xref: /openbmc/qemu/block/mirror.c (revision 96d885b9)
1 /*
2  * Image mirroring
3  *
4  * Copyright Red Hat, Inc. 2012
5  *
6  * Authors:
7  *  Paolo Bonzini  <pbonzini@redhat.com>
8  *
9  * This work is licensed under the terms of the GNU LGPL, version 2 or later.
10  * See the COPYING.LIB file in the top-level directory.
11  *
12  */
13 
14 #include "trace.h"
15 #include "block/blockjob.h"
16 #include "block/block_int.h"
17 #include "qapi/qmp/qerror.h"
18 #include "qemu/ratelimit.h"
19 #include "qemu/bitmap.h"
20 
21 #define SLICE_TIME    100000000ULL /* ns */
22 #define MAX_IN_FLIGHT 16
23 #define DEFAULT_MIRROR_BUF_SIZE   (10 << 20)
24 
25 /* The mirroring buffer is a list of granularity-sized chunks.
26  * Free chunks are organized in a list.
27  */
28 typedef struct MirrorBuffer {
29     QSIMPLEQ_ENTRY(MirrorBuffer) next;
30 } MirrorBuffer;
31 
32 typedef struct MirrorBlockJob {
33     BlockJob common;
34     RateLimit limit;
35     BlockDriverState *target;
36     BlockDriverState *base;
37     /* The name of the graph node to replace */
38     char *replaces;
39     /* The BDS to replace */
40     BlockDriverState *to_replace;
41     /* Used to block operations on the drive-mirror-replace target */
42     Error *replace_blocker;
43     bool is_none_mode;
44     BlockdevOnError on_source_error, on_target_error;
45     bool synced;
46     bool should_complete;
47     int64_t sector_num;
48     int64_t granularity;
49     size_t buf_size;
50     int64_t bdev_length;
51     unsigned long *cow_bitmap;
52     BdrvDirtyBitmap *dirty_bitmap;
53     HBitmapIter hbi;
54     uint8_t *buf;
55     QSIMPLEQ_HEAD(, MirrorBuffer) buf_free;
56     int buf_free_count;
57 
58     unsigned long *in_flight_bitmap;
59     int in_flight;
60     int sectors_in_flight;
61     int ret;
62     bool unmap;
63     bool waiting_for_io;
64 } MirrorBlockJob;
65 
66 typedef struct MirrorOp {
67     MirrorBlockJob *s;
68     QEMUIOVector qiov;
69     int64_t sector_num;
70     int nb_sectors;
71 } MirrorOp;
72 
73 static BlockErrorAction mirror_error_action(MirrorBlockJob *s, bool read,
74                                             int error)
75 {
76     s->synced = false;
77     if (read) {
78         return block_job_error_action(&s->common, s->common.bs,
79                                       s->on_source_error, true, error);
80     } else {
81         return block_job_error_action(&s->common, s->target,
82                                       s->on_target_error, false, error);
83     }
84 }
85 
86 static void mirror_iteration_done(MirrorOp *op, int ret)
87 {
88     MirrorBlockJob *s = op->s;
89     struct iovec *iov;
90     int64_t chunk_num;
91     int i, nb_chunks, sectors_per_chunk;
92 
93     trace_mirror_iteration_done(s, op->sector_num, op->nb_sectors, ret);
94 
95     s->in_flight--;
96     s->sectors_in_flight -= op->nb_sectors;
97     iov = op->qiov.iov;
98     for (i = 0; i < op->qiov.niov; i++) {
99         MirrorBuffer *buf = (MirrorBuffer *) iov[i].iov_base;
100         QSIMPLEQ_INSERT_TAIL(&s->buf_free, buf, next);
101         s->buf_free_count++;
102     }
103 
104     sectors_per_chunk = s->granularity >> BDRV_SECTOR_BITS;
105     chunk_num = op->sector_num / sectors_per_chunk;
106     nb_chunks = op->nb_sectors / sectors_per_chunk;
107     bitmap_clear(s->in_flight_bitmap, chunk_num, nb_chunks);
108     if (ret >= 0) {
109         if (s->cow_bitmap) {
110             bitmap_set(s->cow_bitmap, chunk_num, nb_chunks);
111         }
112         s->common.offset += (uint64_t)op->nb_sectors * BDRV_SECTOR_SIZE;
113     }
114 
115     qemu_iovec_destroy(&op->qiov);
116     g_free(op);
117 
118     if (s->waiting_for_io) {
119         qemu_coroutine_enter(s->common.co, NULL);
120     }
121 }
122 
123 static void mirror_write_complete(void *opaque, int ret)
124 {
125     MirrorOp *op = opaque;
126     MirrorBlockJob *s = op->s;
127     if (ret < 0) {
128         BlockErrorAction action;
129 
130         bdrv_set_dirty_bitmap(s->dirty_bitmap, op->sector_num, op->nb_sectors);
131         action = mirror_error_action(s, false, -ret);
132         if (action == BLOCK_ERROR_ACTION_REPORT && s->ret >= 0) {
133             s->ret = ret;
134         }
135     }
136     mirror_iteration_done(op, ret);
137 }
138 
139 static void mirror_read_complete(void *opaque, int ret)
140 {
141     MirrorOp *op = opaque;
142     MirrorBlockJob *s = op->s;
143     if (ret < 0) {
144         BlockErrorAction action;
145 
146         bdrv_set_dirty_bitmap(s->dirty_bitmap, op->sector_num, op->nb_sectors);
147         action = mirror_error_action(s, true, -ret);
148         if (action == BLOCK_ERROR_ACTION_REPORT && s->ret >= 0) {
149             s->ret = ret;
150         }
151 
152         mirror_iteration_done(op, ret);
153         return;
154     }
155     bdrv_aio_writev(s->target, op->sector_num, &op->qiov, op->nb_sectors,
156                     mirror_write_complete, op);
157 }
158 
159 static uint64_t coroutine_fn mirror_iteration(MirrorBlockJob *s)
160 {
161     BlockDriverState *source = s->common.bs;
162     int nb_sectors, sectors_per_chunk, nb_chunks;
163     int64_t end, sector_num, next_chunk, next_sector, hbitmap_next_sector;
164     uint64_t delay_ns = 0;
165     MirrorOp *op;
166     int pnum;
167     int64_t ret;
168 
169     s->sector_num = hbitmap_iter_next(&s->hbi);
170     if (s->sector_num < 0) {
171         bdrv_dirty_iter_init(s->dirty_bitmap, &s->hbi);
172         s->sector_num = hbitmap_iter_next(&s->hbi);
173         trace_mirror_restart_iter(s, bdrv_get_dirty_count(s->dirty_bitmap));
174         assert(s->sector_num >= 0);
175     }
176 
177     hbitmap_next_sector = s->sector_num;
178     sector_num = s->sector_num;
179     sectors_per_chunk = s->granularity >> BDRV_SECTOR_BITS;
180     end = s->bdev_length / BDRV_SECTOR_SIZE;
181 
182     /* Extend the QEMUIOVector to include all adjacent blocks that will
183      * be copied in this operation.
184      *
185      * We have to do this if we have no backing file yet in the destination,
186      * and the cluster size is very large.  Then we need to do COW ourselves.
187      * The first time a cluster is copied, copy it entirely.  Note that,
188      * because both the granularity and the cluster size are powers of two,
189      * the number of sectors to copy cannot exceed one cluster.
190      *
191      * We also want to extend the QEMUIOVector to include more adjacent
192      * dirty blocks if possible, to limit the number of I/O operations and
193      * run efficiently even with a small granularity.
194      */
195     nb_chunks = 0;
196     nb_sectors = 0;
197     next_sector = sector_num;
198     next_chunk = sector_num / sectors_per_chunk;
199 
200     /* Wait for I/O to this cluster (from a previous iteration) to be done.  */
201     while (test_bit(next_chunk, s->in_flight_bitmap)) {
202         trace_mirror_yield_in_flight(s, sector_num, s->in_flight);
203         s->waiting_for_io = true;
204         qemu_coroutine_yield();
205         s->waiting_for_io = false;
206     }
207 
208     do {
209         int added_sectors, added_chunks;
210 
211         if (!bdrv_get_dirty(source, s->dirty_bitmap, next_sector) ||
212             test_bit(next_chunk, s->in_flight_bitmap)) {
213             assert(nb_sectors > 0);
214             break;
215         }
216 
217         added_sectors = sectors_per_chunk;
218         if (s->cow_bitmap && !test_bit(next_chunk, s->cow_bitmap)) {
219             bdrv_round_to_clusters(s->target,
220                                    next_sector, added_sectors,
221                                    &next_sector, &added_sectors);
222 
223             /* On the first iteration, the rounding may make us copy
224              * sectors before the first dirty one.
225              */
226             if (next_sector < sector_num) {
227                 assert(nb_sectors == 0);
228                 sector_num = next_sector;
229                 next_chunk = next_sector / sectors_per_chunk;
230             }
231         }
232 
233         added_sectors = MIN(added_sectors, end - (sector_num + nb_sectors));
234         added_chunks = (added_sectors + sectors_per_chunk - 1) / sectors_per_chunk;
235 
236         /* When doing COW, it may happen that there is not enough space for
237          * a full cluster.  Wait if that is the case.
238          */
239         while (nb_chunks == 0 && s->buf_free_count < added_chunks) {
240             trace_mirror_yield_buf_busy(s, nb_chunks, s->in_flight);
241             s->waiting_for_io = true;
242             qemu_coroutine_yield();
243             s->waiting_for_io = false;
244         }
245         if (s->buf_free_count < nb_chunks + added_chunks) {
246             trace_mirror_break_buf_busy(s, nb_chunks, s->in_flight);
247             break;
248         }
249         if (IOV_MAX < nb_chunks + added_chunks) {
250             trace_mirror_break_iov_max(s, nb_chunks, added_chunks);
251             break;
252         }
253 
254         /* We have enough free space to copy these sectors.  */
255         bitmap_set(s->in_flight_bitmap, next_chunk, added_chunks);
256 
257         nb_sectors += added_sectors;
258         nb_chunks += added_chunks;
259         next_sector += added_sectors;
260         next_chunk += added_chunks;
261         if (!s->synced && s->common.speed) {
262             delay_ns = ratelimit_calculate_delay(&s->limit, added_sectors);
263         }
264     } while (delay_ns == 0 && next_sector < end);
265 
266     /* Allocate a MirrorOp that is used as an AIO callback.  */
267     op = g_new(MirrorOp, 1);
268     op->s = s;
269     op->sector_num = sector_num;
270     op->nb_sectors = nb_sectors;
271 
272     /* Now make a QEMUIOVector taking enough granularity-sized chunks
273      * from s->buf_free.
274      */
275     qemu_iovec_init(&op->qiov, nb_chunks);
276     next_sector = sector_num;
277     while (nb_chunks-- > 0) {
278         MirrorBuffer *buf = QSIMPLEQ_FIRST(&s->buf_free);
279         size_t remaining = (nb_sectors * BDRV_SECTOR_SIZE) - op->qiov.size;
280 
281         QSIMPLEQ_REMOVE_HEAD(&s->buf_free, next);
282         s->buf_free_count--;
283         qemu_iovec_add(&op->qiov, buf, MIN(s->granularity, remaining));
284 
285         /* Advance the HBitmapIter in parallel, so that we do not examine
286          * the same sector twice.
287          */
288         if (next_sector > hbitmap_next_sector
289             && bdrv_get_dirty(source, s->dirty_bitmap, next_sector)) {
290             hbitmap_next_sector = hbitmap_iter_next(&s->hbi);
291         }
292 
293         next_sector += sectors_per_chunk;
294     }
295 
296     bdrv_reset_dirty_bitmap(s->dirty_bitmap, sector_num, nb_sectors);
297 
298     /* Copy the dirty cluster.  */
299     s->in_flight++;
300     s->sectors_in_flight += nb_sectors;
301     trace_mirror_one_iteration(s, sector_num, nb_sectors);
302 
303     ret = bdrv_get_block_status_above(source, NULL, sector_num,
304                                       nb_sectors, &pnum);
305     if (ret < 0 || pnum < nb_sectors ||
306             (ret & BDRV_BLOCK_DATA && !(ret & BDRV_BLOCK_ZERO))) {
307         bdrv_aio_readv(source, sector_num, &op->qiov, nb_sectors,
308                        mirror_read_complete, op);
309     } else if (ret & BDRV_BLOCK_ZERO) {
310         bdrv_aio_write_zeroes(s->target, sector_num, op->nb_sectors,
311                               s->unmap ? BDRV_REQ_MAY_UNMAP : 0,
312                               mirror_write_complete, op);
313     } else {
314         assert(!(ret & BDRV_BLOCK_DATA));
315         bdrv_aio_discard(s->target, sector_num, op->nb_sectors,
316                          mirror_write_complete, op);
317     }
318     return delay_ns;
319 }
320 
321 static void mirror_free_init(MirrorBlockJob *s)
322 {
323     int granularity = s->granularity;
324     size_t buf_size = s->buf_size;
325     uint8_t *buf = s->buf;
326 
327     assert(s->buf_free_count == 0);
328     QSIMPLEQ_INIT(&s->buf_free);
329     while (buf_size != 0) {
330         MirrorBuffer *cur = (MirrorBuffer *)buf;
331         QSIMPLEQ_INSERT_TAIL(&s->buf_free, cur, next);
332         s->buf_free_count++;
333         buf_size -= granularity;
334         buf += granularity;
335     }
336 }
337 
338 static void mirror_drain(MirrorBlockJob *s)
339 {
340     while (s->in_flight > 0) {
341         s->waiting_for_io = true;
342         qemu_coroutine_yield();
343         s->waiting_for_io = false;
344     }
345 }
346 
347 typedef struct {
348     int ret;
349 } MirrorExitData;
350 
351 static void mirror_exit(BlockJob *job, void *opaque)
352 {
353     MirrorBlockJob *s = container_of(job, MirrorBlockJob, common);
354     MirrorExitData *data = opaque;
355     AioContext *replace_aio_context = NULL;
356 
357     if (s->to_replace) {
358         replace_aio_context = bdrv_get_aio_context(s->to_replace);
359         aio_context_acquire(replace_aio_context);
360     }
361 
362     if (s->should_complete && data->ret == 0) {
363         BlockDriverState *to_replace = s->common.bs;
364         if (s->to_replace) {
365             to_replace = s->to_replace;
366         }
367         if (bdrv_get_flags(s->target) != bdrv_get_flags(to_replace)) {
368             bdrv_reopen(s->target, bdrv_get_flags(to_replace), NULL);
369         }
370         bdrv_swap(s->target, to_replace);
371         if (s->common.driver->job_type == BLOCK_JOB_TYPE_COMMIT) {
372             /* drop the bs loop chain formed by the swap: break the loop then
373              * trigger the unref from the top one */
374             BlockDriverState *p = s->base->backing_hd;
375             bdrv_set_backing_hd(s->base, NULL);
376             bdrv_unref(p);
377         }
378     }
379     if (s->to_replace) {
380         bdrv_op_unblock_all(s->to_replace, s->replace_blocker);
381         error_free(s->replace_blocker);
382         bdrv_unref(s->to_replace);
383     }
384     if (replace_aio_context) {
385         aio_context_release(replace_aio_context);
386     }
387     g_free(s->replaces);
388     bdrv_unref(s->target);
389     block_job_completed(&s->common, data->ret);
390     g_free(data);
391 }
392 
393 static void coroutine_fn mirror_run(void *opaque)
394 {
395     MirrorBlockJob *s = opaque;
396     MirrorExitData *data;
397     BlockDriverState *bs = s->common.bs;
398     int64_t sector_num, end, length;
399     uint64_t last_pause_ns;
400     BlockDriverInfo bdi;
401     char backing_filename[2]; /* we only need 2 characters because we are only
402                                  checking for a NULL string */
403     int ret = 0;
404     int n;
405 
406     if (block_job_is_cancelled(&s->common)) {
407         goto immediate_exit;
408     }
409 
410     s->bdev_length = bdrv_getlength(bs);
411     if (s->bdev_length < 0) {
412         ret = s->bdev_length;
413         goto immediate_exit;
414     } else if (s->bdev_length == 0) {
415         /* Report BLOCK_JOB_READY and wait for complete. */
416         block_job_event_ready(&s->common);
417         s->synced = true;
418         while (!block_job_is_cancelled(&s->common) && !s->should_complete) {
419             block_job_yield(&s->common);
420         }
421         s->common.cancelled = false;
422         goto immediate_exit;
423     }
424 
425     length = DIV_ROUND_UP(s->bdev_length, s->granularity);
426     s->in_flight_bitmap = bitmap_new(length);
427 
428     /* If we have no backing file yet in the destination, we cannot let
429      * the destination do COW.  Instead, we copy sectors around the
430      * dirty data if needed.  We need a bitmap to do that.
431      */
432     bdrv_get_backing_filename(s->target, backing_filename,
433                               sizeof(backing_filename));
434     if (backing_filename[0] && !s->target->backing_hd) {
435         ret = bdrv_get_info(s->target, &bdi);
436         if (ret < 0) {
437             goto immediate_exit;
438         }
439         if (s->granularity < bdi.cluster_size) {
440             s->buf_size = MAX(s->buf_size, bdi.cluster_size);
441             s->cow_bitmap = bitmap_new(length);
442         }
443     }
444 
445     end = s->bdev_length / BDRV_SECTOR_SIZE;
446     s->buf = qemu_try_blockalign(bs, s->buf_size);
447     if (s->buf == NULL) {
448         ret = -ENOMEM;
449         goto immediate_exit;
450     }
451 
452     mirror_free_init(s);
453 
454     last_pause_ns = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
455     if (!s->is_none_mode) {
456         /* First part, loop on the sectors and initialize the dirty bitmap.  */
457         BlockDriverState *base = s->base;
458         bool mark_all_dirty = s->base == NULL && !bdrv_has_zero_init(s->target);
459 
460         for (sector_num = 0; sector_num < end; ) {
461             /* Just to make sure we are not exceeding int limit. */
462             int nb_sectors = MIN(INT_MAX >> BDRV_SECTOR_BITS,
463                                  end - sector_num);
464             int64_t now = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
465 
466             if (now - last_pause_ns > SLICE_TIME) {
467                 last_pause_ns = now;
468                 block_job_sleep_ns(&s->common, QEMU_CLOCK_REALTIME, 0);
469             }
470 
471             if (block_job_is_cancelled(&s->common)) {
472                 goto immediate_exit;
473             }
474 
475             ret = bdrv_is_allocated_above(bs, base, sector_num, nb_sectors, &n);
476 
477             if (ret < 0) {
478                 goto immediate_exit;
479             }
480 
481             assert(n > 0);
482             if (ret == 1 || mark_all_dirty) {
483                 bdrv_set_dirty_bitmap(s->dirty_bitmap, sector_num, n);
484             }
485             sector_num += n;
486         }
487     }
488 
489     bdrv_dirty_iter_init(s->dirty_bitmap, &s->hbi);
490     for (;;) {
491         uint64_t delay_ns = 0;
492         int64_t cnt;
493         bool should_complete;
494 
495         if (s->ret < 0) {
496             ret = s->ret;
497             goto immediate_exit;
498         }
499 
500         cnt = bdrv_get_dirty_count(s->dirty_bitmap);
501         /* s->common.offset contains the number of bytes already processed so
502          * far, cnt is the number of dirty sectors remaining and
503          * s->sectors_in_flight is the number of sectors currently being
504          * processed; together those are the current total operation length */
505         s->common.len = s->common.offset +
506                         (cnt + s->sectors_in_flight) * BDRV_SECTOR_SIZE;
507 
508         /* Note that even when no rate limit is applied we need to yield
509          * periodically with no pending I/O so that bdrv_drain_all() returns.
510          * We do so every SLICE_TIME nanoseconds, or when there is an error,
511          * or when the source is clean, whichever comes first.
512          */
513         if (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - last_pause_ns < SLICE_TIME &&
514             s->common.iostatus == BLOCK_DEVICE_IO_STATUS_OK) {
515             if (s->in_flight == MAX_IN_FLIGHT || s->buf_free_count == 0 ||
516                 (cnt == 0 && s->in_flight > 0)) {
517                 trace_mirror_yield(s, s->in_flight, s->buf_free_count, cnt);
518                 s->waiting_for_io = true;
519                 qemu_coroutine_yield();
520                 s->waiting_for_io = false;
521                 continue;
522             } else if (cnt != 0) {
523                 delay_ns = mirror_iteration(s);
524             }
525         }
526 
527         should_complete = false;
528         if (s->in_flight == 0 && cnt == 0) {
529             trace_mirror_before_flush(s);
530             ret = bdrv_flush(s->target);
531             if (ret < 0) {
532                 if (mirror_error_action(s, false, -ret) ==
533                     BLOCK_ERROR_ACTION_REPORT) {
534                     goto immediate_exit;
535                 }
536             } else {
537                 /* We're out of the streaming phase.  From now on, if the job
538                  * is cancelled we will actually complete all pending I/O and
539                  * report completion.  This way, block-job-cancel will leave
540                  * the target in a consistent state.
541                  */
542                 if (!s->synced) {
543                     block_job_event_ready(&s->common);
544                     s->synced = true;
545                 }
546 
547                 should_complete = s->should_complete ||
548                     block_job_is_cancelled(&s->common);
549                 cnt = bdrv_get_dirty_count(s->dirty_bitmap);
550             }
551         }
552 
553         if (cnt == 0 && should_complete) {
554             /* The dirty bitmap is not updated while operations are pending.
555              * If we're about to exit, wait for pending operations before
556              * calling bdrv_get_dirty_count(bs), or we may exit while the
557              * source has dirty data to copy!
558              *
559              * Note that I/O can be submitted by the guest while
560              * mirror_populate runs.
561              */
562             trace_mirror_before_drain(s, cnt);
563             bdrv_drain(bs);
564             cnt = bdrv_get_dirty_count(s->dirty_bitmap);
565         }
566 
567         ret = 0;
568         trace_mirror_before_sleep(s, cnt, s->synced, delay_ns);
569         if (!s->synced) {
570             block_job_sleep_ns(&s->common, QEMU_CLOCK_REALTIME, delay_ns);
571             if (block_job_is_cancelled(&s->common)) {
572                 break;
573             }
574         } else if (!should_complete) {
575             delay_ns = (s->in_flight == 0 && cnt == 0 ? SLICE_TIME : 0);
576             block_job_sleep_ns(&s->common, QEMU_CLOCK_REALTIME, delay_ns);
577         } else if (cnt == 0) {
578             /* The two disks are in sync.  Exit and report successful
579              * completion.
580              */
581             assert(QLIST_EMPTY(&bs->tracked_requests));
582             s->common.cancelled = false;
583             break;
584         }
585         last_pause_ns = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
586     }
587 
588 immediate_exit:
589     if (s->in_flight > 0) {
590         /* We get here only if something went wrong.  Either the job failed,
591          * or it was cancelled prematurely so that we do not guarantee that
592          * the target is a copy of the source.
593          */
594         assert(ret < 0 || (!s->synced && block_job_is_cancelled(&s->common)));
595         mirror_drain(s);
596     }
597 
598     assert(s->in_flight == 0);
599     qemu_vfree(s->buf);
600     g_free(s->cow_bitmap);
601     g_free(s->in_flight_bitmap);
602     bdrv_release_dirty_bitmap(bs, s->dirty_bitmap);
603     bdrv_iostatus_disable(s->target);
604 
605     data = g_malloc(sizeof(*data));
606     data->ret = ret;
607     block_job_defer_to_main_loop(&s->common, mirror_exit, data);
608 }
609 
610 static void mirror_set_speed(BlockJob *job, int64_t speed, Error **errp)
611 {
612     MirrorBlockJob *s = container_of(job, MirrorBlockJob, common);
613 
614     if (speed < 0) {
615         error_setg(errp, QERR_INVALID_PARAMETER, "speed");
616         return;
617     }
618     ratelimit_set_speed(&s->limit, speed / BDRV_SECTOR_SIZE, SLICE_TIME);
619 }
620 
621 static void mirror_iostatus_reset(BlockJob *job)
622 {
623     MirrorBlockJob *s = container_of(job, MirrorBlockJob, common);
624 
625     bdrv_iostatus_reset(s->target);
626 }
627 
628 static void mirror_complete(BlockJob *job, Error **errp)
629 {
630     MirrorBlockJob *s = container_of(job, MirrorBlockJob, common);
631     Error *local_err = NULL;
632     int ret;
633 
634     ret = bdrv_open_backing_file(s->target, NULL, &local_err);
635     if (ret < 0) {
636         error_propagate(errp, local_err);
637         return;
638     }
639     if (!s->synced) {
640         error_setg(errp, QERR_BLOCK_JOB_NOT_READY,
641                    bdrv_get_device_name(job->bs));
642         return;
643     }
644 
645     /* check the target bs is not blocked and block all operations on it */
646     if (s->replaces) {
647         AioContext *replace_aio_context;
648 
649         s->to_replace = bdrv_find_node(s->replaces);
650         if (!s->to_replace) {
651             error_setg(errp, "Node name '%s' not found", s->replaces);
652             return;
653         }
654 
655         replace_aio_context = bdrv_get_aio_context(s->to_replace);
656         aio_context_acquire(replace_aio_context);
657 
658         error_setg(&s->replace_blocker,
659                    "block device is in use by block-job-complete");
660         bdrv_op_block_all(s->to_replace, s->replace_blocker);
661         bdrv_ref(s->to_replace);
662 
663         aio_context_release(replace_aio_context);
664     }
665 
666     s->should_complete = true;
667     block_job_enter(&s->common);
668 }
669 
670 static const BlockJobDriver mirror_job_driver = {
671     .instance_size = sizeof(MirrorBlockJob),
672     .job_type      = BLOCK_JOB_TYPE_MIRROR,
673     .set_speed     = mirror_set_speed,
674     .iostatus_reset= mirror_iostatus_reset,
675     .complete      = mirror_complete,
676 };
677 
678 static const BlockJobDriver commit_active_job_driver = {
679     .instance_size = sizeof(MirrorBlockJob),
680     .job_type      = BLOCK_JOB_TYPE_COMMIT,
681     .set_speed     = mirror_set_speed,
682     .iostatus_reset
683                    = mirror_iostatus_reset,
684     .complete      = mirror_complete,
685 };
686 
687 static void mirror_start_job(BlockDriverState *bs, BlockDriverState *target,
688                              const char *replaces,
689                              int64_t speed, uint32_t granularity,
690                              int64_t buf_size,
691                              BlockdevOnError on_source_error,
692                              BlockdevOnError on_target_error,
693                              bool unmap,
694                              BlockCompletionFunc *cb,
695                              void *opaque, Error **errp,
696                              const BlockJobDriver *driver,
697                              bool is_none_mode, BlockDriverState *base)
698 {
699     MirrorBlockJob *s;
700 
701     if (granularity == 0) {
702         granularity = bdrv_get_default_bitmap_granularity(target);
703     }
704 
705     assert ((granularity & (granularity - 1)) == 0);
706 
707     if ((on_source_error == BLOCKDEV_ON_ERROR_STOP ||
708          on_source_error == BLOCKDEV_ON_ERROR_ENOSPC) &&
709         !bdrv_iostatus_is_enabled(bs)) {
710         error_setg(errp, QERR_INVALID_PARAMETER, "on-source-error");
711         return;
712     }
713 
714     if (buf_size < 0) {
715         error_setg(errp, "Invalid parameter 'buf-size'");
716         return;
717     }
718 
719     if (buf_size == 0) {
720         buf_size = DEFAULT_MIRROR_BUF_SIZE;
721     }
722 
723     s = block_job_create(driver, bs, speed, cb, opaque, errp);
724     if (!s) {
725         return;
726     }
727 
728     s->replaces = g_strdup(replaces);
729     s->on_source_error = on_source_error;
730     s->on_target_error = on_target_error;
731     s->target = target;
732     s->is_none_mode = is_none_mode;
733     s->base = base;
734     s->granularity = granularity;
735     s->buf_size = ROUND_UP(buf_size, granularity);
736     s->unmap = unmap;
737 
738     s->dirty_bitmap = bdrv_create_dirty_bitmap(bs, granularity, NULL, errp);
739     if (!s->dirty_bitmap) {
740         g_free(s->replaces);
741         block_job_release(bs);
742         return;
743     }
744     bdrv_set_enable_write_cache(s->target, true);
745     bdrv_set_on_error(s->target, on_target_error, on_target_error);
746     bdrv_iostatus_enable(s->target);
747     s->common.co = qemu_coroutine_create(mirror_run);
748     trace_mirror_start(bs, s, s->common.co, opaque);
749     qemu_coroutine_enter(s->common.co, s);
750 }
751 
752 void mirror_start(BlockDriverState *bs, BlockDriverState *target,
753                   const char *replaces,
754                   int64_t speed, uint32_t granularity, int64_t buf_size,
755                   MirrorSyncMode mode, BlockdevOnError on_source_error,
756                   BlockdevOnError on_target_error,
757                   bool unmap,
758                   BlockCompletionFunc *cb,
759                   void *opaque, Error **errp)
760 {
761     bool is_none_mode;
762     BlockDriverState *base;
763 
764     if (mode == MIRROR_SYNC_MODE_INCREMENTAL) {
765         error_setg(errp, "Sync mode 'incremental' not supported");
766         return;
767     }
768     is_none_mode = mode == MIRROR_SYNC_MODE_NONE;
769     base = mode == MIRROR_SYNC_MODE_TOP ? bs->backing_hd : NULL;
770     mirror_start_job(bs, target, replaces,
771                      speed, granularity, buf_size,
772                      on_source_error, on_target_error, unmap, cb, opaque, errp,
773                      &mirror_job_driver, is_none_mode, base);
774 }
775 
776 void commit_active_start(BlockDriverState *bs, BlockDriverState *base,
777                          int64_t speed,
778                          BlockdevOnError on_error,
779                          BlockCompletionFunc *cb,
780                          void *opaque, Error **errp)
781 {
782     int64_t length, base_length;
783     int orig_base_flags;
784     int ret;
785     Error *local_err = NULL;
786 
787     orig_base_flags = bdrv_get_flags(base);
788 
789     if (bdrv_reopen(base, bs->open_flags, errp)) {
790         return;
791     }
792 
793     length = bdrv_getlength(bs);
794     if (length < 0) {
795         error_setg_errno(errp, -length,
796                          "Unable to determine length of %s", bs->filename);
797         goto error_restore_flags;
798     }
799 
800     base_length = bdrv_getlength(base);
801     if (base_length < 0) {
802         error_setg_errno(errp, -base_length,
803                          "Unable to determine length of %s", base->filename);
804         goto error_restore_flags;
805     }
806 
807     if (length > base_length) {
808         ret = bdrv_truncate(base, length);
809         if (ret < 0) {
810             error_setg_errno(errp, -ret,
811                             "Top image %s is larger than base image %s, and "
812                              "resize of base image failed",
813                              bs->filename, base->filename);
814             goto error_restore_flags;
815         }
816     }
817 
818     bdrv_ref(base);
819     mirror_start_job(bs, base, NULL, speed, 0, 0,
820                      on_error, on_error, false, cb, opaque, &local_err,
821                      &commit_active_job_driver, false, base);
822     if (local_err) {
823         error_propagate(errp, local_err);
824         goto error_restore_flags;
825     }
826 
827     return;
828 
829 error_restore_flags:
830     /* ignore error and errp for bdrv_reopen, because we want to propagate
831      * the original error */
832     bdrv_reopen(base, orig_base_flags, NULL);
833     return;
834 }
835