]> Git Repo - qemu.git/blob - block/replication.c
Merge remote-tracking branch 'remotes/sstabellini/tags/xen-20161028-tag' into staging
[qemu.git] / block / replication.c
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 <[email protected]>
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
425     aio_context = bdrv_get_aio_context(bs);
426     aio_context_acquire(aio_context);
427     s = bs->opaque;
428
429     if (s->replication_state != BLOCK_REPLICATION_NONE) {
430         error_setg(errp, "Block replication is running or done");
431         aio_context_release(aio_context);
432         return;
433     }
434
435     if (s->mode != mode) {
436         error_setg(errp, "The parameter mode's value is invalid, needs %d,"
437                    " but got %d", s->mode, mode);
438         aio_context_release(aio_context);
439         return;
440     }
441
442     switch (s->mode) {
443     case REPLICATION_MODE_PRIMARY:
444         break;
445     case REPLICATION_MODE_SECONDARY:
446         s->active_disk = bs->file;
447         if (!s->active_disk || !s->active_disk->bs ||
448                                     !s->active_disk->bs->backing) {
449             error_setg(errp, "Active disk doesn't have backing file");
450             aio_context_release(aio_context);
451             return;
452         }
453
454         s->hidden_disk = s->active_disk->bs->backing;
455         if (!s->hidden_disk->bs || !s->hidden_disk->bs->backing) {
456             error_setg(errp, "Hidden disk doesn't have backing file");
457             aio_context_release(aio_context);
458             return;
459         }
460
461         s->secondary_disk = s->hidden_disk->bs->backing;
462         if (!s->secondary_disk->bs || !bdrv_has_blk(s->secondary_disk->bs)) {
463             error_setg(errp, "The secondary disk doesn't have block backend");
464             aio_context_release(aio_context);
465             return;
466         }
467
468         /* verify the length */
469         active_length = bdrv_getlength(s->active_disk->bs);
470         hidden_length = bdrv_getlength(s->hidden_disk->bs);
471         disk_length = bdrv_getlength(s->secondary_disk->bs);
472         if (active_length < 0 || hidden_length < 0 || disk_length < 0 ||
473             active_length != hidden_length || hidden_length != disk_length) {
474             error_setg(errp, "Active disk, hidden disk, secondary disk's length"
475                        " are not the same");
476             aio_context_release(aio_context);
477             return;
478         }
479
480         if (!s->active_disk->bs->drv->bdrv_make_empty ||
481             !s->hidden_disk->bs->drv->bdrv_make_empty) {
482             error_setg(errp,
483                        "Active disk or hidden disk doesn't support make_empty");
484             aio_context_release(aio_context);
485             return;
486         }
487
488         /* reopen the backing file in r/w mode */
489         reopen_backing_file(bs, true, &local_err);
490         if (local_err) {
491             error_propagate(errp, local_err);
492             aio_context_release(aio_context);
493             return;
494         }
495
496         /* start backup job now */
497         error_setg(&s->blocker,
498                    "Block device is in use by internal backup job");
499
500         top_bs = bdrv_lookup_bs(s->top_id, s->top_id, NULL);
501         if (!top_bs || !bdrv_is_root_node(top_bs) ||
502             !check_top_bs(top_bs, bs)) {
503             error_setg(errp, "No top_bs or it is invalid");
504             reopen_backing_file(bs, false, NULL);
505             aio_context_release(aio_context);
506             return;
507         }
508         bdrv_op_block_all(top_bs, s->blocker);
509         bdrv_op_unblock(top_bs, BLOCK_OP_TYPE_DATAPLANE, s->blocker);
510
511         backup_start("replication-backup", s->secondary_disk->bs,
512                      s->hidden_disk->bs, 0, MIRROR_SYNC_MODE_NONE, NULL, false,
513                      BLOCKDEV_ON_ERROR_REPORT, BLOCKDEV_ON_ERROR_REPORT,
514                      backup_job_completed, bs, NULL, &local_err);
515         if (local_err) {
516             error_propagate(errp, local_err);
517             backup_job_cleanup(bs);
518             aio_context_release(aio_context);
519             return;
520         }
521         break;
522     default:
523         aio_context_release(aio_context);
524         abort();
525     }
526
527     s->replication_state = BLOCK_REPLICATION_RUNNING;
528
529     if (s->mode == REPLICATION_MODE_SECONDARY) {
530         secondary_do_checkpoint(s, errp);
531     }
532
533     s->error = 0;
534     aio_context_release(aio_context);
535 }
536
537 static void replication_do_checkpoint(ReplicationState *rs, Error **errp)
538 {
539     BlockDriverState *bs = rs->opaque;
540     BDRVReplicationState *s;
541     AioContext *aio_context;
542
543     aio_context = bdrv_get_aio_context(bs);
544     aio_context_acquire(aio_context);
545     s = bs->opaque;
546
547     if (s->mode == REPLICATION_MODE_SECONDARY) {
548         secondary_do_checkpoint(s, errp);
549     }
550     aio_context_release(aio_context);
551 }
552
553 static void replication_get_error(ReplicationState *rs, Error **errp)
554 {
555     BlockDriverState *bs = rs->opaque;
556     BDRVReplicationState *s;
557     AioContext *aio_context;
558
559     aio_context = bdrv_get_aio_context(bs);
560     aio_context_acquire(aio_context);
561     s = bs->opaque;
562
563     if (s->replication_state != BLOCK_REPLICATION_RUNNING) {
564         error_setg(errp, "Block replication is not running");
565         aio_context_release(aio_context);
566         return;
567     }
568
569     if (s->error) {
570         error_setg(errp, "I/O error occurred");
571         aio_context_release(aio_context);
572         return;
573     }
574     aio_context_release(aio_context);
575 }
576
577 static void replication_done(void *opaque, int ret)
578 {
579     BlockDriverState *bs = opaque;
580     BDRVReplicationState *s = bs->opaque;
581
582     if (ret == 0) {
583         s->replication_state = BLOCK_REPLICATION_DONE;
584
585         /* refresh top bs's filename */
586         bdrv_refresh_filename(bs);
587         s->active_disk = NULL;
588         s->secondary_disk = NULL;
589         s->hidden_disk = NULL;
590         s->error = 0;
591     } else {
592         s->replication_state = BLOCK_REPLICATION_FAILOVER_FAILED;
593         s->error = -EIO;
594     }
595 }
596
597 static void replication_stop(ReplicationState *rs, bool failover, Error **errp)
598 {
599     BlockDriverState *bs = rs->opaque;
600     BDRVReplicationState *s;
601     AioContext *aio_context;
602
603     aio_context = bdrv_get_aio_context(bs);
604     aio_context_acquire(aio_context);
605     s = bs->opaque;
606
607     if (s->replication_state != BLOCK_REPLICATION_RUNNING) {
608         error_setg(errp, "Block replication is not running");
609         aio_context_release(aio_context);
610         return;
611     }
612
613     switch (s->mode) {
614     case REPLICATION_MODE_PRIMARY:
615         s->replication_state = BLOCK_REPLICATION_DONE;
616         s->error = 0;
617         break;
618     case REPLICATION_MODE_SECONDARY:
619         /*
620          * This BDS will be closed, and the job should be completed
621          * before the BDS is closed, because we will access hidden
622          * disk, secondary disk in backup_job_completed().
623          */
624         if (s->secondary_disk->bs->job) {
625             block_job_cancel_sync(s->secondary_disk->bs->job);
626         }
627
628         if (!failover) {
629             secondary_do_checkpoint(s, errp);
630             s->replication_state = BLOCK_REPLICATION_DONE;
631             aio_context_release(aio_context);
632             return;
633         }
634
635         s->replication_state = BLOCK_REPLICATION_FAILOVER;
636         commit_active_start("replication-commit", s->active_disk->bs,
637                             s->secondary_disk->bs, 0, BLOCKDEV_ON_ERROR_REPORT,
638                             replication_done,
639                             bs, errp, true);
640         break;
641     default:
642         aio_context_release(aio_context);
643         abort();
644     }
645     aio_context_release(aio_context);
646 }
647
648 BlockDriver bdrv_replication = {
649     .format_name                = "replication",
650     .protocol_name              = "replication",
651     .instance_size              = sizeof(BDRVReplicationState),
652
653     .bdrv_open                  = replication_open,
654     .bdrv_close                 = replication_close,
655
656     .bdrv_getlength             = replication_getlength,
657     .bdrv_co_readv              = replication_co_readv,
658     .bdrv_co_writev             = replication_co_writev,
659
660     .is_filter                  = true,
661     .bdrv_recurse_is_first_non_filter = replication_recurse_is_first_non_filter,
662
663     .has_variable_length        = true,
664 };
665
666 static void bdrv_replication_init(void)
667 {
668     bdrv_register(&bdrv_replication);
669 }
670
671 block_init(bdrv_replication_init);
This page took 0.064422 seconds and 4 git commands to generate.