]>
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" |
85a8578e | 51 | #include "io/channel-tls.h" |
35a6ed4f | 52 | #include "migration/colo.h" |
4ffdb337 | 53 | #include "hw/boards.h" |
a27bd6c7 | 54 | #include "hw/qdev-properties.h" |
ce35e229 | 55 | #include "hw/qdev-properties-system.h" |
9d18af93 | 56 | #include "monitor/monitor.h" |
50510ea2 | 57 | #include "net/announce.h" |
c7e0acd5 | 58 | #include "qemu/queue.h" |
d32ca5ad | 59 | #include "multifd.h" |
b5eea99e | 60 | #include "qemu/yank.h" |
6e8c25b4 | 61 | #include "sysemu/cpus.h" |
39675fff | 62 | #include "yank_functions.h" |
1b529d90 | 63 | #include "sysemu/qtest.h" |
065e2813 | 64 | |
7590a2ae | 65 | #define MAX_THROTTLE (128 << 20) /* Migration transfer speed throttling */ |
5bb7910a | 66 | |
5b4e1eb7 JQ |
67 | /* Amount of time to allocate to each "chunk" of bandwidth-throttled |
68 | * data. */ | |
69 | #define BUFFER_DELAY 100 | |
70 | #define XFER_LIMIT_RATIO (1000 / BUFFER_DELAY) | |
71 | ||
2ff30257 AA |
72 | /* Time in milliseconds we are allowed to stop the source, |
73 | * for sending the last part */ | |
74 | #define DEFAULT_MIGRATE_SET_DOWNTIME 300 | |
75 | ||
87c9cc1c DHB |
76 | /* Maximum migrate downtime set to 2000 seconds */ |
77 | #define MAX_MIGRATE_DOWNTIME_SECONDS 2000 | |
78 | #define MAX_MIGRATE_DOWNTIME (MAX_MIGRATE_DOWNTIME_SECONDS * 1000) | |
79 | ||
8706d2d5 LL |
80 | /* Default compression thread count */ |
81 | #define DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT 8 | |
3fcb38c2 LL |
82 | /* Default decompression thread count, usually decompression is at |
83 | * least 4 times as fast as compression.*/ | |
84 | #define DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT 2 | |
8706d2d5 LL |
85 | /*0: means nocompress, 1: best speed, ... 9: best compress ratio */ |
86 | #define DEFAULT_MIGRATE_COMPRESS_LEVEL 1 | |
1626fee3 | 87 | /* Define default autoconverge cpu throttle migration parameters */ |
dc14a470 | 88 | #define DEFAULT_MIGRATE_THROTTLE_TRIGGER_THRESHOLD 50 |
d85a31d1 JH |
89 | #define DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL 20 |
90 | #define DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT 10 | |
4cbc9c7f | 91 | #define DEFAULT_MIGRATE_MAX_CPU_THROTTLE 99 |
8706d2d5 | 92 | |
17ad9b35 | 93 | /* Migration XBZRLE default cache size */ |
73af8dd8 | 94 | #define DEFAULT_MIGRATE_XBZRLE_CACHE_SIZE (64 * 1024 * 1024) |
17ad9b35 | 95 | |
131b2153 ZC |
96 | /* The delay time (in ms) between two COLO checkpoints */ |
97 | #define DEFAULT_MIGRATE_X_CHECKPOINT_DELAY (200 * 100) | |
4075fb1c | 98 | #define DEFAULT_MIGRATE_MULTIFD_CHANNELS 2 |
96eef042 | 99 | #define DEFAULT_MIGRATE_MULTIFD_COMPRESSION MULTIFD_COMPRESSION_NONE |
9004db48 JQ |
100 | /* 0: means nocompress, 1: best speed, ... 9: best compress ratio */ |
101 | #define DEFAULT_MIGRATE_MULTIFD_ZLIB_LEVEL 1 | |
6a9ad154 JQ |
102 | /* 0: means nocompress, 1: best speed, ... 20: best compress ratio */ |
103 | #define DEFAULT_MIGRATE_MULTIFD_ZSTD_LEVEL 1 | |
68b53591 | 104 | |
7e555c6c DDAG |
105 | /* Background transfer rate for postcopy, 0 means unlimited, note |
106 | * that page requests can still exceed this limit. | |
107 | */ | |
108 | #define DEFAULT_MIGRATE_MAX_POSTCOPY_BANDWIDTH 0 | |
109 | ||
ee3d96ba DDAG |
110 | /* |
111 | * Parameters for self_announce_delay giving a stream of RARP/ARP | |
112 | * packets after migration. | |
113 | */ | |
114 | #define DEFAULT_MIGRATE_ANNOUNCE_INITIAL 50 | |
115 | #define DEFAULT_MIGRATE_ANNOUNCE_MAX 550 | |
116 | #define DEFAULT_MIGRATE_ANNOUNCE_ROUNDS 5 | |
117 | #define DEFAULT_MIGRATE_ANNOUNCE_STEP 100 | |
118 | ||
99a0db9b GH |
119 | static NotifierList migration_state_notifiers = |
120 | NOTIFIER_LIST_INITIALIZER(migration_state_notifiers); | |
121 | ||
da6f1790 JQ |
122 | /* Messages sent on the return path from destination to source */ |
123 | enum mig_rp_message_type { | |
124 | MIG_RP_MSG_INVALID = 0, /* Must be 0 */ | |
125 | MIG_RP_MSG_SHUT, /* sibling will not send any more RP messages */ | |
126 | MIG_RP_MSG_PONG, /* Response to a PING; data (seq: be32 ) */ | |
127 | ||
128 | MIG_RP_MSG_REQ_PAGES_ID, /* data (start: be64, len: be32, id: string) */ | |
129 | MIG_RP_MSG_REQ_PAGES, /* data (start: be64, len: be32) */ | |
a335debb | 130 | MIG_RP_MSG_RECV_BITMAP, /* send recved_bitmap back to source */ |
13955b89 | 131 | MIG_RP_MSG_RESUME_ACK, /* tell source that we are ready to resume */ |
da6f1790 JQ |
132 | |
133 | MIG_RP_MSG_MAX | |
134 | }; | |
135 | ||
6e8c25b4 AG |
136 | /* Migration capabilities set */ |
137 | struct MigrateCapsSet { | |
138 | int size; /* Capability set size */ | |
139 | MigrationCapability caps[]; /* Variadic array of capabilities */ | |
140 | }; | |
141 | typedef struct MigrateCapsSet MigrateCapsSet; | |
142 | ||
143 | /* Define and initialize MigrateCapsSet */ | |
144 | #define INITIALIZE_MIGRATE_CAPS_SET(_name, ...) \ | |
145 | MigrateCapsSet _name = { \ | |
146 | .size = sizeof((int []) { __VA_ARGS__ }) / sizeof(int), \ | |
147 | .caps = { __VA_ARGS__ } \ | |
148 | } | |
149 | ||
150 | /* Background-snapshot compatibility check list */ | |
151 | static const | |
152 | INITIALIZE_MIGRATE_CAPS_SET(check_caps_background_snapshot, | |
153 | MIGRATION_CAPABILITY_POSTCOPY_RAM, | |
154 | MIGRATION_CAPABILITY_DIRTY_BITMAPS, | |
155 | MIGRATION_CAPABILITY_POSTCOPY_BLOCKTIME, | |
156 | MIGRATION_CAPABILITY_LATE_BLOCK_ACTIVATE, | |
157 | MIGRATION_CAPABILITY_RETURN_PATH, | |
158 | MIGRATION_CAPABILITY_MULTIFD, | |
159 | MIGRATION_CAPABILITY_PAUSE_BEFORE_SWITCHOVER, | |
160 | MIGRATION_CAPABILITY_AUTO_CONVERGE, | |
161 | MIGRATION_CAPABILITY_RELEASE_RAM, | |
162 | MIGRATION_CAPABILITY_RDMA_PIN_ALL, | |
163 | MIGRATION_CAPABILITY_COMPRESS, | |
164 | MIGRATION_CAPABILITY_XBZRLE, | |
165 | MIGRATION_CAPABILITY_X_COLO, | |
1abaec9a LB |
166 | MIGRATION_CAPABILITY_VALIDATE_UUID, |
167 | MIGRATION_CAPABILITY_ZERO_COPY_SEND); | |
6e8c25b4 | 168 | |
17549e84 JQ |
169 | /* When we add fault tolerance, we could have several |
170 | migrations at once. For now we don't need to add | |
171 | dynamic creation of migration */ | |
172 | ||
e5cb7e76 | 173 | static MigrationState *current_migration; |
e1b1b1bc | 174 | static MigrationIncomingState *current_incoming; |
e5cb7e76 | 175 | |
3af8554b DDAG |
176 | static GSList *migration_blockers; |
177 | ||
8b0b29dc | 178 | static bool migration_object_check(MigrationState *ms, Error **errp); |
0331c8ca DDAG |
179 | static int migration_maybe_pause(MigrationState *s, |
180 | int *current_active_state, | |
181 | int new_state); | |
892ae715 | 182 | static void migrate_fd_cancel(MigrationState *s); |
8b0b29dc | 183 | |
f444eeda PX |
184 | static bool migrate_allow_multi_channels = true; |
185 | ||
186 | void migrate_protocol_allow_multi_channels(bool allow) | |
187 | { | |
188 | migrate_allow_multi_channels = allow; | |
189 | } | |
190 | ||
191 | bool migrate_multi_channels_is_allowed(void) | |
192 | { | |
193 | return migrate_allow_multi_channels; | |
194 | } | |
195 | ||
8f8bfffc PX |
196 | static gint page_request_addr_cmp(gconstpointer ap, gconstpointer bp) |
197 | { | |
198 | uintptr_t a = (uintptr_t) ap, b = (uintptr_t) bp; | |
199 | ||
200 | return (a > b) - (a < b); | |
201 | } | |
202 | ||
e5cb7e76 PX |
203 | void migration_object_init(void) |
204 | { | |
205 | /* This can only be called once. */ | |
206 | assert(!current_migration); | |
207 | current_migration = MIGRATION_OBJ(object_new(TYPE_MIGRATION)); | |
4ffdb337 | 208 | |
e1b1b1bc PX |
209 | /* |
210 | * Init the migrate incoming object as well no matter whether | |
211 | * we'll use it or not. | |
212 | */ | |
213 | assert(!current_incoming); | |
214 | current_incoming = g_new0(MigrationIncomingState, 1); | |
215 | current_incoming->state = MIGRATION_STATUS_NONE; | |
216 | current_incoming->postcopy_remote_fds = | |
217 | g_array_new(FALSE, TRUE, sizeof(struct PostCopyFD)); | |
218 | qemu_mutex_init(¤t_incoming->rp_mutex); | |
60bb3c58 | 219 | qemu_mutex_init(¤t_incoming->postcopy_prio_thread_mutex); |
e1b1b1bc PX |
220 | qemu_event_init(¤t_incoming->main_thread_load_event, false); |
221 | qemu_sem_init(¤t_incoming->postcopy_pause_sem_dst, 0); | |
222 | qemu_sem_init(¤t_incoming->postcopy_pause_sem_fault, 0); | |
60bb3c58 | 223 | qemu_sem_init(¤t_incoming->postcopy_pause_sem_fast_load, 0); |
8f8bfffc PX |
224 | qemu_mutex_init(¤t_incoming->page_request_mutex); |
225 | current_incoming->page_requested = g_tree_new(page_request_addr_cmp); | |
e1b1b1bc | 226 | |
f9734d5d | 227 | migration_object_check(current_migration, &error_fatal); |
e0d17dfd PB |
228 | |
229 | blk_mig_init(); | |
230 | ram_mig_init(); | |
231 | dirty_bitmap_mig_init(); | |
e5cb7e76 PX |
232 | } |
233 | ||
458fecca | 234 | void migration_cancel(const Error *error) |
c7c0e724 | 235 | { |
458fecca LV |
236 | if (error) { |
237 | migrate_set_error(current_migration, error); | |
238 | } | |
c7c0e724 DH |
239 | migrate_fd_cancel(current_migration); |
240 | } | |
241 | ||
892ae715 | 242 | void migration_shutdown(void) |
1f895604 | 243 | { |
795969ab RL |
244 | /* |
245 | * When the QEMU main thread exit, the COLO thread | |
246 | * may wait a semaphore. So, we should wakeup the | |
247 | * COLO thread before migration shutdown. | |
248 | */ | |
249 | colo_shutdown(); | |
892ae715 DDAG |
250 | /* |
251 | * Cancel the current migration - that will (eventually) | |
252 | * stop the migration using this structure | |
253 | */ | |
458fecca | 254 | migration_cancel(NULL); |
1f895604 | 255 | object_unref(OBJECT(current_migration)); |
1499ab09 VSO |
256 | |
257 | /* | |
258 | * Cancel outgoing migration of dirty bitmaps. It should | |
259 | * at least unref used block nodes. | |
260 | */ | |
261 | dirty_bitmap_mig_cancel_outgoing(); | |
262 | ||
263 | /* | |
264 | * Cancel incoming migration of dirty bitmaps. Dirty bitmaps | |
265 | * are non-critical data, and their loss never considered as | |
266 | * something serious. | |
267 | */ | |
268 | dirty_bitmap_mig_cancel_incoming(); | |
1f895604 VSO |
269 | } |
270 | ||
bca7856a | 271 | /* For outgoing */ |
859bc756 | 272 | MigrationState *migrate_get_current(void) |
17549e84 | 273 | { |
e5cb7e76 PX |
274 | /* This can only be called after the object created. */ |
275 | assert(current_migration); | |
276 | return current_migration; | |
17549e84 JQ |
277 | } |
278 | ||
bca7856a DDAG |
279 | MigrationIncomingState *migration_incoming_get_current(void) |
280 | { | |
e1b1b1bc PX |
281 | assert(current_incoming); |
282 | return current_incoming; | |
bca7856a DDAG |
283 | } |
284 | ||
e031149c PX |
285 | void migration_incoming_transport_cleanup(MigrationIncomingState *mis) |
286 | { | |
287 | if (mis->socket_address_list) { | |
288 | qapi_free_SocketAddressList(mis->socket_address_list); | |
289 | mis->socket_address_list = NULL; | |
290 | } | |
291 | ||
292 | if (mis->transport_cleanup) { | |
293 | mis->transport_cleanup(mis->transport_data); | |
294 | mis->transport_data = mis->transport_cleanup = NULL; | |
295 | } | |
296 | } | |
297 | ||
bca7856a DDAG |
298 | void migration_incoming_state_destroy(void) |
299 | { | |
b4b076da JQ |
300 | struct MigrationIncomingState *mis = migration_incoming_get_current(); |
301 | ||
3482655b | 302 | if (mis->to_src_file) { |
660819b1 PX |
303 | /* Tell source that we are done */ |
304 | migrate_send_rp_shut(mis, qemu_file_get_error(mis->from_src_file) != 0); | |
3482655b PX |
305 | qemu_fclose(mis->to_src_file); |
306 | mis->to_src_file = NULL; | |
307 | } | |
308 | ||
660819b1 | 309 | if (mis->from_src_file) { |
39675fff | 310 | migration_ioc_unregister_yank_from_file(mis->from_src_file); |
660819b1 PX |
311 | qemu_fclose(mis->from_src_file); |
312 | mis->from_src_file = NULL; | |
313 | } | |
00fa4fc8 DDAG |
314 | if (mis->postcopy_remote_fds) { |
315 | g_array_free(mis->postcopy_remote_fds, TRUE); | |
316 | mis->postcopy_remote_fds = NULL; | |
317 | } | |
660819b1 | 318 | |
e031149c | 319 | migration_incoming_transport_cleanup(mis); |
1783c00f DDAG |
320 | qemu_event_reset(&mis->main_thread_load_event); |
321 | ||
8f8bfffc PX |
322 | if (mis->page_requested) { |
323 | g_tree_destroy(mis->page_requested); | |
324 | mis->page_requested = NULL; | |
325 | } | |
326 | ||
36f62f11 PX |
327 | if (mis->postcopy_qemufile_dst) { |
328 | migration_ioc_unregister_yank_from_file(mis->postcopy_qemufile_dst); | |
329 | qemu_fclose(mis->postcopy_qemufile_dst); | |
330 | mis->postcopy_qemufile_dst = NULL; | |
331 | } | |
332 | ||
b5eea99e | 333 | yank_unregister_instance(MIGRATION_YANK_INSTANCE); |
bca7856a DDAG |
334 | } |
335 | ||
b05dc723 JQ |
336 | static void migrate_generate_event(int new_state) |
337 | { | |
338 | if (migrate_use_events()) { | |
3ab72385 | 339 | qapi_event_send_migration(new_state); |
b05dc723 JQ |
340 | } |
341 | } | |
342 | ||
0f073f44 DDAG |
343 | static bool migrate_late_block_activate(void) |
344 | { | |
345 | MigrationState *s; | |
346 | ||
347 | s = migrate_get_current(); | |
348 | ||
349 | return s->enabled_capabilities[ | |
350 | MIGRATION_CAPABILITY_LATE_BLOCK_ACTIVATE]; | |
351 | } | |
352 | ||
da6f1790 JQ |
353 | /* |
354 | * Send a message on the return channel back to the source | |
355 | * of the migration. | |
356 | */ | |
d6208e35 PX |
357 | static int migrate_send_rp_message(MigrationIncomingState *mis, |
358 | enum mig_rp_message_type message_type, | |
359 | uint16_t len, void *data) | |
da6f1790 | 360 | { |
d6208e35 PX |
361 | int ret = 0; |
362 | ||
da6f1790 | 363 | trace_migrate_send_rp_message((int)message_type, len); |
37396950 | 364 | QEMU_LOCK_GUARD(&mis->rp_mutex); |
d6208e35 PX |
365 | |
366 | /* | |
367 | * It's possible that the file handle got lost due to network | |
368 | * failures. | |
369 | */ | |
370 | if (!mis->to_src_file) { | |
371 | ret = -EIO; | |
37396950 | 372 | return ret; |
d6208e35 PX |
373 | } |
374 | ||
da6f1790 JQ |
375 | qemu_put_be16(mis->to_src_file, (unsigned int)message_type); |
376 | qemu_put_be16(mis->to_src_file, len); | |
377 | qemu_put_buffer(mis->to_src_file, data, len); | |
378 | qemu_fflush(mis->to_src_file); | |
d6208e35 PX |
379 | |
380 | /* It's possible that qemu file got error during sending */ | |
381 | ret = qemu_file_get_error(mis->to_src_file); | |
382 | ||
d6208e35 | 383 | return ret; |
da6f1790 JQ |
384 | } |
385 | ||
2e2bce16 PX |
386 | /* Request one page from the source VM at the given start address. |
387 | * rb: the RAMBlock to request the page in | |
1e2d90eb DDAG |
388 | * Start: Address offset within the RB |
389 | * Len: Length in bytes required - must be a multiple of pagesize | |
390 | */ | |
7a267fc4 PX |
391 | int migrate_send_rp_message_req_pages(MigrationIncomingState *mis, |
392 | RAMBlock *rb, ram_addr_t start) | |
1e2d90eb | 393 | { |
cb8d4c8f | 394 | uint8_t bufc[12 + 1 + 255]; /* start (8), len (4), rbname up to 256 */ |
1e2d90eb | 395 | size_t msglen = 12; /* start + len */ |
2e2bce16 | 396 | size_t len = qemu_ram_pagesize(rb); |
d6208e35 | 397 | enum mig_rp_message_type msg_type; |
2e2bce16 PX |
398 | const char *rbname; |
399 | int rbname_len; | |
1e2d90eb DDAG |
400 | |
401 | *(uint64_t *)bufc = cpu_to_be64((uint64_t)start); | |
402 | *(uint32_t *)(bufc + 8) = cpu_to_be32((uint32_t)len); | |
403 | ||
2e2bce16 PX |
404 | /* |
405 | * We maintain the last ramblock that we requested for page. Note that we | |
406 | * don't need locking because this function will only be called within the | |
407 | * postcopy ram fault thread. | |
408 | */ | |
409 | if (rb != mis->last_rb) { | |
410 | mis->last_rb = rb; | |
411 | ||
412 | rbname = qemu_ram_get_idstr(rb); | |
413 | rbname_len = strlen(rbname); | |
414 | ||
1e2d90eb DDAG |
415 | assert(rbname_len < 256); |
416 | ||
417 | bufc[msglen++] = rbname_len; | |
418 | memcpy(bufc + msglen, rbname, rbname_len); | |
419 | msglen += rbname_len; | |
d6208e35 | 420 | msg_type = MIG_RP_MSG_REQ_PAGES_ID; |
1e2d90eb | 421 | } else { |
d6208e35 | 422 | msg_type = MIG_RP_MSG_REQ_PAGES; |
1e2d90eb | 423 | } |
d6208e35 PX |
424 | |
425 | return migrate_send_rp_message(mis, msg_type, msglen, bufc); | |
1e2d90eb DDAG |
426 | } |
427 | ||
7a267fc4 | 428 | int migrate_send_rp_req_pages(MigrationIncomingState *mis, |
8f8bfffc | 429 | RAMBlock *rb, ram_addr_t start, uint64_t haddr) |
7a267fc4 | 430 | { |
7648297d | 431 | void *aligned = (void *)(uintptr_t)ROUND_DOWN(haddr, qemu_ram_pagesize(rb)); |
a2429283 | 432 | bool received = false; |
8f8bfffc PX |
433 | |
434 | WITH_QEMU_LOCK_GUARD(&mis->page_request_mutex) { | |
435 | received = ramblock_recv_bitmap_test_byte_offset(rb, start); | |
436 | if (!received && !g_tree_lookup(mis->page_requested, aligned)) { | |
437 | /* | |
438 | * The page has not been received, and it's not yet in the page | |
439 | * request list. Queue it. Set the value of element to 1, so that | |
440 | * things like g_tree_lookup() will return TRUE (1) when found. | |
441 | */ | |
442 | g_tree_insert(mis->page_requested, aligned, (gpointer)1); | |
443 | mis->page_requested_count++; | |
444 | trace_postcopy_page_req_add(aligned, mis->page_requested_count); | |
445 | } | |
446 | } | |
447 | ||
448 | /* | |
449 | * If the page is there, skip sending the message. We don't even need the | |
450 | * lock because as long as the page arrived, it'll be there forever. | |
451 | */ | |
452 | if (received) { | |
453 | return 0; | |
454 | } | |
455 | ||
7a267fc4 PX |
456 | return migrate_send_rp_message_req_pages(mis, rb, start); |
457 | } | |
458 | ||
aad555c2 ZC |
459 | static bool migration_colo_enabled; |
460 | bool migration_incoming_colo_enabled(void) | |
461 | { | |
462 | return migration_colo_enabled; | |
463 | } | |
464 | ||
465 | void migration_incoming_disable_colo(void) | |
466 | { | |
18b1d3c9 | 467 | ram_block_discard_disable(false); |
aad555c2 ZC |
468 | migration_colo_enabled = false; |
469 | } | |
470 | ||
18b1d3c9 | 471 | int migration_incoming_enable_colo(void) |
aad555c2 | 472 | { |
18b1d3c9 DH |
473 | if (ram_block_discard_disable(true)) { |
474 | error_report("COLO: cannot disable RAM discard"); | |
475 | return -EBUSY; | |
476 | } | |
aad555c2 | 477 | migration_colo_enabled = true; |
18b1d3c9 | 478 | return 0; |
aad555c2 ZC |
479 | } |
480 | ||
9aca82ba JQ |
481 | void migrate_add_address(SocketAddress *address) |
482 | { | |
483 | MigrationIncomingState *mis = migration_incoming_get_current(); | |
9aca82ba | 484 | |
54aa3de7 EB |
485 | QAPI_LIST_PREPEND(mis->socket_address_list, |
486 | QAPI_CLONE(SocketAddress, address)); | |
9aca82ba JQ |
487 | } |
488 | ||
e69d50d6 | 489 | static void qemu_start_incoming_migration(const char *uri, Error **errp) |
5bb7910a | 490 | { |
d658f65c | 491 | const char *p = NULL; |
34c9dd8e | 492 | |
f444eeda | 493 | migrate_protocol_allow_multi_channels(false); /* reset it anyway */ |
3ab72385 | 494 | qapi_event_send_migration(MIGRATION_STATUS_SETUP); |
e69d50d6 PB |
495 | if (strstart(uri, "tcp:", &p) || |
496 | strstart(uri, "unix:", NULL) || | |
497 | strstart(uri, "vsock:", NULL)) { | |
f444eeda | 498 | migrate_protocol_allow_multi_channels(true); |
d658f65c | 499 | socket_start_incoming_migration(p ? p : uri, errp); |
2da776db | 500 | #ifdef CONFIG_RDMA |
adde220a | 501 | } else if (strstart(uri, "rdma:", &p)) { |
2da776db MH |
502 | rdma_start_incoming_migration(p, errp); |
503 | #endif | |
adde220a | 504 | } else if (strstart(uri, "exec:", &p)) { |
43eaae28 | 505 | exec_start_incoming_migration(p, errp); |
adde220a | 506 | } else if (strstart(uri, "fd:", &p)) { |
43eaae28 | 507 | fd_start_incoming_migration(p, errp); |
adde220a | 508 | } else { |
312fd5f2 | 509 | error_setg(errp, "unknown migration protocol: %s", uri); |
8ca5e801 | 510 | } |
5bb7910a AL |
511 | } |
512 | ||
0aa6aefc DL |
513 | static void process_incoming_migration_bh(void *opaque) |
514 | { | |
515 | Error *local_err = NULL; | |
516 | MigrationIncomingState *mis = opaque; | |
517 | ||
0f073f44 DDAG |
518 | /* If capability late_block_activate is set: |
519 | * Only fire up the block code now if we're going to restart the | |
520 | * VM, else 'cont' will do it. | |
521 | * This causes file locking to happen; so we don't want it to happen | |
522 | * unless we really are starting the VM. | |
523 | */ | |
524 | if (!migrate_late_block_activate() || | |
525 | (autostart && (!global_state_received() || | |
526 | global_state_get_runstate() == RUN_STATE_RUNNING))) { | |
3b717194 | 527 | /* Make sure all file formats throw away their mutable metadata. |
0f073f44 | 528 | * If we get an error here, just don't restart the VM yet. */ |
3b717194 | 529 | bdrv_activate_all(&local_err); |
0f073f44 DDAG |
530 | if (local_err) { |
531 | error_report_err(local_err); | |
532 | local_err = NULL; | |
533 | autostart = false; | |
534 | } | |
d35ff5e6 KW |
535 | } |
536 | ||
0aa6aefc DL |
537 | /* |
538 | * This must happen after all error conditions are dealt with and | |
539 | * we're sure the VM is going to be running on this host. | |
540 | */ | |
7659505c | 541 | qemu_announce_self(&mis->announce_timer, migrate_announce_params()); |
0aa6aefc | 542 | |
f986c3d2 JQ |
543 | if (multifd_load_cleanup(&local_err) != 0) { |
544 | error_report_err(local_err); | |
545 | autostart = false; | |
546 | } | |
0aa6aefc DL |
547 | /* If global state section was not received or we are in running |
548 | state, we need to obey autostart. Any other state is set with | |
549 | runstate_set. */ | |
550 | ||
b35ebdf0 VSO |
551 | dirty_bitmap_mig_before_vm_start(); |
552 | ||
0aa6aefc DL |
553 | if (!global_state_received() || |
554 | global_state_get_runstate() == RUN_STATE_RUNNING) { | |
555 | if (autostart) { | |
556 | vm_start(); | |
557 | } else { | |
558 | runstate_set(RUN_STATE_PAUSED); | |
559 | } | |
db009729 ZC |
560 | } else if (migration_incoming_colo_enabled()) { |
561 | migration_incoming_disable_colo(); | |
562 | vm_start(); | |
0aa6aefc DL |
563 | } else { |
564 | runstate_set(global_state_get_runstate()); | |
565 | } | |
0aa6aefc DL |
566 | /* |
567 | * This must happen after any state changes since as soon as an external | |
568 | * observer sees this event they might start to prod at the VM assuming | |
569 | * it's ready to use. | |
570 | */ | |
571 | migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE, | |
572 | MIGRATION_STATUS_COMPLETED); | |
573 | qemu_bh_delete(mis->bh); | |
574 | migration_incoming_state_destroy(); | |
575 | } | |
576 | ||
38e8f9af MAL |
577 | static void coroutine_fn |
578 | process_incoming_migration_co(void *opaque) | |
511c0231 | 579 | { |
b4b076da | 580 | MigrationIncomingState *mis = migration_incoming_get_current(); |
e9bef235 | 581 | PostcopyState ps; |
1c12e1f5 | 582 | int ret; |
8e48ac95 | 583 | Error *local_err = NULL; |
1c12e1f5 | 584 | |
4f0fae7f | 585 | assert(mis->from_src_file); |
92370989 | 586 | mis->migration_incoming_co = qemu_coroutine_self(); |
67f11b5c | 587 | mis->largest_page_size = qemu_ram_pagesize_largest(); |
093e3c42 | 588 | postcopy_state_set(POSTCOPY_INCOMING_NONE); |
93d7af6f HZ |
589 | migrate_set_state(&mis->state, MIGRATION_STATUS_NONE, |
590 | MIGRATION_STATUS_ACTIVE); | |
4f0fae7f | 591 | ret = qemu_loadvm_state(mis->from_src_file); |
bca7856a | 592 | |
e9bef235 DDAG |
593 | ps = postcopy_state_get(); |
594 | trace_process_incoming_migration_co_end(ret, ps); | |
595 | if (ps != POSTCOPY_INCOMING_NONE) { | |
596 | if (ps == POSTCOPY_INCOMING_ADVISE) { | |
597 | /* | |
598 | * Where a migration had postcopy enabled (and thus went to advise) | |
599 | * but managed to complete within the precopy period, we can use | |
600 | * the normal exit. | |
601 | */ | |
602 | postcopy_ram_incoming_cleanup(mis); | |
603 | } else if (ret >= 0) { | |
604 | /* | |
605 | * Postcopy was started, cleanup should happen at the end of the | |
606 | * postcopy thread. | |
607 | */ | |
608 | trace_process_incoming_migration_co_postcopy_end_main(); | |
609 | return; | |
610 | } | |
611 | /* Else if something went wrong then just fall out of the normal exit */ | |
612 | } | |
613 | ||
25d0c16f | 614 | /* we get COLO info, and know if we are in COLO mode */ |
aad555c2 | 615 | if (!ret && migration_incoming_colo_enabled()) { |
3b717194 EGE |
616 | /* Make sure all file formats throw away their mutable metadata */ |
617 | bdrv_activate_all(&local_err); | |
8e48ac95 | 618 | if (local_err) { |
8e48ac95 | 619 | error_report_err(local_err); |
6d99c2d4 | 620 | goto fail; |
8e48ac95 ZC |
621 | } |
622 | ||
25d0c16f HZ |
623 | qemu_thread_create(&mis->colo_incoming_thread, "COLO incoming", |
624 | colo_process_incoming_thread, mis, QEMU_THREAD_JOINABLE); | |
625 | mis->have_colo_incoming_thread = true; | |
626 | qemu_coroutine_yield(); | |
627 | ||
684bfd18 | 628 | qemu_mutex_unlock_iothread(); |
25d0c16f HZ |
629 | /* Wait checkpoint incoming thread exit before free resource */ |
630 | qemu_thread_join(&mis->colo_incoming_thread); | |
684bfd18 | 631 | qemu_mutex_lock_iothread(); |
13af18f2 ZC |
632 | /* We hold the global iothread lock, so it is safe here */ |
633 | colo_release_ram_cache(); | |
25d0c16f HZ |
634 | } |
635 | ||
1c12e1f5 | 636 | if (ret < 0) { |
db80face | 637 | error_report("load of migration failed: %s", strerror(-ret)); |
6d99c2d4 | 638 | goto fail; |
511c0231 | 639 | } |
0aa6aefc DL |
640 | mis->bh = qemu_bh_new(process_incoming_migration_bh, mis); |
641 | qemu_bh_schedule(mis->bh); | |
92370989 | 642 | mis->migration_incoming_co = NULL; |
6d99c2d4 FL |
643 | return; |
644 | fail: | |
645 | local_err = NULL; | |
646 | migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE, | |
647 | MIGRATION_STATUS_FAILED); | |
648 | qemu_fclose(mis->from_src_file); | |
649 | if (multifd_load_cleanup(&local_err) != 0) { | |
650 | error_report_err(local_err); | |
651 | } | |
652 | exit(EXIT_FAILURE); | |
511c0231 JQ |
653 | } |
654 | ||
b673eab4 | 655 | /** |
7d6f6933 | 656 | * migration_incoming_setup: Setup incoming migration |
b673eab4 JQ |
657 | * @f: file for main migration channel |
658 | * @errp: where to put errors | |
7d6f6933 MA |
659 | * |
660 | * Returns: %true on success, %false on error. | |
b673eab4 | 661 | */ |
7d6f6933 | 662 | static bool migration_incoming_setup(QEMUFile *f, Error **errp) |
82a4da79 | 663 | { |
4f0fae7f | 664 | MigrationIncomingState *mis = migration_incoming_get_current(); |
82a4da79 | 665 | |
7d6f6933 MA |
666 | if (multifd_load_setup(errp) != 0) { |
667 | return false; | |
f986c3d2 JQ |
668 | } |
669 | ||
4f0fae7f JQ |
670 | if (!mis->from_src_file) { |
671 | mis->from_src_file = f; | |
672 | } | |
06ad5135 | 673 | qemu_file_set_blocking(f, false); |
7d6f6933 | 674 | return true; |
e595a01a JQ |
675 | } |
676 | ||
36c2f8be | 677 | void migration_incoming_process(void) |
e595a01a JQ |
678 | { |
679 | Coroutine *co = qemu_coroutine_create(process_incoming_migration_co, NULL); | |
0b8b8753 | 680 | qemu_coroutine_enter(co); |
82a4da79 PB |
681 | } |
682 | ||
884835fa | 683 | /* Returns true if recovered from a paused migration, otherwise false */ |
a39e9339 | 684 | static bool postcopy_try_recover(void) |
e595a01a | 685 | { |
d96c9e8d PX |
686 | MigrationIncomingState *mis = migration_incoming_get_current(); |
687 | ||
688 | if (mis->state == MIGRATION_STATUS_POSTCOPY_PAUSED) { | |
689 | /* Resumed from a paused postcopy migration */ | |
690 | ||
a39e9339 PX |
691 | /* This should be set already in migration_incoming_setup() */ |
692 | assert(mis->from_src_file); | |
d96c9e8d | 693 | /* Postcopy has standalone thread to do vm load */ |
a39e9339 | 694 | qemu_file_set_blocking(mis->from_src_file, true); |
d96c9e8d PX |
695 | |
696 | /* Re-configure the return path */ | |
a39e9339 | 697 | mis->to_src_file = qemu_file_get_return_path(mis->from_src_file); |
d96c9e8d PX |
698 | |
699 | migrate_set_state(&mis->state, MIGRATION_STATUS_POSTCOPY_PAUSED, | |
700 | MIGRATION_STATUS_POSTCOPY_RECOVER); | |
701 | ||
702 | /* | |
703 | * Here, we only wake up the main loading thread (while the | |
60bb3c58 | 704 | * rest threads will still be waiting), so that we can receive |
d96c9e8d | 705 | * commands from source now, and answer it if needed. The |
60bb3c58 | 706 | * rest threads will be woken up afterwards until we are sure |
d96c9e8d PX |
707 | * that source is ready to reply to page requests. |
708 | */ | |
709 | qemu_sem_post(&mis->postcopy_pause_sem_dst); | |
884835fa PX |
710 | return true; |
711 | } | |
712 | ||
713 | return false; | |
714 | } | |
715 | ||
b673eab4 | 716 | void migration_fd_process_incoming(QEMUFile *f, Error **errp) |
884835fa | 717 | { |
a39e9339 | 718 | if (!migration_incoming_setup(f, errp)) { |
884835fa | 719 | return; |
d96c9e8d | 720 | } |
a39e9339 | 721 | if (postcopy_try_recover()) { |
b673eab4 JQ |
722 | return; |
723 | } | |
884835fa | 724 | migration_incoming_process(); |
e595a01a JQ |
725 | } |
726 | ||
36f62f11 PX |
727 | static bool migration_needs_multiple_sockets(void) |
728 | { | |
729 | return migrate_use_multifd() || migrate_postcopy_preempt(); | |
730 | } | |
731 | ||
49ed0d24 | 732 | void migration_ioc_process_incoming(QIOChannel *ioc, Error **errp) |
4f0fae7f JQ |
733 | { |
734 | MigrationIncomingState *mis = migration_incoming_get_current(); | |
b673eab4 | 735 | Error *local_err = NULL; |
a429e7f4 | 736 | bool start_migration; |
36f62f11 | 737 | QEMUFile *f; |
4f0fae7f JQ |
738 | |
739 | if (!mis->from_src_file) { | |
a429e7f4 | 740 | /* The first connection (multifd may have multiple) */ |
36f62f11 | 741 | f = qemu_file_new_input(ioc); |
a429e7f4 | 742 | |
7d6f6933 | 743 | if (!migration_incoming_setup(f, errp)) { |
b673eab4 JQ |
744 | return; |
745 | } | |
a429e7f4 PX |
746 | |
747 | /* | |
748 | * Common migration only needs one channel, so we can start | |
36f62f11 | 749 | * right now. Some features need more than one channel, we wait. |
a429e7f4 | 750 | */ |
36f62f11 | 751 | start_migration = !migration_needs_multiple_sockets(); |
a429e7f4 PX |
752 | } else { |
753 | /* Multiple connections */ | |
36f62f11 PX |
754 | assert(migration_needs_multiple_sockets()); |
755 | if (migrate_use_multifd()) { | |
756 | start_migration = multifd_recv_new_channel(ioc, &local_err); | |
757 | } else { | |
758 | assert(migrate_postcopy_preempt()); | |
759 | f = qemu_file_new_input(ioc); | |
760 | start_migration = postcopy_preempt_new_channel(mis, f); | |
761 | } | |
49ed0d24 FL |
762 | if (local_err) { |
763 | error_propagate(errp, local_err); | |
764 | return; | |
765 | } | |
4f0fae7f | 766 | } |
81e62053 | 767 | |
a429e7f4 | 768 | if (start_migration) { |
a39e9339 PX |
769 | /* If it's a recovery, we're done */ |
770 | if (postcopy_try_recover()) { | |
771 | return; | |
772 | } | |
81e62053 PX |
773 | migration_incoming_process(); |
774 | } | |
4f0fae7f JQ |
775 | } |
776 | ||
428d8908 JQ |
777 | /** |
778 | * @migration_has_all_channels: We have received all channels that we need | |
779 | * | |
780 | * Returns true when we have got connections to all the channels that | |
781 | * we need for migration. | |
782 | */ | |
783 | bool migration_has_all_channels(void) | |
784 | { | |
ca273df3 | 785 | MigrationIncomingState *mis = migration_incoming_get_current(); |
62c1e0ca | 786 | |
36f62f11 PX |
787 | if (!mis->from_src_file) { |
788 | return false; | |
789 | } | |
62c1e0ca | 790 | |
36f62f11 PX |
791 | if (migrate_use_multifd()) { |
792 | return multifd_recv_all_channels_created(); | |
793 | } | |
794 | ||
795 | if (migrate_postcopy_preempt()) { | |
796 | return mis->postcopy_qemufile_dst != NULL; | |
797 | } | |
798 | ||
799 | return true; | |
428d8908 JQ |
800 | } |
801 | ||
6decec93 DDAG |
802 | /* |
803 | * Send a 'SHUT' message on the return channel with the given value | |
804 | * to indicate that we've finished with the RP. Non-0 value indicates | |
805 | * error. | |
806 | */ | |
807 | void migrate_send_rp_shut(MigrationIncomingState *mis, | |
808 | uint32_t value) | |
809 | { | |
810 | uint32_t buf; | |
811 | ||
812 | buf = cpu_to_be32(value); | |
813 | migrate_send_rp_message(mis, MIG_RP_MSG_SHUT, sizeof(buf), &buf); | |
814 | } | |
815 | ||
816 | /* | |
817 | * Send a 'PONG' message on the return channel with the given value | |
818 | * (normally in response to a 'PING') | |
819 | */ | |
820 | void migrate_send_rp_pong(MigrationIncomingState *mis, | |
821 | uint32_t value) | |
822 | { | |
823 | uint32_t buf; | |
824 | ||
825 | buf = cpu_to_be32(value); | |
826 | migrate_send_rp_message(mis, MIG_RP_MSG_PONG, sizeof(buf), &buf); | |
827 | } | |
828 | ||
a335debb PX |
829 | void migrate_send_rp_recv_bitmap(MigrationIncomingState *mis, |
830 | char *block_name) | |
831 | { | |
832 | char buf[512]; | |
833 | int len; | |
834 | int64_t res; | |
835 | ||
836 | /* | |
837 | * First, we send the header part. It contains only the len of | |
838 | * idstr, and the idstr itself. | |
839 | */ | |
840 | len = strlen(block_name); | |
841 | buf[0] = len; | |
842 | memcpy(buf + 1, block_name, len); | |
843 | ||
844 | if (mis->state != MIGRATION_STATUS_POSTCOPY_RECOVER) { | |
845 | error_report("%s: MSG_RP_RECV_BITMAP only used for recovery", | |
846 | __func__); | |
847 | return; | |
848 | } | |
849 | ||
850 | migrate_send_rp_message(mis, MIG_RP_MSG_RECV_BITMAP, len + 1, buf); | |
851 | ||
852 | /* | |
853 | * Next, we dump the received bitmap to the stream. | |
854 | * | |
855 | * TODO: currently we are safe since we are the only one that is | |
856 | * using the to_src_file handle (fault thread is still paused), | |
857 | * and it's ok even not taking the mutex. However the best way is | |
858 | * to take the lock before sending the message header, and release | |
859 | * the lock after sending the bitmap. | |
860 | */ | |
861 | qemu_mutex_lock(&mis->rp_mutex); | |
862 | res = ramblock_recv_bitmap_send(mis->to_src_file, block_name); | |
863 | qemu_mutex_unlock(&mis->rp_mutex); | |
864 | ||
865 | trace_migrate_send_rp_recv_bitmap(block_name, res); | |
866 | } | |
867 | ||
13955b89 PX |
868 | void migrate_send_rp_resume_ack(MigrationIncomingState *mis, uint32_t value) |
869 | { | |
870 | uint32_t buf; | |
871 | ||
872 | buf = cpu_to_be32(value); | |
873 | migrate_send_rp_message(mis, MIG_RP_MSG_RESUME_ACK, sizeof(buf), &buf); | |
874 | } | |
875 | ||
bbf6da32 OW |
876 | MigrationCapabilityStatusList *qmp_query_migrate_capabilities(Error **errp) |
877 | { | |
95b3a8c8 EB |
878 | MigrationCapabilityStatusList *head = NULL, **tail = &head; |
879 | MigrationCapabilityStatus *caps; | |
bbf6da32 OW |
880 | MigrationState *s = migrate_get_current(); |
881 | int i; | |
882 | ||
7fb1cf16 | 883 | for (i = 0; i < MIGRATION_CAPABILITY__MAX; i++) { |
ed1701c6 DDAG |
884 | #ifndef CONFIG_LIVE_BLOCK_MIGRATION |
885 | if (i == MIGRATION_CAPABILITY_BLOCK) { | |
886 | continue; | |
887 | } | |
888 | #endif | |
95b3a8c8 EB |
889 | caps = g_malloc0(sizeof(*caps)); |
890 | caps->capability = i; | |
891 | caps->state = s->enabled_capabilities[i]; | |
892 | QAPI_LIST_APPEND(tail, caps); | |
bbf6da32 OW |
893 | } |
894 | ||
895 | return head; | |
896 | } | |
897 | ||
85de8323 LL |
898 | MigrationParameters *qmp_query_migrate_parameters(Error **errp) |
899 | { | |
900 | MigrationParameters *params; | |
901 | MigrationState *s = migrate_get_current(); | |
902 | ||
e87fae4c | 903 | /* TODO use QAPI_CLONE() instead of duplicating it inline */ |
85de8323 | 904 | params = g_malloc0(sizeof(*params)); |
de63ab61 | 905 | params->has_compress_level = true; |
2594f56d | 906 | params->compress_level = s->parameters.compress_level; |
de63ab61 | 907 | params->has_compress_threads = true; |
2594f56d | 908 | params->compress_threads = s->parameters.compress_threads; |
1d58872a XG |
909 | params->has_compress_wait_thread = true; |
910 | params->compress_wait_thread = s->parameters.compress_wait_thread; | |
de63ab61 | 911 | params->has_decompress_threads = true; |
2594f56d | 912 | params->decompress_threads = s->parameters.decompress_threads; |
dc14a470 KZ |
913 | params->has_throttle_trigger_threshold = true; |
914 | params->throttle_trigger_threshold = s->parameters.throttle_trigger_threshold; | |
de63ab61 | 915 | params->has_cpu_throttle_initial = true; |
2594f56d | 916 | params->cpu_throttle_initial = s->parameters.cpu_throttle_initial; |
de63ab61 | 917 | params->has_cpu_throttle_increment = true; |
2594f56d | 918 | params->cpu_throttle_increment = s->parameters.cpu_throttle_increment; |
cbbf8182 KZ |
919 | params->has_cpu_throttle_tailslow = true; |
920 | params->cpu_throttle_tailslow = s->parameters.cpu_throttle_tailslow; | |
8cc99dcd | 921 | params->has_tls_creds = true; |
69ef1f36 | 922 | params->tls_creds = g_strdup(s->parameters.tls_creds); |
8cc99dcd | 923 | params->has_tls_hostname = true; |
69ef1f36 | 924 | params->tls_hostname = g_strdup(s->parameters.tls_hostname); |
d2f1d29b | 925 | params->has_tls_authz = true; |
7cd75cbd MZ |
926 | params->tls_authz = g_strdup(s->parameters.tls_authz ? |
927 | s->parameters.tls_authz : ""); | |
2ff30257 AA |
928 | params->has_max_bandwidth = true; |
929 | params->max_bandwidth = s->parameters.max_bandwidth; | |
930 | params->has_downtime_limit = true; | |
931 | params->downtime_limit = s->parameters.downtime_limit; | |
fe39a4d4 | 932 | params->has_x_checkpoint_delay = true; |
68b53591 | 933 | params->x_checkpoint_delay = s->parameters.x_checkpoint_delay; |
2833c59b JQ |
934 | params->has_block_incremental = true; |
935 | params->block_incremental = s->parameters.block_incremental; | |
cbfd6c95 JQ |
936 | params->has_multifd_channels = true; |
937 | params->multifd_channels = s->parameters.multifd_channels; | |
96eef042 JQ |
938 | params->has_multifd_compression = true; |
939 | params->multifd_compression = s->parameters.multifd_compression; | |
9004db48 JQ |
940 | params->has_multifd_zlib_level = true; |
941 | params->multifd_zlib_level = s->parameters.multifd_zlib_level; | |
6a9ad154 JQ |
942 | params->has_multifd_zstd_level = true; |
943 | params->multifd_zstd_level = s->parameters.multifd_zstd_level; | |
73af8dd8 JQ |
944 | params->has_xbzrle_cache_size = true; |
945 | params->xbzrle_cache_size = s->parameters.xbzrle_cache_size; | |
7e555c6c DDAG |
946 | params->has_max_postcopy_bandwidth = true; |
947 | params->max_postcopy_bandwidth = s->parameters.max_postcopy_bandwidth; | |
4cbc9c7f LQ |
948 | params->has_max_cpu_throttle = true; |
949 | params->max_cpu_throttle = s->parameters.max_cpu_throttle; | |
ee3d96ba DDAG |
950 | params->has_announce_initial = true; |
951 | params->announce_initial = s->parameters.announce_initial; | |
952 | params->has_announce_max = true; | |
953 | params->announce_max = s->parameters.announce_max; | |
954 | params->has_announce_rounds = true; | |
955 | params->announce_rounds = s->parameters.announce_rounds; | |
956 | params->has_announce_step = true; | |
957 | params->announce_step = s->parameters.announce_step; | |
85de8323 | 958 | |
31e4c354 HR |
959 | if (s->parameters.has_block_bitmap_mapping) { |
960 | params->has_block_bitmap_mapping = true; | |
961 | params->block_bitmap_mapping = | |
962 | QAPI_CLONE(BitmapMigrationNodeAliasList, | |
963 | s->parameters.block_bitmap_mapping); | |
964 | } | |
965 | ||
85de8323 LL |
966 | return params; |
967 | } | |
968 | ||
ee3d96ba DDAG |
969 | AnnounceParameters *migrate_announce_params(void) |
970 | { | |
971 | static AnnounceParameters ap; | |
972 | ||
973 | MigrationState *s = migrate_get_current(); | |
974 | ||
975 | ap.initial = s->parameters.announce_initial; | |
976 | ap.max = s->parameters.announce_max; | |
977 | ap.rounds = s->parameters.announce_rounds; | |
978 | ap.step = s->parameters.announce_step; | |
979 | ||
980 | return ≈ | |
981 | } | |
982 | ||
f6844b99 DDAG |
983 | /* |
984 | * Return true if we're already in the middle of a migration | |
985 | * (i.e. any of the active or setup states) | |
986 | */ | |
3d63da16 | 987 | bool migration_is_setup_or_active(int state) |
f6844b99 DDAG |
988 | { |
989 | switch (state) { | |
990 | case MIGRATION_STATUS_ACTIVE: | |
9ec055ae | 991 | case MIGRATION_STATUS_POSTCOPY_ACTIVE: |
a688d2c1 | 992 | case MIGRATION_STATUS_POSTCOPY_PAUSED: |
135b87b4 | 993 | case MIGRATION_STATUS_POSTCOPY_RECOVER: |
f6844b99 | 994 | case MIGRATION_STATUS_SETUP: |
31e06077 DDAG |
995 | case MIGRATION_STATUS_PRE_SWITCHOVER: |
996 | case MIGRATION_STATUS_DEVICE: | |
c7e0acd5 | 997 | case MIGRATION_STATUS_WAIT_UNPLUG: |
19dd408a | 998 | case MIGRATION_STATUS_COLO: |
f6844b99 DDAG |
999 | return true; |
1000 | ||
1001 | default: | |
1002 | return false; | |
1003 | ||
1004 | } | |
1005 | } | |
1006 | ||
392d87e2 JQ |
1007 | bool migration_is_running(int state) |
1008 | { | |
1009 | switch (state) { | |
1010 | case MIGRATION_STATUS_ACTIVE: | |
1011 | case MIGRATION_STATUS_POSTCOPY_ACTIVE: | |
1012 | case MIGRATION_STATUS_POSTCOPY_PAUSED: | |
1013 | case MIGRATION_STATUS_POSTCOPY_RECOVER: | |
1014 | case MIGRATION_STATUS_SETUP: | |
1015 | case MIGRATION_STATUS_PRE_SWITCHOVER: | |
1016 | case MIGRATION_STATUS_DEVICE: | |
1017 | case MIGRATION_STATUS_WAIT_UNPLUG: | |
1018 | case MIGRATION_STATUS_CANCELLING: | |
392d87e2 JQ |
1019 | return true; |
1020 | ||
1021 | default: | |
1022 | return false; | |
1023 | ||
1024 | } | |
1025 | } | |
1026 | ||
640dfb14 WY |
1027 | static void populate_time_info(MigrationInfo *info, MigrationState *s) |
1028 | { | |
1029 | info->has_status = true; | |
1030 | info->has_setup_time = true; | |
1031 | info->setup_time = s->setup_time; | |
1032 | if (s->state == MIGRATION_STATUS_COMPLETED) { | |
1033 | info->has_total_time = true; | |
1034 | info->total_time = s->total_time; | |
1035 | info->has_downtime = true; | |
1036 | info->downtime = s->downtime; | |
1037 | } else { | |
1038 | info->has_total_time = true; | |
1039 | info->total_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME) - | |
1040 | s->start_time; | |
1041 | info->has_expected_downtime = true; | |
1042 | info->expected_downtime = s->expected_downtime; | |
1043 | } | |
1044 | } | |
1045 | ||
a22463a5 DDAG |
1046 | static void populate_ram_info(MigrationInfo *info, MigrationState *s) |
1047 | { | |
144fa06b JQ |
1048 | size_t page_size = qemu_target_page_size(); |
1049 | ||
a22463a5 DDAG |
1050 | info->has_ram = true; |
1051 | info->ram = g_malloc0(sizeof(*info->ram)); | |
9360447d | 1052 | info->ram->transferred = ram_counters.transferred; |
a22463a5 | 1053 | info->ram->total = ram_bytes_total(); |
9360447d | 1054 | info->ram->duplicate = ram_counters.duplicate; |
bedf53c1 JQ |
1055 | /* legacy value. It is not used anymore */ |
1056 | info->ram->skipped = 0; | |
9360447d | 1057 | info->ram->normal = ram_counters.normal; |
144fa06b | 1058 | info->ram->normal_bytes = ram_counters.normal * page_size; |
a22463a5 | 1059 | info->ram->mbps = s->mbps; |
9360447d | 1060 | info->ram->dirty_sync_count = ram_counters.dirty_sync_count; |
cf20c897 LB |
1061 | info->ram->dirty_sync_missed_zero_copy = |
1062 | ram_counters.dirty_sync_missed_zero_copy; | |
9360447d | 1063 | info->ram->postcopy_requests = ram_counters.postcopy_requests; |
144fa06b | 1064 | info->ram->page_size = page_size; |
a61c45bd | 1065 | info->ram->multifd_bytes = ram_counters.multifd_bytes; |
aecbfe9c | 1066 | info->ram->pages_per_second = s->pages_per_second; |
ae680668 DE |
1067 | info->ram->precopy_bytes = ram_counters.precopy_bytes; |
1068 | info->ram->downtime_bytes = ram_counters.downtime_bytes; | |
1069 | info->ram->postcopy_bytes = ram_counters.postcopy_bytes; | |
a22463a5 | 1070 | |
114f5aee JQ |
1071 | if (migrate_use_xbzrle()) { |
1072 | info->has_xbzrle_cache = true; | |
1073 | info->xbzrle_cache = g_malloc0(sizeof(*info->xbzrle_cache)); | |
1074 | info->xbzrle_cache->cache_size = migrate_xbzrle_cache_size(); | |
9360447d JQ |
1075 | info->xbzrle_cache->bytes = xbzrle_counters.bytes; |
1076 | info->xbzrle_cache->pages = xbzrle_counters.pages; | |
1077 | info->xbzrle_cache->cache_miss = xbzrle_counters.cache_miss; | |
1078 | info->xbzrle_cache->cache_miss_rate = xbzrle_counters.cache_miss_rate; | |
e460a4b1 | 1079 | info->xbzrle_cache->encoding_rate = xbzrle_counters.encoding_rate; |
9360447d | 1080 | info->xbzrle_cache->overflow = xbzrle_counters.overflow; |
114f5aee JQ |
1081 | } |
1082 | ||
76e03000 XG |
1083 | if (migrate_use_compression()) { |
1084 | info->has_compression = true; | |
1085 | info->compression = g_malloc0(sizeof(*info->compression)); | |
1086 | info->compression->pages = compression_counters.pages; | |
1087 | info->compression->busy = compression_counters.busy; | |
1088 | info->compression->busy_rate = compression_counters.busy_rate; | |
1089 | info->compression->compressed_size = | |
1090 | compression_counters.compressed_size; | |
1091 | info->compression->compression_rate = | |
1092 | compression_counters.compression_rate; | |
1093 | } | |
1094 | ||
338182c8 JQ |
1095 | if (cpu_throttle_active()) { |
1096 | info->has_cpu_throttle_percentage = true; | |
1097 | info->cpu_throttle_percentage = cpu_throttle_get_percentage(); | |
1098 | } | |
1099 | ||
a22463a5 DDAG |
1100 | if (s->state != MIGRATION_STATUS_COMPLETED) { |
1101 | info->ram->remaining = ram_bytes_remaining(); | |
9360447d | 1102 | info->ram->dirty_pages_rate = ram_counters.dirty_pages_rate; |
a22463a5 DDAG |
1103 | } |
1104 | } | |
1105 | ||
930ac04c JQ |
1106 | static void populate_disk_info(MigrationInfo *info) |
1107 | { | |
1108 | if (blk_mig_active()) { | |
1109 | info->has_disk = true; | |
1110 | info->disk = g_malloc0(sizeof(*info->disk)); | |
1111 | info->disk->transferred = blk_mig_bytes_transferred(); | |
1112 | info->disk->remaining = blk_mig_bytes_remaining(); | |
1113 | info->disk->total = blk_mig_bytes_total(); | |
1114 | } | |
1115 | } | |
1116 | ||
65ace060 | 1117 | static void fill_source_migration_info(MigrationInfo *info) |
5bb7910a | 1118 | { |
17549e84 | 1119 | MigrationState *s = migrate_get_current(); |
552de79b | 1120 | int state = qatomic_read(&s->state); |
372043f3 | 1121 | GSList *cur_blocker = migration_blockers; |
17549e84 | 1122 | |
3af8554b | 1123 | info->blocked_reasons = NULL; |
3af8554b | 1124 | |
372043f3 MA |
1125 | /* |
1126 | * There are two types of reasons a migration might be blocked; | |
1127 | * a) devices marked in VMState as non-migratable, and | |
1128 | * b) Explicit migration blockers | |
1129 | * We need to add both of them here. | |
1130 | */ | |
1131 | qemu_savevm_non_migratable_list(&info->blocked_reasons); | |
3af8554b | 1132 | |
372043f3 MA |
1133 | while (cur_blocker) { |
1134 | QAPI_LIST_PREPEND(info->blocked_reasons, | |
1135 | g_strdup(error_get_pretty(cur_blocker->data))); | |
1136 | cur_blocker = g_slist_next(cur_blocker); | |
3af8554b | 1137 | } |
372043f3 | 1138 | info->has_blocked_reasons = info->blocked_reasons != NULL; |
3af8554b | 1139 | |
552de79b | 1140 | switch (state) { |
31194731 | 1141 | case MIGRATION_STATUS_NONE: |
17549e84 | 1142 | /* no migration has happened ever */ |
65ace060 AP |
1143 | /* do not overwrite destination migration status */ |
1144 | return; | |
31194731 | 1145 | case MIGRATION_STATUS_SETUP: |
29ae8a41 | 1146 | info->has_status = true; |
ed4fbd10 | 1147 | info->has_total_time = false; |
29ae8a41 | 1148 | break; |
31194731 HZ |
1149 | case MIGRATION_STATUS_ACTIVE: |
1150 | case MIGRATION_STATUS_CANCELLING: | |
9ec055ae | 1151 | case MIGRATION_STATUS_POSTCOPY_ACTIVE: |
31e06077 DDAG |
1152 | case MIGRATION_STATUS_PRE_SWITCHOVER: |
1153 | case MIGRATION_STATUS_DEVICE: | |
a688d2c1 | 1154 | case MIGRATION_STATUS_POSTCOPY_PAUSED: |
135b87b4 | 1155 | case MIGRATION_STATUS_POSTCOPY_RECOVER: |
640dfb14 WY |
1156 | /* TODO add some postcopy stats */ |
1157 | populate_time_info(info, s); | |
a22463a5 | 1158 | populate_ram_info(info, s); |
930ac04c | 1159 | populate_disk_info(info); |
3710586c | 1160 | populate_vfio_info(info); |
17549e84 | 1161 | break; |
0b827d5e HZ |
1162 | case MIGRATION_STATUS_COLO: |
1163 | info->has_status = true; | |
1164 | /* TODO: display COLO specific information (checkpoint info etc.) */ | |
1165 | break; | |
31194731 | 1166 | case MIGRATION_STATUS_COMPLETED: |
640dfb14 | 1167 | populate_time_info(info, s); |
a22463a5 | 1168 | populate_ram_info(info, s); |
3710586c | 1169 | populate_vfio_info(info); |
17549e84 | 1170 | break; |
31194731 | 1171 | case MIGRATION_STATUS_FAILED: |
791e7c82 | 1172 | info->has_status = true; |
d59ce6f3 DB |
1173 | if (s->error) { |
1174 | info->has_error_desc = true; | |
1175 | info->error_desc = g_strdup(error_get_pretty(s->error)); | |
1176 | } | |
17549e84 | 1177 | break; |
31194731 | 1178 | case MIGRATION_STATUS_CANCELLED: |
791e7c82 | 1179 | info->has_status = true; |
17549e84 | 1180 | break; |
c7e0acd5 JF |
1181 | case MIGRATION_STATUS_WAIT_UNPLUG: |
1182 | info->has_status = true; | |
1183 | break; | |
5bb7910a | 1184 | } |
552de79b | 1185 | info->status = state; |
5bb7910a AL |
1186 | } |
1187 | ||
6e8c25b4 AG |
1188 | typedef enum WriteTrackingSupport { |
1189 | WT_SUPPORT_UNKNOWN = 0, | |
1190 | WT_SUPPORT_ABSENT, | |
1191 | WT_SUPPORT_AVAILABLE, | |
1192 | WT_SUPPORT_COMPATIBLE | |
1193 | } WriteTrackingSupport; | |
1194 | ||
1195 | static | |
1196 | WriteTrackingSupport migrate_query_write_tracking(void) | |
1197 | { | |
1198 | /* Check if kernel supports required UFFD features */ | |
1199 | if (!ram_write_tracking_available()) { | |
1200 | return WT_SUPPORT_ABSENT; | |
1201 | } | |
1202 | /* | |
1203 | * Check if current memory configuration is | |
1204 | * compatible with required UFFD features. | |
1205 | */ | |
1206 | if (!ram_write_tracking_compatible()) { | |
1207 | return WT_SUPPORT_AVAILABLE; | |
1208 | } | |
1209 | ||
1210 | return WT_SUPPORT_COMPATIBLE; | |
1211 | } | |
1212 | ||
4a84214e PX |
1213 | /** |
1214 | * @migration_caps_check - check capability validity | |
1215 | * | |
1216 | * @cap_list: old capability list, array of bool | |
1217 | * @params: new capabilities to be applied soon | |
1218 | * @errp: set *errp if the check failed, with reason | |
1219 | * | |
1220 | * Returns true if check passed, otherwise false. | |
1221 | */ | |
1222 | static bool migrate_caps_check(bool *cap_list, | |
1223 | MigrationCapabilityStatusList *params, | |
1224 | Error **errp) | |
00458433 | 1225 | { |
00458433 | 1226 | MigrationCapabilityStatusList *cap; |
4a84214e | 1227 | bool old_postcopy_cap; |
d7651f15 | 1228 | MigrationIncomingState *mis = migration_incoming_get_current(); |
00458433 | 1229 | |
4a84214e | 1230 | old_postcopy_cap = cap_list[MIGRATION_CAPABILITY_POSTCOPY_RAM]; |
00458433 OW |
1231 | |
1232 | for (cap = params; cap; cap = cap->next) { | |
4a84214e PX |
1233 | cap_list[cap->value->capability] = cap->value->state; |
1234 | } | |
1235 | ||
ed1701c6 | 1236 | #ifndef CONFIG_LIVE_BLOCK_MIGRATION |
4a84214e PX |
1237 | if (cap_list[MIGRATION_CAPABILITY_BLOCK]) { |
1238 | error_setg(errp, "QEMU compiled without old-style (blk/-b, inc/-i) " | |
1239 | "block migration"); | |
1240 | error_append_hint(errp, "Use drive_mirror+NBD instead.\n"); | |
1241 | return false; | |
00458433 | 1242 | } |
4a84214e | 1243 | #endif |
53dd370c | 1244 | |
7e934f5b ZC |
1245 | #ifndef CONFIG_REPLICATION |
1246 | if (cap_list[MIGRATION_CAPABILITY_X_COLO]) { | |
1247 | error_setg(errp, "QEMU compiled without replication module" | |
1248 | " can't enable COLO"); | |
1249 | error_append_hint(errp, "Please enable replication before COLO.\n"); | |
1250 | return false; | |
1251 | } | |
1252 | #endif | |
1253 | ||
4a84214e | 1254 | if (cap_list[MIGRATION_CAPABILITY_POSTCOPY_RAM]) { |
096631bd DDAG |
1255 | /* This check is reasonably expensive, so only when it's being |
1256 | * set the first time, also it's only the destination that needs | |
1257 | * special support. | |
1258 | */ | |
1259 | if (!old_postcopy_cap && runstate_check(RUN_STATE_INMIGRATE) && | |
d7651f15 | 1260 | !postcopy_ram_supported_by_host(mis)) { |
096631bd DDAG |
1261 | /* postcopy_ram_supported_by_host will have emitted a more |
1262 | * detailed message | |
1263 | */ | |
4a84214e PX |
1264 | error_setg(errp, "Postcopy is not supported"); |
1265 | return false; | |
096631bd | 1266 | } |
18269069 YK |
1267 | |
1268 | if (cap_list[MIGRATION_CAPABILITY_X_IGNORE_SHARED]) { | |
1269 | error_setg(errp, "Postcopy is not compatible with ignore-shared"); | |
1270 | return false; | |
1271 | } | |
53dd370c | 1272 | } |
4a84214e | 1273 | |
6e8c25b4 AG |
1274 | if (cap_list[MIGRATION_CAPABILITY_BACKGROUND_SNAPSHOT]) { |
1275 | WriteTrackingSupport wt_support; | |
1276 | int idx; | |
1277 | /* | |
1278 | * Check if 'background-snapshot' capability is supported by | |
1279 | * host kernel and compatible with guest memory configuration. | |
1280 | */ | |
1281 | wt_support = migrate_query_write_tracking(); | |
1282 | if (wt_support < WT_SUPPORT_AVAILABLE) { | |
1283 | error_setg(errp, "Background-snapshot is not supported by host kernel"); | |
1284 | return false; | |
1285 | } | |
1286 | if (wt_support < WT_SUPPORT_COMPATIBLE) { | |
1287 | error_setg(errp, "Background-snapshot is not compatible " | |
1288 | "with guest memory configuration"); | |
1289 | return false; | |
1290 | } | |
1291 | ||
1292 | /* | |
1293 | * Check if there are any migration capabilities | |
1294 | * incompatible with 'background-snapshot'. | |
1295 | */ | |
1296 | for (idx = 0; idx < check_caps_background_snapshot.size; idx++) { | |
1297 | int incomp_cap = check_caps_background_snapshot.caps[idx]; | |
1298 | if (cap_list[incomp_cap]) { | |
1299 | error_setg(errp, | |
1300 | "Background-snapshot is not compatible with %s", | |
1301 | MigrationCapability_str(incomp_cap)); | |
1302 | return false; | |
1303 | } | |
1304 | } | |
1305 | } | |
1306 | ||
1abaec9a LB |
1307 | #ifdef CONFIG_LINUX |
1308 | if (cap_list[MIGRATION_CAPABILITY_ZERO_COPY_SEND] && | |
1309 | (!cap_list[MIGRATION_CAPABILITY_MULTIFD] || | |
90eb69e4 LB |
1310 | cap_list[MIGRATION_CAPABILITY_COMPRESS] || |
1311 | cap_list[MIGRATION_CAPABILITY_XBZRLE] || | |
1312 | migrate_multifd_compression() || | |
1abaec9a LB |
1313 | migrate_use_tls())) { |
1314 | error_setg(errp, | |
1315 | "Zero copy only available for non-compressed non-TLS multifd migration"); | |
1316 | return false; | |
1317 | } | |
1318 | #else | |
1319 | if (cap_list[MIGRATION_CAPABILITY_ZERO_COPY_SEND]) { | |
1320 | error_setg(errp, | |
1321 | "Zero copy currently only available on Linux"); | |
1322 | return false; | |
1323 | } | |
1324 | #endif | |
1325 | ||
1326 | ||
5ad15e86 LZ |
1327 | /* incoming side only */ |
1328 | if (runstate_check(RUN_STATE_INMIGRATE) && | |
f444eeda | 1329 | !migrate_multi_channels_is_allowed() && |
5ad15e86 LZ |
1330 | cap_list[MIGRATION_CAPABILITY_MULTIFD]) { |
1331 | error_setg(errp, "multifd is not supported by current protocol"); | |
1332 | return false; | |
1333 | } | |
1334 | ||
ce5b0f4a PX |
1335 | if (cap_list[MIGRATION_CAPABILITY_POSTCOPY_PREEMPT]) { |
1336 | if (!cap_list[MIGRATION_CAPABILITY_POSTCOPY_RAM]) { | |
1337 | error_setg(errp, "Postcopy preempt requires postcopy-ram"); | |
1338 | return false; | |
1339 | } | |
1340 | } | |
1341 | ||
4a84214e PX |
1342 | return true; |
1343 | } | |
1344 | ||
65ace060 AP |
1345 | static void fill_destination_migration_info(MigrationInfo *info) |
1346 | { | |
1347 | MigrationIncomingState *mis = migration_incoming_get_current(); | |
1348 | ||
9aca82ba JQ |
1349 | if (mis->socket_address_list) { |
1350 | info->has_socket_address = true; | |
1351 | info->socket_address = | |
1352 | QAPI_CLONE(SocketAddressList, mis->socket_address_list); | |
1353 | } | |
1354 | ||
65ace060 AP |
1355 | switch (mis->state) { |
1356 | case MIGRATION_STATUS_NONE: | |
1357 | return; | |
65ace060 AP |
1358 | case MIGRATION_STATUS_SETUP: |
1359 | case MIGRATION_STATUS_CANCELLING: | |
1360 | case MIGRATION_STATUS_CANCELLED: | |
1361 | case MIGRATION_STATUS_ACTIVE: | |
1362 | case MIGRATION_STATUS_POSTCOPY_ACTIVE: | |
3c9928d9 PX |
1363 | case MIGRATION_STATUS_POSTCOPY_PAUSED: |
1364 | case MIGRATION_STATUS_POSTCOPY_RECOVER: | |
65ace060 AP |
1365 | case MIGRATION_STATUS_FAILED: |
1366 | case MIGRATION_STATUS_COLO: | |
1367 | info->has_status = true; | |
1368 | break; | |
1369 | case MIGRATION_STATUS_COMPLETED: | |
1370 | info->has_status = true; | |
1371 | fill_destination_postcopy_migration_info(info); | |
1372 | break; | |
1373 | } | |
1374 | info->status = mis->state; | |
1375 | } | |
1376 | ||
1377 | MigrationInfo *qmp_query_migrate(Error **errp) | |
1378 | { | |
1379 | MigrationInfo *info = g_malloc0(sizeof(*info)); | |
1380 | ||
1381 | fill_destination_migration_info(info); | |
1382 | fill_source_migration_info(info); | |
1383 | ||
1384 | return info; | |
1385 | } | |
1386 | ||
4a84214e PX |
1387 | void qmp_migrate_set_capabilities(MigrationCapabilityStatusList *params, |
1388 | Error **errp) | |
1389 | { | |
1390 | MigrationState *s = migrate_get_current(); | |
1391 | MigrationCapabilityStatusList *cap; | |
dd0ee30c | 1392 | bool cap_list[MIGRATION_CAPABILITY__MAX]; |
4a84214e | 1393 | |
392d87e2 | 1394 | if (migration_is_running(s->state)) { |
4a84214e PX |
1395 | error_setg(errp, QERR_MIGRATION_ACTIVE); |
1396 | return; | |
1397 | } | |
1398 | ||
dd0ee30c PX |
1399 | memcpy(cap_list, s->enabled_capabilities, sizeof(cap_list)); |
1400 | if (!migrate_caps_check(cap_list, params, errp)) { | |
4a84214e PX |
1401 | return; |
1402 | } | |
1403 | ||
1404 | for (cap = params; cap; cap = cap->next) { | |
1405 | s->enabled_capabilities[cap->value->capability] = cap->value->state; | |
1406 | } | |
00458433 OW |
1407 | } |
1408 | ||
16d063bc PX |
1409 | /* |
1410 | * Check whether the parameters are valid. Error will be put into errp | |
1411 | * (if provided). Return true if valid, otherwise false. | |
1412 | */ | |
1413 | static bool migrate_params_check(MigrationParameters *params, Error **errp) | |
85de8323 | 1414 | { |
7f375e04 | 1415 | if (params->has_compress_level && |
741d4086 | 1416 | (params->compress_level > 9)) { |
c6bd8c70 | 1417 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "compress_level", |
54270c45 | 1418 | "a value between 0 and 9"); |
16d063bc | 1419 | return false; |
85de8323 | 1420 | } |
16d063bc | 1421 | |
741d4086 | 1422 | if (params->has_compress_threads && (params->compress_threads < 1)) { |
c6bd8c70 MA |
1423 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, |
1424 | "compress_threads", | |
54270c45 | 1425 | "a value between 1 and 255"); |
16d063bc | 1426 | return false; |
85de8323 | 1427 | } |
16d063bc | 1428 | |
741d4086 | 1429 | if (params->has_decompress_threads && (params->decompress_threads < 1)) { |
c6bd8c70 MA |
1430 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, |
1431 | "decompress_threads", | |
54270c45 | 1432 | "a value between 1 and 255"); |
16d063bc | 1433 | return false; |
85de8323 | 1434 | } |
16d063bc | 1435 | |
dc14a470 KZ |
1436 | if (params->has_throttle_trigger_threshold && |
1437 | (params->throttle_trigger_threshold < 1 || | |
1438 | params->throttle_trigger_threshold > 100)) { | |
1439 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, | |
1440 | "throttle_trigger_threshold", | |
1441 | "an integer in the range of 1 to 100"); | |
1442 | return false; | |
1443 | } | |
1444 | ||
7f375e04 EB |
1445 | if (params->has_cpu_throttle_initial && |
1446 | (params->cpu_throttle_initial < 1 || | |
1447 | params->cpu_throttle_initial > 99)) { | |
1626fee3 | 1448 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, |
d85a31d1 | 1449 | "cpu_throttle_initial", |
1626fee3 | 1450 | "an integer in the range of 1 to 99"); |
16d063bc | 1451 | return false; |
1626fee3 | 1452 | } |
16d063bc | 1453 | |
7f375e04 EB |
1454 | if (params->has_cpu_throttle_increment && |
1455 | (params->cpu_throttle_increment < 1 || | |
1456 | params->cpu_throttle_increment > 99)) { | |
1626fee3 | 1457 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, |
d85a31d1 | 1458 | "cpu_throttle_increment", |
1626fee3 | 1459 | "an integer in the range of 1 to 99"); |
16d063bc | 1460 | return false; |
1626fee3 | 1461 | } |
16d063bc | 1462 | |
741d4086 | 1463 | if (params->has_max_bandwidth && (params->max_bandwidth > SIZE_MAX)) { |
2ee30cf0 MZ |
1464 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, |
1465 | "max_bandwidth", | |
1466 | "an integer in the range of 0 to "stringify(SIZE_MAX) | |
1467 | " bytes/second"); | |
16d063bc | 1468 | return false; |
2ff30257 | 1469 | } |
16d063bc | 1470 | |
2ff30257 | 1471 | if (params->has_downtime_limit && |
741d4086 | 1472 | (params->downtime_limit > MAX_MIGRATE_DOWNTIME)) { |
2ee30cf0 MZ |
1473 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, |
1474 | "downtime_limit", | |
1475 | "an integer in the range of 0 to " | |
7ac5529a | 1476 | stringify(MAX_MIGRATE_DOWNTIME)" ms"); |
16d063bc | 1477 | return false; |
2ff30257 | 1478 | } |
16d063bc | 1479 | |
741d4086 JQ |
1480 | /* x_checkpoint_delay is now always positive */ |
1481 | ||
cbfd6c95 | 1482 | if (params->has_multifd_channels && (params->multifd_channels < 1)) { |
4075fb1c JQ |
1483 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, |
1484 | "multifd_channels", | |
54270c45 | 1485 | "a value between 1 and 255"); |
4075fb1c JQ |
1486 | return false; |
1487 | } | |
16d063bc | 1488 | |
9004db48 JQ |
1489 | if (params->has_multifd_zlib_level && |
1490 | (params->multifd_zlib_level > 9)) { | |
1491 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "multifd_zlib_level", | |
54270c45 | 1492 | "a value between 0 and 9"); |
9004db48 JQ |
1493 | return false; |
1494 | } | |
1495 | ||
6a9ad154 JQ |
1496 | if (params->has_multifd_zstd_level && |
1497 | (params->multifd_zstd_level > 20)) { | |
1498 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "multifd_zstd_level", | |
54270c45 | 1499 | "a value between 0 and 20"); |
6a9ad154 JQ |
1500 | return false; |
1501 | } | |
1502 | ||
73af8dd8 JQ |
1503 | if (params->has_xbzrle_cache_size && |
1504 | (params->xbzrle_cache_size < qemu_target_page_size() || | |
1505 | !is_power_of_2(params->xbzrle_cache_size))) { | |
1506 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, | |
1507 | "xbzrle_cache_size", | |
54270c45 | 1508 | "a power of two no less than the target page size"); |
73af8dd8 JQ |
1509 | return false; |
1510 | } | |
1511 | ||
4cbc9c7f LQ |
1512 | if (params->has_max_cpu_throttle && |
1513 | (params->max_cpu_throttle < params->cpu_throttle_initial || | |
1514 | params->max_cpu_throttle > 99)) { | |
1515 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, | |
1516 | "max_cpu_throttle", | |
1517 | "an integer in the range of cpu_throttle_initial to 99"); | |
1518 | return false; | |
1519 | } | |
1520 | ||
ee3d96ba DDAG |
1521 | if (params->has_announce_initial && |
1522 | params->announce_initial > 100000) { | |
1523 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, | |
1524 | "announce_initial", | |
54270c45 | 1525 | "a value between 0 and 100000"); |
ee3d96ba DDAG |
1526 | return false; |
1527 | } | |
1528 | if (params->has_announce_max && | |
1529 | params->announce_max > 100000) { | |
1530 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, | |
1531 | "announce_max", | |
54270c45 | 1532 | "a value between 0 and 100000"); |
ee3d96ba DDAG |
1533 | return false; |
1534 | } | |
1535 | if (params->has_announce_rounds && | |
1536 | params->announce_rounds > 1000) { | |
1537 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, | |
1538 | "announce_rounds", | |
54270c45 | 1539 | "a value between 0 and 1000"); |
ee3d96ba DDAG |
1540 | return false; |
1541 | } | |
1542 | if (params->has_announce_step && | |
1543 | (params->announce_step < 1 || | |
1544 | params->announce_step > 10000)) { | |
1545 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, | |
1546 | "announce_step", | |
54270c45 | 1547 | "a value between 0 and 10000"); |
ee3d96ba DDAG |
1548 | return false; |
1549 | } | |
31e4c354 HR |
1550 | |
1551 | if (params->has_block_bitmap_mapping && | |
1552 | !check_dirty_bitmap_mig_alias_map(params->block_bitmap_mapping, errp)) { | |
1553 | error_prepend(errp, "Invalid mapping given for block-bitmap-mapping: "); | |
1554 | return false; | |
1555 | } | |
90eb69e4 LB |
1556 | |
1557 | #ifdef CONFIG_LINUX | |
1558 | if (migrate_use_zero_copy_send() && | |
1559 | ((params->has_multifd_compression && params->multifd_compression) || | |
1560 | (params->has_tls_creds && params->tls_creds && *params->tls_creds))) { | |
1561 | error_setg(errp, | |
1562 | "Zero copy only available for non-compressed non-TLS multifd migration"); | |
1563 | return false; | |
1564 | } | |
1565 | #endif | |
1566 | ||
16d063bc PX |
1567 | return true; |
1568 | } | |
1569 | ||
1bda8b3c MA |
1570 | static void migrate_params_test_apply(MigrateSetParameters *params, |
1571 | MigrationParameters *dest) | |
1572 | { | |
1573 | *dest = migrate_get_current()->parameters; | |
1574 | ||
1575 | /* TODO use QAPI_CLONE() instead of duplicating it inline */ | |
1576 | ||
1577 | if (params->has_compress_level) { | |
1578 | dest->compress_level = params->compress_level; | |
1579 | } | |
1580 | ||
1581 | if (params->has_compress_threads) { | |
1582 | dest->compress_threads = params->compress_threads; | |
1583 | } | |
1584 | ||
1d58872a XG |
1585 | if (params->has_compress_wait_thread) { |
1586 | dest->compress_wait_thread = params->compress_wait_thread; | |
1587 | } | |
1588 | ||
1bda8b3c MA |
1589 | if (params->has_decompress_threads) { |
1590 | dest->decompress_threads = params->decompress_threads; | |
1591 | } | |
1592 | ||
dc14a470 KZ |
1593 | if (params->has_throttle_trigger_threshold) { |
1594 | dest->throttle_trigger_threshold = params->throttle_trigger_threshold; | |
1595 | } | |
1596 | ||
1bda8b3c MA |
1597 | if (params->has_cpu_throttle_initial) { |
1598 | dest->cpu_throttle_initial = params->cpu_throttle_initial; | |
1599 | } | |
1600 | ||
1601 | if (params->has_cpu_throttle_increment) { | |
1602 | dest->cpu_throttle_increment = params->cpu_throttle_increment; | |
1603 | } | |
1604 | ||
cbbf8182 KZ |
1605 | if (params->has_cpu_throttle_tailslow) { |
1606 | dest->cpu_throttle_tailslow = params->cpu_throttle_tailslow; | |
1607 | } | |
1608 | ||
1bda8b3c | 1609 | if (params->has_tls_creds) { |
01fa5598 | 1610 | assert(params->tls_creds->type == QTYPE_QSTRING); |
9728ebfb | 1611 | dest->tls_creds = params->tls_creds->u.s; |
1bda8b3c MA |
1612 | } |
1613 | ||
1614 | if (params->has_tls_hostname) { | |
01fa5598 | 1615 | assert(params->tls_hostname->type == QTYPE_QSTRING); |
9728ebfb | 1616 | dest->tls_hostname = params->tls_hostname->u.s; |
1bda8b3c MA |
1617 | } |
1618 | ||
1619 | if (params->has_max_bandwidth) { | |
1620 | dest->max_bandwidth = params->max_bandwidth; | |
1621 | } | |
1622 | ||
1623 | if (params->has_downtime_limit) { | |
1624 | dest->downtime_limit = params->downtime_limit; | |
1625 | } | |
1626 | ||
1627 | if (params->has_x_checkpoint_delay) { | |
1628 | dest->x_checkpoint_delay = params->x_checkpoint_delay; | |
1629 | } | |
1630 | ||
1631 | if (params->has_block_incremental) { | |
1632 | dest->block_incremental = params->block_incremental; | |
1633 | } | |
cbfd6c95 JQ |
1634 | if (params->has_multifd_channels) { |
1635 | dest->multifd_channels = params->multifd_channels; | |
5e7577a1 | 1636 | } |
96eef042 JQ |
1637 | if (params->has_multifd_compression) { |
1638 | dest->multifd_compression = params->multifd_compression; | |
1639 | } | |
73af8dd8 JQ |
1640 | if (params->has_xbzrle_cache_size) { |
1641 | dest->xbzrle_cache_size = params->xbzrle_cache_size; | |
1642 | } | |
7e555c6c DDAG |
1643 | if (params->has_max_postcopy_bandwidth) { |
1644 | dest->max_postcopy_bandwidth = params->max_postcopy_bandwidth; | |
1645 | } | |
4cbc9c7f LQ |
1646 | if (params->has_max_cpu_throttle) { |
1647 | dest->max_cpu_throttle = params->max_cpu_throttle; | |
1648 | } | |
ee3d96ba DDAG |
1649 | if (params->has_announce_initial) { |
1650 | dest->announce_initial = params->announce_initial; | |
1651 | } | |
1652 | if (params->has_announce_max) { | |
1653 | dest->announce_max = params->announce_max; | |
1654 | } | |
1655 | if (params->has_announce_rounds) { | |
1656 | dest->announce_rounds = params->announce_rounds; | |
1657 | } | |
1658 | if (params->has_announce_step) { | |
1659 | dest->announce_step = params->announce_step; | |
1660 | } | |
31e4c354 HR |
1661 | |
1662 | if (params->has_block_bitmap_mapping) { | |
1663 | dest->has_block_bitmap_mapping = true; | |
1664 | dest->block_bitmap_mapping = params->block_bitmap_mapping; | |
1665 | } | |
1bda8b3c MA |
1666 | } |
1667 | ||
73af8dd8 | 1668 | static void migrate_params_apply(MigrateSetParameters *params, Error **errp) |
16d063bc PX |
1669 | { |
1670 | MigrationState *s = migrate_get_current(); | |
1671 | ||
e87fae4c MA |
1672 | /* TODO use QAPI_CLONE() instead of duplicating it inline */ |
1673 | ||
7f375e04 EB |
1674 | if (params->has_compress_level) { |
1675 | s->parameters.compress_level = params->compress_level; | |
85de8323 | 1676 | } |
476c72aa | 1677 | |
7f375e04 EB |
1678 | if (params->has_compress_threads) { |
1679 | s->parameters.compress_threads = params->compress_threads; | |
85de8323 | 1680 | } |
476c72aa | 1681 | |
1d58872a XG |
1682 | if (params->has_compress_wait_thread) { |
1683 | s->parameters.compress_wait_thread = params->compress_wait_thread; | |
1684 | } | |
1685 | ||
7f375e04 EB |
1686 | if (params->has_decompress_threads) { |
1687 | s->parameters.decompress_threads = params->decompress_threads; | |
85de8323 | 1688 | } |
476c72aa | 1689 | |
dc14a470 KZ |
1690 | if (params->has_throttle_trigger_threshold) { |
1691 | s->parameters.throttle_trigger_threshold = params->throttle_trigger_threshold; | |
1692 | } | |
1693 | ||
7f375e04 EB |
1694 | if (params->has_cpu_throttle_initial) { |
1695 | s->parameters.cpu_throttle_initial = params->cpu_throttle_initial; | |
1626fee3 | 1696 | } |
476c72aa | 1697 | |
7f375e04 EB |
1698 | if (params->has_cpu_throttle_increment) { |
1699 | s->parameters.cpu_throttle_increment = params->cpu_throttle_increment; | |
1626fee3 | 1700 | } |
476c72aa | 1701 | |
cbbf8182 KZ |
1702 | if (params->has_cpu_throttle_tailslow) { |
1703 | s->parameters.cpu_throttle_tailslow = params->cpu_throttle_tailslow; | |
1704 | } | |
1705 | ||
7f375e04 | 1706 | if (params->has_tls_creds) { |
69ef1f36 | 1707 | g_free(s->parameters.tls_creds); |
01fa5598 MA |
1708 | assert(params->tls_creds->type == QTYPE_QSTRING); |
1709 | s->parameters.tls_creds = g_strdup(params->tls_creds->u.s); | |
69ef1f36 | 1710 | } |
476c72aa | 1711 | |
7f375e04 | 1712 | if (params->has_tls_hostname) { |
69ef1f36 | 1713 | g_free(s->parameters.tls_hostname); |
01fa5598 MA |
1714 | assert(params->tls_hostname->type == QTYPE_QSTRING); |
1715 | s->parameters.tls_hostname = g_strdup(params->tls_hostname->u.s); | |
69ef1f36 | 1716 | } |
476c72aa | 1717 | |
d2f1d29b DB |
1718 | if (params->has_tls_authz) { |
1719 | g_free(s->parameters.tls_authz); | |
1720 | assert(params->tls_authz->type == QTYPE_QSTRING); | |
1721 | s->parameters.tls_authz = g_strdup(params->tls_authz->u.s); | |
1722 | } | |
1723 | ||
2ff30257 AA |
1724 | if (params->has_max_bandwidth) { |
1725 | s->parameters.max_bandwidth = params->max_bandwidth; | |
c38c1c14 | 1726 | if (s->to_dst_file && !migration_in_postcopy()) { |
2ff30257 AA |
1727 | qemu_file_set_rate_limit(s->to_dst_file, |
1728 | s->parameters.max_bandwidth / XFER_LIMIT_RATIO); | |
1729 | } | |
1730 | } | |
476c72aa | 1731 | |
2ff30257 AA |
1732 | if (params->has_downtime_limit) { |
1733 | s->parameters.downtime_limit = params->downtime_limit; | |
1734 | } | |
68b53591 HZ |
1735 | |
1736 | if (params->has_x_checkpoint_delay) { | |
1737 | s->parameters.x_checkpoint_delay = params->x_checkpoint_delay; | |
479125d5 HZ |
1738 | if (migration_in_colo_state()) { |
1739 | colo_checkpoint_notify(s); | |
1740 | } | |
68b53591 | 1741 | } |
476c72aa | 1742 | |
2833c59b JQ |
1743 | if (params->has_block_incremental) { |
1744 | s->parameters.block_incremental = params->block_incremental; | |
1745 | } | |
cbfd6c95 JQ |
1746 | if (params->has_multifd_channels) { |
1747 | s->parameters.multifd_channels = params->multifd_channels; | |
4075fb1c | 1748 | } |
96eef042 JQ |
1749 | if (params->has_multifd_compression) { |
1750 | s->parameters.multifd_compression = params->multifd_compression; | |
1751 | } | |
73af8dd8 JQ |
1752 | if (params->has_xbzrle_cache_size) { |
1753 | s->parameters.xbzrle_cache_size = params->xbzrle_cache_size; | |
1754 | xbzrle_cache_resize(params->xbzrle_cache_size, errp); | |
1755 | } | |
7e555c6c DDAG |
1756 | if (params->has_max_postcopy_bandwidth) { |
1757 | s->parameters.max_postcopy_bandwidth = params->max_postcopy_bandwidth; | |
c38c1c14 DDAG |
1758 | if (s->to_dst_file && migration_in_postcopy()) { |
1759 | qemu_file_set_rate_limit(s->to_dst_file, | |
1760 | s->parameters.max_postcopy_bandwidth / XFER_LIMIT_RATIO); | |
1761 | } | |
7e555c6c | 1762 | } |
4cbc9c7f LQ |
1763 | if (params->has_max_cpu_throttle) { |
1764 | s->parameters.max_cpu_throttle = params->max_cpu_throttle; | |
1765 | } | |
ee3d96ba DDAG |
1766 | if (params->has_announce_initial) { |
1767 | s->parameters.announce_initial = params->announce_initial; | |
1768 | } | |
1769 | if (params->has_announce_max) { | |
1770 | s->parameters.announce_max = params->announce_max; | |
1771 | } | |
1772 | if (params->has_announce_rounds) { | |
1773 | s->parameters.announce_rounds = params->announce_rounds; | |
1774 | } | |
1775 | if (params->has_announce_step) { | |
1776 | s->parameters.announce_step = params->announce_step; | |
1777 | } | |
31e4c354 HR |
1778 | |
1779 | if (params->has_block_bitmap_mapping) { | |
1780 | qapi_free_BitmapMigrationNodeAliasList( | |
1781 | s->parameters.block_bitmap_mapping); | |
1782 | ||
1783 | s->parameters.has_block_bitmap_mapping = true; | |
1784 | s->parameters.block_bitmap_mapping = | |
1785 | QAPI_CLONE(BitmapMigrationNodeAliasList, | |
1786 | params->block_bitmap_mapping); | |
1787 | } | |
85de8323 LL |
1788 | } |
1789 | ||
1bda8b3c | 1790 | void qmp_migrate_set_parameters(MigrateSetParameters *params, Error **errp) |
476c72aa | 1791 | { |
1bda8b3c MA |
1792 | MigrationParameters tmp; |
1793 | ||
01fa5598 MA |
1794 | /* TODO Rewrite "" to null instead */ |
1795 | if (params->has_tls_creds | |
1796 | && params->tls_creds->type == QTYPE_QNULL) { | |
cb3e7f08 | 1797 | qobject_unref(params->tls_creds->u.n); |
01fa5598 MA |
1798 | params->tls_creds->type = QTYPE_QSTRING; |
1799 | params->tls_creds->u.s = strdup(""); | |
1800 | } | |
1801 | /* TODO Rewrite "" to null instead */ | |
1802 | if (params->has_tls_hostname | |
1803 | && params->tls_hostname->type == QTYPE_QNULL) { | |
cb3e7f08 | 1804 | qobject_unref(params->tls_hostname->u.n); |
01fa5598 MA |
1805 | params->tls_hostname->type = QTYPE_QSTRING; |
1806 | params->tls_hostname->u.s = strdup(""); | |
1807 | } | |
1808 | ||
1bda8b3c MA |
1809 | migrate_params_test_apply(params, &tmp); |
1810 | ||
1811 | if (!migrate_params_check(&tmp, errp)) { | |
476c72aa PX |
1812 | /* Invalid parameter */ |
1813 | return; | |
1814 | } | |
1815 | ||
73af8dd8 | 1816 | migrate_params_apply(params, errp); |
476c72aa PX |
1817 | } |
1818 | ||
2594f56d | 1819 | |
4886a1bc DDAG |
1820 | void qmp_migrate_start_postcopy(Error **errp) |
1821 | { | |
1822 | MigrationState *s = migrate_get_current(); | |
1823 | ||
16b0fd32 | 1824 | if (!migrate_postcopy()) { |
a54d340b | 1825 | error_setg(errp, "Enable postcopy with migrate_set_capability before" |
4886a1bc DDAG |
1826 | " the start of migration"); |
1827 | return; | |
1828 | } | |
1829 | ||
1830 | if (s->state == MIGRATION_STATUS_NONE) { | |
1831 | error_setg(errp, "Postcopy must be started after migration has been" | |
1832 | " started"); | |
1833 | return; | |
1834 | } | |
1835 | /* | |
1836 | * we don't error if migration has finished since that would be racy | |
1837 | * with issuing this command. | |
1838 | */ | |
d73415a3 | 1839 | qatomic_set(&s->start_postcopy, true); |
4886a1bc DDAG |
1840 | } |
1841 | ||
065e2813 AL |
1842 | /* shared migration helpers */ |
1843 | ||
48781e5b | 1844 | void migrate_set_state(int *state, int old_state, int new_state) |
51cf4c1a | 1845 | { |
a31fedee | 1846 | assert(new_state < MIGRATION_STATUS__MAX); |
d73415a3 | 1847 | if (qatomic_cmpxchg(state, old_state, new_state) == old_state) { |
a31fedee | 1848 | trace_migrate_set_state(MigrationStatus_str(new_state)); |
b05dc723 | 1849 | migrate_generate_event(new_state); |
51cf4c1a Z |
1850 | } |
1851 | } | |
1852 | ||
eaedde52 EB |
1853 | static MigrationCapabilityStatus *migrate_cap_add(MigrationCapability index, |
1854 | bool state) | |
2833c59b | 1855 | { |
eaedde52 | 1856 | MigrationCapabilityStatus *cap; |
2833c59b | 1857 | |
eaedde52 EB |
1858 | cap = g_new0(MigrationCapabilityStatus, 1); |
1859 | cap->capability = index; | |
1860 | cap->state = state; | |
4e4a3d3a PX |
1861 | |
1862 | return cap; | |
1863 | } | |
1864 | ||
1865 | void migrate_set_block_enabled(bool value, Error **errp) | |
1866 | { | |
eaedde52 | 1867 | MigrationCapabilityStatusList *cap = NULL; |
4e4a3d3a | 1868 | |
eaedde52 | 1869 | QAPI_LIST_PREPEND(cap, migrate_cap_add(MIGRATION_CAPABILITY_BLOCK, value)); |
2833c59b JQ |
1870 | qmp_migrate_set_capabilities(cap, errp); |
1871 | qapi_free_MigrationCapabilityStatusList(cap); | |
1872 | } | |
1873 | ||
1874 | static void migrate_set_block_incremental(MigrationState *s, bool value) | |
1875 | { | |
1876 | s->parameters.block_incremental = value; | |
1877 | } | |
1878 | ||
1879 | static void block_cleanup_parameters(MigrationState *s) | |
1880 | { | |
1881 | if (s->must_remove_block_options) { | |
1882 | /* setting to false can never fail */ | |
1883 | migrate_set_block_enabled(false, &error_abort); | |
1884 | migrate_set_block_incremental(s, false); | |
1885 | s->must_remove_block_options = false; | |
1886 | } | |
1887 | } | |
1888 | ||
fd392cfa | 1889 | static void migrate_fd_cleanup(MigrationState *s) |
065e2813 | 1890 | { |
bb1fadc4 PB |
1891 | qemu_bh_delete(s->cleanup_bh); |
1892 | s->cleanup_bh = NULL; | |
1893 | ||
83174765 PX |
1894 | g_free(s->hostname); |
1895 | s->hostname = NULL; | |
1896 | ||
0ceccd85 PX |
1897 | qemu_savevm_state_cleanup(); |
1898 | ||
89a02a9f | 1899 | if (s->to_dst_file) { |
62df066f | 1900 | QEMUFile *tmp; |
f986c3d2 | 1901 | |
9013dca5 | 1902 | trace_migrate_fd_cleanup(); |
404a7c05 | 1903 | qemu_mutex_unlock_iothread(); |
1d34e4bf DDAG |
1904 | if (s->migration_thread_running) { |
1905 | qemu_thread_join(&s->thread); | |
1906 | s->migration_thread_running = false; | |
1907 | } | |
404a7c05 PB |
1908 | qemu_mutex_lock_iothread(); |
1909 | ||
1398b2e3 | 1910 | multifd_save_cleanup(); |
62df066f PX |
1911 | qemu_mutex_lock(&s->qemu_file_lock); |
1912 | tmp = s->to_dst_file; | |
89a02a9f | 1913 | s->to_dst_file = NULL; |
62df066f PX |
1914 | qemu_mutex_unlock(&s->qemu_file_lock); |
1915 | /* | |
1916 | * Close the file handle without the lock to make sure the | |
1917 | * critical section won't block for long. | |
1918 | */ | |
39675fff | 1919 | migration_ioc_unregister_yank_from_file(tmp); |
62df066f | 1920 | qemu_fclose(tmp); |
065e2813 AL |
1921 | } |
1922 | ||
36f62f11 PX |
1923 | if (s->postcopy_qemufile_src) { |
1924 | migration_ioc_unregister_yank_from_file(s->postcopy_qemufile_src); | |
1925 | qemu_fclose(s->postcopy_qemufile_src); | |
1926 | s->postcopy_qemufile_src = NULL; | |
1927 | } | |
1928 | ||
8f8d528e | 1929 | assert(!migration_is_active(s)); |
7a2c1721 | 1930 | |
94f5a437 | 1931 | if (s->state == MIGRATION_STATUS_CANCELLING) { |
48781e5b | 1932 | migrate_set_state(&s->state, MIGRATION_STATUS_CANCELLING, |
94f5a437 | 1933 | MIGRATION_STATUS_CANCELLED); |
7a2c1721 | 1934 | } |
a3fa1d78 | 1935 | |
87db1a7d JQ |
1936 | if (s->error) { |
1937 | /* It is used on info migrate. We can't free it */ | |
1938 | error_report_err(error_copy(s->error)); | |
1939 | } | |
a3fa1d78 | 1940 | notifier_list_notify(&migration_state_notifiers, s); |
2833c59b | 1941 | block_cleanup_parameters(s); |
b5eea99e | 1942 | yank_unregister_instance(MIGRATION_YANK_INSTANCE); |
065e2813 AL |
1943 | } |
1944 | ||
fd392cfa YK |
1945 | static void migrate_fd_cleanup_schedule(MigrationState *s) |
1946 | { | |
1947 | /* | |
1948 | * Ref the state for bh, because it may be called when | |
1949 | * there're already no other refs | |
1950 | */ | |
1951 | object_ref(OBJECT(s)); | |
1952 | qemu_bh_schedule(s->cleanup_bh); | |
1953 | } | |
1954 | ||
1955 | static void migrate_fd_cleanup_bh(void *opaque) | |
1956 | { | |
1957 | MigrationState *s = opaque; | |
1958 | migrate_fd_cleanup(s); | |
1959 | object_unref(OBJECT(s)); | |
1960 | } | |
1961 | ||
87db1a7d JQ |
1962 | void migrate_set_error(MigrationState *s, const Error *error) |
1963 | { | |
6e8a355d | 1964 | QEMU_LOCK_GUARD(&s->error_mutex); |
87db1a7d JQ |
1965 | if (!s->error) { |
1966 | s->error = error_copy(error); | |
1967 | } | |
87db1a7d JQ |
1968 | } |
1969 | ||
ca7bd082 PX |
1970 | static void migrate_error_free(MigrationState *s) |
1971 | { | |
1972 | QEMU_LOCK_GUARD(&s->error_mutex); | |
1973 | if (s->error) { | |
1974 | error_free(s->error); | |
1975 | s->error = NULL; | |
1976 | } | |
1977 | } | |
1978 | ||
d59ce6f3 | 1979 | void migrate_fd_error(MigrationState *s, const Error *error) |
065e2813 | 1980 | { |
25174055 | 1981 | trace_migrate_fd_error(error_get_pretty(error)); |
89a02a9f | 1982 | assert(s->to_dst_file == NULL); |
48781e5b HZ |
1983 | migrate_set_state(&s->state, MIGRATION_STATUS_SETUP, |
1984 | MIGRATION_STATUS_FAILED); | |
87db1a7d | 1985 | migrate_set_error(s, error); |
458cf28e JQ |
1986 | } |
1987 | ||
0edda1c4 | 1988 | static void migrate_fd_cancel(MigrationState *s) |
065e2813 | 1989 | { |
6f2b811a | 1990 | int old_state ; |
89a02a9f | 1991 | QEMUFile *f = migrate_get_current()->to_dst_file; |
9013dca5 | 1992 | trace_migrate_fd_cancel(); |
065e2813 | 1993 | |
43044ac0 PX |
1994 | WITH_QEMU_LOCK_GUARD(&s->qemu_file_lock) { |
1995 | if (s->rp_state.from_dst_file) { | |
1996 | /* shutdown the rp socket, so causing the rp thread to shutdown */ | |
1997 | qemu_file_shutdown(s->rp_state.from_dst_file); | |
1998 | } | |
70b20477 DDAG |
1999 | } |
2000 | ||
6f2b811a Z |
2001 | do { |
2002 | old_state = s->state; | |
392d87e2 | 2003 | if (!migration_is_running(old_state)) { |
6f2b811a Z |
2004 | break; |
2005 | } | |
a7b36b48 DDAG |
2006 | /* If the migration is paused, kick it out of the pause */ |
2007 | if (old_state == MIGRATION_STATUS_PRE_SWITCHOVER) { | |
2008 | qemu_sem_post(&s->pause_sem); | |
2009 | } | |
48781e5b | 2010 | migrate_set_state(&s->state, old_state, MIGRATION_STATUS_CANCELLING); |
31194731 | 2011 | } while (s->state != MIGRATION_STATUS_CANCELLING); |
a26ba26e DDAG |
2012 | |
2013 | /* | |
2014 | * If we're unlucky the migration code might be stuck somewhere in a | |
2015 | * send/write while the network has failed and is waiting to timeout; | |
2016 | * if we've got shutdown(2) available then we can force it to quit. | |
2017 | * The outgoing qemu file gets closed in migrate_fd_cleanup that is | |
2018 | * called in a bh, so there is no race against this cancel. | |
2019 | */ | |
31194731 | 2020 | if (s->state == MIGRATION_STATUS_CANCELLING && f) { |
a26ba26e DDAG |
2021 | qemu_file_shutdown(f); |
2022 | } | |
1d2acc31 HZ |
2023 | if (s->state == MIGRATION_STATUS_CANCELLING && s->block_inactive) { |
2024 | Error *local_err = NULL; | |
2025 | ||
3b717194 | 2026 | bdrv_activate_all(&local_err); |
1d2acc31 HZ |
2027 | if (local_err) { |
2028 | error_report_err(local_err); | |
2029 | } else { | |
2030 | s->block_inactive = false; | |
2031 | } | |
2032 | } | |
065e2813 AL |
2033 | } |
2034 | ||
99a0db9b GH |
2035 | void add_migration_state_change_notifier(Notifier *notify) |
2036 | { | |
2037 | notifier_list_add(&migration_state_notifiers, notify); | |
2038 | } | |
2039 | ||
2040 | void remove_migration_state_change_notifier(Notifier *notify) | |
2041 | { | |
31552529 | 2042 | notifier_remove(notify); |
99a0db9b GH |
2043 | } |
2044 | ||
02edd2e7 | 2045 | bool migration_in_setup(MigrationState *s) |
afe2df69 | 2046 | { |
31194731 | 2047 | return s->state == MIGRATION_STATUS_SETUP; |
afe2df69 GH |
2048 | } |
2049 | ||
7073693b | 2050 | bool migration_has_finished(MigrationState *s) |
99a0db9b | 2051 | { |
31194731 | 2052 | return s->state == MIGRATION_STATUS_COMPLETED; |
99a0db9b | 2053 | } |
0edda1c4 | 2054 | |
afe2df69 GH |
2055 | bool migration_has_failed(MigrationState *s) |
2056 | { | |
31194731 HZ |
2057 | return (s->state == MIGRATION_STATUS_CANCELLED || |
2058 | s->state == MIGRATION_STATUS_FAILED); | |
afe2df69 GH |
2059 | } |
2060 | ||
5727309d | 2061 | bool migration_in_postcopy(void) |
9ec055ae | 2062 | { |
5727309d JQ |
2063 | MigrationState *s = migrate_get_current(); |
2064 | ||
3748fef9 DDAG |
2065 | switch (s->state) { |
2066 | case MIGRATION_STATUS_POSTCOPY_ACTIVE: | |
2067 | case MIGRATION_STATUS_POSTCOPY_PAUSED: | |
2068 | case MIGRATION_STATUS_POSTCOPY_RECOVER: | |
2069 | return true; | |
2070 | default: | |
2071 | return false; | |
2072 | } | |
9ec055ae DDAG |
2073 | } |
2074 | ||
b82fc321 DDAG |
2075 | bool migration_in_postcopy_after_devices(MigrationState *s) |
2076 | { | |
5727309d | 2077 | return migration_in_postcopy() && s->postcopy_after_devices; |
b82fc321 DDAG |
2078 | } |
2079 | ||
06df2e69 DH |
2080 | bool migration_in_incoming_postcopy(void) |
2081 | { | |
2082 | PostcopyState ps = postcopy_state_get(); | |
2083 | ||
2084 | return ps >= POSTCOPY_INCOMING_DISCARD && ps < POSTCOPY_INCOMING_END; | |
2085 | } | |
2086 | ||
1a8e44a8 AG |
2087 | bool migration_in_bg_snapshot(void) |
2088 | { | |
2089 | MigrationState *s = migrate_get_current(); | |
2090 | ||
2091 | return migrate_background_snapshot() && | |
2092 | migration_is_setup_or_active(s->state); | |
2093 | } | |
2094 | ||
fab35005 | 2095 | bool migration_is_idle(void) |
fe44dc91 | 2096 | { |
daff7f0b MA |
2097 | MigrationState *s = current_migration; |
2098 | ||
2099 | if (!s) { | |
2100 | return true; | |
2101 | } | |
fe44dc91 AA |
2102 | |
2103 | switch (s->state) { | |
2104 | case MIGRATION_STATUS_NONE: | |
2105 | case MIGRATION_STATUS_CANCELLED: | |
2106 | case MIGRATION_STATUS_COMPLETED: | |
2107 | case MIGRATION_STATUS_FAILED: | |
2108 | return true; | |
2109 | case MIGRATION_STATUS_SETUP: | |
2110 | case MIGRATION_STATUS_CANCELLING: | |
2111 | case MIGRATION_STATUS_ACTIVE: | |
2112 | case MIGRATION_STATUS_POSTCOPY_ACTIVE: | |
2113 | case MIGRATION_STATUS_COLO: | |
31e06077 DDAG |
2114 | case MIGRATION_STATUS_PRE_SWITCHOVER: |
2115 | case MIGRATION_STATUS_DEVICE: | |
c7e0acd5 | 2116 | case MIGRATION_STATUS_WAIT_UNPLUG: |
fe44dc91 AA |
2117 | return false; |
2118 | case MIGRATION_STATUS__MAX: | |
2119 | g_assert_not_reached(); | |
2120 | } | |
2121 | ||
2122 | return false; | |
2123 | } | |
2124 | ||
8f8d528e WY |
2125 | bool migration_is_active(MigrationState *s) |
2126 | { | |
2127 | return (s->state == MIGRATION_STATUS_ACTIVE || | |
2128 | s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE); | |
2129 | } | |
2130 | ||
3e0c8050 | 2131 | void migrate_init(MigrationState *s) |
0edda1c4 | 2132 | { |
389775d1 DDAG |
2133 | /* |
2134 | * Reinitialise all migration state, except | |
2135 | * parameters/capabilities that the user set, and | |
2136 | * locks. | |
2137 | */ | |
389775d1 | 2138 | s->cleanup_bh = 0; |
8518278a | 2139 | s->vm_start_bh = 0; |
89a02a9f | 2140 | s->to_dst_file = NULL; |
389775d1 | 2141 | s->state = MIGRATION_STATUS_NONE; |
389775d1 DDAG |
2142 | s->rp_state.from_dst_file = NULL; |
2143 | s->rp_state.error = false; | |
2144 | s->mbps = 0.0; | |
aecbfe9c | 2145 | s->pages_per_second = 0.0; |
389775d1 DDAG |
2146 | s->downtime = 0; |
2147 | s->expected_downtime = 0; | |
389775d1 | 2148 | s->setup_time = 0; |
389775d1 | 2149 | s->start_postcopy = false; |
b82fc321 | 2150 | s->postcopy_after_devices = false; |
389775d1 | 2151 | s->migration_thread_running = false; |
d59ce6f3 DB |
2152 | error_free(s->error); |
2153 | s->error = NULL; | |
d8053e73 | 2154 | s->hostname = NULL; |
389775d1 | 2155 | |
48781e5b | 2156 | migrate_set_state(&s->state, MIGRATION_STATUS_NONE, MIGRATION_STATUS_SETUP); |
0edda1c4 | 2157 | |
4af246a3 PX |
2158 | s->start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); |
2159 | s->total_time = 0; | |
7287cbd4 | 2160 | s->vm_was_running = false; |
b15df1ae PX |
2161 | s->iteration_initial_bytes = 0; |
2162 | s->threshold_size = 0; | |
0edda1c4 | 2163 | } |
cab30143 | 2164 | |
60fd6801 | 2165 | int migrate_add_blocker_internal(Error *reason, Error **errp) |
fa2756b7 | 2166 | { |
4c170330 PX |
2167 | /* Snapshots are similar to migrations, so check RUN_STATE_SAVE_VM too. */ |
2168 | if (runstate_check(RUN_STATE_SAVE_VM) || !migration_is_idle()) { | |
2169 | error_propagate_prepend(errp, error_copy(reason), | |
2170 | "disallowing migration blocker " | |
2171 | "(migration/snapshot in progress) for: "); | |
2172 | return -EBUSY; | |
fe44dc91 AA |
2173 | } |
2174 | ||
4c170330 PX |
2175 | migration_blockers = g_slist_prepend(migration_blockers, reason); |
2176 | return 0; | |
fa2756b7 AL |
2177 | } |
2178 | ||
60fd6801 PX |
2179 | int migrate_add_blocker(Error *reason, Error **errp) |
2180 | { | |
2181 | if (only_migratable) { | |
2182 | error_propagate_prepend(errp, error_copy(reason), | |
2183 | "disallowing migration blocker " | |
2184 | "(--only-migratable) for: "); | |
2185 | return -EACCES; | |
2186 | } | |
2187 | ||
2188 | return migrate_add_blocker_internal(reason, errp); | |
2189 | } | |
2190 | ||
fa2756b7 AL |
2191 | void migrate_del_blocker(Error *reason) |
2192 | { | |
2193 | migration_blockers = g_slist_remove(migration_blockers, reason); | |
2194 | } | |
2195 | ||
bf1ae1f4 DDAG |
2196 | void qmp_migrate_incoming(const char *uri, Error **errp) |
2197 | { | |
2198 | Error *local_err = NULL; | |
4debb5f5 | 2199 | static bool once = true; |
bf1ae1f4 | 2200 | |
4debb5f5 DDAG |
2201 | if (!once) { |
2202 | error_setg(errp, "The incoming migration has already been started"); | |
603d5a42 | 2203 | return; |
4debb5f5 | 2204 | } |
e69d50d6 PB |
2205 | if (!runstate_check(RUN_STATE_INMIGRATE)) { |
2206 | error_setg(errp, "'-incoming' was not specified on the command line"); | |
2207 | return; | |
2208 | } | |
bf1ae1f4 | 2209 | |
cc48c587 PX |
2210 | if (!yank_register_instance(MIGRATION_YANK_INSTANCE, errp)) { |
2211 | return; | |
2212 | } | |
2213 | ||
bf1ae1f4 DDAG |
2214 | qemu_start_incoming_migration(uri, &local_err); |
2215 | ||
2216 | if (local_err) { | |
cc48c587 | 2217 | yank_unregister_instance(MIGRATION_YANK_INSTANCE); |
bf1ae1f4 DDAG |
2218 | error_propagate(errp, local_err); |
2219 | return; | |
2220 | } | |
2221 | ||
4debb5f5 | 2222 | once = false; |
bf1ae1f4 DDAG |
2223 | } |
2224 | ||
02affd41 PX |
2225 | void qmp_migrate_recover(const char *uri, Error **errp) |
2226 | { | |
2227 | MigrationIncomingState *mis = migration_incoming_get_current(); | |
2228 | ||
b7f9afd4 PX |
2229 | /* |
2230 | * Don't even bother to use ERRP_GUARD() as it _must_ always be set by | |
2231 | * callers (no one should ignore a recover failure); if there is, it's a | |
2232 | * programming error. | |
2233 | */ | |
2234 | assert(errp); | |
2235 | ||
02affd41 PX |
2236 | if (mis->state != MIGRATION_STATUS_POSTCOPY_PAUSED) { |
2237 | error_setg(errp, "Migrate recover can only be run " | |
2238 | "when postcopy is paused."); | |
2239 | return; | |
2240 | } | |
2241 | ||
08401c04 PX |
2242 | /* If there's an existing transport, release it */ |
2243 | migration_incoming_transport_cleanup(mis); | |
02affd41 PX |
2244 | |
2245 | /* | |
2246 | * Note that this call will never start a real migration; it will | |
2247 | * only re-setup the migration stream and poke existing migration | |
2248 | * to continue using that newly established channel. | |
2249 | */ | |
2250 | qemu_start_incoming_migration(uri, errp); | |
2251 | } | |
2252 | ||
bfbf89c2 PX |
2253 | void qmp_migrate_pause(Error **errp) |
2254 | { | |
2255 | MigrationState *ms = migrate_get_current(); | |
2256 | MigrationIncomingState *mis = migration_incoming_get_current(); | |
2257 | int ret; | |
2258 | ||
2259 | if (ms->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) { | |
2260 | /* Source side, during postcopy */ | |
2261 | qemu_mutex_lock(&ms->qemu_file_lock); | |
2262 | ret = qemu_file_shutdown(ms->to_dst_file); | |
2263 | qemu_mutex_unlock(&ms->qemu_file_lock); | |
2264 | if (ret) { | |
2265 | error_setg(errp, "Failed to pause source migration"); | |
2266 | } | |
2267 | return; | |
2268 | } | |
2269 | ||
2270 | if (mis->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) { | |
2271 | ret = qemu_file_shutdown(mis->from_src_file); | |
2272 | if (ret) { | |
2273 | error_setg(errp, "Failed to pause destination migration"); | |
2274 | } | |
2275 | return; | |
2276 | } | |
2277 | ||
2278 | error_setg(errp, "migrate-pause is currently only supported " | |
2279 | "during postcopy-active state"); | |
2280 | } | |
2281 | ||
24f3902b GK |
2282 | bool migration_is_blocked(Error **errp) |
2283 | { | |
2284 | if (qemu_savevm_state_blocked(errp)) { | |
2285 | return true; | |
2286 | } | |
2287 | ||
2288 | if (migration_blockers) { | |
250561e1 | 2289 | error_propagate(errp, error_copy(migration_blockers->data)); |
24f3902b GK |
2290 | return true; |
2291 | } | |
2292 | ||
2293 | return false; | |
2294 | } | |
2295 | ||
d3e35b8f PX |
2296 | /* Returns true if continue to migrate, or false if error detected */ |
2297 | static bool migrate_prepare(MigrationState *s, bool blk, bool blk_inc, | |
2298 | bool resume, Error **errp) | |
cab30143 | 2299 | { |
be7059cd | 2300 | Error *local_err = NULL; |
d3e35b8f PX |
2301 | |
2302 | if (resume) { | |
2303 | if (s->state != MIGRATION_STATUS_POSTCOPY_PAUSED) { | |
2304 | error_setg(errp, "Cannot resume if there is no " | |
2305 | "paused migration"); | |
2306 | return false; | |
2307 | } | |
97ca211c PX |
2308 | |
2309 | /* | |
2310 | * Postcopy recovery won't work well with release-ram | |
2311 | * capability since release-ram will drop the page buffer as | |
2312 | * long as the page is put into the send buffer. So if there | |
2313 | * is a network failure happened, any page buffers that have | |
2314 | * not yet reached the destination VM but have already been | |
2315 | * sent from the source VM will be lost forever. Let's refuse | |
2316 | * the client from resuming such a postcopy migration. | |
2317 | * Luckily release-ram was designed to only be used when src | |
2318 | * and destination VMs are on the same host, so it should be | |
2319 | * fine. | |
2320 | */ | |
2321 | if (migrate_release_ram()) { | |
2322 | error_setg(errp, "Postcopy recovery cannot work " | |
2323 | "when release-ram capability is set"); | |
2324 | return false; | |
2325 | } | |
2326 | ||
d3e35b8f PX |
2327 | /* This is a resume, skip init status */ |
2328 | return true; | |
2329 | } | |
cab30143 | 2330 | |
392d87e2 | 2331 | if (migration_is_running(s->state)) { |
c6bd8c70 | 2332 | error_setg(errp, QERR_MIGRATION_ACTIVE); |
d3e35b8f | 2333 | return false; |
cab30143 | 2334 | } |
d3e35b8f | 2335 | |
ca99993a DDAG |
2336 | if (runstate_check(RUN_STATE_INMIGRATE)) { |
2337 | error_setg(errp, "Guest is waiting for an incoming migration"); | |
d3e35b8f | 2338 | return false; |
ca99993a DDAG |
2339 | } |
2340 | ||
36d0fe65 T |
2341 | if (runstate_check(RUN_STATE_POSTMIGRATE)) { |
2342 | error_setg(errp, "Can't migrate the vm that was paused due to " | |
2343 | "previous migration"); | |
2344 | return false; | |
2345 | } | |
2346 | ||
24f3902b | 2347 | if (migration_is_blocked(errp)) { |
d3e35b8f | 2348 | return false; |
fa2756b7 AL |
2349 | } |
2350 | ||
d3e35b8f | 2351 | if (blk || blk_inc) { |
3ba02445 RL |
2352 | if (migrate_colo_enabled()) { |
2353 | error_setg(errp, "No disk migration is required in COLO mode"); | |
2354 | return false; | |
2355 | } | |
2833c59b JQ |
2356 | if (migrate_use_block() || migrate_use_block_incremental()) { |
2357 | error_setg(errp, "Command options are incompatible with " | |
2358 | "current migration capabilities"); | |
d3e35b8f | 2359 | return false; |
2833c59b JQ |
2360 | } |
2361 | migrate_set_block_enabled(true, &local_err); | |
2362 | if (local_err) { | |
2363 | error_propagate(errp, local_err); | |
d3e35b8f | 2364 | return false; |
2833c59b JQ |
2365 | } |
2366 | s->must_remove_block_options = true; | |
2367 | } | |
2368 | ||
d3e35b8f | 2369 | if (blk_inc) { |
2833c59b JQ |
2370 | migrate_set_block_incremental(s, true); |
2371 | } | |
2372 | ||
3e0c8050 | 2373 | migrate_init(s); |
87f3bd87 | 2374 | /* |
fa0b31d5 | 2375 | * set ram_counters compression_counters memory to zero for a |
87f3bd87 IR |
2376 | * new migration |
2377 | */ | |
2378 | memset(&ram_counters, 0, sizeof(ram_counters)); | |
fa0b31d5 | 2379 | memset(&compression_counters, 0, sizeof(compression_counters)); |
cab30143 | 2380 | |
d3e35b8f PX |
2381 | return true; |
2382 | } | |
2383 | ||
2384 | void qmp_migrate(const char *uri, bool has_blk, bool blk, | |
2385 | bool has_inc, bool inc, bool has_detach, bool detach, | |
2386 | bool has_resume, bool resume, Error **errp) | |
2387 | { | |
2388 | Error *local_err = NULL; | |
2389 | MigrationState *s = migrate_get_current(); | |
d658f65c | 2390 | const char *p = NULL; |
d3e35b8f PX |
2391 | |
2392 | if (!migrate_prepare(s, has_blk && blk, has_inc && inc, | |
2393 | has_resume && resume, errp)) { | |
2394 | /* Error detected, put into errp */ | |
2395 | return; | |
2396 | } | |
2397 | ||
b5eea99e LS |
2398 | if (!(has_resume && resume)) { |
2399 | if (!yank_register_instance(MIGRATION_YANK_INSTANCE, errp)) { | |
2400 | return; | |
2401 | } | |
2402 | } | |
2403 | ||
f444eeda | 2404 | migrate_protocol_allow_multi_channels(false); |
d658f65c | 2405 | if (strstart(uri, "tcp:", &p) || |
9ba3b2ba LM |
2406 | strstart(uri, "unix:", NULL) || |
2407 | strstart(uri, "vsock:", NULL)) { | |
f444eeda | 2408 | migrate_protocol_allow_multi_channels(true); |
d658f65c | 2409 | socket_start_outgoing_migration(s, p ? p : uri, &local_err); |
2da776db | 2410 | #ifdef CONFIG_RDMA |
41310c68 | 2411 | } else if (strstart(uri, "rdma:", &p)) { |
2da776db MH |
2412 | rdma_start_outgoing_migration(s, p, &local_err); |
2413 | #endif | |
cab30143 | 2414 | } else if (strstart(uri, "exec:", &p)) { |
f37afb5a | 2415 | exec_start_outgoing_migration(s, p, &local_err); |
cab30143 | 2416 | } else if (strstart(uri, "fd:", &p)) { |
f37afb5a | 2417 | fd_start_outgoing_migration(s, p, &local_err); |
99a0db9b | 2418 | } else { |
b5eea99e LS |
2419 | if (!(has_resume && resume)) { |
2420 | yank_unregister_instance(MIGRATION_YANK_INSTANCE); | |
2421 | } | |
c6bd8c70 MA |
2422 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "uri", |
2423 | "a valid migration protocol"); | |
48781e5b HZ |
2424 | migrate_set_state(&s->state, MIGRATION_STATUS_SETUP, |
2425 | MIGRATION_STATUS_FAILED); | |
09576e74 | 2426 | block_cleanup_parameters(s); |
e1c37d0e | 2427 | return; |
cab30143 JQ |
2428 | } |
2429 | ||
f37afb5a | 2430 | if (local_err) { |
b5eea99e LS |
2431 | if (!(has_resume && resume)) { |
2432 | yank_unregister_instance(MIGRATION_YANK_INSTANCE); | |
2433 | } | |
d59ce6f3 | 2434 | migrate_fd_error(s, local_err); |
f37afb5a | 2435 | error_propagate(errp, local_err); |
e1c37d0e | 2436 | return; |
1299c631 | 2437 | } |
cab30143 JQ |
2438 | } |
2439 | ||
6cdedb07 | 2440 | void qmp_migrate_cancel(Error **errp) |
cab30143 | 2441 | { |
458fecca | 2442 | migration_cancel(NULL); |
cab30143 JQ |
2443 | } |
2444 | ||
89cfc02c DDAG |
2445 | void qmp_migrate_continue(MigrationStatus state, Error **errp) |
2446 | { | |
2447 | MigrationState *s = migrate_get_current(); | |
2448 | if (s->state != state) { | |
2449 | error_setg(errp, "Migration not in expected state: %s", | |
2450 | MigrationStatus_str(s->state)); | |
2451 | return; | |
2452 | } | |
2453 | qemu_sem_post(&s->pause_sem); | |
2454 | } | |
2455 | ||
53f09a10 PB |
2456 | bool migrate_release_ram(void) |
2457 | { | |
2458 | MigrationState *s; | |
2459 | ||
2460 | s = migrate_get_current(); | |
2461 | ||
2462 | return s->enabled_capabilities[MIGRATION_CAPABILITY_RELEASE_RAM]; | |
2463 | } | |
2464 | ||
53dd370c DDAG |
2465 | bool migrate_postcopy_ram(void) |
2466 | { | |
2467 | MigrationState *s; | |
2468 | ||
2469 | s = migrate_get_current(); | |
2470 | ||
32c3db5b | 2471 | return s->enabled_capabilities[MIGRATION_CAPABILITY_POSTCOPY_RAM]; |
53dd370c DDAG |
2472 | } |
2473 | ||
58110f0a VSO |
2474 | bool migrate_postcopy(void) |
2475 | { | |
dd6bb914 | 2476 | return migrate_postcopy_ram() || migrate_dirty_bitmaps(); |
58110f0a VSO |
2477 | } |
2478 | ||
bde1e2ec CV |
2479 | bool migrate_auto_converge(void) |
2480 | { | |
2481 | MigrationState *s; | |
2482 | ||
2483 | s = migrate_get_current(); | |
2484 | ||
2485 | return s->enabled_capabilities[MIGRATION_CAPABILITY_AUTO_CONVERGE]; | |
2486 | } | |
2487 | ||
323004a3 PL |
2488 | bool migrate_zero_blocks(void) |
2489 | { | |
2490 | MigrationState *s; | |
2491 | ||
2492 | s = migrate_get_current(); | |
2493 | ||
2494 | return s->enabled_capabilities[MIGRATION_CAPABILITY_ZERO_BLOCKS]; | |
2495 | } | |
2496 | ||
f22f928e AP |
2497 | bool migrate_postcopy_blocktime(void) |
2498 | { | |
2499 | MigrationState *s; | |
2500 | ||
2501 | s = migrate_get_current(); | |
2502 | ||
2503 | return s->enabled_capabilities[MIGRATION_CAPABILITY_POSTCOPY_BLOCKTIME]; | |
2504 | } | |
2505 | ||
8706d2d5 LL |
2506 | bool migrate_use_compression(void) |
2507 | { | |
dde4e694 LL |
2508 | MigrationState *s; |
2509 | ||
2510 | s = migrate_get_current(); | |
2511 | ||
2512 | return s->enabled_capabilities[MIGRATION_CAPABILITY_COMPRESS]; | |
8706d2d5 LL |
2513 | } |
2514 | ||
2515 | int migrate_compress_level(void) | |
2516 | { | |
2517 | MigrationState *s; | |
2518 | ||
2519 | s = migrate_get_current(); | |
2520 | ||
2594f56d | 2521 | return s->parameters.compress_level; |
8706d2d5 LL |
2522 | } |
2523 | ||
2524 | int migrate_compress_threads(void) | |
2525 | { | |
2526 | MigrationState *s; | |
2527 | ||
2528 | s = migrate_get_current(); | |
2529 | ||
2594f56d | 2530 | return s->parameters.compress_threads; |
8706d2d5 LL |
2531 | } |
2532 | ||
1d58872a XG |
2533 | int migrate_compress_wait_thread(void) |
2534 | { | |
2535 | MigrationState *s; | |
2536 | ||
2537 | s = migrate_get_current(); | |
2538 | ||
2539 | return s->parameters.compress_wait_thread; | |
2540 | } | |
2541 | ||
3fcb38c2 LL |
2542 | int migrate_decompress_threads(void) |
2543 | { | |
2544 | MigrationState *s; | |
2545 | ||
2546 | s = migrate_get_current(); | |
2547 | ||
2594f56d | 2548 | return s->parameters.decompress_threads; |
3fcb38c2 LL |
2549 | } |
2550 | ||
55efc8c2 VSO |
2551 | bool migrate_dirty_bitmaps(void) |
2552 | { | |
2553 | MigrationState *s; | |
2554 | ||
2555 | s = migrate_get_current(); | |
2556 | ||
2557 | return s->enabled_capabilities[MIGRATION_CAPABILITY_DIRTY_BITMAPS]; | |
2558 | } | |
2559 | ||
18269069 YK |
2560 | bool migrate_ignore_shared(void) |
2561 | { | |
2562 | MigrationState *s; | |
2563 | ||
2564 | s = migrate_get_current(); | |
2565 | ||
2566 | return s->enabled_capabilities[MIGRATION_CAPABILITY_X_IGNORE_SHARED]; | |
2567 | } | |
2568 | ||
b9d68df6 YK |
2569 | bool migrate_validate_uuid(void) |
2570 | { | |
2571 | MigrationState *s; | |
2572 | ||
2573 | s = migrate_get_current(); | |
2574 | ||
2575 | return s->enabled_capabilities[MIGRATION_CAPABILITY_VALIDATE_UUID]; | |
2576 | } | |
2577 | ||
b05dc723 JQ |
2578 | bool migrate_use_events(void) |
2579 | { | |
2580 | MigrationState *s; | |
2581 | ||
2582 | s = migrate_get_current(); | |
2583 | ||
2584 | return s->enabled_capabilities[MIGRATION_CAPABILITY_EVENTS]; | |
2585 | } | |
2586 | ||
30126bbf JQ |
2587 | bool migrate_use_multifd(void) |
2588 | { | |
2589 | MigrationState *s; | |
2590 | ||
2591 | s = migrate_get_current(); | |
2592 | ||
cbfd6c95 | 2593 | return s->enabled_capabilities[MIGRATION_CAPABILITY_MULTIFD]; |
30126bbf JQ |
2594 | } |
2595 | ||
93fbd031 DDAG |
2596 | bool migrate_pause_before_switchover(void) |
2597 | { | |
2598 | MigrationState *s; | |
2599 | ||
2600 | s = migrate_get_current(); | |
2601 | ||
2602 | return s->enabled_capabilities[ | |
2603 | MIGRATION_CAPABILITY_PAUSE_BEFORE_SWITCHOVER]; | |
2604 | } | |
2605 | ||
4075fb1c JQ |
2606 | int migrate_multifd_channels(void) |
2607 | { | |
2608 | MigrationState *s; | |
2609 | ||
2610 | s = migrate_get_current(); | |
2611 | ||
cbfd6c95 | 2612 | return s->parameters.multifd_channels; |
4075fb1c JQ |
2613 | } |
2614 | ||
ab7cbb0b JQ |
2615 | MultiFDCompression migrate_multifd_compression(void) |
2616 | { | |
2617 | MigrationState *s; | |
2618 | ||
2619 | s = migrate_get_current(); | |
2620 | ||
ead34f64 | 2621 | assert(s->parameters.multifd_compression < MULTIFD_COMPRESSION__MAX); |
ab7cbb0b JQ |
2622 | return s->parameters.multifd_compression; |
2623 | } | |
2624 | ||
9004db48 JQ |
2625 | int migrate_multifd_zlib_level(void) |
2626 | { | |
2627 | MigrationState *s; | |
2628 | ||
2629 | s = migrate_get_current(); | |
2630 | ||
2631 | return s->parameters.multifd_zlib_level; | |
2632 | } | |
2633 | ||
6a9ad154 JQ |
2634 | int migrate_multifd_zstd_level(void) |
2635 | { | |
2636 | MigrationState *s; | |
2637 | ||
2638 | s = migrate_get_current(); | |
2639 | ||
2640 | return s->parameters.multifd_zstd_level; | |
2641 | } | |
2642 | ||
abb6295b LB |
2643 | #ifdef CONFIG_LINUX |
2644 | bool migrate_use_zero_copy_send(void) | |
2645 | { | |
2646 | MigrationState *s; | |
2647 | ||
2648 | s = migrate_get_current(); | |
2649 | ||
1abaec9a | 2650 | return s->enabled_capabilities[MIGRATION_CAPABILITY_ZERO_COPY_SEND]; |
abb6295b LB |
2651 | } |
2652 | #endif | |
2653 | ||
d2fafb6a LB |
2654 | int migrate_use_tls(void) |
2655 | { | |
2656 | MigrationState *s; | |
2657 | ||
2658 | s = migrate_get_current(); | |
2659 | ||
2660 | return s->parameters.tls_creds && *s->parameters.tls_creds; | |
2661 | } | |
2662 | ||
17ad9b35 OW |
2663 | int migrate_use_xbzrle(void) |
2664 | { | |
2665 | MigrationState *s; | |
2666 | ||
2667 | s = migrate_get_current(); | |
2668 | ||
2669 | return s->enabled_capabilities[MIGRATION_CAPABILITY_XBZRLE]; | |
2670 | } | |
2671 | ||
8b9407a0 | 2672 | uint64_t migrate_xbzrle_cache_size(void) |
17ad9b35 OW |
2673 | { |
2674 | MigrationState *s; | |
2675 | ||
2676 | s = migrate_get_current(); | |
2677 | ||
73af8dd8 | 2678 | return s->parameters.xbzrle_cache_size; |
17ad9b35 | 2679 | } |
0d82d0e8 | 2680 | |
7e555c6c DDAG |
2681 | static int64_t migrate_max_postcopy_bandwidth(void) |
2682 | { | |
2683 | MigrationState *s; | |
2684 | ||
2685 | s = migrate_get_current(); | |
2686 | ||
2687 | return s->parameters.max_postcopy_bandwidth; | |
2688 | } | |
2689 | ||
2833c59b JQ |
2690 | bool migrate_use_block(void) |
2691 | { | |
2692 | MigrationState *s; | |
2693 | ||
2694 | s = migrate_get_current(); | |
2695 | ||
2696 | return s->enabled_capabilities[MIGRATION_CAPABILITY_BLOCK]; | |
2697 | } | |
2698 | ||
c788ada8 PX |
2699 | bool migrate_use_return_path(void) |
2700 | { | |
2701 | MigrationState *s; | |
2702 | ||
2703 | s = migrate_get_current(); | |
2704 | ||
2705 | return s->enabled_capabilities[MIGRATION_CAPABILITY_RETURN_PATH]; | |
2706 | } | |
2707 | ||
2833c59b JQ |
2708 | bool migrate_use_block_incremental(void) |
2709 | { | |
2710 | MigrationState *s; | |
2711 | ||
2712 | s = migrate_get_current(); | |
2713 | ||
2714 | return s->parameters.block_incremental; | |
2715 | } | |
2716 | ||
6e8c25b4 AG |
2717 | bool migrate_background_snapshot(void) |
2718 | { | |
2719 | MigrationState *s; | |
2720 | ||
2721 | s = migrate_get_current(); | |
2722 | ||
2723 | return s->enabled_capabilities[MIGRATION_CAPABILITY_BACKGROUND_SNAPSHOT]; | |
2724 | } | |
2725 | ||
ce5b0f4a PX |
2726 | bool migrate_postcopy_preempt(void) |
2727 | { | |
2728 | MigrationState *s; | |
2729 | ||
2730 | s = migrate_get_current(); | |
2731 | ||
2732 | return s->enabled_capabilities[MIGRATION_CAPABILITY_POSTCOPY_PREEMPT]; | |
2733 | } | |
2734 | ||
70b20477 DDAG |
2735 | /* migration thread support */ |
2736 | /* | |
2737 | * Something bad happened to the RP stream, mark an error | |
2738 | * The caller shall print or trace something to indicate why | |
2739 | */ | |
2740 | static void mark_source_rp_bad(MigrationState *s) | |
2741 | { | |
2742 | s->rp_state.error = true; | |
2743 | } | |
2744 | ||
2745 | static struct rp_cmd_args { | |
2746 | ssize_t len; /* -1 = variable */ | |
2747 | const char *name; | |
2748 | } rp_cmd_args[] = { | |
2749 | [MIG_RP_MSG_INVALID] = { .len = -1, .name = "INVALID" }, | |
2750 | [MIG_RP_MSG_SHUT] = { .len = 4, .name = "SHUT" }, | |
2751 | [MIG_RP_MSG_PONG] = { .len = 4, .name = "PONG" }, | |
1e2d90eb DDAG |
2752 | [MIG_RP_MSG_REQ_PAGES] = { .len = 12, .name = "REQ_PAGES" }, |
2753 | [MIG_RP_MSG_REQ_PAGES_ID] = { .len = -1, .name = "REQ_PAGES_ID" }, | |
a335debb | 2754 | [MIG_RP_MSG_RECV_BITMAP] = { .len = -1, .name = "RECV_BITMAP" }, |
13955b89 | 2755 | [MIG_RP_MSG_RESUME_ACK] = { .len = 4, .name = "RESUME_ACK" }, |
70b20477 DDAG |
2756 | [MIG_RP_MSG_MAX] = { .len = -1, .name = "MAX" }, |
2757 | }; | |
2758 | ||
1e2d90eb DDAG |
2759 | /* |
2760 | * Process a request for pages received on the return path, | |
2761 | * We're allowed to send more than requested (e.g. to round to our page size) | |
2762 | * and we don't need to send pages that have already been sent. | |
2763 | */ | |
2764 | static void migrate_handle_rp_req_pages(MigrationState *ms, const char* rbname, | |
2765 | ram_addr_t start, size_t len) | |
2766 | { | |
8e3b0cbb | 2767 | long our_host_ps = qemu_real_host_page_size(); |
6c595cde | 2768 | |
1e2d90eb | 2769 | trace_migrate_handle_rp_req_pages(rbname, start, len); |
6c595cde DDAG |
2770 | |
2771 | /* | |
2772 | * Since we currently insist on matching page sizes, just sanity check | |
2773 | * we're being asked for whole host pages. | |
2774 | */ | |
7648297d DH |
2775 | if (!QEMU_IS_ALIGNED(start, our_host_ps) || |
2776 | !QEMU_IS_ALIGNED(len, our_host_ps)) { | |
6c595cde DDAG |
2777 | error_report("%s: Misaligned page request, start: " RAM_ADDR_FMT |
2778 | " len: %zd", __func__, start, len); | |
2779 | mark_source_rp_bad(ms); | |
2780 | return; | |
2781 | } | |
2782 | ||
96506894 | 2783 | if (ram_save_queue_pages(rbname, start, len)) { |
6c595cde DDAG |
2784 | mark_source_rp_bad(ms); |
2785 | } | |
1e2d90eb DDAG |
2786 | } |
2787 | ||
14b1742e PX |
2788 | /* Return true to retry, false to quit */ |
2789 | static bool postcopy_pause_return_path_thread(MigrationState *s) | |
2790 | { | |
2791 | trace_postcopy_pause_return_path(); | |
2792 | ||
2793 | qemu_sem_wait(&s->postcopy_pause_rp_sem); | |
2794 | ||
2795 | trace_postcopy_pause_return_path_continued(); | |
2796 | ||
2797 | return true; | |
2798 | } | |
2799 | ||
a335debb PX |
2800 | static int migrate_handle_rp_recv_bitmap(MigrationState *s, char *block_name) |
2801 | { | |
2802 | RAMBlock *block = qemu_ram_block_by_name(block_name); | |
2803 | ||
2804 | if (!block) { | |
2805 | error_report("%s: invalid block name '%s'", __func__, block_name); | |
2806 | return -EINVAL; | |
2807 | } | |
2808 | ||
2809 | /* Fetch the received bitmap and refresh the dirty bitmap */ | |
2810 | return ram_dirty_bitmap_reload(s, block); | |
2811 | } | |
2812 | ||
13955b89 PX |
2813 | static int migrate_handle_rp_resume_ack(MigrationState *s, uint32_t value) |
2814 | { | |
2815 | trace_source_return_path_thread_resume_ack(value); | |
2816 | ||
2817 | if (value != MIGRATION_RESUME_ACK_VALUE) { | |
2818 | error_report("%s: illegal resume_ack value %"PRIu32, | |
2819 | __func__, value); | |
2820 | return -1; | |
2821 | } | |
2822 | ||
2823 | /* Now both sides are active. */ | |
2824 | migrate_set_state(&s->state, MIGRATION_STATUS_POSTCOPY_RECOVER, | |
2825 | MIGRATION_STATUS_POSTCOPY_ACTIVE); | |
2826 | ||
94190696 PX |
2827 | /* Notify send thread that time to continue send pages */ |
2828 | qemu_sem_post(&s->rp_state.rp_sem); | |
13955b89 PX |
2829 | |
2830 | return 0; | |
2831 | } | |
2832 | ||
43044ac0 PX |
2833 | /* Release ms->rp_state.from_dst_file in a safe way */ |
2834 | static void migration_release_from_dst_file(MigrationState *ms) | |
2835 | { | |
2836 | QEMUFile *file; | |
2837 | ||
2838 | WITH_QEMU_LOCK_GUARD(&ms->qemu_file_lock) { | |
2839 | /* | |
2840 | * Reset the from_dst_file pointer first before releasing it, as we | |
2841 | * can't block within lock section | |
2842 | */ | |
2843 | file = ms->rp_state.from_dst_file; | |
2844 | ms->rp_state.from_dst_file = NULL; | |
2845 | } | |
2846 | ||
2847 | qemu_fclose(file); | |
2848 | } | |
2849 | ||
70b20477 DDAG |
2850 | /* |
2851 | * Handles messages sent on the return path towards the source VM | |
2852 | * | |
2853 | */ | |
2854 | static void *source_return_path_thread(void *opaque) | |
2855 | { | |
2856 | MigrationState *ms = opaque; | |
2857 | QEMUFile *rp = ms->rp_state.from_dst_file; | |
2858 | uint16_t header_len, header_type; | |
568b01ca | 2859 | uint8_t buf[512]; |
70b20477 | 2860 | uint32_t tmp32, sibling_error; |
1e2d90eb DDAG |
2861 | ram_addr_t start = 0; /* =0 to silence warning */ |
2862 | size_t len = 0, expected_len; | |
70b20477 DDAG |
2863 | int res; |
2864 | ||
2865 | trace_source_return_path_thread_entry(); | |
74637e6f | 2866 | rcu_register_thread(); |
14b1742e PX |
2867 | |
2868 | retry: | |
70b20477 DDAG |
2869 | while (!ms->rp_state.error && !qemu_file_get_error(rp) && |
2870 | migration_is_setup_or_active(ms->state)) { | |
2871 | trace_source_return_path_thread_loop_top(); | |
2872 | header_type = qemu_get_be16(rp); | |
2873 | header_len = qemu_get_be16(rp); | |
2874 | ||
7a9ddfbf PX |
2875 | if (qemu_file_get_error(rp)) { |
2876 | mark_source_rp_bad(ms); | |
2877 | goto out; | |
2878 | } | |
2879 | ||
70b20477 DDAG |
2880 | if (header_type >= MIG_RP_MSG_MAX || |
2881 | header_type == MIG_RP_MSG_INVALID) { | |
2882 | error_report("RP: Received invalid message 0x%04x length 0x%04x", | |
ed8b2828 | 2883 | header_type, header_len); |
70b20477 DDAG |
2884 | mark_source_rp_bad(ms); |
2885 | goto out; | |
2886 | } | |
2887 | ||
2888 | if ((rp_cmd_args[header_type].len != -1 && | |
2889 | header_len != rp_cmd_args[header_type].len) || | |
568b01ca | 2890 | header_len > sizeof(buf)) { |
70b20477 | 2891 | error_report("RP: Received '%s' message (0x%04x) with" |
ed8b2828 MZ |
2892 | "incorrect length %d expecting %zu", |
2893 | rp_cmd_args[header_type].name, header_type, header_len, | |
2894 | (size_t)rp_cmd_args[header_type].len); | |
70b20477 DDAG |
2895 | mark_source_rp_bad(ms); |
2896 | goto out; | |
2897 | } | |
2898 | ||
2899 | /* We know we've got a valid header by this point */ | |
2900 | res = qemu_get_buffer(rp, buf, header_len); | |
2901 | if (res != header_len) { | |
2902 | error_report("RP: Failed reading data for message 0x%04x" | |
2903 | " read %d expected %d", | |
2904 | header_type, res, header_len); | |
2905 | mark_source_rp_bad(ms); | |
2906 | goto out; | |
2907 | } | |
2908 | ||
2909 | /* OK, we have the message and the data */ | |
2910 | switch (header_type) { | |
2911 | case MIG_RP_MSG_SHUT: | |
4d885131 | 2912 | sibling_error = ldl_be_p(buf); |
70b20477 DDAG |
2913 | trace_source_return_path_thread_shut(sibling_error); |
2914 | if (sibling_error) { | |
2915 | error_report("RP: Sibling indicated error %d", sibling_error); | |
2916 | mark_source_rp_bad(ms); | |
2917 | } | |
2918 | /* | |
2919 | * We'll let the main thread deal with closing the RP | |
2920 | * we could do a shutdown(2) on it, but we're the only user | |
2921 | * anyway, so there's nothing gained. | |
2922 | */ | |
2923 | goto out; | |
2924 | ||
2925 | case MIG_RP_MSG_PONG: | |
4d885131 | 2926 | tmp32 = ldl_be_p(buf); |
70b20477 DDAG |
2927 | trace_source_return_path_thread_pong(tmp32); |
2928 | break; | |
2929 | ||
1e2d90eb | 2930 | case MIG_RP_MSG_REQ_PAGES: |
4d885131 PM |
2931 | start = ldq_be_p(buf); |
2932 | len = ldl_be_p(buf + 8); | |
1e2d90eb DDAG |
2933 | migrate_handle_rp_req_pages(ms, NULL, start, len); |
2934 | break; | |
2935 | ||
2936 | case MIG_RP_MSG_REQ_PAGES_ID: | |
2937 | expected_len = 12 + 1; /* header + termination */ | |
2938 | ||
2939 | if (header_len >= expected_len) { | |
4d885131 PM |
2940 | start = ldq_be_p(buf); |
2941 | len = ldl_be_p(buf + 8); | |
1e2d90eb DDAG |
2942 | /* Now we expect an idstr */ |
2943 | tmp32 = buf[12]; /* Length of the following idstr */ | |
2944 | buf[13 + tmp32] = '\0'; | |
2945 | expected_len += tmp32; | |
2946 | } | |
2947 | if (header_len != expected_len) { | |
2948 | error_report("RP: Req_Page_id with length %d expecting %zd", | |
ed8b2828 | 2949 | header_len, expected_len); |
1e2d90eb DDAG |
2950 | mark_source_rp_bad(ms); |
2951 | goto out; | |
2952 | } | |
2953 | migrate_handle_rp_req_pages(ms, (char *)&buf[13], start, len); | |
2954 | break; | |
2955 | ||
a335debb PX |
2956 | case MIG_RP_MSG_RECV_BITMAP: |
2957 | if (header_len < 1) { | |
2958 | error_report("%s: missing block name", __func__); | |
2959 | mark_source_rp_bad(ms); | |
2960 | goto out; | |
2961 | } | |
2962 | /* Format: len (1B) + idstr (<255B). This ends the idstr. */ | |
2963 | buf[buf[0] + 1] = '\0'; | |
2964 | if (migrate_handle_rp_recv_bitmap(ms, (char *)(buf + 1))) { | |
2965 | mark_source_rp_bad(ms); | |
2966 | goto out; | |
2967 | } | |
2968 | break; | |
2969 | ||
13955b89 PX |
2970 | case MIG_RP_MSG_RESUME_ACK: |
2971 | tmp32 = ldl_be_p(buf); | |
2972 | if (migrate_handle_rp_resume_ack(ms, tmp32)) { | |
2973 | mark_source_rp_bad(ms); | |
2974 | goto out; | |
2975 | } | |
2976 | break; | |
2977 | ||
70b20477 DDAG |
2978 | default: |
2979 | break; | |
2980 | } | |
2981 | } | |
14b1742e PX |
2982 | |
2983 | out: | |
2984 | res = qemu_file_get_error(rp); | |
2985 | if (res) { | |
d5c8f2af | 2986 | if (res && migration_in_postcopy()) { |
14b1742e PX |
2987 | /* |
2988 | * Maybe there is something we can do: it looks like a | |
2989 | * network down issue, and we pause for a recovery. | |
2990 | */ | |
43044ac0 | 2991 | migration_release_from_dst_file(ms); |
2e3e3da3 | 2992 | rp = NULL; |
14b1742e | 2993 | if (postcopy_pause_return_path_thread(ms)) { |
43044ac0 PX |
2994 | /* |
2995 | * Reload rp, reset the rest. Referencing it is safe since | |
2996 | * it's reset only by us above, or when migration completes | |
2997 | */ | |
2e3e3da3 | 2998 | rp = ms->rp_state.from_dst_file; |
14b1742e PX |
2999 | ms->rp_state.error = false; |
3000 | goto retry; | |
3001 | } | |
3002 | } | |
3003 | ||
70b20477 DDAG |
3004 | trace_source_return_path_thread_bad_end(); |
3005 | mark_source_rp_bad(ms); | |
3006 | } | |
3007 | ||
3008 | trace_source_return_path_thread_end(); | |
43044ac0 | 3009 | migration_release_from_dst_file(ms); |
74637e6f | 3010 | rcu_unregister_thread(); |
70b20477 DDAG |
3011 | return NULL; |
3012 | } | |
3013 | ||
d3e35b8f PX |
3014 | static int open_return_path_on_source(MigrationState *ms, |
3015 | bool create_thread) | |
70b20477 | 3016 | { |
89a02a9f | 3017 | ms->rp_state.from_dst_file = qemu_file_get_return_path(ms->to_dst_file); |
70b20477 DDAG |
3018 | if (!ms->rp_state.from_dst_file) { |
3019 | return -1; | |
3020 | } | |
3021 | ||
3022 | trace_open_return_path_on_source(); | |
d3e35b8f PX |
3023 | |
3024 | if (!create_thread) { | |
3025 | /* We're done */ | |
3026 | return 0; | |
3027 | } | |
3028 | ||
70b20477 DDAG |
3029 | qemu_thread_create(&ms->rp_state.rp_thread, "return path", |
3030 | source_return_path_thread, ms, QEMU_THREAD_JOINABLE); | |
53021ea1 | 3031 | ms->rp_state.rp_thread_created = true; |
70b20477 DDAG |
3032 | |
3033 | trace_open_return_path_on_source_continue(); | |
3034 | ||
3035 | return 0; | |
3036 | } | |
3037 | ||
70b20477 DDAG |
3038 | /* Returns 0 if the RP was ok, otherwise there was an error on the RP */ |
3039 | static int await_return_path_close_on_source(MigrationState *ms) | |
3040 | { | |
3041 | /* | |
3042 | * If this is a normal exit then the destination will send a SHUT and the | |
3043 | * rp_thread will exit, however if there's an error we need to cause | |
3044 | * it to exit. | |
3045 | */ | |
89a02a9f | 3046 | if (qemu_file_get_error(ms->to_dst_file) && ms->rp_state.from_dst_file) { |
70b20477 DDAG |
3047 | /* |
3048 | * shutdown(2), if we have it, will cause it to unblock if it's stuck | |
3049 | * waiting for the destination. | |
3050 | */ | |
3051 | qemu_file_shutdown(ms->rp_state.from_dst_file); | |
3052 | mark_source_rp_bad(ms); | |
3053 | } | |
3054 | trace_await_return_path_close_on_source_joining(); | |
3055 | qemu_thread_join(&ms->rp_state.rp_thread); | |
53021ea1 | 3056 | ms->rp_state.rp_thread_created = false; |
70b20477 DDAG |
3057 | trace_await_return_path_close_on_source_close(); |
3058 | return ms->rp_state.error; | |
3059 | } | |
3060 | ||
1d34e4bf DDAG |
3061 | /* |
3062 | * Switch from normal iteration to postcopy | |
3063 | * Returns non-0 on error | |
3064 | */ | |
7287cbd4 | 3065 | static int postcopy_start(MigrationState *ms) |
1d34e4bf DDAG |
3066 | { |
3067 | int ret; | |
61b67d47 DB |
3068 | QIOChannelBuffer *bioc; |
3069 | QEMUFile *fb; | |
1d34e4bf | 3070 | int64_t time_at_stop = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); |
7e555c6c | 3071 | int64_t bandwidth = migrate_max_postcopy_bandwidth(); |
ef8d6488 | 3072 | bool restart_block = false; |
0331c8ca | 3073 | int cur_state = MIGRATION_STATUS_ACTIVE; |
d0edb8a1 PX |
3074 | |
3075 | if (postcopy_preempt_wait_channel(ms)) { | |
3076 | migrate_set_state(&ms->state, ms->state, MIGRATION_STATUS_FAILED); | |
3077 | return -1; | |
3078 | } | |
3079 | ||
0331c8ca DDAG |
3080 | if (!migrate_pause_before_switchover()) { |
3081 | migrate_set_state(&ms->state, MIGRATION_STATUS_ACTIVE, | |
3082 | MIGRATION_STATUS_POSTCOPY_ACTIVE); | |
3083 | } | |
1d34e4bf DDAG |
3084 | |
3085 | trace_postcopy_start(); | |
3086 | qemu_mutex_lock_iothread(); | |
3087 | trace_postcopy_start_set_run(); | |
3088 | ||
fb064112 | 3089 | qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER, NULL); |
1d34e4bf DDAG |
3090 | global_state_store(); |
3091 | ret = vm_stop_force_state(RUN_STATE_FINISH_MIGRATE); | |
76b1c7fe KW |
3092 | if (ret < 0) { |
3093 | goto fail; | |
3094 | } | |
1d34e4bf | 3095 | |
0331c8ca DDAG |
3096 | ret = migration_maybe_pause(ms, &cur_state, |
3097 | MIGRATION_STATUS_POSTCOPY_ACTIVE); | |
3098 | if (ret < 0) { | |
3099 | goto fail; | |
3100 | } | |
3101 | ||
76b1c7fe | 3102 | ret = bdrv_inactivate_all(); |
1d34e4bf DDAG |
3103 | if (ret < 0) { |
3104 | goto fail; | |
3105 | } | |
ef8d6488 | 3106 | restart_block = true; |
1d34e4bf | 3107 | |
1c0d249d DDAG |
3108 | /* |
3109 | * Cause any non-postcopiable, but iterative devices to | |
3110 | * send out their final data. | |
3111 | */ | |
a1fbe750 | 3112 | qemu_savevm_state_complete_precopy(ms->to_dst_file, true, false); |
1c0d249d | 3113 | |
1d34e4bf DDAG |
3114 | /* |
3115 | * in Finish migrate and with the io-lock held everything should | |
3116 | * be quiet, but we've potentially still got dirty pages and we | |
3117 | * need to tell the destination to throw any pages it's already received | |
3118 | * that are dirty | |
3119 | */ | |
58110f0a | 3120 | if (migrate_postcopy_ram()) { |
739fcc1b | 3121 | ram_postcopy_send_discard_bitmap(ms); |
1d34e4bf DDAG |
3122 | } |
3123 | ||
3124 | /* | |
3125 | * send rest of state - note things that are doing postcopy | |
3126 | * will notice we're in POSTCOPY_ACTIVE and not actually | |
3127 | * wrap their state up here | |
3128 | */ | |
7e555c6c DDAG |
3129 | /* 0 max-postcopy-bandwidth means unlimited */ |
3130 | if (!bandwidth) { | |
3131 | qemu_file_set_rate_limit(ms->to_dst_file, INT64_MAX); | |
3132 | } else { | |
3133 | qemu_file_set_rate_limit(ms->to_dst_file, bandwidth / XFER_LIMIT_RATIO); | |
3134 | } | |
58110f0a VSO |
3135 | if (migrate_postcopy_ram()) { |
3136 | /* Ping just for debugging, helps line traces up */ | |
3137 | qemu_savevm_send_ping(ms->to_dst_file, 2); | |
3138 | } | |
1d34e4bf DDAG |
3139 | |
3140 | /* | |
3141 | * While loading the device state we may trigger page transfer | |
3142 | * requests and the fd must be free to process those, and thus | |
3143 | * the destination must read the whole device state off the fd before | |
3144 | * it starts processing it. Unfortunately the ad-hoc migration format | |
3145 | * doesn't allow the destination to know the size to read without fully | |
3146 | * parsing it through each devices load-state code (especially the open | |
3147 | * coded devices that use get/put). | |
3148 | * So we wrap the device state up in a package with a length at the start; | |
3149 | * to do this we use a qemu_buf to hold the whole of the device state. | |
3150 | */ | |
61b67d47 | 3151 | bioc = qio_channel_buffer_new(4096); |
6f01f136 | 3152 | qio_channel_set_name(QIO_CHANNEL(bioc), "migration-postcopy-buffer"); |
77ef2dc1 | 3153 | fb = qemu_file_new_output(QIO_CHANNEL(bioc)); |
61b67d47 | 3154 | object_unref(OBJECT(bioc)); |
1d34e4bf | 3155 | |
c76201ab DDAG |
3156 | /* |
3157 | * Make sure the receiver can get incoming pages before we send the rest | |
3158 | * of the state | |
3159 | */ | |
3160 | qemu_savevm_send_postcopy_listen(fb); | |
3161 | ||
a1fbe750 | 3162 | qemu_savevm_state_complete_precopy(fb, false, false); |
58110f0a VSO |
3163 | if (migrate_postcopy_ram()) { |
3164 | qemu_savevm_send_ping(fb, 3); | |
3165 | } | |
1d34e4bf DDAG |
3166 | |
3167 | qemu_savevm_send_postcopy_run(fb); | |
3168 | ||
3169 | /* <><> end of stuff going into the package */ | |
1d34e4bf | 3170 | |
ef8d6488 DDAG |
3171 | /* Last point of recovery; as soon as we send the package the destination |
3172 | * can open devices and potentially start running. | |
3173 | * Lets just check again we've not got any errors. | |
3174 | */ | |
3175 | ret = qemu_file_get_error(ms->to_dst_file); | |
3176 | if (ret) { | |
3177 | error_report("postcopy_start: Migration stream errored (pre package)"); | |
3178 | goto fail_closefb; | |
3179 | } | |
3180 | ||
3181 | restart_block = false; | |
3182 | ||
1d34e4bf | 3183 | /* Now send that blob */ |
61b67d47 | 3184 | if (qemu_savevm_send_packaged(ms->to_dst_file, bioc->data, bioc->usage)) { |
1d34e4bf DDAG |
3185 | goto fail_closefb; |
3186 | } | |
3187 | qemu_fclose(fb); | |
b82fc321 DDAG |
3188 | |
3189 | /* Send a notify to give a chance for anything that needs to happen | |
3190 | * at the transition to postcopy and after the device state; in particular | |
3191 | * spice needs to trigger a transition now | |
3192 | */ | |
3193 | ms->postcopy_after_devices = true; | |
3194 | notifier_list_notify(&migration_state_notifiers, ms); | |
3195 | ||
1d34e4bf DDAG |
3196 | ms->downtime = qemu_clock_get_ms(QEMU_CLOCK_REALTIME) - time_at_stop; |
3197 | ||
3198 | qemu_mutex_unlock_iothread(); | |
3199 | ||
58110f0a VSO |
3200 | if (migrate_postcopy_ram()) { |
3201 | /* | |
3202 | * Although this ping is just for debug, it could potentially be | |
3203 | * used for getting a better measurement of downtime at the source. | |
3204 | */ | |
3205 | qemu_savevm_send_ping(ms->to_dst_file, 4); | |
3206 | } | |
1d34e4bf | 3207 | |
ced1c616 PB |
3208 | if (migrate_release_ram()) { |
3209 | ram_postcopy_migrated_memory_release(ms); | |
3210 | } | |
3211 | ||
89a02a9f | 3212 | ret = qemu_file_get_error(ms->to_dst_file); |
1d34e4bf DDAG |
3213 | if (ret) { |
3214 | error_report("postcopy_start: Migration stream errored"); | |
48781e5b | 3215 | migrate_set_state(&ms->state, MIGRATION_STATUS_POSTCOPY_ACTIVE, |
1d34e4bf DDAG |
3216 | MIGRATION_STATUS_FAILED); |
3217 | } | |
3218 | ||
c01b16ed PX |
3219 | trace_postcopy_preempt_enabled(migrate_postcopy_preempt()); |
3220 | ||
1d34e4bf DDAG |
3221 | return ret; |
3222 | ||
3223 | fail_closefb: | |
3224 | qemu_fclose(fb); | |
3225 | fail: | |
48781e5b | 3226 | migrate_set_state(&ms->state, MIGRATION_STATUS_POSTCOPY_ACTIVE, |
1d34e4bf | 3227 | MIGRATION_STATUS_FAILED); |
ef8d6488 DDAG |
3228 | if (restart_block) { |
3229 | /* A failure happened early enough that we know the destination hasn't | |
3230 | * accessed block devices, so we're safe to recover. | |
3231 | */ | |
3232 | Error *local_err = NULL; | |
3233 | ||
3b717194 | 3234 | bdrv_activate_all(&local_err); |
ef8d6488 DDAG |
3235 | if (local_err) { |
3236 | error_report_err(local_err); | |
3237 | } | |
3238 | } | |
1d34e4bf DDAG |
3239 | qemu_mutex_unlock_iothread(); |
3240 | return -1; | |
3241 | } | |
3242 | ||
e91d8951 DDAG |
3243 | /** |
3244 | * migration_maybe_pause: Pause if required to by | |
3245 | * migrate_pause_before_switchover called with the iothread locked | |
3246 | * Returns: 0 on success | |
3247 | */ | |
0331c8ca DDAG |
3248 | static int migration_maybe_pause(MigrationState *s, |
3249 | int *current_active_state, | |
3250 | int new_state) | |
e91d8951 DDAG |
3251 | { |
3252 | if (!migrate_pause_before_switchover()) { | |
3253 | return 0; | |
3254 | } | |
3255 | ||
3256 | /* Since leaving this state is not atomic with posting the semaphore | |
3257 | * it's possible that someone could have issued multiple migrate_continue | |
3258 | * and the semaphore is incorrectly positive at this point; | |
3259 | * the docs say it's undefined to reinit a semaphore that's already | |
3260 | * init'd, so use timedwait to eat up any existing posts. | |
3261 | */ | |
3262 | while (qemu_sem_timedwait(&s->pause_sem, 1) == 0) { | |
3263 | /* This block intentionally left blank */ | |
3264 | } | |
3265 | ||
8958338b ZF |
3266 | /* |
3267 | * If the migration is cancelled when it is in the completion phase, | |
3268 | * the migration state is set to MIGRATION_STATUS_CANCELLING. | |
3269 | * So we don't need to wait a semaphore, otherwise we would always | |
3270 | * wait for the 'pause_sem' semaphore. | |
3271 | */ | |
3272 | if (s->state != MIGRATION_STATUS_CANCELLING) { | |
3273 | qemu_mutex_unlock_iothread(); | |
3274 | migrate_set_state(&s->state, *current_active_state, | |
3275 | MIGRATION_STATUS_PRE_SWITCHOVER); | |
3276 | qemu_sem_wait(&s->pause_sem); | |
3277 | migrate_set_state(&s->state, MIGRATION_STATUS_PRE_SWITCHOVER, | |
3278 | new_state); | |
3279 | *current_active_state = new_state; | |
3280 | qemu_mutex_lock_iothread(); | |
3281 | } | |
e91d8951 | 3282 | |
0331c8ca | 3283 | return s->state == new_state ? 0 : -EINVAL; |
e91d8951 DDAG |
3284 | } |
3285 | ||
09f6c85e DDAG |
3286 | /** |
3287 | * migration_completion: Used by migration_thread when there's not much left. | |
3288 | * The caller 'breaks' the loop when this returns. | |
3289 | * | |
3290 | * @s: Current migration state | |
09f6c85e | 3291 | */ |
2ad87305 | 3292 | static void migration_completion(MigrationState *s) |
09f6c85e DDAG |
3293 | { |
3294 | int ret; | |
2ad87305 | 3295 | int current_active_state = s->state; |
09f6c85e | 3296 | |
b10ac0c4 DDAG |
3297 | if (s->state == MIGRATION_STATUS_ACTIVE) { |
3298 | qemu_mutex_lock_iothread(); | |
64909f97 | 3299 | s->downtime_start = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); |
fb064112 | 3300 | qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER, NULL); |
7287cbd4 | 3301 | s->vm_was_running = runstate_is_running(); |
b10ac0c4 DDAG |
3302 | ret = global_state_store(); |
3303 | ||
3304 | if (!ret) { | |
a1fbe750 | 3305 | bool inactivate = !migrate_colo_enabled(); |
b10ac0c4 | 3306 | ret = vm_stop_force_state(RUN_STATE_FINISH_MIGRATE); |
626ff651 | 3307 | trace_migration_completion_vm_stop(ret); |
e91d8951 | 3308 | if (ret >= 0) { |
0331c8ca DDAG |
3309 | ret = migration_maybe_pause(s, ¤t_active_state, |
3310 | MIGRATION_STATUS_DEVICE); | |
e91d8951 | 3311 | } |
f07fa4cb KW |
3312 | if (ret >= 0) { |
3313 | qemu_file_set_rate_limit(s->to_dst_file, INT64_MAX); | |
a1fbe750 FZ |
3314 | ret = qemu_savevm_state_complete_precopy(s->to_dst_file, false, |
3315 | inactivate); | |
f07fa4cb | 3316 | } |
a1fbe750 FZ |
3317 | if (inactivate && ret >= 0) { |
3318 | s->block_inactive = true; | |
b10ac0c4 DDAG |
3319 | } |
3320 | } | |
3321 | qemu_mutex_unlock_iothread(); | |
09f6c85e | 3322 | |
b10ac0c4 DDAG |
3323 | if (ret < 0) { |
3324 | goto fail; | |
09f6c85e | 3325 | } |
b10ac0c4 DDAG |
3326 | } else if (s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) { |
3327 | trace_migration_completion_postcopy_end(); | |
3328 | ||
68b88468 | 3329 | qemu_mutex_lock_iothread(); |
89a02a9f | 3330 | qemu_savevm_state_complete_postcopy(s->to_dst_file); |
68b88468 EGE |
3331 | qemu_mutex_unlock_iothread(); |
3332 | ||
36f62f11 PX |
3333 | /* Shutdown the postcopy fast path thread */ |
3334 | if (migrate_postcopy_preempt()) { | |
3335 | postcopy_preempt_shutdown_file(s); | |
3336 | } | |
3337 | ||
b10ac0c4 | 3338 | trace_migration_completion_postcopy_end_after_complete(); |
444252b9 | 3339 | } else { |
6ba11211 | 3340 | goto fail; |
09f6c85e | 3341 | } |
09f6c85e | 3342 | |
b10ac0c4 DDAG |
3343 | /* |
3344 | * If rp was opened we must clean up the thread before | |
3345 | * cleaning everything else up (since if there are no failures | |
3346 | * it will wait for the destination to send it's status in | |
3347 | * a SHUT command). | |
b10ac0c4 | 3348 | */ |
53021ea1 | 3349 | if (s->rp_state.rp_thread_created) { |
b10ac0c4 | 3350 | int rp_error; |
0425dc97 | 3351 | trace_migration_return_path_end_before(); |
b10ac0c4 | 3352 | rp_error = await_return_path_close_on_source(s); |
0425dc97 | 3353 | trace_migration_return_path_end_after(rp_error); |
b10ac0c4 | 3354 | if (rp_error) { |
fe904ea8 | 3355 | goto fail_invalidate; |
b10ac0c4 | 3356 | } |
09f6c85e DDAG |
3357 | } |
3358 | ||
89a02a9f | 3359 | if (qemu_file_get_error(s->to_dst_file)) { |
09f6c85e | 3360 | trace_migration_completion_file_err(); |
fe904ea8 | 3361 | goto fail_invalidate; |
09f6c85e DDAG |
3362 | } |
3363 | ||
eeeb48ee ZC |
3364 | if (migrate_colo_enabled() && s->state == MIGRATION_STATUS_ACTIVE) { |
3365 | /* COLO does not support postcopy */ | |
3366 | migrate_set_state(&s->state, MIGRATION_STATUS_ACTIVE, | |
3367 | MIGRATION_STATUS_COLO); | |
3368 | } else { | |
0b827d5e HZ |
3369 | migrate_set_state(&s->state, current_active_state, |
3370 | MIGRATION_STATUS_COMPLETED); | |
3371 | } | |
3372 | ||
09f6c85e DDAG |
3373 | return; |
3374 | ||
fe904ea8 GK |
3375 | fail_invalidate: |
3376 | /* If not doing postcopy, vm_start() will be called: let's regain | |
3377 | * control on images. | |
3378 | */ | |
6039dd5b DDAG |
3379 | if (s->state == MIGRATION_STATUS_ACTIVE || |
3380 | s->state == MIGRATION_STATUS_DEVICE) { | |
fe904ea8 GK |
3381 | Error *local_err = NULL; |
3382 | ||
1d2acc31 | 3383 | qemu_mutex_lock_iothread(); |
3b717194 | 3384 | bdrv_activate_all(&local_err); |
fe904ea8 GK |
3385 | if (local_err) { |
3386 | error_report_err(local_err); | |
1d2acc31 HZ |
3387 | } else { |
3388 | s->block_inactive = false; | |
fe904ea8 | 3389 | } |
1d2acc31 | 3390 | qemu_mutex_unlock_iothread(); |
fe904ea8 GK |
3391 | } |
3392 | ||
09f6c85e | 3393 | fail: |
48781e5b HZ |
3394 | migrate_set_state(&s->state, current_active_state, |
3395 | MIGRATION_STATUS_FAILED); | |
09f6c85e DDAG |
3396 | } |
3397 | ||
8518278a AG |
3398 | /** |
3399 | * bg_migration_completion: Used by bg_migration_thread when after all the | |
3400 | * RAM has been saved. The caller 'breaks' the loop when this returns. | |
3401 | * | |
3402 | * @s: Current migration state | |
3403 | */ | |
3404 | static void bg_migration_completion(MigrationState *s) | |
3405 | { | |
3406 | int current_active_state = s->state; | |
3407 | ||
3408 | /* | |
3409 | * Stop tracking RAM writes - un-protect memory, un-register UFFD | |
3410 | * memory ranges, flush kernel wait queues and wake up threads | |
3411 | * waiting for write fault to be resolved. | |
3412 | */ | |
3413 | ram_write_tracking_stop(); | |
3414 | ||
3415 | if (s->state == MIGRATION_STATUS_ACTIVE) { | |
3416 | /* | |
3417 | * By this moment we have RAM content saved into the migration stream. | |
3418 | * The next step is to flush the non-RAM content (device state) | |
3419 | * right after the ram content. The device state has been stored into | |
3420 | * the temporary buffer before RAM saving started. | |
3421 | */ | |
3422 | qemu_put_buffer(s->to_dst_file, s->bioc->data, s->bioc->usage); | |
3423 | qemu_fflush(s->to_dst_file); | |
3424 | } else if (s->state == MIGRATION_STATUS_CANCELLING) { | |
3425 | goto fail; | |
3426 | } | |
3427 | ||
3428 | if (qemu_file_get_error(s->to_dst_file)) { | |
3429 | trace_migration_completion_file_err(); | |
3430 | goto fail; | |
3431 | } | |
3432 | ||
3433 | migrate_set_state(&s->state, current_active_state, | |
3434 | MIGRATION_STATUS_COMPLETED); | |
3435 | return; | |
3436 | ||
3437 | fail: | |
3438 | migrate_set_state(&s->state, current_active_state, | |
3439 | MIGRATION_STATUS_FAILED); | |
3440 | } | |
3441 | ||
35a6ed4f HZ |
3442 | bool migrate_colo_enabled(void) |
3443 | { | |
3444 | MigrationState *s = migrate_get_current(); | |
3445 | return s->enabled_capabilities[MIGRATION_CAPABILITY_X_COLO]; | |
3446 | } | |
3447 | ||
b23c2ade PX |
3448 | typedef enum MigThrError { |
3449 | /* No error detected */ | |
3450 | MIG_THR_ERR_NONE = 0, | |
3451 | /* Detected error, but resumed successfully */ | |
3452 | MIG_THR_ERR_RECOVERED = 1, | |
3453 | /* Detected fatal error, need to exit */ | |
3454 | MIG_THR_ERR_FATAL = 2, | |
3455 | } MigThrError; | |
3456 | ||
94190696 PX |
3457 | static int postcopy_resume_handshake(MigrationState *s) |
3458 | { | |
3459 | qemu_savevm_send_postcopy_resume(s->to_dst_file); | |
3460 | ||
3461 | while (s->state == MIGRATION_STATUS_POSTCOPY_RECOVER) { | |
3462 | qemu_sem_wait(&s->rp_state.rp_sem); | |
3463 | } | |
3464 | ||
3465 | if (s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) { | |
3466 | return 0; | |
3467 | } | |
3468 | ||
3469 | return -1; | |
3470 | } | |
3471 | ||
135b87b4 PX |
3472 | /* Return zero if success, or <0 for error */ |
3473 | static int postcopy_do_resume(MigrationState *s) | |
3474 | { | |
d1b8eadb PX |
3475 | int ret; |
3476 | ||
3477 | /* | |
3478 | * Call all the resume_prepare() hooks, so that modules can be | |
3479 | * ready for the migration resume. | |
3480 | */ | |
3481 | ret = qemu_savevm_state_resume_prepare(s); | |
3482 | if (ret) { | |
3483 | error_report("%s: resume_prepare() failure detected: %d", | |
3484 | __func__, ret); | |
3485 | return ret; | |
3486 | } | |
3487 | ||
3488 | /* | |
94190696 PX |
3489 | * Last handshake with destination on the resume (destination will |
3490 | * switch to postcopy-active afterwards) | |
d1b8eadb | 3491 | */ |
94190696 PX |
3492 | ret = postcopy_resume_handshake(s); |
3493 | if (ret) { | |
3494 | error_report("%s: handshake failed: %d", __func__, ret); | |
3495 | return ret; | |
3496 | } | |
d1b8eadb | 3497 | |
135b87b4 PX |
3498 | return 0; |
3499 | } | |
3500 | ||
b23c2ade PX |
3501 | /* |
3502 | * We don't return until we are in a safe state to continue current | |
3503 | * postcopy migration. Returns MIG_THR_ERR_RECOVERED if recovered, or | |
3504 | * MIG_THR_ERR_FATAL if unrecovery failure happened. | |
3505 | */ | |
3506 | static MigThrError postcopy_pause(MigrationState *s) | |
3507 | { | |
3508 | assert(s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE); | |
b23c2ade | 3509 | |
135b87b4 | 3510 | while (true) { |
62df066f PX |
3511 | QEMUFile *file; |
3512 | ||
39675fff PX |
3513 | /* |
3514 | * Current channel is possibly broken. Release it. Note that this is | |
3515 | * guaranteed even without lock because to_dst_file should only be | |
3516 | * modified by the migration thread. That also guarantees that the | |
3517 | * unregister of yank is safe too without the lock. It should be safe | |
3518 | * even to be within the qemu_file_lock, but we didn't do that to avoid | |
3519 | * taking more mutex (yank_lock) within qemu_file_lock. TL;DR: we make | |
3520 | * the qemu_file_lock critical section as small as possible. | |
3521 | */ | |
135b87b4 | 3522 | assert(s->to_dst_file); |
39675fff | 3523 | migration_ioc_unregister_yank_from_file(s->to_dst_file); |
62df066f PX |
3524 | qemu_mutex_lock(&s->qemu_file_lock); |
3525 | file = s->to_dst_file; | |
135b87b4 | 3526 | s->to_dst_file = NULL; |
62df066f PX |
3527 | qemu_mutex_unlock(&s->qemu_file_lock); |
3528 | ||
3529 | qemu_file_shutdown(file); | |
3530 | qemu_fclose(file); | |
b23c2ade | 3531 | |
60bb3c58 PX |
3532 | /* |
3533 | * Do the same to postcopy fast path socket too if there is. No | |
3534 | * locking needed because no racer as long as we do this before setting | |
3535 | * status to paused. | |
3536 | */ | |
3537 | if (s->postcopy_qemufile_src) { | |
3538 | migration_ioc_unregister_yank_from_file(s->postcopy_qemufile_src); | |
3539 | qemu_file_shutdown(s->postcopy_qemufile_src); | |
3540 | qemu_fclose(s->postcopy_qemufile_src); | |
3541 | s->postcopy_qemufile_src = NULL; | |
3542 | } | |
3543 | ||
d246ea50 PX |
3544 | migrate_set_state(&s->state, s->state, |
3545 | MIGRATION_STATUS_POSTCOPY_PAUSED); | |
3546 | ||
135b87b4 PX |
3547 | error_report("Detected IO failure for postcopy. " |
3548 | "Migration paused."); | |
b23c2ade | 3549 | |
135b87b4 PX |
3550 | /* |
3551 | * We wait until things fixed up. Then someone will setup the | |
3552 | * status back for us. | |
3553 | */ | |
3554 | while (s->state == MIGRATION_STATUS_POSTCOPY_PAUSED) { | |
3555 | qemu_sem_wait(&s->postcopy_pause_sem); | |
3556 | } | |
3557 | ||
3558 | if (s->state == MIGRATION_STATUS_POSTCOPY_RECOVER) { | |
3559 | /* Woken up by a recover procedure. Give it a shot */ | |
b23c2ade | 3560 | |
d0edb8a1 PX |
3561 | if (postcopy_preempt_wait_channel(s)) { |
3562 | /* | |
3563 | * Preempt enabled, and new channel create failed; loop | |
3564 | * back to wait for another recovery. | |
3565 | */ | |
3566 | continue; | |
3567 | } | |
3568 | ||
135b87b4 PX |
3569 | /* |
3570 | * Firstly, let's wake up the return path now, with a new | |
3571 | * return path channel. | |
3572 | */ | |
3573 | qemu_sem_post(&s->postcopy_pause_rp_sem); | |
3574 | ||
3575 | /* Do the resume logic */ | |
3576 | if (postcopy_do_resume(s) == 0) { | |
3577 | /* Let's continue! */ | |
3578 | trace_postcopy_pause_continued(); | |
3579 | return MIG_THR_ERR_RECOVERED; | |
3580 | } else { | |
3581 | /* | |
3582 | * Something wrong happened during the recovery, let's | |
3583 | * pause again. Pause is always better than throwing | |
3584 | * data away. | |
3585 | */ | |
3586 | continue; | |
3587 | } | |
3588 | } else { | |
3589 | /* This is not right... Time to quit. */ | |
3590 | return MIG_THR_ERR_FATAL; | |
3591 | } | |
3592 | } | |
b23c2ade PX |
3593 | } |
3594 | ||
3595 | static MigThrError migration_detect_error(MigrationState *s) | |
3596 | { | |
3597 | int ret; | |
c3c5eae6 | 3598 | int state = s->state; |
3d661c8a | 3599 | Error *local_error = NULL; |
c3c5eae6 DDAG |
3600 | |
3601 | if (state == MIGRATION_STATUS_CANCELLING || | |
3602 | state == MIGRATION_STATUS_CANCELLED) { | |
3603 | /* End the migration, but don't set the state to failed */ | |
3604 | return MIG_THR_ERR_FATAL; | |
3605 | } | |
b23c2ade | 3606 | |
60bb3c58 PX |
3607 | /* |
3608 | * Try to detect any file errors. Note that postcopy_qemufile_src will | |
3609 | * be NULL when postcopy preempt is not enabled. | |
3610 | */ | |
3611 | ret = qemu_file_get_error_obj_any(s->to_dst_file, | |
3612 | s->postcopy_qemufile_src, | |
3613 | &local_error); | |
b23c2ade PX |
3614 | if (!ret) { |
3615 | /* Everything is fine */ | |
3d661c8a | 3616 | assert(!local_error); |
b23c2ade PX |
3617 | return MIG_THR_ERR_NONE; |
3618 | } | |
3619 | ||
3d661c8a YK |
3620 | if (local_error) { |
3621 | migrate_set_error(s, local_error); | |
3622 | error_free(local_error); | |
3623 | } | |
3624 | ||
d5c8f2af | 3625 | if (state == MIGRATION_STATUS_POSTCOPY_ACTIVE && ret) { |
b23c2ade PX |
3626 | /* |
3627 | * For postcopy, we allow the network to be down for a | |
3628 | * while. After that, it can be continued by a | |
3629 | * recovery phase. | |
3630 | */ | |
3631 | return postcopy_pause(s); | |
3632 | } else { | |
3633 | /* | |
3634 | * For precopy (or postcopy with error outside IO), we fail | |
3635 | * with no time. | |
3636 | */ | |
c3c5eae6 | 3637 | migrate_set_state(&s->state, state, MIGRATION_STATUS_FAILED); |
b23c2ade PX |
3638 | trace_migration_thread_file_err(); |
3639 | ||
3640 | /* Time to stop the migration, now. */ | |
3641 | return MIG_THR_ERR_FATAL; | |
3642 | } | |
3643 | } | |
3644 | ||
268dcd46 | 3645 | /* How many bytes have we transferred since the beginning of the migration */ |
0c8f0efd JQ |
3646 | static uint64_t migration_total_bytes(MigrationState *s) |
3647 | { | |
fbfa6404 DB |
3648 | return qemu_file_total_transferred(s->to_dst_file) + |
3649 | ram_counters.multifd_bytes; | |
0c8f0efd JQ |
3650 | } |
3651 | ||
cf011f08 PX |
3652 | static void migration_calculate_complete(MigrationState *s) |
3653 | { | |
0c8f0efd | 3654 | uint64_t bytes = migration_total_bytes(s); |
cf011f08 | 3655 | int64_t end_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); |
6cde6fbe | 3656 | int64_t transfer_time; |
cf011f08 PX |
3657 | |
3658 | s->total_time = end_time - s->start_time; | |
3659 | if (!s->downtime) { | |
3660 | /* | |
3661 | * It's still not set, so we are precopy migration. For | |
3662 | * postcopy, downtime is calculated during postcopy_start(). | |
3663 | */ | |
3664 | s->downtime = end_time - s->downtime_start; | |
3665 | } | |
3666 | ||
6cde6fbe JQ |
3667 | transfer_time = s->total_time - s->setup_time; |
3668 | if (transfer_time) { | |
3669 | s->mbps = ((double) bytes * 8.0) / transfer_time / 1000; | |
cf011f08 PX |
3670 | } |
3671 | } | |
3672 | ||
87f3bd87 IR |
3673 | static void update_iteration_initial_status(MigrationState *s) |
3674 | { | |
3675 | /* | |
3676 | * Update these three fields at the same time to avoid mismatch info lead | |
3677 | * wrong speed calculation. | |
3678 | */ | |
3679 | s->iteration_start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); | |
3680 | s->iteration_initial_bytes = migration_total_bytes(s); | |
3681 | s->iteration_initial_pages = ram_get_total_transferred_pages(); | |
3682 | } | |
3683 | ||
b15df1ae PX |
3684 | static void migration_update_counters(MigrationState *s, |
3685 | int64_t current_time) | |
3686 | { | |
aecbfe9c | 3687 | uint64_t transferred, transferred_pages, time_spent; |
0c8f0efd | 3688 | uint64_t current_bytes; /* bytes transferred since the beginning */ |
b15df1ae PX |
3689 | double bandwidth; |
3690 | ||
3691 | if (current_time < s->iteration_start_time + BUFFER_DELAY) { | |
3692 | return; | |
3693 | } | |
3694 | ||
0c8f0efd JQ |
3695 | current_bytes = migration_total_bytes(s); |
3696 | transferred = current_bytes - s->iteration_initial_bytes; | |
b15df1ae PX |
3697 | time_spent = current_time - s->iteration_start_time; |
3698 | bandwidth = (double)transferred / time_spent; | |
0781c1ed | 3699 | s->threshold_size = bandwidth * s->parameters.downtime_limit; |
b15df1ae PX |
3700 | |
3701 | s->mbps = (((double) transferred * 8.0) / | |
3702 | ((double) time_spent / 1000.0)) / 1000.0 / 1000.0; | |
3703 | ||
aecbfe9c XG |
3704 | transferred_pages = ram_get_total_transferred_pages() - |
3705 | s->iteration_initial_pages; | |
3706 | s->pages_per_second = (double) transferred_pages / | |
3707 | (((double) time_spent / 1000.0)); | |
3708 | ||
b15df1ae PX |
3709 | /* |
3710 | * if we haven't sent anything, we don't want to | |
3711 | * recalculate. 10000 is a small enough number for our purposes | |
3712 | */ | |
3713 | if (ram_counters.dirty_pages_rate && transferred > 10000) { | |
650af890 | 3714 | s->expected_downtime = ram_counters.remaining / bandwidth; |
b15df1ae PX |
3715 | } |
3716 | ||
3717 | qemu_file_reset_rate_limit(s->to_dst_file); | |
3718 | ||
87f3bd87 | 3719 | update_iteration_initial_status(s); |
b15df1ae PX |
3720 | |
3721 | trace_migrate_transferred(transferred, time_spent, | |
0781c1ed | 3722 | bandwidth, s->threshold_size); |
b15df1ae PX |
3723 | } |
3724 | ||
2ad87305 PX |
3725 | /* Migration thread iteration status */ |
3726 | typedef enum { | |
3727 | MIG_ITERATE_RESUME, /* Resume current iteration */ | |
3728 | MIG_ITERATE_SKIP, /* Skip current iteration */ | |
3729 | MIG_ITERATE_BREAK, /* Break the loop */ | |
3730 | } MigIterateState; | |
3731 | ||
3732 | /* | |
3733 | * Return true if continue to the next iteration directly, false | |
3734 | * otherwise. | |
3735 | */ | |
3736 | static MigIterateState migration_iteration_run(MigrationState *s) | |
3737 | { | |
47995026 | 3738 | uint64_t pending_size, pend_pre, pend_compat, pend_post; |
2ad87305 PX |
3739 | bool in_postcopy = s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE; |
3740 | ||
47995026 VSO |
3741 | qemu_savevm_state_pending(s->to_dst_file, s->threshold_size, &pend_pre, |
3742 | &pend_compat, &pend_post); | |
3743 | pending_size = pend_pre + pend_compat + pend_post; | |
2ad87305 PX |
3744 | |
3745 | trace_migrate_pending(pending_size, s->threshold_size, | |
47995026 | 3746 | pend_pre, pend_compat, pend_post); |
2ad87305 PX |
3747 | |
3748 | if (pending_size && pending_size >= s->threshold_size) { | |
3749 | /* Still a significant amount to transfer */ | |
52aec709 | 3750 | if (!in_postcopy && pend_pre <= s->threshold_size && |
d73415a3 | 3751 | qatomic_read(&s->start_postcopy)) { |
2ad87305 PX |
3752 | if (postcopy_start(s)) { |
3753 | error_report("%s: postcopy failed to start", __func__); | |
3754 | } | |
3755 | return MIG_ITERATE_SKIP; | |
3756 | } | |
3757 | /* Just another iteration step */ | |
17d9351b | 3758 | qemu_savevm_state_iterate(s->to_dst_file, in_postcopy); |
2ad87305 PX |
3759 | } else { |
3760 | trace_migration_thread_low_pending(pending_size); | |
3761 | migration_completion(s); | |
3762 | return MIG_ITERATE_BREAK; | |
3763 | } | |
3764 | ||
3765 | return MIG_ITERATE_RESUME; | |
3766 | } | |
3767 | ||
199aa6d4 PX |
3768 | static void migration_iteration_finish(MigrationState *s) |
3769 | { | |
3770 | /* If we enabled cpu throttling for auto-converge, turn it off. */ | |
3771 | cpu_throttle_stop(); | |
3772 | ||
3773 | qemu_mutex_lock_iothread(); | |
3774 | switch (s->state) { | |
3775 | case MIGRATION_STATUS_COMPLETED: | |
3776 | migration_calculate_complete(s); | |
3777 | runstate_set(RUN_STATE_POSTMIGRATE); | |
3778 | break; | |
751fe4c6 | 3779 | case MIGRATION_STATUS_COLO: |
199aa6d4 PX |
3780 | if (!migrate_colo_enabled()) { |
3781 | error_report("%s: critical error: calling COLO code without " | |
3782 | "COLO enabled", __func__); | |
3783 | } | |
3784 | migrate_start_colo_process(s); | |
199aa6d4 PX |
3785 | s->vm_was_running = true; |
3786 | /* Fallthrough */ | |
3787 | case MIGRATION_STATUS_FAILED: | |
3788 | case MIGRATION_STATUS_CANCELLED: | |
57225e5f | 3789 | case MIGRATION_STATUS_CANCELLING: |
199aa6d4 | 3790 | if (s->vm_was_running) { |
aa505f8e RL |
3791 | if (!runstate_check(RUN_STATE_SHUTDOWN)) { |
3792 | vm_start(); | |
3793 | } | |
199aa6d4 PX |
3794 | } else { |
3795 | if (runstate_check(RUN_STATE_FINISH_MIGRATE)) { | |
3796 | runstate_set(RUN_STATE_POSTMIGRATE); | |
3797 | } | |
3798 | } | |
3799 | break; | |
3800 | ||
3801 | default: | |
3802 | /* Should not reach here, but if so, forgive the VM. */ | |
3803 | error_report("%s: Unknown ending state %d", __func__, s->state); | |
3804 | break; | |
3805 | } | |
fd392cfa | 3806 | migrate_fd_cleanup_schedule(s); |
199aa6d4 PX |
3807 | qemu_mutex_unlock_iothread(); |
3808 | } | |
3809 | ||
8518278a AG |
3810 | static void bg_migration_iteration_finish(MigrationState *s) |
3811 | { | |
3812 | qemu_mutex_lock_iothread(); | |
3813 | switch (s->state) { | |
3814 | case MIGRATION_STATUS_COMPLETED: | |
3815 | migration_calculate_complete(s); | |
3816 | break; | |
3817 | ||
3818 | case MIGRATION_STATUS_ACTIVE: | |
3819 | case MIGRATION_STATUS_FAILED: | |
3820 | case MIGRATION_STATUS_CANCELLED: | |
3821 | case MIGRATION_STATUS_CANCELLING: | |
3822 | break; | |
3823 | ||
3824 | default: | |
3825 | /* Should not reach here, but if so, forgive the VM. */ | |
3826 | error_report("%s: Unknown ending state %d", __func__, s->state); | |
3827 | break; | |
3828 | } | |
3829 | ||
3830 | migrate_fd_cleanup_schedule(s); | |
3831 | qemu_mutex_unlock_iothread(); | |
3832 | } | |
3833 | ||
3834 | /* | |
3835 | * Return true if continue to the next iteration directly, false | |
3836 | * otherwise. | |
3837 | */ | |
3838 | static MigIterateState bg_migration_iteration_run(MigrationState *s) | |
3839 | { | |
3840 | int res; | |
3841 | ||
3842 | res = qemu_savevm_state_iterate(s->to_dst_file, false); | |
3843 | if (res > 0) { | |
3844 | bg_migration_completion(s); | |
3845 | return MIG_ITERATE_BREAK; | |
3846 | } | |
3847 | ||
3848 | return MIG_ITERATE_RESUME; | |
3849 | } | |
3850 | ||
ad767bed DDAG |
3851 | void migration_make_urgent_request(void) |
3852 | { | |
3853 | qemu_sem_post(&migrate_get_current()->rate_limit_sem); | |
3854 | } | |
3855 | ||
3856 | void migration_consume_urgent_request(void) | |
3857 | { | |
3858 | qemu_sem_wait(&migrate_get_current()->rate_limit_sem); | |
3859 | } | |
3860 | ||
97e1e067 DDAG |
3861 | /* Returns true if the rate limiting was broken by an urgent request */ |
3862 | bool migration_rate_limit(void) | |
3863 | { | |
3864 | int64_t now = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); | |
3865 | MigrationState *s = migrate_get_current(); | |
3866 | ||
3867 | bool urgent = false; | |
3868 | migration_update_counters(s, now); | |
3869 | if (qemu_file_rate_limit(s->to_dst_file)) { | |
77386127 LS |
3870 | |
3871 | if (qemu_file_get_error(s->to_dst_file)) { | |
3872 | return false; | |
3873 | } | |
97e1e067 DDAG |
3874 | /* |
3875 | * Wait for a delay to do rate limiting OR | |
3876 | * something urgent to post the semaphore. | |
3877 | */ | |
3878 | int ms = s->iteration_start_time + BUFFER_DELAY - now; | |
3879 | trace_migration_rate_limit_pre(ms); | |
3880 | if (qemu_sem_timedwait(&s->rate_limit_sem, ms) == 0) { | |
3881 | /* | |
3882 | * We were woken by one or more urgent things but | |
3883 | * the timedwait will have consumed one of them. | |
3884 | * The service routine for the urgent wake will dec | |
3885 | * the semaphore itself for each item it consumes, | |
3886 | * so add this one we just eat back. | |
3887 | */ | |
3888 | qemu_sem_post(&s->rate_limit_sem); | |
3889 | urgent = true; | |
3890 | } | |
3891 | trace_migration_rate_limit_post(urgent); | |
3892 | } | |
3893 | return urgent; | |
3894 | } | |
3895 | ||
fde93d99 LV |
3896 | /* |
3897 | * if failover devices are present, wait they are completely | |
3898 | * unplugged | |
3899 | */ | |
3900 | ||
3901 | static void qemu_savevm_wait_unplug(MigrationState *s, int old_state, | |
3902 | int new_state) | |
3903 | { | |
3904 | if (qemu_savevm_state_guest_unplug_pending()) { | |
3905 | migrate_set_state(&s->state, old_state, MIGRATION_STATUS_WAIT_UNPLUG); | |
3906 | ||
3907 | while (s->state == MIGRATION_STATUS_WAIT_UNPLUG && | |
3908 | qemu_savevm_state_guest_unplug_pending()) { | |
3909 | qemu_sem_timedwait(&s->wait_unplug_sem, 250); | |
3910 | } | |
944bc528 LV |
3911 | if (s->state != MIGRATION_STATUS_WAIT_UNPLUG) { |
3912 | int timeout = 120; /* 30 seconds */ | |
3913 | /* | |
3914 | * migration has been canceled | |
3915 | * but as we have started an unplug we must wait the end | |
3916 | * to be able to plug back the card | |
3917 | */ | |
3918 | while (timeout-- && qemu_savevm_state_guest_unplug_pending()) { | |
3919 | qemu_sem_timedwait(&s->wait_unplug_sem, 250); | |
3920 | } | |
1b529d90 LV |
3921 | if (qemu_savevm_state_guest_unplug_pending() && |
3922 | !qtest_enabled()) { | |
a51dcef0 LV |
3923 | warn_report("migration: partially unplugged device on " |
3924 | "failure"); | |
3925 | } | |
944bc528 | 3926 | } |
fde93d99 LV |
3927 | |
3928 | migrate_set_state(&s->state, MIGRATION_STATUS_WAIT_UNPLUG, new_state); | |
3929 | } else { | |
3930 | migrate_set_state(&s->state, old_state, new_state); | |
3931 | } | |
3932 | } | |
3933 | ||
70b20477 DDAG |
3934 | /* |
3935 | * Master migration thread on the source VM. | |
3936 | * It drives the migration and pumps the data down the outgoing channel. | |
3937 | */ | |
5f496a1b | 3938 | static void *migration_thread(void *opaque) |
0d82d0e8 | 3939 | { |
9848a404 | 3940 | MigrationState *s = opaque; |
bc72ad67 | 3941 | int64_t setup_start = qemu_clock_get_ms(QEMU_CLOCK_HOST); |
b23c2ade | 3942 | MigThrError thr_error; |
ad767bed | 3943 | bool urgent = false; |
76f5933a | 3944 | |
ab28bd23 PB |
3945 | rcu_register_thread(); |
3946 | ||
892ae715 | 3947 | object_ref(OBJECT(s)); |
87f3bd87 | 3948 | update_iteration_initial_status(s); |
b15df1ae | 3949 | |
89a02a9f | 3950 | qemu_savevm_state_header(s->to_dst_file); |
1d34e4bf | 3951 | |
62a02658 PX |
3952 | /* |
3953 | * If we opened the return path, we need to make sure dst has it | |
3954 | * opened as well. | |
3955 | */ | |
43044ac0 | 3956 | if (s->rp_state.rp_thread_created) { |
1d34e4bf | 3957 | /* Now tell the dest that it should open its end so it can reply */ |
89a02a9f | 3958 | qemu_savevm_send_open_return_path(s->to_dst_file); |
1d34e4bf DDAG |
3959 | |
3960 | /* And do a ping that will make stuff easier to debug */ | |
89a02a9f | 3961 | qemu_savevm_send_ping(s->to_dst_file, 1); |
0425dc97 | 3962 | } |
1d34e4bf | 3963 | |
58110f0a | 3964 | if (migrate_postcopy()) { |
1d34e4bf DDAG |
3965 | /* |
3966 | * Tell the destination that we *might* want to do postcopy later; | |
3967 | * if the other end can't do postcopy it should fail now, nice and | |
3968 | * early. | |
3969 | */ | |
89a02a9f | 3970 | qemu_savevm_send_postcopy_advise(s->to_dst_file); |
1d34e4bf DDAG |
3971 | } |
3972 | ||
aad555c2 ZC |
3973 | if (migrate_colo_enabled()) { |
3974 | /* Notify migration destination that we enable COLO */ | |
3975 | qemu_savevm_send_colo_enable(s->to_dst_file); | |
3976 | } | |
3977 | ||
9907e842 | 3978 | qemu_savevm_state_setup(s->to_dst_file); |
0d82d0e8 | 3979 | |
fde93d99 LV |
3980 | qemu_savevm_wait_unplug(s, MIGRATION_STATUS_SETUP, |
3981 | MIGRATION_STATUS_ACTIVE); | |
c7e0acd5 | 3982 | |
bc72ad67 | 3983 | s->setup_time = qemu_clock_get_ms(QEMU_CLOCK_HOST) - setup_start; |
29ae8a41 | 3984 | |
9ec055ae DDAG |
3985 | trace_migration_thread_setup_complete(); |
3986 | ||
8f8d528e | 3987 | while (migration_is_active(s)) { |
ad767bed | 3988 | if (urgent || !qemu_file_rate_limit(s->to_dst_file)) { |
2ad87305 PX |
3989 | MigIterateState iter_state = migration_iteration_run(s); |
3990 | if (iter_state == MIG_ITERATE_SKIP) { | |
3991 | continue; | |
3992 | } else if (iter_state == MIG_ITERATE_BREAK) { | |
09f6c85e | 3993 | break; |
c369f40d JQ |
3994 | } |
3995 | } | |
f4410a5d | 3996 | |
b23c2ade PX |
3997 | /* |
3998 | * Try to detect any kind of failures, and see whether we | |
3999 | * should stop the migration now. | |
4000 | */ | |
4001 | thr_error = migration_detect_error(s); | |
4002 | if (thr_error == MIG_THR_ERR_FATAL) { | |
4003 | /* Stop migration */ | |
fd45ee2c | 4004 | break; |
b23c2ade PX |
4005 | } else if (thr_error == MIG_THR_ERR_RECOVERED) { |
4006 | /* | |
4007 | * Just recovered from a e.g. network failure, reset all | |
4008 | * the local variables. This is important to avoid | |
4009 | * breaking transferred_bytes and bandwidth calculation | |
4010 | */ | |
87f3bd87 | 4011 | update_iteration_initial_status(s); |
fd45ee2c | 4012 | } |
b15df1ae | 4013 | |
97e1e067 | 4014 | urgent = migration_rate_limit(); |
a3fa1d78 PB |
4015 | } |
4016 | ||
1d34e4bf | 4017 | trace_migration_thread_after_loop(); |
199aa6d4 | 4018 | migration_iteration_finish(s); |
892ae715 | 4019 | object_unref(OBJECT(s)); |
ab28bd23 | 4020 | rcu_unregister_thread(); |
0d82d0e8 JQ |
4021 | return NULL; |
4022 | } | |
4023 | ||
8518278a AG |
4024 | static void bg_migration_vm_start_bh(void *opaque) |
4025 | { | |
4026 | MigrationState *s = opaque; | |
4027 | ||
4028 | qemu_bh_delete(s->vm_start_bh); | |
4029 | s->vm_start_bh = NULL; | |
4030 | ||
4031 | vm_start(); | |
4032 | s->downtime = qemu_clock_get_ms(QEMU_CLOCK_REALTIME) - s->downtime_start; | |
4033 | } | |
4034 | ||
4035 | /** | |
4036 | * Background snapshot thread, based on live migration code. | |
4037 | * This is an alternative implementation of live migration mechanism | |
4038 | * introduced specifically to support background snapshots. | |
4039 | * | |
4040 | * It takes advantage of userfault_fd write protection mechanism introduced | |
4041 | * in v5.7 kernel. Compared to existing dirty page logging migration much | |
4042 | * lesser stream traffic is produced resulting in smaller snapshot images, | |
4043 | * simply cause of no page duplicates can get into the stream. | |
4044 | * | |
4045 | * Another key point is that generated vmstate stream reflects machine state | |
4046 | * 'frozen' at the beginning of snapshot creation compared to dirty page logging | |
4047 | * mechanism, which effectively results in that saved snapshot is the state of VM | |
4048 | * at the end of the process. | |
4049 | */ | |
4050 | static void *bg_migration_thread(void *opaque) | |
4051 | { | |
4052 | MigrationState *s = opaque; | |
4053 | int64_t setup_start; | |
4054 | MigThrError thr_error; | |
4055 | QEMUFile *fb; | |
4056 | bool early_fail = true; | |
4057 | ||
4058 | rcu_register_thread(); | |
4059 | object_ref(OBJECT(s)); | |
4060 | ||
4061 | qemu_file_set_rate_limit(s->to_dst_file, INT64_MAX); | |
4062 | ||
4063 | setup_start = qemu_clock_get_ms(QEMU_CLOCK_HOST); | |
4064 | /* | |
4065 | * We want to save vmstate for the moment when migration has been | |
4066 | * initiated but also we want to save RAM content while VM is running. | |
4067 | * The RAM content should appear first in the vmstate. So, we first | |
4068 | * stash the non-RAM part of the vmstate to the temporary buffer, | |
4069 | * then write RAM part of the vmstate to the migration stream | |
4070 | * with vCPUs running and, finally, write stashed non-RAM part of | |
4071 | * the vmstate from the buffer to the migration stream. | |
4072 | */ | |
ecb23efe | 4073 | s->bioc = qio_channel_buffer_new(512 * 1024); |
8518278a | 4074 | qio_channel_set_name(QIO_CHANNEL(s->bioc), "vmstate-buffer"); |
77ef2dc1 | 4075 | fb = qemu_file_new_output(QIO_CHANNEL(s->bioc)); |
8518278a AG |
4076 | object_unref(OBJECT(s->bioc)); |
4077 | ||
4078 | update_iteration_initial_status(s); | |
4079 | ||
eeccb99c AG |
4080 | /* |
4081 | * Prepare for tracking memory writes with UFFD-WP - populate | |
4082 | * RAM pages before protecting. | |
4083 | */ | |
4084 | #ifdef __linux__ | |
4085 | ram_write_tracking_prepare(); | |
4086 | #endif | |
4087 | ||
8518278a AG |
4088 | qemu_savevm_state_header(s->to_dst_file); |
4089 | qemu_savevm_state_setup(s->to_dst_file); | |
4090 | ||
fde93d99 LV |
4091 | qemu_savevm_wait_unplug(s, MIGRATION_STATUS_SETUP, |
4092 | MIGRATION_STATUS_ACTIVE); | |
8518278a | 4093 | |
8518278a AG |
4094 | s->setup_time = qemu_clock_get_ms(QEMU_CLOCK_HOST) - setup_start; |
4095 | ||
4096 | trace_migration_thread_setup_complete(); | |
4097 | s->downtime_start = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); | |
4098 | ||
4099 | qemu_mutex_lock_iothread(); | |
4100 | ||
4101 | /* | |
4102 | * If VM is currently in suspended state, then, to make a valid runstate | |
4103 | * transition in vm_stop_force_state() we need to wakeup it up. | |
4104 | */ | |
4105 | qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER, NULL); | |
4106 | s->vm_was_running = runstate_is_running(); | |
4107 | ||
4108 | if (global_state_store()) { | |
4109 | goto fail; | |
4110 | } | |
4111 | /* Forcibly stop VM before saving state of vCPUs and devices */ | |
4112 | if (vm_stop_force_state(RUN_STATE_PAUSED)) { | |
4113 | goto fail; | |
4114 | } | |
4115 | /* | |
4116 | * Put vCPUs in sync with shadow context structures, then | |
4117 | * save their state to channel-buffer along with devices. | |
4118 | */ | |
4119 | cpu_synchronize_all_states(); | |
4120 | if (qemu_savevm_state_complete_precopy_non_iterable(fb, false, false)) { | |
4121 | goto fail; | |
4122 | } | |
ecb23efe AG |
4123 | /* |
4124 | * Since we are going to get non-iterable state data directly | |
4125 | * from s->bioc->data, explicit flush is needed here. | |
4126 | */ | |
4127 | qemu_fflush(fb); | |
4128 | ||
8518278a AG |
4129 | /* Now initialize UFFD context and start tracking RAM writes */ |
4130 | if (ram_write_tracking_start()) { | |
4131 | goto fail; | |
4132 | } | |
4133 | early_fail = false; | |
4134 | ||
4135 | /* | |
4136 | * Start VM from BH handler to avoid write-fault lock here. | |
4137 | * UFFD-WP protection for the whole RAM is already enabled so | |
4138 | * calling VM state change notifiers from vm_start() would initiate | |
4139 | * writes to virtio VQs memory which is in write-protected region. | |
4140 | */ | |
4141 | s->vm_start_bh = qemu_bh_new(bg_migration_vm_start_bh, s); | |
4142 | qemu_bh_schedule(s->vm_start_bh); | |
4143 | ||
4144 | qemu_mutex_unlock_iothread(); | |
4145 | ||
4146 | while (migration_is_active(s)) { | |
4147 | MigIterateState iter_state = bg_migration_iteration_run(s); | |
4148 | if (iter_state == MIG_ITERATE_SKIP) { | |
4149 | continue; | |
4150 | } else if (iter_state == MIG_ITERATE_BREAK) { | |
4151 | break; | |
4152 | } | |
4153 | ||
4154 | /* | |
4155 | * Try to detect any kind of failures, and see whether we | |
4156 | * should stop the migration now. | |
4157 | */ | |
4158 | thr_error = migration_detect_error(s); | |
4159 | if (thr_error == MIG_THR_ERR_FATAL) { | |
4160 | /* Stop migration */ | |
4161 | break; | |
4162 | } | |
4163 | ||
4164 | migration_update_counters(s, qemu_clock_get_ms(QEMU_CLOCK_REALTIME)); | |
4165 | } | |
4166 | ||
4167 | trace_migration_thread_after_loop(); | |
4168 | ||
4169 | fail: | |
4170 | if (early_fail) { | |
4171 | migrate_set_state(&s->state, MIGRATION_STATUS_ACTIVE, | |
4172 | MIGRATION_STATUS_FAILED); | |
4173 | qemu_mutex_unlock_iothread(); | |
4174 | } | |
4175 | ||
4176 | bg_migration_iteration_finish(s); | |
4177 | ||
4178 | qemu_fclose(fb); | |
4179 | object_unref(OBJECT(s)); | |
4180 | rcu_unregister_thread(); | |
4181 | ||
4182 | return NULL; | |
4183 | } | |
4184 | ||
cce8040b | 4185 | void migrate_fd_connect(MigrationState *s, Error *error_in) |
0d82d0e8 | 4186 | { |
00f4b572 | 4187 | Error *local_err = NULL; |
d3e35b8f PX |
4188 | int64_t rate_limit; |
4189 | bool resume = s->state == MIGRATION_STATUS_POSTCOPY_PAUSED; | |
4190 | ||
ca7bd082 PX |
4191 | /* |
4192 | * If there's a previous error, free it and prepare for another one. | |
4193 | * Meanwhile if migration completes successfully, there won't have an error | |
4194 | * dumped when calling migrate_fd_cleanup(). | |
4195 | */ | |
4196 | migrate_error_free(s); | |
4197 | ||
2ff30257 | 4198 | s->expected_downtime = s->parameters.downtime_limit; |
9cbc3649 MAL |
4199 | if (resume) { |
4200 | assert(s->cleanup_bh); | |
4201 | } else { | |
4202 | assert(!s->cleanup_bh); | |
4203 | s->cleanup_bh = qemu_bh_new(migrate_fd_cleanup_bh, s); | |
4204 | } | |
cce8040b DDAG |
4205 | if (error_in) { |
4206 | migrate_fd_error(s, error_in); | |
ca30f24d PX |
4207 | if (resume) { |
4208 | /* | |
4209 | * Don't do cleanup for resume if channel is invalid, but only dump | |
4210 | * the error. We wait for another channel connect from the user. | |
4211 | * The error_report still gives HMP user a hint on what failed. | |
4212 | * It's normally done in migrate_fd_cleanup(), but call it here | |
4213 | * explicitly. | |
4214 | */ | |
4215 | error_report_err(error_copy(s->error)); | |
4216 | } else { | |
4217 | migrate_fd_cleanup(s); | |
4218 | } | |
cce8040b DDAG |
4219 | return; |
4220 | } | |
0d82d0e8 | 4221 | |
d3e35b8f PX |
4222 | if (resume) { |
4223 | /* This is a resumed migration */ | |
8504ddec PX |
4224 | rate_limit = s->parameters.max_postcopy_bandwidth / |
4225 | XFER_LIMIT_RATIO; | |
d3e35b8f PX |
4226 | } else { |
4227 | /* This is a fresh new migration */ | |
4228 | rate_limit = s->parameters.max_bandwidth / XFER_LIMIT_RATIO; | |
442773ce | 4229 | |
d3e35b8f PX |
4230 | /* Notify before starting migration thread */ |
4231 | notifier_list_notify(&migration_state_notifiers, s); | |
4232 | } | |
4233 | ||
4234 | qemu_file_set_rate_limit(s->to_dst_file, rate_limit); | |
4235 | qemu_file_set_blocking(s->to_dst_file, true); | |
9287ac27 | 4236 | |
1d34e4bf | 4237 | /* |
c788ada8 PX |
4238 | * Open the return path. For postcopy, it is used exclusively. For |
4239 | * precopy, only if user specified "return-path" capability would | |
4240 | * QEMU uses the return path. | |
1d34e4bf | 4241 | */ |
c788ada8 | 4242 | if (migrate_postcopy_ram() || migrate_use_return_path()) { |
d3e35b8f | 4243 | if (open_return_path_on_source(s, !resume)) { |
1d34e4bf | 4244 | error_report("Unable to open return-path for postcopy"); |
d3e35b8f | 4245 | migrate_set_state(&s->state, s->state, MIGRATION_STATUS_FAILED); |
1d34e4bf DDAG |
4246 | migrate_fd_cleanup(s); |
4247 | return; | |
4248 | } | |
4249 | } | |
4250 | ||
36f62f11 PX |
4251 | /* This needs to be done before resuming a postcopy */ |
4252 | if (postcopy_preempt_setup(s, &local_err)) { | |
4253 | error_report_err(local_err); | |
4254 | migrate_set_state(&s->state, MIGRATION_STATUS_SETUP, | |
4255 | MIGRATION_STATUS_FAILED); | |
4256 | migrate_fd_cleanup(s); | |
4257 | return; | |
4258 | } | |
4259 | ||
d3e35b8f | 4260 | if (resume) { |
135b87b4 PX |
4261 | /* Wakeup the main migration thread to do the recovery */ |
4262 | migrate_set_state(&s->state, MIGRATION_STATUS_POSTCOPY_PAUSED, | |
4263 | MIGRATION_STATUS_POSTCOPY_RECOVER); | |
4264 | qemu_sem_post(&s->postcopy_pause_sem); | |
d3e35b8f PX |
4265 | return; |
4266 | } | |
4267 | ||
00f4b572 JQ |
4268 | if (multifd_save_setup(&local_err) != 0) { |
4269 | error_report_err(local_err); | |
f986c3d2 JQ |
4270 | migrate_set_state(&s->state, MIGRATION_STATUS_SETUP, |
4271 | MIGRATION_STATUS_FAILED); | |
4272 | migrate_fd_cleanup(s); | |
4273 | return; | |
4274 | } | |
8518278a AG |
4275 | |
4276 | if (migrate_background_snapshot()) { | |
4277 | qemu_thread_create(&s->thread, "bg_snapshot", | |
4278 | bg_migration_thread, s, QEMU_THREAD_JOINABLE); | |
4279 | } else { | |
4280 | qemu_thread_create(&s->thread, "live_migration", | |
4281 | migration_thread, s, QEMU_THREAD_JOINABLE); | |
4282 | } | |
1d34e4bf | 4283 | s->migration_thread_running = true; |
0d82d0e8 | 4284 | } |
093e3c42 | 4285 | |
9d18af93 PX |
4286 | void migration_global_dump(Monitor *mon) |
4287 | { | |
4288 | MigrationState *ms = migrate_get_current(); | |
4289 | ||
6f0f6428 JQ |
4290 | monitor_printf(mon, "globals:\n"); |
4291 | monitor_printf(mon, "store-global-state: %s\n", | |
4292 | ms->store_global_state ? "on" : "off"); | |
4293 | monitor_printf(mon, "only-migratable: %s\n", | |
811f8652 | 4294 | only_migratable ? "on" : "off"); |
6f0f6428 JQ |
4295 | monitor_printf(mon, "send-configuration: %s\n", |
4296 | ms->send_configuration ? "on" : "off"); | |
4297 | monitor_printf(mon, "send-section-footer: %s\n", | |
4298 | ms->send_section_footer ? "on" : "off"); | |
f548222c XG |
4299 | monitor_printf(mon, "decompress-error-check: %s\n", |
4300 | ms->decompress_error_check ? "on" : "off"); | |
002cad6b PX |
4301 | monitor_printf(mon, "clear-bitmap-shift: %u\n", |
4302 | ms->clear_bitmap_shift); | |
9d18af93 PX |
4303 | } |
4304 | ||
20814758 PX |
4305 | #define DEFINE_PROP_MIG_CAP(name, x) \ |
4306 | DEFINE_PROP_BOOL(name, MigrationState, enabled_capabilities[x], false) | |
4307 | ||
5272298c PX |
4308 | static Property migration_properties[] = { |
4309 | DEFINE_PROP_BOOL("store-global-state", MigrationState, | |
4310 | store_global_state, true), | |
71dd4c1a PX |
4311 | DEFINE_PROP_BOOL("send-configuration", MigrationState, |
4312 | send_configuration, true), | |
15c38503 PX |
4313 | DEFINE_PROP_BOOL("send-section-footer", MigrationState, |
4314 | send_section_footer, true), | |
f548222c XG |
4315 | DEFINE_PROP_BOOL("decompress-error-check", MigrationState, |
4316 | decompress_error_check, true), | |
002cad6b PX |
4317 | DEFINE_PROP_UINT8("x-clear-bitmap-shift", MigrationState, |
4318 | clear_bitmap_shift, CLEAR_BITMAP_SHIFT_DEFAULT), | |
89632faf PX |
4319 | |
4320 | /* Migration parameters */ | |
741d4086 | 4321 | DEFINE_PROP_UINT8("x-compress-level", MigrationState, |
89632faf PX |
4322 | parameters.compress_level, |
4323 | DEFAULT_MIGRATE_COMPRESS_LEVEL), | |
741d4086 | 4324 | DEFINE_PROP_UINT8("x-compress-threads", MigrationState, |
89632faf PX |
4325 | parameters.compress_threads, |
4326 | DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT), | |
1d58872a XG |
4327 | DEFINE_PROP_BOOL("x-compress-wait-thread", MigrationState, |
4328 | parameters.compress_wait_thread, true), | |
741d4086 | 4329 | DEFINE_PROP_UINT8("x-decompress-threads", MigrationState, |
89632faf PX |
4330 | parameters.decompress_threads, |
4331 | DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT), | |
dc14a470 KZ |
4332 | DEFINE_PROP_UINT8("x-throttle-trigger-threshold", MigrationState, |
4333 | parameters.throttle_trigger_threshold, | |
4334 | DEFAULT_MIGRATE_THROTTLE_TRIGGER_THRESHOLD), | |
741d4086 | 4335 | DEFINE_PROP_UINT8("x-cpu-throttle-initial", MigrationState, |
89632faf PX |
4336 | parameters.cpu_throttle_initial, |
4337 | DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL), | |
741d4086 | 4338 | DEFINE_PROP_UINT8("x-cpu-throttle-increment", MigrationState, |
89632faf PX |
4339 | parameters.cpu_throttle_increment, |
4340 | DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT), | |
cbbf8182 KZ |
4341 | DEFINE_PROP_BOOL("x-cpu-throttle-tailslow", MigrationState, |
4342 | parameters.cpu_throttle_tailslow, false), | |
741d4086 | 4343 | DEFINE_PROP_SIZE("x-max-bandwidth", MigrationState, |
89632faf | 4344 | parameters.max_bandwidth, MAX_THROTTLE), |
741d4086 | 4345 | DEFINE_PROP_UINT64("x-downtime-limit", MigrationState, |
89632faf PX |
4346 | parameters.downtime_limit, |
4347 | DEFAULT_MIGRATE_SET_DOWNTIME), | |
741d4086 | 4348 | DEFINE_PROP_UINT32("x-checkpoint-delay", MigrationState, |
89632faf PX |
4349 | parameters.x_checkpoint_delay, |
4350 | DEFAULT_MIGRATE_X_CHECKPOINT_DELAY), | |
cbfd6c95 JQ |
4351 | DEFINE_PROP_UINT8("multifd-channels", MigrationState, |
4352 | parameters.multifd_channels, | |
4075fb1c | 4353 | DEFAULT_MIGRATE_MULTIFD_CHANNELS), |
96eef042 JQ |
4354 | DEFINE_PROP_MULTIFD_COMPRESSION("multifd-compression", MigrationState, |
4355 | parameters.multifd_compression, | |
4356 | DEFAULT_MIGRATE_MULTIFD_COMPRESSION), | |
9004db48 JQ |
4357 | DEFINE_PROP_UINT8("multifd-zlib-level", MigrationState, |
4358 | parameters.multifd_zlib_level, | |
4359 | DEFAULT_MIGRATE_MULTIFD_ZLIB_LEVEL), | |
6a9ad154 JQ |
4360 | DEFINE_PROP_UINT8("multifd-zstd-level", MigrationState, |
4361 | parameters.multifd_zstd_level, | |
4362 | DEFAULT_MIGRATE_MULTIFD_ZSTD_LEVEL), | |
73af8dd8 JQ |
4363 | DEFINE_PROP_SIZE("xbzrle-cache-size", MigrationState, |
4364 | parameters.xbzrle_cache_size, | |
4365 | DEFAULT_MIGRATE_XBZRLE_CACHE_SIZE), | |
7e555c6c DDAG |
4366 | DEFINE_PROP_SIZE("max-postcopy-bandwidth", MigrationState, |
4367 | parameters.max_postcopy_bandwidth, | |
4368 | DEFAULT_MIGRATE_MAX_POSTCOPY_BANDWIDTH), | |
4cbc9c7f LQ |
4369 | DEFINE_PROP_UINT8("max-cpu-throttle", MigrationState, |
4370 | parameters.max_cpu_throttle, | |
4371 | DEFAULT_MIGRATE_MAX_CPU_THROTTLE), | |
ee3d96ba DDAG |
4372 | DEFINE_PROP_SIZE("announce-initial", MigrationState, |
4373 | parameters.announce_initial, | |
4374 | DEFAULT_MIGRATE_ANNOUNCE_INITIAL), | |
4375 | DEFINE_PROP_SIZE("announce-max", MigrationState, | |
4376 | parameters.announce_max, | |
4377 | DEFAULT_MIGRATE_ANNOUNCE_MAX), | |
4378 | DEFINE_PROP_SIZE("announce-rounds", MigrationState, | |
4379 | parameters.announce_rounds, | |
4380 | DEFAULT_MIGRATE_ANNOUNCE_ROUNDS), | |
4381 | DEFINE_PROP_SIZE("announce-step", MigrationState, | |
4382 | parameters.announce_step, | |
4383 | DEFAULT_MIGRATE_ANNOUNCE_STEP), | |
c8750de1 PX |
4384 | DEFINE_PROP_BOOL("x-postcopy-preempt-break-huge", MigrationState, |
4385 | postcopy_preempt_break_huge, true), | |
9a266627 PX |
4386 | DEFINE_PROP_STRING("tls-creds", MigrationState, parameters.tls_creds), |
4387 | DEFINE_PROP_STRING("tls-hostname", MigrationState, parameters.tls_hostname), | |
4388 | DEFINE_PROP_STRING("tls-authz", MigrationState, parameters.tls_authz), | |
20814758 PX |
4389 | |
4390 | /* Migration capabilities */ | |
4391 | DEFINE_PROP_MIG_CAP("x-xbzrle", MIGRATION_CAPABILITY_XBZRLE), | |
4392 | DEFINE_PROP_MIG_CAP("x-rdma-pin-all", MIGRATION_CAPABILITY_RDMA_PIN_ALL), | |
4393 | DEFINE_PROP_MIG_CAP("x-auto-converge", MIGRATION_CAPABILITY_AUTO_CONVERGE), | |
4394 | DEFINE_PROP_MIG_CAP("x-zero-blocks", MIGRATION_CAPABILITY_ZERO_BLOCKS), | |
4395 | DEFINE_PROP_MIG_CAP("x-compress", MIGRATION_CAPABILITY_COMPRESS), | |
4396 | DEFINE_PROP_MIG_CAP("x-events", MIGRATION_CAPABILITY_EVENTS), | |
4397 | DEFINE_PROP_MIG_CAP("x-postcopy-ram", MIGRATION_CAPABILITY_POSTCOPY_RAM), | |
ce5b0f4a PX |
4398 | DEFINE_PROP_MIG_CAP("x-postcopy-preempt", |
4399 | MIGRATION_CAPABILITY_POSTCOPY_PREEMPT), | |
20814758 PX |
4400 | DEFINE_PROP_MIG_CAP("x-colo", MIGRATION_CAPABILITY_X_COLO), |
4401 | DEFINE_PROP_MIG_CAP("x-release-ram", MIGRATION_CAPABILITY_RELEASE_RAM), | |
4402 | DEFINE_PROP_MIG_CAP("x-block", MIGRATION_CAPABILITY_BLOCK), | |
4403 | DEFINE_PROP_MIG_CAP("x-return-path", MIGRATION_CAPABILITY_RETURN_PATH), | |
cbfd6c95 | 4404 | DEFINE_PROP_MIG_CAP("x-multifd", MIGRATION_CAPABILITY_MULTIFD), |
6e8c25b4 AG |
4405 | DEFINE_PROP_MIG_CAP("x-background-snapshot", |
4406 | MIGRATION_CAPABILITY_BACKGROUND_SNAPSHOT), | |
1abaec9a LB |
4407 | #ifdef CONFIG_LINUX |
4408 | DEFINE_PROP_MIG_CAP("x-zero-copy-send", | |
4409 | MIGRATION_CAPABILITY_ZERO_COPY_SEND), | |
4410 | #endif | |
20814758 | 4411 | |
5272298c PX |
4412 | DEFINE_PROP_END_OF_LIST(), |
4413 | }; | |
4414 | ||
e5cb7e76 PX |
4415 | static void migration_class_init(ObjectClass *klass, void *data) |
4416 | { | |
4417 | DeviceClass *dc = DEVICE_CLASS(klass); | |
4418 | ||
4419 | dc->user_creatable = false; | |
4f67d30b | 4420 | device_class_set_props(dc, migration_properties); |
e5cb7e76 PX |
4421 | } |
4422 | ||
b91bf5e4 MAL |
4423 | static void migration_instance_finalize(Object *obj) |
4424 | { | |
4425 | MigrationState *ms = MIGRATION_OBJ(obj); | |
b91bf5e4 | 4426 | |
87db1a7d | 4427 | qemu_mutex_destroy(&ms->error_mutex); |
62df066f | 4428 | qemu_mutex_destroy(&ms->qemu_file_lock); |
c7e0acd5 | 4429 | qemu_sem_destroy(&ms->wait_unplug_sem); |
ad767bed | 4430 | qemu_sem_destroy(&ms->rate_limit_sem); |
e91d8951 | 4431 | qemu_sem_destroy(&ms->pause_sem); |
b23c2ade | 4432 | qemu_sem_destroy(&ms->postcopy_pause_sem); |
14b1742e | 4433 | qemu_sem_destroy(&ms->postcopy_pause_rp_sem); |
edd090c7 | 4434 | qemu_sem_destroy(&ms->rp_state.rp_sem); |
d0edb8a1 | 4435 | qemu_sem_destroy(&ms->postcopy_qemufile_src_sem); |
ab105cc1 | 4436 | error_free(ms->error); |
b91bf5e4 MAL |
4437 | } |
4438 | ||
e5cb7e76 PX |
4439 | static void migration_instance_init(Object *obj) |
4440 | { | |
4441 | MigrationState *ms = MIGRATION_OBJ(obj); | |
8b0b29dc | 4442 | MigrationParameters *params = &ms->parameters; |
e5cb7e76 PX |
4443 | |
4444 | ms->state = MIGRATION_STATUS_NONE; | |
e5cb7e76 | 4445 | ms->mbps = -1; |
aecbfe9c | 4446 | ms->pages_per_second = -1; |
e91d8951 | 4447 | qemu_sem_init(&ms->pause_sem, 0); |
87db1a7d | 4448 | qemu_mutex_init(&ms->error_mutex); |
8b0b29dc PX |
4449 | |
4450 | params->tls_hostname = g_strdup(""); | |
4451 | params->tls_creds = g_strdup(""); | |
4452 | ||
4453 | /* Set has_* up only for parameter checks */ | |
4454 | params->has_compress_level = true; | |
4455 | params->has_compress_threads = true; | |
df67aa3e | 4456 | params->has_compress_wait_thread = true; |
8b0b29dc | 4457 | params->has_decompress_threads = true; |
dc14a470 | 4458 | params->has_throttle_trigger_threshold = true; |
8b0b29dc PX |
4459 | params->has_cpu_throttle_initial = true; |
4460 | params->has_cpu_throttle_increment = true; | |
cbbf8182 | 4461 | params->has_cpu_throttle_tailslow = true; |
8b0b29dc PX |
4462 | params->has_max_bandwidth = true; |
4463 | params->has_downtime_limit = true; | |
4464 | params->has_x_checkpoint_delay = true; | |
4465 | params->has_block_incremental = true; | |
cbfd6c95 | 4466 | params->has_multifd_channels = true; |
96eef042 | 4467 | params->has_multifd_compression = true; |
9004db48 | 4468 | params->has_multifd_zlib_level = true; |
6a9ad154 | 4469 | params->has_multifd_zstd_level = true; |
73af8dd8 | 4470 | params->has_xbzrle_cache_size = true; |
7e555c6c | 4471 | params->has_max_postcopy_bandwidth = true; |
4cbc9c7f | 4472 | params->has_max_cpu_throttle = true; |
ee3d96ba DDAG |
4473 | params->has_announce_initial = true; |
4474 | params->has_announce_max = true; | |
4475 | params->has_announce_rounds = true; | |
4476 | params->has_announce_step = true; | |
df67aa3e LB |
4477 | params->has_tls_creds = true; |
4478 | params->has_tls_hostname = true; | |
4479 | params->has_tls_authz = true; | |
b23c2ade PX |
4480 | |
4481 | qemu_sem_init(&ms->postcopy_pause_sem, 0); | |
14b1742e | 4482 | qemu_sem_init(&ms->postcopy_pause_rp_sem, 0); |
edd090c7 | 4483 | qemu_sem_init(&ms->rp_state.rp_sem, 0); |
ad767bed | 4484 | qemu_sem_init(&ms->rate_limit_sem, 0); |
c7e0acd5 | 4485 | qemu_sem_init(&ms->wait_unplug_sem, 0); |
d0edb8a1 | 4486 | qemu_sem_init(&ms->postcopy_qemufile_src_sem, 0); |
62df066f | 4487 | qemu_mutex_init(&ms->qemu_file_lock); |
8b0b29dc PX |
4488 | } |
4489 | ||
4490 | /* | |
4491 | * Return true if check pass, false otherwise. Error will be put | |
4492 | * inside errp if provided. | |
4493 | */ | |
4494 | static bool migration_object_check(MigrationState *ms, Error **errp) | |
4495 | { | |
6b19a7d9 PX |
4496 | MigrationCapabilityStatusList *head = NULL; |
4497 | /* Assuming all off */ | |
4498 | bool cap_list[MIGRATION_CAPABILITY__MAX] = { 0 }, ret; | |
4499 | int i; | |
4500 | ||
8b0b29dc PX |
4501 | if (!migrate_params_check(&ms->parameters, errp)) { |
4502 | return false; | |
4503 | } | |
4504 | ||
6b19a7d9 PX |
4505 | for (i = 0; i < MIGRATION_CAPABILITY__MAX; i++) { |
4506 | if (ms->enabled_capabilities[i]) { | |
eaedde52 | 4507 | QAPI_LIST_PREPEND(head, migrate_cap_add(i, true)); |
6b19a7d9 PX |
4508 | } |
4509 | } | |
4510 | ||
4511 | ret = migrate_caps_check(cap_list, head, errp); | |
4512 | ||
4513 | /* It works with head == NULL */ | |
4514 | qapi_free_MigrationCapabilityStatusList(head); | |
4515 | ||
4516 | return ret; | |
e5cb7e76 PX |
4517 | } |
4518 | ||
4519 | static const TypeInfo migration_type = { | |
4520 | .name = TYPE_MIGRATION, | |
01f6e14c | 4521 | /* |
c8d3ff38 | 4522 | * NOTE: TYPE_MIGRATION is not really a device, as the object is |
2194abd6 | 4523 | * not created using qdev_new(), it is not attached to the qdev |
c8d3ff38 PX |
4524 | * device tree, and it is never realized. |
4525 | * | |
4526 | * TODO: Make this TYPE_OBJECT once QOM provides something like | |
4527 | * TYPE_DEVICE's "-global" properties. | |
01f6e14c | 4528 | */ |
e5cb7e76 PX |
4529 | .parent = TYPE_DEVICE, |
4530 | .class_init = migration_class_init, | |
4531 | .class_size = sizeof(MigrationClass), | |
4532 | .instance_size = sizeof(MigrationState), | |
4533 | .instance_init = migration_instance_init, | |
b91bf5e4 | 4534 | .instance_finalize = migration_instance_finalize, |
e5cb7e76 PX |
4535 | }; |
4536 | ||
4537 | static void register_migration_types(void) | |
4538 | { | |
4539 | type_register_static(&migration_type); | |
4540 | } | |
4541 | ||
4542 | type_init(register_migration_types); |