xref: /qemu/migration/multifd.c (revision a81df1b6)
1 /*
2  * Multifd common code
3  *
4  * Copyright (c) 2019-2020 Red Hat Inc
5  *
6  * Authors:
7  *  Juan Quintela <quintela@redhat.com>
8  *
9  * This work is licensed under the terms of the GNU GPL, version 2 or later.
10  * See the COPYING file in the top-level directory.
11  */
12 
13 #include "qemu/osdep.h"
14 #include "qemu/rcu.h"
15 #include "exec/target_page.h"
16 #include "sysemu/sysemu.h"
17 #include "exec/ramblock.h"
18 #include "qemu/error-report.h"
19 #include "qapi/error.h"
20 #include "ram.h"
21 #include "migration.h"
22 #include "socket.h"
23 #include "qemu-file.h"
24 #include "trace.h"
25 #include "multifd.h"
26 
27 /* Multiple fd's */
28 
29 #define MULTIFD_MAGIC 0x11223344U
30 #define MULTIFD_VERSION 1
31 
32 typedef struct {
33     uint32_t magic;
34     uint32_t version;
35     unsigned char uuid[16]; /* QemuUUID */
36     uint8_t id;
37     uint8_t unused1[7];     /* Reserved for future use */
38     uint64_t unused2[4];    /* Reserved for future use */
39 } __attribute__((packed)) MultiFDInit_t;
40 
41 /* Multifd without compression */
42 
43 /**
44  * nocomp_send_setup: setup send side
45  *
46  * For no compression this function does nothing.
47  *
48  * Returns 0 for success or -1 for error
49  *
50  * @p: Params for the channel that we are using
51  * @errp: pointer to an error
52  */
53 static int nocomp_send_setup(MultiFDSendParams *p, Error **errp)
54 {
55     return 0;
56 }
57 
58 /**
59  * nocomp_send_cleanup: cleanup send side
60  *
61  * For no compression this function does nothing.
62  *
63  * @p: Params for the channel that we are using
64  */
65 static void nocomp_send_cleanup(MultiFDSendParams *p, Error **errp)
66 {
67     return;
68 }
69 
70 /**
71  * nocomp_send_prepare: prepare date to be able to send
72  *
73  * For no compression we just have to calculate the size of the
74  * packet.
75  *
76  * Returns 0 for success or -1 for error
77  *
78  * @p: Params for the channel that we are using
79  * @used: number of pages used
80  * @errp: pointer to an error
81  */
82 static int nocomp_send_prepare(MultiFDSendParams *p, uint32_t used,
83                                Error **errp)
84 {
85     p->next_packet_size = used * qemu_target_page_size();
86     p->flags |= MULTIFD_FLAG_NOCOMP;
87     return 0;
88 }
89 
90 /**
91  * nocomp_send_write: do the actual write of the data
92  *
93  * For no compression we just have to write the data.
94  *
95  * Returns 0 for success or -1 for error
96  *
97  * @p: Params for the channel that we are using
98  * @used: number of pages used
99  * @errp: pointer to an error
100  */
101 static int nocomp_send_write(MultiFDSendParams *p, uint32_t used, Error **errp)
102 {
103     return qio_channel_writev_all(p->c, p->pages->iov, used, errp);
104 }
105 
106 /**
107  * nocomp_recv_setup: setup receive side
108  *
109  * For no compression this function does nothing.
110  *
111  * Returns 0 for success or -1 for error
112  *
113  * @p: Params for the channel that we are using
114  * @errp: pointer to an error
115  */
116 static int nocomp_recv_setup(MultiFDRecvParams *p, Error **errp)
117 {
118     return 0;
119 }
120 
121 /**
122  * nocomp_recv_cleanup: setup receive side
123  *
124  * For no compression this function does nothing.
125  *
126  * @p: Params for the channel that we are using
127  */
128 static void nocomp_recv_cleanup(MultiFDRecvParams *p)
129 {
130 }
131 
132 /**
133  * nocomp_recv_pages: read the data from the channel into actual pages
134  *
135  * For no compression we just need to read things into the correct place.
136  *
137  * Returns 0 for success or -1 for error
138  *
139  * @p: Params for the channel that we are using
140  * @used: number of pages used
141  * @errp: pointer to an error
142  */
143 static int nocomp_recv_pages(MultiFDRecvParams *p, uint32_t used, Error **errp)
144 {
145     uint32_t flags = p->flags & MULTIFD_FLAG_COMPRESSION_MASK;
146 
147     if (flags != MULTIFD_FLAG_NOCOMP) {
148         error_setg(errp, "multifd %d: flags received %x flags expected %x",
149                    p->id, flags, MULTIFD_FLAG_NOCOMP);
150         return -1;
151     }
152     return qio_channel_readv_all(p->c, p->pages->iov, used, errp);
153 }
154 
155 static MultiFDMethods multifd_nocomp_ops = {
156     .send_setup = nocomp_send_setup,
157     .send_cleanup = nocomp_send_cleanup,
158     .send_prepare = nocomp_send_prepare,
159     .send_write = nocomp_send_write,
160     .recv_setup = nocomp_recv_setup,
161     .recv_cleanup = nocomp_recv_cleanup,
162     .recv_pages = nocomp_recv_pages
163 };
164 
165 static MultiFDMethods *multifd_ops[MULTIFD_COMPRESSION__MAX] = {
166     [MULTIFD_COMPRESSION_NONE] = &multifd_nocomp_ops,
167 };
168 
169 void multifd_register_ops(int method, MultiFDMethods *ops)
170 {
171     assert(0 < method && method < MULTIFD_COMPRESSION__MAX);
172     multifd_ops[method] = ops;
173 }
174 
175 static int multifd_send_initial_packet(MultiFDSendParams *p, Error **errp)
176 {
177     MultiFDInit_t msg = {};
178     int ret;
179 
180     msg.magic = cpu_to_be32(MULTIFD_MAGIC);
181     msg.version = cpu_to_be32(MULTIFD_VERSION);
182     msg.id = p->id;
183     memcpy(msg.uuid, &qemu_uuid.data, sizeof(msg.uuid));
184 
185     ret = qio_channel_write_all(p->c, (char *)&msg, sizeof(msg), errp);
186     if (ret != 0) {
187         return -1;
188     }
189     return 0;
190 }
191 
192 static int multifd_recv_initial_packet(QIOChannel *c, Error **errp)
193 {
194     MultiFDInit_t msg;
195     int ret;
196 
197     ret = qio_channel_read_all(c, (char *)&msg, sizeof(msg), errp);
198     if (ret != 0) {
199         return -1;
200     }
201 
202     msg.magic = be32_to_cpu(msg.magic);
203     msg.version = be32_to_cpu(msg.version);
204 
205     if (msg.magic != MULTIFD_MAGIC) {
206         error_setg(errp, "multifd: received packet magic %x "
207                    "expected %x", msg.magic, MULTIFD_MAGIC);
208         return -1;
209     }
210 
211     if (msg.version != MULTIFD_VERSION) {
212         error_setg(errp, "multifd: received packet version %d "
213                    "expected %d", msg.version, MULTIFD_VERSION);
214         return -1;
215     }
216 
217     if (memcmp(msg.uuid, &qemu_uuid, sizeof(qemu_uuid))) {
218         char *uuid = qemu_uuid_unparse_strdup(&qemu_uuid);
219         char *msg_uuid = qemu_uuid_unparse_strdup((const QemuUUID *)msg.uuid);
220 
221         error_setg(errp, "multifd: received uuid '%s' and expected "
222                    "uuid '%s' for channel %hhd", msg_uuid, uuid, msg.id);
223         g_free(uuid);
224         g_free(msg_uuid);
225         return -1;
226     }
227 
228     if (msg.id > migrate_multifd_channels()) {
229         error_setg(errp, "multifd: received channel version %d "
230                    "expected %d", msg.version, MULTIFD_VERSION);
231         return -1;
232     }
233 
234     return msg.id;
235 }
236 
237 static MultiFDPages_t *multifd_pages_init(size_t size)
238 {
239     MultiFDPages_t *pages = g_new0(MultiFDPages_t, 1);
240 
241     pages->allocated = size;
242     pages->iov = g_new0(struct iovec, size);
243     pages->offset = g_new0(ram_addr_t, size);
244 
245     return pages;
246 }
247 
248 static void multifd_pages_clear(MultiFDPages_t *pages)
249 {
250     pages->used = 0;
251     pages->allocated = 0;
252     pages->packet_num = 0;
253     pages->block = NULL;
254     g_free(pages->iov);
255     pages->iov = NULL;
256     g_free(pages->offset);
257     pages->offset = NULL;
258     g_free(pages);
259 }
260 
261 static void multifd_send_fill_packet(MultiFDSendParams *p)
262 {
263     MultiFDPacket_t *packet = p->packet;
264     int i;
265 
266     packet->flags = cpu_to_be32(p->flags);
267     packet->pages_alloc = cpu_to_be32(p->pages->allocated);
268     packet->pages_used = cpu_to_be32(p->pages->used);
269     packet->next_packet_size = cpu_to_be32(p->next_packet_size);
270     packet->packet_num = cpu_to_be64(p->packet_num);
271 
272     if (p->pages->block) {
273         strncpy(packet->ramblock, p->pages->block->idstr, 256);
274     }
275 
276     for (i = 0; i < p->pages->used; i++) {
277         /* there are architectures where ram_addr_t is 32 bit */
278         uint64_t temp = p->pages->offset[i];
279 
280         packet->offset[i] = cpu_to_be64(temp);
281     }
282 }
283 
284 static int multifd_recv_unfill_packet(MultiFDRecvParams *p, Error **errp)
285 {
286     MultiFDPacket_t *packet = p->packet;
287     uint32_t pages_max = MULTIFD_PACKET_SIZE / qemu_target_page_size();
288     RAMBlock *block;
289     int i;
290 
291     packet->magic = be32_to_cpu(packet->magic);
292     if (packet->magic != MULTIFD_MAGIC) {
293         error_setg(errp, "multifd: received packet "
294                    "magic %x and expected magic %x",
295                    packet->magic, MULTIFD_MAGIC);
296         return -1;
297     }
298 
299     packet->version = be32_to_cpu(packet->version);
300     if (packet->version != MULTIFD_VERSION) {
301         error_setg(errp, "multifd: received packet "
302                    "version %d and expected version %d",
303                    packet->version, MULTIFD_VERSION);
304         return -1;
305     }
306 
307     p->flags = be32_to_cpu(packet->flags);
308 
309     packet->pages_alloc = be32_to_cpu(packet->pages_alloc);
310     /*
311      * If we received a packet that is 100 times bigger than expected
312      * just stop migration.  It is a magic number.
313      */
314     if (packet->pages_alloc > pages_max * 100) {
315         error_setg(errp, "multifd: received packet "
316                    "with size %d and expected a maximum size of %d",
317                    packet->pages_alloc, pages_max * 100) ;
318         return -1;
319     }
320     /*
321      * We received a packet that is bigger than expected but inside
322      * reasonable limits (see previous comment).  Just reallocate.
323      */
324     if (packet->pages_alloc > p->pages->allocated) {
325         multifd_pages_clear(p->pages);
326         p->pages = multifd_pages_init(packet->pages_alloc);
327     }
328 
329     p->pages->used = be32_to_cpu(packet->pages_used);
330     if (p->pages->used > packet->pages_alloc) {
331         error_setg(errp, "multifd: received packet "
332                    "with %d pages and expected maximum pages are %d",
333                    p->pages->used, packet->pages_alloc) ;
334         return -1;
335     }
336 
337     p->next_packet_size = be32_to_cpu(packet->next_packet_size);
338     p->packet_num = be64_to_cpu(packet->packet_num);
339 
340     if (p->pages->used == 0) {
341         return 0;
342     }
343 
344     /* make sure that ramblock is 0 terminated */
345     packet->ramblock[255] = 0;
346     block = qemu_ram_block_by_name(packet->ramblock);
347     if (!block) {
348         error_setg(errp, "multifd: unknown ram block %s",
349                    packet->ramblock);
350         return -1;
351     }
352 
353     for (i = 0; i < p->pages->used; i++) {
354         uint64_t offset = be64_to_cpu(packet->offset[i]);
355 
356         if (offset > (block->used_length - qemu_target_page_size())) {
357             error_setg(errp, "multifd: offset too long %" PRIu64
358                        " (max " RAM_ADDR_FMT ")",
359                        offset, block->max_length);
360             return -1;
361         }
362         p->pages->iov[i].iov_base = block->host + offset;
363         p->pages->iov[i].iov_len = qemu_target_page_size();
364     }
365 
366     return 0;
367 }
368 
369 struct {
370     MultiFDSendParams *params;
371     /* array of pages to sent */
372     MultiFDPages_t *pages;
373     /* global number of generated multifd packets */
374     uint64_t packet_num;
375     /* send channels ready */
376     QemuSemaphore channels_ready;
377     /*
378      * Have we already run terminate threads.  There is a race when it
379      * happens that we got one error while we are exiting.
380      * We will use atomic operations.  Only valid values are 0 and 1.
381      */
382     int exiting;
383     /* multifd ops */
384     MultiFDMethods *ops;
385 } *multifd_send_state;
386 
387 /*
388  * How we use multifd_send_state->pages and channel->pages?
389  *
390  * We create a pages for each channel, and a main one.  Each time that
391  * we need to send a batch of pages we interchange the ones between
392  * multifd_send_state and the channel that is sending it.  There are
393  * two reasons for that:
394  *    - to not have to do so many mallocs during migration
395  *    - to make easier to know what to free at the end of migration
396  *
397  * This way we always know who is the owner of each "pages" struct,
398  * and we don't need any locking.  It belongs to the migration thread
399  * or to the channel thread.  Switching is safe because the migration
400  * thread is using the channel mutex when changing it, and the channel
401  * have to had finish with its own, otherwise pending_job can't be
402  * false.
403  */
404 
405 static int multifd_send_pages(QEMUFile *f)
406 {
407     int i;
408     static int next_channel;
409     MultiFDSendParams *p = NULL; /* make happy gcc */
410     MultiFDPages_t *pages = multifd_send_state->pages;
411     uint64_t transferred;
412 
413     if (atomic_read(&multifd_send_state->exiting)) {
414         return -1;
415     }
416 
417     qemu_sem_wait(&multifd_send_state->channels_ready);
418     /*
419      * next_channel can remain from a previous migration that was
420      * using more channels, so ensure it doesn't overflow if the
421      * limit is lower now.
422      */
423     next_channel %= migrate_multifd_channels();
424     for (i = next_channel;; i = (i + 1) % migrate_multifd_channels()) {
425         p = &multifd_send_state->params[i];
426 
427         qemu_mutex_lock(&p->mutex);
428         if (p->quit) {
429             error_report("%s: channel %d has already quit!", __func__, i);
430             qemu_mutex_unlock(&p->mutex);
431             return -1;
432         }
433         if (!p->pending_job) {
434             p->pending_job++;
435             next_channel = (i + 1) % migrate_multifd_channels();
436             break;
437         }
438         qemu_mutex_unlock(&p->mutex);
439     }
440     assert(!p->pages->used);
441     assert(!p->pages->block);
442 
443     p->packet_num = multifd_send_state->packet_num++;
444     multifd_send_state->pages = p->pages;
445     p->pages = pages;
446     transferred = ((uint64_t) pages->used) * qemu_target_page_size()
447                 + p->packet_len;
448     qemu_file_update_transfer(f, transferred);
449     ram_counters.multifd_bytes += transferred;
450     ram_counters.transferred += transferred;;
451     qemu_mutex_unlock(&p->mutex);
452     qemu_sem_post(&p->sem);
453 
454     return 1;
455 }
456 
457 int multifd_queue_page(QEMUFile *f, RAMBlock *block, ram_addr_t offset)
458 {
459     MultiFDPages_t *pages = multifd_send_state->pages;
460 
461     if (!pages->block) {
462         pages->block = block;
463     }
464 
465     if (pages->block == block) {
466         pages->offset[pages->used] = offset;
467         pages->iov[pages->used].iov_base = block->host + offset;
468         pages->iov[pages->used].iov_len = qemu_target_page_size();
469         pages->used++;
470 
471         if (pages->used < pages->allocated) {
472             return 1;
473         }
474     }
475 
476     if (multifd_send_pages(f) < 0) {
477         return -1;
478     }
479 
480     if (pages->block != block) {
481         return  multifd_queue_page(f, block, offset);
482     }
483 
484     return 1;
485 }
486 
487 static void multifd_send_terminate_threads(Error *err)
488 {
489     int i;
490 
491     trace_multifd_send_terminate_threads(err != NULL);
492 
493     if (err) {
494         MigrationState *s = migrate_get_current();
495         migrate_set_error(s, err);
496         if (s->state == MIGRATION_STATUS_SETUP ||
497             s->state == MIGRATION_STATUS_PRE_SWITCHOVER ||
498             s->state == MIGRATION_STATUS_DEVICE ||
499             s->state == MIGRATION_STATUS_ACTIVE) {
500             migrate_set_state(&s->state, s->state,
501                               MIGRATION_STATUS_FAILED);
502         }
503     }
504 
505     /*
506      * We don't want to exit each threads twice.  Depending on where
507      * we get the error, or if there are two independent errors in two
508      * threads at the same time, we can end calling this function
509      * twice.
510      */
511     if (atomic_xchg(&multifd_send_state->exiting, 1)) {
512         return;
513     }
514 
515     for (i = 0; i < migrate_multifd_channels(); i++) {
516         MultiFDSendParams *p = &multifd_send_state->params[i];
517 
518         qemu_mutex_lock(&p->mutex);
519         p->quit = true;
520         qemu_sem_post(&p->sem);
521         qemu_mutex_unlock(&p->mutex);
522     }
523 }
524 
525 void multifd_save_cleanup(void)
526 {
527     int i;
528 
529     if (!migrate_use_multifd()) {
530         return;
531     }
532     multifd_send_terminate_threads(NULL);
533     for (i = 0; i < migrate_multifd_channels(); i++) {
534         MultiFDSendParams *p = &multifd_send_state->params[i];
535 
536         if (p->running) {
537             qemu_thread_join(&p->thread);
538         }
539     }
540     for (i = 0; i < migrate_multifd_channels(); i++) {
541         MultiFDSendParams *p = &multifd_send_state->params[i];
542         Error *local_err = NULL;
543 
544         socket_send_channel_destroy(p->c);
545         p->c = NULL;
546         qemu_mutex_destroy(&p->mutex);
547         qemu_sem_destroy(&p->sem);
548         qemu_sem_destroy(&p->sem_sync);
549         g_free(p->name);
550         p->name = NULL;
551         multifd_pages_clear(p->pages);
552         p->pages = NULL;
553         p->packet_len = 0;
554         g_free(p->packet);
555         p->packet = NULL;
556         multifd_send_state->ops->send_cleanup(p, &local_err);
557         if (local_err) {
558             migrate_set_error(migrate_get_current(), local_err);
559             error_free(local_err);
560         }
561     }
562     qemu_sem_destroy(&multifd_send_state->channels_ready);
563     g_free(multifd_send_state->params);
564     multifd_send_state->params = NULL;
565     multifd_pages_clear(multifd_send_state->pages);
566     multifd_send_state->pages = NULL;
567     g_free(multifd_send_state);
568     multifd_send_state = NULL;
569 }
570 
571 void multifd_send_sync_main(QEMUFile *f)
572 {
573     int i;
574 
575     if (!migrate_use_multifd()) {
576         return;
577     }
578     if (multifd_send_state->pages->used) {
579         if (multifd_send_pages(f) < 0) {
580             error_report("%s: multifd_send_pages fail", __func__);
581             return;
582         }
583     }
584     for (i = 0; i < migrate_multifd_channels(); i++) {
585         MultiFDSendParams *p = &multifd_send_state->params[i];
586 
587         trace_multifd_send_sync_main_signal(p->id);
588 
589         qemu_mutex_lock(&p->mutex);
590 
591         if (p->quit) {
592             error_report("%s: channel %d has already quit", __func__, i);
593             qemu_mutex_unlock(&p->mutex);
594             return;
595         }
596 
597         p->packet_num = multifd_send_state->packet_num++;
598         p->flags |= MULTIFD_FLAG_SYNC;
599         p->pending_job++;
600         qemu_file_update_transfer(f, p->packet_len);
601         ram_counters.multifd_bytes += p->packet_len;
602         ram_counters.transferred += p->packet_len;
603         qemu_mutex_unlock(&p->mutex);
604         qemu_sem_post(&p->sem);
605     }
606     for (i = 0; i < migrate_multifd_channels(); i++) {
607         MultiFDSendParams *p = &multifd_send_state->params[i];
608 
609         trace_multifd_send_sync_main_wait(p->id);
610         qemu_sem_wait(&p->sem_sync);
611     }
612     trace_multifd_send_sync_main(multifd_send_state->packet_num);
613 }
614 
615 static void *multifd_send_thread(void *opaque)
616 {
617     MultiFDSendParams *p = opaque;
618     Error *local_err = NULL;
619     int ret = 0;
620     uint32_t flags = 0;
621 
622     trace_multifd_send_thread_start(p->id);
623     rcu_register_thread();
624 
625     if (multifd_send_initial_packet(p, &local_err) < 0) {
626         ret = -1;
627         goto out;
628     }
629     /* initial packet */
630     p->num_packets = 1;
631 
632     while (true) {
633         qemu_sem_wait(&p->sem);
634 
635         if (atomic_read(&multifd_send_state->exiting)) {
636             break;
637         }
638         qemu_mutex_lock(&p->mutex);
639 
640         if (p->pending_job) {
641             uint32_t used = p->pages->used;
642             uint64_t packet_num = p->packet_num;
643             flags = p->flags;
644 
645             if (used) {
646                 ret = multifd_send_state->ops->send_prepare(p, used,
647                                                             &local_err);
648                 if (ret != 0) {
649                     qemu_mutex_unlock(&p->mutex);
650                     break;
651                 }
652             }
653             multifd_send_fill_packet(p);
654             p->flags = 0;
655             p->num_packets++;
656             p->num_pages += used;
657             p->pages->used = 0;
658             p->pages->block = NULL;
659             qemu_mutex_unlock(&p->mutex);
660 
661             trace_multifd_send(p->id, packet_num, used, flags,
662                                p->next_packet_size);
663 
664             ret = qio_channel_write_all(p->c, (void *)p->packet,
665                                         p->packet_len, &local_err);
666             if (ret != 0) {
667                 break;
668             }
669 
670             if (used) {
671                 ret = multifd_send_state->ops->send_write(p, used, &local_err);
672                 if (ret != 0) {
673                     break;
674                 }
675             }
676 
677             qemu_mutex_lock(&p->mutex);
678             p->pending_job--;
679             qemu_mutex_unlock(&p->mutex);
680 
681             if (flags & MULTIFD_FLAG_SYNC) {
682                 qemu_sem_post(&p->sem_sync);
683             }
684             qemu_sem_post(&multifd_send_state->channels_ready);
685         } else if (p->quit) {
686             qemu_mutex_unlock(&p->mutex);
687             break;
688         } else {
689             qemu_mutex_unlock(&p->mutex);
690             /* sometimes there are spurious wakeups */
691         }
692     }
693 
694 out:
695     if (local_err) {
696         trace_multifd_send_error(p->id);
697         multifd_send_terminate_threads(local_err);
698         error_free(local_err);
699     }
700 
701     /*
702      * Error happen, I will exit, but I can't just leave, tell
703      * who pay attention to me.
704      */
705     if (ret != 0) {
706         qemu_sem_post(&p->sem_sync);
707         qemu_sem_post(&multifd_send_state->channels_ready);
708     }
709 
710     qemu_mutex_lock(&p->mutex);
711     p->running = false;
712     qemu_mutex_unlock(&p->mutex);
713 
714     rcu_unregister_thread();
715     trace_multifd_send_thread_end(p->id, p->num_packets, p->num_pages);
716 
717     return NULL;
718 }
719 
720 static void multifd_new_send_channel_async(QIOTask *task, gpointer opaque)
721 {
722     MultiFDSendParams *p = opaque;
723     QIOChannel *sioc = QIO_CHANNEL(qio_task_get_source(task));
724     Error *local_err = NULL;
725 
726     trace_multifd_new_send_channel_async(p->id);
727     if (qio_task_propagate_error(task, &local_err)) {
728         migrate_set_error(migrate_get_current(), local_err);
729         /* Error happen, we need to tell who pay attention to me */
730         qemu_sem_post(&multifd_send_state->channels_ready);
731         qemu_sem_post(&p->sem_sync);
732         /*
733          * Although multifd_send_thread is not created, but main migration
734          * thread neet to judge whether it is running, so we need to mark
735          * its status.
736          */
737         p->quit = true;
738         object_unref(OBJECT(sioc));
739         error_free(local_err);
740     } else {
741         p->c = QIO_CHANNEL(sioc);
742         qio_channel_set_delay(p->c, false);
743         p->running = true;
744         qemu_thread_create(&p->thread, p->name, multifd_send_thread, p,
745                            QEMU_THREAD_JOINABLE);
746     }
747 }
748 
749 int multifd_save_setup(Error **errp)
750 {
751     int thread_count;
752     uint32_t page_count = MULTIFD_PACKET_SIZE / qemu_target_page_size();
753     uint8_t i;
754 
755     if (!migrate_use_multifd()) {
756         return 0;
757     }
758     thread_count = migrate_multifd_channels();
759     multifd_send_state = g_malloc0(sizeof(*multifd_send_state));
760     multifd_send_state->params = g_new0(MultiFDSendParams, thread_count);
761     multifd_send_state->pages = multifd_pages_init(page_count);
762     qemu_sem_init(&multifd_send_state->channels_ready, 0);
763     atomic_set(&multifd_send_state->exiting, 0);
764     multifd_send_state->ops = multifd_ops[migrate_multifd_compression()];
765 
766     for (i = 0; i < thread_count; i++) {
767         MultiFDSendParams *p = &multifd_send_state->params[i];
768 
769         qemu_mutex_init(&p->mutex);
770         qemu_sem_init(&p->sem, 0);
771         qemu_sem_init(&p->sem_sync, 0);
772         p->quit = false;
773         p->pending_job = 0;
774         p->id = i;
775         p->pages = multifd_pages_init(page_count);
776         p->packet_len = sizeof(MultiFDPacket_t)
777                       + sizeof(uint64_t) * page_count;
778         p->packet = g_malloc0(p->packet_len);
779         p->packet->magic = cpu_to_be32(MULTIFD_MAGIC);
780         p->packet->version = cpu_to_be32(MULTIFD_VERSION);
781         p->name = g_strdup_printf("multifdsend_%d", i);
782         socket_send_channel_create(multifd_new_send_channel_async, p);
783     }
784 
785     for (i = 0; i < thread_count; i++) {
786         MultiFDSendParams *p = &multifd_send_state->params[i];
787         Error *local_err = NULL;
788         int ret;
789 
790         ret = multifd_send_state->ops->send_setup(p, &local_err);
791         if (ret) {
792             error_propagate(errp, local_err);
793             return ret;
794         }
795     }
796     return 0;
797 }
798 
799 struct {
800     MultiFDRecvParams *params;
801     /* number of created threads */
802     int count;
803     /* syncs main thread and channels */
804     QemuSemaphore sem_sync;
805     /* global number of generated multifd packets */
806     uint64_t packet_num;
807     /* multifd ops */
808     MultiFDMethods *ops;
809 } *multifd_recv_state;
810 
811 static void multifd_recv_terminate_threads(Error *err)
812 {
813     int i;
814 
815     trace_multifd_recv_terminate_threads(err != NULL);
816 
817     if (err) {
818         MigrationState *s = migrate_get_current();
819         migrate_set_error(s, err);
820         if (s->state == MIGRATION_STATUS_SETUP ||
821             s->state == MIGRATION_STATUS_ACTIVE) {
822             migrate_set_state(&s->state, s->state,
823                               MIGRATION_STATUS_FAILED);
824         }
825     }
826 
827     for (i = 0; i < migrate_multifd_channels(); i++) {
828         MultiFDRecvParams *p = &multifd_recv_state->params[i];
829 
830         qemu_mutex_lock(&p->mutex);
831         p->quit = true;
832         /*
833          * We could arrive here for two reasons:
834          *  - normal quit, i.e. everything went fine, just finished
835          *  - error quit: We close the channels so the channel threads
836          *    finish the qio_channel_read_all_eof()
837          */
838         if (p->c) {
839             qio_channel_shutdown(p->c, QIO_CHANNEL_SHUTDOWN_BOTH, NULL);
840         }
841         qemu_mutex_unlock(&p->mutex);
842     }
843 }
844 
845 int multifd_load_cleanup(Error **errp)
846 {
847     int i;
848 
849     if (!migrate_use_multifd()) {
850         return 0;
851     }
852     multifd_recv_terminate_threads(NULL);
853     for (i = 0; i < migrate_multifd_channels(); i++) {
854         MultiFDRecvParams *p = &multifd_recv_state->params[i];
855 
856         if (p->running) {
857             p->quit = true;
858             /*
859              * multifd_recv_thread may hung at MULTIFD_FLAG_SYNC handle code,
860              * however try to wakeup it without harm in cleanup phase.
861              */
862             qemu_sem_post(&p->sem_sync);
863             qemu_thread_join(&p->thread);
864         }
865     }
866     for (i = 0; i < migrate_multifd_channels(); i++) {
867         MultiFDRecvParams *p = &multifd_recv_state->params[i];
868 
869         object_unref(OBJECT(p->c));
870         p->c = NULL;
871         qemu_mutex_destroy(&p->mutex);
872         qemu_sem_destroy(&p->sem_sync);
873         g_free(p->name);
874         p->name = NULL;
875         multifd_pages_clear(p->pages);
876         p->pages = NULL;
877         p->packet_len = 0;
878         g_free(p->packet);
879         p->packet = NULL;
880         multifd_recv_state->ops->recv_cleanup(p);
881     }
882     qemu_sem_destroy(&multifd_recv_state->sem_sync);
883     g_free(multifd_recv_state->params);
884     multifd_recv_state->params = NULL;
885     g_free(multifd_recv_state);
886     multifd_recv_state = NULL;
887 
888     return 0;
889 }
890 
891 void multifd_recv_sync_main(void)
892 {
893     int i;
894 
895     if (!migrate_use_multifd()) {
896         return;
897     }
898     for (i = 0; i < migrate_multifd_channels(); i++) {
899         MultiFDRecvParams *p = &multifd_recv_state->params[i];
900 
901         trace_multifd_recv_sync_main_wait(p->id);
902         qemu_sem_wait(&multifd_recv_state->sem_sync);
903     }
904     for (i = 0; i < migrate_multifd_channels(); i++) {
905         MultiFDRecvParams *p = &multifd_recv_state->params[i];
906 
907         WITH_QEMU_LOCK_GUARD(&p->mutex) {
908             if (multifd_recv_state->packet_num < p->packet_num) {
909                 multifd_recv_state->packet_num = p->packet_num;
910             }
911         }
912         trace_multifd_recv_sync_main_signal(p->id);
913         qemu_sem_post(&p->sem_sync);
914     }
915     trace_multifd_recv_sync_main(multifd_recv_state->packet_num);
916 }
917 
918 static void *multifd_recv_thread(void *opaque)
919 {
920     MultiFDRecvParams *p = opaque;
921     Error *local_err = NULL;
922     int ret;
923 
924     trace_multifd_recv_thread_start(p->id);
925     rcu_register_thread();
926 
927     while (true) {
928         uint32_t used;
929         uint32_t flags;
930 
931         if (p->quit) {
932             break;
933         }
934 
935         ret = qio_channel_read_all_eof(p->c, (void *)p->packet,
936                                        p->packet_len, &local_err);
937         if (ret == 0) {   /* EOF */
938             break;
939         }
940         if (ret == -1) {   /* Error */
941             break;
942         }
943 
944         qemu_mutex_lock(&p->mutex);
945         ret = multifd_recv_unfill_packet(p, &local_err);
946         if (ret) {
947             qemu_mutex_unlock(&p->mutex);
948             break;
949         }
950 
951         used = p->pages->used;
952         flags = p->flags;
953         /* recv methods don't know how to handle the SYNC flag */
954         p->flags &= ~MULTIFD_FLAG_SYNC;
955         trace_multifd_recv(p->id, p->packet_num, used, flags,
956                            p->next_packet_size);
957         p->num_packets++;
958         p->num_pages += used;
959         qemu_mutex_unlock(&p->mutex);
960 
961         if (used) {
962             ret = multifd_recv_state->ops->recv_pages(p, used, &local_err);
963             if (ret != 0) {
964                 break;
965             }
966         }
967 
968         if (flags & MULTIFD_FLAG_SYNC) {
969             qemu_sem_post(&multifd_recv_state->sem_sync);
970             qemu_sem_wait(&p->sem_sync);
971         }
972     }
973 
974     if (local_err) {
975         multifd_recv_terminate_threads(local_err);
976         error_free(local_err);
977     }
978     qemu_mutex_lock(&p->mutex);
979     p->running = false;
980     qemu_mutex_unlock(&p->mutex);
981 
982     rcu_unregister_thread();
983     trace_multifd_recv_thread_end(p->id, p->num_packets, p->num_pages);
984 
985     return NULL;
986 }
987 
988 int multifd_load_setup(Error **errp)
989 {
990     int thread_count;
991     uint32_t page_count = MULTIFD_PACKET_SIZE / qemu_target_page_size();
992     uint8_t i;
993 
994     if (!migrate_use_multifd()) {
995         return 0;
996     }
997     thread_count = migrate_multifd_channels();
998     multifd_recv_state = g_malloc0(sizeof(*multifd_recv_state));
999     multifd_recv_state->params = g_new0(MultiFDRecvParams, thread_count);
1000     atomic_set(&multifd_recv_state->count, 0);
1001     qemu_sem_init(&multifd_recv_state->sem_sync, 0);
1002     multifd_recv_state->ops = multifd_ops[migrate_multifd_compression()];
1003 
1004     for (i = 0; i < thread_count; i++) {
1005         MultiFDRecvParams *p = &multifd_recv_state->params[i];
1006 
1007         qemu_mutex_init(&p->mutex);
1008         qemu_sem_init(&p->sem_sync, 0);
1009         p->quit = false;
1010         p->id = i;
1011         p->pages = multifd_pages_init(page_count);
1012         p->packet_len = sizeof(MultiFDPacket_t)
1013                       + sizeof(uint64_t) * page_count;
1014         p->packet = g_malloc0(p->packet_len);
1015         p->name = g_strdup_printf("multifdrecv_%d", i);
1016     }
1017 
1018     for (i = 0; i < thread_count; i++) {
1019         MultiFDRecvParams *p = &multifd_recv_state->params[i];
1020         Error *local_err = NULL;
1021         int ret;
1022 
1023         ret = multifd_recv_state->ops->recv_setup(p, &local_err);
1024         if (ret) {
1025             error_propagate(errp, local_err);
1026             return ret;
1027         }
1028     }
1029     return 0;
1030 }
1031 
1032 bool multifd_recv_all_channels_created(void)
1033 {
1034     int thread_count = migrate_multifd_channels();
1035 
1036     if (!migrate_use_multifd()) {
1037         return true;
1038     }
1039 
1040     return thread_count == atomic_read(&multifd_recv_state->count);
1041 }
1042 
1043 /*
1044  * Try to receive all multifd channels to get ready for the migration.
1045  * - Return true and do not set @errp when correctly receving all channels;
1046  * - Return false and do not set @errp when correctly receiving the current one;
1047  * - Return false and set @errp when failing to receive the current channel.
1048  */
1049 bool multifd_recv_new_channel(QIOChannel *ioc, Error **errp)
1050 {
1051     MultiFDRecvParams *p;
1052     Error *local_err = NULL;
1053     int id;
1054 
1055     id = multifd_recv_initial_packet(ioc, &local_err);
1056     if (id < 0) {
1057         multifd_recv_terminate_threads(local_err);
1058         error_propagate_prepend(errp, local_err,
1059                                 "failed to receive packet"
1060                                 " via multifd channel %d: ",
1061                                 atomic_read(&multifd_recv_state->count));
1062         return false;
1063     }
1064     trace_multifd_recv_new_channel(id);
1065 
1066     p = &multifd_recv_state->params[id];
1067     if (p->c != NULL) {
1068         error_setg(&local_err, "multifd: received id '%d' already setup'",
1069                    id);
1070         multifd_recv_terminate_threads(local_err);
1071         error_propagate(errp, local_err);
1072         return false;
1073     }
1074     p->c = ioc;
1075     object_ref(OBJECT(ioc));
1076     /* initial packet */
1077     p->num_packets = 1;
1078 
1079     p->running = true;
1080     qemu_thread_create(&p->thread, p->name, multifd_recv_thread, p,
1081                        QEMU_THREAD_JOINABLE);
1082     atomic_inc(&multifd_recv_state->count);
1083     return atomic_read(&multifd_recv_state->count) ==
1084            migrate_multifd_channels();
1085 }
1086