xref: /qemu/migration/multifd.c (revision 61dec060)
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 "fd.h"
21 #include "file.h"
22 #include "migration.h"
23 #include "migration-stats.h"
24 #include "socket.h"
25 #include "tls.h"
26 #include "qemu-file.h"
27 #include "trace.h"
28 #include "multifd.h"
29 #include "threadinfo.h"
30 #include "options.h"
31 #include "qemu/yank.h"
32 #include "io/channel-file.h"
33 #include "io/channel-socket.h"
34 #include "yank_functions.h"
35 
36 /* Multiple fd's */
37 
38 #define MULTIFD_MAGIC 0x11223344U
39 #define MULTIFD_VERSION 1
40 
41 typedef struct {
42     uint32_t magic;
43     uint32_t version;
44     unsigned char uuid[16]; /* QemuUUID */
45     uint8_t id;
46     uint8_t unused1[7];     /* Reserved for future use */
47     uint64_t unused2[4];    /* Reserved for future use */
48 } __attribute__((packed)) MultiFDInit_t;
49 
50 struct {
51     MultiFDSendParams *params;
52     /* array of pages to sent */
53     MultiFDPages_t *pages;
54     /*
55      * Global number of generated multifd packets.
56      *
57      * Note that we used 'uintptr_t' because it'll naturally support atomic
58      * operations on both 32bit / 64 bits hosts.  It means on 32bit systems
59      * multifd will overflow the packet_num easier, but that should be
60      * fine.
61      *
62      * Another option is to use QEMU's Stat64 then it'll be 64 bits on all
63      * hosts, however so far it does not support atomic fetch_add() yet.
64      * Make it easy for now.
65      */
66     uintptr_t packet_num;
67     /*
68      * Synchronization point past which no more channels will be
69      * created.
70      */
71     QemuSemaphore channels_created;
72     /* send channels ready */
73     QemuSemaphore channels_ready;
74     /*
75      * Have we already run terminate threads.  There is a race when it
76      * happens that we got one error while we are exiting.
77      * We will use atomic operations.  Only valid values are 0 and 1.
78      */
79     int exiting;
80     /* multifd ops */
81     MultiFDMethods *ops;
82 } *multifd_send_state;
83 
84 struct {
85     MultiFDRecvParams *params;
86     MultiFDRecvData *data;
87     /* number of created threads */
88     int count;
89     /*
90      * This is always posted by the recv threads, the migration thread
91      * uses it to wait for recv threads to finish assigned tasks.
92      */
93     QemuSemaphore sem_sync;
94     /* global number of generated multifd packets */
95     uint64_t packet_num;
96     int exiting;
97     /* multifd ops */
98     MultiFDMethods *ops;
99 } *multifd_recv_state;
100 
101 static bool multifd_use_packets(void)
102 {
103     return !migrate_mapped_ram();
104 }
105 
106 void multifd_send_channel_created(void)
107 {
108     qemu_sem_post(&multifd_send_state->channels_created);
109 }
110 
111 static void multifd_set_file_bitmap(MultiFDSendParams *p)
112 {
113     MultiFDPages_t *pages = p->pages;
114 
115     assert(pages->block);
116 
117     for (int i = 0; i < p->pages->num; i++) {
118         ramblock_set_file_bmap_atomic(pages->block, pages->offset[i]);
119     }
120 }
121 
122 /* Multifd without compression */
123 
124 /**
125  * nocomp_send_setup: setup send side
126  *
127  * @p: Params for the channel that we are using
128  * @errp: pointer to an error
129  */
130 static int nocomp_send_setup(MultiFDSendParams *p, Error **errp)
131 {
132     if (migrate_zero_copy_send()) {
133         p->write_flags |= QIO_CHANNEL_WRITE_FLAG_ZERO_COPY;
134     }
135 
136     return 0;
137 }
138 
139 /**
140  * nocomp_send_cleanup: cleanup send side
141  *
142  * For no compression this function does nothing.
143  *
144  * @p: Params for the channel that we are using
145  * @errp: pointer to an error
146  */
147 static void nocomp_send_cleanup(MultiFDSendParams *p, Error **errp)
148 {
149     return;
150 }
151 
152 static void multifd_send_prepare_iovs(MultiFDSendParams *p)
153 {
154     MultiFDPages_t *pages = p->pages;
155 
156     for (int i = 0; i < pages->num; i++) {
157         p->iov[p->iovs_num].iov_base = pages->block->host + pages->offset[i];
158         p->iov[p->iovs_num].iov_len = p->page_size;
159         p->iovs_num++;
160     }
161 
162     p->next_packet_size = pages->num * p->page_size;
163 }
164 
165 /**
166  * nocomp_send_prepare: prepare date to be able to send
167  *
168  * For no compression we just have to calculate the size of the
169  * packet.
170  *
171  * Returns 0 for success or -1 for error
172  *
173  * @p: Params for the channel that we are using
174  * @errp: pointer to an error
175  */
176 static int nocomp_send_prepare(MultiFDSendParams *p, Error **errp)
177 {
178     bool use_zero_copy_send = migrate_zero_copy_send();
179     int ret;
180 
181     if (!multifd_use_packets()) {
182         multifd_send_prepare_iovs(p);
183         multifd_set_file_bitmap(p);
184 
185         return 0;
186     }
187 
188     if (!use_zero_copy_send) {
189         /*
190          * Only !zerocopy needs the header in IOV; zerocopy will
191          * send it separately.
192          */
193         multifd_send_prepare_header(p);
194     }
195 
196     multifd_send_prepare_iovs(p);
197     p->flags |= MULTIFD_FLAG_NOCOMP;
198 
199     multifd_send_fill_packet(p);
200 
201     if (use_zero_copy_send) {
202         /* Send header first, without zerocopy */
203         ret = qio_channel_write_all(p->c, (void *)p->packet,
204                                     p->packet_len, errp);
205         if (ret != 0) {
206             return -1;
207         }
208     }
209 
210     return 0;
211 }
212 
213 /**
214  * nocomp_recv_setup: setup receive side
215  *
216  * For no compression this function does nothing.
217  *
218  * Returns 0 for success or -1 for error
219  *
220  * @p: Params for the channel that we are using
221  * @errp: pointer to an error
222  */
223 static int nocomp_recv_setup(MultiFDRecvParams *p, Error **errp)
224 {
225     return 0;
226 }
227 
228 /**
229  * nocomp_recv_cleanup: setup receive side
230  *
231  * For no compression this function does nothing.
232  *
233  * @p: Params for the channel that we are using
234  */
235 static void nocomp_recv_cleanup(MultiFDRecvParams *p)
236 {
237 }
238 
239 /**
240  * nocomp_recv: read the data from the channel
241  *
242  * For no compression we just need to read things into the correct place.
243  *
244  * Returns 0 for success or -1 for error
245  *
246  * @p: Params for the channel that we are using
247  * @errp: pointer to an error
248  */
249 static int nocomp_recv(MultiFDRecvParams *p, Error **errp)
250 {
251     uint32_t flags;
252 
253     if (!multifd_use_packets()) {
254         return multifd_file_recv_data(p, errp);
255     }
256 
257     flags = p->flags & MULTIFD_FLAG_COMPRESSION_MASK;
258 
259     if (flags != MULTIFD_FLAG_NOCOMP) {
260         error_setg(errp, "multifd %u: flags received %x flags expected %x",
261                    p->id, flags, MULTIFD_FLAG_NOCOMP);
262         return -1;
263     }
264     for (int i = 0; i < p->normal_num; i++) {
265         p->iov[i].iov_base = p->host + p->normal[i];
266         p->iov[i].iov_len = p->page_size;
267     }
268     return qio_channel_readv_all(p->c, p->iov, p->normal_num, errp);
269 }
270 
271 static MultiFDMethods multifd_nocomp_ops = {
272     .send_setup = nocomp_send_setup,
273     .send_cleanup = nocomp_send_cleanup,
274     .send_prepare = nocomp_send_prepare,
275     .recv_setup = nocomp_recv_setup,
276     .recv_cleanup = nocomp_recv_cleanup,
277     .recv = nocomp_recv
278 };
279 
280 static MultiFDMethods *multifd_ops[MULTIFD_COMPRESSION__MAX] = {
281     [MULTIFD_COMPRESSION_NONE] = &multifd_nocomp_ops,
282 };
283 
284 void multifd_register_ops(int method, MultiFDMethods *ops)
285 {
286     assert(0 < method && method < MULTIFD_COMPRESSION__MAX);
287     multifd_ops[method] = ops;
288 }
289 
290 /* Reset a MultiFDPages_t* object for the next use */
291 static void multifd_pages_reset(MultiFDPages_t *pages)
292 {
293     /*
294      * We don't need to touch offset[] array, because it will be
295      * overwritten later when reused.
296      */
297     pages->num = 0;
298     pages->block = NULL;
299 }
300 
301 static int multifd_send_initial_packet(MultiFDSendParams *p, Error **errp)
302 {
303     MultiFDInit_t msg = {};
304     size_t size = sizeof(msg);
305     int ret;
306 
307     msg.magic = cpu_to_be32(MULTIFD_MAGIC);
308     msg.version = cpu_to_be32(MULTIFD_VERSION);
309     msg.id = p->id;
310     memcpy(msg.uuid, &qemu_uuid.data, sizeof(msg.uuid));
311 
312     ret = qio_channel_write_all(p->c, (char *)&msg, size, errp);
313     if (ret != 0) {
314         return -1;
315     }
316     stat64_add(&mig_stats.multifd_bytes, size);
317     return 0;
318 }
319 
320 static int multifd_recv_initial_packet(QIOChannel *c, Error **errp)
321 {
322     MultiFDInit_t msg;
323     int ret;
324 
325     ret = qio_channel_read_all(c, (char *)&msg, sizeof(msg), errp);
326     if (ret != 0) {
327         return -1;
328     }
329 
330     msg.magic = be32_to_cpu(msg.magic);
331     msg.version = be32_to_cpu(msg.version);
332 
333     if (msg.magic != MULTIFD_MAGIC) {
334         error_setg(errp, "multifd: received packet magic %x "
335                    "expected %x", msg.magic, MULTIFD_MAGIC);
336         return -1;
337     }
338 
339     if (msg.version != MULTIFD_VERSION) {
340         error_setg(errp, "multifd: received packet version %u "
341                    "expected %u", msg.version, MULTIFD_VERSION);
342         return -1;
343     }
344 
345     if (memcmp(msg.uuid, &qemu_uuid, sizeof(qemu_uuid))) {
346         char *uuid = qemu_uuid_unparse_strdup(&qemu_uuid);
347         char *msg_uuid = qemu_uuid_unparse_strdup((const QemuUUID *)msg.uuid);
348 
349         error_setg(errp, "multifd: received uuid '%s' and expected "
350                    "uuid '%s' for channel %hhd", msg_uuid, uuid, msg.id);
351         g_free(uuid);
352         g_free(msg_uuid);
353         return -1;
354     }
355 
356     if (msg.id > migrate_multifd_channels()) {
357         error_setg(errp, "multifd: received channel id %u is greater than "
358                    "number of channels %u", msg.id, migrate_multifd_channels());
359         return -1;
360     }
361 
362     return msg.id;
363 }
364 
365 static MultiFDPages_t *multifd_pages_init(uint32_t n)
366 {
367     MultiFDPages_t *pages = g_new0(MultiFDPages_t, 1);
368 
369     pages->allocated = n;
370     pages->offset = g_new0(ram_addr_t, n);
371 
372     return pages;
373 }
374 
375 static void multifd_pages_clear(MultiFDPages_t *pages)
376 {
377     multifd_pages_reset(pages);
378     pages->allocated = 0;
379     g_free(pages->offset);
380     pages->offset = NULL;
381     g_free(pages);
382 }
383 
384 void multifd_send_fill_packet(MultiFDSendParams *p)
385 {
386     MultiFDPacket_t *packet = p->packet;
387     MultiFDPages_t *pages = p->pages;
388     uint64_t packet_num;
389     int i;
390 
391     packet->flags = cpu_to_be32(p->flags);
392     packet->pages_alloc = cpu_to_be32(p->pages->allocated);
393     packet->normal_pages = cpu_to_be32(pages->num);
394     packet->next_packet_size = cpu_to_be32(p->next_packet_size);
395 
396     packet_num = qatomic_fetch_inc(&multifd_send_state->packet_num);
397     packet->packet_num = cpu_to_be64(packet_num);
398 
399     if (pages->block) {
400         strncpy(packet->ramblock, pages->block->idstr, 256);
401     }
402 
403     for (i = 0; i < pages->num; i++) {
404         /* there are architectures where ram_addr_t is 32 bit */
405         uint64_t temp = pages->offset[i];
406 
407         packet->offset[i] = cpu_to_be64(temp);
408     }
409 
410     p->packets_sent++;
411     p->total_normal_pages += pages->num;
412 
413     trace_multifd_send(p->id, packet_num, pages->num, p->flags,
414                        p->next_packet_size);
415 }
416 
417 static int multifd_recv_unfill_packet(MultiFDRecvParams *p, Error **errp)
418 {
419     MultiFDPacket_t *packet = p->packet;
420     int i;
421 
422     packet->magic = be32_to_cpu(packet->magic);
423     if (packet->magic != MULTIFD_MAGIC) {
424         error_setg(errp, "multifd: received packet "
425                    "magic %x and expected magic %x",
426                    packet->magic, MULTIFD_MAGIC);
427         return -1;
428     }
429 
430     packet->version = be32_to_cpu(packet->version);
431     if (packet->version != MULTIFD_VERSION) {
432         error_setg(errp, "multifd: received packet "
433                    "version %u and expected version %u",
434                    packet->version, MULTIFD_VERSION);
435         return -1;
436     }
437 
438     p->flags = be32_to_cpu(packet->flags);
439 
440     packet->pages_alloc = be32_to_cpu(packet->pages_alloc);
441     /*
442      * If we received a packet that is 100 times bigger than expected
443      * just stop migration.  It is a magic number.
444      */
445     if (packet->pages_alloc > p->page_count) {
446         error_setg(errp, "multifd: received packet "
447                    "with size %u and expected a size of %u",
448                    packet->pages_alloc, p->page_count) ;
449         return -1;
450     }
451 
452     p->normal_num = be32_to_cpu(packet->normal_pages);
453     if (p->normal_num > packet->pages_alloc) {
454         error_setg(errp, "multifd: received packet "
455                    "with %u pages and expected maximum pages are %u",
456                    p->normal_num, packet->pages_alloc) ;
457         return -1;
458     }
459 
460     p->next_packet_size = be32_to_cpu(packet->next_packet_size);
461     p->packet_num = be64_to_cpu(packet->packet_num);
462     p->packets_recved++;
463     p->total_normal_pages += p->normal_num;
464 
465     trace_multifd_recv(p->id, p->packet_num, p->normal_num, p->flags,
466                        p->next_packet_size);
467 
468     if (p->normal_num == 0) {
469         return 0;
470     }
471 
472     /* make sure that ramblock is 0 terminated */
473     packet->ramblock[255] = 0;
474     p->block = qemu_ram_block_by_name(packet->ramblock);
475     if (!p->block) {
476         error_setg(errp, "multifd: unknown ram block %s",
477                    packet->ramblock);
478         return -1;
479     }
480 
481     p->host = p->block->host;
482     for (i = 0; i < p->normal_num; i++) {
483         uint64_t offset = be64_to_cpu(packet->offset[i]);
484 
485         if (offset > (p->block->used_length - p->page_size)) {
486             error_setg(errp, "multifd: offset too long %" PRIu64
487                        " (max " RAM_ADDR_FMT ")",
488                        offset, p->block->used_length);
489             return -1;
490         }
491         p->normal[i] = offset;
492     }
493 
494     return 0;
495 }
496 
497 static bool multifd_send_should_exit(void)
498 {
499     return qatomic_read(&multifd_send_state->exiting);
500 }
501 
502 static bool multifd_recv_should_exit(void)
503 {
504     return qatomic_read(&multifd_recv_state->exiting);
505 }
506 
507 /*
508  * The migration thread can wait on either of the two semaphores.  This
509  * function can be used to kick the main thread out of waiting on either of
510  * them.  Should mostly only be called when something wrong happened with
511  * the current multifd send thread.
512  */
513 static void multifd_send_kick_main(MultiFDSendParams *p)
514 {
515     qemu_sem_post(&p->sem_sync);
516     qemu_sem_post(&multifd_send_state->channels_ready);
517 }
518 
519 /*
520  * How we use multifd_send_state->pages and channel->pages?
521  *
522  * We create a pages for each channel, and a main one.  Each time that
523  * we need to send a batch of pages we interchange the ones between
524  * multifd_send_state and the channel that is sending it.  There are
525  * two reasons for that:
526  *    - to not have to do so many mallocs during migration
527  *    - to make easier to know what to free at the end of migration
528  *
529  * This way we always know who is the owner of each "pages" struct,
530  * and we don't need any locking.  It belongs to the migration thread
531  * or to the channel thread.  Switching is safe because the migration
532  * thread is using the channel mutex when changing it, and the channel
533  * have to had finish with its own, otherwise pending_job can't be
534  * false.
535  *
536  * Returns true if succeed, false otherwise.
537  */
538 static bool multifd_send_pages(void)
539 {
540     int i;
541     static int next_channel;
542     MultiFDSendParams *p = NULL; /* make happy gcc */
543     MultiFDPages_t *pages = multifd_send_state->pages;
544 
545     if (multifd_send_should_exit()) {
546         return false;
547     }
548 
549     /* We wait here, until at least one channel is ready */
550     qemu_sem_wait(&multifd_send_state->channels_ready);
551 
552     /*
553      * next_channel can remain from a previous migration that was
554      * using more channels, so ensure it doesn't overflow if the
555      * limit is lower now.
556      */
557     next_channel %= migrate_multifd_channels();
558     for (i = next_channel;; i = (i + 1) % migrate_multifd_channels()) {
559         if (multifd_send_should_exit()) {
560             return false;
561         }
562         p = &multifd_send_state->params[i];
563         /*
564          * Lockless read to p->pending_job is safe, because only multifd
565          * sender thread can clear it.
566          */
567         if (qatomic_read(&p->pending_job) == false) {
568             next_channel = (i + 1) % migrate_multifd_channels();
569             break;
570         }
571     }
572 
573     /*
574      * Make sure we read p->pending_job before all the rest.  Pairs with
575      * qatomic_store_release() in multifd_send_thread().
576      */
577     smp_mb_acquire();
578     assert(!p->pages->num);
579     multifd_send_state->pages = p->pages;
580     p->pages = pages;
581     /*
582      * Making sure p->pages is setup before marking pending_job=true. Pairs
583      * with the qatomic_load_acquire() in multifd_send_thread().
584      */
585     qatomic_store_release(&p->pending_job, true);
586     qemu_sem_post(&p->sem);
587 
588     return true;
589 }
590 
591 static inline bool multifd_queue_empty(MultiFDPages_t *pages)
592 {
593     return pages->num == 0;
594 }
595 
596 static inline bool multifd_queue_full(MultiFDPages_t *pages)
597 {
598     return pages->num == pages->allocated;
599 }
600 
601 static inline void multifd_enqueue(MultiFDPages_t *pages, ram_addr_t offset)
602 {
603     pages->offset[pages->num++] = offset;
604 }
605 
606 /* Returns true if enqueue successful, false otherwise */
607 bool multifd_queue_page(RAMBlock *block, ram_addr_t offset)
608 {
609     MultiFDPages_t *pages;
610 
611 retry:
612     pages = multifd_send_state->pages;
613 
614     /* If the queue is empty, we can already enqueue now */
615     if (multifd_queue_empty(pages)) {
616         pages->block = block;
617         multifd_enqueue(pages, offset);
618         return true;
619     }
620 
621     /*
622      * Not empty, meanwhile we need a flush.  It can because of either:
623      *
624      * (1) The page is not on the same ramblock of previous ones, or,
625      * (2) The queue is full.
626      *
627      * After flush, always retry.
628      */
629     if (pages->block != block || multifd_queue_full(pages)) {
630         if (!multifd_send_pages()) {
631             return false;
632         }
633         goto retry;
634     }
635 
636     /* Not empty, and we still have space, do it! */
637     multifd_enqueue(pages, offset);
638     return true;
639 }
640 
641 /* Multifd send side hit an error; remember it and prepare to quit */
642 static void multifd_send_set_error(Error *err)
643 {
644     /*
645      * We don't want to exit each threads twice.  Depending on where
646      * we get the error, or if there are two independent errors in two
647      * threads at the same time, we can end calling this function
648      * twice.
649      */
650     if (qatomic_xchg(&multifd_send_state->exiting, 1)) {
651         return;
652     }
653 
654     if (err) {
655         MigrationState *s = migrate_get_current();
656         migrate_set_error(s, err);
657         if (s->state == MIGRATION_STATUS_SETUP ||
658             s->state == MIGRATION_STATUS_PRE_SWITCHOVER ||
659             s->state == MIGRATION_STATUS_DEVICE ||
660             s->state == MIGRATION_STATUS_ACTIVE) {
661             migrate_set_state(&s->state, s->state,
662                               MIGRATION_STATUS_FAILED);
663         }
664     }
665 }
666 
667 static void multifd_send_terminate_threads(void)
668 {
669     int i;
670 
671     trace_multifd_send_terminate_threads();
672 
673     /*
674      * Tell everyone we're quitting.  No xchg() needed here; we simply
675      * always set it.
676      */
677     qatomic_set(&multifd_send_state->exiting, 1);
678 
679     /*
680      * Firstly, kick all threads out; no matter whether they are just idle,
681      * or blocked in an IO system call.
682      */
683     for (i = 0; i < migrate_multifd_channels(); i++) {
684         MultiFDSendParams *p = &multifd_send_state->params[i];
685 
686         qemu_sem_post(&p->sem);
687         if (p->c) {
688             qio_channel_shutdown(p->c, QIO_CHANNEL_SHUTDOWN_BOTH, NULL);
689         }
690     }
691 
692     /*
693      * Finally recycle all the threads.
694      */
695     for (i = 0; i < migrate_multifd_channels(); i++) {
696         MultiFDSendParams *p = &multifd_send_state->params[i];
697 
698         if (p->tls_thread_created) {
699             qemu_thread_join(&p->tls_thread);
700         }
701 
702         if (p->thread_created) {
703             qemu_thread_join(&p->thread);
704         }
705     }
706 }
707 
708 static bool multifd_send_cleanup_channel(MultiFDSendParams *p, Error **errp)
709 {
710     if (p->c) {
711         migration_ioc_unregister_yank(p->c);
712         /*
713          * The object_unref() cannot guarantee the fd will always be
714          * released because finalize() of the iochannel is only
715          * triggered on the last reference and it's not guaranteed
716          * that we always hold the last refcount when reaching here.
717          *
718          * Closing the fd explicitly has the benefit that if there is any
719          * registered I/O handler callbacks on such fd, that will get a
720          * POLLNVAL event and will further trigger the cleanup to finally
721          * release the IOC.
722          *
723          * FIXME: It should logically be guaranteed that all multifd
724          * channels have no I/O handler callback registered when reaching
725          * here, because migration thread will wait for all multifd channel
726          * establishments to complete during setup.  Since
727          * migrate_fd_cleanup() will be scheduled in main thread too, all
728          * previous callbacks should guarantee to be completed when
729          * reaching here.  See multifd_send_state.channels_created and its
730          * usage.  In the future, we could replace this with an assert
731          * making sure we're the last reference, or simply drop it if above
732          * is more clear to be justified.
733          */
734         qio_channel_close(p->c, &error_abort);
735         object_unref(OBJECT(p->c));
736         p->c = NULL;
737     }
738     qemu_sem_destroy(&p->sem);
739     qemu_sem_destroy(&p->sem_sync);
740     g_free(p->name);
741     p->name = NULL;
742     multifd_pages_clear(p->pages);
743     p->pages = NULL;
744     p->packet_len = 0;
745     g_free(p->packet);
746     p->packet = NULL;
747     g_free(p->iov);
748     p->iov = NULL;
749     multifd_send_state->ops->send_cleanup(p, errp);
750 
751     return *errp == NULL;
752 }
753 
754 static void multifd_send_cleanup_state(void)
755 {
756     file_cleanup_outgoing_migration();
757     fd_cleanup_outgoing_migration();
758     socket_cleanup_outgoing_migration();
759     qemu_sem_destroy(&multifd_send_state->channels_created);
760     qemu_sem_destroy(&multifd_send_state->channels_ready);
761     g_free(multifd_send_state->params);
762     multifd_send_state->params = NULL;
763     multifd_pages_clear(multifd_send_state->pages);
764     multifd_send_state->pages = NULL;
765     g_free(multifd_send_state);
766     multifd_send_state = NULL;
767 }
768 
769 void multifd_send_shutdown(void)
770 {
771     int i;
772 
773     if (!migrate_multifd()) {
774         return;
775     }
776 
777     multifd_send_terminate_threads();
778 
779     for (i = 0; i < migrate_multifd_channels(); i++) {
780         MultiFDSendParams *p = &multifd_send_state->params[i];
781         Error *local_err = NULL;
782 
783         if (!multifd_send_cleanup_channel(p, &local_err)) {
784             migrate_set_error(migrate_get_current(), local_err);
785             error_free(local_err);
786         }
787     }
788 
789     multifd_send_cleanup_state();
790 }
791 
792 static int multifd_zero_copy_flush(QIOChannel *c)
793 {
794     int ret;
795     Error *err = NULL;
796 
797     ret = qio_channel_flush(c, &err);
798     if (ret < 0) {
799         error_report_err(err);
800         return -1;
801     }
802     if (ret == 1) {
803         stat64_add(&mig_stats.dirty_sync_missed_zero_copy, 1);
804     }
805 
806     return ret;
807 }
808 
809 int multifd_send_sync_main(void)
810 {
811     int i;
812     bool flush_zero_copy;
813 
814     if (!migrate_multifd()) {
815         return 0;
816     }
817     if (multifd_send_state->pages->num) {
818         if (!multifd_send_pages()) {
819             error_report("%s: multifd_send_pages fail", __func__);
820             return -1;
821         }
822     }
823 
824     flush_zero_copy = migrate_zero_copy_send();
825 
826     for (i = 0; i < migrate_multifd_channels(); i++) {
827         MultiFDSendParams *p = &multifd_send_state->params[i];
828 
829         if (multifd_send_should_exit()) {
830             return -1;
831         }
832 
833         trace_multifd_send_sync_main_signal(p->id);
834 
835         /*
836          * We should be the only user so far, so not possible to be set by
837          * others concurrently.
838          */
839         assert(qatomic_read(&p->pending_sync) == false);
840         qatomic_set(&p->pending_sync, true);
841         qemu_sem_post(&p->sem);
842     }
843     for (i = 0; i < migrate_multifd_channels(); i++) {
844         MultiFDSendParams *p = &multifd_send_state->params[i];
845 
846         if (multifd_send_should_exit()) {
847             return -1;
848         }
849 
850         qemu_sem_wait(&multifd_send_state->channels_ready);
851         trace_multifd_send_sync_main_wait(p->id);
852         qemu_sem_wait(&p->sem_sync);
853 
854         if (flush_zero_copy && p->c && (multifd_zero_copy_flush(p->c) < 0)) {
855             return -1;
856         }
857     }
858     trace_multifd_send_sync_main(multifd_send_state->packet_num);
859 
860     return 0;
861 }
862 
863 static void *multifd_send_thread(void *opaque)
864 {
865     MultiFDSendParams *p = opaque;
866     MigrationThread *thread = NULL;
867     Error *local_err = NULL;
868     int ret = 0;
869     bool use_packets = multifd_use_packets();
870 
871     thread = migration_threads_add(p->name, qemu_get_thread_id());
872 
873     trace_multifd_send_thread_start(p->id);
874     rcu_register_thread();
875 
876     if (use_packets) {
877         if (multifd_send_initial_packet(p, &local_err) < 0) {
878             ret = -1;
879             goto out;
880         }
881     }
882 
883     while (true) {
884         qemu_sem_post(&multifd_send_state->channels_ready);
885         qemu_sem_wait(&p->sem);
886 
887         if (multifd_send_should_exit()) {
888             break;
889         }
890 
891         /*
892          * Read pending_job flag before p->pages.  Pairs with the
893          * qatomic_store_release() in multifd_send_pages().
894          */
895         if (qatomic_load_acquire(&p->pending_job)) {
896             MultiFDPages_t *pages = p->pages;
897 
898             p->iovs_num = 0;
899             assert(pages->num);
900 
901             ret = multifd_send_state->ops->send_prepare(p, &local_err);
902             if (ret != 0) {
903                 break;
904             }
905 
906             if (migrate_mapped_ram()) {
907                 ret = file_write_ramblock_iov(p->c, p->iov, p->iovs_num,
908                                               p->pages->block, &local_err);
909             } else {
910                 ret = qio_channel_writev_full_all(p->c, p->iov, p->iovs_num,
911                                                   NULL, 0, p->write_flags,
912                                                   &local_err);
913             }
914 
915             if (ret != 0) {
916                 break;
917             }
918 
919             stat64_add(&mig_stats.multifd_bytes,
920                        p->next_packet_size + p->packet_len);
921 
922             multifd_pages_reset(p->pages);
923             p->next_packet_size = 0;
924 
925             /*
926              * Making sure p->pages is published before saying "we're
927              * free".  Pairs with the smp_mb_acquire() in
928              * multifd_send_pages().
929              */
930             qatomic_store_release(&p->pending_job, false);
931         } else {
932             /*
933              * If not a normal job, must be a sync request.  Note that
934              * pending_sync is a standalone flag (unlike pending_job), so
935              * it doesn't require explicit memory barriers.
936              */
937             assert(qatomic_read(&p->pending_sync));
938 
939             if (use_packets) {
940                 p->flags = MULTIFD_FLAG_SYNC;
941                 multifd_send_fill_packet(p);
942                 ret = qio_channel_write_all(p->c, (void *)p->packet,
943                                             p->packet_len, &local_err);
944                 if (ret != 0) {
945                     break;
946                 }
947                 /* p->next_packet_size will always be zero for a SYNC packet */
948                 stat64_add(&mig_stats.multifd_bytes, p->packet_len);
949                 p->flags = 0;
950             }
951 
952             qatomic_set(&p->pending_sync, false);
953             qemu_sem_post(&p->sem_sync);
954         }
955     }
956 
957 out:
958     if (ret) {
959         assert(local_err);
960         trace_multifd_send_error(p->id);
961         multifd_send_set_error(local_err);
962         multifd_send_kick_main(p);
963         error_free(local_err);
964     }
965 
966     rcu_unregister_thread();
967     migration_threads_remove(thread);
968     trace_multifd_send_thread_end(p->id, p->packets_sent, p->total_normal_pages);
969 
970     return NULL;
971 }
972 
973 static void multifd_new_send_channel_async(QIOTask *task, gpointer opaque);
974 
975 typedef struct {
976     MultiFDSendParams *p;
977     QIOChannelTLS *tioc;
978 } MultiFDTLSThreadArgs;
979 
980 static void *multifd_tls_handshake_thread(void *opaque)
981 {
982     MultiFDTLSThreadArgs *args = opaque;
983 
984     qio_channel_tls_handshake(args->tioc,
985                               multifd_new_send_channel_async,
986                               args->p,
987                               NULL,
988                               NULL);
989     g_free(args);
990 
991     return NULL;
992 }
993 
994 static bool multifd_tls_channel_connect(MultiFDSendParams *p,
995                                         QIOChannel *ioc,
996                                         Error **errp)
997 {
998     MigrationState *s = migrate_get_current();
999     const char *hostname = s->hostname;
1000     MultiFDTLSThreadArgs *args;
1001     QIOChannelTLS *tioc;
1002 
1003     tioc = migration_tls_client_create(ioc, hostname, errp);
1004     if (!tioc) {
1005         return false;
1006     }
1007 
1008     /*
1009      * Ownership of the socket channel now transfers to the newly
1010      * created TLS channel, which has already taken a reference.
1011      */
1012     object_unref(OBJECT(ioc));
1013     trace_multifd_tls_outgoing_handshake_start(ioc, tioc, hostname);
1014     qio_channel_set_name(QIO_CHANNEL(tioc), "multifd-tls-outgoing");
1015 
1016     args = g_new0(MultiFDTLSThreadArgs, 1);
1017     args->tioc = tioc;
1018     args->p = p;
1019 
1020     p->tls_thread_created = true;
1021     qemu_thread_create(&p->tls_thread, "multifd-tls-handshake-worker",
1022                        multifd_tls_handshake_thread, args,
1023                        QEMU_THREAD_JOINABLE);
1024     return true;
1025 }
1026 
1027 void multifd_channel_connect(MultiFDSendParams *p, QIOChannel *ioc)
1028 {
1029     qio_channel_set_delay(ioc, false);
1030 
1031     migration_ioc_register_yank(ioc);
1032     /* Setup p->c only if the channel is completely setup */
1033     p->c = ioc;
1034 
1035     p->thread_created = true;
1036     qemu_thread_create(&p->thread, p->name, multifd_send_thread, p,
1037                        QEMU_THREAD_JOINABLE);
1038 }
1039 
1040 /*
1041  * When TLS is enabled this function is called once to establish the
1042  * TLS connection and a second time after the TLS handshake to create
1043  * the multifd channel. Without TLS it goes straight into the channel
1044  * creation.
1045  */
1046 static void multifd_new_send_channel_async(QIOTask *task, gpointer opaque)
1047 {
1048     MultiFDSendParams *p = opaque;
1049     QIOChannel *ioc = QIO_CHANNEL(qio_task_get_source(task));
1050     Error *local_err = NULL;
1051     bool ret;
1052 
1053     trace_multifd_new_send_channel_async(p->id);
1054 
1055     if (qio_task_propagate_error(task, &local_err)) {
1056         ret = false;
1057         goto out;
1058     }
1059 
1060     trace_multifd_set_outgoing_channel(ioc, object_get_typename(OBJECT(ioc)),
1061                                        migrate_get_current()->hostname);
1062 
1063     if (migrate_channel_requires_tls_upgrade(ioc)) {
1064         ret = multifd_tls_channel_connect(p, ioc, &local_err);
1065         if (ret) {
1066             return;
1067         }
1068     } else {
1069         multifd_channel_connect(p, ioc);
1070         ret = true;
1071     }
1072 
1073 out:
1074     /*
1075      * Here we're not interested whether creation succeeded, only that
1076      * it happened at all.
1077      */
1078     multifd_send_channel_created();
1079 
1080     if (ret) {
1081         return;
1082     }
1083 
1084     trace_multifd_new_send_channel_async_error(p->id, local_err);
1085     multifd_send_set_error(local_err);
1086     /*
1087      * For error cases (TLS or non-TLS), IO channel is always freed here
1088      * rather than when cleanup multifd: since p->c is not set, multifd
1089      * cleanup code doesn't even know its existence.
1090      */
1091     object_unref(OBJECT(ioc));
1092     error_free(local_err);
1093 }
1094 
1095 static bool multifd_new_send_channel_create(gpointer opaque, Error **errp)
1096 {
1097     if (!multifd_use_packets()) {
1098         return file_send_channel_create(opaque, errp);
1099     }
1100 
1101     socket_send_channel_create(multifd_new_send_channel_async, opaque);
1102     return true;
1103 }
1104 
1105 bool multifd_send_setup(void)
1106 {
1107     MigrationState *s = migrate_get_current();
1108     Error *local_err = NULL;
1109     int thread_count, ret = 0;
1110     uint32_t page_count = MULTIFD_PACKET_SIZE / qemu_target_page_size();
1111     bool use_packets = multifd_use_packets();
1112     uint8_t i;
1113 
1114     if (!migrate_multifd()) {
1115         return true;
1116     }
1117 
1118     thread_count = migrate_multifd_channels();
1119     multifd_send_state = g_malloc0(sizeof(*multifd_send_state));
1120     multifd_send_state->params = g_new0(MultiFDSendParams, thread_count);
1121     multifd_send_state->pages = multifd_pages_init(page_count);
1122     qemu_sem_init(&multifd_send_state->channels_created, 0);
1123     qemu_sem_init(&multifd_send_state->channels_ready, 0);
1124     qatomic_set(&multifd_send_state->exiting, 0);
1125     multifd_send_state->ops = multifd_ops[migrate_multifd_compression()];
1126 
1127     for (i = 0; i < thread_count; i++) {
1128         MultiFDSendParams *p = &multifd_send_state->params[i];
1129 
1130         qemu_sem_init(&p->sem, 0);
1131         qemu_sem_init(&p->sem_sync, 0);
1132         p->id = i;
1133         p->pages = multifd_pages_init(page_count);
1134 
1135         if (use_packets) {
1136             p->packet_len = sizeof(MultiFDPacket_t)
1137                           + sizeof(uint64_t) * page_count;
1138             p->packet = g_malloc0(p->packet_len);
1139             p->packet->magic = cpu_to_be32(MULTIFD_MAGIC);
1140             p->packet->version = cpu_to_be32(MULTIFD_VERSION);
1141 
1142             /* We need one extra place for the packet header */
1143             p->iov = g_new0(struct iovec, page_count + 1);
1144         } else {
1145             p->iov = g_new0(struct iovec, page_count);
1146         }
1147         p->name = g_strdup_printf("multifdsend_%d", i);
1148         p->page_size = qemu_target_page_size();
1149         p->page_count = page_count;
1150         p->write_flags = 0;
1151 
1152         if (!multifd_new_send_channel_create(p, &local_err)) {
1153             return false;
1154         }
1155     }
1156 
1157     /*
1158      * Wait until channel creation has started for all channels. The
1159      * creation can still fail, but no more channels will be created
1160      * past this point.
1161      */
1162     for (i = 0; i < thread_count; i++) {
1163         qemu_sem_wait(&multifd_send_state->channels_created);
1164     }
1165 
1166     for (i = 0; i < thread_count; i++) {
1167         MultiFDSendParams *p = &multifd_send_state->params[i];
1168 
1169         ret = multifd_send_state->ops->send_setup(p, &local_err);
1170         if (ret) {
1171             break;
1172         }
1173     }
1174 
1175     if (ret) {
1176         migrate_set_error(s, local_err);
1177         error_report_err(local_err);
1178         migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
1179                           MIGRATION_STATUS_FAILED);
1180         return false;
1181     }
1182 
1183     return true;
1184 }
1185 
1186 bool multifd_recv(void)
1187 {
1188     int i;
1189     static int next_recv_channel;
1190     MultiFDRecvParams *p = NULL;
1191     MultiFDRecvData *data = multifd_recv_state->data;
1192 
1193     /*
1194      * next_channel can remain from a previous migration that was
1195      * using more channels, so ensure it doesn't overflow if the
1196      * limit is lower now.
1197      */
1198     next_recv_channel %= migrate_multifd_channels();
1199     for (i = next_recv_channel;; i = (i + 1) % migrate_multifd_channels()) {
1200         if (multifd_recv_should_exit()) {
1201             return false;
1202         }
1203 
1204         p = &multifd_recv_state->params[i];
1205 
1206         if (qatomic_read(&p->pending_job) == false) {
1207             next_recv_channel = (i + 1) % migrate_multifd_channels();
1208             break;
1209         }
1210     }
1211 
1212     /*
1213      * Order pending_job read before manipulating p->data below. Pairs
1214      * with qatomic_store_release() at multifd_recv_thread().
1215      */
1216     smp_mb_acquire();
1217 
1218     assert(!p->data->size);
1219     multifd_recv_state->data = p->data;
1220     p->data = data;
1221 
1222     /*
1223      * Order p->data update before setting pending_job. Pairs with
1224      * qatomic_load_acquire() at multifd_recv_thread().
1225      */
1226     qatomic_store_release(&p->pending_job, true);
1227     qemu_sem_post(&p->sem);
1228 
1229     return true;
1230 }
1231 
1232 MultiFDRecvData *multifd_get_recv_data(void)
1233 {
1234     return multifd_recv_state->data;
1235 }
1236 
1237 static void multifd_recv_terminate_threads(Error *err)
1238 {
1239     int i;
1240 
1241     trace_multifd_recv_terminate_threads(err != NULL);
1242 
1243     if (qatomic_xchg(&multifd_recv_state->exiting, 1)) {
1244         return;
1245     }
1246 
1247     if (err) {
1248         MigrationState *s = migrate_get_current();
1249         migrate_set_error(s, err);
1250         if (s->state == MIGRATION_STATUS_SETUP ||
1251             s->state == MIGRATION_STATUS_ACTIVE) {
1252             migrate_set_state(&s->state, s->state,
1253                               MIGRATION_STATUS_FAILED);
1254         }
1255     }
1256 
1257     for (i = 0; i < migrate_multifd_channels(); i++) {
1258         MultiFDRecvParams *p = &multifd_recv_state->params[i];
1259 
1260         /*
1261          * The migration thread and channels interact differently
1262          * depending on the presence of packets.
1263          */
1264         if (multifd_use_packets()) {
1265             /*
1266              * The channel receives as long as there are packets. When
1267              * packets end (i.e. MULTIFD_FLAG_SYNC is reached), the
1268              * channel waits for the migration thread to sync. If the
1269              * sync never happens, do it here.
1270              */
1271             qemu_sem_post(&p->sem_sync);
1272         } else {
1273             /*
1274              * The channel waits for the migration thread to give it
1275              * work. When the migration thread runs out of work, it
1276              * releases the channel and waits for any pending work to
1277              * finish. If we reach here (e.g. due to error) before the
1278              * work runs out, release the channel.
1279              */
1280             qemu_sem_post(&p->sem);
1281         }
1282 
1283         /*
1284          * We could arrive here for two reasons:
1285          *  - normal quit, i.e. everything went fine, just finished
1286          *  - error quit: We close the channels so the channel threads
1287          *    finish the qio_channel_read_all_eof()
1288          */
1289         if (p->c) {
1290             qio_channel_shutdown(p->c, QIO_CHANNEL_SHUTDOWN_BOTH, NULL);
1291         }
1292     }
1293 }
1294 
1295 void multifd_recv_shutdown(void)
1296 {
1297     if (migrate_multifd()) {
1298         multifd_recv_terminate_threads(NULL);
1299     }
1300 }
1301 
1302 static void multifd_recv_cleanup_channel(MultiFDRecvParams *p)
1303 {
1304     migration_ioc_unregister_yank(p->c);
1305     object_unref(OBJECT(p->c));
1306     p->c = NULL;
1307     qemu_mutex_destroy(&p->mutex);
1308     qemu_sem_destroy(&p->sem_sync);
1309     qemu_sem_destroy(&p->sem);
1310     g_free(p->name);
1311     p->name = NULL;
1312     p->packet_len = 0;
1313     g_free(p->packet);
1314     p->packet = NULL;
1315     g_free(p->iov);
1316     p->iov = NULL;
1317     g_free(p->normal);
1318     p->normal = NULL;
1319     multifd_recv_state->ops->recv_cleanup(p);
1320 }
1321 
1322 static void multifd_recv_cleanup_state(void)
1323 {
1324     qemu_sem_destroy(&multifd_recv_state->sem_sync);
1325     g_free(multifd_recv_state->params);
1326     multifd_recv_state->params = NULL;
1327     g_free(multifd_recv_state->data);
1328     multifd_recv_state->data = NULL;
1329     g_free(multifd_recv_state);
1330     multifd_recv_state = NULL;
1331 }
1332 
1333 void multifd_recv_cleanup(void)
1334 {
1335     int i;
1336 
1337     if (!migrate_multifd()) {
1338         return;
1339     }
1340     multifd_recv_terminate_threads(NULL);
1341     for (i = 0; i < migrate_multifd_channels(); i++) {
1342         MultiFDRecvParams *p = &multifd_recv_state->params[i];
1343 
1344         if (p->thread_created) {
1345             qemu_thread_join(&p->thread);
1346         }
1347     }
1348     for (i = 0; i < migrate_multifd_channels(); i++) {
1349         multifd_recv_cleanup_channel(&multifd_recv_state->params[i]);
1350     }
1351     multifd_recv_cleanup_state();
1352 }
1353 
1354 void multifd_recv_sync_main(void)
1355 {
1356     int thread_count = migrate_multifd_channels();
1357     bool file_based = !multifd_use_packets();
1358     int i;
1359 
1360     if (!migrate_multifd()) {
1361         return;
1362     }
1363 
1364     /*
1365      * File-based channels don't use packets and therefore need to
1366      * wait for more work. Release them to start the sync.
1367      */
1368     if (file_based) {
1369         for (i = 0; i < thread_count; i++) {
1370             MultiFDRecvParams *p = &multifd_recv_state->params[i];
1371 
1372             trace_multifd_recv_sync_main_signal(p->id);
1373             qemu_sem_post(&p->sem);
1374         }
1375     }
1376 
1377     /*
1378      * Initiate the synchronization by waiting for all channels.
1379      *
1380      * For socket-based migration this means each channel has received
1381      * the SYNC packet on the stream.
1382      *
1383      * For file-based migration this means each channel is done with
1384      * the work (pending_job=false).
1385      */
1386     for (i = 0; i < thread_count; i++) {
1387         trace_multifd_recv_sync_main_wait(i);
1388         qemu_sem_wait(&multifd_recv_state->sem_sync);
1389     }
1390 
1391     if (file_based) {
1392         /*
1393          * For file-based loading is done in one iteration. We're
1394          * done.
1395          */
1396         return;
1397     }
1398 
1399     /*
1400      * Sync done. Release the channels for the next iteration.
1401      */
1402     for (i = 0; i < thread_count; i++) {
1403         MultiFDRecvParams *p = &multifd_recv_state->params[i];
1404 
1405         WITH_QEMU_LOCK_GUARD(&p->mutex) {
1406             if (multifd_recv_state->packet_num < p->packet_num) {
1407                 multifd_recv_state->packet_num = p->packet_num;
1408             }
1409         }
1410         trace_multifd_recv_sync_main_signal(p->id);
1411         qemu_sem_post(&p->sem_sync);
1412     }
1413     trace_multifd_recv_sync_main(multifd_recv_state->packet_num);
1414 }
1415 
1416 static void *multifd_recv_thread(void *opaque)
1417 {
1418     MultiFDRecvParams *p = opaque;
1419     Error *local_err = NULL;
1420     bool use_packets = multifd_use_packets();
1421     int ret;
1422 
1423     trace_multifd_recv_thread_start(p->id);
1424     rcu_register_thread();
1425 
1426     while (true) {
1427         uint32_t flags = 0;
1428         bool has_data = false;
1429         p->normal_num = 0;
1430 
1431         if (use_packets) {
1432             if (multifd_recv_should_exit()) {
1433                 break;
1434             }
1435 
1436             ret = qio_channel_read_all_eof(p->c, (void *)p->packet,
1437                                            p->packet_len, &local_err);
1438             if (ret == 0 || ret == -1) {   /* 0: EOF  -1: Error */
1439                 break;
1440             }
1441 
1442             qemu_mutex_lock(&p->mutex);
1443             ret = multifd_recv_unfill_packet(p, &local_err);
1444             if (ret) {
1445                 qemu_mutex_unlock(&p->mutex);
1446                 break;
1447             }
1448 
1449             flags = p->flags;
1450             /* recv methods don't know how to handle the SYNC flag */
1451             p->flags &= ~MULTIFD_FLAG_SYNC;
1452             has_data = !!p->normal_num;
1453             qemu_mutex_unlock(&p->mutex);
1454         } else {
1455             /*
1456              * No packets, so we need to wait for the vmstate code to
1457              * give us work.
1458              */
1459             qemu_sem_wait(&p->sem);
1460 
1461             if (multifd_recv_should_exit()) {
1462                 break;
1463             }
1464 
1465             /* pairs with qatomic_store_release() at multifd_recv() */
1466             if (!qatomic_load_acquire(&p->pending_job)) {
1467                 /*
1468                  * Migration thread did not send work, this is
1469                  * equivalent to pending_sync on the sending
1470                  * side. Post sem_sync to notify we reached this
1471                  * point.
1472                  */
1473                 qemu_sem_post(&multifd_recv_state->sem_sync);
1474                 continue;
1475             }
1476 
1477             has_data = !!p->data->size;
1478         }
1479 
1480         if (has_data) {
1481             ret = multifd_recv_state->ops->recv(p, &local_err);
1482             if (ret != 0) {
1483                 break;
1484             }
1485         }
1486 
1487         if (use_packets) {
1488             if (flags & MULTIFD_FLAG_SYNC) {
1489                 qemu_sem_post(&multifd_recv_state->sem_sync);
1490                 qemu_sem_wait(&p->sem_sync);
1491             }
1492         } else {
1493             p->total_normal_pages += p->data->size / qemu_target_page_size();
1494             p->data->size = 0;
1495             /*
1496              * Order data->size update before clearing
1497              * pending_job. Pairs with smp_mb_acquire() at
1498              * multifd_recv().
1499              */
1500             qatomic_store_release(&p->pending_job, false);
1501         }
1502     }
1503 
1504     if (local_err) {
1505         multifd_recv_terminate_threads(local_err);
1506         error_free(local_err);
1507     }
1508 
1509     rcu_unregister_thread();
1510     trace_multifd_recv_thread_end(p->id, p->packets_recved, p->total_normal_pages);
1511 
1512     return NULL;
1513 }
1514 
1515 int multifd_recv_setup(Error **errp)
1516 {
1517     int thread_count;
1518     uint32_t page_count = MULTIFD_PACKET_SIZE / qemu_target_page_size();
1519     bool use_packets = multifd_use_packets();
1520     uint8_t i;
1521 
1522     /*
1523      * Return successfully if multiFD recv state is already initialised
1524      * or multiFD is not enabled.
1525      */
1526     if (multifd_recv_state || !migrate_multifd()) {
1527         return 0;
1528     }
1529 
1530     thread_count = migrate_multifd_channels();
1531     multifd_recv_state = g_malloc0(sizeof(*multifd_recv_state));
1532     multifd_recv_state->params = g_new0(MultiFDRecvParams, thread_count);
1533 
1534     multifd_recv_state->data = g_new0(MultiFDRecvData, 1);
1535     multifd_recv_state->data->size = 0;
1536 
1537     qatomic_set(&multifd_recv_state->count, 0);
1538     qatomic_set(&multifd_recv_state->exiting, 0);
1539     qemu_sem_init(&multifd_recv_state->sem_sync, 0);
1540     multifd_recv_state->ops = multifd_ops[migrate_multifd_compression()];
1541 
1542     for (i = 0; i < thread_count; i++) {
1543         MultiFDRecvParams *p = &multifd_recv_state->params[i];
1544 
1545         qemu_mutex_init(&p->mutex);
1546         qemu_sem_init(&p->sem_sync, 0);
1547         qemu_sem_init(&p->sem, 0);
1548         p->pending_job = false;
1549         p->id = i;
1550 
1551         p->data = g_new0(MultiFDRecvData, 1);
1552         p->data->size = 0;
1553 
1554         if (use_packets) {
1555             p->packet_len = sizeof(MultiFDPacket_t)
1556                 + sizeof(uint64_t) * page_count;
1557             p->packet = g_malloc0(p->packet_len);
1558         }
1559         p->name = g_strdup_printf("multifdrecv_%d", i);
1560         p->iov = g_new0(struct iovec, page_count);
1561         p->normal = g_new0(ram_addr_t, page_count);
1562         p->page_count = page_count;
1563         p->page_size = qemu_target_page_size();
1564     }
1565 
1566     for (i = 0; i < thread_count; i++) {
1567         MultiFDRecvParams *p = &multifd_recv_state->params[i];
1568         int ret;
1569 
1570         ret = multifd_recv_state->ops->recv_setup(p, errp);
1571         if (ret) {
1572             return ret;
1573         }
1574     }
1575     return 0;
1576 }
1577 
1578 bool multifd_recv_all_channels_created(void)
1579 {
1580     int thread_count = migrate_multifd_channels();
1581 
1582     if (!migrate_multifd()) {
1583         return true;
1584     }
1585 
1586     if (!multifd_recv_state) {
1587         /* Called before any connections created */
1588         return false;
1589     }
1590 
1591     return thread_count == qatomic_read(&multifd_recv_state->count);
1592 }
1593 
1594 /*
1595  * Try to receive all multifd channels to get ready for the migration.
1596  * Sets @errp when failing to receive the current channel.
1597  */
1598 void multifd_recv_new_channel(QIOChannel *ioc, Error **errp)
1599 {
1600     MultiFDRecvParams *p;
1601     Error *local_err = NULL;
1602     bool use_packets = multifd_use_packets();
1603     int id;
1604 
1605     if (use_packets) {
1606         id = multifd_recv_initial_packet(ioc, &local_err);
1607         if (id < 0) {
1608             multifd_recv_terminate_threads(local_err);
1609             error_propagate_prepend(errp, local_err,
1610                                     "failed to receive packet"
1611                                     " via multifd channel %d: ",
1612                                     qatomic_read(&multifd_recv_state->count));
1613             return;
1614         }
1615         trace_multifd_recv_new_channel(id);
1616     } else {
1617         id = qatomic_read(&multifd_recv_state->count);
1618     }
1619 
1620     p = &multifd_recv_state->params[id];
1621     if (p->c != NULL) {
1622         error_setg(&local_err, "multifd: received id '%d' already setup'",
1623                    id);
1624         multifd_recv_terminate_threads(local_err);
1625         error_propagate(errp, local_err);
1626         return;
1627     }
1628     p->c = ioc;
1629     object_ref(OBJECT(ioc));
1630 
1631     p->thread_created = true;
1632     qemu_thread_create(&p->thread, p->name, multifd_recv_thread, p,
1633                        QEMU_THREAD_JOINABLE);
1634     qatomic_inc(&multifd_recv_state->count);
1635 }
1636