xref: /qemu/migration/migration.h (revision 96215036f47403438c7c7869b7cd419bd7a11f82)
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 "qobject/json-writer.h"
21 #include "qemu/thread.h"
22 #include "qemu/coroutine.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 "system/runstate.h"
29 #include "migration/misc.h"
30 
31 #define  MIGRATION_THREAD_SNAPSHOT          "mig/snapshot"
32 #define  MIGRATION_THREAD_DIRTY_RATE        "mig/dirtyrate"
33 
34 #define  MIGRATION_THREAD_SRC_MAIN          "mig/src/main"
35 #define  MIGRATION_THREAD_SRC_MULTIFD       "mig/src/send_%d"
36 #define  MIGRATION_THREAD_SRC_RETURN        "mig/src/return"
37 #define  MIGRATION_THREAD_SRC_TLS           "mig/src/tls"
38 
39 #define  MIGRATION_THREAD_DST_COLO          "mig/dst/colo"
40 #define  MIGRATION_THREAD_DST_MULTIFD       "mig/dst/recv_%d"
41 #define  MIGRATION_THREAD_DST_FAULT         "mig/dst/fault"
42 #define  MIGRATION_THREAD_DST_LISTEN        "mig/dst/listen"
43 #define  MIGRATION_THREAD_DST_PREEMPT       "mig/dst/preempt"
44 
45 struct PostcopyBlocktimeContext;
46 typedef struct ThreadPool ThreadPool;
47 
48 #define  MIGRATION_RESUME_ACK_VALUE  (1)
49 
50 /*
51  * 1<<6=64 pages -> 256K chunk when page size is 4K.  This gives us
52  * the benefit that all the chunks are 64 pages aligned then the
53  * bitmaps are always aligned to LONG.
54  */
55 #define CLEAR_BITMAP_SHIFT_MIN             6
56 /*
57  * 1<<18=256K pages -> 1G chunk when page size is 4K.  This is the
58  * default value to use if no one specified.
59  */
60 #define CLEAR_BITMAP_SHIFT_DEFAULT        18
61 /*
62  * 1<<31=2G pages -> 8T chunk when page size is 4K.  This should be
63  * big enough and make sure we won't overflow easily.
64  */
65 #define CLEAR_BITMAP_SHIFT_MAX            31
66 
67 /* This is an abstraction of a "temp huge page" for postcopy's purpose */
68 typedef struct {
69     /*
70      * This points to a temporary huge page as a buffer for UFFDIO_COPY.  It's
71      * mmap()ed and needs to be freed when cleanup.
72      */
73     void *tmp_huge_page;
74     /*
75      * This points to the host page we're going to install for this temp page.
76      * It tells us after we've received the whole page, where we should put it.
77      */
78     void *host_addr;
79     /* Number of small pages copied (in size of TARGET_PAGE_SIZE) */
80     unsigned int target_pages;
81     /* Whether this page contains all zeros */
82     bool all_zero;
83 } PostcopyTmpPage;
84 
85 typedef enum {
86     PREEMPT_THREAD_NONE = 0,
87     PREEMPT_THREAD_CREATED,
88     PREEMPT_THREAD_QUIT,
89 } PreemptThreadStatus;
90 
91 /* State for the incoming migration */
92 struct MigrationIncomingState {
93     QEMUFile *from_src_file;
94     /* Previously received RAM's RAMBlock pointer */
95     RAMBlock *last_recv_block[RAM_CHANNEL_MAX];
96     /* A hook to allow cleanup at the end of incoming migration */
97     void *transport_data;
98     void (*transport_cleanup)(void *data);
99     /*
100      * Used to sync thread creations.  Note that we can't create threads in
101      * parallel with this event.
102      */
103     QemuEvent  thread_sync_event;
104     /*
105      * Free at the start of the main state load, set as the main thread finishes
106      * loading state.
107      */
108     QemuEvent main_thread_load_event;
109 
110     /* For network announces */
111     AnnounceTimer  announce_timer;
112 
113     size_t         largest_page_size;
114     bool           have_fault_thread;
115     QemuThread     fault_thread;
116     /* Set this when we want the fault thread to quit */
117     bool           fault_thread_quit;
118 
119     bool           have_listen_thread;
120     QemuThread     listen_thread;
121 
122     /* For the kernel to send us notifications */
123     int       userfault_fd;
124     /* To notify the fault_thread to wake, e.g., when need to quit */
125     int       userfault_event_fd;
126     QEMUFile *to_src_file;
127     QemuMutex rp_mutex;    /* We send replies from multiple threads */
128     /* RAMBlock of last request sent to source */
129     RAMBlock *last_rb;
130     /*
131      * Number of postcopy channels including the default precopy channel, so
132      * vanilla postcopy will only contain one channel which contain both
133      * precopy and postcopy streams.
134      *
135      * This is calculated when the src requests to enable postcopy but before
136      * it starts.  Its value can depend on e.g. whether postcopy preemption is
137      * enabled.
138      */
139     unsigned int postcopy_channels;
140     /* QEMUFile for postcopy only; it'll be handled by a separate thread */
141     QEMUFile *postcopy_qemufile_dst;
142     /*
143      * When postcopy_qemufile_dst is properly setup, this sem is posted.
144      * One can wait on this semaphore to wait until the preempt channel is
145      * properly setup.
146      */
147     QemuSemaphore postcopy_qemufile_dst_done;
148     /* Postcopy priority thread is used to receive postcopy requested pages */
149     QemuThread postcopy_prio_thread;
150     /*
151      * Always set by the main vm load thread only, but can be read by the
152      * postcopy preempt thread.  "volatile" makes sure all reads will be
153      * up-to-date across cores.
154      */
155     volatile PreemptThreadStatus preempt_thread_status;
156     /*
157      * Used to sync between the ram load main thread and the fast ram load
158      * thread.  It protects postcopy_qemufile_dst, which is the postcopy
159      * fast channel.
160      *
161      * The ram fast load thread will take it mostly for the whole lifecycle
162      * because it needs to continuously read data from the channel, and
163      * it'll only release this mutex if postcopy is interrupted, so that
164      * the ram load main thread will take this mutex over and properly
165      * release the broken channel.
166      */
167     QemuMutex postcopy_prio_thread_mutex;
168     /*
169      * An array of temp host huge pages to be used, one for each postcopy
170      * channel.
171      */
172     PostcopyTmpPage *postcopy_tmp_pages;
173     /* This is shared for all postcopy channels */
174     void     *postcopy_tmp_zero_page;
175     /* PostCopyFD's for external userfaultfds & handlers of shared memory */
176     GArray   *postcopy_remote_fds;
177 
178     MigrationStatus state;
179 
180     /*
181      * The incoming migration coroutine, non-NULL during qemu_loadvm_state().
182      * Used to wake the migration incoming coroutine from rdma code. How much is
183      * it safe - it's a question.
184      */
185     Coroutine *loadvm_co;
186 
187     /* The coroutine we should enter (back) after failover */
188     Coroutine *colo_incoming_co;
189     QemuEvent colo_incoming_event;
190 
191     /* Optional load threads pool and its thread exit request flag */
192     ThreadPool *load_threads;
193     bool load_threads_abort;
194 
195     /*
196      * PostcopyBlocktimeContext to keep information for postcopy
197      * live migration, to calculate vCPU block time
198      * */
199     struct PostcopyBlocktimeContext *blocktime_ctx;
200 
201     /* notify PAUSED postcopy incoming migrations to try to continue */
202     QemuSemaphore postcopy_pause_sem_dst;
203     QemuSemaphore postcopy_pause_sem_fault;
204     /*
205      * This semaphore is used to allow the ram fast load thread (only when
206      * postcopy preempt is enabled) fall into sleep when there's network
207      * interruption detected.  When the recovery is done, the main load
208      * thread will kick the fast ram load thread using this semaphore.
209      */
210     QemuSemaphore postcopy_pause_sem_fast_load;
211 
212     /* List of listening socket addresses  */
213     SocketAddressList *socket_address_list;
214 
215     /* A tree of pages that we requested to the source VM */
216     GTree *page_requested;
217     /*
218      * For postcopy only, count the number of requested page faults that
219      * still haven't been resolved.
220      */
221     int page_requested_count;
222     /*
223      * The mutex helps to maintain the requested pages that we sent to the
224      * source, IOW, to guarantee coherent between the page_requests tree and
225      * the per-ramblock receivedmap.  Note! This does not guarantee consistency
226      * of the real page copy procedures (using UFFDIO_[ZERO]COPY).  E.g., even
227      * if one bit in receivedmap is cleared, UFFDIO_COPY could have happened
228      * for that page already.  This is intended so that the mutex won't
229      * serialize and blocked by slow operations like UFFDIO_* ioctls.  However
230      * this should be enough to make sure the page_requested tree always
231      * contains valid information.
232      */
233     QemuMutex page_request_mutex;
234     /*
235      * If postcopy preempt is enabled, there is a chance that the main
236      * thread finished loading its data before the preempt channel has
237      * finished loading the urgent pages.  If that happens, the two threads
238      * will use this condvar to synchronize, so the main thread will always
239      * wait until all pages received.
240      */
241     QemuCond page_request_cond;
242 
243     /*
244      * Number of devices that have yet to approve switchover. When this reaches
245      * zero an ACK that it's OK to do switchover is sent to the source. No lock
246      * is needed as this field is updated serially.
247      */
248     unsigned int switchover_ack_pending_num;
249 
250     /* Do exit on incoming migration failure */
251     bool exit_on_error;
252 };
253 
254 MigrationIncomingState *migration_incoming_get_current(void);
255 void migration_incoming_state_destroy(void);
256 void migration_incoming_transport_cleanup(MigrationIncomingState *mis);
257 /*
258  * Functions to work with blocktime context
259  */
260 void fill_destination_postcopy_migration_info(MigrationInfo *info);
261 
262 #define TYPE_MIGRATION "migration"
263 
264 typedef struct MigrationClass MigrationClass;
265 DECLARE_OBJ_CHECKERS(MigrationState, MigrationClass,
266                      MIGRATION_OBJ, TYPE_MIGRATION)
267 
268 struct MigrationClass {
269     /*< private >*/
270     DeviceClass parent_class;
271 };
272 
273 struct MigrationState {
274     /*< private >*/
275     DeviceState parent_obj;
276 
277     /*< public >*/
278     QemuThread thread;
279     /* Protected by qemu_file_lock */
280     QEMUFile *to_dst_file;
281     /* Postcopy specific transfer channel */
282     QEMUFile *postcopy_qemufile_src;
283     /*
284      * It is posted when the preempt channel is established.  Note: this is
285      * used for both the start or recover of a postcopy migration.  We'll
286      * post to this sem every time a new preempt channel is created in the
287      * main thread, and we keep post() and wait() in pair.
288      */
289     QemuSemaphore postcopy_qemufile_src_sem;
290     QIOChannelBuffer *bioc;
291     /*
292      * Protects to_dst_file/from_dst_file pointers.  We need to make sure we
293      * won't yield or hang during the critical section, since this lock will be
294      * used in OOB command handler.
295      */
296     QemuMutex qemu_file_lock;
297 
298     /*
299      * Used to allow urgent requests to override rate limiting.
300      */
301     QemuSemaphore rate_limit_sem;
302 
303     /* pages already send at the beginning of current iteration */
304     uint64_t iteration_initial_pages;
305 
306     /* pages transferred per second */
307     double pages_per_second;
308 
309     /* bytes already send at the beginning of current iteration */
310     uint64_t iteration_initial_bytes;
311     /* time at the start of current iteration */
312     int64_t iteration_start_time;
313     /*
314      * The final stage happens when the remaining data is smaller than
315      * this threshold; it's calculated from the requested downtime and
316      * measured bandwidth, or avail-switchover-bandwidth if specified.
317      */
318     uint64_t threshold_size;
319 
320     /* params from 'migrate-set-parameters' */
321     MigrationParameters parameters;
322 
323     MigrationStatus state;
324 
325     /* State related to return path */
326     struct {
327         /* Protected by qemu_file_lock */
328         QEMUFile     *from_dst_file;
329         QemuThread    rp_thread;
330         /*
331          * We can also check non-zero of rp_thread, but there's no "official"
332          * way to do this, so this bool makes it slightly more elegant.
333          * Checking from_dst_file for this is racy because from_dst_file will
334          * be cleared in the rp_thread!
335          */
336         bool          rp_thread_created;
337         /*
338          * Used to synchronize between migration main thread and return
339          * path thread.  The migration thread can wait() on this sem, while
340          * other threads (e.g., return path thread) can kick it using a
341          * post().
342          */
343         QemuSemaphore rp_sem;
344         /*
345          * We post to this when we got one PONG from dest. So far it's an
346          * easy way to know the main channel has successfully established
347          * on dest QEMU.
348          */
349         QemuSemaphore rp_pong_acks;
350     } rp_state;
351 
352     double mbps;
353     /* Timestamp when recent migration starts (ms) */
354     int64_t start_time;
355     /* Total time used by latest migration (ms) */
356     int64_t total_time;
357     /* Timestamp when VM is down (ms) to migrate the last stuff */
358     int64_t downtime_start;
359     int64_t downtime;
360     int64_t expected_downtime;
361     bool capabilities[MIGRATION_CAPABILITY__MAX];
362     int64_t setup_time;
363 
364     /*
365      * State before stopping the vm by vm_stop_force_state().
366      * If migration is interrupted by any reason, we need to continue
367      * running the guest on source if it was running or restore its stopped
368      * state.
369      */
370     RunState vm_old_state;
371 
372     /* Flag set once the migration has been asked to enter postcopy */
373     bool start_postcopy;
374 
375     /* Flag set once the migration thread is running (and needs joining) */
376     bool migration_thread_running;
377 
378     /* Migration is waiting for guest to unplug device */
379     QemuSemaphore wait_unplug_sem;
380 
381     /* Migration is paused due to pause-before-switchover */
382     QemuEvent pause_event;
383 
384     /* The event is used to notify COLO thread that failover is finished */
385     QemuEvent colo_exit_event;
386 
387     /* The event is used to notify COLO thread to do checkpoint */
388     QemuEvent colo_checkpoint_event;
389     int64_t colo_checkpoint_time;
390     QEMUTimer *colo_delay_timer;
391 
392     /* The first error that has occurred.
393        We used the mutex to be able to return the 1st error message */
394     Error *error;
395     /* mutex to protect errp */
396     QemuMutex error_mutex;
397 
398     /*
399      * Global switch on whether we need to store the global state
400      * during migration.
401      */
402     bool store_global_state;
403 
404     /* Whether we send QEMU_VM_CONFIGURATION during migration */
405     bool send_configuration;
406     /* Whether we send section footer during migration */
407     bool send_section_footer;
408     /* Whether we send switchover start notification during migration */
409     bool send_switchover_start;
410 
411     /* Needed by postcopy-pause state */
412     QemuSemaphore postcopy_pause_sem;
413     /*
414      * This variable only affects behavior when postcopy preempt mode is
415      * enabled.
416      *
417      * When set:
418      *
419      * - postcopy preempt src QEMU instance will generate an EOS message at
420      *   the end of migration to shut the preempt channel on dest side.
421      *
422      * - postcopy preempt channel will be created at the setup phase on src
423          QEMU.
424      *
425      * When clear:
426      *
427      * - postcopy preempt src QEMU instance will _not_ generate an EOS
428      *   message at the end of migration, the dest qemu will shutdown the
429      *   channel itself.
430      *
431      * - postcopy preempt channel will be created at the switching phase
432      *   from precopy -> postcopy (to avoid race condition of misordered
433      *   creation of channels).
434      *
435      * NOTE: See message-id <ZBoShWArKDPpX/D7@work-vm> on qemu-devel
436      * mailing list for more information on the possible race.  Everyone
437      * should probably just keep this value untouched after set by the
438      * machine type (or the default).
439      */
440     bool preempt_pre_7_2;
441 
442     /*
443      * flush every channel after each section sent.
444      *
445      * This assures that we can't mix pages from one iteration through
446      * ram pages with pages for the following iteration.  We really
447      * only need to do this flush after we have go through all the
448      * dirty pages.  For historical reasons, we do that after each
449      * section.  This is suboptimal (we flush too many times).
450      * Default value is false. (since 8.1)
451      */
452     bool multifd_flush_after_each_section;
453 
454     /*
455      * This variable only makes sense when set on the machine that is
456      * the destination of a multifd migration with TLS enabled. It
457      * affects the behavior of the last send->recv iteration with
458      * regards to termination of the TLS session.
459      *
460      * When set:
461      *
462      * - the destination QEMU instance can expect to never get a
463      *   GNUTLS_E_PREMATURE_TERMINATION error. Manifested as the error
464      *   message: "The TLS connection was non-properly terminated".
465      *
466      * When clear:
467      *
468      * - the destination QEMU instance can expect to see a
469      *   GNUTLS_E_PREMATURE_TERMINATION error in any multifd channel
470      *   whenever the last recv() call of that channel happens after
471      *   the source QEMU instance has already issued shutdown() on the
472      *   channel.
473      *
474      *   Commit 637280aeb2 (since 9.1) introduced a side effect that
475      *   causes the destination instance to not be affected by the
476      *   premature termination, while commit 1d457daf86 (since 10.0)
477      *   causes the premature termination condition to be once again
478      *   reachable.
479      *
480      * NOTE: Regardless of the state of this option, a premature
481      * termination of the TLS connection might happen due to error at
482      * any moment prior to the last send->recv iteration.
483      */
484     bool multifd_clean_tls_termination;
485 
486     /*
487      * This decides the size of guest memory chunk that will be used
488      * to track dirty bitmap clearing.  The size of memory chunk will
489      * be GUEST_PAGE_SIZE << N.  Say, N=0 means we will clear dirty
490      * bitmap for each page to send (1<<0=1); N=10 means we will clear
491      * dirty bitmap only once for 1<<10=1K continuous guest pages
492      * (which is in 4M chunk).
493      */
494     uint8_t clear_bitmap_shift;
495 
496     /*
497      * This save hostname when out-going migration starts
498      */
499     char *hostname;
500 
501     /* QEMU_VM_VMDESCRIPTION content filled for all non-iterable devices. */
502     JSONWriter *vmdesc;
503 
504     /*
505      * Indicates whether an ACK from the destination that it's OK to do
506      * switchover has been received.
507      */
508     bool switchover_acked;
509     /* Is this a rdma migration */
510     bool rdma_migration;
511 
512     GSource *hup_source;
513 };
514 
515 void migrate_set_state(MigrationStatus *state, MigrationStatus old_state,
516                        MigrationStatus new_state);
517 
518 void migration_fd_process_incoming(QEMUFile *f);
519 void migration_ioc_process_incoming(QIOChannel *ioc, Error **errp);
520 void migration_incoming_process(void);
521 
522 bool  migration_has_all_channels(void);
523 
524 void migrate_set_error(MigrationState *s, const Error *error);
525 bool migrate_has_error(MigrationState *s);
526 
527 void migration_connect(MigrationState *s, Error *error_in);
528 
529 int migration_call_notifiers(MigrationState *s, MigrationEventType type,
530                              Error **errp);
531 
532 int migrate_init(MigrationState *s, Error **errp);
533 bool migration_is_blocked(Error **errp);
534 /* True if outgoing migration has entered postcopy phase */
535 bool migration_in_postcopy(void);
536 bool migration_postcopy_is_alive(MigrationStatus state);
537 MigrationState *migrate_get_current(void);
538 bool migration_has_failed(MigrationState *);
539 bool migrate_mode_is_cpr(MigrationState *);
540 
541 uint64_t ram_get_total_transferred_pages(void);
542 
543 /* Sending on the return path - generic and then for each message type */
544 void migrate_send_rp_shut(MigrationIncomingState *mis,
545                           uint32_t value);
546 void migrate_send_rp_pong(MigrationIncomingState *mis,
547                           uint32_t value);
548 int migrate_send_rp_req_pages(MigrationIncomingState *mis, RAMBlock *rb,
549                               ram_addr_t start, uint64_t haddr);
550 int migrate_send_rp_message_req_pages(MigrationIncomingState *mis,
551                                       RAMBlock *rb, ram_addr_t start);
552 void migrate_send_rp_recv_bitmap(MigrationIncomingState *mis,
553                                  char *block_name);
554 void migrate_send_rp_resume_ack(MigrationIncomingState *mis, uint32_t value);
555 int migrate_send_rp_switchover_ack(MigrationIncomingState *mis);
556 
557 void dirty_bitmap_mig_before_vm_start(void);
558 void dirty_bitmap_mig_cancel_outgoing(void);
559 void dirty_bitmap_mig_cancel_incoming(void);
560 bool check_dirty_bitmap_mig_alias_map(const BitmapMigrationNodeAliasList *bbm,
561                                       Error **errp);
562 
563 void migrate_add_address(SocketAddress *address);
564 int foreach_not_ignored_block(RAMBlockIterFunc func, void *opaque);
565 
566 #define qemu_ram_foreach_block \
567   #warning "Use foreach_not_ignored_block in migration code"
568 
569 void migration_make_urgent_request(void);
570 void migration_consume_urgent_request(void);
571 bool migration_rate_limit(void);
572 void migration_bh_schedule(QEMUBHFunc *cb, void *opaque);
573 void migration_cancel(void);
574 
575 void migration_populate_vfio_info(MigrationInfo *info);
576 void migration_reset_vfio_bytes_transferred(void);
577 void postcopy_temp_page_reset(PostcopyTmpPage *tmp_page);
578 
579 /*
580  * Migration thread waiting for return path thread.  Return non-zero if an
581  * error is detected.
582  */
583 int migration_rp_wait(MigrationState *s);
584 /*
585  * Kick the migration thread waiting for return path messages.  NOTE: the
586  * name can be slightly confusing (when read as "kick the rp thread"), just
587  * to remember the target is always the migration thread.
588  */
589 void migration_rp_kick(MigrationState *s);
590 
591 void migration_bitmap_sync_precopy(bool last_stage);
592 
593 /* migration/block-dirty-bitmap.c */
594 void dirty_bitmap_mig_init(void);
595 bool should_send_vmdesc(void);
596 
597 #endif
598