xref: /openbmc/qemu/block/mirror.c (revision 2f95279a)
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 "block/dirty-bitmap.h"
22 #include "sysemu/block-backend.h"
23 #include "qapi/error.h"
24 #include "qemu/ratelimit.h"
25 #include "qemu/bitmap.h"
26 #include "qemu/memalign.h"
27 
28 #define MAX_IN_FLIGHT 16
29 #define MAX_IO_BYTES (1 << 20) /* 1 Mb */
30 #define DEFAULT_MIRROR_BUF_SIZE (MAX_IN_FLIGHT * MAX_IO_BYTES)
31 
32 /* The mirroring buffer is a list of granularity-sized chunks.
33  * Free chunks are organized in a list.
34  */
35 typedef struct MirrorBuffer {
36     QSIMPLEQ_ENTRY(MirrorBuffer) next;
37 } MirrorBuffer;
38 
39 typedef struct MirrorOp MirrorOp;
40 
41 typedef struct MirrorBlockJob {
42     BlockJob common;
43     BlockBackend *target;
44     BlockDriverState *mirror_top_bs;
45     BlockDriverState *base;
46     BlockDriverState *base_overlay;
47 
48     /* The name of the graph node to replace */
49     char *replaces;
50     /* The BDS to replace */
51     BlockDriverState *to_replace;
52     /* Used to block operations on the drive-mirror-replace target */
53     Error *replace_blocker;
54     bool is_none_mode;
55     BlockMirrorBackingMode backing_mode;
56     /* Whether the target image requires explicit zero-initialization */
57     bool zero_target;
58     /*
59      * To be accesssed with atomics. Written only under the BQL (required by the
60      * current implementation of mirror_change()).
61      */
62     MirrorCopyMode copy_mode;
63     BlockdevOnError on_source_error, on_target_error;
64     /*
65      * To be accessed with atomics.
66      *
67      * Set when the target is synced (dirty bitmap is clean, nothing in flight)
68      * and the job is running in active mode.
69      */
70     bool actively_synced;
71     bool should_complete;
72     int64_t granularity;
73     size_t buf_size;
74     int64_t bdev_length;
75     unsigned long *cow_bitmap;
76     BdrvDirtyBitmap *dirty_bitmap;
77     BdrvDirtyBitmapIter *dbi;
78     uint8_t *buf;
79     QSIMPLEQ_HEAD(, MirrorBuffer) buf_free;
80     int buf_free_count;
81 
82     uint64_t last_pause_ns;
83     unsigned long *in_flight_bitmap;
84     unsigned in_flight;
85     int64_t bytes_in_flight;
86     QTAILQ_HEAD(, MirrorOp) ops_in_flight;
87     int ret;
88     bool unmap;
89     int target_cluster_size;
90     int max_iov;
91     bool initial_zeroing_ongoing;
92     int in_active_write_counter;
93     int64_t active_write_bytes_in_flight;
94     bool prepared;
95     bool in_drain;
96     bool base_ro;
97 } MirrorBlockJob;
98 
99 typedef struct MirrorBDSOpaque {
100     MirrorBlockJob *job;
101     bool stop;
102     bool is_commit;
103 } MirrorBDSOpaque;
104 
105 struct MirrorOp {
106     MirrorBlockJob *s;
107     QEMUIOVector qiov;
108     int64_t offset;
109     uint64_t bytes;
110 
111     /* The pointee is set by mirror_co_read(), mirror_co_zero(), and
112      * mirror_co_discard() before yielding for the first time */
113     int64_t *bytes_handled;
114 
115     bool is_pseudo_op;
116     bool is_active_write;
117     bool is_in_flight;
118     CoQueue waiting_requests;
119     Coroutine *co;
120     MirrorOp *waiting_for_op;
121 
122     QTAILQ_ENTRY(MirrorOp) next;
123 };
124 
125 typedef enum MirrorMethod {
126     MIRROR_METHOD_COPY,
127     MIRROR_METHOD_ZERO,
128     MIRROR_METHOD_DISCARD,
129 } MirrorMethod;
130 
131 static BlockErrorAction mirror_error_action(MirrorBlockJob *s, bool read,
132                                             int error)
133 {
134     qatomic_set(&s->actively_synced, false);
135     if (read) {
136         return block_job_error_action(&s->common, s->on_source_error,
137                                       true, error);
138     } else {
139         return block_job_error_action(&s->common, s->on_target_error,
140                                       false, error);
141     }
142 }
143 
144 static void coroutine_fn mirror_wait_on_conflicts(MirrorOp *self,
145                                                   MirrorBlockJob *s,
146                                                   uint64_t offset,
147                                                   uint64_t bytes)
148 {
149     uint64_t self_start_chunk = offset / s->granularity;
150     uint64_t self_end_chunk = DIV_ROUND_UP(offset + bytes, s->granularity);
151     uint64_t self_nb_chunks = self_end_chunk - self_start_chunk;
152 
153     while (find_next_bit(s->in_flight_bitmap, self_end_chunk,
154                          self_start_chunk) < self_end_chunk &&
155            s->ret >= 0)
156     {
157         MirrorOp *op;
158 
159         QTAILQ_FOREACH(op, &s->ops_in_flight, next) {
160             uint64_t op_start_chunk = op->offset / s->granularity;
161             uint64_t op_nb_chunks = DIV_ROUND_UP(op->offset + op->bytes,
162                                                  s->granularity) -
163                                     op_start_chunk;
164 
165             if (op == self) {
166                 continue;
167             }
168 
169             if (ranges_overlap(self_start_chunk, self_nb_chunks,
170                                op_start_chunk, op_nb_chunks))
171             {
172                 if (self) {
173                     /*
174                      * If the operation is already (indirectly) waiting for us,
175                      * or will wait for us as soon as it wakes up, then just go
176                      * on (instead of producing a deadlock in the former case).
177                      */
178                     if (op->waiting_for_op) {
179                         continue;
180                     }
181 
182                     self->waiting_for_op = op;
183                 }
184 
185                 qemu_co_queue_wait(&op->waiting_requests, NULL);
186 
187                 if (self) {
188                     self->waiting_for_op = NULL;
189                 }
190 
191                 break;
192             }
193         }
194     }
195 }
196 
197 static void coroutine_fn mirror_iteration_done(MirrorOp *op, int ret)
198 {
199     MirrorBlockJob *s = op->s;
200     struct iovec *iov;
201     int64_t chunk_num;
202     int i, nb_chunks;
203 
204     trace_mirror_iteration_done(s, op->offset, op->bytes, ret);
205 
206     s->in_flight--;
207     s->bytes_in_flight -= op->bytes;
208     iov = op->qiov.iov;
209     for (i = 0; i < op->qiov.niov; i++) {
210         MirrorBuffer *buf = (MirrorBuffer *) iov[i].iov_base;
211         QSIMPLEQ_INSERT_TAIL(&s->buf_free, buf, next);
212         s->buf_free_count++;
213     }
214 
215     chunk_num = op->offset / s->granularity;
216     nb_chunks = DIV_ROUND_UP(op->bytes, s->granularity);
217 
218     bitmap_clear(s->in_flight_bitmap, chunk_num, nb_chunks);
219     QTAILQ_REMOVE(&s->ops_in_flight, op, next);
220     if (ret >= 0) {
221         if (s->cow_bitmap) {
222             bitmap_set(s->cow_bitmap, chunk_num, nb_chunks);
223         }
224         if (!s->initial_zeroing_ongoing) {
225             job_progress_update(&s->common.job, op->bytes);
226         }
227     }
228     qemu_iovec_destroy(&op->qiov);
229 
230     qemu_co_queue_restart_all(&op->waiting_requests);
231     g_free(op);
232 }
233 
234 static void coroutine_fn mirror_write_complete(MirrorOp *op, int ret)
235 {
236     MirrorBlockJob *s = op->s;
237 
238     if (ret < 0) {
239         BlockErrorAction action;
240 
241         bdrv_set_dirty_bitmap(s->dirty_bitmap, op->offset, op->bytes);
242         action = mirror_error_action(s, false, -ret);
243         if (action == BLOCK_ERROR_ACTION_REPORT && s->ret >= 0) {
244             s->ret = ret;
245         }
246     }
247 
248     mirror_iteration_done(op, ret);
249 }
250 
251 static void coroutine_fn mirror_read_complete(MirrorOp *op, int ret)
252 {
253     MirrorBlockJob *s = op->s;
254 
255     if (ret < 0) {
256         BlockErrorAction action;
257 
258         bdrv_set_dirty_bitmap(s->dirty_bitmap, op->offset, op->bytes);
259         action = mirror_error_action(s, true, -ret);
260         if (action == BLOCK_ERROR_ACTION_REPORT && s->ret >= 0) {
261             s->ret = ret;
262         }
263 
264         mirror_iteration_done(op, ret);
265         return;
266     }
267 
268     ret = blk_co_pwritev(s->target, op->offset, op->qiov.size, &op->qiov, 0);
269     mirror_write_complete(op, ret);
270 }
271 
272 /* Clip bytes relative to offset to not exceed end-of-file */
273 static inline int64_t mirror_clip_bytes(MirrorBlockJob *s,
274                                         int64_t offset,
275                                         int64_t bytes)
276 {
277     return MIN(bytes, s->bdev_length - offset);
278 }
279 
280 /* Round offset and/or bytes to target cluster if COW is needed, and
281  * return the offset of the adjusted tail against original. */
282 static int coroutine_fn mirror_cow_align(MirrorBlockJob *s, int64_t *offset,
283                                          uint64_t *bytes)
284 {
285     bool need_cow;
286     int ret = 0;
287     int64_t align_offset = *offset;
288     int64_t align_bytes = *bytes;
289     int max_bytes = s->granularity * s->max_iov;
290 
291     need_cow = !test_bit(*offset / s->granularity, s->cow_bitmap);
292     need_cow |= !test_bit((*offset + *bytes - 1) / s->granularity,
293                           s->cow_bitmap);
294     if (need_cow) {
295         bdrv_round_to_subclusters(blk_bs(s->target), *offset, *bytes,
296                                   &align_offset, &align_bytes);
297     }
298 
299     if (align_bytes > max_bytes) {
300         align_bytes = max_bytes;
301         if (need_cow) {
302             align_bytes = QEMU_ALIGN_DOWN(align_bytes, s->target_cluster_size);
303         }
304     }
305     /* Clipping may result in align_bytes unaligned to chunk boundary, but
306      * that doesn't matter because it's already the end of source image. */
307     align_bytes = mirror_clip_bytes(s, align_offset, align_bytes);
308 
309     ret = align_offset + align_bytes - (*offset + *bytes);
310     *offset = align_offset;
311     *bytes = align_bytes;
312     assert(ret >= 0);
313     return ret;
314 }
315 
316 static inline void coroutine_fn
317 mirror_wait_for_free_in_flight_slot(MirrorBlockJob *s)
318 {
319     MirrorOp *op;
320 
321     QTAILQ_FOREACH(op, &s->ops_in_flight, next) {
322         /*
323          * Do not wait on pseudo ops, because it may in turn wait on
324          * some other operation to start, which may in fact be the
325          * caller of this function.  Since there is only one pseudo op
326          * at any given time, we will always find some real operation
327          * to wait on.
328          * Also, do not wait on active operations, because they do not
329          * use up in-flight slots.
330          */
331         if (!op->is_pseudo_op && op->is_in_flight && !op->is_active_write) {
332             qemu_co_queue_wait(&op->waiting_requests, NULL);
333             return;
334         }
335     }
336     abort();
337 }
338 
339 /* Perform a mirror copy operation.
340  *
341  * *op->bytes_handled is set to the number of bytes copied after and
342  * including offset, excluding any bytes copied prior to offset due
343  * to alignment.  This will be op->bytes if no alignment is necessary,
344  * or (new_end - op->offset) if the tail is rounded up or down due to
345  * alignment or buffer limit.
346  */
347 static void coroutine_fn mirror_co_read(void *opaque)
348 {
349     MirrorOp *op = opaque;
350     MirrorBlockJob *s = op->s;
351     int nb_chunks;
352     uint64_t ret;
353     uint64_t max_bytes;
354 
355     max_bytes = s->granularity * s->max_iov;
356 
357     /* We can only handle as much as buf_size at a time. */
358     op->bytes = MIN(s->buf_size, MIN(max_bytes, op->bytes));
359     assert(op->bytes);
360     assert(op->bytes < BDRV_REQUEST_MAX_BYTES);
361     *op->bytes_handled = op->bytes;
362 
363     if (s->cow_bitmap) {
364         *op->bytes_handled += mirror_cow_align(s, &op->offset, &op->bytes);
365     }
366     /* Cannot exceed BDRV_REQUEST_MAX_BYTES + INT_MAX */
367     assert(*op->bytes_handled <= UINT_MAX);
368     assert(op->bytes <= s->buf_size);
369     /* The offset is granularity-aligned because:
370      * 1) Caller passes in aligned values;
371      * 2) mirror_cow_align is used only when target cluster is larger. */
372     assert(QEMU_IS_ALIGNED(op->offset, s->granularity));
373     /* The range is sector-aligned, since bdrv_getlength() rounds up. */
374     assert(QEMU_IS_ALIGNED(op->bytes, BDRV_SECTOR_SIZE));
375     nb_chunks = DIV_ROUND_UP(op->bytes, s->granularity);
376 
377     while (s->buf_free_count < nb_chunks) {
378         trace_mirror_yield_in_flight(s, op->offset, s->in_flight);
379         mirror_wait_for_free_in_flight_slot(s);
380     }
381 
382     /* Now make a QEMUIOVector taking enough granularity-sized chunks
383      * from s->buf_free.
384      */
385     qemu_iovec_init(&op->qiov, nb_chunks);
386     while (nb_chunks-- > 0) {
387         MirrorBuffer *buf = QSIMPLEQ_FIRST(&s->buf_free);
388         size_t remaining = op->bytes - op->qiov.size;
389 
390         QSIMPLEQ_REMOVE_HEAD(&s->buf_free, next);
391         s->buf_free_count--;
392         qemu_iovec_add(&op->qiov, buf, MIN(s->granularity, remaining));
393     }
394 
395     /* Copy the dirty cluster.  */
396     s->in_flight++;
397     s->bytes_in_flight += op->bytes;
398     op->is_in_flight = true;
399     trace_mirror_one_iteration(s, op->offset, op->bytes);
400 
401     WITH_GRAPH_RDLOCK_GUARD() {
402         ret = bdrv_co_preadv(s->mirror_top_bs->backing, op->offset, op->bytes,
403                              &op->qiov, 0);
404     }
405     mirror_read_complete(op, ret);
406 }
407 
408 static void coroutine_fn mirror_co_zero(void *opaque)
409 {
410     MirrorOp *op = opaque;
411     int ret;
412 
413     op->s->in_flight++;
414     op->s->bytes_in_flight += op->bytes;
415     *op->bytes_handled = op->bytes;
416     op->is_in_flight = true;
417 
418     ret = blk_co_pwrite_zeroes(op->s->target, op->offset, op->bytes,
419                                op->s->unmap ? BDRV_REQ_MAY_UNMAP : 0);
420     mirror_write_complete(op, ret);
421 }
422 
423 static void coroutine_fn mirror_co_discard(void *opaque)
424 {
425     MirrorOp *op = opaque;
426     int ret;
427 
428     op->s->in_flight++;
429     op->s->bytes_in_flight += op->bytes;
430     *op->bytes_handled = op->bytes;
431     op->is_in_flight = true;
432 
433     ret = blk_co_pdiscard(op->s->target, op->offset, op->bytes);
434     mirror_write_complete(op, ret);
435 }
436 
437 static unsigned mirror_perform(MirrorBlockJob *s, int64_t offset,
438                                unsigned bytes, MirrorMethod mirror_method)
439 {
440     MirrorOp *op;
441     Coroutine *co;
442     int64_t bytes_handled = -1;
443 
444     op = g_new(MirrorOp, 1);
445     *op = (MirrorOp){
446         .s              = s,
447         .offset         = offset,
448         .bytes          = bytes,
449         .bytes_handled  = &bytes_handled,
450     };
451     qemu_co_queue_init(&op->waiting_requests);
452 
453     switch (mirror_method) {
454     case MIRROR_METHOD_COPY:
455         co = qemu_coroutine_create(mirror_co_read, op);
456         break;
457     case MIRROR_METHOD_ZERO:
458         co = qemu_coroutine_create(mirror_co_zero, op);
459         break;
460     case MIRROR_METHOD_DISCARD:
461         co = qemu_coroutine_create(mirror_co_discard, op);
462         break;
463     default:
464         abort();
465     }
466     op->co = co;
467 
468     QTAILQ_INSERT_TAIL(&s->ops_in_flight, op, next);
469     qemu_coroutine_enter(co);
470     /* At this point, ownership of op has been moved to the coroutine
471      * and the object may already be freed */
472 
473     /* Assert that this value has been set */
474     assert(bytes_handled >= 0);
475 
476     /* Same assertion as in mirror_co_read() (and for mirror_co_read()
477      * and mirror_co_discard(), bytes_handled == op->bytes, which
478      * is the @bytes parameter given to this function) */
479     assert(bytes_handled <= UINT_MAX);
480     return bytes_handled;
481 }
482 
483 static void coroutine_fn GRAPH_UNLOCKED mirror_iteration(MirrorBlockJob *s)
484 {
485     BlockDriverState *source;
486     MirrorOp *pseudo_op;
487     int64_t offset;
488     /* At least the first dirty chunk is mirrored in one iteration. */
489     int nb_chunks = 1;
490     bool write_zeroes_ok = bdrv_can_write_zeroes_with_unmap(blk_bs(s->target));
491     int max_io_bytes = MAX(s->buf_size / MAX_IN_FLIGHT, MAX_IO_BYTES);
492 
493     bdrv_graph_co_rdlock();
494     source = s->mirror_top_bs->backing->bs;
495     bdrv_graph_co_rdunlock();
496 
497     bdrv_dirty_bitmap_lock(s->dirty_bitmap);
498     offset = bdrv_dirty_iter_next(s->dbi);
499     if (offset < 0) {
500         bdrv_set_dirty_iter(s->dbi, 0);
501         offset = bdrv_dirty_iter_next(s->dbi);
502         trace_mirror_restart_iter(s, bdrv_get_dirty_count(s->dirty_bitmap));
503         assert(offset >= 0);
504     }
505     bdrv_dirty_bitmap_unlock(s->dirty_bitmap);
506 
507     /*
508      * Wait for concurrent requests to @offset.  The next loop will limit the
509      * copied area based on in_flight_bitmap so we only copy an area that does
510      * not overlap with concurrent in-flight requests.  Still, we would like to
511      * copy something, so wait until there are at least no more requests to the
512      * very beginning of the area.
513      */
514     mirror_wait_on_conflicts(NULL, s, offset, 1);
515 
516     job_pause_point(&s->common.job);
517 
518     /* Find the number of consecutive dirty chunks following the first dirty
519      * one, and wait for in flight requests in them. */
520     bdrv_dirty_bitmap_lock(s->dirty_bitmap);
521     while (nb_chunks * s->granularity < s->buf_size) {
522         int64_t next_dirty;
523         int64_t next_offset = offset + nb_chunks * s->granularity;
524         int64_t next_chunk = next_offset / s->granularity;
525         if (next_offset >= s->bdev_length ||
526             !bdrv_dirty_bitmap_get_locked(s->dirty_bitmap, next_offset)) {
527             break;
528         }
529         if (test_bit(next_chunk, s->in_flight_bitmap)) {
530             break;
531         }
532 
533         next_dirty = bdrv_dirty_iter_next(s->dbi);
534         if (next_dirty > next_offset || next_dirty < 0) {
535             /* The bitmap iterator's cache is stale, refresh it */
536             bdrv_set_dirty_iter(s->dbi, next_offset);
537             next_dirty = bdrv_dirty_iter_next(s->dbi);
538         }
539         assert(next_dirty == next_offset);
540         nb_chunks++;
541     }
542 
543     /* Clear dirty bits before querying the block status, because
544      * calling bdrv_block_status_above could yield - if some blocks are
545      * marked dirty in this window, we need to know.
546      */
547     bdrv_reset_dirty_bitmap_locked(s->dirty_bitmap, offset,
548                                    nb_chunks * s->granularity);
549     bdrv_dirty_bitmap_unlock(s->dirty_bitmap);
550 
551     /* Before claiming an area in the in-flight bitmap, we have to
552      * create a MirrorOp for it so that conflicting requests can wait
553      * for it.  mirror_perform() will create the real MirrorOps later,
554      * for now we just create a pseudo operation that will wake up all
555      * conflicting requests once all real operations have been
556      * launched. */
557     pseudo_op = g_new(MirrorOp, 1);
558     *pseudo_op = (MirrorOp){
559         .offset         = offset,
560         .bytes          = nb_chunks * s->granularity,
561         .is_pseudo_op   = true,
562     };
563     qemu_co_queue_init(&pseudo_op->waiting_requests);
564     QTAILQ_INSERT_TAIL(&s->ops_in_flight, pseudo_op, next);
565 
566     bitmap_set(s->in_flight_bitmap, offset / s->granularity, nb_chunks);
567     while (nb_chunks > 0 && offset < s->bdev_length) {
568         int ret;
569         int64_t io_bytes;
570         int64_t io_bytes_acct;
571         MirrorMethod mirror_method = MIRROR_METHOD_COPY;
572 
573         assert(!(offset % s->granularity));
574         WITH_GRAPH_RDLOCK_GUARD() {
575             ret = bdrv_co_block_status_above(source, NULL, offset,
576                                              nb_chunks * s->granularity,
577                                              &io_bytes, NULL, NULL);
578         }
579         if (ret < 0) {
580             io_bytes = MIN(nb_chunks * s->granularity, max_io_bytes);
581         } else if (ret & BDRV_BLOCK_DATA) {
582             io_bytes = MIN(io_bytes, max_io_bytes);
583         }
584 
585         io_bytes -= io_bytes % s->granularity;
586         if (io_bytes < s->granularity) {
587             io_bytes = s->granularity;
588         } else if (ret >= 0 && !(ret & BDRV_BLOCK_DATA)) {
589             int64_t target_offset;
590             int64_t target_bytes;
591             WITH_GRAPH_RDLOCK_GUARD() {
592                 bdrv_round_to_subclusters(blk_bs(s->target), offset, io_bytes,
593                                           &target_offset, &target_bytes);
594             }
595             if (target_offset == offset &&
596                 target_bytes == io_bytes) {
597                 mirror_method = ret & BDRV_BLOCK_ZERO ?
598                                     MIRROR_METHOD_ZERO :
599                                     MIRROR_METHOD_DISCARD;
600             }
601         }
602 
603         while (s->in_flight >= MAX_IN_FLIGHT) {
604             trace_mirror_yield_in_flight(s, offset, s->in_flight);
605             mirror_wait_for_free_in_flight_slot(s);
606         }
607 
608         if (s->ret < 0) {
609             ret = 0;
610             goto fail;
611         }
612 
613         io_bytes = mirror_clip_bytes(s, offset, io_bytes);
614         io_bytes = mirror_perform(s, offset, io_bytes, mirror_method);
615         if (mirror_method != MIRROR_METHOD_COPY && write_zeroes_ok) {
616             io_bytes_acct = 0;
617         } else {
618             io_bytes_acct = io_bytes;
619         }
620         assert(io_bytes);
621         offset += io_bytes;
622         nb_chunks -= DIV_ROUND_UP(io_bytes, s->granularity);
623         block_job_ratelimit_processed_bytes(&s->common, io_bytes_acct);
624     }
625 
626 fail:
627     QTAILQ_REMOVE(&s->ops_in_flight, pseudo_op, next);
628     qemu_co_queue_restart_all(&pseudo_op->waiting_requests);
629     g_free(pseudo_op);
630 }
631 
632 static void mirror_free_init(MirrorBlockJob *s)
633 {
634     int granularity = s->granularity;
635     size_t buf_size = s->buf_size;
636     uint8_t *buf = s->buf;
637 
638     assert(s->buf_free_count == 0);
639     QSIMPLEQ_INIT(&s->buf_free);
640     while (buf_size != 0) {
641         MirrorBuffer *cur = (MirrorBuffer *)buf;
642         QSIMPLEQ_INSERT_TAIL(&s->buf_free, cur, next);
643         s->buf_free_count++;
644         buf_size -= granularity;
645         buf += granularity;
646     }
647 }
648 
649 /* This is also used for the .pause callback. There is no matching
650  * mirror_resume() because mirror_run() will begin iterating again
651  * when the job is resumed.
652  */
653 static void coroutine_fn mirror_wait_for_all_io(MirrorBlockJob *s)
654 {
655     while (s->in_flight > 0) {
656         mirror_wait_for_free_in_flight_slot(s);
657     }
658 }
659 
660 /**
661  * mirror_exit_common: handle both abort() and prepare() cases.
662  * for .prepare, returns 0 on success and -errno on failure.
663  * for .abort cases, denoted by abort = true, MUST return 0.
664  */
665 static int mirror_exit_common(Job *job)
666 {
667     MirrorBlockJob *s = container_of(job, MirrorBlockJob, common.job);
668     BlockJob *bjob = &s->common;
669     MirrorBDSOpaque *bs_opaque;
670     BlockDriverState *src;
671     BlockDriverState *target_bs;
672     BlockDriverState *mirror_top_bs;
673     Error *local_err = NULL;
674     bool abort = job->ret < 0;
675     int ret = 0;
676 
677     GLOBAL_STATE_CODE();
678 
679     if (s->prepared) {
680         return 0;
681     }
682     s->prepared = true;
683 
684     bdrv_graph_rdlock_main_loop();
685 
686     mirror_top_bs = s->mirror_top_bs;
687     bs_opaque = mirror_top_bs->opaque;
688     src = mirror_top_bs->backing->bs;
689     target_bs = blk_bs(s->target);
690 
691     if (bdrv_chain_contains(src, target_bs)) {
692         bdrv_unfreeze_backing_chain(mirror_top_bs, target_bs);
693     }
694 
695     bdrv_release_dirty_bitmap(s->dirty_bitmap);
696 
697     /* Make sure that the source BDS doesn't go away during bdrv_replace_node,
698      * before we can call bdrv_drained_end */
699     bdrv_ref(src);
700     bdrv_ref(mirror_top_bs);
701     bdrv_ref(target_bs);
702 
703     bdrv_graph_rdunlock_main_loop();
704 
705     /*
706      * Remove target parent that still uses BLK_PERM_WRITE/RESIZE before
707      * inserting target_bs at s->to_replace, where we might not be able to get
708      * these permissions.
709      */
710     blk_unref(s->target);
711     s->target = NULL;
712 
713     /* We don't access the source any more. Dropping any WRITE/RESIZE is
714      * required before it could become a backing file of target_bs. Not having
715      * these permissions any more means that we can't allow any new requests on
716      * mirror_top_bs from now on, so keep it drained. */
717     bdrv_drained_begin(mirror_top_bs);
718     bdrv_drained_begin(target_bs);
719     bs_opaque->stop = true;
720 
721     bdrv_graph_rdlock_main_loop();
722     bdrv_child_refresh_perms(mirror_top_bs, mirror_top_bs->backing,
723                              &error_abort);
724 
725     if (!abort && s->backing_mode == MIRROR_SOURCE_BACKING_CHAIN) {
726         BlockDriverState *backing = s->is_none_mode ? src : s->base;
727         BlockDriverState *unfiltered_target = bdrv_skip_filters(target_bs);
728 
729         if (bdrv_cow_bs(unfiltered_target) != backing) {
730             bdrv_set_backing_hd(unfiltered_target, backing, &local_err);
731             if (local_err) {
732                 error_report_err(local_err);
733                 local_err = NULL;
734                 ret = -EPERM;
735             }
736         }
737     } else if (!abort && s->backing_mode == MIRROR_OPEN_BACKING_CHAIN) {
738         assert(!bdrv_backing_chain_next(target_bs));
739         ret = bdrv_open_backing_file(bdrv_skip_filters(target_bs), NULL,
740                                      "backing", &local_err);
741         if (ret < 0) {
742             error_report_err(local_err);
743             local_err = NULL;
744         }
745     }
746     bdrv_graph_rdunlock_main_loop();
747 
748     if (s->should_complete && !abort) {
749         BlockDriverState *to_replace = s->to_replace ?: src;
750         bool ro = bdrv_is_read_only(to_replace);
751 
752         if (ro != bdrv_is_read_only(target_bs)) {
753             bdrv_reopen_set_read_only(target_bs, ro, NULL);
754         }
755 
756         /* The mirror job has no requests in flight any more, but we need to
757          * drain potential other users of the BDS before changing the graph. */
758         assert(s->in_drain);
759         bdrv_drained_begin(to_replace);
760         /*
761          * Cannot use check_to_replace_node() here, because that would
762          * check for an op blocker on @to_replace, and we have our own
763          * there.
764          */
765         bdrv_graph_wrlock();
766         if (bdrv_recurse_can_replace(src, to_replace)) {
767             bdrv_replace_node(to_replace, target_bs, &local_err);
768         } else {
769             error_setg(&local_err, "Can no longer replace '%s' by '%s', "
770                        "because it can no longer be guaranteed that doing so "
771                        "would not lead to an abrupt change of visible data",
772                        to_replace->node_name, target_bs->node_name);
773         }
774         bdrv_graph_wrunlock();
775         bdrv_drained_end(to_replace);
776         if (local_err) {
777             error_report_err(local_err);
778             ret = -EPERM;
779         }
780     }
781     if (s->to_replace) {
782         bdrv_op_unblock_all(s->to_replace, s->replace_blocker);
783         error_free(s->replace_blocker);
784         bdrv_unref(s->to_replace);
785     }
786     g_free(s->replaces);
787 
788     /*
789      * Remove the mirror filter driver from the graph. Before this, get rid of
790      * the blockers on the intermediate nodes so that the resulting state is
791      * valid.
792      */
793     block_job_remove_all_bdrv(bjob);
794     bdrv_graph_wrlock();
795     bdrv_replace_node(mirror_top_bs, mirror_top_bs->backing->bs, &error_abort);
796     bdrv_graph_wrunlock();
797 
798     if (abort && s->base_ro && !bdrv_is_read_only(target_bs)) {
799         bdrv_reopen_set_read_only(target_bs, true, NULL);
800     }
801 
802     bdrv_drained_end(target_bs);
803     bdrv_unref(target_bs);
804 
805     bs_opaque->job = NULL;
806 
807     bdrv_drained_end(src);
808     bdrv_drained_end(mirror_top_bs);
809     s->in_drain = false;
810     bdrv_unref(mirror_top_bs);
811     bdrv_unref(src);
812 
813     return ret;
814 }
815 
816 static int mirror_prepare(Job *job)
817 {
818     return mirror_exit_common(job);
819 }
820 
821 static void mirror_abort(Job *job)
822 {
823     int ret = mirror_exit_common(job);
824     assert(ret == 0);
825 }
826 
827 static void coroutine_fn mirror_throttle(MirrorBlockJob *s)
828 {
829     int64_t now = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
830 
831     if (now - s->last_pause_ns > BLOCK_JOB_SLICE_TIME) {
832         s->last_pause_ns = now;
833         job_sleep_ns(&s->common.job, 0);
834     } else {
835         job_pause_point(&s->common.job);
836     }
837 }
838 
839 static int coroutine_fn GRAPH_UNLOCKED mirror_dirty_init(MirrorBlockJob *s)
840 {
841     int64_t offset;
842     BlockDriverState *bs;
843     BlockDriverState *target_bs = blk_bs(s->target);
844     int ret;
845     int64_t count;
846 
847     bdrv_graph_co_rdlock();
848     bs = s->mirror_top_bs->backing->bs;
849     bdrv_graph_co_rdunlock();
850 
851     if (s->zero_target) {
852         if (!bdrv_can_write_zeroes_with_unmap(target_bs)) {
853             bdrv_set_dirty_bitmap(s->dirty_bitmap, 0, s->bdev_length);
854             return 0;
855         }
856 
857         s->initial_zeroing_ongoing = true;
858         for (offset = 0; offset < s->bdev_length; ) {
859             int bytes = MIN(s->bdev_length - offset,
860                             QEMU_ALIGN_DOWN(INT_MAX, s->granularity));
861 
862             mirror_throttle(s);
863 
864             if (job_is_cancelled(&s->common.job)) {
865                 s->initial_zeroing_ongoing = false;
866                 return 0;
867             }
868 
869             if (s->in_flight >= MAX_IN_FLIGHT) {
870                 trace_mirror_yield(s, UINT64_MAX, s->buf_free_count,
871                                    s->in_flight);
872                 mirror_wait_for_free_in_flight_slot(s);
873                 continue;
874             }
875 
876             mirror_perform(s, offset, bytes, MIRROR_METHOD_ZERO);
877             offset += bytes;
878         }
879 
880         mirror_wait_for_all_io(s);
881         s->initial_zeroing_ongoing = false;
882     }
883 
884     /* First part, loop on the sectors and initialize the dirty bitmap.  */
885     for (offset = 0; offset < s->bdev_length; ) {
886         /* Just to make sure we are not exceeding int limit. */
887         int bytes = MIN(s->bdev_length - offset,
888                         QEMU_ALIGN_DOWN(INT_MAX, s->granularity));
889 
890         mirror_throttle(s);
891 
892         if (job_is_cancelled(&s->common.job)) {
893             return 0;
894         }
895 
896         WITH_GRAPH_RDLOCK_GUARD() {
897             ret = bdrv_co_is_allocated_above(bs, s->base_overlay, true, offset,
898                                              bytes, &count);
899         }
900         if (ret < 0) {
901             return ret;
902         }
903 
904         assert(count);
905         if (ret > 0) {
906             bdrv_set_dirty_bitmap(s->dirty_bitmap, offset, count);
907         }
908         offset += count;
909     }
910     return 0;
911 }
912 
913 /* Called when going out of the streaming phase to flush the bulk of the
914  * data to the medium, or just before completing.
915  */
916 static int coroutine_fn mirror_flush(MirrorBlockJob *s)
917 {
918     int ret = blk_co_flush(s->target);
919     if (ret < 0) {
920         if (mirror_error_action(s, false, -ret) == BLOCK_ERROR_ACTION_REPORT) {
921             s->ret = ret;
922         }
923     }
924     return ret;
925 }
926 
927 static int coroutine_fn mirror_run(Job *job, Error **errp)
928 {
929     MirrorBlockJob *s = container_of(job, MirrorBlockJob, common.job);
930     BlockDriverState *bs;
931     MirrorBDSOpaque *mirror_top_opaque = s->mirror_top_bs->opaque;
932     BlockDriverState *target_bs = blk_bs(s->target);
933     bool need_drain = true;
934     BlockDeviceIoStatus iostatus;
935     int64_t length;
936     int64_t target_length;
937     BlockDriverInfo bdi;
938     char backing_filename[2]; /* we only need 2 characters because we are only
939                                  checking for a NULL string */
940     int ret = 0;
941 
942     bdrv_graph_co_rdlock();
943     bs = bdrv_filter_bs(s->mirror_top_bs);
944     bdrv_graph_co_rdunlock();
945 
946     if (job_is_cancelled(&s->common.job)) {
947         goto immediate_exit;
948     }
949 
950     bdrv_graph_co_rdlock();
951     s->bdev_length = bdrv_co_getlength(bs);
952     bdrv_graph_co_rdunlock();
953 
954     if (s->bdev_length < 0) {
955         ret = s->bdev_length;
956         goto immediate_exit;
957     }
958 
959     target_length = blk_co_getlength(s->target);
960     if (target_length < 0) {
961         ret = target_length;
962         goto immediate_exit;
963     }
964 
965     /* Active commit must resize the base image if its size differs from the
966      * active layer. */
967     if (s->base == blk_bs(s->target)) {
968         if (s->bdev_length > target_length) {
969             ret = blk_co_truncate(s->target, s->bdev_length, false,
970                                   PREALLOC_MODE_OFF, 0, NULL);
971             if (ret < 0) {
972                 goto immediate_exit;
973             }
974         }
975     } else if (s->bdev_length != target_length) {
976         error_setg(errp, "Source and target image have different sizes");
977         ret = -EINVAL;
978         goto immediate_exit;
979     }
980 
981     if (s->bdev_length == 0) {
982         /* Transition to the READY state and wait for complete. */
983         job_transition_to_ready(&s->common.job);
984         qatomic_set(&s->actively_synced, true);
985         while (!job_cancel_requested(&s->common.job) && !s->should_complete) {
986             job_yield(&s->common.job);
987         }
988         goto immediate_exit;
989     }
990 
991     length = DIV_ROUND_UP(s->bdev_length, s->granularity);
992     s->in_flight_bitmap = bitmap_new(length);
993 
994     /* If we have no backing file yet in the destination, we cannot let
995      * the destination do COW.  Instead, we copy sectors around the
996      * dirty data if needed.  We need a bitmap to do that.
997      */
998     bdrv_get_backing_filename(target_bs, backing_filename,
999                               sizeof(backing_filename));
1000     bdrv_graph_co_rdlock();
1001     if (!bdrv_co_get_info(target_bs, &bdi) && bdi.cluster_size) {
1002         s->target_cluster_size = bdi.cluster_size;
1003     } else {
1004         s->target_cluster_size = BDRV_SECTOR_SIZE;
1005     }
1006     if (backing_filename[0] && !bdrv_backing_chain_next(target_bs) &&
1007         s->granularity < s->target_cluster_size) {
1008         s->buf_size = MAX(s->buf_size, s->target_cluster_size);
1009         s->cow_bitmap = bitmap_new(length);
1010     }
1011     s->max_iov = MIN(bs->bl.max_iov, target_bs->bl.max_iov);
1012     bdrv_graph_co_rdunlock();
1013 
1014     s->buf = qemu_try_blockalign(bs, s->buf_size);
1015     if (s->buf == NULL) {
1016         ret = -ENOMEM;
1017         goto immediate_exit;
1018     }
1019 
1020     mirror_free_init(s);
1021 
1022     s->last_pause_ns = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
1023     if (!s->is_none_mode) {
1024         ret = mirror_dirty_init(s);
1025         if (ret < 0 || job_is_cancelled(&s->common.job)) {
1026             goto immediate_exit;
1027         }
1028     }
1029 
1030     /*
1031      * Only now the job is fully initialised and mirror_top_bs should start
1032      * accessing it.
1033      */
1034     mirror_top_opaque->job = s;
1035 
1036     assert(!s->dbi);
1037     s->dbi = bdrv_dirty_iter_new(s->dirty_bitmap);
1038     for (;;) {
1039         int64_t cnt, delta;
1040         bool should_complete;
1041 
1042         if (s->ret < 0) {
1043             ret = s->ret;
1044             goto immediate_exit;
1045         }
1046 
1047         job_pause_point(&s->common.job);
1048 
1049         if (job_is_cancelled(&s->common.job)) {
1050             ret = 0;
1051             goto immediate_exit;
1052         }
1053 
1054         cnt = bdrv_get_dirty_count(s->dirty_bitmap);
1055         /* cnt is the number of dirty bytes remaining and s->bytes_in_flight is
1056          * the number of bytes currently being processed; together those are
1057          * the current remaining operation length */
1058         job_progress_set_remaining(&s->common.job,
1059                                    s->bytes_in_flight + cnt +
1060                                    s->active_write_bytes_in_flight);
1061 
1062         /* Note that even when no rate limit is applied we need to yield
1063          * periodically with no pending I/O so that bdrv_drain_all() returns.
1064          * We do so every BLKOCK_JOB_SLICE_TIME nanoseconds, or when there is
1065          * an error, or when the source is clean, whichever comes first. */
1066         delta = qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - s->last_pause_ns;
1067         WITH_JOB_LOCK_GUARD() {
1068             iostatus = s->common.iostatus;
1069         }
1070         if (delta < BLOCK_JOB_SLICE_TIME &&
1071             iostatus == BLOCK_DEVICE_IO_STATUS_OK) {
1072             if (s->in_flight >= MAX_IN_FLIGHT || s->buf_free_count == 0 ||
1073                 (cnt == 0 && s->in_flight > 0)) {
1074                 trace_mirror_yield(s, cnt, s->buf_free_count, s->in_flight);
1075                 mirror_wait_for_free_in_flight_slot(s);
1076                 continue;
1077             } else if (cnt != 0) {
1078                 mirror_iteration(s);
1079             }
1080         }
1081 
1082         should_complete = false;
1083         if (s->in_flight == 0 && cnt == 0) {
1084             trace_mirror_before_flush(s);
1085             if (!job_is_ready(&s->common.job)) {
1086                 if (mirror_flush(s) < 0) {
1087                     /* Go check s->ret.  */
1088                     continue;
1089                 }
1090                 /* We're out of the streaming phase.  From now on, if the job
1091                  * is cancelled we will actually complete all pending I/O and
1092                  * report completion.  This way, block-job-cancel will leave
1093                  * the target in a consistent state.
1094                  */
1095                 job_transition_to_ready(&s->common.job);
1096             }
1097             if (qatomic_read(&s->copy_mode) != MIRROR_COPY_MODE_BACKGROUND) {
1098                 qatomic_set(&s->actively_synced, true);
1099             }
1100 
1101             should_complete = s->should_complete ||
1102                 job_cancel_requested(&s->common.job);
1103             cnt = bdrv_get_dirty_count(s->dirty_bitmap);
1104         }
1105 
1106         if (cnt == 0 && should_complete) {
1107             /* The dirty bitmap is not updated while operations are pending.
1108              * If we're about to exit, wait for pending operations before
1109              * calling bdrv_get_dirty_count(bs), or we may exit while the
1110              * source has dirty data to copy!
1111              *
1112              * Note that I/O can be submitted by the guest while
1113              * mirror_populate runs, so pause it now.  Before deciding
1114              * whether to switch to target check one last time if I/O has
1115              * come in the meanwhile, and if not flush the data to disk.
1116              */
1117             trace_mirror_before_drain(s, cnt);
1118 
1119             s->in_drain = true;
1120             bdrv_drained_begin(bs);
1121 
1122             /* Must be zero because we are drained */
1123             assert(s->in_active_write_counter == 0);
1124 
1125             cnt = bdrv_get_dirty_count(s->dirty_bitmap);
1126             if (cnt > 0 || mirror_flush(s) < 0) {
1127                 bdrv_drained_end(bs);
1128                 s->in_drain = false;
1129                 continue;
1130             }
1131 
1132             /* The two disks are in sync.  Exit and report successful
1133              * completion.
1134              */
1135             assert(QLIST_EMPTY(&bs->tracked_requests));
1136             need_drain = false;
1137             break;
1138         }
1139 
1140         if (job_is_ready(&s->common.job) && !should_complete) {
1141             if (s->in_flight == 0 && cnt == 0) {
1142                 trace_mirror_before_sleep(s, cnt, job_is_ready(&s->common.job),
1143                                           BLOCK_JOB_SLICE_TIME);
1144                 job_sleep_ns(&s->common.job, BLOCK_JOB_SLICE_TIME);
1145             }
1146         } else {
1147             block_job_ratelimit_sleep(&s->common);
1148         }
1149         s->last_pause_ns = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
1150     }
1151 
1152 immediate_exit:
1153     if (s->in_flight > 0) {
1154         /* We get here only if something went wrong.  Either the job failed,
1155          * or it was cancelled prematurely so that we do not guarantee that
1156          * the target is a copy of the source.
1157          */
1158         assert(ret < 0 || job_is_cancelled(&s->common.job));
1159         assert(need_drain);
1160         mirror_wait_for_all_io(s);
1161     }
1162 
1163     assert(s->in_flight == 0);
1164     qemu_vfree(s->buf);
1165     g_free(s->cow_bitmap);
1166     g_free(s->in_flight_bitmap);
1167     bdrv_dirty_iter_free(s->dbi);
1168 
1169     if (need_drain) {
1170         s->in_drain = true;
1171         bdrv_drained_begin(bs);
1172     }
1173 
1174     return ret;
1175 }
1176 
1177 static void mirror_complete(Job *job, Error **errp)
1178 {
1179     MirrorBlockJob *s = container_of(job, MirrorBlockJob, common.job);
1180 
1181     if (!job_is_ready(job)) {
1182         error_setg(errp, "The active block job '%s' cannot be completed",
1183                    job->id);
1184         return;
1185     }
1186 
1187     /* block all operations on to_replace bs */
1188     if (s->replaces) {
1189         s->to_replace = bdrv_find_node(s->replaces);
1190         if (!s->to_replace) {
1191             error_setg(errp, "Node name '%s' not found", s->replaces);
1192             return;
1193         }
1194 
1195         /* TODO Translate this into child freeze system. */
1196         error_setg(&s->replace_blocker,
1197                    "block device is in use by block-job-complete");
1198         bdrv_op_block_all(s->to_replace, s->replace_blocker);
1199         bdrv_ref(s->to_replace);
1200     }
1201 
1202     s->should_complete = true;
1203 
1204     /* If the job is paused, it will be re-entered when it is resumed */
1205     WITH_JOB_LOCK_GUARD() {
1206         if (!job->paused) {
1207             job_enter_cond_locked(job, NULL);
1208         }
1209     }
1210 }
1211 
1212 static void coroutine_fn mirror_pause(Job *job)
1213 {
1214     MirrorBlockJob *s = container_of(job, MirrorBlockJob, common.job);
1215 
1216     mirror_wait_for_all_io(s);
1217 }
1218 
1219 static bool mirror_drained_poll(BlockJob *job)
1220 {
1221     MirrorBlockJob *s = container_of(job, MirrorBlockJob, common);
1222 
1223     /* If the job isn't paused nor cancelled, we can't be sure that it won't
1224      * issue more requests. We make an exception if we've reached this point
1225      * from one of our own drain sections, to avoid a deadlock waiting for
1226      * ourselves.
1227      */
1228     WITH_JOB_LOCK_GUARD() {
1229         if (!s->common.job.paused && !job_is_cancelled_locked(&job->job)
1230             && !s->in_drain) {
1231             return true;
1232         }
1233     }
1234 
1235     return !!s->in_flight;
1236 }
1237 
1238 static bool mirror_cancel(Job *job, bool force)
1239 {
1240     MirrorBlockJob *s = container_of(job, MirrorBlockJob, common.job);
1241     BlockDriverState *target = blk_bs(s->target);
1242 
1243     /*
1244      * Before the job is READY, we treat any cancellation like a
1245      * force-cancellation.
1246      */
1247     force = force || !job_is_ready(job);
1248 
1249     if (force) {
1250         bdrv_cancel_in_flight(target);
1251     }
1252     return force;
1253 }
1254 
1255 static bool commit_active_cancel(Job *job, bool force)
1256 {
1257     /* Same as above in mirror_cancel() */
1258     return force || !job_is_ready(job);
1259 }
1260 
1261 static void mirror_change(BlockJob *job, BlockJobChangeOptions *opts,
1262                           Error **errp)
1263 {
1264     MirrorBlockJob *s = container_of(job, MirrorBlockJob, common);
1265     BlockJobChangeOptionsMirror *change_opts = &opts->u.mirror;
1266     MirrorCopyMode current;
1267 
1268     /*
1269      * The implementation relies on the fact that copy_mode is only written
1270      * under the BQL. Otherwise, further synchronization would be required.
1271      */
1272 
1273     GLOBAL_STATE_CODE();
1274 
1275     if (qatomic_read(&s->copy_mode) == change_opts->copy_mode) {
1276         return;
1277     }
1278 
1279     if (change_opts->copy_mode != MIRROR_COPY_MODE_WRITE_BLOCKING) {
1280         error_setg(errp, "Change to copy mode '%s' is not implemented",
1281                    MirrorCopyMode_str(change_opts->copy_mode));
1282         return;
1283     }
1284 
1285     current = qatomic_cmpxchg(&s->copy_mode, MIRROR_COPY_MODE_BACKGROUND,
1286                               change_opts->copy_mode);
1287     if (current != MIRROR_COPY_MODE_BACKGROUND) {
1288         error_setg(errp, "Expected current copy mode '%s', got '%s'",
1289                    MirrorCopyMode_str(MIRROR_COPY_MODE_BACKGROUND),
1290                    MirrorCopyMode_str(current));
1291     }
1292 }
1293 
1294 static void mirror_query(BlockJob *job, BlockJobInfo *info)
1295 {
1296     MirrorBlockJob *s = container_of(job, MirrorBlockJob, common);
1297 
1298     info->u.mirror = (BlockJobInfoMirror) {
1299         .actively_synced = qatomic_read(&s->actively_synced),
1300     };
1301 }
1302 
1303 static const BlockJobDriver mirror_job_driver = {
1304     .job_driver = {
1305         .instance_size          = sizeof(MirrorBlockJob),
1306         .job_type               = JOB_TYPE_MIRROR,
1307         .free                   = block_job_free,
1308         .user_resume            = block_job_user_resume,
1309         .run                    = mirror_run,
1310         .prepare                = mirror_prepare,
1311         .abort                  = mirror_abort,
1312         .pause                  = mirror_pause,
1313         .complete               = mirror_complete,
1314         .cancel                 = mirror_cancel,
1315     },
1316     .drained_poll           = mirror_drained_poll,
1317     .change                 = mirror_change,
1318     .query                  = mirror_query,
1319 };
1320 
1321 static const BlockJobDriver commit_active_job_driver = {
1322     .job_driver = {
1323         .instance_size          = sizeof(MirrorBlockJob),
1324         .job_type               = JOB_TYPE_COMMIT,
1325         .free                   = block_job_free,
1326         .user_resume            = block_job_user_resume,
1327         .run                    = mirror_run,
1328         .prepare                = mirror_prepare,
1329         .abort                  = mirror_abort,
1330         .pause                  = mirror_pause,
1331         .complete               = mirror_complete,
1332         .cancel                 = commit_active_cancel,
1333     },
1334     .drained_poll           = mirror_drained_poll,
1335 };
1336 
1337 static void coroutine_fn
1338 do_sync_target_write(MirrorBlockJob *job, MirrorMethod method,
1339                      uint64_t offset, uint64_t bytes,
1340                      QEMUIOVector *qiov, int flags)
1341 {
1342     int ret;
1343     size_t qiov_offset = 0;
1344     int64_t bitmap_offset, bitmap_end;
1345 
1346     if (!QEMU_IS_ALIGNED(offset, job->granularity) &&
1347         bdrv_dirty_bitmap_get(job->dirty_bitmap, offset))
1348     {
1349             /*
1350              * Dirty unaligned padding: ignore it.
1351              *
1352              * Reasoning:
1353              * 1. If we copy it, we can't reset corresponding bit in
1354              *    dirty_bitmap as there may be some "dirty" bytes still not
1355              *    copied.
1356              * 2. It's already dirty, so skipping it we don't diverge mirror
1357              *    progress.
1358              *
1359              * Note, that because of this, guest write may have no contribution
1360              * into mirror converge, but that's not bad, as we have background
1361              * process of mirroring. If under some bad circumstances (high guest
1362              * IO load) background process starve, we will not converge anyway,
1363              * even if each write will contribute, as guest is not guaranteed to
1364              * rewrite the whole disk.
1365              */
1366             qiov_offset = QEMU_ALIGN_UP(offset, job->granularity) - offset;
1367             if (bytes <= qiov_offset) {
1368                 /* nothing to do after shrink */
1369                 return;
1370             }
1371             offset += qiov_offset;
1372             bytes -= qiov_offset;
1373     }
1374 
1375     if (!QEMU_IS_ALIGNED(offset + bytes, job->granularity) &&
1376         bdrv_dirty_bitmap_get(job->dirty_bitmap, offset + bytes - 1))
1377     {
1378         uint64_t tail = (offset + bytes) % job->granularity;
1379 
1380         if (bytes <= tail) {
1381             /* nothing to do after shrink */
1382             return;
1383         }
1384         bytes -= tail;
1385     }
1386 
1387     /*
1388      * Tails are either clean or shrunk, so for bitmap resetting
1389      * we safely align the range down.
1390      */
1391     bitmap_offset = QEMU_ALIGN_UP(offset, job->granularity);
1392     bitmap_end = QEMU_ALIGN_DOWN(offset + bytes, job->granularity);
1393     if (bitmap_offset < bitmap_end) {
1394         bdrv_reset_dirty_bitmap(job->dirty_bitmap, bitmap_offset,
1395                                 bitmap_end - bitmap_offset);
1396     }
1397 
1398     job_progress_increase_remaining(&job->common.job, bytes);
1399     job->active_write_bytes_in_flight += bytes;
1400 
1401     switch (method) {
1402     case MIRROR_METHOD_COPY:
1403         ret = blk_co_pwritev_part(job->target, offset, bytes,
1404                                   qiov, qiov_offset, flags);
1405         break;
1406 
1407     case MIRROR_METHOD_ZERO:
1408         assert(!qiov);
1409         ret = blk_co_pwrite_zeroes(job->target, offset, bytes, flags);
1410         break;
1411 
1412     case MIRROR_METHOD_DISCARD:
1413         assert(!qiov);
1414         ret = blk_co_pdiscard(job->target, offset, bytes);
1415         break;
1416 
1417     default:
1418         abort();
1419     }
1420 
1421     job->active_write_bytes_in_flight -= bytes;
1422     if (ret >= 0) {
1423         job_progress_update(&job->common.job, bytes);
1424     } else {
1425         BlockErrorAction action;
1426 
1427         /*
1428          * We failed, so we should mark dirty the whole area, aligned up.
1429          * Note that we don't care about shrunk tails if any: they were dirty
1430          * at function start, and they must be still dirty, as we've locked
1431          * the region for in-flight op.
1432          */
1433         bitmap_offset = QEMU_ALIGN_DOWN(offset, job->granularity);
1434         bitmap_end = QEMU_ALIGN_UP(offset + bytes, job->granularity);
1435         bdrv_set_dirty_bitmap(job->dirty_bitmap, bitmap_offset,
1436                               bitmap_end - bitmap_offset);
1437         qatomic_set(&job->actively_synced, false);
1438 
1439         action = mirror_error_action(job, false, -ret);
1440         if (action == BLOCK_ERROR_ACTION_REPORT) {
1441             if (!job->ret) {
1442                 job->ret = ret;
1443             }
1444         }
1445     }
1446 }
1447 
1448 static MirrorOp *coroutine_fn active_write_prepare(MirrorBlockJob *s,
1449                                                    uint64_t offset,
1450                                                    uint64_t bytes)
1451 {
1452     MirrorOp *op;
1453     uint64_t start_chunk = offset / s->granularity;
1454     uint64_t end_chunk = DIV_ROUND_UP(offset + bytes, s->granularity);
1455 
1456     op = g_new(MirrorOp, 1);
1457     *op = (MirrorOp){
1458         .s                  = s,
1459         .offset             = offset,
1460         .bytes              = bytes,
1461         .is_active_write    = true,
1462         .is_in_flight       = true,
1463         .co                 = qemu_coroutine_self(),
1464     };
1465     qemu_co_queue_init(&op->waiting_requests);
1466     QTAILQ_INSERT_TAIL(&s->ops_in_flight, op, next);
1467 
1468     s->in_active_write_counter++;
1469 
1470     /*
1471      * Wait for concurrent requests affecting the area.  If there are already
1472      * running requests that are copying off now-to-be stale data in the area,
1473      * we must wait for them to finish before we begin writing fresh data to the
1474      * target so that the write operations appear in the correct order.
1475      * Note that background requests (see mirror_iteration()) in contrast only
1476      * wait for conflicting requests at the start of the dirty area, and then
1477      * (based on the in_flight_bitmap) truncate the area to copy so it will not
1478      * conflict with any requests beyond that.  For active writes, however, we
1479      * cannot truncate that area.  The request from our parent must be blocked
1480      * until the area is copied in full.  Therefore, we must wait for the whole
1481      * area to become free of concurrent requests.
1482      */
1483     mirror_wait_on_conflicts(op, s, offset, bytes);
1484 
1485     bitmap_set(s->in_flight_bitmap, start_chunk, end_chunk - start_chunk);
1486 
1487     return op;
1488 }
1489 
1490 static void coroutine_fn GRAPH_RDLOCK active_write_settle(MirrorOp *op)
1491 {
1492     uint64_t start_chunk = op->offset / op->s->granularity;
1493     uint64_t end_chunk = DIV_ROUND_UP(op->offset + op->bytes,
1494                                       op->s->granularity);
1495 
1496     if (!--op->s->in_active_write_counter &&
1497         qatomic_read(&op->s->actively_synced)) {
1498         BdrvChild *source = op->s->mirror_top_bs->backing;
1499 
1500         if (QLIST_FIRST(&source->bs->parents) == source &&
1501             QLIST_NEXT(source, next_parent) == NULL)
1502         {
1503             /* Assert that we are back in sync once all active write
1504              * operations are settled.
1505              * Note that we can only assert this if the mirror node
1506              * is the source node's only parent. */
1507             assert(!bdrv_get_dirty_count(op->s->dirty_bitmap));
1508         }
1509     }
1510     bitmap_clear(op->s->in_flight_bitmap, start_chunk, end_chunk - start_chunk);
1511     QTAILQ_REMOVE(&op->s->ops_in_flight, op, next);
1512     qemu_co_queue_restart_all(&op->waiting_requests);
1513     g_free(op);
1514 }
1515 
1516 static int coroutine_fn GRAPH_RDLOCK
1517 bdrv_mirror_top_preadv(BlockDriverState *bs, int64_t offset, int64_t bytes,
1518                        QEMUIOVector *qiov, BdrvRequestFlags flags)
1519 {
1520     return bdrv_co_preadv(bs->backing, offset, bytes, qiov, flags);
1521 }
1522 
1523 static bool should_copy_to_target(MirrorBDSOpaque *s)
1524 {
1525     return s->job && s->job->ret >= 0 &&
1526         !job_is_cancelled(&s->job->common.job) &&
1527         qatomic_read(&s->job->copy_mode) == MIRROR_COPY_MODE_WRITE_BLOCKING;
1528 }
1529 
1530 static int coroutine_fn GRAPH_RDLOCK
1531 bdrv_mirror_top_do_write(BlockDriverState *bs, MirrorMethod method,
1532                          bool copy_to_target, uint64_t offset, uint64_t bytes,
1533                          QEMUIOVector *qiov, int flags)
1534 {
1535     MirrorOp *op = NULL;
1536     MirrorBDSOpaque *s = bs->opaque;
1537     int ret = 0;
1538 
1539     if (copy_to_target) {
1540         op = active_write_prepare(s->job, offset, bytes);
1541     }
1542 
1543     switch (method) {
1544     case MIRROR_METHOD_COPY:
1545         ret = bdrv_co_pwritev(bs->backing, offset, bytes, qiov, flags);
1546         break;
1547 
1548     case MIRROR_METHOD_ZERO:
1549         ret = bdrv_co_pwrite_zeroes(bs->backing, offset, bytes, flags);
1550         break;
1551 
1552     case MIRROR_METHOD_DISCARD:
1553         ret = bdrv_co_pdiscard(bs->backing, offset, bytes);
1554         break;
1555 
1556     default:
1557         abort();
1558     }
1559 
1560     if (!copy_to_target && s->job && s->job->dirty_bitmap) {
1561         qatomic_set(&s->job->actively_synced, false);
1562         bdrv_set_dirty_bitmap(s->job->dirty_bitmap, offset, bytes);
1563     }
1564 
1565     if (ret < 0) {
1566         goto out;
1567     }
1568 
1569     if (copy_to_target) {
1570         do_sync_target_write(s->job, method, offset, bytes, qiov, flags);
1571     }
1572 
1573 out:
1574     if (copy_to_target) {
1575         active_write_settle(op);
1576     }
1577     return ret;
1578 }
1579 
1580 static int coroutine_fn GRAPH_RDLOCK
1581 bdrv_mirror_top_pwritev(BlockDriverState *bs, int64_t offset, int64_t bytes,
1582                         QEMUIOVector *qiov, BdrvRequestFlags flags)
1583 {
1584     QEMUIOVector bounce_qiov;
1585     void *bounce_buf;
1586     int ret = 0;
1587     bool copy_to_target = should_copy_to_target(bs->opaque);
1588 
1589     if (copy_to_target) {
1590         /* The guest might concurrently modify the data to write; but
1591          * the data on source and destination must match, so we have
1592          * to use a bounce buffer if we are going to write to the
1593          * target now. */
1594         bounce_buf = qemu_blockalign(bs, bytes);
1595         iov_to_buf_full(qiov->iov, qiov->niov, 0, bounce_buf, bytes);
1596 
1597         qemu_iovec_init(&bounce_qiov, 1);
1598         qemu_iovec_add(&bounce_qiov, bounce_buf, bytes);
1599         qiov = &bounce_qiov;
1600 
1601         flags &= ~BDRV_REQ_REGISTERED_BUF;
1602     }
1603 
1604     ret = bdrv_mirror_top_do_write(bs, MIRROR_METHOD_COPY, copy_to_target,
1605                                    offset, bytes, qiov, flags);
1606 
1607     if (copy_to_target) {
1608         qemu_iovec_destroy(&bounce_qiov);
1609         qemu_vfree(bounce_buf);
1610     }
1611 
1612     return ret;
1613 }
1614 
1615 static int coroutine_fn GRAPH_RDLOCK bdrv_mirror_top_flush(BlockDriverState *bs)
1616 {
1617     if (bs->backing == NULL) {
1618         /* we can be here after failed bdrv_append in mirror_start_job */
1619         return 0;
1620     }
1621     return bdrv_co_flush(bs->backing->bs);
1622 }
1623 
1624 static int coroutine_fn GRAPH_RDLOCK
1625 bdrv_mirror_top_pwrite_zeroes(BlockDriverState *bs, int64_t offset,
1626                               int64_t bytes, BdrvRequestFlags flags)
1627 {
1628     bool copy_to_target = should_copy_to_target(bs->opaque);
1629     return bdrv_mirror_top_do_write(bs, MIRROR_METHOD_ZERO, copy_to_target,
1630                                     offset, bytes, NULL, flags);
1631 }
1632 
1633 static int coroutine_fn GRAPH_RDLOCK
1634 bdrv_mirror_top_pdiscard(BlockDriverState *bs, int64_t offset, int64_t bytes)
1635 {
1636     bool copy_to_target = should_copy_to_target(bs->opaque);
1637     return bdrv_mirror_top_do_write(bs, MIRROR_METHOD_DISCARD, copy_to_target,
1638                                     offset, bytes, NULL, 0);
1639 }
1640 
1641 static void GRAPH_RDLOCK bdrv_mirror_top_refresh_filename(BlockDriverState *bs)
1642 {
1643     if (bs->backing == NULL) {
1644         /* we can be here after failed bdrv_attach_child in
1645          * bdrv_set_backing_hd */
1646         return;
1647     }
1648     pstrcpy(bs->exact_filename, sizeof(bs->exact_filename),
1649             bs->backing->bs->filename);
1650 }
1651 
1652 static void bdrv_mirror_top_child_perm(BlockDriverState *bs, BdrvChild *c,
1653                                        BdrvChildRole role,
1654                                        BlockReopenQueue *reopen_queue,
1655                                        uint64_t perm, uint64_t shared,
1656                                        uint64_t *nperm, uint64_t *nshared)
1657 {
1658     MirrorBDSOpaque *s = bs->opaque;
1659 
1660     if (s->stop) {
1661         /*
1662          * If the job is to be stopped, we do not need to forward
1663          * anything to the real image.
1664          */
1665         *nperm = 0;
1666         *nshared = BLK_PERM_ALL;
1667         return;
1668     }
1669 
1670     bdrv_default_perms(bs, c, role, reopen_queue,
1671                        perm, shared, nperm, nshared);
1672 
1673     if (s->is_commit) {
1674         /*
1675          * For commit jobs, we cannot take CONSISTENT_READ, because
1676          * that permission is unshared for everything above the base
1677          * node (except for filters on the base node).
1678          * We also have to force-share the WRITE permission, or
1679          * otherwise we would block ourselves at the base node (if
1680          * writes are blocked for a node, they are also blocked for
1681          * its backing file).
1682          * (We could also share RESIZE, because it may be needed for
1683          * the target if its size is less than the top node's; but
1684          * bdrv_default_perms_for_cow() automatically shares RESIZE
1685          * for backing nodes if WRITE is shared, so there is no need
1686          * to do it here.)
1687          */
1688         *nperm &= ~BLK_PERM_CONSISTENT_READ;
1689         *nshared |= BLK_PERM_WRITE;
1690     }
1691 }
1692 
1693 /* Dummy node that provides consistent read to its users without requiring it
1694  * from its backing file and that allows writes on the backing file chain. */
1695 static BlockDriver bdrv_mirror_top = {
1696     .format_name                = "mirror_top",
1697     .bdrv_co_preadv             = bdrv_mirror_top_preadv,
1698     .bdrv_co_pwritev            = bdrv_mirror_top_pwritev,
1699     .bdrv_co_pwrite_zeroes      = bdrv_mirror_top_pwrite_zeroes,
1700     .bdrv_co_pdiscard           = bdrv_mirror_top_pdiscard,
1701     .bdrv_co_flush              = bdrv_mirror_top_flush,
1702     .bdrv_refresh_filename      = bdrv_mirror_top_refresh_filename,
1703     .bdrv_child_perm            = bdrv_mirror_top_child_perm,
1704 
1705     .is_filter                  = true,
1706     .filtered_child_is_backing  = true,
1707 };
1708 
1709 static BlockJob *mirror_start_job(
1710                              const char *job_id, BlockDriverState *bs,
1711                              int creation_flags, BlockDriverState *target,
1712                              const char *replaces, int64_t speed,
1713                              uint32_t granularity, int64_t buf_size,
1714                              BlockMirrorBackingMode backing_mode,
1715                              bool zero_target,
1716                              BlockdevOnError on_source_error,
1717                              BlockdevOnError on_target_error,
1718                              bool unmap,
1719                              BlockCompletionFunc *cb,
1720                              void *opaque,
1721                              const BlockJobDriver *driver,
1722                              bool is_none_mode, BlockDriverState *base,
1723                              bool auto_complete, const char *filter_node_name,
1724                              bool is_mirror, MirrorCopyMode copy_mode,
1725                              bool base_ro,
1726                              Error **errp)
1727 {
1728     MirrorBlockJob *s;
1729     MirrorBDSOpaque *bs_opaque;
1730     BlockDriverState *mirror_top_bs;
1731     bool target_is_backing;
1732     uint64_t target_perms, target_shared_perms;
1733     int ret;
1734 
1735     GLOBAL_STATE_CODE();
1736 
1737     if (granularity == 0) {
1738         granularity = bdrv_get_default_bitmap_granularity(target);
1739     }
1740 
1741     assert(is_power_of_2(granularity));
1742 
1743     if (buf_size < 0) {
1744         error_setg(errp, "Invalid parameter 'buf-size'");
1745         return NULL;
1746     }
1747 
1748     if (buf_size == 0) {
1749         buf_size = DEFAULT_MIRROR_BUF_SIZE;
1750     }
1751 
1752     bdrv_graph_rdlock_main_loop();
1753     if (bdrv_skip_filters(bs) == bdrv_skip_filters(target)) {
1754         error_setg(errp, "Can't mirror node into itself");
1755         bdrv_graph_rdunlock_main_loop();
1756         return NULL;
1757     }
1758 
1759     target_is_backing = bdrv_chain_contains(bs, target);
1760     bdrv_graph_rdunlock_main_loop();
1761 
1762     /* In the case of active commit, add dummy driver to provide consistent
1763      * reads on the top, while disabling it in the intermediate nodes, and make
1764      * the backing chain writable. */
1765     mirror_top_bs = bdrv_new_open_driver(&bdrv_mirror_top, filter_node_name,
1766                                          BDRV_O_RDWR, errp);
1767     if (mirror_top_bs == NULL) {
1768         return NULL;
1769     }
1770     if (!filter_node_name) {
1771         mirror_top_bs->implicit = true;
1772     }
1773 
1774     /* So that we can always drop this node */
1775     mirror_top_bs->never_freeze = true;
1776 
1777     mirror_top_bs->total_sectors = bs->total_sectors;
1778     mirror_top_bs->supported_write_flags = BDRV_REQ_WRITE_UNCHANGED;
1779     mirror_top_bs->supported_zero_flags = BDRV_REQ_WRITE_UNCHANGED |
1780                                           BDRV_REQ_NO_FALLBACK;
1781     bs_opaque = g_new0(MirrorBDSOpaque, 1);
1782     mirror_top_bs->opaque = bs_opaque;
1783 
1784     bs_opaque->is_commit = target_is_backing;
1785 
1786     bdrv_drained_begin(bs);
1787     ret = bdrv_append(mirror_top_bs, bs, errp);
1788     bdrv_drained_end(bs);
1789 
1790     if (ret < 0) {
1791         bdrv_unref(mirror_top_bs);
1792         return NULL;
1793     }
1794 
1795     /* Make sure that the source is not resized while the job is running */
1796     s = block_job_create(job_id, driver, NULL, mirror_top_bs,
1797                          BLK_PERM_CONSISTENT_READ,
1798                          BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE_UNCHANGED |
1799                          BLK_PERM_WRITE, speed,
1800                          creation_flags, cb, opaque, errp);
1801     if (!s) {
1802         goto fail;
1803     }
1804 
1805     /* The block job now has a reference to this node */
1806     bdrv_unref(mirror_top_bs);
1807 
1808     s->mirror_top_bs = mirror_top_bs;
1809     s->base_ro = base_ro;
1810 
1811     /* No resize for the target either; while the mirror is still running, a
1812      * consistent read isn't necessarily possible. We could possibly allow
1813      * writes and graph modifications, though it would likely defeat the
1814      * purpose of a mirror, so leave them blocked for now.
1815      *
1816      * In the case of active commit, things look a bit different, though,
1817      * because the target is an already populated backing file in active use.
1818      * We can allow anything except resize there.*/
1819 
1820     target_perms = BLK_PERM_WRITE;
1821     target_shared_perms = BLK_PERM_WRITE_UNCHANGED;
1822 
1823     if (target_is_backing) {
1824         int64_t bs_size, target_size;
1825         bs_size = bdrv_getlength(bs);
1826         if (bs_size < 0) {
1827             error_setg_errno(errp, -bs_size,
1828                              "Could not inquire top image size");
1829             goto fail;
1830         }
1831 
1832         target_size = bdrv_getlength(target);
1833         if (target_size < 0) {
1834             error_setg_errno(errp, -target_size,
1835                              "Could not inquire base image size");
1836             goto fail;
1837         }
1838 
1839         if (target_size < bs_size) {
1840             target_perms |= BLK_PERM_RESIZE;
1841         }
1842 
1843         target_shared_perms |= BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE;
1844     } else {
1845         bdrv_graph_rdlock_main_loop();
1846         if (bdrv_chain_contains(bs, bdrv_skip_filters(target))) {
1847             /*
1848              * We may want to allow this in the future, but it would
1849              * require taking some extra care.
1850              */
1851             error_setg(errp, "Cannot mirror to a filter on top of a node in "
1852                        "the source's backing chain");
1853             bdrv_graph_rdunlock_main_loop();
1854             goto fail;
1855         }
1856         bdrv_graph_rdunlock_main_loop();
1857     }
1858 
1859     s->target = blk_new(s->common.job.aio_context,
1860                         target_perms, target_shared_perms);
1861     ret = blk_insert_bs(s->target, target, errp);
1862     if (ret < 0) {
1863         goto fail;
1864     }
1865     if (is_mirror) {
1866         /* XXX: Mirror target could be a NBD server of target QEMU in the case
1867          * of non-shared block migration. To allow migration completion, we
1868          * have to allow "inactivate" of the target BB.  When that happens, we
1869          * know the job is drained, and the vcpus are stopped, so no write
1870          * operation will be performed. Block layer already has assertions to
1871          * ensure that. */
1872         blk_set_force_allow_inactivate(s->target);
1873     }
1874     blk_set_allow_aio_context_change(s->target, true);
1875     blk_set_disable_request_queuing(s->target, true);
1876 
1877     bdrv_graph_rdlock_main_loop();
1878     s->replaces = g_strdup(replaces);
1879     s->on_source_error = on_source_error;
1880     s->on_target_error = on_target_error;
1881     s->is_none_mode = is_none_mode;
1882     s->backing_mode = backing_mode;
1883     s->zero_target = zero_target;
1884     qatomic_set(&s->copy_mode, copy_mode);
1885     s->base = base;
1886     s->base_overlay = bdrv_find_overlay(bs, base);
1887     s->granularity = granularity;
1888     s->buf_size = ROUND_UP(buf_size, granularity);
1889     s->unmap = unmap;
1890     if (auto_complete) {
1891         s->should_complete = true;
1892     }
1893     bdrv_graph_rdunlock_main_loop();
1894 
1895     s->dirty_bitmap = bdrv_create_dirty_bitmap(s->mirror_top_bs, granularity,
1896                                                NULL, errp);
1897     if (!s->dirty_bitmap) {
1898         goto fail;
1899     }
1900 
1901     /*
1902      * The dirty bitmap is set by bdrv_mirror_top_do_write() when not in active
1903      * mode.
1904      */
1905     bdrv_disable_dirty_bitmap(s->dirty_bitmap);
1906 
1907     bdrv_graph_wrlock();
1908     ret = block_job_add_bdrv(&s->common, "source", bs, 0,
1909                              BLK_PERM_WRITE_UNCHANGED | BLK_PERM_WRITE |
1910                              BLK_PERM_CONSISTENT_READ,
1911                              errp);
1912     if (ret < 0) {
1913         bdrv_graph_wrunlock();
1914         goto fail;
1915     }
1916 
1917     /* Required permissions are already taken with blk_new() */
1918     block_job_add_bdrv(&s->common, "target", target, 0, BLK_PERM_ALL,
1919                        &error_abort);
1920 
1921     /* In commit_active_start() all intermediate nodes disappear, so
1922      * any jobs in them must be blocked */
1923     if (target_is_backing) {
1924         BlockDriverState *iter, *filtered_target;
1925         uint64_t iter_shared_perms;
1926 
1927         /*
1928          * The topmost node with
1929          * bdrv_skip_filters(filtered_target) == bdrv_skip_filters(target)
1930          */
1931         filtered_target = bdrv_cow_bs(bdrv_find_overlay(bs, target));
1932 
1933         assert(bdrv_skip_filters(filtered_target) ==
1934                bdrv_skip_filters(target));
1935 
1936         /*
1937          * XXX BLK_PERM_WRITE needs to be allowed so we don't block
1938          * ourselves at s->base (if writes are blocked for a node, they are
1939          * also blocked for its backing file). The other options would be a
1940          * second filter driver above s->base (== target).
1941          */
1942         iter_shared_perms = BLK_PERM_WRITE_UNCHANGED | BLK_PERM_WRITE;
1943 
1944         for (iter = bdrv_filter_or_cow_bs(bs); iter != target;
1945              iter = bdrv_filter_or_cow_bs(iter))
1946         {
1947             if (iter == filtered_target) {
1948                 /*
1949                  * From here on, all nodes are filters on the base.
1950                  * This allows us to share BLK_PERM_CONSISTENT_READ.
1951                  */
1952                 iter_shared_perms |= BLK_PERM_CONSISTENT_READ;
1953             }
1954 
1955             ret = block_job_add_bdrv(&s->common, "intermediate node", iter, 0,
1956                                      iter_shared_perms, errp);
1957             if (ret < 0) {
1958                 bdrv_graph_wrunlock();
1959                 goto fail;
1960             }
1961         }
1962 
1963         if (bdrv_freeze_backing_chain(mirror_top_bs, target, errp) < 0) {
1964             bdrv_graph_wrunlock();
1965             goto fail;
1966         }
1967     }
1968     bdrv_graph_wrunlock();
1969 
1970     QTAILQ_INIT(&s->ops_in_flight);
1971 
1972     trace_mirror_start(bs, s, opaque);
1973     job_start(&s->common.job);
1974 
1975     return &s->common;
1976 
1977 fail:
1978     if (s) {
1979         /* Make sure this BDS does not go away until we have completed the graph
1980          * changes below */
1981         bdrv_ref(mirror_top_bs);
1982 
1983         g_free(s->replaces);
1984         blk_unref(s->target);
1985         bs_opaque->job = NULL;
1986         if (s->dirty_bitmap) {
1987             bdrv_release_dirty_bitmap(s->dirty_bitmap);
1988         }
1989         job_early_fail(&s->common.job);
1990     }
1991 
1992     bs_opaque->stop = true;
1993     bdrv_drained_begin(bs);
1994     bdrv_graph_wrlock();
1995     assert(mirror_top_bs->backing->bs == bs);
1996     bdrv_child_refresh_perms(mirror_top_bs, mirror_top_bs->backing,
1997                              &error_abort);
1998     bdrv_replace_node(mirror_top_bs, bs, &error_abort);
1999     bdrv_graph_wrunlock();
2000     bdrv_drained_end(bs);
2001 
2002     bdrv_unref(mirror_top_bs);
2003 
2004     return NULL;
2005 }
2006 
2007 void mirror_start(const char *job_id, BlockDriverState *bs,
2008                   BlockDriverState *target, const char *replaces,
2009                   int creation_flags, int64_t speed,
2010                   uint32_t granularity, int64_t buf_size,
2011                   MirrorSyncMode mode, BlockMirrorBackingMode backing_mode,
2012                   bool zero_target,
2013                   BlockdevOnError on_source_error,
2014                   BlockdevOnError on_target_error,
2015                   bool unmap, const char *filter_node_name,
2016                   MirrorCopyMode copy_mode, Error **errp)
2017 {
2018     bool is_none_mode;
2019     BlockDriverState *base;
2020 
2021     GLOBAL_STATE_CODE();
2022 
2023     if ((mode == MIRROR_SYNC_MODE_INCREMENTAL) ||
2024         (mode == MIRROR_SYNC_MODE_BITMAP)) {
2025         error_setg(errp, "Sync mode '%s' not supported",
2026                    MirrorSyncMode_str(mode));
2027         return;
2028     }
2029 
2030     bdrv_graph_rdlock_main_loop();
2031     is_none_mode = mode == MIRROR_SYNC_MODE_NONE;
2032     base = mode == MIRROR_SYNC_MODE_TOP ? bdrv_backing_chain_next(bs) : NULL;
2033     bdrv_graph_rdunlock_main_loop();
2034 
2035     mirror_start_job(job_id, bs, creation_flags, target, replaces,
2036                      speed, granularity, buf_size, backing_mode, zero_target,
2037                      on_source_error, on_target_error, unmap, NULL, NULL,
2038                      &mirror_job_driver, is_none_mode, base, false,
2039                      filter_node_name, true, copy_mode, false, errp);
2040 }
2041 
2042 BlockJob *commit_active_start(const char *job_id, BlockDriverState *bs,
2043                               BlockDriverState *base, int creation_flags,
2044                               int64_t speed, BlockdevOnError on_error,
2045                               const char *filter_node_name,
2046                               BlockCompletionFunc *cb, void *opaque,
2047                               bool auto_complete, Error **errp)
2048 {
2049     bool base_read_only;
2050     BlockJob *job;
2051 
2052     GLOBAL_STATE_CODE();
2053 
2054     base_read_only = bdrv_is_read_only(base);
2055 
2056     if (base_read_only) {
2057         if (bdrv_reopen_set_read_only(base, false, errp) < 0) {
2058             return NULL;
2059         }
2060     }
2061 
2062     job = mirror_start_job(
2063                      job_id, bs, creation_flags, base, NULL, speed, 0, 0,
2064                      MIRROR_LEAVE_BACKING_CHAIN, false,
2065                      on_error, on_error, true, cb, opaque,
2066                      &commit_active_job_driver, false, base, auto_complete,
2067                      filter_node_name, false, MIRROR_COPY_MODE_BACKGROUND,
2068                      base_read_only, errp);
2069     if (!job) {
2070         goto error_restore_flags;
2071     }
2072 
2073     return job;
2074 
2075 error_restore_flags:
2076     /* ignore error and errp for bdrv_reopen, because we want to propagate
2077      * the original error */
2078     if (base_read_only) {
2079         bdrv_reopen_set_read_only(base, true, NULL);
2080     }
2081     return NULL;
2082 }
2083