xref: /openbmc/qemu/blockdev.c (revision 2b3912f1350971fbc2c04d986a1d0c60ae757c78)
1 /*
2  * QEMU host block devices
3  *
4  * Copyright (c) 2003-2008 Fabrice Bellard
5  *
6  * This work is licensed under the terms of the GNU GPL, version 2 or
7  * later.  See the COPYING file in the top-level directory.
8  *
9  * This file incorporates work covered by the following copyright and
10  * permission notice:
11  *
12  * Copyright (c) 2003-2008 Fabrice Bellard
13  *
14  * Permission is hereby granted, free of charge, to any person obtaining a copy
15  * of this software and associated documentation files (the "Software"), to deal
16  * in the Software without restriction, including without limitation the rights
17  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
18  * copies of the Software, and to permit persons to whom the Software is
19  * furnished to do so, subject to the following conditions:
20  *
21  * The above copyright notice and this permission notice shall be included in
22  * all copies or substantial portions of the Software.
23  *
24  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
25  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
26  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
27  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
28  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
29  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
30  * THE SOFTWARE.
31  */
32 
33 #include "qemu/osdep.h"
34 #include "sysemu/block-backend.h"
35 #include "sysemu/blockdev.h"
36 #include "hw/block/block.h"
37 #include "block/blockjob.h"
38 #include "block/dirty-bitmap.h"
39 #include "block/qdict.h"
40 #include "block/throttle-groups.h"
41 #include "monitor/monitor.h"
42 #include "qemu/error-report.h"
43 #include "qemu/option.h"
44 #include "qemu/qemu-print.h"
45 #include "qemu/config-file.h"
46 #include "qapi/qapi-commands-block.h"
47 #include "qapi/qapi-commands-transaction.h"
48 #include "qapi/qapi-visit-block-core.h"
49 #include "qapi/qmp/qdict.h"
50 #include "qapi/qmp/qnum.h"
51 #include "qapi/qmp/qstring.h"
52 #include "qapi/error.h"
53 #include "qapi/qmp/qerror.h"
54 #include "qapi/qmp/qlist.h"
55 #include "qapi/qobject-output-visitor.h"
56 #include "sysemu/sysemu.h"
57 #include "sysemu/iothread.h"
58 #include "block/block_int.h"
59 #include "block/trace.h"
60 #include "sysemu/runstate.h"
61 #include "sysemu/replay.h"
62 #include "qemu/cutils.h"
63 #include "qemu/help_option.h"
64 #include "qemu/main-loop.h"
65 #include "qemu/throttle-options.h"
66 
67 /* Protected by BQL */
68 QTAILQ_HEAD(, BlockDriverState) monitor_bdrv_states =
69     QTAILQ_HEAD_INITIALIZER(monitor_bdrv_states);
70 
71 void bdrv_set_monitor_owned(BlockDriverState *bs)
72 {
73     GLOBAL_STATE_CODE();
74     QTAILQ_INSERT_TAIL(&monitor_bdrv_states, bs, monitor_list);
75 }
76 
77 static const char *const if_name[IF_COUNT] = {
78     [IF_NONE] = "none",
79     [IF_IDE] = "ide",
80     [IF_SCSI] = "scsi",
81     [IF_FLOPPY] = "floppy",
82     [IF_PFLASH] = "pflash",
83     [IF_MTD] = "mtd",
84     [IF_SD] = "sd",
85     [IF_VIRTIO] = "virtio",
86     [IF_XEN] = "xen",
87 };
88 
89 static int if_max_devs[IF_COUNT] = {
90     /*
91      * Do not change these numbers!  They govern how drive option
92      * index maps to unit and bus.  That mapping is ABI.
93      *
94      * All controllers used to implement if=T drives need to support
95      * if_max_devs[T] units, for any T with if_max_devs[T] != 0.
96      * Otherwise, some index values map to "impossible" bus, unit
97      * values.
98      *
99      * For instance, if you change [IF_SCSI] to 255, -drive
100      * if=scsi,index=12 no longer means bus=1,unit=5, but
101      * bus=0,unit=12.  With an lsi53c895a controller (7 units max),
102      * the drive can't be set up.  Regression.
103      */
104     [IF_IDE] = 2,
105     [IF_SCSI] = 7,
106 };
107 
108 /**
109  * Boards may call this to offer board-by-board overrides
110  * of the default, global values.
111  */
112 void override_max_devs(BlockInterfaceType type, int max_devs)
113 {
114     BlockBackend *blk;
115     DriveInfo *dinfo;
116 
117     GLOBAL_STATE_CODE();
118 
119     if (max_devs <= 0) {
120         return;
121     }
122 
123     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
124         dinfo = blk_legacy_dinfo(blk);
125         if (dinfo->type == type) {
126             fprintf(stderr, "Cannot override units-per-bus property of"
127                     " the %s interface, because a drive of that type has"
128                     " already been added.\n", if_name[type]);
129             g_assert_not_reached();
130         }
131     }
132 
133     if_max_devs[type] = max_devs;
134 }
135 
136 /*
137  * We automatically delete the drive when a device using it gets
138  * unplugged.  Questionable feature, but we can't just drop it.
139  * Device models call blockdev_mark_auto_del() to schedule the
140  * automatic deletion, and generic qdev code calls blockdev_auto_del()
141  * when deletion is actually safe.
142  */
143 void blockdev_mark_auto_del(BlockBackend *blk)
144 {
145     DriveInfo *dinfo = blk_legacy_dinfo(blk);
146     BlockJob *job;
147 
148     GLOBAL_STATE_CODE();
149 
150     if (!dinfo) {
151         return;
152     }
153 
154     JOB_LOCK_GUARD();
155 
156     do {
157         job = block_job_next_locked(NULL);
158         while (job && (job->job.cancelled ||
159                        job->job.deferred_to_main_loop ||
160                        !block_job_has_bdrv(job, blk_bs(blk))))
161         {
162             job = block_job_next_locked(job);
163         }
164         if (job) {
165             /*
166              * This drops the job lock temporarily and polls, so we need to
167              * restart processing the list from the start after this.
168              */
169             job_cancel_locked(&job->job, false);
170         }
171     } while (job);
172 
173     dinfo->auto_del = 1;
174 }
175 
176 void blockdev_auto_del(BlockBackend *blk)
177 {
178     DriveInfo *dinfo = blk_legacy_dinfo(blk);
179     GLOBAL_STATE_CODE();
180 
181     if (dinfo && dinfo->auto_del) {
182         monitor_remove_blk(blk);
183         blk_unref(blk);
184     }
185 }
186 
187 static int drive_index_to_bus_id(BlockInterfaceType type, int index)
188 {
189     int max_devs = if_max_devs[type];
190     return max_devs ? index / max_devs : 0;
191 }
192 
193 static int drive_index_to_unit_id(BlockInterfaceType type, int index)
194 {
195     int max_devs = if_max_devs[type];
196     return max_devs ? index % max_devs : index;
197 }
198 
199 QemuOpts *drive_add(BlockInterfaceType type, int index, const char *file,
200                     const char *optstr)
201 {
202     QemuOpts *opts;
203 
204     GLOBAL_STATE_CODE();
205 
206     opts = qemu_opts_parse_noisily(qemu_find_opts("drive"), optstr, false);
207     if (!opts) {
208         return NULL;
209     }
210     if (type != IF_DEFAULT) {
211         qemu_opt_set(opts, "if", if_name[type], &error_abort);
212     }
213     if (index >= 0) {
214         qemu_opt_set_number(opts, "index", index, &error_abort);
215     }
216     if (file)
217         qemu_opt_set(opts, "file", file, &error_abort);
218     return opts;
219 }
220 
221 DriveInfo *drive_get(BlockInterfaceType type, int bus, int unit)
222 {
223     BlockBackend *blk;
224     DriveInfo *dinfo;
225 
226     GLOBAL_STATE_CODE();
227 
228     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
229         dinfo = blk_legacy_dinfo(blk);
230         if (dinfo && dinfo->type == type
231             && dinfo->bus == bus && dinfo->unit == unit) {
232             return dinfo;
233         }
234     }
235 
236     return NULL;
237 }
238 
239 /*
240  * Check board claimed all -drive that are meant to be claimed.
241  * Fatal error if any remain unclaimed.
242  */
243 void drive_check_orphaned(void)
244 {
245     BlockBackend *blk;
246     DriveInfo *dinfo;
247     Location loc;
248     bool orphans = false;
249 
250     GLOBAL_STATE_CODE();
251 
252     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
253         dinfo = blk_legacy_dinfo(blk);
254         /*
255          * Ignore default drives, because we create certain default
256          * drives unconditionally, then leave them unclaimed.  Not the
257          * users fault.
258          * Ignore IF_VIRTIO, because it gets desugared into -device,
259          * so we can leave failing to -device.
260          * Ignore IF_NONE, because leaving unclaimed IF_NONE remains
261          * available for device_add is a feature.
262          */
263         if (dinfo->is_default || dinfo->type == IF_VIRTIO
264             || dinfo->type == IF_NONE) {
265             continue;
266         }
267         if (!blk_get_attached_dev(blk)) {
268             loc_push_none(&loc);
269             qemu_opts_loc_restore(dinfo->opts);
270             error_report("machine type does not support"
271                          " if=%s,bus=%d,unit=%d",
272                          if_name[dinfo->type], dinfo->bus, dinfo->unit);
273             loc_pop(&loc);
274             orphans = true;
275         }
276     }
277 
278     if (orphans) {
279         exit(1);
280     }
281 }
282 
283 DriveInfo *drive_get_by_index(BlockInterfaceType type, int index)
284 {
285     GLOBAL_STATE_CODE();
286     return drive_get(type,
287                      drive_index_to_bus_id(type, index),
288                      drive_index_to_unit_id(type, index));
289 }
290 
291 int drive_get_max_bus(BlockInterfaceType type)
292 {
293     int max_bus;
294     BlockBackend *blk;
295     DriveInfo *dinfo;
296 
297     GLOBAL_STATE_CODE();
298 
299     max_bus = -1;
300     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
301         dinfo = blk_legacy_dinfo(blk);
302         if (dinfo && dinfo->type == type && dinfo->bus > max_bus) {
303             max_bus = dinfo->bus;
304         }
305     }
306     return max_bus;
307 }
308 
309 static void bdrv_format_print(void *opaque, const char *name)
310 {
311     qemu_printf(" %s", name);
312 }
313 
314 typedef struct {
315     QEMUBH *bh;
316     BlockDriverState *bs;
317 } BDRVPutRefBH;
318 
319 static int parse_block_error_action(const char *buf, bool is_read, Error **errp)
320 {
321     if (!strcmp(buf, "ignore")) {
322         return BLOCKDEV_ON_ERROR_IGNORE;
323     } else if (!is_read && !strcmp(buf, "enospc")) {
324         return BLOCKDEV_ON_ERROR_ENOSPC;
325     } else if (!strcmp(buf, "stop")) {
326         return BLOCKDEV_ON_ERROR_STOP;
327     } else if (!strcmp(buf, "report")) {
328         return BLOCKDEV_ON_ERROR_REPORT;
329     } else {
330         error_setg(errp, "'%s' invalid %s error action",
331                    buf, is_read ? "read" : "write");
332         return -1;
333     }
334 }
335 
336 static bool parse_stats_intervals(BlockAcctStats *stats, QList *intervals,
337                                   Error **errp)
338 {
339     const QListEntry *entry;
340     for (entry = qlist_first(intervals); entry; entry = qlist_next(entry)) {
341         switch (qobject_type(entry->value)) {
342 
343         case QTYPE_QSTRING: {
344             uint64_t length;
345             const char *str = qstring_get_str(qobject_to(QString,
346                                                          entry->value));
347             if (parse_uint_full(str, 10, &length) == 0 &&
348                 length > 0 && length <= UINT_MAX) {
349                 block_acct_add_interval(stats, (unsigned) length);
350             } else {
351                 error_setg(errp, "Invalid interval length: %s", str);
352                 return false;
353             }
354             break;
355         }
356 
357         case QTYPE_QNUM: {
358             int64_t length = qnum_get_int(qobject_to(QNum, entry->value));
359 
360             if (length > 0 && length <= UINT_MAX) {
361                 block_acct_add_interval(stats, (unsigned) length);
362             } else {
363                 error_setg(errp, "Invalid interval length: %" PRId64, length);
364                 return false;
365             }
366             break;
367         }
368 
369         default:
370             error_setg(errp, "The specification of stats-intervals is invalid");
371             return false;
372         }
373     }
374     return true;
375 }
376 
377 typedef enum { MEDIA_DISK, MEDIA_CDROM } DriveMediaType;
378 
379 /* All parameters but @opts are optional and may be set to NULL. */
380 static void extract_common_blockdev_options(QemuOpts *opts, int *bdrv_flags,
381     const char **throttling_group, ThrottleConfig *throttle_cfg,
382     BlockdevDetectZeroesOptions *detect_zeroes, Error **errp)
383 {
384     Error *local_error = NULL;
385     const char *aio;
386 
387     if (bdrv_flags) {
388         if (qemu_opt_get_bool(opts, "copy-on-read", false)) {
389             *bdrv_flags |= BDRV_O_COPY_ON_READ;
390         }
391 
392         if ((aio = qemu_opt_get(opts, "aio")) != NULL) {
393             if (bdrv_parse_aio(aio, bdrv_flags) < 0) {
394                 error_setg(errp, "invalid aio option");
395                 return;
396             }
397         }
398     }
399 
400     /* disk I/O throttling */
401     if (throttling_group) {
402         *throttling_group = qemu_opt_get(opts, "throttling.group");
403     }
404 
405     if (throttle_cfg) {
406         throttle_config_init(throttle_cfg);
407         throttle_cfg->buckets[THROTTLE_BPS_TOTAL].avg =
408             qemu_opt_get_number(opts, "throttling.bps-total", 0);
409         throttle_cfg->buckets[THROTTLE_BPS_READ].avg  =
410             qemu_opt_get_number(opts, "throttling.bps-read", 0);
411         throttle_cfg->buckets[THROTTLE_BPS_WRITE].avg =
412             qemu_opt_get_number(opts, "throttling.bps-write", 0);
413         throttle_cfg->buckets[THROTTLE_OPS_TOTAL].avg =
414             qemu_opt_get_number(opts, "throttling.iops-total", 0);
415         throttle_cfg->buckets[THROTTLE_OPS_READ].avg =
416             qemu_opt_get_number(opts, "throttling.iops-read", 0);
417         throttle_cfg->buckets[THROTTLE_OPS_WRITE].avg =
418             qemu_opt_get_number(opts, "throttling.iops-write", 0);
419 
420         throttle_cfg->buckets[THROTTLE_BPS_TOTAL].max =
421             qemu_opt_get_number(opts, "throttling.bps-total-max", 0);
422         throttle_cfg->buckets[THROTTLE_BPS_READ].max  =
423             qemu_opt_get_number(opts, "throttling.bps-read-max", 0);
424         throttle_cfg->buckets[THROTTLE_BPS_WRITE].max =
425             qemu_opt_get_number(opts, "throttling.bps-write-max", 0);
426         throttle_cfg->buckets[THROTTLE_OPS_TOTAL].max =
427             qemu_opt_get_number(opts, "throttling.iops-total-max", 0);
428         throttle_cfg->buckets[THROTTLE_OPS_READ].max =
429             qemu_opt_get_number(opts, "throttling.iops-read-max", 0);
430         throttle_cfg->buckets[THROTTLE_OPS_WRITE].max =
431             qemu_opt_get_number(opts, "throttling.iops-write-max", 0);
432 
433         throttle_cfg->buckets[THROTTLE_BPS_TOTAL].burst_length =
434             qemu_opt_get_number(opts, "throttling.bps-total-max-length", 1);
435         throttle_cfg->buckets[THROTTLE_BPS_READ].burst_length  =
436             qemu_opt_get_number(opts, "throttling.bps-read-max-length", 1);
437         throttle_cfg->buckets[THROTTLE_BPS_WRITE].burst_length =
438             qemu_opt_get_number(opts, "throttling.bps-write-max-length", 1);
439         throttle_cfg->buckets[THROTTLE_OPS_TOTAL].burst_length =
440             qemu_opt_get_number(opts, "throttling.iops-total-max-length", 1);
441         throttle_cfg->buckets[THROTTLE_OPS_READ].burst_length =
442             qemu_opt_get_number(opts, "throttling.iops-read-max-length", 1);
443         throttle_cfg->buckets[THROTTLE_OPS_WRITE].burst_length =
444             qemu_opt_get_number(opts, "throttling.iops-write-max-length", 1);
445 
446         throttle_cfg->op_size =
447             qemu_opt_get_number(opts, "throttling.iops-size", 0);
448 
449         if (!throttle_is_valid(throttle_cfg, errp)) {
450             return;
451         }
452     }
453 
454     if (detect_zeroes) {
455         *detect_zeroes =
456             qapi_enum_parse(&BlockdevDetectZeroesOptions_lookup,
457                             qemu_opt_get(opts, "detect-zeroes"),
458                             BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF,
459                             &local_error);
460         if (local_error) {
461             error_propagate(errp, local_error);
462             return;
463         }
464     }
465 }
466 
467 static OnOffAuto account_get_opt(QemuOpts *opts, const char *name)
468 {
469     if (!qemu_opt_find(opts, name)) {
470         return ON_OFF_AUTO_AUTO;
471     }
472     if (qemu_opt_get_bool(opts, name, true)) {
473         return ON_OFF_AUTO_ON;
474     }
475     return ON_OFF_AUTO_OFF;
476 }
477 
478 /* Takes the ownership of bs_opts */
479 static BlockBackend *blockdev_init(const char *file, QDict *bs_opts,
480                                    Error **errp)
481 {
482     const char *buf;
483     int bdrv_flags = 0;
484     int on_read_error, on_write_error;
485     OnOffAuto account_invalid, account_failed;
486     bool writethrough, read_only;
487     BlockBackend *blk;
488     BlockDriverState *bs;
489     ThrottleConfig cfg;
490     int snapshot = 0;
491     Error *error = NULL;
492     QemuOpts *opts;
493     QDict *interval_dict = NULL;
494     QList *interval_list = NULL;
495     const char *id;
496     BlockdevDetectZeroesOptions detect_zeroes =
497         BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF;
498     const char *throttling_group = NULL;
499 
500     /* Check common options by copying from bs_opts to opts, all other options
501      * stay in bs_opts for processing by bdrv_open(). */
502     id = qdict_get_try_str(bs_opts, "id");
503     opts = qemu_opts_create(&qemu_common_drive_opts, id, 1, errp);
504     if (!opts) {
505         goto err_no_opts;
506     }
507 
508     if (!qemu_opts_absorb_qdict(opts, bs_opts, errp)) {
509         goto early_err;
510     }
511 
512     if (id) {
513         qdict_del(bs_opts, "id");
514     }
515 
516     /* extract parameters */
517     snapshot = qemu_opt_get_bool(opts, "snapshot", 0);
518 
519     account_invalid = account_get_opt(opts, "stats-account-invalid");
520     account_failed = account_get_opt(opts, "stats-account-failed");
521 
522     writethrough = !qemu_opt_get_bool(opts, BDRV_OPT_CACHE_WB, true);
523 
524     id = qemu_opts_id(opts);
525 
526     qdict_extract_subqdict(bs_opts, &interval_dict, "stats-intervals.");
527     qdict_array_split(interval_dict, &interval_list);
528 
529     if (qdict_size(interval_dict) != 0) {
530         error_setg(errp, "Invalid option stats-intervals.%s",
531                    qdict_first(interval_dict)->key);
532         goto early_err;
533     }
534 
535     extract_common_blockdev_options(opts, &bdrv_flags, &throttling_group, &cfg,
536                                     &detect_zeroes, &error);
537     if (error) {
538         error_propagate(errp, error);
539         goto early_err;
540     }
541 
542     if ((buf = qemu_opt_get(opts, "format")) != NULL) {
543         if (is_help_option(buf)) {
544             qemu_printf("Supported formats:");
545             bdrv_iterate_format(bdrv_format_print, NULL, false);
546             qemu_printf("\nSupported formats (read-only):");
547             bdrv_iterate_format(bdrv_format_print, NULL, true);
548             qemu_printf("\n");
549             goto early_err;
550         }
551 
552         if (qdict_haskey(bs_opts, "driver")) {
553             error_setg(errp, "Cannot specify both 'driver' and 'format'");
554             goto early_err;
555         }
556         qdict_put_str(bs_opts, "driver", buf);
557     }
558 
559     on_write_error = BLOCKDEV_ON_ERROR_ENOSPC;
560     if ((buf = qemu_opt_get(opts, "werror")) != NULL) {
561         on_write_error = parse_block_error_action(buf, 0, &error);
562         if (error) {
563             error_propagate(errp, error);
564             goto early_err;
565         }
566     }
567 
568     on_read_error = BLOCKDEV_ON_ERROR_REPORT;
569     if ((buf = qemu_opt_get(opts, "rerror")) != NULL) {
570         on_read_error = parse_block_error_action(buf, 1, &error);
571         if (error) {
572             error_propagate(errp, error);
573             goto early_err;
574         }
575     }
576 
577     if (snapshot) {
578         bdrv_flags |= BDRV_O_SNAPSHOT;
579     }
580 
581     read_only = qemu_opt_get_bool(opts, BDRV_OPT_READ_ONLY, false);
582 
583     /* init */
584     if ((!file || !*file) && !qdict_size(bs_opts)) {
585         BlockBackendRootState *blk_rs;
586 
587         blk = blk_new(qemu_get_aio_context(), 0, BLK_PERM_ALL);
588         blk_rs = blk_get_root_state(blk);
589         blk_rs->open_flags    = bdrv_flags | (read_only ? 0 : BDRV_O_RDWR);
590         blk_rs->detect_zeroes = detect_zeroes;
591 
592         qobject_unref(bs_opts);
593     } else {
594         if (file && !*file) {
595             file = NULL;
596         }
597 
598         /* bdrv_open() defaults to the values in bdrv_flags (for compatibility
599          * with other callers) rather than what we want as the real defaults.
600          * Apply the defaults here instead. */
601         qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_DIRECT, "off");
602         qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_NO_FLUSH, "off");
603         qdict_set_default_str(bs_opts, BDRV_OPT_READ_ONLY,
604                               read_only ? "on" : "off");
605         qdict_set_default_str(bs_opts, BDRV_OPT_AUTO_READ_ONLY, "on");
606         assert((bdrv_flags & BDRV_O_CACHE_MASK) == 0);
607 
608         if (runstate_check(RUN_STATE_INMIGRATE)) {
609             bdrv_flags |= BDRV_O_INACTIVE;
610         }
611 
612         blk = blk_new_open(file, NULL, bs_opts, bdrv_flags, errp);
613         if (!blk) {
614             goto err_no_bs_opts;
615         }
616         bs = blk_bs(blk);
617 
618         bs->detect_zeroes = detect_zeroes;
619 
620         block_acct_setup(blk_get_stats(blk), account_invalid, account_failed);
621 
622         if (!parse_stats_intervals(blk_get_stats(blk), interval_list, errp)) {
623             blk_unref(blk);
624             blk = NULL;
625             goto err_no_bs_opts;
626         }
627     }
628 
629     /* disk I/O throttling */
630     if (throttle_enabled(&cfg)) {
631         if (!throttling_group) {
632             throttling_group = id;
633         }
634         blk_io_limits_enable(blk, throttling_group);
635         blk_set_io_limits(blk, &cfg);
636     }
637 
638     blk_set_enable_write_cache(blk, !writethrough);
639     blk_set_on_error(blk, on_read_error, on_write_error);
640 
641     if (!monitor_add_blk(blk, id, errp)) {
642         blk_unref(blk);
643         blk = NULL;
644         goto err_no_bs_opts;
645     }
646 
647 err_no_bs_opts:
648     qemu_opts_del(opts);
649     qobject_unref(interval_dict);
650     qobject_unref(interval_list);
651     return blk;
652 
653 early_err:
654     qemu_opts_del(opts);
655     qobject_unref(interval_dict);
656     qobject_unref(interval_list);
657 err_no_opts:
658     qobject_unref(bs_opts);
659     return NULL;
660 }
661 
662 /* Takes the ownership of bs_opts */
663 BlockDriverState *bds_tree_init(QDict *bs_opts, Error **errp)
664 {
665     BlockDriverState *bs;
666     int bdrv_flags = 0;
667 
668     GLOBAL_STATE_CODE();
669     /* bdrv_open() defaults to the values in bdrv_flags (for compatibility
670      * with other callers) rather than what we want as the real defaults.
671      * Apply the defaults here instead. */
672     qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_DIRECT, "off");
673     qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_NO_FLUSH, "off");
674     qdict_set_default_str(bs_opts, BDRV_OPT_READ_ONLY, "off");
675 
676     if (runstate_check(RUN_STATE_INMIGRATE)) {
677         bdrv_flags |= BDRV_O_INACTIVE;
678     }
679 
680     aio_context_acquire(qemu_get_aio_context());
681     bs = bdrv_open(NULL, NULL, bs_opts, bdrv_flags, errp);
682     aio_context_release(qemu_get_aio_context());
683 
684     return bs;
685 }
686 
687 void blockdev_close_all_bdrv_states(void)
688 {
689     BlockDriverState *bs, *next_bs;
690 
691     GLOBAL_STATE_CODE();
692     QTAILQ_FOREACH_SAFE(bs, &monitor_bdrv_states, monitor_list, next_bs) {
693         AioContext *ctx = bdrv_get_aio_context(bs);
694 
695         aio_context_acquire(ctx);
696         bdrv_unref(bs);
697         aio_context_release(ctx);
698     }
699 }
700 
701 /* Iterates over the list of monitor-owned BlockDriverStates */
702 BlockDriverState *bdrv_next_monitor_owned(BlockDriverState *bs)
703 {
704     GLOBAL_STATE_CODE();
705     return bs ? QTAILQ_NEXT(bs, monitor_list)
706               : QTAILQ_FIRST(&monitor_bdrv_states);
707 }
708 
709 static bool qemu_opt_rename(QemuOpts *opts, const char *from, const char *to,
710                             Error **errp)
711 {
712     const char *value;
713 
714     value = qemu_opt_get(opts, from);
715     if (value) {
716         if (qemu_opt_find(opts, to)) {
717             error_setg(errp, "'%s' and its alias '%s' can't be used at the "
718                        "same time", to, from);
719             return false;
720         }
721     }
722 
723     /* rename all items in opts */
724     while ((value = qemu_opt_get(opts, from))) {
725         qemu_opt_set(opts, to, value, &error_abort);
726         qemu_opt_unset(opts, from);
727     }
728     return true;
729 }
730 
731 QemuOptsList qemu_legacy_drive_opts = {
732     .name = "drive",
733     .head = QTAILQ_HEAD_INITIALIZER(qemu_legacy_drive_opts.head),
734     .desc = {
735         {
736             .name = "bus",
737             .type = QEMU_OPT_NUMBER,
738             .help = "bus number",
739         },{
740             .name = "unit",
741             .type = QEMU_OPT_NUMBER,
742             .help = "unit number (i.e. lun for scsi)",
743         },{
744             .name = "index",
745             .type = QEMU_OPT_NUMBER,
746             .help = "index number",
747         },{
748             .name = "media",
749             .type = QEMU_OPT_STRING,
750             .help = "media type (disk, cdrom)",
751         },{
752             .name = "if",
753             .type = QEMU_OPT_STRING,
754             .help = "interface (ide, scsi, sd, mtd, floppy, pflash, virtio)",
755         },{
756             .name = "file",
757             .type = QEMU_OPT_STRING,
758             .help = "file name",
759         },
760 
761         /* Options that are passed on, but have special semantics with -drive */
762         {
763             .name = BDRV_OPT_READ_ONLY,
764             .type = QEMU_OPT_BOOL,
765             .help = "open drive file as read-only",
766         },{
767             .name = "rerror",
768             .type = QEMU_OPT_STRING,
769             .help = "read error action",
770         },{
771             .name = "werror",
772             .type = QEMU_OPT_STRING,
773             .help = "write error action",
774         },{
775             .name = "copy-on-read",
776             .type = QEMU_OPT_BOOL,
777             .help = "copy read data from backing file into image file",
778         },
779 
780         { /* end of list */ }
781     },
782 };
783 
784 DriveInfo *drive_new(QemuOpts *all_opts, BlockInterfaceType block_default_type,
785                      Error **errp)
786 {
787     const char *value;
788     BlockBackend *blk;
789     DriveInfo *dinfo = NULL;
790     QDict *bs_opts;
791     QemuOpts *legacy_opts;
792     DriveMediaType media = MEDIA_DISK;
793     BlockInterfaceType type;
794     int max_devs, bus_id, unit_id, index;
795     const char *werror, *rerror;
796     bool read_only = false;
797     bool copy_on_read;
798     const char *filename;
799     int i;
800 
801     GLOBAL_STATE_CODE();
802 
803     /* Change legacy command line options into QMP ones */
804     static const struct {
805         const char *from;
806         const char *to;
807     } opt_renames[] = {
808         { "iops",           "throttling.iops-total" },
809         { "iops_rd",        "throttling.iops-read" },
810         { "iops_wr",        "throttling.iops-write" },
811 
812         { "bps",            "throttling.bps-total" },
813         { "bps_rd",         "throttling.bps-read" },
814         { "bps_wr",         "throttling.bps-write" },
815 
816         { "iops_max",       "throttling.iops-total-max" },
817         { "iops_rd_max",    "throttling.iops-read-max" },
818         { "iops_wr_max",    "throttling.iops-write-max" },
819 
820         { "bps_max",        "throttling.bps-total-max" },
821         { "bps_rd_max",     "throttling.bps-read-max" },
822         { "bps_wr_max",     "throttling.bps-write-max" },
823 
824         { "iops_size",      "throttling.iops-size" },
825 
826         { "group",          "throttling.group" },
827 
828         { "readonly",       BDRV_OPT_READ_ONLY },
829     };
830 
831     for (i = 0; i < ARRAY_SIZE(opt_renames); i++) {
832         if (!qemu_opt_rename(all_opts, opt_renames[i].from,
833                              opt_renames[i].to, errp)) {
834             return NULL;
835         }
836     }
837 
838     value = qemu_opt_get(all_opts, "cache");
839     if (value) {
840         int flags = 0;
841         bool writethrough;
842 
843         if (bdrv_parse_cache_mode(value, &flags, &writethrough) != 0) {
844             error_setg(errp, "invalid cache option");
845             return NULL;
846         }
847 
848         /* Specific options take precedence */
849         if (!qemu_opt_get(all_opts, BDRV_OPT_CACHE_WB)) {
850             qemu_opt_set_bool(all_opts, BDRV_OPT_CACHE_WB,
851                               !writethrough, &error_abort);
852         }
853         if (!qemu_opt_get(all_opts, BDRV_OPT_CACHE_DIRECT)) {
854             qemu_opt_set_bool(all_opts, BDRV_OPT_CACHE_DIRECT,
855                               !!(flags & BDRV_O_NOCACHE), &error_abort);
856         }
857         if (!qemu_opt_get(all_opts, BDRV_OPT_CACHE_NO_FLUSH)) {
858             qemu_opt_set_bool(all_opts, BDRV_OPT_CACHE_NO_FLUSH,
859                               !!(flags & BDRV_O_NO_FLUSH), &error_abort);
860         }
861         qemu_opt_unset(all_opts, "cache");
862     }
863 
864     /* Get a QDict for processing the options */
865     bs_opts = qdict_new();
866     qemu_opts_to_qdict(all_opts, bs_opts);
867 
868     legacy_opts = qemu_opts_create(&qemu_legacy_drive_opts, NULL, 0,
869                                    &error_abort);
870     if (!qemu_opts_absorb_qdict(legacy_opts, bs_opts, errp)) {
871         goto fail;
872     }
873 
874     /* Media type */
875     value = qemu_opt_get(legacy_opts, "media");
876     if (value) {
877         if (!strcmp(value, "disk")) {
878             media = MEDIA_DISK;
879         } else if (!strcmp(value, "cdrom")) {
880             media = MEDIA_CDROM;
881             read_only = true;
882         } else {
883             error_setg(errp, "'%s' invalid media", value);
884             goto fail;
885         }
886     }
887 
888     /* copy-on-read is disabled with a warning for read-only devices */
889     read_only |= qemu_opt_get_bool(legacy_opts, BDRV_OPT_READ_ONLY, false);
890     copy_on_read = qemu_opt_get_bool(legacy_opts, "copy-on-read", false);
891 
892     if (read_only && copy_on_read) {
893         warn_report("disabling copy-on-read on read-only drive");
894         copy_on_read = false;
895     }
896 
897     qdict_put_str(bs_opts, BDRV_OPT_READ_ONLY, read_only ? "on" : "off");
898     qdict_put_str(bs_opts, "copy-on-read", copy_on_read ? "on" : "off");
899 
900     /* Controller type */
901     value = qemu_opt_get(legacy_opts, "if");
902     if (value) {
903         for (type = 0;
904              type < IF_COUNT && strcmp(value, if_name[type]);
905              type++) {
906         }
907         if (type == IF_COUNT) {
908             error_setg(errp, "unsupported bus type '%s'", value);
909             goto fail;
910         }
911     } else {
912         type = block_default_type;
913     }
914 
915     /* Device address specified by bus/unit or index.
916      * If none was specified, try to find the first free one. */
917     bus_id  = qemu_opt_get_number(legacy_opts, "bus", 0);
918     unit_id = qemu_opt_get_number(legacy_opts, "unit", -1);
919     index   = qemu_opt_get_number(legacy_opts, "index", -1);
920 
921     max_devs = if_max_devs[type];
922 
923     if (index != -1) {
924         if (bus_id != 0 || unit_id != -1) {
925             error_setg(errp, "index cannot be used with bus and unit");
926             goto fail;
927         }
928         bus_id = drive_index_to_bus_id(type, index);
929         unit_id = drive_index_to_unit_id(type, index);
930     }
931 
932     if (unit_id == -1) {
933        unit_id = 0;
934        while (drive_get(type, bus_id, unit_id) != NULL) {
935            unit_id++;
936            if (max_devs && unit_id >= max_devs) {
937                unit_id -= max_devs;
938                bus_id++;
939            }
940        }
941     }
942 
943     if (max_devs && unit_id >= max_devs) {
944         error_setg(errp, "unit %d too big (max is %d)", unit_id, max_devs - 1);
945         goto fail;
946     }
947 
948     if (drive_get(type, bus_id, unit_id) != NULL) {
949         error_setg(errp, "drive with bus=%d, unit=%d (index=%d) exists",
950                    bus_id, unit_id, index);
951         goto fail;
952     }
953 
954     /* no id supplied -> create one */
955     if (qemu_opts_id(all_opts) == NULL) {
956         char *new_id;
957         const char *mediastr = "";
958         if (type == IF_IDE || type == IF_SCSI) {
959             mediastr = (media == MEDIA_CDROM) ? "-cd" : "-hd";
960         }
961         if (max_devs) {
962             new_id = g_strdup_printf("%s%i%s%i", if_name[type], bus_id,
963                                      mediastr, unit_id);
964         } else {
965             new_id = g_strdup_printf("%s%s%i", if_name[type],
966                                      mediastr, unit_id);
967         }
968         qdict_put_str(bs_opts, "id", new_id);
969         g_free(new_id);
970     }
971 
972     /* Add virtio block device */
973     if (type == IF_VIRTIO) {
974         QemuOpts *devopts;
975         devopts = qemu_opts_create(qemu_find_opts("device"), NULL, 0,
976                                    &error_abort);
977         qemu_opt_set(devopts, "driver", "virtio-blk", &error_abort);
978         qemu_opt_set(devopts, "drive", qdict_get_str(bs_opts, "id"),
979                      &error_abort);
980     }
981 
982     filename = qemu_opt_get(legacy_opts, "file");
983 
984     /* Check werror/rerror compatibility with if=... */
985     werror = qemu_opt_get(legacy_opts, "werror");
986     if (werror != NULL) {
987         if (type != IF_IDE && type != IF_SCSI && type != IF_VIRTIO &&
988             type != IF_NONE) {
989             error_setg(errp, "werror is not supported by this bus type");
990             goto fail;
991         }
992         qdict_put_str(bs_opts, "werror", werror);
993     }
994 
995     rerror = qemu_opt_get(legacy_opts, "rerror");
996     if (rerror != NULL) {
997         if (type != IF_IDE && type != IF_VIRTIO && type != IF_SCSI &&
998             type != IF_NONE) {
999             error_setg(errp, "rerror is not supported by this bus type");
1000             goto fail;
1001         }
1002         qdict_put_str(bs_opts, "rerror", rerror);
1003     }
1004 
1005     /* Actual block device init: Functionality shared with blockdev-add */
1006     blk = blockdev_init(filename, bs_opts, errp);
1007     bs_opts = NULL;
1008     if (!blk) {
1009         goto fail;
1010     }
1011 
1012     /* Create legacy DriveInfo */
1013     dinfo = g_malloc0(sizeof(*dinfo));
1014     dinfo->opts = all_opts;
1015 
1016     dinfo->type = type;
1017     dinfo->bus = bus_id;
1018     dinfo->unit = unit_id;
1019 
1020     blk_set_legacy_dinfo(blk, dinfo);
1021 
1022     switch(type) {
1023     case IF_IDE:
1024     case IF_SCSI:
1025     case IF_XEN:
1026     case IF_NONE:
1027         dinfo->media_cd = media == MEDIA_CDROM;
1028         break;
1029     default:
1030         break;
1031     }
1032 
1033 fail:
1034     qemu_opts_del(legacy_opts);
1035     qobject_unref(bs_opts);
1036     return dinfo;
1037 }
1038 
1039 static BlockDriverState *qmp_get_root_bs(const char *name, Error **errp)
1040 {
1041     BlockDriverState *bs;
1042     AioContext *aio_context;
1043 
1044     GRAPH_RDLOCK_GUARD_MAINLOOP();
1045 
1046     bs = bdrv_lookup_bs(name, name, errp);
1047     if (bs == NULL) {
1048         return NULL;
1049     }
1050 
1051     if (!bdrv_is_root_node(bs)) {
1052         error_setg(errp, "Need a root block node");
1053         return NULL;
1054     }
1055 
1056     aio_context = bdrv_get_aio_context(bs);
1057     aio_context_acquire(aio_context);
1058 
1059     if (!bdrv_is_inserted(bs)) {
1060         error_setg(errp, "Device has no medium");
1061         bs = NULL;
1062     }
1063 
1064     aio_context_release(aio_context);
1065 
1066     return bs;
1067 }
1068 
1069 static void blockdev_do_action(TransactionAction *action, Error **errp)
1070 {
1071     TransactionActionList list;
1072 
1073     list.value = action;
1074     list.next = NULL;
1075     qmp_transaction(&list, NULL, errp);
1076 }
1077 
1078 void qmp_blockdev_snapshot_sync(const char *device, const char *node_name,
1079                                 const char *snapshot_file,
1080                                 const char *snapshot_node_name,
1081                                 const char *format,
1082                                 bool has_mode, NewImageMode mode, Error **errp)
1083 {
1084     BlockdevSnapshotSync snapshot = {
1085         .device = (char *) device,
1086         .node_name = (char *) node_name,
1087         .snapshot_file = (char *) snapshot_file,
1088         .snapshot_node_name = (char *) snapshot_node_name,
1089         .format = (char *) format,
1090         .has_mode = has_mode,
1091         .mode = mode,
1092     };
1093     TransactionAction action = {
1094         .type = TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC,
1095         .u.blockdev_snapshot_sync.data = &snapshot,
1096     };
1097     blockdev_do_action(&action, errp);
1098 }
1099 
1100 void qmp_blockdev_snapshot(const char *node, const char *overlay,
1101                            Error **errp)
1102 {
1103     BlockdevSnapshot snapshot_data = {
1104         .node = (char *) node,
1105         .overlay = (char *) overlay
1106     };
1107     TransactionAction action = {
1108         .type = TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT,
1109         .u.blockdev_snapshot.data = &snapshot_data,
1110     };
1111     blockdev_do_action(&action, errp);
1112 }
1113 
1114 void qmp_blockdev_snapshot_internal_sync(const char *device,
1115                                          const char *name,
1116                                          Error **errp)
1117 {
1118     BlockdevSnapshotInternal snapshot = {
1119         .device = (char *) device,
1120         .name = (char *) name
1121     };
1122     TransactionAction action = {
1123         .type = TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC,
1124         .u.blockdev_snapshot_internal_sync.data = &snapshot,
1125     };
1126     blockdev_do_action(&action, errp);
1127 }
1128 
1129 SnapshotInfo *qmp_blockdev_snapshot_delete_internal_sync(const char *device,
1130                                                          const char *id,
1131                                                          const char *name,
1132                                                          Error **errp)
1133 {
1134     BlockDriverState *bs;
1135     AioContext *aio_context;
1136     QEMUSnapshotInfo sn;
1137     Error *local_err = NULL;
1138     SnapshotInfo *info = NULL;
1139     int ret;
1140 
1141     bs = qmp_get_root_bs(device, errp);
1142     if (!bs) {
1143         return NULL;
1144     }
1145     aio_context = bdrv_get_aio_context(bs);
1146     aio_context_acquire(aio_context);
1147 
1148     if (!id && !name) {
1149         error_setg(errp, "Name or id must be provided");
1150         goto out_aio_context;
1151     }
1152 
1153     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_INTERNAL_SNAPSHOT_DELETE, errp)) {
1154         goto out_aio_context;
1155     }
1156 
1157     ret = bdrv_snapshot_find_by_id_and_name(bs, id, name, &sn, &local_err);
1158     if (local_err) {
1159         error_propagate(errp, local_err);
1160         goto out_aio_context;
1161     }
1162     if (!ret) {
1163         error_setg(errp,
1164                    "Snapshot with id '%s' and name '%s' does not exist on "
1165                    "device '%s'",
1166                    STR_OR_NULL(id), STR_OR_NULL(name), device);
1167         goto out_aio_context;
1168     }
1169 
1170     bdrv_snapshot_delete(bs, id, name, &local_err);
1171     if (local_err) {
1172         error_propagate(errp, local_err);
1173         goto out_aio_context;
1174     }
1175 
1176     aio_context_release(aio_context);
1177 
1178     info = g_new0(SnapshotInfo, 1);
1179     info->id = g_strdup(sn.id_str);
1180     info->name = g_strdup(sn.name);
1181     info->date_nsec = sn.date_nsec;
1182     info->date_sec = sn.date_sec;
1183     info->vm_state_size = sn.vm_state_size;
1184     info->vm_clock_nsec = sn.vm_clock_nsec % 1000000000;
1185     info->vm_clock_sec = sn.vm_clock_nsec / 1000000000;
1186     if (sn.icount != -1ULL) {
1187         info->icount = sn.icount;
1188         info->has_icount = true;
1189     }
1190 
1191     return info;
1192 
1193 out_aio_context:
1194     aio_context_release(aio_context);
1195     return NULL;
1196 }
1197 
1198 /* internal snapshot private data */
1199 typedef struct InternalSnapshotState {
1200     BlockDriverState *bs;
1201     QEMUSnapshotInfo sn;
1202     bool created;
1203 } InternalSnapshotState;
1204 
1205 static void internal_snapshot_abort(void *opaque);
1206 static void internal_snapshot_clean(void *opaque);
1207 TransactionActionDrv internal_snapshot_drv = {
1208     .abort = internal_snapshot_abort,
1209     .clean = internal_snapshot_clean,
1210 };
1211 
1212 static void internal_snapshot_action(BlockdevSnapshotInternal *internal,
1213                                      Transaction *tran, Error **errp)
1214 {
1215     Error *local_err = NULL;
1216     const char *device;
1217     const char *name;
1218     BlockDriverState *bs;
1219     QEMUSnapshotInfo old_sn, *sn;
1220     bool ret;
1221     int64_t rt;
1222     InternalSnapshotState *state = g_new0(InternalSnapshotState, 1);
1223     AioContext *aio_context;
1224     int ret1;
1225 
1226     tran_add(tran, &internal_snapshot_drv, state);
1227 
1228     device = internal->device;
1229     name = internal->name;
1230 
1231     bs = qmp_get_root_bs(device, errp);
1232     if (!bs) {
1233         return;
1234     }
1235 
1236     aio_context = bdrv_get_aio_context(bs);
1237     aio_context_acquire(aio_context);
1238 
1239     state->bs = bs;
1240 
1241     /* Paired with .clean() */
1242     bdrv_drained_begin(bs);
1243 
1244     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_INTERNAL_SNAPSHOT, errp)) {
1245         goto out;
1246     }
1247 
1248     if (bdrv_is_read_only(bs)) {
1249         error_setg(errp, "Device '%s' is read only", device);
1250         goto out;
1251     }
1252 
1253     if (!bdrv_can_snapshot(bs)) {
1254         error_setg(errp, "Block format '%s' used by device '%s' "
1255                    "does not support internal snapshots",
1256                    bs->drv->format_name, device);
1257         goto out;
1258     }
1259 
1260     if (!strlen(name)) {
1261         error_setg(errp, "Name is empty");
1262         goto out;
1263     }
1264 
1265     /* check whether a snapshot with name exist */
1266     ret = bdrv_snapshot_find_by_id_and_name(bs, NULL, name, &old_sn,
1267                                             &local_err);
1268     if (local_err) {
1269         error_propagate(errp, local_err);
1270         goto out;
1271     } else if (ret) {
1272         error_setg(errp,
1273                    "Snapshot with name '%s' already exists on device '%s'",
1274                    name, device);
1275         goto out;
1276     }
1277 
1278     /* 3. take the snapshot */
1279     sn = &state->sn;
1280     pstrcpy(sn->name, sizeof(sn->name), name);
1281     rt = g_get_real_time();
1282     sn->date_sec = rt / G_USEC_PER_SEC;
1283     sn->date_nsec = (rt % G_USEC_PER_SEC) * 1000;
1284     sn->vm_clock_nsec = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL);
1285     if (replay_mode != REPLAY_MODE_NONE) {
1286         sn->icount = replay_get_current_icount();
1287     } else {
1288         sn->icount = -1ULL;
1289     }
1290 
1291     ret1 = bdrv_snapshot_create(bs, sn);
1292     if (ret1 < 0) {
1293         error_setg_errno(errp, -ret1,
1294                          "Failed to create snapshot '%s' on device '%s'",
1295                          name, device);
1296         goto out;
1297     }
1298 
1299     /* 4. succeed, mark a snapshot is created */
1300     state->created = true;
1301 
1302 out:
1303     aio_context_release(aio_context);
1304 }
1305 
1306 static void internal_snapshot_abort(void *opaque)
1307 {
1308     InternalSnapshotState *state = opaque;
1309     BlockDriverState *bs = state->bs;
1310     QEMUSnapshotInfo *sn = &state->sn;
1311     AioContext *aio_context;
1312     Error *local_error = NULL;
1313 
1314     if (!state->created) {
1315         return;
1316     }
1317 
1318     aio_context = bdrv_get_aio_context(state->bs);
1319     aio_context_acquire(aio_context);
1320 
1321     if (bdrv_snapshot_delete(bs, sn->id_str, sn->name, &local_error) < 0) {
1322         error_reportf_err(local_error,
1323                           "Failed to delete snapshot with id '%s' and "
1324                           "name '%s' on device '%s' in abort: ",
1325                           sn->id_str, sn->name,
1326                           bdrv_get_device_name(bs));
1327     }
1328 
1329     aio_context_release(aio_context);
1330 }
1331 
1332 static void internal_snapshot_clean(void *opaque)
1333 {
1334     g_autofree InternalSnapshotState *state = opaque;
1335     AioContext *aio_context;
1336 
1337     if (!state->bs) {
1338         return;
1339     }
1340 
1341     aio_context = bdrv_get_aio_context(state->bs);
1342     aio_context_acquire(aio_context);
1343 
1344     bdrv_drained_end(state->bs);
1345 
1346     aio_context_release(aio_context);
1347 }
1348 
1349 /* external snapshot private data */
1350 typedef struct ExternalSnapshotState {
1351     BlockDriverState *old_bs;
1352     BlockDriverState *new_bs;
1353     bool overlay_appended;
1354 } ExternalSnapshotState;
1355 
1356 static void external_snapshot_commit(void *opaque);
1357 static void external_snapshot_abort(void *opaque);
1358 static void external_snapshot_clean(void *opaque);
1359 TransactionActionDrv external_snapshot_drv = {
1360     .commit = external_snapshot_commit,
1361     .abort = external_snapshot_abort,
1362     .clean = external_snapshot_clean,
1363 };
1364 
1365 static void external_snapshot_action(TransactionAction *action,
1366                                      Transaction *tran, Error **errp)
1367 {
1368     int ret;
1369     int flags = 0;
1370     QDict *options = NULL;
1371     Error *local_err = NULL;
1372     /* Device and node name of the image to generate the snapshot from */
1373     const char *device;
1374     const char *node_name;
1375     /* Reference to the new image (for 'blockdev-snapshot') */
1376     const char *snapshot_ref;
1377     /* File name of the new image (for 'blockdev-snapshot-sync') */
1378     const char *new_image_file;
1379     ExternalSnapshotState *state = g_new0(ExternalSnapshotState, 1);
1380     AioContext *aio_context;
1381     uint64_t perm, shared;
1382 
1383     /* TODO We'll eventually have to take a writer lock in this function */
1384     GRAPH_RDLOCK_GUARD_MAINLOOP();
1385 
1386     tran_add(tran, &external_snapshot_drv, state);
1387 
1388     /* 'blockdev-snapshot' and 'blockdev-snapshot-sync' have similar
1389      * purpose but a different set of parameters */
1390     switch (action->type) {
1391     case TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT:
1392         {
1393             BlockdevSnapshot *s = action->u.blockdev_snapshot.data;
1394             device = s->node;
1395             node_name = s->node;
1396             new_image_file = NULL;
1397             snapshot_ref = s->overlay;
1398         }
1399         break;
1400     case TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC:
1401         {
1402             BlockdevSnapshotSync *s = action->u.blockdev_snapshot_sync.data;
1403             device = s->device;
1404             node_name = s->node_name;
1405             new_image_file = s->snapshot_file;
1406             snapshot_ref = NULL;
1407         }
1408         break;
1409     default:
1410         g_assert_not_reached();
1411     }
1412 
1413     /* start processing */
1414 
1415     state->old_bs = bdrv_lookup_bs(device, node_name, errp);
1416     if (!state->old_bs) {
1417         return;
1418     }
1419 
1420     aio_context = bdrv_get_aio_context(state->old_bs);
1421     aio_context_acquire(aio_context);
1422 
1423     /* Paired with .clean() */
1424     bdrv_drained_begin(state->old_bs);
1425 
1426     if (!bdrv_is_inserted(state->old_bs)) {
1427         error_setg(errp, QERR_DEVICE_HAS_NO_MEDIUM, device);
1428         goto out;
1429     }
1430 
1431     if (bdrv_op_is_blocked(state->old_bs,
1432                            BLOCK_OP_TYPE_EXTERNAL_SNAPSHOT, errp)) {
1433         goto out;
1434     }
1435 
1436     if (!bdrv_is_read_only(state->old_bs)) {
1437         if (bdrv_flush(state->old_bs)) {
1438             error_setg(errp, QERR_IO_ERROR);
1439             goto out;
1440         }
1441     }
1442 
1443     if (action->type == TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC) {
1444         BlockdevSnapshotSync *s = action->u.blockdev_snapshot_sync.data;
1445         const char *format = s->format ?: "qcow2";
1446         enum NewImageMode mode;
1447         const char *snapshot_node_name = s->snapshot_node_name;
1448 
1449         if (node_name && !snapshot_node_name) {
1450             error_setg(errp, "New overlay node-name missing");
1451             goto out;
1452         }
1453 
1454         if (snapshot_node_name &&
1455             bdrv_lookup_bs(snapshot_node_name, snapshot_node_name, NULL)) {
1456             error_setg(errp, "New overlay node-name already in use");
1457             goto out;
1458         }
1459 
1460         flags = state->old_bs->open_flags;
1461         flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_COPY_ON_READ);
1462         flags |= BDRV_O_NO_BACKING;
1463 
1464         /* create new image w/backing file */
1465         mode = s->has_mode ? s->mode : NEW_IMAGE_MODE_ABSOLUTE_PATHS;
1466         if (mode != NEW_IMAGE_MODE_EXISTING) {
1467             int64_t size = bdrv_getlength(state->old_bs);
1468             if (size < 0) {
1469                 error_setg_errno(errp, -size, "bdrv_getlength failed");
1470                 goto out;
1471             }
1472             bdrv_refresh_filename(state->old_bs);
1473 
1474             aio_context_release(aio_context);
1475             bdrv_img_create(new_image_file, format,
1476                             state->old_bs->filename,
1477                             state->old_bs->drv->format_name,
1478                             NULL, size, flags, false, &local_err);
1479             aio_context_acquire(aio_context);
1480 
1481             if (local_err) {
1482                 error_propagate(errp, local_err);
1483                 goto out;
1484             }
1485         }
1486 
1487         options = qdict_new();
1488         if (snapshot_node_name) {
1489             qdict_put_str(options, "node-name", snapshot_node_name);
1490         }
1491         qdict_put_str(options, "driver", format);
1492     }
1493     aio_context_release(aio_context);
1494 
1495     aio_context_acquire(qemu_get_aio_context());
1496     state->new_bs = bdrv_open(new_image_file, snapshot_ref, options, flags,
1497                               errp);
1498     aio_context_release(qemu_get_aio_context());
1499 
1500     /* We will manually add the backing_hd field to the bs later */
1501     if (!state->new_bs) {
1502         return;
1503     }
1504 
1505     aio_context_acquire(aio_context);
1506 
1507     /*
1508      * Allow attaching a backing file to an overlay that's already in use only
1509      * if the parents don't assume that they are already seeing a valid image.
1510      * (Specifically, allow it as a mirror target, which is write-only access.)
1511      */
1512     bdrv_get_cumulative_perm(state->new_bs, &perm, &shared);
1513     if (perm & BLK_PERM_CONSISTENT_READ) {
1514         error_setg(errp, "The overlay is already in use");
1515         goto out;
1516     }
1517 
1518     if (state->new_bs->drv->is_filter) {
1519         error_setg(errp, "Filters cannot be used as overlays");
1520         goto out;
1521     }
1522 
1523     if (bdrv_cow_child(state->new_bs)) {
1524         error_setg(errp, "The overlay already has a backing image");
1525         goto out;
1526     }
1527 
1528     if (!state->new_bs->drv->supports_backing) {
1529         error_setg(errp, "The overlay does not support backing images");
1530         goto out;
1531     }
1532 
1533     ret = bdrv_append(state->new_bs, state->old_bs, errp);
1534     if (ret < 0) {
1535         goto out;
1536     }
1537     state->overlay_appended = true;
1538 
1539 out:
1540     aio_context_release(aio_context);
1541 }
1542 
1543 static void external_snapshot_commit(void *opaque)
1544 {
1545     ExternalSnapshotState *state = opaque;
1546     AioContext *aio_context;
1547 
1548     aio_context = bdrv_get_aio_context(state->old_bs);
1549     aio_context_acquire(aio_context);
1550 
1551     /* We don't need (or want) to use the transactional
1552      * bdrv_reopen_multiple() across all the entries at once, because we
1553      * don't want to abort all of them if one of them fails the reopen */
1554     if (!qatomic_read(&state->old_bs->copy_on_read)) {
1555         bdrv_reopen_set_read_only(state->old_bs, true, NULL);
1556     }
1557 
1558     aio_context_release(aio_context);
1559 }
1560 
1561 static void external_snapshot_abort(void *opaque)
1562 {
1563     ExternalSnapshotState *state = opaque;
1564     if (state->new_bs) {
1565         if (state->overlay_appended) {
1566             AioContext *aio_context;
1567             AioContext *tmp_context;
1568             int ret;
1569 
1570             aio_context = bdrv_get_aio_context(state->old_bs);
1571             aio_context_acquire(aio_context);
1572 
1573             bdrv_ref(state->old_bs);   /* we can't let bdrv_set_backind_hd()
1574                                           close state->old_bs; we need it */
1575             bdrv_set_backing_hd(state->new_bs, NULL, &error_abort);
1576 
1577             /*
1578              * The call to bdrv_set_backing_hd() above returns state->old_bs to
1579              * the main AioContext. As we're still going to be using it, return
1580              * it to the AioContext it was before.
1581              */
1582             tmp_context = bdrv_get_aio_context(state->old_bs);
1583             if (aio_context != tmp_context) {
1584                 aio_context_release(aio_context);
1585                 aio_context_acquire(tmp_context);
1586 
1587                 ret = bdrv_try_change_aio_context(state->old_bs,
1588                                                   aio_context, NULL, NULL);
1589                 assert(ret == 0);
1590 
1591                 aio_context_release(tmp_context);
1592                 aio_context_acquire(aio_context);
1593             }
1594 
1595             bdrv_replace_node(state->new_bs, state->old_bs, &error_abort);
1596             bdrv_unref(state->old_bs); /* bdrv_replace_node() ref'ed old_bs */
1597 
1598             aio_context_release(aio_context);
1599         }
1600     }
1601 }
1602 
1603 static void external_snapshot_clean(void *opaque)
1604 {
1605     g_autofree ExternalSnapshotState *state = opaque;
1606     AioContext *aio_context;
1607 
1608     if (!state->old_bs) {
1609         return;
1610     }
1611 
1612     aio_context = bdrv_get_aio_context(state->old_bs);
1613     aio_context_acquire(aio_context);
1614 
1615     bdrv_drained_end(state->old_bs);
1616     bdrv_unref(state->new_bs);
1617 
1618     aio_context_release(aio_context);
1619 }
1620 
1621 typedef struct DriveBackupState {
1622     BlockDriverState *bs;
1623     BlockJob *job;
1624 } DriveBackupState;
1625 
1626 static BlockJob *do_backup_common(BackupCommon *backup,
1627                                   BlockDriverState *bs,
1628                                   BlockDriverState *target_bs,
1629                                   AioContext *aio_context,
1630                                   JobTxn *txn, Error **errp);
1631 
1632 static void drive_backup_commit(void *opaque);
1633 static void drive_backup_abort(void *opaque);
1634 static void drive_backup_clean(void *opaque);
1635 TransactionActionDrv drive_backup_drv = {
1636     .commit = drive_backup_commit,
1637     .abort = drive_backup_abort,
1638     .clean = drive_backup_clean,
1639 };
1640 
1641 static void drive_backup_action(DriveBackup *backup,
1642                                 JobTxn *block_job_txn,
1643                                 Transaction *tran, Error **errp)
1644 {
1645     DriveBackupState *state = g_new0(DriveBackupState, 1);
1646     BlockDriverState *bs;
1647     BlockDriverState *target_bs;
1648     BlockDriverState *source = NULL;
1649     AioContext *aio_context;
1650     AioContext *old_context;
1651     const char *format;
1652     QDict *options;
1653     Error *local_err = NULL;
1654     int flags;
1655     int64_t size;
1656     bool set_backing_hd = false;
1657     int ret;
1658 
1659     tran_add(tran, &drive_backup_drv, state);
1660 
1661     if (!backup->has_mode) {
1662         backup->mode = NEW_IMAGE_MODE_ABSOLUTE_PATHS;
1663     }
1664 
1665     bs = bdrv_lookup_bs(backup->device, backup->device, errp);
1666     if (!bs) {
1667         return;
1668     }
1669 
1670     if (!bs->drv) {
1671         error_setg(errp, "Device has no medium");
1672         return;
1673     }
1674 
1675     aio_context = bdrv_get_aio_context(bs);
1676     aio_context_acquire(aio_context);
1677 
1678     state->bs = bs;
1679     /* Paired with .clean() */
1680     bdrv_drained_begin(bs);
1681 
1682     format = backup->format;
1683     if (!format && backup->mode != NEW_IMAGE_MODE_EXISTING) {
1684         format = bs->drv->format_name;
1685     }
1686 
1687     /* Early check to avoid creating target */
1688     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_BACKUP_SOURCE, errp)) {
1689         goto out;
1690     }
1691 
1692     flags = bs->open_flags | BDRV_O_RDWR;
1693 
1694     /*
1695      * See if we have a backing HD we can use to create our new image
1696      * on top of.
1697      */
1698     if (backup->sync == MIRROR_SYNC_MODE_TOP) {
1699         /*
1700          * Backup will not replace the source by the target, so none
1701          * of the filters skipped here will be removed (in contrast to
1702          * mirror).  Therefore, we can skip all of them when looking
1703          * for the first COW relationship.
1704          */
1705         source = bdrv_cow_bs(bdrv_skip_filters(bs));
1706         if (!source) {
1707             backup->sync = MIRROR_SYNC_MODE_FULL;
1708         }
1709     }
1710     if (backup->sync == MIRROR_SYNC_MODE_NONE) {
1711         source = bs;
1712         flags |= BDRV_O_NO_BACKING;
1713         set_backing_hd = true;
1714     }
1715 
1716     size = bdrv_getlength(bs);
1717     if (size < 0) {
1718         error_setg_errno(errp, -size, "bdrv_getlength failed");
1719         goto out;
1720     }
1721 
1722     if (backup->mode != NEW_IMAGE_MODE_EXISTING) {
1723         assert(format);
1724         if (source) {
1725             /* Implicit filters should not appear in the filename */
1726             BlockDriverState *explicit_backing =
1727                 bdrv_skip_implicit_filters(source);
1728 
1729             bdrv_refresh_filename(explicit_backing);
1730             bdrv_img_create(backup->target, format,
1731                             explicit_backing->filename,
1732                             explicit_backing->drv->format_name, NULL,
1733                             size, flags, false, &local_err);
1734         } else {
1735             bdrv_img_create(backup->target, format, NULL, NULL, NULL,
1736                             size, flags, false, &local_err);
1737         }
1738     }
1739 
1740     if (local_err) {
1741         error_propagate(errp, local_err);
1742         goto out;
1743     }
1744 
1745     options = qdict_new();
1746     qdict_put_str(options, "discard", "unmap");
1747     qdict_put_str(options, "detect-zeroes", "unmap");
1748     if (format) {
1749         qdict_put_str(options, "driver", format);
1750     }
1751     aio_context_release(aio_context);
1752 
1753     aio_context_acquire(qemu_get_aio_context());
1754     target_bs = bdrv_open(backup->target, NULL, options, flags, errp);
1755     aio_context_release(qemu_get_aio_context());
1756 
1757     if (!target_bs) {
1758         return;
1759     }
1760 
1761     /* Honor bdrv_try_change_aio_context() context acquisition requirements. */
1762     old_context = bdrv_get_aio_context(target_bs);
1763     aio_context_acquire(old_context);
1764 
1765     ret = bdrv_try_change_aio_context(target_bs, aio_context, NULL, errp);
1766     if (ret < 0) {
1767         bdrv_unref(target_bs);
1768         aio_context_release(old_context);
1769         return;
1770     }
1771 
1772     aio_context_release(old_context);
1773     aio_context_acquire(aio_context);
1774 
1775     if (set_backing_hd) {
1776         if (bdrv_set_backing_hd(target_bs, source, errp) < 0) {
1777             goto unref;
1778         }
1779     }
1780 
1781     state->job = do_backup_common(qapi_DriveBackup_base(backup),
1782                                   bs, target_bs, aio_context,
1783                                   block_job_txn, errp);
1784 
1785 unref:
1786     bdrv_unref(target_bs);
1787 out:
1788     aio_context_release(aio_context);
1789 }
1790 
1791 static void drive_backup_commit(void *opaque)
1792 {
1793     DriveBackupState *state = opaque;
1794     AioContext *aio_context;
1795 
1796     aio_context = bdrv_get_aio_context(state->bs);
1797     aio_context_acquire(aio_context);
1798 
1799     assert(state->job);
1800     job_start(&state->job->job);
1801 
1802     aio_context_release(aio_context);
1803 }
1804 
1805 static void drive_backup_abort(void *opaque)
1806 {
1807     DriveBackupState *state = opaque;
1808 
1809     if (state->job) {
1810         job_cancel_sync(&state->job->job, true);
1811     }
1812 }
1813 
1814 static void drive_backup_clean(void *opaque)
1815 {
1816     g_autofree DriveBackupState *state = opaque;
1817     AioContext *aio_context;
1818 
1819     if (!state->bs) {
1820         return;
1821     }
1822 
1823     aio_context = bdrv_get_aio_context(state->bs);
1824     aio_context_acquire(aio_context);
1825 
1826     bdrv_drained_end(state->bs);
1827 
1828     aio_context_release(aio_context);
1829 }
1830 
1831 typedef struct BlockdevBackupState {
1832     BlockDriverState *bs;
1833     BlockJob *job;
1834 } BlockdevBackupState;
1835 
1836 static void blockdev_backup_commit(void *opaque);
1837 static void blockdev_backup_abort(void *opaque);
1838 static void blockdev_backup_clean(void *opaque);
1839 TransactionActionDrv blockdev_backup_drv = {
1840     .commit = blockdev_backup_commit,
1841     .abort = blockdev_backup_abort,
1842     .clean = blockdev_backup_clean,
1843 };
1844 
1845 static void blockdev_backup_action(BlockdevBackup *backup,
1846                                    JobTxn *block_job_txn,
1847                                    Transaction *tran, Error **errp)
1848 {
1849     BlockdevBackupState *state = g_new0(BlockdevBackupState, 1);
1850     BlockDriverState *bs;
1851     BlockDriverState *target_bs;
1852     AioContext *aio_context;
1853     AioContext *old_context;
1854     int ret;
1855 
1856     tran_add(tran, &blockdev_backup_drv, state);
1857 
1858     bs = bdrv_lookup_bs(backup->device, backup->device, errp);
1859     if (!bs) {
1860         return;
1861     }
1862 
1863     target_bs = bdrv_lookup_bs(backup->target, backup->target, errp);
1864     if (!target_bs) {
1865         return;
1866     }
1867 
1868     /* Honor bdrv_try_change_aio_context() context acquisition requirements. */
1869     aio_context = bdrv_get_aio_context(bs);
1870     old_context = bdrv_get_aio_context(target_bs);
1871     aio_context_acquire(old_context);
1872 
1873     ret = bdrv_try_change_aio_context(target_bs, aio_context, NULL, errp);
1874     if (ret < 0) {
1875         aio_context_release(old_context);
1876         return;
1877     }
1878 
1879     aio_context_release(old_context);
1880     aio_context_acquire(aio_context);
1881     state->bs = bs;
1882 
1883     /* Paired with .clean() */
1884     bdrv_drained_begin(state->bs);
1885 
1886     state->job = do_backup_common(qapi_BlockdevBackup_base(backup),
1887                                   bs, target_bs, aio_context,
1888                                   block_job_txn, errp);
1889 
1890     aio_context_release(aio_context);
1891 }
1892 
1893 static void blockdev_backup_commit(void *opaque)
1894 {
1895     BlockdevBackupState *state = opaque;
1896     AioContext *aio_context;
1897 
1898     aio_context = bdrv_get_aio_context(state->bs);
1899     aio_context_acquire(aio_context);
1900 
1901     assert(state->job);
1902     job_start(&state->job->job);
1903 
1904     aio_context_release(aio_context);
1905 }
1906 
1907 static void blockdev_backup_abort(void *opaque)
1908 {
1909     BlockdevBackupState *state = opaque;
1910 
1911     if (state->job) {
1912         job_cancel_sync(&state->job->job, true);
1913     }
1914 }
1915 
1916 static void blockdev_backup_clean(void *opaque)
1917 {
1918     g_autofree BlockdevBackupState *state = opaque;
1919     AioContext *aio_context;
1920 
1921     if (!state->bs) {
1922         return;
1923     }
1924 
1925     aio_context = bdrv_get_aio_context(state->bs);
1926     aio_context_acquire(aio_context);
1927 
1928     bdrv_drained_end(state->bs);
1929 
1930     aio_context_release(aio_context);
1931 }
1932 
1933 typedef struct BlockDirtyBitmapState {
1934     BdrvDirtyBitmap *bitmap;
1935     BlockDriverState *bs;
1936     HBitmap *backup;
1937     bool was_enabled;
1938 } BlockDirtyBitmapState;
1939 
1940 static void block_dirty_bitmap_add_abort(void *opaque);
1941 TransactionActionDrv block_dirty_bitmap_add_drv = {
1942     .abort = block_dirty_bitmap_add_abort,
1943     .clean = g_free,
1944 };
1945 
1946 static void block_dirty_bitmap_add_action(BlockDirtyBitmapAdd *action,
1947                                           Transaction *tran, Error **errp)
1948 {
1949     Error *local_err = NULL;
1950     BlockDirtyBitmapState *state = g_new0(BlockDirtyBitmapState, 1);
1951 
1952     tran_add(tran, &block_dirty_bitmap_add_drv, state);
1953 
1954     /* AIO context taken and released within qmp_block_dirty_bitmap_add */
1955     qmp_block_dirty_bitmap_add(action->node, action->name,
1956                                action->has_granularity, action->granularity,
1957                                action->has_persistent, action->persistent,
1958                                action->has_disabled, action->disabled,
1959                                &local_err);
1960 
1961     if (!local_err) {
1962         state->bitmap = block_dirty_bitmap_lookup(action->node, action->name,
1963                                                   NULL, &error_abort);
1964     } else {
1965         error_propagate(errp, local_err);
1966     }
1967 }
1968 
1969 static void block_dirty_bitmap_add_abort(void *opaque)
1970 {
1971     BlockDirtyBitmapState *state = opaque;
1972 
1973     if (state->bitmap) {
1974         bdrv_release_dirty_bitmap(state->bitmap);
1975     }
1976 }
1977 
1978 static void block_dirty_bitmap_restore(void *opaque);
1979 static void block_dirty_bitmap_free_backup(void *opaque);
1980 TransactionActionDrv block_dirty_bitmap_clear_drv = {
1981     .abort = block_dirty_bitmap_restore,
1982     .commit = block_dirty_bitmap_free_backup,
1983     .clean = g_free,
1984 };
1985 
1986 static void block_dirty_bitmap_clear_action(BlockDirtyBitmap *action,
1987                                             Transaction *tran, Error **errp)
1988 {
1989     BlockDirtyBitmapState *state = g_new0(BlockDirtyBitmapState, 1);
1990 
1991     tran_add(tran, &block_dirty_bitmap_clear_drv, state);
1992 
1993     state->bitmap = block_dirty_bitmap_lookup(action->node,
1994                                               action->name,
1995                                               &state->bs,
1996                                               errp);
1997     if (!state->bitmap) {
1998         return;
1999     }
2000 
2001     if (bdrv_dirty_bitmap_check(state->bitmap, BDRV_BITMAP_DEFAULT, errp)) {
2002         return;
2003     }
2004 
2005     bdrv_clear_dirty_bitmap(state->bitmap, &state->backup);
2006 }
2007 
2008 static void block_dirty_bitmap_restore(void *opaque)
2009 {
2010     BlockDirtyBitmapState *state = opaque;
2011 
2012     if (state->backup) {
2013         bdrv_restore_dirty_bitmap(state->bitmap, state->backup);
2014     }
2015 }
2016 
2017 static void block_dirty_bitmap_free_backup(void *opaque)
2018 {
2019     BlockDirtyBitmapState *state = opaque;
2020 
2021     hbitmap_free(state->backup);
2022 }
2023 
2024 static void block_dirty_bitmap_enable_abort(void *opaque);
2025 TransactionActionDrv block_dirty_bitmap_enable_drv = {
2026     .abort = block_dirty_bitmap_enable_abort,
2027     .clean = g_free,
2028 };
2029 
2030 static void block_dirty_bitmap_enable_action(BlockDirtyBitmap *action,
2031                                              Transaction *tran, Error **errp)
2032 {
2033     BlockDirtyBitmapState *state = g_new0(BlockDirtyBitmapState, 1);
2034 
2035     tran_add(tran, &block_dirty_bitmap_enable_drv, state);
2036 
2037     state->bitmap = block_dirty_bitmap_lookup(action->node,
2038                                               action->name,
2039                                               NULL,
2040                                               errp);
2041     if (!state->bitmap) {
2042         return;
2043     }
2044 
2045     if (bdrv_dirty_bitmap_check(state->bitmap, BDRV_BITMAP_ALLOW_RO, errp)) {
2046         return;
2047     }
2048 
2049     state->was_enabled = bdrv_dirty_bitmap_enabled(state->bitmap);
2050     bdrv_enable_dirty_bitmap(state->bitmap);
2051 }
2052 
2053 static void block_dirty_bitmap_enable_abort(void *opaque)
2054 {
2055     BlockDirtyBitmapState *state = opaque;
2056 
2057     if (!state->was_enabled) {
2058         bdrv_disable_dirty_bitmap(state->bitmap);
2059     }
2060 }
2061 
2062 static void block_dirty_bitmap_disable_abort(void *opaque);
2063 TransactionActionDrv block_dirty_bitmap_disable_drv = {
2064     .abort = block_dirty_bitmap_disable_abort,
2065     .clean = g_free,
2066 };
2067 
2068 static void block_dirty_bitmap_disable_action(BlockDirtyBitmap *action,
2069                                               Transaction *tran, Error **errp)
2070 {
2071     BlockDirtyBitmapState *state = g_new0(BlockDirtyBitmapState, 1);
2072 
2073     tran_add(tran, &block_dirty_bitmap_disable_drv, state);
2074 
2075     state->bitmap = block_dirty_bitmap_lookup(action->node,
2076                                               action->name,
2077                                               NULL,
2078                                               errp);
2079     if (!state->bitmap) {
2080         return;
2081     }
2082 
2083     if (bdrv_dirty_bitmap_check(state->bitmap, BDRV_BITMAP_ALLOW_RO, errp)) {
2084         return;
2085     }
2086 
2087     state->was_enabled = bdrv_dirty_bitmap_enabled(state->bitmap);
2088     bdrv_disable_dirty_bitmap(state->bitmap);
2089 }
2090 
2091 static void block_dirty_bitmap_disable_abort(void *opaque)
2092 {
2093     BlockDirtyBitmapState *state = opaque;
2094 
2095     if (state->was_enabled) {
2096         bdrv_enable_dirty_bitmap(state->bitmap);
2097     }
2098 }
2099 
2100 TransactionActionDrv block_dirty_bitmap_merge_drv = {
2101     .commit = block_dirty_bitmap_free_backup,
2102     .abort = block_dirty_bitmap_restore,
2103     .clean = g_free,
2104 };
2105 
2106 static void block_dirty_bitmap_merge_action(BlockDirtyBitmapMerge *action,
2107                                             Transaction *tran, Error **errp)
2108 {
2109     BlockDirtyBitmapState *state = g_new0(BlockDirtyBitmapState, 1);
2110 
2111     tran_add(tran, &block_dirty_bitmap_merge_drv, state);
2112 
2113     state->bitmap = block_dirty_bitmap_merge(action->node, action->target,
2114                                              action->bitmaps, &state->backup,
2115                                              errp);
2116 }
2117 
2118 static void block_dirty_bitmap_remove_commit(void *opaque);
2119 static void block_dirty_bitmap_remove_abort(void *opaque);
2120 TransactionActionDrv block_dirty_bitmap_remove_drv = {
2121     .commit = block_dirty_bitmap_remove_commit,
2122     .abort = block_dirty_bitmap_remove_abort,
2123     .clean = g_free,
2124 };
2125 
2126 static void block_dirty_bitmap_remove_action(BlockDirtyBitmap *action,
2127                                              Transaction *tran, Error **errp)
2128 {
2129     BlockDirtyBitmapState *state = g_new0(BlockDirtyBitmapState, 1);
2130 
2131     tran_add(tran, &block_dirty_bitmap_remove_drv, state);
2132 
2133 
2134     state->bitmap = block_dirty_bitmap_remove(action->node, action->name,
2135                                               false, &state->bs, errp);
2136     if (state->bitmap) {
2137         bdrv_dirty_bitmap_skip_store(state->bitmap, true);
2138         bdrv_dirty_bitmap_set_busy(state->bitmap, true);
2139     }
2140 }
2141 
2142 static void block_dirty_bitmap_remove_abort(void *opaque)
2143 {
2144     BlockDirtyBitmapState *state = opaque;
2145 
2146     if (state->bitmap) {
2147         bdrv_dirty_bitmap_skip_store(state->bitmap, false);
2148         bdrv_dirty_bitmap_set_busy(state->bitmap, false);
2149     }
2150 }
2151 
2152 static void block_dirty_bitmap_remove_commit(void *opaque)
2153 {
2154     BlockDirtyBitmapState *state = opaque;
2155 
2156     bdrv_dirty_bitmap_set_busy(state->bitmap, false);
2157     bdrv_release_dirty_bitmap(state->bitmap);
2158 }
2159 
2160 static void abort_commit(void *opaque);
2161 TransactionActionDrv abort_drv = {
2162     .commit = abort_commit,
2163 };
2164 
2165 static void abort_action(Transaction *tran, Error **errp)
2166 {
2167     tran_add(tran, &abort_drv, NULL);
2168     error_setg(errp, "Transaction aborted using Abort action");
2169 }
2170 
2171 static void abort_commit(void *opaque)
2172 {
2173     g_assert_not_reached(); /* this action never succeeds */
2174 }
2175 
2176 static void transaction_action(TransactionAction *act, JobTxn *block_job_txn,
2177                                Transaction *tran, Error **errp)
2178 {
2179     switch (act->type) {
2180     case TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT:
2181     case TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC:
2182         external_snapshot_action(act, tran, errp);
2183         return;
2184     case TRANSACTION_ACTION_KIND_DRIVE_BACKUP:
2185         drive_backup_action(act->u.drive_backup.data,
2186                             block_job_txn, tran, errp);
2187         return;
2188     case TRANSACTION_ACTION_KIND_BLOCKDEV_BACKUP:
2189         blockdev_backup_action(act->u.blockdev_backup.data,
2190                                block_job_txn, tran, errp);
2191         return;
2192     case TRANSACTION_ACTION_KIND_ABORT:
2193         abort_action(tran, errp);
2194         return;
2195     case TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC:
2196         internal_snapshot_action(act->u.blockdev_snapshot_internal_sync.data,
2197                                  tran, errp);
2198         return;
2199     case TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_ADD:
2200         block_dirty_bitmap_add_action(act->u.block_dirty_bitmap_add.data,
2201                                       tran, errp);
2202         return;
2203     case TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_CLEAR:
2204         block_dirty_bitmap_clear_action(act->u.block_dirty_bitmap_clear.data,
2205                                         tran, errp);
2206         return;
2207     case TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_ENABLE:
2208         block_dirty_bitmap_enable_action(act->u.block_dirty_bitmap_enable.data,
2209                                          tran, errp);
2210         return;
2211     case TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_DISABLE:
2212         block_dirty_bitmap_disable_action(
2213                 act->u.block_dirty_bitmap_disable.data, tran, errp);
2214         return;
2215     case TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_MERGE:
2216         block_dirty_bitmap_merge_action(act->u.block_dirty_bitmap_merge.data,
2217                                         tran, errp);
2218         return;
2219     case TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_REMOVE:
2220         block_dirty_bitmap_remove_action(act->u.block_dirty_bitmap_remove.data,
2221                                          tran, errp);
2222         return;
2223     /*
2224      * Where are transactions for MIRROR, COMMIT and STREAM?
2225      * Although these blockjobs use transaction callbacks like the backup job,
2226      * these jobs do not necessarily adhere to transaction semantics.
2227      * These jobs may not fully undo all of their actions on abort, nor do they
2228      * necessarily work in transactions with more than one job in them.
2229      */
2230     case TRANSACTION_ACTION_KIND__MAX:
2231     default:
2232         g_assert_not_reached();
2233     };
2234 }
2235 
2236 
2237 /*
2238  * 'Atomic' group operations.  The operations are performed as a set, and if
2239  * any fail then we roll back all operations in the group.
2240  *
2241  * Always run under BQL.
2242  */
2243 void qmp_transaction(TransactionActionList *actions,
2244                      struct TransactionProperties *properties,
2245                      Error **errp)
2246 {
2247     TransactionActionList *act;
2248     JobTxn *block_job_txn = NULL;
2249     Error *local_err = NULL;
2250     Transaction *tran;
2251     ActionCompletionMode comp_mode =
2252         properties ? properties->completion_mode :
2253         ACTION_COMPLETION_MODE_INDIVIDUAL;
2254 
2255     GLOBAL_STATE_CODE();
2256 
2257     /* Does this transaction get canceled as a group on failure?
2258      * If not, we don't really need to make a JobTxn.
2259      */
2260     if (comp_mode != ACTION_COMPLETION_MODE_INDIVIDUAL) {
2261         for (act = actions; act; act = act->next) {
2262             TransactionActionKind type = act->value->type;
2263 
2264             if (type != TRANSACTION_ACTION_KIND_BLOCKDEV_BACKUP &&
2265                 type != TRANSACTION_ACTION_KIND_DRIVE_BACKUP)
2266             {
2267                 error_setg(errp,
2268                            "Action '%s' does not support transaction property "
2269                            "completion-mode = %s",
2270                            TransactionActionKind_str(type),
2271                            ActionCompletionMode_str(comp_mode));
2272                 return;
2273             }
2274         }
2275 
2276         block_job_txn = job_txn_new();
2277     }
2278 
2279     /* drain all i/o before any operations */
2280     bdrv_drain_all();
2281 
2282     tran = tran_new();
2283 
2284     /* We don't do anything in this loop that commits us to the operations */
2285     for (act = actions; act; act = act->next) {
2286         transaction_action(act->value, block_job_txn, tran, &local_err);
2287         if (local_err) {
2288             error_propagate(errp, local_err);
2289             goto delete_and_fail;
2290         }
2291     }
2292 
2293     tran_commit(tran);
2294 
2295     /* success */
2296     goto exit;
2297 
2298 delete_and_fail:
2299     /* failure, and it is all-or-none; roll back all operations */
2300     tran_abort(tran);
2301 exit:
2302     job_txn_unref(block_job_txn);
2303 }
2304 
2305 BlockDirtyBitmapSha256 *qmp_x_debug_block_dirty_bitmap_sha256(const char *node,
2306                                                               const char *name,
2307                                                               Error **errp)
2308 {
2309     BdrvDirtyBitmap *bitmap;
2310     BlockDriverState *bs;
2311     BlockDirtyBitmapSha256 *ret = NULL;
2312     char *sha256;
2313 
2314     bitmap = block_dirty_bitmap_lookup(node, name, &bs, errp);
2315     if (!bitmap || !bs) {
2316         return NULL;
2317     }
2318 
2319     sha256 = bdrv_dirty_bitmap_sha256(bitmap, errp);
2320     if (sha256 == NULL) {
2321         return NULL;
2322     }
2323 
2324     ret = g_new(BlockDirtyBitmapSha256, 1);
2325     ret->sha256 = sha256;
2326 
2327     return ret;
2328 }
2329 
2330 void coroutine_fn qmp_block_resize(const char *device, const char *node_name,
2331                                    int64_t size, Error **errp)
2332 {
2333     Error *local_err = NULL;
2334     BlockBackend *blk;
2335     BlockDriverState *bs;
2336     AioContext *old_ctx;
2337 
2338     bs = bdrv_lookup_bs(device, node_name, &local_err);
2339     if (local_err) {
2340         error_propagate(errp, local_err);
2341         return;
2342     }
2343 
2344     if (size < 0) {
2345         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "size", "a >0 size");
2346         return;
2347     }
2348 
2349     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_RESIZE, NULL)) {
2350         error_setg(errp, QERR_DEVICE_IN_USE, device);
2351         return;
2352     }
2353 
2354     blk = blk_co_new_with_bs(bs, BLK_PERM_RESIZE, BLK_PERM_ALL, errp);
2355     if (!blk) {
2356         return;
2357     }
2358 
2359     bdrv_co_lock(bs);
2360     bdrv_drained_begin(bs);
2361     bdrv_co_unlock(bs);
2362 
2363     old_ctx = bdrv_co_enter(bs);
2364     blk_co_truncate(blk, size, false, PREALLOC_MODE_OFF, 0, errp);
2365     bdrv_co_leave(bs, old_ctx);
2366 
2367     bdrv_co_lock(bs);
2368     bdrv_drained_end(bs);
2369     blk_co_unref(blk);
2370     bdrv_co_unlock(bs);
2371 }
2372 
2373 void qmp_block_stream(const char *job_id, const char *device,
2374                       const char *base,
2375                       const char *base_node,
2376                       const char *backing_file,
2377                       const char *bottom,
2378                       bool has_speed, int64_t speed,
2379                       bool has_on_error, BlockdevOnError on_error,
2380                       const char *filter_node_name,
2381                       bool has_auto_finalize, bool auto_finalize,
2382                       bool has_auto_dismiss, bool auto_dismiss,
2383                       Error **errp)
2384 {
2385     BlockDriverState *bs, *iter, *iter_end;
2386     BlockDriverState *base_bs = NULL;
2387     BlockDriverState *bottom_bs = NULL;
2388     AioContext *aio_context;
2389     Error *local_err = NULL;
2390     int job_flags = JOB_DEFAULT;
2391 
2392     if (base && base_node) {
2393         error_setg(errp, "'base' and 'base-node' cannot be specified "
2394                    "at the same time");
2395         return;
2396     }
2397 
2398     if (base && bottom) {
2399         error_setg(errp, "'base' and 'bottom' cannot be specified "
2400                    "at the same time");
2401         return;
2402     }
2403 
2404     if (bottom && base_node) {
2405         error_setg(errp, "'bottom' and 'base-node' cannot be specified "
2406                    "at the same time");
2407         return;
2408     }
2409 
2410     if (!has_on_error) {
2411         on_error = BLOCKDEV_ON_ERROR_REPORT;
2412     }
2413 
2414     bs = bdrv_lookup_bs(device, device, errp);
2415     if (!bs) {
2416         return;
2417     }
2418 
2419     aio_context = bdrv_get_aio_context(bs);
2420     aio_context_acquire(aio_context);
2421 
2422     if (base) {
2423         base_bs = bdrv_find_backing_image(bs, base);
2424         if (base_bs == NULL) {
2425             error_setg(errp, "Can't find '%s' in the backing chain", base);
2426             goto out;
2427         }
2428         assert(bdrv_get_aio_context(base_bs) == aio_context);
2429     }
2430 
2431     if (base_node) {
2432         base_bs = bdrv_lookup_bs(NULL, base_node, errp);
2433         if (!base_bs) {
2434             goto out;
2435         }
2436         if (bs == base_bs || !bdrv_chain_contains(bs, base_bs)) {
2437             error_setg(errp, "Node '%s' is not a backing image of '%s'",
2438                        base_node, device);
2439             goto out;
2440         }
2441         assert(bdrv_get_aio_context(base_bs) == aio_context);
2442         bdrv_refresh_filename(base_bs);
2443     }
2444 
2445     if (bottom) {
2446         bottom_bs = bdrv_lookup_bs(NULL, bottom, errp);
2447         if (!bottom_bs) {
2448             goto out;
2449         }
2450         if (!bottom_bs->drv) {
2451             error_setg(errp, "Node '%s' is not open", bottom);
2452             goto out;
2453         }
2454         if (bottom_bs->drv->is_filter) {
2455             error_setg(errp, "Node '%s' is a filter, use a non-filter node "
2456                        "as 'bottom'", bottom);
2457             goto out;
2458         }
2459         if (!bdrv_chain_contains(bs, bottom_bs)) {
2460             error_setg(errp, "Node '%s' is not in a chain starting from '%s'",
2461                        bottom, device);
2462             goto out;
2463         }
2464         assert(bdrv_get_aio_context(bottom_bs) == aio_context);
2465     }
2466 
2467     /*
2468      * Check for op blockers in the whole chain between bs and base (or bottom)
2469      */
2470     iter_end = bottom ? bdrv_filter_or_cow_bs(bottom_bs) : base_bs;
2471     for (iter = bs; iter && iter != iter_end;
2472          iter = bdrv_filter_or_cow_bs(iter))
2473     {
2474         if (bdrv_op_is_blocked(iter, BLOCK_OP_TYPE_STREAM, errp)) {
2475             goto out;
2476         }
2477     }
2478 
2479     /* if we are streaming the entire chain, the result will have no backing
2480      * file, and specifying one is therefore an error */
2481     if (!base_bs && backing_file) {
2482         error_setg(errp, "backing file specified, but streaming the "
2483                          "entire chain");
2484         goto out;
2485     }
2486 
2487     if (has_auto_finalize && !auto_finalize) {
2488         job_flags |= JOB_MANUAL_FINALIZE;
2489     }
2490     if (has_auto_dismiss && !auto_dismiss) {
2491         job_flags |= JOB_MANUAL_DISMISS;
2492     }
2493 
2494     stream_start(job_id, bs, base_bs, backing_file,
2495                  bottom_bs, job_flags, has_speed ? speed : 0, on_error,
2496                  filter_node_name, &local_err);
2497     if (local_err) {
2498         error_propagate(errp, local_err);
2499         goto out;
2500     }
2501 
2502     trace_qmp_block_stream(bs);
2503 
2504 out:
2505     aio_context_release(aio_context);
2506 }
2507 
2508 void qmp_block_commit(const char *job_id, const char *device,
2509                       const char *base_node,
2510                       const char *base,
2511                       const char *top_node,
2512                       const char *top,
2513                       const char *backing_file,
2514                       bool has_speed, int64_t speed,
2515                       bool has_on_error, BlockdevOnError on_error,
2516                       const char *filter_node_name,
2517                       bool has_auto_finalize, bool auto_finalize,
2518                       bool has_auto_dismiss, bool auto_dismiss,
2519                       Error **errp)
2520 {
2521     BlockDriverState *bs;
2522     BlockDriverState *iter;
2523     BlockDriverState *base_bs, *top_bs;
2524     AioContext *aio_context;
2525     Error *local_err = NULL;
2526     int job_flags = JOB_DEFAULT;
2527     uint64_t top_perm, top_shared;
2528 
2529     /* TODO We'll eventually have to take a writer lock in this function */
2530     GRAPH_RDLOCK_GUARD_MAINLOOP();
2531 
2532     if (!has_speed) {
2533         speed = 0;
2534     }
2535     if (!has_on_error) {
2536         on_error = BLOCKDEV_ON_ERROR_REPORT;
2537     }
2538     if (has_auto_finalize && !auto_finalize) {
2539         job_flags |= JOB_MANUAL_FINALIZE;
2540     }
2541     if (has_auto_dismiss && !auto_dismiss) {
2542         job_flags |= JOB_MANUAL_DISMISS;
2543     }
2544 
2545     /* Important Note:
2546      *  libvirt relies on the DeviceNotFound error class in order to probe for
2547      *  live commit feature versions; for this to work, we must make sure to
2548      *  perform the device lookup before any generic errors that may occur in a
2549      *  scenario in which all optional arguments are omitted. */
2550     bs = qmp_get_root_bs(device, &local_err);
2551     if (!bs) {
2552         bs = bdrv_lookup_bs(device, device, NULL);
2553         if (!bs) {
2554             error_free(local_err);
2555             error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
2556                       "Device '%s' not found", device);
2557         } else {
2558             error_propagate(errp, local_err);
2559         }
2560         return;
2561     }
2562 
2563     aio_context = bdrv_get_aio_context(bs);
2564     aio_context_acquire(aio_context);
2565 
2566     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_COMMIT_SOURCE, errp)) {
2567         goto out;
2568     }
2569 
2570     /* default top_bs is the active layer */
2571     top_bs = bs;
2572 
2573     if (top_node && top) {
2574         error_setg(errp, "'top-node' and 'top' are mutually exclusive");
2575         goto out;
2576     } else if (top_node) {
2577         top_bs = bdrv_lookup_bs(NULL, top_node, errp);
2578         if (top_bs == NULL) {
2579             goto out;
2580         }
2581         if (!bdrv_chain_contains(bs, top_bs)) {
2582             error_setg(errp, "'%s' is not in this backing file chain",
2583                        top_node);
2584             goto out;
2585         }
2586     } else if (top) {
2587         /* This strcmp() is just a shortcut, there is no need to
2588          * refresh @bs's filename.  If it mismatches,
2589          * bdrv_find_backing_image() will do the refresh and may still
2590          * return @bs. */
2591         if (strcmp(bs->filename, top) != 0) {
2592             top_bs = bdrv_find_backing_image(bs, top);
2593         }
2594     }
2595 
2596     if (top_bs == NULL) {
2597         error_setg(errp, "Top image file %s not found", top ? top : "NULL");
2598         goto out;
2599     }
2600 
2601     assert(bdrv_get_aio_context(top_bs) == aio_context);
2602 
2603     if (base_node && base) {
2604         error_setg(errp, "'base-node' and 'base' are mutually exclusive");
2605         goto out;
2606     } else if (base_node) {
2607         base_bs = bdrv_lookup_bs(NULL, base_node, errp);
2608         if (base_bs == NULL) {
2609             goto out;
2610         }
2611         if (!bdrv_chain_contains(top_bs, base_bs)) {
2612             error_setg(errp, "'%s' is not in this backing file chain",
2613                        base_node);
2614             goto out;
2615         }
2616     } else if (base) {
2617         base_bs = bdrv_find_backing_image(top_bs, base);
2618         if (base_bs == NULL) {
2619             error_setg(errp, "Can't find '%s' in the backing chain", base);
2620             goto out;
2621         }
2622     } else {
2623         base_bs = bdrv_find_base(top_bs);
2624         if (base_bs == NULL) {
2625             error_setg(errp, "There is no backimg image");
2626             goto out;
2627         }
2628     }
2629 
2630     assert(bdrv_get_aio_context(base_bs) == aio_context);
2631 
2632     for (iter = top_bs; iter != bdrv_filter_or_cow_bs(base_bs);
2633          iter = bdrv_filter_or_cow_bs(iter))
2634     {
2635         if (bdrv_op_is_blocked(iter, BLOCK_OP_TYPE_COMMIT_TARGET, errp)) {
2636             goto out;
2637         }
2638     }
2639 
2640     /* Do not allow attempts to commit an image into itself */
2641     if (top_bs == base_bs) {
2642         error_setg(errp, "cannot commit an image into itself");
2643         goto out;
2644     }
2645 
2646     /*
2647      * Active commit is required if and only if someone has taken a
2648      * WRITE permission on the top node.  Historically, we have always
2649      * used active commit for top nodes, so continue that practice
2650      * lest we possibly break clients that rely on this behavior, e.g.
2651      * to later attach this node to a writing parent.
2652      * (Active commit is never really wrong.)
2653      */
2654     bdrv_get_cumulative_perm(top_bs, &top_perm, &top_shared);
2655     if (top_perm & BLK_PERM_WRITE ||
2656         bdrv_skip_filters(top_bs) == bdrv_skip_filters(bs))
2657     {
2658         if (backing_file) {
2659             if (bdrv_skip_filters(top_bs) == bdrv_skip_filters(bs)) {
2660                 error_setg(errp, "'backing-file' specified,"
2661                                  " but 'top' is the active layer");
2662             } else {
2663                 error_setg(errp, "'backing-file' specified, but 'top' has a "
2664                                  "writer on it");
2665             }
2666             goto out;
2667         }
2668         if (!job_id) {
2669             /*
2670              * Emulate here what block_job_create() does, because it
2671              * is possible that @bs != @top_bs (the block job should
2672              * be named after @bs, even if @top_bs is the actual
2673              * source)
2674              */
2675             job_id = bdrv_get_device_name(bs);
2676         }
2677         commit_active_start(job_id, top_bs, base_bs, job_flags, speed, on_error,
2678                             filter_node_name, NULL, NULL, false, &local_err);
2679     } else {
2680         BlockDriverState *overlay_bs = bdrv_find_overlay(bs, top_bs);
2681         if (bdrv_op_is_blocked(overlay_bs, BLOCK_OP_TYPE_COMMIT_TARGET, errp)) {
2682             goto out;
2683         }
2684         commit_start(job_id, bs, base_bs, top_bs, job_flags,
2685                      speed, on_error, backing_file,
2686                      filter_node_name, &local_err);
2687     }
2688     if (local_err != NULL) {
2689         error_propagate(errp, local_err);
2690         goto out;
2691     }
2692 
2693 out:
2694     aio_context_release(aio_context);
2695 }
2696 
2697 /* Common QMP interface for drive-backup and blockdev-backup */
2698 static BlockJob *do_backup_common(BackupCommon *backup,
2699                                   BlockDriverState *bs,
2700                                   BlockDriverState *target_bs,
2701                                   AioContext *aio_context,
2702                                   JobTxn *txn, Error **errp)
2703 {
2704     BlockJob *job = NULL;
2705     BdrvDirtyBitmap *bmap = NULL;
2706     BackupPerf perf = { .max_workers = 64 };
2707     int job_flags = JOB_DEFAULT;
2708 
2709     if (!backup->has_speed) {
2710         backup->speed = 0;
2711     }
2712     if (!backup->has_on_source_error) {
2713         backup->on_source_error = BLOCKDEV_ON_ERROR_REPORT;
2714     }
2715     if (!backup->has_on_target_error) {
2716         backup->on_target_error = BLOCKDEV_ON_ERROR_REPORT;
2717     }
2718     if (!backup->has_auto_finalize) {
2719         backup->auto_finalize = true;
2720     }
2721     if (!backup->has_auto_dismiss) {
2722         backup->auto_dismiss = true;
2723     }
2724     if (!backup->has_compress) {
2725         backup->compress = false;
2726     }
2727 
2728     if (backup->x_perf) {
2729         if (backup->x_perf->has_use_copy_range) {
2730             perf.use_copy_range = backup->x_perf->use_copy_range;
2731         }
2732         if (backup->x_perf->has_max_workers) {
2733             perf.max_workers = backup->x_perf->max_workers;
2734         }
2735         if (backup->x_perf->has_max_chunk) {
2736             perf.max_chunk = backup->x_perf->max_chunk;
2737         }
2738     }
2739 
2740     if ((backup->sync == MIRROR_SYNC_MODE_BITMAP) ||
2741         (backup->sync == MIRROR_SYNC_MODE_INCREMENTAL)) {
2742         /* done before desugaring 'incremental' to print the right message */
2743         if (!backup->bitmap) {
2744             error_setg(errp, "must provide a valid bitmap name for "
2745                        "'%s' sync mode", MirrorSyncMode_str(backup->sync));
2746             return NULL;
2747         }
2748     }
2749 
2750     if (backup->sync == MIRROR_SYNC_MODE_INCREMENTAL) {
2751         if (backup->has_bitmap_mode &&
2752             backup->bitmap_mode != BITMAP_SYNC_MODE_ON_SUCCESS) {
2753             error_setg(errp, "Bitmap sync mode must be '%s' "
2754                        "when using sync mode '%s'",
2755                        BitmapSyncMode_str(BITMAP_SYNC_MODE_ON_SUCCESS),
2756                        MirrorSyncMode_str(backup->sync));
2757             return NULL;
2758         }
2759         backup->has_bitmap_mode = true;
2760         backup->sync = MIRROR_SYNC_MODE_BITMAP;
2761         backup->bitmap_mode = BITMAP_SYNC_MODE_ON_SUCCESS;
2762     }
2763 
2764     if (backup->bitmap) {
2765         bmap = bdrv_find_dirty_bitmap(bs, backup->bitmap);
2766         if (!bmap) {
2767             error_setg(errp, "Bitmap '%s' could not be found", backup->bitmap);
2768             return NULL;
2769         }
2770         if (!backup->has_bitmap_mode) {
2771             error_setg(errp, "Bitmap sync mode must be given "
2772                        "when providing a bitmap");
2773             return NULL;
2774         }
2775         if (bdrv_dirty_bitmap_check(bmap, BDRV_BITMAP_ALLOW_RO, errp)) {
2776             return NULL;
2777         }
2778 
2779         /* This does not produce a useful bitmap artifact: */
2780         if (backup->sync == MIRROR_SYNC_MODE_NONE) {
2781             error_setg(errp, "sync mode '%s' does not produce meaningful bitmap"
2782                        " outputs", MirrorSyncMode_str(backup->sync));
2783             return NULL;
2784         }
2785 
2786         /* If the bitmap isn't used for input or output, this is useless: */
2787         if (backup->bitmap_mode == BITMAP_SYNC_MODE_NEVER &&
2788             backup->sync != MIRROR_SYNC_MODE_BITMAP) {
2789             error_setg(errp, "Bitmap sync mode '%s' has no meaningful effect"
2790                        " when combined with sync mode '%s'",
2791                        BitmapSyncMode_str(backup->bitmap_mode),
2792                        MirrorSyncMode_str(backup->sync));
2793             return NULL;
2794         }
2795     }
2796 
2797     if (!backup->bitmap && backup->has_bitmap_mode) {
2798         error_setg(errp, "Cannot specify bitmap sync mode without a bitmap");
2799         return NULL;
2800     }
2801 
2802     if (!backup->auto_finalize) {
2803         job_flags |= JOB_MANUAL_FINALIZE;
2804     }
2805     if (!backup->auto_dismiss) {
2806         job_flags |= JOB_MANUAL_DISMISS;
2807     }
2808 
2809     job = backup_job_create(backup->job_id, bs, target_bs, backup->speed,
2810                             backup->sync, bmap, backup->bitmap_mode,
2811                             backup->compress,
2812                             backup->filter_node_name,
2813                             &perf,
2814                             backup->on_source_error,
2815                             backup->on_target_error,
2816                             job_flags, NULL, NULL, txn, errp);
2817     return job;
2818 }
2819 
2820 void qmp_drive_backup(DriveBackup *backup, Error **errp)
2821 {
2822     TransactionAction action = {
2823         .type = TRANSACTION_ACTION_KIND_DRIVE_BACKUP,
2824         .u.drive_backup.data = backup,
2825     };
2826     blockdev_do_action(&action, errp);
2827 }
2828 
2829 BlockDeviceInfoList *qmp_query_named_block_nodes(bool has_flat,
2830                                                  bool flat,
2831                                                  Error **errp)
2832 {
2833     bool return_flat = has_flat && flat;
2834 
2835     return bdrv_named_nodes_list(return_flat, errp);
2836 }
2837 
2838 XDbgBlockGraph *qmp_x_debug_query_block_graph(Error **errp)
2839 {
2840     return bdrv_get_xdbg_block_graph(errp);
2841 }
2842 
2843 void qmp_blockdev_backup(BlockdevBackup *backup, Error **errp)
2844 {
2845     TransactionAction action = {
2846         .type = TRANSACTION_ACTION_KIND_BLOCKDEV_BACKUP,
2847         .u.blockdev_backup.data = backup,
2848     };
2849     blockdev_do_action(&action, errp);
2850 }
2851 
2852 /* Parameter check and block job starting for drive mirroring.
2853  * Caller should hold @device and @target's aio context (must be the same).
2854  **/
2855 static void blockdev_mirror_common(const char *job_id, BlockDriverState *bs,
2856                                    BlockDriverState *target,
2857                                    const char *replaces,
2858                                    enum MirrorSyncMode sync,
2859                                    BlockMirrorBackingMode backing_mode,
2860                                    bool zero_target,
2861                                    bool has_speed, int64_t speed,
2862                                    bool has_granularity, uint32_t granularity,
2863                                    bool has_buf_size, int64_t buf_size,
2864                                    bool has_on_source_error,
2865                                    BlockdevOnError on_source_error,
2866                                    bool has_on_target_error,
2867                                    BlockdevOnError on_target_error,
2868                                    bool has_unmap, bool unmap,
2869                                    const char *filter_node_name,
2870                                    bool has_copy_mode, MirrorCopyMode copy_mode,
2871                                    bool has_auto_finalize, bool auto_finalize,
2872                                    bool has_auto_dismiss, bool auto_dismiss,
2873                                    Error **errp)
2874 {
2875     BlockDriverState *unfiltered_bs;
2876     int job_flags = JOB_DEFAULT;
2877 
2878     GLOBAL_STATE_CODE();
2879     GRAPH_RDLOCK_GUARD_MAINLOOP();
2880 
2881     if (!has_speed) {
2882         speed = 0;
2883     }
2884     if (!has_on_source_error) {
2885         on_source_error = BLOCKDEV_ON_ERROR_REPORT;
2886     }
2887     if (!has_on_target_error) {
2888         on_target_error = BLOCKDEV_ON_ERROR_REPORT;
2889     }
2890     if (!has_granularity) {
2891         granularity = 0;
2892     }
2893     if (!has_buf_size) {
2894         buf_size = 0;
2895     }
2896     if (!has_unmap) {
2897         unmap = true;
2898     }
2899     if (!has_copy_mode) {
2900         copy_mode = MIRROR_COPY_MODE_BACKGROUND;
2901     }
2902     if (has_auto_finalize && !auto_finalize) {
2903         job_flags |= JOB_MANUAL_FINALIZE;
2904     }
2905     if (has_auto_dismiss && !auto_dismiss) {
2906         job_flags |= JOB_MANUAL_DISMISS;
2907     }
2908 
2909     if (granularity != 0 && (granularity < 512 || granularity > 1048576 * 64)) {
2910         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "granularity",
2911                    "a value in range [512B, 64MB]");
2912         return;
2913     }
2914     if (granularity & (granularity - 1)) {
2915         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "granularity",
2916                    "a power of 2");
2917         return;
2918     }
2919 
2920     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_MIRROR_SOURCE, errp)) {
2921         return;
2922     }
2923     if (bdrv_op_is_blocked(target, BLOCK_OP_TYPE_MIRROR_TARGET, errp)) {
2924         return;
2925     }
2926 
2927     if (!bdrv_backing_chain_next(bs) && sync == MIRROR_SYNC_MODE_TOP) {
2928         sync = MIRROR_SYNC_MODE_FULL;
2929     }
2930 
2931     if (!replaces) {
2932         /* We want to mirror from @bs, but keep implicit filters on top */
2933         unfiltered_bs = bdrv_skip_implicit_filters(bs);
2934         if (unfiltered_bs != bs) {
2935             replaces = unfiltered_bs->node_name;
2936         }
2937     }
2938 
2939     if (replaces) {
2940         BlockDriverState *to_replace_bs;
2941         AioContext *replace_aio_context;
2942         int64_t bs_size, replace_size;
2943 
2944         bs_size = bdrv_getlength(bs);
2945         if (bs_size < 0) {
2946             error_setg_errno(errp, -bs_size, "Failed to query device's size");
2947             return;
2948         }
2949 
2950         to_replace_bs = check_to_replace_node(bs, replaces, errp);
2951         if (!to_replace_bs) {
2952             return;
2953         }
2954 
2955         replace_aio_context = bdrv_get_aio_context(to_replace_bs);
2956         aio_context_acquire(replace_aio_context);
2957         replace_size = bdrv_getlength(to_replace_bs);
2958         aio_context_release(replace_aio_context);
2959 
2960         if (replace_size < 0) {
2961             error_setg_errno(errp, -replace_size,
2962                              "Failed to query the replacement node's size");
2963             return;
2964         }
2965         if (bs_size != replace_size) {
2966             error_setg(errp, "cannot replace image with a mirror image of "
2967                              "different size");
2968             return;
2969         }
2970     }
2971 
2972     /* pass the node name to replace to mirror start since it's loose coupling
2973      * and will allow to check whether the node still exist at mirror completion
2974      */
2975     mirror_start(job_id, bs, target,
2976                  replaces, job_flags,
2977                  speed, granularity, buf_size, sync, backing_mode, zero_target,
2978                  on_source_error, on_target_error, unmap, filter_node_name,
2979                  copy_mode, errp);
2980 }
2981 
2982 void qmp_drive_mirror(DriveMirror *arg, Error **errp)
2983 {
2984     BlockDriverState *bs;
2985     BlockDriverState *target_backing_bs, *target_bs;
2986     AioContext *aio_context;
2987     AioContext *old_context;
2988     BlockMirrorBackingMode backing_mode;
2989     Error *local_err = NULL;
2990     QDict *options = NULL;
2991     int flags;
2992     int64_t size;
2993     const char *format = arg->format;
2994     bool zero_target;
2995     int ret;
2996 
2997     bs = qmp_get_root_bs(arg->device, errp);
2998     if (!bs) {
2999         return;
3000     }
3001 
3002     /* Early check to avoid creating target */
3003     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_MIRROR_SOURCE, errp)) {
3004         return;
3005     }
3006 
3007     aio_context = bdrv_get_aio_context(bs);
3008     aio_context_acquire(aio_context);
3009 
3010     if (!arg->has_mode) {
3011         arg->mode = NEW_IMAGE_MODE_ABSOLUTE_PATHS;
3012     }
3013 
3014     if (!arg->format) {
3015         format = (arg->mode == NEW_IMAGE_MODE_EXISTING
3016                   ? NULL : bs->drv->format_name);
3017     }
3018 
3019     flags = bs->open_flags | BDRV_O_RDWR;
3020     target_backing_bs = bdrv_cow_bs(bdrv_skip_filters(bs));
3021     if (!target_backing_bs && arg->sync == MIRROR_SYNC_MODE_TOP) {
3022         arg->sync = MIRROR_SYNC_MODE_FULL;
3023     }
3024     if (arg->sync == MIRROR_SYNC_MODE_NONE) {
3025         target_backing_bs = bs;
3026     }
3027 
3028     size = bdrv_getlength(bs);
3029     if (size < 0) {
3030         error_setg_errno(errp, -size, "bdrv_getlength failed");
3031         goto out;
3032     }
3033 
3034     if (arg->replaces) {
3035         if (!arg->node_name) {
3036             error_setg(errp, "a node-name must be provided when replacing a"
3037                              " named node of the graph");
3038             goto out;
3039         }
3040     }
3041 
3042     if (arg->mode == NEW_IMAGE_MODE_ABSOLUTE_PATHS) {
3043         backing_mode = MIRROR_SOURCE_BACKING_CHAIN;
3044     } else {
3045         backing_mode = MIRROR_OPEN_BACKING_CHAIN;
3046     }
3047 
3048     /* Don't open backing image in create() */
3049     flags |= BDRV_O_NO_BACKING;
3050 
3051     if ((arg->sync == MIRROR_SYNC_MODE_FULL || !target_backing_bs)
3052         && arg->mode != NEW_IMAGE_MODE_EXISTING)
3053     {
3054         /* create new image w/o backing file */
3055         assert(format);
3056         bdrv_img_create(arg->target, format,
3057                         NULL, NULL, NULL, size, flags, false, &local_err);
3058     } else {
3059         /* Implicit filters should not appear in the filename */
3060         BlockDriverState *explicit_backing =
3061             bdrv_skip_implicit_filters(target_backing_bs);
3062 
3063         switch (arg->mode) {
3064         case NEW_IMAGE_MODE_EXISTING:
3065             break;
3066         case NEW_IMAGE_MODE_ABSOLUTE_PATHS:
3067             /* create new image with backing file */
3068             bdrv_refresh_filename(explicit_backing);
3069             bdrv_img_create(arg->target, format,
3070                             explicit_backing->filename,
3071                             explicit_backing->drv->format_name,
3072                             NULL, size, flags, false, &local_err);
3073             break;
3074         default:
3075             abort();
3076         }
3077     }
3078 
3079     if (local_err) {
3080         error_propagate(errp, local_err);
3081         goto out;
3082     }
3083 
3084     options = qdict_new();
3085     if (arg->node_name) {
3086         qdict_put_str(options, "node-name", arg->node_name);
3087     }
3088     if (format) {
3089         qdict_put_str(options, "driver", format);
3090     }
3091     aio_context_release(aio_context);
3092 
3093     /* Mirroring takes care of copy-on-write using the source's backing
3094      * file.
3095      */
3096     aio_context_acquire(qemu_get_aio_context());
3097     target_bs = bdrv_open(arg->target, NULL, options, flags, errp);
3098     aio_context_release(qemu_get_aio_context());
3099 
3100     if (!target_bs) {
3101         return;
3102     }
3103 
3104     zero_target = (arg->sync == MIRROR_SYNC_MODE_FULL &&
3105                    (arg->mode == NEW_IMAGE_MODE_EXISTING ||
3106                     !bdrv_has_zero_init(target_bs)));
3107 
3108 
3109     /* Honor bdrv_try_change_aio_context() context acquisition requirements. */
3110     old_context = bdrv_get_aio_context(target_bs);
3111     aio_context_acquire(old_context);
3112 
3113     ret = bdrv_try_change_aio_context(target_bs, aio_context, NULL, errp);
3114     if (ret < 0) {
3115         bdrv_unref(target_bs);
3116         aio_context_release(old_context);
3117         return;
3118     }
3119 
3120     aio_context_release(old_context);
3121     aio_context_acquire(aio_context);
3122 
3123     blockdev_mirror_common(arg->job_id, bs, target_bs,
3124                            arg->replaces, arg->sync,
3125                            backing_mode, zero_target,
3126                            arg->has_speed, arg->speed,
3127                            arg->has_granularity, arg->granularity,
3128                            arg->has_buf_size, arg->buf_size,
3129                            arg->has_on_source_error, arg->on_source_error,
3130                            arg->has_on_target_error, arg->on_target_error,
3131                            arg->has_unmap, arg->unmap,
3132                            NULL,
3133                            arg->has_copy_mode, arg->copy_mode,
3134                            arg->has_auto_finalize, arg->auto_finalize,
3135                            arg->has_auto_dismiss, arg->auto_dismiss,
3136                            errp);
3137     bdrv_unref(target_bs);
3138 out:
3139     aio_context_release(aio_context);
3140 }
3141 
3142 void qmp_blockdev_mirror(const char *job_id,
3143                          const char *device, const char *target,
3144                          const char *replaces,
3145                          MirrorSyncMode sync,
3146                          bool has_speed, int64_t speed,
3147                          bool has_granularity, uint32_t granularity,
3148                          bool has_buf_size, int64_t buf_size,
3149                          bool has_on_source_error,
3150                          BlockdevOnError on_source_error,
3151                          bool has_on_target_error,
3152                          BlockdevOnError on_target_error,
3153                          const char *filter_node_name,
3154                          bool has_copy_mode, MirrorCopyMode copy_mode,
3155                          bool has_auto_finalize, bool auto_finalize,
3156                          bool has_auto_dismiss, bool auto_dismiss,
3157                          Error **errp)
3158 {
3159     BlockDriverState *bs;
3160     BlockDriverState *target_bs;
3161     AioContext *aio_context;
3162     AioContext *old_context;
3163     BlockMirrorBackingMode backing_mode = MIRROR_LEAVE_BACKING_CHAIN;
3164     bool zero_target;
3165     int ret;
3166 
3167     bs = qmp_get_root_bs(device, errp);
3168     if (!bs) {
3169         return;
3170     }
3171 
3172     target_bs = bdrv_lookup_bs(target, target, errp);
3173     if (!target_bs) {
3174         return;
3175     }
3176 
3177     zero_target = (sync == MIRROR_SYNC_MODE_FULL);
3178 
3179     /* Honor bdrv_try_change_aio_context() context acquisition requirements. */
3180     old_context = bdrv_get_aio_context(target_bs);
3181     aio_context = bdrv_get_aio_context(bs);
3182     aio_context_acquire(old_context);
3183 
3184     ret = bdrv_try_change_aio_context(target_bs, aio_context, NULL, errp);
3185 
3186     aio_context_release(old_context);
3187     aio_context_acquire(aio_context);
3188 
3189     if (ret < 0) {
3190         goto out;
3191     }
3192 
3193     blockdev_mirror_common(job_id, bs, target_bs,
3194                            replaces, sync, backing_mode,
3195                            zero_target, has_speed, speed,
3196                            has_granularity, granularity,
3197                            has_buf_size, buf_size,
3198                            has_on_source_error, on_source_error,
3199                            has_on_target_error, on_target_error,
3200                            true, true, filter_node_name,
3201                            has_copy_mode, copy_mode,
3202                            has_auto_finalize, auto_finalize,
3203                            has_auto_dismiss, auto_dismiss,
3204                            errp);
3205 out:
3206     aio_context_release(aio_context);
3207 }
3208 
3209 /*
3210  * Get a block job using its ID. Called with job_mutex held.
3211  */
3212 static BlockJob *find_block_job_locked(const char *id, Error **errp)
3213 {
3214     BlockJob *job;
3215 
3216     assert(id != NULL);
3217 
3218     job = block_job_get_locked(id);
3219 
3220     if (!job) {
3221         error_set(errp, ERROR_CLASS_DEVICE_NOT_ACTIVE,
3222                   "Block job '%s' not found", id);
3223         return NULL;
3224     }
3225 
3226     return job;
3227 }
3228 
3229 void qmp_block_job_set_speed(const char *device, int64_t speed, Error **errp)
3230 {
3231     BlockJob *job;
3232 
3233     JOB_LOCK_GUARD();
3234     job = find_block_job_locked(device, errp);
3235 
3236     if (!job) {
3237         return;
3238     }
3239 
3240     block_job_set_speed_locked(job, speed, errp);
3241 }
3242 
3243 void qmp_block_job_cancel(const char *device,
3244                           bool has_force, bool force, Error **errp)
3245 {
3246     BlockJob *job;
3247 
3248     JOB_LOCK_GUARD();
3249     job = find_block_job_locked(device, errp);
3250 
3251     if (!job) {
3252         return;
3253     }
3254 
3255     if (!has_force) {
3256         force = false;
3257     }
3258 
3259     if (job_user_paused_locked(&job->job) && !force) {
3260         error_setg(errp, "The block job for device '%s' is currently paused",
3261                    device);
3262         return;
3263     }
3264 
3265     trace_qmp_block_job_cancel(job);
3266     job_user_cancel_locked(&job->job, force, errp);
3267 }
3268 
3269 void qmp_block_job_pause(const char *device, Error **errp)
3270 {
3271     BlockJob *job;
3272 
3273     JOB_LOCK_GUARD();
3274     job = find_block_job_locked(device, errp);
3275 
3276     if (!job) {
3277         return;
3278     }
3279 
3280     trace_qmp_block_job_pause(job);
3281     job_user_pause_locked(&job->job, errp);
3282 }
3283 
3284 void qmp_block_job_resume(const char *device, Error **errp)
3285 {
3286     BlockJob *job;
3287 
3288     JOB_LOCK_GUARD();
3289     job = find_block_job_locked(device, errp);
3290 
3291     if (!job) {
3292         return;
3293     }
3294 
3295     trace_qmp_block_job_resume(job);
3296     job_user_resume_locked(&job->job, errp);
3297 }
3298 
3299 void qmp_block_job_complete(const char *device, Error **errp)
3300 {
3301     BlockJob *job;
3302 
3303     JOB_LOCK_GUARD();
3304     job = find_block_job_locked(device, errp);
3305 
3306     if (!job) {
3307         return;
3308     }
3309 
3310     trace_qmp_block_job_complete(job);
3311     job_complete_locked(&job->job, errp);
3312 }
3313 
3314 void qmp_block_job_finalize(const char *id, Error **errp)
3315 {
3316     BlockJob *job;
3317 
3318     JOB_LOCK_GUARD();
3319     job = find_block_job_locked(id, errp);
3320 
3321     if (!job) {
3322         return;
3323     }
3324 
3325     trace_qmp_block_job_finalize(job);
3326     job_ref_locked(&job->job);
3327     job_finalize_locked(&job->job, errp);
3328 
3329     job_unref_locked(&job->job);
3330 }
3331 
3332 void qmp_block_job_dismiss(const char *id, Error **errp)
3333 {
3334     BlockJob *bjob;
3335     Job *job;
3336 
3337     JOB_LOCK_GUARD();
3338     bjob = find_block_job_locked(id, errp);
3339 
3340     if (!bjob) {
3341         return;
3342     }
3343 
3344     trace_qmp_block_job_dismiss(bjob);
3345     job = &bjob->job;
3346     job_dismiss_locked(&job, errp);
3347 }
3348 
3349 void qmp_change_backing_file(const char *device,
3350                              const char *image_node_name,
3351                              const char *backing_file,
3352                              Error **errp)
3353 {
3354     BlockDriverState *bs = NULL;
3355     AioContext *aio_context;
3356     BlockDriverState *image_bs = NULL;
3357     Error *local_err = NULL;
3358     bool ro;
3359     int ret;
3360 
3361     bs = qmp_get_root_bs(device, errp);
3362     if (!bs) {
3363         return;
3364     }
3365 
3366     aio_context = bdrv_get_aio_context(bs);
3367     aio_context_acquire(aio_context);
3368 
3369     image_bs = bdrv_lookup_bs(NULL, image_node_name, &local_err);
3370     if (local_err) {
3371         error_propagate(errp, local_err);
3372         goto out;
3373     }
3374 
3375     if (!image_bs) {
3376         error_setg(errp, "image file not found");
3377         goto out;
3378     }
3379 
3380     if (bdrv_find_base(image_bs) == image_bs) {
3381         error_setg(errp, "not allowing backing file change on an image "
3382                          "without a backing file");
3383         goto out;
3384     }
3385 
3386     /* even though we are not necessarily operating on bs, we need it to
3387      * determine if block ops are currently prohibited on the chain */
3388     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_CHANGE, errp)) {
3389         goto out;
3390     }
3391 
3392     /* final sanity check */
3393     if (!bdrv_chain_contains(bs, image_bs)) {
3394         error_setg(errp, "'%s' and image file are not in the same chain",
3395                    device);
3396         goto out;
3397     }
3398 
3399     /* if not r/w, reopen to make r/w */
3400     ro = bdrv_is_read_only(image_bs);
3401 
3402     if (ro) {
3403         if (bdrv_reopen_set_read_only(image_bs, false, errp) != 0) {
3404             goto out;
3405         }
3406     }
3407 
3408     ret = bdrv_change_backing_file(image_bs, backing_file,
3409                                    image_bs->drv ? image_bs->drv->format_name : "",
3410                                    false);
3411 
3412     if (ret < 0) {
3413         error_setg_errno(errp, -ret, "Could not change backing file to '%s'",
3414                          backing_file);
3415         /* don't exit here, so we can try to restore open flags if
3416          * appropriate */
3417     }
3418 
3419     if (ro) {
3420         bdrv_reopen_set_read_only(image_bs, true, errp);
3421     }
3422 
3423 out:
3424     aio_context_release(aio_context);
3425 }
3426 
3427 void qmp_blockdev_add(BlockdevOptions *options, Error **errp)
3428 {
3429     BlockDriverState *bs;
3430     QObject *obj;
3431     Visitor *v = qobject_output_visitor_new(&obj);
3432     QDict *qdict;
3433 
3434     visit_type_BlockdevOptions(v, NULL, &options, &error_abort);
3435     visit_complete(v, &obj);
3436     qdict = qobject_to(QDict, obj);
3437 
3438     qdict_flatten(qdict);
3439 
3440     if (!qdict_get_try_str(qdict, "node-name")) {
3441         error_setg(errp, "'node-name' must be specified for the root node");
3442         goto fail;
3443     }
3444 
3445     bs = bds_tree_init(qdict, errp);
3446     if (!bs) {
3447         goto fail;
3448     }
3449 
3450     bdrv_set_monitor_owned(bs);
3451 
3452 fail:
3453     visit_free(v);
3454 }
3455 
3456 void qmp_blockdev_reopen(BlockdevOptionsList *reopen_list, Error **errp)
3457 {
3458     BlockReopenQueue *queue = NULL;
3459 
3460     /* Add each one of the BDS that we want to reopen to the queue */
3461     for (; reopen_list != NULL; reopen_list = reopen_list->next) {
3462         BlockdevOptions *options = reopen_list->value;
3463         BlockDriverState *bs;
3464         AioContext *ctx;
3465         QObject *obj;
3466         Visitor *v;
3467         QDict *qdict;
3468 
3469         /* Check for the selected node name */
3470         if (!options->node_name) {
3471             error_setg(errp, "node-name not specified");
3472             goto fail;
3473         }
3474 
3475         bs = bdrv_find_node(options->node_name);
3476         if (!bs) {
3477             error_setg(errp, "Failed to find node with node-name='%s'",
3478                        options->node_name);
3479             goto fail;
3480         }
3481 
3482         /* Put all options in a QDict and flatten it */
3483         v = qobject_output_visitor_new(&obj);
3484         visit_type_BlockdevOptions(v, NULL, &options, &error_abort);
3485         visit_complete(v, &obj);
3486         visit_free(v);
3487 
3488         qdict = qobject_to(QDict, obj);
3489 
3490         qdict_flatten(qdict);
3491 
3492         ctx = bdrv_get_aio_context(bs);
3493         aio_context_acquire(ctx);
3494 
3495         queue = bdrv_reopen_queue(queue, bs, qdict, false);
3496 
3497         aio_context_release(ctx);
3498     }
3499 
3500     /* Perform the reopen operation */
3501     bdrv_reopen_multiple(queue, errp);
3502     queue = NULL;
3503 
3504 fail:
3505     bdrv_reopen_queue_free(queue);
3506 }
3507 
3508 void qmp_blockdev_del(const char *node_name, Error **errp)
3509 {
3510     AioContext *aio_context;
3511     BlockDriverState *bs;
3512 
3513     GLOBAL_STATE_CODE();
3514     GRAPH_RDLOCK_GUARD_MAINLOOP();
3515 
3516     bs = bdrv_find_node(node_name);
3517     if (!bs) {
3518         error_setg(errp, "Failed to find node with node-name='%s'", node_name);
3519         return;
3520     }
3521     if (bdrv_has_blk(bs)) {
3522         error_setg(errp, "Node %s is in use", node_name);
3523         return;
3524     }
3525     aio_context = bdrv_get_aio_context(bs);
3526     aio_context_acquire(aio_context);
3527 
3528     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_DRIVE_DEL, errp)) {
3529         goto out;
3530     }
3531 
3532     if (!QTAILQ_IN_USE(bs, monitor_list)) {
3533         error_setg(errp, "Node %s is not owned by the monitor",
3534                    bs->node_name);
3535         goto out;
3536     }
3537 
3538     if (bs->refcnt > 1) {
3539         error_setg(errp, "Block device %s is in use",
3540                    bdrv_get_device_or_node_name(bs));
3541         goto out;
3542     }
3543 
3544     QTAILQ_REMOVE(&monitor_bdrv_states, bs, monitor_list);
3545     bdrv_unref(bs);
3546 
3547 out:
3548     aio_context_release(aio_context);
3549 }
3550 
3551 static BdrvChild * GRAPH_RDLOCK
3552 bdrv_find_child(BlockDriverState *parent_bs, const char *child_name)
3553 {
3554     BdrvChild *child;
3555 
3556     QLIST_FOREACH(child, &parent_bs->children, next) {
3557         if (strcmp(child->name, child_name) == 0) {
3558             return child;
3559         }
3560     }
3561 
3562     return NULL;
3563 }
3564 
3565 void qmp_x_blockdev_change(const char *parent, const char *child,
3566                            const char *node, Error **errp)
3567 {
3568     BlockDriverState *parent_bs, *new_bs = NULL;
3569     BdrvChild *p_child;
3570 
3571     bdrv_graph_wrlock(NULL);
3572 
3573     parent_bs = bdrv_lookup_bs(parent, parent, errp);
3574     if (!parent_bs) {
3575         goto out;
3576     }
3577 
3578     if (!child == !node) {
3579         if (child) {
3580             error_setg(errp, "The parameters child and node are in conflict");
3581         } else {
3582             error_setg(errp, "Either child or node must be specified");
3583         }
3584         goto out;
3585     }
3586 
3587     if (child) {
3588         p_child = bdrv_find_child(parent_bs, child);
3589         if (!p_child) {
3590             error_setg(errp, "Node '%s' does not have child '%s'",
3591                        parent, child);
3592             goto out;
3593         }
3594         bdrv_del_child(parent_bs, p_child, errp);
3595     }
3596 
3597     if (node) {
3598         new_bs = bdrv_find_node(node);
3599         if (!new_bs) {
3600             error_setg(errp, "Node '%s' not found", node);
3601             goto out;
3602         }
3603         bdrv_add_child(parent_bs, new_bs, errp);
3604     }
3605 
3606 out:
3607     bdrv_graph_wrunlock();
3608 }
3609 
3610 BlockJobInfoList *qmp_query_block_jobs(Error **errp)
3611 {
3612     BlockJobInfoList *head = NULL, **tail = &head;
3613     BlockJob *job;
3614 
3615     JOB_LOCK_GUARD();
3616 
3617     for (job = block_job_next_locked(NULL); job;
3618          job = block_job_next_locked(job)) {
3619         BlockJobInfo *value;
3620 
3621         if (block_job_is_internal(job)) {
3622             continue;
3623         }
3624         value = block_job_query_locked(job, errp);
3625         if (!value) {
3626             qapi_free_BlockJobInfoList(head);
3627             return NULL;
3628         }
3629         QAPI_LIST_APPEND(tail, value);
3630     }
3631 
3632     return head;
3633 }
3634 
3635 void qmp_x_blockdev_set_iothread(const char *node_name, StrOrNull *iothread,
3636                                  bool has_force, bool force, Error **errp)
3637 {
3638     AioContext *old_context;
3639     AioContext *new_context;
3640     BlockDriverState *bs;
3641 
3642     GRAPH_RDLOCK_GUARD_MAINLOOP();
3643 
3644     bs = bdrv_find_node(node_name);
3645     if (!bs) {
3646         error_setg(errp, "Failed to find node with node-name='%s'", node_name);
3647         return;
3648     }
3649 
3650     /* Protects against accidents. */
3651     if (!(has_force && force) && bdrv_has_blk(bs)) {
3652         error_setg(errp, "Node %s is associated with a BlockBackend and could "
3653                          "be in use (use force=true to override this check)",
3654                          node_name);
3655         return;
3656     }
3657 
3658     if (iothread->type == QTYPE_QSTRING) {
3659         IOThread *obj = iothread_by_id(iothread->u.s);
3660         if (!obj) {
3661             error_setg(errp, "Cannot find iothread %s", iothread->u.s);
3662             return;
3663         }
3664 
3665         new_context = iothread_get_aio_context(obj);
3666     } else {
3667         new_context = qemu_get_aio_context();
3668     }
3669 
3670     old_context = bdrv_get_aio_context(bs);
3671     aio_context_acquire(old_context);
3672 
3673     bdrv_try_change_aio_context(bs, new_context, NULL, errp);
3674 
3675     aio_context_release(old_context);
3676 }
3677 
3678 QemuOptsList qemu_common_drive_opts = {
3679     .name = "drive",
3680     .head = QTAILQ_HEAD_INITIALIZER(qemu_common_drive_opts.head),
3681     .desc = {
3682         {
3683             .name = "snapshot",
3684             .type = QEMU_OPT_BOOL,
3685             .help = "enable/disable snapshot mode",
3686         },{
3687             .name = "aio",
3688             .type = QEMU_OPT_STRING,
3689             .help = "host AIO implementation (threads, native, io_uring)",
3690         },{
3691             .name = BDRV_OPT_CACHE_WB,
3692             .type = QEMU_OPT_BOOL,
3693             .help = "Enable writeback mode",
3694         },{
3695             .name = "format",
3696             .type = QEMU_OPT_STRING,
3697             .help = "disk format (raw, qcow2, ...)",
3698         },{
3699             .name = "rerror",
3700             .type = QEMU_OPT_STRING,
3701             .help = "read error action",
3702         },{
3703             .name = "werror",
3704             .type = QEMU_OPT_STRING,
3705             .help = "write error action",
3706         },{
3707             .name = BDRV_OPT_READ_ONLY,
3708             .type = QEMU_OPT_BOOL,
3709             .help = "open drive file as read-only",
3710         },
3711 
3712         THROTTLE_OPTS,
3713 
3714         {
3715             .name = "throttling.group",
3716             .type = QEMU_OPT_STRING,
3717             .help = "name of the block throttling group",
3718         },{
3719             .name = "copy-on-read",
3720             .type = QEMU_OPT_BOOL,
3721             .help = "copy read data from backing file into image file",
3722         },{
3723             .name = "detect-zeroes",
3724             .type = QEMU_OPT_STRING,
3725             .help = "try to optimize zero writes (off, on, unmap)",
3726         },{
3727             .name = "stats-account-invalid",
3728             .type = QEMU_OPT_BOOL,
3729             .help = "whether to account for invalid I/O operations "
3730                     "in the statistics",
3731         },{
3732             .name = "stats-account-failed",
3733             .type = QEMU_OPT_BOOL,
3734             .help = "whether to account for failed I/O operations "
3735                     "in the statistics",
3736         },
3737         { /* end of list */ }
3738     },
3739 };
3740 
3741 QemuOptsList qemu_drive_opts = {
3742     .name = "drive",
3743     .head = QTAILQ_HEAD_INITIALIZER(qemu_drive_opts.head),
3744     .desc = {
3745         /*
3746          * no elements => accept any params
3747          * validation will happen later
3748          */
3749         { /* end of list */ }
3750     },
3751 };
3752