]>
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" |
987772d9 | 34 | #include "migration/vmstate.h" |
737e150e | 35 | #include "block/block.h" |
e688df6b | 36 | #include "qapi/error.h" |
9aca82ba | 37 | #include "qapi/clone-visitor.h" |
31e4c354 | 38 | #include "qapi/qapi-visit-migration.h" |
9aca82ba | 39 | #include "qapi/qapi-visit-sockets.h" |
9af23989 MA |
40 | #include "qapi/qapi-commands-migration.h" |
41 | #include "qapi/qapi-events-migration.h" | |
cc7a8ea7 | 42 | #include "qapi/qmp/qerror.h" |
15280c36 | 43 | #include "qapi/qmp/qnull.h" |
ab28bd23 | 44 | #include "qemu/rcu.h" |
2c9e6fec | 45 | #include "block.h" |
be07b0ac | 46 | #include "postcopy-ram.h" |
766bd176 | 47 | #include "qemu/thread.h" |
c09e5bb1 | 48 | #include "trace.h" |
51180423 | 49 | #include "exec/target_page.h" |
61b67d47 | 50 | #include "io/channel-buffer.h" |
35a6ed4f | 51 | #include "migration/colo.h" |
4ffdb337 | 52 | #include "hw/boards.h" |
a27bd6c7 | 53 | #include "hw/qdev-properties.h" |
ce35e229 | 54 | #include "hw/qdev-properties-system.h" |
9d18af93 | 55 | #include "monitor/monitor.h" |
50510ea2 | 56 | #include "net/announce.h" |
c7e0acd5 | 57 | #include "qemu/queue.h" |
d32ca5ad | 58 | #include "multifd.h" |
b5eea99e | 59 | #include "qemu/yank.h" |
6e8c25b4 | 60 | #include "sysemu/cpus.h" |
39675fff | 61 | #include "yank_functions.h" |
1b529d90 | 62 | #include "sysemu/qtest.h" |
065e2813 | 63 | |
7590a2ae | 64 | #define MAX_THROTTLE (128 << 20) /* Migration transfer speed throttling */ |
5bb7910a | 65 | |
5b4e1eb7 JQ |
66 | /* Amount of time to allocate to each "chunk" of bandwidth-throttled |
67 | * data. */ | |
68 | #define BUFFER_DELAY 100 | |
69 | #define XFER_LIMIT_RATIO (1000 / BUFFER_DELAY) | |
70 | ||
2ff30257 AA |
71 | /* Time in milliseconds we are allowed to stop the source, |
72 | * for sending the last part */ | |
73 | #define DEFAULT_MIGRATE_SET_DOWNTIME 300 | |
74 | ||
87c9cc1c DHB |
75 | /* Maximum migrate downtime set to 2000 seconds */ |
76 | #define MAX_MIGRATE_DOWNTIME_SECONDS 2000 | |
77 | #define MAX_MIGRATE_DOWNTIME (MAX_MIGRATE_DOWNTIME_SECONDS * 1000) | |
78 | ||
8706d2d5 LL |
79 | /* Default compression thread count */ |
80 | #define DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT 8 | |
3fcb38c2 LL |
81 | /* Default decompression thread count, usually decompression is at |
82 | * least 4 times as fast as compression.*/ | |
83 | #define DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT 2 | |
8706d2d5 LL |
84 | /*0: means nocompress, 1: best speed, ... 9: best compress ratio */ |
85 | #define DEFAULT_MIGRATE_COMPRESS_LEVEL 1 | |
1626fee3 | 86 | /* Define default autoconverge cpu throttle migration parameters */ |
dc14a470 | 87 | #define DEFAULT_MIGRATE_THROTTLE_TRIGGER_THRESHOLD 50 |
d85a31d1 JH |
88 | #define DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL 20 |
89 | #define DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT 10 | |
4cbc9c7f | 90 | #define DEFAULT_MIGRATE_MAX_CPU_THROTTLE 99 |
8706d2d5 | 91 | |
17ad9b35 | 92 | /* Migration XBZRLE default cache size */ |
73af8dd8 | 93 | #define DEFAULT_MIGRATE_XBZRLE_CACHE_SIZE (64 * 1024 * 1024) |
17ad9b35 | 94 | |
131b2153 ZC |
95 | /* The delay time (in ms) between two COLO checkpoints */ |
96 | #define DEFAULT_MIGRATE_X_CHECKPOINT_DELAY (200 * 100) | |
4075fb1c | 97 | #define DEFAULT_MIGRATE_MULTIFD_CHANNELS 2 |
96eef042 | 98 | #define DEFAULT_MIGRATE_MULTIFD_COMPRESSION MULTIFD_COMPRESSION_NONE |
9004db48 JQ |
99 | /* 0: means nocompress, 1: best speed, ... 9: best compress ratio */ |
100 | #define DEFAULT_MIGRATE_MULTIFD_ZLIB_LEVEL 1 | |
6a9ad154 JQ |
101 | /* 0: means nocompress, 1: best speed, ... 20: best compress ratio */ |
102 | #define DEFAULT_MIGRATE_MULTIFD_ZSTD_LEVEL 1 | |
68b53591 | 103 | |
7e555c6c DDAG |
104 | /* Background transfer rate for postcopy, 0 means unlimited, note |
105 | * that page requests can still exceed this limit. | |
106 | */ | |
107 | #define DEFAULT_MIGRATE_MAX_POSTCOPY_BANDWIDTH 0 | |
108 | ||
ee3d96ba DDAG |
109 | /* |
110 | * Parameters for self_announce_delay giving a stream of RARP/ARP | |
111 | * packets after migration. | |
112 | */ | |
113 | #define DEFAULT_MIGRATE_ANNOUNCE_INITIAL 50 | |
114 | #define DEFAULT_MIGRATE_ANNOUNCE_MAX 550 | |
115 | #define DEFAULT_MIGRATE_ANNOUNCE_ROUNDS 5 | |
116 | #define DEFAULT_MIGRATE_ANNOUNCE_STEP 100 | |
117 | ||
99a0db9b GH |
118 | static NotifierList migration_state_notifiers = |
119 | NOTIFIER_LIST_INITIALIZER(migration_state_notifiers); | |
120 | ||
da6f1790 JQ |
121 | /* Messages sent on the return path from destination to source */ |
122 | enum mig_rp_message_type { | |
123 | MIG_RP_MSG_INVALID = 0, /* Must be 0 */ | |
124 | MIG_RP_MSG_SHUT, /* sibling will not send any more RP messages */ | |
125 | MIG_RP_MSG_PONG, /* Response to a PING; data (seq: be32 ) */ | |
126 | ||
127 | MIG_RP_MSG_REQ_PAGES_ID, /* data (start: be64, len: be32, id: string) */ | |
128 | MIG_RP_MSG_REQ_PAGES, /* data (start: be64, len: be32) */ | |
a335debb | 129 | MIG_RP_MSG_RECV_BITMAP, /* send recved_bitmap back to source */ |
13955b89 | 130 | MIG_RP_MSG_RESUME_ACK, /* tell source that we are ready to resume */ |
da6f1790 JQ |
131 | |
132 | MIG_RP_MSG_MAX | |
133 | }; | |
134 | ||
6e8c25b4 AG |
135 | /* Migration capabilities set */ |
136 | struct MigrateCapsSet { | |
137 | int size; /* Capability set size */ | |
138 | MigrationCapability caps[]; /* Variadic array of capabilities */ | |
139 | }; | |
140 | typedef struct MigrateCapsSet MigrateCapsSet; | |
141 | ||
142 | /* Define and initialize MigrateCapsSet */ | |
143 | #define INITIALIZE_MIGRATE_CAPS_SET(_name, ...) \ | |
144 | MigrateCapsSet _name = { \ | |
145 | .size = sizeof((int []) { __VA_ARGS__ }) / sizeof(int), \ | |
146 | .caps = { __VA_ARGS__ } \ | |
147 | } | |
148 | ||
149 | /* Background-snapshot compatibility check list */ | |
150 | static const | |
151 | INITIALIZE_MIGRATE_CAPS_SET(check_caps_background_snapshot, | |
152 | MIGRATION_CAPABILITY_POSTCOPY_RAM, | |
153 | MIGRATION_CAPABILITY_DIRTY_BITMAPS, | |
154 | MIGRATION_CAPABILITY_POSTCOPY_BLOCKTIME, | |
155 | MIGRATION_CAPABILITY_LATE_BLOCK_ACTIVATE, | |
156 | MIGRATION_CAPABILITY_RETURN_PATH, | |
157 | MIGRATION_CAPABILITY_MULTIFD, | |
158 | MIGRATION_CAPABILITY_PAUSE_BEFORE_SWITCHOVER, | |
159 | MIGRATION_CAPABILITY_AUTO_CONVERGE, | |
160 | MIGRATION_CAPABILITY_RELEASE_RAM, | |
161 | MIGRATION_CAPABILITY_RDMA_PIN_ALL, | |
162 | MIGRATION_CAPABILITY_COMPRESS, | |
163 | MIGRATION_CAPABILITY_XBZRLE, | |
164 | MIGRATION_CAPABILITY_X_COLO, | |
1abaec9a LB |
165 | MIGRATION_CAPABILITY_VALIDATE_UUID, |
166 | MIGRATION_CAPABILITY_ZERO_COPY_SEND); | |
6e8c25b4 | 167 | |
17549e84 JQ |
168 | /* When we add fault tolerance, we could have several |
169 | migrations at once. For now we don't need to add | |
170 | dynamic creation of migration */ | |
171 | ||
e5cb7e76 | 172 | static MigrationState *current_migration; |
e1b1b1bc | 173 | static MigrationIncomingState *current_incoming; |
e5cb7e76 | 174 | |
3af8554b DDAG |
175 | static GSList *migration_blockers; |
176 | ||
8b0b29dc | 177 | static bool migration_object_check(MigrationState *ms, Error **errp); |
0331c8ca DDAG |
178 | static int migration_maybe_pause(MigrationState *s, |
179 | int *current_active_state, | |
180 | int new_state); | |
892ae715 | 181 | static void migrate_fd_cancel(MigrationState *s); |
8b0b29dc | 182 | |
f444eeda PX |
183 | static bool migrate_allow_multi_channels = true; |
184 | ||
185 | void migrate_protocol_allow_multi_channels(bool allow) | |
186 | { | |
187 | migrate_allow_multi_channels = allow; | |
188 | } | |
189 | ||
190 | bool migrate_multi_channels_is_allowed(void) | |
191 | { | |
192 | return migrate_allow_multi_channels; | |
193 | } | |
194 | ||
8f8bfffc PX |
195 | static gint page_request_addr_cmp(gconstpointer ap, gconstpointer bp) |
196 | { | |
197 | uintptr_t a = (uintptr_t) ap, b = (uintptr_t) bp; | |
198 | ||
199 | return (a > b) - (a < b); | |
200 | } | |
201 | ||
e5cb7e76 PX |
202 | void migration_object_init(void) |
203 | { | |
204 | /* This can only be called once. */ | |
205 | assert(!current_migration); | |
206 | current_migration = MIGRATION_OBJ(object_new(TYPE_MIGRATION)); | |
4ffdb337 | 207 | |
e1b1b1bc PX |
208 | /* |
209 | * Init the migrate incoming object as well no matter whether | |
210 | * we'll use it or not. | |
211 | */ | |
212 | assert(!current_incoming); | |
213 | current_incoming = g_new0(MigrationIncomingState, 1); | |
214 | current_incoming->state = MIGRATION_STATUS_NONE; | |
215 | current_incoming->postcopy_remote_fds = | |
216 | g_array_new(FALSE, TRUE, sizeof(struct PostCopyFD)); | |
217 | qemu_mutex_init(¤t_incoming->rp_mutex); | |
218 | qemu_event_init(¤t_incoming->main_thread_load_event, false); | |
219 | qemu_sem_init(¤t_incoming->postcopy_pause_sem_dst, 0); | |
220 | qemu_sem_init(¤t_incoming->postcopy_pause_sem_fault, 0); | |
8f8bfffc PX |
221 | qemu_mutex_init(¤t_incoming->page_request_mutex); |
222 | current_incoming->page_requested = g_tree_new(page_request_addr_cmp); | |
e1b1b1bc | 223 | |
f9734d5d | 224 | migration_object_check(current_migration, &error_fatal); |
e0d17dfd PB |
225 | |
226 | blk_mig_init(); | |
227 | ram_mig_init(); | |
228 | dirty_bitmap_mig_init(); | |
e5cb7e76 PX |
229 | } |
230 | ||
458fecca | 231 | void migration_cancel(const Error *error) |
c7c0e724 | 232 | { |
458fecca LV |
233 | if (error) { |
234 | migrate_set_error(current_migration, error); | |
235 | } | |
c7c0e724 DH |
236 | migrate_fd_cancel(current_migration); |
237 | } | |
238 | ||
892ae715 | 239 | void migration_shutdown(void) |
1f895604 | 240 | { |
795969ab RL |
241 | /* |
242 | * When the QEMU main thread exit, the COLO thread | |
243 | * may wait a semaphore. So, we should wakeup the | |
244 | * COLO thread before migration shutdown. | |
245 | */ | |
246 | colo_shutdown(); | |
892ae715 DDAG |
247 | /* |
248 | * Cancel the current migration - that will (eventually) | |
249 | * stop the migration using this structure | |
250 | */ | |
458fecca | 251 | migration_cancel(NULL); |
1f895604 | 252 | object_unref(OBJECT(current_migration)); |
1499ab09 VSO |
253 | |
254 | /* | |
255 | * Cancel outgoing migration of dirty bitmaps. It should | |
256 | * at least unref used block nodes. | |
257 | */ | |
258 | dirty_bitmap_mig_cancel_outgoing(); | |
259 | ||
260 | /* | |
261 | * Cancel incoming migration of dirty bitmaps. Dirty bitmaps | |
262 | * are non-critical data, and their loss never considered as | |
263 | * something serious. | |
264 | */ | |
265 | dirty_bitmap_mig_cancel_incoming(); | |
1f895604 VSO |
266 | } |
267 | ||
bca7856a | 268 | /* For outgoing */ |
859bc756 | 269 | MigrationState *migrate_get_current(void) |
17549e84 | 270 | { |
e5cb7e76 PX |
271 | /* This can only be called after the object created. */ |
272 | assert(current_migration); | |
273 | return current_migration; | |
17549e84 JQ |
274 | } |
275 | ||
bca7856a DDAG |
276 | MigrationIncomingState *migration_incoming_get_current(void) |
277 | { | |
e1b1b1bc PX |
278 | assert(current_incoming); |
279 | return current_incoming; | |
bca7856a DDAG |
280 | } |
281 | ||
e031149c PX |
282 | void migration_incoming_transport_cleanup(MigrationIncomingState *mis) |
283 | { | |
284 | if (mis->socket_address_list) { | |
285 | qapi_free_SocketAddressList(mis->socket_address_list); | |
286 | mis->socket_address_list = NULL; | |
287 | } | |
288 | ||
289 | if (mis->transport_cleanup) { | |
290 | mis->transport_cleanup(mis->transport_data); | |
291 | mis->transport_data = mis->transport_cleanup = NULL; | |
292 | } | |
293 | } | |
294 | ||
bca7856a DDAG |
295 | void migration_incoming_state_destroy(void) |
296 | { | |
b4b076da JQ |
297 | struct MigrationIncomingState *mis = migration_incoming_get_current(); |
298 | ||
3482655b | 299 | if (mis->to_src_file) { |
660819b1 PX |
300 | /* Tell source that we are done */ |
301 | migrate_send_rp_shut(mis, qemu_file_get_error(mis->from_src_file) != 0); | |
3482655b PX |
302 | qemu_fclose(mis->to_src_file); |
303 | mis->to_src_file = NULL; | |
304 | } | |
305 | ||
660819b1 | 306 | if (mis->from_src_file) { |
39675fff | 307 | migration_ioc_unregister_yank_from_file(mis->from_src_file); |
660819b1 PX |
308 | qemu_fclose(mis->from_src_file); |
309 | mis->from_src_file = NULL; | |
310 | } | |
00fa4fc8 DDAG |
311 | if (mis->postcopy_remote_fds) { |
312 | g_array_free(mis->postcopy_remote_fds, TRUE); | |
313 | mis->postcopy_remote_fds = NULL; | |
314 | } | |
660819b1 | 315 | |
e031149c | 316 | migration_incoming_transport_cleanup(mis); |
1783c00f DDAG |
317 | qemu_event_reset(&mis->main_thread_load_event); |
318 | ||
8f8bfffc PX |
319 | if (mis->page_requested) { |
320 | g_tree_destroy(mis->page_requested); | |
321 | mis->page_requested = NULL; | |
322 | } | |
323 | ||
b5eea99e | 324 | yank_unregister_instance(MIGRATION_YANK_INSTANCE); |
bca7856a DDAG |
325 | } |
326 | ||
b05dc723 JQ |
327 | static void migrate_generate_event(int new_state) |
328 | { | |
329 | if (migrate_use_events()) { | |
3ab72385 | 330 | qapi_event_send_migration(new_state); |
b05dc723 JQ |
331 | } |
332 | } | |
333 | ||
0f073f44 DDAG |
334 | static bool migrate_late_block_activate(void) |
335 | { | |
336 | MigrationState *s; | |
337 | ||
338 | s = migrate_get_current(); | |
339 | ||
340 | return s->enabled_capabilities[ | |
341 | MIGRATION_CAPABILITY_LATE_BLOCK_ACTIVATE]; | |
342 | } | |
343 | ||
da6f1790 JQ |
344 | /* |
345 | * Send a message on the return channel back to the source | |
346 | * of the migration. | |
347 | */ | |
d6208e35 PX |
348 | static int migrate_send_rp_message(MigrationIncomingState *mis, |
349 | enum mig_rp_message_type message_type, | |
350 | uint16_t len, void *data) | |
da6f1790 | 351 | { |
d6208e35 PX |
352 | int ret = 0; |
353 | ||
da6f1790 | 354 | trace_migrate_send_rp_message((int)message_type, len); |
37396950 | 355 | QEMU_LOCK_GUARD(&mis->rp_mutex); |
d6208e35 PX |
356 | |
357 | /* | |
358 | * It's possible that the file handle got lost due to network | |
359 | * failures. | |
360 | */ | |
361 | if (!mis->to_src_file) { | |
362 | ret = -EIO; | |
37396950 | 363 | return ret; |
d6208e35 PX |
364 | } |
365 | ||
da6f1790 JQ |
366 | qemu_put_be16(mis->to_src_file, (unsigned int)message_type); |
367 | qemu_put_be16(mis->to_src_file, len); | |
368 | qemu_put_buffer(mis->to_src_file, data, len); | |
369 | qemu_fflush(mis->to_src_file); | |
d6208e35 PX |
370 | |
371 | /* It's possible that qemu file got error during sending */ | |
372 | ret = qemu_file_get_error(mis->to_src_file); | |
373 | ||
d6208e35 | 374 | return ret; |
da6f1790 JQ |
375 | } |
376 | ||
2e2bce16 PX |
377 | /* Request one page from the source VM at the given start address. |
378 | * rb: the RAMBlock to request the page in | |
1e2d90eb DDAG |
379 | * Start: Address offset within the RB |
380 | * Len: Length in bytes required - must be a multiple of pagesize | |
381 | */ | |
7a267fc4 PX |
382 | int migrate_send_rp_message_req_pages(MigrationIncomingState *mis, |
383 | RAMBlock *rb, ram_addr_t start) | |
1e2d90eb | 384 | { |
cb8d4c8f | 385 | uint8_t bufc[12 + 1 + 255]; /* start (8), len (4), rbname up to 256 */ |
1e2d90eb | 386 | size_t msglen = 12; /* start + len */ |
2e2bce16 | 387 | size_t len = qemu_ram_pagesize(rb); |
d6208e35 | 388 | enum mig_rp_message_type msg_type; |
2e2bce16 PX |
389 | const char *rbname; |
390 | int rbname_len; | |
1e2d90eb DDAG |
391 | |
392 | *(uint64_t *)bufc = cpu_to_be64((uint64_t)start); | |
393 | *(uint32_t *)(bufc + 8) = cpu_to_be32((uint32_t)len); | |
394 | ||
2e2bce16 PX |
395 | /* |
396 | * We maintain the last ramblock that we requested for page. Note that we | |
397 | * don't need locking because this function will only be called within the | |
398 | * postcopy ram fault thread. | |
399 | */ | |
400 | if (rb != mis->last_rb) { | |
401 | mis->last_rb = rb; | |
402 | ||
403 | rbname = qemu_ram_get_idstr(rb); | |
404 | rbname_len = strlen(rbname); | |
405 | ||
1e2d90eb DDAG |
406 | assert(rbname_len < 256); |
407 | ||
408 | bufc[msglen++] = rbname_len; | |
409 | memcpy(bufc + msglen, rbname, rbname_len); | |
410 | msglen += rbname_len; | |
d6208e35 | 411 | msg_type = MIG_RP_MSG_REQ_PAGES_ID; |
1e2d90eb | 412 | } else { |
d6208e35 | 413 | msg_type = MIG_RP_MSG_REQ_PAGES; |
1e2d90eb | 414 | } |
d6208e35 PX |
415 | |
416 | return migrate_send_rp_message(mis, msg_type, msglen, bufc); | |
1e2d90eb DDAG |
417 | } |
418 | ||
7a267fc4 | 419 | int migrate_send_rp_req_pages(MigrationIncomingState *mis, |
8f8bfffc | 420 | RAMBlock *rb, ram_addr_t start, uint64_t haddr) |
7a267fc4 | 421 | { |
7648297d | 422 | void *aligned = (void *)(uintptr_t)ROUND_DOWN(haddr, qemu_ram_pagesize(rb)); |
a2429283 | 423 | bool received = false; |
8f8bfffc PX |
424 | |
425 | WITH_QEMU_LOCK_GUARD(&mis->page_request_mutex) { | |
426 | received = ramblock_recv_bitmap_test_byte_offset(rb, start); | |
427 | if (!received && !g_tree_lookup(mis->page_requested, aligned)) { | |
428 | /* | |
429 | * The page has not been received, and it's not yet in the page | |
430 | * request list. Queue it. Set the value of element to 1, so that | |
431 | * things like g_tree_lookup() will return TRUE (1) when found. | |
432 | */ | |
433 | g_tree_insert(mis->page_requested, aligned, (gpointer)1); | |
434 | mis->page_requested_count++; | |
435 | trace_postcopy_page_req_add(aligned, mis->page_requested_count); | |
436 | } | |
437 | } | |
438 | ||
439 | /* | |
440 | * If the page is there, skip sending the message. We don't even need the | |
441 | * lock because as long as the page arrived, it'll be there forever. | |
442 | */ | |
443 | if (received) { | |
444 | return 0; | |
445 | } | |
446 | ||
7a267fc4 PX |
447 | return migrate_send_rp_message_req_pages(mis, rb, start); |
448 | } | |
449 | ||
aad555c2 ZC |
450 | static bool migration_colo_enabled; |
451 | bool migration_incoming_colo_enabled(void) | |
452 | { | |
453 | return migration_colo_enabled; | |
454 | } | |
455 | ||
456 | void migration_incoming_disable_colo(void) | |
457 | { | |
18b1d3c9 | 458 | ram_block_discard_disable(false); |
aad555c2 ZC |
459 | migration_colo_enabled = false; |
460 | } | |
461 | ||
18b1d3c9 | 462 | int migration_incoming_enable_colo(void) |
aad555c2 | 463 | { |
18b1d3c9 DH |
464 | if (ram_block_discard_disable(true)) { |
465 | error_report("COLO: cannot disable RAM discard"); | |
466 | return -EBUSY; | |
467 | } | |
aad555c2 | 468 | migration_colo_enabled = true; |
18b1d3c9 | 469 | return 0; |
aad555c2 ZC |
470 | } |
471 | ||
9aca82ba JQ |
472 | void migrate_add_address(SocketAddress *address) |
473 | { | |
474 | MigrationIncomingState *mis = migration_incoming_get_current(); | |
9aca82ba | 475 | |
54aa3de7 EB |
476 | QAPI_LIST_PREPEND(mis->socket_address_list, |
477 | QAPI_CLONE(SocketAddress, address)); | |
9aca82ba JQ |
478 | } |
479 | ||
e69d50d6 | 480 | static void qemu_start_incoming_migration(const char *uri, Error **errp) |
5bb7910a | 481 | { |
d658f65c | 482 | const char *p = NULL; |
34c9dd8e | 483 | |
f444eeda | 484 | migrate_protocol_allow_multi_channels(false); /* reset it anyway */ |
3ab72385 | 485 | qapi_event_send_migration(MIGRATION_STATUS_SETUP); |
e69d50d6 PB |
486 | if (strstart(uri, "tcp:", &p) || |
487 | strstart(uri, "unix:", NULL) || | |
488 | strstart(uri, "vsock:", NULL)) { | |
f444eeda | 489 | migrate_protocol_allow_multi_channels(true); |
d658f65c | 490 | socket_start_incoming_migration(p ? p : uri, errp); |
2da776db | 491 | #ifdef CONFIG_RDMA |
adde220a | 492 | } else if (strstart(uri, "rdma:", &p)) { |
2da776db MH |
493 | rdma_start_incoming_migration(p, errp); |
494 | #endif | |
adde220a | 495 | } else if (strstart(uri, "exec:", &p)) { |
43eaae28 | 496 | exec_start_incoming_migration(p, errp); |
adde220a | 497 | } else if (strstart(uri, "fd:", &p)) { |
43eaae28 | 498 | fd_start_incoming_migration(p, errp); |
adde220a | 499 | } else { |
312fd5f2 | 500 | error_setg(errp, "unknown migration protocol: %s", uri); |
8ca5e801 | 501 | } |
5bb7910a AL |
502 | } |
503 | ||
0aa6aefc DL |
504 | static void process_incoming_migration_bh(void *opaque) |
505 | { | |
506 | Error *local_err = NULL; | |
507 | MigrationIncomingState *mis = opaque; | |
508 | ||
0f073f44 DDAG |
509 | /* If capability late_block_activate is set: |
510 | * Only fire up the block code now if we're going to restart the | |
511 | * VM, else 'cont' will do it. | |
512 | * This causes file locking to happen; so we don't want it to happen | |
513 | * unless we really are starting the VM. | |
514 | */ | |
515 | if (!migrate_late_block_activate() || | |
516 | (autostart && (!global_state_received() || | |
517 | global_state_get_runstate() == RUN_STATE_RUNNING))) { | |
3b717194 | 518 | /* Make sure all file formats throw away their mutable metadata. |
0f073f44 | 519 | * If we get an error here, just don't restart the VM yet. */ |
3b717194 | 520 | bdrv_activate_all(&local_err); |
0f073f44 DDAG |
521 | if (local_err) { |
522 | error_report_err(local_err); | |
523 | local_err = NULL; | |
524 | autostart = false; | |
525 | } | |
d35ff5e6 KW |
526 | } |
527 | ||
0aa6aefc DL |
528 | /* |
529 | * This must happen after all error conditions are dealt with and | |
530 | * we're sure the VM is going to be running on this host. | |
531 | */ | |
7659505c | 532 | qemu_announce_self(&mis->announce_timer, migrate_announce_params()); |
0aa6aefc | 533 | |
f986c3d2 JQ |
534 | if (multifd_load_cleanup(&local_err) != 0) { |
535 | error_report_err(local_err); | |
536 | autostart = false; | |
537 | } | |
0aa6aefc DL |
538 | /* If global state section was not received or we are in running |
539 | state, we need to obey autostart. Any other state is set with | |
540 | runstate_set. */ | |
541 | ||
b35ebdf0 VSO |
542 | dirty_bitmap_mig_before_vm_start(); |
543 | ||
0aa6aefc DL |
544 | if (!global_state_received() || |
545 | global_state_get_runstate() == RUN_STATE_RUNNING) { | |
546 | if (autostart) { | |
547 | vm_start(); | |
548 | } else { | |
549 | runstate_set(RUN_STATE_PAUSED); | |
550 | } | |
db009729 ZC |
551 | } else if (migration_incoming_colo_enabled()) { |
552 | migration_incoming_disable_colo(); | |
553 | vm_start(); | |
0aa6aefc DL |
554 | } else { |
555 | runstate_set(global_state_get_runstate()); | |
556 | } | |
0aa6aefc DL |
557 | /* |
558 | * This must happen after any state changes since as soon as an external | |
559 | * observer sees this event they might start to prod at the VM assuming | |
560 | * it's ready to use. | |
561 | */ | |
562 | migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE, | |
563 | MIGRATION_STATUS_COMPLETED); | |
564 | qemu_bh_delete(mis->bh); | |
565 | migration_incoming_state_destroy(); | |
566 | } | |
567 | ||
82a4da79 | 568 | static void process_incoming_migration_co(void *opaque) |
511c0231 | 569 | { |
b4b076da | 570 | MigrationIncomingState *mis = migration_incoming_get_current(); |
e9bef235 | 571 | PostcopyState ps; |
1c12e1f5 | 572 | int ret; |
8e48ac95 | 573 | Error *local_err = NULL; |
1c12e1f5 | 574 | |
4f0fae7f | 575 | assert(mis->from_src_file); |
92370989 | 576 | mis->migration_incoming_co = qemu_coroutine_self(); |
67f11b5c | 577 | mis->largest_page_size = qemu_ram_pagesize_largest(); |
093e3c42 | 578 | postcopy_state_set(POSTCOPY_INCOMING_NONE); |
93d7af6f HZ |
579 | migrate_set_state(&mis->state, MIGRATION_STATUS_NONE, |
580 | MIGRATION_STATUS_ACTIVE); | |
4f0fae7f | 581 | ret = qemu_loadvm_state(mis->from_src_file); |
bca7856a | 582 | |
e9bef235 DDAG |
583 | ps = postcopy_state_get(); |
584 | trace_process_incoming_migration_co_end(ret, ps); | |
585 | if (ps != POSTCOPY_INCOMING_NONE) { | |
586 | if (ps == POSTCOPY_INCOMING_ADVISE) { | |
587 | /* | |
588 | * Where a migration had postcopy enabled (and thus went to advise) | |
589 | * but managed to complete within the precopy period, we can use | |
590 | * the normal exit. | |
591 | */ | |
592 | postcopy_ram_incoming_cleanup(mis); | |
593 | } else if (ret >= 0) { | |
594 | /* | |
595 | * Postcopy was started, cleanup should happen at the end of the | |
596 | * postcopy thread. | |
597 | */ | |
598 | trace_process_incoming_migration_co_postcopy_end_main(); | |
599 | return; | |
600 | } | |
601 | /* Else if something went wrong then just fall out of the normal exit */ | |
602 | } | |
603 | ||
25d0c16f | 604 | /* we get COLO info, and know if we are in COLO mode */ |
aad555c2 | 605 | if (!ret && migration_incoming_colo_enabled()) { |
3b717194 EGE |
606 | /* Make sure all file formats throw away their mutable metadata */ |
607 | bdrv_activate_all(&local_err); | |
8e48ac95 | 608 | if (local_err) { |
8e48ac95 | 609 | error_report_err(local_err); |
6d99c2d4 | 610 | goto fail; |
8e48ac95 ZC |
611 | } |
612 | ||
25d0c16f HZ |
613 | qemu_thread_create(&mis->colo_incoming_thread, "COLO incoming", |
614 | colo_process_incoming_thread, mis, QEMU_THREAD_JOINABLE); | |
615 | mis->have_colo_incoming_thread = true; | |
616 | qemu_coroutine_yield(); | |
617 | ||
684bfd18 | 618 | qemu_mutex_unlock_iothread(); |
25d0c16f HZ |
619 | /* Wait checkpoint incoming thread exit before free resource */ |
620 | qemu_thread_join(&mis->colo_incoming_thread); | |
684bfd18 | 621 | qemu_mutex_lock_iothread(); |
13af18f2 ZC |
622 | /* We hold the global iothread lock, so it is safe here */ |
623 | colo_release_ram_cache(); | |
25d0c16f HZ |
624 | } |
625 | ||
1c12e1f5 | 626 | if (ret < 0) { |
db80face | 627 | error_report("load of migration failed: %s", strerror(-ret)); |
6d99c2d4 | 628 | goto fail; |
511c0231 | 629 | } |
0aa6aefc DL |
630 | mis->bh = qemu_bh_new(process_incoming_migration_bh, mis); |
631 | qemu_bh_schedule(mis->bh); | |
92370989 | 632 | mis->migration_incoming_co = NULL; |
6d99c2d4 FL |
633 | return; |
634 | fail: | |
635 | local_err = NULL; | |
636 | migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE, | |
637 | MIGRATION_STATUS_FAILED); | |
638 | qemu_fclose(mis->from_src_file); | |
639 | if (multifd_load_cleanup(&local_err) != 0) { | |
640 | error_report_err(local_err); | |
641 | } | |
642 | exit(EXIT_FAILURE); | |
511c0231 JQ |
643 | } |
644 | ||
b673eab4 | 645 | /** |
7d6f6933 | 646 | * migration_incoming_setup: Setup incoming migration |
b673eab4 JQ |
647 | * @f: file for main migration channel |
648 | * @errp: where to put errors | |
7d6f6933 MA |
649 | * |
650 | * Returns: %true on success, %false on error. | |
b673eab4 | 651 | */ |
7d6f6933 | 652 | static bool migration_incoming_setup(QEMUFile *f, Error **errp) |
82a4da79 | 653 | { |
4f0fae7f | 654 | MigrationIncomingState *mis = migration_incoming_get_current(); |
82a4da79 | 655 | |
7d6f6933 MA |
656 | if (multifd_load_setup(errp) != 0) { |
657 | return false; | |
f986c3d2 JQ |
658 | } |
659 | ||
4f0fae7f JQ |
660 | if (!mis->from_src_file) { |
661 | mis->from_src_file = f; | |
662 | } | |
06ad5135 | 663 | qemu_file_set_blocking(f, false); |
7d6f6933 | 664 | return true; |
e595a01a JQ |
665 | } |
666 | ||
36c2f8be | 667 | void migration_incoming_process(void) |
e595a01a JQ |
668 | { |
669 | Coroutine *co = qemu_coroutine_create(process_incoming_migration_co, NULL); | |
0b8b8753 | 670 | qemu_coroutine_enter(co); |
82a4da79 PB |
671 | } |
672 | ||
884835fa | 673 | /* Returns true if recovered from a paused migration, otherwise false */ |
a39e9339 | 674 | static bool postcopy_try_recover(void) |
e595a01a | 675 | { |
d96c9e8d PX |
676 | MigrationIncomingState *mis = migration_incoming_get_current(); |
677 | ||
678 | if (mis->state == MIGRATION_STATUS_POSTCOPY_PAUSED) { | |
679 | /* Resumed from a paused postcopy migration */ | |
680 | ||
a39e9339 PX |
681 | /* This should be set already in migration_incoming_setup() */ |
682 | assert(mis->from_src_file); | |
d96c9e8d | 683 | /* Postcopy has standalone thread to do vm load */ |
a39e9339 | 684 | qemu_file_set_blocking(mis->from_src_file, true); |
d96c9e8d PX |
685 | |
686 | /* Re-configure the return path */ | |
a39e9339 | 687 | mis->to_src_file = qemu_file_get_return_path(mis->from_src_file); |
d96c9e8d PX |
688 | |
689 | migrate_set_state(&mis->state, MIGRATION_STATUS_POSTCOPY_PAUSED, | |
690 | MIGRATION_STATUS_POSTCOPY_RECOVER); | |
691 | ||
692 | /* | |
693 | * Here, we only wake up the main loading thread (while the | |
694 | * fault thread will still be waiting), so that we can receive | |
695 | * commands from source now, and answer it if needed. The | |
696 | * fault thread will be woken up afterwards until we are sure | |
697 | * that source is ready to reply to page requests. | |
698 | */ | |
699 | qemu_sem_post(&mis->postcopy_pause_sem_dst); | |
884835fa PX |
700 | return true; |
701 | } | |
702 | ||
703 | return false; | |
704 | } | |
705 | ||
b673eab4 | 706 | void migration_fd_process_incoming(QEMUFile *f, Error **errp) |
884835fa | 707 | { |
a39e9339 | 708 | if (!migration_incoming_setup(f, errp)) { |
884835fa | 709 | return; |
d96c9e8d | 710 | } |
a39e9339 | 711 | if (postcopy_try_recover()) { |
b673eab4 JQ |
712 | return; |
713 | } | |
884835fa | 714 | migration_incoming_process(); |
e595a01a JQ |
715 | } |
716 | ||
49ed0d24 | 717 | void migration_ioc_process_incoming(QIOChannel *ioc, Error **errp) |
4f0fae7f JQ |
718 | { |
719 | MigrationIncomingState *mis = migration_incoming_get_current(); | |
b673eab4 | 720 | Error *local_err = NULL; |
a429e7f4 | 721 | bool start_migration; |
4f0fae7f JQ |
722 | |
723 | if (!mis->from_src_file) { | |
a429e7f4 | 724 | /* The first connection (multifd may have multiple) */ |
77ef2dc1 | 725 | QEMUFile *f = qemu_file_new_input(ioc); |
a429e7f4 | 726 | |
7d6f6933 | 727 | if (!migration_incoming_setup(f, errp)) { |
b673eab4 JQ |
728 | return; |
729 | } | |
a429e7f4 PX |
730 | |
731 | /* | |
732 | * Common migration only needs one channel, so we can start | |
733 | * right now. Multifd needs more than one channel, we wait. | |
734 | */ | |
735 | start_migration = !migrate_use_multifd(); | |
736 | } else { | |
737 | /* Multiple connections */ | |
738 | assert(migrate_use_multifd()); | |
49ed0d24 FL |
739 | start_migration = multifd_recv_new_channel(ioc, &local_err); |
740 | if (local_err) { | |
741 | error_propagate(errp, local_err); | |
742 | return; | |
743 | } | |
4f0fae7f | 744 | } |
81e62053 | 745 | |
a429e7f4 | 746 | if (start_migration) { |
a39e9339 PX |
747 | /* If it's a recovery, we're done */ |
748 | if (postcopy_try_recover()) { | |
749 | return; | |
750 | } | |
81e62053 PX |
751 | migration_incoming_process(); |
752 | } | |
4f0fae7f JQ |
753 | } |
754 | ||
428d8908 JQ |
755 | /** |
756 | * @migration_has_all_channels: We have received all channels that we need | |
757 | * | |
758 | * Returns true when we have got connections to all the channels that | |
759 | * we need for migration. | |
760 | */ | |
761 | bool migration_has_all_channels(void) | |
762 | { | |
ca273df3 | 763 | MigrationIncomingState *mis = migration_incoming_get_current(); |
62c1e0ca JQ |
764 | bool all_channels; |
765 | ||
766 | all_channels = multifd_recv_all_channels_created(); | |
767 | ||
ca273df3 | 768 | return all_channels && mis->from_src_file != NULL; |
428d8908 JQ |
769 | } |
770 | ||
6decec93 DDAG |
771 | /* |
772 | * Send a 'SHUT' message on the return channel with the given value | |
773 | * to indicate that we've finished with the RP. Non-0 value indicates | |
774 | * error. | |
775 | */ | |
776 | void migrate_send_rp_shut(MigrationIncomingState *mis, | |
777 | uint32_t value) | |
778 | { | |
779 | uint32_t buf; | |
780 | ||
781 | buf = cpu_to_be32(value); | |
782 | migrate_send_rp_message(mis, MIG_RP_MSG_SHUT, sizeof(buf), &buf); | |
783 | } | |
784 | ||
785 | /* | |
786 | * Send a 'PONG' message on the return channel with the given value | |
787 | * (normally in response to a 'PING') | |
788 | */ | |
789 | void migrate_send_rp_pong(MigrationIncomingState *mis, | |
790 | uint32_t value) | |
791 | { | |
792 | uint32_t buf; | |
793 | ||
794 | buf = cpu_to_be32(value); | |
795 | migrate_send_rp_message(mis, MIG_RP_MSG_PONG, sizeof(buf), &buf); | |
796 | } | |
797 | ||
a335debb PX |
798 | void migrate_send_rp_recv_bitmap(MigrationIncomingState *mis, |
799 | char *block_name) | |
800 | { | |
801 | char buf[512]; | |
802 | int len; | |
803 | int64_t res; | |
804 | ||
805 | /* | |
806 | * First, we send the header part. It contains only the len of | |
807 | * idstr, and the idstr itself. | |
808 | */ | |
809 | len = strlen(block_name); | |
810 | buf[0] = len; | |
811 | memcpy(buf + 1, block_name, len); | |
812 | ||
813 | if (mis->state != MIGRATION_STATUS_POSTCOPY_RECOVER) { | |
814 | error_report("%s: MSG_RP_RECV_BITMAP only used for recovery", | |
815 | __func__); | |
816 | return; | |
817 | } | |
818 | ||
819 | migrate_send_rp_message(mis, MIG_RP_MSG_RECV_BITMAP, len + 1, buf); | |
820 | ||
821 | /* | |
822 | * Next, we dump the received bitmap to the stream. | |
823 | * | |
824 | * TODO: currently we are safe since we are the only one that is | |
825 | * using the to_src_file handle (fault thread is still paused), | |
826 | * and it's ok even not taking the mutex. However the best way is | |
827 | * to take the lock before sending the message header, and release | |
828 | * the lock after sending the bitmap. | |
829 | */ | |
830 | qemu_mutex_lock(&mis->rp_mutex); | |
831 | res = ramblock_recv_bitmap_send(mis->to_src_file, block_name); | |
832 | qemu_mutex_unlock(&mis->rp_mutex); | |
833 | ||
834 | trace_migrate_send_rp_recv_bitmap(block_name, res); | |
835 | } | |
836 | ||
13955b89 PX |
837 | void migrate_send_rp_resume_ack(MigrationIncomingState *mis, uint32_t value) |
838 | { | |
839 | uint32_t buf; | |
840 | ||
841 | buf = cpu_to_be32(value); | |
842 | migrate_send_rp_message(mis, MIG_RP_MSG_RESUME_ACK, sizeof(buf), &buf); | |
843 | } | |
844 | ||
bbf6da32 OW |
845 | MigrationCapabilityStatusList *qmp_query_migrate_capabilities(Error **errp) |
846 | { | |
95b3a8c8 EB |
847 | MigrationCapabilityStatusList *head = NULL, **tail = &head; |
848 | MigrationCapabilityStatus *caps; | |
bbf6da32 OW |
849 | MigrationState *s = migrate_get_current(); |
850 | int i; | |
851 | ||
7fb1cf16 | 852 | for (i = 0; i < MIGRATION_CAPABILITY__MAX; i++) { |
ed1701c6 DDAG |
853 | #ifndef CONFIG_LIVE_BLOCK_MIGRATION |
854 | if (i == MIGRATION_CAPABILITY_BLOCK) { | |
855 | continue; | |
856 | } | |
857 | #endif | |
95b3a8c8 EB |
858 | caps = g_malloc0(sizeof(*caps)); |
859 | caps->capability = i; | |
860 | caps->state = s->enabled_capabilities[i]; | |
861 | QAPI_LIST_APPEND(tail, caps); | |
bbf6da32 OW |
862 | } |
863 | ||
864 | return head; | |
865 | } | |
866 | ||
85de8323 LL |
867 | MigrationParameters *qmp_query_migrate_parameters(Error **errp) |
868 | { | |
869 | MigrationParameters *params; | |
870 | MigrationState *s = migrate_get_current(); | |
871 | ||
e87fae4c | 872 | /* TODO use QAPI_CLONE() instead of duplicating it inline */ |
85de8323 | 873 | params = g_malloc0(sizeof(*params)); |
de63ab61 | 874 | params->has_compress_level = true; |
2594f56d | 875 | params->compress_level = s->parameters.compress_level; |
de63ab61 | 876 | params->has_compress_threads = true; |
2594f56d | 877 | params->compress_threads = s->parameters.compress_threads; |
1d58872a XG |
878 | params->has_compress_wait_thread = true; |
879 | params->compress_wait_thread = s->parameters.compress_wait_thread; | |
de63ab61 | 880 | params->has_decompress_threads = true; |
2594f56d | 881 | params->decompress_threads = s->parameters.decompress_threads; |
dc14a470 KZ |
882 | params->has_throttle_trigger_threshold = true; |
883 | params->throttle_trigger_threshold = s->parameters.throttle_trigger_threshold; | |
de63ab61 | 884 | params->has_cpu_throttle_initial = true; |
2594f56d | 885 | params->cpu_throttle_initial = s->parameters.cpu_throttle_initial; |
de63ab61 | 886 | params->has_cpu_throttle_increment = true; |
2594f56d | 887 | params->cpu_throttle_increment = s->parameters.cpu_throttle_increment; |
cbbf8182 KZ |
888 | params->has_cpu_throttle_tailslow = true; |
889 | params->cpu_throttle_tailslow = s->parameters.cpu_throttle_tailslow; | |
8cc99dcd | 890 | params->has_tls_creds = true; |
69ef1f36 | 891 | params->tls_creds = g_strdup(s->parameters.tls_creds); |
8cc99dcd | 892 | params->has_tls_hostname = true; |
69ef1f36 | 893 | params->tls_hostname = g_strdup(s->parameters.tls_hostname); |
d2f1d29b | 894 | params->has_tls_authz = true; |
7cd75cbd MZ |
895 | params->tls_authz = g_strdup(s->parameters.tls_authz ? |
896 | s->parameters.tls_authz : ""); | |
2ff30257 AA |
897 | params->has_max_bandwidth = true; |
898 | params->max_bandwidth = s->parameters.max_bandwidth; | |
899 | params->has_downtime_limit = true; | |
900 | params->downtime_limit = s->parameters.downtime_limit; | |
fe39a4d4 | 901 | params->has_x_checkpoint_delay = true; |
68b53591 | 902 | params->x_checkpoint_delay = s->parameters.x_checkpoint_delay; |
2833c59b JQ |
903 | params->has_block_incremental = true; |
904 | params->block_incremental = s->parameters.block_incremental; | |
cbfd6c95 JQ |
905 | params->has_multifd_channels = true; |
906 | params->multifd_channels = s->parameters.multifd_channels; | |
96eef042 JQ |
907 | params->has_multifd_compression = true; |
908 | params->multifd_compression = s->parameters.multifd_compression; | |
9004db48 JQ |
909 | params->has_multifd_zlib_level = true; |
910 | params->multifd_zlib_level = s->parameters.multifd_zlib_level; | |
6a9ad154 JQ |
911 | params->has_multifd_zstd_level = true; |
912 | params->multifd_zstd_level = s->parameters.multifd_zstd_level; | |
73af8dd8 JQ |
913 | params->has_xbzrle_cache_size = true; |
914 | params->xbzrle_cache_size = s->parameters.xbzrle_cache_size; | |
7e555c6c DDAG |
915 | params->has_max_postcopy_bandwidth = true; |
916 | params->max_postcopy_bandwidth = s->parameters.max_postcopy_bandwidth; | |
4cbc9c7f LQ |
917 | params->has_max_cpu_throttle = true; |
918 | params->max_cpu_throttle = s->parameters.max_cpu_throttle; | |
ee3d96ba DDAG |
919 | params->has_announce_initial = true; |
920 | params->announce_initial = s->parameters.announce_initial; | |
921 | params->has_announce_max = true; | |
922 | params->announce_max = s->parameters.announce_max; | |
923 | params->has_announce_rounds = true; | |
924 | params->announce_rounds = s->parameters.announce_rounds; | |
925 | params->has_announce_step = true; | |
926 | params->announce_step = s->parameters.announce_step; | |
85de8323 | 927 | |
31e4c354 HR |
928 | if (s->parameters.has_block_bitmap_mapping) { |
929 | params->has_block_bitmap_mapping = true; | |
930 | params->block_bitmap_mapping = | |
931 | QAPI_CLONE(BitmapMigrationNodeAliasList, | |
932 | s->parameters.block_bitmap_mapping); | |
933 | } | |
934 | ||
85de8323 LL |
935 | return params; |
936 | } | |
937 | ||
ee3d96ba DDAG |
938 | AnnounceParameters *migrate_announce_params(void) |
939 | { | |
940 | static AnnounceParameters ap; | |
941 | ||
942 | MigrationState *s = migrate_get_current(); | |
943 | ||
944 | ap.initial = s->parameters.announce_initial; | |
945 | ap.max = s->parameters.announce_max; | |
946 | ap.rounds = s->parameters.announce_rounds; | |
947 | ap.step = s->parameters.announce_step; | |
948 | ||
949 | return ≈ | |
950 | } | |
951 | ||
f6844b99 DDAG |
952 | /* |
953 | * Return true if we're already in the middle of a migration | |
954 | * (i.e. any of the active or setup states) | |
955 | */ | |
3d63da16 | 956 | bool migration_is_setup_or_active(int state) |
f6844b99 DDAG |
957 | { |
958 | switch (state) { | |
959 | case MIGRATION_STATUS_ACTIVE: | |
9ec055ae | 960 | case MIGRATION_STATUS_POSTCOPY_ACTIVE: |
a688d2c1 | 961 | case MIGRATION_STATUS_POSTCOPY_PAUSED: |
135b87b4 | 962 | case MIGRATION_STATUS_POSTCOPY_RECOVER: |
f6844b99 | 963 | case MIGRATION_STATUS_SETUP: |
31e06077 DDAG |
964 | case MIGRATION_STATUS_PRE_SWITCHOVER: |
965 | case MIGRATION_STATUS_DEVICE: | |
c7e0acd5 | 966 | case MIGRATION_STATUS_WAIT_UNPLUG: |
19dd408a | 967 | case MIGRATION_STATUS_COLO: |
f6844b99 DDAG |
968 | return true; |
969 | ||
970 | default: | |
971 | return false; | |
972 | ||
973 | } | |
974 | } | |
975 | ||
392d87e2 JQ |
976 | bool migration_is_running(int state) |
977 | { | |
978 | switch (state) { | |
979 | case MIGRATION_STATUS_ACTIVE: | |
980 | case MIGRATION_STATUS_POSTCOPY_ACTIVE: | |
981 | case MIGRATION_STATUS_POSTCOPY_PAUSED: | |
982 | case MIGRATION_STATUS_POSTCOPY_RECOVER: | |
983 | case MIGRATION_STATUS_SETUP: | |
984 | case MIGRATION_STATUS_PRE_SWITCHOVER: | |
985 | case MIGRATION_STATUS_DEVICE: | |
986 | case MIGRATION_STATUS_WAIT_UNPLUG: | |
987 | case MIGRATION_STATUS_CANCELLING: | |
392d87e2 JQ |
988 | return true; |
989 | ||
990 | default: | |
991 | return false; | |
992 | ||
993 | } | |
994 | } | |
995 | ||
640dfb14 WY |
996 | static void populate_time_info(MigrationInfo *info, MigrationState *s) |
997 | { | |
998 | info->has_status = true; | |
999 | info->has_setup_time = true; | |
1000 | info->setup_time = s->setup_time; | |
1001 | if (s->state == MIGRATION_STATUS_COMPLETED) { | |
1002 | info->has_total_time = true; | |
1003 | info->total_time = s->total_time; | |
1004 | info->has_downtime = true; | |
1005 | info->downtime = s->downtime; | |
1006 | } else { | |
1007 | info->has_total_time = true; | |
1008 | info->total_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME) - | |
1009 | s->start_time; | |
1010 | info->has_expected_downtime = true; | |
1011 | info->expected_downtime = s->expected_downtime; | |
1012 | } | |
1013 | } | |
1014 | ||
a22463a5 DDAG |
1015 | static void populate_ram_info(MigrationInfo *info, MigrationState *s) |
1016 | { | |
144fa06b JQ |
1017 | size_t page_size = qemu_target_page_size(); |
1018 | ||
a22463a5 DDAG |
1019 | info->has_ram = true; |
1020 | info->ram = g_malloc0(sizeof(*info->ram)); | |
9360447d | 1021 | info->ram->transferred = ram_counters.transferred; |
a22463a5 | 1022 | info->ram->total = ram_bytes_total(); |
9360447d | 1023 | info->ram->duplicate = ram_counters.duplicate; |
bedf53c1 JQ |
1024 | /* legacy value. It is not used anymore */ |
1025 | info->ram->skipped = 0; | |
9360447d | 1026 | info->ram->normal = ram_counters.normal; |
144fa06b | 1027 | info->ram->normal_bytes = ram_counters.normal * page_size; |
a22463a5 | 1028 | info->ram->mbps = s->mbps; |
9360447d JQ |
1029 | info->ram->dirty_sync_count = ram_counters.dirty_sync_count; |
1030 | info->ram->postcopy_requests = ram_counters.postcopy_requests; | |
144fa06b | 1031 | info->ram->page_size = page_size; |
a61c45bd | 1032 | info->ram->multifd_bytes = ram_counters.multifd_bytes; |
aecbfe9c | 1033 | info->ram->pages_per_second = s->pages_per_second; |
ae680668 DE |
1034 | info->ram->precopy_bytes = ram_counters.precopy_bytes; |
1035 | info->ram->downtime_bytes = ram_counters.downtime_bytes; | |
1036 | info->ram->postcopy_bytes = ram_counters.postcopy_bytes; | |
a22463a5 | 1037 | |
114f5aee JQ |
1038 | if (migrate_use_xbzrle()) { |
1039 | info->has_xbzrle_cache = true; | |
1040 | info->xbzrle_cache = g_malloc0(sizeof(*info->xbzrle_cache)); | |
1041 | info->xbzrle_cache->cache_size = migrate_xbzrle_cache_size(); | |
9360447d JQ |
1042 | info->xbzrle_cache->bytes = xbzrle_counters.bytes; |
1043 | info->xbzrle_cache->pages = xbzrle_counters.pages; | |
1044 | info->xbzrle_cache->cache_miss = xbzrle_counters.cache_miss; | |
1045 | info->xbzrle_cache->cache_miss_rate = xbzrle_counters.cache_miss_rate; | |
e460a4b1 | 1046 | info->xbzrle_cache->encoding_rate = xbzrle_counters.encoding_rate; |
9360447d | 1047 | info->xbzrle_cache->overflow = xbzrle_counters.overflow; |
114f5aee JQ |
1048 | } |
1049 | ||
76e03000 XG |
1050 | if (migrate_use_compression()) { |
1051 | info->has_compression = true; | |
1052 | info->compression = g_malloc0(sizeof(*info->compression)); | |
1053 | info->compression->pages = compression_counters.pages; | |
1054 | info->compression->busy = compression_counters.busy; | |
1055 | info->compression->busy_rate = compression_counters.busy_rate; | |
1056 | info->compression->compressed_size = | |
1057 | compression_counters.compressed_size; | |
1058 | info->compression->compression_rate = | |
1059 | compression_counters.compression_rate; | |
1060 | } | |
1061 | ||
338182c8 JQ |
1062 | if (cpu_throttle_active()) { |
1063 | info->has_cpu_throttle_percentage = true; | |
1064 | info->cpu_throttle_percentage = cpu_throttle_get_percentage(); | |
1065 | } | |
1066 | ||
a22463a5 DDAG |
1067 | if (s->state != MIGRATION_STATUS_COMPLETED) { |
1068 | info->ram->remaining = ram_bytes_remaining(); | |
9360447d | 1069 | info->ram->dirty_pages_rate = ram_counters.dirty_pages_rate; |
a22463a5 DDAG |
1070 | } |
1071 | } | |
1072 | ||
930ac04c JQ |
1073 | static void populate_disk_info(MigrationInfo *info) |
1074 | { | |
1075 | if (blk_mig_active()) { | |
1076 | info->has_disk = true; | |
1077 | info->disk = g_malloc0(sizeof(*info->disk)); | |
1078 | info->disk->transferred = blk_mig_bytes_transferred(); | |
1079 | info->disk->remaining = blk_mig_bytes_remaining(); | |
1080 | info->disk->total = blk_mig_bytes_total(); | |
1081 | } | |
1082 | } | |
1083 | ||
65ace060 | 1084 | static void fill_source_migration_info(MigrationInfo *info) |
5bb7910a | 1085 | { |
17549e84 | 1086 | MigrationState *s = migrate_get_current(); |
552de79b | 1087 | int state = qatomic_read(&s->state); |
372043f3 | 1088 | GSList *cur_blocker = migration_blockers; |
17549e84 | 1089 | |
3af8554b | 1090 | info->blocked_reasons = NULL; |
3af8554b | 1091 | |
372043f3 MA |
1092 | /* |
1093 | * There are two types of reasons a migration might be blocked; | |
1094 | * a) devices marked in VMState as non-migratable, and | |
1095 | * b) Explicit migration blockers | |
1096 | * We need to add both of them here. | |
1097 | */ | |
1098 | qemu_savevm_non_migratable_list(&info->blocked_reasons); | |
3af8554b | 1099 | |
372043f3 MA |
1100 | while (cur_blocker) { |
1101 | QAPI_LIST_PREPEND(info->blocked_reasons, | |
1102 | g_strdup(error_get_pretty(cur_blocker->data))); | |
1103 | cur_blocker = g_slist_next(cur_blocker); | |
3af8554b | 1104 | } |
372043f3 | 1105 | info->has_blocked_reasons = info->blocked_reasons != NULL; |
3af8554b | 1106 | |
552de79b | 1107 | switch (state) { |
31194731 | 1108 | case MIGRATION_STATUS_NONE: |
17549e84 | 1109 | /* no migration has happened ever */ |
65ace060 AP |
1110 | /* do not overwrite destination migration status */ |
1111 | return; | |
31194731 | 1112 | case MIGRATION_STATUS_SETUP: |
29ae8a41 | 1113 | info->has_status = true; |
ed4fbd10 | 1114 | info->has_total_time = false; |
29ae8a41 | 1115 | break; |
31194731 HZ |
1116 | case MIGRATION_STATUS_ACTIVE: |
1117 | case MIGRATION_STATUS_CANCELLING: | |
9ec055ae | 1118 | case MIGRATION_STATUS_POSTCOPY_ACTIVE: |
31e06077 DDAG |
1119 | case MIGRATION_STATUS_PRE_SWITCHOVER: |
1120 | case MIGRATION_STATUS_DEVICE: | |
a688d2c1 | 1121 | case MIGRATION_STATUS_POSTCOPY_PAUSED: |
135b87b4 | 1122 | case MIGRATION_STATUS_POSTCOPY_RECOVER: |
640dfb14 WY |
1123 | /* TODO add some postcopy stats */ |
1124 | populate_time_info(info, s); | |
a22463a5 | 1125 | populate_ram_info(info, s); |
930ac04c | 1126 | populate_disk_info(info); |
3710586c | 1127 | populate_vfio_info(info); |
17549e84 | 1128 | break; |
0b827d5e HZ |
1129 | case MIGRATION_STATUS_COLO: |
1130 | info->has_status = true; | |
1131 | /* TODO: display COLO specific information (checkpoint info etc.) */ | |
1132 | break; | |
31194731 | 1133 | case MIGRATION_STATUS_COMPLETED: |
640dfb14 | 1134 | populate_time_info(info, s); |
a22463a5 | 1135 | populate_ram_info(info, s); |
3710586c | 1136 | populate_vfio_info(info); |
17549e84 | 1137 | break; |
31194731 | 1138 | case MIGRATION_STATUS_FAILED: |
791e7c82 | 1139 | info->has_status = true; |
d59ce6f3 DB |
1140 | if (s->error) { |
1141 | info->has_error_desc = true; | |
1142 | info->error_desc = g_strdup(error_get_pretty(s->error)); | |
1143 | } | |
17549e84 | 1144 | break; |
31194731 | 1145 | case MIGRATION_STATUS_CANCELLED: |
791e7c82 | 1146 | info->has_status = true; |
17549e84 | 1147 | break; |
c7e0acd5 JF |
1148 | case MIGRATION_STATUS_WAIT_UNPLUG: |
1149 | info->has_status = true; | |
1150 | break; | |
5bb7910a | 1151 | } |
552de79b | 1152 | info->status = state; |
5bb7910a AL |
1153 | } |
1154 | ||
6e8c25b4 AG |
1155 | typedef enum WriteTrackingSupport { |
1156 | WT_SUPPORT_UNKNOWN = 0, | |
1157 | WT_SUPPORT_ABSENT, | |
1158 | WT_SUPPORT_AVAILABLE, | |
1159 | WT_SUPPORT_COMPATIBLE | |
1160 | } WriteTrackingSupport; | |
1161 | ||
1162 | static | |
1163 | WriteTrackingSupport migrate_query_write_tracking(void) | |
1164 | { | |
1165 | /* Check if kernel supports required UFFD features */ | |
1166 | if (!ram_write_tracking_available()) { | |
1167 | return WT_SUPPORT_ABSENT; | |
1168 | } | |
1169 | /* | |
1170 | * Check if current memory configuration is | |
1171 | * compatible with required UFFD features. | |
1172 | */ | |
1173 | if (!ram_write_tracking_compatible()) { | |
1174 | return WT_SUPPORT_AVAILABLE; | |
1175 | } | |
1176 | ||
1177 | return WT_SUPPORT_COMPATIBLE; | |
1178 | } | |
1179 | ||
4a84214e PX |
1180 | /** |
1181 | * @migration_caps_check - check capability validity | |
1182 | * | |
1183 | * @cap_list: old capability list, array of bool | |
1184 | * @params: new capabilities to be applied soon | |
1185 | * @errp: set *errp if the check failed, with reason | |
1186 | * | |
1187 | * Returns true if check passed, otherwise false. | |
1188 | */ | |
1189 | static bool migrate_caps_check(bool *cap_list, | |
1190 | MigrationCapabilityStatusList *params, | |
1191 | Error **errp) | |
00458433 | 1192 | { |
00458433 | 1193 | MigrationCapabilityStatusList *cap; |
4a84214e | 1194 | bool old_postcopy_cap; |
d7651f15 | 1195 | MigrationIncomingState *mis = migration_incoming_get_current(); |
00458433 | 1196 | |
4a84214e | 1197 | old_postcopy_cap = cap_list[MIGRATION_CAPABILITY_POSTCOPY_RAM]; |
00458433 OW |
1198 | |
1199 | for (cap = params; cap; cap = cap->next) { | |
4a84214e PX |
1200 | cap_list[cap->value->capability] = cap->value->state; |
1201 | } | |
1202 | ||
ed1701c6 | 1203 | #ifndef CONFIG_LIVE_BLOCK_MIGRATION |
4a84214e PX |
1204 | if (cap_list[MIGRATION_CAPABILITY_BLOCK]) { |
1205 | error_setg(errp, "QEMU compiled without old-style (blk/-b, inc/-i) " | |
1206 | "block migration"); | |
1207 | error_append_hint(errp, "Use drive_mirror+NBD instead.\n"); | |
1208 | return false; | |
00458433 | 1209 | } |
4a84214e | 1210 | #endif |
53dd370c | 1211 | |
7e934f5b ZC |
1212 | #ifndef CONFIG_REPLICATION |
1213 | if (cap_list[MIGRATION_CAPABILITY_X_COLO]) { | |
1214 | error_setg(errp, "QEMU compiled without replication module" | |
1215 | " can't enable COLO"); | |
1216 | error_append_hint(errp, "Please enable replication before COLO.\n"); | |
1217 | return false; | |
1218 | } | |
1219 | #endif | |
1220 | ||
4a84214e | 1221 | if (cap_list[MIGRATION_CAPABILITY_POSTCOPY_RAM]) { |
096631bd DDAG |
1222 | /* This check is reasonably expensive, so only when it's being |
1223 | * set the first time, also it's only the destination that needs | |
1224 | * special support. | |
1225 | */ | |
1226 | if (!old_postcopy_cap && runstate_check(RUN_STATE_INMIGRATE) && | |
d7651f15 | 1227 | !postcopy_ram_supported_by_host(mis)) { |
096631bd DDAG |
1228 | /* postcopy_ram_supported_by_host will have emitted a more |
1229 | * detailed message | |
1230 | */ | |
4a84214e PX |
1231 | error_setg(errp, "Postcopy is not supported"); |
1232 | return false; | |
096631bd | 1233 | } |
18269069 YK |
1234 | |
1235 | if (cap_list[MIGRATION_CAPABILITY_X_IGNORE_SHARED]) { | |
1236 | error_setg(errp, "Postcopy is not compatible with ignore-shared"); | |
1237 | return false; | |
1238 | } | |
53dd370c | 1239 | } |
4a84214e | 1240 | |
6e8c25b4 AG |
1241 | if (cap_list[MIGRATION_CAPABILITY_BACKGROUND_SNAPSHOT]) { |
1242 | WriteTrackingSupport wt_support; | |
1243 | int idx; | |
1244 | /* | |
1245 | * Check if 'background-snapshot' capability is supported by | |
1246 | * host kernel and compatible with guest memory configuration. | |
1247 | */ | |
1248 | wt_support = migrate_query_write_tracking(); | |
1249 | if (wt_support < WT_SUPPORT_AVAILABLE) { | |
1250 | error_setg(errp, "Background-snapshot is not supported by host kernel"); | |
1251 | return false; | |
1252 | } | |
1253 | if (wt_support < WT_SUPPORT_COMPATIBLE) { | |
1254 | error_setg(errp, "Background-snapshot is not compatible " | |
1255 | "with guest memory configuration"); | |
1256 | return false; | |
1257 | } | |
1258 | ||
1259 | /* | |
1260 | * Check if there are any migration capabilities | |
1261 | * incompatible with 'background-snapshot'. | |
1262 | */ | |
1263 | for (idx = 0; idx < check_caps_background_snapshot.size; idx++) { | |
1264 | int incomp_cap = check_caps_background_snapshot.caps[idx]; | |
1265 | if (cap_list[incomp_cap]) { | |
1266 | error_setg(errp, | |
1267 | "Background-snapshot is not compatible with %s", | |
1268 | MigrationCapability_str(incomp_cap)); | |
1269 | return false; | |
1270 | } | |
1271 | } | |
1272 | } | |
1273 | ||
1abaec9a LB |
1274 | #ifdef CONFIG_LINUX |
1275 | if (cap_list[MIGRATION_CAPABILITY_ZERO_COPY_SEND] && | |
1276 | (!cap_list[MIGRATION_CAPABILITY_MULTIFD] || | |
1277 | migrate_use_compression() || | |
1278 | migrate_use_tls())) { | |
1279 | error_setg(errp, | |
1280 | "Zero copy only available for non-compressed non-TLS multifd migration"); | |
1281 | return false; | |
1282 | } | |
1283 | #else | |
1284 | if (cap_list[MIGRATION_CAPABILITY_ZERO_COPY_SEND]) { | |
1285 | error_setg(errp, | |
1286 | "Zero copy currently only available on Linux"); | |
1287 | return false; | |
1288 | } | |
1289 | #endif | |
1290 | ||
1291 | ||
5ad15e86 LZ |
1292 | /* incoming side only */ |
1293 | if (runstate_check(RUN_STATE_INMIGRATE) && | |
f444eeda | 1294 | !migrate_multi_channels_is_allowed() && |
5ad15e86 LZ |
1295 | cap_list[MIGRATION_CAPABILITY_MULTIFD]) { |
1296 | error_setg(errp, "multifd is not supported by current protocol"); | |
1297 | return false; | |
1298 | } | |
1299 | ||
4a84214e PX |
1300 | return true; |
1301 | } | |
1302 | ||
65ace060 AP |
1303 | static void fill_destination_migration_info(MigrationInfo *info) |
1304 | { | |
1305 | MigrationIncomingState *mis = migration_incoming_get_current(); | |
1306 | ||
9aca82ba JQ |
1307 | if (mis->socket_address_list) { |
1308 | info->has_socket_address = true; | |
1309 | info->socket_address = | |
1310 | QAPI_CLONE(SocketAddressList, mis->socket_address_list); | |
1311 | } | |
1312 | ||
65ace060 AP |
1313 | switch (mis->state) { |
1314 | case MIGRATION_STATUS_NONE: | |
1315 | return; | |
65ace060 AP |
1316 | case MIGRATION_STATUS_SETUP: |
1317 | case MIGRATION_STATUS_CANCELLING: | |
1318 | case MIGRATION_STATUS_CANCELLED: | |
1319 | case MIGRATION_STATUS_ACTIVE: | |
1320 | case MIGRATION_STATUS_POSTCOPY_ACTIVE: | |
3c9928d9 PX |
1321 | case MIGRATION_STATUS_POSTCOPY_PAUSED: |
1322 | case MIGRATION_STATUS_POSTCOPY_RECOVER: | |
65ace060 AP |
1323 | case MIGRATION_STATUS_FAILED: |
1324 | case MIGRATION_STATUS_COLO: | |
1325 | info->has_status = true; | |
1326 | break; | |
1327 | case MIGRATION_STATUS_COMPLETED: | |
1328 | info->has_status = true; | |
1329 | fill_destination_postcopy_migration_info(info); | |
1330 | break; | |
1331 | } | |
1332 | info->status = mis->state; | |
1333 | } | |
1334 | ||
1335 | MigrationInfo *qmp_query_migrate(Error **errp) | |
1336 | { | |
1337 | MigrationInfo *info = g_malloc0(sizeof(*info)); | |
1338 | ||
1339 | fill_destination_migration_info(info); | |
1340 | fill_source_migration_info(info); | |
1341 | ||
1342 | return info; | |
1343 | } | |
1344 | ||
4a84214e PX |
1345 | void qmp_migrate_set_capabilities(MigrationCapabilityStatusList *params, |
1346 | Error **errp) | |
1347 | { | |
1348 | MigrationState *s = migrate_get_current(); | |
1349 | MigrationCapabilityStatusList *cap; | |
dd0ee30c | 1350 | bool cap_list[MIGRATION_CAPABILITY__MAX]; |
4a84214e | 1351 | |
392d87e2 | 1352 | if (migration_is_running(s->state)) { |
4a84214e PX |
1353 | error_setg(errp, QERR_MIGRATION_ACTIVE); |
1354 | return; | |
1355 | } | |
1356 | ||
dd0ee30c PX |
1357 | memcpy(cap_list, s->enabled_capabilities, sizeof(cap_list)); |
1358 | if (!migrate_caps_check(cap_list, params, errp)) { | |
4a84214e PX |
1359 | return; |
1360 | } | |
1361 | ||
1362 | for (cap = params; cap; cap = cap->next) { | |
1363 | s->enabled_capabilities[cap->value->capability] = cap->value->state; | |
1364 | } | |
00458433 OW |
1365 | } |
1366 | ||
16d063bc PX |
1367 | /* |
1368 | * Check whether the parameters are valid. Error will be put into errp | |
1369 | * (if provided). Return true if valid, otherwise false. | |
1370 | */ | |
1371 | static bool migrate_params_check(MigrationParameters *params, Error **errp) | |
85de8323 | 1372 | { |
7f375e04 | 1373 | if (params->has_compress_level && |
741d4086 | 1374 | (params->compress_level > 9)) { |
c6bd8c70 | 1375 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "compress_level", |
54270c45 | 1376 | "a value between 0 and 9"); |
16d063bc | 1377 | return false; |
85de8323 | 1378 | } |
16d063bc | 1379 | |
741d4086 | 1380 | if (params->has_compress_threads && (params->compress_threads < 1)) { |
c6bd8c70 MA |
1381 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, |
1382 | "compress_threads", | |
54270c45 | 1383 | "a value between 1 and 255"); |
16d063bc | 1384 | return false; |
85de8323 | 1385 | } |
16d063bc | 1386 | |
741d4086 | 1387 | if (params->has_decompress_threads && (params->decompress_threads < 1)) { |
c6bd8c70 MA |
1388 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, |
1389 | "decompress_threads", | |
54270c45 | 1390 | "a value between 1 and 255"); |
16d063bc | 1391 | return false; |
85de8323 | 1392 | } |
16d063bc | 1393 | |
dc14a470 KZ |
1394 | if (params->has_throttle_trigger_threshold && |
1395 | (params->throttle_trigger_threshold < 1 || | |
1396 | params->throttle_trigger_threshold > 100)) { | |
1397 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, | |
1398 | "throttle_trigger_threshold", | |
1399 | "an integer in the range of 1 to 100"); | |
1400 | return false; | |
1401 | } | |
1402 | ||
7f375e04 EB |
1403 | if (params->has_cpu_throttle_initial && |
1404 | (params->cpu_throttle_initial < 1 || | |
1405 | params->cpu_throttle_initial > 99)) { | |
1626fee3 | 1406 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, |
d85a31d1 | 1407 | "cpu_throttle_initial", |
1626fee3 | 1408 | "an integer in the range of 1 to 99"); |
16d063bc | 1409 | return false; |
1626fee3 | 1410 | } |
16d063bc | 1411 | |
7f375e04 EB |
1412 | if (params->has_cpu_throttle_increment && |
1413 | (params->cpu_throttle_increment < 1 || | |
1414 | params->cpu_throttle_increment > 99)) { | |
1626fee3 | 1415 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, |
d85a31d1 | 1416 | "cpu_throttle_increment", |
1626fee3 | 1417 | "an integer in the range of 1 to 99"); |
16d063bc | 1418 | return false; |
1626fee3 | 1419 | } |
16d063bc | 1420 | |
741d4086 | 1421 | if (params->has_max_bandwidth && (params->max_bandwidth > SIZE_MAX)) { |
2ee30cf0 MZ |
1422 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, |
1423 | "max_bandwidth", | |
1424 | "an integer in the range of 0 to "stringify(SIZE_MAX) | |
1425 | " bytes/second"); | |
16d063bc | 1426 | return false; |
2ff30257 | 1427 | } |
16d063bc | 1428 | |
2ff30257 | 1429 | if (params->has_downtime_limit && |
741d4086 | 1430 | (params->downtime_limit > MAX_MIGRATE_DOWNTIME)) { |
2ee30cf0 MZ |
1431 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, |
1432 | "downtime_limit", | |
1433 | "an integer in the range of 0 to " | |
7ac5529a | 1434 | stringify(MAX_MIGRATE_DOWNTIME)" ms"); |
16d063bc | 1435 | return false; |
2ff30257 | 1436 | } |
16d063bc | 1437 | |
741d4086 JQ |
1438 | /* x_checkpoint_delay is now always positive */ |
1439 | ||
cbfd6c95 | 1440 | if (params->has_multifd_channels && (params->multifd_channels < 1)) { |
4075fb1c JQ |
1441 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, |
1442 | "multifd_channels", | |
54270c45 | 1443 | "a value between 1 and 255"); |
4075fb1c JQ |
1444 | return false; |
1445 | } | |
16d063bc | 1446 | |
9004db48 JQ |
1447 | if (params->has_multifd_zlib_level && |
1448 | (params->multifd_zlib_level > 9)) { | |
1449 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "multifd_zlib_level", | |
54270c45 | 1450 | "a value between 0 and 9"); |
9004db48 JQ |
1451 | return false; |
1452 | } | |
1453 | ||
6a9ad154 JQ |
1454 | if (params->has_multifd_zstd_level && |
1455 | (params->multifd_zstd_level > 20)) { | |
1456 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "multifd_zstd_level", | |
54270c45 | 1457 | "a value between 0 and 20"); |
6a9ad154 JQ |
1458 | return false; |
1459 | } | |
1460 | ||
73af8dd8 JQ |
1461 | if (params->has_xbzrle_cache_size && |
1462 | (params->xbzrle_cache_size < qemu_target_page_size() || | |
1463 | !is_power_of_2(params->xbzrle_cache_size))) { | |
1464 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, | |
1465 | "xbzrle_cache_size", | |
54270c45 | 1466 | "a power of two no less than the target page size"); |
73af8dd8 JQ |
1467 | return false; |
1468 | } | |
1469 | ||
4cbc9c7f LQ |
1470 | if (params->has_max_cpu_throttle && |
1471 | (params->max_cpu_throttle < params->cpu_throttle_initial || | |
1472 | params->max_cpu_throttle > 99)) { | |
1473 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, | |
1474 | "max_cpu_throttle", | |
1475 | "an integer in the range of cpu_throttle_initial to 99"); | |
1476 | return false; | |
1477 | } | |
1478 | ||
ee3d96ba DDAG |
1479 | if (params->has_announce_initial && |
1480 | params->announce_initial > 100000) { | |
1481 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, | |
1482 | "announce_initial", | |
54270c45 | 1483 | "a value between 0 and 100000"); |
ee3d96ba DDAG |
1484 | return false; |
1485 | } | |
1486 | if (params->has_announce_max && | |
1487 | params->announce_max > 100000) { | |
1488 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, | |
1489 | "announce_max", | |
54270c45 | 1490 | "a value between 0 and 100000"); |
ee3d96ba DDAG |
1491 | return false; |
1492 | } | |
1493 | if (params->has_announce_rounds && | |
1494 | params->announce_rounds > 1000) { | |
1495 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, | |
1496 | "announce_rounds", | |
54270c45 | 1497 | "a value between 0 and 1000"); |
ee3d96ba DDAG |
1498 | return false; |
1499 | } | |
1500 | if (params->has_announce_step && | |
1501 | (params->announce_step < 1 || | |
1502 | params->announce_step > 10000)) { | |
1503 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, | |
1504 | "announce_step", | |
54270c45 | 1505 | "a value between 0 and 10000"); |
ee3d96ba DDAG |
1506 | return false; |
1507 | } | |
31e4c354 HR |
1508 | |
1509 | if (params->has_block_bitmap_mapping && | |
1510 | !check_dirty_bitmap_mig_alias_map(params->block_bitmap_mapping, errp)) { | |
1511 | error_prepend(errp, "Invalid mapping given for block-bitmap-mapping: "); | |
1512 | return false; | |
1513 | } | |
16d063bc PX |
1514 | return true; |
1515 | } | |
1516 | ||
1bda8b3c MA |
1517 | static void migrate_params_test_apply(MigrateSetParameters *params, |
1518 | MigrationParameters *dest) | |
1519 | { | |
1520 | *dest = migrate_get_current()->parameters; | |
1521 | ||
1522 | /* TODO use QAPI_CLONE() instead of duplicating it inline */ | |
1523 | ||
1524 | if (params->has_compress_level) { | |
1525 | dest->compress_level = params->compress_level; | |
1526 | } | |
1527 | ||
1528 | if (params->has_compress_threads) { | |
1529 | dest->compress_threads = params->compress_threads; | |
1530 | } | |
1531 | ||
1d58872a XG |
1532 | if (params->has_compress_wait_thread) { |
1533 | dest->compress_wait_thread = params->compress_wait_thread; | |
1534 | } | |
1535 | ||
1bda8b3c MA |
1536 | if (params->has_decompress_threads) { |
1537 | dest->decompress_threads = params->decompress_threads; | |
1538 | } | |
1539 | ||
dc14a470 KZ |
1540 | if (params->has_throttle_trigger_threshold) { |
1541 | dest->throttle_trigger_threshold = params->throttle_trigger_threshold; | |
1542 | } | |
1543 | ||
1bda8b3c MA |
1544 | if (params->has_cpu_throttle_initial) { |
1545 | dest->cpu_throttle_initial = params->cpu_throttle_initial; | |
1546 | } | |
1547 | ||
1548 | if (params->has_cpu_throttle_increment) { | |
1549 | dest->cpu_throttle_increment = params->cpu_throttle_increment; | |
1550 | } | |
1551 | ||
cbbf8182 KZ |
1552 | if (params->has_cpu_throttle_tailslow) { |
1553 | dest->cpu_throttle_tailslow = params->cpu_throttle_tailslow; | |
1554 | } | |
1555 | ||
1bda8b3c | 1556 | if (params->has_tls_creds) { |
01fa5598 | 1557 | assert(params->tls_creds->type == QTYPE_QSTRING); |
9728ebfb | 1558 | dest->tls_creds = params->tls_creds->u.s; |
1bda8b3c MA |
1559 | } |
1560 | ||
1561 | if (params->has_tls_hostname) { | |
01fa5598 | 1562 | assert(params->tls_hostname->type == QTYPE_QSTRING); |
9728ebfb | 1563 | dest->tls_hostname = params->tls_hostname->u.s; |
1bda8b3c MA |
1564 | } |
1565 | ||
1566 | if (params->has_max_bandwidth) { | |
1567 | dest->max_bandwidth = params->max_bandwidth; | |
1568 | } | |
1569 | ||
1570 | if (params->has_downtime_limit) { | |
1571 | dest->downtime_limit = params->downtime_limit; | |
1572 | } | |
1573 | ||
1574 | if (params->has_x_checkpoint_delay) { | |
1575 | dest->x_checkpoint_delay = params->x_checkpoint_delay; | |
1576 | } | |
1577 | ||
1578 | if (params->has_block_incremental) { | |
1579 | dest->block_incremental = params->block_incremental; | |
1580 | } | |
cbfd6c95 JQ |
1581 | if (params->has_multifd_channels) { |
1582 | dest->multifd_channels = params->multifd_channels; | |
5e7577a1 | 1583 | } |
96eef042 JQ |
1584 | if (params->has_multifd_compression) { |
1585 | dest->multifd_compression = params->multifd_compression; | |
1586 | } | |
73af8dd8 JQ |
1587 | if (params->has_xbzrle_cache_size) { |
1588 | dest->xbzrle_cache_size = params->xbzrle_cache_size; | |
1589 | } | |
7e555c6c DDAG |
1590 | if (params->has_max_postcopy_bandwidth) { |
1591 | dest->max_postcopy_bandwidth = params->max_postcopy_bandwidth; | |
1592 | } | |
4cbc9c7f LQ |
1593 | if (params->has_max_cpu_throttle) { |
1594 | dest->max_cpu_throttle = params->max_cpu_throttle; | |
1595 | } | |
ee3d96ba DDAG |
1596 | if (params->has_announce_initial) { |
1597 | dest->announce_initial = params->announce_initial; | |
1598 | } | |
1599 | if (params->has_announce_max) { | |
1600 | dest->announce_max = params->announce_max; | |
1601 | } | |
1602 | if (params->has_announce_rounds) { | |
1603 | dest->announce_rounds = params->announce_rounds; | |
1604 | } | |
1605 | if (params->has_announce_step) { | |
1606 | dest->announce_step = params->announce_step; | |
1607 | } | |
31e4c354 HR |
1608 | |
1609 | if (params->has_block_bitmap_mapping) { | |
1610 | dest->has_block_bitmap_mapping = true; | |
1611 | dest->block_bitmap_mapping = params->block_bitmap_mapping; | |
1612 | } | |
1bda8b3c MA |
1613 | } |
1614 | ||
73af8dd8 | 1615 | static void migrate_params_apply(MigrateSetParameters *params, Error **errp) |
16d063bc PX |
1616 | { |
1617 | MigrationState *s = migrate_get_current(); | |
1618 | ||
e87fae4c MA |
1619 | /* TODO use QAPI_CLONE() instead of duplicating it inline */ |
1620 | ||
7f375e04 EB |
1621 | if (params->has_compress_level) { |
1622 | s->parameters.compress_level = params->compress_level; | |
85de8323 | 1623 | } |
476c72aa | 1624 | |
7f375e04 EB |
1625 | if (params->has_compress_threads) { |
1626 | s->parameters.compress_threads = params->compress_threads; | |
85de8323 | 1627 | } |
476c72aa | 1628 | |
1d58872a XG |
1629 | if (params->has_compress_wait_thread) { |
1630 | s->parameters.compress_wait_thread = params->compress_wait_thread; | |
1631 | } | |
1632 | ||
7f375e04 EB |
1633 | if (params->has_decompress_threads) { |
1634 | s->parameters.decompress_threads = params->decompress_threads; | |
85de8323 | 1635 | } |
476c72aa | 1636 | |
dc14a470 KZ |
1637 | if (params->has_throttle_trigger_threshold) { |
1638 | s->parameters.throttle_trigger_threshold = params->throttle_trigger_threshold; | |
1639 | } | |
1640 | ||
7f375e04 EB |
1641 | if (params->has_cpu_throttle_initial) { |
1642 | s->parameters.cpu_throttle_initial = params->cpu_throttle_initial; | |
1626fee3 | 1643 | } |
476c72aa | 1644 | |
7f375e04 EB |
1645 | if (params->has_cpu_throttle_increment) { |
1646 | s->parameters.cpu_throttle_increment = params->cpu_throttle_increment; | |
1626fee3 | 1647 | } |
476c72aa | 1648 | |
cbbf8182 KZ |
1649 | if (params->has_cpu_throttle_tailslow) { |
1650 | s->parameters.cpu_throttle_tailslow = params->cpu_throttle_tailslow; | |
1651 | } | |
1652 | ||
7f375e04 | 1653 | if (params->has_tls_creds) { |
69ef1f36 | 1654 | g_free(s->parameters.tls_creds); |
01fa5598 MA |
1655 | assert(params->tls_creds->type == QTYPE_QSTRING); |
1656 | s->parameters.tls_creds = g_strdup(params->tls_creds->u.s); | |
69ef1f36 | 1657 | } |
476c72aa | 1658 | |
7f375e04 | 1659 | if (params->has_tls_hostname) { |
69ef1f36 | 1660 | g_free(s->parameters.tls_hostname); |
01fa5598 MA |
1661 | assert(params->tls_hostname->type == QTYPE_QSTRING); |
1662 | s->parameters.tls_hostname = g_strdup(params->tls_hostname->u.s); | |
69ef1f36 | 1663 | } |
476c72aa | 1664 | |
d2f1d29b DB |
1665 | if (params->has_tls_authz) { |
1666 | g_free(s->parameters.tls_authz); | |
1667 | assert(params->tls_authz->type == QTYPE_QSTRING); | |
1668 | s->parameters.tls_authz = g_strdup(params->tls_authz->u.s); | |
1669 | } | |
1670 | ||
2ff30257 AA |
1671 | if (params->has_max_bandwidth) { |
1672 | s->parameters.max_bandwidth = params->max_bandwidth; | |
c38c1c14 | 1673 | if (s->to_dst_file && !migration_in_postcopy()) { |
2ff30257 AA |
1674 | qemu_file_set_rate_limit(s->to_dst_file, |
1675 | s->parameters.max_bandwidth / XFER_LIMIT_RATIO); | |
1676 | } | |
1677 | } | |
476c72aa | 1678 | |
2ff30257 AA |
1679 | if (params->has_downtime_limit) { |
1680 | s->parameters.downtime_limit = params->downtime_limit; | |
1681 | } | |
68b53591 HZ |
1682 | |
1683 | if (params->has_x_checkpoint_delay) { | |
1684 | s->parameters.x_checkpoint_delay = params->x_checkpoint_delay; | |
479125d5 HZ |
1685 | if (migration_in_colo_state()) { |
1686 | colo_checkpoint_notify(s); | |
1687 | } | |
68b53591 | 1688 | } |
476c72aa | 1689 | |
2833c59b JQ |
1690 | if (params->has_block_incremental) { |
1691 | s->parameters.block_incremental = params->block_incremental; | |
1692 | } | |
cbfd6c95 JQ |
1693 | if (params->has_multifd_channels) { |
1694 | s->parameters.multifd_channels = params->multifd_channels; | |
4075fb1c | 1695 | } |
96eef042 JQ |
1696 | if (params->has_multifd_compression) { |
1697 | s->parameters.multifd_compression = params->multifd_compression; | |
1698 | } | |
73af8dd8 JQ |
1699 | if (params->has_xbzrle_cache_size) { |
1700 | s->parameters.xbzrle_cache_size = params->xbzrle_cache_size; | |
1701 | xbzrle_cache_resize(params->xbzrle_cache_size, errp); | |
1702 | } | |
7e555c6c DDAG |
1703 | if (params->has_max_postcopy_bandwidth) { |
1704 | s->parameters.max_postcopy_bandwidth = params->max_postcopy_bandwidth; | |
c38c1c14 DDAG |
1705 | if (s->to_dst_file && migration_in_postcopy()) { |
1706 | qemu_file_set_rate_limit(s->to_dst_file, | |
1707 | s->parameters.max_postcopy_bandwidth / XFER_LIMIT_RATIO); | |
1708 | } | |
7e555c6c | 1709 | } |
4cbc9c7f LQ |
1710 | if (params->has_max_cpu_throttle) { |
1711 | s->parameters.max_cpu_throttle = params->max_cpu_throttle; | |
1712 | } | |
ee3d96ba DDAG |
1713 | if (params->has_announce_initial) { |
1714 | s->parameters.announce_initial = params->announce_initial; | |
1715 | } | |
1716 | if (params->has_announce_max) { | |
1717 | s->parameters.announce_max = params->announce_max; | |
1718 | } | |
1719 | if (params->has_announce_rounds) { | |
1720 | s->parameters.announce_rounds = params->announce_rounds; | |
1721 | } | |
1722 | if (params->has_announce_step) { | |
1723 | s->parameters.announce_step = params->announce_step; | |
1724 | } | |
31e4c354 HR |
1725 | |
1726 | if (params->has_block_bitmap_mapping) { | |
1727 | qapi_free_BitmapMigrationNodeAliasList( | |
1728 | s->parameters.block_bitmap_mapping); | |
1729 | ||
1730 | s->parameters.has_block_bitmap_mapping = true; | |
1731 | s->parameters.block_bitmap_mapping = | |
1732 | QAPI_CLONE(BitmapMigrationNodeAliasList, | |
1733 | params->block_bitmap_mapping); | |
1734 | } | |
85de8323 LL |
1735 | } |
1736 | ||
1bda8b3c | 1737 | void qmp_migrate_set_parameters(MigrateSetParameters *params, Error **errp) |
476c72aa | 1738 | { |
1bda8b3c MA |
1739 | MigrationParameters tmp; |
1740 | ||
01fa5598 MA |
1741 | /* TODO Rewrite "" to null instead */ |
1742 | if (params->has_tls_creds | |
1743 | && params->tls_creds->type == QTYPE_QNULL) { | |
cb3e7f08 | 1744 | qobject_unref(params->tls_creds->u.n); |
01fa5598 MA |
1745 | params->tls_creds->type = QTYPE_QSTRING; |
1746 | params->tls_creds->u.s = strdup(""); | |
1747 | } | |
1748 | /* TODO Rewrite "" to null instead */ | |
1749 | if (params->has_tls_hostname | |
1750 | && params->tls_hostname->type == QTYPE_QNULL) { | |
cb3e7f08 | 1751 | qobject_unref(params->tls_hostname->u.n); |
01fa5598 MA |
1752 | params->tls_hostname->type = QTYPE_QSTRING; |
1753 | params->tls_hostname->u.s = strdup(""); | |
1754 | } | |
1755 | ||
1bda8b3c MA |
1756 | migrate_params_test_apply(params, &tmp); |
1757 | ||
1758 | if (!migrate_params_check(&tmp, errp)) { | |
476c72aa PX |
1759 | /* Invalid parameter */ |
1760 | return; | |
1761 | } | |
1762 | ||
73af8dd8 | 1763 | migrate_params_apply(params, errp); |
476c72aa PX |
1764 | } |
1765 | ||
2594f56d | 1766 | |
4886a1bc DDAG |
1767 | void qmp_migrate_start_postcopy(Error **errp) |
1768 | { | |
1769 | MigrationState *s = migrate_get_current(); | |
1770 | ||
16b0fd32 | 1771 | if (!migrate_postcopy()) { |
a54d340b | 1772 | error_setg(errp, "Enable postcopy with migrate_set_capability before" |
4886a1bc DDAG |
1773 | " the start of migration"); |
1774 | return; | |
1775 | } | |
1776 | ||
1777 | if (s->state == MIGRATION_STATUS_NONE) { | |
1778 | error_setg(errp, "Postcopy must be started after migration has been" | |
1779 | " started"); | |
1780 | return; | |
1781 | } | |
1782 | /* | |
1783 | * we don't error if migration has finished since that would be racy | |
1784 | * with issuing this command. | |
1785 | */ | |
d73415a3 | 1786 | qatomic_set(&s->start_postcopy, true); |
4886a1bc DDAG |
1787 | } |
1788 | ||
065e2813 AL |
1789 | /* shared migration helpers */ |
1790 | ||
48781e5b | 1791 | void migrate_set_state(int *state, int old_state, int new_state) |
51cf4c1a | 1792 | { |
a31fedee | 1793 | assert(new_state < MIGRATION_STATUS__MAX); |
d73415a3 | 1794 | if (qatomic_cmpxchg(state, old_state, new_state) == old_state) { |
a31fedee | 1795 | trace_migrate_set_state(MigrationStatus_str(new_state)); |
b05dc723 | 1796 | migrate_generate_event(new_state); |
51cf4c1a Z |
1797 | } |
1798 | } | |
1799 | ||
eaedde52 EB |
1800 | static MigrationCapabilityStatus *migrate_cap_add(MigrationCapability index, |
1801 | bool state) | |
2833c59b | 1802 | { |
eaedde52 | 1803 | MigrationCapabilityStatus *cap; |
2833c59b | 1804 | |
eaedde52 EB |
1805 | cap = g_new0(MigrationCapabilityStatus, 1); |
1806 | cap->capability = index; | |
1807 | cap->state = state; | |
4e4a3d3a PX |
1808 | |
1809 | return cap; | |
1810 | } | |
1811 | ||
1812 | void migrate_set_block_enabled(bool value, Error **errp) | |
1813 | { | |
eaedde52 | 1814 | MigrationCapabilityStatusList *cap = NULL; |
4e4a3d3a | 1815 | |
eaedde52 | 1816 | QAPI_LIST_PREPEND(cap, migrate_cap_add(MIGRATION_CAPABILITY_BLOCK, value)); |
2833c59b JQ |
1817 | qmp_migrate_set_capabilities(cap, errp); |
1818 | qapi_free_MigrationCapabilityStatusList(cap); | |
1819 | } | |
1820 | ||
1821 | static void migrate_set_block_incremental(MigrationState *s, bool value) | |
1822 | { | |
1823 | s->parameters.block_incremental = value; | |
1824 | } | |
1825 | ||
1826 | static void block_cleanup_parameters(MigrationState *s) | |
1827 | { | |
1828 | if (s->must_remove_block_options) { | |
1829 | /* setting to false can never fail */ | |
1830 | migrate_set_block_enabled(false, &error_abort); | |
1831 | migrate_set_block_incremental(s, false); | |
1832 | s->must_remove_block_options = false; | |
1833 | } | |
1834 | } | |
1835 | ||
fd392cfa | 1836 | static void migrate_fd_cleanup(MigrationState *s) |
065e2813 | 1837 | { |
bb1fadc4 PB |
1838 | qemu_bh_delete(s->cleanup_bh); |
1839 | s->cleanup_bh = NULL; | |
1840 | ||
83174765 PX |
1841 | g_free(s->hostname); |
1842 | s->hostname = NULL; | |
1843 | ||
0ceccd85 PX |
1844 | qemu_savevm_state_cleanup(); |
1845 | ||
89a02a9f | 1846 | if (s->to_dst_file) { |
62df066f | 1847 | QEMUFile *tmp; |
f986c3d2 | 1848 | |
9013dca5 | 1849 | trace_migrate_fd_cleanup(); |
404a7c05 | 1850 | qemu_mutex_unlock_iothread(); |
1d34e4bf DDAG |
1851 | if (s->migration_thread_running) { |
1852 | qemu_thread_join(&s->thread); | |
1853 | s->migration_thread_running = false; | |
1854 | } | |
404a7c05 PB |
1855 | qemu_mutex_lock_iothread(); |
1856 | ||
1398b2e3 | 1857 | multifd_save_cleanup(); |
62df066f PX |
1858 | qemu_mutex_lock(&s->qemu_file_lock); |
1859 | tmp = s->to_dst_file; | |
89a02a9f | 1860 | s->to_dst_file = NULL; |
62df066f PX |
1861 | qemu_mutex_unlock(&s->qemu_file_lock); |
1862 | /* | |
1863 | * Close the file handle without the lock to make sure the | |
1864 | * critical section won't block for long. | |
1865 | */ | |
39675fff | 1866 | migration_ioc_unregister_yank_from_file(tmp); |
62df066f | 1867 | qemu_fclose(tmp); |
065e2813 AL |
1868 | } |
1869 | ||
8f8d528e | 1870 | assert(!migration_is_active(s)); |
7a2c1721 | 1871 | |
94f5a437 | 1872 | if (s->state == MIGRATION_STATUS_CANCELLING) { |
48781e5b | 1873 | migrate_set_state(&s->state, MIGRATION_STATUS_CANCELLING, |
94f5a437 | 1874 | MIGRATION_STATUS_CANCELLED); |
7a2c1721 | 1875 | } |
a3fa1d78 | 1876 | |
87db1a7d JQ |
1877 | if (s->error) { |
1878 | /* It is used on info migrate. We can't free it */ | |
1879 | error_report_err(error_copy(s->error)); | |
1880 | } | |
a3fa1d78 | 1881 | notifier_list_notify(&migration_state_notifiers, s); |
2833c59b | 1882 | block_cleanup_parameters(s); |
b5eea99e | 1883 | yank_unregister_instance(MIGRATION_YANK_INSTANCE); |
065e2813 AL |
1884 | } |
1885 | ||
fd392cfa YK |
1886 | static void migrate_fd_cleanup_schedule(MigrationState *s) |
1887 | { | |
1888 | /* | |
1889 | * Ref the state for bh, because it may be called when | |
1890 | * there're already no other refs | |
1891 | */ | |
1892 | object_ref(OBJECT(s)); | |
1893 | qemu_bh_schedule(s->cleanup_bh); | |
1894 | } | |
1895 | ||
1896 | static void migrate_fd_cleanup_bh(void *opaque) | |
1897 | { | |
1898 | MigrationState *s = opaque; | |
1899 | migrate_fd_cleanup(s); | |
1900 | object_unref(OBJECT(s)); | |
1901 | } | |
1902 | ||
87db1a7d JQ |
1903 | void migrate_set_error(MigrationState *s, const Error *error) |
1904 | { | |
6e8a355d | 1905 | QEMU_LOCK_GUARD(&s->error_mutex); |
87db1a7d JQ |
1906 | if (!s->error) { |
1907 | s->error = error_copy(error); | |
1908 | } | |
87db1a7d JQ |
1909 | } |
1910 | ||
ca7bd082 PX |
1911 | static void migrate_error_free(MigrationState *s) |
1912 | { | |
1913 | QEMU_LOCK_GUARD(&s->error_mutex); | |
1914 | if (s->error) { | |
1915 | error_free(s->error); | |
1916 | s->error = NULL; | |
1917 | } | |
1918 | } | |
1919 | ||
d59ce6f3 | 1920 | void migrate_fd_error(MigrationState *s, const Error *error) |
065e2813 | 1921 | { |
25174055 | 1922 | trace_migrate_fd_error(error_get_pretty(error)); |
89a02a9f | 1923 | assert(s->to_dst_file == NULL); |
48781e5b HZ |
1924 | migrate_set_state(&s->state, MIGRATION_STATUS_SETUP, |
1925 | MIGRATION_STATUS_FAILED); | |
87db1a7d | 1926 | migrate_set_error(s, error); |
458cf28e JQ |
1927 | } |
1928 | ||
0edda1c4 | 1929 | static void migrate_fd_cancel(MigrationState *s) |
065e2813 | 1930 | { |
6f2b811a | 1931 | int old_state ; |
89a02a9f | 1932 | QEMUFile *f = migrate_get_current()->to_dst_file; |
9013dca5 | 1933 | trace_migrate_fd_cancel(); |
065e2813 | 1934 | |
43044ac0 PX |
1935 | WITH_QEMU_LOCK_GUARD(&s->qemu_file_lock) { |
1936 | if (s->rp_state.from_dst_file) { | |
1937 | /* shutdown the rp socket, so causing the rp thread to shutdown */ | |
1938 | qemu_file_shutdown(s->rp_state.from_dst_file); | |
1939 | } | |
70b20477 DDAG |
1940 | } |
1941 | ||
6f2b811a Z |
1942 | do { |
1943 | old_state = s->state; | |
392d87e2 | 1944 | if (!migration_is_running(old_state)) { |
6f2b811a Z |
1945 | break; |
1946 | } | |
a7b36b48 DDAG |
1947 | /* If the migration is paused, kick it out of the pause */ |
1948 | if (old_state == MIGRATION_STATUS_PRE_SWITCHOVER) { | |
1949 | qemu_sem_post(&s->pause_sem); | |
1950 | } | |
48781e5b | 1951 | migrate_set_state(&s->state, old_state, MIGRATION_STATUS_CANCELLING); |
31194731 | 1952 | } while (s->state != MIGRATION_STATUS_CANCELLING); |
a26ba26e DDAG |
1953 | |
1954 | /* | |
1955 | * If we're unlucky the migration code might be stuck somewhere in a | |
1956 | * send/write while the network has failed and is waiting to timeout; | |
1957 | * if we've got shutdown(2) available then we can force it to quit. | |
1958 | * The outgoing qemu file gets closed in migrate_fd_cleanup that is | |
1959 | * called in a bh, so there is no race against this cancel. | |
1960 | */ | |
31194731 | 1961 | if (s->state == MIGRATION_STATUS_CANCELLING && f) { |
a26ba26e DDAG |
1962 | qemu_file_shutdown(f); |
1963 | } | |
1d2acc31 HZ |
1964 | if (s->state == MIGRATION_STATUS_CANCELLING && s->block_inactive) { |
1965 | Error *local_err = NULL; | |
1966 | ||
3b717194 | 1967 | bdrv_activate_all(&local_err); |
1d2acc31 HZ |
1968 | if (local_err) { |
1969 | error_report_err(local_err); | |
1970 | } else { | |
1971 | s->block_inactive = false; | |
1972 | } | |
1973 | } | |
065e2813 AL |
1974 | } |
1975 | ||
99a0db9b GH |
1976 | void add_migration_state_change_notifier(Notifier *notify) |
1977 | { | |
1978 | notifier_list_add(&migration_state_notifiers, notify); | |
1979 | } | |
1980 | ||
1981 | void remove_migration_state_change_notifier(Notifier *notify) | |
1982 | { | |
31552529 | 1983 | notifier_remove(notify); |
99a0db9b GH |
1984 | } |
1985 | ||
02edd2e7 | 1986 | bool migration_in_setup(MigrationState *s) |
afe2df69 | 1987 | { |
31194731 | 1988 | return s->state == MIGRATION_STATUS_SETUP; |
afe2df69 GH |
1989 | } |
1990 | ||
7073693b | 1991 | bool migration_has_finished(MigrationState *s) |
99a0db9b | 1992 | { |
31194731 | 1993 | return s->state == MIGRATION_STATUS_COMPLETED; |
99a0db9b | 1994 | } |
0edda1c4 | 1995 | |
afe2df69 GH |
1996 | bool migration_has_failed(MigrationState *s) |
1997 | { | |
31194731 HZ |
1998 | return (s->state == MIGRATION_STATUS_CANCELLED || |
1999 | s->state == MIGRATION_STATUS_FAILED); | |
afe2df69 GH |
2000 | } |
2001 | ||
5727309d | 2002 | bool migration_in_postcopy(void) |
9ec055ae | 2003 | { |
5727309d JQ |
2004 | MigrationState *s = migrate_get_current(); |
2005 | ||
3748fef9 DDAG |
2006 | switch (s->state) { |
2007 | case MIGRATION_STATUS_POSTCOPY_ACTIVE: | |
2008 | case MIGRATION_STATUS_POSTCOPY_PAUSED: | |
2009 | case MIGRATION_STATUS_POSTCOPY_RECOVER: | |
2010 | return true; | |
2011 | default: | |
2012 | return false; | |
2013 | } | |
9ec055ae DDAG |
2014 | } |
2015 | ||
b82fc321 DDAG |
2016 | bool migration_in_postcopy_after_devices(MigrationState *s) |
2017 | { | |
5727309d | 2018 | return migration_in_postcopy() && s->postcopy_after_devices; |
b82fc321 DDAG |
2019 | } |
2020 | ||
06df2e69 DH |
2021 | bool migration_in_incoming_postcopy(void) |
2022 | { | |
2023 | PostcopyState ps = postcopy_state_get(); | |
2024 | ||
2025 | return ps >= POSTCOPY_INCOMING_DISCARD && ps < POSTCOPY_INCOMING_END; | |
2026 | } | |
2027 | ||
1a8e44a8 AG |
2028 | bool migration_in_bg_snapshot(void) |
2029 | { | |
2030 | MigrationState *s = migrate_get_current(); | |
2031 | ||
2032 | return migrate_background_snapshot() && | |
2033 | migration_is_setup_or_active(s->state); | |
2034 | } | |
2035 | ||
fab35005 | 2036 | bool migration_is_idle(void) |
fe44dc91 | 2037 | { |
daff7f0b MA |
2038 | MigrationState *s = current_migration; |
2039 | ||
2040 | if (!s) { | |
2041 | return true; | |
2042 | } | |
fe44dc91 AA |
2043 | |
2044 | switch (s->state) { | |
2045 | case MIGRATION_STATUS_NONE: | |
2046 | case MIGRATION_STATUS_CANCELLED: | |
2047 | case MIGRATION_STATUS_COMPLETED: | |
2048 | case MIGRATION_STATUS_FAILED: | |
2049 | return true; | |
2050 | case MIGRATION_STATUS_SETUP: | |
2051 | case MIGRATION_STATUS_CANCELLING: | |
2052 | case MIGRATION_STATUS_ACTIVE: | |
2053 | case MIGRATION_STATUS_POSTCOPY_ACTIVE: | |
2054 | case MIGRATION_STATUS_COLO: | |
31e06077 DDAG |
2055 | case MIGRATION_STATUS_PRE_SWITCHOVER: |
2056 | case MIGRATION_STATUS_DEVICE: | |
c7e0acd5 | 2057 | case MIGRATION_STATUS_WAIT_UNPLUG: |
fe44dc91 AA |
2058 | return false; |
2059 | case MIGRATION_STATUS__MAX: | |
2060 | g_assert_not_reached(); | |
2061 | } | |
2062 | ||
2063 | return false; | |
2064 | } | |
2065 | ||
8f8d528e WY |
2066 | bool migration_is_active(MigrationState *s) |
2067 | { | |
2068 | return (s->state == MIGRATION_STATUS_ACTIVE || | |
2069 | s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE); | |
2070 | } | |
2071 | ||
3e0c8050 | 2072 | void migrate_init(MigrationState *s) |
0edda1c4 | 2073 | { |
389775d1 DDAG |
2074 | /* |
2075 | * Reinitialise all migration state, except | |
2076 | * parameters/capabilities that the user set, and | |
2077 | * locks. | |
2078 | */ | |
389775d1 | 2079 | s->cleanup_bh = 0; |
8518278a | 2080 | s->vm_start_bh = 0; |
89a02a9f | 2081 | s->to_dst_file = NULL; |
389775d1 | 2082 | s->state = MIGRATION_STATUS_NONE; |
389775d1 DDAG |
2083 | s->rp_state.from_dst_file = NULL; |
2084 | s->rp_state.error = false; | |
2085 | s->mbps = 0.0; | |
aecbfe9c | 2086 | s->pages_per_second = 0.0; |
389775d1 DDAG |
2087 | s->downtime = 0; |
2088 | s->expected_downtime = 0; | |
389775d1 | 2089 | s->setup_time = 0; |
389775d1 | 2090 | s->start_postcopy = false; |
b82fc321 | 2091 | s->postcopy_after_devices = false; |
389775d1 | 2092 | s->migration_thread_running = false; |
d59ce6f3 DB |
2093 | error_free(s->error); |
2094 | s->error = NULL; | |
d8053e73 | 2095 | s->hostname = NULL; |
389775d1 | 2096 | |
48781e5b | 2097 | migrate_set_state(&s->state, MIGRATION_STATUS_NONE, MIGRATION_STATUS_SETUP); |
0edda1c4 | 2098 | |
4af246a3 PX |
2099 | s->start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); |
2100 | s->total_time = 0; | |
7287cbd4 | 2101 | s->vm_was_running = false; |
b15df1ae PX |
2102 | s->iteration_initial_bytes = 0; |
2103 | s->threshold_size = 0; | |
0edda1c4 | 2104 | } |
cab30143 | 2105 | |
60fd6801 | 2106 | int migrate_add_blocker_internal(Error *reason, Error **errp) |
fa2756b7 | 2107 | { |
4c170330 PX |
2108 | /* Snapshots are similar to migrations, so check RUN_STATE_SAVE_VM too. */ |
2109 | if (runstate_check(RUN_STATE_SAVE_VM) || !migration_is_idle()) { | |
2110 | error_propagate_prepend(errp, error_copy(reason), | |
2111 | "disallowing migration blocker " | |
2112 | "(migration/snapshot in progress) for: "); | |
2113 | return -EBUSY; | |
fe44dc91 AA |
2114 | } |
2115 | ||
4c170330 PX |
2116 | migration_blockers = g_slist_prepend(migration_blockers, reason); |
2117 | return 0; | |
fa2756b7 AL |
2118 | } |
2119 | ||
60fd6801 PX |
2120 | int migrate_add_blocker(Error *reason, Error **errp) |
2121 | { | |
2122 | if (only_migratable) { | |
2123 | error_propagate_prepend(errp, error_copy(reason), | |
2124 | "disallowing migration blocker " | |
2125 | "(--only-migratable) for: "); | |
2126 | return -EACCES; | |
2127 | } | |
2128 | ||
2129 | return migrate_add_blocker_internal(reason, errp); | |
2130 | } | |
2131 | ||
fa2756b7 AL |
2132 | void migrate_del_blocker(Error *reason) |
2133 | { | |
2134 | migration_blockers = g_slist_remove(migration_blockers, reason); | |
2135 | } | |
2136 | ||
bf1ae1f4 DDAG |
2137 | void qmp_migrate_incoming(const char *uri, Error **errp) |
2138 | { | |
2139 | Error *local_err = NULL; | |
4debb5f5 | 2140 | static bool once = true; |
bf1ae1f4 | 2141 | |
4debb5f5 DDAG |
2142 | if (!once) { |
2143 | error_setg(errp, "The incoming migration has already been started"); | |
603d5a42 | 2144 | return; |
4debb5f5 | 2145 | } |
e69d50d6 PB |
2146 | if (!runstate_check(RUN_STATE_INMIGRATE)) { |
2147 | error_setg(errp, "'-incoming' was not specified on the command line"); | |
2148 | return; | |
2149 | } | |
bf1ae1f4 | 2150 | |
cc48c587 PX |
2151 | if (!yank_register_instance(MIGRATION_YANK_INSTANCE, errp)) { |
2152 | return; | |
2153 | } | |
2154 | ||
bf1ae1f4 DDAG |
2155 | qemu_start_incoming_migration(uri, &local_err); |
2156 | ||
2157 | if (local_err) { | |
cc48c587 | 2158 | yank_unregister_instance(MIGRATION_YANK_INSTANCE); |
bf1ae1f4 DDAG |
2159 | error_propagate(errp, local_err); |
2160 | return; | |
2161 | } | |
2162 | ||
4debb5f5 | 2163 | once = false; |
bf1ae1f4 DDAG |
2164 | } |
2165 | ||
02affd41 PX |
2166 | void qmp_migrate_recover(const char *uri, Error **errp) |
2167 | { | |
2168 | MigrationIncomingState *mis = migration_incoming_get_current(); | |
2169 | ||
b7f9afd4 PX |
2170 | /* |
2171 | * Don't even bother to use ERRP_GUARD() as it _must_ always be set by | |
2172 | * callers (no one should ignore a recover failure); if there is, it's a | |
2173 | * programming error. | |
2174 | */ | |
2175 | assert(errp); | |
2176 | ||
02affd41 PX |
2177 | if (mis->state != MIGRATION_STATUS_POSTCOPY_PAUSED) { |
2178 | error_setg(errp, "Migrate recover can only be run " | |
2179 | "when postcopy is paused."); | |
2180 | return; | |
2181 | } | |
2182 | ||
08401c04 PX |
2183 | /* If there's an existing transport, release it */ |
2184 | migration_incoming_transport_cleanup(mis); | |
02affd41 PX |
2185 | |
2186 | /* | |
2187 | * Note that this call will never start a real migration; it will | |
2188 | * only re-setup the migration stream and poke existing migration | |
2189 | * to continue using that newly established channel. | |
2190 | */ | |
2191 | qemu_start_incoming_migration(uri, errp); | |
2192 | } | |
2193 | ||
bfbf89c2 PX |
2194 | void qmp_migrate_pause(Error **errp) |
2195 | { | |
2196 | MigrationState *ms = migrate_get_current(); | |
2197 | MigrationIncomingState *mis = migration_incoming_get_current(); | |
2198 | int ret; | |
2199 | ||
2200 | if (ms->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) { | |
2201 | /* Source side, during postcopy */ | |
2202 | qemu_mutex_lock(&ms->qemu_file_lock); | |
2203 | ret = qemu_file_shutdown(ms->to_dst_file); | |
2204 | qemu_mutex_unlock(&ms->qemu_file_lock); | |
2205 | if (ret) { | |
2206 | error_setg(errp, "Failed to pause source migration"); | |
2207 | } | |
2208 | return; | |
2209 | } | |
2210 | ||
2211 | if (mis->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) { | |
2212 | ret = qemu_file_shutdown(mis->from_src_file); | |
2213 | if (ret) { | |
2214 | error_setg(errp, "Failed to pause destination migration"); | |
2215 | } | |
2216 | return; | |
2217 | } | |
2218 | ||
2219 | error_setg(errp, "migrate-pause is currently only supported " | |
2220 | "during postcopy-active state"); | |
2221 | } | |
2222 | ||
24f3902b GK |
2223 | bool migration_is_blocked(Error **errp) |
2224 | { | |
2225 | if (qemu_savevm_state_blocked(errp)) { | |
2226 | return true; | |
2227 | } | |
2228 | ||
2229 | if (migration_blockers) { | |
250561e1 | 2230 | error_propagate(errp, error_copy(migration_blockers->data)); |
24f3902b GK |
2231 | return true; |
2232 | } | |
2233 | ||
2234 | return false; | |
2235 | } | |
2236 | ||
d3e35b8f PX |
2237 | /* Returns true if continue to migrate, or false if error detected */ |
2238 | static bool migrate_prepare(MigrationState *s, bool blk, bool blk_inc, | |
2239 | bool resume, Error **errp) | |
cab30143 | 2240 | { |
be7059cd | 2241 | Error *local_err = NULL; |
d3e35b8f PX |
2242 | |
2243 | if (resume) { | |
2244 | if (s->state != MIGRATION_STATUS_POSTCOPY_PAUSED) { | |
2245 | error_setg(errp, "Cannot resume if there is no " | |
2246 | "paused migration"); | |
2247 | return false; | |
2248 | } | |
97ca211c PX |
2249 | |
2250 | /* | |
2251 | * Postcopy recovery won't work well with release-ram | |
2252 | * capability since release-ram will drop the page buffer as | |
2253 | * long as the page is put into the send buffer. So if there | |
2254 | * is a network failure happened, any page buffers that have | |
2255 | * not yet reached the destination VM but have already been | |
2256 | * sent from the source VM will be lost forever. Let's refuse | |
2257 | * the client from resuming such a postcopy migration. | |
2258 | * Luckily release-ram was designed to only be used when src | |
2259 | * and destination VMs are on the same host, so it should be | |
2260 | * fine. | |
2261 | */ | |
2262 | if (migrate_release_ram()) { | |
2263 | error_setg(errp, "Postcopy recovery cannot work " | |
2264 | "when release-ram capability is set"); | |
2265 | return false; | |
2266 | } | |
2267 | ||
d3e35b8f PX |
2268 | /* This is a resume, skip init status */ |
2269 | return true; | |
2270 | } | |
cab30143 | 2271 | |
392d87e2 | 2272 | if (migration_is_running(s->state)) { |
c6bd8c70 | 2273 | error_setg(errp, QERR_MIGRATION_ACTIVE); |
d3e35b8f | 2274 | return false; |
cab30143 | 2275 | } |
d3e35b8f | 2276 | |
ca99993a DDAG |
2277 | if (runstate_check(RUN_STATE_INMIGRATE)) { |
2278 | error_setg(errp, "Guest is waiting for an incoming migration"); | |
d3e35b8f | 2279 | return false; |
ca99993a DDAG |
2280 | } |
2281 | ||
36d0fe65 T |
2282 | if (runstate_check(RUN_STATE_POSTMIGRATE)) { |
2283 | error_setg(errp, "Can't migrate the vm that was paused due to " | |
2284 | "previous migration"); | |
2285 | return false; | |
2286 | } | |
2287 | ||
24f3902b | 2288 | if (migration_is_blocked(errp)) { |
d3e35b8f | 2289 | return false; |
fa2756b7 AL |
2290 | } |
2291 | ||
d3e35b8f | 2292 | if (blk || blk_inc) { |
3ba02445 RL |
2293 | if (migrate_colo_enabled()) { |
2294 | error_setg(errp, "No disk migration is required in COLO mode"); | |
2295 | return false; | |
2296 | } | |
2833c59b JQ |
2297 | if (migrate_use_block() || migrate_use_block_incremental()) { |
2298 | error_setg(errp, "Command options are incompatible with " | |
2299 | "current migration capabilities"); | |
d3e35b8f | 2300 | return false; |
2833c59b JQ |
2301 | } |
2302 | migrate_set_block_enabled(true, &local_err); | |
2303 | if (local_err) { | |
2304 | error_propagate(errp, local_err); | |
d3e35b8f | 2305 | return false; |
2833c59b JQ |
2306 | } |
2307 | s->must_remove_block_options = true; | |
2308 | } | |
2309 | ||
d3e35b8f | 2310 | if (blk_inc) { |
2833c59b JQ |
2311 | migrate_set_block_incremental(s, true); |
2312 | } | |
2313 | ||
3e0c8050 | 2314 | migrate_init(s); |
87f3bd87 | 2315 | /* |
fa0b31d5 | 2316 | * set ram_counters compression_counters memory to zero for a |
87f3bd87 IR |
2317 | * new migration |
2318 | */ | |
2319 | memset(&ram_counters, 0, sizeof(ram_counters)); | |
fa0b31d5 | 2320 | memset(&compression_counters, 0, sizeof(compression_counters)); |
cab30143 | 2321 | |
d3e35b8f PX |
2322 | return true; |
2323 | } | |
2324 | ||
2325 | void qmp_migrate(const char *uri, bool has_blk, bool blk, | |
2326 | bool has_inc, bool inc, bool has_detach, bool detach, | |
2327 | bool has_resume, bool resume, Error **errp) | |
2328 | { | |
2329 | Error *local_err = NULL; | |
2330 | MigrationState *s = migrate_get_current(); | |
d658f65c | 2331 | const char *p = NULL; |
d3e35b8f PX |
2332 | |
2333 | if (!migrate_prepare(s, has_blk && blk, has_inc && inc, | |
2334 | has_resume && resume, errp)) { | |
2335 | /* Error detected, put into errp */ | |
2336 | return; | |
2337 | } | |
2338 | ||
b5eea99e LS |
2339 | if (!(has_resume && resume)) { |
2340 | if (!yank_register_instance(MIGRATION_YANK_INSTANCE, errp)) { | |
2341 | return; | |
2342 | } | |
2343 | } | |
2344 | ||
f444eeda | 2345 | migrate_protocol_allow_multi_channels(false); |
d658f65c | 2346 | if (strstart(uri, "tcp:", &p) || |
9ba3b2ba LM |
2347 | strstart(uri, "unix:", NULL) || |
2348 | strstart(uri, "vsock:", NULL)) { | |
f444eeda | 2349 | migrate_protocol_allow_multi_channels(true); |
d658f65c | 2350 | socket_start_outgoing_migration(s, p ? p : uri, &local_err); |
2da776db | 2351 | #ifdef CONFIG_RDMA |
41310c68 | 2352 | } else if (strstart(uri, "rdma:", &p)) { |
2da776db MH |
2353 | rdma_start_outgoing_migration(s, p, &local_err); |
2354 | #endif | |
cab30143 | 2355 | } else if (strstart(uri, "exec:", &p)) { |
f37afb5a | 2356 | exec_start_outgoing_migration(s, p, &local_err); |
cab30143 | 2357 | } else if (strstart(uri, "fd:", &p)) { |
f37afb5a | 2358 | fd_start_outgoing_migration(s, p, &local_err); |
99a0db9b | 2359 | } else { |
b5eea99e LS |
2360 | if (!(has_resume && resume)) { |
2361 | yank_unregister_instance(MIGRATION_YANK_INSTANCE); | |
2362 | } | |
c6bd8c70 MA |
2363 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "uri", |
2364 | "a valid migration protocol"); | |
48781e5b HZ |
2365 | migrate_set_state(&s->state, MIGRATION_STATUS_SETUP, |
2366 | MIGRATION_STATUS_FAILED); | |
09576e74 | 2367 | block_cleanup_parameters(s); |
e1c37d0e | 2368 | return; |
cab30143 JQ |
2369 | } |
2370 | ||
f37afb5a | 2371 | if (local_err) { |
b5eea99e LS |
2372 | if (!(has_resume && resume)) { |
2373 | yank_unregister_instance(MIGRATION_YANK_INSTANCE); | |
2374 | } | |
d59ce6f3 | 2375 | migrate_fd_error(s, local_err); |
f37afb5a | 2376 | error_propagate(errp, local_err); |
e1c37d0e | 2377 | return; |
1299c631 | 2378 | } |
cab30143 JQ |
2379 | } |
2380 | ||
6cdedb07 | 2381 | void qmp_migrate_cancel(Error **errp) |
cab30143 | 2382 | { |
458fecca | 2383 | migration_cancel(NULL); |
cab30143 JQ |
2384 | } |
2385 | ||
89cfc02c DDAG |
2386 | void qmp_migrate_continue(MigrationStatus state, Error **errp) |
2387 | { | |
2388 | MigrationState *s = migrate_get_current(); | |
2389 | if (s->state != state) { | |
2390 | error_setg(errp, "Migration not in expected state: %s", | |
2391 | MigrationStatus_str(s->state)); | |
2392 | return; | |
2393 | } | |
2394 | qemu_sem_post(&s->pause_sem); | |
2395 | } | |
2396 | ||
53f09a10 PB |
2397 | bool migrate_release_ram(void) |
2398 | { | |
2399 | MigrationState *s; | |
2400 | ||
2401 | s = migrate_get_current(); | |
2402 | ||
2403 | return s->enabled_capabilities[MIGRATION_CAPABILITY_RELEASE_RAM]; | |
2404 | } | |
2405 | ||
53dd370c DDAG |
2406 | bool migrate_postcopy_ram(void) |
2407 | { | |
2408 | MigrationState *s; | |
2409 | ||
2410 | s = migrate_get_current(); | |
2411 | ||
32c3db5b | 2412 | return s->enabled_capabilities[MIGRATION_CAPABILITY_POSTCOPY_RAM]; |
53dd370c DDAG |
2413 | } |
2414 | ||
58110f0a VSO |
2415 | bool migrate_postcopy(void) |
2416 | { | |
dd6bb914 | 2417 | return migrate_postcopy_ram() || migrate_dirty_bitmaps(); |
58110f0a VSO |
2418 | } |
2419 | ||
bde1e2ec CV |
2420 | bool migrate_auto_converge(void) |
2421 | { | |
2422 | MigrationState *s; | |
2423 | ||
2424 | s = migrate_get_current(); | |
2425 | ||
2426 | return s->enabled_capabilities[MIGRATION_CAPABILITY_AUTO_CONVERGE]; | |
2427 | } | |
2428 | ||
323004a3 PL |
2429 | bool migrate_zero_blocks(void) |
2430 | { | |
2431 | MigrationState *s; | |
2432 | ||
2433 | s = migrate_get_current(); | |
2434 | ||
2435 | return s->enabled_capabilities[MIGRATION_CAPABILITY_ZERO_BLOCKS]; | |
2436 | } | |
2437 | ||
f22f928e AP |
2438 | bool migrate_postcopy_blocktime(void) |
2439 | { | |
2440 | MigrationState *s; | |
2441 | ||
2442 | s = migrate_get_current(); | |
2443 | ||
2444 | return s->enabled_capabilities[MIGRATION_CAPABILITY_POSTCOPY_BLOCKTIME]; | |
2445 | } | |
2446 | ||
8706d2d5 LL |
2447 | bool migrate_use_compression(void) |
2448 | { | |
dde4e694 LL |
2449 | MigrationState *s; |
2450 | ||
2451 | s = migrate_get_current(); | |
2452 | ||
2453 | return s->enabled_capabilities[MIGRATION_CAPABILITY_COMPRESS]; | |
8706d2d5 LL |
2454 | } |
2455 | ||
2456 | int migrate_compress_level(void) | |
2457 | { | |
2458 | MigrationState *s; | |
2459 | ||
2460 | s = migrate_get_current(); | |
2461 | ||
2594f56d | 2462 | return s->parameters.compress_level; |
8706d2d5 LL |
2463 | } |
2464 | ||
2465 | int migrate_compress_threads(void) | |
2466 | { | |
2467 | MigrationState *s; | |
2468 | ||
2469 | s = migrate_get_current(); | |
2470 | ||
2594f56d | 2471 | return s->parameters.compress_threads; |
8706d2d5 LL |
2472 | } |
2473 | ||
1d58872a XG |
2474 | int migrate_compress_wait_thread(void) |
2475 | { | |
2476 | MigrationState *s; | |
2477 | ||
2478 | s = migrate_get_current(); | |
2479 | ||
2480 | return s->parameters.compress_wait_thread; | |
2481 | } | |
2482 | ||
3fcb38c2 LL |
2483 | int migrate_decompress_threads(void) |
2484 | { | |
2485 | MigrationState *s; | |
2486 | ||
2487 | s = migrate_get_current(); | |
2488 | ||
2594f56d | 2489 | return s->parameters.decompress_threads; |
3fcb38c2 LL |
2490 | } |
2491 | ||
55efc8c2 VSO |
2492 | bool migrate_dirty_bitmaps(void) |
2493 | { | |
2494 | MigrationState *s; | |
2495 | ||
2496 | s = migrate_get_current(); | |
2497 | ||
2498 | return s->enabled_capabilities[MIGRATION_CAPABILITY_DIRTY_BITMAPS]; | |
2499 | } | |
2500 | ||
18269069 YK |
2501 | bool migrate_ignore_shared(void) |
2502 | { | |
2503 | MigrationState *s; | |
2504 | ||
2505 | s = migrate_get_current(); | |
2506 | ||
2507 | return s->enabled_capabilities[MIGRATION_CAPABILITY_X_IGNORE_SHARED]; | |
2508 | } | |
2509 | ||
b9d68df6 YK |
2510 | bool migrate_validate_uuid(void) |
2511 | { | |
2512 | MigrationState *s; | |
2513 | ||
2514 | s = migrate_get_current(); | |
2515 | ||
2516 | return s->enabled_capabilities[MIGRATION_CAPABILITY_VALIDATE_UUID]; | |
2517 | } | |
2518 | ||
b05dc723 JQ |
2519 | bool migrate_use_events(void) |
2520 | { | |
2521 | MigrationState *s; | |
2522 | ||
2523 | s = migrate_get_current(); | |
2524 | ||
2525 | return s->enabled_capabilities[MIGRATION_CAPABILITY_EVENTS]; | |
2526 | } | |
2527 | ||
30126bbf JQ |
2528 | bool migrate_use_multifd(void) |
2529 | { | |
2530 | MigrationState *s; | |
2531 | ||
2532 | s = migrate_get_current(); | |
2533 | ||
cbfd6c95 | 2534 | return s->enabled_capabilities[MIGRATION_CAPABILITY_MULTIFD]; |
30126bbf JQ |
2535 | } |
2536 | ||
93fbd031 DDAG |
2537 | bool migrate_pause_before_switchover(void) |
2538 | { | |
2539 | MigrationState *s; | |
2540 | ||
2541 | s = migrate_get_current(); | |
2542 | ||
2543 | return s->enabled_capabilities[ | |
2544 | MIGRATION_CAPABILITY_PAUSE_BEFORE_SWITCHOVER]; | |
2545 | } | |
2546 | ||
4075fb1c JQ |
2547 | int migrate_multifd_channels(void) |
2548 | { | |
2549 | MigrationState *s; | |
2550 | ||
2551 | s = migrate_get_current(); | |
2552 | ||
cbfd6c95 | 2553 | return s->parameters.multifd_channels; |
4075fb1c JQ |
2554 | } |
2555 | ||
ab7cbb0b JQ |
2556 | MultiFDCompression migrate_multifd_compression(void) |
2557 | { | |
2558 | MigrationState *s; | |
2559 | ||
2560 | s = migrate_get_current(); | |
2561 | ||
2562 | return s->parameters.multifd_compression; | |
2563 | } | |
2564 | ||
9004db48 JQ |
2565 | int migrate_multifd_zlib_level(void) |
2566 | { | |
2567 | MigrationState *s; | |
2568 | ||
2569 | s = migrate_get_current(); | |
2570 | ||
2571 | return s->parameters.multifd_zlib_level; | |
2572 | } | |
2573 | ||
6a9ad154 JQ |
2574 | int migrate_multifd_zstd_level(void) |
2575 | { | |
2576 | MigrationState *s; | |
2577 | ||
2578 | s = migrate_get_current(); | |
2579 | ||
2580 | return s->parameters.multifd_zstd_level; | |
2581 | } | |
2582 | ||
abb6295b LB |
2583 | #ifdef CONFIG_LINUX |
2584 | bool migrate_use_zero_copy_send(void) | |
2585 | { | |
2586 | MigrationState *s; | |
2587 | ||
2588 | s = migrate_get_current(); | |
2589 | ||
1abaec9a | 2590 | return s->enabled_capabilities[MIGRATION_CAPABILITY_ZERO_COPY_SEND]; |
abb6295b LB |
2591 | } |
2592 | #endif | |
2593 | ||
d2fafb6a LB |
2594 | int migrate_use_tls(void) |
2595 | { | |
2596 | MigrationState *s; | |
2597 | ||
2598 | s = migrate_get_current(); | |
2599 | ||
2600 | return s->parameters.tls_creds && *s->parameters.tls_creds; | |
2601 | } | |
2602 | ||
17ad9b35 OW |
2603 | int migrate_use_xbzrle(void) |
2604 | { | |
2605 | MigrationState *s; | |
2606 | ||
2607 | s = migrate_get_current(); | |
2608 | ||
2609 | return s->enabled_capabilities[MIGRATION_CAPABILITY_XBZRLE]; | |
2610 | } | |
2611 | ||
8b9407a0 | 2612 | uint64_t migrate_xbzrle_cache_size(void) |
17ad9b35 OW |
2613 | { |
2614 | MigrationState *s; | |
2615 | ||
2616 | s = migrate_get_current(); | |
2617 | ||
73af8dd8 | 2618 | return s->parameters.xbzrle_cache_size; |
17ad9b35 | 2619 | } |
0d82d0e8 | 2620 | |
7e555c6c DDAG |
2621 | static int64_t migrate_max_postcopy_bandwidth(void) |
2622 | { | |
2623 | MigrationState *s; | |
2624 | ||
2625 | s = migrate_get_current(); | |
2626 | ||
2627 | return s->parameters.max_postcopy_bandwidth; | |
2628 | } | |
2629 | ||
2833c59b JQ |
2630 | bool migrate_use_block(void) |
2631 | { | |
2632 | MigrationState *s; | |
2633 | ||
2634 | s = migrate_get_current(); | |
2635 | ||
2636 | return s->enabled_capabilities[MIGRATION_CAPABILITY_BLOCK]; | |
2637 | } | |
2638 | ||
c788ada8 PX |
2639 | bool migrate_use_return_path(void) |
2640 | { | |
2641 | MigrationState *s; | |
2642 | ||
2643 | s = migrate_get_current(); | |
2644 | ||
2645 | return s->enabled_capabilities[MIGRATION_CAPABILITY_RETURN_PATH]; | |
2646 | } | |
2647 | ||
2833c59b JQ |
2648 | bool migrate_use_block_incremental(void) |
2649 | { | |
2650 | MigrationState *s; | |
2651 | ||
2652 | s = migrate_get_current(); | |
2653 | ||
2654 | return s->parameters.block_incremental; | |
2655 | } | |
2656 | ||
6e8c25b4 AG |
2657 | bool migrate_background_snapshot(void) |
2658 | { | |
2659 | MigrationState *s; | |
2660 | ||
2661 | s = migrate_get_current(); | |
2662 | ||
2663 | return s->enabled_capabilities[MIGRATION_CAPABILITY_BACKGROUND_SNAPSHOT]; | |
2664 | } | |
2665 | ||
70b20477 DDAG |
2666 | /* migration thread support */ |
2667 | /* | |
2668 | * Something bad happened to the RP stream, mark an error | |
2669 | * The caller shall print or trace something to indicate why | |
2670 | */ | |
2671 | static void mark_source_rp_bad(MigrationState *s) | |
2672 | { | |
2673 | s->rp_state.error = true; | |
2674 | } | |
2675 | ||
2676 | static struct rp_cmd_args { | |
2677 | ssize_t len; /* -1 = variable */ | |
2678 | const char *name; | |
2679 | } rp_cmd_args[] = { | |
2680 | [MIG_RP_MSG_INVALID] = { .len = -1, .name = "INVALID" }, | |
2681 | [MIG_RP_MSG_SHUT] = { .len = 4, .name = "SHUT" }, | |
2682 | [MIG_RP_MSG_PONG] = { .len = 4, .name = "PONG" }, | |
1e2d90eb DDAG |
2683 | [MIG_RP_MSG_REQ_PAGES] = { .len = 12, .name = "REQ_PAGES" }, |
2684 | [MIG_RP_MSG_REQ_PAGES_ID] = { .len = -1, .name = "REQ_PAGES_ID" }, | |
a335debb | 2685 | [MIG_RP_MSG_RECV_BITMAP] = { .len = -1, .name = "RECV_BITMAP" }, |
13955b89 | 2686 | [MIG_RP_MSG_RESUME_ACK] = { .len = 4, .name = "RESUME_ACK" }, |
70b20477 DDAG |
2687 | [MIG_RP_MSG_MAX] = { .len = -1, .name = "MAX" }, |
2688 | }; | |
2689 | ||
1e2d90eb DDAG |
2690 | /* |
2691 | * Process a request for pages received on the return path, | |
2692 | * We're allowed to send more than requested (e.g. to round to our page size) | |
2693 | * and we don't need to send pages that have already been sent. | |
2694 | */ | |
2695 | static void migrate_handle_rp_req_pages(MigrationState *ms, const char* rbname, | |
2696 | ram_addr_t start, size_t len) | |
2697 | { | |
8e3b0cbb | 2698 | long our_host_ps = qemu_real_host_page_size(); |
6c595cde | 2699 | |
1e2d90eb | 2700 | trace_migrate_handle_rp_req_pages(rbname, start, len); |
6c595cde DDAG |
2701 | |
2702 | /* | |
2703 | * Since we currently insist on matching page sizes, just sanity check | |
2704 | * we're being asked for whole host pages. | |
2705 | */ | |
7648297d DH |
2706 | if (!QEMU_IS_ALIGNED(start, our_host_ps) || |
2707 | !QEMU_IS_ALIGNED(len, our_host_ps)) { | |
6c595cde DDAG |
2708 | error_report("%s: Misaligned page request, start: " RAM_ADDR_FMT |
2709 | " len: %zd", __func__, start, len); | |
2710 | mark_source_rp_bad(ms); | |
2711 | return; | |
2712 | } | |
2713 | ||
96506894 | 2714 | if (ram_save_queue_pages(rbname, start, len)) { |
6c595cde DDAG |
2715 | mark_source_rp_bad(ms); |
2716 | } | |
1e2d90eb DDAG |
2717 | } |
2718 | ||
14b1742e PX |
2719 | /* Return true to retry, false to quit */ |
2720 | static bool postcopy_pause_return_path_thread(MigrationState *s) | |
2721 | { | |
2722 | trace_postcopy_pause_return_path(); | |
2723 | ||
2724 | qemu_sem_wait(&s->postcopy_pause_rp_sem); | |
2725 | ||
2726 | trace_postcopy_pause_return_path_continued(); | |
2727 | ||
2728 | return true; | |
2729 | } | |
2730 | ||
a335debb PX |
2731 | static int migrate_handle_rp_recv_bitmap(MigrationState *s, char *block_name) |
2732 | { | |
2733 | RAMBlock *block = qemu_ram_block_by_name(block_name); | |
2734 | ||
2735 | if (!block) { | |
2736 | error_report("%s: invalid block name '%s'", __func__, block_name); | |
2737 | return -EINVAL; | |
2738 | } | |
2739 | ||
2740 | /* Fetch the received bitmap and refresh the dirty bitmap */ | |
2741 | return ram_dirty_bitmap_reload(s, block); | |
2742 | } | |
2743 | ||
13955b89 PX |
2744 | static int migrate_handle_rp_resume_ack(MigrationState *s, uint32_t value) |
2745 | { | |
2746 | trace_source_return_path_thread_resume_ack(value); | |
2747 | ||
2748 | if (value != MIGRATION_RESUME_ACK_VALUE) { | |
2749 | error_report("%s: illegal resume_ack value %"PRIu32, | |
2750 | __func__, value); | |
2751 | return -1; | |
2752 | } | |
2753 | ||
2754 | /* Now both sides are active. */ | |
2755 | migrate_set_state(&s->state, MIGRATION_STATUS_POSTCOPY_RECOVER, | |
2756 | MIGRATION_STATUS_POSTCOPY_ACTIVE); | |
2757 | ||
94190696 PX |
2758 | /* Notify send thread that time to continue send pages */ |
2759 | qemu_sem_post(&s->rp_state.rp_sem); | |
13955b89 PX |
2760 | |
2761 | return 0; | |
2762 | } | |
2763 | ||
43044ac0 PX |
2764 | /* Release ms->rp_state.from_dst_file in a safe way */ |
2765 | static void migration_release_from_dst_file(MigrationState *ms) | |
2766 | { | |
2767 | QEMUFile *file; | |
2768 | ||
2769 | WITH_QEMU_LOCK_GUARD(&ms->qemu_file_lock) { | |
2770 | /* | |
2771 | * Reset the from_dst_file pointer first before releasing it, as we | |
2772 | * can't block within lock section | |
2773 | */ | |
2774 | file = ms->rp_state.from_dst_file; | |
2775 | ms->rp_state.from_dst_file = NULL; | |
2776 | } | |
2777 | ||
2778 | qemu_fclose(file); | |
2779 | } | |
2780 | ||
70b20477 DDAG |
2781 | /* |
2782 | * Handles messages sent on the return path towards the source VM | |
2783 | * | |
2784 | */ | |
2785 | static void *source_return_path_thread(void *opaque) | |
2786 | { | |
2787 | MigrationState *ms = opaque; | |
2788 | QEMUFile *rp = ms->rp_state.from_dst_file; | |
2789 | uint16_t header_len, header_type; | |
568b01ca | 2790 | uint8_t buf[512]; |
70b20477 | 2791 | uint32_t tmp32, sibling_error; |
1e2d90eb DDAG |
2792 | ram_addr_t start = 0; /* =0 to silence warning */ |
2793 | size_t len = 0, expected_len; | |
70b20477 DDAG |
2794 | int res; |
2795 | ||
2796 | trace_source_return_path_thread_entry(); | |
74637e6f | 2797 | rcu_register_thread(); |
14b1742e PX |
2798 | |
2799 | retry: | |
70b20477 DDAG |
2800 | while (!ms->rp_state.error && !qemu_file_get_error(rp) && |
2801 | migration_is_setup_or_active(ms->state)) { | |
2802 | trace_source_return_path_thread_loop_top(); | |
2803 | header_type = qemu_get_be16(rp); | |
2804 | header_len = qemu_get_be16(rp); | |
2805 | ||
7a9ddfbf PX |
2806 | if (qemu_file_get_error(rp)) { |
2807 | mark_source_rp_bad(ms); | |
2808 | goto out; | |
2809 | } | |
2810 | ||
70b20477 DDAG |
2811 | if (header_type >= MIG_RP_MSG_MAX || |
2812 | header_type == MIG_RP_MSG_INVALID) { | |
2813 | error_report("RP: Received invalid message 0x%04x length 0x%04x", | |
ed8b2828 | 2814 | header_type, header_len); |
70b20477 DDAG |
2815 | mark_source_rp_bad(ms); |
2816 | goto out; | |
2817 | } | |
2818 | ||
2819 | if ((rp_cmd_args[header_type].len != -1 && | |
2820 | header_len != rp_cmd_args[header_type].len) || | |
568b01ca | 2821 | header_len > sizeof(buf)) { |
70b20477 | 2822 | error_report("RP: Received '%s' message (0x%04x) with" |
ed8b2828 MZ |
2823 | "incorrect length %d expecting %zu", |
2824 | rp_cmd_args[header_type].name, header_type, header_len, | |
2825 | (size_t)rp_cmd_args[header_type].len); | |
70b20477 DDAG |
2826 | mark_source_rp_bad(ms); |
2827 | goto out; | |
2828 | } | |
2829 | ||
2830 | /* We know we've got a valid header by this point */ | |
2831 | res = qemu_get_buffer(rp, buf, header_len); | |
2832 | if (res != header_len) { | |
2833 | error_report("RP: Failed reading data for message 0x%04x" | |
2834 | " read %d expected %d", | |
2835 | header_type, res, header_len); | |
2836 | mark_source_rp_bad(ms); | |
2837 | goto out; | |
2838 | } | |
2839 | ||
2840 | /* OK, we have the message and the data */ | |
2841 | switch (header_type) { | |
2842 | case MIG_RP_MSG_SHUT: | |
4d885131 | 2843 | sibling_error = ldl_be_p(buf); |
70b20477 DDAG |
2844 | trace_source_return_path_thread_shut(sibling_error); |
2845 | if (sibling_error) { | |
2846 | error_report("RP: Sibling indicated error %d", sibling_error); | |
2847 | mark_source_rp_bad(ms); | |
2848 | } | |
2849 | /* | |
2850 | * We'll let the main thread deal with closing the RP | |
2851 | * we could do a shutdown(2) on it, but we're the only user | |
2852 | * anyway, so there's nothing gained. | |
2853 | */ | |
2854 | goto out; | |
2855 | ||
2856 | case MIG_RP_MSG_PONG: | |
4d885131 | 2857 | tmp32 = ldl_be_p(buf); |
70b20477 DDAG |
2858 | trace_source_return_path_thread_pong(tmp32); |
2859 | break; | |
2860 | ||
1e2d90eb | 2861 | case MIG_RP_MSG_REQ_PAGES: |
4d885131 PM |
2862 | start = ldq_be_p(buf); |
2863 | len = ldl_be_p(buf + 8); | |
1e2d90eb DDAG |
2864 | migrate_handle_rp_req_pages(ms, NULL, start, len); |
2865 | break; | |
2866 | ||
2867 | case MIG_RP_MSG_REQ_PAGES_ID: | |
2868 | expected_len = 12 + 1; /* header + termination */ | |
2869 | ||
2870 | if (header_len >= expected_len) { | |
4d885131 PM |
2871 | start = ldq_be_p(buf); |
2872 | len = ldl_be_p(buf + 8); | |
1e2d90eb DDAG |
2873 | /* Now we expect an idstr */ |
2874 | tmp32 = buf[12]; /* Length of the following idstr */ | |
2875 | buf[13 + tmp32] = '\0'; | |
2876 | expected_len += tmp32; | |
2877 | } | |
2878 | if (header_len != expected_len) { | |
2879 | error_report("RP: Req_Page_id with length %d expecting %zd", | |
ed8b2828 | 2880 | header_len, expected_len); |
1e2d90eb DDAG |
2881 | mark_source_rp_bad(ms); |
2882 | goto out; | |
2883 | } | |
2884 | migrate_handle_rp_req_pages(ms, (char *)&buf[13], start, len); | |
2885 | break; | |
2886 | ||
a335debb PX |
2887 | case MIG_RP_MSG_RECV_BITMAP: |
2888 | if (header_len < 1) { | |
2889 | error_report("%s: missing block name", __func__); | |
2890 | mark_source_rp_bad(ms); | |
2891 | goto out; | |
2892 | } | |
2893 | /* Format: len (1B) + idstr (<255B). This ends the idstr. */ | |
2894 | buf[buf[0] + 1] = '\0'; | |
2895 | if (migrate_handle_rp_recv_bitmap(ms, (char *)(buf + 1))) { | |
2896 | mark_source_rp_bad(ms); | |
2897 | goto out; | |
2898 | } | |
2899 | break; | |
2900 | ||
13955b89 PX |
2901 | case MIG_RP_MSG_RESUME_ACK: |
2902 | tmp32 = ldl_be_p(buf); | |
2903 | if (migrate_handle_rp_resume_ack(ms, tmp32)) { | |
2904 | mark_source_rp_bad(ms); | |
2905 | goto out; | |
2906 | } | |
2907 | break; | |
2908 | ||
70b20477 DDAG |
2909 | default: |
2910 | break; | |
2911 | } | |
2912 | } | |
14b1742e PX |
2913 | |
2914 | out: | |
2915 | res = qemu_file_get_error(rp); | |
2916 | if (res) { | |
d5c8f2af | 2917 | if (res && migration_in_postcopy()) { |
14b1742e PX |
2918 | /* |
2919 | * Maybe there is something we can do: it looks like a | |
2920 | * network down issue, and we pause for a recovery. | |
2921 | */ | |
43044ac0 | 2922 | migration_release_from_dst_file(ms); |
2e3e3da3 | 2923 | rp = NULL; |
14b1742e | 2924 | if (postcopy_pause_return_path_thread(ms)) { |
43044ac0 PX |
2925 | /* |
2926 | * Reload rp, reset the rest. Referencing it is safe since | |
2927 | * it's reset only by us above, or when migration completes | |
2928 | */ | |
2e3e3da3 | 2929 | rp = ms->rp_state.from_dst_file; |
14b1742e PX |
2930 | ms->rp_state.error = false; |
2931 | goto retry; | |
2932 | } | |
2933 | } | |
2934 | ||
70b20477 DDAG |
2935 | trace_source_return_path_thread_bad_end(); |
2936 | mark_source_rp_bad(ms); | |
2937 | } | |
2938 | ||
2939 | trace_source_return_path_thread_end(); | |
43044ac0 | 2940 | migration_release_from_dst_file(ms); |
74637e6f | 2941 | rcu_unregister_thread(); |
70b20477 DDAG |
2942 | return NULL; |
2943 | } | |
2944 | ||
d3e35b8f PX |
2945 | static int open_return_path_on_source(MigrationState *ms, |
2946 | bool create_thread) | |
70b20477 | 2947 | { |
89a02a9f | 2948 | ms->rp_state.from_dst_file = qemu_file_get_return_path(ms->to_dst_file); |
70b20477 DDAG |
2949 | if (!ms->rp_state.from_dst_file) { |
2950 | return -1; | |
2951 | } | |
2952 | ||
2953 | trace_open_return_path_on_source(); | |
d3e35b8f PX |
2954 | |
2955 | if (!create_thread) { | |
2956 | /* We're done */ | |
2957 | return 0; | |
2958 | } | |
2959 | ||
70b20477 DDAG |
2960 | qemu_thread_create(&ms->rp_state.rp_thread, "return path", |
2961 | source_return_path_thread, ms, QEMU_THREAD_JOINABLE); | |
53021ea1 | 2962 | ms->rp_state.rp_thread_created = true; |
70b20477 DDAG |
2963 | |
2964 | trace_open_return_path_on_source_continue(); | |
2965 | ||
2966 | return 0; | |
2967 | } | |
2968 | ||
70b20477 DDAG |
2969 | /* Returns 0 if the RP was ok, otherwise there was an error on the RP */ |
2970 | static int await_return_path_close_on_source(MigrationState *ms) | |
2971 | { | |
2972 | /* | |
2973 | * If this is a normal exit then the destination will send a SHUT and the | |
2974 | * rp_thread will exit, however if there's an error we need to cause | |
2975 | * it to exit. | |
2976 | */ | |
89a02a9f | 2977 | if (qemu_file_get_error(ms->to_dst_file) && ms->rp_state.from_dst_file) { |
70b20477 DDAG |
2978 | /* |
2979 | * shutdown(2), if we have it, will cause it to unblock if it's stuck | |
2980 | * waiting for the destination. | |
2981 | */ | |
2982 | qemu_file_shutdown(ms->rp_state.from_dst_file); | |
2983 | mark_source_rp_bad(ms); | |
2984 | } | |
2985 | trace_await_return_path_close_on_source_joining(); | |
2986 | qemu_thread_join(&ms->rp_state.rp_thread); | |
53021ea1 | 2987 | ms->rp_state.rp_thread_created = false; |
70b20477 DDAG |
2988 | trace_await_return_path_close_on_source_close(); |
2989 | return ms->rp_state.error; | |
2990 | } | |
2991 | ||
1d34e4bf DDAG |
2992 | /* |
2993 | * Switch from normal iteration to postcopy | |
2994 | * Returns non-0 on error | |
2995 | */ | |
7287cbd4 | 2996 | static int postcopy_start(MigrationState *ms) |
1d34e4bf DDAG |
2997 | { |
2998 | int ret; | |
61b67d47 DB |
2999 | QIOChannelBuffer *bioc; |
3000 | QEMUFile *fb; | |
1d34e4bf | 3001 | int64_t time_at_stop = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); |
7e555c6c | 3002 | int64_t bandwidth = migrate_max_postcopy_bandwidth(); |
ef8d6488 | 3003 | bool restart_block = false; |
0331c8ca DDAG |
3004 | int cur_state = MIGRATION_STATUS_ACTIVE; |
3005 | if (!migrate_pause_before_switchover()) { | |
3006 | migrate_set_state(&ms->state, MIGRATION_STATUS_ACTIVE, | |
3007 | MIGRATION_STATUS_POSTCOPY_ACTIVE); | |
3008 | } | |
1d34e4bf DDAG |
3009 | |
3010 | trace_postcopy_start(); | |
3011 | qemu_mutex_lock_iothread(); | |
3012 | trace_postcopy_start_set_run(); | |
3013 | ||
fb064112 | 3014 | qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER, NULL); |
1d34e4bf DDAG |
3015 | global_state_store(); |
3016 | ret = vm_stop_force_state(RUN_STATE_FINISH_MIGRATE); | |
76b1c7fe KW |
3017 | if (ret < 0) { |
3018 | goto fail; | |
3019 | } | |
1d34e4bf | 3020 | |
0331c8ca DDAG |
3021 | ret = migration_maybe_pause(ms, &cur_state, |
3022 | MIGRATION_STATUS_POSTCOPY_ACTIVE); | |
3023 | if (ret < 0) { | |
3024 | goto fail; | |
3025 | } | |
3026 | ||
76b1c7fe | 3027 | ret = bdrv_inactivate_all(); |
1d34e4bf DDAG |
3028 | if (ret < 0) { |
3029 | goto fail; | |
3030 | } | |
ef8d6488 | 3031 | restart_block = true; |
1d34e4bf | 3032 | |
1c0d249d DDAG |
3033 | /* |
3034 | * Cause any non-postcopiable, but iterative devices to | |
3035 | * send out their final data. | |
3036 | */ | |
a1fbe750 | 3037 | qemu_savevm_state_complete_precopy(ms->to_dst_file, true, false); |
1c0d249d | 3038 | |
1d34e4bf DDAG |
3039 | /* |
3040 | * in Finish migrate and with the io-lock held everything should | |
3041 | * be quiet, but we've potentially still got dirty pages and we | |
3042 | * need to tell the destination to throw any pages it's already received | |
3043 | * that are dirty | |
3044 | */ | |
58110f0a | 3045 | if (migrate_postcopy_ram()) { |
739fcc1b | 3046 | ram_postcopy_send_discard_bitmap(ms); |
1d34e4bf DDAG |
3047 | } |
3048 | ||
3049 | /* | |
3050 | * send rest of state - note things that are doing postcopy | |
3051 | * will notice we're in POSTCOPY_ACTIVE and not actually | |
3052 | * wrap their state up here | |
3053 | */ | |
7e555c6c DDAG |
3054 | /* 0 max-postcopy-bandwidth means unlimited */ |
3055 | if (!bandwidth) { | |
3056 | qemu_file_set_rate_limit(ms->to_dst_file, INT64_MAX); | |
3057 | } else { | |
3058 | qemu_file_set_rate_limit(ms->to_dst_file, bandwidth / XFER_LIMIT_RATIO); | |
3059 | } | |
58110f0a VSO |
3060 | if (migrate_postcopy_ram()) { |
3061 | /* Ping just for debugging, helps line traces up */ | |
3062 | qemu_savevm_send_ping(ms->to_dst_file, 2); | |
3063 | } | |
1d34e4bf DDAG |
3064 | |
3065 | /* | |
3066 | * While loading the device state we may trigger page transfer | |
3067 | * requests and the fd must be free to process those, and thus | |
3068 | * the destination must read the whole device state off the fd before | |
3069 | * it starts processing it. Unfortunately the ad-hoc migration format | |
3070 | * doesn't allow the destination to know the size to read without fully | |
3071 | * parsing it through each devices load-state code (especially the open | |
3072 | * coded devices that use get/put). | |
3073 | * So we wrap the device state up in a package with a length at the start; | |
3074 | * to do this we use a qemu_buf to hold the whole of the device state. | |
3075 | */ | |
61b67d47 | 3076 | bioc = qio_channel_buffer_new(4096); |
6f01f136 | 3077 | qio_channel_set_name(QIO_CHANNEL(bioc), "migration-postcopy-buffer"); |
77ef2dc1 | 3078 | fb = qemu_file_new_output(QIO_CHANNEL(bioc)); |
61b67d47 | 3079 | object_unref(OBJECT(bioc)); |
1d34e4bf | 3080 | |
c76201ab DDAG |
3081 | /* |
3082 | * Make sure the receiver can get incoming pages before we send the rest | |
3083 | * of the state | |
3084 | */ | |
3085 | qemu_savevm_send_postcopy_listen(fb); | |
3086 | ||
a1fbe750 | 3087 | qemu_savevm_state_complete_precopy(fb, false, false); |
58110f0a VSO |
3088 | if (migrate_postcopy_ram()) { |
3089 | qemu_savevm_send_ping(fb, 3); | |
3090 | } | |
1d34e4bf DDAG |
3091 | |
3092 | qemu_savevm_send_postcopy_run(fb); | |
3093 | ||
3094 | /* <><> end of stuff going into the package */ | |
1d34e4bf | 3095 | |
ef8d6488 DDAG |
3096 | /* Last point of recovery; as soon as we send the package the destination |
3097 | * can open devices and potentially start running. | |
3098 | * Lets just check again we've not got any errors. | |
3099 | */ | |
3100 | ret = qemu_file_get_error(ms->to_dst_file); | |
3101 | if (ret) { | |
3102 | error_report("postcopy_start: Migration stream errored (pre package)"); | |
3103 | goto fail_closefb; | |
3104 | } | |
3105 | ||
3106 | restart_block = false; | |
3107 | ||
1d34e4bf | 3108 | /* Now send that blob */ |
61b67d47 | 3109 | if (qemu_savevm_send_packaged(ms->to_dst_file, bioc->data, bioc->usage)) { |
1d34e4bf DDAG |
3110 | goto fail_closefb; |
3111 | } | |
3112 | qemu_fclose(fb); | |
b82fc321 DDAG |
3113 | |
3114 | /* Send a notify to give a chance for anything that needs to happen | |
3115 | * at the transition to postcopy and after the device state; in particular | |
3116 | * spice needs to trigger a transition now | |
3117 | */ | |
3118 | ms->postcopy_after_devices = true; | |
3119 | notifier_list_notify(&migration_state_notifiers, ms); | |
3120 | ||
1d34e4bf DDAG |
3121 | ms->downtime = qemu_clock_get_ms(QEMU_CLOCK_REALTIME) - time_at_stop; |
3122 | ||
3123 | qemu_mutex_unlock_iothread(); | |
3124 | ||
58110f0a VSO |
3125 | if (migrate_postcopy_ram()) { |
3126 | /* | |
3127 | * Although this ping is just for debug, it could potentially be | |
3128 | * used for getting a better measurement of downtime at the source. | |
3129 | */ | |
3130 | qemu_savevm_send_ping(ms->to_dst_file, 4); | |
3131 | } | |
1d34e4bf | 3132 | |
ced1c616 PB |
3133 | if (migrate_release_ram()) { |
3134 | ram_postcopy_migrated_memory_release(ms); | |
3135 | } | |
3136 | ||
89a02a9f | 3137 | ret = qemu_file_get_error(ms->to_dst_file); |
1d34e4bf DDAG |
3138 | if (ret) { |
3139 | error_report("postcopy_start: Migration stream errored"); | |
48781e5b | 3140 | migrate_set_state(&ms->state, MIGRATION_STATUS_POSTCOPY_ACTIVE, |
1d34e4bf DDAG |
3141 | MIGRATION_STATUS_FAILED); |
3142 | } | |
3143 | ||
3144 | return ret; | |
3145 | ||
3146 | fail_closefb: | |
3147 | qemu_fclose(fb); | |
3148 | fail: | |
48781e5b | 3149 | migrate_set_state(&ms->state, MIGRATION_STATUS_POSTCOPY_ACTIVE, |
1d34e4bf | 3150 | MIGRATION_STATUS_FAILED); |
ef8d6488 DDAG |
3151 | if (restart_block) { |
3152 | /* A failure happened early enough that we know the destination hasn't | |
3153 | * accessed block devices, so we're safe to recover. | |
3154 | */ | |
3155 | Error *local_err = NULL; | |
3156 | ||
3b717194 | 3157 | bdrv_activate_all(&local_err); |
ef8d6488 DDAG |
3158 | if (local_err) { |
3159 | error_report_err(local_err); | |
3160 | } | |
3161 | } | |
1d34e4bf DDAG |
3162 | qemu_mutex_unlock_iothread(); |
3163 | return -1; | |
3164 | } | |
3165 | ||
e91d8951 DDAG |
3166 | /** |
3167 | * migration_maybe_pause: Pause if required to by | |
3168 | * migrate_pause_before_switchover called with the iothread locked | |
3169 | * Returns: 0 on success | |
3170 | */ | |
0331c8ca DDAG |
3171 | static int migration_maybe_pause(MigrationState *s, |
3172 | int *current_active_state, | |
3173 | int new_state) | |
e91d8951 DDAG |
3174 | { |
3175 | if (!migrate_pause_before_switchover()) { | |
3176 | return 0; | |
3177 | } | |
3178 | ||
3179 | /* Since leaving this state is not atomic with posting the semaphore | |
3180 | * it's possible that someone could have issued multiple migrate_continue | |
3181 | * and the semaphore is incorrectly positive at this point; | |
3182 | * the docs say it's undefined to reinit a semaphore that's already | |
3183 | * init'd, so use timedwait to eat up any existing posts. | |
3184 | */ | |
3185 | while (qemu_sem_timedwait(&s->pause_sem, 1) == 0) { | |
3186 | /* This block intentionally left blank */ | |
3187 | } | |
3188 | ||
8958338b ZF |
3189 | /* |
3190 | * If the migration is cancelled when it is in the completion phase, | |
3191 | * the migration state is set to MIGRATION_STATUS_CANCELLING. | |
3192 | * So we don't need to wait a semaphore, otherwise we would always | |
3193 | * wait for the 'pause_sem' semaphore. | |
3194 | */ | |
3195 | if (s->state != MIGRATION_STATUS_CANCELLING) { | |
3196 | qemu_mutex_unlock_iothread(); | |
3197 | migrate_set_state(&s->state, *current_active_state, | |
3198 | MIGRATION_STATUS_PRE_SWITCHOVER); | |
3199 | qemu_sem_wait(&s->pause_sem); | |
3200 | migrate_set_state(&s->state, MIGRATION_STATUS_PRE_SWITCHOVER, | |
3201 | new_state); | |
3202 | *current_active_state = new_state; | |
3203 | qemu_mutex_lock_iothread(); | |
3204 | } | |
e91d8951 | 3205 | |
0331c8ca | 3206 | return s->state == new_state ? 0 : -EINVAL; |
e91d8951 DDAG |
3207 | } |
3208 | ||
09f6c85e DDAG |
3209 | /** |
3210 | * migration_completion: Used by migration_thread when there's not much left. | |
3211 | * The caller 'breaks' the loop when this returns. | |
3212 | * | |
3213 | * @s: Current migration state | |
09f6c85e | 3214 | */ |
2ad87305 | 3215 | static void migration_completion(MigrationState *s) |
09f6c85e DDAG |
3216 | { |
3217 | int ret; | |
2ad87305 | 3218 | int current_active_state = s->state; |
09f6c85e | 3219 | |
b10ac0c4 DDAG |
3220 | if (s->state == MIGRATION_STATUS_ACTIVE) { |
3221 | qemu_mutex_lock_iothread(); | |
64909f97 | 3222 | s->downtime_start = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); |
fb064112 | 3223 | qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER, NULL); |
7287cbd4 | 3224 | s->vm_was_running = runstate_is_running(); |
b10ac0c4 DDAG |
3225 | ret = global_state_store(); |
3226 | ||
3227 | if (!ret) { | |
a1fbe750 | 3228 | bool inactivate = !migrate_colo_enabled(); |
b10ac0c4 | 3229 | ret = vm_stop_force_state(RUN_STATE_FINISH_MIGRATE); |
626ff651 | 3230 | trace_migration_completion_vm_stop(ret); |
e91d8951 | 3231 | if (ret >= 0) { |
0331c8ca DDAG |
3232 | ret = migration_maybe_pause(s, ¤t_active_state, |
3233 | MIGRATION_STATUS_DEVICE); | |
e91d8951 | 3234 | } |
f07fa4cb KW |
3235 | if (ret >= 0) { |
3236 | qemu_file_set_rate_limit(s->to_dst_file, INT64_MAX); | |
a1fbe750 FZ |
3237 | ret = qemu_savevm_state_complete_precopy(s->to_dst_file, false, |
3238 | inactivate); | |
f07fa4cb | 3239 | } |
a1fbe750 FZ |
3240 | if (inactivate && ret >= 0) { |
3241 | s->block_inactive = true; | |
b10ac0c4 DDAG |
3242 | } |
3243 | } | |
3244 | qemu_mutex_unlock_iothread(); | |
09f6c85e | 3245 | |
b10ac0c4 DDAG |
3246 | if (ret < 0) { |
3247 | goto fail; | |
09f6c85e | 3248 | } |
b10ac0c4 DDAG |
3249 | } else if (s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) { |
3250 | trace_migration_completion_postcopy_end(); | |
3251 | ||
68b88468 | 3252 | qemu_mutex_lock_iothread(); |
89a02a9f | 3253 | qemu_savevm_state_complete_postcopy(s->to_dst_file); |
68b88468 EGE |
3254 | qemu_mutex_unlock_iothread(); |
3255 | ||
b10ac0c4 | 3256 | trace_migration_completion_postcopy_end_after_complete(); |
444252b9 | 3257 | } else { |
6ba11211 | 3258 | goto fail; |
09f6c85e | 3259 | } |
09f6c85e | 3260 | |
b10ac0c4 DDAG |
3261 | /* |
3262 | * If rp was opened we must clean up the thread before | |
3263 | * cleaning everything else up (since if there are no failures | |
3264 | * it will wait for the destination to send it's status in | |
3265 | * a SHUT command). | |
b10ac0c4 | 3266 | */ |
53021ea1 | 3267 | if (s->rp_state.rp_thread_created) { |
b10ac0c4 | 3268 | int rp_error; |
0425dc97 | 3269 | trace_migration_return_path_end_before(); |
b10ac0c4 | 3270 | rp_error = await_return_path_close_on_source(s); |
0425dc97 | 3271 | trace_migration_return_path_end_after(rp_error); |
b10ac0c4 | 3272 | if (rp_error) { |
fe904ea8 | 3273 | goto fail_invalidate; |
b10ac0c4 | 3274 | } |
09f6c85e DDAG |
3275 | } |
3276 | ||
89a02a9f | 3277 | if (qemu_file_get_error(s->to_dst_file)) { |
09f6c85e | 3278 | trace_migration_completion_file_err(); |
fe904ea8 | 3279 | goto fail_invalidate; |
09f6c85e DDAG |
3280 | } |
3281 | ||
eeeb48ee ZC |
3282 | if (migrate_colo_enabled() && s->state == MIGRATION_STATUS_ACTIVE) { |
3283 | /* COLO does not support postcopy */ | |
3284 | migrate_set_state(&s->state, MIGRATION_STATUS_ACTIVE, | |
3285 | MIGRATION_STATUS_COLO); | |
3286 | } else { | |
0b827d5e HZ |
3287 | migrate_set_state(&s->state, current_active_state, |
3288 | MIGRATION_STATUS_COMPLETED); | |
3289 | } | |
3290 | ||
09f6c85e DDAG |
3291 | return; |
3292 | ||
fe904ea8 GK |
3293 | fail_invalidate: |
3294 | /* If not doing postcopy, vm_start() will be called: let's regain | |
3295 | * control on images. | |
3296 | */ | |
6039dd5b DDAG |
3297 | if (s->state == MIGRATION_STATUS_ACTIVE || |
3298 | s->state == MIGRATION_STATUS_DEVICE) { | |
fe904ea8 GK |
3299 | Error *local_err = NULL; |
3300 | ||
1d2acc31 | 3301 | qemu_mutex_lock_iothread(); |
3b717194 | 3302 | bdrv_activate_all(&local_err); |
fe904ea8 GK |
3303 | if (local_err) { |
3304 | error_report_err(local_err); | |
1d2acc31 HZ |
3305 | } else { |
3306 | s->block_inactive = false; | |
fe904ea8 | 3307 | } |
1d2acc31 | 3308 | qemu_mutex_unlock_iothread(); |
fe904ea8 GK |
3309 | } |
3310 | ||
09f6c85e | 3311 | fail: |
48781e5b HZ |
3312 | migrate_set_state(&s->state, current_active_state, |
3313 | MIGRATION_STATUS_FAILED); | |
09f6c85e DDAG |
3314 | } |
3315 | ||
8518278a AG |
3316 | /** |
3317 | * bg_migration_completion: Used by bg_migration_thread when after all the | |
3318 | * RAM has been saved. The caller 'breaks' the loop when this returns. | |
3319 | * | |
3320 | * @s: Current migration state | |
3321 | */ | |
3322 | static void bg_migration_completion(MigrationState *s) | |
3323 | { | |
3324 | int current_active_state = s->state; | |
3325 | ||
3326 | /* | |
3327 | * Stop tracking RAM writes - un-protect memory, un-register UFFD | |
3328 | * memory ranges, flush kernel wait queues and wake up threads | |
3329 | * waiting for write fault to be resolved. | |
3330 | */ | |
3331 | ram_write_tracking_stop(); | |
3332 | ||
3333 | if (s->state == MIGRATION_STATUS_ACTIVE) { | |
3334 | /* | |
3335 | * By this moment we have RAM content saved into the migration stream. | |
3336 | * The next step is to flush the non-RAM content (device state) | |
3337 | * right after the ram content. The device state has been stored into | |
3338 | * the temporary buffer before RAM saving started. | |
3339 | */ | |
3340 | qemu_put_buffer(s->to_dst_file, s->bioc->data, s->bioc->usage); | |
3341 | qemu_fflush(s->to_dst_file); | |
3342 | } else if (s->state == MIGRATION_STATUS_CANCELLING) { | |
3343 | goto fail; | |
3344 | } | |
3345 | ||
3346 | if (qemu_file_get_error(s->to_dst_file)) { | |
3347 | trace_migration_completion_file_err(); | |
3348 | goto fail; | |
3349 | } | |
3350 | ||
3351 | migrate_set_state(&s->state, current_active_state, | |
3352 | MIGRATION_STATUS_COMPLETED); | |
3353 | return; | |
3354 | ||
3355 | fail: | |
3356 | migrate_set_state(&s->state, current_active_state, | |
3357 | MIGRATION_STATUS_FAILED); | |
3358 | } | |
3359 | ||
35a6ed4f HZ |
3360 | bool migrate_colo_enabled(void) |
3361 | { | |
3362 | MigrationState *s = migrate_get_current(); | |
3363 | return s->enabled_capabilities[MIGRATION_CAPABILITY_X_COLO]; | |
3364 | } | |
3365 | ||
b23c2ade PX |
3366 | typedef enum MigThrError { |
3367 | /* No error detected */ | |
3368 | MIG_THR_ERR_NONE = 0, | |
3369 | /* Detected error, but resumed successfully */ | |
3370 | MIG_THR_ERR_RECOVERED = 1, | |
3371 | /* Detected fatal error, need to exit */ | |
3372 | MIG_THR_ERR_FATAL = 2, | |
3373 | } MigThrError; | |
3374 | ||
94190696 PX |
3375 | static int postcopy_resume_handshake(MigrationState *s) |
3376 | { | |
3377 | qemu_savevm_send_postcopy_resume(s->to_dst_file); | |
3378 | ||
3379 | while (s->state == MIGRATION_STATUS_POSTCOPY_RECOVER) { | |
3380 | qemu_sem_wait(&s->rp_state.rp_sem); | |
3381 | } | |
3382 | ||
3383 | if (s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) { | |
3384 | return 0; | |
3385 | } | |
3386 | ||
3387 | return -1; | |
3388 | } | |
3389 | ||
135b87b4 PX |
3390 | /* Return zero if success, or <0 for error */ |
3391 | static int postcopy_do_resume(MigrationState *s) | |
3392 | { | |
d1b8eadb PX |
3393 | int ret; |
3394 | ||
3395 | /* | |
3396 | * Call all the resume_prepare() hooks, so that modules can be | |
3397 | * ready for the migration resume. | |
3398 | */ | |
3399 | ret = qemu_savevm_state_resume_prepare(s); | |
3400 | if (ret) { | |
3401 | error_report("%s: resume_prepare() failure detected: %d", | |
3402 | __func__, ret); | |
3403 | return ret; | |
3404 | } | |
3405 | ||
3406 | /* | |
94190696 PX |
3407 | * Last handshake with destination on the resume (destination will |
3408 | * switch to postcopy-active afterwards) | |
d1b8eadb | 3409 | */ |
94190696 PX |
3410 | ret = postcopy_resume_handshake(s); |
3411 | if (ret) { | |
3412 | error_report("%s: handshake failed: %d", __func__, ret); | |
3413 | return ret; | |
3414 | } | |
d1b8eadb | 3415 | |
135b87b4 PX |
3416 | return 0; |
3417 | } | |
3418 | ||
b23c2ade PX |
3419 | /* |
3420 | * We don't return until we are in a safe state to continue current | |
3421 | * postcopy migration. Returns MIG_THR_ERR_RECOVERED if recovered, or | |
3422 | * MIG_THR_ERR_FATAL if unrecovery failure happened. | |
3423 | */ | |
3424 | static MigThrError postcopy_pause(MigrationState *s) | |
3425 | { | |
3426 | assert(s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE); | |
b23c2ade | 3427 | |
135b87b4 | 3428 | while (true) { |
62df066f PX |
3429 | QEMUFile *file; |
3430 | ||
39675fff PX |
3431 | /* |
3432 | * Current channel is possibly broken. Release it. Note that this is | |
3433 | * guaranteed even without lock because to_dst_file should only be | |
3434 | * modified by the migration thread. That also guarantees that the | |
3435 | * unregister of yank is safe too without the lock. It should be safe | |
3436 | * even to be within the qemu_file_lock, but we didn't do that to avoid | |
3437 | * taking more mutex (yank_lock) within qemu_file_lock. TL;DR: we make | |
3438 | * the qemu_file_lock critical section as small as possible. | |
3439 | */ | |
135b87b4 | 3440 | assert(s->to_dst_file); |
39675fff | 3441 | migration_ioc_unregister_yank_from_file(s->to_dst_file); |
62df066f PX |
3442 | qemu_mutex_lock(&s->qemu_file_lock); |
3443 | file = s->to_dst_file; | |
135b87b4 | 3444 | s->to_dst_file = NULL; |
62df066f PX |
3445 | qemu_mutex_unlock(&s->qemu_file_lock); |
3446 | ||
3447 | qemu_file_shutdown(file); | |
3448 | qemu_fclose(file); | |
b23c2ade | 3449 | |
d246ea50 PX |
3450 | migrate_set_state(&s->state, s->state, |
3451 | MIGRATION_STATUS_POSTCOPY_PAUSED); | |
3452 | ||
135b87b4 PX |
3453 | error_report("Detected IO failure for postcopy. " |
3454 | "Migration paused."); | |
b23c2ade | 3455 | |
135b87b4 PX |
3456 | /* |
3457 | * We wait until things fixed up. Then someone will setup the | |
3458 | * status back for us. | |
3459 | */ | |
3460 | while (s->state == MIGRATION_STATUS_POSTCOPY_PAUSED) { | |
3461 | qemu_sem_wait(&s->postcopy_pause_sem); | |
3462 | } | |
3463 | ||
3464 | if (s->state == MIGRATION_STATUS_POSTCOPY_RECOVER) { | |
3465 | /* Woken up by a recover procedure. Give it a shot */ | |
b23c2ade | 3466 | |
135b87b4 PX |
3467 | /* |
3468 | * Firstly, let's wake up the return path now, with a new | |
3469 | * return path channel. | |
3470 | */ | |
3471 | qemu_sem_post(&s->postcopy_pause_rp_sem); | |
3472 | ||
3473 | /* Do the resume logic */ | |
3474 | if (postcopy_do_resume(s) == 0) { | |
3475 | /* Let's continue! */ | |
3476 | trace_postcopy_pause_continued(); | |
3477 | return MIG_THR_ERR_RECOVERED; | |
3478 | } else { | |
3479 | /* | |
3480 | * Something wrong happened during the recovery, let's | |
3481 | * pause again. Pause is always better than throwing | |
3482 | * data away. | |
3483 | */ | |
3484 | continue; | |
3485 | } | |
3486 | } else { | |
3487 | /* This is not right... Time to quit. */ | |
3488 | return MIG_THR_ERR_FATAL; | |
3489 | } | |
3490 | } | |
b23c2ade PX |
3491 | } |
3492 | ||
3493 | static MigThrError migration_detect_error(MigrationState *s) | |
3494 | { | |
3495 | int ret; | |
c3c5eae6 | 3496 | int state = s->state; |
3d661c8a | 3497 | Error *local_error = NULL; |
c3c5eae6 DDAG |
3498 | |
3499 | if (state == MIGRATION_STATUS_CANCELLING || | |
3500 | state == MIGRATION_STATUS_CANCELLED) { | |
3501 | /* End the migration, but don't set the state to failed */ | |
3502 | return MIG_THR_ERR_FATAL; | |
3503 | } | |
b23c2ade PX |
3504 | |
3505 | /* Try to detect any file errors */ | |
3d661c8a | 3506 | ret = qemu_file_get_error_obj(s->to_dst_file, &local_error); |
b23c2ade PX |
3507 | if (!ret) { |
3508 | /* Everything is fine */ | |
3d661c8a | 3509 | assert(!local_error); |
b23c2ade PX |
3510 | return MIG_THR_ERR_NONE; |
3511 | } | |
3512 | ||
3d661c8a YK |
3513 | if (local_error) { |
3514 | migrate_set_error(s, local_error); | |
3515 | error_free(local_error); | |
3516 | } | |
3517 | ||
d5c8f2af | 3518 | if (state == MIGRATION_STATUS_POSTCOPY_ACTIVE && ret) { |
b23c2ade PX |
3519 | /* |
3520 | * For postcopy, we allow the network to be down for a | |
3521 | * while. After that, it can be continued by a | |
3522 | * recovery phase. | |
3523 | */ | |
3524 | return postcopy_pause(s); | |
3525 | } else { | |
3526 | /* | |
3527 | * For precopy (or postcopy with error outside IO), we fail | |
3528 | * with no time. | |
3529 | */ | |
c3c5eae6 | 3530 | migrate_set_state(&s->state, state, MIGRATION_STATUS_FAILED); |
b23c2ade PX |
3531 | trace_migration_thread_file_err(); |
3532 | ||
3533 | /* Time to stop the migration, now. */ | |
3534 | return MIG_THR_ERR_FATAL; | |
3535 | } | |
3536 | } | |
3537 | ||
268dcd46 | 3538 | /* How many bytes have we transferred since the beginning of the migration */ |
0c8f0efd JQ |
3539 | static uint64_t migration_total_bytes(MigrationState *s) |
3540 | { | |
fbfa6404 DB |
3541 | return qemu_file_total_transferred(s->to_dst_file) + |
3542 | ram_counters.multifd_bytes; | |
0c8f0efd JQ |
3543 | } |
3544 | ||
cf011f08 PX |
3545 | static void migration_calculate_complete(MigrationState *s) |
3546 | { | |
0c8f0efd | 3547 | uint64_t bytes = migration_total_bytes(s); |
cf011f08 | 3548 | int64_t end_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); |
6cde6fbe | 3549 | int64_t transfer_time; |
cf011f08 PX |
3550 | |
3551 | s->total_time = end_time - s->start_time; | |
3552 | if (!s->downtime) { | |
3553 | /* | |
3554 | * It's still not set, so we are precopy migration. For | |
3555 | * postcopy, downtime is calculated during postcopy_start(). | |
3556 | */ | |
3557 | s->downtime = end_time - s->downtime_start; | |
3558 | } | |
3559 | ||
6cde6fbe JQ |
3560 | transfer_time = s->total_time - s->setup_time; |
3561 | if (transfer_time) { | |
3562 | s->mbps = ((double) bytes * 8.0) / transfer_time / 1000; | |
cf011f08 PX |
3563 | } |
3564 | } | |
3565 | ||
87f3bd87 IR |
3566 | static void update_iteration_initial_status(MigrationState *s) |
3567 | { | |
3568 | /* | |
3569 | * Update these three fields at the same time to avoid mismatch info lead | |
3570 | * wrong speed calculation. | |
3571 | */ | |
3572 | s->iteration_start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); | |
3573 | s->iteration_initial_bytes = migration_total_bytes(s); | |
3574 | s->iteration_initial_pages = ram_get_total_transferred_pages(); | |
3575 | } | |
3576 | ||
b15df1ae PX |
3577 | static void migration_update_counters(MigrationState *s, |
3578 | int64_t current_time) | |
3579 | { | |
aecbfe9c | 3580 | uint64_t transferred, transferred_pages, time_spent; |
0c8f0efd | 3581 | uint64_t current_bytes; /* bytes transferred since the beginning */ |
b15df1ae PX |
3582 | double bandwidth; |
3583 | ||
3584 | if (current_time < s->iteration_start_time + BUFFER_DELAY) { | |
3585 | return; | |
3586 | } | |
3587 | ||
0c8f0efd JQ |
3588 | current_bytes = migration_total_bytes(s); |
3589 | transferred = current_bytes - s->iteration_initial_bytes; | |
b15df1ae PX |
3590 | time_spent = current_time - s->iteration_start_time; |
3591 | bandwidth = (double)transferred / time_spent; | |
0781c1ed | 3592 | s->threshold_size = bandwidth * s->parameters.downtime_limit; |
b15df1ae PX |
3593 | |
3594 | s->mbps = (((double) transferred * 8.0) / | |
3595 | ((double) time_spent / 1000.0)) / 1000.0 / 1000.0; | |
3596 | ||
aecbfe9c XG |
3597 | transferred_pages = ram_get_total_transferred_pages() - |
3598 | s->iteration_initial_pages; | |
3599 | s->pages_per_second = (double) transferred_pages / | |
3600 | (((double) time_spent / 1000.0)); | |
3601 | ||
b15df1ae PX |
3602 | /* |
3603 | * if we haven't sent anything, we don't want to | |
3604 | * recalculate. 10000 is a small enough number for our purposes | |
3605 | */ | |
3606 | if (ram_counters.dirty_pages_rate && transferred > 10000) { | |
650af890 | 3607 | s->expected_downtime = ram_counters.remaining / bandwidth; |
b15df1ae PX |
3608 | } |
3609 | ||
3610 | qemu_file_reset_rate_limit(s->to_dst_file); | |
3611 | ||
87f3bd87 | 3612 | update_iteration_initial_status(s); |
b15df1ae PX |
3613 | |
3614 | trace_migrate_transferred(transferred, time_spent, | |
0781c1ed | 3615 | bandwidth, s->threshold_size); |
b15df1ae PX |
3616 | } |
3617 | ||
2ad87305 PX |
3618 | /* Migration thread iteration status */ |
3619 | typedef enum { | |
3620 | MIG_ITERATE_RESUME, /* Resume current iteration */ | |
3621 | MIG_ITERATE_SKIP, /* Skip current iteration */ | |
3622 | MIG_ITERATE_BREAK, /* Break the loop */ | |
3623 | } MigIterateState; | |
3624 | ||
3625 | /* | |
3626 | * Return true if continue to the next iteration directly, false | |
3627 | * otherwise. | |
3628 | */ | |
3629 | static MigIterateState migration_iteration_run(MigrationState *s) | |
3630 | { | |
47995026 | 3631 | uint64_t pending_size, pend_pre, pend_compat, pend_post; |
2ad87305 PX |
3632 | bool in_postcopy = s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE; |
3633 | ||
47995026 VSO |
3634 | qemu_savevm_state_pending(s->to_dst_file, s->threshold_size, &pend_pre, |
3635 | &pend_compat, &pend_post); | |
3636 | pending_size = pend_pre + pend_compat + pend_post; | |
2ad87305 PX |
3637 | |
3638 | trace_migrate_pending(pending_size, s->threshold_size, | |
47995026 | 3639 | pend_pre, pend_compat, pend_post); |
2ad87305 PX |
3640 | |
3641 | if (pending_size && pending_size >= s->threshold_size) { | |
3642 | /* Still a significant amount to transfer */ | |
52aec709 | 3643 | if (!in_postcopy && pend_pre <= s->threshold_size && |
d73415a3 | 3644 | qatomic_read(&s->start_postcopy)) { |
2ad87305 PX |
3645 | if (postcopy_start(s)) { |
3646 | error_report("%s: postcopy failed to start", __func__); | |
3647 | } | |
3648 | return MIG_ITERATE_SKIP; | |
3649 | } | |
3650 | /* Just another iteration step */ | |
17d9351b | 3651 | qemu_savevm_state_iterate(s->to_dst_file, in_postcopy); |
2ad87305 PX |
3652 | } else { |
3653 | trace_migration_thread_low_pending(pending_size); | |
3654 | migration_completion(s); | |
3655 | return MIG_ITERATE_BREAK; | |
3656 | } | |
3657 | ||
3658 | return MIG_ITERATE_RESUME; | |
3659 | } | |
3660 | ||
199aa6d4 PX |
3661 | static void migration_iteration_finish(MigrationState *s) |
3662 | { | |
3663 | /* If we enabled cpu throttling for auto-converge, turn it off. */ | |
3664 | cpu_throttle_stop(); | |
3665 | ||
3666 | qemu_mutex_lock_iothread(); | |
3667 | switch (s->state) { | |
3668 | case MIGRATION_STATUS_COMPLETED: | |
3669 | migration_calculate_complete(s); | |
3670 | runstate_set(RUN_STATE_POSTMIGRATE); | |
3671 | break; | |
751fe4c6 | 3672 | case MIGRATION_STATUS_COLO: |
199aa6d4 PX |
3673 | if (!migrate_colo_enabled()) { |
3674 | error_report("%s: critical error: calling COLO code without " | |
3675 | "COLO enabled", __func__); | |
3676 | } | |
3677 | migrate_start_colo_process(s); | |
199aa6d4 PX |
3678 | s->vm_was_running = true; |
3679 | /* Fallthrough */ | |
3680 | case MIGRATION_STATUS_FAILED: | |
3681 | case MIGRATION_STATUS_CANCELLED: | |
57225e5f | 3682 | case MIGRATION_STATUS_CANCELLING: |
199aa6d4 | 3683 | if (s->vm_was_running) { |
aa505f8e RL |
3684 | if (!runstate_check(RUN_STATE_SHUTDOWN)) { |
3685 | vm_start(); | |
3686 | } | |
199aa6d4 PX |
3687 | } else { |
3688 | if (runstate_check(RUN_STATE_FINISH_MIGRATE)) { | |
3689 | runstate_set(RUN_STATE_POSTMIGRATE); | |
3690 | } | |
3691 | } | |
3692 | break; | |
3693 | ||
3694 | default: | |
3695 | /* Should not reach here, but if so, forgive the VM. */ | |
3696 | error_report("%s: Unknown ending state %d", __func__, s->state); | |
3697 | break; | |
3698 | } | |
fd392cfa | 3699 | migrate_fd_cleanup_schedule(s); |
199aa6d4 PX |
3700 | qemu_mutex_unlock_iothread(); |
3701 | } | |
3702 | ||
8518278a AG |
3703 | static void bg_migration_iteration_finish(MigrationState *s) |
3704 | { | |
3705 | qemu_mutex_lock_iothread(); | |
3706 | switch (s->state) { | |
3707 | case MIGRATION_STATUS_COMPLETED: | |
3708 | migration_calculate_complete(s); | |
3709 | break; | |
3710 | ||
3711 | case MIGRATION_STATUS_ACTIVE: | |
3712 | case MIGRATION_STATUS_FAILED: | |
3713 | case MIGRATION_STATUS_CANCELLED: | |
3714 | case MIGRATION_STATUS_CANCELLING: | |
3715 | break; | |
3716 | ||
3717 | default: | |
3718 | /* Should not reach here, but if so, forgive the VM. */ | |
3719 | error_report("%s: Unknown ending state %d", __func__, s->state); | |
3720 | break; | |
3721 | } | |
3722 | ||
3723 | migrate_fd_cleanup_schedule(s); | |
3724 | qemu_mutex_unlock_iothread(); | |
3725 | } | |
3726 | ||
3727 | /* | |
3728 | * Return true if continue to the next iteration directly, false | |
3729 | * otherwise. | |
3730 | */ | |
3731 | static MigIterateState bg_migration_iteration_run(MigrationState *s) | |
3732 | { | |
3733 | int res; | |
3734 | ||
3735 | res = qemu_savevm_state_iterate(s->to_dst_file, false); | |
3736 | if (res > 0) { | |
3737 | bg_migration_completion(s); | |
3738 | return MIG_ITERATE_BREAK; | |
3739 | } | |
3740 | ||
3741 | return MIG_ITERATE_RESUME; | |
3742 | } | |
3743 | ||
ad767bed DDAG |
3744 | void migration_make_urgent_request(void) |
3745 | { | |
3746 | qemu_sem_post(&migrate_get_current()->rate_limit_sem); | |
3747 | } | |
3748 | ||
3749 | void migration_consume_urgent_request(void) | |
3750 | { | |
3751 | qemu_sem_wait(&migrate_get_current()->rate_limit_sem); | |
3752 | } | |
3753 | ||
97e1e067 DDAG |
3754 | /* Returns true if the rate limiting was broken by an urgent request */ |
3755 | bool migration_rate_limit(void) | |
3756 | { | |
3757 | int64_t now = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); | |
3758 | MigrationState *s = migrate_get_current(); | |
3759 | ||
3760 | bool urgent = false; | |
3761 | migration_update_counters(s, now); | |
3762 | if (qemu_file_rate_limit(s->to_dst_file)) { | |
77386127 LS |
3763 | |
3764 | if (qemu_file_get_error(s->to_dst_file)) { | |
3765 | return false; | |
3766 | } | |
97e1e067 DDAG |
3767 | /* |
3768 | * Wait for a delay to do rate limiting OR | |
3769 | * something urgent to post the semaphore. | |
3770 | */ | |
3771 | int ms = s->iteration_start_time + BUFFER_DELAY - now; | |
3772 | trace_migration_rate_limit_pre(ms); | |
3773 | if (qemu_sem_timedwait(&s->rate_limit_sem, ms) == 0) { | |
3774 | /* | |
3775 | * We were woken by one or more urgent things but | |
3776 | * the timedwait will have consumed one of them. | |
3777 | * The service routine for the urgent wake will dec | |
3778 | * the semaphore itself for each item it consumes, | |
3779 | * so add this one we just eat back. | |
3780 | */ | |
3781 | qemu_sem_post(&s->rate_limit_sem); | |
3782 | urgent = true; | |
3783 | } | |
3784 | trace_migration_rate_limit_post(urgent); | |
3785 | } | |
3786 | return urgent; | |
3787 | } | |
3788 | ||
fde93d99 LV |
3789 | /* |
3790 | * if failover devices are present, wait they are completely | |
3791 | * unplugged | |
3792 | */ | |
3793 | ||
3794 | static void qemu_savevm_wait_unplug(MigrationState *s, int old_state, | |
3795 | int new_state) | |
3796 | { | |
3797 | if (qemu_savevm_state_guest_unplug_pending()) { | |
3798 | migrate_set_state(&s->state, old_state, MIGRATION_STATUS_WAIT_UNPLUG); | |
3799 | ||
3800 | while (s->state == MIGRATION_STATUS_WAIT_UNPLUG && | |
3801 | qemu_savevm_state_guest_unplug_pending()) { | |
3802 | qemu_sem_timedwait(&s->wait_unplug_sem, 250); | |
3803 | } | |
944bc528 LV |
3804 | if (s->state != MIGRATION_STATUS_WAIT_UNPLUG) { |
3805 | int timeout = 120; /* 30 seconds */ | |
3806 | /* | |
3807 | * migration has been canceled | |
3808 | * but as we have started an unplug we must wait the end | |
3809 | * to be able to plug back the card | |
3810 | */ | |
3811 | while (timeout-- && qemu_savevm_state_guest_unplug_pending()) { | |
3812 | qemu_sem_timedwait(&s->wait_unplug_sem, 250); | |
3813 | } | |
1b529d90 LV |
3814 | if (qemu_savevm_state_guest_unplug_pending() && |
3815 | !qtest_enabled()) { | |
a51dcef0 LV |
3816 | warn_report("migration: partially unplugged device on " |
3817 | "failure"); | |
3818 | } | |
944bc528 | 3819 | } |
fde93d99 LV |
3820 | |
3821 | migrate_set_state(&s->state, MIGRATION_STATUS_WAIT_UNPLUG, new_state); | |
3822 | } else { | |
3823 | migrate_set_state(&s->state, old_state, new_state); | |
3824 | } | |
3825 | } | |
3826 | ||
70b20477 DDAG |
3827 | /* |
3828 | * Master migration thread on the source VM. | |
3829 | * It drives the migration and pumps the data down the outgoing channel. | |
3830 | */ | |
5f496a1b | 3831 | static void *migration_thread(void *opaque) |
0d82d0e8 | 3832 | { |
9848a404 | 3833 | MigrationState *s = opaque; |
bc72ad67 | 3834 | int64_t setup_start = qemu_clock_get_ms(QEMU_CLOCK_HOST); |
b23c2ade | 3835 | MigThrError thr_error; |
ad767bed | 3836 | bool urgent = false; |
76f5933a | 3837 | |
ab28bd23 PB |
3838 | rcu_register_thread(); |
3839 | ||
892ae715 | 3840 | object_ref(OBJECT(s)); |
87f3bd87 | 3841 | update_iteration_initial_status(s); |
b15df1ae | 3842 | |
89a02a9f | 3843 | qemu_savevm_state_header(s->to_dst_file); |
1d34e4bf | 3844 | |
62a02658 PX |
3845 | /* |
3846 | * If we opened the return path, we need to make sure dst has it | |
3847 | * opened as well. | |
3848 | */ | |
43044ac0 | 3849 | if (s->rp_state.rp_thread_created) { |
1d34e4bf | 3850 | /* Now tell the dest that it should open its end so it can reply */ |
89a02a9f | 3851 | qemu_savevm_send_open_return_path(s->to_dst_file); |
1d34e4bf DDAG |
3852 | |
3853 | /* And do a ping that will make stuff easier to debug */ | |
89a02a9f | 3854 | qemu_savevm_send_ping(s->to_dst_file, 1); |
0425dc97 | 3855 | } |
1d34e4bf | 3856 | |
58110f0a | 3857 | if (migrate_postcopy()) { |
1d34e4bf DDAG |
3858 | /* |
3859 | * Tell the destination that we *might* want to do postcopy later; | |
3860 | * if the other end can't do postcopy it should fail now, nice and | |
3861 | * early. | |
3862 | */ | |
89a02a9f | 3863 | qemu_savevm_send_postcopy_advise(s->to_dst_file); |
1d34e4bf DDAG |
3864 | } |
3865 | ||
aad555c2 ZC |
3866 | if (migrate_colo_enabled()) { |
3867 | /* Notify migration destination that we enable COLO */ | |
3868 | qemu_savevm_send_colo_enable(s->to_dst_file); | |
3869 | } | |
3870 | ||
9907e842 | 3871 | qemu_savevm_state_setup(s->to_dst_file); |
0d82d0e8 | 3872 | |
fde93d99 LV |
3873 | qemu_savevm_wait_unplug(s, MIGRATION_STATUS_SETUP, |
3874 | MIGRATION_STATUS_ACTIVE); | |
c7e0acd5 | 3875 | |
bc72ad67 | 3876 | s->setup_time = qemu_clock_get_ms(QEMU_CLOCK_HOST) - setup_start; |
29ae8a41 | 3877 | |
9ec055ae DDAG |
3878 | trace_migration_thread_setup_complete(); |
3879 | ||
8f8d528e | 3880 | while (migration_is_active(s)) { |
ad767bed | 3881 | if (urgent || !qemu_file_rate_limit(s->to_dst_file)) { |
2ad87305 PX |
3882 | MigIterateState iter_state = migration_iteration_run(s); |
3883 | if (iter_state == MIG_ITERATE_SKIP) { | |
3884 | continue; | |
3885 | } else if (iter_state == MIG_ITERATE_BREAK) { | |
09f6c85e | 3886 | break; |
c369f40d JQ |
3887 | } |
3888 | } | |
f4410a5d | 3889 | |
b23c2ade PX |
3890 | /* |
3891 | * Try to detect any kind of failures, and see whether we | |
3892 | * should stop the migration now. | |
3893 | */ | |
3894 | thr_error = migration_detect_error(s); | |
3895 | if (thr_error == MIG_THR_ERR_FATAL) { | |
3896 | /* Stop migration */ | |
fd45ee2c | 3897 | break; |
b23c2ade PX |
3898 | } else if (thr_error == MIG_THR_ERR_RECOVERED) { |
3899 | /* | |
3900 | * Just recovered from a e.g. network failure, reset all | |
3901 | * the local variables. This is important to avoid | |
3902 | * breaking transferred_bytes and bandwidth calculation | |
3903 | */ | |
87f3bd87 | 3904 | update_iteration_initial_status(s); |
fd45ee2c | 3905 | } |
b15df1ae | 3906 | |
97e1e067 | 3907 | urgent = migration_rate_limit(); |
a3fa1d78 PB |
3908 | } |
3909 | ||
1d34e4bf | 3910 | trace_migration_thread_after_loop(); |
199aa6d4 | 3911 | migration_iteration_finish(s); |
892ae715 | 3912 | object_unref(OBJECT(s)); |
ab28bd23 | 3913 | rcu_unregister_thread(); |
0d82d0e8 JQ |
3914 | return NULL; |
3915 | } | |
3916 | ||
8518278a AG |
3917 | static void bg_migration_vm_start_bh(void *opaque) |
3918 | { | |
3919 | MigrationState *s = opaque; | |
3920 | ||
3921 | qemu_bh_delete(s->vm_start_bh); | |
3922 | s->vm_start_bh = NULL; | |
3923 | ||
3924 | vm_start(); | |
3925 | s->downtime = qemu_clock_get_ms(QEMU_CLOCK_REALTIME) - s->downtime_start; | |
3926 | } | |
3927 | ||
3928 | /** | |
3929 | * Background snapshot thread, based on live migration code. | |
3930 | * This is an alternative implementation of live migration mechanism | |
3931 | * introduced specifically to support background snapshots. | |
3932 | * | |
3933 | * It takes advantage of userfault_fd write protection mechanism introduced | |
3934 | * in v5.7 kernel. Compared to existing dirty page logging migration much | |
3935 | * lesser stream traffic is produced resulting in smaller snapshot images, | |
3936 | * simply cause of no page duplicates can get into the stream. | |
3937 | * | |
3938 | * Another key point is that generated vmstate stream reflects machine state | |
3939 | * 'frozen' at the beginning of snapshot creation compared to dirty page logging | |
3940 | * mechanism, which effectively results in that saved snapshot is the state of VM | |
3941 | * at the end of the process. | |
3942 | */ | |
3943 | static void *bg_migration_thread(void *opaque) | |
3944 | { | |
3945 | MigrationState *s = opaque; | |
3946 | int64_t setup_start; | |
3947 | MigThrError thr_error; | |
3948 | QEMUFile *fb; | |
3949 | bool early_fail = true; | |
3950 | ||
3951 | rcu_register_thread(); | |
3952 | object_ref(OBJECT(s)); | |
3953 | ||
3954 | qemu_file_set_rate_limit(s->to_dst_file, INT64_MAX); | |
3955 | ||
3956 | setup_start = qemu_clock_get_ms(QEMU_CLOCK_HOST); | |
3957 | /* | |
3958 | * We want to save vmstate for the moment when migration has been | |
3959 | * initiated but also we want to save RAM content while VM is running. | |
3960 | * The RAM content should appear first in the vmstate. So, we first | |
3961 | * stash the non-RAM part of the vmstate to the temporary buffer, | |
3962 | * then write RAM part of the vmstate to the migration stream | |
3963 | * with vCPUs running and, finally, write stashed non-RAM part of | |
3964 | * the vmstate from the buffer to the migration stream. | |
3965 | */ | |
ecb23efe | 3966 | s->bioc = qio_channel_buffer_new(512 * 1024); |
8518278a | 3967 | qio_channel_set_name(QIO_CHANNEL(s->bioc), "vmstate-buffer"); |
77ef2dc1 | 3968 | fb = qemu_file_new_output(QIO_CHANNEL(s->bioc)); |
8518278a AG |
3969 | object_unref(OBJECT(s->bioc)); |
3970 | ||
3971 | update_iteration_initial_status(s); | |
3972 | ||
eeccb99c AG |
3973 | /* |
3974 | * Prepare for tracking memory writes with UFFD-WP - populate | |
3975 | * RAM pages before protecting. | |
3976 | */ | |
3977 | #ifdef __linux__ | |
3978 | ram_write_tracking_prepare(); | |
3979 | #endif | |
3980 | ||
8518278a AG |
3981 | qemu_savevm_state_header(s->to_dst_file); |
3982 | qemu_savevm_state_setup(s->to_dst_file); | |
3983 | ||
fde93d99 LV |
3984 | qemu_savevm_wait_unplug(s, MIGRATION_STATUS_SETUP, |
3985 | MIGRATION_STATUS_ACTIVE); | |
8518278a | 3986 | |
8518278a AG |
3987 | s->setup_time = qemu_clock_get_ms(QEMU_CLOCK_HOST) - setup_start; |
3988 | ||
3989 | trace_migration_thread_setup_complete(); | |
3990 | s->downtime_start = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); | |
3991 | ||
3992 | qemu_mutex_lock_iothread(); | |
3993 | ||
3994 | /* | |
3995 | * If VM is currently in suspended state, then, to make a valid runstate | |
3996 | * transition in vm_stop_force_state() we need to wakeup it up. | |
3997 | */ | |
3998 | qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER, NULL); | |
3999 | s->vm_was_running = runstate_is_running(); | |
4000 | ||
4001 | if (global_state_store()) { | |
4002 | goto fail; | |
4003 | } | |
4004 | /* Forcibly stop VM before saving state of vCPUs and devices */ | |
4005 | if (vm_stop_force_state(RUN_STATE_PAUSED)) { | |
4006 | goto fail; | |
4007 | } | |
4008 | /* | |
4009 | * Put vCPUs in sync with shadow context structures, then | |
4010 | * save their state to channel-buffer along with devices. | |
4011 | */ | |
4012 | cpu_synchronize_all_states(); | |
4013 | if (qemu_savevm_state_complete_precopy_non_iterable(fb, false, false)) { | |
4014 | goto fail; | |
4015 | } | |
ecb23efe AG |
4016 | /* |
4017 | * Since we are going to get non-iterable state data directly | |
4018 | * from s->bioc->data, explicit flush is needed here. | |
4019 | */ | |
4020 | qemu_fflush(fb); | |
4021 | ||
8518278a AG |
4022 | /* Now initialize UFFD context and start tracking RAM writes */ |
4023 | if (ram_write_tracking_start()) { | |
4024 | goto fail; | |
4025 | } | |
4026 | early_fail = false; | |
4027 | ||
4028 | /* | |
4029 | * Start VM from BH handler to avoid write-fault lock here. | |
4030 | * UFFD-WP protection for the whole RAM is already enabled so | |
4031 | * calling VM state change notifiers from vm_start() would initiate | |
4032 | * writes to virtio VQs memory which is in write-protected region. | |
4033 | */ | |
4034 | s->vm_start_bh = qemu_bh_new(bg_migration_vm_start_bh, s); | |
4035 | qemu_bh_schedule(s->vm_start_bh); | |
4036 | ||
4037 | qemu_mutex_unlock_iothread(); | |
4038 | ||
4039 | while (migration_is_active(s)) { | |
4040 | MigIterateState iter_state = bg_migration_iteration_run(s); | |
4041 | if (iter_state == MIG_ITERATE_SKIP) { | |
4042 | continue; | |
4043 | } else if (iter_state == MIG_ITERATE_BREAK) { | |
4044 | break; | |
4045 | } | |
4046 | ||
4047 | /* | |
4048 | * Try to detect any kind of failures, and see whether we | |
4049 | * should stop the migration now. | |
4050 | */ | |
4051 | thr_error = migration_detect_error(s); | |
4052 | if (thr_error == MIG_THR_ERR_FATAL) { | |
4053 | /* Stop migration */ | |
4054 | break; | |
4055 | } | |
4056 | ||
4057 | migration_update_counters(s, qemu_clock_get_ms(QEMU_CLOCK_REALTIME)); | |
4058 | } | |
4059 | ||
4060 | trace_migration_thread_after_loop(); | |
4061 | ||
4062 | fail: | |
4063 | if (early_fail) { | |
4064 | migrate_set_state(&s->state, MIGRATION_STATUS_ACTIVE, | |
4065 | MIGRATION_STATUS_FAILED); | |
4066 | qemu_mutex_unlock_iothread(); | |
4067 | } | |
4068 | ||
4069 | bg_migration_iteration_finish(s); | |
4070 | ||
4071 | qemu_fclose(fb); | |
4072 | object_unref(OBJECT(s)); | |
4073 | rcu_unregister_thread(); | |
4074 | ||
4075 | return NULL; | |
4076 | } | |
4077 | ||
cce8040b | 4078 | void migrate_fd_connect(MigrationState *s, Error *error_in) |
0d82d0e8 | 4079 | { |
00f4b572 | 4080 | Error *local_err = NULL; |
d3e35b8f PX |
4081 | int64_t rate_limit; |
4082 | bool resume = s->state == MIGRATION_STATUS_POSTCOPY_PAUSED; | |
4083 | ||
ca7bd082 PX |
4084 | /* |
4085 | * If there's a previous error, free it and prepare for another one. | |
4086 | * Meanwhile if migration completes successfully, there won't have an error | |
4087 | * dumped when calling migrate_fd_cleanup(). | |
4088 | */ | |
4089 | migrate_error_free(s); | |
4090 | ||
2ff30257 | 4091 | s->expected_downtime = s->parameters.downtime_limit; |
9cbc3649 MAL |
4092 | if (resume) { |
4093 | assert(s->cleanup_bh); | |
4094 | } else { | |
4095 | assert(!s->cleanup_bh); | |
4096 | s->cleanup_bh = qemu_bh_new(migrate_fd_cleanup_bh, s); | |
4097 | } | |
cce8040b DDAG |
4098 | if (error_in) { |
4099 | migrate_fd_error(s, error_in); | |
ca30f24d PX |
4100 | if (resume) { |
4101 | /* | |
4102 | * Don't do cleanup for resume if channel is invalid, but only dump | |
4103 | * the error. We wait for another channel connect from the user. | |
4104 | * The error_report still gives HMP user a hint on what failed. | |
4105 | * It's normally done in migrate_fd_cleanup(), but call it here | |
4106 | * explicitly. | |
4107 | */ | |
4108 | error_report_err(error_copy(s->error)); | |
4109 | } else { | |
4110 | migrate_fd_cleanup(s); | |
4111 | } | |
cce8040b DDAG |
4112 | return; |
4113 | } | |
0d82d0e8 | 4114 | |
d3e35b8f PX |
4115 | if (resume) { |
4116 | /* This is a resumed migration */ | |
8504ddec PX |
4117 | rate_limit = s->parameters.max_postcopy_bandwidth / |
4118 | XFER_LIMIT_RATIO; | |
d3e35b8f PX |
4119 | } else { |
4120 | /* This is a fresh new migration */ | |
4121 | rate_limit = s->parameters.max_bandwidth / XFER_LIMIT_RATIO; | |
442773ce | 4122 | |
d3e35b8f PX |
4123 | /* Notify before starting migration thread */ |
4124 | notifier_list_notify(&migration_state_notifiers, s); | |
4125 | } | |
4126 | ||
4127 | qemu_file_set_rate_limit(s->to_dst_file, rate_limit); | |
4128 | qemu_file_set_blocking(s->to_dst_file, true); | |
9287ac27 | 4129 | |
1d34e4bf | 4130 | /* |
c788ada8 PX |
4131 | * Open the return path. For postcopy, it is used exclusively. For |
4132 | * precopy, only if user specified "return-path" capability would | |
4133 | * QEMU uses the return path. | |
1d34e4bf | 4134 | */ |
c788ada8 | 4135 | if (migrate_postcopy_ram() || migrate_use_return_path()) { |
d3e35b8f | 4136 | if (open_return_path_on_source(s, !resume)) { |
1d34e4bf | 4137 | error_report("Unable to open return-path for postcopy"); |
d3e35b8f | 4138 | migrate_set_state(&s->state, s->state, MIGRATION_STATUS_FAILED); |
1d34e4bf DDAG |
4139 | migrate_fd_cleanup(s); |
4140 | return; | |
4141 | } | |
4142 | } | |
4143 | ||
d3e35b8f | 4144 | if (resume) { |
135b87b4 PX |
4145 | /* Wakeup the main migration thread to do the recovery */ |
4146 | migrate_set_state(&s->state, MIGRATION_STATUS_POSTCOPY_PAUSED, | |
4147 | MIGRATION_STATUS_POSTCOPY_RECOVER); | |
4148 | qemu_sem_post(&s->postcopy_pause_sem); | |
d3e35b8f PX |
4149 | return; |
4150 | } | |
4151 | ||
00f4b572 JQ |
4152 | if (multifd_save_setup(&local_err) != 0) { |
4153 | error_report_err(local_err); | |
f986c3d2 JQ |
4154 | migrate_set_state(&s->state, MIGRATION_STATUS_SETUP, |
4155 | MIGRATION_STATUS_FAILED); | |
4156 | migrate_fd_cleanup(s); | |
4157 | return; | |
4158 | } | |
8518278a AG |
4159 | |
4160 | if (migrate_background_snapshot()) { | |
4161 | qemu_thread_create(&s->thread, "bg_snapshot", | |
4162 | bg_migration_thread, s, QEMU_THREAD_JOINABLE); | |
4163 | } else { | |
4164 | qemu_thread_create(&s->thread, "live_migration", | |
4165 | migration_thread, s, QEMU_THREAD_JOINABLE); | |
4166 | } | |
1d34e4bf | 4167 | s->migration_thread_running = true; |
0d82d0e8 | 4168 | } |
093e3c42 | 4169 | |
9d18af93 PX |
4170 | void migration_global_dump(Monitor *mon) |
4171 | { | |
4172 | MigrationState *ms = migrate_get_current(); | |
4173 | ||
6f0f6428 JQ |
4174 | monitor_printf(mon, "globals:\n"); |
4175 | monitor_printf(mon, "store-global-state: %s\n", | |
4176 | ms->store_global_state ? "on" : "off"); | |
4177 | monitor_printf(mon, "only-migratable: %s\n", | |
811f8652 | 4178 | only_migratable ? "on" : "off"); |
6f0f6428 JQ |
4179 | monitor_printf(mon, "send-configuration: %s\n", |
4180 | ms->send_configuration ? "on" : "off"); | |
4181 | monitor_printf(mon, "send-section-footer: %s\n", | |
4182 | ms->send_section_footer ? "on" : "off"); | |
f548222c XG |
4183 | monitor_printf(mon, "decompress-error-check: %s\n", |
4184 | ms->decompress_error_check ? "on" : "off"); | |
002cad6b PX |
4185 | monitor_printf(mon, "clear-bitmap-shift: %u\n", |
4186 | ms->clear_bitmap_shift); | |
9d18af93 PX |
4187 | } |
4188 | ||
20814758 PX |
4189 | #define DEFINE_PROP_MIG_CAP(name, x) \ |
4190 | DEFINE_PROP_BOOL(name, MigrationState, enabled_capabilities[x], false) | |
4191 | ||
5272298c PX |
4192 | static Property migration_properties[] = { |
4193 | DEFINE_PROP_BOOL("store-global-state", MigrationState, | |
4194 | store_global_state, true), | |
71dd4c1a PX |
4195 | DEFINE_PROP_BOOL("send-configuration", MigrationState, |
4196 | send_configuration, true), | |
15c38503 PX |
4197 | DEFINE_PROP_BOOL("send-section-footer", MigrationState, |
4198 | send_section_footer, true), | |
f548222c XG |
4199 | DEFINE_PROP_BOOL("decompress-error-check", MigrationState, |
4200 | decompress_error_check, true), | |
002cad6b PX |
4201 | DEFINE_PROP_UINT8("x-clear-bitmap-shift", MigrationState, |
4202 | clear_bitmap_shift, CLEAR_BITMAP_SHIFT_DEFAULT), | |
89632faf PX |
4203 | |
4204 | /* Migration parameters */ | |
741d4086 | 4205 | DEFINE_PROP_UINT8("x-compress-level", MigrationState, |
89632faf PX |
4206 | parameters.compress_level, |
4207 | DEFAULT_MIGRATE_COMPRESS_LEVEL), | |
741d4086 | 4208 | DEFINE_PROP_UINT8("x-compress-threads", MigrationState, |
89632faf PX |
4209 | parameters.compress_threads, |
4210 | DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT), | |
1d58872a XG |
4211 | DEFINE_PROP_BOOL("x-compress-wait-thread", MigrationState, |
4212 | parameters.compress_wait_thread, true), | |
741d4086 | 4213 | DEFINE_PROP_UINT8("x-decompress-threads", MigrationState, |
89632faf PX |
4214 | parameters.decompress_threads, |
4215 | DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT), | |
dc14a470 KZ |
4216 | DEFINE_PROP_UINT8("x-throttle-trigger-threshold", MigrationState, |
4217 | parameters.throttle_trigger_threshold, | |
4218 | DEFAULT_MIGRATE_THROTTLE_TRIGGER_THRESHOLD), | |
741d4086 | 4219 | DEFINE_PROP_UINT8("x-cpu-throttle-initial", MigrationState, |
89632faf PX |
4220 | parameters.cpu_throttle_initial, |
4221 | DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL), | |
741d4086 | 4222 | DEFINE_PROP_UINT8("x-cpu-throttle-increment", MigrationState, |
89632faf PX |
4223 | parameters.cpu_throttle_increment, |
4224 | DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT), | |
cbbf8182 KZ |
4225 | DEFINE_PROP_BOOL("x-cpu-throttle-tailslow", MigrationState, |
4226 | parameters.cpu_throttle_tailslow, false), | |
741d4086 | 4227 | DEFINE_PROP_SIZE("x-max-bandwidth", MigrationState, |
89632faf | 4228 | parameters.max_bandwidth, MAX_THROTTLE), |
741d4086 | 4229 | DEFINE_PROP_UINT64("x-downtime-limit", MigrationState, |
89632faf PX |
4230 | parameters.downtime_limit, |
4231 | DEFAULT_MIGRATE_SET_DOWNTIME), | |
741d4086 | 4232 | DEFINE_PROP_UINT32("x-checkpoint-delay", MigrationState, |
89632faf PX |
4233 | parameters.x_checkpoint_delay, |
4234 | DEFAULT_MIGRATE_X_CHECKPOINT_DELAY), | |
cbfd6c95 JQ |
4235 | DEFINE_PROP_UINT8("multifd-channels", MigrationState, |
4236 | parameters.multifd_channels, | |
4075fb1c | 4237 | DEFAULT_MIGRATE_MULTIFD_CHANNELS), |
96eef042 JQ |
4238 | DEFINE_PROP_MULTIFD_COMPRESSION("multifd-compression", MigrationState, |
4239 | parameters.multifd_compression, | |
4240 | DEFAULT_MIGRATE_MULTIFD_COMPRESSION), | |
9004db48 JQ |
4241 | DEFINE_PROP_UINT8("multifd-zlib-level", MigrationState, |
4242 | parameters.multifd_zlib_level, | |
4243 | DEFAULT_MIGRATE_MULTIFD_ZLIB_LEVEL), | |
6a9ad154 JQ |
4244 | DEFINE_PROP_UINT8("multifd-zstd-level", MigrationState, |
4245 | parameters.multifd_zstd_level, | |
4246 | DEFAULT_MIGRATE_MULTIFD_ZSTD_LEVEL), | |
73af8dd8 JQ |
4247 | DEFINE_PROP_SIZE("xbzrle-cache-size", MigrationState, |
4248 | parameters.xbzrle_cache_size, | |
4249 | DEFAULT_MIGRATE_XBZRLE_CACHE_SIZE), | |
7e555c6c DDAG |
4250 | DEFINE_PROP_SIZE("max-postcopy-bandwidth", MigrationState, |
4251 | parameters.max_postcopy_bandwidth, | |
4252 | DEFAULT_MIGRATE_MAX_POSTCOPY_BANDWIDTH), | |
4cbc9c7f LQ |
4253 | DEFINE_PROP_UINT8("max-cpu-throttle", MigrationState, |
4254 | parameters.max_cpu_throttle, | |
4255 | DEFAULT_MIGRATE_MAX_CPU_THROTTLE), | |
ee3d96ba DDAG |
4256 | DEFINE_PROP_SIZE("announce-initial", MigrationState, |
4257 | parameters.announce_initial, | |
4258 | DEFAULT_MIGRATE_ANNOUNCE_INITIAL), | |
4259 | DEFINE_PROP_SIZE("announce-max", MigrationState, | |
4260 | parameters.announce_max, | |
4261 | DEFAULT_MIGRATE_ANNOUNCE_MAX), | |
4262 | DEFINE_PROP_SIZE("announce-rounds", MigrationState, | |
4263 | parameters.announce_rounds, | |
4264 | DEFAULT_MIGRATE_ANNOUNCE_ROUNDS), | |
4265 | DEFINE_PROP_SIZE("announce-step", MigrationState, | |
4266 | parameters.announce_step, | |
4267 | DEFAULT_MIGRATE_ANNOUNCE_STEP), | |
20814758 PX |
4268 | |
4269 | /* Migration capabilities */ | |
4270 | DEFINE_PROP_MIG_CAP("x-xbzrle", MIGRATION_CAPABILITY_XBZRLE), | |
4271 | DEFINE_PROP_MIG_CAP("x-rdma-pin-all", MIGRATION_CAPABILITY_RDMA_PIN_ALL), | |
4272 | DEFINE_PROP_MIG_CAP("x-auto-converge", MIGRATION_CAPABILITY_AUTO_CONVERGE), | |
4273 | DEFINE_PROP_MIG_CAP("x-zero-blocks", MIGRATION_CAPABILITY_ZERO_BLOCKS), | |
4274 | DEFINE_PROP_MIG_CAP("x-compress", MIGRATION_CAPABILITY_COMPRESS), | |
4275 | DEFINE_PROP_MIG_CAP("x-events", MIGRATION_CAPABILITY_EVENTS), | |
4276 | DEFINE_PROP_MIG_CAP("x-postcopy-ram", MIGRATION_CAPABILITY_POSTCOPY_RAM), | |
4277 | DEFINE_PROP_MIG_CAP("x-colo", MIGRATION_CAPABILITY_X_COLO), | |
4278 | DEFINE_PROP_MIG_CAP("x-release-ram", MIGRATION_CAPABILITY_RELEASE_RAM), | |
4279 | DEFINE_PROP_MIG_CAP("x-block", MIGRATION_CAPABILITY_BLOCK), | |
4280 | DEFINE_PROP_MIG_CAP("x-return-path", MIGRATION_CAPABILITY_RETURN_PATH), | |
cbfd6c95 | 4281 | DEFINE_PROP_MIG_CAP("x-multifd", MIGRATION_CAPABILITY_MULTIFD), |
6e8c25b4 AG |
4282 | DEFINE_PROP_MIG_CAP("x-background-snapshot", |
4283 | MIGRATION_CAPABILITY_BACKGROUND_SNAPSHOT), | |
1abaec9a LB |
4284 | #ifdef CONFIG_LINUX |
4285 | DEFINE_PROP_MIG_CAP("x-zero-copy-send", | |
4286 | MIGRATION_CAPABILITY_ZERO_COPY_SEND), | |
4287 | #endif | |
20814758 | 4288 | |
5272298c PX |
4289 | DEFINE_PROP_END_OF_LIST(), |
4290 | }; | |
4291 | ||
e5cb7e76 PX |
4292 | static void migration_class_init(ObjectClass *klass, void *data) |
4293 | { | |
4294 | DeviceClass *dc = DEVICE_CLASS(klass); | |
4295 | ||
4296 | dc->user_creatable = false; | |
4f67d30b | 4297 | device_class_set_props(dc, migration_properties); |
e5cb7e76 PX |
4298 | } |
4299 | ||
b91bf5e4 MAL |
4300 | static void migration_instance_finalize(Object *obj) |
4301 | { | |
4302 | MigrationState *ms = MIGRATION_OBJ(obj); | |
4303 | MigrationParameters *params = &ms->parameters; | |
4304 | ||
87db1a7d | 4305 | qemu_mutex_destroy(&ms->error_mutex); |
62df066f | 4306 | qemu_mutex_destroy(&ms->qemu_file_lock); |
b91bf5e4 MAL |
4307 | g_free(params->tls_hostname); |
4308 | g_free(params->tls_creds); | |
c7e0acd5 | 4309 | qemu_sem_destroy(&ms->wait_unplug_sem); |
ad767bed | 4310 | qemu_sem_destroy(&ms->rate_limit_sem); |
e91d8951 | 4311 | qemu_sem_destroy(&ms->pause_sem); |
b23c2ade | 4312 | qemu_sem_destroy(&ms->postcopy_pause_sem); |
14b1742e | 4313 | qemu_sem_destroy(&ms->postcopy_pause_rp_sem); |
edd090c7 | 4314 | qemu_sem_destroy(&ms->rp_state.rp_sem); |
ab105cc1 | 4315 | error_free(ms->error); |
b91bf5e4 MAL |
4316 | } |
4317 | ||
e5cb7e76 PX |
4318 | static void migration_instance_init(Object *obj) |
4319 | { | |
4320 | MigrationState *ms = MIGRATION_OBJ(obj); | |
8b0b29dc | 4321 | MigrationParameters *params = &ms->parameters; |
e5cb7e76 PX |
4322 | |
4323 | ms->state = MIGRATION_STATUS_NONE; | |
e5cb7e76 | 4324 | ms->mbps = -1; |
aecbfe9c | 4325 | ms->pages_per_second = -1; |
e91d8951 | 4326 | qemu_sem_init(&ms->pause_sem, 0); |
87db1a7d | 4327 | qemu_mutex_init(&ms->error_mutex); |
8b0b29dc PX |
4328 | |
4329 | params->tls_hostname = g_strdup(""); | |
4330 | params->tls_creds = g_strdup(""); | |
4331 | ||
4332 | /* Set has_* up only for parameter checks */ | |
4333 | params->has_compress_level = true; | |
4334 | params->has_compress_threads = true; | |
4335 | params->has_decompress_threads = true; | |
dc14a470 | 4336 | params->has_throttle_trigger_threshold = true; |
8b0b29dc PX |
4337 | params->has_cpu_throttle_initial = true; |
4338 | params->has_cpu_throttle_increment = true; | |
cbbf8182 | 4339 | params->has_cpu_throttle_tailslow = true; |
8b0b29dc PX |
4340 | params->has_max_bandwidth = true; |
4341 | params->has_downtime_limit = true; | |
4342 | params->has_x_checkpoint_delay = true; | |
4343 | params->has_block_incremental = true; | |
cbfd6c95 | 4344 | params->has_multifd_channels = true; |
96eef042 | 4345 | params->has_multifd_compression = true; |
9004db48 | 4346 | params->has_multifd_zlib_level = true; |
6a9ad154 | 4347 | params->has_multifd_zstd_level = true; |
73af8dd8 | 4348 | params->has_xbzrle_cache_size = true; |
7e555c6c | 4349 | params->has_max_postcopy_bandwidth = true; |
4cbc9c7f | 4350 | params->has_max_cpu_throttle = true; |
ee3d96ba DDAG |
4351 | params->has_announce_initial = true; |
4352 | params->has_announce_max = true; | |
4353 | params->has_announce_rounds = true; | |
4354 | params->has_announce_step = true; | |
b23c2ade PX |
4355 | |
4356 | qemu_sem_init(&ms->postcopy_pause_sem, 0); | |
14b1742e | 4357 | qemu_sem_init(&ms->postcopy_pause_rp_sem, 0); |
edd090c7 | 4358 | qemu_sem_init(&ms->rp_state.rp_sem, 0); |
ad767bed | 4359 | qemu_sem_init(&ms->rate_limit_sem, 0); |
c7e0acd5 | 4360 | qemu_sem_init(&ms->wait_unplug_sem, 0); |
62df066f | 4361 | qemu_mutex_init(&ms->qemu_file_lock); |
8b0b29dc PX |
4362 | } |
4363 | ||
4364 | /* | |
4365 | * Return true if check pass, false otherwise. Error will be put | |
4366 | * inside errp if provided. | |
4367 | */ | |
4368 | static bool migration_object_check(MigrationState *ms, Error **errp) | |
4369 | { | |
6b19a7d9 PX |
4370 | MigrationCapabilityStatusList *head = NULL; |
4371 | /* Assuming all off */ | |
4372 | bool cap_list[MIGRATION_CAPABILITY__MAX] = { 0 }, ret; | |
4373 | int i; | |
4374 | ||
8b0b29dc PX |
4375 | if (!migrate_params_check(&ms->parameters, errp)) { |
4376 | return false; | |
4377 | } | |
4378 | ||
6b19a7d9 PX |
4379 | for (i = 0; i < MIGRATION_CAPABILITY__MAX; i++) { |
4380 | if (ms->enabled_capabilities[i]) { | |
eaedde52 | 4381 | QAPI_LIST_PREPEND(head, migrate_cap_add(i, true)); |
6b19a7d9 PX |
4382 | } |
4383 | } | |
4384 | ||
4385 | ret = migrate_caps_check(cap_list, head, errp); | |
4386 | ||
4387 | /* It works with head == NULL */ | |
4388 | qapi_free_MigrationCapabilityStatusList(head); | |
4389 | ||
4390 | return ret; | |
e5cb7e76 PX |
4391 | } |
4392 | ||
4393 | static const TypeInfo migration_type = { | |
4394 | .name = TYPE_MIGRATION, | |
01f6e14c | 4395 | /* |
c8d3ff38 | 4396 | * NOTE: TYPE_MIGRATION is not really a device, as the object is |
2194abd6 | 4397 | * not created using qdev_new(), it is not attached to the qdev |
c8d3ff38 PX |
4398 | * device tree, and it is never realized. |
4399 | * | |
4400 | * TODO: Make this TYPE_OBJECT once QOM provides something like | |
4401 | * TYPE_DEVICE's "-global" properties. | |
01f6e14c | 4402 | */ |
e5cb7e76 PX |
4403 | .parent = TYPE_DEVICE, |
4404 | .class_init = migration_class_init, | |
4405 | .class_size = sizeof(MigrationClass), | |
4406 | .instance_size = sizeof(MigrationState), | |
4407 | .instance_init = migration_instance_init, | |
b91bf5e4 | 4408 | .instance_finalize = migration_instance_finalize, |
e5cb7e76 PX |
4409 | }; |
4410 | ||
4411 | static void register_migration_types(void) | |
4412 | { | |
4413 | type_register_static(&migration_type); | |
4414 | } | |
4415 | ||
4416 | type_init(register_migration_types); |