]>
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" |
61e8b148 | 23 | #include "socket.h" |
54d31236 | 24 | #include "sysemu/runstate.h" |
46517dd4 | 25 | #include "sysemu/sysemu.h" |
b0c3cf94 | 26 | #include "sysemu/cpu-throttle.h" |
e1a3ecee | 27 | #include "rdma.h" |
7b1e1a22 | 28 | #include "ram.h" |
84a899de | 29 | #include "migration/global_state.h" |
c4b63b7c | 30 | #include "migration/misc.h" |
6666c96a | 31 | #include "migration.h" |
20a519a0 | 32 | #include "savevm.h" |
08a0aee1 | 33 | #include "qemu-file.h" |
6720c2b3 | 34 | #include "channel.h" |
987772d9 | 35 | #include "migration/vmstate.h" |
737e150e | 36 | #include "block/block.h" |
e688df6b | 37 | #include "qapi/error.h" |
9aca82ba | 38 | #include "qapi/clone-visitor.h" |
31e4c354 | 39 | #include "qapi/qapi-visit-migration.h" |
9aca82ba | 40 | #include "qapi/qapi-visit-sockets.h" |
9af23989 MA |
41 | #include "qapi/qapi-commands-migration.h" |
42 | #include "qapi/qapi-events-migration.h" | |
cc7a8ea7 | 43 | #include "qapi/qmp/qerror.h" |
15280c36 | 44 | #include "qapi/qmp/qnull.h" |
ab28bd23 | 45 | #include "qemu/rcu.h" |
2c9e6fec | 46 | #include "block.h" |
be07b0ac | 47 | #include "postcopy-ram.h" |
766bd176 | 48 | #include "qemu/thread.h" |
c09e5bb1 | 49 | #include "trace.h" |
51180423 | 50 | #include "exec/target_page.h" |
61b67d47 | 51 | #include "io/channel-buffer.h" |
85a8578e | 52 | #include "io/channel-tls.h" |
35a6ed4f | 53 | #include "migration/colo.h" |
4ffdb337 | 54 | #include "hw/boards.h" |
a27bd6c7 | 55 | #include "hw/qdev-properties.h" |
ce35e229 | 56 | #include "hw/qdev-properties-system.h" |
9d18af93 | 57 | #include "monitor/monitor.h" |
50510ea2 | 58 | #include "net/announce.h" |
c7e0acd5 | 59 | #include "qemu/queue.h" |
d32ca5ad | 60 | #include "multifd.h" |
1b1f4ab6 | 61 | #include "threadinfo.h" |
b5eea99e | 62 | #include "qemu/yank.h" |
6e8c25b4 | 63 | #include "sysemu/cpus.h" |
39675fff | 64 | #include "yank_functions.h" |
1b529d90 | 65 | #include "sysemu/qtest.h" |
1f0776f1 | 66 | #include "options.h" |
065e2813 | 67 | |
7590a2ae | 68 | #define MAX_THROTTLE (128 << 20) /* Migration transfer speed throttling */ |
5bb7910a | 69 | |
5b4e1eb7 JQ |
70 | /* Amount of time to allocate to each "chunk" of bandwidth-throttled |
71 | * data. */ | |
72 | #define BUFFER_DELAY 100 | |
73 | #define XFER_LIMIT_RATIO (1000 / BUFFER_DELAY) | |
74 | ||
2ff30257 AA |
75 | /* Time in milliseconds we are allowed to stop the source, |
76 | * for sending the last part */ | |
77 | #define DEFAULT_MIGRATE_SET_DOWNTIME 300 | |
78 | ||
87c9cc1c DHB |
79 | /* Maximum migrate downtime set to 2000 seconds */ |
80 | #define MAX_MIGRATE_DOWNTIME_SECONDS 2000 | |
81 | #define MAX_MIGRATE_DOWNTIME (MAX_MIGRATE_DOWNTIME_SECONDS * 1000) | |
82 | ||
8706d2d5 LL |
83 | /* Default compression thread count */ |
84 | #define DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT 8 | |
3fcb38c2 LL |
85 | /* Default decompression thread count, usually decompression is at |
86 | * least 4 times as fast as compression.*/ | |
87 | #define DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT 2 | |
8706d2d5 LL |
88 | /*0: means nocompress, 1: best speed, ... 9: best compress ratio */ |
89 | #define DEFAULT_MIGRATE_COMPRESS_LEVEL 1 | |
1626fee3 | 90 | /* Define default autoconverge cpu throttle migration parameters */ |
dc14a470 | 91 | #define DEFAULT_MIGRATE_THROTTLE_TRIGGER_THRESHOLD 50 |
d85a31d1 JH |
92 | #define DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL 20 |
93 | #define DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT 10 | |
4cbc9c7f | 94 | #define DEFAULT_MIGRATE_MAX_CPU_THROTTLE 99 |
8706d2d5 | 95 | |
17ad9b35 | 96 | /* Migration XBZRLE default cache size */ |
73af8dd8 | 97 | #define DEFAULT_MIGRATE_XBZRLE_CACHE_SIZE (64 * 1024 * 1024) |
17ad9b35 | 98 | |
131b2153 ZC |
99 | /* The delay time (in ms) between two COLO checkpoints */ |
100 | #define DEFAULT_MIGRATE_X_CHECKPOINT_DELAY (200 * 100) | |
4075fb1c | 101 | #define DEFAULT_MIGRATE_MULTIFD_CHANNELS 2 |
96eef042 | 102 | #define DEFAULT_MIGRATE_MULTIFD_COMPRESSION MULTIFD_COMPRESSION_NONE |
9004db48 JQ |
103 | /* 0: means nocompress, 1: best speed, ... 9: best compress ratio */ |
104 | #define DEFAULT_MIGRATE_MULTIFD_ZLIB_LEVEL 1 | |
6a9ad154 JQ |
105 | /* 0: means nocompress, 1: best speed, ... 20: best compress ratio */ |
106 | #define DEFAULT_MIGRATE_MULTIFD_ZSTD_LEVEL 1 | |
68b53591 | 107 | |
7e555c6c DDAG |
108 | /* Background transfer rate for postcopy, 0 means unlimited, note |
109 | * that page requests can still exceed this limit. | |
110 | */ | |
111 | #define DEFAULT_MIGRATE_MAX_POSTCOPY_BANDWIDTH 0 | |
112 | ||
ee3d96ba DDAG |
113 | /* |
114 | * Parameters for self_announce_delay giving a stream of RARP/ARP | |
115 | * packets after migration. | |
116 | */ | |
117 | #define DEFAULT_MIGRATE_ANNOUNCE_INITIAL 50 | |
118 | #define DEFAULT_MIGRATE_ANNOUNCE_MAX 550 | |
119 | #define DEFAULT_MIGRATE_ANNOUNCE_ROUNDS 5 | |
120 | #define DEFAULT_MIGRATE_ANNOUNCE_STEP 100 | |
121 | ||
99a0db9b GH |
122 | static NotifierList migration_state_notifiers = |
123 | NOTIFIER_LIST_INITIALIZER(migration_state_notifiers); | |
124 | ||
da6f1790 JQ |
125 | /* Messages sent on the return path from destination to source */ |
126 | enum mig_rp_message_type { | |
127 | MIG_RP_MSG_INVALID = 0, /* Must be 0 */ | |
128 | MIG_RP_MSG_SHUT, /* sibling will not send any more RP messages */ | |
129 | MIG_RP_MSG_PONG, /* Response to a PING; data (seq: be32 ) */ | |
130 | ||
131 | MIG_RP_MSG_REQ_PAGES_ID, /* data (start: be64, len: be32, id: string) */ | |
132 | MIG_RP_MSG_REQ_PAGES, /* data (start: be64, len: be32) */ | |
a335debb | 133 | MIG_RP_MSG_RECV_BITMAP, /* send recved_bitmap back to source */ |
13955b89 | 134 | MIG_RP_MSG_RESUME_ACK, /* tell source that we are ready to resume */ |
da6f1790 JQ |
135 | |
136 | MIG_RP_MSG_MAX | |
137 | }; | |
138 | ||
17549e84 JQ |
139 | /* When we add fault tolerance, we could have several |
140 | migrations at once. For now we don't need to add | |
141 | dynamic creation of migration */ | |
142 | ||
e5cb7e76 | 143 | static MigrationState *current_migration; |
e1b1b1bc | 144 | static MigrationIncomingState *current_incoming; |
e5cb7e76 | 145 | |
3af8554b DDAG |
146 | static GSList *migration_blockers; |
147 | ||
8b0b29dc | 148 | static bool migration_object_check(MigrationState *ms, Error **errp); |
0331c8ca DDAG |
149 | static int migration_maybe_pause(MigrationState *s, |
150 | int *current_active_state, | |
151 | int new_state); | |
892ae715 | 152 | static void migrate_fd_cancel(MigrationState *s); |
8b0b29dc | 153 | |
d6f74fd1 PX |
154 | static bool migration_needs_multiple_sockets(void) |
155 | { | |
51b07548 | 156 | return migrate_multifd() || migrate_postcopy_preempt(); |
d6f74fd1 | 157 | } |
f444eeda | 158 | |
d6f74fd1 | 159 | static bool uri_supports_multi_channels(const char *uri) |
f444eeda | 160 | { |
d6f74fd1 PX |
161 | return strstart(uri, "tcp:", NULL) || strstart(uri, "unix:", NULL) || |
162 | strstart(uri, "vsock:", NULL); | |
f444eeda PX |
163 | } |
164 | ||
d6f74fd1 PX |
165 | static bool |
166 | migration_channels_and_uri_compatible(const char *uri, Error **errp) | |
f444eeda | 167 | { |
d6f74fd1 PX |
168 | if (migration_needs_multiple_sockets() && |
169 | !uri_supports_multi_channels(uri)) { | |
170 | error_setg(errp, "Migration requires multi-channel URIs (e.g. tcp)"); | |
171 | return false; | |
172 | } | |
173 | ||
174 | return true; | |
f444eeda PX |
175 | } |
176 | ||
8f8bfffc PX |
177 | static gint page_request_addr_cmp(gconstpointer ap, gconstpointer bp) |
178 | { | |
179 | uintptr_t a = (uintptr_t) ap, b = (uintptr_t) bp; | |
180 | ||
181 | return (a > b) - (a < b); | |
182 | } | |
183 | ||
e5cb7e76 PX |
184 | void migration_object_init(void) |
185 | { | |
186 | /* This can only be called once. */ | |
187 | assert(!current_migration); | |
188 | current_migration = MIGRATION_OBJ(object_new(TYPE_MIGRATION)); | |
4ffdb337 | 189 | |
e1b1b1bc PX |
190 | /* |
191 | * Init the migrate incoming object as well no matter whether | |
192 | * we'll use it or not. | |
193 | */ | |
194 | assert(!current_incoming); | |
195 | current_incoming = g_new0(MigrationIncomingState, 1); | |
196 | current_incoming->state = MIGRATION_STATUS_NONE; | |
197 | current_incoming->postcopy_remote_fds = | |
198 | g_array_new(FALSE, TRUE, sizeof(struct PostCopyFD)); | |
199 | qemu_mutex_init(¤t_incoming->rp_mutex); | |
60bb3c58 | 200 | qemu_mutex_init(¤t_incoming->postcopy_prio_thread_mutex); |
e1b1b1bc PX |
201 | qemu_event_init(¤t_incoming->main_thread_load_event, false); |
202 | qemu_sem_init(¤t_incoming->postcopy_pause_sem_dst, 0); | |
203 | qemu_sem_init(¤t_incoming->postcopy_pause_sem_fault, 0); | |
60bb3c58 | 204 | qemu_sem_init(¤t_incoming->postcopy_pause_sem_fast_load, 0); |
5655aab0 PX |
205 | qemu_sem_init(¤t_incoming->postcopy_qemufile_dst_done, 0); |
206 | ||
8f8bfffc PX |
207 | qemu_mutex_init(¤t_incoming->page_request_mutex); |
208 | current_incoming->page_requested = g_tree_new(page_request_addr_cmp); | |
e1b1b1bc | 209 | |
f9734d5d | 210 | migration_object_check(current_migration, &error_fatal); |
e0d17dfd PB |
211 | |
212 | blk_mig_init(); | |
213 | ram_mig_init(); | |
214 | dirty_bitmap_mig_init(); | |
e5cb7e76 PX |
215 | } |
216 | ||
458fecca | 217 | void migration_cancel(const Error *error) |
c7c0e724 | 218 | { |
458fecca LV |
219 | if (error) { |
220 | migrate_set_error(current_migration, error); | |
221 | } | |
c7c0e724 DH |
222 | migrate_fd_cancel(current_migration); |
223 | } | |
224 | ||
892ae715 | 225 | void migration_shutdown(void) |
1f895604 | 226 | { |
795969ab RL |
227 | /* |
228 | * When the QEMU main thread exit, the COLO thread | |
229 | * may wait a semaphore. So, we should wakeup the | |
230 | * COLO thread before migration shutdown. | |
231 | */ | |
232 | colo_shutdown(); | |
892ae715 DDAG |
233 | /* |
234 | * Cancel the current migration - that will (eventually) | |
235 | * stop the migration using this structure | |
236 | */ | |
458fecca | 237 | migration_cancel(NULL); |
1f895604 | 238 | object_unref(OBJECT(current_migration)); |
1499ab09 VSO |
239 | |
240 | /* | |
241 | * Cancel outgoing migration of dirty bitmaps. It should | |
242 | * at least unref used block nodes. | |
243 | */ | |
244 | dirty_bitmap_mig_cancel_outgoing(); | |
245 | ||
246 | /* | |
247 | * Cancel incoming migration of dirty bitmaps. Dirty bitmaps | |
248 | * are non-critical data, and their loss never considered as | |
249 | * something serious. | |
250 | */ | |
251 | dirty_bitmap_mig_cancel_incoming(); | |
1f895604 VSO |
252 | } |
253 | ||
bca7856a | 254 | /* For outgoing */ |
859bc756 | 255 | MigrationState *migrate_get_current(void) |
17549e84 | 256 | { |
e5cb7e76 PX |
257 | /* This can only be called after the object created. */ |
258 | assert(current_migration); | |
259 | return current_migration; | |
17549e84 JQ |
260 | } |
261 | ||
bca7856a DDAG |
262 | MigrationIncomingState *migration_incoming_get_current(void) |
263 | { | |
e1b1b1bc PX |
264 | assert(current_incoming); |
265 | return current_incoming; | |
bca7856a DDAG |
266 | } |
267 | ||
e031149c PX |
268 | void migration_incoming_transport_cleanup(MigrationIncomingState *mis) |
269 | { | |
270 | if (mis->socket_address_list) { | |
271 | qapi_free_SocketAddressList(mis->socket_address_list); | |
272 | mis->socket_address_list = NULL; | |
273 | } | |
274 | ||
275 | if (mis->transport_cleanup) { | |
276 | mis->transport_cleanup(mis->transport_data); | |
277 | mis->transport_data = mis->transport_cleanup = NULL; | |
278 | } | |
279 | } | |
280 | ||
bca7856a DDAG |
281 | void migration_incoming_state_destroy(void) |
282 | { | |
b4b076da JQ |
283 | struct MigrationIncomingState *mis = migration_incoming_get_current(); |
284 | ||
cfc3bcf3 LB |
285 | multifd_load_cleanup(); |
286 | ||
3482655b | 287 | if (mis->to_src_file) { |
660819b1 PX |
288 | /* Tell source that we are done */ |
289 | migrate_send_rp_shut(mis, qemu_file_get_error(mis->from_src_file) != 0); | |
3482655b PX |
290 | qemu_fclose(mis->to_src_file); |
291 | mis->to_src_file = NULL; | |
292 | } | |
293 | ||
660819b1 | 294 | if (mis->from_src_file) { |
39675fff | 295 | migration_ioc_unregister_yank_from_file(mis->from_src_file); |
660819b1 PX |
296 | qemu_fclose(mis->from_src_file); |
297 | mis->from_src_file = NULL; | |
298 | } | |
00fa4fc8 DDAG |
299 | if (mis->postcopy_remote_fds) { |
300 | g_array_free(mis->postcopy_remote_fds, TRUE); | |
301 | mis->postcopy_remote_fds = NULL; | |
302 | } | |
660819b1 | 303 | |
e031149c | 304 | migration_incoming_transport_cleanup(mis); |
1783c00f DDAG |
305 | qemu_event_reset(&mis->main_thread_load_event); |
306 | ||
8f8bfffc PX |
307 | if (mis->page_requested) { |
308 | g_tree_destroy(mis->page_requested); | |
309 | mis->page_requested = NULL; | |
310 | } | |
311 | ||
36f62f11 PX |
312 | if (mis->postcopy_qemufile_dst) { |
313 | migration_ioc_unregister_yank_from_file(mis->postcopy_qemufile_dst); | |
314 | qemu_fclose(mis->postcopy_qemufile_dst); | |
315 | mis->postcopy_qemufile_dst = NULL; | |
316 | } | |
317 | ||
b5eea99e | 318 | yank_unregister_instance(MIGRATION_YANK_INSTANCE); |
bca7856a DDAG |
319 | } |
320 | ||
b05dc723 JQ |
321 | static void migrate_generate_event(int new_state) |
322 | { | |
b890902c | 323 | if (migrate_events()) { |
3ab72385 | 324 | qapi_event_send_migration(new_state); |
b05dc723 JQ |
325 | } |
326 | } | |
327 | ||
da6f1790 JQ |
328 | /* |
329 | * Send a message on the return channel back to the source | |
330 | * of the migration. | |
331 | */ | |
d6208e35 PX |
332 | static int migrate_send_rp_message(MigrationIncomingState *mis, |
333 | enum mig_rp_message_type message_type, | |
334 | uint16_t len, void *data) | |
da6f1790 | 335 | { |
d6208e35 PX |
336 | int ret = 0; |
337 | ||
da6f1790 | 338 | trace_migrate_send_rp_message((int)message_type, len); |
37396950 | 339 | QEMU_LOCK_GUARD(&mis->rp_mutex); |
d6208e35 PX |
340 | |
341 | /* | |
342 | * It's possible that the file handle got lost due to network | |
343 | * failures. | |
344 | */ | |
345 | if (!mis->to_src_file) { | |
346 | ret = -EIO; | |
37396950 | 347 | return ret; |
d6208e35 PX |
348 | } |
349 | ||
da6f1790 JQ |
350 | qemu_put_be16(mis->to_src_file, (unsigned int)message_type); |
351 | qemu_put_be16(mis->to_src_file, len); | |
352 | qemu_put_buffer(mis->to_src_file, data, len); | |
353 | qemu_fflush(mis->to_src_file); | |
d6208e35 PX |
354 | |
355 | /* It's possible that qemu file got error during sending */ | |
356 | ret = qemu_file_get_error(mis->to_src_file); | |
357 | ||
d6208e35 | 358 | return ret; |
da6f1790 JQ |
359 | } |
360 | ||
2e2bce16 PX |
361 | /* Request one page from the source VM at the given start address. |
362 | * rb: the RAMBlock to request the page in | |
1e2d90eb DDAG |
363 | * Start: Address offset within the RB |
364 | * Len: Length in bytes required - must be a multiple of pagesize | |
365 | */ | |
7a267fc4 PX |
366 | int migrate_send_rp_message_req_pages(MigrationIncomingState *mis, |
367 | RAMBlock *rb, ram_addr_t start) | |
1e2d90eb | 368 | { |
cb8d4c8f | 369 | uint8_t bufc[12 + 1 + 255]; /* start (8), len (4), rbname up to 256 */ |
1e2d90eb | 370 | size_t msglen = 12; /* start + len */ |
2e2bce16 | 371 | size_t len = qemu_ram_pagesize(rb); |
d6208e35 | 372 | enum mig_rp_message_type msg_type; |
2e2bce16 PX |
373 | const char *rbname; |
374 | int rbname_len; | |
1e2d90eb DDAG |
375 | |
376 | *(uint64_t *)bufc = cpu_to_be64((uint64_t)start); | |
377 | *(uint32_t *)(bufc + 8) = cpu_to_be32((uint32_t)len); | |
378 | ||
2e2bce16 PX |
379 | /* |
380 | * We maintain the last ramblock that we requested for page. Note that we | |
381 | * don't need locking because this function will only be called within the | |
382 | * postcopy ram fault thread. | |
383 | */ | |
384 | if (rb != mis->last_rb) { | |
385 | mis->last_rb = rb; | |
386 | ||
387 | rbname = qemu_ram_get_idstr(rb); | |
388 | rbname_len = strlen(rbname); | |
389 | ||
1e2d90eb DDAG |
390 | assert(rbname_len < 256); |
391 | ||
392 | bufc[msglen++] = rbname_len; | |
393 | memcpy(bufc + msglen, rbname, rbname_len); | |
394 | msglen += rbname_len; | |
d6208e35 | 395 | msg_type = MIG_RP_MSG_REQ_PAGES_ID; |
1e2d90eb | 396 | } else { |
d6208e35 | 397 | msg_type = MIG_RP_MSG_REQ_PAGES; |
1e2d90eb | 398 | } |
d6208e35 PX |
399 | |
400 | return migrate_send_rp_message(mis, msg_type, msglen, bufc); | |
1e2d90eb DDAG |
401 | } |
402 | ||
7a267fc4 | 403 | int migrate_send_rp_req_pages(MigrationIncomingState *mis, |
8f8bfffc | 404 | RAMBlock *rb, ram_addr_t start, uint64_t haddr) |
7a267fc4 | 405 | { |
7648297d | 406 | void *aligned = (void *)(uintptr_t)ROUND_DOWN(haddr, qemu_ram_pagesize(rb)); |
a2429283 | 407 | bool received = false; |
8f8bfffc PX |
408 | |
409 | WITH_QEMU_LOCK_GUARD(&mis->page_request_mutex) { | |
410 | received = ramblock_recv_bitmap_test_byte_offset(rb, start); | |
411 | if (!received && !g_tree_lookup(mis->page_requested, aligned)) { | |
412 | /* | |
413 | * The page has not been received, and it's not yet in the page | |
414 | * request list. Queue it. Set the value of element to 1, so that | |
415 | * things like g_tree_lookup() will return TRUE (1) when found. | |
416 | */ | |
417 | g_tree_insert(mis->page_requested, aligned, (gpointer)1); | |
418 | mis->page_requested_count++; | |
419 | trace_postcopy_page_req_add(aligned, mis->page_requested_count); | |
420 | } | |
421 | } | |
422 | ||
423 | /* | |
424 | * If the page is there, skip sending the message. We don't even need the | |
425 | * lock because as long as the page arrived, it'll be there forever. | |
426 | */ | |
427 | if (received) { | |
428 | return 0; | |
429 | } | |
430 | ||
7a267fc4 PX |
431 | return migrate_send_rp_message_req_pages(mis, rb, start); |
432 | } | |
433 | ||
aad555c2 ZC |
434 | static bool migration_colo_enabled; |
435 | bool migration_incoming_colo_enabled(void) | |
436 | { | |
437 | return migration_colo_enabled; | |
438 | } | |
439 | ||
440 | void migration_incoming_disable_colo(void) | |
441 | { | |
18b1d3c9 | 442 | ram_block_discard_disable(false); |
aad555c2 ZC |
443 | migration_colo_enabled = false; |
444 | } | |
445 | ||
18b1d3c9 | 446 | int migration_incoming_enable_colo(void) |
aad555c2 | 447 | { |
18b1d3c9 DH |
448 | if (ram_block_discard_disable(true)) { |
449 | error_report("COLO: cannot disable RAM discard"); | |
450 | return -EBUSY; | |
451 | } | |
aad555c2 | 452 | migration_colo_enabled = true; |
18b1d3c9 | 453 | return 0; |
aad555c2 ZC |
454 | } |
455 | ||
9aca82ba JQ |
456 | void migrate_add_address(SocketAddress *address) |
457 | { | |
458 | MigrationIncomingState *mis = migration_incoming_get_current(); | |
9aca82ba | 459 | |
54aa3de7 EB |
460 | QAPI_LIST_PREPEND(mis->socket_address_list, |
461 | QAPI_CLONE(SocketAddress, address)); | |
9aca82ba JQ |
462 | } |
463 | ||
e69d50d6 | 464 | static void qemu_start_incoming_migration(const char *uri, Error **errp) |
5bb7910a | 465 | { |
d658f65c | 466 | const char *p = NULL; |
34c9dd8e | 467 | |
d6f74fd1 PX |
468 | /* URI is not suitable for migration? */ |
469 | if (!migration_channels_and_uri_compatible(uri, errp)) { | |
470 | return; | |
471 | } | |
472 | ||
3ab72385 | 473 | qapi_event_send_migration(MIGRATION_STATUS_SETUP); |
e69d50d6 PB |
474 | if (strstart(uri, "tcp:", &p) || |
475 | strstart(uri, "unix:", NULL) || | |
476 | strstart(uri, "vsock:", NULL)) { | |
d658f65c | 477 | socket_start_incoming_migration(p ? p : uri, errp); |
2da776db | 478 | #ifdef CONFIG_RDMA |
adde220a | 479 | } else if (strstart(uri, "rdma:", &p)) { |
2da776db MH |
480 | rdma_start_incoming_migration(p, errp); |
481 | #endif | |
adde220a | 482 | } else if (strstart(uri, "exec:", &p)) { |
43eaae28 | 483 | exec_start_incoming_migration(p, errp); |
adde220a | 484 | } else if (strstart(uri, "fd:", &p)) { |
43eaae28 | 485 | fd_start_incoming_migration(p, errp); |
adde220a | 486 | } else { |
312fd5f2 | 487 | error_setg(errp, "unknown migration protocol: %s", uri); |
8ca5e801 | 488 | } |
5bb7910a AL |
489 | } |
490 | ||
0aa6aefc DL |
491 | static void process_incoming_migration_bh(void *opaque) |
492 | { | |
493 | Error *local_err = NULL; | |
494 | MigrationIncomingState *mis = opaque; | |
495 | ||
0f073f44 DDAG |
496 | /* If capability late_block_activate is set: |
497 | * Only fire up the block code now if we're going to restart the | |
498 | * VM, else 'cont' will do it. | |
499 | * This causes file locking to happen; so we don't want it to happen | |
500 | * unless we really are starting the VM. | |
501 | */ | |
502 | if (!migrate_late_block_activate() || | |
503 | (autostart && (!global_state_received() || | |
504 | global_state_get_runstate() == RUN_STATE_RUNNING))) { | |
3b717194 | 505 | /* Make sure all file formats throw away their mutable metadata. |
0f073f44 | 506 | * If we get an error here, just don't restart the VM yet. */ |
3b717194 | 507 | bdrv_activate_all(&local_err); |
0f073f44 DDAG |
508 | if (local_err) { |
509 | error_report_err(local_err); | |
510 | local_err = NULL; | |
511 | autostart = false; | |
512 | } | |
d35ff5e6 KW |
513 | } |
514 | ||
0aa6aefc DL |
515 | /* |
516 | * This must happen after all error conditions are dealt with and | |
517 | * we're sure the VM is going to be running on this host. | |
518 | */ | |
7659505c | 519 | qemu_announce_self(&mis->announce_timer, migrate_announce_params()); |
0aa6aefc | 520 | |
cfc3bcf3 | 521 | multifd_load_shutdown(); |
0aa6aefc | 522 | |
b35ebdf0 VSO |
523 | dirty_bitmap_mig_before_vm_start(); |
524 | ||
0aa6aefc DL |
525 | if (!global_state_received() || |
526 | global_state_get_runstate() == RUN_STATE_RUNNING) { | |
527 | if (autostart) { | |
528 | vm_start(); | |
529 | } else { | |
530 | runstate_set(RUN_STATE_PAUSED); | |
531 | } | |
db009729 ZC |
532 | } else if (migration_incoming_colo_enabled()) { |
533 | migration_incoming_disable_colo(); | |
534 | vm_start(); | |
0aa6aefc DL |
535 | } else { |
536 | runstate_set(global_state_get_runstate()); | |
537 | } | |
0aa6aefc DL |
538 | /* |
539 | * This must happen after any state changes since as soon as an external | |
540 | * observer sees this event they might start to prod at the VM assuming | |
541 | * it's ready to use. | |
542 | */ | |
543 | migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE, | |
544 | MIGRATION_STATUS_COMPLETED); | |
545 | qemu_bh_delete(mis->bh); | |
546 | migration_incoming_state_destroy(); | |
547 | } | |
548 | ||
38e8f9af MAL |
549 | static void coroutine_fn |
550 | process_incoming_migration_co(void *opaque) | |
511c0231 | 551 | { |
b4b076da | 552 | MigrationIncomingState *mis = migration_incoming_get_current(); |
e9bef235 | 553 | PostcopyState ps; |
1c12e1f5 | 554 | int ret; |
8e48ac95 | 555 | Error *local_err = NULL; |
1c12e1f5 | 556 | |
4f0fae7f | 557 | assert(mis->from_src_file); |
92370989 | 558 | mis->migration_incoming_co = qemu_coroutine_self(); |
67f11b5c | 559 | mis->largest_page_size = qemu_ram_pagesize_largest(); |
093e3c42 | 560 | postcopy_state_set(POSTCOPY_INCOMING_NONE); |
93d7af6f HZ |
561 | migrate_set_state(&mis->state, MIGRATION_STATUS_NONE, |
562 | MIGRATION_STATUS_ACTIVE); | |
4f0fae7f | 563 | ret = qemu_loadvm_state(mis->from_src_file); |
bca7856a | 564 | |
e9bef235 DDAG |
565 | ps = postcopy_state_get(); |
566 | trace_process_incoming_migration_co_end(ret, ps); | |
567 | if (ps != POSTCOPY_INCOMING_NONE) { | |
568 | if (ps == POSTCOPY_INCOMING_ADVISE) { | |
569 | /* | |
570 | * Where a migration had postcopy enabled (and thus went to advise) | |
571 | * but managed to complete within the precopy period, we can use | |
572 | * the normal exit. | |
573 | */ | |
574 | postcopy_ram_incoming_cleanup(mis); | |
575 | } else if (ret >= 0) { | |
576 | /* | |
577 | * Postcopy was started, cleanup should happen at the end of the | |
578 | * postcopy thread. | |
579 | */ | |
580 | trace_process_incoming_migration_co_postcopy_end_main(); | |
581 | return; | |
582 | } | |
583 | /* Else if something went wrong then just fall out of the normal exit */ | |
584 | } | |
585 | ||
25d0c16f | 586 | /* we get COLO info, and know if we are in COLO mode */ |
aad555c2 | 587 | if (!ret && migration_incoming_colo_enabled()) { |
3b717194 EGE |
588 | /* Make sure all file formats throw away their mutable metadata */ |
589 | bdrv_activate_all(&local_err); | |
8e48ac95 | 590 | if (local_err) { |
8e48ac95 | 591 | error_report_err(local_err); |
6d99c2d4 | 592 | goto fail; |
8e48ac95 ZC |
593 | } |
594 | ||
25d0c16f HZ |
595 | qemu_thread_create(&mis->colo_incoming_thread, "COLO incoming", |
596 | colo_process_incoming_thread, mis, QEMU_THREAD_JOINABLE); | |
597 | mis->have_colo_incoming_thread = true; | |
598 | qemu_coroutine_yield(); | |
599 | ||
684bfd18 | 600 | qemu_mutex_unlock_iothread(); |
25d0c16f HZ |
601 | /* Wait checkpoint incoming thread exit before free resource */ |
602 | qemu_thread_join(&mis->colo_incoming_thread); | |
684bfd18 | 603 | qemu_mutex_lock_iothread(); |
13af18f2 ZC |
604 | /* We hold the global iothread lock, so it is safe here */ |
605 | colo_release_ram_cache(); | |
25d0c16f HZ |
606 | } |
607 | ||
1c12e1f5 | 608 | if (ret < 0) { |
db80face | 609 | error_report("load of migration failed: %s", strerror(-ret)); |
6d99c2d4 | 610 | goto fail; |
511c0231 | 611 | } |
0aa6aefc DL |
612 | mis->bh = qemu_bh_new(process_incoming_migration_bh, mis); |
613 | qemu_bh_schedule(mis->bh); | |
92370989 | 614 | mis->migration_incoming_co = NULL; |
6d99c2d4 FL |
615 | return; |
616 | fail: | |
617 | local_err = NULL; | |
618 | migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE, | |
619 | MIGRATION_STATUS_FAILED); | |
620 | qemu_fclose(mis->from_src_file); | |
e5bac1f5 LB |
621 | |
622 | multifd_load_cleanup(); | |
623 | ||
6d99c2d4 | 624 | exit(EXIT_FAILURE); |
511c0231 JQ |
625 | } |
626 | ||
b673eab4 | 627 | /** |
7d6f6933 | 628 | * migration_incoming_setup: Setup incoming migration |
b673eab4 JQ |
629 | * @f: file for main migration channel |
630 | * @errp: where to put errors | |
7d6f6933 MA |
631 | * |
632 | * Returns: %true on success, %false on error. | |
b673eab4 | 633 | */ |
7d6f6933 | 634 | static bool migration_incoming_setup(QEMUFile *f, Error **errp) |
82a4da79 | 635 | { |
4f0fae7f | 636 | MigrationIncomingState *mis = migration_incoming_get_current(); |
82a4da79 | 637 | |
4f0fae7f JQ |
638 | if (!mis->from_src_file) { |
639 | mis->from_src_file = f; | |
640 | } | |
06ad5135 | 641 | qemu_file_set_blocking(f, false); |
7d6f6933 | 642 | return true; |
e595a01a JQ |
643 | } |
644 | ||
36c2f8be | 645 | void migration_incoming_process(void) |
e595a01a JQ |
646 | { |
647 | Coroutine *co = qemu_coroutine_create(process_incoming_migration_co, NULL); | |
0b8b8753 | 648 | qemu_coroutine_enter(co); |
82a4da79 PB |
649 | } |
650 | ||
884835fa | 651 | /* Returns true if recovered from a paused migration, otherwise false */ |
a39e9339 | 652 | static bool postcopy_try_recover(void) |
e595a01a | 653 | { |
d96c9e8d PX |
654 | MigrationIncomingState *mis = migration_incoming_get_current(); |
655 | ||
656 | if (mis->state == MIGRATION_STATUS_POSTCOPY_PAUSED) { | |
657 | /* Resumed from a paused postcopy migration */ | |
658 | ||
a39e9339 PX |
659 | /* This should be set already in migration_incoming_setup() */ |
660 | assert(mis->from_src_file); | |
d96c9e8d | 661 | /* Postcopy has standalone thread to do vm load */ |
a39e9339 | 662 | qemu_file_set_blocking(mis->from_src_file, true); |
d96c9e8d PX |
663 | |
664 | /* Re-configure the return path */ | |
a39e9339 | 665 | mis->to_src_file = qemu_file_get_return_path(mis->from_src_file); |
d96c9e8d PX |
666 | |
667 | migrate_set_state(&mis->state, MIGRATION_STATUS_POSTCOPY_PAUSED, | |
668 | MIGRATION_STATUS_POSTCOPY_RECOVER); | |
669 | ||
670 | /* | |
671 | * Here, we only wake up the main loading thread (while the | |
60bb3c58 | 672 | * rest threads will still be waiting), so that we can receive |
d96c9e8d | 673 | * commands from source now, and answer it if needed. The |
60bb3c58 | 674 | * rest threads will be woken up afterwards until we are sure |
d96c9e8d PX |
675 | * that source is ready to reply to page requests. |
676 | */ | |
677 | qemu_sem_post(&mis->postcopy_pause_sem_dst); | |
884835fa PX |
678 | return true; |
679 | } | |
680 | ||
681 | return false; | |
682 | } | |
683 | ||
b673eab4 | 684 | void migration_fd_process_incoming(QEMUFile *f, Error **errp) |
884835fa | 685 | { |
a39e9339 | 686 | if (!migration_incoming_setup(f, errp)) { |
884835fa | 687 | return; |
d96c9e8d | 688 | } |
a39e9339 | 689 | if (postcopy_try_recover()) { |
b673eab4 JQ |
690 | return; |
691 | } | |
884835fa | 692 | migration_incoming_process(); |
e595a01a JQ |
693 | } |
694 | ||
5655aab0 PX |
695 | /* |
696 | * Returns true when we want to start a new incoming migration process, | |
697 | * false otherwise. | |
698 | */ | |
699 | static bool migration_should_start_incoming(bool main_channel) | |
700 | { | |
701 | /* Multifd doesn't start unless all channels are established */ | |
51b07548 | 702 | if (migrate_multifd()) { |
5655aab0 PX |
703 | return migration_has_all_channels(); |
704 | } | |
705 | ||
706 | /* Preempt channel only starts when the main channel is created */ | |
707 | if (migrate_postcopy_preempt()) { | |
708 | return main_channel; | |
709 | } | |
710 | ||
711 | /* | |
712 | * For all the rest types of migration, we should only reach here when | |
713 | * it's the main channel that's being created, and we should always | |
714 | * proceed with this channel. | |
715 | */ | |
716 | assert(main_channel); | |
717 | return true; | |
718 | } | |
719 | ||
49ed0d24 | 720 | void migration_ioc_process_incoming(QIOChannel *ioc, Error **errp) |
4f0fae7f JQ |
721 | { |
722 | MigrationIncomingState *mis = migration_incoming_get_current(); | |
b673eab4 | 723 | Error *local_err = NULL; |
36f62f11 | 724 | QEMUFile *f; |
6720c2b3 | 725 | bool default_channel = true; |
726 | uint32_t channel_magic = 0; | |
727 | int ret = 0; | |
4f0fae7f | 728 | |
51b07548 | 729 | if (migrate_multifd() && !migrate_postcopy_ram() && |
6720c2b3 | 730 | qio_channel_has_feature(ioc, QIO_CHANNEL_FEATURE_READ_MSG_PEEK)) { |
731 | /* | |
732 | * With multiple channels, it is possible that we receive channels | |
733 | * out of order on destination side, causing incorrect mapping of | |
734 | * source channels on destination side. Check channel MAGIC to | |
735 | * decide type of channel. Please note this is best effort, postcopy | |
736 | * preempt channel does not send any magic number so avoid it for | |
737 | * postcopy live migration. Also tls live migration already does | |
738 | * tls handshake while initializing main channel so with tls this | |
739 | * issue is not possible. | |
740 | */ | |
741 | ret = migration_channel_read_peek(ioc, (void *)&channel_magic, | |
742 | sizeof(channel_magic), &local_err); | |
743 | ||
744 | if (ret != 0) { | |
745 | error_propagate(errp, local_err); | |
746 | return; | |
747 | } | |
748 | ||
749 | default_channel = (channel_magic == cpu_to_be32(QEMU_VM_FILE_MAGIC)); | |
750 | } else { | |
751 | default_channel = !mis->from_src_file; | |
752 | } | |
753 | ||
754 | if (multifd_load_setup(errp) != 0) { | |
755 | error_setg(errp, "Failed to setup multifd channels"); | |
756 | return; | |
757 | } | |
758 | ||
759 | if (default_channel) { | |
36f62f11 | 760 | f = qemu_file_new_input(ioc); |
a429e7f4 | 761 | |
7d6f6933 | 762 | if (!migration_incoming_setup(f, errp)) { |
b673eab4 JQ |
763 | return; |
764 | } | |
a429e7f4 PX |
765 | } else { |
766 | /* Multiple connections */ | |
36f62f11 | 767 | assert(migration_needs_multiple_sockets()); |
51b07548 | 768 | if (migrate_multifd()) { |
6720c2b3 | 769 | multifd_recv_new_channel(ioc, &local_err); |
36f62f11 PX |
770 | } else { |
771 | assert(migrate_postcopy_preempt()); | |
772 | f = qemu_file_new_input(ioc); | |
6720c2b3 | 773 | postcopy_preempt_new_channel(mis, f); |
36f62f11 | 774 | } |
49ed0d24 FL |
775 | if (local_err) { |
776 | error_propagate(errp, local_err); | |
777 | return; | |
778 | } | |
4f0fae7f | 779 | } |
81e62053 | 780 | |
5655aab0 | 781 | if (migration_should_start_incoming(default_channel)) { |
a39e9339 PX |
782 | /* If it's a recovery, we're done */ |
783 | if (postcopy_try_recover()) { | |
784 | return; | |
785 | } | |
81e62053 PX |
786 | migration_incoming_process(); |
787 | } | |
4f0fae7f JQ |
788 | } |
789 | ||
428d8908 JQ |
790 | /** |
791 | * @migration_has_all_channels: We have received all channels that we need | |
792 | * | |
793 | * Returns true when we have got connections to all the channels that | |
794 | * we need for migration. | |
795 | */ | |
796 | bool migration_has_all_channels(void) | |
797 | { | |
ca273df3 | 798 | MigrationIncomingState *mis = migration_incoming_get_current(); |
62c1e0ca | 799 | |
36f62f11 PX |
800 | if (!mis->from_src_file) { |
801 | return false; | |
802 | } | |
62c1e0ca | 803 | |
51b07548 | 804 | if (migrate_multifd()) { |
36f62f11 PX |
805 | return multifd_recv_all_channels_created(); |
806 | } | |
807 | ||
808 | if (migrate_postcopy_preempt()) { | |
809 | return mis->postcopy_qemufile_dst != NULL; | |
810 | } | |
811 | ||
812 | return true; | |
428d8908 JQ |
813 | } |
814 | ||
6decec93 DDAG |
815 | /* |
816 | * Send a 'SHUT' message on the return channel with the given value | |
817 | * to indicate that we've finished with the RP. Non-0 value indicates | |
818 | * error. | |
819 | */ | |
820 | void migrate_send_rp_shut(MigrationIncomingState *mis, | |
821 | uint32_t value) | |
822 | { | |
823 | uint32_t buf; | |
824 | ||
825 | buf = cpu_to_be32(value); | |
826 | migrate_send_rp_message(mis, MIG_RP_MSG_SHUT, sizeof(buf), &buf); | |
827 | } | |
828 | ||
829 | /* | |
830 | * Send a 'PONG' message on the return channel with the given value | |
831 | * (normally in response to a 'PING') | |
832 | */ | |
833 | void migrate_send_rp_pong(MigrationIncomingState *mis, | |
834 | uint32_t value) | |
835 | { | |
836 | uint32_t buf; | |
837 | ||
838 | buf = cpu_to_be32(value); | |
839 | migrate_send_rp_message(mis, MIG_RP_MSG_PONG, sizeof(buf), &buf); | |
840 | } | |
841 | ||
a335debb PX |
842 | void migrate_send_rp_recv_bitmap(MigrationIncomingState *mis, |
843 | char *block_name) | |
844 | { | |
845 | char buf[512]; | |
846 | int len; | |
847 | int64_t res; | |
848 | ||
849 | /* | |
850 | * First, we send the header part. It contains only the len of | |
851 | * idstr, and the idstr itself. | |
852 | */ | |
853 | len = strlen(block_name); | |
854 | buf[0] = len; | |
855 | memcpy(buf + 1, block_name, len); | |
856 | ||
857 | if (mis->state != MIGRATION_STATUS_POSTCOPY_RECOVER) { | |
858 | error_report("%s: MSG_RP_RECV_BITMAP only used for recovery", | |
859 | __func__); | |
860 | return; | |
861 | } | |
862 | ||
863 | migrate_send_rp_message(mis, MIG_RP_MSG_RECV_BITMAP, len + 1, buf); | |
864 | ||
865 | /* | |
866 | * Next, we dump the received bitmap to the stream. | |
867 | * | |
868 | * TODO: currently we are safe since we are the only one that is | |
869 | * using the to_src_file handle (fault thread is still paused), | |
870 | * and it's ok even not taking the mutex. However the best way is | |
871 | * to take the lock before sending the message header, and release | |
872 | * the lock after sending the bitmap. | |
873 | */ | |
874 | qemu_mutex_lock(&mis->rp_mutex); | |
875 | res = ramblock_recv_bitmap_send(mis->to_src_file, block_name); | |
876 | qemu_mutex_unlock(&mis->rp_mutex); | |
877 | ||
878 | trace_migrate_send_rp_recv_bitmap(block_name, res); | |
879 | } | |
880 | ||
13955b89 PX |
881 | void migrate_send_rp_resume_ack(MigrationIncomingState *mis, uint32_t value) |
882 | { | |
883 | uint32_t buf; | |
884 | ||
885 | buf = cpu_to_be32(value); | |
886 | migrate_send_rp_message(mis, MIG_RP_MSG_RESUME_ACK, sizeof(buf), &buf); | |
887 | } | |
888 | ||
85de8323 LL |
889 | MigrationParameters *qmp_query_migrate_parameters(Error **errp) |
890 | { | |
891 | MigrationParameters *params; | |
892 | MigrationState *s = migrate_get_current(); | |
893 | ||
e87fae4c | 894 | /* TODO use QAPI_CLONE() instead of duplicating it inline */ |
85de8323 | 895 | params = g_malloc0(sizeof(*params)); |
de63ab61 | 896 | params->has_compress_level = true; |
2594f56d | 897 | params->compress_level = s->parameters.compress_level; |
de63ab61 | 898 | params->has_compress_threads = true; |
2594f56d | 899 | params->compress_threads = s->parameters.compress_threads; |
1d58872a XG |
900 | params->has_compress_wait_thread = true; |
901 | params->compress_wait_thread = s->parameters.compress_wait_thread; | |
de63ab61 | 902 | params->has_decompress_threads = true; |
2594f56d | 903 | params->decompress_threads = s->parameters.decompress_threads; |
dc14a470 KZ |
904 | params->has_throttle_trigger_threshold = true; |
905 | params->throttle_trigger_threshold = s->parameters.throttle_trigger_threshold; | |
de63ab61 | 906 | params->has_cpu_throttle_initial = true; |
2594f56d | 907 | params->cpu_throttle_initial = s->parameters.cpu_throttle_initial; |
de63ab61 | 908 | params->has_cpu_throttle_increment = true; |
2594f56d | 909 | params->cpu_throttle_increment = s->parameters.cpu_throttle_increment; |
cbbf8182 KZ |
910 | params->has_cpu_throttle_tailslow = true; |
911 | params->cpu_throttle_tailslow = s->parameters.cpu_throttle_tailslow; | |
69ef1f36 DB |
912 | params->tls_creds = g_strdup(s->parameters.tls_creds); |
913 | params->tls_hostname = g_strdup(s->parameters.tls_hostname); | |
7cd75cbd MZ |
914 | params->tls_authz = g_strdup(s->parameters.tls_authz ? |
915 | s->parameters.tls_authz : ""); | |
2ff30257 AA |
916 | params->has_max_bandwidth = true; |
917 | params->max_bandwidth = s->parameters.max_bandwidth; | |
918 | params->has_downtime_limit = true; | |
919 | params->downtime_limit = s->parameters.downtime_limit; | |
fe39a4d4 | 920 | params->has_x_checkpoint_delay = true; |
68b53591 | 921 | params->x_checkpoint_delay = s->parameters.x_checkpoint_delay; |
2833c59b JQ |
922 | params->has_block_incremental = true; |
923 | params->block_incremental = s->parameters.block_incremental; | |
cbfd6c95 JQ |
924 | params->has_multifd_channels = true; |
925 | params->multifd_channels = s->parameters.multifd_channels; | |
96eef042 JQ |
926 | params->has_multifd_compression = true; |
927 | params->multifd_compression = s->parameters.multifd_compression; | |
9004db48 JQ |
928 | params->has_multifd_zlib_level = true; |
929 | params->multifd_zlib_level = s->parameters.multifd_zlib_level; | |
6a9ad154 JQ |
930 | params->has_multifd_zstd_level = true; |
931 | params->multifd_zstd_level = s->parameters.multifd_zstd_level; | |
73af8dd8 JQ |
932 | params->has_xbzrle_cache_size = true; |
933 | params->xbzrle_cache_size = s->parameters.xbzrle_cache_size; | |
7e555c6c DDAG |
934 | params->has_max_postcopy_bandwidth = true; |
935 | params->max_postcopy_bandwidth = s->parameters.max_postcopy_bandwidth; | |
4cbc9c7f LQ |
936 | params->has_max_cpu_throttle = true; |
937 | params->max_cpu_throttle = s->parameters.max_cpu_throttle; | |
ee3d96ba DDAG |
938 | params->has_announce_initial = true; |
939 | params->announce_initial = s->parameters.announce_initial; | |
940 | params->has_announce_max = true; | |
941 | params->announce_max = s->parameters.announce_max; | |
942 | params->has_announce_rounds = true; | |
943 | params->announce_rounds = s->parameters.announce_rounds; | |
944 | params->has_announce_step = true; | |
945 | params->announce_step = s->parameters.announce_step; | |
85de8323 | 946 | |
31e4c354 HR |
947 | if (s->parameters.has_block_bitmap_mapping) { |
948 | params->has_block_bitmap_mapping = true; | |
949 | params->block_bitmap_mapping = | |
950 | QAPI_CLONE(BitmapMigrationNodeAliasList, | |
951 | s->parameters.block_bitmap_mapping); | |
952 | } | |
953 | ||
85de8323 LL |
954 | return params; |
955 | } | |
956 | ||
f6844b99 DDAG |
957 | /* |
958 | * Return true if we're already in the middle of a migration | |
959 | * (i.e. any of the active or setup states) | |
960 | */ | |
3d63da16 | 961 | bool migration_is_setup_or_active(int state) |
f6844b99 DDAG |
962 | { |
963 | switch (state) { | |
964 | case MIGRATION_STATUS_ACTIVE: | |
9ec055ae | 965 | case MIGRATION_STATUS_POSTCOPY_ACTIVE: |
a688d2c1 | 966 | case MIGRATION_STATUS_POSTCOPY_PAUSED: |
135b87b4 | 967 | case MIGRATION_STATUS_POSTCOPY_RECOVER: |
f6844b99 | 968 | case MIGRATION_STATUS_SETUP: |
31e06077 DDAG |
969 | case MIGRATION_STATUS_PRE_SWITCHOVER: |
970 | case MIGRATION_STATUS_DEVICE: | |
c7e0acd5 | 971 | case MIGRATION_STATUS_WAIT_UNPLUG: |
19dd408a | 972 | case MIGRATION_STATUS_COLO: |
f6844b99 DDAG |
973 | return true; |
974 | ||
975 | default: | |
976 | return false; | |
977 | ||
978 | } | |
979 | } | |
980 | ||
392d87e2 JQ |
981 | bool migration_is_running(int state) |
982 | { | |
983 | switch (state) { | |
984 | case MIGRATION_STATUS_ACTIVE: | |
985 | case MIGRATION_STATUS_POSTCOPY_ACTIVE: | |
986 | case MIGRATION_STATUS_POSTCOPY_PAUSED: | |
987 | case MIGRATION_STATUS_POSTCOPY_RECOVER: | |
988 | case MIGRATION_STATUS_SETUP: | |
989 | case MIGRATION_STATUS_PRE_SWITCHOVER: | |
990 | case MIGRATION_STATUS_DEVICE: | |
991 | case MIGRATION_STATUS_WAIT_UNPLUG: | |
992 | case MIGRATION_STATUS_CANCELLING: | |
392d87e2 JQ |
993 | return true; |
994 | ||
995 | default: | |
996 | return false; | |
997 | ||
998 | } | |
999 | } | |
1000 | ||
db18dee7 PX |
1001 | static bool migrate_show_downtime(MigrationState *s) |
1002 | { | |
1003 | return (s->state == MIGRATION_STATUS_COMPLETED) || migration_in_postcopy(); | |
1004 | } | |
1005 | ||
640dfb14 WY |
1006 | static void populate_time_info(MigrationInfo *info, MigrationState *s) |
1007 | { | |
1008 | info->has_status = true; | |
1009 | info->has_setup_time = true; | |
1010 | info->setup_time = s->setup_time; | |
db18dee7 | 1011 | |
640dfb14 WY |
1012 | if (s->state == MIGRATION_STATUS_COMPLETED) { |
1013 | info->has_total_time = true; | |
1014 | info->total_time = s->total_time; | |
640dfb14 WY |
1015 | } else { |
1016 | info->has_total_time = true; | |
1017 | info->total_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME) - | |
1018 | s->start_time; | |
db18dee7 PX |
1019 | } |
1020 | ||
1021 | if (migrate_show_downtime(s)) { | |
1022 | info->has_downtime = true; | |
1023 | info->downtime = s->downtime; | |
1024 | } else { | |
640dfb14 WY |
1025 | info->has_expected_downtime = true; |
1026 | info->expected_downtime = s->expected_downtime; | |
1027 | } | |
1028 | } | |
1029 | ||
a22463a5 DDAG |
1030 | static void populate_ram_info(MigrationInfo *info, MigrationState *s) |
1031 | { | |
144fa06b JQ |
1032 | size_t page_size = qemu_target_page_size(); |
1033 | ||
a22463a5 | 1034 | info->ram = g_malloc0(sizeof(*info->ram)); |
abce5fa1 | 1035 | info->ram->transferred = stat64_get(&ram_counters.transferred); |
a22463a5 | 1036 | info->ram->total = ram_bytes_total(); |
1a386e8d | 1037 | info->ram->duplicate = stat64_get(&ram_counters.zero_pages); |
bedf53c1 JQ |
1038 | /* legacy value. It is not used anymore */ |
1039 | info->ram->skipped = 0; | |
8c0cda8f | 1040 | info->ram->normal = stat64_get(&ram_counters.normal_pages); |
23b7576d | 1041 | info->ram->normal_bytes = info->ram->normal * page_size; |
a22463a5 | 1042 | info->ram->mbps = s->mbps; |
536b5a4e JQ |
1043 | info->ram->dirty_sync_count = |
1044 | stat64_get(&ram_counters.dirty_sync_count); | |
cf20c897 | 1045 | info->ram->dirty_sync_missed_zero_copy = |
42918236 | 1046 | stat64_get(&ram_counters.dirty_sync_missed_zero_copy); |
3c764f9b JQ |
1047 | info->ram->postcopy_requests = |
1048 | stat64_get(&ram_counters.postcopy_requests); | |
144fa06b | 1049 | info->ram->page_size = page_size; |
cf671116 | 1050 | info->ram->multifd_bytes = stat64_get(&ram_counters.multifd_bytes); |
aecbfe9c | 1051 | info->ram->pages_per_second = s->pages_per_second; |
b013b5d1 | 1052 | info->ram->precopy_bytes = stat64_get(&ram_counters.precopy_bytes); |
296a4ac2 | 1053 | info->ram->downtime_bytes = stat64_get(&ram_counters.downtime_bytes); |
abce5fa1 | 1054 | info->ram->postcopy_bytes = stat64_get(&ram_counters.postcopy_bytes); |
a22463a5 | 1055 | |
87dca0c9 | 1056 | if (migrate_xbzrle()) { |
114f5aee JQ |
1057 | info->xbzrle_cache = g_malloc0(sizeof(*info->xbzrle_cache)); |
1058 | info->xbzrle_cache->cache_size = migrate_xbzrle_cache_size(); | |
9360447d JQ |
1059 | info->xbzrle_cache->bytes = xbzrle_counters.bytes; |
1060 | info->xbzrle_cache->pages = xbzrle_counters.pages; | |
1061 | info->xbzrle_cache->cache_miss = xbzrle_counters.cache_miss; | |
1062 | info->xbzrle_cache->cache_miss_rate = xbzrle_counters.cache_miss_rate; | |
e460a4b1 | 1063 | info->xbzrle_cache->encoding_rate = xbzrle_counters.encoding_rate; |
9360447d | 1064 | info->xbzrle_cache->overflow = xbzrle_counters.overflow; |
114f5aee JQ |
1065 | } |
1066 | ||
a7a94d14 | 1067 | if (migrate_compress()) { |
76e03000 XG |
1068 | info->compression = g_malloc0(sizeof(*info->compression)); |
1069 | info->compression->pages = compression_counters.pages; | |
1070 | info->compression->busy = compression_counters.busy; | |
1071 | info->compression->busy_rate = compression_counters.busy_rate; | |
1072 | info->compression->compressed_size = | |
1073 | compression_counters.compressed_size; | |
1074 | info->compression->compression_rate = | |
1075 | compression_counters.compression_rate; | |
1076 | } | |
1077 | ||
338182c8 JQ |
1078 | if (cpu_throttle_active()) { |
1079 | info->has_cpu_throttle_percentage = true; | |
1080 | info->cpu_throttle_percentage = cpu_throttle_get_percentage(); | |
1081 | } | |
1082 | ||
a22463a5 DDAG |
1083 | if (s->state != MIGRATION_STATUS_COMPLETED) { |
1084 | info->ram->remaining = ram_bytes_remaining(); | |
9360447d | 1085 | info->ram->dirty_pages_rate = ram_counters.dirty_pages_rate; |
a22463a5 DDAG |
1086 | } |
1087 | } | |
1088 | ||
930ac04c JQ |
1089 | static void populate_disk_info(MigrationInfo *info) |
1090 | { | |
1091 | if (blk_mig_active()) { | |
930ac04c JQ |
1092 | info->disk = g_malloc0(sizeof(*info->disk)); |
1093 | info->disk->transferred = blk_mig_bytes_transferred(); | |
1094 | info->disk->remaining = blk_mig_bytes_remaining(); | |
1095 | info->disk->total = blk_mig_bytes_total(); | |
1096 | } | |
1097 | } | |
1098 | ||
65ace060 | 1099 | static void fill_source_migration_info(MigrationInfo *info) |
5bb7910a | 1100 | { |
17549e84 | 1101 | MigrationState *s = migrate_get_current(); |
552de79b | 1102 | int state = qatomic_read(&s->state); |
372043f3 | 1103 | GSList *cur_blocker = migration_blockers; |
17549e84 | 1104 | |
3af8554b | 1105 | info->blocked_reasons = NULL; |
3af8554b | 1106 | |
372043f3 MA |
1107 | /* |
1108 | * There are two types of reasons a migration might be blocked; | |
1109 | * a) devices marked in VMState as non-migratable, and | |
1110 | * b) Explicit migration blockers | |
1111 | * We need to add both of them here. | |
1112 | */ | |
1113 | qemu_savevm_non_migratable_list(&info->blocked_reasons); | |
3af8554b | 1114 | |
372043f3 MA |
1115 | while (cur_blocker) { |
1116 | QAPI_LIST_PREPEND(info->blocked_reasons, | |
1117 | g_strdup(error_get_pretty(cur_blocker->data))); | |
1118 | cur_blocker = g_slist_next(cur_blocker); | |
3af8554b | 1119 | } |
372043f3 | 1120 | info->has_blocked_reasons = info->blocked_reasons != NULL; |
3af8554b | 1121 | |
552de79b | 1122 | switch (state) { |
31194731 | 1123 | case MIGRATION_STATUS_NONE: |
17549e84 | 1124 | /* no migration has happened ever */ |
65ace060 AP |
1125 | /* do not overwrite destination migration status */ |
1126 | return; | |
31194731 | 1127 | case MIGRATION_STATUS_SETUP: |
29ae8a41 | 1128 | info->has_status = true; |
ed4fbd10 | 1129 | info->has_total_time = false; |
29ae8a41 | 1130 | break; |
31194731 HZ |
1131 | case MIGRATION_STATUS_ACTIVE: |
1132 | case MIGRATION_STATUS_CANCELLING: | |
9ec055ae | 1133 | case MIGRATION_STATUS_POSTCOPY_ACTIVE: |
31e06077 DDAG |
1134 | case MIGRATION_STATUS_PRE_SWITCHOVER: |
1135 | case MIGRATION_STATUS_DEVICE: | |
a688d2c1 | 1136 | case MIGRATION_STATUS_POSTCOPY_PAUSED: |
135b87b4 | 1137 | case MIGRATION_STATUS_POSTCOPY_RECOVER: |
640dfb14 WY |
1138 | /* TODO add some postcopy stats */ |
1139 | populate_time_info(info, s); | |
a22463a5 | 1140 | populate_ram_info(info, s); |
930ac04c | 1141 | populate_disk_info(info); |
3710586c | 1142 | populate_vfio_info(info); |
17549e84 | 1143 | break; |
0b827d5e HZ |
1144 | case MIGRATION_STATUS_COLO: |
1145 | info->has_status = true; | |
1146 | /* TODO: display COLO specific information (checkpoint info etc.) */ | |
1147 | break; | |
31194731 | 1148 | case MIGRATION_STATUS_COMPLETED: |
640dfb14 | 1149 | populate_time_info(info, s); |
a22463a5 | 1150 | populate_ram_info(info, s); |
3710586c | 1151 | populate_vfio_info(info); |
17549e84 | 1152 | break; |
31194731 | 1153 | case MIGRATION_STATUS_FAILED: |
791e7c82 | 1154 | info->has_status = true; |
d59ce6f3 | 1155 | if (s->error) { |
d59ce6f3 DB |
1156 | info->error_desc = g_strdup(error_get_pretty(s->error)); |
1157 | } | |
17549e84 | 1158 | break; |
31194731 | 1159 | case MIGRATION_STATUS_CANCELLED: |
791e7c82 | 1160 | info->has_status = true; |
17549e84 | 1161 | break; |
c7e0acd5 JF |
1162 | case MIGRATION_STATUS_WAIT_UNPLUG: |
1163 | info->has_status = true; | |
1164 | break; | |
5bb7910a | 1165 | } |
552de79b | 1166 | info->status = state; |
5bb7910a AL |
1167 | } |
1168 | ||
65ace060 AP |
1169 | static void fill_destination_migration_info(MigrationInfo *info) |
1170 | { | |
1171 | MigrationIncomingState *mis = migration_incoming_get_current(); | |
1172 | ||
9aca82ba JQ |
1173 | if (mis->socket_address_list) { |
1174 | info->has_socket_address = true; | |
1175 | info->socket_address = | |
1176 | QAPI_CLONE(SocketAddressList, mis->socket_address_list); | |
1177 | } | |
1178 | ||
65ace060 AP |
1179 | switch (mis->state) { |
1180 | case MIGRATION_STATUS_NONE: | |
1181 | return; | |
65ace060 AP |
1182 | case MIGRATION_STATUS_SETUP: |
1183 | case MIGRATION_STATUS_CANCELLING: | |
1184 | case MIGRATION_STATUS_CANCELLED: | |
1185 | case MIGRATION_STATUS_ACTIVE: | |
1186 | case MIGRATION_STATUS_POSTCOPY_ACTIVE: | |
3c9928d9 PX |
1187 | case MIGRATION_STATUS_POSTCOPY_PAUSED: |
1188 | case MIGRATION_STATUS_POSTCOPY_RECOVER: | |
65ace060 AP |
1189 | case MIGRATION_STATUS_FAILED: |
1190 | case MIGRATION_STATUS_COLO: | |
1191 | info->has_status = true; | |
1192 | break; | |
1193 | case MIGRATION_STATUS_COMPLETED: | |
1194 | info->has_status = true; | |
1195 | fill_destination_postcopy_migration_info(info); | |
1196 | break; | |
1197 | } | |
1198 | info->status = mis->state; | |
1199 | } | |
1200 | ||
1201 | MigrationInfo *qmp_query_migrate(Error **errp) | |
1202 | { | |
1203 | MigrationInfo *info = g_malloc0(sizeof(*info)); | |
1204 | ||
1205 | fill_destination_migration_info(info); | |
1206 | fill_source_migration_info(info); | |
1207 | ||
1208 | return info; | |
1209 | } | |
1210 | ||
16d063bc PX |
1211 | /* |
1212 | * Check whether the parameters are valid. Error will be put into errp | |
1213 | * (if provided). Return true if valid, otherwise false. | |
1214 | */ | |
1215 | static bool migrate_params_check(MigrationParameters *params, Error **errp) | |
85de8323 | 1216 | { |
7f375e04 | 1217 | if (params->has_compress_level && |
741d4086 | 1218 | (params->compress_level > 9)) { |
c6bd8c70 | 1219 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "compress_level", |
54270c45 | 1220 | "a value between 0 and 9"); |
16d063bc | 1221 | return false; |
85de8323 | 1222 | } |
16d063bc | 1223 | |
741d4086 | 1224 | if (params->has_compress_threads && (params->compress_threads < 1)) { |
c6bd8c70 MA |
1225 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, |
1226 | "compress_threads", | |
54270c45 | 1227 | "a value between 1 and 255"); |
16d063bc | 1228 | return false; |
85de8323 | 1229 | } |
16d063bc | 1230 | |
741d4086 | 1231 | if (params->has_decompress_threads && (params->decompress_threads < 1)) { |
c6bd8c70 MA |
1232 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, |
1233 | "decompress_threads", | |
54270c45 | 1234 | "a value between 1 and 255"); |
16d063bc | 1235 | return false; |
85de8323 | 1236 | } |
16d063bc | 1237 | |
dc14a470 KZ |
1238 | if (params->has_throttle_trigger_threshold && |
1239 | (params->throttle_trigger_threshold < 1 || | |
1240 | params->throttle_trigger_threshold > 100)) { | |
1241 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, | |
1242 | "throttle_trigger_threshold", | |
1243 | "an integer in the range of 1 to 100"); | |
1244 | return false; | |
1245 | } | |
1246 | ||
7f375e04 EB |
1247 | if (params->has_cpu_throttle_initial && |
1248 | (params->cpu_throttle_initial < 1 || | |
1249 | params->cpu_throttle_initial > 99)) { | |
1626fee3 | 1250 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, |
d85a31d1 | 1251 | "cpu_throttle_initial", |
1626fee3 | 1252 | "an integer in the range of 1 to 99"); |
16d063bc | 1253 | return false; |
1626fee3 | 1254 | } |
16d063bc | 1255 | |
7f375e04 EB |
1256 | if (params->has_cpu_throttle_increment && |
1257 | (params->cpu_throttle_increment < 1 || | |
1258 | params->cpu_throttle_increment > 99)) { | |
1626fee3 | 1259 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, |
d85a31d1 | 1260 | "cpu_throttle_increment", |
1626fee3 | 1261 | "an integer in the range of 1 to 99"); |
16d063bc | 1262 | return false; |
1626fee3 | 1263 | } |
16d063bc | 1264 | |
741d4086 | 1265 | if (params->has_max_bandwidth && (params->max_bandwidth > SIZE_MAX)) { |
2ee30cf0 MZ |
1266 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, |
1267 | "max_bandwidth", | |
1268 | "an integer in the range of 0 to "stringify(SIZE_MAX) | |
1269 | " bytes/second"); | |
16d063bc | 1270 | return false; |
2ff30257 | 1271 | } |
16d063bc | 1272 | |
2ff30257 | 1273 | if (params->has_downtime_limit && |
741d4086 | 1274 | (params->downtime_limit > MAX_MIGRATE_DOWNTIME)) { |
2ee30cf0 MZ |
1275 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, |
1276 | "downtime_limit", | |
1277 | "an integer in the range of 0 to " | |
7ac5529a | 1278 | stringify(MAX_MIGRATE_DOWNTIME)" ms"); |
16d063bc | 1279 | return false; |
2ff30257 | 1280 | } |
16d063bc | 1281 | |
741d4086 JQ |
1282 | /* x_checkpoint_delay is now always positive */ |
1283 | ||
cbfd6c95 | 1284 | if (params->has_multifd_channels && (params->multifd_channels < 1)) { |
4075fb1c JQ |
1285 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, |
1286 | "multifd_channels", | |
54270c45 | 1287 | "a value between 1 and 255"); |
4075fb1c JQ |
1288 | return false; |
1289 | } | |
16d063bc | 1290 | |
9004db48 JQ |
1291 | if (params->has_multifd_zlib_level && |
1292 | (params->multifd_zlib_level > 9)) { | |
1293 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "multifd_zlib_level", | |
54270c45 | 1294 | "a value between 0 and 9"); |
9004db48 JQ |
1295 | return false; |
1296 | } | |
1297 | ||
6a9ad154 JQ |
1298 | if (params->has_multifd_zstd_level && |
1299 | (params->multifd_zstd_level > 20)) { | |
1300 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "multifd_zstd_level", | |
54270c45 | 1301 | "a value between 0 and 20"); |
6a9ad154 JQ |
1302 | return false; |
1303 | } | |
1304 | ||
73af8dd8 JQ |
1305 | if (params->has_xbzrle_cache_size && |
1306 | (params->xbzrle_cache_size < qemu_target_page_size() || | |
1307 | !is_power_of_2(params->xbzrle_cache_size))) { | |
1308 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, | |
1309 | "xbzrle_cache_size", | |
54270c45 | 1310 | "a power of two no less than the target page size"); |
73af8dd8 JQ |
1311 | return false; |
1312 | } | |
1313 | ||
4cbc9c7f LQ |
1314 | if (params->has_max_cpu_throttle && |
1315 | (params->max_cpu_throttle < params->cpu_throttle_initial || | |
1316 | params->max_cpu_throttle > 99)) { | |
1317 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, | |
1318 | "max_cpu_throttle", | |
1319 | "an integer in the range of cpu_throttle_initial to 99"); | |
1320 | return false; | |
1321 | } | |
1322 | ||
ee3d96ba DDAG |
1323 | if (params->has_announce_initial && |
1324 | params->announce_initial > 100000) { | |
1325 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, | |
1326 | "announce_initial", | |
54270c45 | 1327 | "a value between 0 and 100000"); |
ee3d96ba DDAG |
1328 | return false; |
1329 | } | |
1330 | if (params->has_announce_max && | |
1331 | params->announce_max > 100000) { | |
1332 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, | |
1333 | "announce_max", | |
54270c45 | 1334 | "a value between 0 and 100000"); |
ee3d96ba DDAG |
1335 | return false; |
1336 | } | |
1337 | if (params->has_announce_rounds && | |
1338 | params->announce_rounds > 1000) { | |
1339 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, | |
1340 | "announce_rounds", | |
54270c45 | 1341 | "a value between 0 and 1000"); |
ee3d96ba DDAG |
1342 | return false; |
1343 | } | |
1344 | if (params->has_announce_step && | |
1345 | (params->announce_step < 1 || | |
1346 | params->announce_step > 10000)) { | |
1347 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, | |
1348 | "announce_step", | |
54270c45 | 1349 | "a value between 0 and 10000"); |
ee3d96ba DDAG |
1350 | return false; |
1351 | } | |
31e4c354 HR |
1352 | |
1353 | if (params->has_block_bitmap_mapping && | |
1354 | !check_dirty_bitmap_mig_alias_map(params->block_bitmap_mapping, errp)) { | |
1355 | error_prepend(errp, "Invalid mapping given for block-bitmap-mapping: "); | |
1356 | return false; | |
1357 | } | |
90eb69e4 LB |
1358 | |
1359 | #ifdef CONFIG_LINUX | |
b4bc342c | 1360 | if (migrate_zero_copy_send() && |
90eb69e4 | 1361 | ((params->has_multifd_compression && params->multifd_compression) || |
720a252c | 1362 | (params->tls_creds && *params->tls_creds))) { |
90eb69e4 LB |
1363 | error_setg(errp, |
1364 | "Zero copy only available for non-compressed non-TLS multifd migration"); | |
1365 | return false; | |
1366 | } | |
1367 | #endif | |
1368 | ||
16d063bc PX |
1369 | return true; |
1370 | } | |
1371 | ||
1bda8b3c MA |
1372 | static void migrate_params_test_apply(MigrateSetParameters *params, |
1373 | MigrationParameters *dest) | |
1374 | { | |
1375 | *dest = migrate_get_current()->parameters; | |
1376 | ||
1377 | /* TODO use QAPI_CLONE() instead of duplicating it inline */ | |
1378 | ||
1379 | if (params->has_compress_level) { | |
1380 | dest->compress_level = params->compress_level; | |
1381 | } | |
1382 | ||
1383 | if (params->has_compress_threads) { | |
1384 | dest->compress_threads = params->compress_threads; | |
1385 | } | |
1386 | ||
1d58872a XG |
1387 | if (params->has_compress_wait_thread) { |
1388 | dest->compress_wait_thread = params->compress_wait_thread; | |
1389 | } | |
1390 | ||
1bda8b3c MA |
1391 | if (params->has_decompress_threads) { |
1392 | dest->decompress_threads = params->decompress_threads; | |
1393 | } | |
1394 | ||
dc14a470 KZ |
1395 | if (params->has_throttle_trigger_threshold) { |
1396 | dest->throttle_trigger_threshold = params->throttle_trigger_threshold; | |
1397 | } | |
1398 | ||
1bda8b3c MA |
1399 | if (params->has_cpu_throttle_initial) { |
1400 | dest->cpu_throttle_initial = params->cpu_throttle_initial; | |
1401 | } | |
1402 | ||
1403 | if (params->has_cpu_throttle_increment) { | |
1404 | dest->cpu_throttle_increment = params->cpu_throttle_increment; | |
1405 | } | |
1406 | ||
cbbf8182 KZ |
1407 | if (params->has_cpu_throttle_tailslow) { |
1408 | dest->cpu_throttle_tailslow = params->cpu_throttle_tailslow; | |
1409 | } | |
1410 | ||
720a252c | 1411 | if (params->tls_creds) { |
01fa5598 | 1412 | assert(params->tls_creds->type == QTYPE_QSTRING); |
9728ebfb | 1413 | dest->tls_creds = params->tls_creds->u.s; |
1bda8b3c MA |
1414 | } |
1415 | ||
720a252c | 1416 | if (params->tls_hostname) { |
01fa5598 | 1417 | assert(params->tls_hostname->type == QTYPE_QSTRING); |
9728ebfb | 1418 | dest->tls_hostname = params->tls_hostname->u.s; |
1bda8b3c MA |
1419 | } |
1420 | ||
1421 | if (params->has_max_bandwidth) { | |
1422 | dest->max_bandwidth = params->max_bandwidth; | |
1423 | } | |
1424 | ||
1425 | if (params->has_downtime_limit) { | |
1426 | dest->downtime_limit = params->downtime_limit; | |
1427 | } | |
1428 | ||
1429 | if (params->has_x_checkpoint_delay) { | |
1430 | dest->x_checkpoint_delay = params->x_checkpoint_delay; | |
1431 | } | |
1432 | ||
1433 | if (params->has_block_incremental) { | |
1434 | dest->block_incremental = params->block_incremental; | |
1435 | } | |
cbfd6c95 JQ |
1436 | if (params->has_multifd_channels) { |
1437 | dest->multifd_channels = params->multifd_channels; | |
5e7577a1 | 1438 | } |
96eef042 JQ |
1439 | if (params->has_multifd_compression) { |
1440 | dest->multifd_compression = params->multifd_compression; | |
1441 | } | |
73af8dd8 JQ |
1442 | if (params->has_xbzrle_cache_size) { |
1443 | dest->xbzrle_cache_size = params->xbzrle_cache_size; | |
1444 | } | |
7e555c6c DDAG |
1445 | if (params->has_max_postcopy_bandwidth) { |
1446 | dest->max_postcopy_bandwidth = params->max_postcopy_bandwidth; | |
1447 | } | |
4cbc9c7f LQ |
1448 | if (params->has_max_cpu_throttle) { |
1449 | dest->max_cpu_throttle = params->max_cpu_throttle; | |
1450 | } | |
ee3d96ba DDAG |
1451 | if (params->has_announce_initial) { |
1452 | dest->announce_initial = params->announce_initial; | |
1453 | } | |
1454 | if (params->has_announce_max) { | |
1455 | dest->announce_max = params->announce_max; | |
1456 | } | |
1457 | if (params->has_announce_rounds) { | |
1458 | dest->announce_rounds = params->announce_rounds; | |
1459 | } | |
1460 | if (params->has_announce_step) { | |
1461 | dest->announce_step = params->announce_step; | |
1462 | } | |
31e4c354 HR |
1463 | |
1464 | if (params->has_block_bitmap_mapping) { | |
1465 | dest->has_block_bitmap_mapping = true; | |
1466 | dest->block_bitmap_mapping = params->block_bitmap_mapping; | |
1467 | } | |
1bda8b3c MA |
1468 | } |
1469 | ||
73af8dd8 | 1470 | static void migrate_params_apply(MigrateSetParameters *params, Error **errp) |
16d063bc PX |
1471 | { |
1472 | MigrationState *s = migrate_get_current(); | |
1473 | ||
e87fae4c MA |
1474 | /* TODO use QAPI_CLONE() instead of duplicating it inline */ |
1475 | ||
7f375e04 EB |
1476 | if (params->has_compress_level) { |
1477 | s->parameters.compress_level = params->compress_level; | |
85de8323 | 1478 | } |
476c72aa | 1479 | |
7f375e04 EB |
1480 | if (params->has_compress_threads) { |
1481 | s->parameters.compress_threads = params->compress_threads; | |
85de8323 | 1482 | } |
476c72aa | 1483 | |
1d58872a XG |
1484 | if (params->has_compress_wait_thread) { |
1485 | s->parameters.compress_wait_thread = params->compress_wait_thread; | |
1486 | } | |
1487 | ||
7f375e04 EB |
1488 | if (params->has_decompress_threads) { |
1489 | s->parameters.decompress_threads = params->decompress_threads; | |
85de8323 | 1490 | } |
476c72aa | 1491 | |
dc14a470 KZ |
1492 | if (params->has_throttle_trigger_threshold) { |
1493 | s->parameters.throttle_trigger_threshold = params->throttle_trigger_threshold; | |
1494 | } | |
1495 | ||
7f375e04 EB |
1496 | if (params->has_cpu_throttle_initial) { |
1497 | s->parameters.cpu_throttle_initial = params->cpu_throttle_initial; | |
1626fee3 | 1498 | } |
476c72aa | 1499 | |
7f375e04 EB |
1500 | if (params->has_cpu_throttle_increment) { |
1501 | s->parameters.cpu_throttle_increment = params->cpu_throttle_increment; | |
1626fee3 | 1502 | } |
476c72aa | 1503 | |
cbbf8182 KZ |
1504 | if (params->has_cpu_throttle_tailslow) { |
1505 | s->parameters.cpu_throttle_tailslow = params->cpu_throttle_tailslow; | |
1506 | } | |
1507 | ||
720a252c | 1508 | if (params->tls_creds) { |
69ef1f36 | 1509 | g_free(s->parameters.tls_creds); |
01fa5598 MA |
1510 | assert(params->tls_creds->type == QTYPE_QSTRING); |
1511 | s->parameters.tls_creds = g_strdup(params->tls_creds->u.s); | |
69ef1f36 | 1512 | } |
476c72aa | 1513 | |
720a252c | 1514 | if (params->tls_hostname) { |
69ef1f36 | 1515 | g_free(s->parameters.tls_hostname); |
01fa5598 MA |
1516 | assert(params->tls_hostname->type == QTYPE_QSTRING); |
1517 | s->parameters.tls_hostname = g_strdup(params->tls_hostname->u.s); | |
69ef1f36 | 1518 | } |
476c72aa | 1519 | |
720a252c | 1520 | if (params->tls_authz) { |
d2f1d29b DB |
1521 | g_free(s->parameters.tls_authz); |
1522 | assert(params->tls_authz->type == QTYPE_QSTRING); | |
1523 | s->parameters.tls_authz = g_strdup(params->tls_authz->u.s); | |
1524 | } | |
1525 | ||
2ff30257 AA |
1526 | if (params->has_max_bandwidth) { |
1527 | s->parameters.max_bandwidth = params->max_bandwidth; | |
c38c1c14 | 1528 | if (s->to_dst_file && !migration_in_postcopy()) { |
2ff30257 AA |
1529 | qemu_file_set_rate_limit(s->to_dst_file, |
1530 | s->parameters.max_bandwidth / XFER_LIMIT_RATIO); | |
1531 | } | |
1532 | } | |
476c72aa | 1533 | |
2ff30257 AA |
1534 | if (params->has_downtime_limit) { |
1535 | s->parameters.downtime_limit = params->downtime_limit; | |
1536 | } | |
68b53591 HZ |
1537 | |
1538 | if (params->has_x_checkpoint_delay) { | |
1539 | s->parameters.x_checkpoint_delay = params->x_checkpoint_delay; | |
479125d5 HZ |
1540 | if (migration_in_colo_state()) { |
1541 | colo_checkpoint_notify(s); | |
1542 | } | |
68b53591 | 1543 | } |
476c72aa | 1544 | |
2833c59b JQ |
1545 | if (params->has_block_incremental) { |
1546 | s->parameters.block_incremental = params->block_incremental; | |
1547 | } | |
cbfd6c95 JQ |
1548 | if (params->has_multifd_channels) { |
1549 | s->parameters.multifd_channels = params->multifd_channels; | |
4075fb1c | 1550 | } |
96eef042 JQ |
1551 | if (params->has_multifd_compression) { |
1552 | s->parameters.multifd_compression = params->multifd_compression; | |
1553 | } | |
73af8dd8 JQ |
1554 | if (params->has_xbzrle_cache_size) { |
1555 | s->parameters.xbzrle_cache_size = params->xbzrle_cache_size; | |
1556 | xbzrle_cache_resize(params->xbzrle_cache_size, errp); | |
1557 | } | |
7e555c6c DDAG |
1558 | if (params->has_max_postcopy_bandwidth) { |
1559 | s->parameters.max_postcopy_bandwidth = params->max_postcopy_bandwidth; | |
c38c1c14 DDAG |
1560 | if (s->to_dst_file && migration_in_postcopy()) { |
1561 | qemu_file_set_rate_limit(s->to_dst_file, | |
1562 | s->parameters.max_postcopy_bandwidth / XFER_LIMIT_RATIO); | |
1563 | } | |
7e555c6c | 1564 | } |
4cbc9c7f LQ |
1565 | if (params->has_max_cpu_throttle) { |
1566 | s->parameters.max_cpu_throttle = params->max_cpu_throttle; | |
1567 | } | |
ee3d96ba DDAG |
1568 | if (params->has_announce_initial) { |
1569 | s->parameters.announce_initial = params->announce_initial; | |
1570 | } | |
1571 | if (params->has_announce_max) { | |
1572 | s->parameters.announce_max = params->announce_max; | |
1573 | } | |
1574 | if (params->has_announce_rounds) { | |
1575 | s->parameters.announce_rounds = params->announce_rounds; | |
1576 | } | |
1577 | if (params->has_announce_step) { | |
1578 | s->parameters.announce_step = params->announce_step; | |
1579 | } | |
31e4c354 HR |
1580 | |
1581 | if (params->has_block_bitmap_mapping) { | |
1582 | qapi_free_BitmapMigrationNodeAliasList( | |
1583 | s->parameters.block_bitmap_mapping); | |
1584 | ||
1585 | s->parameters.has_block_bitmap_mapping = true; | |
1586 | s->parameters.block_bitmap_mapping = | |
1587 | QAPI_CLONE(BitmapMigrationNodeAliasList, | |
1588 | params->block_bitmap_mapping); | |
1589 | } | |
85de8323 LL |
1590 | } |
1591 | ||
1bda8b3c | 1592 | void qmp_migrate_set_parameters(MigrateSetParameters *params, Error **errp) |
476c72aa | 1593 | { |
1bda8b3c MA |
1594 | MigrationParameters tmp; |
1595 | ||
01fa5598 | 1596 | /* TODO Rewrite "" to null instead */ |
720a252c | 1597 | if (params->tls_creds |
01fa5598 | 1598 | && params->tls_creds->type == QTYPE_QNULL) { |
cb3e7f08 | 1599 | qobject_unref(params->tls_creds->u.n); |
01fa5598 MA |
1600 | params->tls_creds->type = QTYPE_QSTRING; |
1601 | params->tls_creds->u.s = strdup(""); | |
1602 | } | |
1603 | /* TODO Rewrite "" to null instead */ | |
720a252c | 1604 | if (params->tls_hostname |
01fa5598 | 1605 | && params->tls_hostname->type == QTYPE_QNULL) { |
cb3e7f08 | 1606 | qobject_unref(params->tls_hostname->u.n); |
01fa5598 MA |
1607 | params->tls_hostname->type = QTYPE_QSTRING; |
1608 | params->tls_hostname->u.s = strdup(""); | |
1609 | } | |
1610 | ||
1bda8b3c MA |
1611 | migrate_params_test_apply(params, &tmp); |
1612 | ||
1613 | if (!migrate_params_check(&tmp, errp)) { | |
476c72aa PX |
1614 | /* Invalid parameter */ |
1615 | return; | |
1616 | } | |
1617 | ||
73af8dd8 | 1618 | migrate_params_apply(params, errp); |
476c72aa PX |
1619 | } |
1620 | ||
2594f56d | 1621 | |
4886a1bc DDAG |
1622 | void qmp_migrate_start_postcopy(Error **errp) |
1623 | { | |
1624 | MigrationState *s = migrate_get_current(); | |
1625 | ||
16b0fd32 | 1626 | if (!migrate_postcopy()) { |
a54d340b | 1627 | error_setg(errp, "Enable postcopy with migrate_set_capability before" |
4886a1bc DDAG |
1628 | " the start of migration"); |
1629 | return; | |
1630 | } | |
1631 | ||
1632 | if (s->state == MIGRATION_STATUS_NONE) { | |
1633 | error_setg(errp, "Postcopy must be started after migration has been" | |
1634 | " started"); | |
1635 | return; | |
1636 | } | |
1637 | /* | |
1638 | * we don't error if migration has finished since that would be racy | |
1639 | * with issuing this command. | |
1640 | */ | |
d73415a3 | 1641 | qatomic_set(&s->start_postcopy, true); |
4886a1bc DDAG |
1642 | } |
1643 | ||
065e2813 AL |
1644 | /* shared migration helpers */ |
1645 | ||
48781e5b | 1646 | void migrate_set_state(int *state, int old_state, int new_state) |
51cf4c1a | 1647 | { |
a31fedee | 1648 | assert(new_state < MIGRATION_STATUS__MAX); |
d73415a3 | 1649 | if (qatomic_cmpxchg(state, old_state, new_state) == old_state) { |
a31fedee | 1650 | trace_migrate_set_state(MigrationStatus_str(new_state)); |
b05dc723 | 1651 | migrate_generate_event(new_state); |
51cf4c1a Z |
1652 | } |
1653 | } | |
1654 | ||
2833c59b JQ |
1655 | static void migrate_set_block_incremental(MigrationState *s, bool value) |
1656 | { | |
1657 | s->parameters.block_incremental = value; | |
1658 | } | |
1659 | ||
1660 | static void block_cleanup_parameters(MigrationState *s) | |
1661 | { | |
1662 | if (s->must_remove_block_options) { | |
1663 | /* setting to false can never fail */ | |
9eb1109c | 1664 | migrate_cap_set(MIGRATION_CAPABILITY_BLOCK, false, &error_abort); |
2833c59b JQ |
1665 | migrate_set_block_incremental(s, false); |
1666 | s->must_remove_block_options = false; | |
1667 | } | |
1668 | } | |
1669 | ||
fd392cfa | 1670 | static void migrate_fd_cleanup(MigrationState *s) |
065e2813 | 1671 | { |
bb1fadc4 PB |
1672 | qemu_bh_delete(s->cleanup_bh); |
1673 | s->cleanup_bh = NULL; | |
1674 | ||
83174765 PX |
1675 | g_free(s->hostname); |
1676 | s->hostname = NULL; | |
e3bf5e68 DH |
1677 | json_writer_free(s->vmdesc); |
1678 | s->vmdesc = NULL; | |
83174765 | 1679 | |
0ceccd85 PX |
1680 | qemu_savevm_state_cleanup(); |
1681 | ||
89a02a9f | 1682 | if (s->to_dst_file) { |
62df066f | 1683 | QEMUFile *tmp; |
f986c3d2 | 1684 | |
9013dca5 | 1685 | trace_migrate_fd_cleanup(); |
404a7c05 | 1686 | qemu_mutex_unlock_iothread(); |
1d34e4bf DDAG |
1687 | if (s->migration_thread_running) { |
1688 | qemu_thread_join(&s->thread); | |
1689 | s->migration_thread_running = false; | |
1690 | } | |
404a7c05 PB |
1691 | qemu_mutex_lock_iothread(); |
1692 | ||
1398b2e3 | 1693 | multifd_save_cleanup(); |
62df066f PX |
1694 | qemu_mutex_lock(&s->qemu_file_lock); |
1695 | tmp = s->to_dst_file; | |
89a02a9f | 1696 | s->to_dst_file = NULL; |
62df066f PX |
1697 | qemu_mutex_unlock(&s->qemu_file_lock); |
1698 | /* | |
1699 | * Close the file handle without the lock to make sure the | |
1700 | * critical section won't block for long. | |
1701 | */ | |
39675fff | 1702 | migration_ioc_unregister_yank_from_file(tmp); |
62df066f | 1703 | qemu_fclose(tmp); |
065e2813 AL |
1704 | } |
1705 | ||
36f62f11 PX |
1706 | if (s->postcopy_qemufile_src) { |
1707 | migration_ioc_unregister_yank_from_file(s->postcopy_qemufile_src); | |
1708 | qemu_fclose(s->postcopy_qemufile_src); | |
1709 | s->postcopy_qemufile_src = NULL; | |
1710 | } | |
1711 | ||
8f8d528e | 1712 | assert(!migration_is_active(s)); |
7a2c1721 | 1713 | |
94f5a437 | 1714 | if (s->state == MIGRATION_STATUS_CANCELLING) { |
48781e5b | 1715 | migrate_set_state(&s->state, MIGRATION_STATUS_CANCELLING, |
94f5a437 | 1716 | MIGRATION_STATUS_CANCELLED); |
7a2c1721 | 1717 | } |
a3fa1d78 | 1718 | |
87db1a7d JQ |
1719 | if (s->error) { |
1720 | /* It is used on info migrate. We can't free it */ | |
1721 | error_report_err(error_copy(s->error)); | |
1722 | } | |
a3fa1d78 | 1723 | notifier_list_notify(&migration_state_notifiers, s); |
2833c59b | 1724 | block_cleanup_parameters(s); |
b5eea99e | 1725 | yank_unregister_instance(MIGRATION_YANK_INSTANCE); |
065e2813 AL |
1726 | } |
1727 | ||
fd392cfa YK |
1728 | static void migrate_fd_cleanup_schedule(MigrationState *s) |
1729 | { | |
1730 | /* | |
1731 | * Ref the state for bh, because it may be called when | |
1732 | * there're already no other refs | |
1733 | */ | |
1734 | object_ref(OBJECT(s)); | |
1735 | qemu_bh_schedule(s->cleanup_bh); | |
1736 | } | |
1737 | ||
1738 | static void migrate_fd_cleanup_bh(void *opaque) | |
1739 | { | |
1740 | MigrationState *s = opaque; | |
1741 | migrate_fd_cleanup(s); | |
1742 | object_unref(OBJECT(s)); | |
1743 | } | |
1744 | ||
87db1a7d JQ |
1745 | void migrate_set_error(MigrationState *s, const Error *error) |
1746 | { | |
6e8a355d | 1747 | QEMU_LOCK_GUARD(&s->error_mutex); |
87db1a7d JQ |
1748 | if (!s->error) { |
1749 | s->error = error_copy(error); | |
1750 | } | |
87db1a7d JQ |
1751 | } |
1752 | ||
ca7bd082 PX |
1753 | static void migrate_error_free(MigrationState *s) |
1754 | { | |
1755 | QEMU_LOCK_GUARD(&s->error_mutex); | |
1756 | if (s->error) { | |
1757 | error_free(s->error); | |
1758 | s->error = NULL; | |
1759 | } | |
1760 | } | |
1761 | ||
d59ce6f3 | 1762 | void migrate_fd_error(MigrationState *s, const Error *error) |
065e2813 | 1763 | { |
25174055 | 1764 | trace_migrate_fd_error(error_get_pretty(error)); |
89a02a9f | 1765 | assert(s->to_dst_file == NULL); |
48781e5b HZ |
1766 | migrate_set_state(&s->state, MIGRATION_STATUS_SETUP, |
1767 | MIGRATION_STATUS_FAILED); | |
87db1a7d | 1768 | migrate_set_error(s, error); |
458cf28e JQ |
1769 | } |
1770 | ||
0edda1c4 | 1771 | static void migrate_fd_cancel(MigrationState *s) |
065e2813 | 1772 | { |
6f2b811a | 1773 | int old_state ; |
89a02a9f | 1774 | QEMUFile *f = migrate_get_current()->to_dst_file; |
9013dca5 | 1775 | trace_migrate_fd_cancel(); |
065e2813 | 1776 | |
43044ac0 PX |
1777 | WITH_QEMU_LOCK_GUARD(&s->qemu_file_lock) { |
1778 | if (s->rp_state.from_dst_file) { | |
1779 | /* shutdown the rp socket, so causing the rp thread to shutdown */ | |
1780 | qemu_file_shutdown(s->rp_state.from_dst_file); | |
1781 | } | |
70b20477 DDAG |
1782 | } |
1783 | ||
6f2b811a Z |
1784 | do { |
1785 | old_state = s->state; | |
392d87e2 | 1786 | if (!migration_is_running(old_state)) { |
6f2b811a Z |
1787 | break; |
1788 | } | |
a7b36b48 DDAG |
1789 | /* If the migration is paused, kick it out of the pause */ |
1790 | if (old_state == MIGRATION_STATUS_PRE_SWITCHOVER) { | |
1791 | qemu_sem_post(&s->pause_sem); | |
1792 | } | |
48781e5b | 1793 | migrate_set_state(&s->state, old_state, MIGRATION_STATUS_CANCELLING); |
31194731 | 1794 | } while (s->state != MIGRATION_STATUS_CANCELLING); |
a26ba26e DDAG |
1795 | |
1796 | /* | |
1797 | * If we're unlucky the migration code might be stuck somewhere in a | |
1798 | * send/write while the network has failed and is waiting to timeout; | |
1799 | * if we've got shutdown(2) available then we can force it to quit. | |
1800 | * The outgoing qemu file gets closed in migrate_fd_cleanup that is | |
1801 | * called in a bh, so there is no race against this cancel. | |
1802 | */ | |
31194731 | 1803 | if (s->state == MIGRATION_STATUS_CANCELLING && f) { |
a26ba26e DDAG |
1804 | qemu_file_shutdown(f); |
1805 | } | |
1d2acc31 HZ |
1806 | if (s->state == MIGRATION_STATUS_CANCELLING && s->block_inactive) { |
1807 | Error *local_err = NULL; | |
1808 | ||
3b717194 | 1809 | bdrv_activate_all(&local_err); |
1d2acc31 HZ |
1810 | if (local_err) { |
1811 | error_report_err(local_err); | |
1812 | } else { | |
1813 | s->block_inactive = false; | |
1814 | } | |
1815 | } | |
065e2813 AL |
1816 | } |
1817 | ||
99a0db9b GH |
1818 | void add_migration_state_change_notifier(Notifier *notify) |
1819 | { | |
1820 | notifier_list_add(&migration_state_notifiers, notify); | |
1821 | } | |
1822 | ||
1823 | void remove_migration_state_change_notifier(Notifier *notify) | |
1824 | { | |
31552529 | 1825 | notifier_remove(notify); |
99a0db9b GH |
1826 | } |
1827 | ||
02edd2e7 | 1828 | bool migration_in_setup(MigrationState *s) |
afe2df69 | 1829 | { |
31194731 | 1830 | return s->state == MIGRATION_STATUS_SETUP; |
afe2df69 GH |
1831 | } |
1832 | ||
7073693b | 1833 | bool migration_has_finished(MigrationState *s) |
99a0db9b | 1834 | { |
31194731 | 1835 | return s->state == MIGRATION_STATUS_COMPLETED; |
99a0db9b | 1836 | } |
0edda1c4 | 1837 | |
afe2df69 GH |
1838 | bool migration_has_failed(MigrationState *s) |
1839 | { | |
31194731 HZ |
1840 | return (s->state == MIGRATION_STATUS_CANCELLED || |
1841 | s->state == MIGRATION_STATUS_FAILED); | |
afe2df69 GH |
1842 | } |
1843 | ||
5727309d | 1844 | bool migration_in_postcopy(void) |
9ec055ae | 1845 | { |
5727309d JQ |
1846 | MigrationState *s = migrate_get_current(); |
1847 | ||
3748fef9 DDAG |
1848 | switch (s->state) { |
1849 | case MIGRATION_STATUS_POSTCOPY_ACTIVE: | |
1850 | case MIGRATION_STATUS_POSTCOPY_PAUSED: | |
1851 | case MIGRATION_STATUS_POSTCOPY_RECOVER: | |
1852 | return true; | |
1853 | default: | |
1854 | return false; | |
1855 | } | |
9ec055ae DDAG |
1856 | } |
1857 | ||
b82fc321 DDAG |
1858 | bool migration_in_postcopy_after_devices(MigrationState *s) |
1859 | { | |
5727309d | 1860 | return migration_in_postcopy() && s->postcopy_after_devices; |
b82fc321 DDAG |
1861 | } |
1862 | ||
06df2e69 DH |
1863 | bool migration_in_incoming_postcopy(void) |
1864 | { | |
1865 | PostcopyState ps = postcopy_state_get(); | |
1866 | ||
1867 | return ps >= POSTCOPY_INCOMING_DISCARD && ps < POSTCOPY_INCOMING_END; | |
1868 | } | |
1869 | ||
80fe315c DH |
1870 | bool migration_incoming_postcopy_advised(void) |
1871 | { | |
1872 | PostcopyState ps = postcopy_state_get(); | |
1873 | ||
1874 | return ps >= POSTCOPY_INCOMING_ADVISE && ps < POSTCOPY_INCOMING_END; | |
1875 | } | |
1876 | ||
1a8e44a8 AG |
1877 | bool migration_in_bg_snapshot(void) |
1878 | { | |
1879 | MigrationState *s = migrate_get_current(); | |
1880 | ||
1881 | return migrate_background_snapshot() && | |
1882 | migration_is_setup_or_active(s->state); | |
1883 | } | |
1884 | ||
fab35005 | 1885 | bool migration_is_idle(void) |
fe44dc91 | 1886 | { |
daff7f0b MA |
1887 | MigrationState *s = current_migration; |
1888 | ||
1889 | if (!s) { | |
1890 | return true; | |
1891 | } | |
fe44dc91 AA |
1892 | |
1893 | switch (s->state) { | |
1894 | case MIGRATION_STATUS_NONE: | |
1895 | case MIGRATION_STATUS_CANCELLED: | |
1896 | case MIGRATION_STATUS_COMPLETED: | |
1897 | case MIGRATION_STATUS_FAILED: | |
1898 | return true; | |
1899 | case MIGRATION_STATUS_SETUP: | |
1900 | case MIGRATION_STATUS_CANCELLING: | |
1901 | case MIGRATION_STATUS_ACTIVE: | |
1902 | case MIGRATION_STATUS_POSTCOPY_ACTIVE: | |
1903 | case MIGRATION_STATUS_COLO: | |
31e06077 DDAG |
1904 | case MIGRATION_STATUS_PRE_SWITCHOVER: |
1905 | case MIGRATION_STATUS_DEVICE: | |
c7e0acd5 | 1906 | case MIGRATION_STATUS_WAIT_UNPLUG: |
fe44dc91 AA |
1907 | return false; |
1908 | case MIGRATION_STATUS__MAX: | |
1909 | g_assert_not_reached(); | |
1910 | } | |
1911 | ||
1912 | return false; | |
1913 | } | |
1914 | ||
8f8d528e WY |
1915 | bool migration_is_active(MigrationState *s) |
1916 | { | |
1917 | return (s->state == MIGRATION_STATUS_ACTIVE || | |
1918 | s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE); | |
1919 | } | |
1920 | ||
3e0c8050 | 1921 | void migrate_init(MigrationState *s) |
0edda1c4 | 1922 | { |
389775d1 DDAG |
1923 | /* |
1924 | * Reinitialise all migration state, except | |
1925 | * parameters/capabilities that the user set, and | |
1926 | * locks. | |
1927 | */ | |
389775d1 | 1928 | s->cleanup_bh = 0; |
8518278a | 1929 | s->vm_start_bh = 0; |
89a02a9f | 1930 | s->to_dst_file = NULL; |
389775d1 | 1931 | s->state = MIGRATION_STATUS_NONE; |
389775d1 DDAG |
1932 | s->rp_state.from_dst_file = NULL; |
1933 | s->rp_state.error = false; | |
1934 | s->mbps = 0.0; | |
aecbfe9c | 1935 | s->pages_per_second = 0.0; |
389775d1 DDAG |
1936 | s->downtime = 0; |
1937 | s->expected_downtime = 0; | |
389775d1 | 1938 | s->setup_time = 0; |
389775d1 | 1939 | s->start_postcopy = false; |
b82fc321 | 1940 | s->postcopy_after_devices = false; |
389775d1 | 1941 | s->migration_thread_running = false; |
d59ce6f3 DB |
1942 | error_free(s->error); |
1943 | s->error = NULL; | |
d8053e73 | 1944 | s->hostname = NULL; |
389775d1 | 1945 | |
48781e5b | 1946 | migrate_set_state(&s->state, MIGRATION_STATUS_NONE, MIGRATION_STATUS_SETUP); |
0edda1c4 | 1947 | |
4af246a3 PX |
1948 | s->start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); |
1949 | s->total_time = 0; | |
7287cbd4 | 1950 | s->vm_was_running = false; |
b15df1ae PX |
1951 | s->iteration_initial_bytes = 0; |
1952 | s->threshold_size = 0; | |
0edda1c4 | 1953 | } |
cab30143 | 1954 | |
60fd6801 | 1955 | int migrate_add_blocker_internal(Error *reason, Error **errp) |
fa2756b7 | 1956 | { |
4c170330 PX |
1957 | /* Snapshots are similar to migrations, so check RUN_STATE_SAVE_VM too. */ |
1958 | if (runstate_check(RUN_STATE_SAVE_VM) || !migration_is_idle()) { | |
1959 | error_propagate_prepend(errp, error_copy(reason), | |
1960 | "disallowing migration blocker " | |
1961 | "(migration/snapshot in progress) for: "); | |
1962 | return -EBUSY; | |
fe44dc91 AA |
1963 | } |
1964 | ||
4c170330 PX |
1965 | migration_blockers = g_slist_prepend(migration_blockers, reason); |
1966 | return 0; | |
fa2756b7 AL |
1967 | } |
1968 | ||
60fd6801 PX |
1969 | int migrate_add_blocker(Error *reason, Error **errp) |
1970 | { | |
1971 | if (only_migratable) { | |
1972 | error_propagate_prepend(errp, error_copy(reason), | |
1973 | "disallowing migration blocker " | |
1974 | "(--only-migratable) for: "); | |
1975 | return -EACCES; | |
1976 | } | |
1977 | ||
1978 | return migrate_add_blocker_internal(reason, errp); | |
1979 | } | |
1980 | ||
fa2756b7 AL |
1981 | void migrate_del_blocker(Error *reason) |
1982 | { | |
1983 | migration_blockers = g_slist_remove(migration_blockers, reason); | |
1984 | } | |
1985 | ||
bf1ae1f4 DDAG |
1986 | void qmp_migrate_incoming(const char *uri, Error **errp) |
1987 | { | |
1988 | Error *local_err = NULL; | |
4debb5f5 | 1989 | static bool once = true; |
bf1ae1f4 | 1990 | |
4debb5f5 DDAG |
1991 | if (!once) { |
1992 | error_setg(errp, "The incoming migration has already been started"); | |
603d5a42 | 1993 | return; |
4debb5f5 | 1994 | } |
e69d50d6 PB |
1995 | if (!runstate_check(RUN_STATE_INMIGRATE)) { |
1996 | error_setg(errp, "'-incoming' was not specified on the command line"); | |
1997 | return; | |
1998 | } | |
bf1ae1f4 | 1999 | |
cc48c587 PX |
2000 | if (!yank_register_instance(MIGRATION_YANK_INSTANCE, errp)) { |
2001 | return; | |
2002 | } | |
2003 | ||
bf1ae1f4 DDAG |
2004 | qemu_start_incoming_migration(uri, &local_err); |
2005 | ||
2006 | if (local_err) { | |
cc48c587 | 2007 | yank_unregister_instance(MIGRATION_YANK_INSTANCE); |
bf1ae1f4 DDAG |
2008 | error_propagate(errp, local_err); |
2009 | return; | |
2010 | } | |
2011 | ||
4debb5f5 | 2012 | once = false; |
bf1ae1f4 DDAG |
2013 | } |
2014 | ||
02affd41 PX |
2015 | void qmp_migrate_recover(const char *uri, Error **errp) |
2016 | { | |
2017 | MigrationIncomingState *mis = migration_incoming_get_current(); | |
2018 | ||
b7f9afd4 PX |
2019 | /* |
2020 | * Don't even bother to use ERRP_GUARD() as it _must_ always be set by | |
2021 | * callers (no one should ignore a recover failure); if there is, it's a | |
2022 | * programming error. | |
2023 | */ | |
2024 | assert(errp); | |
2025 | ||
02affd41 PX |
2026 | if (mis->state != MIGRATION_STATUS_POSTCOPY_PAUSED) { |
2027 | error_setg(errp, "Migrate recover can only be run " | |
2028 | "when postcopy is paused."); | |
2029 | return; | |
2030 | } | |
2031 | ||
08401c04 PX |
2032 | /* If there's an existing transport, release it */ |
2033 | migration_incoming_transport_cleanup(mis); | |
02affd41 PX |
2034 | |
2035 | /* | |
2036 | * Note that this call will never start a real migration; it will | |
2037 | * only re-setup the migration stream and poke existing migration | |
2038 | * to continue using that newly established channel. | |
2039 | */ | |
2040 | qemu_start_incoming_migration(uri, errp); | |
2041 | } | |
2042 | ||
bfbf89c2 PX |
2043 | void qmp_migrate_pause(Error **errp) |
2044 | { | |
2045 | MigrationState *ms = migrate_get_current(); | |
2046 | MigrationIncomingState *mis = migration_incoming_get_current(); | |
2047 | int ret; | |
2048 | ||
2049 | if (ms->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) { | |
2050 | /* Source side, during postcopy */ | |
2051 | qemu_mutex_lock(&ms->qemu_file_lock); | |
2052 | ret = qemu_file_shutdown(ms->to_dst_file); | |
2053 | qemu_mutex_unlock(&ms->qemu_file_lock); | |
2054 | if (ret) { | |
2055 | error_setg(errp, "Failed to pause source migration"); | |
2056 | } | |
2057 | return; | |
2058 | } | |
2059 | ||
2060 | if (mis->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) { | |
2061 | ret = qemu_file_shutdown(mis->from_src_file); | |
2062 | if (ret) { | |
2063 | error_setg(errp, "Failed to pause destination migration"); | |
2064 | } | |
2065 | return; | |
2066 | } | |
2067 | ||
2068 | error_setg(errp, "migrate-pause is currently only supported " | |
2069 | "during postcopy-active state"); | |
2070 | } | |
2071 | ||
24f3902b GK |
2072 | bool migration_is_blocked(Error **errp) |
2073 | { | |
2074 | if (qemu_savevm_state_blocked(errp)) { | |
2075 | return true; | |
2076 | } | |
2077 | ||
2078 | if (migration_blockers) { | |
250561e1 | 2079 | error_propagate(errp, error_copy(migration_blockers->data)); |
24f3902b GK |
2080 | return true; |
2081 | } | |
2082 | ||
2083 | return false; | |
2084 | } | |
2085 | ||
d3e35b8f PX |
2086 | /* Returns true if continue to migrate, or false if error detected */ |
2087 | static bool migrate_prepare(MigrationState *s, bool blk, bool blk_inc, | |
2088 | bool resume, Error **errp) | |
cab30143 | 2089 | { |
be7059cd | 2090 | Error *local_err = NULL; |
d3e35b8f PX |
2091 | |
2092 | if (resume) { | |
2093 | if (s->state != MIGRATION_STATUS_POSTCOPY_PAUSED) { | |
2094 | error_setg(errp, "Cannot resume if there is no " | |
2095 | "paused migration"); | |
2096 | return false; | |
2097 | } | |
97ca211c PX |
2098 | |
2099 | /* | |
2100 | * Postcopy recovery won't work well with release-ram | |
2101 | * capability since release-ram will drop the page buffer as | |
2102 | * long as the page is put into the send buffer. So if there | |
2103 | * is a network failure happened, any page buffers that have | |
2104 | * not yet reached the destination VM but have already been | |
2105 | * sent from the source VM will be lost forever. Let's refuse | |
2106 | * the client from resuming such a postcopy migration. | |
2107 | * Luckily release-ram was designed to only be used when src | |
2108 | * and destination VMs are on the same host, so it should be | |
2109 | * fine. | |
2110 | */ | |
2111 | if (migrate_release_ram()) { | |
2112 | error_setg(errp, "Postcopy recovery cannot work " | |
2113 | "when release-ram capability is set"); | |
2114 | return false; | |
2115 | } | |
2116 | ||
d3e35b8f PX |
2117 | /* This is a resume, skip init status */ |
2118 | return true; | |
2119 | } | |
cab30143 | 2120 | |
392d87e2 | 2121 | if (migration_is_running(s->state)) { |
c6bd8c70 | 2122 | error_setg(errp, QERR_MIGRATION_ACTIVE); |
d3e35b8f | 2123 | return false; |
cab30143 | 2124 | } |
d3e35b8f | 2125 | |
ca99993a DDAG |
2126 | if (runstate_check(RUN_STATE_INMIGRATE)) { |
2127 | error_setg(errp, "Guest is waiting for an incoming migration"); | |
d3e35b8f | 2128 | return false; |
ca99993a DDAG |
2129 | } |
2130 | ||
36d0fe65 T |
2131 | if (runstate_check(RUN_STATE_POSTMIGRATE)) { |
2132 | error_setg(errp, "Can't migrate the vm that was paused due to " | |
2133 | "previous migration"); | |
2134 | return false; | |
2135 | } | |
2136 | ||
24f3902b | 2137 | if (migration_is_blocked(errp)) { |
d3e35b8f | 2138 | return false; |
fa2756b7 AL |
2139 | } |
2140 | ||
d3e35b8f | 2141 | if (blk || blk_inc) { |
5e804644 | 2142 | if (migrate_colo()) { |
3ba02445 RL |
2143 | error_setg(errp, "No disk migration is required in COLO mode"); |
2144 | return false; | |
2145 | } | |
6f8be708 | 2146 | if (migrate_block() || migrate_block_incremental()) { |
2833c59b JQ |
2147 | error_setg(errp, "Command options are incompatible with " |
2148 | "current migration capabilities"); | |
d3e35b8f | 2149 | return false; |
2833c59b | 2150 | } |
9eb1109c | 2151 | if (!migrate_cap_set(MIGRATION_CAPABILITY_BLOCK, true, &local_err)) { |
2833c59b | 2152 | error_propagate(errp, local_err); |
d3e35b8f | 2153 | return false; |
2833c59b JQ |
2154 | } |
2155 | s->must_remove_block_options = true; | |
2156 | } | |
2157 | ||
d3e35b8f | 2158 | if (blk_inc) { |
2833c59b JQ |
2159 | migrate_set_block_incremental(s, true); |
2160 | } | |
2161 | ||
3e0c8050 | 2162 | migrate_init(s); |
87f3bd87 | 2163 | /* |
fa0b31d5 | 2164 | * set ram_counters compression_counters memory to zero for a |
87f3bd87 IR |
2165 | * new migration |
2166 | */ | |
2167 | memset(&ram_counters, 0, sizeof(ram_counters)); | |
fa0b31d5 | 2168 | memset(&compression_counters, 0, sizeof(compression_counters)); |
cab30143 | 2169 | |
d3e35b8f PX |
2170 | return true; |
2171 | } | |
2172 | ||
2173 | void qmp_migrate(const char *uri, bool has_blk, bool blk, | |
2174 | bool has_inc, bool inc, bool has_detach, bool detach, | |
2175 | bool has_resume, bool resume, Error **errp) | |
2176 | { | |
2177 | Error *local_err = NULL; | |
2178 | MigrationState *s = migrate_get_current(); | |
d658f65c | 2179 | const char *p = NULL; |
d3e35b8f | 2180 | |
d6f74fd1 PX |
2181 | /* URI is not suitable for migration? */ |
2182 | if (!migration_channels_and_uri_compatible(uri, errp)) { | |
2183 | return; | |
2184 | } | |
2185 | ||
d3e35b8f PX |
2186 | if (!migrate_prepare(s, has_blk && blk, has_inc && inc, |
2187 | has_resume && resume, errp)) { | |
2188 | /* Error detected, put into errp */ | |
2189 | return; | |
2190 | } | |
2191 | ||
b5eea99e LS |
2192 | if (!(has_resume && resume)) { |
2193 | if (!yank_register_instance(MIGRATION_YANK_INSTANCE, errp)) { | |
2194 | return; | |
2195 | } | |
2196 | } | |
2197 | ||
d658f65c | 2198 | if (strstart(uri, "tcp:", &p) || |
9ba3b2ba LM |
2199 | strstart(uri, "unix:", NULL) || |
2200 | strstart(uri, "vsock:", NULL)) { | |
d658f65c | 2201 | socket_start_outgoing_migration(s, p ? p : uri, &local_err); |
2da776db | 2202 | #ifdef CONFIG_RDMA |
41310c68 | 2203 | } else if (strstart(uri, "rdma:", &p)) { |
2da776db MH |
2204 | rdma_start_outgoing_migration(s, p, &local_err); |
2205 | #endif | |
cab30143 | 2206 | } else if (strstart(uri, "exec:", &p)) { |
f37afb5a | 2207 | exec_start_outgoing_migration(s, p, &local_err); |
cab30143 | 2208 | } else if (strstart(uri, "fd:", &p)) { |
f37afb5a | 2209 | fd_start_outgoing_migration(s, p, &local_err); |
99a0db9b | 2210 | } else { |
b5eea99e LS |
2211 | if (!(has_resume && resume)) { |
2212 | yank_unregister_instance(MIGRATION_YANK_INSTANCE); | |
2213 | } | |
c6bd8c70 MA |
2214 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "uri", |
2215 | "a valid migration protocol"); | |
48781e5b HZ |
2216 | migrate_set_state(&s->state, MIGRATION_STATUS_SETUP, |
2217 | MIGRATION_STATUS_FAILED); | |
09576e74 | 2218 | block_cleanup_parameters(s); |
e1c37d0e | 2219 | return; |
cab30143 JQ |
2220 | } |
2221 | ||
f37afb5a | 2222 | if (local_err) { |
b5eea99e LS |
2223 | if (!(has_resume && resume)) { |
2224 | yank_unregister_instance(MIGRATION_YANK_INSTANCE); | |
2225 | } | |
d59ce6f3 | 2226 | migrate_fd_error(s, local_err); |
f37afb5a | 2227 | error_propagate(errp, local_err); |
e1c37d0e | 2228 | return; |
1299c631 | 2229 | } |
cab30143 JQ |
2230 | } |
2231 | ||
6cdedb07 | 2232 | void qmp_migrate_cancel(Error **errp) |
cab30143 | 2233 | { |
458fecca | 2234 | migration_cancel(NULL); |
cab30143 JQ |
2235 | } |
2236 | ||
89cfc02c DDAG |
2237 | void qmp_migrate_continue(MigrationStatus state, Error **errp) |
2238 | { | |
2239 | MigrationState *s = migrate_get_current(); | |
2240 | if (s->state != state) { | |
2241 | error_setg(errp, "Migration not in expected state: %s", | |
2242 | MigrationStatus_str(s->state)); | |
2243 | return; | |
2244 | } | |
2245 | qemu_sem_post(&s->pause_sem); | |
2246 | } | |
2247 | ||
d2fafb6a LB |
2248 | int migrate_use_tls(void) |
2249 | { | |
2250 | MigrationState *s; | |
2251 | ||
2252 | s = migrate_get_current(); | |
2253 | ||
2254 | return s->parameters.tls_creds && *s->parameters.tls_creds; | |
2255 | } | |
2256 | ||
70b20477 DDAG |
2257 | /* migration thread support */ |
2258 | /* | |
2259 | * Something bad happened to the RP stream, mark an error | |
2260 | * The caller shall print or trace something to indicate why | |
2261 | */ | |
2262 | static void mark_source_rp_bad(MigrationState *s) | |
2263 | { | |
2264 | s->rp_state.error = true; | |
2265 | } | |
2266 | ||
2267 | static struct rp_cmd_args { | |
2268 | ssize_t len; /* -1 = variable */ | |
2269 | const char *name; | |
2270 | } rp_cmd_args[] = { | |
2271 | [MIG_RP_MSG_INVALID] = { .len = -1, .name = "INVALID" }, | |
2272 | [MIG_RP_MSG_SHUT] = { .len = 4, .name = "SHUT" }, | |
2273 | [MIG_RP_MSG_PONG] = { .len = 4, .name = "PONG" }, | |
1e2d90eb DDAG |
2274 | [MIG_RP_MSG_REQ_PAGES] = { .len = 12, .name = "REQ_PAGES" }, |
2275 | [MIG_RP_MSG_REQ_PAGES_ID] = { .len = -1, .name = "REQ_PAGES_ID" }, | |
a335debb | 2276 | [MIG_RP_MSG_RECV_BITMAP] = { .len = -1, .name = "RECV_BITMAP" }, |
13955b89 | 2277 | [MIG_RP_MSG_RESUME_ACK] = { .len = 4, .name = "RESUME_ACK" }, |
70b20477 DDAG |
2278 | [MIG_RP_MSG_MAX] = { .len = -1, .name = "MAX" }, |
2279 | }; | |
2280 | ||
1e2d90eb DDAG |
2281 | /* |
2282 | * Process a request for pages received on the return path, | |
2283 | * We're allowed to send more than requested (e.g. to round to our page size) | |
2284 | * and we don't need to send pages that have already been sent. | |
2285 | */ | |
2286 | static void migrate_handle_rp_req_pages(MigrationState *ms, const char* rbname, | |
2287 | ram_addr_t start, size_t len) | |
2288 | { | |
8e3b0cbb | 2289 | long our_host_ps = qemu_real_host_page_size(); |
6c595cde | 2290 | |
1e2d90eb | 2291 | trace_migrate_handle_rp_req_pages(rbname, start, len); |
6c595cde DDAG |
2292 | |
2293 | /* | |
2294 | * Since we currently insist on matching page sizes, just sanity check | |
2295 | * we're being asked for whole host pages. | |
2296 | */ | |
7648297d DH |
2297 | if (!QEMU_IS_ALIGNED(start, our_host_ps) || |
2298 | !QEMU_IS_ALIGNED(len, our_host_ps)) { | |
6c595cde DDAG |
2299 | error_report("%s: Misaligned page request, start: " RAM_ADDR_FMT |
2300 | " len: %zd", __func__, start, len); | |
2301 | mark_source_rp_bad(ms); | |
2302 | return; | |
2303 | } | |
2304 | ||
96506894 | 2305 | if (ram_save_queue_pages(rbname, start, len)) { |
6c595cde DDAG |
2306 | mark_source_rp_bad(ms); |
2307 | } | |
1e2d90eb DDAG |
2308 | } |
2309 | ||
14b1742e PX |
2310 | /* Return true to retry, false to quit */ |
2311 | static bool postcopy_pause_return_path_thread(MigrationState *s) | |
2312 | { | |
2313 | trace_postcopy_pause_return_path(); | |
2314 | ||
2315 | qemu_sem_wait(&s->postcopy_pause_rp_sem); | |
2316 | ||
2317 | trace_postcopy_pause_return_path_continued(); | |
2318 | ||
2319 | return true; | |
2320 | } | |
2321 | ||
a335debb PX |
2322 | static int migrate_handle_rp_recv_bitmap(MigrationState *s, char *block_name) |
2323 | { | |
2324 | RAMBlock *block = qemu_ram_block_by_name(block_name); | |
2325 | ||
2326 | if (!block) { | |
2327 | error_report("%s: invalid block name '%s'", __func__, block_name); | |
2328 | return -EINVAL; | |
2329 | } | |
2330 | ||
2331 | /* Fetch the received bitmap and refresh the dirty bitmap */ | |
2332 | return ram_dirty_bitmap_reload(s, block); | |
2333 | } | |
2334 | ||
13955b89 PX |
2335 | static int migrate_handle_rp_resume_ack(MigrationState *s, uint32_t value) |
2336 | { | |
2337 | trace_source_return_path_thread_resume_ack(value); | |
2338 | ||
2339 | if (value != MIGRATION_RESUME_ACK_VALUE) { | |
2340 | error_report("%s: illegal resume_ack value %"PRIu32, | |
2341 | __func__, value); | |
2342 | return -1; | |
2343 | } | |
2344 | ||
2345 | /* Now both sides are active. */ | |
2346 | migrate_set_state(&s->state, MIGRATION_STATUS_POSTCOPY_RECOVER, | |
2347 | MIGRATION_STATUS_POSTCOPY_ACTIVE); | |
2348 | ||
94190696 PX |
2349 | /* Notify send thread that time to continue send pages */ |
2350 | qemu_sem_post(&s->rp_state.rp_sem); | |
13955b89 PX |
2351 | |
2352 | return 0; | |
2353 | } | |
2354 | ||
93589827 PX |
2355 | /* |
2356 | * Release ms->rp_state.from_dst_file (and postcopy_qemufile_src if | |
2357 | * existed) in a safe way. | |
2358 | */ | |
2359 | static void migration_release_dst_files(MigrationState *ms) | |
43044ac0 PX |
2360 | { |
2361 | QEMUFile *file; | |
2362 | ||
2363 | WITH_QEMU_LOCK_GUARD(&ms->qemu_file_lock) { | |
2364 | /* | |
2365 | * Reset the from_dst_file pointer first before releasing it, as we | |
2366 | * can't block within lock section | |
2367 | */ | |
2368 | file = ms->rp_state.from_dst_file; | |
2369 | ms->rp_state.from_dst_file = NULL; | |
2370 | } | |
2371 | ||
93589827 PX |
2372 | /* |
2373 | * Do the same to postcopy fast path socket too if there is. No | |
2374 | * locking needed because this qemufile should only be managed by | |
2375 | * return path thread. | |
2376 | */ | |
2377 | if (ms->postcopy_qemufile_src) { | |
2378 | migration_ioc_unregister_yank_from_file(ms->postcopy_qemufile_src); | |
2379 | qemu_file_shutdown(ms->postcopy_qemufile_src); | |
2380 | qemu_fclose(ms->postcopy_qemufile_src); | |
2381 | ms->postcopy_qemufile_src = NULL; | |
2382 | } | |
2383 | ||
43044ac0 PX |
2384 | qemu_fclose(file); |
2385 | } | |
2386 | ||
70b20477 DDAG |
2387 | /* |
2388 | * Handles messages sent on the return path towards the source VM | |
2389 | * | |
2390 | */ | |
2391 | static void *source_return_path_thread(void *opaque) | |
2392 | { | |
2393 | MigrationState *ms = opaque; | |
2394 | QEMUFile *rp = ms->rp_state.from_dst_file; | |
2395 | uint16_t header_len, header_type; | |
568b01ca | 2396 | uint8_t buf[512]; |
70b20477 | 2397 | uint32_t tmp32, sibling_error; |
1e2d90eb DDAG |
2398 | ram_addr_t start = 0; /* =0 to silence warning */ |
2399 | size_t len = 0, expected_len; | |
70b20477 DDAG |
2400 | int res; |
2401 | ||
2402 | trace_source_return_path_thread_entry(); | |
74637e6f | 2403 | rcu_register_thread(); |
14b1742e PX |
2404 | |
2405 | retry: | |
70b20477 DDAG |
2406 | while (!ms->rp_state.error && !qemu_file_get_error(rp) && |
2407 | migration_is_setup_or_active(ms->state)) { | |
2408 | trace_source_return_path_thread_loop_top(); | |
2409 | header_type = qemu_get_be16(rp); | |
2410 | header_len = qemu_get_be16(rp); | |
2411 | ||
7a9ddfbf PX |
2412 | if (qemu_file_get_error(rp)) { |
2413 | mark_source_rp_bad(ms); | |
2414 | goto out; | |
2415 | } | |
2416 | ||
70b20477 DDAG |
2417 | if (header_type >= MIG_RP_MSG_MAX || |
2418 | header_type == MIG_RP_MSG_INVALID) { | |
2419 | error_report("RP: Received invalid message 0x%04x length 0x%04x", | |
ed8b2828 | 2420 | header_type, header_len); |
70b20477 DDAG |
2421 | mark_source_rp_bad(ms); |
2422 | goto out; | |
2423 | } | |
2424 | ||
2425 | if ((rp_cmd_args[header_type].len != -1 && | |
2426 | header_len != rp_cmd_args[header_type].len) || | |
568b01ca | 2427 | header_len > sizeof(buf)) { |
70b20477 | 2428 | error_report("RP: Received '%s' message (0x%04x) with" |
ed8b2828 MZ |
2429 | "incorrect length %d expecting %zu", |
2430 | rp_cmd_args[header_type].name, header_type, header_len, | |
2431 | (size_t)rp_cmd_args[header_type].len); | |
70b20477 DDAG |
2432 | mark_source_rp_bad(ms); |
2433 | goto out; | |
2434 | } | |
2435 | ||
2436 | /* We know we've got a valid header by this point */ | |
2437 | res = qemu_get_buffer(rp, buf, header_len); | |
2438 | if (res != header_len) { | |
2439 | error_report("RP: Failed reading data for message 0x%04x" | |
2440 | " read %d expected %d", | |
2441 | header_type, res, header_len); | |
2442 | mark_source_rp_bad(ms); | |
2443 | goto out; | |
2444 | } | |
2445 | ||
2446 | /* OK, we have the message and the data */ | |
2447 | switch (header_type) { | |
2448 | case MIG_RP_MSG_SHUT: | |
4d885131 | 2449 | sibling_error = ldl_be_p(buf); |
70b20477 DDAG |
2450 | trace_source_return_path_thread_shut(sibling_error); |
2451 | if (sibling_error) { | |
2452 | error_report("RP: Sibling indicated error %d", sibling_error); | |
2453 | mark_source_rp_bad(ms); | |
2454 | } | |
2455 | /* | |
2456 | * We'll let the main thread deal with closing the RP | |
2457 | * we could do a shutdown(2) on it, but we're the only user | |
2458 | * anyway, so there's nothing gained. | |
2459 | */ | |
2460 | goto out; | |
2461 | ||
2462 | case MIG_RP_MSG_PONG: | |
4d885131 | 2463 | tmp32 = ldl_be_p(buf); |
70b20477 | 2464 | trace_source_return_path_thread_pong(tmp32); |
b28fb582 | 2465 | qemu_sem_post(&ms->rp_state.rp_pong_acks); |
70b20477 DDAG |
2466 | break; |
2467 | ||
1e2d90eb | 2468 | case MIG_RP_MSG_REQ_PAGES: |
4d885131 PM |
2469 | start = ldq_be_p(buf); |
2470 | len = ldl_be_p(buf + 8); | |
1e2d90eb DDAG |
2471 | migrate_handle_rp_req_pages(ms, NULL, start, len); |
2472 | break; | |
2473 | ||
2474 | case MIG_RP_MSG_REQ_PAGES_ID: | |
2475 | expected_len = 12 + 1; /* header + termination */ | |
2476 | ||
2477 | if (header_len >= expected_len) { | |
4d885131 PM |
2478 | start = ldq_be_p(buf); |
2479 | len = ldl_be_p(buf + 8); | |
1e2d90eb DDAG |
2480 | /* Now we expect an idstr */ |
2481 | tmp32 = buf[12]; /* Length of the following idstr */ | |
2482 | buf[13 + tmp32] = '\0'; | |
2483 | expected_len += tmp32; | |
2484 | } | |
2485 | if (header_len != expected_len) { | |
2486 | error_report("RP: Req_Page_id with length %d expecting %zd", | |
ed8b2828 | 2487 | header_len, expected_len); |
1e2d90eb DDAG |
2488 | mark_source_rp_bad(ms); |
2489 | goto out; | |
2490 | } | |
2491 | migrate_handle_rp_req_pages(ms, (char *)&buf[13], start, len); | |
2492 | break; | |
2493 | ||
a335debb PX |
2494 | case MIG_RP_MSG_RECV_BITMAP: |
2495 | if (header_len < 1) { | |
2496 | error_report("%s: missing block name", __func__); | |
2497 | mark_source_rp_bad(ms); | |
2498 | goto out; | |
2499 | } | |
2500 | /* Format: len (1B) + idstr (<255B). This ends the idstr. */ | |
2501 | buf[buf[0] + 1] = '\0'; | |
2502 | if (migrate_handle_rp_recv_bitmap(ms, (char *)(buf + 1))) { | |
2503 | mark_source_rp_bad(ms); | |
2504 | goto out; | |
2505 | } | |
2506 | break; | |
2507 | ||
13955b89 PX |
2508 | case MIG_RP_MSG_RESUME_ACK: |
2509 | tmp32 = ldl_be_p(buf); | |
2510 | if (migrate_handle_rp_resume_ack(ms, tmp32)) { | |
2511 | mark_source_rp_bad(ms); | |
2512 | goto out; | |
2513 | } | |
2514 | break; | |
2515 | ||
70b20477 DDAG |
2516 | default: |
2517 | break; | |
2518 | } | |
2519 | } | |
14b1742e PX |
2520 | |
2521 | out: | |
2522 | res = qemu_file_get_error(rp); | |
2523 | if (res) { | |
d5c8f2af | 2524 | if (res && migration_in_postcopy()) { |
14b1742e PX |
2525 | /* |
2526 | * Maybe there is something we can do: it looks like a | |
2527 | * network down issue, and we pause for a recovery. | |
2528 | */ | |
93589827 | 2529 | migration_release_dst_files(ms); |
2e3e3da3 | 2530 | rp = NULL; |
14b1742e | 2531 | if (postcopy_pause_return_path_thread(ms)) { |
43044ac0 PX |
2532 | /* |
2533 | * Reload rp, reset the rest. Referencing it is safe since | |
2534 | * it's reset only by us above, or when migration completes | |
2535 | */ | |
2e3e3da3 | 2536 | rp = ms->rp_state.from_dst_file; |
14b1742e PX |
2537 | ms->rp_state.error = false; |
2538 | goto retry; | |
2539 | } | |
2540 | } | |
2541 | ||
70b20477 DDAG |
2542 | trace_source_return_path_thread_bad_end(); |
2543 | mark_source_rp_bad(ms); | |
2544 | } | |
2545 | ||
2546 | trace_source_return_path_thread_end(); | |
93589827 | 2547 | migration_release_dst_files(ms); |
74637e6f | 2548 | rcu_unregister_thread(); |
70b20477 DDAG |
2549 | return NULL; |
2550 | } | |
2551 | ||
d3e35b8f PX |
2552 | static int open_return_path_on_source(MigrationState *ms, |
2553 | bool create_thread) | |
70b20477 | 2554 | { |
89a02a9f | 2555 | ms->rp_state.from_dst_file = qemu_file_get_return_path(ms->to_dst_file); |
70b20477 DDAG |
2556 | if (!ms->rp_state.from_dst_file) { |
2557 | return -1; | |
2558 | } | |
2559 | ||
2560 | trace_open_return_path_on_source(); | |
d3e35b8f PX |
2561 | |
2562 | if (!create_thread) { | |
2563 | /* We're done */ | |
2564 | return 0; | |
2565 | } | |
2566 | ||
70b20477 DDAG |
2567 | qemu_thread_create(&ms->rp_state.rp_thread, "return path", |
2568 | source_return_path_thread, ms, QEMU_THREAD_JOINABLE); | |
53021ea1 | 2569 | ms->rp_state.rp_thread_created = true; |
70b20477 DDAG |
2570 | |
2571 | trace_open_return_path_on_source_continue(); | |
2572 | ||
2573 | return 0; | |
2574 | } | |
2575 | ||
70b20477 DDAG |
2576 | /* Returns 0 if the RP was ok, otherwise there was an error on the RP */ |
2577 | static int await_return_path_close_on_source(MigrationState *ms) | |
2578 | { | |
2579 | /* | |
2580 | * If this is a normal exit then the destination will send a SHUT and the | |
2581 | * rp_thread will exit, however if there's an error we need to cause | |
2582 | * it to exit. | |
2583 | */ | |
89a02a9f | 2584 | if (qemu_file_get_error(ms->to_dst_file) && ms->rp_state.from_dst_file) { |
70b20477 DDAG |
2585 | /* |
2586 | * shutdown(2), if we have it, will cause it to unblock if it's stuck | |
2587 | * waiting for the destination. | |
2588 | */ | |
2589 | qemu_file_shutdown(ms->rp_state.from_dst_file); | |
2590 | mark_source_rp_bad(ms); | |
2591 | } | |
2592 | trace_await_return_path_close_on_source_joining(); | |
2593 | qemu_thread_join(&ms->rp_state.rp_thread); | |
53021ea1 | 2594 | ms->rp_state.rp_thread_created = false; |
70b20477 DDAG |
2595 | trace_await_return_path_close_on_source_close(); |
2596 | return ms->rp_state.error; | |
2597 | } | |
2598 | ||
5655aab0 PX |
2599 | static inline void |
2600 | migration_wait_main_channel(MigrationState *ms) | |
2601 | { | |
2602 | /* Wait until one PONG message received */ | |
2603 | qemu_sem_wait(&ms->rp_state.rp_pong_acks); | |
2604 | } | |
2605 | ||
1d34e4bf DDAG |
2606 | /* |
2607 | * Switch from normal iteration to postcopy | |
2608 | * Returns non-0 on error | |
2609 | */ | |
7287cbd4 | 2610 | static int postcopy_start(MigrationState *ms) |
1d34e4bf DDAG |
2611 | { |
2612 | int ret; | |
61b67d47 DB |
2613 | QIOChannelBuffer *bioc; |
2614 | QEMUFile *fb; | |
1d34e4bf | 2615 | int64_t time_at_stop = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); |
7e555c6c | 2616 | int64_t bandwidth = migrate_max_postcopy_bandwidth(); |
ef8d6488 | 2617 | bool restart_block = false; |
0331c8ca | 2618 | int cur_state = MIGRATION_STATUS_ACTIVE; |
d0edb8a1 | 2619 | |
5655aab0 PX |
2620 | if (migrate_postcopy_preempt()) { |
2621 | migration_wait_main_channel(ms); | |
2622 | if (postcopy_preempt_establish_channel(ms)) { | |
2623 | migrate_set_state(&ms->state, ms->state, MIGRATION_STATUS_FAILED); | |
2624 | return -1; | |
2625 | } | |
d0edb8a1 PX |
2626 | } |
2627 | ||
0331c8ca DDAG |
2628 | if (!migrate_pause_before_switchover()) { |
2629 | migrate_set_state(&ms->state, MIGRATION_STATUS_ACTIVE, | |
2630 | MIGRATION_STATUS_POSTCOPY_ACTIVE); | |
2631 | } | |
1d34e4bf DDAG |
2632 | |
2633 | trace_postcopy_start(); | |
2634 | qemu_mutex_lock_iothread(); | |
2635 | trace_postcopy_start_set_run(); | |
2636 | ||
fb064112 | 2637 | qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER, NULL); |
1d34e4bf DDAG |
2638 | global_state_store(); |
2639 | ret = vm_stop_force_state(RUN_STATE_FINISH_MIGRATE); | |
76b1c7fe KW |
2640 | if (ret < 0) { |
2641 | goto fail; | |
2642 | } | |
1d34e4bf | 2643 | |
0331c8ca DDAG |
2644 | ret = migration_maybe_pause(ms, &cur_state, |
2645 | MIGRATION_STATUS_POSTCOPY_ACTIVE); | |
2646 | if (ret < 0) { | |
2647 | goto fail; | |
2648 | } | |
2649 | ||
76b1c7fe | 2650 | ret = bdrv_inactivate_all(); |
1d34e4bf DDAG |
2651 | if (ret < 0) { |
2652 | goto fail; | |
2653 | } | |
ef8d6488 | 2654 | restart_block = true; |
1d34e4bf | 2655 | |
1c0d249d DDAG |
2656 | /* |
2657 | * Cause any non-postcopiable, but iterative devices to | |
2658 | * send out their final data. | |
2659 | */ | |
a1fbe750 | 2660 | qemu_savevm_state_complete_precopy(ms->to_dst_file, true, false); |
1c0d249d | 2661 | |
1d34e4bf DDAG |
2662 | /* |
2663 | * in Finish migrate and with the io-lock held everything should | |
2664 | * be quiet, but we've potentially still got dirty pages and we | |
2665 | * need to tell the destination to throw any pages it's already received | |
2666 | * that are dirty | |
2667 | */ | |
58110f0a | 2668 | if (migrate_postcopy_ram()) { |
739fcc1b | 2669 | ram_postcopy_send_discard_bitmap(ms); |
1d34e4bf DDAG |
2670 | } |
2671 | ||
2672 | /* | |
2673 | * send rest of state - note things that are doing postcopy | |
2674 | * will notice we're in POSTCOPY_ACTIVE and not actually | |
2675 | * wrap their state up here | |
2676 | */ | |
7e555c6c DDAG |
2677 | /* 0 max-postcopy-bandwidth means unlimited */ |
2678 | if (!bandwidth) { | |
2679 | qemu_file_set_rate_limit(ms->to_dst_file, INT64_MAX); | |
2680 | } else { | |
2681 | qemu_file_set_rate_limit(ms->to_dst_file, bandwidth / XFER_LIMIT_RATIO); | |
2682 | } | |
58110f0a VSO |
2683 | if (migrate_postcopy_ram()) { |
2684 | /* Ping just for debugging, helps line traces up */ | |
2685 | qemu_savevm_send_ping(ms->to_dst_file, 2); | |
2686 | } | |
1d34e4bf DDAG |
2687 | |
2688 | /* | |
2689 | * While loading the device state we may trigger page transfer | |
2690 | * requests and the fd must be free to process those, and thus | |
2691 | * the destination must read the whole device state off the fd before | |
2692 | * it starts processing it. Unfortunately the ad-hoc migration format | |
2693 | * doesn't allow the destination to know the size to read without fully | |
2694 | * parsing it through each devices load-state code (especially the open | |
2695 | * coded devices that use get/put). | |
2696 | * So we wrap the device state up in a package with a length at the start; | |
2697 | * to do this we use a qemu_buf to hold the whole of the device state. | |
2698 | */ | |
61b67d47 | 2699 | bioc = qio_channel_buffer_new(4096); |
6f01f136 | 2700 | qio_channel_set_name(QIO_CHANNEL(bioc), "migration-postcopy-buffer"); |
77ef2dc1 | 2701 | fb = qemu_file_new_output(QIO_CHANNEL(bioc)); |
61b67d47 | 2702 | object_unref(OBJECT(bioc)); |
1d34e4bf | 2703 | |
c76201ab DDAG |
2704 | /* |
2705 | * Make sure the receiver can get incoming pages before we send the rest | |
2706 | * of the state | |
2707 | */ | |
2708 | qemu_savevm_send_postcopy_listen(fb); | |
2709 | ||
a1fbe750 | 2710 | qemu_savevm_state_complete_precopy(fb, false, false); |
58110f0a VSO |
2711 | if (migrate_postcopy_ram()) { |
2712 | qemu_savevm_send_ping(fb, 3); | |
2713 | } | |
1d34e4bf DDAG |
2714 | |
2715 | qemu_savevm_send_postcopy_run(fb); | |
2716 | ||
2717 | /* <><> end of stuff going into the package */ | |
1d34e4bf | 2718 | |
ef8d6488 DDAG |
2719 | /* Last point of recovery; as soon as we send the package the destination |
2720 | * can open devices and potentially start running. | |
2721 | * Lets just check again we've not got any errors. | |
2722 | */ | |
2723 | ret = qemu_file_get_error(ms->to_dst_file); | |
2724 | if (ret) { | |
2725 | error_report("postcopy_start: Migration stream errored (pre package)"); | |
2726 | goto fail_closefb; | |
2727 | } | |
2728 | ||
2729 | restart_block = false; | |
2730 | ||
1d34e4bf | 2731 | /* Now send that blob */ |
61b67d47 | 2732 | if (qemu_savevm_send_packaged(ms->to_dst_file, bioc->data, bioc->usage)) { |
1d34e4bf DDAG |
2733 | goto fail_closefb; |
2734 | } | |
2735 | qemu_fclose(fb); | |
b82fc321 DDAG |
2736 | |
2737 | /* Send a notify to give a chance for anything that needs to happen | |
2738 | * at the transition to postcopy and after the device state; in particular | |
2739 | * spice needs to trigger a transition now | |
2740 | */ | |
2741 | ms->postcopy_after_devices = true; | |
2742 | notifier_list_notify(&migration_state_notifiers, ms); | |
2743 | ||
1d34e4bf DDAG |
2744 | ms->downtime = qemu_clock_get_ms(QEMU_CLOCK_REALTIME) - time_at_stop; |
2745 | ||
2746 | qemu_mutex_unlock_iothread(); | |
2747 | ||
58110f0a VSO |
2748 | if (migrate_postcopy_ram()) { |
2749 | /* | |
2750 | * Although this ping is just for debug, it could potentially be | |
2751 | * used for getting a better measurement of downtime at the source. | |
2752 | */ | |
2753 | qemu_savevm_send_ping(ms->to_dst_file, 4); | |
2754 | } | |
1d34e4bf | 2755 | |
ced1c616 PB |
2756 | if (migrate_release_ram()) { |
2757 | ram_postcopy_migrated_memory_release(ms); | |
2758 | } | |
2759 | ||
89a02a9f | 2760 | ret = qemu_file_get_error(ms->to_dst_file); |
1d34e4bf DDAG |
2761 | if (ret) { |
2762 | error_report("postcopy_start: Migration stream errored"); | |
48781e5b | 2763 | migrate_set_state(&ms->state, MIGRATION_STATUS_POSTCOPY_ACTIVE, |
1d34e4bf DDAG |
2764 | MIGRATION_STATUS_FAILED); |
2765 | } | |
2766 | ||
c01b16ed PX |
2767 | trace_postcopy_preempt_enabled(migrate_postcopy_preempt()); |
2768 | ||
1d34e4bf DDAG |
2769 | return ret; |
2770 | ||
2771 | fail_closefb: | |
2772 | qemu_fclose(fb); | |
2773 | fail: | |
48781e5b | 2774 | migrate_set_state(&ms->state, MIGRATION_STATUS_POSTCOPY_ACTIVE, |
1d34e4bf | 2775 | MIGRATION_STATUS_FAILED); |
ef8d6488 DDAG |
2776 | if (restart_block) { |
2777 | /* A failure happened early enough that we know the destination hasn't | |
2778 | * accessed block devices, so we're safe to recover. | |
2779 | */ | |
2780 | Error *local_err = NULL; | |
2781 | ||
3b717194 | 2782 | bdrv_activate_all(&local_err); |
ef8d6488 DDAG |
2783 | if (local_err) { |
2784 | error_report_err(local_err); | |
2785 | } | |
2786 | } | |
1d34e4bf DDAG |
2787 | qemu_mutex_unlock_iothread(); |
2788 | return -1; | |
2789 | } | |
2790 | ||
e91d8951 DDAG |
2791 | /** |
2792 | * migration_maybe_pause: Pause if required to by | |
2793 | * migrate_pause_before_switchover called with the iothread locked | |
2794 | * Returns: 0 on success | |
2795 | */ | |
0331c8ca DDAG |
2796 | static int migration_maybe_pause(MigrationState *s, |
2797 | int *current_active_state, | |
2798 | int new_state) | |
e91d8951 DDAG |
2799 | { |
2800 | if (!migrate_pause_before_switchover()) { | |
2801 | return 0; | |
2802 | } | |
2803 | ||
2804 | /* Since leaving this state is not atomic with posting the semaphore | |
2805 | * it's possible that someone could have issued multiple migrate_continue | |
2806 | * and the semaphore is incorrectly positive at this point; | |
2807 | * the docs say it's undefined to reinit a semaphore that's already | |
2808 | * init'd, so use timedwait to eat up any existing posts. | |
2809 | */ | |
2810 | while (qemu_sem_timedwait(&s->pause_sem, 1) == 0) { | |
2811 | /* This block intentionally left blank */ | |
2812 | } | |
2813 | ||
8958338b ZF |
2814 | /* |
2815 | * If the migration is cancelled when it is in the completion phase, | |
2816 | * the migration state is set to MIGRATION_STATUS_CANCELLING. | |
2817 | * So we don't need to wait a semaphore, otherwise we would always | |
2818 | * wait for the 'pause_sem' semaphore. | |
2819 | */ | |
2820 | if (s->state != MIGRATION_STATUS_CANCELLING) { | |
2821 | qemu_mutex_unlock_iothread(); | |
2822 | migrate_set_state(&s->state, *current_active_state, | |
2823 | MIGRATION_STATUS_PRE_SWITCHOVER); | |
2824 | qemu_sem_wait(&s->pause_sem); | |
2825 | migrate_set_state(&s->state, MIGRATION_STATUS_PRE_SWITCHOVER, | |
2826 | new_state); | |
2827 | *current_active_state = new_state; | |
2828 | qemu_mutex_lock_iothread(); | |
2829 | } | |
e91d8951 | 2830 | |
0331c8ca | 2831 | return s->state == new_state ? 0 : -EINVAL; |
e91d8951 DDAG |
2832 | } |
2833 | ||
09f6c85e DDAG |
2834 | /** |
2835 | * migration_completion: Used by migration_thread when there's not much left. | |
2836 | * The caller 'breaks' the loop when this returns. | |
2837 | * | |
2838 | * @s: Current migration state | |
09f6c85e | 2839 | */ |
2ad87305 | 2840 | static void migration_completion(MigrationState *s) |
09f6c85e DDAG |
2841 | { |
2842 | int ret; | |
2ad87305 | 2843 | int current_active_state = s->state; |
09f6c85e | 2844 | |
b10ac0c4 DDAG |
2845 | if (s->state == MIGRATION_STATUS_ACTIVE) { |
2846 | qemu_mutex_lock_iothread(); | |
64909f97 | 2847 | s->downtime_start = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); |
fb064112 | 2848 | qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER, NULL); |
7287cbd4 | 2849 | s->vm_was_running = runstate_is_running(); |
b10ac0c4 DDAG |
2850 | ret = global_state_store(); |
2851 | ||
2852 | if (!ret) { | |
2853 | ret = vm_stop_force_state(RUN_STATE_FINISH_MIGRATE); | |
626ff651 | 2854 | trace_migration_completion_vm_stop(ret); |
e91d8951 | 2855 | if (ret >= 0) { |
0331c8ca DDAG |
2856 | ret = migration_maybe_pause(s, ¤t_active_state, |
2857 | MIGRATION_STATUS_DEVICE); | |
e91d8951 | 2858 | } |
f07fa4cb | 2859 | if (ret >= 0) { |
5e804644 | 2860 | s->block_inactive = !migrate_colo(); |
f07fa4cb | 2861 | qemu_file_set_rate_limit(s->to_dst_file, INT64_MAX); |
a1fbe750 | 2862 | ret = qemu_savevm_state_complete_precopy(s->to_dst_file, false, |
5d39f44d | 2863 | s->block_inactive); |
f07fa4cb | 2864 | } |
b10ac0c4 DDAG |
2865 | } |
2866 | qemu_mutex_unlock_iothread(); | |
09f6c85e | 2867 | |
b10ac0c4 DDAG |
2868 | if (ret < 0) { |
2869 | goto fail; | |
09f6c85e | 2870 | } |
b10ac0c4 DDAG |
2871 | } else if (s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) { |
2872 | trace_migration_completion_postcopy_end(); | |
2873 | ||
68b88468 | 2874 | qemu_mutex_lock_iothread(); |
89a02a9f | 2875 | qemu_savevm_state_complete_postcopy(s->to_dst_file); |
68b88468 EGE |
2876 | qemu_mutex_unlock_iothread(); |
2877 | ||
6621883f PX |
2878 | /* |
2879 | * Shutdown the postcopy fast path thread. This is only needed | |
2880 | * when dest QEMU binary is old (7.1/7.2). QEMU 8.0+ doesn't need | |
2881 | * this. | |
2882 | */ | |
2883 | if (migrate_postcopy_preempt() && s->preempt_pre_7_2) { | |
36f62f11 PX |
2884 | postcopy_preempt_shutdown_file(s); |
2885 | } | |
2886 | ||
b10ac0c4 | 2887 | trace_migration_completion_postcopy_end_after_complete(); |
444252b9 | 2888 | } else { |
6ba11211 | 2889 | goto fail; |
09f6c85e | 2890 | } |
09f6c85e | 2891 | |
b10ac0c4 DDAG |
2892 | /* |
2893 | * If rp was opened we must clean up the thread before | |
2894 | * cleaning everything else up (since if there are no failures | |
2895 | * it will wait for the destination to send it's status in | |
2896 | * a SHUT command). | |
b10ac0c4 | 2897 | */ |
53021ea1 | 2898 | if (s->rp_state.rp_thread_created) { |
b10ac0c4 | 2899 | int rp_error; |
0425dc97 | 2900 | trace_migration_return_path_end_before(); |
b10ac0c4 | 2901 | rp_error = await_return_path_close_on_source(s); |
0425dc97 | 2902 | trace_migration_return_path_end_after(rp_error); |
b10ac0c4 | 2903 | if (rp_error) { |
fe904ea8 | 2904 | goto fail_invalidate; |
b10ac0c4 | 2905 | } |
09f6c85e DDAG |
2906 | } |
2907 | ||
89a02a9f | 2908 | if (qemu_file_get_error(s->to_dst_file)) { |
09f6c85e | 2909 | trace_migration_completion_file_err(); |
fe904ea8 | 2910 | goto fail_invalidate; |
09f6c85e DDAG |
2911 | } |
2912 | ||
5e804644 | 2913 | if (migrate_colo() && s->state == MIGRATION_STATUS_ACTIVE) { |
eeeb48ee ZC |
2914 | /* COLO does not support postcopy */ |
2915 | migrate_set_state(&s->state, MIGRATION_STATUS_ACTIVE, | |
2916 | MIGRATION_STATUS_COLO); | |
2917 | } else { | |
0b827d5e HZ |
2918 | migrate_set_state(&s->state, current_active_state, |
2919 | MIGRATION_STATUS_COMPLETED); | |
2920 | } | |
2921 | ||
09f6c85e DDAG |
2922 | return; |
2923 | ||
fe904ea8 GK |
2924 | fail_invalidate: |
2925 | /* If not doing postcopy, vm_start() will be called: let's regain | |
2926 | * control on images. | |
2927 | */ | |
6039dd5b DDAG |
2928 | if (s->state == MIGRATION_STATUS_ACTIVE || |
2929 | s->state == MIGRATION_STATUS_DEVICE) { | |
fe904ea8 GK |
2930 | Error *local_err = NULL; |
2931 | ||
1d2acc31 | 2932 | qemu_mutex_lock_iothread(); |
3b717194 | 2933 | bdrv_activate_all(&local_err); |
fe904ea8 GK |
2934 | if (local_err) { |
2935 | error_report_err(local_err); | |
403d18ae | 2936 | s->block_inactive = true; |
1d2acc31 HZ |
2937 | } else { |
2938 | s->block_inactive = false; | |
fe904ea8 | 2939 | } |
1d2acc31 | 2940 | qemu_mutex_unlock_iothread(); |
fe904ea8 GK |
2941 | } |
2942 | ||
09f6c85e | 2943 | fail: |
48781e5b HZ |
2944 | migrate_set_state(&s->state, current_active_state, |
2945 | MIGRATION_STATUS_FAILED); | |
09f6c85e DDAG |
2946 | } |
2947 | ||
8518278a AG |
2948 | /** |
2949 | * bg_migration_completion: Used by bg_migration_thread when after all the | |
2950 | * RAM has been saved. The caller 'breaks' the loop when this returns. | |
2951 | * | |
2952 | * @s: Current migration state | |
2953 | */ | |
2954 | static void bg_migration_completion(MigrationState *s) | |
2955 | { | |
2956 | int current_active_state = s->state; | |
2957 | ||
2958 | /* | |
2959 | * Stop tracking RAM writes - un-protect memory, un-register UFFD | |
2960 | * memory ranges, flush kernel wait queues and wake up threads | |
2961 | * waiting for write fault to be resolved. | |
2962 | */ | |
2963 | ram_write_tracking_stop(); | |
2964 | ||
2965 | if (s->state == MIGRATION_STATUS_ACTIVE) { | |
2966 | /* | |
2967 | * By this moment we have RAM content saved into the migration stream. | |
2968 | * The next step is to flush the non-RAM content (device state) | |
2969 | * right after the ram content. The device state has been stored into | |
2970 | * the temporary buffer before RAM saving started. | |
2971 | */ | |
2972 | qemu_put_buffer(s->to_dst_file, s->bioc->data, s->bioc->usage); | |
2973 | qemu_fflush(s->to_dst_file); | |
2974 | } else if (s->state == MIGRATION_STATUS_CANCELLING) { | |
2975 | goto fail; | |
2976 | } | |
2977 | ||
2978 | if (qemu_file_get_error(s->to_dst_file)) { | |
2979 | trace_migration_completion_file_err(); | |
2980 | goto fail; | |
2981 | } | |
2982 | ||
2983 | migrate_set_state(&s->state, current_active_state, | |
2984 | MIGRATION_STATUS_COMPLETED); | |
2985 | return; | |
2986 | ||
2987 | fail: | |
2988 | migrate_set_state(&s->state, current_active_state, | |
2989 | MIGRATION_STATUS_FAILED); | |
2990 | } | |
2991 | ||
b23c2ade PX |
2992 | typedef enum MigThrError { |
2993 | /* No error detected */ | |
2994 | MIG_THR_ERR_NONE = 0, | |
2995 | /* Detected error, but resumed successfully */ | |
2996 | MIG_THR_ERR_RECOVERED = 1, | |
2997 | /* Detected fatal error, need to exit */ | |
2998 | MIG_THR_ERR_FATAL = 2, | |
2999 | } MigThrError; | |
3000 | ||
94190696 PX |
3001 | static int postcopy_resume_handshake(MigrationState *s) |
3002 | { | |
3003 | qemu_savevm_send_postcopy_resume(s->to_dst_file); | |
3004 | ||
3005 | while (s->state == MIGRATION_STATUS_POSTCOPY_RECOVER) { | |
3006 | qemu_sem_wait(&s->rp_state.rp_sem); | |
3007 | } | |
3008 | ||
3009 | if (s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) { | |
3010 | return 0; | |
3011 | } | |
3012 | ||
3013 | return -1; | |
3014 | } | |
3015 | ||
135b87b4 PX |
3016 | /* Return zero if success, or <0 for error */ |
3017 | static int postcopy_do_resume(MigrationState *s) | |
3018 | { | |
d1b8eadb PX |
3019 | int ret; |
3020 | ||
3021 | /* | |
3022 | * Call all the resume_prepare() hooks, so that modules can be | |
3023 | * ready for the migration resume. | |
3024 | */ | |
3025 | ret = qemu_savevm_state_resume_prepare(s); | |
3026 | if (ret) { | |
3027 | error_report("%s: resume_prepare() failure detected: %d", | |
3028 | __func__, ret); | |
3029 | return ret; | |
3030 | } | |
3031 | ||
5655aab0 PX |
3032 | /* |
3033 | * If preempt is enabled, re-establish the preempt channel. Note that | |
3034 | * we do it after resume prepare to make sure the main channel will be | |
3035 | * created before the preempt channel. E.g. with weak network, the | |
3036 | * dest QEMU may get messed up with the preempt and main channels on | |
3037 | * the order of connection setup. This guarantees the correct order. | |
3038 | */ | |
3039 | ret = postcopy_preempt_establish_channel(s); | |
3040 | if (ret) { | |
3041 | error_report("%s: postcopy_preempt_establish_channel(): %d", | |
3042 | __func__, ret); | |
3043 | return ret; | |
3044 | } | |
3045 | ||
d1b8eadb | 3046 | /* |
94190696 PX |
3047 | * Last handshake with destination on the resume (destination will |
3048 | * switch to postcopy-active afterwards) | |
d1b8eadb | 3049 | */ |
94190696 PX |
3050 | ret = postcopy_resume_handshake(s); |
3051 | if (ret) { | |
3052 | error_report("%s: handshake failed: %d", __func__, ret); | |
3053 | return ret; | |
3054 | } | |
d1b8eadb | 3055 | |
135b87b4 PX |
3056 | return 0; |
3057 | } | |
3058 | ||
b23c2ade PX |
3059 | /* |
3060 | * We don't return until we are in a safe state to continue current | |
3061 | * postcopy migration. Returns MIG_THR_ERR_RECOVERED if recovered, or | |
3062 | * MIG_THR_ERR_FATAL if unrecovery failure happened. | |
3063 | */ | |
3064 | static MigThrError postcopy_pause(MigrationState *s) | |
3065 | { | |
3066 | assert(s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE); | |
b23c2ade | 3067 | |
135b87b4 | 3068 | while (true) { |
62df066f PX |
3069 | QEMUFile *file; |
3070 | ||
39675fff PX |
3071 | /* |
3072 | * Current channel is possibly broken. Release it. Note that this is | |
3073 | * guaranteed even without lock because to_dst_file should only be | |
3074 | * modified by the migration thread. That also guarantees that the | |
3075 | * unregister of yank is safe too without the lock. It should be safe | |
3076 | * even to be within the qemu_file_lock, but we didn't do that to avoid | |
3077 | * taking more mutex (yank_lock) within qemu_file_lock. TL;DR: we make | |
3078 | * the qemu_file_lock critical section as small as possible. | |
3079 | */ | |
135b87b4 | 3080 | assert(s->to_dst_file); |
39675fff | 3081 | migration_ioc_unregister_yank_from_file(s->to_dst_file); |
62df066f PX |
3082 | qemu_mutex_lock(&s->qemu_file_lock); |
3083 | file = s->to_dst_file; | |
135b87b4 | 3084 | s->to_dst_file = NULL; |
62df066f PX |
3085 | qemu_mutex_unlock(&s->qemu_file_lock); |
3086 | ||
3087 | qemu_file_shutdown(file); | |
3088 | qemu_fclose(file); | |
b23c2ade | 3089 | |
d246ea50 PX |
3090 | migrate_set_state(&s->state, s->state, |
3091 | MIGRATION_STATUS_POSTCOPY_PAUSED); | |
3092 | ||
135b87b4 PX |
3093 | error_report("Detected IO failure for postcopy. " |
3094 | "Migration paused."); | |
b23c2ade | 3095 | |
135b87b4 PX |
3096 | /* |
3097 | * We wait until things fixed up. Then someone will setup the | |
3098 | * status back for us. | |
3099 | */ | |
3100 | while (s->state == MIGRATION_STATUS_POSTCOPY_PAUSED) { | |
3101 | qemu_sem_wait(&s->postcopy_pause_sem); | |
3102 | } | |
3103 | ||
3104 | if (s->state == MIGRATION_STATUS_POSTCOPY_RECOVER) { | |
3105 | /* Woken up by a recover procedure. Give it a shot */ | |
b23c2ade | 3106 | |
135b87b4 PX |
3107 | /* |
3108 | * Firstly, let's wake up the return path now, with a new | |
3109 | * return path channel. | |
3110 | */ | |
3111 | qemu_sem_post(&s->postcopy_pause_rp_sem); | |
3112 | ||
3113 | /* Do the resume logic */ | |
3114 | if (postcopy_do_resume(s) == 0) { | |
3115 | /* Let's continue! */ | |
3116 | trace_postcopy_pause_continued(); | |
3117 | return MIG_THR_ERR_RECOVERED; | |
3118 | } else { | |
3119 | /* | |
3120 | * Something wrong happened during the recovery, let's | |
3121 | * pause again. Pause is always better than throwing | |
3122 | * data away. | |
3123 | */ | |
3124 | continue; | |
3125 | } | |
3126 | } else { | |
3127 | /* This is not right... Time to quit. */ | |
3128 | return MIG_THR_ERR_FATAL; | |
3129 | } | |
3130 | } | |
b23c2ade PX |
3131 | } |
3132 | ||
3133 | static MigThrError migration_detect_error(MigrationState *s) | |
3134 | { | |
3135 | int ret; | |
c3c5eae6 | 3136 | int state = s->state; |
3d661c8a | 3137 | Error *local_error = NULL; |
c3c5eae6 DDAG |
3138 | |
3139 | if (state == MIGRATION_STATUS_CANCELLING || | |
3140 | state == MIGRATION_STATUS_CANCELLED) { | |
3141 | /* End the migration, but don't set the state to failed */ | |
3142 | return MIG_THR_ERR_FATAL; | |
3143 | } | |
b23c2ade | 3144 | |
60bb3c58 PX |
3145 | /* |
3146 | * Try to detect any file errors. Note that postcopy_qemufile_src will | |
3147 | * be NULL when postcopy preempt is not enabled. | |
3148 | */ | |
3149 | ret = qemu_file_get_error_obj_any(s->to_dst_file, | |
3150 | s->postcopy_qemufile_src, | |
3151 | &local_error); | |
b23c2ade PX |
3152 | if (!ret) { |
3153 | /* Everything is fine */ | |
3d661c8a | 3154 | assert(!local_error); |
b23c2ade PX |
3155 | return MIG_THR_ERR_NONE; |
3156 | } | |
3157 | ||
3d661c8a YK |
3158 | if (local_error) { |
3159 | migrate_set_error(s, local_error); | |
3160 | error_free(local_error); | |
3161 | } | |
3162 | ||
d5c8f2af | 3163 | if (state == MIGRATION_STATUS_POSTCOPY_ACTIVE && ret) { |
b23c2ade PX |
3164 | /* |
3165 | * For postcopy, we allow the network to be down for a | |
3166 | * while. After that, it can be continued by a | |
3167 | * recovery phase. | |
3168 | */ | |
3169 | return postcopy_pause(s); | |
3170 | } else { | |
3171 | /* | |
3172 | * For precopy (or postcopy with error outside IO), we fail | |
3173 | * with no time. | |
3174 | */ | |
c3c5eae6 | 3175 | migrate_set_state(&s->state, state, MIGRATION_STATUS_FAILED); |
b23c2ade PX |
3176 | trace_migration_thread_file_err(); |
3177 | ||
3178 | /* Time to stop the migration, now. */ | |
3179 | return MIG_THR_ERR_FATAL; | |
3180 | } | |
3181 | } | |
3182 | ||
268dcd46 | 3183 | /* How many bytes have we transferred since the beginning of the migration */ |
0c8f0efd JQ |
3184 | static uint64_t migration_total_bytes(MigrationState *s) |
3185 | { | |
fbfa6404 | 3186 | return qemu_file_total_transferred(s->to_dst_file) + |
cf671116 | 3187 | stat64_get(&ram_counters.multifd_bytes); |
0c8f0efd JQ |
3188 | } |
3189 | ||
cf011f08 PX |
3190 | static void migration_calculate_complete(MigrationState *s) |
3191 | { | |
0c8f0efd | 3192 | uint64_t bytes = migration_total_bytes(s); |
cf011f08 | 3193 | int64_t end_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); |
6cde6fbe | 3194 | int64_t transfer_time; |
cf011f08 PX |
3195 | |
3196 | s->total_time = end_time - s->start_time; | |
3197 | if (!s->downtime) { | |
3198 | /* | |
3199 | * It's still not set, so we are precopy migration. For | |
3200 | * postcopy, downtime is calculated during postcopy_start(). | |
3201 | */ | |
3202 | s->downtime = end_time - s->downtime_start; | |
3203 | } | |
3204 | ||
6cde6fbe JQ |
3205 | transfer_time = s->total_time - s->setup_time; |
3206 | if (transfer_time) { | |
3207 | s->mbps = ((double) bytes * 8.0) / transfer_time / 1000; | |
cf011f08 PX |
3208 | } |
3209 | } | |
3210 | ||
87f3bd87 IR |
3211 | static void update_iteration_initial_status(MigrationState *s) |
3212 | { | |
3213 | /* | |
3214 | * Update these three fields at the same time to avoid mismatch info lead | |
3215 | * wrong speed calculation. | |
3216 | */ | |
3217 | s->iteration_start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); | |
3218 | s->iteration_initial_bytes = migration_total_bytes(s); | |
3219 | s->iteration_initial_pages = ram_get_total_transferred_pages(); | |
3220 | } | |
3221 | ||
b15df1ae PX |
3222 | static void migration_update_counters(MigrationState *s, |
3223 | int64_t current_time) | |
3224 | { | |
aecbfe9c | 3225 | uint64_t transferred, transferred_pages, time_spent; |
0c8f0efd | 3226 | uint64_t current_bytes; /* bytes transferred since the beginning */ |
b15df1ae PX |
3227 | double bandwidth; |
3228 | ||
3229 | if (current_time < s->iteration_start_time + BUFFER_DELAY) { | |
3230 | return; | |
3231 | } | |
3232 | ||
0c8f0efd JQ |
3233 | current_bytes = migration_total_bytes(s); |
3234 | transferred = current_bytes - s->iteration_initial_bytes; | |
b15df1ae PX |
3235 | time_spent = current_time - s->iteration_start_time; |
3236 | bandwidth = (double)transferred / time_spent; | |
0781c1ed | 3237 | s->threshold_size = bandwidth * s->parameters.downtime_limit; |
b15df1ae PX |
3238 | |
3239 | s->mbps = (((double) transferred * 8.0) / | |
3240 | ((double) time_spent / 1000.0)) / 1000.0 / 1000.0; | |
3241 | ||
aecbfe9c XG |
3242 | transferred_pages = ram_get_total_transferred_pages() - |
3243 | s->iteration_initial_pages; | |
3244 | s->pages_per_second = (double) transferred_pages / | |
3245 | (((double) time_spent / 1000.0)); | |
3246 | ||
b15df1ae PX |
3247 | /* |
3248 | * if we haven't sent anything, we don't want to | |
3249 | * recalculate. 10000 is a small enough number for our purposes | |
3250 | */ | |
3251 | if (ram_counters.dirty_pages_rate && transferred > 10000) { | |
650af890 | 3252 | s->expected_downtime = ram_counters.remaining / bandwidth; |
b15df1ae PX |
3253 | } |
3254 | ||
3255 | qemu_file_reset_rate_limit(s->to_dst_file); | |
3256 | ||
87f3bd87 | 3257 | update_iteration_initial_status(s); |
b15df1ae PX |
3258 | |
3259 | trace_migrate_transferred(transferred, time_spent, | |
0781c1ed | 3260 | bandwidth, s->threshold_size); |
b15df1ae PX |
3261 | } |
3262 | ||
2ad87305 PX |
3263 | /* Migration thread iteration status */ |
3264 | typedef enum { | |
3265 | MIG_ITERATE_RESUME, /* Resume current iteration */ | |
3266 | MIG_ITERATE_SKIP, /* Skip current iteration */ | |
3267 | MIG_ITERATE_BREAK, /* Break the loop */ | |
3268 | } MigIterateState; | |
3269 | ||
3270 | /* | |
3271 | * Return true if continue to the next iteration directly, false | |
3272 | * otherwise. | |
3273 | */ | |
3274 | static MigIterateState migration_iteration_run(MigrationState *s) | |
3275 | { | |
24beea4e | 3276 | uint64_t must_precopy, can_postcopy; |
2ad87305 PX |
3277 | bool in_postcopy = s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE; |
3278 | ||
24beea4e JQ |
3279 | qemu_savevm_state_pending_estimate(&must_precopy, &can_postcopy); |
3280 | uint64_t pending_size = must_precopy + can_postcopy; | |
2ad87305 | 3281 | |
24beea4e | 3282 | trace_migrate_pending_estimate(pending_size, must_precopy, can_postcopy); |
c8df4a7a | 3283 | |
24beea4e JQ |
3284 | if (must_precopy <= s->threshold_size) { |
3285 | qemu_savevm_state_pending_exact(&must_precopy, &can_postcopy); | |
3286 | pending_size = must_precopy + can_postcopy; | |
3287 | trace_migrate_pending_exact(pending_size, must_precopy, can_postcopy); | |
c8df4a7a | 3288 | } |
2ad87305 | 3289 | |
d9df9292 | 3290 | if (!pending_size || pending_size < s->threshold_size) { |
2ad87305 PX |
3291 | trace_migration_thread_low_pending(pending_size); |
3292 | migration_completion(s); | |
3293 | return MIG_ITERATE_BREAK; | |
3294 | } | |
3295 | ||
d9df9292 | 3296 | /* Still a significant amount to transfer */ |
24beea4e | 3297 | if (!in_postcopy && must_precopy <= s->threshold_size && |
d9df9292 JQ |
3298 | qatomic_read(&s->start_postcopy)) { |
3299 | if (postcopy_start(s)) { | |
3300 | error_report("%s: postcopy failed to start", __func__); | |
3301 | } | |
3302 | return MIG_ITERATE_SKIP; | |
3303 | } | |
3304 | ||
3305 | /* Just another iteration step */ | |
3306 | qemu_savevm_state_iterate(s->to_dst_file, in_postcopy); | |
2ad87305 PX |
3307 | return MIG_ITERATE_RESUME; |
3308 | } | |
3309 | ||
199aa6d4 PX |
3310 | static void migration_iteration_finish(MigrationState *s) |
3311 | { | |
3312 | /* If we enabled cpu throttling for auto-converge, turn it off. */ | |
3313 | cpu_throttle_stop(); | |
3314 | ||
3315 | qemu_mutex_lock_iothread(); | |
3316 | switch (s->state) { | |
3317 | case MIGRATION_STATUS_COMPLETED: | |
3318 | migration_calculate_complete(s); | |
3319 | runstate_set(RUN_STATE_POSTMIGRATE); | |
3320 | break; | |
751fe4c6 | 3321 | case MIGRATION_STATUS_COLO: |
5e804644 | 3322 | if (!migrate_colo()) { |
199aa6d4 PX |
3323 | error_report("%s: critical error: calling COLO code without " |
3324 | "COLO enabled", __func__); | |
3325 | } | |
3326 | migrate_start_colo_process(s); | |
199aa6d4 PX |
3327 | s->vm_was_running = true; |
3328 | /* Fallthrough */ | |
3329 | case MIGRATION_STATUS_FAILED: | |
3330 | case MIGRATION_STATUS_CANCELLED: | |
57225e5f | 3331 | case MIGRATION_STATUS_CANCELLING: |
199aa6d4 | 3332 | if (s->vm_was_running) { |
aa505f8e RL |
3333 | if (!runstate_check(RUN_STATE_SHUTDOWN)) { |
3334 | vm_start(); | |
3335 | } | |
199aa6d4 PX |
3336 | } else { |
3337 | if (runstate_check(RUN_STATE_FINISH_MIGRATE)) { | |
3338 | runstate_set(RUN_STATE_POSTMIGRATE); | |
3339 | } | |
3340 | } | |
3341 | break; | |
3342 | ||
3343 | default: | |
3344 | /* Should not reach here, but if so, forgive the VM. */ | |
3345 | error_report("%s: Unknown ending state %d", __func__, s->state); | |
3346 | break; | |
3347 | } | |
fd392cfa | 3348 | migrate_fd_cleanup_schedule(s); |
199aa6d4 PX |
3349 | qemu_mutex_unlock_iothread(); |
3350 | } | |
3351 | ||
8518278a AG |
3352 | static void bg_migration_iteration_finish(MigrationState *s) |
3353 | { | |
3354 | qemu_mutex_lock_iothread(); | |
3355 | switch (s->state) { | |
3356 | case MIGRATION_STATUS_COMPLETED: | |
3357 | migration_calculate_complete(s); | |
3358 | break; | |
3359 | ||
3360 | case MIGRATION_STATUS_ACTIVE: | |
3361 | case MIGRATION_STATUS_FAILED: | |
3362 | case MIGRATION_STATUS_CANCELLED: | |
3363 | case MIGRATION_STATUS_CANCELLING: | |
3364 | break; | |
3365 | ||
3366 | default: | |
3367 | /* Should not reach here, but if so, forgive the VM. */ | |
3368 | error_report("%s: Unknown ending state %d", __func__, s->state); | |
3369 | break; | |
3370 | } | |
3371 | ||
3372 | migrate_fd_cleanup_schedule(s); | |
3373 | qemu_mutex_unlock_iothread(); | |
3374 | } | |
3375 | ||
3376 | /* | |
3377 | * Return true if continue to the next iteration directly, false | |
3378 | * otherwise. | |
3379 | */ | |
3380 | static MigIterateState bg_migration_iteration_run(MigrationState *s) | |
3381 | { | |
3382 | int res; | |
3383 | ||
3384 | res = qemu_savevm_state_iterate(s->to_dst_file, false); | |
3385 | if (res > 0) { | |
3386 | bg_migration_completion(s); | |
3387 | return MIG_ITERATE_BREAK; | |
3388 | } | |
3389 | ||
3390 | return MIG_ITERATE_RESUME; | |
3391 | } | |
3392 | ||
ad767bed DDAG |
3393 | void migration_make_urgent_request(void) |
3394 | { | |
3395 | qemu_sem_post(&migrate_get_current()->rate_limit_sem); | |
3396 | } | |
3397 | ||
3398 | void migration_consume_urgent_request(void) | |
3399 | { | |
3400 | qemu_sem_wait(&migrate_get_current()->rate_limit_sem); | |
3401 | } | |
3402 | ||
97e1e067 DDAG |
3403 | /* Returns true if the rate limiting was broken by an urgent request */ |
3404 | bool migration_rate_limit(void) | |
3405 | { | |
3406 | int64_t now = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); | |
3407 | MigrationState *s = migrate_get_current(); | |
3408 | ||
3409 | bool urgent = false; | |
3410 | migration_update_counters(s, now); | |
3411 | if (qemu_file_rate_limit(s->to_dst_file)) { | |
77386127 LS |
3412 | |
3413 | if (qemu_file_get_error(s->to_dst_file)) { | |
3414 | return false; | |
3415 | } | |
97e1e067 DDAG |
3416 | /* |
3417 | * Wait for a delay to do rate limiting OR | |
3418 | * something urgent to post the semaphore. | |
3419 | */ | |
3420 | int ms = s->iteration_start_time + BUFFER_DELAY - now; | |
3421 | trace_migration_rate_limit_pre(ms); | |
3422 | if (qemu_sem_timedwait(&s->rate_limit_sem, ms) == 0) { | |
3423 | /* | |
3424 | * We were woken by one or more urgent things but | |
3425 | * the timedwait will have consumed one of them. | |
3426 | * The service routine for the urgent wake will dec | |
3427 | * the semaphore itself for each item it consumes, | |
3428 | * so add this one we just eat back. | |
3429 | */ | |
3430 | qemu_sem_post(&s->rate_limit_sem); | |
3431 | urgent = true; | |
3432 | } | |
3433 | trace_migration_rate_limit_post(urgent); | |
3434 | } | |
3435 | return urgent; | |
3436 | } | |
3437 | ||
fde93d99 LV |
3438 | /* |
3439 | * if failover devices are present, wait they are completely | |
3440 | * unplugged | |
3441 | */ | |
3442 | ||
3443 | static void qemu_savevm_wait_unplug(MigrationState *s, int old_state, | |
3444 | int new_state) | |
3445 | { | |
3446 | if (qemu_savevm_state_guest_unplug_pending()) { | |
3447 | migrate_set_state(&s->state, old_state, MIGRATION_STATUS_WAIT_UNPLUG); | |
3448 | ||
3449 | while (s->state == MIGRATION_STATUS_WAIT_UNPLUG && | |
3450 | qemu_savevm_state_guest_unplug_pending()) { | |
3451 | qemu_sem_timedwait(&s->wait_unplug_sem, 250); | |
3452 | } | |
944bc528 LV |
3453 | if (s->state != MIGRATION_STATUS_WAIT_UNPLUG) { |
3454 | int timeout = 120; /* 30 seconds */ | |
3455 | /* | |
3456 | * migration has been canceled | |
3457 | * but as we have started an unplug we must wait the end | |
3458 | * to be able to plug back the card | |
3459 | */ | |
3460 | while (timeout-- && qemu_savevm_state_guest_unplug_pending()) { | |
3461 | qemu_sem_timedwait(&s->wait_unplug_sem, 250); | |
3462 | } | |
1b529d90 LV |
3463 | if (qemu_savevm_state_guest_unplug_pending() && |
3464 | !qtest_enabled()) { | |
a51dcef0 LV |
3465 | warn_report("migration: partially unplugged device on " |
3466 | "failure"); | |
3467 | } | |
944bc528 | 3468 | } |
fde93d99 LV |
3469 | |
3470 | migrate_set_state(&s->state, MIGRATION_STATUS_WAIT_UNPLUG, new_state); | |
3471 | } else { | |
3472 | migrate_set_state(&s->state, old_state, new_state); | |
3473 | } | |
3474 | } | |
3475 | ||
70b20477 DDAG |
3476 | /* |
3477 | * Master migration thread on the source VM. | |
3478 | * It drives the migration and pumps the data down the outgoing channel. | |
3479 | */ | |
5f496a1b | 3480 | static void *migration_thread(void *opaque) |
0d82d0e8 | 3481 | { |
9848a404 | 3482 | MigrationState *s = opaque; |
1b1f4ab6 | 3483 | MigrationThread *thread = NULL; |
bc72ad67 | 3484 | int64_t setup_start = qemu_clock_get_ms(QEMU_CLOCK_HOST); |
b23c2ade | 3485 | MigThrError thr_error; |
ad767bed | 3486 | bool urgent = false; |
76f5933a | 3487 | |
1b1f4ab6 JJ |
3488 | thread = MigrationThreadAdd("live_migration", qemu_get_thread_id()); |
3489 | ||
ab28bd23 PB |
3490 | rcu_register_thread(); |
3491 | ||
892ae715 | 3492 | object_ref(OBJECT(s)); |
87f3bd87 | 3493 | update_iteration_initial_status(s); |
b15df1ae | 3494 | |
89a02a9f | 3495 | qemu_savevm_state_header(s->to_dst_file); |
1d34e4bf | 3496 | |
62a02658 PX |
3497 | /* |
3498 | * If we opened the return path, we need to make sure dst has it | |
3499 | * opened as well. | |
3500 | */ | |
43044ac0 | 3501 | if (s->rp_state.rp_thread_created) { |
1d34e4bf | 3502 | /* Now tell the dest that it should open its end so it can reply */ |
89a02a9f | 3503 | qemu_savevm_send_open_return_path(s->to_dst_file); |
1d34e4bf DDAG |
3504 | |
3505 | /* And do a ping that will make stuff easier to debug */ | |
89a02a9f | 3506 | qemu_savevm_send_ping(s->to_dst_file, 1); |
0425dc97 | 3507 | } |
1d34e4bf | 3508 | |
58110f0a | 3509 | if (migrate_postcopy()) { |
1d34e4bf DDAG |
3510 | /* |
3511 | * Tell the destination that we *might* want to do postcopy later; | |
3512 | * if the other end can't do postcopy it should fail now, nice and | |
3513 | * early. | |
3514 | */ | |
89a02a9f | 3515 | qemu_savevm_send_postcopy_advise(s->to_dst_file); |
1d34e4bf DDAG |
3516 | } |
3517 | ||
5e804644 | 3518 | if (migrate_colo()) { |
aad555c2 ZC |
3519 | /* Notify migration destination that we enable COLO */ |
3520 | qemu_savevm_send_colo_enable(s->to_dst_file); | |
3521 | } | |
3522 | ||
9907e842 | 3523 | qemu_savevm_state_setup(s->to_dst_file); |
0d82d0e8 | 3524 | |
fde93d99 LV |
3525 | qemu_savevm_wait_unplug(s, MIGRATION_STATUS_SETUP, |
3526 | MIGRATION_STATUS_ACTIVE); | |
c7e0acd5 | 3527 | |
bc72ad67 | 3528 | s->setup_time = qemu_clock_get_ms(QEMU_CLOCK_HOST) - setup_start; |
29ae8a41 | 3529 | |
9ec055ae DDAG |
3530 | trace_migration_thread_setup_complete(); |
3531 | ||
8f8d528e | 3532 | while (migration_is_active(s)) { |
ad767bed | 3533 | if (urgent || !qemu_file_rate_limit(s->to_dst_file)) { |
2ad87305 PX |
3534 | MigIterateState iter_state = migration_iteration_run(s); |
3535 | if (iter_state == MIG_ITERATE_SKIP) { | |
3536 | continue; | |
3537 | } else if (iter_state == MIG_ITERATE_BREAK) { | |
09f6c85e | 3538 | break; |
c369f40d JQ |
3539 | } |
3540 | } | |
f4410a5d | 3541 | |
b23c2ade PX |
3542 | /* |
3543 | * Try to detect any kind of failures, and see whether we | |
3544 | * should stop the migration now. | |
3545 | */ | |
3546 | thr_error = migration_detect_error(s); | |
3547 | if (thr_error == MIG_THR_ERR_FATAL) { | |
3548 | /* Stop migration */ | |
fd45ee2c | 3549 | break; |
b23c2ade PX |
3550 | } else if (thr_error == MIG_THR_ERR_RECOVERED) { |
3551 | /* | |
3552 | * Just recovered from a e.g. network failure, reset all | |
3553 | * the local variables. This is important to avoid | |
3554 | * breaking transferred_bytes and bandwidth calculation | |
3555 | */ | |
87f3bd87 | 3556 | update_iteration_initial_status(s); |
fd45ee2c | 3557 | } |
b15df1ae | 3558 | |
97e1e067 | 3559 | urgent = migration_rate_limit(); |
a3fa1d78 PB |
3560 | } |
3561 | ||
1d34e4bf | 3562 | trace_migration_thread_after_loop(); |
199aa6d4 | 3563 | migration_iteration_finish(s); |
892ae715 | 3564 | object_unref(OBJECT(s)); |
ab28bd23 | 3565 | rcu_unregister_thread(); |
1b1f4ab6 | 3566 | MigrationThreadDel(thread); |
0d82d0e8 JQ |
3567 | return NULL; |
3568 | } | |
3569 | ||
8518278a AG |
3570 | static void bg_migration_vm_start_bh(void *opaque) |
3571 | { | |
3572 | MigrationState *s = opaque; | |
3573 | ||
3574 | qemu_bh_delete(s->vm_start_bh); | |
3575 | s->vm_start_bh = NULL; | |
3576 | ||
3577 | vm_start(); | |
3578 | s->downtime = qemu_clock_get_ms(QEMU_CLOCK_REALTIME) - s->downtime_start; | |
3579 | } | |
3580 | ||
3581 | /** | |
3582 | * Background snapshot thread, based on live migration code. | |
3583 | * This is an alternative implementation of live migration mechanism | |
3584 | * introduced specifically to support background snapshots. | |
3585 | * | |
3586 | * It takes advantage of userfault_fd write protection mechanism introduced | |
3587 | * in v5.7 kernel. Compared to existing dirty page logging migration much | |
3588 | * lesser stream traffic is produced resulting in smaller snapshot images, | |
3589 | * simply cause of no page duplicates can get into the stream. | |
3590 | * | |
3591 | * Another key point is that generated vmstate stream reflects machine state | |
3592 | * 'frozen' at the beginning of snapshot creation compared to dirty page logging | |
3593 | * mechanism, which effectively results in that saved snapshot is the state of VM | |
3594 | * at the end of the process. | |
3595 | */ | |
3596 | static void *bg_migration_thread(void *opaque) | |
3597 | { | |
3598 | MigrationState *s = opaque; | |
3599 | int64_t setup_start; | |
3600 | MigThrError thr_error; | |
3601 | QEMUFile *fb; | |
3602 | bool early_fail = true; | |
3603 | ||
3604 | rcu_register_thread(); | |
3605 | object_ref(OBJECT(s)); | |
3606 | ||
3607 | qemu_file_set_rate_limit(s->to_dst_file, INT64_MAX); | |
3608 | ||
3609 | setup_start = qemu_clock_get_ms(QEMU_CLOCK_HOST); | |
3610 | /* | |
3611 | * We want to save vmstate for the moment when migration has been | |
3612 | * initiated but also we want to save RAM content while VM is running. | |
3613 | * The RAM content should appear first in the vmstate. So, we first | |
3614 | * stash the non-RAM part of the vmstate to the temporary buffer, | |
3615 | * then write RAM part of the vmstate to the migration stream | |
3616 | * with vCPUs running and, finally, write stashed non-RAM part of | |
3617 | * the vmstate from the buffer to the migration stream. | |
3618 | */ | |
ecb23efe | 3619 | s->bioc = qio_channel_buffer_new(512 * 1024); |
8518278a | 3620 | qio_channel_set_name(QIO_CHANNEL(s->bioc), "vmstate-buffer"); |
77ef2dc1 | 3621 | fb = qemu_file_new_output(QIO_CHANNEL(s->bioc)); |
8518278a AG |
3622 | object_unref(OBJECT(s->bioc)); |
3623 | ||
3624 | update_iteration_initial_status(s); | |
3625 | ||
eeccb99c AG |
3626 | /* |
3627 | * Prepare for tracking memory writes with UFFD-WP - populate | |
3628 | * RAM pages before protecting. | |
3629 | */ | |
3630 | #ifdef __linux__ | |
3631 | ram_write_tracking_prepare(); | |
3632 | #endif | |
3633 | ||
8518278a AG |
3634 | qemu_savevm_state_header(s->to_dst_file); |
3635 | qemu_savevm_state_setup(s->to_dst_file); | |
3636 | ||
fde93d99 LV |
3637 | qemu_savevm_wait_unplug(s, MIGRATION_STATUS_SETUP, |
3638 | MIGRATION_STATUS_ACTIVE); | |
8518278a | 3639 | |
8518278a AG |
3640 | s->setup_time = qemu_clock_get_ms(QEMU_CLOCK_HOST) - setup_start; |
3641 | ||
3642 | trace_migration_thread_setup_complete(); | |
3643 | s->downtime_start = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); | |
3644 | ||
3645 | qemu_mutex_lock_iothread(); | |
3646 | ||
3647 | /* | |
3648 | * If VM is currently in suspended state, then, to make a valid runstate | |
3649 | * transition in vm_stop_force_state() we need to wakeup it up. | |
3650 | */ | |
3651 | qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER, NULL); | |
3652 | s->vm_was_running = runstate_is_running(); | |
3653 | ||
3654 | if (global_state_store()) { | |
3655 | goto fail; | |
3656 | } | |
3657 | /* Forcibly stop VM before saving state of vCPUs and devices */ | |
3658 | if (vm_stop_force_state(RUN_STATE_PAUSED)) { | |
3659 | goto fail; | |
3660 | } | |
3661 | /* | |
3662 | * Put vCPUs in sync with shadow context structures, then | |
3663 | * save their state to channel-buffer along with devices. | |
3664 | */ | |
3665 | cpu_synchronize_all_states(); | |
3666 | if (qemu_savevm_state_complete_precopy_non_iterable(fb, false, false)) { | |
3667 | goto fail; | |
3668 | } | |
ecb23efe AG |
3669 | /* |
3670 | * Since we are going to get non-iterable state data directly | |
3671 | * from s->bioc->data, explicit flush is needed here. | |
3672 | */ | |
3673 | qemu_fflush(fb); | |
3674 | ||
8518278a AG |
3675 | /* Now initialize UFFD context and start tracking RAM writes */ |
3676 | if (ram_write_tracking_start()) { | |
3677 | goto fail; | |
3678 | } | |
3679 | early_fail = false; | |
3680 | ||
3681 | /* | |
3682 | * Start VM from BH handler to avoid write-fault lock here. | |
3683 | * UFFD-WP protection for the whole RAM is already enabled so | |
3684 | * calling VM state change notifiers from vm_start() would initiate | |
3685 | * writes to virtio VQs memory which is in write-protected region. | |
3686 | */ | |
3687 | s->vm_start_bh = qemu_bh_new(bg_migration_vm_start_bh, s); | |
3688 | qemu_bh_schedule(s->vm_start_bh); | |
3689 | ||
3690 | qemu_mutex_unlock_iothread(); | |
3691 | ||
3692 | while (migration_is_active(s)) { | |
3693 | MigIterateState iter_state = bg_migration_iteration_run(s); | |
3694 | if (iter_state == MIG_ITERATE_SKIP) { | |
3695 | continue; | |
3696 | } else if (iter_state == MIG_ITERATE_BREAK) { | |
3697 | break; | |
3698 | } | |
3699 | ||
3700 | /* | |
3701 | * Try to detect any kind of failures, and see whether we | |
3702 | * should stop the migration now. | |
3703 | */ | |
3704 | thr_error = migration_detect_error(s); | |
3705 | if (thr_error == MIG_THR_ERR_FATAL) { | |
3706 | /* Stop migration */ | |
3707 | break; | |
3708 | } | |
3709 | ||
3710 | migration_update_counters(s, qemu_clock_get_ms(QEMU_CLOCK_REALTIME)); | |
3711 | } | |
3712 | ||
3713 | trace_migration_thread_after_loop(); | |
3714 | ||
3715 | fail: | |
3716 | if (early_fail) { | |
3717 | migrate_set_state(&s->state, MIGRATION_STATUS_ACTIVE, | |
3718 | MIGRATION_STATUS_FAILED); | |
3719 | qemu_mutex_unlock_iothread(); | |
3720 | } | |
3721 | ||
3722 | bg_migration_iteration_finish(s); | |
3723 | ||
3724 | qemu_fclose(fb); | |
3725 | object_unref(OBJECT(s)); | |
3726 | rcu_unregister_thread(); | |
3727 | ||
3728 | return NULL; | |
3729 | } | |
3730 | ||
cce8040b | 3731 | void migrate_fd_connect(MigrationState *s, Error *error_in) |
0d82d0e8 | 3732 | { |
00f4b572 | 3733 | Error *local_err = NULL; |
d3e35b8f PX |
3734 | int64_t rate_limit; |
3735 | bool resume = s->state == MIGRATION_STATUS_POSTCOPY_PAUSED; | |
3736 | ||
ca7bd082 PX |
3737 | /* |
3738 | * If there's a previous error, free it and prepare for another one. | |
3739 | * Meanwhile if migration completes successfully, there won't have an error | |
3740 | * dumped when calling migrate_fd_cleanup(). | |
3741 | */ | |
3742 | migrate_error_free(s); | |
3743 | ||
2ff30257 | 3744 | s->expected_downtime = s->parameters.downtime_limit; |
9cbc3649 MAL |
3745 | if (resume) { |
3746 | assert(s->cleanup_bh); | |
3747 | } else { | |
3748 | assert(!s->cleanup_bh); | |
3749 | s->cleanup_bh = qemu_bh_new(migrate_fd_cleanup_bh, s); | |
3750 | } | |
cce8040b DDAG |
3751 | if (error_in) { |
3752 | migrate_fd_error(s, error_in); | |
ca30f24d PX |
3753 | if (resume) { |
3754 | /* | |
3755 | * Don't do cleanup for resume if channel is invalid, but only dump | |
3756 | * the error. We wait for another channel connect from the user. | |
3757 | * The error_report still gives HMP user a hint on what failed. | |
3758 | * It's normally done in migrate_fd_cleanup(), but call it here | |
3759 | * explicitly. | |
3760 | */ | |
3761 | error_report_err(error_copy(s->error)); | |
3762 | } else { | |
3763 | migrate_fd_cleanup(s); | |
3764 | } | |
cce8040b DDAG |
3765 | return; |
3766 | } | |
0d82d0e8 | 3767 | |
d3e35b8f PX |
3768 | if (resume) { |
3769 | /* This is a resumed migration */ | |
5390adec | 3770 | rate_limit = migrate_max_postcopy_bandwidth() / |
8504ddec | 3771 | XFER_LIMIT_RATIO; |
d3e35b8f PX |
3772 | } else { |
3773 | /* This is a fresh new migration */ | |
3774 | rate_limit = s->parameters.max_bandwidth / XFER_LIMIT_RATIO; | |
442773ce | 3775 | |
d3e35b8f PX |
3776 | /* Notify before starting migration thread */ |
3777 | notifier_list_notify(&migration_state_notifiers, s); | |
3778 | } | |
3779 | ||
3780 | qemu_file_set_rate_limit(s->to_dst_file, rate_limit); | |
3781 | qemu_file_set_blocking(s->to_dst_file, true); | |
9287ac27 | 3782 | |
1d34e4bf | 3783 | /* |
c788ada8 PX |
3784 | * Open the return path. For postcopy, it is used exclusively. For |
3785 | * precopy, only if user specified "return-path" capability would | |
3786 | * QEMU uses the return path. | |
1d34e4bf | 3787 | */ |
38ad1110 | 3788 | if (migrate_postcopy_ram() || migrate_return_path()) { |
d3e35b8f | 3789 | if (open_return_path_on_source(s, !resume)) { |
1d34e4bf | 3790 | error_report("Unable to open return-path for postcopy"); |
d3e35b8f | 3791 | migrate_set_state(&s->state, s->state, MIGRATION_STATUS_FAILED); |
1d34e4bf DDAG |
3792 | migrate_fd_cleanup(s); |
3793 | return; | |
3794 | } | |
3795 | } | |
3796 | ||
06064a67 PX |
3797 | /* |
3798 | * This needs to be done before resuming a postcopy. Note: for newer | |
3799 | * QEMUs we will delay the channel creation until postcopy_start(), to | |
3800 | * avoid disorder of channel creations. | |
3801 | */ | |
3802 | if (migrate_postcopy_preempt() && s->preempt_pre_7_2) { | |
3803 | postcopy_preempt_setup(s); | |
3804 | } | |
3805 | ||
d3e35b8f | 3806 | if (resume) { |
135b87b4 PX |
3807 | /* Wakeup the main migration thread to do the recovery */ |
3808 | migrate_set_state(&s->state, MIGRATION_STATUS_POSTCOPY_PAUSED, | |
3809 | MIGRATION_STATUS_POSTCOPY_RECOVER); | |
3810 | qemu_sem_post(&s->postcopy_pause_sem); | |
d3e35b8f PX |
3811 | return; |
3812 | } | |
3813 | ||
00f4b572 JQ |
3814 | if (multifd_save_setup(&local_err) != 0) { |
3815 | error_report_err(local_err); | |
f986c3d2 JQ |
3816 | migrate_set_state(&s->state, MIGRATION_STATUS_SETUP, |
3817 | MIGRATION_STATUS_FAILED); | |
3818 | migrate_fd_cleanup(s); | |
3819 | return; | |
3820 | } | |
8518278a AG |
3821 | |
3822 | if (migrate_background_snapshot()) { | |
3823 | qemu_thread_create(&s->thread, "bg_snapshot", | |
3824 | bg_migration_thread, s, QEMU_THREAD_JOINABLE); | |
3825 | } else { | |
3826 | qemu_thread_create(&s->thread, "live_migration", | |
3827 | migration_thread, s, QEMU_THREAD_JOINABLE); | |
3828 | } | |
1d34e4bf | 3829 | s->migration_thread_running = true; |
0d82d0e8 | 3830 | } |
093e3c42 | 3831 | |
20814758 | 3832 | #define DEFINE_PROP_MIG_CAP(name, x) \ |
0cec2056 | 3833 | DEFINE_PROP_BOOL(name, MigrationState, capabilities[x], false) |
20814758 | 3834 | |
5272298c PX |
3835 | static Property migration_properties[] = { |
3836 | DEFINE_PROP_BOOL("store-global-state", MigrationState, | |
3837 | store_global_state, true), | |
71dd4c1a PX |
3838 | DEFINE_PROP_BOOL("send-configuration", MigrationState, |
3839 | send_configuration, true), | |
15c38503 PX |
3840 | DEFINE_PROP_BOOL("send-section-footer", MigrationState, |
3841 | send_section_footer, true), | |
f548222c XG |
3842 | DEFINE_PROP_BOOL("decompress-error-check", MigrationState, |
3843 | decompress_error_check, true), | |
002cad6b PX |
3844 | DEFINE_PROP_UINT8("x-clear-bitmap-shift", MigrationState, |
3845 | clear_bitmap_shift, CLEAR_BITMAP_SHIFT_DEFAULT), | |
6621883f PX |
3846 | DEFINE_PROP_BOOL("x-preempt-pre-7-2", MigrationState, |
3847 | preempt_pre_7_2, false), | |
89632faf PX |
3848 | |
3849 | /* Migration parameters */ | |
741d4086 | 3850 | DEFINE_PROP_UINT8("x-compress-level", MigrationState, |
89632faf PX |
3851 | parameters.compress_level, |
3852 | DEFAULT_MIGRATE_COMPRESS_LEVEL), | |
741d4086 | 3853 | DEFINE_PROP_UINT8("x-compress-threads", MigrationState, |
89632faf PX |
3854 | parameters.compress_threads, |
3855 | DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT), | |
1d58872a XG |
3856 | DEFINE_PROP_BOOL("x-compress-wait-thread", MigrationState, |
3857 | parameters.compress_wait_thread, true), | |
741d4086 | 3858 | DEFINE_PROP_UINT8("x-decompress-threads", MigrationState, |
89632faf PX |
3859 | parameters.decompress_threads, |
3860 | DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT), | |
dc14a470 KZ |
3861 | DEFINE_PROP_UINT8("x-throttle-trigger-threshold", MigrationState, |
3862 | parameters.throttle_trigger_threshold, | |
3863 | DEFAULT_MIGRATE_THROTTLE_TRIGGER_THRESHOLD), | |
741d4086 | 3864 | DEFINE_PROP_UINT8("x-cpu-throttle-initial", MigrationState, |
89632faf PX |
3865 | parameters.cpu_throttle_initial, |
3866 | DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL), | |
741d4086 | 3867 | DEFINE_PROP_UINT8("x-cpu-throttle-increment", MigrationState, |
89632faf PX |
3868 | parameters.cpu_throttle_increment, |
3869 | DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT), | |
cbbf8182 KZ |
3870 | DEFINE_PROP_BOOL("x-cpu-throttle-tailslow", MigrationState, |
3871 | parameters.cpu_throttle_tailslow, false), | |
741d4086 | 3872 | DEFINE_PROP_SIZE("x-max-bandwidth", MigrationState, |
89632faf | 3873 | parameters.max_bandwidth, MAX_THROTTLE), |
741d4086 | 3874 | DEFINE_PROP_UINT64("x-downtime-limit", MigrationState, |
89632faf PX |
3875 | parameters.downtime_limit, |
3876 | DEFAULT_MIGRATE_SET_DOWNTIME), | |
741d4086 | 3877 | DEFINE_PROP_UINT32("x-checkpoint-delay", MigrationState, |
89632faf PX |
3878 | parameters.x_checkpoint_delay, |
3879 | DEFAULT_MIGRATE_X_CHECKPOINT_DELAY), | |
cbfd6c95 JQ |
3880 | DEFINE_PROP_UINT8("multifd-channels", MigrationState, |
3881 | parameters.multifd_channels, | |
4075fb1c | 3882 | DEFAULT_MIGRATE_MULTIFD_CHANNELS), |
96eef042 JQ |
3883 | DEFINE_PROP_MULTIFD_COMPRESSION("multifd-compression", MigrationState, |
3884 | parameters.multifd_compression, | |
3885 | DEFAULT_MIGRATE_MULTIFD_COMPRESSION), | |
9004db48 JQ |
3886 | DEFINE_PROP_UINT8("multifd-zlib-level", MigrationState, |
3887 | parameters.multifd_zlib_level, | |
3888 | DEFAULT_MIGRATE_MULTIFD_ZLIB_LEVEL), | |
6a9ad154 JQ |
3889 | DEFINE_PROP_UINT8("multifd-zstd-level", MigrationState, |
3890 | parameters.multifd_zstd_level, | |
3891 | DEFAULT_MIGRATE_MULTIFD_ZSTD_LEVEL), | |
73af8dd8 JQ |
3892 | DEFINE_PROP_SIZE("xbzrle-cache-size", MigrationState, |
3893 | parameters.xbzrle_cache_size, | |
3894 | DEFAULT_MIGRATE_XBZRLE_CACHE_SIZE), | |
7e555c6c DDAG |
3895 | DEFINE_PROP_SIZE("max-postcopy-bandwidth", MigrationState, |
3896 | parameters.max_postcopy_bandwidth, | |
3897 | DEFAULT_MIGRATE_MAX_POSTCOPY_BANDWIDTH), | |
4cbc9c7f LQ |
3898 | DEFINE_PROP_UINT8("max-cpu-throttle", MigrationState, |
3899 | parameters.max_cpu_throttle, | |
3900 | DEFAULT_MIGRATE_MAX_CPU_THROTTLE), | |
ee3d96ba DDAG |
3901 | DEFINE_PROP_SIZE("announce-initial", MigrationState, |
3902 | parameters.announce_initial, | |
3903 | DEFAULT_MIGRATE_ANNOUNCE_INITIAL), | |
3904 | DEFINE_PROP_SIZE("announce-max", MigrationState, | |
3905 | parameters.announce_max, | |
3906 | DEFAULT_MIGRATE_ANNOUNCE_MAX), | |
3907 | DEFINE_PROP_SIZE("announce-rounds", MigrationState, | |
3908 | parameters.announce_rounds, | |
3909 | DEFAULT_MIGRATE_ANNOUNCE_ROUNDS), | |
3910 | DEFINE_PROP_SIZE("announce-step", MigrationState, | |
3911 | parameters.announce_step, | |
3912 | DEFAULT_MIGRATE_ANNOUNCE_STEP), | |
9a266627 PX |
3913 | DEFINE_PROP_STRING("tls-creds", MigrationState, parameters.tls_creds), |
3914 | DEFINE_PROP_STRING("tls-hostname", MigrationState, parameters.tls_hostname), | |
3915 | DEFINE_PROP_STRING("tls-authz", MigrationState, parameters.tls_authz), | |
20814758 PX |
3916 | |
3917 | /* Migration capabilities */ | |
3918 | DEFINE_PROP_MIG_CAP("x-xbzrle", MIGRATION_CAPABILITY_XBZRLE), | |
3919 | DEFINE_PROP_MIG_CAP("x-rdma-pin-all", MIGRATION_CAPABILITY_RDMA_PIN_ALL), | |
3920 | DEFINE_PROP_MIG_CAP("x-auto-converge", MIGRATION_CAPABILITY_AUTO_CONVERGE), | |
3921 | DEFINE_PROP_MIG_CAP("x-zero-blocks", MIGRATION_CAPABILITY_ZERO_BLOCKS), | |
3922 | DEFINE_PROP_MIG_CAP("x-compress", MIGRATION_CAPABILITY_COMPRESS), | |
3923 | DEFINE_PROP_MIG_CAP("x-events", MIGRATION_CAPABILITY_EVENTS), | |
3924 | DEFINE_PROP_MIG_CAP("x-postcopy-ram", MIGRATION_CAPABILITY_POSTCOPY_RAM), | |
ce5b0f4a PX |
3925 | DEFINE_PROP_MIG_CAP("x-postcopy-preempt", |
3926 | MIGRATION_CAPABILITY_POSTCOPY_PREEMPT), | |
20814758 PX |
3927 | DEFINE_PROP_MIG_CAP("x-colo", MIGRATION_CAPABILITY_X_COLO), |
3928 | DEFINE_PROP_MIG_CAP("x-release-ram", MIGRATION_CAPABILITY_RELEASE_RAM), | |
3929 | DEFINE_PROP_MIG_CAP("x-block", MIGRATION_CAPABILITY_BLOCK), | |
3930 | DEFINE_PROP_MIG_CAP("x-return-path", MIGRATION_CAPABILITY_RETURN_PATH), | |
cbfd6c95 | 3931 | DEFINE_PROP_MIG_CAP("x-multifd", MIGRATION_CAPABILITY_MULTIFD), |
6e8c25b4 AG |
3932 | DEFINE_PROP_MIG_CAP("x-background-snapshot", |
3933 | MIGRATION_CAPABILITY_BACKGROUND_SNAPSHOT), | |
1abaec9a LB |
3934 | #ifdef CONFIG_LINUX |
3935 | DEFINE_PROP_MIG_CAP("x-zero-copy-send", | |
3936 | MIGRATION_CAPABILITY_ZERO_COPY_SEND), | |
3937 | #endif | |
20814758 | 3938 | |
5272298c PX |
3939 | DEFINE_PROP_END_OF_LIST(), |
3940 | }; | |
3941 | ||
e5cb7e76 PX |
3942 | static void migration_class_init(ObjectClass *klass, void *data) |
3943 | { | |
3944 | DeviceClass *dc = DEVICE_CLASS(klass); | |
3945 | ||
3946 | dc->user_creatable = false; | |
4f67d30b | 3947 | device_class_set_props(dc, migration_properties); |
e5cb7e76 PX |
3948 | } |
3949 | ||
b91bf5e4 MAL |
3950 | static void migration_instance_finalize(Object *obj) |
3951 | { | |
3952 | MigrationState *ms = MIGRATION_OBJ(obj); | |
b91bf5e4 | 3953 | |
87db1a7d | 3954 | qemu_mutex_destroy(&ms->error_mutex); |
62df066f | 3955 | qemu_mutex_destroy(&ms->qemu_file_lock); |
c7e0acd5 | 3956 | qemu_sem_destroy(&ms->wait_unplug_sem); |
ad767bed | 3957 | qemu_sem_destroy(&ms->rate_limit_sem); |
e91d8951 | 3958 | qemu_sem_destroy(&ms->pause_sem); |
b23c2ade | 3959 | qemu_sem_destroy(&ms->postcopy_pause_sem); |
14b1742e | 3960 | qemu_sem_destroy(&ms->postcopy_pause_rp_sem); |
edd090c7 | 3961 | qemu_sem_destroy(&ms->rp_state.rp_sem); |
b28fb582 | 3962 | qemu_sem_destroy(&ms->rp_state.rp_pong_acks); |
d0edb8a1 | 3963 | qemu_sem_destroy(&ms->postcopy_qemufile_src_sem); |
ab105cc1 | 3964 | error_free(ms->error); |
b91bf5e4 MAL |
3965 | } |
3966 | ||
e5cb7e76 PX |
3967 | static void migration_instance_init(Object *obj) |
3968 | { | |
3969 | MigrationState *ms = MIGRATION_OBJ(obj); | |
8b0b29dc | 3970 | MigrationParameters *params = &ms->parameters; |
e5cb7e76 PX |
3971 | |
3972 | ms->state = MIGRATION_STATUS_NONE; | |
e5cb7e76 | 3973 | ms->mbps = -1; |
aecbfe9c | 3974 | ms->pages_per_second = -1; |
e91d8951 | 3975 | qemu_sem_init(&ms->pause_sem, 0); |
87db1a7d | 3976 | qemu_mutex_init(&ms->error_mutex); |
8b0b29dc PX |
3977 | |
3978 | params->tls_hostname = g_strdup(""); | |
3979 | params->tls_creds = g_strdup(""); | |
3980 | ||
3981 | /* Set has_* up only for parameter checks */ | |
3982 | params->has_compress_level = true; | |
3983 | params->has_compress_threads = true; | |
df67aa3e | 3984 | params->has_compress_wait_thread = true; |
8b0b29dc | 3985 | params->has_decompress_threads = true; |
dc14a470 | 3986 | params->has_throttle_trigger_threshold = true; |
8b0b29dc PX |
3987 | params->has_cpu_throttle_initial = true; |
3988 | params->has_cpu_throttle_increment = true; | |
cbbf8182 | 3989 | params->has_cpu_throttle_tailslow = true; |
8b0b29dc PX |
3990 | params->has_max_bandwidth = true; |
3991 | params->has_downtime_limit = true; | |
3992 | params->has_x_checkpoint_delay = true; | |
3993 | params->has_block_incremental = true; | |
cbfd6c95 | 3994 | params->has_multifd_channels = true; |
96eef042 | 3995 | params->has_multifd_compression = true; |
9004db48 | 3996 | params->has_multifd_zlib_level = true; |
6a9ad154 | 3997 | params->has_multifd_zstd_level = true; |
73af8dd8 | 3998 | params->has_xbzrle_cache_size = true; |
7e555c6c | 3999 | params->has_max_postcopy_bandwidth = true; |
4cbc9c7f | 4000 | params->has_max_cpu_throttle = true; |
ee3d96ba DDAG |
4001 | params->has_announce_initial = true; |
4002 | params->has_announce_max = true; | |
4003 | params->has_announce_rounds = true; | |
4004 | params->has_announce_step = true; | |
b23c2ade PX |
4005 | |
4006 | qemu_sem_init(&ms->postcopy_pause_sem, 0); | |
14b1742e | 4007 | qemu_sem_init(&ms->postcopy_pause_rp_sem, 0); |
edd090c7 | 4008 | qemu_sem_init(&ms->rp_state.rp_sem, 0); |
b28fb582 | 4009 | qemu_sem_init(&ms->rp_state.rp_pong_acks, 0); |
ad767bed | 4010 | qemu_sem_init(&ms->rate_limit_sem, 0); |
c7e0acd5 | 4011 | qemu_sem_init(&ms->wait_unplug_sem, 0); |
d0edb8a1 | 4012 | qemu_sem_init(&ms->postcopy_qemufile_src_sem, 0); |
62df066f | 4013 | qemu_mutex_init(&ms->qemu_file_lock); |
8b0b29dc PX |
4014 | } |
4015 | ||
4016 | /* | |
4017 | * Return true if check pass, false otherwise. Error will be put | |
4018 | * inside errp if provided. | |
4019 | */ | |
4020 | static bool migration_object_check(MigrationState *ms, Error **errp) | |
4021 | { | |
6b19a7d9 | 4022 | /* Assuming all off */ |
b02c7fc9 | 4023 | bool old_caps[MIGRATION_CAPABILITY__MAX] = { 0 }; |
6b19a7d9 | 4024 | |
8b0b29dc PX |
4025 | if (!migrate_params_check(&ms->parameters, errp)) { |
4026 | return false; | |
4027 | } | |
4028 | ||
b02c7fc9 | 4029 | return migrate_caps_check(old_caps, ms->capabilities, errp); |
e5cb7e76 PX |
4030 | } |
4031 | ||
4032 | static const TypeInfo migration_type = { | |
4033 | .name = TYPE_MIGRATION, | |
01f6e14c | 4034 | /* |
c8d3ff38 | 4035 | * NOTE: TYPE_MIGRATION is not really a device, as the object is |
2194abd6 | 4036 | * not created using qdev_new(), it is not attached to the qdev |
c8d3ff38 PX |
4037 | * device tree, and it is never realized. |
4038 | * | |
4039 | * TODO: Make this TYPE_OBJECT once QOM provides something like | |
4040 | * TYPE_DEVICE's "-global" properties. | |
01f6e14c | 4041 | */ |
e5cb7e76 PX |
4042 | .parent = TYPE_DEVICE, |
4043 | .class_init = migration_class_init, | |
4044 | .class_size = sizeof(MigrationClass), | |
4045 | .instance_size = sizeof(MigrationState), | |
4046 | .instance_init = migration_instance_init, | |
b91bf5e4 | 4047 | .instance_finalize = migration_instance_finalize, |
e5cb7e76 PX |
4048 | }; |
4049 | ||
4050 | static void register_migration_types(void) | |
4051 | { | |
4052 | type_register_static(&migration_type); | |
4053 | } | |
4054 | ||
4055 | type_init(register_migration_types); |