b1d3c9 | 397 | ram_block_discard_disable(false); |
aad555c2 ZC |
398 | migration_colo_enabled = false; |
399 | } | |
400 | ||
18b1d3c9 | 401 | int migration_incoming_enable_colo(void) |
aad555c2 | 402 | { |
51e47cf8 VSO |
403 | #ifndef CONFIG_REPLICATION |
404 | error_report("ENABLE_COLO command come in migration stream, but COLO " | |
405 | "module is not built in"); | |
406 | return -ENOTSUP; | |
407 | #endif | |
408 | ||
121ccedc VSO |
409 | if (!migrate_colo()) { |
410 | error_report("ENABLE_COLO command come in migration stream, but c-colo " | |
411 | "capability is not set"); | |
412 | return -EINVAL; | |
413 | } | |
414 | ||
18b1d3c9 DH |
415 | if (ram_block_discard_disable(true)) { |
416 | error_report("COLO: cannot disable RAM discard"); | |
417 | return -EBUSY; | |
418 | } | |
aad555c2 | 419 | migration_colo_enabled = true; |
18b1d3c9 | 420 | return 0; |
aad555c2 ZC |
421 | } |
422 | ||
9aca82ba JQ |
423 | void migrate_add_address(SocketAddress *address) |
424 | { | |
425 | MigrationIncomingState *mis = migration_incoming_get_current(); | |
9aca82ba | 426 | |
54aa3de7 EB |
427 | QAPI_LIST_PREPEND(mis->socket_address_list, |
428 | QAPI_CLONE(SocketAddress, address)); | |
9aca82ba JQ |
429 | } |
430 | ||
e69d50d6 | 431 | static void qemu_start_incoming_migration(const char *uri, Error **errp) |
5bb7910a | 432 | { |
d658f65c | 433 | const char *p = NULL; |
4111a732 | 434 | MigrationIncomingState *mis = migration_incoming_get_current(); |
34c9dd8e | 435 | |
d6f74fd1 PX |
436 | /* URI is not suitable for migration? */ |
437 | if (!migration_channels_and_uri_compatible(uri, errp)) { | |
438 | return; | |
439 | } | |
440 | ||
4111a732 FR |
441 | migrate_set_state(&mis->state, MIGRATION_STATUS_NONE, |
442 | MIGRATION_STATUS_SETUP); | |
443 | ||
e69d50d6 PB |
444 | if (strstart(uri, "tcp:", &p) || |
445 | strstart(uri, "unix:", NULL) || | |
446 | strstart(uri, "vsock:", NULL)) { | |
d658f65c | 447 | socket_start_incoming_migration(p ? p : uri, errp); |
2da776db | 448 | #ifdef CONFIG_RDMA |
adde220a | 449 | } else if (strstart(uri, "rdma:", &p)) { |
b88a3306 JQ |
450 | if (migrate_compress()) { |
451 | error_setg(errp, "RDMA and compression can't be used together"); | |
452 | return; | |
453 | } | |
454 | if (migrate_xbzrle()) { | |
455 | error_setg(errp, "RDMA and XBZRLE can't be used together"); | |
456 | return; | |
457 | } | |
458 | if (migrate_multifd()) { | |
459 | error_setg(errp, "RDMA and multifd can't be used together"); | |
460 | return; | |
461 | } | |
2da776db MH |
462 | rdma_start_incoming_migration(p, errp); |
463 | #endif | |
adde220a | 464 | } else if (strstart(uri, "exec:", &p)) { |
43eaae28 | 465 | exec_start_incoming_migration(p, errp); |
adde220a | 466 | } else if (strstart(uri, "fd:", &p)) { |
43eaae28 | 467 | fd_start_incoming_migration(p, errp); |
2a9e2e59 SS |
468 | } else if (strstart(uri, "file:", &p)) { |
469 | file_start_incoming_migration(p, errp); | |
adde220a | 470 | } else { |
312fd5f2 | 471 | error_setg(errp, "unknown migration protocol: %s", uri); |
8ca5e801 | 472 | } |
5bb7910a AL |
473 | } |
474 | ||
0aa6aefc DL |
475 | static void process_incoming_migration_bh(void *opaque) |
476 | { | |
477 | Error *local_err = NULL; | |
478 | MigrationIncomingState *mis = opaque; | |
479 | ||
0f073f44 DDAG |
480 | /* If capability late_block_activate is set: |
481 | * Only fire up the block code now if we're going to restart the | |
482 | * VM, else 'cont' will do it. | |
483 | * This causes file locking to happen; so we don't want it to happen | |
484 | * unless we really are starting the VM. | |
485 | */ | |
486 | if (!migrate_late_block_activate() || | |
487 | (autostart && (!global_state_received() || | |
488 | global_state_get_runstate() == RUN_STATE_RUNNING))) { | |
3b717194 | 489 | /* Make sure all file formats throw away their mutable metadata. |
0f073f44 | 490 | * If we get an error here, just don't restart the VM yet. */ |
3b717194 | 491 | bdrv_activate_all(&local_err); |
0f073f44 DDAG |
492 | if (local_err) { |
493 | error_report_err(local_err); | |
494 | local_err = NULL; | |
495 | autostart = false; | |
496 | } | |
d35ff5e6 KW |
497 | } |
498 | ||
0aa6aefc DL |
499 | /* |
500 | * This must happen after all error conditions are dealt with and | |
501 | * we're sure the VM is going to be running on this host. | |
502 | */ | |
7659505c | 503 | qemu_announce_self(&mis->announce_timer, migrate_announce_params()); |
0aa6aefc | 504 | |
cfc3bcf3 | 505 | multifd_load_shutdown(); |
0aa6aefc | 506 | |
b35ebdf0 VSO |
507 | dirty_bitmap_mig_before_vm_start(); |
508 | ||
0aa6aefc DL |
509 | if (!global_state_received() || |
510 | global_state_get_runstate() == RUN_STATE_RUNNING) { | |
511 | if (autostart) { | |
512 | vm_start(); | |
513 | } else { | |
514 | runstate_set(RUN_STATE_PAUSED); | |
515 | } | |
db009729 ZC |
516 | } else if (migration_incoming_colo_enabled()) { |
517 | migration_incoming_disable_colo(); | |
518 | vm_start(); | |
0aa6aefc DL |
519 | } else { |
520 | runstate_set(global_state_get_runstate()); | |
521 | } | |
0aa6aefc DL |
522 | /* |
523 | * This must happen after any state changes since as soon as an external | |
524 | * observer sees this event they might start to prod at the VM assuming | |
525 | * it's ready to use. | |
526 | */ | |
527 | migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE, | |
528 | MIGRATION_STATUS_COMPLETED); | |
529 | qemu_bh_delete(mis->bh); | |
530 | migration_incoming_state_destroy(); | |
531 | } | |
532 | ||
38e8f9af MAL |
533 | static void coroutine_fn |
534 | process_incoming_migration_co(void *opaque) | |
511c0231 | 535 | { |
b4b076da | 536 | MigrationIncomingState *mis = migration_incoming_get_current(); |
e9bef235 | 537 | PostcopyState ps; |
1c12e1f5 PB |
538 | int ret; |
539 | ||
4f0fae7f | 540 | assert(mis->from_src_file); |
c323518a LS |
541 | |
542 | if (compress_threads_load_setup(mis->from_src_file)) { | |
543 | error_report("Failed to setup decompress threads"); | |
544 | goto fail; | |
545 | } | |
546 | ||
67f11b5c | 547 | mis->largest_page_size = qemu_ram_pagesize_largest(); |
093e3c42 | 548 | postcopy_state_set(POSTCOPY_INCOMING_NONE); |
4111a732 | 549 | migrate_set_state(&mis->state, MIGRATION_STATUS_SETUP, |
93d7af6f | 550 | MIGRATION_STATUS_ACTIVE); |
dd42ce24 VSO |
551 | |
552 | mis->loadvm_co = qemu_coroutine_self(); | |
4f0fae7f | 553 | ret = qemu_loadvm_state(mis->from_src_file); |
dd42ce24 | 554 | mis->loadvm_co = NULL; |
bca7856a | 555 | |
e9bef235 DDAG |
556 | ps = postcopy_state_get(); |
557 | trace_process_incoming_migration_co_end(ret, ps); | |
558 | if (ps != POSTCOPY_INCOMING_NONE) { | |
559 | if (ps == POSTCOPY_INCOMING_ADVISE) { | |
560 | /* | |
561 | * Where a migration had postcopy enabled (and thus went to advise) | |
562 | * but managed to complete within the precopy period, we can use | |
563 | * the normal exit. | |
564 | */ | |
565 | postcopy_ram_incoming_cleanup(mis); | |
566 | } else if (ret >= 0) { | |
567 | /* | |
568 | * Postcopy was started, cleanup should happen at the end of the | |
569 | * postcopy thread. | |
570 | */ | |
571 | trace_process_incoming_migration_co_postcopy_end_main(); | |
572 | return; | |
573 | } | |
574 | /* Else if something went wrong then just fall out of the normal exit */ | |
575 | } | |
576 | ||
ecbfec6d VSO |
577 | if (ret < 0) { |
578 | error_report("load of migration failed: %s", strerror(-ret)); | |
579 | goto fail; | |
580 | } | |
581 | ||
d0a14a2b VSO |
582 | if (colo_incoming_co() < 0) { |
583 | goto fail; | |
25d0c16f HZ |
584 | } |
585 | ||
0aa6aefc DL |
586 | mis->bh = qemu_bh_new(process_incoming_migration_bh, mis); |
587 | qemu_bh_schedule(mis->bh); | |
6d99c2d4 FL |
588 | return; |
589 | fail: | |
6d99c2d4 FL |
590 | migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE, |
591 | MIGRATION_STATUS_FAILED); | |
592 | qemu_fclose(mis->from_src_file); | |
e5bac1f5 LB |
593 | |
594 | multifd_load_cleanup(); | |
c323518a | 595 | compress_threads_load_cleanup(); |
e5bac1f5 | 596 | |
6d99c2d4 | 597 | exit(EXIT_FAILURE); |
511c0231 JQ |
598 | } |
599 | ||
b673eab4 | 600 | /** |
7d6f6933 | 601 | * migration_incoming_setup: Setup incoming migration |
b673eab4 JQ |
602 | * @f: file for main migration channel |
603 | * @errp: where to put errors | |
7d6f6933 MA |
604 | * |
605 | * Returns: %true on success, %false on error. | |
b673eab4 | 606 | */ |
7d6f6933 | 607 | static bool migration_incoming_setup(QEMUFile *f, Error **errp) |
82a4da79 | 608 | { |
4f0fae7f | 609 | MigrationIncomingState *mis = migration_incoming_get_current(); |
82a4da79 | 610 | |
4f0fae7f JQ |
611 | if (!mis->from_src_file) { |
612 | mis->from_src_file = f; | |
613 | } | |
06ad5135 | 614 | qemu_file_set_blocking(f, false); |
7d6f6933 | 615 | return true; |
e595a01a JQ |
616 | } |
617 | ||
36c2f8be | 618 | void migration_incoming_process(void) |
e595a01a JQ |
619 | { |
620 | Coroutine *co = qemu_coroutine_create(process_incoming_migration_co, NULL); | |
0b8b8753 | 621 | qemu_coroutine_enter(co); |
82a4da79 PB |
622 | } |
623 | ||
884835fa | 624 | /* Returns true if recovered from a paused migration, otherwise false */ |
a39e9339 | 625 | static bool postcopy_try_recover(void) |
e595a01a | 626 | { |
d96c9e8d PX |
627 | MigrationIncomingState *mis = migration_incoming_get_current(); |
628 | ||
629 | if (mis->state == MIGRATION_STATUS_POSTCOPY_PAUSED) { | |
630 | /* Resumed from a paused postcopy migration */ | |
631 | ||
a39e9339 PX |
632 | /* This should be set already in migration_incoming_setup() */ |
633 | assert(mis->from_src_file); | |
d96c9e8d | 634 | /* Postcopy has standalone thread to do vm load */ |
a39e9339 | 635 | qemu_file_set_blocking(mis->from_src_file, true); |
d96c9e8d PX |
636 | |
637 | /* Re-configure the return path */ | |
a39e9339 | 638 | mis->to_src_file = qemu_file_get_return_path(mis->from_src_file); |
d96c9e8d PX |
639 | |
640 | migrate_set_state(&mis->state, MIGRATION_STATUS_POSTCOPY_PAUSED, | |
641 | MIGRATION_STATUS_POSTCOPY_RECOVER); | |
642 | ||
643 | /* | |
644 | * Here, we only wake up the main loading thread (while the | |
60bb3c58 | 645 | * rest threads will still be waiting), so that we can receive |
d96c9e8d | 646 | * commands from source now, and answer it if needed. The |
60bb3c58 | 647 | * rest threads will be woken up afterwards until we are sure |
d96c9e8d PX |
648 | * that source is ready to reply to page requests. |
649 | */ | |
650 | qemu_sem_post(&mis->postcopy_pause_sem_dst); | |
884835fa PX |
651 | return true; |
652 | } | |
653 | ||
654 | return false; | |
655 | } | |
656 | ||
b673eab4 | 657 | void migration_fd_process_incoming(QEMUFile *f, Error **errp) |
884835fa | 658 | { |
a39e9339 | 659 | if (!migration_incoming_setup(f, errp)) { |
884835fa | 660 | return; |
d96c9e8d | 661 | } |
a39e9339 | 662 | if (postcopy_try_recover()) { |
b673eab4 JQ |
663 | return; |
664 | } | |
884835fa | 665 | migration_incoming_process(); |
e595a01a JQ |
666 | } |
667 | ||
5655aab0 PX |
668 | /* |
669 | * Returns true when we want to start a new incoming migration process, | |
670 | * false otherwise. | |
671 | */ | |
672 | static bool migration_should_start_incoming(bool main_channel) | |
673 | { | |
674 | /* Multifd doesn't start unless all channels are established */ | |
51b07548 | 675 | if (migrate_multifd()) { |
5655aab0 PX |
676 | return migration_has_all_channels(); |
677 | } | |
678 | ||
679 | /* Preempt channel only starts when the main channel is created */ | |
680 | if (migrate_postcopy_preempt()) { | |
681 | return main_channel; | |
682 | } | |
683 | ||
684 | /* | |
685 | * For all the rest types of migration, we should only reach here when | |
686 | * it's the main channel that's being created, and we should always | |
687 | * proceed with this channel. | |
688 | */ | |
689 | assert(main_channel); | |
690 | return true; | |
691 | } | |
692 | ||
49ed0d24 | 693 | void migration_ioc_process_incoming(QIOChannel *ioc, Error **errp) |
4f0fae7f JQ |
694 | { |
695 | MigrationIncomingState *mis = migration_incoming_get_current(); | |
b673eab4 | 696 | Error *local_err = NULL; |
36f62f11 | 697 | QEMUFile *f; |
6720c2b3 | 698 | bool default_channel = true; |
699 | uint32_t channel_magic = 0; | |
700 | int ret = 0; | |
4f0fae7f | 701 | |
51b07548 | 702 | if (migrate_multifd() && !migrate_postcopy_ram() && |
6720c2b3 | 703 | qio_channel_has_feature(ioc, QIO_CHANNEL_FEATURE_READ_MSG_PEEK)) { |
704 | /* | |
705 | * With multiple channels, it is possible that we receive channels | |
706 | * out of order on destination side, causing incorrect mapping of | |
707 | * source channels on destination side. Check channel MAGIC to | |
708 | * decide type of channel. Please note this is best effort, postcopy | |
709 | * preempt channel does not send any magic number so avoid it for | |
710 | * postcopy live migration. Also tls live migration already does | |
711 | * tls handshake while initializing main channel so with tls this | |
712 | * issue is not possible. | |
713 | */ | |
714 | ret = migration_channel_read_peek(ioc, (void *)&channel_magic, | |
715 | sizeof(channel_magic), &local_err); | |
716 | ||
717 | if (ret != 0) { | |
718 | error_propagate(errp, local_err); | |
719 | return; | |
720 | } | |
721 | ||
722 | default_channel = (channel_magic == cpu_to_be32(QEMU_VM_FILE_MAGIC)); | |
723 | } else { | |
724 | default_channel = !mis->from_src_file; | |
725 | } | |
726 | ||
727 | if (multifd_load_setup(errp) != 0) { | |
728 | error_setg(errp, "Failed to setup multifd channels"); | |
729 | return; | |
730 | } | |
731 | ||
732 | if (default_channel) { | |
36f62f11 | 733 | f = qemu_file_new_input(ioc); |
a429e7f4 | 734 | |
7d6f6933 | 735 | if (!migration_incoming_setup(f, errp)) { |
b673eab4 JQ |
736 | return; |
737 | } | |
a429e7f4 PX |
738 | } else { |
739 | /* Multiple connections */ | |
36f62f11 | 740 | assert(migration_needs_multiple_sockets()); |
51b07548 | 741 | if (migrate_multifd()) { |
6720c2b3 | 742 | multifd_recv_new_channel(ioc, &local_err); |
36f62f11 PX |
743 | } else { |
744 | assert(migrate_postcopy_preempt()); | |
745 | f = qemu_file_new_input(ioc); | |
6720c2b3 | 746 | postcopy_preempt_new_channel(mis, f); |
36f62f11 | 747 | } |
49ed0d24 FL |
748 | if (local_err) { |
749 | error_propagate(errp, local_err); | |
750 | return; | |
751 | } | |
4f0fae7f | 752 | } |
81e62053 | 753 | |
5655aab0 | 754 | if (migration_should_start_incoming(default_channel)) { |
a39e9339 PX |
755 | /* If it's a recovery, we're done */ |
756 | if (postcopy_try_recover()) { | |
757 | return; | |
758 | } | |
81e62053 PX |
759 | migration_incoming_process(); |
760 | } | |
4f0fae7f JQ |
761 | } |
762 | ||
428d8908 JQ |
763 | /** |
764 | * @migration_has_all_channels: We have received all channels that we need | |
765 | * | |
766 | * Returns true when we have got connections to all the channels that | |
767 | * we need for migration. | |
768 | */ | |
769 | bool migration_has_all_channels(void) | |
770 | { | |
ca273df3 | 771 | MigrationIncomingState *mis = migration_incoming_get_current(); |
62c1e0ca | 772 | |
36f62f11 PX |
773 | if (!mis->from_src_file) { |
774 | return false; | |
775 | } | |
62c1e0ca | 776 | |
51b07548 | 777 | if (migrate_multifd()) { |
36f62f11 PX |
778 | return multifd_recv_all_channels_created(); |
779 | } | |
780 | ||
781 | if (migrate_postcopy_preempt()) { | |
782 | return mis->postcopy_qemufile_dst != NULL; | |
783 | } | |
784 | ||
785 | return true; | |
428d8908 JQ |
786 | } |
787 | ||
1b4adb10 AH |
788 | int migrate_send_rp_switchover_ack(MigrationIncomingState *mis) |
789 | { | |
790 | return migrate_send_rp_message(mis, MIG_RP_MSG_SWITCHOVER_ACK, 0, NULL); | |
791 | } | |
792 | ||
6decec93 DDAG |
793 | /* |
794 | * Send a 'SHUT' message on the return channel with the given value | |
795 | * to indicate that we've finished with the RP. Non-0 value indicates | |
796 | * error. | |
797 | */ | |
798 | void migrate_send_rp_shut(MigrationIncomingState *mis, | |
799 | uint32_t value) | |
800 | { | |
801 | uint32_t buf; | |
802 | ||
803 | buf = cpu_to_be32(value); | |
804 | migrate_send_rp_message(mis, MIG_RP_MSG_SHUT, sizeof(buf), &buf); | |
805 | } | |
806 | ||
807 | /* | |
808 | * Send a 'PONG' message on the return channel with the given value | |
809 | * (normally in response to a 'PING') | |
810 | */ | |
811 | void migrate_send_rp_pong(MigrationIncomingState *mis, | |
812 | uint32_t value) | |
813 | { | |
814 | uint32_t buf; | |
815 | ||
816 | buf = cpu_to_be32(value); | |
817 | migrate_send_rp_message(mis, MIG_RP_MSG_PONG, sizeof(buf), &buf); | |
818 | } | |
819 | ||
a335debb PX |
820 | void migrate_send_rp_recv_bitmap(MigrationIncomingState *mis, |
821 | char *block_name) | |
822 | { | |
823 | char buf[512]; | |
824 | int len; | |
825 | int64_t res; | |
826 | ||
827 | /* | |
828 | * First, we send the header part. It contains only the len of | |
829 | * idstr, and the idstr itself. | |
830 | */ | |
831 | len = strlen(block_name); | |
832 | buf[0] = len; | |
833 | memcpy(buf + 1, block_name, len); | |
834 | ||
835 | if (mis->state != MIGRATION_STATUS_POSTCOPY_RECOVER) { | |
836 | error_report("%s: MSG_RP_RECV_BITMAP only used for recovery", | |
837 | __func__); | |
838 | return; | |
839 | } | |
840 | ||
841 | migrate_send_rp_message(mis, MIG_RP_MSG_RECV_BITMAP, len + 1, buf); | |
842 | ||
843 | /* | |
844 | * Next, we dump the received bitmap to the stream. | |
845 | * | |
846 | * TODO: currently we are safe since we are the only one that is | |
847 | * using the to_src_file handle (fault thread is still paused), | |
848 | * and it's ok even not taking the mutex. However the best way is | |
849 | * to take the lock before sending the message header, and release | |
850 | * the lock after sending the bitmap. | |
851 | */ | |
852 | qemu_mutex_lock(&mis->rp_mutex); | |
853 | res = ramblock_recv_bitmap_send(mis->to_src_file, block_name); | |
854 | qemu_mutex_unlock(&mis->rp_mutex); | |
855 | ||
856 | trace_migrate_send_rp_recv_bitmap(block_name, res); | |
857 | } | |
858 | ||
13955b89 PX |
859 | void migrate_send_rp_resume_ack(MigrationIncomingState *mis, uint32_t value) |
860 | { | |
861 | uint32_t buf; | |
862 | ||
863 | buf = cpu_to_be32(value); | |
864 | migrate_send_rp_message(mis, MIG_RP_MSG_RESUME_ACK, sizeof(buf), &buf); | |
865 | } | |
866 | ||
f6844b99 DDAG |
867 | /* |
868 | * Return true if we're already in the middle of a migration | |
869 | * (i.e. any of the active or setup states) | |
870 | */ | |
3d63da16 | 871 | bool migration_is_setup_or_active(int state) |
f6844b99 DDAG |
872 | { |
873 | switch (state) { | |
874 | case MIGRATION_STATUS_ACTIVE: | |
9ec055ae | 875 | case MIGRATION_STATUS_POSTCOPY_ACTIVE: |
a688d2c1 | 876 | case MIGRATION_STATUS_POSTCOPY_PAUSED: |
135b87b4 | 877 | case MIGRATION_STATUS_POSTCOPY_RECOVER: |
f6844b99 | 878 | case MIGRATION_STATUS_SETUP: |
31e06077 DDAG |
879 | case MIGRATION_STATUS_PRE_SWITCHOVER: |
880 | case MIGRATION_STATUS_DEVICE: | |
c7e0acd5 | 881 | case MIGRATION_STATUS_WAIT_UNPLUG: |
19dd408a | 882 | case MIGRATION_STATUS_COLO: |
f6844b99 DDAG |
883 | return true; |
884 | ||
885 | default: | |
886 | return false; | |
887 | ||
888 | } | |
889 | } | |
890 | ||
392d87e2 JQ |
891 | bool migration_is_running(int state) |
892 | { | |
893 | switch (state) { | |
894 | case MIGRATION_STATUS_ACTIVE: | |
895 | case MIGRATION_STATUS_POSTCOPY_ACTIVE: | |
896 | case MIGRATION_STATUS_POSTCOPY_PAUSED: | |
897 | case MIGRATION_STATUS_POSTCOPY_RECOVER: | |
898 | case MIGRATION_STATUS_SETUP: | |
899 | case MIGRATION_STATUS_PRE_SWITCHOVER: | |
900 | case MIGRATION_STATUS_DEVICE: | |
901 | case MIGRATION_STATUS_WAIT_UNPLUG: | |
902 | case MIGRATION_STATUS_CANCELLING: | |
392d87e2 JQ |
903 | return true; |
904 | ||
905 | default: | |
906 | return false; | |
907 | ||
908 | } | |
909 | } | |
910 | ||
db18dee7 PX |
911 | static bool migrate_show_downtime(MigrationState *s) |
912 | { | |
913 | return (s->state == MIGRATION_STATUS_COMPLETED) || migration_in_postcopy(); | |
914 | } | |
915 | ||
640dfb14 WY |
916 | static void populate_time_info(MigrationInfo *info, MigrationState *s) |
917 | { | |
918 | info->has_status = true; | |
919 | info->has_setup_time = true; | |
920 | info->setup_time = s->setup_time; | |
db18dee7 | 921 | |
640dfb14 WY |
922 | if (s->state == MIGRATION_STATUS_COMPLETED) { |
923 | info->has_total_time = true; | |
924 | info->total_time = s->total_time; | |
640dfb14 WY |
925 | } else { |
926 | info->has_total_time = true; | |
927 | info->total_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME) - | |
928 | s->start_time; | |
db18dee7 PX |
929 | } |
930 | ||
931 | if (migrate_show_downtime(s)) { | |
932 | info->has_downtime = true; | |
933 | info->downtime = s->downtime; | |
934 | } else { | |
640dfb14 WY |
935 | info->has_expected_downtime = true; |
936 | info->expected_downtime = s->expected_downtime; | |
937 | } | |
938 | } | |
939 | ||
a22463a5 DDAG |
940 | static void populate_ram_info(MigrationInfo *info, MigrationState *s) |
941 | { | |
144fa06b JQ |
942 | size_t page_size = qemu_target_page_size(); |
943 | ||
a22463a5 | 944 | info->ram = g_malloc0(sizeof(*info->ram)); |
aff3f660 | 945 | info->ram->transferred = stat64_get(&mig_stats.transferred); |
a22463a5 | 946 | info->ram->total = ram_bytes_total(); |
aff3f660 | 947 | info->ram->duplicate = stat64_get(&mig_stats.zero_pages); |
bedf53c1 JQ |
948 | /* legacy value. It is not used anymore */ |
949 | info->ram->skipped = 0; | |
aff3f660 | 950 | info->ram->normal = stat64_get(&mig_stats.normal_pages); |
23b7576d | 951 | info->ram->normal_bytes = info->ram->normal * page_size; |
a22463a5 | 952 | info->ram->mbps = s->mbps; |
536b5a4e | 953 | info->ram->dirty_sync_count = |
aff3f660 | 954 | stat64_get(&mig_stats.dirty_sync_count); |
cf20c897 | 955 | info->ram->dirty_sync_missed_zero_copy = |
aff3f660 | 956 | stat64_get(&mig_stats.dirty_sync_missed_zero_copy); |
3c764f9b | 957 | info->ram->postcopy_requests = |
aff3f660 | 958 | stat64_get(&mig_stats.postcopy_requests); |
144fa06b | 959 | info->ram->page_size = page_size; |
aff3f660 | 960 | info->ram->multifd_bytes = stat64_get(&mig_stats.multifd_bytes); |
aecbfe9c | 961 | info->ram->pages_per_second = s->pages_per_second; |
aff3f660 JQ |
962 | info->ram->precopy_bytes = stat64_get(&mig_stats.precopy_bytes); |
963 | info->ram->downtime_bytes = stat64_get(&mig_stats.downtime_bytes); | |
964 | info->ram->postcopy_bytes = stat64_get(&mig_stats.postcopy_bytes); | |
a22463a5 | 965 | |
87dca0c9 | 966 | if (migrate_xbzrle()) { |
114f5aee JQ |
967 | info->xbzrle_cache = g_malloc0(sizeof(*info->xbzrle_cache)); |
968 | info->xbzrle_cache->cache_size = migrate_xbzrle_cache_size(); | |
9360447d JQ |
969 | info->xbzrle_cache->bytes = xbzrle_counters.bytes; |
970 | info->xbzrle_cache->pages = xbzrle_counters.pages; | |
971 | info->xbzrle_cache->cache_miss = xbzrle_counters.cache_miss; | |
972 | info->xbzrle_cache->cache_miss_rate = xbzrle_counters.cache_miss_rate; | |
e460a4b1 | 973 | info->xbzrle_cache->encoding_rate = xbzrle_counters.encoding_rate; |
9360447d | 974 | info->xbzrle_cache->overflow = xbzrle_counters.overflow; |
114f5aee JQ |
975 | } |
976 | ||
6f609005 | 977 | populate_compress(info); |
76e03000 | 978 | |
338182c8 JQ |
979 | if (cpu_throttle_active()) { |
980 | info->has_cpu_throttle_percentage = true; | |
981 | info->cpu_throttle_percentage = cpu_throttle_get_percentage(); | |
982 | } | |
983 | ||
a22463a5 DDAG |
984 | if (s->state != MIGRATION_STATUS_COMPLETED) { |
985 | info->ram->remaining = ram_bytes_remaining(); | |
72f8e587 | 986 | info->ram->dirty_pages_rate = |
aff3f660 | 987 | stat64_get(&mig_stats.dirty_pages_rate); |
a22463a5 | 988 | } |
15699cf5 HH |
989 | |
990 | if (migrate_dirty_limit() && dirtylimit_in_service()) { | |
991 | info->has_dirty_limit_throttle_time_per_round = true; | |
992 | info->dirty_limit_throttle_time_per_round = | |
993 | dirtylimit_throttle_time_per_round(); | |
994 | ||
995 | info->has_dirty_limit_ring_full_time = true; | |
996 | info->dirty_limit_ring_full_time = dirtylimit_ring_full_time(); | |
997 | } | |
a22463a5 DDAG |
998 | } |
999 | ||
930ac04c JQ |
1000 | static void populate_disk_info(MigrationInfo *info) |
1001 | { | |
1002 | if (blk_mig_active()) { | |
930ac04c JQ |
1003 | info->disk = g_malloc0(sizeof(*info->disk)); |
1004 | info->disk->transferred = blk_mig_bytes_transferred(); | |
1005 | info->disk->remaining = blk_mig_bytes_remaining(); | |
1006 | info->disk->total = blk_mig_bytes_total(); | |
1007 | } | |
1008 | } | |
1009 | ||
65ace060 | 1010 | static void fill_source_migration_info(MigrationInfo *info) |
5bb7910a | 1011 | { |
17549e84 | 1012 | MigrationState *s = migrate_get_current(); |
552de79b | 1013 | int state = qatomic_read(&s->state); |
372043f3 | 1014 | GSList *cur_blocker = migration_blockers; |
17549e84 | 1015 | |
3af8554b | 1016 | info->blocked_reasons = NULL; |
3af8554b | 1017 | |
372043f3 MA |
1018 | /* |
1019 | * There are two types of reasons a migration might be blocked; | |
1020 | * a) devices marked in VMState as non-migratable, and | |
1021 | * b) Explicit migration blockers | |
1022 | * We need to add both of them here. | |
1023 | */ | |
1024 | qemu_savevm_non_migratable_list(&info->blocked_reasons); | |
3af8554b | 1025 | |
372043f3 MA |
1026 | while (cur_blocker) { |
1027 | QAPI_LIST_PREPEND(info->blocked_reasons, | |
1028 | g_strdup(error_get_pretty(cur_blocker->data))); | |
1029 | cur_blocker = g_slist_next(cur_blocker); | |
3af8554b | 1030 | } |
372043f3 | 1031 | info->has_blocked_reasons = info->blocked_reasons != NULL; |
3af8554b | 1032 | |
552de79b | 1033 | switch (state) { |
31194731 | 1034 | case MIGRATION_STATUS_NONE: |
17549e84 | 1035 | /* no migration has happened ever */ |
65ace060 AP |
1036 | /* do not overwrite destination migration status */ |
1037 | return; | |
31194731 | 1038 | case MIGRATION_STATUS_SETUP: |
29ae8a41 | 1039 | info->has_status = true; |
ed4fbd10 | 1040 | info->has_total_time = false; |
29ae8a41 | 1041 | break; |
31194731 HZ |
1042 | case MIGRATION_STATUS_ACTIVE: |
1043 | case MIGRATION_STATUS_CANCELLING: | |
9ec055ae | 1044 | case MIGRATION_STATUS_POSTCOPY_ACTIVE: |
31e06077 DDAG |
1045 | case MIGRATION_STATUS_PRE_SWITCHOVER: |
1046 | case MIGRATION_STATUS_DEVICE: | |
a688d2c1 | 1047 | case MIGRATION_STATUS_POSTCOPY_PAUSED: |
135b87b4 | 1048 | case MIGRATION_STATUS_POSTCOPY_RECOVER: |
640dfb14 WY |
1049 | /* TODO add some postcopy stats */ |
1050 | populate_time_info(info, s); | |
a22463a5 | 1051 | populate_ram_info(info, s); |
930ac04c | 1052 | populate_disk_info(info); |
38c482b4 | 1053 | migration_populate_vfio_info(info); |
17549e84 | 1054 | break; |
0b827d5e HZ |
1055 | case MIGRATION_STATUS_COLO: |
1056 | info->has_status = true; | |
1057 | /* TODO: display COLO specific information (checkpoint info etc.) */ | |
1058 | break; | |
31194731 | 1059 | case MIGRATION_STATUS_COMPLETED: |
640dfb14 | 1060 | populate_time_info(info, s); |
a22463a5 | 1061 | populate_ram_info(info, s); |
38c482b4 | 1062 | migration_populate_vfio_info(info); |
17549e84 | 1063 | break; |
31194731 | 1064 | case MIGRATION_STATUS_FAILED: |
791e7c82 | 1065 | info->has_status = true; |
17549e84 | 1066 | break; |
31194731 | 1067 | case MIGRATION_STATUS_CANCELLED: |
791e7c82 | 1068 | info->has_status = true; |
17549e84 | 1069 | break; |
c7e0acd5 JF |
1070 | case MIGRATION_STATUS_WAIT_UNPLUG: |
1071 | info->has_status = true; | |
1072 | break; | |
5bb7910a | 1073 | } |
552de79b | 1074 | info->status = state; |
c94143e5 PX |
1075 | |
1076 | QEMU_LOCK_GUARD(&s->error_mutex); | |
1077 | if (s->error) { | |
1078 | info->error_desc = g_strdup(error_get_pretty(s->error)); | |
1079 | } | |
5bb7910a AL |
1080 | } |
1081 | ||
65ace060 AP |
1082 | static void fill_destination_migration_info(MigrationInfo *info) |
1083 | { | |
1084 | MigrationIncomingState *mis = migration_incoming_get_current(); | |
1085 | ||
9aca82ba JQ |
1086 | if (mis->socket_address_list) { |
1087 | info->has_socket_address = true; | |
1088 | info->socket_address = | |
1089 | QAPI_CLONE(SocketAddressList, mis->socket_address_list); | |
1090 | } | |
1091 | ||
65ace060 AP |
1092 | switch (mis->state) { |
1093 | case MIGRATION_STATUS_NONE: | |
1094 | return; | |
65ace060 AP |
1095 | case MIGRATION_STATUS_SETUP: |
1096 | case MIGRATION_STATUS_CANCELLING: | |
1097 | case MIGRATION_STATUS_CANCELLED: | |
1098 | case MIGRATION_STATUS_ACTIVE: | |
1099 | case MIGRATION_STATUS_POSTCOPY_ACTIVE: | |
3c9928d9 PX |
1100 | case MIGRATION_STATUS_POSTCOPY_PAUSED: |
1101 | case MIGRATION_STATUS_POSTCOPY_RECOVER: | |
65ace060 AP |
1102 | case MIGRATION_STATUS_FAILED: |
1103 | case MIGRATION_STATUS_COLO: | |
1104 | info->has_status = true; | |
1105 | break; | |
1106 | case MIGRATION_STATUS_COMPLETED: | |
1107 | info->has_status = true; | |
1108 | fill_destination_postcopy_migration_info(info); | |
1109 | break; | |
1110 | } | |
1111 | info->status = mis->state; | |
1112 | } | |
1113 | ||
1114 | MigrationInfo *qmp_query_migrate(Error **errp) | |
1115 | { | |
1116 | MigrationInfo *info = g_malloc0(sizeof(*info)); | |
1117 | ||
1118 | fill_destination_migration_info(info); | |
1119 | fill_source_migration_info(info); | |
1120 | ||
1121 | return info; | |
1122 | } | |
1123 | ||
4886a1bc DDAG |
1124 | void qmp_migrate_start_postcopy(Error **errp) |
1125 | { | |
1126 | MigrationState *s = migrate_get_current(); | |
1127 | ||
16b0fd32 | 1128 | if (!migrate_postcopy()) { |
a54d340b | 1129 | error_setg(errp, "Enable postcopy with migrate_set_capability before" |
4886a1bc DDAG |
1130 | " the start of migration"); |
1131 | return; | |
1132 | } | |
1133 | ||
1134 | if (s->state == MIGRATION_STATUS_NONE) { | |
1135 | error_setg(errp, "Postcopy must be started after migration has been" | |
1136 | " started"); | |
1137 | return; | |
1138 | } | |
1139 | /* | |
1140 | * we don't error if migration has finished since that would be racy | |
1141 | * with issuing this command. | |
1142 | */ | |
d73415a3 | 1143 | qatomic_set(&s->start_postcopy, true); |
4886a1bc DDAG |
1144 | } |
1145 | ||
065e2813 AL |
1146 | /* shared migration helpers */ |
1147 | ||
48781e5b | 1148 | void migrate_set_state(int *state, int old_state, int new_state) |
51cf4c1a | 1149 | { |
a31fedee | 1150 | assert(new_state < MIGRATION_STATUS__MAX); |
d73415a3 | 1151 | if (qatomic_cmpxchg(state, old_state, new_state) == old_state) { |
a31fedee | 1152 | trace_migrate_set_state(MigrationStatus_str(new_state)); |
b05dc723 | 1153 | migrate_generate_event(new_state); |
51cf4c1a Z |
1154 | } |
1155 | } | |
1156 | ||
fd392cfa | 1157 | static void migrate_fd_cleanup(MigrationState *s) |
065e2813 | 1158 | { |
bb1fadc4 PB |
1159 | qemu_bh_delete(s->cleanup_bh); |
1160 | s->cleanup_bh = NULL; | |
1161 | ||
83174765 PX |
1162 | g_free(s->hostname); |
1163 | s->hostname = NULL; | |
e3bf5e68 DH |
1164 | json_writer_free(s->vmdesc); |
1165 | s->vmdesc = NULL; | |
83174765 | 1166 | |
0ceccd85 PX |
1167 | qemu_savevm_state_cleanup(); |
1168 | ||
89a02a9f | 1169 | if (s->to_dst_file) { |
62df066f | 1170 | QEMUFile *tmp; |
f986c3d2 | 1171 | |
9013dca5 | 1172 | trace_migrate_fd_cleanup(); |
404a7c05 | 1173 | qemu_mutex_unlock_iothread(); |
1d34e4bf DDAG |
1174 | if (s->migration_thread_running) { |
1175 | qemu_thread_join(&s->thread); | |
1176 | s->migration_thread_running = false; | |
1177 | } | |
404a7c05 PB |
1178 | qemu_mutex_lock_iothread(); |
1179 | ||
1398b2e3 | 1180 | multifd_save_cleanup(); |
62df066f PX |
1181 | qemu_mutex_lock(&s->qemu_file_lock); |
1182 | tmp = s->to_dst_file; | |
89a02a9f | 1183 | s->to_dst_file = NULL; |
62df066f PX |
1184 | qemu_mutex_unlock(&s->qemu_file_lock); |
1185 | /* | |
1186 | * Close the file handle without the lock to make sure the | |
1187 | * critical section won't block for long. | |
1188 | */ | |
39675fff | 1189 | migration_ioc_unregister_yank_from_file(tmp); |
62df066f | 1190 | qemu_fclose(tmp); |
065e2813 AL |
1191 | } |
1192 | ||
36e9aab3 FR |
1193 | /* |
1194 | * We already cleaned up to_dst_file, so errors from the return | |
1195 | * path might be due to that, ignore them. | |
1196 | */ | |
d7f5a043 | 1197 | close_return_path_on_source(s); |
36e9aab3 | 1198 | |
8f8d528e | 1199 | assert(!migration_is_active(s)); |
7a2c1721 | 1200 | |
94f5a437 | 1201 | if (s->state == MIGRATION_STATUS_CANCELLING) { |
48781e5b | 1202 | migrate_set_state(&s->state, MIGRATION_STATUS_CANCELLING, |
94f5a437 | 1203 | MIGRATION_STATUS_CANCELLED); |
7a2c1721 | 1204 | } |
a3fa1d78 | 1205 | |
87db1a7d JQ |
1206 | if (s->error) { |
1207 | /* It is used on info migrate. We can't free it */ | |
1208 | error_report_err(error_copy(s->error)); | |
1209 | } | |
d9cda213 | 1210 | migration_call_notifiers(s); |
b1a87956 | 1211 | block_cleanup_parameters(); |
b5eea99e | 1212 | yank_unregister_instance(MIGRATION_YANK_INSTANCE); |
065e2813 AL |
1213 | } |
1214 | ||
fd392cfa YK |
1215 | static void migrate_fd_cleanup_schedule(MigrationState *s) |
1216 | { | |
1217 | /* | |
1218 | * Ref the state for bh, because it may be called when | |
1219 | * there're already no other refs | |
1220 | */ | |
1221 | object_ref(OBJECT(s)); | |
1222 | qemu_bh_schedule(s->cleanup_bh); | |
1223 | } | |
1224 | ||
1225 | static void migrate_fd_cleanup_bh(void *opaque) | |
1226 | { | |
1227 | MigrationState *s = opaque; | |
1228 | migrate_fd_cleanup(s); | |
1229 | object_unref(OBJECT(s)); | |
1230 | } | |
1231 | ||
87db1a7d JQ |
1232 | void migrate_set_error(MigrationState *s, const Error *error) |
1233 | { | |
6e8a355d | 1234 | QEMU_LOCK_GUARD(&s->error_mutex); |
87db1a7d JQ |
1235 | if (!s->error) { |
1236 | s->error = error_copy(error); | |
1237 | } | |
87db1a7d JQ |
1238 | } |
1239 | ||
2b2f6f41 PX |
1240 | bool migrate_has_error(MigrationState *s) |
1241 | { | |
1242 | /* The lock is not helpful here, but still follow the rule */ | |
1243 | QEMU_LOCK_GUARD(&s->error_mutex); | |
1244 | return qatomic_read(&s->error); | |
1245 | } | |
1246 | ||
ca7bd082 PX |
1247 | static void migrate_error_free(MigrationState *s) |
1248 | { | |
1249 | QEMU_LOCK_GUARD(&s->error_mutex); | |
1250 | if (s->error) { | |
1251 | error_free(s->error); | |
1252 | s->error = NULL; | |
1253 | } | |
1254 | } | |
1255 | ||
aaf26bd3 | 1256 | static void migrate_fd_error(MigrationState *s, const Error *error) |
065e2813 | 1257 | { |
25174055 | 1258 | trace_migrate_fd_error(error_get_pretty(error)); |
89a02a9f | 1259 | assert(s->to_dst_file == NULL); |
48781e5b HZ |
1260 | migrate_set_state(&s->state, MIGRATION_STATUS_SETUP, |
1261 | MIGRATION_STATUS_FAILED); | |
87db1a7d | 1262 | migrate_set_error(s, error); |
458cf28e JQ |
1263 | } |
1264 | ||
0edda1c4 | 1265 | static void migrate_fd_cancel(MigrationState *s) |
065e2813 | 1266 | { |
6f2b811a | 1267 | int old_state ; |
7478fb0d | 1268 | |
9013dca5 | 1269 | trace_migrate_fd_cancel(); |
065e2813 | 1270 | |
43044ac0 PX |
1271 | WITH_QEMU_LOCK_GUARD(&s->qemu_file_lock) { |
1272 | if (s->rp_state.from_dst_file) { | |
1273 | /* shutdown the rp socket, so causing the rp thread to shutdown */ | |
1274 | qemu_file_shutdown(s->rp_state.from_dst_file); | |
1275 | } | |
70b20477 DDAG |
1276 | } |
1277 | ||
6f2b811a Z |
1278 | do { |
1279 | old_state = s->state; | |
392d87e2 | 1280 | if (!migration_is_running(old_state)) { |
6f2b811a Z |
1281 | break; |
1282 | } | |
a7b36b48 DDAG |
1283 | /* If the migration is paused, kick it out of the pause */ |
1284 | if (old_state == MIGRATION_STATUS_PRE_SWITCHOVER) { | |
1285 | qemu_sem_post(&s->pause_sem); | |
1286 | } | |
48781e5b | 1287 | migrate_set_state(&s->state, old_state, MIGRATION_STATUS_CANCELLING); |
31194731 | 1288 | } while (s->state != MIGRATION_STATUS_CANCELLING); |
a26ba26e DDAG |
1289 | |
1290 | /* | |
1291 | * If we're unlucky the migration code might be stuck somewhere in a | |
1292 | * send/write while the network has failed and is waiting to timeout; | |
1293 | * if we've got shutdown(2) available then we can force it to quit. | |
a26ba26e | 1294 | */ |
7478fb0d FR |
1295 | if (s->state == MIGRATION_STATUS_CANCELLING) { |
1296 | WITH_QEMU_LOCK_GUARD(&s->qemu_file_lock) { | |
1297 | if (s->to_dst_file) { | |
1298 | qemu_file_shutdown(s->to_dst_file); | |
1299 | } | |
1300 | } | |
a26ba26e | 1301 | } |
1d2acc31 HZ |
1302 | if (s->state == MIGRATION_STATUS_CANCELLING && s->block_inactive) { |
1303 | Error *local_err = NULL; | |
1304 | ||
3b717194 | 1305 | bdrv_activate_all(&local_err); |
1d2acc31 HZ |
1306 | if (local_err) { |
1307 | error_report_err(local_err); | |
1308 | } else { | |
1309 | s->block_inactive = false; | |
1310 | } | |
1311 | } | |
065e2813 AL |
1312 | } |
1313 | ||
d9cda213 SS |
1314 | void migration_add_notifier(Notifier *notify, |
1315 | void (*func)(Notifier *notifier, void *data)) | |
99a0db9b | 1316 | { |
d9cda213 | 1317 | notify->notify = func; |
99a0db9b GH |
1318 | notifier_list_add(&migration_state_notifiers, notify); |
1319 | } | |
1320 | ||
d9cda213 SS |
1321 | void migration_remove_notifier(Notifier *notify) |
1322 | { | |
1323 | if (notify->notify) { | |
1324 | notifier_remove(notify); | |
1325 | notify->notify = NULL; | |
1326 | } | |
1327 | } | |
1328 | ||
1329 | void migration_call_notifiers(MigrationState *s) | |
99a0db9b | 1330 | { |
d9cda213 | 1331 | notifier_list_notify(&migration_state_notifiers, s); |
99a0db9b GH |
1332 | } |
1333 | ||
02edd2e7 | 1334 | bool migration_in_setup(MigrationState *s) |
afe2df69 | 1335 | { |
31194731 | 1336 | return s->state == MIGRATION_STATUS_SETUP; |
afe2df69 GH |
1337 | } |
1338 | ||
7073693b | 1339 | bool migration_has_finished(MigrationState *s) |
99a0db9b | 1340 | { |
31194731 | 1341 | return s->state == MIGRATION_STATUS_COMPLETED; |
99a0db9b | 1342 | } |
0edda1c4 | 1343 | |
afe2df69 GH |
1344 | bool migration_has_failed(MigrationState *s) |
1345 | { | |
31194731 HZ |
1346 | return (s->state == MIGRATION_STATUS_CANCELLED || |
1347 | s->state == MIGRATION_STATUS_FAILED); | |
afe2df69 GH |
1348 | } |
1349 | ||
5727309d | 1350 | bool migration_in_postcopy(void) |
9ec055ae | 1351 | { |
5727309d JQ |
1352 | MigrationState *s = migrate_get_current(); |
1353 | ||
3748fef9 DDAG |
1354 | switch (s->state) { |
1355 | case MIGRATION_STATUS_POSTCOPY_ACTIVE: | |
1356 | case MIGRATION_STATUS_POSTCOPY_PAUSED: | |
1357 | case MIGRATION_STATUS_POSTCOPY_RECOVER: | |
1358 | return true; | |
1359 | default: | |
1360 | return false; | |
1361 | } | |
9ec055ae DDAG |
1362 | } |
1363 | ||
b82fc321 DDAG |
1364 | bool migration_in_postcopy_after_devices(MigrationState *s) |
1365 | { | |
5727309d | 1366 | return migration_in_postcopy() && s->postcopy_after_devices; |
b82fc321 DDAG |
1367 | } |
1368 | ||
06df2e69 DH |
1369 | bool migration_in_incoming_postcopy(void) |
1370 | { | |
1371 | PostcopyState ps = postcopy_state_get(); | |
1372 | ||
1373 | return ps >= POSTCOPY_INCOMING_DISCARD && ps < POSTCOPY_INCOMING_END; | |
1374 | } | |
1375 | ||
80fe315c DH |
1376 | bool migration_incoming_postcopy_advised(void) |
1377 | { | |
1378 | PostcopyState ps = postcopy_state_get(); | |
1379 | ||
1380 | return ps >= POSTCOPY_INCOMING_ADVISE && ps < POSTCOPY_INCOMING_END; | |
1381 | } | |
1382 | ||
1a8e44a8 AG |
1383 | bool migration_in_bg_snapshot(void) |
1384 | { | |
1385 | MigrationState *s = migrate_get_current(); | |
1386 | ||
1387 | return migrate_background_snapshot() && | |
1388 | migration_is_setup_or_active(s->state); | |
1389 | } | |
1390 | ||
fab35005 | 1391 | bool migration_is_idle(void) |
fe44dc91 | 1392 | { |
daff7f0b MA |
1393 | MigrationState *s = current_migration; |
1394 | ||
1395 | if (!s) { | |
1396 | return true; | |
1397 | } | |
fe44dc91 AA |
1398 | |
1399 | switch (s->state) { | |
1400 | case MIGRATION_STATUS_NONE: | |
1401 | case MIGRATION_STATUS_CANCELLED: | |
1402 | case MIGRATION_STATUS_COMPLETED: | |
1403 | case MIGRATION_STATUS_FAILED: | |
1404 | return true; | |
1405 | case MIGRATION_STATUS_SETUP: | |
1406 | case MIGRATION_STATUS_CANCELLING: | |
1407 | case MIGRATION_STATUS_ACTIVE: | |
1408 | case MIGRATION_STATUS_POSTCOPY_ACTIVE: | |
1409 | case MIGRATION_STATUS_COLO: | |
31e06077 DDAG |
1410 | case MIGRATION_STATUS_PRE_SWITCHOVER: |
1411 | case MIGRATION_STATUS_DEVICE: | |
c7e0acd5 | 1412 | case MIGRATION_STATUS_WAIT_UNPLUG: |
fe44dc91 AA |
1413 | return false; |
1414 | case MIGRATION_STATUS__MAX: | |
1415 | g_assert_not_reached(); | |
1416 | } | |
1417 | ||
1418 | return false; | |
1419 | } | |
1420 | ||
8f8d528e WY |
1421 | bool migration_is_active(MigrationState *s) |
1422 | { | |
1423 | return (s->state == MIGRATION_STATUS_ACTIVE || | |
1424 | s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE); | |
1425 | } | |
1426 | ||
08fc4cb5 | 1427 | int migrate_init(MigrationState *s, Error **errp) |
0edda1c4 | 1428 | { |
08fc4cb5 AH |
1429 | int ret; |
1430 | ||
1431 | ret = qemu_savevm_state_prepare(errp); | |
1432 | if (ret) { | |
1433 | return ret; | |
1434 | } | |
1435 | ||
389775d1 DDAG |
1436 | /* |
1437 | * Reinitialise all migration state, except | |
1438 | * parameters/capabilities that the user set, and | |
1439 | * locks. | |
1440 | */ | |
389775d1 | 1441 | s->cleanup_bh = 0; |
8518278a | 1442 | s->vm_start_bh = 0; |
89a02a9f | 1443 | s->to_dst_file = NULL; |
389775d1 | 1444 | s->state = MIGRATION_STATUS_NONE; |
389775d1 DDAG |
1445 | s->rp_state.from_dst_file = NULL; |
1446 | s->rp_state.error = false; | |
1447 | s->mbps = 0.0; | |
aecbfe9c | 1448 | s->pages_per_second = 0.0; |
389775d1 DDAG |
1449 | s->downtime = 0; |
1450 | s->expected_downtime = 0; | |
389775d1 | 1451 | s->setup_time = 0; |
389775d1 | 1452 | s->start_postcopy = false; |
b82fc321 | 1453 | s->postcopy_after_devices = false; |
389775d1 | 1454 | s->migration_thread_running = false; |
d59ce6f3 DB |
1455 | error_free(s->error); |
1456 | s->error = NULL; | |
d8053e73 | 1457 | s->hostname = NULL; |
2aae1eb8 | 1458 | s->vmdesc = NULL; |
389775d1 | 1459 | |
48781e5b | 1460 | migrate_set_state(&s->state, MIGRATION_STATUS_NONE, MIGRATION_STATUS_SETUP); |
0edda1c4 | 1461 | |
4af246a3 PX |
1462 | s->start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); |
1463 | s->total_time = 0; | |
f4584076 | 1464 | s->vm_old_state = -1; |
b15df1ae PX |
1465 | s->iteration_initial_bytes = 0; |
1466 | s->threshold_size = 0; | |
1b4adb10 | 1467 | s->switchover_acked = false; |
27fd25b0 | 1468 | s->rdma_migration = false; |
f543aa22 | 1469 | /* |
809f188a | 1470 | * set mig_stats memory to zero for a new migration |
f543aa22 AH |
1471 | */ |
1472 | memset(&mig_stats, 0, sizeof(mig_stats)); | |
f543aa22 | 1473 | migration_reset_vfio_bytes_transferred(); |
08fc4cb5 AH |
1474 | |
1475 | return 0; | |
0edda1c4 | 1476 | } |
cab30143 | 1477 | |
c8a7fc51 | 1478 | int migrate_add_blocker_internal(Error **reasonp, Error **errp) |
fa2756b7 | 1479 | { |
4c170330 PX |
1480 | /* Snapshots are similar to migrations, so check RUN_STATE_SAVE_VM too. */ |
1481 | if (runstate_check(RUN_STATE_SAVE_VM) || !migration_is_idle()) { | |
c8a7fc51 | 1482 | error_propagate_prepend(errp, *reasonp, |
4c170330 PX |
1483 | "disallowing migration blocker " |
1484 | "(migration/snapshot in progress) for: "); | |
c8a7fc51 | 1485 | *reasonp = NULL; |
4c170330 | 1486 | return -EBUSY; |
fe44dc91 AA |
1487 | } |
1488 | ||
c8a7fc51 | 1489 | migration_blockers = g_slist_prepend(migration_blockers, *reasonp); |
4c170330 | 1490 | return 0; |
fa2756b7 AL |
1491 | } |
1492 | ||
c8a7fc51 | 1493 | int migrate_add_blocker(Error **reasonp, Error **errp) |
60fd6801 PX |
1494 | { |
1495 | if (only_migratable) { | |
c8a7fc51 | 1496 | error_propagate_prepend(errp, *reasonp, |
60fd6801 PX |
1497 | "disallowing migration blocker " |
1498 | "(--only-migratable) for: "); | |
c8a7fc51 | 1499 | *reasonp = NULL; |
60fd6801 PX |
1500 | return -EACCES; |
1501 | } | |
1502 | ||
c8a7fc51 | 1503 | return migrate_add_blocker_internal(reasonp, errp); |
60fd6801 PX |
1504 | } |
1505 | ||
c8a7fc51 | 1506 | void migrate_del_blocker(Error **reasonp) |
fa2756b7 | 1507 | { |
c8a7fc51 SS |
1508 | if (*reasonp) { |
1509 | migration_blockers = g_slist_remove(migration_blockers, *reasonp); | |
1510 | error_free(*reasonp); | |
1511 | *reasonp = NULL; | |
1512 | } | |
fa2756b7 AL |
1513 | } |
1514 | ||
bf1ae1f4 DDAG |
1515 | void qmp_migrate_incoming(const char *uri, Error **errp) |
1516 | { | |
1517 | Error *local_err = NULL; | |
4debb5f5 | 1518 | static bool once = true; |
bf1ae1f4 | 1519 | |
4debb5f5 DDAG |
1520 | if (!once) { |
1521 | error_setg(errp, "The incoming migration has already been started"); | |
603d5a42 | 1522 | return; |
4debb5f5 | 1523 | } |
e69d50d6 PB |
1524 | if (!runstate_check(RUN_STATE_INMIGRATE)) { |
1525 | error_setg(errp, "'-incoming' was not specified on the command line"); | |
1526 | return; | |
1527 | } | |
bf1ae1f4 | 1528 | |
cc48c587 PX |
1529 | if (!yank_register_instance(MIGRATION_YANK_INSTANCE, errp)) { |
1530 | return; | |
1531 | } | |
1532 | ||
bf1ae1f4 DDAG |
1533 | qemu_start_incoming_migration(uri, &local_err); |
1534 | ||
1535 | if (local_err) { | |
cc48c587 | 1536 | yank_unregister_instance(MIGRATION_YANK_INSTANCE); |
bf1ae1f4 DDAG |
1537 | error_propagate(errp, local_err); |
1538 | return; | |
1539 | } | |
1540 | ||
4debb5f5 | 1541 | once = false; |
bf1ae1f4 DDAG |
1542 | } |
1543 | ||
02affd41 PX |
1544 | void qmp_migrate_recover(const char *uri, Error **errp) |
1545 | { | |
1546 | MigrationIncomingState *mis = migration_incoming_get_current(); | |
1547 | ||
b7f9afd4 PX |
1548 | /* |
1549 | * Don't even bother to use ERRP_GUARD() as it _must_ always be set by | |
1550 | * callers (no one should ignore a recover failure); if there is, it's a | |
1551 | * programming error. | |
1552 | */ | |
1553 | assert(errp); | |
1554 | ||
02affd41 PX |
1555 | if (mis->state != MIGRATION_STATUS_POSTCOPY_PAUSED) { |
1556 | error_setg(errp, "Migrate recover can only be run " | |
1557 | "when postcopy is paused."); | |
1558 | return; | |
1559 | } | |
1560 | ||
08401c04 PX |
1561 | /* If there's an existing transport, release it */ |
1562 | migration_incoming_transport_cleanup(mis); | |
02affd41 PX |
1563 | |
1564 | /* | |
1565 | * Note that this call will never start a real migration; it will | |
1566 | * only re-setup the migration stream and poke existing migration | |
1567 | * to continue using that newly established channel. | |
1568 | */ | |
1569 | qemu_start_incoming_migration(uri, errp); | |
1570 | } | |
1571 | ||
bfbf89c2 PX |
1572 | void qmp_migrate_pause(Error **errp) |
1573 | { | |
1574 | MigrationState *ms = migrate_get_current(); | |
1575 | MigrationIncomingState *mis = migration_incoming_get_current(); | |
7478fb0d | 1576 | int ret = 0; |
bfbf89c2 PX |
1577 | |
1578 | if (ms->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) { | |
1579 | /* Source side, during postcopy */ | |
1580 | qemu_mutex_lock(&ms->qemu_file_lock); | |
7478fb0d FR |
1581 | if (ms->to_dst_file) { |
1582 | ret = qemu_file_shutdown(ms->to_dst_file); | |
1583 | } | |
bfbf89c2 PX |
1584 | qemu_mutex_unlock(&ms->qemu_file_lock); |
1585 | if (ret) { | |
1586 | error_setg(errp, "Failed to pause source migration"); | |
1587 | } | |
1588 | return; | |
1589 | } | |
1590 | ||
1591 | if (mis->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) { | |
1592 | ret = qemu_file_shutdown(mis->from_src_file); | |
1593 | if (ret) { | |
1594 | error_setg(errp, "Failed to pause destination migration"); | |
1595 | } | |
1596 | return; | |
1597 | } | |
1598 | ||
1599 | error_setg(errp, "migrate-pause is currently only supported " | |
1600 | "during postcopy-active state"); | |
1601 | } | |
1602 | ||
24f3902b GK |
1603 | bool migration_is_blocked(Error **errp) |
1604 | { | |
1605 | if (qemu_savevm_state_blocked(errp)) { | |
1606 | return true; | |
1607 | } | |
1608 | ||
1609 | if (migration_blockers) { | |
250561e1 | 1610 | error_propagate(errp, error_copy(migration_blockers->data)); |
24f3902b GK |
1611 | return true; |
1612 | } | |
1613 | ||
1614 | return false; | |
1615 | } | |
1616 | ||
d3e35b8f PX |
1617 | /* Returns true if continue to migrate, or false if error detected */ |
1618 | static bool migrate_prepare(MigrationState *s, bool blk, bool blk_inc, | |
1619 | bool resume, Error **errp) | |
cab30143 | 1620 | { |
be7059cd | 1621 | Error *local_err = NULL; |
d3e35b8f PX |
1622 | |
1623 | if (resume) { | |
1624 | if (s->state != MIGRATION_STATUS_POSTCOPY_PAUSED) { | |
1625 | error_setg(errp, "Cannot resume if there is no " | |
1626 | "paused migration"); | |
1627 | return false; | |
1628 | } | |
97ca211c PX |
1629 | |
1630 | /* | |
1631 | * Postcopy recovery won't work well with release-ram | |
1632 | * capability since release-ram will drop the page buffer as | |
1633 | * long as the page is put into the send buffer. So if there | |
1634 | * is a network failure happened, any page buffers that have | |
1635 | * not yet reached the destination VM but have already been | |
1636 | * sent from the source VM will be lost forever. Let's refuse | |
1637 | * the client from resuming such a postcopy migration. | |
1638 | * Luckily release-ram was designed to only be used when src | |
1639 | * and destination VMs are on the same host, so it should be | |
1640 | * fine. | |
1641 | */ | |
1642 | if (migrate_release_ram()) { | |
1643 | error_setg(errp, "Postcopy recovery cannot work " | |
1644 | "when release-ram capability is set"); | |
1645 | return false; | |
1646 | } | |
1647 | ||
d3e35b8f PX |
1648 | /* This is a resume, skip init status */ |
1649 | return true; | |
1650 | } | |
cab30143 | 1651 | |
392d87e2 | 1652 | if (migration_is_running(s->state)) { |
c6bd8c70 | 1653 | error_setg(errp, QERR_MIGRATION_ACTIVE); |
d3e35b8f | 1654 | return false; |
cab30143 | 1655 | } |
d3e35b8f | 1656 | |
ca99993a DDAG |
1657 | if (runstate_check(RUN_STATE_INMIGRATE)) { |
1658 | error_setg(errp, "Guest is waiting for an incoming migration"); | |
d3e35b8f | 1659 | return false; |
ca99993a DDAG |
1660 | } |
1661 | ||
36d0fe65 T |
1662 | if (runstate_check(RUN_STATE_POSTMIGRATE)) { |
1663 | error_setg(errp, "Can't migrate the vm that was paused due to " | |
1664 | "previous migration"); | |
1665 | return false; | |
1666 | } | |
1667 | ||
24f3902b | 1668 | if (migration_is_blocked(errp)) { |
d3e35b8f | 1669 | return false; |
fa2756b7 AL |
1670 | } |
1671 | ||
d3e35b8f | 1672 | if (blk || blk_inc) { |
5e804644 | 1673 | if (migrate_colo()) { |
3ba02445 RL |
1674 | error_setg(errp, "No disk migration is required in COLO mode"); |
1675 | return false; | |
1676 | } | |
6f8be708 | 1677 | if (migrate_block() || migrate_block_incremental()) { |
2833c59b JQ |
1678 | error_setg(errp, "Command options are incompatible with " |
1679 | "current migration capabilities"); | |
d3e35b8f | 1680 | return false; |
2833c59b | 1681 | } |
9eb1109c | 1682 | if (!migrate_cap_set(MIGRATION_CAPABILITY_BLOCK, true, &local_err)) { |
2833c59b | 1683 | error_propagate(errp, local_err); |
d3e35b8f | 1684 | return false; |
2833c59b JQ |
1685 | } |
1686 | s->must_remove_block_options = true; | |
1687 | } | |
1688 | ||
d3e35b8f | 1689 | if (blk_inc) { |
87c22901 | 1690 | migrate_set_block_incremental(true); |
2833c59b JQ |
1691 | } |
1692 | ||
08fc4cb5 AH |
1693 | if (migrate_init(s, errp)) { |
1694 | return false; | |
1695 | } | |
cab30143 | 1696 | |
d3e35b8f PX |
1697 | return true; |
1698 | } | |
1699 | ||
1700 | void qmp_migrate(const char *uri, bool has_blk, bool blk, | |
1701 | bool has_inc, bool inc, bool has_detach, bool detach, | |
1702 | bool has_resume, bool resume, Error **errp) | |
1703 | { | |
8c69ae9e | 1704 | bool resume_requested; |
d3e35b8f PX |
1705 | Error *local_err = NULL; |
1706 | MigrationState *s = migrate_get_current(); | |
d658f65c | 1707 | const char *p = NULL; |
d3e35b8f | 1708 | |
d6f74fd1 PX |
1709 | /* URI is not suitable for migration? */ |
1710 | if (!migration_channels_and_uri_compatible(uri, errp)) { | |
1711 | return; | |
1712 | } | |
1713 | ||
8c69ae9e | 1714 | resume_requested = has_resume && resume; |
d3e35b8f | 1715 | if (!migrate_prepare(s, has_blk && blk, has_inc && inc, |
8c69ae9e | 1716 | resume_requested, errp)) { |
d3e35b8f PX |
1717 | /* Error detected, put into errp */ |
1718 | return; | |
1719 | } | |
1720 | ||
8c69ae9e | 1721 | if (!resume_requested) { |
b5eea99e LS |
1722 | if (!yank_register_instance(MIGRATION_YANK_INSTANCE, errp)) { |
1723 | return; | |
1724 | } | |
1725 | } | |
1726 | ||
d658f65c | 1727 | if (strstart(uri, "tcp:", &p) || |
9ba3b2ba LM |
1728 | strstart(uri, "unix:", NULL) || |
1729 | strstart(uri, "vsock:", NULL)) { | |
d658f65c | 1730 | socket_start_outgoing_migration(s, p ? p : uri, &local_err); |
2da776db | 1731 | #ifdef CONFIG_RDMA |
41310c68 | 1732 | } else if (strstart(uri, "rdma:", &p)) { |
2da776db MH |
1733 | rdma_start_outgoing_migration(s, p, &local_err); |
1734 | #endif | |
cab30143 | 1735 | } else if (strstart(uri, "exec:", &p)) { |
f37afb5a | 1736 | exec_start_outgoing_migration(s, p, &local_err); |
cab30143 | 1737 | } else if (strstart(uri, "fd:", &p)) { |
f37afb5a | 1738 | fd_start_outgoing_migration(s, p, &local_err); |
2a9e2e59 SS |
1739 | } else if (strstart(uri, "file:", &p)) { |
1740 | file_start_outgoing_migration(s, p, &local_err); | |
99a0db9b | 1741 | } else { |
908927db | 1742 | error_setg(&local_err, QERR_INVALID_PARAMETER_VALUE, "uri", |
c6bd8c70 | 1743 | "a valid migration protocol"); |
48781e5b HZ |
1744 | migrate_set_state(&s->state, MIGRATION_STATUS_SETUP, |
1745 | MIGRATION_STATUS_FAILED); | |
b1a87956 | 1746 | block_cleanup_parameters(); |
cab30143 JQ |
1747 | } |
1748 | ||
f37afb5a | 1749 | if (local_err) { |
8c69ae9e | 1750 | if (!resume_requested) { |
b5eea99e LS |
1751 | yank_unregister_instance(MIGRATION_YANK_INSTANCE); |
1752 | } | |
d59ce6f3 | 1753 | migrate_fd_error(s, local_err); |
f37afb5a | 1754 | error_propagate(errp, local_err); |
e1c37d0e | 1755 | return; |
1299c631 | 1756 | } |
cab30143 JQ |
1757 | } |
1758 | ||
6cdedb07 | 1759 | void qmp_migrate_cancel(Error **errp) |
cab30143 | 1760 | { |
458fecca | 1761 | migration_cancel(NULL); |
cab30143 JQ |
1762 | } |
1763 | ||
89cfc02c DDAG |
1764 | void qmp_migrate_continue(MigrationStatus state, Error **errp) |
1765 | { | |
1766 | MigrationState *s = migrate_get_current(); | |
1767 | if (s->state != state) { | |
1768 | error_setg(errp, "Migration not in expected state: %s", | |
1769 | MigrationStatus_str(s->state)); | |
1770 | return; | |
1771 | } | |
1772 | qemu_sem_post(&s->pause_sem); | |
1773 | } | |
1774 | ||
70b20477 DDAG |
1775 | /* migration thread support */ |
1776 | /* | |
1777 | * Something bad happened to the RP stream, mark an error | |
1778 | * The caller shall print or trace something to indicate why | |
1779 | */ | |
1780 | static void mark_source_rp_bad(MigrationState *s) | |
1781 | { | |
1782 | s->rp_state.error = true; | |
1783 | } | |
1784 | ||
5e79a4bf PX |
1785 | void migration_rp_wait(MigrationState *s) |
1786 | { | |
1787 | qemu_sem_wait(&s->rp_state.rp_sem); | |
1788 | } | |
1789 | ||
1790 | void migration_rp_kick(MigrationState *s) | |
1791 | { | |
1792 | qemu_sem_post(&s->rp_state.rp_sem); | |
1793 | } | |
1794 | ||
70b20477 DDAG |
1795 | static struct rp_cmd_args { |
1796 | ssize_t len; /* -1 = variable */ | |
1797 | const char *name; | |
1798 | } rp_cmd_args[] = { | |
1799 | [MIG_RP_MSG_INVALID] = { .len = -1, .name = "INVALID" }, | |
1800 | [MIG_RP_MSG_SHUT] = { .len = 4, .name = "SHUT" }, | |
1801 | [MIG_RP_MSG_PONG] = { .len = 4, .name = "PONG" }, | |
1e2d90eb DDAG |
1802 | [MIG_RP_MSG_REQ_PAGES] = { .len = 12, .name = "REQ_PAGES" }, |
1803 | [MIG_RP_MSG_REQ_PAGES_ID] = { .len = -1, .name = "REQ_PAGES_ID" }, | |
a335debb | 1804 | [MIG_RP_MSG_RECV_BITMAP] = { .len = -1, .name = "RECV_BITMAP" }, |
13955b89 | 1805 | [MIG_RP_MSG_RESUME_ACK] = { .len = 4, .name = "RESUME_ACK" }, |
1b4adb10 | 1806 | [MIG_RP_MSG_SWITCHOVER_ACK] = { .len = 0, .name = "SWITCHOVER_ACK" }, |
70b20477 DDAG |
1807 | [MIG_RP_MSG_MAX] = { .len = -1, .name = "MAX" }, |
1808 | }; | |
1809 | ||
1e2d90eb DDAG |
1810 | /* |
1811 | * Process a request for pages received on the return path, | |
1812 | * We're allowed to send more than requested (e.g. to round to our page size) | |
1813 | * and we don't need to send pages that have already been sent. | |
1814 | */ | |
1815 | static void migrate_handle_rp_req_pages(MigrationState *ms, const char* rbname, | |
1816 | ram_addr_t start, size_t len) | |
1817 | { | |
8e3b0cbb | 1818 | long our_host_ps = qemu_real_host_page_size(); |
6c595cde | 1819 | |
1e2d90eb | 1820 | trace_migrate_handle_rp_req_pages(rbname, start, len); |
6c595cde DDAG |
1821 | |
1822 | /* | |
1823 | * Since we currently insist on matching page sizes, just sanity check | |
1824 | * we're being asked for whole host pages. | |
1825 | */ | |
7648297d DH |
1826 | if (!QEMU_IS_ALIGNED(start, our_host_ps) || |
1827 | !QEMU_IS_ALIGNED(len, our_host_ps)) { | |
6c595cde DDAG |
1828 | error_report("%s: Misaligned page request, start: " RAM_ADDR_FMT |
1829 | " len: %zd", __func__, start, len); | |
1830 | mark_source_rp_bad(ms); | |
1831 | return; | |
1832 | } | |
1833 | ||
96506894 | 1834 | if (ram_save_queue_pages(rbname, start, len)) { |
6c595cde DDAG |
1835 | mark_source_rp_bad(ms); |
1836 | } | |
1e2d90eb DDAG |
1837 | } |
1838 | ||
a335debb PX |
1839 | static int migrate_handle_rp_recv_bitmap(MigrationState *s, char *block_name) |
1840 | { | |
1841 | RAMBlock *block = qemu_ram_block_by_name(block_name); | |
1842 | ||
1843 | if (!block) { | |
1844 | error_report("%s: invalid block name '%s'", __func__, block_name); | |
1845 | return -EINVAL; | |
1846 | } | |
1847 | ||
1848 | /* Fetch the received bitmap and refresh the dirty bitmap */ | |
1849 | return ram_dirty_bitmap_reload(s, block); | |
1850 | } | |
1851 | ||
13955b89 PX |
1852 | static int migrate_handle_rp_resume_ack(MigrationState *s, uint32_t value) |
1853 | { | |
1854 | trace_source_return_path_thread_resume_ack(value); | |
1855 | ||
1856 | if (value != MIGRATION_RESUME_ACK_VALUE) { | |
1857 | error_report("%s: illegal resume_ack value %"PRIu32, | |
1858 | __func__, value); | |
1859 | return -1; | |
1860 | } | |
1861 | ||
1862 | /* Now both sides are active. */ | |
1863 | migrate_set_state(&s->state, MIGRATION_STATUS_POSTCOPY_RECOVER, | |
1864 | MIGRATION_STATUS_POSTCOPY_ACTIVE); | |
1865 | ||
94190696 | 1866 | /* Notify send thread that time to continue send pages */ |
5e79a4bf | 1867 | migration_rp_kick(s); |
13955b89 PX |
1868 | |
1869 | return 0; | |
1870 | } | |
1871 | ||
93589827 PX |
1872 | /* |
1873 | * Release ms->rp_state.from_dst_file (and postcopy_qemufile_src if | |
1874 | * existed) in a safe way. | |
1875 | */ | |
1876 | static void migration_release_dst_files(MigrationState *ms) | |
43044ac0 PX |
1877 | { |
1878 | QEMUFile *file; | |
1879 | ||
1880 | WITH_QEMU_LOCK_GUARD(&ms->qemu_file_lock) { | |
1881 | /* | |
1882 | * Reset the from_dst_file pointer first before releasing it, as we | |
1883 | * can't block within lock section | |
1884 | */ | |
1885 | file = ms->rp_state.from_dst_file; | |
1886 | ms->rp_state.from_dst_file = NULL; | |
1887 | } | |
1888 | ||
93589827 PX |
1889 | /* |
1890 | * Do the same to postcopy fast path socket too if there is. No | |
1891 | * locking needed because this qemufile should only be managed by | |
1892 | * return path thread. | |
1893 | */ | |
1894 | if (ms->postcopy_qemufile_src) { | |
1895 | migration_ioc_unregister_yank_from_file(ms->postcopy_qemufile_src); | |
1896 | qemu_file_shutdown(ms->postcopy_qemufile_src); | |
1897 | qemu_fclose(ms->postcopy_qemufile_src); | |
1898 | ms->postcopy_qemufile_src = NULL; | |
1899 | } | |
1900 | ||
43044ac0 PX |
1901 | qemu_fclose(file); |
1902 | } | |
1903 | ||
70b20477 DDAG |
1904 | /* |
1905 | * Handles messages sent on the return path towards the source VM | |
1906 | * | |
1907 | */ | |
1908 | static void *source_return_path_thread(void *opaque) | |
1909 | { | |
1910 | MigrationState *ms = opaque; | |
1911 | QEMUFile *rp = ms->rp_state.from_dst_file; | |
1912 | uint16_t header_len, header_type; | |
568b01ca | 1913 | uint8_t buf[512]; |
70b20477 | 1914 | uint32_t tmp32, sibling_error; |
1e2d90eb DDAG |
1915 | ram_addr_t start = 0; /* =0 to silence warning */ |
1916 | size_t len = 0, expected_len; | |
70b20477 DDAG |
1917 | int res; |
1918 | ||
1919 | trace_source_return_path_thread_entry(); | |
74637e6f | 1920 | rcu_register_thread(); |
14b1742e | 1921 | |
70b20477 DDAG |
1922 | while (!ms->rp_state.error && !qemu_file_get_error(rp) && |
1923 | migration_is_setup_or_active(ms->state)) { | |
1924 | trace_source_return_path_thread_loop_top(); | |
1925 | header_type = qemu_get_be16(rp); | |
1926 | header_len = qemu_get_be16(rp); | |
1927 | ||
7a9ddfbf PX |
1928 | if (qemu_file_get_error(rp)) { |
1929 | mark_source_rp_bad(ms); | |
1930 | goto out; | |
1931 | } | |
1932 | ||
70b20477 DDAG |
1933 | if (header_type >= MIG_RP_MSG_MAX || |
1934 | header_type == MIG_RP_MSG_INVALID) { | |
1935 | error_report("RP: Received invalid message 0x%04x length 0x%04x", | |
ed8b2828 | 1936 | header_type, header_len); |
70b20477 DDAG |
1937 | mark_source_rp_bad(ms); |
1938 | goto out; | |
1939 | } | |
1940 | ||
1941 | if ((rp_cmd_args[header_type].len != -1 && | |
1942 | header_len != rp_cmd_args[header_type].len) || | |
568b01ca | 1943 | header_len > sizeof(buf)) { |
70b20477 | 1944 | error_report("RP: Received '%s' message (0x%04x) with" |
ed8b2828 MZ |
1945 | "incorrect length %d expecting %zu", |
1946 | rp_cmd_args[header_type].name, header_type, header_len, | |
1947 | (size_t)rp_cmd_args[header_type].len); | |
70b20477 DDAG |
1948 | mark_source_rp_bad(ms); |
1949 | goto out; | |
1950 | } | |
1951 | ||
1952 | /* We know we've got a valid header by this point */ | |
1953 | res = qemu_get_buffer(rp, buf, header_len); | |
1954 | if (res != header_len) { | |
1955 | error_report("RP: Failed reading data for message 0x%04x" | |
1956 | " read %d expected %d", | |
1957 | header_type, res, header_len); | |
1958 | mark_source_rp_bad(ms); | |
1959 | goto out; | |
1960 | } | |
1961 | ||
1962 | /* OK, we have the message and the data */ | |
1963 | switch (header_type) { | |
1964 | case MIG_RP_MSG_SHUT: | |
4d885131 | 1965 | sibling_error = ldl_be_p(buf); |
70b20477 DDAG |
1966 | trace_source_return_path_thread_shut(sibling_error); |
1967 | if (sibling_error) { | |
1968 | error_report("RP: Sibling indicated error %d", sibling_error); | |
1969 | mark_source_rp_bad(ms); | |
1970 | } | |
1971 | /* | |
1972 | * We'll let the main thread deal with closing the RP | |
1973 | * we could do a shutdown(2) on it, but we're the only user | |
1974 | * anyway, so there's nothing gained. | |
1975 | */ | |
1976 | goto out; | |
1977 | ||
1978 | case MIG_RP_MSG_PONG: | |
4d885131 | 1979 | tmp32 = ldl_be_p(buf); |
70b20477 | 1980 | trace_source_return_path_thread_pong(tmp32); |
b28fb582 | 1981 | qemu_sem_post(&ms->rp_state.rp_pong_acks); |
70b20477 DDAG |
1982 | break; |
1983 | ||
1e2d90eb | 1984 | case MIG_RP_MSG_REQ_PAGES: |
4d885131 PM |
1985 | start = ldq_be_p(buf); |
1986 | len = ldl_be_p(buf + 8); | |
1e2d90eb DDAG |
1987 | migrate_handle_rp_req_pages(ms, NULL, start, len); |
1988 | break; | |
1989 | ||
1990 | case MIG_RP_MSG_REQ_PAGES_ID: | |
1991 | expected_len = 12 + 1; /* header + termination */ | |
1992 | ||
1993 | if (header_len >= expected_len) { | |
4d885131 PM |
1994 | start = ldq_be_p(buf); |
1995 | len = ldl_be_p(buf + 8); | |
1e2d90eb DDAG |
1996 | /* Now we expect an idstr */ |
1997 | tmp32 = buf[12]; /* Length of the following idstr */ | |
1998 | buf[13 + tmp32] = '\0'; | |
1999 | expected_len += tmp32; | |
2000 | } | |
2001 | if (header_len != expected_len) { | |
2002 | error_report("RP: Req_Page_id with length %d expecting %zd", | |
ed8b2828 | 2003 | header_len, expected_len); |
1e2d90eb DDAG |
2004 | mark_source_rp_bad(ms); |
2005 | goto out; | |
2006 | } | |
2007 | migrate_handle_rp_req_pages(ms, (char *)&buf[13], start, len); | |
2008 | break; | |
2009 | ||
a335debb PX |
2010 | case MIG_RP_MSG_RECV_BITMAP: |
2011 | if (header_len < 1) { | |
2012 | error_report("%s: missing block name", __func__); | |
2013 | mark_source_rp_bad(ms); | |
2014 | goto out; | |
2015 | } | |
2016 | /* Format: len (1B) + idstr (<255B). This ends the idstr. */ | |
2017 | buf[buf[0] + 1] = '\0'; | |
2018 | if (migrate_handle_rp_recv_bitmap(ms, (char *)(buf + 1))) { | |
2019 | mark_source_rp_bad(ms); | |
2020 | goto out; | |
2021 | } | |
2022 | break; | |
2023 | ||
13955b89 PX |
2024 | case MIG_RP_MSG_RESUME_ACK: |
2025 | tmp32 = ldl_be_p(buf); | |
2026 | if (migrate_handle_rp_resume_ack(ms, tmp32)) { | |
2027 | mark_source_rp_bad(ms); | |
2028 | goto out; | |
2029 | } | |
2030 | break; | |
2031 | ||
1b4adb10 AH |
2032 | case MIG_RP_MSG_SWITCHOVER_ACK: |
2033 | ms->switchover_acked = true; | |
2034 | trace_source_return_path_thread_switchover_acked(); | |
2035 | break; | |
2036 | ||
70b20477 DDAG |
2037 | default: |
2038 | break; | |
2039 | } | |
2040 | } | |
14b1742e PX |
2041 | |
2042 | out: | |
ef796ee9 | 2043 | if (qemu_file_get_error(rp)) { |
70b20477 DDAG |
2044 | trace_source_return_path_thread_bad_end(); |
2045 | mark_source_rp_bad(ms); | |
2046 | } | |
2047 | ||
2048 | trace_source_return_path_thread_end(); | |
74637e6f | 2049 | rcu_unregister_thread(); |
70b20477 DDAG |
2050 | return NULL; |
2051 | } | |
2052 | ||
ef796ee9 | 2053 | static int open_return_path_on_source(MigrationState *ms) |
70b20477 | 2054 | { |
89a02a9f | 2055 | ms->rp_state.from_dst_file = qemu_file_get_return_path(ms->to_dst_file); |
70b20477 DDAG |
2056 | if (!ms->rp_state.from_dst_file) { |
2057 | return -1; | |
2058 | } | |
2059 | ||
2060 | trace_open_return_path_on_source(); | |
d3e35b8f | 2061 | |
70b20477 DDAG |
2062 | qemu_thread_create(&ms->rp_state.rp_thread, "return path", |
2063 | source_return_path_thread, ms, QEMU_THREAD_JOINABLE); | |
53021ea1 | 2064 | ms->rp_state.rp_thread_created = true; |
70b20477 DDAG |
2065 | |
2066 | trace_open_return_path_on_source_continue(); | |
2067 | ||
2068 | return 0; | |
2069 | } | |
2070 | ||
d7f5a043 | 2071 | static int close_return_path_on_source(MigrationState *ms) |
70b20477 | 2072 | { |
d50f5dc0 FR |
2073 | int ret; |
2074 | ||
2075 | if (!ms->rp_state.rp_thread_created) { | |
2076 | return 0; | |
2077 | } | |
2078 | ||
2079 | trace_migration_return_path_end_before(); | |
2080 | ||
70b20477 | 2081 | /* |
639decf5 FR |
2082 | * If this is a normal exit then the destination will send a SHUT |
2083 | * and the rp_thread will exit, however if there's an error we | |
2084 | * need to cause it to exit. shutdown(2), if we have it, will | |
2085 | * cause it to unblock if it's stuck waiting for the destination. | |
70b20477 | 2086 | */ |
639decf5 FR |
2087 | WITH_QEMU_LOCK_GUARD(&ms->qemu_file_lock) { |
2088 | if (ms->to_dst_file && ms->rp_state.from_dst_file && | |
2089 | qemu_file_get_error(ms->to_dst_file)) { | |
2090 | qemu_file_shutdown(ms->rp_state.from_dst_file); | |
2091 | } | |
70b20477 | 2092 | } |
639decf5 | 2093 | |
70b20477 DDAG |
2094 | trace_await_return_path_close_on_source_joining(); |
2095 | qemu_thread_join(&ms->rp_state.rp_thread); | |
53021ea1 | 2096 | ms->rp_state.rp_thread_created = false; |
70b20477 | 2097 | trace_await_return_path_close_on_source_close(); |
d50f5dc0 FR |
2098 | |
2099 | ret = ms->rp_state.error; | |
ef796ee9 | 2100 | ms->rp_state.error = false; |
36e9aab3 FR |
2101 | |
2102 | migration_release_dst_files(ms); | |
2103 | ||
d50f5dc0 FR |
2104 | trace_migration_return_path_end_after(ret); |
2105 | return ret; | |
70b20477 DDAG |
2106 | } |
2107 | ||
5655aab0 PX |
2108 | static inline void |
2109 | migration_wait_main_channel(MigrationState *ms) | |
2110 | { | |
2111 | /* Wait until one PONG message received */ | |
2112 | qemu_sem_wait(&ms->rp_state.rp_pong_acks); | |
2113 | } | |
2114 | ||
1d34e4bf DDAG |
2115 | /* |
2116 | * Switch from normal iteration to postcopy | |
2117 | * Returns non-0 on error | |
2118 | */ | |
908927db | 2119 | static int postcopy_start(MigrationState *ms, Error **errp) |
1d34e4bf DDAG |
2120 | { |
2121 | int ret; | |
61b67d47 DB |
2122 | QIOChannelBuffer *bioc; |
2123 | QEMUFile *fb; | |
1d34e4bf | 2124 | int64_t time_at_stop = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); |
52033349 | 2125 | uint64_t bandwidth = migrate_max_postcopy_bandwidth(); |
ef8d6488 | 2126 | bool restart_block = false; |
0331c8ca | 2127 | int cur_state = MIGRATION_STATUS_ACTIVE; |
d0edb8a1 | 2128 | |
5655aab0 PX |
2129 | if (migrate_postcopy_preempt()) { |
2130 | migration_wait_main_channel(ms); | |
2131 | if (postcopy_preempt_establish_channel(ms)) { | |
2132 | migrate_set_state(&ms->state, ms->state, MIGRATION_STATUS_FAILED); | |
2133 | return -1; | |
2134 | } | |
d0edb8a1 PX |
2135 | } |
2136 | ||
0331c8ca DDAG |
2137 | if (!migrate_pause_before_switchover()) { |
2138 | migrate_set_state(&ms->state, MIGRATION_STATUS_ACTIVE, | |
2139 | MIGRATION_STATUS_POSTCOPY_ACTIVE); | |
2140 | } | |
1d34e4bf DDAG |
2141 | |
2142 | trace_postcopy_start(); | |
2143 | qemu_mutex_lock_iothread(); | |
2144 | trace_postcopy_start_set_run(); | |
2145 | ||
fb064112 | 2146 | qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER, NULL); |
1d34e4bf DDAG |
2147 | global_state_store(); |
2148 | ret = vm_stop_force_state(RUN_STATE_FINISH_MIGRATE); | |
76b1c7fe KW |
2149 | if (ret < 0) { |
2150 | goto fail; | |
2151 | } | |
1d34e4bf | 2152 | |
0331c8ca DDAG |
2153 | ret = migration_maybe_pause(ms, &cur_state, |
2154 | MIGRATION_STATUS_POSTCOPY_ACTIVE); | |
2155 | if (ret < 0) { | |
2156 | goto fail; | |
2157 | } | |
2158 | ||
76b1c7fe | 2159 | ret = bdrv_inactivate_all(); |
1d34e4bf DDAG |
2160 | if (ret < 0) { |
2161 | goto fail; | |
2162 | } | |
ef8d6488 | 2163 | restart_block = true; |
1d34e4bf | 2164 | |
1c0d249d DDAG |
2165 | /* |
2166 | * Cause any non-postcopiable, but iterative devices to | |
2167 | * send out their final data. | |
2168 | */ | |
a1fbe750 | 2169 | qemu_savevm_state_complete_precopy(ms->to_dst_file, true, false); |
1c0d249d | 2170 | |
1d34e4bf DDAG |
2171 | /* |
2172 | * in Finish migrate and with the io-lock held everything should | |
2173 | * be quiet, but we've potentially still got dirty pages and we | |
2174 | * need to tell the destination to throw any pages it's already received | |
2175 | * that are dirty | |
2176 | */ | |
58110f0a | 2177 | if (migrate_postcopy_ram()) { |
739fcc1b | 2178 | ram_postcopy_send_discard_bitmap(ms); |
1d34e4bf DDAG |
2179 | } |
2180 | ||
2181 | /* | |
2182 | * send rest of state - note things that are doing postcopy | |
2183 | * will notice we're in POSTCOPY_ACTIVE and not actually | |
2184 | * wrap their state up here | |
2185 | */ | |
e1fde0e0 | 2186 | migration_rate_set(bandwidth); |
58110f0a VSO |
2187 | if (migrate_postcopy_ram()) { |
2188 | /* Ping just for debugging, helps line traces up */ | |
2189 | qemu_savevm_send_ping(ms->to_dst_file, 2); | |
2190 | } | |
1d34e4bf DDAG |
2191 | |
2192 | /* | |
2193 | * While loading the device state we may trigger page transfer | |
2194 | * requests and the fd must be free to process those, and thus | |
2195 | * the destination must read the whole device state off the fd before | |
2196 | * it starts processing it. Unfortunately the ad-hoc migration format | |
2197 | * doesn't allow the destination to know the size to read without fully | |
2198 | * parsing it through each devices load-state code (especially the open | |
2199 | * coded devices that use get/put). | |
2200 | * So we wrap the device state up in a package with a length at the start; | |
2201 | * to do this we use a qemu_buf to hold the whole of the device state. | |
2202 | */ | |
61b67d47 | 2203 | bioc = qio_channel_buffer_new(4096); |
6f01f136 | 2204 | qio_channel_set_name(QIO_CHANNEL(bioc), "migration-postcopy-buffer"); |
77ef2dc1 | 2205 | fb = qemu_file_new_output(QIO_CHANNEL(bioc)); |
61b67d47 | 2206 | object_unref(OBJECT(bioc)); |
1d34e4bf | 2207 | |
c76201ab DDAG |
2208 | /* |
2209 | * Make sure the receiver can get incoming pages before we send the rest | |
2210 | * of the state | |
2211 | */ | |
2212 | qemu_savevm_send_postcopy_listen(fb); | |
2213 | ||
a1fbe750 | 2214 | qemu_savevm_state_complete_precopy(fb, false, false); |
58110f0a VSO |
2215 | if (migrate_postcopy_ram()) { |
2216 | qemu_savevm_send_ping(fb, 3); | |
2217 | } | |
1d34e4bf DDAG |
2218 | |
2219 | qemu_savevm_send_postcopy_run(fb); | |
2220 | ||
2221 | /* <><> end of stuff going into the package */ | |
1d34e4bf | 2222 | |
ef8d6488 DDAG |
2223 | /* Last point of recovery; as soon as we send the package the destination |
2224 | * can open devices and potentially start running. | |
2225 | * Lets just check again we've not got any errors. | |
2226 | */ | |
2227 | ret = qemu_file_get_error(ms->to_dst_file); | |
2228 | if (ret) { | |
908927db | 2229 | error_setg(errp, "postcopy_start: Migration stream errored (pre package)"); |
ef8d6488 DDAG |
2230 | goto fail_closefb; |
2231 | } | |
2232 | ||
2233 | restart_block = false; | |
2234 | ||
1d34e4bf | 2235 | /* Now send that blob */ |
61b67d47 | 2236 | if (qemu_savevm_send_packaged(ms->to_dst_file, bioc->data, bioc->usage)) { |
1d34e4bf DDAG |
2237 | goto fail_closefb; |
2238 | } | |
2239 | qemu_fclose(fb); | |
b82fc321 DDAG |
2240 | |
2241 | /* Send a notify to give a chance for anything that needs to happen | |
2242 | * at the transition to postcopy and after the device state; in particular | |
2243 | * spice needs to trigger a transition now | |
2244 | */ | |
2245 | ms->postcopy_after_devices = true; | |
d9cda213 | 2246 | migration_call_notifiers(ms); |
b82fc321 | 2247 | |
1d34e4bf DDAG |
2248 | ms->downtime = qemu_clock_get_ms(QEMU_CLOCK_REALTIME) - time_at_stop; |
2249 | ||
2250 | qemu_mutex_unlock_iothread(); | |
2251 | ||
58110f0a VSO |
2252 | if (migrate_postcopy_ram()) { |
2253 | /* | |
2254 | * Although this ping is just for debug, it could potentially be | |
2255 | * used for getting a better measurement of downtime at the source. | |
2256 | */ | |
2257 | qemu_savevm_send_ping(ms->to_dst_file, 4); | |
2258 | } | |
1d34e4bf | 2259 | |
ced1c616 PB |
2260 | if (migrate_release_ram()) { |
2261 | ram_postcopy_migrated_memory_release(ms); | |
2262 | } | |
2263 | ||
89a02a9f | 2264 | ret = qemu_file_get_error(ms->to_dst_file); |
1d34e4bf | 2265 | if (ret) { |
908927db | 2266 | error_setg(errp, "postcopy_start: Migration stream errored"); |
48781e5b | 2267 | migrate_set_state(&ms->state, MIGRATION_STATUS_POSTCOPY_ACTIVE, |
1d34e4bf DDAG |
2268 | MIGRATION_STATUS_FAILED); |
2269 | } | |
2270 | ||
c01b16ed PX |
2271 | trace_postcopy_preempt_enabled(migrate_postcopy_preempt()); |
2272 | ||
1d34e4bf DDAG |
2273 | return ret; |
2274 | ||
2275 | fail_closefb: | |
2276 | qemu_fclose(fb); | |
2277 | fail: | |
48781e5b | 2278 | migrate_set_state(&ms->state, MIGRATION_STATUS_POSTCOPY_ACTIVE, |
1d34e4bf | 2279 | MIGRATION_STATUS_FAILED); |
ef8d6488 DDAG |
2280 | if (restart_block) { |
2281 | /* A failure happened early enough that we know the destination hasn't | |
2282 | * accessed block devices, so we're safe to recover. | |
2283 | */ | |
2284 | Error *local_err = NULL; | |
2285 | ||
3b717194 | 2286 | bdrv_activate_all(&local_err); |
ef8d6488 DDAG |
2287 | if (local_err) { |
2288 | error_report_err(local_err); | |
2289 | } | |
2290 | } | |
1d34e4bf DDAG |
2291 | qemu_mutex_unlock_iothread(); |
2292 | return -1; | |
2293 | } | |
2294 | ||
e91d8951 DDAG |
2295 | /** |
2296 | * migration_maybe_pause: Pause if required to by | |
2297 | * migrate_pause_before_switchover called with the iothread locked | |
2298 | * Returns: 0 on success | |
2299 | */ | |
0331c8ca DDAG |
2300 | static int migration_maybe_pause(MigrationState *s, |
2301 | int *current_active_state, | |
2302 | int new_state) | |
e91d8951 DDAG |
2303 | { |
2304 | if (!migrate_pause_before_switchover()) { | |
2305 | return 0; | |
2306 | } | |
2307 | ||
2308 | /* Since leaving this state is not atomic with posting the semaphore | |
2309 | * it's possible that someone could have issued multiple migrate_continue | |
2310 | * and the semaphore is incorrectly positive at this point; | |
2311 | * the docs say it's undefined to reinit a semaphore that's already | |
2312 | * init'd, so use timedwait to eat up any existing posts. | |
2313 | */ | |
2314 | while (qemu_sem_timedwait(&s->pause_sem, 1) == 0) { | |
2315 | /* This block intentionally left blank */ | |
2316 | } | |
2317 | ||
8958338b ZF |
2318 | /* |
2319 | * If the migration is cancelled when it is in the completion phase, | |
2320 | * the migration state is set to MIGRATION_STATUS_CANCELLING. | |
2321 | * So we don't need to wait a semaphore, otherwise we would always | |
2322 | * wait for the 'pause_sem' semaphore. | |
2323 | */ | |
2324 | if (s->state != MIGRATION_STATUS_CANCELLING) { | |
2325 | qemu_mutex_unlock_iothread(); | |
2326 | migrate_set_state(&s->state, *current_active_state, | |
2327 | MIGRATION_STATUS_PRE_SWITCHOVER); | |
2328 | qemu_sem_wait(&s->pause_sem); | |
2329 | migrate_set_state(&s->state, MIGRATION_STATUS_PRE_SWITCHOVER, | |
2330 | new_state); | |
2331 | *current_active_state = new_state; | |
2332 | qemu_mutex_lock_iothread(); | |
2333 | } | |
e91d8951 | 2334 | |
0331c8ca | 2335 | return s->state == new_state ? 0 : -EINVAL; |
e91d8951 DDAG |
2336 | } |
2337 | ||
d7f5a043 WW |
2338 | static int migration_completion_precopy(MigrationState *s, |
2339 | int *current_active_state) | |
09f6c85e DDAG |
2340 | { |
2341 | int ret; | |
2342 | ||
d7f5a043 WW |
2343 | qemu_mutex_lock_iothread(); |
2344 | s->downtime_start = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); | |
2345 | qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER, NULL); | |
f4584076 | 2346 | |
d7f5a043 WW |
2347 | s->vm_old_state = runstate_get(); |
2348 | global_state_store(); | |
c33f1829 | 2349 | |
d7f5a043 WW |
2350 | ret = vm_stop_force_state(RUN_STATE_FINISH_MIGRATE); |
2351 | trace_migration_completion_vm_stop(ret); | |
2352 | if (ret < 0) { | |
2353 | goto out_unlock; | |
2354 | } | |
c33f1829 | 2355 | |
d7f5a043 WW |
2356 | ret = migration_maybe_pause(s, current_active_state, |
2357 | MIGRATION_STATUS_DEVICE); | |
2358 | if (ret < 0) { | |
2359 | goto out_unlock; | |
2360 | } | |
09f6c85e | 2361 | |
d7f5a043 WW |
2362 | /* |
2363 | * Inactivate disks except in COLO, and track that we have done so in order | |
2364 | * to remember to reactivate them if migration fails or is cancelled. | |
2365 | */ | |
2366 | s->block_inactive = !migrate_colo(); | |
2367 | migration_rate_set(RATE_LIMIT_DISABLED); | |
2368 | ret = qemu_savevm_state_complete_precopy(s->to_dst_file, false, | |
2369 | s->block_inactive); | |
2370 | out_unlock: | |
2371 | qemu_mutex_unlock_iothread(); | |
2372 | return ret; | |
2373 | } | |
b10ac0c4 | 2374 | |
d7f5a043 WW |
2375 | static void migration_completion_postcopy(MigrationState *s) |
2376 | { | |
2377 | trace_migration_completion_postcopy_end(); | |
2378 | ||
2379 | qemu_mutex_lock_iothread(); | |
2380 | qemu_savevm_state_complete_postcopy(s->to_dst_file); | |
2381 | qemu_mutex_unlock_iothread(); | |
2382 | ||
2383 | /* | |
2384 | * Shutdown the postcopy fast path thread. This is only needed when dest | |
2385 | * QEMU binary is old (7.1/7.2). QEMU 8.0+ doesn't need this. | |
2386 | */ | |
2387 | if (migrate_postcopy_preempt() && s->preempt_pre_7_2) { | |
2388 | postcopy_preempt_shutdown_file(s); | |
2389 | } | |
68b88468 | 2390 | |
d7f5a043 WW |
2391 | trace_migration_completion_postcopy_end_after_complete(); |
2392 | } | |
2393 | ||
2394 | static void migration_completion_failed(MigrationState *s, | |
2395 | int current_active_state) | |
2396 | { | |
2397 | if (s->block_inactive && (s->state == MIGRATION_STATUS_ACTIVE || | |
2398 | s->state == MIGRATION_STATUS_DEVICE)) { | |
6621883f | 2399 | /* |
d7f5a043 WW |
2400 | * If not doing postcopy, vm_start() will be called: let's |
2401 | * regain control on images. | |
6621883f | 2402 | */ |
d7f5a043 WW |
2403 | Error *local_err = NULL; |
2404 | ||
2405 | qemu_mutex_lock_iothread(); | |
2406 | bdrv_activate_all(&local_err); | |
2407 | if (local_err) { | |
2408 | error_report_err(local_err); | |
2409 | } else { | |
2410 | s->block_inactive = false; | |
36f62f11 | 2411 | } |
d7f5a043 WW |
2412 | qemu_mutex_unlock_iothread(); |
2413 | } | |
2414 | ||
2415 | migrate_set_state(&s->state, current_active_state, | |
2416 | MIGRATION_STATUS_FAILED); | |
2417 | } | |
36f62f11 | 2418 | |
d7f5a043 WW |
2419 | /** |
2420 | * migration_completion: Used by migration_thread when there's not much left. | |
2421 | * The caller 'breaks' the loop when this returns. | |
2422 | * | |
2423 | * @s: Current migration state | |
2424 | */ | |
2425 | static void migration_completion(MigrationState *s) | |
2426 | { | |
2427 | int ret = 0; | |
2428 | int current_active_state = s->state; | |
2429 | ||
2430 | if (s->state == MIGRATION_STATUS_ACTIVE) { | |
2431 | ret = migration_completion_precopy(s, ¤t_active_state); | |
2432 | } else if (s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) { | |
2433 | migration_completion_postcopy(s); | |
444252b9 | 2434 | } else { |
d7f5a043 WW |
2435 | ret = -1; |
2436 | } | |
2437 | ||
2438 | if (ret < 0) { | |
6ba11211 | 2439 | goto fail; |
09f6c85e | 2440 | } |
09f6c85e | 2441 | |
d7f5a043 | 2442 | if (close_return_path_on_source(s)) { |
d50f5dc0 | 2443 | goto fail; |
09f6c85e DDAG |
2444 | } |
2445 | ||
89a02a9f | 2446 | if (qemu_file_get_error(s->to_dst_file)) { |
09f6c85e | 2447 | trace_migration_completion_file_err(); |
6dab4c93 | 2448 | goto fail; |
09f6c85e DDAG |
2449 | } |
2450 | ||
5e804644 | 2451 | if (migrate_colo() && s->state == MIGRATION_STATUS_ACTIVE) { |
eeeb48ee ZC |
2452 | /* COLO does not support postcopy */ |
2453 | migrate_set_state(&s->state, MIGRATION_STATUS_ACTIVE, | |
2454 | MIGRATION_STATUS_COLO); | |
2455 | } else { | |
0b827d5e HZ |
2456 | migrate_set_state(&s->state, current_active_state, |
2457 | MIGRATION_STATUS_COMPLETED); | |
2458 | } | |
2459 | ||
09f6c85e DDAG |
2460 | return; |
2461 | ||
6dab4c93 | 2462 | fail: |
d7f5a043 | 2463 | migration_completion_failed(s, current_active_state); |
09f6c85e DDAG |
2464 | } |
2465 | ||
8518278a AG |
2466 | /** |
2467 | * bg_migration_completion: Used by bg_migration_thread when after all the | |
2468 | * RAM has been saved. The caller 'breaks' the loop when this returns. | |
2469 | * | |
2470 | * @s: Current migration state | |
2471 | */ | |
2472 | static void bg_migration_completion(MigrationState *s) | |
2473 | { | |
2474 | int current_active_state = s->state; | |
2475 | ||
8518278a AG |
2476 | if (s->state == MIGRATION_STATUS_ACTIVE) { |
2477 | /* | |
2478 | * By this moment we have RAM content saved into the migration stream. | |
2479 | * The next step is to flush the non-RAM content (device state) | |
2480 | * right after the ram content. The device state has been stored into | |
2481 | * the temporary buffer before RAM saving started. | |
2482 | */ | |
2483 | qemu_put_buffer(s->to_dst_file, s->bioc->data, s->bioc->usage); | |
2484 | qemu_fflush(s->to_dst_file); | |
2485 | } else if (s->state == MIGRATION_STATUS_CANCELLING) { | |
2486 | goto fail; | |
2487 | } | |
2488 | ||
2489 | if (qemu_file_get_error(s->to_dst_file)) { | |
2490 | trace_migration_completion_file_err(); | |
2491 | goto fail; | |
2492 | } | |
2493 | ||
2494 | migrate_set_state(&s->state, current_active_state, | |
2495 | MIGRATION_STATUS_COMPLETED); | |
2496 | return; | |
2497 | ||
2498 | fail: | |
2499 | migrate_set_state(&s->state, current_active_state, | |
2500 | MIGRATION_STATUS_FAILED); | |
2501 | } | |
2502 | ||
b23c2ade PX |
2503 | typedef enum MigThrError { |
2504 | /* No error detected */ | |
2505 | MIG_THR_ERR_NONE = 0, | |
2506 | /* Detected error, but resumed successfully */ | |
2507 | MIG_THR_ERR_RECOVERED = 1, | |
2508 | /* Detected fatal error, need to exit */ | |
2509 | MIG_THR_ERR_FATAL = 2, | |
2510 | } MigThrError; | |
2511 | ||
94190696 PX |
2512 | static int postcopy_resume_handshake(MigrationState *s) |
2513 | { | |
2514 | qemu_savevm_send_postcopy_resume(s->to_dst_file); | |
2515 | ||
2516 | while (s->state == MIGRATION_STATUS_POSTCOPY_RECOVER) { | |
5e79a4bf | 2517 | migration_rp_wait(s); |
94190696 PX |
2518 | } |
2519 | ||
2520 | if (s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) { | |
2521 | return 0; | |
2522 | } | |
2523 | ||
2524 | return -1; | |
2525 | } | |
2526 | ||
135b87b4 PX |
2527 | /* Return zero if success, or <0 for error */ |
2528 | static int postcopy_do_resume(MigrationState *s) | |
2529 | { | |
d1b8eadb PX |
2530 | int ret; |
2531 | ||
2532 | /* | |
2533 | * Call all the resume_prepare() hooks, so that modules can be | |
2534 | * ready for the migration resume. | |
2535 | */ | |
2536 | ret = qemu_savevm_state_resume_prepare(s); | |
2537 | if (ret) { | |
2538 | error_report("%s: resume_prepare() failure detected: %d", | |
2539 | __func__, ret); | |
2540 | return ret; | |
2541 | } | |
2542 | ||
5655aab0 PX |
2543 | /* |
2544 | * If preempt is enabled, re-establish the preempt channel. Note that | |
2545 | * we do it after resume prepare to make sure the main channel will be | |
2546 | * created before the preempt channel. E.g. with weak network, the | |
2547 | * dest QEMU may get messed up with the preempt and main channels on | |
2548 | * the order of connection setup. This guarantees the correct order. | |
2549 | */ | |
2550 | ret = postcopy_preempt_establish_channel(s); | |
2551 | if (ret) { | |
2552 | error_report("%s: postcopy_preempt_establish_channel(): %d", | |
2553 | __func__, ret); | |
2554 | return ret; | |
2555 | } | |
2556 | ||
d1b8eadb | 2557 | /* |
94190696 PX |
2558 | * Last handshake with destination on the resume (destination will |
2559 | * switch to postcopy-active afterwards) | |
d1b8eadb | 2560 | */ |
94190696 PX |
2561 | ret = postcopy_resume_handshake(s); |
2562 | if (ret) { | |
2563 | error_report("%s: handshake failed: %d", __func__, ret); | |
2564 | return ret; | |
2565 | } | |
d1b8eadb | 2566 | |
135b87b4 PX |
2567 | return 0; |
2568 | } | |
2569 | ||
b23c2ade PX |
2570 | /* |
2571 | * We don't return until we are in a safe state to continue current | |
2572 | * postcopy migration. Returns MIG_THR_ERR_RECOVERED if recovered, or | |
2573 | * MIG_THR_ERR_FATAL if unrecovery failure happened. | |
2574 | */ | |
2575 | static MigThrError postcopy_pause(MigrationState *s) | |
2576 | { | |
2577 | assert(s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE); | |
b23c2ade | 2578 | |
135b87b4 | 2579 | while (true) { |
62df066f PX |
2580 | QEMUFile *file; |
2581 | ||
39675fff PX |
2582 | /* |
2583 | * Current channel is possibly broken. Release it. Note that this is | |
2584 | * guaranteed even without lock because to_dst_file should only be | |
2585 | * modified by the migration thread. That also guarantees that the | |
2586 | * unregister of yank is safe too without the lock. It should be safe | |
2587 | * even to be within the qemu_file_lock, but we didn't do that to avoid | |
2588 | * taking more mutex (yank_lock) within qemu_file_lock. TL;DR: we make | |
2589 | * the qemu_file_lock critical section as small as possible. | |
2590 | */ | |
135b87b4 | 2591 | assert(s->to_dst_file); |
39675fff | 2592 | migration_ioc_unregister_yank_from_file(s->to_dst_file); |
62df066f PX |
2593 | qemu_mutex_lock(&s->qemu_file_lock); |
2594 | file = s->to_dst_file; | |
135b87b4 | 2595 | s->to_dst_file = NULL; |
62df066f PX |
2596 | qemu_mutex_unlock(&s->qemu_file_lock); |
2597 | ||
2598 | qemu_file_shutdown(file); | |
2599 | qemu_fclose(file); | |
b23c2ade | 2600 | |
ef796ee9 FR |
2601 | /* |
2602 | * We're already pausing, so ignore any errors on the return | |
2603 | * path and just wait for the thread to finish. It will be | |
2604 | * re-created when we resume. | |
2605 | */ | |
d7f5a043 | 2606 | close_return_path_on_source(s); |
ef796ee9 | 2607 | |
d246ea50 PX |
2608 | migrate_set_state(&s->state, s->state, |
2609 | MIGRATION_STATUS_POSTCOPY_PAUSED); | |
2610 | ||
135b87b4 PX |
2611 | error_report("Detected IO failure for postcopy. " |
2612 | "Migration paused."); | |
b23c2ade | 2613 | |
135b87b4 PX |
2614 | /* |
2615 | * We wait until things fixed up. Then someone will setup the | |
2616 | * status back for us. | |
2617 | */ | |
2618 | while (s->state == MIGRATION_STATUS_POSTCOPY_PAUSED) { | |
2619 | qemu_sem_wait(&s->postcopy_pause_sem); | |
2620 | } | |
2621 | ||
2622 | if (s->state == MIGRATION_STATUS_POSTCOPY_RECOVER) { | |
2623 | /* Woken up by a recover procedure. Give it a shot */ | |
b23c2ade | 2624 | |
135b87b4 PX |
2625 | /* Do the resume logic */ |
2626 | if (postcopy_do_resume(s) == 0) { | |
2627 | /* Let's continue! */ | |
2628 | trace_postcopy_pause_continued(); | |
2629 | return MIG_THR_ERR_RECOVERED; | |
2630 | } else { | |
2631 | /* | |
2632 | * Something wrong happened during the recovery, let's | |
2633 | * pause again. Pause is always better than throwing | |
2634 | * data away. | |
2635 | */ | |
2636 | continue; | |
2637 | } | |
2638 | } else { | |
2639 | /* This is not right... Time to quit. */ | |
2640 | return MIG_THR_ERR_FATAL; | |
2641 | } | |
2642 | } | |
b23c2ade PX |
2643 | } |
2644 | ||
2645 | static MigThrError migration_detect_error(MigrationState *s) | |
2646 | { | |
2647 | int ret; | |
c3c5eae6 | 2648 | int state = s->state; |
3d661c8a | 2649 | Error *local_error = NULL; |
c3c5eae6 DDAG |
2650 | |
2651 | if (state == MIGRATION_STATUS_CANCELLING || | |
2652 | state == MIGRATION_STATUS_CANCELLED) { | |
2653 | /* End the migration, but don't set the state to failed */ | |
2654 | return MIG_THR_ERR_FATAL; | |
2655 | } | |
b23c2ade | 2656 | |
60bb3c58 PX |
2657 | /* |
2658 | * Try to detect any file errors. Note that postcopy_qemufile_src will | |
2659 | * be NULL when postcopy preempt is not enabled. | |
2660 | */ | |
2661 | ret = qemu_file_get_error_obj_any(s->to_dst_file, | |
2662 | s->postcopy_qemufile_src, | |
2663 | &local_error); | |
b23c2ade PX |
2664 | if (!ret) { |
2665 | /* Everything is fine */ | |
3d661c8a | 2666 | assert(!local_error); |
b23c2ade PX |
2667 | return MIG_THR_ERR_NONE; |
2668 | } | |
2669 | ||
3d661c8a YK |
2670 | if (local_error) { |
2671 | migrate_set_error(s, local_error); | |
2672 | error_free(local_error); | |
2673 | } | |
2674 | ||
d5c8f2af | 2675 | if (state == MIGRATION_STATUS_POSTCOPY_ACTIVE && ret) { |
b23c2ade PX |
2676 | /* |
2677 | * For postcopy, we allow the network to be down for a | |
2678 | * while. After that, it can be continued by a | |
2679 | * recovery phase. | |
2680 | */ | |
2681 | return postcopy_pause(s); | |
2682 | } else { | |
2683 | /* | |
2684 | * For precopy (or postcopy with error outside IO), we fail | |
2685 | * with no time. | |
2686 | */ | |
c3c5eae6 | 2687 | migrate_set_state(&s->state, state, MIGRATION_STATUS_FAILED); |
b23c2ade PX |
2688 | trace_migration_thread_file_err(); |
2689 | ||
2690 | /* Time to stop the migration, now. */ | |
2691 | return MIG_THR_ERR_FATAL; | |
2692 | } | |
2693 | } | |
2694 | ||
cf011f08 PX |
2695 | static void migration_calculate_complete(MigrationState *s) |
2696 | { | |
99319e2d | 2697 | uint64_t bytes = migration_transferred_bytes(s->to_dst_file); |
cf011f08 | 2698 | int64_t end_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); |
6cde6fbe | 2699 | int64_t transfer_time; |
cf011f08 PX |
2700 | |
2701 | s->total_time = end_time - s->start_time; | |
2702 | if (!s->downtime) { | |
2703 | /* | |
2704 | * It's still not set, so we are precopy migration. For | |
2705 | * postcopy, downtime is calculated during postcopy_start(). | |
2706 | */ | |
2707 | s->downtime = end_time - s->downtime_start; | |
2708 | } | |
2709 | ||
6cde6fbe JQ |
2710 | transfer_time = s->total_time - s->setup_time; |
2711 | if (transfer_time) { | |
2712 | s->mbps = ((double) bytes * 8.0) / transfer_time / 1000; | |
cf011f08 PX |
2713 | } |
2714 | } | |
2715 | ||
87f3bd87 IR |
2716 | static void update_iteration_initial_status(MigrationState *s) |
2717 | { | |
2718 | /* | |
2719 | * Update these three fields at the same time to avoid mismatch info lead | |
2720 | * wrong speed calculation. | |
2721 | */ | |
2722 | s->iteration_start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); | |
99319e2d | 2723 | s->iteration_initial_bytes = migration_transferred_bytes(s->to_dst_file); |
87f3bd87 IR |
2724 | s->iteration_initial_pages = ram_get_total_transferred_pages(); |
2725 | } | |
2726 | ||
b15df1ae PX |
2727 | static void migration_update_counters(MigrationState *s, |
2728 | int64_t current_time) | |
2729 | { | |
aecbfe9c | 2730 | uint64_t transferred, transferred_pages, time_spent; |
0c8f0efd | 2731 | uint64_t current_bytes; /* bytes transferred since the beginning */ |
8b239597 PX |
2732 | uint64_t switchover_bw; |
2733 | /* Expected bandwidth when switching over to destination QEMU */ | |
2734 | double expected_bw_per_ms; | |
b15df1ae PX |
2735 | double bandwidth; |
2736 | ||
2737 | if (current_time < s->iteration_start_time + BUFFER_DELAY) { | |
2738 | return; | |
2739 | } | |
2740 | ||
8b239597 | 2741 | switchover_bw = migrate_avail_switchover_bandwidth(); |
99319e2d | 2742 | current_bytes = migration_transferred_bytes(s->to_dst_file); |
0c8f0efd | 2743 | transferred = current_bytes - s->iteration_initial_bytes; |
b15df1ae PX |
2744 | time_spent = current_time - s->iteration_start_time; |
2745 | bandwidth = (double)transferred / time_spent; | |
8b239597 PX |
2746 | |
2747 | if (switchover_bw) { | |
2748 | /* | |
2749 | * If the user specified a switchover bandwidth, let's trust the | |
2750 | * user so that can be more accurate than what we estimated. | |
2751 | */ | |
2752 | expected_bw_per_ms = switchover_bw / 1000; | |
2753 | } else { | |
2754 | /* If the user doesn't specify bandwidth, we use the estimated */ | |
2755 | expected_bw_per_ms = bandwidth; | |
2756 | } | |
2757 | ||
2758 | s->threshold_size = expected_bw_per_ms * migrate_downtime_limit(); | |
b15df1ae PX |
2759 | |
2760 | s->mbps = (((double) transferred * 8.0) / | |
2761 | ((double) time_spent / 1000.0)) / 1000.0 / 1000.0; | |
2762 | ||
aecbfe9c XG |
2763 | transferred_pages = ram_get_total_transferred_pages() - |
2764 | s->iteration_initial_pages; | |
2765 | s->pages_per_second = (double) transferred_pages / | |
2766 | (((double) time_spent / 1000.0)); | |
2767 | ||
b15df1ae PX |
2768 | /* |
2769 | * if we haven't sent anything, we don't want to | |
2770 | * recalculate. 10000 is a small enough number for our purposes | |
2771 | */ | |
aff3f660 | 2772 | if (stat64_get(&mig_stats.dirty_pages_rate) && |
72f8e587 | 2773 | transferred > 10000) { |
73208a33 | 2774 | s->expected_downtime = |
8b239597 | 2775 | stat64_get(&mig_stats.dirty_bytes_last_sync) / expected_bw_per_ms; |
b15df1ae PX |
2776 | } |
2777 | ||
813cd616 | 2778 | migration_rate_reset(s->to_dst_file); |
b15df1ae | 2779 | |
87f3bd87 | 2780 | update_iteration_initial_status(s); |
b15df1ae PX |
2781 | |
2782 | trace_migrate_transferred(transferred, time_spent, | |
8b239597 PX |
2783 | /* Both in unit bytes/ms */ |
2784 | bandwidth, switchover_bw / 1000, | |
2785 | s->threshold_size); | |
b15df1ae PX |
2786 | } |
2787 | ||
1b4adb10 AH |
2788 | static bool migration_can_switchover(MigrationState *s) |
2789 | { | |
2790 | if (!migrate_switchover_ack()) { | |
2791 | return true; | |
2792 | } | |
2793 | ||
2794 | /* No reason to wait for switchover ACK if VM is stopped */ | |
2795 | if (!runstate_is_running()) { | |
2796 | return true; | |
2797 | } | |
2798 | ||
2799 | return s->switchover_acked; | |
2800 | } | |
2801 | ||
2ad87305 PX |
2802 | /* Migration thread iteration status */ |
2803 | typedef enum { | |
2804 | MIG_ITERATE_RESUME, /* Resume current iteration */ | |
2805 | MIG_ITERATE_SKIP, /* Skip current iteration */ | |
2806 | MIG_ITERATE_BREAK, /* Break the loop */ | |
2807 | } MigIterateState; | |
2808 | ||
2809 | /* | |
2810 | * Return true if continue to the next iteration directly, false | |
2811 | * otherwise. | |
2812 | */ | |
2813 | static MigIterateState migration_iteration_run(MigrationState *s) | |
2814 | { | |
24beea4e | 2815 | uint64_t must_precopy, can_postcopy; |
908927db | 2816 | Error *local_err = NULL; |
2ad87305 | 2817 | bool in_postcopy = s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE; |
1b4adb10 | 2818 | bool can_switchover = migration_can_switchover(s); |
2ad87305 | 2819 | |
24beea4e JQ |
2820 | qemu_savevm_state_pending_estimate(&must_precopy, &can_postcopy); |
2821 | uint64_t pending_size = must_precopy + can_postcopy; | |
2ad87305 | 2822 | |
24beea4e | 2823 | trace_migrate_pending_estimate(pending_size, must_precopy, can_postcopy); |
c8df4a7a | 2824 | |
24beea4e JQ |
2825 | if (must_precopy <= s->threshold_size) { |
2826 | qemu_savevm_state_pending_exact(&must_precopy, &can_postcopy); | |
2827 | pending_size = must_precopy + can_postcopy; | |
2828 | trace_migrate_pending_exact(pending_size, must_precopy, can_postcopy); | |
c8df4a7a | 2829 | } |
2ad87305 | 2830 | |
1b4adb10 | 2831 | if ((!pending_size || pending_size < s->threshold_size) && can_switchover) { |
2ad87305 PX |
2832 | trace_migration_thread_low_pending(pending_size); |
2833 | migration_completion(s); | |
2834 | return MIG_ITERATE_BREAK; | |
2835 | } | |
2836 | ||
d9df9292 | 2837 | /* Still a significant amount to transfer */ |
1b4adb10 | 2838 | if (!in_postcopy && must_precopy <= s->threshold_size && can_switchover && |
d9df9292 | 2839 | qatomic_read(&s->start_postcopy)) { |
908927db TG |
2840 | if (postcopy_start(s, &local_err)) { |
2841 | migrate_set_error(s, local_err); | |
2842 | error_report_err(local_err); | |
d9df9292 JQ |
2843 | } |
2844 | return MIG_ITERATE_SKIP; | |
2845 | } | |
2846 | ||
2847 | /* Just another iteration step */ | |
2848 | qemu_savevm_state_iterate(s->to_dst_file, in_postcopy); | |
2ad87305 PX |
2849 | return MIG_ITERATE_RESUME; |
2850 | } | |
2851 | ||
199aa6d4 PX |
2852 | static void migration_iteration_finish(MigrationState *s) |
2853 | { | |
2854 | /* If we enabled cpu throttling for auto-converge, turn it off. */ | |
2855 | cpu_throttle_stop(); | |
2856 | ||
2857 | qemu_mutex_lock_iothread(); | |
2858 | switch (s->state) { | |
2859 | case MIGRATION_STATUS_COMPLETED: | |
2860 | migration_calculate_complete(s); | |
2861 | runstate_set(RUN_STATE_POSTMIGRATE); | |
2862 | break; | |
751fe4c6 | 2863 | case MIGRATION_STATUS_COLO: |
d70178a8 | 2864 | assert(migrate_colo()); |
199aa6d4 | 2865 | migrate_start_colo_process(s); |
f4584076 | 2866 | s->vm_old_state = RUN_STATE_RUNNING; |
199aa6d4 PX |
2867 | /* Fallthrough */ |
2868 | case MIGRATION_STATUS_FAILED: | |
2869 | case MIGRATION_STATUS_CANCELLED: | |
57225e5f | 2870 | case MIGRATION_STATUS_CANCELLING: |
f4584076 | 2871 | if (s->vm_old_state == RUN_STATE_RUNNING) { |
aa505f8e RL |
2872 | if (!runstate_check(RUN_STATE_SHUTDOWN)) { |
2873 | vm_start(); | |
2874 | } | |
199aa6d4 PX |
2875 | } else { |
2876 | if (runstate_check(RUN_STATE_FINISH_MIGRATE)) { | |
a4c6275a | 2877 | runstate_set(s->vm_old_state); |
199aa6d4 PX |
2878 | } |
2879 | } | |
2880 | break; | |
2881 | ||
2882 | default: | |
2883 | /* Should not reach here, but if so, forgive the VM. */ | |
2884 | error_report("%s: Unknown ending state %d", __func__, s->state); | |
2885 | break; | |
2886 | } | |
fd392cfa | 2887 | migrate_fd_cleanup_schedule(s); |
199aa6d4 PX |
2888 | qemu_mutex_unlock_iothread(); |
2889 | } | |
2890 | ||
8518278a AG |
2891 | static void bg_migration_iteration_finish(MigrationState *s) |
2892 | { | |
3a8b81f2 FE |
2893 | /* |
2894 | * Stop tracking RAM writes - un-protect memory, un-register UFFD | |
2895 | * memory ranges, flush kernel wait queues and wake up threads | |
2896 | * waiting for write fault to be resolved. | |
2897 | */ | |
2898 | ram_write_tracking_stop(); | |
2899 | ||
8518278a AG |
2900 | qemu_mutex_lock_iothread(); |
2901 | switch (s->state) { | |
2902 | case MIGRATION_STATUS_COMPLETED: | |
2903 | migration_calculate_complete(s); | |
2904 | break; | |
2905 | ||
2906 | case MIGRATION_STATUS_ACTIVE: | |
2907 | case MIGRATION_STATUS_FAILED: | |
2908 | case MIGRATION_STATUS_CANCELLED: | |
2909 | case MIGRATION_STATUS_CANCELLING: | |
2910 | break; | |
2911 | ||
2912 | default: | |
2913 | /* Should not reach here, but if so, forgive the VM. */ | |
2914 | error_report("%s: Unknown ending state %d", __func__, s->state); | |
2915 | break; | |
2916 | } | |
2917 | ||
2918 | migrate_fd_cleanup_schedule(s); | |
2919 | qemu_mutex_unlock_iothread(); | |
2920 | } | |
2921 | ||
2922 | /* | |
2923 | * Return true if continue to the next iteration directly, false | |
2924 | * otherwise. | |
2925 | */ | |
2926 | static MigIterateState bg_migration_iteration_run(MigrationState *s) | |
2927 | { | |
2928 | int res; | |
2929 | ||
2930 | res = qemu_savevm_state_iterate(s->to_dst_file, false); | |
2931 | if (res > 0) { | |
2932 | bg_migration_completion(s); | |
2933 | return MIG_ITERATE_BREAK; | |
2934 | } | |
2935 | ||
2936 | return MIG_ITERATE_RESUME; | |
2937 | } | |
2938 | ||
ad767bed DDAG |
2939 | void migration_make_urgent_request(void) |
2940 | { | |
2941 | qemu_sem_post(&migrate_get_current()->rate_limit_sem); | |
2942 | } | |
2943 | ||
2944 | void migration_consume_urgent_request(void) | |
2945 | { | |
2946 | qemu_sem_wait(&migrate_get_current()->rate_limit_sem); | |
2947 | } | |
2948 | ||
97e1e067 DDAG |
2949 | /* Returns true if the rate limiting was broken by an urgent request */ |
2950 | bool migration_rate_limit(void) | |
2951 | { | |
2952 | int64_t now = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); | |
2953 | MigrationState *s = migrate_get_current(); | |
2954 | ||
2955 | bool urgent = false; | |
2956 | migration_update_counters(s, now); | |
e1fde0e0 | 2957 | if (migration_rate_exceeded(s->to_dst_file)) { |
77386127 LS |
2958 | |
2959 | if (qemu_file_get_error(s->to_dst_file)) { | |
2960 | return false; | |
2961 | } | |
97e1e067 DDAG |
2962 | /* |
2963 | * Wait for a delay to do rate limiting OR | |
2964 | * something urgent to post the semaphore. | |
2965 | */ | |
2966 | int ms = s->iteration_start_time + BUFFER_DELAY - now; | |
2967 | trace_migration_rate_limit_pre(ms); | |
2968 | if (qemu_sem_timedwait(&s->rate_limit_sem, ms) == 0) { | |
2969 | /* | |
2970 | * We were woken by one or more urgent things but | |
2971 | * the timedwait will have consumed one of them. | |
2972 | * The service routine for the urgent wake will dec | |
2973 | * the semaphore itself for each item it consumes, | |
2974 | * so add this one we just eat back. | |
2975 | */ | |
2976 | qemu_sem_post(&s->rate_limit_sem); | |
2977 | urgent = true; | |
2978 | } | |
2979 | trace_migration_rate_limit_post(urgent); | |
2980 | } | |
2981 | return urgent; | |
2982 | } | |
2983 | ||
fde93d99 LV |
2984 | /* |
2985 | * if failover devices are present, wait they are completely | |
2986 | * unplugged | |
2987 | */ | |
2988 | ||
2989 | static void qemu_savevm_wait_unplug(MigrationState *s, int old_state, | |
2990 | int new_state) | |
2991 | { | |
2992 | if (qemu_savevm_state_guest_unplug_pending()) { | |
2993 | migrate_set_state(&s->state, old_state, MIGRATION_STATUS_WAIT_UNPLUG); | |
2994 | ||
2995 | while (s->state == MIGRATION_STATUS_WAIT_UNPLUG && | |
2996 | qemu_savevm_state_guest_unplug_pending()) { | |
2997 | qemu_sem_timedwait(&s->wait_unplug_sem, 250); | |
2998 | } | |
944bc528 LV |
2999 | if (s->state != MIGRATION_STATUS_WAIT_UNPLUG) { |
3000 | int timeout = 120; /* 30 seconds */ | |
3001 | /* | |
3002 | * migration has been canceled | |
3003 | * but as we have started an unplug we must wait the end | |
3004 | * to be able to plug back the card | |
3005 | */ | |
3006 | while (timeout-- && qemu_savevm_state_guest_unplug_pending()) { | |
3007 | qemu_sem_timedwait(&s->wait_unplug_sem, 250); | |
3008 | } | |
1b529d90 LV |
3009 | if (qemu_savevm_state_guest_unplug_pending() && |
3010 | !qtest_enabled()) { | |
a51dcef0 LV |
3011 | warn_report("migration: partially unplugged device on " |
3012 | "failure"); | |
3013 | } | |
944bc528 | 3014 | } |
fde93d99 LV |
3015 | |
3016 | migrate_set_state(&s->state, MIGRATION_STATUS_WAIT_UNPLUG, new_state); | |
3017 | } else { | |
3018 | migrate_set_state(&s->state, old_state, new_state); | |
3019 | } | |
3020 | } | |
3021 | ||
70b20477 DDAG |
3022 | /* |
3023 | * Master migration thread on the source VM. | |
3024 | * It drives the migration and pumps the data down the outgoing channel. | |
3025 | */ | |
5f496a1b | 3026 | static void *migration_thread(void *opaque) |
0d82d0e8 | 3027 | { |
9848a404 | 3028 | MigrationState *s = opaque; |
1b1f4ab6 | 3029 | MigrationThread *thread = NULL; |
bc72ad67 | 3030 | int64_t setup_start = qemu_clock_get_ms(QEMU_CLOCK_HOST); |
b23c2ade | 3031 | MigThrError thr_error; |
ad767bed | 3032 | bool urgent = false; |
76f5933a | 3033 | |
788fa680 | 3034 | thread = migration_threads_add("live_migration", qemu_get_thread_id()); |
1b1f4ab6 | 3035 | |
ab28bd23 PB |
3036 | rcu_register_thread(); |
3037 | ||
892ae715 | 3038 | object_ref(OBJECT(s)); |
87f3bd87 | 3039 | update_iteration_initial_status(s); |
b15df1ae | 3040 | |
930e239d | 3041 | qemu_mutex_lock_iothread(); |
89a02a9f | 3042 | qemu_savevm_state_header(s->to_dst_file); |
930e239d | 3043 | qemu_mutex_unlock_iothread(); |
1d34e4bf | 3044 | |
62a02658 PX |
3045 | /* |
3046 | * If we opened the return path, we need to make sure dst has it | |
3047 | * opened as well. | |
3048 | */ | |
43044ac0 | 3049 | if (s->rp_state.rp_thread_created) { |
1d34e4bf | 3050 | /* Now tell the dest that it should open its end so it can reply */ |
89a02a9f | 3051 | qemu_savevm_send_open_return_path(s->to_dst_file); |
1d34e4bf DDAG |
3052 | |
3053 | /* And do a ping that will make stuff easier to debug */ | |
89a02a9f | 3054 | qemu_savevm_send_ping(s->to_dst_file, 1); |
0425dc97 | 3055 | } |
1d34e4bf | 3056 | |
58110f0a | 3057 | if (migrate_postcopy()) { |
1d34e4bf DDAG |
3058 | /* |
3059 | * Tell the destination that we *might* want to do postcopy later; | |
3060 | * if the other end can't do postcopy it should fail now, nice and | |
3061 | * early. | |
3062 | */ | |
89a02a9f | 3063 | qemu_savevm_send_postcopy_advise(s->to_dst_file); |
1d34e4bf DDAG |
3064 | } |
3065 | ||
5e804644 | 3066 | if (migrate_colo()) { |
aad555c2 ZC |
3067 | /* Notify migration destination that we enable COLO */ |
3068 | qemu_savevm_send_colo_enable(s->to_dst_file); | |
3069 | } | |
3070 | ||
930e239d | 3071 | qemu_mutex_lock_iothread(); |
9907e842 | 3072 | qemu_savevm_state_setup(s->to_dst_file); |
930e239d | 3073 | qemu_mutex_unlock_iothread(); |
0d82d0e8 | 3074 | |
fde93d99 LV |
3075 | qemu_savevm_wait_unplug(s, MIGRATION_STATUS_SETUP, |
3076 | MIGRATION_STATUS_ACTIVE); | |
c7e0acd5 | 3077 | |
bc72ad67 | 3078 | s->setup_time = qemu_clock_get_ms(QEMU_CLOCK_HOST) - setup_start; |
29ae8a41 | 3079 | |
9ec055ae DDAG |
3080 | trace_migration_thread_setup_complete(); |
3081 | ||
8f8d528e | 3082 | while (migration_is_active(s)) { |
e1fde0e0 | 3083 | if (urgent || !migration_rate_exceeded(s->to_dst_file)) { |
2ad87305 PX |
3084 | MigIterateState iter_state = migration_iteration_run(s); |
3085 | if (iter_state == MIG_ITERATE_SKIP) { | |
3086 | continue; | |
3087 | } else if (iter_state == MIG_ITERATE_BREAK) { | |
09f6c85e | 3088 | break; |
c369f40d JQ |
3089 | } |
3090 | } | |
f4410a5d | 3091 | |
b23c2ade PX |
3092 | /* |
3093 | * Try to detect any kind of failures, and see whether we | |
3094 | * should stop the migration now. | |
3095 | */ | |
3096 | thr_error = migration_detect_error(s); | |
3097 | if (thr_error == MIG_THR_ERR_FATAL) { | |
3098 | /* Stop migration */ | |
fd45ee2c | 3099 | break; |
b23c2ade PX |
3100 | } else if (thr_error == MIG_THR_ERR_RECOVERED) { |
3101 | /* | |
3102 | * Just recovered from a e.g. network failure, reset all | |
3103 | * the local variables. This is important to avoid | |
3104 | * breaking transferred_bytes and bandwidth calculation | |
3105 | */ | |
87f3bd87 | 3106 | update_iteration_initial_status(s); |
fd45ee2c | 3107 | } |
b15df1ae | 3108 | |
97e1e067 | 3109 | urgent = migration_rate_limit(); |
a3fa1d78 PB |
3110 | } |
3111 | ||
1d34e4bf | 3112 | trace_migration_thread_after_loop(); |
199aa6d4 | 3113 | migration_iteration_finish(s); |
892ae715 | 3114 | object_unref(OBJECT(s)); |
ab28bd23 | 3115 | rcu_unregister_thread(); |
788fa680 | 3116 | migration_threads_remove(thread); |
0d82d0e8 JQ |
3117 | return NULL; |
3118 | } | |
3119 | ||
8518278a AG |
3120 | static void bg_migration_vm_start_bh(void *opaque) |
3121 | { | |
3122 | MigrationState *s = opaque; | |
3123 | ||
3124 | qemu_bh_delete(s->vm_start_bh); | |
3125 | s->vm_start_bh = NULL; | |
3126 | ||
3127 | vm_start(); | |
3128 | s->downtime = qemu_clock_get_ms(QEMU_CLOCK_REALTIME) - s->downtime_start; | |
3129 | } | |
3130 | ||
3131 | /** | |
3132 | * Background snapshot thread, based on live migration code. | |
3133 | * This is an alternative implementation of live migration mechanism | |
3134 | * introduced specifically to support background snapshots. | |
3135 | * | |
3136 | * It takes advantage of userfault_fd write protection mechanism introduced | |
3137 | * in v5.7 kernel. Compared to existing dirty page logging migration much | |
3138 | * lesser stream traffic is produced resulting in smaller snapshot images, | |
3139 | * simply cause of no page duplicates can get into the stream. | |
3140 | * | |
3141 | * Another key point is that generated vmstate stream reflects machine state | |
3142 | * 'frozen' at the beginning of snapshot creation compared to dirty page logging | |
3143 | * mechanism, which effectively results in that saved snapshot is the state of VM | |
3144 | * at the end of the process. | |
3145 | */ | |
3146 | static void *bg_migration_thread(void *opaque) | |
3147 | { | |
3148 | MigrationState *s = opaque; | |
3149 | int64_t setup_start; | |
3150 | MigThrError thr_error; | |
3151 | QEMUFile *fb; | |
3152 | bool early_fail = true; | |
3153 | ||
3154 | rcu_register_thread(); | |
3155 | object_ref(OBJECT(s)); | |
3156 | ||
e1fde0e0 | 3157 | migration_rate_set(RATE_LIMIT_DISABLED); |
8518278a AG |
3158 | |
3159 | setup_start = qemu_clock_get_ms(QEMU_CLOCK_HOST); | |
3160 | /* | |
3161 | * We want to save vmstate for the moment when migration has been | |
3162 | * initiated but also we want to save RAM content while VM is running. | |
3163 | * The RAM content should appear first in the vmstate. So, we first | |
3164 | * stash the non-RAM part of the vmstate to the temporary buffer, | |
3165 | * then write RAM part of the vmstate to the migration stream | |
3166 | * with vCPUs running and, finally, write stashed non-RAM part of | |
3167 | * the vmstate from the buffer to the migration stream. | |
3168 | */ | |
ecb23efe | 3169 | s->bioc = qio_channel_buffer_new(512 * 1024); |
8518278a | 3170 | qio_channel_set_name(QIO_CHANNEL(s->bioc), "vmstate-buffer"); |
77ef2dc1 | 3171 | fb = qemu_file_new_output(QIO_CHANNEL(s->bioc)); |
8518278a AG |
3172 | object_unref(OBJECT(s->bioc)); |
3173 | ||
3174 | update_iteration_initial_status(s); | |
3175 | ||
eeccb99c AG |
3176 | /* |
3177 | * Prepare for tracking memory writes with UFFD-WP - populate | |
3178 | * RAM pages before protecting. | |
3179 | */ | |
3180 | #ifdef __linux__ | |
3181 | ram_write_tracking_prepare(); | |
3182 | #endif | |
3183 | ||
930e239d | 3184 | qemu_mutex_lock_iothread(); |
8518278a AG |
3185 | qemu_savevm_state_header(s->to_dst_file); |
3186 | qemu_savevm_state_setup(s->to_dst_file); | |
930e239d | 3187 | qemu_mutex_unlock_iothread(); |
8518278a | 3188 | |
fde93d99 LV |
3189 | qemu_savevm_wait_unplug(s, MIGRATION_STATUS_SETUP, |
3190 | MIGRATION_STATUS_ACTIVE); | |
8518278a | 3191 | |
8518278a AG |
3192 | s->setup_time = qemu_clock_get_ms(QEMU_CLOCK_HOST) - setup_start; |
3193 | ||
3194 | trace_migration_thread_setup_complete(); | |
3195 | s->downtime_start = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); | |
3196 | ||
3197 | qemu_mutex_lock_iothread(); | |
3198 | ||
3199 | /* | |
3200 | * If VM is currently in suspended state, then, to make a valid runstate | |
3201 | * transition in vm_stop_force_state() we need to wakeup it up. | |
3202 | */ | |
3203 | qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER, NULL); | |
f4584076 | 3204 | s->vm_old_state = runstate_get(); |
8518278a | 3205 | |
c33f1829 | 3206 | global_state_store(); |
8518278a AG |
3207 | /* Forcibly stop VM before saving state of vCPUs and devices */ |
3208 | if (vm_stop_force_state(RUN_STATE_PAUSED)) { | |
3209 | goto fail; | |
3210 | } | |
3211 | /* | |
3212 | * Put vCPUs in sync with shadow context structures, then | |
3213 | * save their state to channel-buffer along with devices. | |
3214 | */ | |
3215 | cpu_synchronize_all_states(); | |
3216 | if (qemu_savevm_state_complete_precopy_non_iterable(fb, false, false)) { | |
3217 | goto fail; | |
3218 | } | |
ecb23efe AG |
3219 | /* |
3220 | * Since we are going to get non-iterable state data directly | |
3221 | * from s->bioc->data, explicit flush is needed here. | |
3222 | */ | |
3223 | qemu_fflush(fb); | |
3224 | ||
8518278a AG |
3225 | /* Now initialize UFFD context and start tracking RAM writes */ |
3226 | if (ram_write_tracking_start()) { | |
3227 | goto fail; | |
3228 | } | |
3229 | early_fail = false; | |
3230 | ||
3231 | /* | |
3232 | * Start VM from BH handler to avoid write-fault lock here. | |
3233 | * UFFD-WP protection for the whole RAM is already enabled so | |
3234 | * calling VM state change notifiers from vm_start() would initiate | |
3235 | * writes to virtio VQs memory which is in write-protected region. | |
3236 | */ | |
3237 | s->vm_start_bh = qemu_bh_new(bg_migration_vm_start_bh, s); | |
3238 | qemu_bh_schedule(s->vm_start_bh); | |
3239 | ||
3240 | qemu_mutex_unlock_iothread(); | |
3241 | ||
3242 | while (migration_is_active(s)) { | |
3243 | MigIterateState iter_state = bg_migration_iteration_run(s); | |
3244 | if (iter_state == MIG_ITERATE_SKIP) { | |
3245 | continue; | |
3246 | } else if (iter_state == MIG_ITERATE_BREAK) { | |
3247 | break; | |
3248 | } | |
3249 | ||
3250 | /* | |
3251 | * Try to detect any kind of failures, and see whether we | |
3252 | * should stop the migration now. | |
3253 | */ | |
3254 | thr_error = migration_detect_error(s); | |
3255 | if (thr_error == MIG_THR_ERR_FATAL) { | |
3256 | /* Stop migration */ | |
3257 | break; | |
3258 | } | |
3259 | ||
3260 | migration_update_counters(s, qemu_clock_get_ms(QEMU_CLOCK_REALTIME)); | |
3261 | } | |
3262 | ||
3263 | trace_migration_thread_after_loop(); | |
3264 | ||
3265 | fail: | |
3266 | if (early_fail) { | |
3267 | migrate_set_state(&s->state, MIGRATION_STATUS_ACTIVE, | |
3268 | MIGRATION_STATUS_FAILED); | |
3269 | qemu_mutex_unlock_iothread(); | |
3270 | } | |
3271 | ||
3272 | bg_migration_iteration_finish(s); | |
3273 | ||
3274 | qemu_fclose(fb); | |
3275 | object_unref(OBJECT(s)); | |
3276 | rcu_unregister_thread(); | |
3277 | ||
3278 | return NULL; | |
3279 | } | |
3280 | ||
cce8040b | 3281 | void migrate_fd_connect(MigrationState *s, Error *error_in) |
0d82d0e8 | 3282 | { |
00f4b572 | 3283 | Error *local_err = NULL; |
52033349 | 3284 | uint64_t rate_limit; |
d3e35b8f PX |
3285 | bool resume = s->state == MIGRATION_STATUS_POSTCOPY_PAUSED; |
3286 | ||
ca7bd082 PX |
3287 | /* |
3288 | * If there's a previous error, free it and prepare for another one. | |
3289 | * Meanwhile if migration completes successfully, there won't have an error | |
3290 | * dumped when calling migrate_fd_cleanup(). | |
3291 | */ | |
3292 | migrate_error_free(s); | |
3293 | ||
f5da8ba4 | 3294 | s->expected_downtime = migrate_downtime_limit(); |
9cbc3649 MAL |
3295 | if (resume) { |
3296 | assert(s->cleanup_bh); | |
3297 | } else { | |
3298 | assert(!s->cleanup_bh); | |
3299 | s->cleanup_bh = qemu_bh_new(migrate_fd_cleanup_bh, s); | |
3300 | } | |
cce8040b DDAG |
3301 | if (error_in) { |
3302 | migrate_fd_error(s, error_in); | |
ca30f24d PX |
3303 | if (resume) { |
3304 | /* | |
3305 | * Don't do cleanup for resume if channel is invalid, but only dump | |
3306 | * the error. We wait for another channel connect from the user. | |
3307 | * The error_report still gives HMP user a hint on what failed. | |
3308 | * It's normally done in migrate_fd_cleanup(), but call it here | |
3309 | * explicitly. | |
3310 | */ | |
3311 | error_report_err(error_copy(s->error)); | |
3312 | } else { | |
3313 | migrate_fd_cleanup(s); | |
3314 | } | |
cce8040b DDAG |
3315 | return; |
3316 | } | |
0d82d0e8 | 3317 | |
d3e35b8f PX |
3318 | if (resume) { |
3319 | /* This is a resumed migration */ | |
9d3ebbe2 | 3320 | rate_limit = migrate_max_postcopy_bandwidth(); |
d3e35b8f PX |
3321 | } else { |
3322 | /* This is a fresh new migration */ | |
9d3ebbe2 | 3323 | rate_limit = migrate_max_bandwidth(); |
442773ce | 3324 | |
d3e35b8f | 3325 | /* Notify before starting migration thread */ |
d9cda213 | 3326 | migration_call_notifiers(s); |
d3e35b8f PX |
3327 | } |
3328 | ||
e1fde0e0 | 3329 | migration_rate_set(rate_limit); |
d3e35b8f | 3330 | qemu_file_set_blocking(s->to_dst_file, true); |
9287ac27 | 3331 | |
1d34e4bf | 3332 | /* |
c788ada8 PX |
3333 | * Open the return path. For postcopy, it is used exclusively. For |
3334 | * precopy, only if user specified "return-path" capability would | |
3335 | * QEMU uses the return path. | |
1d34e4bf | 3336 | */ |
38ad1110 | 3337 | if (migrate_postcopy_ram() || migrate_return_path()) { |
ef796ee9 | 3338 | if (open_return_path_on_source(s)) { |
908927db | 3339 | error_setg(&local_err, "Unable to open return-path for postcopy"); |
d3e35b8f | 3340 | migrate_set_state(&s->state, s->state, MIGRATION_STATUS_FAILED); |
908927db TG |
3341 | migrate_set_error(s, local_err); |
3342 | error_report_err(local_err); | |
1d34e4bf DDAG |
3343 | migrate_fd_cleanup(s); |
3344 | return; | |
3345 | } | |
3346 | } | |
3347 | ||
06064a67 PX |
3348 | /* |
3349 | * This needs to be done before resuming a postcopy. Note: for newer | |
3350 | * QEMUs we will delay the channel creation until postcopy_start(), to | |
3351 | * avoid disorder of channel creations. | |
3352 | */ | |
3353 | if (migrate_postcopy_preempt() && s->preempt_pre_7_2) { | |
3354 | postcopy_preempt_setup(s); | |
3355 | } | |
3356 | ||
d3e35b8f | 3357 | if (resume) { |
135b87b4 PX |
3358 | /* Wakeup the main migration thread to do the recovery */ |
3359 | migrate_set_state(&s->state, MIGRATION_STATUS_POSTCOPY_PAUSED, | |
3360 | MIGRATION_STATUS_POSTCOPY_RECOVER); | |
3361 | qemu_sem_post(&s->postcopy_pause_sem); | |
d3e35b8f PX |
3362 | return; |
3363 | } | |
3364 | ||
00f4b572 | 3365 | if (multifd_save_setup(&local_err) != 0) { |
908927db | 3366 | migrate_set_error(s, local_err); |
00f4b572 | 3367 | error_report_err(local_err); |
f986c3d2 JQ |
3368 | migrate_set_state(&s->state, MIGRATION_STATUS_SETUP, |
3369 | MIGRATION_STATUS_FAILED); | |
3370 | migrate_fd_cleanup(s); | |
3371 | return; | |
3372 | } | |
8518278a AG |
3373 | |
3374 | if (migrate_background_snapshot()) { | |
3375 | qemu_thread_create(&s->thread, "bg_snapshot", | |
3376 | bg_migration_thread, s, QEMU_THREAD_JOINABLE); | |
3377 | } else { | |
3378 | qemu_thread_create(&s->thread, "live_migration", | |
3379 | migration_thread, s, QEMU_THREAD_JOINABLE); | |
3380 | } | |
1d34e4bf | 3381 | s->migration_thread_running = true; |
0d82d0e8 | 3382 | } |
093e3c42 | 3383 | |
e5cb7e76 PX |
3384 | static void migration_class_init(ObjectClass *klass, void *data) |
3385 | { | |
3386 | DeviceClass *dc = DEVICE_CLASS(klass); | |
3387 | ||
3388 | dc->user_creatable = false; | |
4f67d30b | 3389 | device_class_set_props(dc, migration_properties); |
e5cb7e76 PX |
3390 | } |
3391 | ||
b91bf5e4 MAL |
3392 | static void migration_instance_finalize(Object *obj) |
3393 | { | |
3394 | MigrationState *ms = MIGRATION_OBJ(obj); | |
b91bf5e4 | 3395 | |
87db1a7d | 3396 | qemu_mutex_destroy(&ms->error_mutex); |
62df066f | 3397 | qemu_mutex_destroy(&ms->qemu_file_lock); |
c7e0acd5 | 3398 | qemu_sem_destroy(&ms->wait_unplug_sem); |
ad767bed | 3399 | qemu_sem_destroy(&ms->rate_limit_sem); |
e91d8951 | 3400 | qemu_sem_destroy(&ms->pause_sem); |
b23c2ade | 3401 | qemu_sem_destroy(&ms->postcopy_pause_sem); |
edd090c7 | 3402 | qemu_sem_destroy(&ms->rp_state.rp_sem); |
b28fb582 | 3403 | qemu_sem_destroy(&ms->rp_state.rp_pong_acks); |
d0edb8a1 | 3404 | qemu_sem_destroy(&ms->postcopy_qemufile_src_sem); |
ab105cc1 | 3405 | error_free(ms->error); |
b91bf5e4 MAL |
3406 | } |
3407 | ||
e5cb7e76 PX |
3408 | static void migration_instance_init(Object *obj) |
3409 | { | |
3410 | MigrationState *ms = MIGRATION_OBJ(obj); | |
3411 | ||
3412 | ms->state = MIGRATION_STATUS_NONE; | |
e5cb7e76 | 3413 | ms->mbps = -1; |
aecbfe9c | 3414 | ms->pages_per_second = -1; |
e91d8951 | 3415 | qemu_sem_init(&ms->pause_sem, 0); |
87db1a7d | 3416 | qemu_mutex_init(&ms->error_mutex); |
8b0b29dc | 3417 | |
61a174e2 | 3418 | migrate_params_init(&ms->parameters); |
b23c2ade PX |
3419 | |
3420 | qemu_sem_init(&ms->postcopy_pause_sem, 0); | |
edd090c7 | 3421 | qemu_sem_init(&ms->rp_state.rp_sem, 0); |
b28fb582 | 3422 | qemu_sem_init(&ms->rp_state.rp_pong_acks, 0); |
ad767bed | 3423 | qemu_sem_init(&ms->rate_limit_sem, 0); |
c7e0acd5 | 3424 | qemu_sem_init(&ms->wait_unplug_sem, 0); |
d0edb8a1 | 3425 | qemu_sem_init(&ms->postcopy_qemufile_src_sem, 0); |
62df066f | 3426 | qemu_mutex_init(&ms->qemu_file_lock); |
8b0b29dc PX |
3427 | } |
3428 | ||
3429 | /* | |
3430 | * Return true if check pass, false otherwise. Error will be put | |
3431 | * inside errp if provided. | |
3432 | */ | |
3433 | static bool migration_object_check(MigrationState *ms, Error **errp) | |
3434 | { | |
6b19a7d9 | 3435 | /* Assuming all off */ |
b02c7fc9 | 3436 | bool old_caps[MIGRATION_CAPABILITY__MAX] = { 0 }; |
6b19a7d9 | 3437 | |
8b0b29dc PX |
3438 | if (!migrate_params_check(&ms->parameters, errp)) { |
3439 | return false; | |
3440 | } | |
3441 | ||
b02c7fc9 | 3442 | return migrate_caps_check(old_caps, ms->capabilities, errp); |
e5cb7e76 PX |
3443 | } |
3444 | ||
3445 | static const TypeInfo migration_type = { | |
3446 | .name = TYPE_MIGRATION, | |
01f6e14c | 3447 | /* |
c8d3ff38 | 3448 | * NOTE: TYPE_MIGRATION is not really a device, as the object is |
2194abd6 | 3449 | * not created using qdev_new(), it is not attached to the qdev |
c8d3ff38 PX |
3450 | * device tree, and it is never realized. |
3451 | * | |
3452 | * TODO: Make this TYPE_OBJECT once QOM provides something like | |
3453 | * TYPE_DEVICE's "-global" properties. | |
01f6e14c | 3454 | */ |
e5cb7e76 PX |
3455 | .parent = TYPE_DEVICE, |
3456 | .class_init = migration_class_init, | |
3457 | .class_size = sizeof(MigrationClass), | |
3458 | .instance_size = sizeof(MigrationState), | |
3459 | .instance_init = migration_instance_init, | |
b91bf5e4 | 3460 | .instance_finalize = migration_instance_finalize, |
e5cb7e76 PX |
3461 | }; |
3462 | ||
3463 | static void register_migration_types(void) | |
3464 | { | |
3465 | type_register_static(&migration_type); | |
3466 | } | |
3467 | ||
3468 | type_init(register_migration_types); |