xref: /qemu/block/replication.c (revision b30d1886)
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-common.h"
17 #include "block/nbd.h"
18 #include "block/blockjob.h"
19 #include "block/block_int.h"
20 #include "block/block_backup.h"
21 #include "sysemu/block-backend.h"
22 #include "qapi/error.h"
23 #include "replication.h"
24 
25 typedef struct BDRVReplicationState {
26     ReplicationMode mode;
27     int replication_state;
28     BdrvChild *active_disk;
29     BdrvChild *hidden_disk;
30     BdrvChild *secondary_disk;
31     char *top_id;
32     ReplicationState *rs;
33     Error *blocker;
34     int orig_hidden_flags;
35     int orig_secondary_flags;
36     int error;
37 } BDRVReplicationState;
38 
39 enum {
40     BLOCK_REPLICATION_NONE,             /* block replication is not started */
41     BLOCK_REPLICATION_RUNNING,          /* block replication is running */
42     BLOCK_REPLICATION_FAILOVER,         /* failover is running in background */
43     BLOCK_REPLICATION_FAILOVER_FAILED,  /* failover failed */
44     BLOCK_REPLICATION_DONE,             /* block replication is done */
45 };
46 
47 static void replication_start(ReplicationState *rs, ReplicationMode mode,
48                               Error **errp);
49 static void replication_do_checkpoint(ReplicationState *rs, Error **errp);
50 static void replication_get_error(ReplicationState *rs, Error **errp);
51 static void replication_stop(ReplicationState *rs, bool failover,
52                              Error **errp);
53 
54 #define REPLICATION_MODE        "mode"
55 #define REPLICATION_TOP_ID      "top-id"
56 static QemuOptsList replication_runtime_opts = {
57     .name = "replication",
58     .head = QTAILQ_HEAD_INITIALIZER(replication_runtime_opts.head),
59     .desc = {
60         {
61             .name = REPLICATION_MODE,
62             .type = QEMU_OPT_STRING,
63         },
64         {
65             .name = REPLICATION_TOP_ID,
66             .type = QEMU_OPT_STRING,
67         },
68         { /* end of list */ }
69     },
70 };
71 
72 static ReplicationOps replication_ops = {
73     .start = replication_start,
74     .checkpoint = replication_do_checkpoint,
75     .get_error = replication_get_error,
76     .stop = replication_stop,
77 };
78 
79 static int replication_open(BlockDriverState *bs, QDict *options,
80                             int flags, Error **errp)
81 {
82     int ret;
83     BDRVReplicationState *s = bs->opaque;
84     Error *local_err = NULL;
85     QemuOpts *opts = NULL;
86     const char *mode;
87     const char *top_id;
88 
89     ret = -EINVAL;
90     opts = qemu_opts_create(&replication_runtime_opts, NULL, 0, &error_abort);
91     qemu_opts_absorb_qdict(opts, options, &local_err);
92     if (local_err) {
93         goto fail;
94     }
95 
96     mode = qemu_opt_get(opts, REPLICATION_MODE);
97     if (!mode) {
98         error_setg(&local_err, "Missing the option mode");
99         goto fail;
100     }
101 
102     if (!strcmp(mode, "primary")) {
103         s->mode = REPLICATION_MODE_PRIMARY;
104         top_id = qemu_opt_get(opts, REPLICATION_TOP_ID);
105         if (top_id) {
106             error_setg(&local_err, "The primary side does not support option top-id");
107             goto fail;
108         }
109     } else if (!strcmp(mode, "secondary")) {
110         s->mode = REPLICATION_MODE_SECONDARY;
111         top_id = qemu_opt_get(opts, REPLICATION_TOP_ID);
112         s->top_id = g_strdup(top_id);
113         if (!s->top_id) {
114             error_setg(&local_err, "Missing the option top-id");
115             goto fail;
116         }
117     } else {
118         error_setg(&local_err,
119                    "The option mode's value should be primary or secondary");
120         goto fail;
121     }
122 
123     s->rs = replication_new(bs, &replication_ops);
124 
125     ret = 0;
126 
127 fail:
128     qemu_opts_del(opts);
129     error_propagate(errp, local_err);
130 
131     return ret;
132 }
133 
134 static void replication_close(BlockDriverState *bs)
135 {
136     BDRVReplicationState *s = bs->opaque;
137 
138     if (s->replication_state == BLOCK_REPLICATION_RUNNING) {
139         replication_stop(s->rs, false, NULL);
140     }
141     if (s->replication_state == BLOCK_REPLICATION_FAILOVER) {
142         block_job_cancel_sync(s->active_disk->bs->job);
143     }
144 
145     if (s->mode == REPLICATION_MODE_SECONDARY) {
146         g_free(s->top_id);
147     }
148 
149     replication_remove(s->rs);
150 }
151 
152 static int64_t replication_getlength(BlockDriverState *bs)
153 {
154     return bdrv_getlength(bs->file->bs);
155 }
156 
157 static int replication_get_io_status(BDRVReplicationState *s)
158 {
159     switch (s->replication_state) {
160     case BLOCK_REPLICATION_NONE:
161         return -EIO;
162     case BLOCK_REPLICATION_RUNNING:
163         return 0;
164     case BLOCK_REPLICATION_FAILOVER:
165         return s->mode == REPLICATION_MODE_PRIMARY ? -EIO : 0;
166     case BLOCK_REPLICATION_FAILOVER_FAILED:
167         return s->mode == REPLICATION_MODE_PRIMARY ? -EIO : 1;
168     case BLOCK_REPLICATION_DONE:
169         /*
170          * active commit job completes, and active disk and secondary_disk
171          * is swapped, so we can operate bs->file directly
172          */
173         return s->mode == REPLICATION_MODE_PRIMARY ? -EIO : 0;
174     default:
175         abort();
176     }
177 }
178 
179 static int replication_return_value(BDRVReplicationState *s, int ret)
180 {
181     if (s->mode == REPLICATION_MODE_SECONDARY) {
182         return ret;
183     }
184 
185     if (ret < 0) {
186         s->error = ret;
187         ret = 0;
188     }
189 
190     return ret;
191 }
192 
193 static coroutine_fn int replication_co_readv(BlockDriverState *bs,
194                                              int64_t sector_num,
195                                              int remaining_sectors,
196                                              QEMUIOVector *qiov)
197 {
198     BDRVReplicationState *s = bs->opaque;
199     BdrvChild *child = s->secondary_disk;
200     BlockJob *job = NULL;
201     CowRequest req;
202     int ret;
203 
204     if (s->mode == REPLICATION_MODE_PRIMARY) {
205         /* We only use it to forward primary write requests */
206         return -EIO;
207     }
208 
209     ret = replication_get_io_status(s);
210     if (ret < 0) {
211         return ret;
212     }
213 
214     if (child && child->bs) {
215         job = child->bs->job;
216     }
217 
218     if (job) {
219         backup_wait_for_overlapping_requests(child->bs->job, sector_num,
220                                              remaining_sectors);
221         backup_cow_request_begin(&req, child->bs->job, sector_num,
222                                  remaining_sectors);
223         ret = bdrv_co_readv(bs->file, sector_num, remaining_sectors,
224                             qiov);
225         backup_cow_request_end(&req);
226         goto out;
227     }
228 
229     ret = bdrv_co_readv(bs->file, sector_num, remaining_sectors, qiov);
230 out:
231     return replication_return_value(s, ret);
232 }
233 
234 static coroutine_fn int replication_co_writev(BlockDriverState *bs,
235                                               int64_t sector_num,
236                                               int remaining_sectors,
237                                               QEMUIOVector *qiov)
238 {
239     BDRVReplicationState *s = bs->opaque;
240     QEMUIOVector hd_qiov;
241     uint64_t bytes_done = 0;
242     BdrvChild *top = bs->file;
243     BdrvChild *base = s->secondary_disk;
244     BdrvChild *target;
245     int ret, n;
246 
247     ret = replication_get_io_status(s);
248     if (ret < 0) {
249         goto out;
250     }
251 
252     if (ret == 0) {
253         ret = bdrv_co_writev(top, sector_num,
254                              remaining_sectors, qiov);
255         return replication_return_value(s, ret);
256     }
257 
258     /*
259      * Failover failed, only write to active disk if the sectors
260      * have already been allocated in active disk/hidden disk.
261      */
262     qemu_iovec_init(&hd_qiov, qiov->niov);
263     while (remaining_sectors > 0) {
264         ret = bdrv_is_allocated_above(top->bs, base->bs, sector_num,
265                                       remaining_sectors, &n);
266         if (ret < 0) {
267             goto out1;
268         }
269 
270         qemu_iovec_reset(&hd_qiov);
271         qemu_iovec_concat(&hd_qiov, qiov, bytes_done, n * BDRV_SECTOR_SIZE);
272 
273         target = ret ? top : base;
274         ret = bdrv_co_writev(target, sector_num, n, &hd_qiov);
275         if (ret < 0) {
276             goto out1;
277         }
278 
279         remaining_sectors -= n;
280         sector_num += n;
281         bytes_done += n * BDRV_SECTOR_SIZE;
282     }
283 
284 out1:
285     qemu_iovec_destroy(&hd_qiov);
286 out:
287     return ret;
288 }
289 
290 static bool replication_recurse_is_first_non_filter(BlockDriverState *bs,
291                                                     BlockDriverState *candidate)
292 {
293     return bdrv_recurse_is_first_non_filter(bs->file->bs, candidate);
294 }
295 
296 static void secondary_do_checkpoint(BDRVReplicationState *s, Error **errp)
297 {
298     Error *local_err = NULL;
299     int ret;
300 
301     if (!s->secondary_disk->bs->job) {
302         error_setg(errp, "Backup job was cancelled unexpectedly");
303         return;
304     }
305 
306     backup_do_checkpoint(s->secondary_disk->bs->job, &local_err);
307     if (local_err) {
308         error_propagate(errp, local_err);
309         return;
310     }
311 
312     ret = s->active_disk->bs->drv->bdrv_make_empty(s->active_disk->bs);
313     if (ret < 0) {
314         error_setg(errp, "Cannot make active disk empty");
315         return;
316     }
317 
318     ret = s->hidden_disk->bs->drv->bdrv_make_empty(s->hidden_disk->bs);
319     if (ret < 0) {
320         error_setg(errp, "Cannot make hidden disk empty");
321         return;
322     }
323 }
324 
325 static void reopen_backing_file(BlockDriverState *bs, bool writable,
326                                 Error **errp)
327 {
328     BDRVReplicationState *s = bs->opaque;
329     BlockReopenQueue *reopen_queue = NULL;
330     int orig_hidden_flags, orig_secondary_flags;
331     int new_hidden_flags, new_secondary_flags;
332     Error *local_err = NULL;
333 
334     if (writable) {
335         orig_hidden_flags = s->orig_hidden_flags =
336                                 bdrv_get_flags(s->hidden_disk->bs);
337         new_hidden_flags = (orig_hidden_flags | BDRV_O_RDWR) &
338                                                     ~BDRV_O_INACTIVE;
339         orig_secondary_flags = s->orig_secondary_flags =
340                                 bdrv_get_flags(s->secondary_disk->bs);
341         new_secondary_flags = (orig_secondary_flags | BDRV_O_RDWR) &
342                                                      ~BDRV_O_INACTIVE;
343     } else {
344         orig_hidden_flags = (s->orig_hidden_flags | BDRV_O_RDWR) &
345                                                     ~BDRV_O_INACTIVE;
346         new_hidden_flags = s->orig_hidden_flags;
347         orig_secondary_flags = (s->orig_secondary_flags | BDRV_O_RDWR) &
348                                                     ~BDRV_O_INACTIVE;
349         new_secondary_flags = s->orig_secondary_flags;
350     }
351 
352     if (orig_hidden_flags != new_hidden_flags) {
353         reopen_queue = bdrv_reopen_queue(reopen_queue, s->hidden_disk->bs, NULL,
354                                          new_hidden_flags);
355     }
356 
357     if (!(orig_secondary_flags & BDRV_O_RDWR)) {
358         reopen_queue = bdrv_reopen_queue(reopen_queue, s->secondary_disk->bs,
359                                          NULL, new_secondary_flags);
360     }
361 
362     if (reopen_queue) {
363         bdrv_reopen_multiple(bdrv_get_aio_context(bs),
364                              reopen_queue, &local_err);
365         error_propagate(errp, local_err);
366     }
367 }
368 
369 static void backup_job_cleanup(BlockDriverState *bs)
370 {
371     BDRVReplicationState *s = bs->opaque;
372     BlockDriverState *top_bs;
373 
374     top_bs = bdrv_lookup_bs(s->top_id, s->top_id, NULL);
375     if (!top_bs) {
376         return;
377     }
378     bdrv_op_unblock_all(top_bs, s->blocker);
379     error_free(s->blocker);
380     reopen_backing_file(bs, false, NULL);
381 }
382 
383 static void backup_job_completed(void *opaque, int ret)
384 {
385     BlockDriverState *bs = opaque;
386     BDRVReplicationState *s = bs->opaque;
387 
388     if (s->replication_state != BLOCK_REPLICATION_FAILOVER) {
389         /* The backup job is cancelled unexpectedly */
390         s->error = -EIO;
391     }
392 
393     backup_job_cleanup(bs);
394 }
395 
396 static bool check_top_bs(BlockDriverState *top_bs, BlockDriverState *bs)
397 {
398     BdrvChild *child;
399 
400     /* The bs itself is the top_bs */
401     if (top_bs == bs) {
402         return true;
403     }
404 
405     /* Iterate over top_bs's children */
406     QLIST_FOREACH(child, &top_bs->children, next) {
407         if (child->bs == bs || check_top_bs(child->bs, bs)) {
408             return true;
409         }
410     }
411 
412     return false;
413 }
414 
415 static void replication_start(ReplicationState *rs, ReplicationMode mode,
416                               Error **errp)
417 {
418     BlockDriverState *bs = rs->opaque;
419     BDRVReplicationState *s;
420     BlockDriverState *top_bs;
421     int64_t active_length, hidden_length, disk_length;
422     AioContext *aio_context;
423     Error *local_err = NULL;
424     BlockJob *job;
425 
426     aio_context = bdrv_get_aio_context(bs);
427     aio_context_acquire(aio_context);
428     s = bs->opaque;
429 
430     if (s->replication_state != BLOCK_REPLICATION_NONE) {
431         error_setg(errp, "Block replication is running or done");
432         aio_context_release(aio_context);
433         return;
434     }
435 
436     if (s->mode != mode) {
437         error_setg(errp, "The parameter mode's value is invalid, needs %d,"
438                    " but got %d", s->mode, mode);
439         aio_context_release(aio_context);
440         return;
441     }
442 
443     switch (s->mode) {
444     case REPLICATION_MODE_PRIMARY:
445         break;
446     case REPLICATION_MODE_SECONDARY:
447         s->active_disk = bs->file;
448         if (!s->active_disk || !s->active_disk->bs ||
449                                     !s->active_disk->bs->backing) {
450             error_setg(errp, "Active disk doesn't have backing file");
451             aio_context_release(aio_context);
452             return;
453         }
454 
455         s->hidden_disk = s->active_disk->bs->backing;
456         if (!s->hidden_disk->bs || !s->hidden_disk->bs->backing) {
457             error_setg(errp, "Hidden disk doesn't have backing file");
458             aio_context_release(aio_context);
459             return;
460         }
461 
462         s->secondary_disk = s->hidden_disk->bs->backing;
463         if (!s->secondary_disk->bs || !bdrv_has_blk(s->secondary_disk->bs)) {
464             error_setg(errp, "The secondary disk doesn't have block backend");
465             aio_context_release(aio_context);
466             return;
467         }
468 
469         /* verify the length */
470         active_length = bdrv_getlength(s->active_disk->bs);
471         hidden_length = bdrv_getlength(s->hidden_disk->bs);
472         disk_length = bdrv_getlength(s->secondary_disk->bs);
473         if (active_length < 0 || hidden_length < 0 || disk_length < 0 ||
474             active_length != hidden_length || hidden_length != disk_length) {
475             error_setg(errp, "Active disk, hidden disk, secondary disk's length"
476                        " are not the same");
477             aio_context_release(aio_context);
478             return;
479         }
480 
481         if (!s->active_disk->bs->drv->bdrv_make_empty ||
482             !s->hidden_disk->bs->drv->bdrv_make_empty) {
483             error_setg(errp,
484                        "Active disk or hidden disk doesn't support make_empty");
485             aio_context_release(aio_context);
486             return;
487         }
488 
489         /* reopen the backing file in r/w mode */
490         reopen_backing_file(bs, true, &local_err);
491         if (local_err) {
492             error_propagate(errp, local_err);
493             aio_context_release(aio_context);
494             return;
495         }
496 
497         /* start backup job now */
498         error_setg(&s->blocker,
499                    "Block device is in use by internal backup job");
500 
501         top_bs = bdrv_lookup_bs(s->top_id, s->top_id, NULL);
502         if (!top_bs || !bdrv_is_root_node(top_bs) ||
503             !check_top_bs(top_bs, bs)) {
504             error_setg(errp, "No top_bs or it is invalid");
505             reopen_backing_file(bs, false, NULL);
506             aio_context_release(aio_context);
507             return;
508         }
509         bdrv_op_block_all(top_bs, s->blocker);
510         bdrv_op_unblock(top_bs, BLOCK_OP_TYPE_DATAPLANE, s->blocker);
511 
512         job = backup_job_create(NULL, s->secondary_disk->bs, s->hidden_disk->bs,
513                                 0, MIRROR_SYNC_MODE_NONE, NULL, false,
514                                 BLOCKDEV_ON_ERROR_REPORT,
515                                 BLOCKDEV_ON_ERROR_REPORT, BLOCK_JOB_INTERNAL,
516                                 backup_job_completed, bs, NULL, &local_err);
517         if (local_err) {
518             error_propagate(errp, local_err);
519             backup_job_cleanup(bs);
520             aio_context_release(aio_context);
521             return;
522         }
523         block_job_start(job);
524         break;
525     default:
526         aio_context_release(aio_context);
527         abort();
528     }
529 
530     s->replication_state = BLOCK_REPLICATION_RUNNING;
531 
532     if (s->mode == REPLICATION_MODE_SECONDARY) {
533         secondary_do_checkpoint(s, errp);
534     }
535 
536     s->error = 0;
537     aio_context_release(aio_context);
538 }
539 
540 static void replication_do_checkpoint(ReplicationState *rs, Error **errp)
541 {
542     BlockDriverState *bs = rs->opaque;
543     BDRVReplicationState *s;
544     AioContext *aio_context;
545 
546     aio_context = bdrv_get_aio_context(bs);
547     aio_context_acquire(aio_context);
548     s = bs->opaque;
549 
550     if (s->mode == REPLICATION_MODE_SECONDARY) {
551         secondary_do_checkpoint(s, errp);
552     }
553     aio_context_release(aio_context);
554 }
555 
556 static void replication_get_error(ReplicationState *rs, Error **errp)
557 {
558     BlockDriverState *bs = rs->opaque;
559     BDRVReplicationState *s;
560     AioContext *aio_context;
561 
562     aio_context = bdrv_get_aio_context(bs);
563     aio_context_acquire(aio_context);
564     s = bs->opaque;
565 
566     if (s->replication_state != BLOCK_REPLICATION_RUNNING) {
567         error_setg(errp, "Block replication is not running");
568         aio_context_release(aio_context);
569         return;
570     }
571 
572     if (s->error) {
573         error_setg(errp, "I/O error occurred");
574         aio_context_release(aio_context);
575         return;
576     }
577     aio_context_release(aio_context);
578 }
579 
580 static void replication_done(void *opaque, int ret)
581 {
582     BlockDriverState *bs = opaque;
583     BDRVReplicationState *s = bs->opaque;
584 
585     if (ret == 0) {
586         s->replication_state = BLOCK_REPLICATION_DONE;
587 
588         /* refresh top bs's filename */
589         bdrv_refresh_filename(bs);
590         s->active_disk = NULL;
591         s->secondary_disk = NULL;
592         s->hidden_disk = NULL;
593         s->error = 0;
594     } else {
595         s->replication_state = BLOCK_REPLICATION_FAILOVER_FAILED;
596         s->error = -EIO;
597     }
598 }
599 
600 static void replication_stop(ReplicationState *rs, bool failover, Error **errp)
601 {
602     BlockDriverState *bs = rs->opaque;
603     BDRVReplicationState *s;
604     AioContext *aio_context;
605 
606     aio_context = bdrv_get_aio_context(bs);
607     aio_context_acquire(aio_context);
608     s = bs->opaque;
609 
610     if (s->replication_state != BLOCK_REPLICATION_RUNNING) {
611         error_setg(errp, "Block replication is not running");
612         aio_context_release(aio_context);
613         return;
614     }
615 
616     switch (s->mode) {
617     case REPLICATION_MODE_PRIMARY:
618         s->replication_state = BLOCK_REPLICATION_DONE;
619         s->error = 0;
620         break;
621     case REPLICATION_MODE_SECONDARY:
622         /*
623          * This BDS will be closed, and the job should be completed
624          * before the BDS is closed, because we will access hidden
625          * disk, secondary disk in backup_job_completed().
626          */
627         if (s->secondary_disk->bs->job) {
628             block_job_cancel_sync(s->secondary_disk->bs->job);
629         }
630 
631         if (!failover) {
632             secondary_do_checkpoint(s, errp);
633             s->replication_state = BLOCK_REPLICATION_DONE;
634             aio_context_release(aio_context);
635             return;
636         }
637 
638         s->replication_state = BLOCK_REPLICATION_FAILOVER;
639         commit_active_start(NULL, s->active_disk->bs, s->secondary_disk->bs,
640                             BLOCK_JOB_INTERNAL, 0, BLOCKDEV_ON_ERROR_REPORT,
641                             replication_done, bs, errp, true);
642         break;
643     default:
644         aio_context_release(aio_context);
645         abort();
646     }
647     aio_context_release(aio_context);
648 }
649 
650 BlockDriver bdrv_replication = {
651     .format_name                = "replication",
652     .protocol_name              = "replication",
653     .instance_size              = sizeof(BDRVReplicationState),
654 
655     .bdrv_open                  = replication_open,
656     .bdrv_close                 = replication_close,
657 
658     .bdrv_getlength             = replication_getlength,
659     .bdrv_co_readv              = replication_co_readv,
660     .bdrv_co_writev             = replication_co_writev,
661 
662     .is_filter                  = true,
663     .bdrv_recurse_is_first_non_filter = replication_recurse_is_first_non_filter,
664 
665     .has_variable_length        = true,
666 };
667 
668 static void bdrv_replication_init(void)
669 {
670     bdrv_register(&bdrv_replication);
671 }
672 
673 block_init(bdrv_replication_init);
674