xref: /openbmc/qemu/block/replication.c (revision 7200fb21)
1 /*
2  * Replication Block filter
3  *
4  * Copyright (c) 2016 HUAWEI TECHNOLOGIES CO., LTD.
5  * Copyright (c) 2016 Intel Corporation
6  * Copyright (c) 2016 FUJITSU LIMITED
7  *
8  * Author:
9  *   Wen Congyang <wency@cn.fujitsu.com>
10  *
11  * This work is licensed under the terms of the GNU GPL, version 2 or later.
12  * See the COPYING file in the top-level directory.
13  */
14 
15 #include "qemu/osdep.h"
16 #include "qemu/module.h"
17 #include "qemu/option.h"
18 #include "block/nbd.h"
19 #include "block/blockjob.h"
20 #include "block/block_int.h"
21 #include "block/block_backup.h"
22 #include "sysemu/block-backend.h"
23 #include "qapi/error.h"
24 #include "qapi/qmp/qdict.h"
25 #include "block/replication.h"
26 
27 typedef enum {
28     BLOCK_REPLICATION_NONE,             /* block replication is not started */
29     BLOCK_REPLICATION_RUNNING,          /* block replication is running */
30     BLOCK_REPLICATION_FAILOVER,         /* failover is running in background */
31     BLOCK_REPLICATION_FAILOVER_FAILED,  /* failover failed */
32     BLOCK_REPLICATION_DONE,             /* block replication is done */
33 } ReplicationStage;
34 
35 typedef struct BDRVReplicationState {
36     ReplicationMode mode;
37     ReplicationStage stage;
38     BlockJob *commit_job;
39     BdrvChild *hidden_disk;
40     BdrvChild *secondary_disk;
41     BlockJob *backup_job;
42     char *top_id;
43     ReplicationState *rs;
44     Error *blocker;
45     bool orig_hidden_read_only;
46     bool orig_secondary_read_only;
47     int error;
48 } BDRVReplicationState;
49 
50 static void replication_start(ReplicationState *rs, ReplicationMode mode,
51                               Error **errp);
52 static void replication_do_checkpoint(ReplicationState *rs, Error **errp);
53 static void replication_get_error(ReplicationState *rs, Error **errp);
54 static void replication_stop(ReplicationState *rs, bool failover,
55                              Error **errp);
56 
57 #define REPLICATION_MODE        "mode"
58 #define REPLICATION_TOP_ID      "top-id"
59 static QemuOptsList replication_runtime_opts = {
60     .name = "replication",
61     .head = QTAILQ_HEAD_INITIALIZER(replication_runtime_opts.head),
62     .desc = {
63         {
64             .name = REPLICATION_MODE,
65             .type = QEMU_OPT_STRING,
66         },
67         {
68             .name = REPLICATION_TOP_ID,
69             .type = QEMU_OPT_STRING,
70         },
71         { /* end of list */ }
72     },
73 };
74 
75 static ReplicationOps replication_ops = {
76     .start = replication_start,
77     .checkpoint = replication_do_checkpoint,
78     .get_error = replication_get_error,
79     .stop = replication_stop,
80 };
81 
82 static int replication_open(BlockDriverState *bs, QDict *options,
83                             int flags, Error **errp)
84 {
85     int ret;
86     BDRVReplicationState *s = bs->opaque;
87     QemuOpts *opts = NULL;
88     const char *mode;
89     const char *top_id;
90 
91     ret = bdrv_open_file_child(NULL, options, "file", bs, errp);
92     if (ret < 0) {
93         return ret;
94     }
95 
96     ret = -EINVAL;
97     opts = qemu_opts_create(&replication_runtime_opts, NULL, 0, &error_abort);
98     if (!qemu_opts_absorb_qdict(opts, options, errp)) {
99         goto fail;
100     }
101 
102     mode = qemu_opt_get(opts, REPLICATION_MODE);
103     if (!mode) {
104         error_setg(errp, "Missing the option mode");
105         goto fail;
106     }
107 
108     if (!strcmp(mode, "primary")) {
109         s->mode = REPLICATION_MODE_PRIMARY;
110         top_id = qemu_opt_get(opts, REPLICATION_TOP_ID);
111         if (top_id) {
112             error_setg(errp,
113                        "The primary side does not support option top-id");
114             goto fail;
115         }
116     } else if (!strcmp(mode, "secondary")) {
117         s->mode = REPLICATION_MODE_SECONDARY;
118         top_id = qemu_opt_get(opts, REPLICATION_TOP_ID);
119         s->top_id = g_strdup(top_id);
120         if (!s->top_id) {
121             error_setg(errp, "Missing the option top-id");
122             goto fail;
123         }
124     } else {
125         error_setg(errp,
126                    "The option mode's value should be primary or secondary");
127         goto fail;
128     }
129 
130     s->rs = replication_new(bs, &replication_ops);
131 
132     ret = 0;
133 
134 fail:
135     qemu_opts_del(opts);
136     return ret;
137 }
138 
139 static void replication_close(BlockDriverState *bs)
140 {
141     BDRVReplicationState *s = bs->opaque;
142     Job *commit_job;
143     GLOBAL_STATE_CODE();
144 
145     if (s->stage == BLOCK_REPLICATION_RUNNING) {
146         replication_stop(s->rs, false, NULL);
147     }
148     if (s->stage == BLOCK_REPLICATION_FAILOVER) {
149         commit_job = &s->commit_job->job;
150         assert(commit_job->aio_context == qemu_get_current_aio_context());
151         job_cancel_sync(commit_job, false);
152     }
153 
154     if (s->mode == REPLICATION_MODE_SECONDARY) {
155         g_free(s->top_id);
156     }
157 
158     replication_remove(s->rs);
159 }
160 
161 static void replication_child_perm(BlockDriverState *bs, BdrvChild *c,
162                                    BdrvChildRole role,
163                                    BlockReopenQueue *reopen_queue,
164                                    uint64_t perm, uint64_t shared,
165                                    uint64_t *nperm, uint64_t *nshared)
166 {
167     if (role & BDRV_CHILD_PRIMARY) {
168         *nperm = BLK_PERM_CONSISTENT_READ;
169     } else {
170         *nperm = 0;
171     }
172 
173     if ((bs->open_flags & (BDRV_O_INACTIVE | BDRV_O_RDWR)) == BDRV_O_RDWR) {
174         *nperm |= BLK_PERM_WRITE;
175     }
176     *nshared = BLK_PERM_CONSISTENT_READ
177                | BLK_PERM_WRITE
178                | BLK_PERM_WRITE_UNCHANGED;
179     return;
180 }
181 
182 static int64_t coroutine_fn GRAPH_RDLOCK
183 replication_co_getlength(BlockDriverState *bs)
184 {
185     return bdrv_co_getlength(bs->file->bs);
186 }
187 
188 static int replication_get_io_status(BDRVReplicationState *s)
189 {
190     switch (s->stage) {
191     case BLOCK_REPLICATION_NONE:
192         return -EIO;
193     case BLOCK_REPLICATION_RUNNING:
194         return 0;
195     case BLOCK_REPLICATION_FAILOVER:
196         return s->mode == REPLICATION_MODE_PRIMARY ? -EIO : 0;
197     case BLOCK_REPLICATION_FAILOVER_FAILED:
198         return s->mode == REPLICATION_MODE_PRIMARY ? -EIO : 1;
199     case BLOCK_REPLICATION_DONE:
200         /*
201          * active commit job completes, and active disk and secondary_disk
202          * is swapped, so we can operate bs->file directly
203          */
204         return s->mode == REPLICATION_MODE_PRIMARY ? -EIO : 0;
205     default:
206         abort();
207     }
208 }
209 
210 static int replication_return_value(BDRVReplicationState *s, int ret)
211 {
212     if (s->mode == REPLICATION_MODE_SECONDARY) {
213         return ret;
214     }
215 
216     if (ret < 0) {
217         s->error = ret;
218         ret = 0;
219     }
220 
221     return ret;
222 }
223 
224 static int coroutine_fn GRAPH_RDLOCK
225 replication_co_readv(BlockDriverState *bs, int64_t sector_num,
226                      int remaining_sectors, QEMUIOVector *qiov)
227 {
228     BDRVReplicationState *s = bs->opaque;
229     int ret;
230 
231     if (s->mode == REPLICATION_MODE_PRIMARY) {
232         /* We only use it to forward primary write requests */
233         return -EIO;
234     }
235 
236     ret = replication_get_io_status(s);
237     if (ret < 0) {
238         return ret;
239     }
240 
241     ret = bdrv_co_preadv(bs->file, sector_num * BDRV_SECTOR_SIZE,
242                          remaining_sectors * BDRV_SECTOR_SIZE, qiov, 0);
243 
244     return replication_return_value(s, ret);
245 }
246 
247 static int coroutine_fn GRAPH_RDLOCK
248 replication_co_writev(BlockDriverState *bs, int64_t sector_num,
249                       int remaining_sectors, QEMUIOVector *qiov, int flags)
250 {
251     BDRVReplicationState *s = bs->opaque;
252     QEMUIOVector hd_qiov;
253     uint64_t bytes_done = 0;
254     BdrvChild *top = bs->file;
255     BdrvChild *base = s->secondary_disk;
256     BdrvChild *target;
257     int ret;
258     int64_t n;
259 
260     ret = replication_get_io_status(s);
261     if (ret < 0) {
262         goto out;
263     }
264 
265     if (ret == 0) {
266         ret = bdrv_co_pwritev(top, sector_num * BDRV_SECTOR_SIZE,
267                               remaining_sectors * BDRV_SECTOR_SIZE, qiov, 0);
268         return replication_return_value(s, ret);
269     }
270 
271     /*
272      * Failover failed, only write to active disk if the sectors
273      * have already been allocated in active disk/hidden disk.
274      */
275     qemu_iovec_init(&hd_qiov, qiov->niov);
276     while (remaining_sectors > 0) {
277         int64_t count;
278 
279         ret = bdrv_co_is_allocated_above(top->bs, base->bs, false,
280                                          sector_num * BDRV_SECTOR_SIZE,
281                                          remaining_sectors * BDRV_SECTOR_SIZE,
282                                          &count);
283         if (ret < 0) {
284             goto out1;
285         }
286 
287         assert(QEMU_IS_ALIGNED(count, BDRV_SECTOR_SIZE));
288         n = count >> BDRV_SECTOR_BITS;
289         qemu_iovec_reset(&hd_qiov);
290         qemu_iovec_concat(&hd_qiov, qiov, bytes_done, count);
291 
292         target = ret ? top : base;
293         ret = bdrv_co_pwritev(target, sector_num * BDRV_SECTOR_SIZE,
294                               n * BDRV_SECTOR_SIZE, &hd_qiov, 0);
295         if (ret < 0) {
296             goto out1;
297         }
298 
299         remaining_sectors -= n;
300         sector_num += n;
301         bytes_done += count;
302     }
303 
304 out1:
305     qemu_iovec_destroy(&hd_qiov);
306 out:
307     return ret;
308 }
309 
310 static void GRAPH_UNLOCKED
311 secondary_do_checkpoint(BlockDriverState *bs, Error **errp)
312 {
313     BDRVReplicationState *s = bs->opaque;
314     BdrvChild *active_disk;
315     Error *local_err = NULL;
316     int ret;
317 
318     GRAPH_RDLOCK_GUARD_MAINLOOP();
319 
320     if (!s->backup_job) {
321         error_setg(errp, "Backup job was cancelled unexpectedly");
322         return;
323     }
324 
325     backup_do_checkpoint(s->backup_job, &local_err);
326     if (local_err) {
327         error_propagate(errp, local_err);
328         return;
329     }
330 
331     active_disk = bs->file;
332     if (!active_disk->bs->drv) {
333         error_setg(errp, "Active disk %s is ejected",
334                    active_disk->bs->node_name);
335         return;
336     }
337 
338     ret = bdrv_make_empty(active_disk, errp);
339     if (ret < 0) {
340         return;
341     }
342 
343     if (!s->hidden_disk->bs->drv) {
344         error_setg(errp, "Hidden disk %s is ejected",
345                    s->hidden_disk->bs->node_name);
346         return;
347     }
348 
349     ret = bdrv_make_empty(s->hidden_disk, errp);
350     if (ret < 0) {
351         return;
352     }
353 }
354 
355 /* This function is supposed to be called twice:
356  * first with writable = true, then with writable = false.
357  * The first call puts s->hidden_disk and s->secondary_disk in
358  * r/w mode, and the second puts them back in their original state.
359  */
360 static void reopen_backing_file(BlockDriverState *bs, bool writable,
361                                 Error **errp)
362 {
363     BDRVReplicationState *s = bs->opaque;
364     BdrvChild *hidden_disk, *secondary_disk;
365     BlockReopenQueue *reopen_queue = NULL;
366 
367     GLOBAL_STATE_CODE();
368     GRAPH_RDLOCK_GUARD_MAINLOOP();
369 
370     /*
371      * s->hidden_disk and s->secondary_disk may not be set yet, as they will
372      * only be set after the children are writable.
373      */
374     hidden_disk = bs->file->bs->backing;
375     secondary_disk = hidden_disk->bs->backing;
376 
377     if (writable) {
378         s->orig_hidden_read_only = bdrv_is_read_only(hidden_disk->bs);
379         s->orig_secondary_read_only = bdrv_is_read_only(secondary_disk->bs);
380     }
381 
382     if (s->orig_hidden_read_only) {
383         QDict *opts = qdict_new();
384         qdict_put_bool(opts, BDRV_OPT_READ_ONLY, !writable);
385         reopen_queue = bdrv_reopen_queue(reopen_queue, hidden_disk->bs,
386                                          opts, true);
387     }
388 
389     if (s->orig_secondary_read_only) {
390         QDict *opts = qdict_new();
391         qdict_put_bool(opts, BDRV_OPT_READ_ONLY, !writable);
392         reopen_queue = bdrv_reopen_queue(reopen_queue, secondary_disk->bs,
393                                          opts, true);
394     }
395 
396     if (reopen_queue) {
397         AioContext *ctx = bdrv_get_aio_context(bs);
398         if (ctx != qemu_get_aio_context()) {
399             aio_context_release(ctx);
400         }
401         bdrv_reopen_multiple(reopen_queue, errp);
402         if (ctx != qemu_get_aio_context()) {
403             aio_context_acquire(ctx);
404         }
405     }
406 }
407 
408 static void backup_job_cleanup(BlockDriverState *bs)
409 {
410     BDRVReplicationState *s = bs->opaque;
411     BlockDriverState *top_bs;
412 
413     s->backup_job = NULL;
414 
415     top_bs = bdrv_lookup_bs(s->top_id, s->top_id, NULL);
416     if (!top_bs) {
417         return;
418     }
419     bdrv_op_unblock_all(top_bs, s->blocker);
420     error_free(s->blocker);
421     reopen_backing_file(bs, false, NULL);
422 }
423 
424 static void backup_job_completed(void *opaque, int ret)
425 {
426     BlockDriverState *bs = opaque;
427     BDRVReplicationState *s = bs->opaque;
428 
429     if (s->stage != BLOCK_REPLICATION_FAILOVER) {
430         /* The backup job is cancelled unexpectedly */
431         s->error = -EIO;
432     }
433 
434     backup_job_cleanup(bs);
435 }
436 
437 static bool GRAPH_RDLOCK
438 check_top_bs(BlockDriverState *top_bs, BlockDriverState *bs)
439 {
440     BdrvChild *child;
441 
442     /* The bs itself is the top_bs */
443     if (top_bs == bs) {
444         return true;
445     }
446 
447     /* Iterate over top_bs's children */
448     QLIST_FOREACH(child, &top_bs->children, next) {
449         if (child->bs == bs || check_top_bs(child->bs, bs)) {
450             return true;
451         }
452     }
453 
454     return false;
455 }
456 
457 static void replication_start(ReplicationState *rs, ReplicationMode mode,
458                               Error **errp)
459 {
460     BlockDriverState *bs = rs->opaque;
461     BDRVReplicationState *s;
462     BlockDriverState *top_bs;
463     BdrvChild *active_disk, *hidden_disk, *secondary_disk;
464     int64_t active_length, hidden_length, disk_length;
465     AioContext *aio_context;
466     Error *local_err = NULL;
467     BackupPerf perf = { .use_copy_range = true, .max_workers = 1 };
468 
469     GLOBAL_STATE_CODE();
470 
471     aio_context = bdrv_get_aio_context(bs);
472     aio_context_acquire(aio_context);
473     s = bs->opaque;
474 
475     if (s->stage == BLOCK_REPLICATION_DONE ||
476         s->stage == BLOCK_REPLICATION_FAILOVER) {
477         /*
478          * This case happens when a secondary is promoted to primary.
479          * Ignore the request because the secondary side of replication
480          * doesn't have to do anything anymore.
481          */
482         aio_context_release(aio_context);
483         return;
484     }
485 
486     if (s->stage != BLOCK_REPLICATION_NONE) {
487         error_setg(errp, "Block replication is running or done");
488         aio_context_release(aio_context);
489         return;
490     }
491 
492     if (s->mode != mode) {
493         error_setg(errp, "The parameter mode's value is invalid, needs %d,"
494                    " but got %d", s->mode, mode);
495         aio_context_release(aio_context);
496         return;
497     }
498 
499     switch (s->mode) {
500     case REPLICATION_MODE_PRIMARY:
501         break;
502     case REPLICATION_MODE_SECONDARY:
503         bdrv_graph_rdlock_main_loop();
504         active_disk = bs->file;
505         if (!active_disk || !active_disk->bs || !active_disk->bs->backing) {
506             error_setg(errp, "Active disk doesn't have backing file");
507             bdrv_graph_rdunlock_main_loop();
508             aio_context_release(aio_context);
509             return;
510         }
511 
512         hidden_disk = active_disk->bs->backing;
513         if (!hidden_disk->bs || !hidden_disk->bs->backing) {
514             error_setg(errp, "Hidden disk doesn't have backing file");
515             bdrv_graph_rdunlock_main_loop();
516             aio_context_release(aio_context);
517             return;
518         }
519 
520         secondary_disk = hidden_disk->bs->backing;
521         if (!secondary_disk->bs || !bdrv_has_blk(secondary_disk->bs)) {
522             error_setg(errp, "The secondary disk doesn't have block backend");
523             bdrv_graph_rdunlock_main_loop();
524             aio_context_release(aio_context);
525             return;
526         }
527         bdrv_graph_rdunlock_main_loop();
528 
529         /* verify the length */
530         active_length = bdrv_getlength(active_disk->bs);
531         hidden_length = bdrv_getlength(hidden_disk->bs);
532         disk_length = bdrv_getlength(secondary_disk->bs);
533         if (active_length < 0 || hidden_length < 0 || disk_length < 0 ||
534             active_length != hidden_length || hidden_length != disk_length) {
535             error_setg(errp, "Active disk, hidden disk, secondary disk's length"
536                        " are not the same");
537             aio_context_release(aio_context);
538             return;
539         }
540 
541         /* Must be true, or the bdrv_getlength() calls would have failed */
542         assert(active_disk->bs->drv && hidden_disk->bs->drv);
543 
544         bdrv_graph_rdlock_main_loop();
545         if (!active_disk->bs->drv->bdrv_make_empty ||
546             !hidden_disk->bs->drv->bdrv_make_empty) {
547             error_setg(errp,
548                        "Active disk or hidden disk doesn't support make_empty");
549             aio_context_release(aio_context);
550             bdrv_graph_rdunlock_main_loop();
551             return;
552         }
553         bdrv_graph_rdunlock_main_loop();
554 
555         /* reopen the backing file in r/w mode */
556         reopen_backing_file(bs, true, &local_err);
557         if (local_err) {
558             error_propagate(errp, local_err);
559             aio_context_release(aio_context);
560             return;
561         }
562 
563         bdrv_graph_wrlock(bs);
564 
565         bdrv_ref(hidden_disk->bs);
566         s->hidden_disk = bdrv_attach_child(bs, hidden_disk->bs, "hidden disk",
567                                            &child_of_bds, BDRV_CHILD_DATA,
568                                            &local_err);
569         if (local_err) {
570             error_propagate(errp, local_err);
571             bdrv_graph_wrunlock();
572             aio_context_release(aio_context);
573             return;
574         }
575 
576         bdrv_ref(secondary_disk->bs);
577         s->secondary_disk = bdrv_attach_child(bs, secondary_disk->bs,
578                                               "secondary disk", &child_of_bds,
579                                               BDRV_CHILD_DATA, &local_err);
580         if (local_err) {
581             error_propagate(errp, local_err);
582             bdrv_graph_wrunlock();
583             aio_context_release(aio_context);
584             return;
585         }
586 
587         /* start backup job now */
588         error_setg(&s->blocker,
589                    "Block device is in use by internal backup job");
590 
591         top_bs = bdrv_lookup_bs(s->top_id, s->top_id, NULL);
592         if (!top_bs || !bdrv_is_root_node(top_bs) ||
593             !check_top_bs(top_bs, bs)) {
594             error_setg(errp, "No top_bs or it is invalid");
595             bdrv_graph_wrunlock();
596             reopen_backing_file(bs, false, NULL);
597             aio_context_release(aio_context);
598             return;
599         }
600         bdrv_op_block_all(top_bs, s->blocker);
601         bdrv_op_unblock(top_bs, BLOCK_OP_TYPE_DATAPLANE, s->blocker);
602 
603         bdrv_graph_wrunlock();
604 
605         s->backup_job = backup_job_create(
606                                 NULL, s->secondary_disk->bs, s->hidden_disk->bs,
607                                 0, MIRROR_SYNC_MODE_NONE, NULL, 0, false, NULL,
608                                 &perf,
609                                 BLOCKDEV_ON_ERROR_REPORT,
610                                 BLOCKDEV_ON_ERROR_REPORT, JOB_INTERNAL,
611                                 backup_job_completed, bs, NULL, &local_err);
612         if (local_err) {
613             error_propagate(errp, local_err);
614             backup_job_cleanup(bs);
615             aio_context_release(aio_context);
616             return;
617         }
618         job_start(&s->backup_job->job);
619         break;
620     default:
621         aio_context_release(aio_context);
622         abort();
623     }
624 
625     s->stage = BLOCK_REPLICATION_RUNNING;
626 
627     if (s->mode == REPLICATION_MODE_SECONDARY) {
628         secondary_do_checkpoint(bs, errp);
629     }
630 
631     s->error = 0;
632     aio_context_release(aio_context);
633 }
634 
635 static void replication_do_checkpoint(ReplicationState *rs, Error **errp)
636 {
637     BlockDriverState *bs = rs->opaque;
638     BDRVReplicationState *s;
639     AioContext *aio_context;
640 
641     aio_context = bdrv_get_aio_context(bs);
642     aio_context_acquire(aio_context);
643     s = bs->opaque;
644 
645     if (s->stage == BLOCK_REPLICATION_DONE ||
646         s->stage == BLOCK_REPLICATION_FAILOVER) {
647         /*
648          * This case happens when a secondary was promoted to primary.
649          * Ignore the request because the secondary side of replication
650          * doesn't have to do anything anymore.
651          */
652         aio_context_release(aio_context);
653         return;
654     }
655 
656     if (s->mode == REPLICATION_MODE_SECONDARY) {
657         secondary_do_checkpoint(bs, errp);
658     }
659     aio_context_release(aio_context);
660 }
661 
662 static void replication_get_error(ReplicationState *rs, Error **errp)
663 {
664     BlockDriverState *bs = rs->opaque;
665     BDRVReplicationState *s;
666     AioContext *aio_context;
667 
668     aio_context = bdrv_get_aio_context(bs);
669     aio_context_acquire(aio_context);
670     s = bs->opaque;
671 
672     if (s->stage == BLOCK_REPLICATION_NONE) {
673         error_setg(errp, "Block replication is not running");
674         aio_context_release(aio_context);
675         return;
676     }
677 
678     if (s->error) {
679         error_setg(errp, "I/O error occurred");
680         aio_context_release(aio_context);
681         return;
682     }
683     aio_context_release(aio_context);
684 }
685 
686 static void replication_done(void *opaque, int ret)
687 {
688     BlockDriverState *bs = opaque;
689     BDRVReplicationState *s = bs->opaque;
690 
691     if (ret == 0) {
692         s->stage = BLOCK_REPLICATION_DONE;
693 
694         bdrv_graph_wrlock(NULL);
695         bdrv_unref_child(bs, s->secondary_disk);
696         s->secondary_disk = NULL;
697         bdrv_unref_child(bs, s->hidden_disk);
698         s->hidden_disk = NULL;
699         bdrv_graph_wrunlock();
700 
701         s->error = 0;
702     } else {
703         s->stage = BLOCK_REPLICATION_FAILOVER_FAILED;
704         s->error = -EIO;
705     }
706 }
707 
708 static void replication_stop(ReplicationState *rs, bool failover, Error **errp)
709 {
710     BlockDriverState *bs = rs->opaque;
711     BDRVReplicationState *s;
712     AioContext *aio_context;
713 
714     aio_context = bdrv_get_aio_context(bs);
715     aio_context_acquire(aio_context);
716     s = bs->opaque;
717 
718     if (s->stage == BLOCK_REPLICATION_DONE ||
719         s->stage == BLOCK_REPLICATION_FAILOVER) {
720         /*
721          * This case happens when a secondary was promoted to primary.
722          * Ignore the request because the secondary side of replication
723          * doesn't have to do anything anymore.
724          */
725         aio_context_release(aio_context);
726         return;
727     }
728 
729     if (s->stage != BLOCK_REPLICATION_RUNNING) {
730         error_setg(errp, "Block replication is not running");
731         aio_context_release(aio_context);
732         return;
733     }
734 
735     switch (s->mode) {
736     case REPLICATION_MODE_PRIMARY:
737         s->stage = BLOCK_REPLICATION_DONE;
738         s->error = 0;
739         break;
740     case REPLICATION_MODE_SECONDARY:
741         /*
742          * This BDS will be closed, and the job should be completed
743          * before the BDS is closed, because we will access hidden
744          * disk, secondary disk in backup_job_completed().
745          */
746         if (s->backup_job) {
747             aio_context_release(aio_context);
748             job_cancel_sync(&s->backup_job->job, true);
749             aio_context_acquire(aio_context);
750         }
751 
752         if (!failover) {
753             secondary_do_checkpoint(bs, errp);
754             s->stage = BLOCK_REPLICATION_DONE;
755             aio_context_release(aio_context);
756             return;
757         }
758 
759         bdrv_graph_rdlock_main_loop();
760         s->stage = BLOCK_REPLICATION_FAILOVER;
761         s->commit_job = commit_active_start(
762                             NULL, bs->file->bs, s->secondary_disk->bs,
763                             JOB_INTERNAL, 0, BLOCKDEV_ON_ERROR_REPORT,
764                             NULL, replication_done, bs, true, errp);
765         bdrv_graph_rdunlock_main_loop();
766         break;
767     default:
768         aio_context_release(aio_context);
769         abort();
770     }
771     aio_context_release(aio_context);
772 }
773 
774 static const char *const replication_strong_runtime_opts[] = {
775     REPLICATION_MODE,
776     REPLICATION_TOP_ID,
777 
778     NULL
779 };
780 
781 static BlockDriver bdrv_replication = {
782     .format_name                = "replication",
783     .instance_size              = sizeof(BDRVReplicationState),
784 
785     .bdrv_open                  = replication_open,
786     .bdrv_close                 = replication_close,
787     .bdrv_child_perm            = replication_child_perm,
788 
789     .bdrv_co_getlength          = replication_co_getlength,
790     .bdrv_co_readv              = replication_co_readv,
791     .bdrv_co_writev             = replication_co_writev,
792 
793     .is_filter                  = true,
794 
795     .strong_runtime_opts        = replication_strong_runtime_opts,
796 };
797 
798 static void bdrv_replication_init(void)
799 {
800     bdrv_register(&bdrv_replication);
801 }
802 
803 block_init(bdrv_replication_init);
804