xref: /openbmc/qemu/migration/migration.h (revision 9951ba94)
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  *
12  */
13 
14 #ifndef QEMU_MIGRATION_H
15 #define QEMU_MIGRATION_H
16 
17 #include "exec/cpu-common.h"
18 #include "hw/qdev-core.h"
19 #include "qapi/qapi-types-migration.h"
20 #include "qemu/thread.h"
21 #include "qemu/coroutine_int.h"
22 #include "io/channel.h"
23 #include "io/channel-buffer.h"
24 #include "net/announce.h"
25 #include "qom/object.h"
26 
27 struct PostcopyBlocktimeContext;
28 
29 #define  MIGRATION_RESUME_ACK_VALUE  (1)
30 
31 /*
32  * 1<<6=64 pages -> 256K chunk when page size is 4K.  This gives us
33  * the benefit that all the chunks are 64 pages aligned then the
34  * bitmaps are always aligned to LONG.
35  */
36 #define CLEAR_BITMAP_SHIFT_MIN             6
37 /*
38  * 1<<18=256K pages -> 1G chunk when page size is 4K.  This is the
39  * default value to use if no one specified.
40  */
41 #define CLEAR_BITMAP_SHIFT_DEFAULT        18
42 /*
43  * 1<<31=2G pages -> 8T chunk when page size is 4K.  This should be
44  * big enough and make sure we won't overflow easily.
45  */
46 #define CLEAR_BITMAP_SHIFT_MAX            31
47 
48 /* This is an abstraction of a "temp huge page" for postcopy's purpose */
49 typedef struct {
50     /*
51      * This points to a temporary huge page as a buffer for UFFDIO_COPY.  It's
52      * mmap()ed and needs to be freed when cleanup.
53      */
54     void *tmp_huge_page;
55     /*
56      * This points to the host page we're going to install for this temp page.
57      * It tells us after we've received the whole page, where we should put it.
58      */
59     void *host_addr;
60     /* Number of small pages copied (in size of TARGET_PAGE_SIZE) */
61     unsigned int target_pages;
62     /* Whether this page contains all zeros */
63     bool all_zero;
64 } PostcopyTmpPage;
65 
66 /* State for the incoming migration */
67 struct MigrationIncomingState {
68     QEMUFile *from_src_file;
69     /* Previously received RAM's RAMBlock pointer */
70     RAMBlock *last_recv_block;
71     /* A hook to allow cleanup at the end of incoming migration */
72     void *transport_data;
73     void (*transport_cleanup)(void *data);
74     /*
75      * Used to sync thread creations.  Note that we can't create threads in
76      * parallel with this sem.
77      */
78     QemuSemaphore  thread_sync_sem;
79     /*
80      * Free at the start of the main state load, set as the main thread finishes
81      * loading state.
82      */
83     QemuEvent main_thread_load_event;
84 
85     /* For network announces */
86     AnnounceTimer  announce_timer;
87 
88     size_t         largest_page_size;
89     bool           have_fault_thread;
90     QemuThread     fault_thread;
91     /* Set this when we want the fault thread to quit */
92     bool           fault_thread_quit;
93 
94     bool           have_listen_thread;
95     QemuThread     listen_thread;
96 
97     /* For the kernel to send us notifications */
98     int       userfault_fd;
99     /* To notify the fault_thread to wake, e.g., when need to quit */
100     int       userfault_event_fd;
101     QEMUFile *to_src_file;
102     QemuMutex rp_mutex;    /* We send replies from multiple threads */
103     /* RAMBlock of last request sent to source */
104     RAMBlock *last_rb;
105     /*
106      * Number of postcopy channels including the default precopy channel, so
107      * vanilla postcopy will only contain one channel which contain both
108      * precopy and postcopy streams.
109      *
110      * This is calculated when the src requests to enable postcopy but before
111      * it starts.  Its value can depend on e.g. whether postcopy preemption is
112      * enabled.
113      */
114     unsigned int postcopy_channels;
115     /*
116      * An array of temp host huge pages to be used, one for each postcopy
117      * channel.
118      */
119     PostcopyTmpPage *postcopy_tmp_pages;
120     /* This is shared for all postcopy channels */
121     void     *postcopy_tmp_zero_page;
122     /* PostCopyFD's for external userfaultfds & handlers of shared memory */
123     GArray   *postcopy_remote_fds;
124 
125     QEMUBH *bh;
126 
127     int state;
128 
129     bool have_colo_incoming_thread;
130     QemuThread colo_incoming_thread;
131     /* The coroutine we should enter (back) after failover */
132     Coroutine *migration_incoming_co;
133     QemuSemaphore colo_incoming_sem;
134 
135     /*
136      * PostcopyBlocktimeContext to keep information for postcopy
137      * live migration, to calculate vCPU block time
138      * */
139     struct PostcopyBlocktimeContext *blocktime_ctx;
140 
141     /* notify PAUSED postcopy incoming migrations to try to continue */
142     QemuSemaphore postcopy_pause_sem_dst;
143     QemuSemaphore postcopy_pause_sem_fault;
144 
145     /* List of listening socket addresses  */
146     SocketAddressList *socket_address_list;
147 
148     /* A tree of pages that we requested to the source VM */
149     GTree *page_requested;
150     /* For debugging purpose only, but would be nice to keep */
151     int page_requested_count;
152     /*
153      * The mutex helps to maintain the requested pages that we sent to the
154      * source, IOW, to guarantee coherent between the page_requests tree and
155      * the per-ramblock receivedmap.  Note! This does not guarantee consistency
156      * of the real page copy procedures (using UFFDIO_[ZERO]COPY).  E.g., even
157      * if one bit in receivedmap is cleared, UFFDIO_COPY could have happened
158      * for that page already.  This is intended so that the mutex won't
159      * serialize and blocked by slow operations like UFFDIO_* ioctls.  However
160      * this should be enough to make sure the page_requested tree always
161      * contains valid information.
162      */
163     QemuMutex page_request_mutex;
164 };
165 
166 MigrationIncomingState *migration_incoming_get_current(void);
167 void migration_incoming_state_destroy(void);
168 void migration_incoming_transport_cleanup(MigrationIncomingState *mis);
169 /*
170  * Functions to work with blocktime context
171  */
172 void fill_destination_postcopy_migration_info(MigrationInfo *info);
173 
174 #define TYPE_MIGRATION "migration"
175 
176 typedef struct MigrationClass MigrationClass;
177 DECLARE_OBJ_CHECKERS(MigrationState, MigrationClass,
178                      MIGRATION_OBJ, TYPE_MIGRATION)
179 
180 struct MigrationClass {
181     /*< private >*/
182     DeviceClass parent_class;
183 };
184 
185 struct MigrationState {
186     /*< private >*/
187     DeviceState parent_obj;
188 
189     /*< public >*/
190     QemuThread thread;
191     QEMUBH *vm_start_bh;
192     QEMUBH *cleanup_bh;
193     /* Protected by qemu_file_lock */
194     QEMUFile *to_dst_file;
195     QIOChannelBuffer *bioc;
196     /*
197      * Protects to_dst_file/from_dst_file pointers.  We need to make sure we
198      * won't yield or hang during the critical section, since this lock will be
199      * used in OOB command handler.
200      */
201     QemuMutex qemu_file_lock;
202 
203     /*
204      * Used to allow urgent requests to override rate limiting.
205      */
206     QemuSemaphore rate_limit_sem;
207 
208     /* pages already send at the beginning of current iteration */
209     uint64_t iteration_initial_pages;
210 
211     /* pages transferred per second */
212     double pages_per_second;
213 
214     /* bytes already send at the beginning of current iteration */
215     uint64_t iteration_initial_bytes;
216     /* time at the start of current iteration */
217     int64_t iteration_start_time;
218     /*
219      * The final stage happens when the remaining data is smaller than
220      * this threshold; it's calculated from the requested downtime and
221      * measured bandwidth
222      */
223     int64_t threshold_size;
224 
225     /* params from 'migrate-set-parameters' */
226     MigrationParameters parameters;
227 
228     int state;
229 
230     /* State related to return path */
231     struct {
232         /* Protected by qemu_file_lock */
233         QEMUFile     *from_dst_file;
234         QemuThread    rp_thread;
235         bool          error;
236         /*
237          * We can also check non-zero of rp_thread, but there's no "official"
238          * way to do this, so this bool makes it slightly more elegant.
239          * Checking from_dst_file for this is racy because from_dst_file will
240          * be cleared in the rp_thread!
241          */
242         bool          rp_thread_created;
243         QemuSemaphore rp_sem;
244     } rp_state;
245 
246     double mbps;
247     /* Timestamp when recent migration starts (ms) */
248     int64_t start_time;
249     /* Total time used by latest migration (ms) */
250     int64_t total_time;
251     /* Timestamp when VM is down (ms) to migrate the last stuff */
252     int64_t downtime_start;
253     int64_t downtime;
254     int64_t expected_downtime;
255     bool enabled_capabilities[MIGRATION_CAPABILITY__MAX];
256     int64_t setup_time;
257     /*
258      * Whether guest was running when we enter the completion stage.
259      * If migration is interrupted by any reason, we need to continue
260      * running the guest on source.
261      */
262     bool vm_was_running;
263 
264     /* Flag set once the migration has been asked to enter postcopy */
265     bool start_postcopy;
266     /* Flag set after postcopy has sent the device state */
267     bool postcopy_after_devices;
268 
269     /* Flag set once the migration thread is running (and needs joining) */
270     bool migration_thread_running;
271 
272     /* Flag set once the migration thread called bdrv_inactivate_all */
273     bool block_inactive;
274 
275     /* Migration is waiting for guest to unplug device */
276     QemuSemaphore wait_unplug_sem;
277 
278     /* Migration is paused due to pause-before-switchover */
279     QemuSemaphore pause_sem;
280 
281     /* The semaphore is used to notify COLO thread that failover is finished */
282     QemuSemaphore colo_exit_sem;
283 
284     /* The event is used to notify COLO thread to do checkpoint */
285     QemuEvent colo_checkpoint_event;
286     int64_t colo_checkpoint_time;
287     QEMUTimer *colo_delay_timer;
288 
289     /* The first error that has occurred.
290        We used the mutex to be able to return the 1st error message */
291     Error *error;
292     /* mutex to protect errp */
293     QemuMutex error_mutex;
294 
295     /* Do we have to clean up -b/-i from old migrate parameters */
296     /* This feature is deprecated and will be removed */
297     bool must_remove_block_options;
298 
299     /*
300      * Global switch on whether we need to store the global state
301      * during migration.
302      */
303     bool store_global_state;
304 
305     /* Whether we send QEMU_VM_CONFIGURATION during migration */
306     bool send_configuration;
307     /* Whether we send section footer during migration */
308     bool send_section_footer;
309 
310     /* Needed by postcopy-pause state */
311     QemuSemaphore postcopy_pause_sem;
312     QemuSemaphore postcopy_pause_rp_sem;
313     /*
314      * Whether we abort the migration if decompression errors are
315      * detected at the destination. It is left at false for qemu
316      * older than 3.0, since only newer qemu sends streams that
317      * do not trigger spurious decompression errors.
318      */
319     bool decompress_error_check;
320 
321     /*
322      * This decides the size of guest memory chunk that will be used
323      * to track dirty bitmap clearing.  The size of memory chunk will
324      * be GUEST_PAGE_SIZE << N.  Say, N=0 means we will clear dirty
325      * bitmap for each page to send (1<<0=1); N=10 means we will clear
326      * dirty bitmap only once for 1<<10=1K continuous guest pages
327      * (which is in 4M chunk).
328      */
329     uint8_t clear_bitmap_shift;
330 
331     /*
332      * This save hostname when out-going migration starts
333      */
334     char *hostname;
335 };
336 
337 void migrate_set_state(int *state, int old_state, int new_state);
338 
339 void migration_fd_process_incoming(QEMUFile *f, Error **errp);
340 void migration_ioc_process_incoming(QIOChannel *ioc, Error **errp);
341 void migration_incoming_process(void);
342 
343 bool  migration_has_all_channels(void);
344 
345 uint64_t migrate_max_downtime(void);
346 
347 void migrate_set_error(MigrationState *s, const Error *error);
348 void migrate_fd_error(MigrationState *s, const Error *error);
349 
350 void migrate_fd_connect(MigrationState *s, Error *error_in);
351 
352 bool migration_is_setup_or_active(int state);
353 bool migration_is_running(int state);
354 
355 void migrate_init(MigrationState *s);
356 bool migration_is_blocked(Error **errp);
357 /* True if outgoing migration has entered postcopy phase */
358 bool migration_in_postcopy(void);
359 MigrationState *migrate_get_current(void);
360 
361 bool migrate_postcopy(void);
362 
363 bool migrate_release_ram(void);
364 bool migrate_postcopy_ram(void);
365 bool migrate_zero_blocks(void);
366 bool migrate_dirty_bitmaps(void);
367 bool migrate_ignore_shared(void);
368 bool migrate_validate_uuid(void);
369 
370 bool migrate_auto_converge(void);
371 bool migrate_use_multifd(void);
372 bool migrate_pause_before_switchover(void);
373 int migrate_multifd_channels(void);
374 MultiFDCompression migrate_multifd_compression(void);
375 int migrate_multifd_zlib_level(void);
376 int migrate_multifd_zstd_level(void);
377 
378 int migrate_use_xbzrle(void);
379 uint64_t migrate_xbzrle_cache_size(void);
380 bool migrate_colo_enabled(void);
381 
382 bool migrate_use_block(void);
383 bool migrate_use_block_incremental(void);
384 int migrate_max_cpu_throttle(void);
385 bool migrate_use_return_path(void);
386 
387 uint64_t ram_get_total_transferred_pages(void);
388 
389 bool migrate_use_compression(void);
390 int migrate_compress_level(void);
391 int migrate_compress_threads(void);
392 int migrate_compress_wait_thread(void);
393 int migrate_decompress_threads(void);
394 bool migrate_use_events(void);
395 bool migrate_postcopy_blocktime(void);
396 bool migrate_background_snapshot(void);
397 
398 /* Sending on the return path - generic and then for each message type */
399 void migrate_send_rp_shut(MigrationIncomingState *mis,
400                           uint32_t value);
401 void migrate_send_rp_pong(MigrationIncomingState *mis,
402                           uint32_t value);
403 int migrate_send_rp_req_pages(MigrationIncomingState *mis, RAMBlock *rb,
404                               ram_addr_t start, uint64_t haddr);
405 int migrate_send_rp_message_req_pages(MigrationIncomingState *mis,
406                                       RAMBlock *rb, ram_addr_t start);
407 void migrate_send_rp_recv_bitmap(MigrationIncomingState *mis,
408                                  char *block_name);
409 void migrate_send_rp_resume_ack(MigrationIncomingState *mis, uint32_t value);
410 
411 void dirty_bitmap_mig_before_vm_start(void);
412 void dirty_bitmap_mig_cancel_outgoing(void);
413 void dirty_bitmap_mig_cancel_incoming(void);
414 bool check_dirty_bitmap_mig_alias_map(const BitmapMigrationNodeAliasList *bbm,
415                                       Error **errp);
416 
417 void migrate_add_address(SocketAddress *address);
418 
419 int foreach_not_ignored_block(RAMBlockIterFunc func, void *opaque);
420 
421 #define qemu_ram_foreach_block \
422   #warning "Use foreach_not_ignored_block in migration code"
423 
424 void migration_make_urgent_request(void);
425 void migration_consume_urgent_request(void);
426 bool migration_rate_limit(void);
427 void migration_cancel(const Error *error);
428 
429 void populate_vfio_info(MigrationInfo *info);
430 void postcopy_temp_page_reset(PostcopyTmpPage *tmp_page);
431 
432 bool migrate_multi_channels_is_allowed(void);
433 void migrate_protocol_allow_multi_channels(bool allow);
434 
435 #endif
436