xref: /openbmc/qemu/block/mirror.c (revision 6e8a355d)
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 "qemu/osdep.h"
15 #include "qemu/cutils.h"
16 #include "qemu/coroutine.h"
17 #include "qemu/range.h"
18 #include "trace.h"
19 #include "block/blockjob_int.h"
20 #include "block/block_int.h"
21 #include "sysemu/block-backend.h"
22 #include "qapi/error.h"
23 #include "qapi/qmp/qerror.h"
24 #include "qemu/ratelimit.h"
25 #include "qemu/bitmap.h"
26 
27 #define MAX_IN_FLIGHT 16
28 #define MAX_IO_BYTES (1 << 20) /* 1 Mb */
29 #define DEFAULT_MIRROR_BUF_SIZE (MAX_IN_FLIGHT * MAX_IO_BYTES)
30 
31 /* The mirroring buffer is a list of granularity-sized chunks.
32  * Free chunks are organized in a list.
33  */
34 typedef struct MirrorBuffer {
35     QSIMPLEQ_ENTRY(MirrorBuffer) next;
36 } MirrorBuffer;
37 
38 typedef struct MirrorOp MirrorOp;
39 
40 typedef struct MirrorBlockJob {
41     BlockJob common;
42     BlockBackend *target;
43     BlockDriverState *mirror_top_bs;
44     BlockDriverState *base;
45 
46     /* The name of the graph node to replace */
47     char *replaces;
48     /* The BDS to replace */
49     BlockDriverState *to_replace;
50     /* Used to block operations on the drive-mirror-replace target */
51     Error *replace_blocker;
52     bool is_none_mode;
53     BlockMirrorBackingMode backing_mode;
54     /* Whether the target image requires explicit zero-initialization */
55     bool zero_target;
56     MirrorCopyMode copy_mode;
57     BlockdevOnError on_source_error, on_target_error;
58     bool synced;
59     /* Set when the target is synced (dirty bitmap is clean, nothing
60      * in flight) and the job is running in active mode */
61     bool actively_synced;
62     bool should_complete;
63     int64_t granularity;
64     size_t buf_size;
65     int64_t bdev_length;
66     unsigned long *cow_bitmap;
67     BdrvDirtyBitmap *dirty_bitmap;
68     BdrvDirtyBitmapIter *dbi;
69     uint8_t *buf;
70     QSIMPLEQ_HEAD(, MirrorBuffer) buf_free;
71     int buf_free_count;
72 
73     uint64_t last_pause_ns;
74     unsigned long *in_flight_bitmap;
75     int in_flight;
76     int64_t bytes_in_flight;
77     QTAILQ_HEAD(, MirrorOp) ops_in_flight;
78     int ret;
79     bool unmap;
80     int target_cluster_size;
81     int max_iov;
82     bool initial_zeroing_ongoing;
83     int in_active_write_counter;
84     bool prepared;
85     bool in_drain;
86 } MirrorBlockJob;
87 
88 typedef struct MirrorBDSOpaque {
89     MirrorBlockJob *job;
90     bool stop;
91 } MirrorBDSOpaque;
92 
93 struct MirrorOp {
94     MirrorBlockJob *s;
95     QEMUIOVector qiov;
96     int64_t offset;
97     uint64_t bytes;
98 
99     /* The pointee is set by mirror_co_read(), mirror_co_zero(), and
100      * mirror_co_discard() before yielding for the first time */
101     int64_t *bytes_handled;
102 
103     bool is_pseudo_op;
104     bool is_active_write;
105     bool is_in_flight;
106     CoQueue waiting_requests;
107     Coroutine *co;
108 
109     QTAILQ_ENTRY(MirrorOp) next;
110 };
111 
112 typedef enum MirrorMethod {
113     MIRROR_METHOD_COPY,
114     MIRROR_METHOD_ZERO,
115     MIRROR_METHOD_DISCARD,
116 } MirrorMethod;
117 
118 static BlockErrorAction mirror_error_action(MirrorBlockJob *s, bool read,
119                                             int error)
120 {
121     s->synced = false;
122     s->actively_synced = false;
123     if (read) {
124         return block_job_error_action(&s->common, s->on_source_error,
125                                       true, error);
126     } else {
127         return block_job_error_action(&s->common, s->on_target_error,
128                                       false, error);
129     }
130 }
131 
132 static void coroutine_fn mirror_wait_on_conflicts(MirrorOp *self,
133                                                   MirrorBlockJob *s,
134                                                   uint64_t offset,
135                                                   uint64_t bytes)
136 {
137     uint64_t self_start_chunk = offset / s->granularity;
138     uint64_t self_end_chunk = DIV_ROUND_UP(offset + bytes, s->granularity);
139     uint64_t self_nb_chunks = self_end_chunk - self_start_chunk;
140 
141     while (find_next_bit(s->in_flight_bitmap, self_end_chunk,
142                          self_start_chunk) < self_end_chunk &&
143            s->ret >= 0)
144     {
145         MirrorOp *op;
146 
147         QTAILQ_FOREACH(op, &s->ops_in_flight, next) {
148             uint64_t op_start_chunk = op->offset / s->granularity;
149             uint64_t op_nb_chunks = DIV_ROUND_UP(op->offset + op->bytes,
150                                                  s->granularity) -
151                                     op_start_chunk;
152 
153             if (op == self) {
154                 continue;
155             }
156 
157             if (ranges_overlap(self_start_chunk, self_nb_chunks,
158                                op_start_chunk, op_nb_chunks))
159             {
160                 qemu_co_queue_wait(&op->waiting_requests, NULL);
161                 break;
162             }
163         }
164     }
165 }
166 
167 static void coroutine_fn mirror_iteration_done(MirrorOp *op, int ret)
168 {
169     MirrorBlockJob *s = op->s;
170     struct iovec *iov;
171     int64_t chunk_num;
172     int i, nb_chunks;
173 
174     trace_mirror_iteration_done(s, op->offset, op->bytes, ret);
175 
176     s->in_flight--;
177     s->bytes_in_flight -= op->bytes;
178     iov = op->qiov.iov;
179     for (i = 0; i < op->qiov.niov; i++) {
180         MirrorBuffer *buf = (MirrorBuffer *) iov[i].iov_base;
181         QSIMPLEQ_INSERT_TAIL(&s->buf_free, buf, next);
182         s->buf_free_count++;
183     }
184 
185     chunk_num = op->offset / s->granularity;
186     nb_chunks = DIV_ROUND_UP(op->bytes, s->granularity);
187 
188     bitmap_clear(s->in_flight_bitmap, chunk_num, nb_chunks);
189     QTAILQ_REMOVE(&s->ops_in_flight, op, next);
190     if (ret >= 0) {
191         if (s->cow_bitmap) {
192             bitmap_set(s->cow_bitmap, chunk_num, nb_chunks);
193         }
194         if (!s->initial_zeroing_ongoing) {
195             job_progress_update(&s->common.job, op->bytes);
196         }
197     }
198     qemu_iovec_destroy(&op->qiov);
199 
200     qemu_co_queue_restart_all(&op->waiting_requests);
201     g_free(op);
202 }
203 
204 static void coroutine_fn mirror_write_complete(MirrorOp *op, int ret)
205 {
206     MirrorBlockJob *s = op->s;
207 
208     if (ret < 0) {
209         BlockErrorAction action;
210 
211         bdrv_set_dirty_bitmap(s->dirty_bitmap, op->offset, op->bytes);
212         action = mirror_error_action(s, false, -ret);
213         if (action == BLOCK_ERROR_ACTION_REPORT && s->ret >= 0) {
214             s->ret = ret;
215         }
216     }
217 
218     mirror_iteration_done(op, ret);
219 }
220 
221 static void coroutine_fn mirror_read_complete(MirrorOp *op, int ret)
222 {
223     MirrorBlockJob *s = op->s;
224 
225     if (ret < 0) {
226         BlockErrorAction action;
227 
228         bdrv_set_dirty_bitmap(s->dirty_bitmap, op->offset, op->bytes);
229         action = mirror_error_action(s, true, -ret);
230         if (action == BLOCK_ERROR_ACTION_REPORT && s->ret >= 0) {
231             s->ret = ret;
232         }
233 
234         mirror_iteration_done(op, ret);
235         return;
236     }
237 
238     ret = blk_co_pwritev(s->target, op->offset, op->qiov.size, &op->qiov, 0);
239     mirror_write_complete(op, ret);
240 }
241 
242 /* Clip bytes relative to offset to not exceed end-of-file */
243 static inline int64_t mirror_clip_bytes(MirrorBlockJob *s,
244                                         int64_t offset,
245                                         int64_t bytes)
246 {
247     return MIN(bytes, s->bdev_length - offset);
248 }
249 
250 /* Round offset and/or bytes to target cluster if COW is needed, and
251  * return the offset of the adjusted tail against original. */
252 static int mirror_cow_align(MirrorBlockJob *s, int64_t *offset,
253                             uint64_t *bytes)
254 {
255     bool need_cow;
256     int ret = 0;
257     int64_t align_offset = *offset;
258     int64_t align_bytes = *bytes;
259     int max_bytes = s->granularity * s->max_iov;
260 
261     need_cow = !test_bit(*offset / s->granularity, s->cow_bitmap);
262     need_cow |= !test_bit((*offset + *bytes - 1) / s->granularity,
263                           s->cow_bitmap);
264     if (need_cow) {
265         bdrv_round_to_clusters(blk_bs(s->target), *offset, *bytes,
266                                &align_offset, &align_bytes);
267     }
268 
269     if (align_bytes > max_bytes) {
270         align_bytes = max_bytes;
271         if (need_cow) {
272             align_bytes = QEMU_ALIGN_DOWN(align_bytes, s->target_cluster_size);
273         }
274     }
275     /* Clipping may result in align_bytes unaligned to chunk boundary, but
276      * that doesn't matter because it's already the end of source image. */
277     align_bytes = mirror_clip_bytes(s, align_offset, align_bytes);
278 
279     ret = align_offset + align_bytes - (*offset + *bytes);
280     *offset = align_offset;
281     *bytes = align_bytes;
282     assert(ret >= 0);
283     return ret;
284 }
285 
286 static inline void coroutine_fn
287 mirror_wait_for_any_operation(MirrorBlockJob *s, bool active)
288 {
289     MirrorOp *op;
290 
291     QTAILQ_FOREACH(op, &s->ops_in_flight, next) {
292         /* Do not wait on pseudo ops, because it may in turn wait on
293          * some other operation to start, which may in fact be the
294          * caller of this function.  Since there is only one pseudo op
295          * at any given time, we will always find some real operation
296          * to wait on. */
297         if (!op->is_pseudo_op && op->is_in_flight &&
298             op->is_active_write == active)
299         {
300             qemu_co_queue_wait(&op->waiting_requests, NULL);
301             return;
302         }
303     }
304     abort();
305 }
306 
307 static inline void coroutine_fn
308 mirror_wait_for_free_in_flight_slot(MirrorBlockJob *s)
309 {
310     /* Only non-active operations use up in-flight slots */
311     mirror_wait_for_any_operation(s, false);
312 }
313 
314 /* Perform a mirror copy operation.
315  *
316  * *op->bytes_handled is set to the number of bytes copied after and
317  * including offset, excluding any bytes copied prior to offset due
318  * to alignment.  This will be op->bytes if no alignment is necessary,
319  * or (new_end - op->offset) if the tail is rounded up or down due to
320  * alignment or buffer limit.
321  */
322 static void coroutine_fn mirror_co_read(void *opaque)
323 {
324     MirrorOp *op = opaque;
325     MirrorBlockJob *s = op->s;
326     int nb_chunks;
327     uint64_t ret;
328     uint64_t max_bytes;
329 
330     max_bytes = s->granularity * s->max_iov;
331 
332     /* We can only handle as much as buf_size at a time. */
333     op->bytes = MIN(s->buf_size, MIN(max_bytes, op->bytes));
334     assert(op->bytes);
335     assert(op->bytes < BDRV_REQUEST_MAX_BYTES);
336     *op->bytes_handled = op->bytes;
337 
338     if (s->cow_bitmap) {
339         *op->bytes_handled += mirror_cow_align(s, &op->offset, &op->bytes);
340     }
341     /* Cannot exceed BDRV_REQUEST_MAX_BYTES + INT_MAX */
342     assert(*op->bytes_handled <= UINT_MAX);
343     assert(op->bytes <= s->buf_size);
344     /* The offset is granularity-aligned because:
345      * 1) Caller passes in aligned values;
346      * 2) mirror_cow_align is used only when target cluster is larger. */
347     assert(QEMU_IS_ALIGNED(op->offset, s->granularity));
348     /* The range is sector-aligned, since bdrv_getlength() rounds up. */
349     assert(QEMU_IS_ALIGNED(op->bytes, BDRV_SECTOR_SIZE));
350     nb_chunks = DIV_ROUND_UP(op->bytes, s->granularity);
351 
352     while (s->buf_free_count < nb_chunks) {
353         trace_mirror_yield_in_flight(s, op->offset, s->in_flight);
354         mirror_wait_for_free_in_flight_slot(s);
355     }
356 
357     /* Now make a QEMUIOVector taking enough granularity-sized chunks
358      * from s->buf_free.
359      */
360     qemu_iovec_init(&op->qiov, nb_chunks);
361     while (nb_chunks-- > 0) {
362         MirrorBuffer *buf = QSIMPLEQ_FIRST(&s->buf_free);
363         size_t remaining = op->bytes - op->qiov.size;
364 
365         QSIMPLEQ_REMOVE_HEAD(&s->buf_free, next);
366         s->buf_free_count--;
367         qemu_iovec_add(&op->qiov, buf, MIN(s->granularity, remaining));
368     }
369 
370     /* Copy the dirty cluster.  */
371     s->in_flight++;
372     s->bytes_in_flight += op->bytes;
373     op->is_in_flight = true;
374     trace_mirror_one_iteration(s, op->offset, op->bytes);
375 
376     ret = bdrv_co_preadv(s->mirror_top_bs->backing, op->offset, op->bytes,
377                          &op->qiov, 0);
378     mirror_read_complete(op, ret);
379 }
380 
381 static void coroutine_fn mirror_co_zero(void *opaque)
382 {
383     MirrorOp *op = opaque;
384     int ret;
385 
386     op->s->in_flight++;
387     op->s->bytes_in_flight += op->bytes;
388     *op->bytes_handled = op->bytes;
389     op->is_in_flight = true;
390 
391     ret = blk_co_pwrite_zeroes(op->s->target, op->offset, op->bytes,
392                                op->s->unmap ? BDRV_REQ_MAY_UNMAP : 0);
393     mirror_write_complete(op, ret);
394 }
395 
396 static void coroutine_fn mirror_co_discard(void *opaque)
397 {
398     MirrorOp *op = opaque;
399     int ret;
400 
401     op->s->in_flight++;
402     op->s->bytes_in_flight += op->bytes;
403     *op->bytes_handled = op->bytes;
404     op->is_in_flight = true;
405 
406     ret = blk_co_pdiscard(op->s->target, op->offset, op->bytes);
407     mirror_write_complete(op, ret);
408 }
409 
410 static unsigned mirror_perform(MirrorBlockJob *s, int64_t offset,
411                                unsigned bytes, MirrorMethod mirror_method)
412 {
413     MirrorOp *op;
414     Coroutine *co;
415     int64_t bytes_handled = -1;
416 
417     op = g_new(MirrorOp, 1);
418     *op = (MirrorOp){
419         .s              = s,
420         .offset         = offset,
421         .bytes          = bytes,
422         .bytes_handled  = &bytes_handled,
423     };
424     qemu_co_queue_init(&op->waiting_requests);
425 
426     switch (mirror_method) {
427     case MIRROR_METHOD_COPY:
428         co = qemu_coroutine_create(mirror_co_read, op);
429         break;
430     case MIRROR_METHOD_ZERO:
431         co = qemu_coroutine_create(mirror_co_zero, op);
432         break;
433     case MIRROR_METHOD_DISCARD:
434         co = qemu_coroutine_create(mirror_co_discard, op);
435         break;
436     default:
437         abort();
438     }
439     op->co = co;
440 
441     QTAILQ_INSERT_TAIL(&s->ops_in_flight, op, next);
442     qemu_coroutine_enter(co);
443     /* At this point, ownership of op has been moved to the coroutine
444      * and the object may already be freed */
445 
446     /* Assert that this value has been set */
447     assert(bytes_handled >= 0);
448 
449     /* Same assertion as in mirror_co_read() (and for mirror_co_read()
450      * and mirror_co_discard(), bytes_handled == op->bytes, which
451      * is the @bytes parameter given to this function) */
452     assert(bytes_handled <= UINT_MAX);
453     return bytes_handled;
454 }
455 
456 static uint64_t coroutine_fn mirror_iteration(MirrorBlockJob *s)
457 {
458     BlockDriverState *source = s->mirror_top_bs->backing->bs;
459     MirrorOp *pseudo_op;
460     int64_t offset;
461     uint64_t delay_ns = 0, ret = 0;
462     /* At least the first dirty chunk is mirrored in one iteration. */
463     int nb_chunks = 1;
464     bool write_zeroes_ok = bdrv_can_write_zeroes_with_unmap(blk_bs(s->target));
465     int max_io_bytes = MAX(s->buf_size / MAX_IN_FLIGHT, MAX_IO_BYTES);
466 
467     bdrv_dirty_bitmap_lock(s->dirty_bitmap);
468     offset = bdrv_dirty_iter_next(s->dbi);
469     if (offset < 0) {
470         bdrv_set_dirty_iter(s->dbi, 0);
471         offset = bdrv_dirty_iter_next(s->dbi);
472         trace_mirror_restart_iter(s, bdrv_get_dirty_count(s->dirty_bitmap));
473         assert(offset >= 0);
474     }
475     bdrv_dirty_bitmap_unlock(s->dirty_bitmap);
476 
477     mirror_wait_on_conflicts(NULL, s, offset, 1);
478 
479     job_pause_point(&s->common.job);
480 
481     /* Find the number of consective dirty chunks following the first dirty
482      * one, and wait for in flight requests in them. */
483     bdrv_dirty_bitmap_lock(s->dirty_bitmap);
484     while (nb_chunks * s->granularity < s->buf_size) {
485         int64_t next_dirty;
486         int64_t next_offset = offset + nb_chunks * s->granularity;
487         int64_t next_chunk = next_offset / s->granularity;
488         if (next_offset >= s->bdev_length ||
489             !bdrv_dirty_bitmap_get_locked(s->dirty_bitmap, next_offset)) {
490             break;
491         }
492         if (test_bit(next_chunk, s->in_flight_bitmap)) {
493             break;
494         }
495 
496         next_dirty = bdrv_dirty_iter_next(s->dbi);
497         if (next_dirty > next_offset || next_dirty < 0) {
498             /* The bitmap iterator's cache is stale, refresh it */
499             bdrv_set_dirty_iter(s->dbi, next_offset);
500             next_dirty = bdrv_dirty_iter_next(s->dbi);
501         }
502         assert(next_dirty == next_offset);
503         nb_chunks++;
504     }
505 
506     /* Clear dirty bits before querying the block status, because
507      * calling bdrv_block_status_above could yield - if some blocks are
508      * marked dirty in this window, we need to know.
509      */
510     bdrv_reset_dirty_bitmap_locked(s->dirty_bitmap, offset,
511                                    nb_chunks * s->granularity);
512     bdrv_dirty_bitmap_unlock(s->dirty_bitmap);
513 
514     /* Before claiming an area in the in-flight bitmap, we have to
515      * create a MirrorOp for it so that conflicting requests can wait
516      * for it.  mirror_perform() will create the real MirrorOps later,
517      * for now we just create a pseudo operation that will wake up all
518      * conflicting requests once all real operations have been
519      * launched. */
520     pseudo_op = g_new(MirrorOp, 1);
521     *pseudo_op = (MirrorOp){
522         .offset         = offset,
523         .bytes          = nb_chunks * s->granularity,
524         .is_pseudo_op   = true,
525     };
526     qemu_co_queue_init(&pseudo_op->waiting_requests);
527     QTAILQ_INSERT_TAIL(&s->ops_in_flight, pseudo_op, next);
528 
529     bitmap_set(s->in_flight_bitmap, offset / s->granularity, nb_chunks);
530     while (nb_chunks > 0 && offset < s->bdev_length) {
531         int ret;
532         int64_t io_bytes;
533         int64_t io_bytes_acct;
534         MirrorMethod mirror_method = MIRROR_METHOD_COPY;
535 
536         assert(!(offset % s->granularity));
537         ret = bdrv_block_status_above(source, NULL, offset,
538                                       nb_chunks * s->granularity,
539                                       &io_bytes, NULL, NULL);
540         if (ret < 0) {
541             io_bytes = MIN(nb_chunks * s->granularity, max_io_bytes);
542         } else if (ret & BDRV_BLOCK_DATA) {
543             io_bytes = MIN(io_bytes, max_io_bytes);
544         }
545 
546         io_bytes -= io_bytes % s->granularity;
547         if (io_bytes < s->granularity) {
548             io_bytes = s->granularity;
549         } else if (ret >= 0 && !(ret & BDRV_BLOCK_DATA)) {
550             int64_t target_offset;
551             int64_t target_bytes;
552             bdrv_round_to_clusters(blk_bs(s->target), offset, io_bytes,
553                                    &target_offset, &target_bytes);
554             if (target_offset == offset &&
555                 target_bytes == io_bytes) {
556                 mirror_method = ret & BDRV_BLOCK_ZERO ?
557                                     MIRROR_METHOD_ZERO :
558                                     MIRROR_METHOD_DISCARD;
559             }
560         }
561 
562         while (s->in_flight >= MAX_IN_FLIGHT) {
563             trace_mirror_yield_in_flight(s, offset, s->in_flight);
564             mirror_wait_for_free_in_flight_slot(s);
565         }
566 
567         if (s->ret < 0) {
568             ret = 0;
569             goto fail;
570         }
571 
572         io_bytes = mirror_clip_bytes(s, offset, io_bytes);
573         io_bytes = mirror_perform(s, offset, io_bytes, mirror_method);
574         if (mirror_method != MIRROR_METHOD_COPY && write_zeroes_ok) {
575             io_bytes_acct = 0;
576         } else {
577             io_bytes_acct = io_bytes;
578         }
579         assert(io_bytes);
580         offset += io_bytes;
581         nb_chunks -= DIV_ROUND_UP(io_bytes, s->granularity);
582         delay_ns = block_job_ratelimit_get_delay(&s->common, io_bytes_acct);
583     }
584 
585     ret = delay_ns;
586 fail:
587     QTAILQ_REMOVE(&s->ops_in_flight, pseudo_op, next);
588     qemu_co_queue_restart_all(&pseudo_op->waiting_requests);
589     g_free(pseudo_op);
590 
591     return ret;
592 }
593 
594 static void mirror_free_init(MirrorBlockJob *s)
595 {
596     int granularity = s->granularity;
597     size_t buf_size = s->buf_size;
598     uint8_t *buf = s->buf;
599 
600     assert(s->buf_free_count == 0);
601     QSIMPLEQ_INIT(&s->buf_free);
602     while (buf_size != 0) {
603         MirrorBuffer *cur = (MirrorBuffer *)buf;
604         QSIMPLEQ_INSERT_TAIL(&s->buf_free, cur, next);
605         s->buf_free_count++;
606         buf_size -= granularity;
607         buf += granularity;
608     }
609 }
610 
611 /* This is also used for the .pause callback. There is no matching
612  * mirror_resume() because mirror_run() will begin iterating again
613  * when the job is resumed.
614  */
615 static void coroutine_fn mirror_wait_for_all_io(MirrorBlockJob *s)
616 {
617     while (s->in_flight > 0) {
618         mirror_wait_for_free_in_flight_slot(s);
619     }
620 }
621 
622 /**
623  * mirror_exit_common: handle both abort() and prepare() cases.
624  * for .prepare, returns 0 on success and -errno on failure.
625  * for .abort cases, denoted by abort = true, MUST return 0.
626  */
627 static int mirror_exit_common(Job *job)
628 {
629     MirrorBlockJob *s = container_of(job, MirrorBlockJob, common.job);
630     BlockJob *bjob = &s->common;
631     MirrorBDSOpaque *bs_opaque;
632     AioContext *replace_aio_context = NULL;
633     BlockDriverState *src;
634     BlockDriverState *target_bs;
635     BlockDriverState *mirror_top_bs;
636     Error *local_err = NULL;
637     bool abort = job->ret < 0;
638     int ret = 0;
639 
640     if (s->prepared) {
641         return 0;
642     }
643     s->prepared = true;
644 
645     mirror_top_bs = s->mirror_top_bs;
646     bs_opaque = mirror_top_bs->opaque;
647     src = mirror_top_bs->backing->bs;
648     target_bs = blk_bs(s->target);
649 
650     if (bdrv_chain_contains(src, target_bs)) {
651         bdrv_unfreeze_backing_chain(mirror_top_bs, target_bs);
652     }
653 
654     bdrv_release_dirty_bitmap(s->dirty_bitmap);
655 
656     /* Make sure that the source BDS doesn't go away during bdrv_replace_node,
657      * before we can call bdrv_drained_end */
658     bdrv_ref(src);
659     bdrv_ref(mirror_top_bs);
660     bdrv_ref(target_bs);
661 
662     /*
663      * Remove target parent that still uses BLK_PERM_WRITE/RESIZE before
664      * inserting target_bs at s->to_replace, where we might not be able to get
665      * these permissions.
666      */
667     blk_unref(s->target);
668     s->target = NULL;
669 
670     /* We don't access the source any more. Dropping any WRITE/RESIZE is
671      * required before it could become a backing file of target_bs. Not having
672      * these permissions any more means that we can't allow any new requests on
673      * mirror_top_bs from now on, so keep it drained. */
674     bdrv_drained_begin(mirror_top_bs);
675     bs_opaque->stop = true;
676     bdrv_child_refresh_perms(mirror_top_bs, mirror_top_bs->backing,
677                              &error_abort);
678     if (!abort && s->backing_mode == MIRROR_SOURCE_BACKING_CHAIN) {
679         BlockDriverState *backing = s->is_none_mode ? src : s->base;
680         if (backing_bs(target_bs) != backing) {
681             bdrv_set_backing_hd(target_bs, backing, &local_err);
682             if (local_err) {
683                 error_report_err(local_err);
684                 local_err = NULL;
685                 ret = -EPERM;
686             }
687         }
688     }
689 
690     if (s->to_replace) {
691         replace_aio_context = bdrv_get_aio_context(s->to_replace);
692         aio_context_acquire(replace_aio_context);
693     }
694 
695     if (s->should_complete && !abort) {
696         BlockDriverState *to_replace = s->to_replace ?: src;
697         bool ro = bdrv_is_read_only(to_replace);
698 
699         if (ro != bdrv_is_read_only(target_bs)) {
700             bdrv_reopen_set_read_only(target_bs, ro, NULL);
701         }
702 
703         /* The mirror job has no requests in flight any more, but we need to
704          * drain potential other users of the BDS before changing the graph. */
705         assert(s->in_drain);
706         bdrv_drained_begin(target_bs);
707         /*
708          * Cannot use check_to_replace_node() here, because that would
709          * check for an op blocker on @to_replace, and we have our own
710          * there.
711          */
712         if (bdrv_recurse_can_replace(src, to_replace)) {
713             bdrv_replace_node(to_replace, target_bs, &local_err);
714         } else {
715             error_setg(&local_err, "Can no longer replace '%s' by '%s', "
716                        "because it can no longer be guaranteed that doing so "
717                        "would not lead to an abrupt change of visible data",
718                        to_replace->node_name, target_bs->node_name);
719         }
720         bdrv_drained_end(target_bs);
721         if (local_err) {
722             error_report_err(local_err);
723             ret = -EPERM;
724         }
725     }
726     if (s->to_replace) {
727         bdrv_op_unblock_all(s->to_replace, s->replace_blocker);
728         error_free(s->replace_blocker);
729         bdrv_unref(s->to_replace);
730     }
731     if (replace_aio_context) {
732         aio_context_release(replace_aio_context);
733     }
734     g_free(s->replaces);
735     bdrv_unref(target_bs);
736 
737     /*
738      * Remove the mirror filter driver from the graph. Before this, get rid of
739      * the blockers on the intermediate nodes so that the resulting state is
740      * valid.
741      */
742     block_job_remove_all_bdrv(bjob);
743     bdrv_replace_node(mirror_top_bs, backing_bs(mirror_top_bs), &error_abort);
744 
745     /* We just changed the BDS the job BB refers to (with either or both of the
746      * bdrv_replace_node() calls), so switch the BB back so the cleanup does
747      * the right thing. We don't need any permissions any more now. */
748     blk_remove_bs(bjob->blk);
749     blk_set_perm(bjob->blk, 0, BLK_PERM_ALL, &error_abort);
750     blk_insert_bs(bjob->blk, mirror_top_bs, &error_abort);
751 
752     bs_opaque->job = NULL;
753 
754     bdrv_drained_end(src);
755     bdrv_drained_end(mirror_top_bs);
756     s->in_drain = false;
757     bdrv_unref(mirror_top_bs);
758     bdrv_unref(src);
759 
760     return ret;
761 }
762 
763 static int mirror_prepare(Job *job)
764 {
765     return mirror_exit_common(job);
766 }
767 
768 static void mirror_abort(Job *job)
769 {
770     int ret = mirror_exit_common(job);
771     assert(ret == 0);
772 }
773 
774 static void coroutine_fn mirror_throttle(MirrorBlockJob *s)
775 {
776     int64_t now = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
777 
778     if (now - s->last_pause_ns > BLOCK_JOB_SLICE_TIME) {
779         s->last_pause_ns = now;
780         job_sleep_ns(&s->common.job, 0);
781     } else {
782         job_pause_point(&s->common.job);
783     }
784 }
785 
786 static int coroutine_fn mirror_dirty_init(MirrorBlockJob *s)
787 {
788     int64_t offset;
789     BlockDriverState *base = s->base;
790     BlockDriverState *bs = s->mirror_top_bs->backing->bs;
791     BlockDriverState *target_bs = blk_bs(s->target);
792     int ret;
793     int64_t count;
794 
795     if (s->zero_target) {
796         if (!bdrv_can_write_zeroes_with_unmap(target_bs)) {
797             bdrv_set_dirty_bitmap(s->dirty_bitmap, 0, s->bdev_length);
798             return 0;
799         }
800 
801         s->initial_zeroing_ongoing = true;
802         for (offset = 0; offset < s->bdev_length; ) {
803             int bytes = MIN(s->bdev_length - offset,
804                             QEMU_ALIGN_DOWN(INT_MAX, s->granularity));
805 
806             mirror_throttle(s);
807 
808             if (job_is_cancelled(&s->common.job)) {
809                 s->initial_zeroing_ongoing = false;
810                 return 0;
811             }
812 
813             if (s->in_flight >= MAX_IN_FLIGHT) {
814                 trace_mirror_yield(s, UINT64_MAX, s->buf_free_count,
815                                    s->in_flight);
816                 mirror_wait_for_free_in_flight_slot(s);
817                 continue;
818             }
819 
820             mirror_perform(s, offset, bytes, MIRROR_METHOD_ZERO);
821             offset += bytes;
822         }
823 
824         mirror_wait_for_all_io(s);
825         s->initial_zeroing_ongoing = false;
826     }
827 
828     /* First part, loop on the sectors and initialize the dirty bitmap.  */
829     for (offset = 0; offset < s->bdev_length; ) {
830         /* Just to make sure we are not exceeding int limit. */
831         int bytes = MIN(s->bdev_length - offset,
832                         QEMU_ALIGN_DOWN(INT_MAX, s->granularity));
833 
834         mirror_throttle(s);
835 
836         if (job_is_cancelled(&s->common.job)) {
837             return 0;
838         }
839 
840         ret = bdrv_is_allocated_above(bs, base, false, offset, bytes, &count);
841         if (ret < 0) {
842             return ret;
843         }
844 
845         assert(count);
846         if (ret == 1) {
847             bdrv_set_dirty_bitmap(s->dirty_bitmap, offset, count);
848         }
849         offset += count;
850     }
851     return 0;
852 }
853 
854 /* Called when going out of the streaming phase to flush the bulk of the
855  * data to the medium, or just before completing.
856  */
857 static int mirror_flush(MirrorBlockJob *s)
858 {
859     int ret = blk_flush(s->target);
860     if (ret < 0) {
861         if (mirror_error_action(s, false, -ret) == BLOCK_ERROR_ACTION_REPORT) {
862             s->ret = ret;
863         }
864     }
865     return ret;
866 }
867 
868 static int coroutine_fn mirror_run(Job *job, Error **errp)
869 {
870     MirrorBlockJob *s = container_of(job, MirrorBlockJob, common.job);
871     BlockDriverState *bs = s->mirror_top_bs->backing->bs;
872     BlockDriverState *target_bs = blk_bs(s->target);
873     bool need_drain = true;
874     int64_t length;
875     BlockDriverInfo bdi;
876     char backing_filename[2]; /* we only need 2 characters because we are only
877                                  checking for a NULL string */
878     int ret = 0;
879 
880     if (job_is_cancelled(&s->common.job)) {
881         goto immediate_exit;
882     }
883 
884     s->bdev_length = bdrv_getlength(bs);
885     if (s->bdev_length < 0) {
886         ret = s->bdev_length;
887         goto immediate_exit;
888     }
889 
890     /* Active commit must resize the base image if its size differs from the
891      * active layer. */
892     if (s->base == blk_bs(s->target)) {
893         int64_t base_length;
894 
895         base_length = blk_getlength(s->target);
896         if (base_length < 0) {
897             ret = base_length;
898             goto immediate_exit;
899         }
900 
901         if (s->bdev_length > base_length) {
902             ret = blk_truncate(s->target, s->bdev_length, false,
903                                PREALLOC_MODE_OFF, 0, NULL);
904             if (ret < 0) {
905                 goto immediate_exit;
906             }
907         }
908     }
909 
910     if (s->bdev_length == 0) {
911         /* Transition to the READY state and wait for complete. */
912         job_transition_to_ready(&s->common.job);
913         s->synced = true;
914         s->actively_synced = true;
915         while (!job_is_cancelled(&s->common.job) && !s->should_complete) {
916             job_yield(&s->common.job);
917         }
918         s->common.job.cancelled = false;
919         goto immediate_exit;
920     }
921 
922     length = DIV_ROUND_UP(s->bdev_length, s->granularity);
923     s->in_flight_bitmap = bitmap_new(length);
924 
925     /* If we have no backing file yet in the destination, we cannot let
926      * the destination do COW.  Instead, we copy sectors around the
927      * dirty data if needed.  We need a bitmap to do that.
928      */
929     bdrv_get_backing_filename(target_bs, backing_filename,
930                               sizeof(backing_filename));
931     if (!bdrv_get_info(target_bs, &bdi) && bdi.cluster_size) {
932         s->target_cluster_size = bdi.cluster_size;
933     } else {
934         s->target_cluster_size = BDRV_SECTOR_SIZE;
935     }
936     if (backing_filename[0] && !target_bs->backing &&
937         s->granularity < s->target_cluster_size) {
938         s->buf_size = MAX(s->buf_size, s->target_cluster_size);
939         s->cow_bitmap = bitmap_new(length);
940     }
941     s->max_iov = MIN(bs->bl.max_iov, target_bs->bl.max_iov);
942 
943     s->buf = qemu_try_blockalign(bs, s->buf_size);
944     if (s->buf == NULL) {
945         ret = -ENOMEM;
946         goto immediate_exit;
947     }
948 
949     mirror_free_init(s);
950 
951     s->last_pause_ns = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
952     if (!s->is_none_mode) {
953         ret = mirror_dirty_init(s);
954         if (ret < 0 || job_is_cancelled(&s->common.job)) {
955             goto immediate_exit;
956         }
957     }
958 
959     assert(!s->dbi);
960     s->dbi = bdrv_dirty_iter_new(s->dirty_bitmap);
961     for (;;) {
962         uint64_t delay_ns = 0;
963         int64_t cnt, delta;
964         bool should_complete;
965 
966         /* Do not start passive operations while there are active
967          * writes in progress */
968         while (s->in_active_write_counter) {
969             mirror_wait_for_any_operation(s, true);
970         }
971 
972         if (s->ret < 0) {
973             ret = s->ret;
974             goto immediate_exit;
975         }
976 
977         job_pause_point(&s->common.job);
978 
979         cnt = bdrv_get_dirty_count(s->dirty_bitmap);
980         /* cnt is the number of dirty bytes remaining and s->bytes_in_flight is
981          * the number of bytes currently being processed; together those are
982          * the current remaining operation length */
983         job_progress_set_remaining(&s->common.job, s->bytes_in_flight + cnt);
984 
985         /* Note that even when no rate limit is applied we need to yield
986          * periodically with no pending I/O so that bdrv_drain_all() returns.
987          * We do so every BLKOCK_JOB_SLICE_TIME nanoseconds, or when there is
988          * an error, or when the source is clean, whichever comes first. */
989         delta = qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - s->last_pause_ns;
990         if (delta < BLOCK_JOB_SLICE_TIME &&
991             s->common.iostatus == BLOCK_DEVICE_IO_STATUS_OK) {
992             if (s->in_flight >= MAX_IN_FLIGHT || s->buf_free_count == 0 ||
993                 (cnt == 0 && s->in_flight > 0)) {
994                 trace_mirror_yield(s, cnt, s->buf_free_count, s->in_flight);
995                 mirror_wait_for_free_in_flight_slot(s);
996                 continue;
997             } else if (cnt != 0) {
998                 delay_ns = mirror_iteration(s);
999             }
1000         }
1001 
1002         should_complete = false;
1003         if (s->in_flight == 0 && cnt == 0) {
1004             trace_mirror_before_flush(s);
1005             if (!s->synced) {
1006                 if (mirror_flush(s) < 0) {
1007                     /* Go check s->ret.  */
1008                     continue;
1009                 }
1010                 /* We're out of the streaming phase.  From now on, if the job
1011                  * is cancelled we will actually complete all pending I/O and
1012                  * report completion.  This way, block-job-cancel will leave
1013                  * the target in a consistent state.
1014                  */
1015                 job_transition_to_ready(&s->common.job);
1016                 s->synced = true;
1017                 if (s->copy_mode != MIRROR_COPY_MODE_BACKGROUND) {
1018                     s->actively_synced = true;
1019                 }
1020             }
1021 
1022             should_complete = s->should_complete ||
1023                 job_is_cancelled(&s->common.job);
1024             cnt = bdrv_get_dirty_count(s->dirty_bitmap);
1025         }
1026 
1027         if (cnt == 0 && should_complete) {
1028             /* The dirty bitmap is not updated while operations are pending.
1029              * If we're about to exit, wait for pending operations before
1030              * calling bdrv_get_dirty_count(bs), or we may exit while the
1031              * source has dirty data to copy!
1032              *
1033              * Note that I/O can be submitted by the guest while
1034              * mirror_populate runs, so pause it now.  Before deciding
1035              * whether to switch to target check one last time if I/O has
1036              * come in the meanwhile, and if not flush the data to disk.
1037              */
1038             trace_mirror_before_drain(s, cnt);
1039 
1040             s->in_drain = true;
1041             bdrv_drained_begin(bs);
1042             cnt = bdrv_get_dirty_count(s->dirty_bitmap);
1043             if (cnt > 0 || mirror_flush(s) < 0) {
1044                 bdrv_drained_end(bs);
1045                 s->in_drain = false;
1046                 continue;
1047             }
1048 
1049             /* The two disks are in sync.  Exit and report successful
1050              * completion.
1051              */
1052             assert(QLIST_EMPTY(&bs->tracked_requests));
1053             s->common.job.cancelled = false;
1054             need_drain = false;
1055             break;
1056         }
1057 
1058         ret = 0;
1059 
1060         if (s->synced && !should_complete) {
1061             delay_ns = (s->in_flight == 0 &&
1062                         cnt == 0 ? BLOCK_JOB_SLICE_TIME : 0);
1063         }
1064         trace_mirror_before_sleep(s, cnt, s->synced, delay_ns);
1065         job_sleep_ns(&s->common.job, delay_ns);
1066         if (job_is_cancelled(&s->common.job) &&
1067             (!s->synced || s->common.job.force_cancel))
1068         {
1069             break;
1070         }
1071         s->last_pause_ns = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
1072     }
1073 
1074 immediate_exit:
1075     if (s->in_flight > 0) {
1076         /* We get here only if something went wrong.  Either the job failed,
1077          * or it was cancelled prematurely so that we do not guarantee that
1078          * the target is a copy of the source.
1079          */
1080         assert(ret < 0 || ((s->common.job.force_cancel || !s->synced) &&
1081                job_is_cancelled(&s->common.job)));
1082         assert(need_drain);
1083         mirror_wait_for_all_io(s);
1084     }
1085 
1086     assert(s->in_flight == 0);
1087     qemu_vfree(s->buf);
1088     g_free(s->cow_bitmap);
1089     g_free(s->in_flight_bitmap);
1090     bdrv_dirty_iter_free(s->dbi);
1091 
1092     if (need_drain) {
1093         s->in_drain = true;
1094         bdrv_drained_begin(bs);
1095     }
1096 
1097     return ret;
1098 }
1099 
1100 static void mirror_complete(Job *job, Error **errp)
1101 {
1102     MirrorBlockJob *s = container_of(job, MirrorBlockJob, common.job);
1103     BlockDriverState *target;
1104 
1105     target = blk_bs(s->target);
1106 
1107     if (!s->synced) {
1108         error_setg(errp, "The active block job '%s' cannot be completed",
1109                    job->id);
1110         return;
1111     }
1112 
1113     if (s->backing_mode == MIRROR_OPEN_BACKING_CHAIN) {
1114         int ret;
1115 
1116         assert(!target->backing);
1117         ret = bdrv_open_backing_file(target, NULL, "backing", errp);
1118         if (ret < 0) {
1119             return;
1120         }
1121     }
1122 
1123     /* block all operations on to_replace bs */
1124     if (s->replaces) {
1125         AioContext *replace_aio_context;
1126 
1127         s->to_replace = bdrv_find_node(s->replaces);
1128         if (!s->to_replace) {
1129             error_setg(errp, "Node name '%s' not found", s->replaces);
1130             return;
1131         }
1132 
1133         replace_aio_context = bdrv_get_aio_context(s->to_replace);
1134         aio_context_acquire(replace_aio_context);
1135 
1136         /* TODO Translate this into permission system. Current definition of
1137          * GRAPH_MOD would require to request it for the parents; they might
1138          * not even be BlockDriverStates, however, so a BdrvChild can't address
1139          * them. May need redefinition of GRAPH_MOD. */
1140         error_setg(&s->replace_blocker,
1141                    "block device is in use by block-job-complete");
1142         bdrv_op_block_all(s->to_replace, s->replace_blocker);
1143         bdrv_ref(s->to_replace);
1144 
1145         aio_context_release(replace_aio_context);
1146     }
1147 
1148     s->should_complete = true;
1149     job_enter(job);
1150 }
1151 
1152 static void coroutine_fn mirror_pause(Job *job)
1153 {
1154     MirrorBlockJob *s = container_of(job, MirrorBlockJob, common.job);
1155 
1156     mirror_wait_for_all_io(s);
1157 }
1158 
1159 static bool mirror_drained_poll(BlockJob *job)
1160 {
1161     MirrorBlockJob *s = container_of(job, MirrorBlockJob, common);
1162 
1163     /* If the job isn't paused nor cancelled, we can't be sure that it won't
1164      * issue more requests. We make an exception if we've reached this point
1165      * from one of our own drain sections, to avoid a deadlock waiting for
1166      * ourselves.
1167      */
1168     if (!s->common.job.paused && !s->common.job.cancelled && !s->in_drain) {
1169         return true;
1170     }
1171 
1172     return !!s->in_flight;
1173 }
1174 
1175 static const BlockJobDriver mirror_job_driver = {
1176     .job_driver = {
1177         .instance_size          = sizeof(MirrorBlockJob),
1178         .job_type               = JOB_TYPE_MIRROR,
1179         .free                   = block_job_free,
1180         .user_resume            = block_job_user_resume,
1181         .run                    = mirror_run,
1182         .prepare                = mirror_prepare,
1183         .abort                  = mirror_abort,
1184         .pause                  = mirror_pause,
1185         .complete               = mirror_complete,
1186     },
1187     .drained_poll           = mirror_drained_poll,
1188 };
1189 
1190 static const BlockJobDriver commit_active_job_driver = {
1191     .job_driver = {
1192         .instance_size          = sizeof(MirrorBlockJob),
1193         .job_type               = JOB_TYPE_COMMIT,
1194         .free                   = block_job_free,
1195         .user_resume            = block_job_user_resume,
1196         .run                    = mirror_run,
1197         .prepare                = mirror_prepare,
1198         .abort                  = mirror_abort,
1199         .pause                  = mirror_pause,
1200         .complete               = mirror_complete,
1201     },
1202     .drained_poll           = mirror_drained_poll,
1203 };
1204 
1205 static void coroutine_fn
1206 do_sync_target_write(MirrorBlockJob *job, MirrorMethod method,
1207                      uint64_t offset, uint64_t bytes,
1208                      QEMUIOVector *qiov, int flags)
1209 {
1210     int ret;
1211     size_t qiov_offset = 0;
1212     int64_t bitmap_offset, bitmap_end;
1213 
1214     if (!QEMU_IS_ALIGNED(offset, job->granularity) &&
1215         bdrv_dirty_bitmap_get(job->dirty_bitmap, offset))
1216     {
1217             /*
1218              * Dirty unaligned padding: ignore it.
1219              *
1220              * Reasoning:
1221              * 1. If we copy it, we can't reset corresponding bit in
1222              *    dirty_bitmap as there may be some "dirty" bytes still not
1223              *    copied.
1224              * 2. It's already dirty, so skipping it we don't diverge mirror
1225              *    progress.
1226              *
1227              * Note, that because of this, guest write may have no contribution
1228              * into mirror converge, but that's not bad, as we have background
1229              * process of mirroring. If under some bad circumstances (high guest
1230              * IO load) background process starve, we will not converge anyway,
1231              * even if each write will contribute, as guest is not guaranteed to
1232              * rewrite the whole disk.
1233              */
1234             qiov_offset = QEMU_ALIGN_UP(offset, job->granularity) - offset;
1235             if (bytes <= qiov_offset) {
1236                 /* nothing to do after shrink */
1237                 return;
1238             }
1239             offset += qiov_offset;
1240             bytes -= qiov_offset;
1241     }
1242 
1243     if (!QEMU_IS_ALIGNED(offset + bytes, job->granularity) &&
1244         bdrv_dirty_bitmap_get(job->dirty_bitmap, offset + bytes - 1))
1245     {
1246         uint64_t tail = (offset + bytes) % job->granularity;
1247 
1248         if (bytes <= tail) {
1249             /* nothing to do after shrink */
1250             return;
1251         }
1252         bytes -= tail;
1253     }
1254 
1255     /*
1256      * Tails are either clean or shrunk, so for bitmap resetting
1257      * we safely align the range down.
1258      */
1259     bitmap_offset = QEMU_ALIGN_UP(offset, job->granularity);
1260     bitmap_end = QEMU_ALIGN_DOWN(offset + bytes, job->granularity);
1261     if (bitmap_offset < bitmap_end) {
1262         bdrv_reset_dirty_bitmap(job->dirty_bitmap, bitmap_offset,
1263                                 bitmap_end - bitmap_offset);
1264     }
1265 
1266     job_progress_increase_remaining(&job->common.job, bytes);
1267 
1268     switch (method) {
1269     case MIRROR_METHOD_COPY:
1270         ret = blk_co_pwritev_part(job->target, offset, bytes,
1271                                   qiov, qiov_offset, flags);
1272         break;
1273 
1274     case MIRROR_METHOD_ZERO:
1275         assert(!qiov);
1276         ret = blk_co_pwrite_zeroes(job->target, offset, bytes, flags);
1277         break;
1278 
1279     case MIRROR_METHOD_DISCARD:
1280         assert(!qiov);
1281         ret = blk_co_pdiscard(job->target, offset, bytes);
1282         break;
1283 
1284     default:
1285         abort();
1286     }
1287 
1288     if (ret >= 0) {
1289         job_progress_update(&job->common.job, bytes);
1290     } else {
1291         BlockErrorAction action;
1292 
1293         /*
1294          * We failed, so we should mark dirty the whole area, aligned up.
1295          * Note that we don't care about shrunk tails if any: they were dirty
1296          * at function start, and they must be still dirty, as we've locked
1297          * the region for in-flight op.
1298          */
1299         bitmap_offset = QEMU_ALIGN_DOWN(offset, job->granularity);
1300         bitmap_end = QEMU_ALIGN_UP(offset + bytes, job->granularity);
1301         bdrv_set_dirty_bitmap(job->dirty_bitmap, bitmap_offset,
1302                               bitmap_end - bitmap_offset);
1303         job->actively_synced = false;
1304 
1305         action = mirror_error_action(job, false, -ret);
1306         if (action == BLOCK_ERROR_ACTION_REPORT) {
1307             if (!job->ret) {
1308                 job->ret = ret;
1309             }
1310         }
1311     }
1312 }
1313 
1314 static MirrorOp *coroutine_fn active_write_prepare(MirrorBlockJob *s,
1315                                                    uint64_t offset,
1316                                                    uint64_t bytes)
1317 {
1318     MirrorOp *op;
1319     uint64_t start_chunk = offset / s->granularity;
1320     uint64_t end_chunk = DIV_ROUND_UP(offset + bytes, s->granularity);
1321 
1322     op = g_new(MirrorOp, 1);
1323     *op = (MirrorOp){
1324         .s                  = s,
1325         .offset             = offset,
1326         .bytes              = bytes,
1327         .is_active_write    = true,
1328         .is_in_flight       = true,
1329     };
1330     qemu_co_queue_init(&op->waiting_requests);
1331     QTAILQ_INSERT_TAIL(&s->ops_in_flight, op, next);
1332 
1333     s->in_active_write_counter++;
1334 
1335     mirror_wait_on_conflicts(op, s, offset, bytes);
1336 
1337     bitmap_set(s->in_flight_bitmap, start_chunk, end_chunk - start_chunk);
1338 
1339     return op;
1340 }
1341 
1342 static void coroutine_fn active_write_settle(MirrorOp *op)
1343 {
1344     uint64_t start_chunk = op->offset / op->s->granularity;
1345     uint64_t end_chunk = DIV_ROUND_UP(op->offset + op->bytes,
1346                                       op->s->granularity);
1347 
1348     if (!--op->s->in_active_write_counter && op->s->actively_synced) {
1349         BdrvChild *source = op->s->mirror_top_bs->backing;
1350 
1351         if (QLIST_FIRST(&source->bs->parents) == source &&
1352             QLIST_NEXT(source, next_parent) == NULL)
1353         {
1354             /* Assert that we are back in sync once all active write
1355              * operations are settled.
1356              * Note that we can only assert this if the mirror node
1357              * is the source node's only parent. */
1358             assert(!bdrv_get_dirty_count(op->s->dirty_bitmap));
1359         }
1360     }
1361     bitmap_clear(op->s->in_flight_bitmap, start_chunk, end_chunk - start_chunk);
1362     QTAILQ_REMOVE(&op->s->ops_in_flight, op, next);
1363     qemu_co_queue_restart_all(&op->waiting_requests);
1364     g_free(op);
1365 }
1366 
1367 static int coroutine_fn bdrv_mirror_top_preadv(BlockDriverState *bs,
1368     uint64_t offset, uint64_t bytes, QEMUIOVector *qiov, int flags)
1369 {
1370     return bdrv_co_preadv(bs->backing, offset, bytes, qiov, flags);
1371 }
1372 
1373 static int coroutine_fn bdrv_mirror_top_do_write(BlockDriverState *bs,
1374     MirrorMethod method, uint64_t offset, uint64_t bytes, QEMUIOVector *qiov,
1375     int flags)
1376 {
1377     MirrorOp *op = NULL;
1378     MirrorBDSOpaque *s = bs->opaque;
1379     int ret = 0;
1380     bool copy_to_target;
1381 
1382     copy_to_target = s->job->ret >= 0 &&
1383                      s->job->copy_mode == MIRROR_COPY_MODE_WRITE_BLOCKING;
1384 
1385     if (copy_to_target) {
1386         op = active_write_prepare(s->job, offset, bytes);
1387     }
1388 
1389     switch (method) {
1390     case MIRROR_METHOD_COPY:
1391         ret = bdrv_co_pwritev(bs->backing, offset, bytes, qiov, flags);
1392         break;
1393 
1394     case MIRROR_METHOD_ZERO:
1395         ret = bdrv_co_pwrite_zeroes(bs->backing, offset, bytes, flags);
1396         break;
1397 
1398     case MIRROR_METHOD_DISCARD:
1399         ret = bdrv_co_pdiscard(bs->backing, offset, bytes);
1400         break;
1401 
1402     default:
1403         abort();
1404     }
1405 
1406     if (ret < 0) {
1407         goto out;
1408     }
1409 
1410     if (copy_to_target) {
1411         do_sync_target_write(s->job, method, offset, bytes, qiov, flags);
1412     }
1413 
1414 out:
1415     if (copy_to_target) {
1416         active_write_settle(op);
1417     }
1418     return ret;
1419 }
1420 
1421 static int coroutine_fn bdrv_mirror_top_pwritev(BlockDriverState *bs,
1422     uint64_t offset, uint64_t bytes, QEMUIOVector *qiov, int flags)
1423 {
1424     MirrorBDSOpaque *s = bs->opaque;
1425     QEMUIOVector bounce_qiov;
1426     void *bounce_buf;
1427     int ret = 0;
1428     bool copy_to_target;
1429 
1430     copy_to_target = s->job->ret >= 0 &&
1431                      s->job->copy_mode == MIRROR_COPY_MODE_WRITE_BLOCKING;
1432 
1433     if (copy_to_target) {
1434         /* The guest might concurrently modify the data to write; but
1435          * the data on source and destination must match, so we have
1436          * to use a bounce buffer if we are going to write to the
1437          * target now. */
1438         bounce_buf = qemu_blockalign(bs, bytes);
1439         iov_to_buf_full(qiov->iov, qiov->niov, 0, bounce_buf, bytes);
1440 
1441         qemu_iovec_init(&bounce_qiov, 1);
1442         qemu_iovec_add(&bounce_qiov, bounce_buf, bytes);
1443         qiov = &bounce_qiov;
1444     }
1445 
1446     ret = bdrv_mirror_top_do_write(bs, MIRROR_METHOD_COPY, offset, bytes, qiov,
1447                                    flags);
1448 
1449     if (copy_to_target) {
1450         qemu_iovec_destroy(&bounce_qiov);
1451         qemu_vfree(bounce_buf);
1452     }
1453 
1454     return ret;
1455 }
1456 
1457 static int coroutine_fn bdrv_mirror_top_flush(BlockDriverState *bs)
1458 {
1459     if (bs->backing == NULL) {
1460         /* we can be here after failed bdrv_append in mirror_start_job */
1461         return 0;
1462     }
1463     return bdrv_co_flush(bs->backing->bs);
1464 }
1465 
1466 static int coroutine_fn bdrv_mirror_top_pwrite_zeroes(BlockDriverState *bs,
1467     int64_t offset, int bytes, BdrvRequestFlags flags)
1468 {
1469     return bdrv_mirror_top_do_write(bs, MIRROR_METHOD_ZERO, offset, bytes, NULL,
1470                                     flags);
1471 }
1472 
1473 static int coroutine_fn bdrv_mirror_top_pdiscard(BlockDriverState *bs,
1474     int64_t offset, int bytes)
1475 {
1476     return bdrv_mirror_top_do_write(bs, MIRROR_METHOD_DISCARD, offset, bytes,
1477                                     NULL, 0);
1478 }
1479 
1480 static void bdrv_mirror_top_refresh_filename(BlockDriverState *bs)
1481 {
1482     if (bs->backing == NULL) {
1483         /* we can be here after failed bdrv_attach_child in
1484          * bdrv_set_backing_hd */
1485         return;
1486     }
1487     pstrcpy(bs->exact_filename, sizeof(bs->exact_filename),
1488             bs->backing->bs->filename);
1489 }
1490 
1491 static void bdrv_mirror_top_child_perm(BlockDriverState *bs, BdrvChild *c,
1492                                        const BdrvChildRole *role,
1493                                        BlockReopenQueue *reopen_queue,
1494                                        uint64_t perm, uint64_t shared,
1495                                        uint64_t *nperm, uint64_t *nshared)
1496 {
1497     MirrorBDSOpaque *s = bs->opaque;
1498 
1499     if (s->stop) {
1500         /*
1501          * If the job is to be stopped, we do not need to forward
1502          * anything to the real image.
1503          */
1504         *nperm = 0;
1505         *nshared = BLK_PERM_ALL;
1506         return;
1507     }
1508 
1509     /* Must be able to forward guest writes to the real image */
1510     *nperm = 0;
1511     if (perm & BLK_PERM_WRITE) {
1512         *nperm |= BLK_PERM_WRITE;
1513     }
1514 
1515     *nshared = BLK_PERM_ALL;
1516 }
1517 
1518 /* Dummy node that provides consistent read to its users without requiring it
1519  * from its backing file and that allows writes on the backing file chain. */
1520 static BlockDriver bdrv_mirror_top = {
1521     .format_name                = "mirror_top",
1522     .bdrv_co_preadv             = bdrv_mirror_top_preadv,
1523     .bdrv_co_pwritev            = bdrv_mirror_top_pwritev,
1524     .bdrv_co_pwrite_zeroes      = bdrv_mirror_top_pwrite_zeroes,
1525     .bdrv_co_pdiscard           = bdrv_mirror_top_pdiscard,
1526     .bdrv_co_flush              = bdrv_mirror_top_flush,
1527     .bdrv_co_block_status       = bdrv_co_block_status_from_backing,
1528     .bdrv_refresh_filename      = bdrv_mirror_top_refresh_filename,
1529     .bdrv_child_perm            = bdrv_mirror_top_child_perm,
1530 };
1531 
1532 static BlockJob *mirror_start_job(
1533                              const char *job_id, BlockDriverState *bs,
1534                              int creation_flags, BlockDriverState *target,
1535                              const char *replaces, int64_t speed,
1536                              uint32_t granularity, int64_t buf_size,
1537                              BlockMirrorBackingMode backing_mode,
1538                              bool zero_target,
1539                              BlockdevOnError on_source_error,
1540                              BlockdevOnError on_target_error,
1541                              bool unmap,
1542                              BlockCompletionFunc *cb,
1543                              void *opaque,
1544                              const BlockJobDriver *driver,
1545                              bool is_none_mode, BlockDriverState *base,
1546                              bool auto_complete, const char *filter_node_name,
1547                              bool is_mirror, MirrorCopyMode copy_mode,
1548                              Error **errp)
1549 {
1550     MirrorBlockJob *s;
1551     MirrorBDSOpaque *bs_opaque;
1552     BlockDriverState *mirror_top_bs;
1553     bool target_graph_mod;
1554     bool target_is_backing;
1555     Error *local_err = NULL;
1556     int ret;
1557 
1558     if (granularity == 0) {
1559         granularity = bdrv_get_default_bitmap_granularity(target);
1560     }
1561 
1562     assert(is_power_of_2(granularity));
1563 
1564     if (buf_size < 0) {
1565         error_setg(errp, "Invalid parameter 'buf-size'");
1566         return NULL;
1567     }
1568 
1569     if (buf_size == 0) {
1570         buf_size = DEFAULT_MIRROR_BUF_SIZE;
1571     }
1572 
1573     if (bs == target) {
1574         error_setg(errp, "Can't mirror node into itself");
1575         return NULL;
1576     }
1577 
1578     /* In the case of active commit, add dummy driver to provide consistent
1579      * reads on the top, while disabling it in the intermediate nodes, and make
1580      * the backing chain writable. */
1581     mirror_top_bs = bdrv_new_open_driver(&bdrv_mirror_top, filter_node_name,
1582                                          BDRV_O_RDWR, errp);
1583     if (mirror_top_bs == NULL) {
1584         return NULL;
1585     }
1586     if (!filter_node_name) {
1587         mirror_top_bs->implicit = true;
1588     }
1589 
1590     /* So that we can always drop this node */
1591     mirror_top_bs->never_freeze = true;
1592 
1593     mirror_top_bs->total_sectors = bs->total_sectors;
1594     mirror_top_bs->supported_write_flags = BDRV_REQ_WRITE_UNCHANGED;
1595     mirror_top_bs->supported_zero_flags = BDRV_REQ_WRITE_UNCHANGED |
1596                                           BDRV_REQ_NO_FALLBACK;
1597     bs_opaque = g_new0(MirrorBDSOpaque, 1);
1598     mirror_top_bs->opaque = bs_opaque;
1599 
1600     /* bdrv_append takes ownership of the mirror_top_bs reference, need to keep
1601      * it alive until block_job_create() succeeds even if bs has no parent. */
1602     bdrv_ref(mirror_top_bs);
1603     bdrv_drained_begin(bs);
1604     bdrv_append(mirror_top_bs, bs, &local_err);
1605     bdrv_drained_end(bs);
1606 
1607     if (local_err) {
1608         bdrv_unref(mirror_top_bs);
1609         error_propagate(errp, local_err);
1610         return NULL;
1611     }
1612 
1613     /* Make sure that the source is not resized while the job is running */
1614     s = block_job_create(job_id, driver, NULL, mirror_top_bs,
1615                          BLK_PERM_CONSISTENT_READ,
1616                          BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE_UNCHANGED |
1617                          BLK_PERM_WRITE | BLK_PERM_GRAPH_MOD, speed,
1618                          creation_flags, cb, opaque, errp);
1619     if (!s) {
1620         goto fail;
1621     }
1622     bs_opaque->job = s;
1623 
1624     /* The block job now has a reference to this node */
1625     bdrv_unref(mirror_top_bs);
1626 
1627     s->mirror_top_bs = mirror_top_bs;
1628 
1629     /* No resize for the target either; while the mirror is still running, a
1630      * consistent read isn't necessarily possible. We could possibly allow
1631      * writes and graph modifications, though it would likely defeat the
1632      * purpose of a mirror, so leave them blocked for now.
1633      *
1634      * In the case of active commit, things look a bit different, though,
1635      * because the target is an already populated backing file in active use.
1636      * We can allow anything except resize there.*/
1637     target_is_backing = bdrv_chain_contains(bs, target);
1638     target_graph_mod = (backing_mode != MIRROR_LEAVE_BACKING_CHAIN);
1639     s->target = blk_new(s->common.job.aio_context,
1640                         BLK_PERM_WRITE | BLK_PERM_RESIZE |
1641                         (target_graph_mod ? BLK_PERM_GRAPH_MOD : 0),
1642                         BLK_PERM_WRITE_UNCHANGED |
1643                         (target_is_backing ? BLK_PERM_CONSISTENT_READ |
1644                                              BLK_PERM_WRITE |
1645                                              BLK_PERM_GRAPH_MOD : 0));
1646     ret = blk_insert_bs(s->target, target, errp);
1647     if (ret < 0) {
1648         goto fail;
1649     }
1650     if (is_mirror) {
1651         /* XXX: Mirror target could be a NBD server of target QEMU in the case
1652          * of non-shared block migration. To allow migration completion, we
1653          * have to allow "inactivate" of the target BB.  When that happens, we
1654          * know the job is drained, and the vcpus are stopped, so no write
1655          * operation will be performed. Block layer already has assertions to
1656          * ensure that. */
1657         blk_set_force_allow_inactivate(s->target);
1658     }
1659     blk_set_allow_aio_context_change(s->target, true);
1660     blk_set_disable_request_queuing(s->target, true);
1661 
1662     s->replaces = g_strdup(replaces);
1663     s->on_source_error = on_source_error;
1664     s->on_target_error = on_target_error;
1665     s->is_none_mode = is_none_mode;
1666     s->backing_mode = backing_mode;
1667     s->zero_target = zero_target;
1668     s->copy_mode = copy_mode;
1669     s->base = base;
1670     s->granularity = granularity;
1671     s->buf_size = ROUND_UP(buf_size, granularity);
1672     s->unmap = unmap;
1673     if (auto_complete) {
1674         s->should_complete = true;
1675     }
1676 
1677     s->dirty_bitmap = bdrv_create_dirty_bitmap(bs, granularity, NULL, errp);
1678     if (!s->dirty_bitmap) {
1679         goto fail;
1680     }
1681     if (s->copy_mode == MIRROR_COPY_MODE_WRITE_BLOCKING) {
1682         bdrv_disable_dirty_bitmap(s->dirty_bitmap);
1683     }
1684 
1685     ret = block_job_add_bdrv(&s->common, "source", bs, 0,
1686                              BLK_PERM_WRITE_UNCHANGED | BLK_PERM_WRITE |
1687                              BLK_PERM_CONSISTENT_READ,
1688                              errp);
1689     if (ret < 0) {
1690         goto fail;
1691     }
1692 
1693     /* Required permissions are already taken with blk_new() */
1694     block_job_add_bdrv(&s->common, "target", target, 0, BLK_PERM_ALL,
1695                        &error_abort);
1696 
1697     /* In commit_active_start() all intermediate nodes disappear, so
1698      * any jobs in them must be blocked */
1699     if (target_is_backing) {
1700         BlockDriverState *iter;
1701         for (iter = backing_bs(bs); iter != target; iter = backing_bs(iter)) {
1702             /* XXX BLK_PERM_WRITE needs to be allowed so we don't block
1703              * ourselves at s->base (if writes are blocked for a node, they are
1704              * also blocked for its backing file). The other options would be a
1705              * second filter driver above s->base (== target). */
1706             ret = block_job_add_bdrv(&s->common, "intermediate node", iter, 0,
1707                                      BLK_PERM_WRITE_UNCHANGED | BLK_PERM_WRITE,
1708                                      errp);
1709             if (ret < 0) {
1710                 goto fail;
1711             }
1712         }
1713 
1714         if (bdrv_freeze_backing_chain(mirror_top_bs, target, errp) < 0) {
1715             goto fail;
1716         }
1717     }
1718 
1719     QTAILQ_INIT(&s->ops_in_flight);
1720 
1721     trace_mirror_start(bs, s, opaque);
1722     job_start(&s->common.job);
1723 
1724     return &s->common;
1725 
1726 fail:
1727     if (s) {
1728         /* Make sure this BDS does not go away until we have completed the graph
1729          * changes below */
1730         bdrv_ref(mirror_top_bs);
1731 
1732         g_free(s->replaces);
1733         blk_unref(s->target);
1734         bs_opaque->job = NULL;
1735         if (s->dirty_bitmap) {
1736             bdrv_release_dirty_bitmap(s->dirty_bitmap);
1737         }
1738         job_early_fail(&s->common.job);
1739     }
1740 
1741     bs_opaque->stop = true;
1742     bdrv_child_refresh_perms(mirror_top_bs, mirror_top_bs->backing,
1743                              &error_abort);
1744     bdrv_replace_node(mirror_top_bs, backing_bs(mirror_top_bs), &error_abort);
1745 
1746     bdrv_unref(mirror_top_bs);
1747 
1748     return NULL;
1749 }
1750 
1751 void mirror_start(const char *job_id, BlockDriverState *bs,
1752                   BlockDriverState *target, const char *replaces,
1753                   int creation_flags, int64_t speed,
1754                   uint32_t granularity, int64_t buf_size,
1755                   MirrorSyncMode mode, BlockMirrorBackingMode backing_mode,
1756                   bool zero_target,
1757                   BlockdevOnError on_source_error,
1758                   BlockdevOnError on_target_error,
1759                   bool unmap, const char *filter_node_name,
1760                   MirrorCopyMode copy_mode, Error **errp)
1761 {
1762     bool is_none_mode;
1763     BlockDriverState *base;
1764 
1765     if ((mode == MIRROR_SYNC_MODE_INCREMENTAL) ||
1766         (mode == MIRROR_SYNC_MODE_BITMAP)) {
1767         error_setg(errp, "Sync mode '%s' not supported",
1768                    MirrorSyncMode_str(mode));
1769         return;
1770     }
1771     is_none_mode = mode == MIRROR_SYNC_MODE_NONE;
1772     base = mode == MIRROR_SYNC_MODE_TOP ? backing_bs(bs) : NULL;
1773     mirror_start_job(job_id, bs, creation_flags, target, replaces,
1774                      speed, granularity, buf_size, backing_mode, zero_target,
1775                      on_source_error, on_target_error, unmap, NULL, NULL,
1776                      &mirror_job_driver, is_none_mode, base, false,
1777                      filter_node_name, true, copy_mode, errp);
1778 }
1779 
1780 BlockJob *commit_active_start(const char *job_id, BlockDriverState *bs,
1781                               BlockDriverState *base, int creation_flags,
1782                               int64_t speed, BlockdevOnError on_error,
1783                               const char *filter_node_name,
1784                               BlockCompletionFunc *cb, void *opaque,
1785                               bool auto_complete, Error **errp)
1786 {
1787     bool base_read_only;
1788     Error *local_err = NULL;
1789     BlockJob *ret;
1790 
1791     base_read_only = bdrv_is_read_only(base);
1792 
1793     if (base_read_only) {
1794         if (bdrv_reopen_set_read_only(base, false, errp) < 0) {
1795             return NULL;
1796         }
1797     }
1798 
1799     ret = mirror_start_job(
1800                      job_id, bs, creation_flags, base, NULL, speed, 0, 0,
1801                      MIRROR_LEAVE_BACKING_CHAIN, false,
1802                      on_error, on_error, true, cb, opaque,
1803                      &commit_active_job_driver, false, base, auto_complete,
1804                      filter_node_name, false, MIRROR_COPY_MODE_BACKGROUND,
1805                      &local_err);
1806     if (local_err) {
1807         error_propagate(errp, local_err);
1808         goto error_restore_flags;
1809     }
1810 
1811     return ret;
1812 
1813 error_restore_flags:
1814     /* ignore error and errp for bdrv_reopen, because we want to propagate
1815      * the original error */
1816     if (base_read_only) {
1817         bdrv_reopen_set_read_only(base, true, NULL);
1818     }
1819     return NULL;
1820 }
1821