xref: /openbmc/qemu/migration/multifd.c (revision 6a094d62)
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 "tls.h"
24 #include "qemu-file.h"
25 #include "trace.h"
26 #include "multifd.h"
27 
28 #include "qemu/yank.h"
29 #include "io/channel-socket.h"
30 #include "yank_functions.h"
31 
32 /* Multiple fd's */
33 
34 #define MULTIFD_MAGIC 0x11223344U
35 #define MULTIFD_VERSION 1
36 
37 typedef struct {
38     uint32_t magic;
39     uint32_t version;
40     unsigned char uuid[16]; /* QemuUUID */
41     uint8_t id;
42     uint8_t unused1[7];     /* Reserved for future use */
43     uint64_t unused2[4];    /* Reserved for future use */
44 } __attribute__((packed)) MultiFDInit_t;
45 
46 /* Multifd without compression */
47 
48 /**
49  * nocomp_send_setup: setup send side
50  *
51  * For no compression this function does nothing.
52  *
53  * Returns 0 for success or -1 for error
54  *
55  * @p: Params for the channel that we are using
56  * @errp: pointer to an error
57  */
58 static int nocomp_send_setup(MultiFDSendParams *p, Error **errp)
59 {
60     return 0;
61 }
62 
63 /**
64  * nocomp_send_cleanup: cleanup send side
65  *
66  * For no compression this function does nothing.
67  *
68  * @p: Params for the channel that we are using
69  * @errp: pointer to an error
70  */
71 static void nocomp_send_cleanup(MultiFDSendParams *p, Error **errp)
72 {
73     return;
74 }
75 
76 /**
77  * nocomp_send_prepare: prepare date to be able to send
78  *
79  * For no compression we just have to calculate the size of the
80  * packet.
81  *
82  * Returns 0 for success or -1 for error
83  *
84  * @p: Params for the channel that we are using
85  * @errp: pointer to an error
86  */
87 static int nocomp_send_prepare(MultiFDSendParams *p, Error **errp)
88 {
89     p->next_packet_size = p->pages->num * qemu_target_page_size();
90     p->flags |= MULTIFD_FLAG_NOCOMP;
91     return 0;
92 }
93 
94 /**
95  * nocomp_send_write: do the actual write of the data
96  *
97  * For no compression we just have to write the data.
98  *
99  * Returns 0 for success or -1 for error
100  *
101  * @p: Params for the channel that we are using
102  * @used: number of pages used
103  * @errp: pointer to an error
104  */
105 static int nocomp_send_write(MultiFDSendParams *p, uint32_t used, Error **errp)
106 {
107     return qio_channel_writev_all(p->c, p->pages->iov, used, errp);
108 }
109 
110 /**
111  * nocomp_recv_setup: setup receive side
112  *
113  * For no compression this function does nothing.
114  *
115  * Returns 0 for success or -1 for error
116  *
117  * @p: Params for the channel that we are using
118  * @errp: pointer to an error
119  */
120 static int nocomp_recv_setup(MultiFDRecvParams *p, Error **errp)
121 {
122     return 0;
123 }
124 
125 /**
126  * nocomp_recv_cleanup: setup receive side
127  *
128  * For no compression this function does nothing.
129  *
130  * @p: Params for the channel that we are using
131  */
132 static void nocomp_recv_cleanup(MultiFDRecvParams *p)
133 {
134 }
135 
136 /**
137  * nocomp_recv_pages: read the data from the channel into actual pages
138  *
139  * For no compression we just need to read things into the correct place.
140  *
141  * Returns 0 for success or -1 for error
142  *
143  * @p: Params for the channel that we are using
144  * @errp: pointer to an error
145  */
146 static int nocomp_recv_pages(MultiFDRecvParams *p, Error **errp)
147 {
148     uint32_t flags = p->flags & MULTIFD_FLAG_COMPRESSION_MASK;
149 
150     if (flags != MULTIFD_FLAG_NOCOMP) {
151         error_setg(errp, "multifd %d: flags received %x flags expected %x",
152                    p->id, flags, MULTIFD_FLAG_NOCOMP);
153         return -1;
154     }
155     return qio_channel_readv_all(p->c, p->pages->iov, p->pages->num, errp);
156 }
157 
158 static MultiFDMethods multifd_nocomp_ops = {
159     .send_setup = nocomp_send_setup,
160     .send_cleanup = nocomp_send_cleanup,
161     .send_prepare = nocomp_send_prepare,
162     .send_write = nocomp_send_write,
163     .recv_setup = nocomp_recv_setup,
164     .recv_cleanup = nocomp_recv_cleanup,
165     .recv_pages = nocomp_recv_pages
166 };
167 
168 static MultiFDMethods *multifd_ops[MULTIFD_COMPRESSION__MAX] = {
169     [MULTIFD_COMPRESSION_NONE] = &multifd_nocomp_ops,
170 };
171 
172 void multifd_register_ops(int method, MultiFDMethods *ops)
173 {
174     assert(0 < method && method < MULTIFD_COMPRESSION__MAX);
175     multifd_ops[method] = ops;
176 }
177 
178 static int multifd_send_initial_packet(MultiFDSendParams *p, Error **errp)
179 {
180     MultiFDInit_t msg = {};
181     int ret;
182 
183     msg.magic = cpu_to_be32(MULTIFD_MAGIC);
184     msg.version = cpu_to_be32(MULTIFD_VERSION);
185     msg.id = p->id;
186     memcpy(msg.uuid, &qemu_uuid.data, sizeof(msg.uuid));
187 
188     ret = qio_channel_write_all(p->c, (char *)&msg, sizeof(msg), errp);
189     if (ret != 0) {
190         return -1;
191     }
192     return 0;
193 }
194 
195 static int multifd_recv_initial_packet(QIOChannel *c, Error **errp)
196 {
197     MultiFDInit_t msg;
198     int ret;
199 
200     ret = qio_channel_read_all(c, (char *)&msg, sizeof(msg), errp);
201     if (ret != 0) {
202         return -1;
203     }
204 
205     msg.magic = be32_to_cpu(msg.magic);
206     msg.version = be32_to_cpu(msg.version);
207 
208     if (msg.magic != MULTIFD_MAGIC) {
209         error_setg(errp, "multifd: received packet magic %x "
210                    "expected %x", msg.magic, MULTIFD_MAGIC);
211         return -1;
212     }
213 
214     if (msg.version != MULTIFD_VERSION) {
215         error_setg(errp, "multifd: received packet version %d "
216                    "expected %d", msg.version, MULTIFD_VERSION);
217         return -1;
218     }
219 
220     if (memcmp(msg.uuid, &qemu_uuid, sizeof(qemu_uuid))) {
221         char *uuid = qemu_uuid_unparse_strdup(&qemu_uuid);
222         char *msg_uuid = qemu_uuid_unparse_strdup((const QemuUUID *)msg.uuid);
223 
224         error_setg(errp, "multifd: received uuid '%s' and expected "
225                    "uuid '%s' for channel %hhd", msg_uuid, uuid, msg.id);
226         g_free(uuid);
227         g_free(msg_uuid);
228         return -1;
229     }
230 
231     if (msg.id > migrate_multifd_channels()) {
232         error_setg(errp, "multifd: received channel version %d "
233                    "expected %d", msg.version, MULTIFD_VERSION);
234         return -1;
235     }
236 
237     return msg.id;
238 }
239 
240 static MultiFDPages_t *multifd_pages_init(size_t size)
241 {
242     MultiFDPages_t *pages = g_new0(MultiFDPages_t, 1);
243 
244     pages->allocated = size;
245     pages->iov = g_new0(struct iovec, size);
246     pages->offset = g_new0(ram_addr_t, size);
247 
248     return pages;
249 }
250 
251 static void multifd_pages_clear(MultiFDPages_t *pages)
252 {
253     pages->num = 0;
254     pages->allocated = 0;
255     pages->packet_num = 0;
256     pages->block = NULL;
257     g_free(pages->iov);
258     pages->iov = NULL;
259     g_free(pages->offset);
260     pages->offset = NULL;
261     g_free(pages);
262 }
263 
264 static void multifd_send_fill_packet(MultiFDSendParams *p)
265 {
266     MultiFDPacket_t *packet = p->packet;
267     int i;
268 
269     packet->flags = cpu_to_be32(p->flags);
270     packet->pages_alloc = cpu_to_be32(p->pages->allocated);
271     packet->pages_used = cpu_to_be32(p->pages->num);
272     packet->next_packet_size = cpu_to_be32(p->next_packet_size);
273     packet->packet_num = cpu_to_be64(p->packet_num);
274 
275     if (p->pages->block) {
276         strncpy(packet->ramblock, p->pages->block->idstr, 256);
277     }
278 
279     for (i = 0; i < p->pages->num; i++) {
280         /* there are architectures where ram_addr_t is 32 bit */
281         uint64_t temp = p->pages->offset[i];
282 
283         packet->offset[i] = cpu_to_be64(temp);
284     }
285 }
286 
287 static int multifd_recv_unfill_packet(MultiFDRecvParams *p, Error **errp)
288 {
289     MultiFDPacket_t *packet = p->packet;
290     size_t page_size = qemu_target_page_size();
291     uint32_t pages_max = MULTIFD_PACKET_SIZE / page_size;
292     RAMBlock *block;
293     int i;
294 
295     packet->magic = be32_to_cpu(packet->magic);
296     if (packet->magic != MULTIFD_MAGIC) {
297         error_setg(errp, "multifd: received packet "
298                    "magic %x and expected magic %x",
299                    packet->magic, MULTIFD_MAGIC);
300         return -1;
301     }
302 
303     packet->version = be32_to_cpu(packet->version);
304     if (packet->version != MULTIFD_VERSION) {
305         error_setg(errp, "multifd: received packet "
306                    "version %d and expected version %d",
307                    packet->version, MULTIFD_VERSION);
308         return -1;
309     }
310 
311     p->flags = be32_to_cpu(packet->flags);
312 
313     packet->pages_alloc = be32_to_cpu(packet->pages_alloc);
314     /*
315      * If we received a packet that is 100 times bigger than expected
316      * just stop migration.  It is a magic number.
317      */
318     if (packet->pages_alloc > pages_max * 100) {
319         error_setg(errp, "multifd: received packet "
320                    "with size %d and expected a maximum size of %d",
321                    packet->pages_alloc, pages_max * 100) ;
322         return -1;
323     }
324     /*
325      * We received a packet that is bigger than expected but inside
326      * reasonable limits (see previous comment).  Just reallocate.
327      */
328     if (packet->pages_alloc > p->pages->allocated) {
329         multifd_pages_clear(p->pages);
330         p->pages = multifd_pages_init(packet->pages_alloc);
331     }
332 
333     p->pages->num = be32_to_cpu(packet->pages_used);
334     if (p->pages->num > packet->pages_alloc) {
335         error_setg(errp, "multifd: received packet "
336                    "with %d pages and expected maximum pages are %d",
337                    p->pages->num, packet->pages_alloc) ;
338         return -1;
339     }
340 
341     p->next_packet_size = be32_to_cpu(packet->next_packet_size);
342     p->packet_num = be64_to_cpu(packet->packet_num);
343 
344     if (p->pages->num == 0) {
345         return 0;
346     }
347 
348     /* make sure that ramblock is 0 terminated */
349     packet->ramblock[255] = 0;
350     block = qemu_ram_block_by_name(packet->ramblock);
351     if (!block) {
352         error_setg(errp, "multifd: unknown ram block %s",
353                    packet->ramblock);
354         return -1;
355     }
356 
357     p->pages->block = block;
358     for (i = 0; i < p->pages->num; i++) {
359         uint64_t offset = be64_to_cpu(packet->offset[i]);
360 
361         if (offset > (block->used_length - page_size)) {
362             error_setg(errp, "multifd: offset too long %" PRIu64
363                        " (max " RAM_ADDR_FMT ")",
364                        offset, block->used_length);
365             return -1;
366         }
367         p->pages->offset[i] = offset;
368         p->pages->iov[i].iov_base = block->host + offset;
369         p->pages->iov[i].iov_len = page_size;
370     }
371 
372     return 0;
373 }
374 
375 struct {
376     MultiFDSendParams *params;
377     /* array of pages to sent */
378     MultiFDPages_t *pages;
379     /* global number of generated multifd packets */
380     uint64_t packet_num;
381     /* send channels ready */
382     QemuSemaphore channels_ready;
383     /*
384      * Have we already run terminate threads.  There is a race when it
385      * happens that we got one error while we are exiting.
386      * We will use atomic operations.  Only valid values are 0 and 1.
387      */
388     int exiting;
389     /* multifd ops */
390     MultiFDMethods *ops;
391 } *multifd_send_state;
392 
393 /*
394  * How we use multifd_send_state->pages and channel->pages?
395  *
396  * We create a pages for each channel, and a main one.  Each time that
397  * we need to send a batch of pages we interchange the ones between
398  * multifd_send_state and the channel that is sending it.  There are
399  * two reasons for that:
400  *    - to not have to do so many mallocs during migration
401  *    - to make easier to know what to free at the end of migration
402  *
403  * This way we always know who is the owner of each "pages" struct,
404  * and we don't need any locking.  It belongs to the migration thread
405  * or to the channel thread.  Switching is safe because the migration
406  * thread is using the channel mutex when changing it, and the channel
407  * have to had finish with its own, otherwise pending_job can't be
408  * false.
409  */
410 
411 static int multifd_send_pages(QEMUFile *f)
412 {
413     int i;
414     static int next_channel;
415     MultiFDSendParams *p = NULL; /* make happy gcc */
416     MultiFDPages_t *pages = multifd_send_state->pages;
417     uint64_t transferred;
418 
419     if (qatomic_read(&multifd_send_state->exiting)) {
420         return -1;
421     }
422 
423     qemu_sem_wait(&multifd_send_state->channels_ready);
424     /*
425      * next_channel can remain from a previous migration that was
426      * using more channels, so ensure it doesn't overflow if the
427      * limit is lower now.
428      */
429     next_channel %= migrate_multifd_channels();
430     for (i = next_channel;; i = (i + 1) % migrate_multifd_channels()) {
431         p = &multifd_send_state->params[i];
432 
433         qemu_mutex_lock(&p->mutex);
434         if (p->quit) {
435             error_report("%s: channel %d has already quit!", __func__, i);
436             qemu_mutex_unlock(&p->mutex);
437             return -1;
438         }
439         if (!p->pending_job) {
440             p->pending_job++;
441             next_channel = (i + 1) % migrate_multifd_channels();
442             break;
443         }
444         qemu_mutex_unlock(&p->mutex);
445     }
446     assert(!p->pages->num);
447     assert(!p->pages->block);
448 
449     p->packet_num = multifd_send_state->packet_num++;
450     multifd_send_state->pages = p->pages;
451     p->pages = pages;
452     transferred = ((uint64_t) pages->num) * qemu_target_page_size()
453                 + p->packet_len;
454     qemu_file_update_transfer(f, transferred);
455     ram_counters.multifd_bytes += transferred;
456     ram_counters.transferred += transferred;
457     qemu_mutex_unlock(&p->mutex);
458     qemu_sem_post(&p->sem);
459 
460     return 1;
461 }
462 
463 int multifd_queue_page(QEMUFile *f, RAMBlock *block, ram_addr_t offset)
464 {
465     MultiFDPages_t *pages = multifd_send_state->pages;
466 
467     if (!pages->block) {
468         pages->block = block;
469     }
470 
471     if (pages->block == block) {
472         pages->offset[pages->num] = offset;
473         pages->iov[pages->num].iov_base = block->host + offset;
474         pages->iov[pages->num].iov_len = qemu_target_page_size();
475         pages->num++;
476 
477         if (pages->num < pages->allocated) {
478             return 1;
479         }
480     }
481 
482     if (multifd_send_pages(f) < 0) {
483         return -1;
484     }
485 
486     if (pages->block != block) {
487         return  multifd_queue_page(f, block, offset);
488     }
489 
490     return 1;
491 }
492 
493 static void multifd_send_terminate_threads(Error *err)
494 {
495     int i;
496 
497     trace_multifd_send_terminate_threads(err != NULL);
498 
499     if (err) {
500         MigrationState *s = migrate_get_current();
501         migrate_set_error(s, err);
502         if (s->state == MIGRATION_STATUS_SETUP ||
503             s->state == MIGRATION_STATUS_PRE_SWITCHOVER ||
504             s->state == MIGRATION_STATUS_DEVICE ||
505             s->state == MIGRATION_STATUS_ACTIVE) {
506             migrate_set_state(&s->state, s->state,
507                               MIGRATION_STATUS_FAILED);
508         }
509     }
510 
511     /*
512      * We don't want to exit each threads twice.  Depending on where
513      * we get the error, or if there are two independent errors in two
514      * threads at the same time, we can end calling this function
515      * twice.
516      */
517     if (qatomic_xchg(&multifd_send_state->exiting, 1)) {
518         return;
519     }
520 
521     for (i = 0; i < migrate_multifd_channels(); i++) {
522         MultiFDSendParams *p = &multifd_send_state->params[i];
523 
524         qemu_mutex_lock(&p->mutex);
525         p->quit = true;
526         qemu_sem_post(&p->sem);
527         if (p->c) {
528             qio_channel_shutdown(p->c, QIO_CHANNEL_SHUTDOWN_BOTH, NULL);
529         }
530         qemu_mutex_unlock(&p->mutex);
531     }
532 }
533 
534 void multifd_save_cleanup(void)
535 {
536     int i;
537 
538     if (!migrate_use_multifd() || !migrate_multifd_is_allowed()) {
539         return;
540     }
541     multifd_send_terminate_threads(NULL);
542     for (i = 0; i < migrate_multifd_channels(); i++) {
543         MultiFDSendParams *p = &multifd_send_state->params[i];
544 
545         if (p->running) {
546             qemu_thread_join(&p->thread);
547         }
548     }
549     for (i = 0; i < migrate_multifd_channels(); i++) {
550         MultiFDSendParams *p = &multifd_send_state->params[i];
551         Error *local_err = NULL;
552 
553         if (p->registered_yank) {
554             migration_ioc_unregister_yank(p->c);
555         }
556         socket_send_channel_destroy(p->c);
557         p->c = NULL;
558         qemu_mutex_destroy(&p->mutex);
559         qemu_sem_destroy(&p->sem);
560         qemu_sem_destroy(&p->sem_sync);
561         g_free(p->name);
562         p->name = NULL;
563         g_free(p->tls_hostname);
564         p->tls_hostname = NULL;
565         multifd_pages_clear(p->pages);
566         p->pages = NULL;
567         p->packet_len = 0;
568         g_free(p->packet);
569         p->packet = NULL;
570         multifd_send_state->ops->send_cleanup(p, &local_err);
571         if (local_err) {
572             migrate_set_error(migrate_get_current(), local_err);
573             error_free(local_err);
574         }
575     }
576     qemu_sem_destroy(&multifd_send_state->channels_ready);
577     g_free(multifd_send_state->params);
578     multifd_send_state->params = NULL;
579     multifd_pages_clear(multifd_send_state->pages);
580     multifd_send_state->pages = NULL;
581     g_free(multifd_send_state);
582     multifd_send_state = NULL;
583 }
584 
585 void multifd_send_sync_main(QEMUFile *f)
586 {
587     int i;
588 
589     if (!migrate_use_multifd()) {
590         return;
591     }
592     if (multifd_send_state->pages->num) {
593         if (multifd_send_pages(f) < 0) {
594             error_report("%s: multifd_send_pages fail", __func__);
595             return;
596         }
597     }
598     for (i = 0; i < migrate_multifd_channels(); i++) {
599         MultiFDSendParams *p = &multifd_send_state->params[i];
600 
601         trace_multifd_send_sync_main_signal(p->id);
602 
603         qemu_mutex_lock(&p->mutex);
604 
605         if (p->quit) {
606             error_report("%s: channel %d has already quit", __func__, i);
607             qemu_mutex_unlock(&p->mutex);
608             return;
609         }
610 
611         p->packet_num = multifd_send_state->packet_num++;
612         p->flags |= MULTIFD_FLAG_SYNC;
613         p->pending_job++;
614         qemu_file_update_transfer(f, p->packet_len);
615         ram_counters.multifd_bytes += p->packet_len;
616         ram_counters.transferred += p->packet_len;
617         qemu_mutex_unlock(&p->mutex);
618         qemu_sem_post(&p->sem);
619     }
620     for (i = 0; i < migrate_multifd_channels(); i++) {
621         MultiFDSendParams *p = &multifd_send_state->params[i];
622 
623         trace_multifd_send_sync_main_wait(p->id);
624         qemu_sem_wait(&p->sem_sync);
625     }
626     trace_multifd_send_sync_main(multifd_send_state->packet_num);
627 }
628 
629 static void *multifd_send_thread(void *opaque)
630 {
631     MultiFDSendParams *p = opaque;
632     Error *local_err = NULL;
633     int ret = 0;
634 
635     trace_multifd_send_thread_start(p->id);
636     rcu_register_thread();
637 
638     if (multifd_send_initial_packet(p, &local_err) < 0) {
639         ret = -1;
640         goto out;
641     }
642     /* initial packet */
643     p->num_packets = 1;
644 
645     while (true) {
646         qemu_sem_wait(&p->sem);
647 
648         if (qatomic_read(&multifd_send_state->exiting)) {
649             break;
650         }
651         qemu_mutex_lock(&p->mutex);
652 
653         if (p->pending_job) {
654             uint32_t used = p->pages->num;
655             uint64_t packet_num = p->packet_num;
656             uint32_t flags = p->flags;
657 
658             if (used) {
659                 ret = multifd_send_state->ops->send_prepare(p, &local_err);
660                 if (ret != 0) {
661                     qemu_mutex_unlock(&p->mutex);
662                     break;
663                 }
664             }
665             multifd_send_fill_packet(p);
666             p->flags = 0;
667             p->num_packets++;
668             p->num_pages += used;
669             p->pages->num = 0;
670             p->pages->block = NULL;
671             qemu_mutex_unlock(&p->mutex);
672 
673             trace_multifd_send(p->id, packet_num, used, flags,
674                                p->next_packet_size);
675 
676             ret = qio_channel_write_all(p->c, (void *)p->packet,
677                                         p->packet_len, &local_err);
678             if (ret != 0) {
679                 break;
680             }
681 
682             if (used) {
683                 ret = multifd_send_state->ops->send_write(p, used, &local_err);
684                 if (ret != 0) {
685                     break;
686                 }
687             }
688 
689             qemu_mutex_lock(&p->mutex);
690             p->pending_job--;
691             qemu_mutex_unlock(&p->mutex);
692 
693             if (flags & MULTIFD_FLAG_SYNC) {
694                 qemu_sem_post(&p->sem_sync);
695             }
696             qemu_sem_post(&multifd_send_state->channels_ready);
697         } else if (p->quit) {
698             qemu_mutex_unlock(&p->mutex);
699             break;
700         } else {
701             qemu_mutex_unlock(&p->mutex);
702             /* sometimes there are spurious wakeups */
703         }
704     }
705 
706 out:
707     if (local_err) {
708         trace_multifd_send_error(p->id);
709         multifd_send_terminate_threads(local_err);
710         error_free(local_err);
711     }
712 
713     /*
714      * Error happen, I will exit, but I can't just leave, tell
715      * who pay attention to me.
716      */
717     if (ret != 0) {
718         qemu_sem_post(&p->sem_sync);
719         qemu_sem_post(&multifd_send_state->channels_ready);
720     }
721 
722     qemu_mutex_lock(&p->mutex);
723     p->running = false;
724     qemu_mutex_unlock(&p->mutex);
725 
726     rcu_unregister_thread();
727     trace_multifd_send_thread_end(p->id, p->num_packets, p->num_pages);
728 
729     return NULL;
730 }
731 
732 static bool multifd_channel_connect(MultiFDSendParams *p,
733                                     QIOChannel *ioc,
734                                     Error *error);
735 
736 static void multifd_tls_outgoing_handshake(QIOTask *task,
737                                            gpointer opaque)
738 {
739     MultiFDSendParams *p = opaque;
740     QIOChannel *ioc = QIO_CHANNEL(qio_task_get_source(task));
741     Error *err = NULL;
742 
743     if (qio_task_propagate_error(task, &err)) {
744         trace_multifd_tls_outgoing_handshake_error(ioc, error_get_pretty(err));
745     } else {
746         trace_multifd_tls_outgoing_handshake_complete(ioc);
747     }
748 
749     if (!multifd_channel_connect(p, ioc, err)) {
750         /*
751          * Error happen, mark multifd_send_thread status as 'quit' although it
752          * is not created, and then tell who pay attention to me.
753          */
754         p->quit = true;
755         qemu_sem_post(&multifd_send_state->channels_ready);
756         qemu_sem_post(&p->sem_sync);
757     }
758 }
759 
760 static void *multifd_tls_handshake_thread(void *opaque)
761 {
762     MultiFDSendParams *p = opaque;
763     QIOChannelTLS *tioc = QIO_CHANNEL_TLS(p->c);
764 
765     qio_channel_tls_handshake(tioc,
766                               multifd_tls_outgoing_handshake,
767                               p,
768                               NULL,
769                               NULL);
770     return NULL;
771 }
772 
773 static void multifd_tls_channel_connect(MultiFDSendParams *p,
774                                         QIOChannel *ioc,
775                                         Error **errp)
776 {
777     MigrationState *s = migrate_get_current();
778     const char *hostname = p->tls_hostname;
779     QIOChannelTLS *tioc;
780 
781     tioc = migration_tls_client_create(s, ioc, hostname, errp);
782     if (!tioc) {
783         return;
784     }
785 
786     object_unref(OBJECT(ioc));
787     trace_multifd_tls_outgoing_handshake_start(ioc, tioc, hostname);
788     qio_channel_set_name(QIO_CHANNEL(tioc), "multifd-tls-outgoing");
789     p->c = QIO_CHANNEL(tioc);
790     qemu_thread_create(&p->thread, "multifd-tls-handshake-worker",
791                        multifd_tls_handshake_thread, p,
792                        QEMU_THREAD_JOINABLE);
793 }
794 
795 static bool multifd_channel_connect(MultiFDSendParams *p,
796                                     QIOChannel *ioc,
797                                     Error *error)
798 {
799     MigrationState *s = migrate_get_current();
800 
801     trace_multifd_set_outgoing_channel(
802         ioc, object_get_typename(OBJECT(ioc)), p->tls_hostname, error);
803 
804     if (!error) {
805         if (s->parameters.tls_creds &&
806             *s->parameters.tls_creds &&
807             !object_dynamic_cast(OBJECT(ioc),
808                                  TYPE_QIO_CHANNEL_TLS)) {
809             multifd_tls_channel_connect(p, ioc, &error);
810             if (!error) {
811                 /*
812                  * tls_channel_connect will call back to this
813                  * function after the TLS handshake,
814                  * so we mustn't call multifd_send_thread until then
815                  */
816                 return true;
817             } else {
818                 return false;
819             }
820         } else {
821             migration_ioc_register_yank(ioc);
822             p->registered_yank = true;
823             p->c = ioc;
824             qemu_thread_create(&p->thread, p->name, multifd_send_thread, p,
825                                    QEMU_THREAD_JOINABLE);
826        }
827        return true;
828     }
829 
830     return false;
831 }
832 
833 static void multifd_new_send_channel_cleanup(MultiFDSendParams *p,
834                                              QIOChannel *ioc, Error *err)
835 {
836      migrate_set_error(migrate_get_current(), err);
837      /* Error happen, we need to tell who pay attention to me */
838      qemu_sem_post(&multifd_send_state->channels_ready);
839      qemu_sem_post(&p->sem_sync);
840      /*
841       * Although multifd_send_thread is not created, but main migration
842       * thread neet to judge whether it is running, so we need to mark
843       * its status.
844       */
845      p->quit = true;
846      object_unref(OBJECT(ioc));
847      error_free(err);
848 }
849 
850 static void multifd_new_send_channel_async(QIOTask *task, gpointer opaque)
851 {
852     MultiFDSendParams *p = opaque;
853     QIOChannel *sioc = QIO_CHANNEL(qio_task_get_source(task));
854     Error *local_err = NULL;
855 
856     trace_multifd_new_send_channel_async(p->id);
857     if (qio_task_propagate_error(task, &local_err)) {
858         goto cleanup;
859     } else {
860         p->c = QIO_CHANNEL(sioc);
861         qio_channel_set_delay(p->c, false);
862         p->running = true;
863         if (!multifd_channel_connect(p, sioc, local_err)) {
864             goto cleanup;
865         }
866         return;
867     }
868 
869 cleanup:
870     multifd_new_send_channel_cleanup(p, sioc, local_err);
871 }
872 
873 static bool migrate_allow_multifd = true;
874 void migrate_protocol_allow_multifd(bool allow)
875 {
876     migrate_allow_multifd = allow;
877 }
878 
879 bool migrate_multifd_is_allowed(void)
880 {
881     return migrate_allow_multifd;
882 }
883 
884 int multifd_save_setup(Error **errp)
885 {
886     int thread_count;
887     uint32_t page_count = MULTIFD_PACKET_SIZE / qemu_target_page_size();
888     uint8_t i;
889     MigrationState *s;
890 
891     if (!migrate_use_multifd()) {
892         return 0;
893     }
894     if (!migrate_multifd_is_allowed()) {
895         error_setg(errp, "multifd is not supported by current protocol");
896         return -1;
897     }
898 
899     s = migrate_get_current();
900     thread_count = migrate_multifd_channels();
901     multifd_send_state = g_malloc0(sizeof(*multifd_send_state));
902     multifd_send_state->params = g_new0(MultiFDSendParams, thread_count);
903     multifd_send_state->pages = multifd_pages_init(page_count);
904     qemu_sem_init(&multifd_send_state->channels_ready, 0);
905     qatomic_set(&multifd_send_state->exiting, 0);
906     multifd_send_state->ops = multifd_ops[migrate_multifd_compression()];
907 
908     for (i = 0; i < thread_count; i++) {
909         MultiFDSendParams *p = &multifd_send_state->params[i];
910 
911         qemu_mutex_init(&p->mutex);
912         qemu_sem_init(&p->sem, 0);
913         qemu_sem_init(&p->sem_sync, 0);
914         p->quit = false;
915         p->pending_job = 0;
916         p->id = i;
917         p->pages = multifd_pages_init(page_count);
918         p->packet_len = sizeof(MultiFDPacket_t)
919                       + sizeof(uint64_t) * page_count;
920         p->packet = g_malloc0(p->packet_len);
921         p->packet->magic = cpu_to_be32(MULTIFD_MAGIC);
922         p->packet->version = cpu_to_be32(MULTIFD_VERSION);
923         p->name = g_strdup_printf("multifdsend_%d", i);
924         p->tls_hostname = g_strdup(s->hostname);
925         socket_send_channel_create(multifd_new_send_channel_async, p);
926     }
927 
928     for (i = 0; i < thread_count; i++) {
929         MultiFDSendParams *p = &multifd_send_state->params[i];
930         Error *local_err = NULL;
931         int ret;
932 
933         ret = multifd_send_state->ops->send_setup(p, &local_err);
934         if (ret) {
935             error_propagate(errp, local_err);
936             return ret;
937         }
938     }
939     return 0;
940 }
941 
942 struct {
943     MultiFDRecvParams *params;
944     /* number of created threads */
945     int count;
946     /* syncs main thread and channels */
947     QemuSemaphore sem_sync;
948     /* global number of generated multifd packets */
949     uint64_t packet_num;
950     /* multifd ops */
951     MultiFDMethods *ops;
952 } *multifd_recv_state;
953 
954 static void multifd_recv_terminate_threads(Error *err)
955 {
956     int i;
957 
958     trace_multifd_recv_terminate_threads(err != NULL);
959 
960     if (err) {
961         MigrationState *s = migrate_get_current();
962         migrate_set_error(s, err);
963         if (s->state == MIGRATION_STATUS_SETUP ||
964             s->state == MIGRATION_STATUS_ACTIVE) {
965             migrate_set_state(&s->state, s->state,
966                               MIGRATION_STATUS_FAILED);
967         }
968     }
969 
970     for (i = 0; i < migrate_multifd_channels(); i++) {
971         MultiFDRecvParams *p = &multifd_recv_state->params[i];
972 
973         qemu_mutex_lock(&p->mutex);
974         p->quit = true;
975         /*
976          * We could arrive here for two reasons:
977          *  - normal quit, i.e. everything went fine, just finished
978          *  - error quit: We close the channels so the channel threads
979          *    finish the qio_channel_read_all_eof()
980          */
981         if (p->c) {
982             qio_channel_shutdown(p->c, QIO_CHANNEL_SHUTDOWN_BOTH, NULL);
983         }
984         qemu_mutex_unlock(&p->mutex);
985     }
986 }
987 
988 int multifd_load_cleanup(Error **errp)
989 {
990     int i;
991 
992     if (!migrate_use_multifd() || !migrate_multifd_is_allowed()) {
993         return 0;
994     }
995     multifd_recv_terminate_threads(NULL);
996     for (i = 0; i < migrate_multifd_channels(); i++) {
997         MultiFDRecvParams *p = &multifd_recv_state->params[i];
998 
999         if (p->running) {
1000             p->quit = true;
1001             /*
1002              * multifd_recv_thread may hung at MULTIFD_FLAG_SYNC handle code,
1003              * however try to wakeup it without harm in cleanup phase.
1004              */
1005             qemu_sem_post(&p->sem_sync);
1006             qemu_thread_join(&p->thread);
1007         }
1008     }
1009     for (i = 0; i < migrate_multifd_channels(); i++) {
1010         MultiFDRecvParams *p = &multifd_recv_state->params[i];
1011 
1012         migration_ioc_unregister_yank(p->c);
1013         object_unref(OBJECT(p->c));
1014         p->c = NULL;
1015         qemu_mutex_destroy(&p->mutex);
1016         qemu_sem_destroy(&p->sem_sync);
1017         g_free(p->name);
1018         p->name = NULL;
1019         multifd_pages_clear(p->pages);
1020         p->pages = NULL;
1021         p->packet_len = 0;
1022         g_free(p->packet);
1023         p->packet = NULL;
1024         multifd_recv_state->ops->recv_cleanup(p);
1025     }
1026     qemu_sem_destroy(&multifd_recv_state->sem_sync);
1027     g_free(multifd_recv_state->params);
1028     multifd_recv_state->params = NULL;
1029     g_free(multifd_recv_state);
1030     multifd_recv_state = NULL;
1031 
1032     return 0;
1033 }
1034 
1035 void multifd_recv_sync_main(void)
1036 {
1037     int i;
1038 
1039     if (!migrate_use_multifd()) {
1040         return;
1041     }
1042     for (i = 0; i < migrate_multifd_channels(); i++) {
1043         MultiFDRecvParams *p = &multifd_recv_state->params[i];
1044 
1045         trace_multifd_recv_sync_main_wait(p->id);
1046         qemu_sem_wait(&multifd_recv_state->sem_sync);
1047     }
1048     for (i = 0; i < migrate_multifd_channels(); i++) {
1049         MultiFDRecvParams *p = &multifd_recv_state->params[i];
1050 
1051         WITH_QEMU_LOCK_GUARD(&p->mutex) {
1052             if (multifd_recv_state->packet_num < p->packet_num) {
1053                 multifd_recv_state->packet_num = p->packet_num;
1054             }
1055         }
1056         trace_multifd_recv_sync_main_signal(p->id);
1057         qemu_sem_post(&p->sem_sync);
1058     }
1059     trace_multifd_recv_sync_main(multifd_recv_state->packet_num);
1060 }
1061 
1062 static void *multifd_recv_thread(void *opaque)
1063 {
1064     MultiFDRecvParams *p = opaque;
1065     Error *local_err = NULL;
1066     int ret;
1067 
1068     trace_multifd_recv_thread_start(p->id);
1069     rcu_register_thread();
1070 
1071     while (true) {
1072         uint32_t used;
1073         uint32_t flags;
1074 
1075         if (p->quit) {
1076             break;
1077         }
1078 
1079         ret = qio_channel_read_all_eof(p->c, (void *)p->packet,
1080                                        p->packet_len, &local_err);
1081         if (ret == 0) {   /* EOF */
1082             break;
1083         }
1084         if (ret == -1) {   /* Error */
1085             break;
1086         }
1087 
1088         qemu_mutex_lock(&p->mutex);
1089         ret = multifd_recv_unfill_packet(p, &local_err);
1090         if (ret) {
1091             qemu_mutex_unlock(&p->mutex);
1092             break;
1093         }
1094 
1095         used = p->pages->num;
1096         flags = p->flags;
1097         /* recv methods don't know how to handle the SYNC flag */
1098         p->flags &= ~MULTIFD_FLAG_SYNC;
1099         trace_multifd_recv(p->id, p->packet_num, used, flags,
1100                            p->next_packet_size);
1101         p->num_packets++;
1102         p->num_pages += used;
1103         qemu_mutex_unlock(&p->mutex);
1104 
1105         if (used) {
1106             ret = multifd_recv_state->ops->recv_pages(p, &local_err);
1107             if (ret != 0) {
1108                 break;
1109             }
1110         }
1111 
1112         if (flags & MULTIFD_FLAG_SYNC) {
1113             qemu_sem_post(&multifd_recv_state->sem_sync);
1114             qemu_sem_wait(&p->sem_sync);
1115         }
1116     }
1117 
1118     if (local_err) {
1119         multifd_recv_terminate_threads(local_err);
1120         error_free(local_err);
1121     }
1122     qemu_mutex_lock(&p->mutex);
1123     p->running = false;
1124     qemu_mutex_unlock(&p->mutex);
1125 
1126     rcu_unregister_thread();
1127     trace_multifd_recv_thread_end(p->id, p->num_packets, p->num_pages);
1128 
1129     return NULL;
1130 }
1131 
1132 int multifd_load_setup(Error **errp)
1133 {
1134     int thread_count;
1135     uint32_t page_count = MULTIFD_PACKET_SIZE / qemu_target_page_size();
1136     uint8_t i;
1137 
1138     if (!migrate_use_multifd()) {
1139         return 0;
1140     }
1141     if (!migrate_multifd_is_allowed()) {
1142         error_setg(errp, "multifd is not supported by current protocol");
1143         return -1;
1144     }
1145     thread_count = migrate_multifd_channels();
1146     multifd_recv_state = g_malloc0(sizeof(*multifd_recv_state));
1147     multifd_recv_state->params = g_new0(MultiFDRecvParams, thread_count);
1148     qatomic_set(&multifd_recv_state->count, 0);
1149     qemu_sem_init(&multifd_recv_state->sem_sync, 0);
1150     multifd_recv_state->ops = multifd_ops[migrate_multifd_compression()];
1151 
1152     for (i = 0; i < thread_count; i++) {
1153         MultiFDRecvParams *p = &multifd_recv_state->params[i];
1154 
1155         qemu_mutex_init(&p->mutex);
1156         qemu_sem_init(&p->sem_sync, 0);
1157         p->quit = false;
1158         p->id = i;
1159         p->pages = multifd_pages_init(page_count);
1160         p->packet_len = sizeof(MultiFDPacket_t)
1161                       + sizeof(uint64_t) * page_count;
1162         p->packet = g_malloc0(p->packet_len);
1163         p->name = g_strdup_printf("multifdrecv_%d", i);
1164     }
1165 
1166     for (i = 0; i < thread_count; i++) {
1167         MultiFDRecvParams *p = &multifd_recv_state->params[i];
1168         Error *local_err = NULL;
1169         int ret;
1170 
1171         ret = multifd_recv_state->ops->recv_setup(p, &local_err);
1172         if (ret) {
1173             error_propagate(errp, local_err);
1174             return ret;
1175         }
1176     }
1177     return 0;
1178 }
1179 
1180 bool multifd_recv_all_channels_created(void)
1181 {
1182     int thread_count = migrate_multifd_channels();
1183 
1184     if (!migrate_use_multifd()) {
1185         return true;
1186     }
1187 
1188     if (!multifd_recv_state) {
1189         /* Called before any connections created */
1190         return false;
1191     }
1192 
1193     return thread_count == qatomic_read(&multifd_recv_state->count);
1194 }
1195 
1196 /*
1197  * Try to receive all multifd channels to get ready for the migration.
1198  * - Return true and do not set @errp when correctly receiving all channels;
1199  * - Return false and do not set @errp when correctly receiving the current one;
1200  * - Return false and set @errp when failing to receive the current channel.
1201  */
1202 bool multifd_recv_new_channel(QIOChannel *ioc, Error **errp)
1203 {
1204     MultiFDRecvParams *p;
1205     Error *local_err = NULL;
1206     int id;
1207 
1208     id = multifd_recv_initial_packet(ioc, &local_err);
1209     if (id < 0) {
1210         multifd_recv_terminate_threads(local_err);
1211         error_propagate_prepend(errp, local_err,
1212                                 "failed to receive packet"
1213                                 " via multifd channel %d: ",
1214                                 qatomic_read(&multifd_recv_state->count));
1215         return false;
1216     }
1217     trace_multifd_recv_new_channel(id);
1218 
1219     p = &multifd_recv_state->params[id];
1220     if (p->c != NULL) {
1221         error_setg(&local_err, "multifd: received id '%d' already setup'",
1222                    id);
1223         multifd_recv_terminate_threads(local_err);
1224         error_propagate(errp, local_err);
1225         return false;
1226     }
1227     p->c = ioc;
1228     object_ref(OBJECT(ioc));
1229     /* initial packet */
1230     p->num_packets = 1;
1231 
1232     p->running = true;
1233     qemu_thread_create(&p->thread, p->name, multifd_recv_thread, p,
1234                        QEMU_THREAD_JOINABLE);
1235     qatomic_inc(&multifd_recv_state->count);
1236     return qatomic_read(&multifd_recv_state->count) ==
1237            migrate_multifd_channels();
1238 }
1239