xref: /openbmc/qemu/block/replication.c (revision 5d721b78)
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 enum {
26     BLOCK_REPLICATION_NONE,             /* block replication is not started */
27     BLOCK_REPLICATION_RUNNING,          /* block replication is running */
28     BLOCK_REPLICATION_FAILOVER,         /* failover is running in background */
29     BLOCK_REPLICATION_FAILOVER_FAILED,  /* failover failed */
30     BLOCK_REPLICATION_DONE,             /* block replication is done */
31 } ReplicationStage;
32 
33 typedef struct BDRVReplicationState {
34     ReplicationMode mode;
35     ReplicationStage stage;
36     BdrvChild *active_disk;
37     BdrvChild *hidden_disk;
38     BdrvChild *secondary_disk;
39     char *top_id;
40     ReplicationState *rs;
41     Error *blocker;
42     int orig_hidden_flags;
43     int orig_secondary_flags;
44     int error;
45 } BDRVReplicationState;
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     bs->file = bdrv_open_child(NULL, options, "file", bs, &child_file,
90                                false, errp);
91     if (!bs->file) {
92         return -EINVAL;
93     }
94 
95     ret = -EINVAL;
96     opts = qemu_opts_create(&replication_runtime_opts, NULL, 0, &error_abort);
97     qemu_opts_absorb_qdict(opts, options, &local_err);
98     if (local_err) {
99         goto fail;
100     }
101 
102     mode = qemu_opt_get(opts, REPLICATION_MODE);
103     if (!mode) {
104         error_setg(&local_err, "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(&local_err, "The primary side does not support option top-id");
113             goto fail;
114         }
115     } else if (!strcmp(mode, "secondary")) {
116         s->mode = REPLICATION_MODE_SECONDARY;
117         top_id = qemu_opt_get(opts, REPLICATION_TOP_ID);
118         s->top_id = g_strdup(top_id);
119         if (!s->top_id) {
120             error_setg(&local_err, "Missing the option top-id");
121             goto fail;
122         }
123     } else {
124         error_setg(&local_err,
125                    "The option mode's value should be primary or secondary");
126         goto fail;
127     }
128 
129     s->rs = replication_new(bs, &replication_ops);
130 
131     ret = 0;
132 
133 fail:
134     qemu_opts_del(opts);
135     error_propagate(errp, local_err);
136 
137     return ret;
138 }
139 
140 static void replication_close(BlockDriverState *bs)
141 {
142     BDRVReplicationState *s = bs->opaque;
143 
144     if (s->stage == BLOCK_REPLICATION_RUNNING) {
145         replication_stop(s->rs, false, NULL);
146     }
147     if (s->stage == BLOCK_REPLICATION_FAILOVER) {
148         block_job_cancel_sync(s->active_disk->bs->job);
149     }
150 
151     if (s->mode == REPLICATION_MODE_SECONDARY) {
152         g_free(s->top_id);
153     }
154 
155     replication_remove(s->rs);
156 }
157 
158 static void replication_child_perm(BlockDriverState *bs, BdrvChild *c,
159                                    const BdrvChildRole *role,
160                                    uint64_t perm, uint64_t shared,
161                                    uint64_t *nperm, uint64_t *nshared)
162 {
163     *nperm = *nshared = BLK_PERM_CONSISTENT_READ \
164                         | BLK_PERM_WRITE \
165                         | BLK_PERM_WRITE_UNCHANGED;
166 
167     return;
168 }
169 
170 static int64_t replication_getlength(BlockDriverState *bs)
171 {
172     return bdrv_getlength(bs->file->bs);
173 }
174 
175 static int replication_get_io_status(BDRVReplicationState *s)
176 {
177     switch (s->stage) {
178     case BLOCK_REPLICATION_NONE:
179         return -EIO;
180     case BLOCK_REPLICATION_RUNNING:
181         return 0;
182     case BLOCK_REPLICATION_FAILOVER:
183         return s->mode == REPLICATION_MODE_PRIMARY ? -EIO : 0;
184     case BLOCK_REPLICATION_FAILOVER_FAILED:
185         return s->mode == REPLICATION_MODE_PRIMARY ? -EIO : 1;
186     case BLOCK_REPLICATION_DONE:
187         /*
188          * active commit job completes, and active disk and secondary_disk
189          * is swapped, so we can operate bs->file directly
190          */
191         return s->mode == REPLICATION_MODE_PRIMARY ? -EIO : 0;
192     default:
193         abort();
194     }
195 }
196 
197 static int replication_return_value(BDRVReplicationState *s, int ret)
198 {
199     if (s->mode == REPLICATION_MODE_SECONDARY) {
200         return ret;
201     }
202 
203     if (ret < 0) {
204         s->error = ret;
205         ret = 0;
206     }
207 
208     return ret;
209 }
210 
211 static coroutine_fn int replication_co_readv(BlockDriverState *bs,
212                                              int64_t sector_num,
213                                              int remaining_sectors,
214                                              QEMUIOVector *qiov)
215 {
216     BDRVReplicationState *s = bs->opaque;
217     BdrvChild *child = s->secondary_disk;
218     BlockJob *job = NULL;
219     CowRequest req;
220     int ret;
221 
222     if (s->mode == REPLICATION_MODE_PRIMARY) {
223         /* We only use it to forward primary write requests */
224         return -EIO;
225     }
226 
227     ret = replication_get_io_status(s);
228     if (ret < 0) {
229         return ret;
230     }
231 
232     if (child && child->bs) {
233         job = child->bs->job;
234     }
235 
236     if (job) {
237         uint64_t remaining_bytes = remaining_sectors * BDRV_SECTOR_SIZE;
238 
239         backup_wait_for_overlapping_requests(child->bs->job,
240                                              sector_num * BDRV_SECTOR_SIZE,
241                                              remaining_bytes);
242         backup_cow_request_begin(&req, child->bs->job,
243                                  sector_num * BDRV_SECTOR_SIZE,
244                                  remaining_bytes);
245         ret = bdrv_co_readv(bs->file, sector_num, remaining_sectors,
246                             qiov);
247         backup_cow_request_end(&req);
248         goto out;
249     }
250 
251     ret = bdrv_co_readv(bs->file, sector_num, remaining_sectors, qiov);
252 out:
253     return replication_return_value(s, ret);
254 }
255 
256 static coroutine_fn int replication_co_writev(BlockDriverState *bs,
257                                               int64_t sector_num,
258                                               int remaining_sectors,
259                                               QEMUIOVector *qiov)
260 {
261     BDRVReplicationState *s = bs->opaque;
262     QEMUIOVector hd_qiov;
263     uint64_t bytes_done = 0;
264     BdrvChild *top = bs->file;
265     BdrvChild *base = s->secondary_disk;
266     BdrvChild *target;
267     int ret;
268     int64_t n;
269 
270     ret = replication_get_io_status(s);
271     if (ret < 0) {
272         goto out;
273     }
274 
275     if (ret == 0) {
276         ret = bdrv_co_writev(top, sector_num,
277                              remaining_sectors, qiov);
278         return replication_return_value(s, ret);
279     }
280 
281     /*
282      * Failover failed, only write to active disk if the sectors
283      * have already been allocated in active disk/hidden disk.
284      */
285     qemu_iovec_init(&hd_qiov, qiov->niov);
286     while (remaining_sectors > 0) {
287         int64_t count;
288 
289         ret = bdrv_is_allocated_above(top->bs, base->bs,
290                                       sector_num * BDRV_SECTOR_SIZE,
291                                       remaining_sectors * BDRV_SECTOR_SIZE,
292                                       &count);
293         if (ret < 0) {
294             goto out1;
295         }
296 
297         assert(QEMU_IS_ALIGNED(count, BDRV_SECTOR_SIZE));
298         n = count >> BDRV_SECTOR_BITS;
299         qemu_iovec_reset(&hd_qiov);
300         qemu_iovec_concat(&hd_qiov, qiov, bytes_done, count);
301 
302         target = ret ? top : base;
303         ret = bdrv_co_writev(target, sector_num, n, &hd_qiov);
304         if (ret < 0) {
305             goto out1;
306         }
307 
308         remaining_sectors -= n;
309         sector_num += n;
310         bytes_done += count;
311     }
312 
313 out1:
314     qemu_iovec_destroy(&hd_qiov);
315 out:
316     return ret;
317 }
318 
319 static bool replication_recurse_is_first_non_filter(BlockDriverState *bs,
320                                                     BlockDriverState *candidate)
321 {
322     return bdrv_recurse_is_first_non_filter(bs->file->bs, candidate);
323 }
324 
325 static void secondary_do_checkpoint(BDRVReplicationState *s, Error **errp)
326 {
327     Error *local_err = NULL;
328     int ret;
329 
330     if (!s->secondary_disk->bs->job) {
331         error_setg(errp, "Backup job was cancelled unexpectedly");
332         return;
333     }
334 
335     backup_do_checkpoint(s->secondary_disk->bs->job, &local_err);
336     if (local_err) {
337         error_propagate(errp, local_err);
338         return;
339     }
340 
341     ret = s->active_disk->bs->drv->bdrv_make_empty(s->active_disk->bs);
342     if (ret < 0) {
343         error_setg(errp, "Cannot make active disk empty");
344         return;
345     }
346 
347     ret = s->hidden_disk->bs->drv->bdrv_make_empty(s->hidden_disk->bs);
348     if (ret < 0) {
349         error_setg(errp, "Cannot make hidden disk empty");
350         return;
351     }
352 }
353 
354 static void reopen_backing_file(BlockDriverState *bs, bool writable,
355                                 Error **errp)
356 {
357     BDRVReplicationState *s = bs->opaque;
358     BlockReopenQueue *reopen_queue = NULL;
359     int orig_hidden_flags, orig_secondary_flags;
360     int new_hidden_flags, new_secondary_flags;
361     Error *local_err = NULL;
362 
363     if (writable) {
364         orig_hidden_flags = s->orig_hidden_flags =
365                                 bdrv_get_flags(s->hidden_disk->bs);
366         new_hidden_flags = (orig_hidden_flags | BDRV_O_RDWR) &
367                                                     ~BDRV_O_INACTIVE;
368         orig_secondary_flags = s->orig_secondary_flags =
369                                 bdrv_get_flags(s->secondary_disk->bs);
370         new_secondary_flags = (orig_secondary_flags | BDRV_O_RDWR) &
371                                                      ~BDRV_O_INACTIVE;
372     } else {
373         orig_hidden_flags = (s->orig_hidden_flags | BDRV_O_RDWR) &
374                                                     ~BDRV_O_INACTIVE;
375         new_hidden_flags = s->orig_hidden_flags;
376         orig_secondary_flags = (s->orig_secondary_flags | BDRV_O_RDWR) &
377                                                     ~BDRV_O_INACTIVE;
378         new_secondary_flags = s->orig_secondary_flags;
379     }
380 
381     if (orig_hidden_flags != new_hidden_flags) {
382         reopen_queue = bdrv_reopen_queue(reopen_queue, s->hidden_disk->bs, NULL,
383                                          new_hidden_flags);
384     }
385 
386     if (!(orig_secondary_flags & BDRV_O_RDWR)) {
387         reopen_queue = bdrv_reopen_queue(reopen_queue, s->secondary_disk->bs,
388                                          NULL, new_secondary_flags);
389     }
390 
391     if (reopen_queue) {
392         bdrv_reopen_multiple(bdrv_get_aio_context(bs),
393                              reopen_queue, &local_err);
394         error_propagate(errp, local_err);
395     }
396 }
397 
398 static void backup_job_cleanup(BlockDriverState *bs)
399 {
400     BDRVReplicationState *s = bs->opaque;
401     BlockDriverState *top_bs;
402 
403     top_bs = bdrv_lookup_bs(s->top_id, s->top_id, NULL);
404     if (!top_bs) {
405         return;
406     }
407     bdrv_op_unblock_all(top_bs, s->blocker);
408     error_free(s->blocker);
409     reopen_backing_file(bs, false, NULL);
410 }
411 
412 static void backup_job_completed(void *opaque, int ret)
413 {
414     BlockDriverState *bs = opaque;
415     BDRVReplicationState *s = bs->opaque;
416 
417     if (s->stage != BLOCK_REPLICATION_FAILOVER) {
418         /* The backup job is cancelled unexpectedly */
419         s->error = -EIO;
420     }
421 
422     backup_job_cleanup(bs);
423 }
424 
425 static bool check_top_bs(BlockDriverState *top_bs, BlockDriverState *bs)
426 {
427     BdrvChild *child;
428 
429     /* The bs itself is the top_bs */
430     if (top_bs == bs) {
431         return true;
432     }
433 
434     /* Iterate over top_bs's children */
435     QLIST_FOREACH(child, &top_bs->children, next) {
436         if (child->bs == bs || check_top_bs(child->bs, bs)) {
437             return true;
438         }
439     }
440 
441     return false;
442 }
443 
444 static void replication_start(ReplicationState *rs, ReplicationMode mode,
445                               Error **errp)
446 {
447     BlockDriverState *bs = rs->opaque;
448     BDRVReplicationState *s;
449     BlockDriverState *top_bs;
450     int64_t active_length, hidden_length, disk_length;
451     AioContext *aio_context;
452     Error *local_err = NULL;
453     BlockJob *job;
454 
455     aio_context = bdrv_get_aio_context(bs);
456     aio_context_acquire(aio_context);
457     s = bs->opaque;
458 
459     if (s->stage != BLOCK_REPLICATION_NONE) {
460         error_setg(errp, "Block replication is running or done");
461         aio_context_release(aio_context);
462         return;
463     }
464 
465     if (s->mode != mode) {
466         error_setg(errp, "The parameter mode's value is invalid, needs %d,"
467                    " but got %d", s->mode, mode);
468         aio_context_release(aio_context);
469         return;
470     }
471 
472     switch (s->mode) {
473     case REPLICATION_MODE_PRIMARY:
474         break;
475     case REPLICATION_MODE_SECONDARY:
476         s->active_disk = bs->file;
477         if (!s->active_disk || !s->active_disk->bs ||
478                                     !s->active_disk->bs->backing) {
479             error_setg(errp, "Active disk doesn't have backing file");
480             aio_context_release(aio_context);
481             return;
482         }
483 
484         s->hidden_disk = s->active_disk->bs->backing;
485         if (!s->hidden_disk->bs || !s->hidden_disk->bs->backing) {
486             error_setg(errp, "Hidden disk doesn't have backing file");
487             aio_context_release(aio_context);
488             return;
489         }
490 
491         s->secondary_disk = s->hidden_disk->bs->backing;
492         if (!s->secondary_disk->bs || !bdrv_has_blk(s->secondary_disk->bs)) {
493             error_setg(errp, "The secondary disk doesn't have block backend");
494             aio_context_release(aio_context);
495             return;
496         }
497 
498         /* verify the length */
499         active_length = bdrv_getlength(s->active_disk->bs);
500         hidden_length = bdrv_getlength(s->hidden_disk->bs);
501         disk_length = bdrv_getlength(s->secondary_disk->bs);
502         if (active_length < 0 || hidden_length < 0 || disk_length < 0 ||
503             active_length != hidden_length || hidden_length != disk_length) {
504             error_setg(errp, "Active disk, hidden disk, secondary disk's length"
505                        " are not the same");
506             aio_context_release(aio_context);
507             return;
508         }
509 
510         if (!s->active_disk->bs->drv->bdrv_make_empty ||
511             !s->hidden_disk->bs->drv->bdrv_make_empty) {
512             error_setg(errp,
513                        "Active disk or hidden disk doesn't support make_empty");
514             aio_context_release(aio_context);
515             return;
516         }
517 
518         /* reopen the backing file in r/w mode */
519         reopen_backing_file(bs, true, &local_err);
520         if (local_err) {
521             error_propagate(errp, local_err);
522             aio_context_release(aio_context);
523             return;
524         }
525 
526         /* start backup job now */
527         error_setg(&s->blocker,
528                    "Block device is in use by internal backup job");
529 
530         top_bs = bdrv_lookup_bs(s->top_id, s->top_id, NULL);
531         if (!top_bs || !bdrv_is_root_node(top_bs) ||
532             !check_top_bs(top_bs, bs)) {
533             error_setg(errp, "No top_bs or it is invalid");
534             reopen_backing_file(bs, false, NULL);
535             aio_context_release(aio_context);
536             return;
537         }
538         bdrv_op_block_all(top_bs, s->blocker);
539         bdrv_op_unblock(top_bs, BLOCK_OP_TYPE_DATAPLANE, s->blocker);
540 
541         job = backup_job_create(NULL, s->secondary_disk->bs, s->hidden_disk->bs,
542                                 0, MIRROR_SYNC_MODE_NONE, NULL, false,
543                                 BLOCKDEV_ON_ERROR_REPORT,
544                                 BLOCKDEV_ON_ERROR_REPORT, BLOCK_JOB_INTERNAL,
545                                 backup_job_completed, bs, NULL, &local_err);
546         if (local_err) {
547             error_propagate(errp, local_err);
548             backup_job_cleanup(bs);
549             aio_context_release(aio_context);
550             return;
551         }
552         block_job_start(job);
553         break;
554     default:
555         aio_context_release(aio_context);
556         abort();
557     }
558 
559     s->stage = BLOCK_REPLICATION_RUNNING;
560 
561     if (s->mode == REPLICATION_MODE_SECONDARY) {
562         secondary_do_checkpoint(s, errp);
563     }
564 
565     s->error = 0;
566     aio_context_release(aio_context);
567 }
568 
569 static void replication_do_checkpoint(ReplicationState *rs, Error **errp)
570 {
571     BlockDriverState *bs = rs->opaque;
572     BDRVReplicationState *s;
573     AioContext *aio_context;
574 
575     aio_context = bdrv_get_aio_context(bs);
576     aio_context_acquire(aio_context);
577     s = bs->opaque;
578 
579     if (s->mode == REPLICATION_MODE_SECONDARY) {
580         secondary_do_checkpoint(s, errp);
581     }
582     aio_context_release(aio_context);
583 }
584 
585 static void replication_get_error(ReplicationState *rs, Error **errp)
586 {
587     BlockDriverState *bs = rs->opaque;
588     BDRVReplicationState *s;
589     AioContext *aio_context;
590 
591     aio_context = bdrv_get_aio_context(bs);
592     aio_context_acquire(aio_context);
593     s = bs->opaque;
594 
595     if (s->stage != BLOCK_REPLICATION_RUNNING) {
596         error_setg(errp, "Block replication is not running");
597         aio_context_release(aio_context);
598         return;
599     }
600 
601     if (s->error) {
602         error_setg(errp, "I/O error occurred");
603         aio_context_release(aio_context);
604         return;
605     }
606     aio_context_release(aio_context);
607 }
608 
609 static void replication_done(void *opaque, int ret)
610 {
611     BlockDriverState *bs = opaque;
612     BDRVReplicationState *s = bs->opaque;
613 
614     if (ret == 0) {
615         s->stage = BLOCK_REPLICATION_DONE;
616 
617         /* refresh top bs's filename */
618         bdrv_refresh_filename(bs);
619         s->active_disk = NULL;
620         s->secondary_disk = NULL;
621         s->hidden_disk = NULL;
622         s->error = 0;
623     } else {
624         s->stage = BLOCK_REPLICATION_FAILOVER_FAILED;
625         s->error = -EIO;
626     }
627 }
628 
629 static void replication_stop(ReplicationState *rs, bool failover, Error **errp)
630 {
631     BlockDriverState *bs = rs->opaque;
632     BDRVReplicationState *s;
633     AioContext *aio_context;
634 
635     aio_context = bdrv_get_aio_context(bs);
636     aio_context_acquire(aio_context);
637     s = bs->opaque;
638 
639     if (s->stage != BLOCK_REPLICATION_RUNNING) {
640         error_setg(errp, "Block replication is not running");
641         aio_context_release(aio_context);
642         return;
643     }
644 
645     switch (s->mode) {
646     case REPLICATION_MODE_PRIMARY:
647         s->stage = BLOCK_REPLICATION_DONE;
648         s->error = 0;
649         break;
650     case REPLICATION_MODE_SECONDARY:
651         /*
652          * This BDS will be closed, and the job should be completed
653          * before the BDS is closed, because we will access hidden
654          * disk, secondary disk in backup_job_completed().
655          */
656         if (s->secondary_disk->bs->job) {
657             block_job_cancel_sync(s->secondary_disk->bs->job);
658         }
659 
660         if (!failover) {
661             secondary_do_checkpoint(s, errp);
662             s->stage = BLOCK_REPLICATION_DONE;
663             aio_context_release(aio_context);
664             return;
665         }
666 
667         s->stage = BLOCK_REPLICATION_FAILOVER;
668         commit_active_start(NULL, s->active_disk->bs, s->secondary_disk->bs,
669                             BLOCK_JOB_INTERNAL, 0, BLOCKDEV_ON_ERROR_REPORT,
670                             NULL, replication_done, bs, true, errp);
671         break;
672     default:
673         aio_context_release(aio_context);
674         abort();
675     }
676     aio_context_release(aio_context);
677 }
678 
679 BlockDriver bdrv_replication = {
680     .format_name                = "replication",
681     .protocol_name              = "replication",
682     .instance_size              = sizeof(BDRVReplicationState),
683 
684     .bdrv_open                  = replication_open,
685     .bdrv_close                 = replication_close,
686     .bdrv_child_perm            = replication_child_perm,
687 
688     .bdrv_getlength             = replication_getlength,
689     .bdrv_co_readv              = replication_co_readv,
690     .bdrv_co_writev             = replication_co_writev,
691 
692     .is_filter                  = true,
693     .bdrv_recurse_is_first_non_filter = replication_recurse_is_first_non_filter,
694 
695     .has_variable_length        = true,
696 };
697 
698 static void bdrv_replication_init(void)
699 {
700     bdrv_register(&bdrv_replication);
701 }
702 
703 block_init(bdrv_replication_init);
704