xref: /openbmc/qemu/migration/migration.h (revision 136cb9cc)
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 "qapi/qmp/json-writer.h"
21 #include "qemu/thread.h"
22 #include "qemu/coroutine_int.h"
23 #include "io/channel.h"
24 #include "io/channel-buffer.h"
25 #include "net/announce.h"
26 #include "qom/object.h"
27 #include "postcopy-ram.h"
28 #include "sysemu/runstate.h"
29 
30 struct PostcopyBlocktimeContext;
31 
32 #define  MIGRATION_RESUME_ACK_VALUE  (1)
33 
34 /*
35  * 1<<6=64 pages -> 256K chunk when page size is 4K.  This gives us
36  * the benefit that all the chunks are 64 pages aligned then the
37  * bitmaps are always aligned to LONG.
38  */
39 #define CLEAR_BITMAP_SHIFT_MIN             6
40 /*
41  * 1<<18=256K pages -> 1G chunk when page size is 4K.  This is the
42  * default value to use if no one specified.
43  */
44 #define CLEAR_BITMAP_SHIFT_DEFAULT        18
45 /*
46  * 1<<31=2G pages -> 8T chunk when page size is 4K.  This should be
47  * big enough and make sure we won't overflow easily.
48  */
49 #define CLEAR_BITMAP_SHIFT_MAX            31
50 
51 /* This is an abstraction of a "temp huge page" for postcopy's purpose */
52 typedef struct {
53     /*
54      * This points to a temporary huge page as a buffer for UFFDIO_COPY.  It's
55      * mmap()ed and needs to be freed when cleanup.
56      */
57     void *tmp_huge_page;
58     /*
59      * This points to the host page we're going to install for this temp page.
60      * It tells us after we've received the whole page, where we should put it.
61      */
62     void *host_addr;
63     /* Number of small pages copied (in size of TARGET_PAGE_SIZE) */
64     unsigned int target_pages;
65     /* Whether this page contains all zeros */
66     bool all_zero;
67 } PostcopyTmpPage;
68 
69 typedef enum {
70     PREEMPT_THREAD_NONE = 0,
71     PREEMPT_THREAD_CREATED,
72     PREEMPT_THREAD_QUIT,
73 } PreemptThreadStatus;
74 
75 /* State for the incoming migration */
76 struct MigrationIncomingState {
77     QEMUFile *from_src_file;
78     /* Previously received RAM's RAMBlock pointer */
79     RAMBlock *last_recv_block[RAM_CHANNEL_MAX];
80     /* A hook to allow cleanup at the end of incoming migration */
81     void *transport_data;
82     void (*transport_cleanup)(void *data);
83     /*
84      * Used to sync thread creations.  Note that we can't create threads in
85      * parallel with this sem.
86      */
87     QemuSemaphore  thread_sync_sem;
88     /*
89      * Free at the start of the main state load, set as the main thread finishes
90      * loading state.
91      */
92     QemuEvent main_thread_load_event;
93 
94     /* For network announces */
95     AnnounceTimer  announce_timer;
96 
97     size_t         largest_page_size;
98     bool           have_fault_thread;
99     QemuThread     fault_thread;
100     /* Set this when we want the fault thread to quit */
101     bool           fault_thread_quit;
102 
103     bool           have_listen_thread;
104     QemuThread     listen_thread;
105 
106     /* For the kernel to send us notifications */
107     int       userfault_fd;
108     /* To notify the fault_thread to wake, e.g., when need to quit */
109     int       userfault_event_fd;
110     QEMUFile *to_src_file;
111     QemuMutex rp_mutex;    /* We send replies from multiple threads */
112     /* RAMBlock of last request sent to source */
113     RAMBlock *last_rb;
114     /*
115      * Number of postcopy channels including the default precopy channel, so
116      * vanilla postcopy will only contain one channel which contain both
117      * precopy and postcopy streams.
118      *
119      * This is calculated when the src requests to enable postcopy but before
120      * it starts.  Its value can depend on e.g. whether postcopy preemption is
121      * enabled.
122      */
123     unsigned int postcopy_channels;
124     /* QEMUFile for postcopy only; it'll be handled by a separate thread */
125     QEMUFile *postcopy_qemufile_dst;
126     /*
127      * When postcopy_qemufile_dst is properly setup, this sem is posted.
128      * One can wait on this semaphore to wait until the preempt channel is
129      * properly setup.
130      */
131     QemuSemaphore postcopy_qemufile_dst_done;
132     /* Postcopy priority thread is used to receive postcopy requested pages */
133     QemuThread postcopy_prio_thread;
134     /*
135      * Always set by the main vm load thread only, but can be read by the
136      * postcopy preempt thread.  "volatile" makes sure all reads will be
137      * up-to-date across cores.
138      */
139     volatile PreemptThreadStatus preempt_thread_status;
140     /*
141      * Used to sync between the ram load main thread and the fast ram load
142      * thread.  It protects postcopy_qemufile_dst, which is the postcopy
143      * fast channel.
144      *
145      * The ram fast load thread will take it mostly for the whole lifecycle
146      * because it needs to continuously read data from the channel, and
147      * it'll only release this mutex if postcopy is interrupted, so that
148      * the ram load main thread will take this mutex over and properly
149      * release the broken channel.
150      */
151     QemuMutex postcopy_prio_thread_mutex;
152     /*
153      * An array of temp host huge pages to be used, one for each postcopy
154      * channel.
155      */
156     PostcopyTmpPage *postcopy_tmp_pages;
157     /* This is shared for all postcopy channels */
158     void     *postcopy_tmp_zero_page;
159     /* PostCopyFD's for external userfaultfds & handlers of shared memory */
160     GArray   *postcopy_remote_fds;
161 
162     QEMUBH *bh;
163 
164     int state;
165 
166     /*
167      * The incoming migration coroutine, non-NULL during qemu_loadvm_state().
168      * Used to wake the migration incoming coroutine from rdma code. How much is
169      * it safe - it's a question.
170      */
171     Coroutine *loadvm_co;
172 
173     /* The coroutine we should enter (back) after failover */
174     Coroutine *colo_incoming_co;
175     QemuSemaphore colo_incoming_sem;
176 
177     /*
178      * PostcopyBlocktimeContext to keep information for postcopy
179      * live migration, to calculate vCPU block time
180      * */
181     struct PostcopyBlocktimeContext *blocktime_ctx;
182 
183     /* notify PAUSED postcopy incoming migrations to try to continue */
184     QemuSemaphore postcopy_pause_sem_dst;
185     QemuSemaphore postcopy_pause_sem_fault;
186     /*
187      * This semaphore is used to allow the ram fast load thread (only when
188      * postcopy preempt is enabled) fall into sleep when there's network
189      * interruption detected.  When the recovery is done, the main load
190      * thread will kick the fast ram load thread using this semaphore.
191      */
192     QemuSemaphore postcopy_pause_sem_fast_load;
193 
194     /* List of listening socket addresses  */
195     SocketAddressList *socket_address_list;
196 
197     /* A tree of pages that we requested to the source VM */
198     GTree *page_requested;
199     /* For debugging purpose only, but would be nice to keep */
200     int page_requested_count;
201     /*
202      * The mutex helps to maintain the requested pages that we sent to the
203      * source, IOW, to guarantee coherent between the page_requests tree and
204      * the per-ramblock receivedmap.  Note! This does not guarantee consistency
205      * of the real page copy procedures (using UFFDIO_[ZERO]COPY).  E.g., even
206      * if one bit in receivedmap is cleared, UFFDIO_COPY could have happened
207      * for that page already.  This is intended so that the mutex won't
208      * serialize and blocked by slow operations like UFFDIO_* ioctls.  However
209      * this should be enough to make sure the page_requested tree always
210      * contains valid information.
211      */
212     QemuMutex page_request_mutex;
213 
214     /*
215      * Number of devices that have yet to approve switchover. When this reaches
216      * zero an ACK that it's OK to do switchover is sent to the source. No lock
217      * is needed as this field is updated serially.
218      */
219     unsigned int switchover_ack_pending_num;
220 };
221 
222 MigrationIncomingState *migration_incoming_get_current(void);
223 void migration_incoming_state_destroy(void);
224 void migration_incoming_transport_cleanup(MigrationIncomingState *mis);
225 /*
226  * Functions to work with blocktime context
227  */
228 void fill_destination_postcopy_migration_info(MigrationInfo *info);
229 
230 #define TYPE_MIGRATION "migration"
231 
232 typedef struct MigrationClass MigrationClass;
233 DECLARE_OBJ_CHECKERS(MigrationState, MigrationClass,
234                      MIGRATION_OBJ, TYPE_MIGRATION)
235 
236 struct MigrationClass {
237     /*< private >*/
238     DeviceClass parent_class;
239 };
240 
241 struct MigrationState {
242     /*< private >*/
243     DeviceState parent_obj;
244 
245     /*< public >*/
246     QemuThread thread;
247     QEMUBH *vm_start_bh;
248     QEMUBH *cleanup_bh;
249     /* Protected by qemu_file_lock */
250     QEMUFile *to_dst_file;
251     /* Postcopy specific transfer channel */
252     QEMUFile *postcopy_qemufile_src;
253     /*
254      * It is posted when the preempt channel is established.  Note: this is
255      * used for both the start or recover of a postcopy migration.  We'll
256      * post to this sem every time a new preempt channel is created in the
257      * main thread, and we keep post() and wait() in pair.
258      */
259     QemuSemaphore postcopy_qemufile_src_sem;
260     QIOChannelBuffer *bioc;
261     /*
262      * Protects to_dst_file/from_dst_file pointers.  We need to make sure we
263      * won't yield or hang during the critical section, since this lock will be
264      * used in OOB command handler.
265      */
266     QemuMutex qemu_file_lock;
267 
268     /*
269      * Used to allow urgent requests to override rate limiting.
270      */
271     QemuSemaphore rate_limit_sem;
272 
273     /* pages already send at the beginning of current iteration */
274     uint64_t iteration_initial_pages;
275 
276     /* pages transferred per second */
277     double pages_per_second;
278 
279     /* bytes already send at the beginning of current iteration */
280     uint64_t iteration_initial_bytes;
281     /* time at the start of current iteration */
282     int64_t iteration_start_time;
283     /*
284      * The final stage happens when the remaining data is smaller than
285      * this threshold; it's calculated from the requested downtime and
286      * measured bandwidth
287      */
288     int64_t threshold_size;
289 
290     /* params from 'migrate-set-parameters' */
291     MigrationParameters parameters;
292 
293     int state;
294 
295     /* State related to return path */
296     struct {
297         /* Protected by qemu_file_lock */
298         QEMUFile     *from_dst_file;
299         QemuThread    rp_thread;
300         bool          error;
301         /*
302          * We can also check non-zero of rp_thread, but there's no "official"
303          * way to do this, so this bool makes it slightly more elegant.
304          * Checking from_dst_file for this is racy because from_dst_file will
305          * be cleared in the rp_thread!
306          */
307         bool          rp_thread_created;
308         QemuSemaphore rp_sem;
309         /*
310          * We post to this when we got one PONG from dest. So far it's an
311          * easy way to know the main channel has successfully established
312          * on dest QEMU.
313          */
314         QemuSemaphore rp_pong_acks;
315     } rp_state;
316 
317     double mbps;
318     /* Timestamp when recent migration starts (ms) */
319     int64_t start_time;
320     /* Total time used by latest migration (ms) */
321     int64_t total_time;
322     /* Timestamp when VM is down (ms) to migrate the last stuff */
323     int64_t downtime_start;
324     int64_t downtime;
325     int64_t expected_downtime;
326     bool capabilities[MIGRATION_CAPABILITY__MAX];
327     int64_t setup_time;
328 
329     /*
330      * State before stopping the vm by vm_stop_force_state().
331      * If migration is interrupted by any reason, we need to continue
332      * running the guest on source if it was running or restore its stopped
333      * state.
334      */
335     RunState vm_old_state;
336 
337     /* Flag set once the migration has been asked to enter postcopy */
338     bool start_postcopy;
339     /* Flag set after postcopy has sent the device state */
340     bool postcopy_after_devices;
341 
342     /* Flag set once the migration thread is running (and needs joining) */
343     bool migration_thread_running;
344 
345     /* Flag set once the migration thread called bdrv_inactivate_all */
346     bool block_inactive;
347 
348     /* Migration is waiting for guest to unplug device */
349     QemuSemaphore wait_unplug_sem;
350 
351     /* Migration is paused due to pause-before-switchover */
352     QemuSemaphore pause_sem;
353 
354     /* The semaphore is used to notify COLO thread that failover is finished */
355     QemuSemaphore colo_exit_sem;
356 
357     /* The event is used to notify COLO thread to do checkpoint */
358     QemuEvent colo_checkpoint_event;
359     int64_t colo_checkpoint_time;
360     QEMUTimer *colo_delay_timer;
361 
362     /* The first error that has occurred.
363        We used the mutex to be able to return the 1st error message */
364     Error *error;
365     /* mutex to protect errp */
366     QemuMutex error_mutex;
367 
368     /* Do we have to clean up -b/-i from old migrate parameters */
369     /* This feature is deprecated and will be removed */
370     bool must_remove_block_options;
371 
372     /*
373      * Global switch on whether we need to store the global state
374      * during migration.
375      */
376     bool store_global_state;
377 
378     /* Whether we send QEMU_VM_CONFIGURATION during migration */
379     bool send_configuration;
380     /* Whether we send section footer during migration */
381     bool send_section_footer;
382 
383     /* Needed by postcopy-pause state */
384     QemuSemaphore postcopy_pause_sem;
385     QemuSemaphore postcopy_pause_rp_sem;
386     /*
387      * Whether we abort the migration if decompression errors are
388      * detected at the destination. It is left at false for qemu
389      * older than 3.0, since only newer qemu sends streams that
390      * do not trigger spurious decompression errors.
391      */
392     bool decompress_error_check;
393     /*
394      * This variable only affects behavior when postcopy preempt mode is
395      * enabled.
396      *
397      * When set:
398      *
399      * - postcopy preempt src QEMU instance will generate an EOS message at
400      *   the end of migration to shut the preempt channel on dest side.
401      *
402      * - postcopy preempt channel will be created at the setup phase on src
403          QEMU.
404      *
405      * When clear:
406      *
407      * - postcopy preempt src QEMU instance will _not_ generate an EOS
408      *   message at the end of migration, the dest qemu will shutdown the
409      *   channel itself.
410      *
411      * - postcopy preempt channel will be created at the switching phase
412      *   from precopy -> postcopy (to avoid race condition of misordered
413      *   creation of channels).
414      *
415      * NOTE: See message-id <ZBoShWArKDPpX/D7@work-vm> on qemu-devel
416      * mailing list for more information on the possible race.  Everyone
417      * should probably just keep this value untouched after set by the
418      * machine type (or the default).
419      */
420     bool preempt_pre_7_2;
421 
422     /*
423      * flush every channel after each section sent.
424      *
425      * This assures that we can't mix pages from one iteration through
426      * ram pages with pages for the following iteration.  We really
427      * only need to do this flush after we have go through all the
428      * dirty pages.  For historical reasons, we do that after each
429      * section.  This is suboptimal (we flush too many times).
430      * Default value is false. (since 8.1)
431      */
432     bool multifd_flush_after_each_section;
433     /*
434      * This decides the size of guest memory chunk that will be used
435      * to track dirty bitmap clearing.  The size of memory chunk will
436      * be GUEST_PAGE_SIZE << N.  Say, N=0 means we will clear dirty
437      * bitmap for each page to send (1<<0=1); N=10 means we will clear
438      * dirty bitmap only once for 1<<10=1K continuous guest pages
439      * (which is in 4M chunk).
440      */
441     uint8_t clear_bitmap_shift;
442 
443     /*
444      * This save hostname when out-going migration starts
445      */
446     char *hostname;
447 
448     /* QEMU_VM_VMDESCRIPTION content filled for all non-iterable devices. */
449     JSONWriter *vmdesc;
450 
451     /*
452      * Indicates whether an ACK from the destination that it's OK to do
453      * switchover has been received.
454      */
455     bool switchover_acked;
456 };
457 
458 void migrate_set_state(int *state, int old_state, int new_state);
459 
460 void migration_fd_process_incoming(QEMUFile *f, Error **errp);
461 void migration_ioc_process_incoming(QIOChannel *ioc, Error **errp);
462 void migration_incoming_process(void);
463 
464 bool  migration_has_all_channels(void);
465 
466 uint64_t migrate_max_downtime(void);
467 
468 void migrate_set_error(MigrationState *s, const Error *error);
469 
470 void migrate_fd_connect(MigrationState *s, Error *error_in);
471 
472 bool migration_is_setup_or_active(int state);
473 bool migration_is_running(int state);
474 
475 void migrate_init(MigrationState *s);
476 bool migration_is_blocked(Error **errp);
477 /* True if outgoing migration has entered postcopy phase */
478 bool migration_in_postcopy(void);
479 MigrationState *migrate_get_current(void);
480 
481 uint64_t ram_get_total_transferred_pages(void);
482 
483 /* Sending on the return path - generic and then for each message type */
484 void migrate_send_rp_shut(MigrationIncomingState *mis,
485                           uint32_t value);
486 void migrate_send_rp_pong(MigrationIncomingState *mis,
487                           uint32_t value);
488 int migrate_send_rp_req_pages(MigrationIncomingState *mis, RAMBlock *rb,
489                               ram_addr_t start, uint64_t haddr);
490 int migrate_send_rp_message_req_pages(MigrationIncomingState *mis,
491                                       RAMBlock *rb, ram_addr_t start);
492 void migrate_send_rp_recv_bitmap(MigrationIncomingState *mis,
493                                  char *block_name);
494 void migrate_send_rp_resume_ack(MigrationIncomingState *mis, uint32_t value);
495 int migrate_send_rp_switchover_ack(MigrationIncomingState *mis);
496 
497 void dirty_bitmap_mig_before_vm_start(void);
498 void dirty_bitmap_mig_cancel_outgoing(void);
499 void dirty_bitmap_mig_cancel_incoming(void);
500 bool check_dirty_bitmap_mig_alias_map(const BitmapMigrationNodeAliasList *bbm,
501                                       Error **errp);
502 
503 void migrate_add_address(SocketAddress *address);
504 
505 int foreach_not_ignored_block(RAMBlockIterFunc func, void *opaque);
506 
507 #define qemu_ram_foreach_block \
508   #warning "Use foreach_not_ignored_block in migration code"
509 
510 void migration_make_urgent_request(void);
511 void migration_consume_urgent_request(void);
512 bool migration_rate_limit(void);
513 void migration_cancel(const Error *error);
514 
515 void populate_vfio_info(MigrationInfo *info);
516 void reset_vfio_bytes_transferred(void);
517 void postcopy_temp_page_reset(PostcopyTmpPage *tmp_page);
518 
519 #endif
520