xref: /qemu/migration/multifd.c (revision e8c5503a)
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/cutils.h"
15 #include "qemu/rcu.h"
16 #include "exec/target_page.h"
17 #include "sysemu/sysemu.h"
18 #include "exec/ramblock.h"
19 #include "qemu/error-report.h"
20 #include "qapi/error.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->normal_num; i++) {
118         ramblock_set_file_bmap_atomic(pages->block, pages->offset[i], true);
119     }
120 
121     for (int i = p->pages->normal_num; i < p->pages->num; i++) {
122         ramblock_set_file_bmap_atomic(pages->block, pages->offset[i], false);
123     }
124 }
125 
126 /* Multifd without compression */
127 
128 /**
129  * nocomp_send_setup: setup send side
130  *
131  * @p: Params for the channel that we are using
132  * @errp: pointer to an error
133  */
134 static int nocomp_send_setup(MultiFDSendParams *p, Error **errp)
135 {
136     if (migrate_zero_copy_send()) {
137         p->write_flags |= QIO_CHANNEL_WRITE_FLAG_ZERO_COPY;
138     }
139 
140     return 0;
141 }
142 
143 /**
144  * nocomp_send_cleanup: cleanup send side
145  *
146  * For no compression this function does nothing.
147  *
148  * @p: Params for the channel that we are using
149  * @errp: pointer to an error
150  */
151 static void nocomp_send_cleanup(MultiFDSendParams *p, Error **errp)
152 {
153     return;
154 }
155 
156 static void multifd_send_prepare_iovs(MultiFDSendParams *p)
157 {
158     MultiFDPages_t *pages = p->pages;
159 
160     for (int i = 0; i < pages->normal_num; i++) {
161         p->iov[p->iovs_num].iov_base = pages->block->host + pages->offset[i];
162         p->iov[p->iovs_num].iov_len = p->page_size;
163         p->iovs_num++;
164     }
165 
166     p->next_packet_size = pages->normal_num * p->page_size;
167 }
168 
169 /**
170  * nocomp_send_prepare: prepare date to be able to send
171  *
172  * For no compression we just have to calculate the size of the
173  * packet.
174  *
175  * Returns 0 for success or -1 for error
176  *
177  * @p: Params for the channel that we are using
178  * @errp: pointer to an error
179  */
180 static int nocomp_send_prepare(MultiFDSendParams *p, Error **errp)
181 {
182     bool use_zero_copy_send = migrate_zero_copy_send();
183     int ret;
184 
185     multifd_send_zero_page_detect(p);
186 
187     if (!multifd_use_packets()) {
188         multifd_send_prepare_iovs(p);
189         multifd_set_file_bitmap(p);
190 
191         return 0;
192     }
193 
194     if (!use_zero_copy_send) {
195         /*
196          * Only !zerocopy needs the header in IOV; zerocopy will
197          * send it separately.
198          */
199         multifd_send_prepare_header(p);
200     }
201 
202     multifd_send_prepare_iovs(p);
203     p->flags |= MULTIFD_FLAG_NOCOMP;
204 
205     multifd_send_fill_packet(p);
206 
207     if (use_zero_copy_send) {
208         /* Send header first, without zerocopy */
209         ret = qio_channel_write_all(p->c, (void *)p->packet,
210                                     p->packet_len, errp);
211         if (ret != 0) {
212             return -1;
213         }
214     }
215 
216     return 0;
217 }
218 
219 /**
220  * nocomp_recv_setup: setup receive side
221  *
222  * For no compression this function does nothing.
223  *
224  * Returns 0 for success or -1 for error
225  *
226  * @p: Params for the channel that we are using
227  * @errp: pointer to an error
228  */
229 static int nocomp_recv_setup(MultiFDRecvParams *p, Error **errp)
230 {
231     return 0;
232 }
233 
234 /**
235  * nocomp_recv_cleanup: setup receive side
236  *
237  * For no compression this function does nothing.
238  *
239  * @p: Params for the channel that we are using
240  */
241 static void nocomp_recv_cleanup(MultiFDRecvParams *p)
242 {
243 }
244 
245 /**
246  * nocomp_recv: read the data from the channel
247  *
248  * For no compression we just need to read things into the correct place.
249  *
250  * Returns 0 for success or -1 for error
251  *
252  * @p: Params for the channel that we are using
253  * @errp: pointer to an error
254  */
255 static int nocomp_recv(MultiFDRecvParams *p, Error **errp)
256 {
257     uint32_t flags;
258 
259     if (!multifd_use_packets()) {
260         return multifd_file_recv_data(p, errp);
261     }
262 
263     flags = p->flags & MULTIFD_FLAG_COMPRESSION_MASK;
264 
265     if (flags != MULTIFD_FLAG_NOCOMP) {
266         error_setg(errp, "multifd %u: flags received %x flags expected %x",
267                    p->id, flags, MULTIFD_FLAG_NOCOMP);
268         return -1;
269     }
270 
271     multifd_recv_zero_page_process(p);
272 
273     if (!p->normal_num) {
274         return 0;
275     }
276 
277     for (int i = 0; i < p->normal_num; i++) {
278         p->iov[i].iov_base = p->host + p->normal[i];
279         p->iov[i].iov_len = p->page_size;
280     }
281     return qio_channel_readv_all(p->c, p->iov, p->normal_num, errp);
282 }
283 
284 static MultiFDMethods multifd_nocomp_ops = {
285     .send_setup = nocomp_send_setup,
286     .send_cleanup = nocomp_send_cleanup,
287     .send_prepare = nocomp_send_prepare,
288     .recv_setup = nocomp_recv_setup,
289     .recv_cleanup = nocomp_recv_cleanup,
290     .recv = nocomp_recv
291 };
292 
293 static MultiFDMethods *multifd_ops[MULTIFD_COMPRESSION__MAX] = {
294     [MULTIFD_COMPRESSION_NONE] = &multifd_nocomp_ops,
295 };
296 
297 void multifd_register_ops(int method, MultiFDMethods *ops)
298 {
299     assert(0 < method && method < MULTIFD_COMPRESSION__MAX);
300     multifd_ops[method] = ops;
301 }
302 
303 /* Reset a MultiFDPages_t* object for the next use */
304 static void multifd_pages_reset(MultiFDPages_t *pages)
305 {
306     /*
307      * We don't need to touch offset[] array, because it will be
308      * overwritten later when reused.
309      */
310     pages->num = 0;
311     pages->normal_num = 0;
312     pages->block = NULL;
313 }
314 
315 static int multifd_send_initial_packet(MultiFDSendParams *p, Error **errp)
316 {
317     MultiFDInit_t msg = {};
318     size_t size = sizeof(msg);
319     int ret;
320 
321     msg.magic = cpu_to_be32(MULTIFD_MAGIC);
322     msg.version = cpu_to_be32(MULTIFD_VERSION);
323     msg.id = p->id;
324     memcpy(msg.uuid, &qemu_uuid.data, sizeof(msg.uuid));
325 
326     ret = qio_channel_write_all(p->c, (char *)&msg, size, errp);
327     if (ret != 0) {
328         return -1;
329     }
330     stat64_add(&mig_stats.multifd_bytes, size);
331     return 0;
332 }
333 
334 static int multifd_recv_initial_packet(QIOChannel *c, Error **errp)
335 {
336     MultiFDInit_t msg;
337     int ret;
338 
339     ret = qio_channel_read_all(c, (char *)&msg, sizeof(msg), errp);
340     if (ret != 0) {
341         return -1;
342     }
343 
344     msg.magic = be32_to_cpu(msg.magic);
345     msg.version = be32_to_cpu(msg.version);
346 
347     if (msg.magic != MULTIFD_MAGIC) {
348         error_setg(errp, "multifd: received packet magic %x "
349                    "expected %x", msg.magic, MULTIFD_MAGIC);
350         return -1;
351     }
352 
353     if (msg.version != MULTIFD_VERSION) {
354         error_setg(errp, "multifd: received packet version %u "
355                    "expected %u", msg.version, MULTIFD_VERSION);
356         return -1;
357     }
358 
359     if (memcmp(msg.uuid, &qemu_uuid, sizeof(qemu_uuid))) {
360         char *uuid = qemu_uuid_unparse_strdup(&qemu_uuid);
361         char *msg_uuid = qemu_uuid_unparse_strdup((const QemuUUID *)msg.uuid);
362 
363         error_setg(errp, "multifd: received uuid '%s' and expected "
364                    "uuid '%s' for channel %hhd", msg_uuid, uuid, msg.id);
365         g_free(uuid);
366         g_free(msg_uuid);
367         return -1;
368     }
369 
370     if (msg.id > migrate_multifd_channels()) {
371         error_setg(errp, "multifd: received channel id %u is greater than "
372                    "number of channels %u", msg.id, migrate_multifd_channels());
373         return -1;
374     }
375 
376     return msg.id;
377 }
378 
379 static MultiFDPages_t *multifd_pages_init(uint32_t n)
380 {
381     MultiFDPages_t *pages = g_new0(MultiFDPages_t, 1);
382 
383     pages->allocated = n;
384     pages->offset = g_new0(ram_addr_t, n);
385 
386     return pages;
387 }
388 
389 static void multifd_pages_clear(MultiFDPages_t *pages)
390 {
391     multifd_pages_reset(pages);
392     pages->allocated = 0;
393     g_free(pages->offset);
394     pages->offset = NULL;
395     g_free(pages);
396 }
397 
398 void multifd_send_fill_packet(MultiFDSendParams *p)
399 {
400     MultiFDPacket_t *packet = p->packet;
401     MultiFDPages_t *pages = p->pages;
402     uint64_t packet_num;
403     uint32_t zero_num = pages->num - pages->normal_num;
404     int i;
405 
406     packet->flags = cpu_to_be32(p->flags);
407     packet->pages_alloc = cpu_to_be32(p->pages->allocated);
408     packet->normal_pages = cpu_to_be32(pages->normal_num);
409     packet->zero_pages = cpu_to_be32(zero_num);
410     packet->next_packet_size = cpu_to_be32(p->next_packet_size);
411 
412     packet_num = qatomic_fetch_inc(&multifd_send_state->packet_num);
413     packet->packet_num = cpu_to_be64(packet_num);
414 
415     if (pages->block) {
416         strncpy(packet->ramblock, pages->block->idstr, 256);
417     }
418 
419     for (i = 0; i < pages->num; i++) {
420         /* there are architectures where ram_addr_t is 32 bit */
421         uint64_t temp = pages->offset[i];
422 
423         packet->offset[i] = cpu_to_be64(temp);
424     }
425 
426     p->packets_sent++;
427     p->total_normal_pages += pages->normal_num;
428     p->total_zero_pages += zero_num;
429 
430     trace_multifd_send(p->id, packet_num, pages->normal_num, zero_num,
431                        p->flags, p->next_packet_size);
432 }
433 
434 static int multifd_recv_unfill_packet(MultiFDRecvParams *p, Error **errp)
435 {
436     MultiFDPacket_t *packet = p->packet;
437     int i;
438 
439     packet->magic = be32_to_cpu(packet->magic);
440     if (packet->magic != MULTIFD_MAGIC) {
441         error_setg(errp, "multifd: received packet "
442                    "magic %x and expected magic %x",
443                    packet->magic, MULTIFD_MAGIC);
444         return -1;
445     }
446 
447     packet->version = be32_to_cpu(packet->version);
448     if (packet->version != MULTIFD_VERSION) {
449         error_setg(errp, "multifd: received packet "
450                    "version %u and expected version %u",
451                    packet->version, MULTIFD_VERSION);
452         return -1;
453     }
454 
455     p->flags = be32_to_cpu(packet->flags);
456 
457     packet->pages_alloc = be32_to_cpu(packet->pages_alloc);
458     /*
459      * If we received a packet that is 100 times bigger than expected
460      * just stop migration.  It is a magic number.
461      */
462     if (packet->pages_alloc > p->page_count) {
463         error_setg(errp, "multifd: received packet "
464                    "with size %u and expected a size of %u",
465                    packet->pages_alloc, p->page_count) ;
466         return -1;
467     }
468 
469     p->normal_num = be32_to_cpu(packet->normal_pages);
470     if (p->normal_num > packet->pages_alloc) {
471         error_setg(errp, "multifd: received packet "
472                    "with %u normal pages and expected maximum pages are %u",
473                    p->normal_num, packet->pages_alloc) ;
474         return -1;
475     }
476 
477     p->zero_num = be32_to_cpu(packet->zero_pages);
478     if (p->zero_num > packet->pages_alloc - p->normal_num) {
479         error_setg(errp, "multifd: received packet "
480                    "with %u zero pages and expected maximum zero pages are %u",
481                    p->zero_num, packet->pages_alloc - p->normal_num) ;
482         return -1;
483     }
484 
485     p->next_packet_size = be32_to_cpu(packet->next_packet_size);
486     p->packet_num = be64_to_cpu(packet->packet_num);
487     p->packets_recved++;
488     p->total_normal_pages += p->normal_num;
489     p->total_zero_pages += p->zero_num;
490 
491     trace_multifd_recv(p->id, p->packet_num, p->normal_num, p->zero_num,
492                        p->flags, p->next_packet_size);
493 
494     if (p->normal_num == 0 && p->zero_num == 0) {
495         return 0;
496     }
497 
498     /* make sure that ramblock is 0 terminated */
499     packet->ramblock[255] = 0;
500     p->block = qemu_ram_block_by_name(packet->ramblock);
501     if (!p->block) {
502         error_setg(errp, "multifd: unknown ram block %s",
503                    packet->ramblock);
504         return -1;
505     }
506 
507     p->host = p->block->host;
508     for (i = 0; i < p->normal_num; i++) {
509         uint64_t offset = be64_to_cpu(packet->offset[i]);
510 
511         if (offset > (p->block->used_length - p->page_size)) {
512             error_setg(errp, "multifd: offset too long %" PRIu64
513                        " (max " RAM_ADDR_FMT ")",
514                        offset, p->block->used_length);
515             return -1;
516         }
517         p->normal[i] = offset;
518     }
519 
520     for (i = 0; i < p->zero_num; i++) {
521         uint64_t offset = be64_to_cpu(packet->offset[p->normal_num + i]);
522 
523         if (offset > (p->block->used_length - p->page_size)) {
524             error_setg(errp, "multifd: offset too long %" PRIu64
525                        " (max " RAM_ADDR_FMT ")",
526                        offset, p->block->used_length);
527             return -1;
528         }
529         p->zero[i] = offset;
530     }
531 
532     return 0;
533 }
534 
535 static bool multifd_send_should_exit(void)
536 {
537     return qatomic_read(&multifd_send_state->exiting);
538 }
539 
540 static bool multifd_recv_should_exit(void)
541 {
542     return qatomic_read(&multifd_recv_state->exiting);
543 }
544 
545 /*
546  * The migration thread can wait on either of the two semaphores.  This
547  * function can be used to kick the main thread out of waiting on either of
548  * them.  Should mostly only be called when something wrong happened with
549  * the current multifd send thread.
550  */
551 static void multifd_send_kick_main(MultiFDSendParams *p)
552 {
553     qemu_sem_post(&p->sem_sync);
554     qemu_sem_post(&multifd_send_state->channels_ready);
555 }
556 
557 /*
558  * How we use multifd_send_state->pages and channel->pages?
559  *
560  * We create a pages for each channel, and a main one.  Each time that
561  * we need to send a batch of pages we interchange the ones between
562  * multifd_send_state and the channel that is sending it.  There are
563  * two reasons for that:
564  *    - to not have to do so many mallocs during migration
565  *    - to make easier to know what to free at the end of migration
566  *
567  * This way we always know who is the owner of each "pages" struct,
568  * and we don't need any locking.  It belongs to the migration thread
569  * or to the channel thread.  Switching is safe because the migration
570  * thread is using the channel mutex when changing it, and the channel
571  * have to had finish with its own, otherwise pending_job can't be
572  * false.
573  *
574  * Returns true if succeed, false otherwise.
575  */
576 static bool multifd_send_pages(void)
577 {
578     int i;
579     static int next_channel;
580     MultiFDSendParams *p = NULL; /* make happy gcc */
581     MultiFDPages_t *pages = multifd_send_state->pages;
582 
583     if (multifd_send_should_exit()) {
584         return false;
585     }
586 
587     /* We wait here, until at least one channel is ready */
588     qemu_sem_wait(&multifd_send_state->channels_ready);
589 
590     /*
591      * next_channel can remain from a previous migration that was
592      * using more channels, so ensure it doesn't overflow if the
593      * limit is lower now.
594      */
595     next_channel %= migrate_multifd_channels();
596     for (i = next_channel;; i = (i + 1) % migrate_multifd_channels()) {
597         if (multifd_send_should_exit()) {
598             return false;
599         }
600         p = &multifd_send_state->params[i];
601         /*
602          * Lockless read to p->pending_job is safe, because only multifd
603          * sender thread can clear it.
604          */
605         if (qatomic_read(&p->pending_job) == false) {
606             next_channel = (i + 1) % migrate_multifd_channels();
607             break;
608         }
609     }
610 
611     /*
612      * Make sure we read p->pending_job before all the rest.  Pairs with
613      * qatomic_store_release() in multifd_send_thread().
614      */
615     smp_mb_acquire();
616     assert(!p->pages->num);
617     multifd_send_state->pages = p->pages;
618     p->pages = pages;
619     /*
620      * Making sure p->pages is setup before marking pending_job=true. Pairs
621      * with the qatomic_load_acquire() in multifd_send_thread().
622      */
623     qatomic_store_release(&p->pending_job, true);
624     qemu_sem_post(&p->sem);
625 
626     return true;
627 }
628 
629 static inline bool multifd_queue_empty(MultiFDPages_t *pages)
630 {
631     return pages->num == 0;
632 }
633 
634 static inline bool multifd_queue_full(MultiFDPages_t *pages)
635 {
636     return pages->num == pages->allocated;
637 }
638 
639 static inline void multifd_enqueue(MultiFDPages_t *pages, ram_addr_t offset)
640 {
641     pages->offset[pages->num++] = offset;
642 }
643 
644 /* Returns true if enqueue successful, false otherwise */
645 bool multifd_queue_page(RAMBlock *block, ram_addr_t offset)
646 {
647     MultiFDPages_t *pages;
648 
649 retry:
650     pages = multifd_send_state->pages;
651 
652     /* If the queue is empty, we can already enqueue now */
653     if (multifd_queue_empty(pages)) {
654         pages->block = block;
655         multifd_enqueue(pages, offset);
656         return true;
657     }
658 
659     /*
660      * Not empty, meanwhile we need a flush.  It can because of either:
661      *
662      * (1) The page is not on the same ramblock of previous ones, or,
663      * (2) The queue is full.
664      *
665      * After flush, always retry.
666      */
667     if (pages->block != block || multifd_queue_full(pages)) {
668         if (!multifd_send_pages()) {
669             return false;
670         }
671         goto retry;
672     }
673 
674     /* Not empty, and we still have space, do it! */
675     multifd_enqueue(pages, offset);
676     return true;
677 }
678 
679 /* Multifd send side hit an error; remember it and prepare to quit */
680 static void multifd_send_set_error(Error *err)
681 {
682     /*
683      * We don't want to exit each threads twice.  Depending on where
684      * we get the error, or if there are two independent errors in two
685      * threads at the same time, we can end calling this function
686      * twice.
687      */
688     if (qatomic_xchg(&multifd_send_state->exiting, 1)) {
689         return;
690     }
691 
692     if (err) {
693         MigrationState *s = migrate_get_current();
694         migrate_set_error(s, err);
695         if (s->state == MIGRATION_STATUS_SETUP ||
696             s->state == MIGRATION_STATUS_PRE_SWITCHOVER ||
697             s->state == MIGRATION_STATUS_DEVICE ||
698             s->state == MIGRATION_STATUS_ACTIVE) {
699             migrate_set_state(&s->state, s->state,
700                               MIGRATION_STATUS_FAILED);
701         }
702     }
703 }
704 
705 static void multifd_send_terminate_threads(void)
706 {
707     int i;
708 
709     trace_multifd_send_terminate_threads();
710 
711     /*
712      * Tell everyone we're quitting.  No xchg() needed here; we simply
713      * always set it.
714      */
715     qatomic_set(&multifd_send_state->exiting, 1);
716 
717     /*
718      * Firstly, kick all threads out; no matter whether they are just idle,
719      * or blocked in an IO system call.
720      */
721     for (i = 0; i < migrate_multifd_channels(); i++) {
722         MultiFDSendParams *p = &multifd_send_state->params[i];
723 
724         qemu_sem_post(&p->sem);
725         if (p->c) {
726             qio_channel_shutdown(p->c, QIO_CHANNEL_SHUTDOWN_BOTH, NULL);
727         }
728     }
729 
730     /*
731      * Finally recycle all the threads.
732      */
733     for (i = 0; i < migrate_multifd_channels(); i++) {
734         MultiFDSendParams *p = &multifd_send_state->params[i];
735 
736         if (p->tls_thread_created) {
737             qemu_thread_join(&p->tls_thread);
738         }
739 
740         if (p->thread_created) {
741             qemu_thread_join(&p->thread);
742         }
743     }
744 }
745 
746 static bool multifd_send_cleanup_channel(MultiFDSendParams *p, Error **errp)
747 {
748     if (p->c) {
749         migration_ioc_unregister_yank(p->c);
750         /*
751          * The object_unref() cannot guarantee the fd will always be
752          * released because finalize() of the iochannel is only
753          * triggered on the last reference and it's not guaranteed
754          * that we always hold the last refcount when reaching here.
755          *
756          * Closing the fd explicitly has the benefit that if there is any
757          * registered I/O handler callbacks on such fd, that will get a
758          * POLLNVAL event and will further trigger the cleanup to finally
759          * release the IOC.
760          *
761          * FIXME: It should logically be guaranteed that all multifd
762          * channels have no I/O handler callback registered when reaching
763          * here, because migration thread will wait for all multifd channel
764          * establishments to complete during setup.  Since
765          * migrate_fd_cleanup() will be scheduled in main thread too, all
766          * previous callbacks should guarantee to be completed when
767          * reaching here.  See multifd_send_state.channels_created and its
768          * usage.  In the future, we could replace this with an assert
769          * making sure we're the last reference, or simply drop it if above
770          * is more clear to be justified.
771          */
772         qio_channel_close(p->c, &error_abort);
773         object_unref(OBJECT(p->c));
774         p->c = NULL;
775     }
776     qemu_sem_destroy(&p->sem);
777     qemu_sem_destroy(&p->sem_sync);
778     g_free(p->name);
779     p->name = NULL;
780     multifd_pages_clear(p->pages);
781     p->pages = NULL;
782     p->packet_len = 0;
783     g_free(p->packet);
784     p->packet = NULL;
785     g_free(p->iov);
786     p->iov = NULL;
787     multifd_send_state->ops->send_cleanup(p, errp);
788 
789     return *errp == NULL;
790 }
791 
792 static void multifd_send_cleanup_state(void)
793 {
794     file_cleanup_outgoing_migration();
795     socket_cleanup_outgoing_migration();
796     qemu_sem_destroy(&multifd_send_state->channels_created);
797     qemu_sem_destroy(&multifd_send_state->channels_ready);
798     g_free(multifd_send_state->params);
799     multifd_send_state->params = NULL;
800     multifd_pages_clear(multifd_send_state->pages);
801     multifd_send_state->pages = NULL;
802     g_free(multifd_send_state);
803     multifd_send_state = NULL;
804 }
805 
806 void multifd_send_shutdown(void)
807 {
808     int i;
809 
810     if (!migrate_multifd()) {
811         return;
812     }
813 
814     multifd_send_terminate_threads();
815 
816     for (i = 0; i < migrate_multifd_channels(); i++) {
817         MultiFDSendParams *p = &multifd_send_state->params[i];
818         Error *local_err = NULL;
819 
820         if (!multifd_send_cleanup_channel(p, &local_err)) {
821             migrate_set_error(migrate_get_current(), local_err);
822             error_free(local_err);
823         }
824     }
825 
826     multifd_send_cleanup_state();
827 }
828 
829 static int multifd_zero_copy_flush(QIOChannel *c)
830 {
831     int ret;
832     Error *err = NULL;
833 
834     ret = qio_channel_flush(c, &err);
835     if (ret < 0) {
836         error_report_err(err);
837         return -1;
838     }
839     if (ret == 1) {
840         stat64_add(&mig_stats.dirty_sync_missed_zero_copy, 1);
841     }
842 
843     return ret;
844 }
845 
846 int multifd_send_sync_main(void)
847 {
848     int i;
849     bool flush_zero_copy;
850 
851     if (!migrate_multifd()) {
852         return 0;
853     }
854     if (multifd_send_state->pages->num) {
855         if (!multifd_send_pages()) {
856             error_report("%s: multifd_send_pages fail", __func__);
857             return -1;
858         }
859     }
860 
861     flush_zero_copy = migrate_zero_copy_send();
862 
863     for (i = 0; i < migrate_multifd_channels(); i++) {
864         MultiFDSendParams *p = &multifd_send_state->params[i];
865 
866         if (multifd_send_should_exit()) {
867             return -1;
868         }
869 
870         trace_multifd_send_sync_main_signal(p->id);
871 
872         /*
873          * We should be the only user so far, so not possible to be set by
874          * others concurrently.
875          */
876         assert(qatomic_read(&p->pending_sync) == false);
877         qatomic_set(&p->pending_sync, true);
878         qemu_sem_post(&p->sem);
879     }
880     for (i = 0; i < migrate_multifd_channels(); i++) {
881         MultiFDSendParams *p = &multifd_send_state->params[i];
882 
883         if (multifd_send_should_exit()) {
884             return -1;
885         }
886 
887         qemu_sem_wait(&multifd_send_state->channels_ready);
888         trace_multifd_send_sync_main_wait(p->id);
889         qemu_sem_wait(&p->sem_sync);
890 
891         if (flush_zero_copy && p->c && (multifd_zero_copy_flush(p->c) < 0)) {
892             return -1;
893         }
894     }
895     trace_multifd_send_sync_main(multifd_send_state->packet_num);
896 
897     return 0;
898 }
899 
900 static void *multifd_send_thread(void *opaque)
901 {
902     MultiFDSendParams *p = opaque;
903     MigrationThread *thread = NULL;
904     Error *local_err = NULL;
905     int ret = 0;
906     bool use_packets = multifd_use_packets();
907 
908     thread = migration_threads_add(p->name, qemu_get_thread_id());
909 
910     trace_multifd_send_thread_start(p->id);
911     rcu_register_thread();
912 
913     if (use_packets) {
914         if (multifd_send_initial_packet(p, &local_err) < 0) {
915             ret = -1;
916             goto out;
917         }
918     }
919 
920     while (true) {
921         qemu_sem_post(&multifd_send_state->channels_ready);
922         qemu_sem_wait(&p->sem);
923 
924         if (multifd_send_should_exit()) {
925             break;
926         }
927 
928         /*
929          * Read pending_job flag before p->pages.  Pairs with the
930          * qatomic_store_release() in multifd_send_pages().
931          */
932         if (qatomic_load_acquire(&p->pending_job)) {
933             MultiFDPages_t *pages = p->pages;
934 
935             p->iovs_num = 0;
936             assert(pages->num);
937 
938             ret = multifd_send_state->ops->send_prepare(p, &local_err);
939             if (ret != 0) {
940                 break;
941             }
942 
943             if (migrate_mapped_ram()) {
944                 ret = file_write_ramblock_iov(p->c, p->iov, p->iovs_num,
945                                               p->pages->block, &local_err);
946             } else {
947                 ret = qio_channel_writev_full_all(p->c, p->iov, p->iovs_num,
948                                                   NULL, 0, p->write_flags,
949                                                   &local_err);
950             }
951 
952             if (ret != 0) {
953                 break;
954             }
955 
956             stat64_add(&mig_stats.multifd_bytes,
957                        p->next_packet_size + p->packet_len);
958             stat64_add(&mig_stats.normal_pages, pages->normal_num);
959             stat64_add(&mig_stats.zero_pages, pages->num - pages->normal_num);
960 
961             multifd_pages_reset(p->pages);
962             p->next_packet_size = 0;
963 
964             /*
965              * Making sure p->pages is published before saying "we're
966              * free".  Pairs with the smp_mb_acquire() in
967              * multifd_send_pages().
968              */
969             qatomic_store_release(&p->pending_job, false);
970         } else {
971             /*
972              * If not a normal job, must be a sync request.  Note that
973              * pending_sync is a standalone flag (unlike pending_job), so
974              * it doesn't require explicit memory barriers.
975              */
976             assert(qatomic_read(&p->pending_sync));
977 
978             if (use_packets) {
979                 p->flags = MULTIFD_FLAG_SYNC;
980                 multifd_send_fill_packet(p);
981                 ret = qio_channel_write_all(p->c, (void *)p->packet,
982                                             p->packet_len, &local_err);
983                 if (ret != 0) {
984                     break;
985                 }
986                 /* p->next_packet_size will always be zero for a SYNC packet */
987                 stat64_add(&mig_stats.multifd_bytes, p->packet_len);
988                 p->flags = 0;
989             }
990 
991             qatomic_set(&p->pending_sync, false);
992             qemu_sem_post(&p->sem_sync);
993         }
994     }
995 
996 out:
997     if (ret) {
998         assert(local_err);
999         trace_multifd_send_error(p->id);
1000         multifd_send_set_error(local_err);
1001         multifd_send_kick_main(p);
1002         error_free(local_err);
1003     }
1004 
1005     rcu_unregister_thread();
1006     migration_threads_remove(thread);
1007     trace_multifd_send_thread_end(p->id, p->packets_sent, p->total_normal_pages,
1008                                   p->total_zero_pages);
1009 
1010     return NULL;
1011 }
1012 
1013 static void multifd_new_send_channel_async(QIOTask *task, gpointer opaque);
1014 
1015 typedef struct {
1016     MultiFDSendParams *p;
1017     QIOChannelTLS *tioc;
1018 } MultiFDTLSThreadArgs;
1019 
1020 static void *multifd_tls_handshake_thread(void *opaque)
1021 {
1022     MultiFDTLSThreadArgs *args = opaque;
1023 
1024     qio_channel_tls_handshake(args->tioc,
1025                               multifd_new_send_channel_async,
1026                               args->p,
1027                               NULL,
1028                               NULL);
1029     g_free(args);
1030 
1031     return NULL;
1032 }
1033 
1034 static bool multifd_tls_channel_connect(MultiFDSendParams *p,
1035                                         QIOChannel *ioc,
1036                                         Error **errp)
1037 {
1038     MigrationState *s = migrate_get_current();
1039     const char *hostname = s->hostname;
1040     MultiFDTLSThreadArgs *args;
1041     QIOChannelTLS *tioc;
1042 
1043     tioc = migration_tls_client_create(ioc, hostname, errp);
1044     if (!tioc) {
1045         return false;
1046     }
1047 
1048     /*
1049      * Ownership of the socket channel now transfers to the newly
1050      * created TLS channel, which has already taken a reference.
1051      */
1052     object_unref(OBJECT(ioc));
1053     trace_multifd_tls_outgoing_handshake_start(ioc, tioc, hostname);
1054     qio_channel_set_name(QIO_CHANNEL(tioc), "multifd-tls-outgoing");
1055 
1056     args = g_new0(MultiFDTLSThreadArgs, 1);
1057     args->tioc = tioc;
1058     args->p = p;
1059 
1060     p->tls_thread_created = true;
1061     qemu_thread_create(&p->tls_thread, "multifd-tls-handshake-worker",
1062                        multifd_tls_handshake_thread, args,
1063                        QEMU_THREAD_JOINABLE);
1064     return true;
1065 }
1066 
1067 void multifd_channel_connect(MultiFDSendParams *p, QIOChannel *ioc)
1068 {
1069     qio_channel_set_delay(ioc, false);
1070 
1071     migration_ioc_register_yank(ioc);
1072     /* Setup p->c only if the channel is completely setup */
1073     p->c = ioc;
1074 
1075     p->thread_created = true;
1076     qemu_thread_create(&p->thread, p->name, multifd_send_thread, p,
1077                        QEMU_THREAD_JOINABLE);
1078 }
1079 
1080 /*
1081  * When TLS is enabled this function is called once to establish the
1082  * TLS connection and a second time after the TLS handshake to create
1083  * the multifd channel. Without TLS it goes straight into the channel
1084  * creation.
1085  */
1086 static void multifd_new_send_channel_async(QIOTask *task, gpointer opaque)
1087 {
1088     MultiFDSendParams *p = opaque;
1089     QIOChannel *ioc = QIO_CHANNEL(qio_task_get_source(task));
1090     Error *local_err = NULL;
1091     bool ret;
1092 
1093     trace_multifd_new_send_channel_async(p->id);
1094 
1095     if (qio_task_propagate_error(task, &local_err)) {
1096         ret = false;
1097         goto out;
1098     }
1099 
1100     trace_multifd_set_outgoing_channel(ioc, object_get_typename(OBJECT(ioc)),
1101                                        migrate_get_current()->hostname);
1102 
1103     if (migrate_channel_requires_tls_upgrade(ioc)) {
1104         ret = multifd_tls_channel_connect(p, ioc, &local_err);
1105         if (ret) {
1106             return;
1107         }
1108     } else {
1109         multifd_channel_connect(p, ioc);
1110         ret = true;
1111     }
1112 
1113 out:
1114     /*
1115      * Here we're not interested whether creation succeeded, only that
1116      * it happened at all.
1117      */
1118     multifd_send_channel_created();
1119 
1120     if (ret) {
1121         return;
1122     }
1123 
1124     trace_multifd_new_send_channel_async_error(p->id, local_err);
1125     multifd_send_set_error(local_err);
1126     /*
1127      * For error cases (TLS or non-TLS), IO channel is always freed here
1128      * rather than when cleanup multifd: since p->c is not set, multifd
1129      * cleanup code doesn't even know its existence.
1130      */
1131     object_unref(OBJECT(ioc));
1132     error_free(local_err);
1133 }
1134 
1135 static bool multifd_new_send_channel_create(gpointer opaque, Error **errp)
1136 {
1137     if (!multifd_use_packets()) {
1138         return file_send_channel_create(opaque, errp);
1139     }
1140 
1141     socket_send_channel_create(multifd_new_send_channel_async, opaque);
1142     return true;
1143 }
1144 
1145 bool multifd_send_setup(void)
1146 {
1147     MigrationState *s = migrate_get_current();
1148     Error *local_err = NULL;
1149     int thread_count, ret = 0;
1150     uint32_t page_count = MULTIFD_PACKET_SIZE / qemu_target_page_size();
1151     bool use_packets = multifd_use_packets();
1152     uint8_t i;
1153 
1154     if (!migrate_multifd()) {
1155         return true;
1156     }
1157 
1158     thread_count = migrate_multifd_channels();
1159     multifd_send_state = g_malloc0(sizeof(*multifd_send_state));
1160     multifd_send_state->params = g_new0(MultiFDSendParams, thread_count);
1161     multifd_send_state->pages = multifd_pages_init(page_count);
1162     qemu_sem_init(&multifd_send_state->channels_created, 0);
1163     qemu_sem_init(&multifd_send_state->channels_ready, 0);
1164     qatomic_set(&multifd_send_state->exiting, 0);
1165     multifd_send_state->ops = multifd_ops[migrate_multifd_compression()];
1166 
1167     for (i = 0; i < thread_count; i++) {
1168         MultiFDSendParams *p = &multifd_send_state->params[i];
1169 
1170         qemu_sem_init(&p->sem, 0);
1171         qemu_sem_init(&p->sem_sync, 0);
1172         p->id = i;
1173         p->pages = multifd_pages_init(page_count);
1174 
1175         if (use_packets) {
1176             p->packet_len = sizeof(MultiFDPacket_t)
1177                           + sizeof(uint64_t) * page_count;
1178             p->packet = g_malloc0(p->packet_len);
1179             p->packet->magic = cpu_to_be32(MULTIFD_MAGIC);
1180             p->packet->version = cpu_to_be32(MULTIFD_VERSION);
1181 
1182             /* We need one extra place for the packet header */
1183             p->iov = g_new0(struct iovec, page_count + 1);
1184         } else {
1185             p->iov = g_new0(struct iovec, page_count);
1186         }
1187         p->name = g_strdup_printf("multifdsend_%d", i);
1188         p->page_size = qemu_target_page_size();
1189         p->page_count = page_count;
1190         p->write_flags = 0;
1191 
1192         if (!multifd_new_send_channel_create(p, &local_err)) {
1193             return false;
1194         }
1195     }
1196 
1197     /*
1198      * Wait until channel creation has started for all channels. The
1199      * creation can still fail, but no more channels will be created
1200      * past this point.
1201      */
1202     for (i = 0; i < thread_count; i++) {
1203         qemu_sem_wait(&multifd_send_state->channels_created);
1204     }
1205 
1206     for (i = 0; i < thread_count; i++) {
1207         MultiFDSendParams *p = &multifd_send_state->params[i];
1208 
1209         ret = multifd_send_state->ops->send_setup(p, &local_err);
1210         if (ret) {
1211             break;
1212         }
1213     }
1214 
1215     if (ret) {
1216         migrate_set_error(s, local_err);
1217         error_report_err(local_err);
1218         migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
1219                           MIGRATION_STATUS_FAILED);
1220         return false;
1221     }
1222 
1223     return true;
1224 }
1225 
1226 bool multifd_recv(void)
1227 {
1228     int i;
1229     static int next_recv_channel;
1230     MultiFDRecvParams *p = NULL;
1231     MultiFDRecvData *data = multifd_recv_state->data;
1232 
1233     /*
1234      * next_channel can remain from a previous migration that was
1235      * using more channels, so ensure it doesn't overflow if the
1236      * limit is lower now.
1237      */
1238     next_recv_channel %= migrate_multifd_channels();
1239     for (i = next_recv_channel;; i = (i + 1) % migrate_multifd_channels()) {
1240         if (multifd_recv_should_exit()) {
1241             return false;
1242         }
1243 
1244         p = &multifd_recv_state->params[i];
1245 
1246         if (qatomic_read(&p->pending_job) == false) {
1247             next_recv_channel = (i + 1) % migrate_multifd_channels();
1248             break;
1249         }
1250     }
1251 
1252     /*
1253      * Order pending_job read before manipulating p->data below. Pairs
1254      * with qatomic_store_release() at multifd_recv_thread().
1255      */
1256     smp_mb_acquire();
1257 
1258     assert(!p->data->size);
1259     multifd_recv_state->data = p->data;
1260     p->data = data;
1261 
1262     /*
1263      * Order p->data update before setting pending_job. Pairs with
1264      * qatomic_load_acquire() at multifd_recv_thread().
1265      */
1266     qatomic_store_release(&p->pending_job, true);
1267     qemu_sem_post(&p->sem);
1268 
1269     return true;
1270 }
1271 
1272 MultiFDRecvData *multifd_get_recv_data(void)
1273 {
1274     return multifd_recv_state->data;
1275 }
1276 
1277 static void multifd_recv_terminate_threads(Error *err)
1278 {
1279     int i;
1280 
1281     trace_multifd_recv_terminate_threads(err != NULL);
1282 
1283     if (qatomic_xchg(&multifd_recv_state->exiting, 1)) {
1284         return;
1285     }
1286 
1287     if (err) {
1288         MigrationState *s = migrate_get_current();
1289         migrate_set_error(s, err);
1290         if (s->state == MIGRATION_STATUS_SETUP ||
1291             s->state == MIGRATION_STATUS_ACTIVE) {
1292             migrate_set_state(&s->state, s->state,
1293                               MIGRATION_STATUS_FAILED);
1294         }
1295     }
1296 
1297     for (i = 0; i < migrate_multifd_channels(); i++) {
1298         MultiFDRecvParams *p = &multifd_recv_state->params[i];
1299 
1300         /*
1301          * The migration thread and channels interact differently
1302          * depending on the presence of packets.
1303          */
1304         if (multifd_use_packets()) {
1305             /*
1306              * The channel receives as long as there are packets. When
1307              * packets end (i.e. MULTIFD_FLAG_SYNC is reached), the
1308              * channel waits for the migration thread to sync. If the
1309              * sync never happens, do it here.
1310              */
1311             qemu_sem_post(&p->sem_sync);
1312         } else {
1313             /*
1314              * The channel waits for the migration thread to give it
1315              * work. When the migration thread runs out of work, it
1316              * releases the channel and waits for any pending work to
1317              * finish. If we reach here (e.g. due to error) before the
1318              * work runs out, release the channel.
1319              */
1320             qemu_sem_post(&p->sem);
1321         }
1322 
1323         /*
1324          * We could arrive here for two reasons:
1325          *  - normal quit, i.e. everything went fine, just finished
1326          *  - error quit: We close the channels so the channel threads
1327          *    finish the qio_channel_read_all_eof()
1328          */
1329         if (p->c) {
1330             qio_channel_shutdown(p->c, QIO_CHANNEL_SHUTDOWN_BOTH, NULL);
1331         }
1332     }
1333 }
1334 
1335 void multifd_recv_shutdown(void)
1336 {
1337     if (migrate_multifd()) {
1338         multifd_recv_terminate_threads(NULL);
1339     }
1340 }
1341 
1342 static void multifd_recv_cleanup_channel(MultiFDRecvParams *p)
1343 {
1344     migration_ioc_unregister_yank(p->c);
1345     object_unref(OBJECT(p->c));
1346     p->c = NULL;
1347     qemu_mutex_destroy(&p->mutex);
1348     qemu_sem_destroy(&p->sem_sync);
1349     qemu_sem_destroy(&p->sem);
1350     g_free(p->name);
1351     p->name = NULL;
1352     p->packet_len = 0;
1353     g_free(p->packet);
1354     p->packet = NULL;
1355     g_free(p->iov);
1356     p->iov = NULL;
1357     g_free(p->normal);
1358     p->normal = NULL;
1359     g_free(p->zero);
1360     p->zero = NULL;
1361     multifd_recv_state->ops->recv_cleanup(p);
1362 }
1363 
1364 static void multifd_recv_cleanup_state(void)
1365 {
1366     qemu_sem_destroy(&multifd_recv_state->sem_sync);
1367     g_free(multifd_recv_state->params);
1368     multifd_recv_state->params = NULL;
1369     g_free(multifd_recv_state->data);
1370     multifd_recv_state->data = NULL;
1371     g_free(multifd_recv_state);
1372     multifd_recv_state = NULL;
1373 }
1374 
1375 void multifd_recv_cleanup(void)
1376 {
1377     int i;
1378 
1379     if (!migrate_multifd()) {
1380         return;
1381     }
1382     multifd_recv_terminate_threads(NULL);
1383     for (i = 0; i < migrate_multifd_channels(); i++) {
1384         MultiFDRecvParams *p = &multifd_recv_state->params[i];
1385 
1386         if (p->thread_created) {
1387             qemu_thread_join(&p->thread);
1388         }
1389     }
1390     for (i = 0; i < migrate_multifd_channels(); i++) {
1391         multifd_recv_cleanup_channel(&multifd_recv_state->params[i]);
1392     }
1393     multifd_recv_cleanup_state();
1394 }
1395 
1396 void multifd_recv_sync_main(void)
1397 {
1398     int thread_count = migrate_multifd_channels();
1399     bool file_based = !multifd_use_packets();
1400     int i;
1401 
1402     if (!migrate_multifd()) {
1403         return;
1404     }
1405 
1406     /*
1407      * File-based channels don't use packets and therefore need to
1408      * wait for more work. Release them to start the sync.
1409      */
1410     if (file_based) {
1411         for (i = 0; i < thread_count; i++) {
1412             MultiFDRecvParams *p = &multifd_recv_state->params[i];
1413 
1414             trace_multifd_recv_sync_main_signal(p->id);
1415             qemu_sem_post(&p->sem);
1416         }
1417     }
1418 
1419     /*
1420      * Initiate the synchronization by waiting for all channels.
1421      *
1422      * For socket-based migration this means each channel has received
1423      * the SYNC packet on the stream.
1424      *
1425      * For file-based migration this means each channel is done with
1426      * the work (pending_job=false).
1427      */
1428     for (i = 0; i < thread_count; i++) {
1429         trace_multifd_recv_sync_main_wait(i);
1430         qemu_sem_wait(&multifd_recv_state->sem_sync);
1431     }
1432 
1433     if (file_based) {
1434         /*
1435          * For file-based loading is done in one iteration. We're
1436          * done.
1437          */
1438         return;
1439     }
1440 
1441     /*
1442      * Sync done. Release the channels for the next iteration.
1443      */
1444     for (i = 0; i < thread_count; i++) {
1445         MultiFDRecvParams *p = &multifd_recv_state->params[i];
1446 
1447         WITH_QEMU_LOCK_GUARD(&p->mutex) {
1448             if (multifd_recv_state->packet_num < p->packet_num) {
1449                 multifd_recv_state->packet_num = p->packet_num;
1450             }
1451         }
1452         trace_multifd_recv_sync_main_signal(p->id);
1453         qemu_sem_post(&p->sem_sync);
1454     }
1455     trace_multifd_recv_sync_main(multifd_recv_state->packet_num);
1456 }
1457 
1458 static void *multifd_recv_thread(void *opaque)
1459 {
1460     MultiFDRecvParams *p = opaque;
1461     Error *local_err = NULL;
1462     bool use_packets = multifd_use_packets();
1463     int ret;
1464 
1465     trace_multifd_recv_thread_start(p->id);
1466     rcu_register_thread();
1467 
1468     while (true) {
1469         uint32_t flags = 0;
1470         bool has_data = false;
1471         p->normal_num = 0;
1472 
1473         if (use_packets) {
1474             if (multifd_recv_should_exit()) {
1475                 break;
1476             }
1477 
1478             ret = qio_channel_read_all_eof(p->c, (void *)p->packet,
1479                                            p->packet_len, &local_err);
1480             if (ret == 0 || ret == -1) {   /* 0: EOF  -1: Error */
1481                 break;
1482             }
1483 
1484             qemu_mutex_lock(&p->mutex);
1485             ret = multifd_recv_unfill_packet(p, &local_err);
1486             if (ret) {
1487                 qemu_mutex_unlock(&p->mutex);
1488                 break;
1489             }
1490 
1491             flags = p->flags;
1492             /* recv methods don't know how to handle the SYNC flag */
1493             p->flags &= ~MULTIFD_FLAG_SYNC;
1494             has_data = p->normal_num || p->zero_num;
1495             qemu_mutex_unlock(&p->mutex);
1496         } else {
1497             /*
1498              * No packets, so we need to wait for the vmstate code to
1499              * give us work.
1500              */
1501             qemu_sem_wait(&p->sem);
1502 
1503             if (multifd_recv_should_exit()) {
1504                 break;
1505             }
1506 
1507             /* pairs with qatomic_store_release() at multifd_recv() */
1508             if (!qatomic_load_acquire(&p->pending_job)) {
1509                 /*
1510                  * Migration thread did not send work, this is
1511                  * equivalent to pending_sync on the sending
1512                  * side. Post sem_sync to notify we reached this
1513                  * point.
1514                  */
1515                 qemu_sem_post(&multifd_recv_state->sem_sync);
1516                 continue;
1517             }
1518 
1519             has_data = !!p->data->size;
1520         }
1521 
1522         if (has_data) {
1523             ret = multifd_recv_state->ops->recv(p, &local_err);
1524             if (ret != 0) {
1525                 break;
1526             }
1527         }
1528 
1529         if (use_packets) {
1530             if (flags & MULTIFD_FLAG_SYNC) {
1531                 qemu_sem_post(&multifd_recv_state->sem_sync);
1532                 qemu_sem_wait(&p->sem_sync);
1533             }
1534         } else {
1535             p->total_normal_pages += p->data->size / qemu_target_page_size();
1536             p->data->size = 0;
1537             /*
1538              * Order data->size update before clearing
1539              * pending_job. Pairs with smp_mb_acquire() at
1540              * multifd_recv().
1541              */
1542             qatomic_store_release(&p->pending_job, false);
1543         }
1544     }
1545 
1546     if (local_err) {
1547         multifd_recv_terminate_threads(local_err);
1548         error_free(local_err);
1549     }
1550 
1551     rcu_unregister_thread();
1552     trace_multifd_recv_thread_end(p->id, p->packets_recved,
1553                                   p->total_normal_pages,
1554                                   p->total_zero_pages);
1555 
1556     return NULL;
1557 }
1558 
1559 int multifd_recv_setup(Error **errp)
1560 {
1561     int thread_count;
1562     uint32_t page_count = MULTIFD_PACKET_SIZE / qemu_target_page_size();
1563     bool use_packets = multifd_use_packets();
1564     uint8_t i;
1565 
1566     /*
1567      * Return successfully if multiFD recv state is already initialised
1568      * or multiFD is not enabled.
1569      */
1570     if (multifd_recv_state || !migrate_multifd()) {
1571         return 0;
1572     }
1573 
1574     thread_count = migrate_multifd_channels();
1575     multifd_recv_state = g_malloc0(sizeof(*multifd_recv_state));
1576     multifd_recv_state->params = g_new0(MultiFDRecvParams, thread_count);
1577 
1578     multifd_recv_state->data = g_new0(MultiFDRecvData, 1);
1579     multifd_recv_state->data->size = 0;
1580 
1581     qatomic_set(&multifd_recv_state->count, 0);
1582     qatomic_set(&multifd_recv_state->exiting, 0);
1583     qemu_sem_init(&multifd_recv_state->sem_sync, 0);
1584     multifd_recv_state->ops = multifd_ops[migrate_multifd_compression()];
1585 
1586     for (i = 0; i < thread_count; i++) {
1587         MultiFDRecvParams *p = &multifd_recv_state->params[i];
1588 
1589         qemu_mutex_init(&p->mutex);
1590         qemu_sem_init(&p->sem_sync, 0);
1591         qemu_sem_init(&p->sem, 0);
1592         p->pending_job = false;
1593         p->id = i;
1594 
1595         p->data = g_new0(MultiFDRecvData, 1);
1596         p->data->size = 0;
1597 
1598         if (use_packets) {
1599             p->packet_len = sizeof(MultiFDPacket_t)
1600                 + sizeof(uint64_t) * page_count;
1601             p->packet = g_malloc0(p->packet_len);
1602         }
1603         p->name = g_strdup_printf("multifdrecv_%d", i);
1604         p->iov = g_new0(struct iovec, page_count);
1605         p->normal = g_new0(ram_addr_t, page_count);
1606         p->zero = g_new0(ram_addr_t, page_count);
1607         p->page_count = page_count;
1608         p->page_size = qemu_target_page_size();
1609     }
1610 
1611     for (i = 0; i < thread_count; i++) {
1612         MultiFDRecvParams *p = &multifd_recv_state->params[i];
1613         int ret;
1614 
1615         ret = multifd_recv_state->ops->recv_setup(p, errp);
1616         if (ret) {
1617             return ret;
1618         }
1619     }
1620     return 0;
1621 }
1622 
1623 bool multifd_recv_all_channels_created(void)
1624 {
1625     int thread_count = migrate_multifd_channels();
1626 
1627     if (!migrate_multifd()) {
1628         return true;
1629     }
1630 
1631     if (!multifd_recv_state) {
1632         /* Called before any connections created */
1633         return false;
1634     }
1635 
1636     return thread_count == qatomic_read(&multifd_recv_state->count);
1637 }
1638 
1639 /*
1640  * Try to receive all multifd channels to get ready for the migration.
1641  * Sets @errp when failing to receive the current channel.
1642  */
1643 void multifd_recv_new_channel(QIOChannel *ioc, Error **errp)
1644 {
1645     MultiFDRecvParams *p;
1646     Error *local_err = NULL;
1647     bool use_packets = multifd_use_packets();
1648     int id;
1649 
1650     if (use_packets) {
1651         id = multifd_recv_initial_packet(ioc, &local_err);
1652         if (id < 0) {
1653             multifd_recv_terminate_threads(local_err);
1654             error_propagate_prepend(errp, local_err,
1655                                     "failed to receive packet"
1656                                     " via multifd channel %d: ",
1657                                     qatomic_read(&multifd_recv_state->count));
1658             return;
1659         }
1660         trace_multifd_recv_new_channel(id);
1661     } else {
1662         id = qatomic_read(&multifd_recv_state->count);
1663     }
1664 
1665     p = &multifd_recv_state->params[id];
1666     if (p->c != NULL) {
1667         error_setg(&local_err, "multifd: received id '%d' already setup'",
1668                    id);
1669         multifd_recv_terminate_threads(local_err);
1670         error_propagate(errp, local_err);
1671         return;
1672     }
1673     p->c = ioc;
1674     object_ref(OBJECT(ioc));
1675 
1676     p->thread_created = true;
1677     qemu_thread_create(&p->thread, p->name, multifd_recv_thread, p,
1678                        QEMU_THREAD_JOINABLE);
1679     qatomic_inc(&multifd_recv_state->count);
1680 }
1681 
1682 bool multifd_send_prepare_common(MultiFDSendParams *p)
1683 {
1684     multifd_send_zero_page_detect(p);
1685 
1686     if (!p->pages->normal_num) {
1687         p->next_packet_size = 0;
1688         return false;
1689     }
1690 
1691     multifd_send_prepare_header(p);
1692 
1693     return true;
1694 }
1695