xref: /openbmc/qemu/migration/migration.h (revision 6f03770d)
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 /* State for the incoming migration */
49 struct MigrationIncomingState {
50     QEMUFile *from_src_file;
51 
52     /*
53      * Free at the start of the main state load, set as the main thread finishes
54      * loading state.
55      */
56     QemuEvent main_thread_load_event;
57 
58     /* For network announces */
59     AnnounceTimer  announce_timer;
60 
61     size_t         largest_page_size;
62     bool           have_fault_thread;
63     QemuThread     fault_thread;
64     QemuSemaphore  fault_thread_sem;
65     /* Set this when we want the fault thread to quit */
66     bool           fault_thread_quit;
67 
68     bool           have_listen_thread;
69     QemuThread     listen_thread;
70     QemuSemaphore  listen_thread_sem;
71 
72     /* For the kernel to send us notifications */
73     int       userfault_fd;
74     /* To notify the fault_thread to wake, e.g., when need to quit */
75     int       userfault_event_fd;
76     QEMUFile *to_src_file;
77     QemuMutex rp_mutex;    /* We send replies from multiple threads */
78     /* RAMBlock of last request sent to source */
79     RAMBlock *last_rb;
80     void     *postcopy_tmp_page;
81     void     *postcopy_tmp_zero_page;
82     /* PostCopyFD's for external userfaultfds & handlers of shared memory */
83     GArray   *postcopy_remote_fds;
84 
85     QEMUBH *bh;
86 
87     int state;
88 
89     bool have_colo_incoming_thread;
90     QemuThread colo_incoming_thread;
91     /* The coroutine we should enter (back) after failover */
92     Coroutine *migration_incoming_co;
93     QemuSemaphore colo_incoming_sem;
94 
95     /*
96      * PostcopyBlocktimeContext to keep information for postcopy
97      * live migration, to calculate vCPU block time
98      * */
99     struct PostcopyBlocktimeContext *blocktime_ctx;
100 
101     /* notify PAUSED postcopy incoming migrations to try to continue */
102     bool postcopy_recover_triggered;
103     QemuSemaphore postcopy_pause_sem_dst;
104     QemuSemaphore postcopy_pause_sem_fault;
105 
106     /* List of listening socket addresses  */
107     SocketAddressList *socket_address_list;
108 
109     /* A tree of pages that we requested to the source VM */
110     GTree *page_requested;
111     /* For debugging purpose only, but would be nice to keep */
112     int page_requested_count;
113     /*
114      * The mutex helps to maintain the requested pages that we sent to the
115      * source, IOW, to guarantee coherent between the page_requests tree and
116      * the per-ramblock receivedmap.  Note! This does not guarantee consistency
117      * of the real page copy procedures (using UFFDIO_[ZERO]COPY).  E.g., even
118      * if one bit in receivedmap is cleared, UFFDIO_COPY could have happened
119      * for that page already.  This is intended so that the mutex won't
120      * serialize and blocked by slow operations like UFFDIO_* ioctls.  However
121      * this should be enough to make sure the page_requested tree always
122      * contains valid information.
123      */
124     QemuMutex page_request_mutex;
125 };
126 
127 MigrationIncomingState *migration_incoming_get_current(void);
128 void migration_incoming_state_destroy(void);
129 /*
130  * Functions to work with blocktime context
131  */
132 void fill_destination_postcopy_migration_info(MigrationInfo *info);
133 
134 #define TYPE_MIGRATION "migration"
135 
136 typedef struct MigrationClass MigrationClass;
137 DECLARE_OBJ_CHECKERS(MigrationState, MigrationClass,
138                      MIGRATION_OBJ, TYPE_MIGRATION)
139 
140 struct MigrationClass {
141     /*< private >*/
142     DeviceClass parent_class;
143 };
144 
145 struct MigrationState {
146     /*< private >*/
147     DeviceState parent_obj;
148 
149     /*< public >*/
150     QemuThread thread;
151     QEMUBH *vm_start_bh;
152     QEMUBH *cleanup_bh;
153     QEMUFile *to_dst_file;
154     QIOChannelBuffer *bioc;
155     /*
156      * Protects to_dst_file pointer.  We need to make sure we won't
157      * yield or hang during the critical section, since this lock will
158      * be used in OOB command handler.
159      */
160     QemuMutex qemu_file_lock;
161 
162     /*
163      * Used to allow urgent requests to override rate limiting.
164      */
165     QemuSemaphore rate_limit_sem;
166 
167     /* pages already send at the beginning of current iteration */
168     uint64_t iteration_initial_pages;
169 
170     /* pages transferred per second */
171     double pages_per_second;
172 
173     /* bytes already send at the beginning of current iteration */
174     uint64_t iteration_initial_bytes;
175     /* time at the start of current iteration */
176     int64_t iteration_start_time;
177     /*
178      * The final stage happens when the remaining data is smaller than
179      * this threshold; it's calculated from the requested downtime and
180      * measured bandwidth
181      */
182     int64_t threshold_size;
183 
184     /* params from 'migrate-set-parameters' */
185     MigrationParameters parameters;
186 
187     int state;
188 
189     /* State related to return path */
190     struct {
191         QEMUFile     *from_dst_file;
192         QemuThread    rp_thread;
193         bool          error;
194         QemuSemaphore rp_sem;
195     } rp_state;
196 
197     double mbps;
198     /* Timestamp when recent migration starts (ms) */
199     int64_t start_time;
200     /* Total time used by latest migration (ms) */
201     int64_t total_time;
202     /* Timestamp when VM is down (ms) to migrate the last stuff */
203     int64_t downtime_start;
204     int64_t downtime;
205     int64_t expected_downtime;
206     bool enabled_capabilities[MIGRATION_CAPABILITY__MAX];
207     int64_t setup_time;
208     /*
209      * Whether guest was running when we enter the completion stage.
210      * If migration is interrupted by any reason, we need to continue
211      * running the guest on source.
212      */
213     bool vm_was_running;
214 
215     /* Flag set once the migration has been asked to enter postcopy */
216     bool start_postcopy;
217     /* Flag set after postcopy has sent the device state */
218     bool postcopy_after_devices;
219 
220     /* Flag set once the migration thread is running (and needs joining) */
221     bool migration_thread_running;
222 
223     /* Flag set once the migration thread called bdrv_inactivate_all */
224     bool block_inactive;
225 
226     /* Migration is waiting for guest to unplug device */
227     QemuSemaphore wait_unplug_sem;
228 
229     /* Migration is paused due to pause-before-switchover */
230     QemuSemaphore pause_sem;
231 
232     /* The semaphore is used to notify COLO thread that failover is finished */
233     QemuSemaphore colo_exit_sem;
234 
235     /* The event is used to notify COLO thread to do checkpoint */
236     QemuEvent colo_checkpoint_event;
237     int64_t colo_checkpoint_time;
238     QEMUTimer *colo_delay_timer;
239 
240     /* The first error that has occurred.
241        We used the mutex to be able to return the 1st error message */
242     Error *error;
243     /* mutex to protect errp */
244     QemuMutex error_mutex;
245 
246     /* Do we have to clean up -b/-i from old migrate parameters */
247     /* This feature is deprecated and will be removed */
248     bool must_remove_block_options;
249 
250     /*
251      * Global switch on whether we need to store the global state
252      * during migration.
253      */
254     bool store_global_state;
255 
256     /* Whether we send QEMU_VM_CONFIGURATION during migration */
257     bool send_configuration;
258     /* Whether we send section footer during migration */
259     bool send_section_footer;
260 
261     /* Needed by postcopy-pause state */
262     QemuSemaphore postcopy_pause_sem;
263     QemuSemaphore postcopy_pause_rp_sem;
264     /*
265      * Whether we abort the migration if decompression errors are
266      * detected at the destination. It is left at false for qemu
267      * older than 3.0, since only newer qemu sends streams that
268      * do not trigger spurious decompression errors.
269      */
270     bool decompress_error_check;
271 
272     /*
273      * This decides the size of guest memory chunk that will be used
274      * to track dirty bitmap clearing.  The size of memory chunk will
275      * be GUEST_PAGE_SIZE << N.  Say, N=0 means we will clear dirty
276      * bitmap for each page to send (1<<0=1); N=10 means we will clear
277      * dirty bitmap only once for 1<<10=1K continuous guest pages
278      * (which is in 4M chunk).
279      */
280     uint8_t clear_bitmap_shift;
281 
282     /*
283      * This save hostname when out-going migration starts
284      */
285     char *hostname;
286 };
287 
288 void migrate_set_state(int *state, int old_state, int new_state);
289 
290 void migration_fd_process_incoming(QEMUFile *f, Error **errp);
291 void migration_ioc_process_incoming(QIOChannel *ioc, Error **errp);
292 void migration_incoming_process(void);
293 
294 bool  migration_has_all_channels(void);
295 
296 uint64_t migrate_max_downtime(void);
297 
298 void migrate_set_error(MigrationState *s, const Error *error);
299 void migrate_fd_error(MigrationState *s, const Error *error);
300 
301 void migrate_fd_connect(MigrationState *s, Error *error_in);
302 
303 bool migration_is_setup_or_active(int state);
304 bool migration_is_running(int state);
305 
306 void migrate_init(MigrationState *s);
307 bool migration_is_blocked(Error **errp);
308 /* True if outgoing migration has entered postcopy phase */
309 bool migration_in_postcopy(void);
310 MigrationState *migrate_get_current(void);
311 
312 bool migrate_postcopy(void);
313 
314 bool migrate_release_ram(void);
315 bool migrate_postcopy_ram(void);
316 bool migrate_zero_blocks(void);
317 bool migrate_dirty_bitmaps(void);
318 bool migrate_ignore_shared(void);
319 bool migrate_validate_uuid(void);
320 
321 bool migrate_auto_converge(void);
322 bool migrate_use_multifd(void);
323 bool migrate_pause_before_switchover(void);
324 int migrate_multifd_channels(void);
325 MultiFDCompression migrate_multifd_compression(void);
326 int migrate_multifd_zlib_level(void);
327 int migrate_multifd_zstd_level(void);
328 
329 int migrate_use_xbzrle(void);
330 uint64_t migrate_xbzrle_cache_size(void);
331 bool migrate_colo_enabled(void);
332 
333 bool migrate_use_block(void);
334 bool migrate_use_block_incremental(void);
335 int migrate_max_cpu_throttle(void);
336 bool migrate_use_return_path(void);
337 
338 uint64_t ram_get_total_transferred_pages(void);
339 
340 bool migrate_use_compression(void);
341 int migrate_compress_level(void);
342 int migrate_compress_threads(void);
343 int migrate_compress_wait_thread(void);
344 int migrate_decompress_threads(void);
345 bool migrate_use_events(void);
346 bool migrate_postcopy_blocktime(void);
347 bool migrate_background_snapshot(void);
348 
349 /* Sending on the return path - generic and then for each message type */
350 void migrate_send_rp_shut(MigrationIncomingState *mis,
351                           uint32_t value);
352 void migrate_send_rp_pong(MigrationIncomingState *mis,
353                           uint32_t value);
354 int migrate_send_rp_req_pages(MigrationIncomingState *mis, RAMBlock *rb,
355                               ram_addr_t start, uint64_t haddr);
356 int migrate_send_rp_message_req_pages(MigrationIncomingState *mis,
357                                       RAMBlock *rb, ram_addr_t start);
358 void migrate_send_rp_recv_bitmap(MigrationIncomingState *mis,
359                                  char *block_name);
360 void migrate_send_rp_resume_ack(MigrationIncomingState *mis, uint32_t value);
361 
362 void dirty_bitmap_mig_before_vm_start(void);
363 void dirty_bitmap_mig_cancel_outgoing(void);
364 void dirty_bitmap_mig_cancel_incoming(void);
365 bool check_dirty_bitmap_mig_alias_map(const BitmapMigrationNodeAliasList *bbm,
366                                       Error **errp);
367 
368 void migrate_add_address(SocketAddress *address);
369 
370 int foreach_not_ignored_block(RAMBlockIterFunc func, void *opaque);
371 
372 #define qemu_ram_foreach_block \
373   #warning "Use foreach_not_ignored_block in migration code"
374 
375 void migration_make_urgent_request(void);
376 void migration_consume_urgent_request(void);
377 bool migration_rate_limit(void);
378 
379 #endif
380