xref: /openbmc/qemu/migration/migration.h (revision 3011c1dd)
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     int state;
163 
164     /*
165      * The incoming migration coroutine, non-NULL during qemu_loadvm_state().
166      * Used to wake the migration incoming coroutine from rdma code. How much is
167      * it safe - it's a question.
168      */
169     Coroutine *loadvm_co;
170 
171     /* The coroutine we should enter (back) after failover */
172     Coroutine *colo_incoming_co;
173     QemuSemaphore colo_incoming_sem;
174 
175     /*
176      * PostcopyBlocktimeContext to keep information for postcopy
177      * live migration, to calculate vCPU block time
178      * */
179     struct PostcopyBlocktimeContext *blocktime_ctx;
180 
181     /* notify PAUSED postcopy incoming migrations to try to continue */
182     QemuSemaphore postcopy_pause_sem_dst;
183     QemuSemaphore postcopy_pause_sem_fault;
184     /*
185      * This semaphore is used to allow the ram fast load thread (only when
186      * postcopy preempt is enabled) fall into sleep when there's network
187      * interruption detected.  When the recovery is done, the main load
188      * thread will kick the fast ram load thread using this semaphore.
189      */
190     QemuSemaphore postcopy_pause_sem_fast_load;
191 
192     /* List of listening socket addresses  */
193     SocketAddressList *socket_address_list;
194 
195     /* A tree of pages that we requested to the source VM */
196     GTree *page_requested;
197     /*
198      * For postcopy only, count the number of requested page faults that
199      * still haven't been resolved.
200      */
201     int page_requested_count;
202     /*
203      * The mutex helps to maintain the requested pages that we sent to the
204      * source, IOW, to guarantee coherent between the page_requests tree and
205      * the per-ramblock receivedmap.  Note! This does not guarantee consistency
206      * of the real page copy procedures (using UFFDIO_[ZERO]COPY).  E.g., even
207      * if one bit in receivedmap is cleared, UFFDIO_COPY could have happened
208      * for that page already.  This is intended so that the mutex won't
209      * serialize and blocked by slow operations like UFFDIO_* ioctls.  However
210      * this should be enough to make sure the page_requested tree always
211      * contains valid information.
212      */
213     QemuMutex page_request_mutex;
214     /*
215      * If postcopy preempt is enabled, there is a chance that the main
216      * thread finished loading its data before the preempt channel has
217      * finished loading the urgent pages.  If that happens, the two threads
218      * will use this condvar to synchronize, so the main thread will always
219      * wait until all pages received.
220      */
221     QemuCond page_request_cond;
222 
223     /*
224      * Number of devices that have yet to approve switchover. When this reaches
225      * zero an ACK that it's OK to do switchover is sent to the source. No lock
226      * is needed as this field is updated serially.
227      */
228     unsigned int switchover_ack_pending_num;
229 };
230 
231 MigrationIncomingState *migration_incoming_get_current(void);
232 void migration_incoming_state_destroy(void);
233 void migration_incoming_transport_cleanup(MigrationIncomingState *mis);
234 /*
235  * Functions to work with blocktime context
236  */
237 void fill_destination_postcopy_migration_info(MigrationInfo *info);
238 
239 #define TYPE_MIGRATION "migration"
240 
241 typedef struct MigrationClass MigrationClass;
242 DECLARE_OBJ_CHECKERS(MigrationState, MigrationClass,
243                      MIGRATION_OBJ, TYPE_MIGRATION)
244 
245 struct MigrationClass {
246     /*< private >*/
247     DeviceClass parent_class;
248 };
249 
250 struct MigrationState {
251     /*< private >*/
252     DeviceState parent_obj;
253 
254     /*< public >*/
255     QemuThread thread;
256     /* Protected by qemu_file_lock */
257     QEMUFile *to_dst_file;
258     /* Postcopy specific transfer channel */
259     QEMUFile *postcopy_qemufile_src;
260     /*
261      * It is posted when the preempt channel is established.  Note: this is
262      * used for both the start or recover of a postcopy migration.  We'll
263      * post to this sem every time a new preempt channel is created in the
264      * main thread, and we keep post() and wait() in pair.
265      */
266     QemuSemaphore postcopy_qemufile_src_sem;
267     QIOChannelBuffer *bioc;
268     /*
269      * Protects to_dst_file/from_dst_file pointers.  We need to make sure we
270      * won't yield or hang during the critical section, since this lock will be
271      * used in OOB command handler.
272      */
273     QemuMutex qemu_file_lock;
274 
275     /*
276      * Used to allow urgent requests to override rate limiting.
277      */
278     QemuSemaphore rate_limit_sem;
279 
280     /* pages already send at the beginning of current iteration */
281     uint64_t iteration_initial_pages;
282 
283     /* pages transferred per second */
284     double pages_per_second;
285 
286     /* bytes already send at the beginning of current iteration */
287     uint64_t iteration_initial_bytes;
288     /* time at the start of current iteration */
289     int64_t iteration_start_time;
290     /*
291      * The final stage happens when the remaining data is smaller than
292      * this threshold; it's calculated from the requested downtime and
293      * measured bandwidth, or avail-switchover-bandwidth if specified.
294      */
295     uint64_t threshold_size;
296 
297     /* params from 'migrate-set-parameters' */
298     MigrationParameters parameters;
299 
300     int state;
301 
302     /* State related to return path */
303     struct {
304         /* Protected by qemu_file_lock */
305         QEMUFile     *from_dst_file;
306         QemuThread    rp_thread;
307         /*
308          * We can also check non-zero of rp_thread, but there's no "official"
309          * way to do this, so this bool makes it slightly more elegant.
310          * Checking from_dst_file for this is racy because from_dst_file will
311          * be cleared in the rp_thread!
312          */
313         bool          rp_thread_created;
314         /*
315          * Used to synchronize between migration main thread and return
316          * path thread.  The migration thread can wait() on this sem, while
317          * other threads (e.g., return path thread) can kick it using a
318          * post().
319          */
320         QemuSemaphore rp_sem;
321         /*
322          * We post to this when we got one PONG from dest. So far it's an
323          * easy way to know the main channel has successfully established
324          * on dest QEMU.
325          */
326         QemuSemaphore rp_pong_acks;
327     } rp_state;
328 
329     double mbps;
330     /* Timestamp when recent migration starts (ms) */
331     int64_t start_time;
332     /* Total time used by latest migration (ms) */
333     int64_t total_time;
334     /* Timestamp when VM is down (ms) to migrate the last stuff */
335     int64_t downtime_start;
336     int64_t downtime;
337     int64_t expected_downtime;
338     bool capabilities[MIGRATION_CAPABILITY__MAX];
339     int64_t setup_time;
340 
341     /*
342      * State before stopping the vm by vm_stop_force_state().
343      * If migration is interrupted by any reason, we need to continue
344      * running the guest on source if it was running or restore its stopped
345      * state.
346      */
347     RunState vm_old_state;
348 
349     /* Flag set once the migration has been asked to enter postcopy */
350     bool start_postcopy;
351 
352     /* Flag set once the migration thread is running (and needs joining) */
353     bool migration_thread_running;
354 
355     /* Flag set once the migration thread called bdrv_inactivate_all */
356     bool block_inactive;
357 
358     /* Migration is waiting for guest to unplug device */
359     QemuSemaphore wait_unplug_sem;
360 
361     /* Migration is paused due to pause-before-switchover */
362     QemuSemaphore pause_sem;
363 
364     /* The semaphore is used to notify COLO thread that failover is finished */
365     QemuSemaphore colo_exit_sem;
366 
367     /* The event is used to notify COLO thread to do checkpoint */
368     QemuEvent colo_checkpoint_event;
369     int64_t colo_checkpoint_time;
370     QEMUTimer *colo_delay_timer;
371 
372     /* The first error that has occurred.
373        We used the mutex to be able to return the 1st error message */
374     Error *error;
375     /* mutex to protect errp */
376     QemuMutex error_mutex;
377 
378     /* Do we have to clean up -b/-i from old migrate parameters */
379     /* This feature is deprecated and will be removed */
380     bool must_remove_block_options;
381 
382     /*
383      * Global switch on whether we need to store the global state
384      * during migration.
385      */
386     bool store_global_state;
387 
388     /* Whether we send QEMU_VM_CONFIGURATION during migration */
389     bool send_configuration;
390     /* Whether we send section footer during migration */
391     bool send_section_footer;
392 
393     /* Needed by postcopy-pause state */
394     QemuSemaphore postcopy_pause_sem;
395     /*
396      * Whether we abort the migration if decompression errors are
397      * detected at the destination. It is left at false for qemu
398      * older than 3.0, since only newer qemu sends streams that
399      * do not trigger spurious decompression errors.
400      */
401     bool decompress_error_check;
402     /*
403      * This variable only affects behavior when postcopy preempt mode is
404      * enabled.
405      *
406      * When set:
407      *
408      * - postcopy preempt src QEMU instance will generate an EOS message at
409      *   the end of migration to shut the preempt channel on dest side.
410      *
411      * - postcopy preempt channel will be created at the setup phase on src
412          QEMU.
413      *
414      * When clear:
415      *
416      * - postcopy preempt src QEMU instance will _not_ generate an EOS
417      *   message at the end of migration, the dest qemu will shutdown the
418      *   channel itself.
419      *
420      * - postcopy preempt channel will be created at the switching phase
421      *   from precopy -> postcopy (to avoid race condition of misordered
422      *   creation of channels).
423      *
424      * NOTE: See message-id <ZBoShWArKDPpX/D7@work-vm> on qemu-devel
425      * mailing list for more information on the possible race.  Everyone
426      * should probably just keep this value untouched after set by the
427      * machine type (or the default).
428      */
429     bool preempt_pre_7_2;
430 
431     /*
432      * flush every channel after each section sent.
433      *
434      * This assures that we can't mix pages from one iteration through
435      * ram pages with pages for the following iteration.  We really
436      * only need to do this flush after we have go through all the
437      * dirty pages.  For historical reasons, we do that after each
438      * section.  This is suboptimal (we flush too many times).
439      * Default value is false. (since 8.1)
440      */
441     bool multifd_flush_after_each_section;
442     /*
443      * This decides the size of guest memory chunk that will be used
444      * to track dirty bitmap clearing.  The size of memory chunk will
445      * be GUEST_PAGE_SIZE << N.  Say, N=0 means we will clear dirty
446      * bitmap for each page to send (1<<0=1); N=10 means we will clear
447      * dirty bitmap only once for 1<<10=1K continuous guest pages
448      * (which is in 4M chunk).
449      */
450     uint8_t clear_bitmap_shift;
451 
452     /*
453      * This save hostname when out-going migration starts
454      */
455     char *hostname;
456 
457     /* QEMU_VM_VMDESCRIPTION content filled for all non-iterable devices. */
458     JSONWriter *vmdesc;
459 
460     /*
461      * Indicates whether an ACK from the destination that it's OK to do
462      * switchover has been received.
463      */
464     bool switchover_acked;
465     /* Is this a rdma migration */
466     bool rdma_migration;
467 };
468 
469 void migrate_set_state(int *state, int old_state, int new_state);
470 
471 void migration_fd_process_incoming(QEMUFile *f);
472 void migration_ioc_process_incoming(QIOChannel *ioc, Error **errp);
473 void migration_incoming_process(void);
474 
475 bool  migration_has_all_channels(void);
476 
477 void migrate_set_error(MigrationState *s, const Error *error);
478 bool migrate_has_error(MigrationState *s);
479 
480 void migrate_fd_connect(MigrationState *s, Error *error_in);
481 
482 bool migration_is_setup_or_active(int state);
483 bool migration_is_running(int state);
484 
485 int migrate_init(MigrationState *s, Error **errp);
486 bool migration_is_blocked(Error **errp);
487 /* True if outgoing migration has entered postcopy phase */
488 bool migration_in_postcopy(void);
489 bool migration_postcopy_is_alive(int state);
490 MigrationState *migrate_get_current(void);
491 
492 uint64_t ram_get_total_transferred_pages(void);
493 
494 /* Sending on the return path - generic and then for each message type */
495 void migrate_send_rp_shut(MigrationIncomingState *mis,
496                           uint32_t value);
497 void migrate_send_rp_pong(MigrationIncomingState *mis,
498                           uint32_t value);
499 int migrate_send_rp_req_pages(MigrationIncomingState *mis, RAMBlock *rb,
500                               ram_addr_t start, uint64_t haddr);
501 int migrate_send_rp_message_req_pages(MigrationIncomingState *mis,
502                                       RAMBlock *rb, ram_addr_t start);
503 void migrate_send_rp_recv_bitmap(MigrationIncomingState *mis,
504                                  char *block_name);
505 void migrate_send_rp_resume_ack(MigrationIncomingState *mis, uint32_t value);
506 int migrate_send_rp_switchover_ack(MigrationIncomingState *mis);
507 
508 void dirty_bitmap_mig_before_vm_start(void);
509 void dirty_bitmap_mig_cancel_outgoing(void);
510 void dirty_bitmap_mig_cancel_incoming(void);
511 bool check_dirty_bitmap_mig_alias_map(const BitmapMigrationNodeAliasList *bbm,
512                                       Error **errp);
513 
514 void migrate_add_address(SocketAddress *address);
515 bool migrate_uri_parse(const char *uri, MigrationChannel **channel,
516                        Error **errp);
517 int foreach_not_ignored_block(RAMBlockIterFunc func, void *opaque);
518 
519 #define qemu_ram_foreach_block \
520   #warning "Use foreach_not_ignored_block in migration code"
521 
522 void migration_make_urgent_request(void);
523 void migration_consume_urgent_request(void);
524 bool migration_rate_limit(void);
525 void migration_bh_schedule(QEMUBHFunc *cb, void *opaque);
526 void migration_cancel(const Error *error);
527 
528 void migration_populate_vfio_info(MigrationInfo *info);
529 void migration_reset_vfio_bytes_transferred(void);
530 void postcopy_temp_page_reset(PostcopyTmpPage *tmp_page);
531 
532 /*
533  * Migration thread waiting for return path thread.  Return non-zero if an
534  * error is detected.
535  */
536 int migration_rp_wait(MigrationState *s);
537 /*
538  * Kick the migration thread waiting for return path messages.  NOTE: the
539  * name can be slightly confusing (when read as "kick the rp thread"), just
540  * to remember the target is always the migration thread.
541  */
542 void migration_rp_kick(MigrationState *s);
543 
544 #endif
545