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 "net/announce.h"
24
25struct PostcopyBlocktimeContext;
26
27#define MIGRATION_RESUME_ACK_VALUE (1)
28
29/*
30 * 1<<6=64 pages -> 256K chunk when page size is 4K. This gives us
31 * the benefit that all the chunks are 64 pages aligned then the
32 * bitmaps are always aligned to LONG.
33 */
34#define CLEAR_BITMAP_SHIFT_MIN 6
35/*
36 * 1<<18=256K pages -> 1G chunk when page size is 4K. This is the
37 * default value to use if no one specified.
38 */
39#define CLEAR_BITMAP_SHIFT_DEFAULT 18
40/*
41 * 1<<31=2G pages -> 8T chunk when page size is 4K. This should be
42 * big enough and make sure we won't overflow easily.
43 */
44#define CLEAR_BITMAP_SHIFT_MAX 31
45
46/* State for the incoming migration */
47struct MigrationIncomingState {
48 QEMUFile *from_src_file;
49
50 /*
51 * Free at the start of the main state load, set as the main thread finishes
52 * loading state.
53 */
54 QemuEvent main_thread_load_event;
55
56 /* For network announces */
57 AnnounceTimer announce_timer;
58
59 size_t largest_page_size;
60 bool have_fault_thread;
61 QemuThread fault_thread;
62 QemuSemaphore fault_thread_sem;
63 /* Set this when we want the fault thread to quit */
64 bool fault_thread_quit;
65
66 bool have_listen_thread;
67 QemuThread listen_thread;
68 QemuSemaphore listen_thread_sem;
69
70 /* For the kernel to send us notifications */
71 int userfault_fd;
72 /* To notify the fault_thread to wake, e.g., when need to quit */
73 int userfault_event_fd;
74 QEMUFile *to_src_file;
75 QemuMutex rp_mutex; /* We send replies from multiple threads */
76 /* RAMBlock of last request sent to source */
77 RAMBlock *last_rb;
78 void *postcopy_tmp_page;
79 void *postcopy_tmp_zero_page;
80 /* PostCopyFD's for external userfaultfds & handlers of shared memory */
81 GArray *postcopy_remote_fds;
82
83 QEMUBH *bh;
84
85 int state;
86
87 bool have_colo_incoming_thread;
88 QemuThread colo_incoming_thread;
89 /* The coroutine we should enter (back) after failover */
90 Coroutine *migration_incoming_co;
91 QemuSemaphore colo_incoming_sem;
92
93 /*
94 * PostcopyBlocktimeContext to keep information for postcopy
95 * live migration, to calculate vCPU block time
96 * */
97 struct PostcopyBlocktimeContext *blocktime_ctx;
98
99 /* notify PAUSED postcopy incoming migrations to try to continue */
100 bool postcopy_recover_triggered;
101 QemuSemaphore postcopy_pause_sem_dst;
102 QemuSemaphore postcopy_pause_sem_fault;
103
104 /* List of listening socket addresses */
105 SocketAddressList *socket_address_list;
106};
107
108MigrationIncomingState *migration_incoming_get_current(void);
109void migration_incoming_state_destroy(void);
110/*
111 * Functions to work with blocktime context
112 */
113void fill_destination_postcopy_migration_info(MigrationInfo *info);
114
115#define TYPE_MIGRATION "migration"
116
117#define MIGRATION_CLASS(klass) \
118 OBJECT_CLASS_CHECK(MigrationClass, (klass), TYPE_MIGRATION)
119#define MIGRATION_OBJ(obj) \
120 OBJECT_CHECK(MigrationState, (obj), TYPE_MIGRATION)
121#define MIGRATION_GET_CLASS(obj) \
122 OBJECT_GET_CLASS(MigrationClass, (obj), TYPE_MIGRATION)
123
124typedef struct MigrationClass {
125 /*< private >*/
126 DeviceClass parent_class;
127} MigrationClass;
128
129struct MigrationState
130{
131 /*< private >*/
132 DeviceState parent_obj;
133
134 /*< public >*/
135 QemuThread thread;
136 QEMUBH *cleanup_bh;
137 QEMUFile *to_dst_file;
138 /*
139 * Protects to_dst_file pointer. We need to make sure we won't
140 * yield or hang during the critical section, since this lock will
141 * be used in OOB command handler.
142 */
143 QemuMutex qemu_file_lock;
144
145 /*
146 * Used to allow urgent requests to override rate limiting.
147 */
148 QemuSemaphore rate_limit_sem;
149
150 /* pages already send at the beginning of current iteration */
151 uint64_t iteration_initial_pages;
152
153 /* pages transferred per second */
154 double pages_per_second;
155
156 /* bytes already send at the beginning of current iteration */
157 uint64_t iteration_initial_bytes;
158 /* time at the start of current iteration */
159 int64_t iteration_start_time;
160 /*
161 * The final stage happens when the remaining data is smaller than
162 * this threshold; it's calculated from the requested downtime and
163 * measured bandwidth
164 */
165 int64_t threshold_size;
166
167 /* params from 'migrate-set-parameters' */
168 MigrationParameters parameters;
169
170 int state;
171
172 /* State related to return path */
173 struct {
174 QEMUFile *from_dst_file;
175 QemuThread rp_thread;
176 bool error;
177 QemuSemaphore rp_sem;
178 } rp_state;
179
180 double mbps;
181 /* Timestamp when recent migration starts (ms) */
182 int64_t start_time;
183 /* Total time used by latest migration (ms) */
184 int64_t total_time;
185 /* Timestamp when VM is down (ms) to migrate the last stuff */
186 int64_t downtime_start;
187 int64_t downtime;
188 int64_t expected_downtime;
189 bool enabled_capabilities[MIGRATION_CAPABILITY__MAX];
190 int64_t setup_time;
191 /*
192 * Whether guest was running when we enter the completion stage.
193 * If migration is interrupted by any reason, we need to continue
194 * running the guest on source.
195 */
196 bool vm_was_running;
197
198 /* Flag set once the migration has been asked to enter postcopy */
199 bool start_postcopy;
200 /* Flag set after postcopy has sent the device state */
201 bool postcopy_after_devices;
202
203 /* Flag set once the migration thread is running (and needs joining) */
204 bool migration_thread_running;
205
206 /* Flag set once the migration thread called bdrv_inactivate_all */
207 bool block_inactive;
208
209 /* Migration is paused due to pause-before-switchover */
210 QemuSemaphore pause_sem;
211
212 /* The semaphore is used to notify COLO thread that failover is finished */
213 QemuSemaphore colo_exit_sem;
214
215 /* The semaphore is used to notify COLO thread to do checkpoint */
216 QemuSemaphore colo_checkpoint_sem;
217 int64_t colo_checkpoint_time;
218 QEMUTimer *colo_delay_timer;
219
220 /* The first error that has occurred.
221 We used the mutex to be able to return the 1st error message */
222 Error *error;
223 /* mutex to protect errp */
224 QemuMutex error_mutex;
225
226 /* Do we have to clean up -b/-i from old migrate parameters */
227 /* This feature is deprecated and will be removed */
228 bool must_remove_block_options;
229
230 /*
231 * Global switch on whether we need to store the global state
232 * during migration.
233 */
234 bool store_global_state;
235
236 /* Whether we send QEMU_VM_CONFIGURATION during migration */
237 bool send_configuration;
238 /* Whether we send section footer during migration */
239 bool send_section_footer;
240
241 /* Needed by postcopy-pause state */
242 QemuSemaphore postcopy_pause_sem;
243 QemuSemaphore postcopy_pause_rp_sem;
244 /*
245 * Whether we abort the migration if decompression errors are
246 * detected at the destination. It is left at false for qemu
247 * older than 3.0, since only newer qemu sends streams that
248 * do not trigger spurious decompression errors.
249 */
250 bool decompress_error_check;
251
252 /*
253 * This decides the size of guest memory chunk that will be used
254 * to track dirty bitmap clearing. The size of memory chunk will
255 * be GUEST_PAGE_SIZE << N. Say, N=0 means we will clear dirty
256 * bitmap for each page to send (1<<0=1); N=10 means we will clear
257 * dirty bitmap only once for 1<<10=1K continuous guest pages
258 * (which is in 4M chunk).
259 */
260 uint8_t clear_bitmap_shift;
261};
262
263void migrate_set_state(int *state, int old_state, int new_state);
264
265void migration_fd_process_incoming(QEMUFile *f);
266void migration_ioc_process_incoming(QIOChannel *ioc, Error **errp);
267void migration_incoming_process(void);
268
269bool migration_has_all_channels(void);
270
271uint64_t migrate_max_downtime(void);
272
273void migrate_set_error(MigrationState *s, const Error *error);
274void migrate_fd_error(MigrationState *s, const Error *error);
275
276void migrate_fd_connect(MigrationState *s, Error *error_in);
277
278bool migration_is_setup_or_active(int state);
279
280void migrate_init(MigrationState *s);
281bool migration_is_blocked(Error **errp);
282/* True if outgoing migration has entered postcopy phase */
283bool migration_in_postcopy(void);
284MigrationState *migrate_get_current(void);
285
286bool migrate_postcopy(void);
287
288bool migrate_release_ram(void);
289bool migrate_postcopy_ram(void);
290bool migrate_zero_blocks(void);
291bool migrate_dirty_bitmaps(void);
292bool migrate_ignore_shared(void);
293
294bool migrate_auto_converge(void);
295bool migrate_use_multifd(void);
296bool migrate_pause_before_switchover(void);
297int migrate_multifd_channels(void);
298
299int migrate_use_xbzrle(void);
300int64_t migrate_xbzrle_cache_size(void);
301bool migrate_colo_enabled(void);
302
303bool migrate_use_block(void);
304bool migrate_use_block_incremental(void);
305int migrate_max_cpu_throttle(void);
306bool migrate_use_return_path(void);
307
308uint64_t ram_get_total_transferred_pages(void);
309
310bool migrate_use_compression(void);
311int migrate_compress_level(void);
312int migrate_compress_threads(void);
313int migrate_compress_wait_thread(void);
314int migrate_decompress_threads(void);
315bool migrate_use_events(void);
316bool migrate_postcopy_blocktime(void);
317
318/* Sending on the return path - generic and then for each message type */
319void migrate_send_rp_shut(MigrationIncomingState *mis,
320 uint32_t value);
321void migrate_send_rp_pong(MigrationIncomingState *mis,
322 uint32_t value);
323int migrate_send_rp_req_pages(MigrationIncomingState *mis, const char* rbname,
324 ram_addr_t start, size_t len);
325void migrate_send_rp_recv_bitmap(MigrationIncomingState *mis,
326 char *block_name);
327void migrate_send_rp_resume_ack(MigrationIncomingState *mis, uint32_t value);
328
329void dirty_bitmap_mig_before_vm_start(void);
330void init_dirty_bitmap_incoming_migration(void);
331void migrate_add_address(SocketAddress *address);
332
333int foreach_not_ignored_block(RAMBlockIterFunc func, void *opaque);
334
335#define qemu_ram_foreach_block \
336 #warning "Use foreach_not_ignored_block in migration code"
337
338void migration_make_urgent_request(void);
339void migration_consume_urgent_request(void);
340
341#endif
342