xref: /openbmc/qemu/blockdev.c (revision d0bf492f)
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     bs = bdrv_lookup_bs(name, name, errp);
1045     if (bs == NULL) {
1046         return NULL;
1047     }
1048 
1049     if (!bdrv_is_root_node(bs)) {
1050         error_setg(errp, "Need a root block node");
1051         return NULL;
1052     }
1053 
1054     aio_context = bdrv_get_aio_context(bs);
1055     aio_context_acquire(aio_context);
1056 
1057     if (!bdrv_is_inserted(bs)) {
1058         error_setg(errp, "Device has no medium");
1059         bs = NULL;
1060     }
1061 
1062     aio_context_release(aio_context);
1063 
1064     return bs;
1065 }
1066 
1067 static void blockdev_do_action(TransactionAction *action, Error **errp)
1068 {
1069     TransactionActionList list;
1070 
1071     list.value = action;
1072     list.next = NULL;
1073     qmp_transaction(&list, NULL, errp);
1074 }
1075 
1076 void qmp_blockdev_snapshot_sync(const char *device, const char *node_name,
1077                                 const char *snapshot_file,
1078                                 const char *snapshot_node_name,
1079                                 const char *format,
1080                                 bool has_mode, NewImageMode mode, Error **errp)
1081 {
1082     BlockdevSnapshotSync snapshot = {
1083         .device = (char *) device,
1084         .node_name = (char *) node_name,
1085         .snapshot_file = (char *) snapshot_file,
1086         .snapshot_node_name = (char *) snapshot_node_name,
1087         .format = (char *) format,
1088         .has_mode = has_mode,
1089         .mode = mode,
1090     };
1091     TransactionAction action = {
1092         .type = TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC,
1093         .u.blockdev_snapshot_sync.data = &snapshot,
1094     };
1095     blockdev_do_action(&action, errp);
1096 }
1097 
1098 void qmp_blockdev_snapshot(const char *node, const char *overlay,
1099                            Error **errp)
1100 {
1101     BlockdevSnapshot snapshot_data = {
1102         .node = (char *) node,
1103         .overlay = (char *) overlay
1104     };
1105     TransactionAction action = {
1106         .type = TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT,
1107         .u.blockdev_snapshot.data = &snapshot_data,
1108     };
1109     blockdev_do_action(&action, errp);
1110 }
1111 
1112 void qmp_blockdev_snapshot_internal_sync(const char *device,
1113                                          const char *name,
1114                                          Error **errp)
1115 {
1116     BlockdevSnapshotInternal snapshot = {
1117         .device = (char *) device,
1118         .name = (char *) name
1119     };
1120     TransactionAction action = {
1121         .type = TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC,
1122         .u.blockdev_snapshot_internal_sync.data = &snapshot,
1123     };
1124     blockdev_do_action(&action, errp);
1125 }
1126 
1127 SnapshotInfo *qmp_blockdev_snapshot_delete_internal_sync(const char *device,
1128                                                          const char *id,
1129                                                          const char *name,
1130                                                          Error **errp)
1131 {
1132     BlockDriverState *bs;
1133     AioContext *aio_context;
1134     QEMUSnapshotInfo sn;
1135     Error *local_err = NULL;
1136     SnapshotInfo *info = NULL;
1137     int ret;
1138 
1139     bs = qmp_get_root_bs(device, errp);
1140     if (!bs) {
1141         return NULL;
1142     }
1143     aio_context = bdrv_get_aio_context(bs);
1144     aio_context_acquire(aio_context);
1145 
1146     if (!id && !name) {
1147         error_setg(errp, "Name or id must be provided");
1148         goto out_aio_context;
1149     }
1150 
1151     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_INTERNAL_SNAPSHOT_DELETE, errp)) {
1152         goto out_aio_context;
1153     }
1154 
1155     ret = bdrv_snapshot_find_by_id_and_name(bs, id, name, &sn, &local_err);
1156     if (local_err) {
1157         error_propagate(errp, local_err);
1158         goto out_aio_context;
1159     }
1160     if (!ret) {
1161         error_setg(errp,
1162                    "Snapshot with id '%s' and name '%s' does not exist on "
1163                    "device '%s'",
1164                    STR_OR_NULL(id), STR_OR_NULL(name), device);
1165         goto out_aio_context;
1166     }
1167 
1168     bdrv_snapshot_delete(bs, id, name, &local_err);
1169     if (local_err) {
1170         error_propagate(errp, local_err);
1171         goto out_aio_context;
1172     }
1173 
1174     aio_context_release(aio_context);
1175 
1176     info = g_new0(SnapshotInfo, 1);
1177     info->id = g_strdup(sn.id_str);
1178     info->name = g_strdup(sn.name);
1179     info->date_nsec = sn.date_nsec;
1180     info->date_sec = sn.date_sec;
1181     info->vm_state_size = sn.vm_state_size;
1182     info->vm_clock_nsec = sn.vm_clock_nsec % 1000000000;
1183     info->vm_clock_sec = sn.vm_clock_nsec / 1000000000;
1184     if (sn.icount != -1ULL) {
1185         info->icount = sn.icount;
1186         info->has_icount = true;
1187     }
1188 
1189     return info;
1190 
1191 out_aio_context:
1192     aio_context_release(aio_context);
1193     return NULL;
1194 }
1195 
1196 /* internal snapshot private data */
1197 typedef struct InternalSnapshotState {
1198     BlockDriverState *bs;
1199     QEMUSnapshotInfo sn;
1200     bool created;
1201 } InternalSnapshotState;
1202 
1203 static void internal_snapshot_abort(void *opaque);
1204 static void internal_snapshot_clean(void *opaque);
1205 TransactionActionDrv internal_snapshot_drv = {
1206     .abort = internal_snapshot_abort,
1207     .clean = internal_snapshot_clean,
1208 };
1209 
1210 static void internal_snapshot_action(BlockdevSnapshotInternal *internal,
1211                                      Transaction *tran, Error **errp)
1212 {
1213     Error *local_err = NULL;
1214     const char *device;
1215     const char *name;
1216     BlockDriverState *bs;
1217     QEMUSnapshotInfo old_sn, *sn;
1218     bool ret;
1219     int64_t rt;
1220     InternalSnapshotState *state = g_new0(InternalSnapshotState, 1);
1221     AioContext *aio_context;
1222     int ret1;
1223 
1224     tran_add(tran, &internal_snapshot_drv, state);
1225 
1226     device = internal->device;
1227     name = internal->name;
1228 
1229     bs = qmp_get_root_bs(device, errp);
1230     if (!bs) {
1231         return;
1232     }
1233 
1234     aio_context = bdrv_get_aio_context(bs);
1235     aio_context_acquire(aio_context);
1236 
1237     state->bs = bs;
1238 
1239     /* Paired with .clean() */
1240     bdrv_drained_begin(bs);
1241 
1242     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_INTERNAL_SNAPSHOT, errp)) {
1243         goto out;
1244     }
1245 
1246     if (bdrv_is_read_only(bs)) {
1247         error_setg(errp, "Device '%s' is read only", device);
1248         goto out;
1249     }
1250 
1251     if (!bdrv_can_snapshot(bs)) {
1252         error_setg(errp, "Block format '%s' used by device '%s' "
1253                    "does not support internal snapshots",
1254                    bs->drv->format_name, device);
1255         goto out;
1256     }
1257 
1258     if (!strlen(name)) {
1259         error_setg(errp, "Name is empty");
1260         goto out;
1261     }
1262 
1263     /* check whether a snapshot with name exist */
1264     ret = bdrv_snapshot_find_by_id_and_name(bs, NULL, name, &old_sn,
1265                                             &local_err);
1266     if (local_err) {
1267         error_propagate(errp, local_err);
1268         goto out;
1269     } else if (ret) {
1270         error_setg(errp,
1271                    "Snapshot with name '%s' already exists on device '%s'",
1272                    name, device);
1273         goto out;
1274     }
1275 
1276     /* 3. take the snapshot */
1277     sn = &state->sn;
1278     pstrcpy(sn->name, sizeof(sn->name), name);
1279     rt = g_get_real_time();
1280     sn->date_sec = rt / G_USEC_PER_SEC;
1281     sn->date_nsec = (rt % G_USEC_PER_SEC) * 1000;
1282     sn->vm_clock_nsec = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL);
1283     if (replay_mode != REPLAY_MODE_NONE) {
1284         sn->icount = replay_get_current_icount();
1285     } else {
1286         sn->icount = -1ULL;
1287     }
1288 
1289     ret1 = bdrv_snapshot_create(bs, sn);
1290     if (ret1 < 0) {
1291         error_setg_errno(errp, -ret1,
1292                          "Failed to create snapshot '%s' on device '%s'",
1293                          name, device);
1294         goto out;
1295     }
1296 
1297     /* 4. succeed, mark a snapshot is created */
1298     state->created = true;
1299 
1300 out:
1301     aio_context_release(aio_context);
1302 }
1303 
1304 static void internal_snapshot_abort(void *opaque)
1305 {
1306     InternalSnapshotState *state = opaque;
1307     BlockDriverState *bs = state->bs;
1308     QEMUSnapshotInfo *sn = &state->sn;
1309     AioContext *aio_context;
1310     Error *local_error = NULL;
1311 
1312     if (!state->created) {
1313         return;
1314     }
1315 
1316     aio_context = bdrv_get_aio_context(state->bs);
1317     aio_context_acquire(aio_context);
1318 
1319     if (bdrv_snapshot_delete(bs, sn->id_str, sn->name, &local_error) < 0) {
1320         error_reportf_err(local_error,
1321                           "Failed to delete snapshot with id '%s' and "
1322                           "name '%s' on device '%s' in abort: ",
1323                           sn->id_str, sn->name,
1324                           bdrv_get_device_name(bs));
1325     }
1326 
1327     aio_context_release(aio_context);
1328 }
1329 
1330 static void internal_snapshot_clean(void *opaque)
1331 {
1332     g_autofree InternalSnapshotState *state = opaque;
1333     AioContext *aio_context;
1334 
1335     if (!state->bs) {
1336         return;
1337     }
1338 
1339     aio_context = bdrv_get_aio_context(state->bs);
1340     aio_context_acquire(aio_context);
1341 
1342     bdrv_drained_end(state->bs);
1343 
1344     aio_context_release(aio_context);
1345 }
1346 
1347 /* external snapshot private data */
1348 typedef struct ExternalSnapshotState {
1349     BlockDriverState *old_bs;
1350     BlockDriverState *new_bs;
1351     bool overlay_appended;
1352 } ExternalSnapshotState;
1353 
1354 static void external_snapshot_commit(void *opaque);
1355 static void external_snapshot_abort(void *opaque);
1356 static void external_snapshot_clean(void *opaque);
1357 TransactionActionDrv external_snapshot_drv = {
1358     .commit = external_snapshot_commit,
1359     .abort = external_snapshot_abort,
1360     .clean = external_snapshot_clean,
1361 };
1362 
1363 static void external_snapshot_action(TransactionAction *action,
1364                                      Transaction *tran, Error **errp)
1365 {
1366     int ret;
1367     int flags = 0;
1368     QDict *options = NULL;
1369     Error *local_err = NULL;
1370     /* Device and node name of the image to generate the snapshot from */
1371     const char *device;
1372     const char *node_name;
1373     /* Reference to the new image (for 'blockdev-snapshot') */
1374     const char *snapshot_ref;
1375     /* File name of the new image (for 'blockdev-snapshot-sync') */
1376     const char *new_image_file;
1377     ExternalSnapshotState *state = g_new0(ExternalSnapshotState, 1);
1378     AioContext *aio_context;
1379     uint64_t perm, shared;
1380 
1381     tran_add(tran, &external_snapshot_drv, state);
1382 
1383     /* 'blockdev-snapshot' and 'blockdev-snapshot-sync' have similar
1384      * purpose but a different set of parameters */
1385     switch (action->type) {
1386     case TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT:
1387         {
1388             BlockdevSnapshot *s = action->u.blockdev_snapshot.data;
1389             device = s->node;
1390             node_name = s->node;
1391             new_image_file = NULL;
1392             snapshot_ref = s->overlay;
1393         }
1394         break;
1395     case TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC:
1396         {
1397             BlockdevSnapshotSync *s = action->u.blockdev_snapshot_sync.data;
1398             device = s->device;
1399             node_name = s->node_name;
1400             new_image_file = s->snapshot_file;
1401             snapshot_ref = NULL;
1402         }
1403         break;
1404     default:
1405         g_assert_not_reached();
1406     }
1407 
1408     /* start processing */
1409 
1410     state->old_bs = bdrv_lookup_bs(device, node_name, errp);
1411     if (!state->old_bs) {
1412         return;
1413     }
1414 
1415     aio_context = bdrv_get_aio_context(state->old_bs);
1416     aio_context_acquire(aio_context);
1417 
1418     /* Paired with .clean() */
1419     bdrv_drained_begin(state->old_bs);
1420 
1421     if (!bdrv_is_inserted(state->old_bs)) {
1422         error_setg(errp, QERR_DEVICE_HAS_NO_MEDIUM, device);
1423         goto out;
1424     }
1425 
1426     if (bdrv_op_is_blocked(state->old_bs,
1427                            BLOCK_OP_TYPE_EXTERNAL_SNAPSHOT, errp)) {
1428         goto out;
1429     }
1430 
1431     if (!bdrv_is_read_only(state->old_bs)) {
1432         if (bdrv_flush(state->old_bs)) {
1433             error_setg(errp, QERR_IO_ERROR);
1434             goto out;
1435         }
1436     }
1437 
1438     if (action->type == TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC) {
1439         BlockdevSnapshotSync *s = action->u.blockdev_snapshot_sync.data;
1440         const char *format = s->format ?: "qcow2";
1441         enum NewImageMode mode;
1442         const char *snapshot_node_name = s->snapshot_node_name;
1443 
1444         if (node_name && !snapshot_node_name) {
1445             error_setg(errp, "New overlay node-name missing");
1446             goto out;
1447         }
1448 
1449         if (snapshot_node_name &&
1450             bdrv_lookup_bs(snapshot_node_name, snapshot_node_name, NULL)) {
1451             error_setg(errp, "New overlay node-name already in use");
1452             goto out;
1453         }
1454 
1455         flags = state->old_bs->open_flags;
1456         flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_COPY_ON_READ);
1457         flags |= BDRV_O_NO_BACKING;
1458 
1459         /* create new image w/backing file */
1460         mode = s->has_mode ? s->mode : NEW_IMAGE_MODE_ABSOLUTE_PATHS;
1461         if (mode != NEW_IMAGE_MODE_EXISTING) {
1462             int64_t size = bdrv_getlength(state->old_bs);
1463             if (size < 0) {
1464                 error_setg_errno(errp, -size, "bdrv_getlength failed");
1465                 goto out;
1466             }
1467             bdrv_refresh_filename(state->old_bs);
1468 
1469             aio_context_release(aio_context);
1470             bdrv_img_create(new_image_file, format,
1471                             state->old_bs->filename,
1472                             state->old_bs->drv->format_name,
1473                             NULL, size, flags, false, &local_err);
1474             aio_context_acquire(aio_context);
1475 
1476             if (local_err) {
1477                 error_propagate(errp, local_err);
1478                 goto out;
1479             }
1480         }
1481 
1482         options = qdict_new();
1483         if (snapshot_node_name) {
1484             qdict_put_str(options, "node-name", snapshot_node_name);
1485         }
1486         qdict_put_str(options, "driver", format);
1487     }
1488     aio_context_release(aio_context);
1489 
1490     aio_context_acquire(qemu_get_aio_context());
1491     state->new_bs = bdrv_open(new_image_file, snapshot_ref, options, flags,
1492                               errp);
1493     aio_context_release(qemu_get_aio_context());
1494 
1495     /* We will manually add the backing_hd field to the bs later */
1496     if (!state->new_bs) {
1497         return;
1498     }
1499 
1500     aio_context_acquire(aio_context);
1501 
1502     /*
1503      * Allow attaching a backing file to an overlay that's already in use only
1504      * if the parents don't assume that they are already seeing a valid image.
1505      * (Specifically, allow it as a mirror target, which is write-only access.)
1506      */
1507     bdrv_get_cumulative_perm(state->new_bs, &perm, &shared);
1508     if (perm & BLK_PERM_CONSISTENT_READ) {
1509         error_setg(errp, "The overlay is already in use");
1510         goto out;
1511     }
1512 
1513     if (state->new_bs->drv->is_filter) {
1514         error_setg(errp, "Filters cannot be used as overlays");
1515         goto out;
1516     }
1517 
1518     if (bdrv_cow_child(state->new_bs)) {
1519         error_setg(errp, "The overlay already has a backing image");
1520         goto out;
1521     }
1522 
1523     if (!state->new_bs->drv->supports_backing) {
1524         error_setg(errp, "The overlay does not support backing images");
1525         goto out;
1526     }
1527 
1528     ret = bdrv_append(state->new_bs, state->old_bs, errp);
1529     if (ret < 0) {
1530         goto out;
1531     }
1532     state->overlay_appended = true;
1533 
1534 out:
1535     aio_context_release(aio_context);
1536 }
1537 
1538 static void external_snapshot_commit(void *opaque)
1539 {
1540     ExternalSnapshotState *state = opaque;
1541     AioContext *aio_context;
1542 
1543     aio_context = bdrv_get_aio_context(state->old_bs);
1544     aio_context_acquire(aio_context);
1545 
1546     /* We don't need (or want) to use the transactional
1547      * bdrv_reopen_multiple() across all the entries at once, because we
1548      * don't want to abort all of them if one of them fails the reopen */
1549     if (!qatomic_read(&state->old_bs->copy_on_read)) {
1550         bdrv_reopen_set_read_only(state->old_bs, true, NULL);
1551     }
1552 
1553     aio_context_release(aio_context);
1554 }
1555 
1556 static void external_snapshot_abort(void *opaque)
1557 {
1558     ExternalSnapshotState *state = opaque;
1559     if (state->new_bs) {
1560         if (state->overlay_appended) {
1561             AioContext *aio_context;
1562             AioContext *tmp_context;
1563             int ret;
1564 
1565             aio_context = bdrv_get_aio_context(state->old_bs);
1566             aio_context_acquire(aio_context);
1567 
1568             bdrv_ref(state->old_bs);   /* we can't let bdrv_set_backind_hd()
1569                                           close state->old_bs; we need it */
1570             bdrv_set_backing_hd(state->new_bs, NULL, &error_abort);
1571 
1572             /*
1573              * The call to bdrv_set_backing_hd() above returns state->old_bs to
1574              * the main AioContext. As we're still going to be using it, return
1575              * it to the AioContext it was before.
1576              */
1577             tmp_context = bdrv_get_aio_context(state->old_bs);
1578             if (aio_context != tmp_context) {
1579                 aio_context_release(aio_context);
1580                 aio_context_acquire(tmp_context);
1581 
1582                 ret = bdrv_try_change_aio_context(state->old_bs,
1583                                                   aio_context, NULL, NULL);
1584                 assert(ret == 0);
1585 
1586                 aio_context_release(tmp_context);
1587                 aio_context_acquire(aio_context);
1588             }
1589 
1590             bdrv_replace_node(state->new_bs, state->old_bs, &error_abort);
1591             bdrv_unref(state->old_bs); /* bdrv_replace_node() ref'ed old_bs */
1592 
1593             aio_context_release(aio_context);
1594         }
1595     }
1596 }
1597 
1598 static void external_snapshot_clean(void *opaque)
1599 {
1600     g_autofree ExternalSnapshotState *state = opaque;
1601     AioContext *aio_context;
1602 
1603     if (!state->old_bs) {
1604         return;
1605     }
1606 
1607     aio_context = bdrv_get_aio_context(state->old_bs);
1608     aio_context_acquire(aio_context);
1609 
1610     bdrv_drained_end(state->old_bs);
1611     bdrv_unref(state->new_bs);
1612 
1613     aio_context_release(aio_context);
1614 }
1615 
1616 typedef struct DriveBackupState {
1617     BlockDriverState *bs;
1618     BlockJob *job;
1619 } DriveBackupState;
1620 
1621 static BlockJob *do_backup_common(BackupCommon *backup,
1622                                   BlockDriverState *bs,
1623                                   BlockDriverState *target_bs,
1624                                   AioContext *aio_context,
1625                                   JobTxn *txn, Error **errp);
1626 
1627 static void drive_backup_commit(void *opaque);
1628 static void drive_backup_abort(void *opaque);
1629 static void drive_backup_clean(void *opaque);
1630 TransactionActionDrv drive_backup_drv = {
1631     .commit = drive_backup_commit,
1632     .abort = drive_backup_abort,
1633     .clean = drive_backup_clean,
1634 };
1635 
1636 static void drive_backup_action(DriveBackup *backup,
1637                                 JobTxn *block_job_txn,
1638                                 Transaction *tran, Error **errp)
1639 {
1640     DriveBackupState *state = g_new0(DriveBackupState, 1);
1641     BlockDriverState *bs;
1642     BlockDriverState *target_bs;
1643     BlockDriverState *source = NULL;
1644     AioContext *aio_context;
1645     AioContext *old_context;
1646     const char *format;
1647     QDict *options;
1648     Error *local_err = NULL;
1649     int flags;
1650     int64_t size;
1651     bool set_backing_hd = false;
1652     int ret;
1653 
1654     tran_add(tran, &drive_backup_drv, state);
1655 
1656     if (!backup->has_mode) {
1657         backup->mode = NEW_IMAGE_MODE_ABSOLUTE_PATHS;
1658     }
1659 
1660     bs = bdrv_lookup_bs(backup->device, backup->device, errp);
1661     if (!bs) {
1662         return;
1663     }
1664 
1665     if (!bs->drv) {
1666         error_setg(errp, "Device has no medium");
1667         return;
1668     }
1669 
1670     aio_context = bdrv_get_aio_context(bs);
1671     aio_context_acquire(aio_context);
1672 
1673     state->bs = bs;
1674     /* Paired with .clean() */
1675     bdrv_drained_begin(bs);
1676 
1677     format = backup->format;
1678     if (!format && backup->mode != NEW_IMAGE_MODE_EXISTING) {
1679         format = bs->drv->format_name;
1680     }
1681 
1682     /* Early check to avoid creating target */
1683     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_BACKUP_SOURCE, errp)) {
1684         goto out;
1685     }
1686 
1687     flags = bs->open_flags | BDRV_O_RDWR;
1688 
1689     /*
1690      * See if we have a backing HD we can use to create our new image
1691      * on top of.
1692      */
1693     if (backup->sync == MIRROR_SYNC_MODE_TOP) {
1694         /*
1695          * Backup will not replace the source by the target, so none
1696          * of the filters skipped here will be removed (in contrast to
1697          * mirror).  Therefore, we can skip all of them when looking
1698          * for the first COW relationship.
1699          */
1700         source = bdrv_cow_bs(bdrv_skip_filters(bs));
1701         if (!source) {
1702             backup->sync = MIRROR_SYNC_MODE_FULL;
1703         }
1704     }
1705     if (backup->sync == MIRROR_SYNC_MODE_NONE) {
1706         source = bs;
1707         flags |= BDRV_O_NO_BACKING;
1708         set_backing_hd = true;
1709     }
1710 
1711     size = bdrv_getlength(bs);
1712     if (size < 0) {
1713         error_setg_errno(errp, -size, "bdrv_getlength failed");
1714         goto out;
1715     }
1716 
1717     if (backup->mode != NEW_IMAGE_MODE_EXISTING) {
1718         assert(format);
1719         if (source) {
1720             /* Implicit filters should not appear in the filename */
1721             BlockDriverState *explicit_backing =
1722                 bdrv_skip_implicit_filters(source);
1723 
1724             bdrv_refresh_filename(explicit_backing);
1725             bdrv_img_create(backup->target, format,
1726                             explicit_backing->filename,
1727                             explicit_backing->drv->format_name, NULL,
1728                             size, flags, false, &local_err);
1729         } else {
1730             bdrv_img_create(backup->target, format, NULL, NULL, NULL,
1731                             size, flags, false, &local_err);
1732         }
1733     }
1734 
1735     if (local_err) {
1736         error_propagate(errp, local_err);
1737         goto out;
1738     }
1739 
1740     options = qdict_new();
1741     qdict_put_str(options, "discard", "unmap");
1742     qdict_put_str(options, "detect-zeroes", "unmap");
1743     if (format) {
1744         qdict_put_str(options, "driver", format);
1745     }
1746     aio_context_release(aio_context);
1747 
1748     aio_context_acquire(qemu_get_aio_context());
1749     target_bs = bdrv_open(backup->target, NULL, options, flags, errp);
1750     aio_context_release(qemu_get_aio_context());
1751 
1752     if (!target_bs) {
1753         return;
1754     }
1755 
1756     /* Honor bdrv_try_change_aio_context() context acquisition requirements. */
1757     old_context = bdrv_get_aio_context(target_bs);
1758     aio_context_acquire(old_context);
1759 
1760     ret = bdrv_try_change_aio_context(target_bs, aio_context, NULL, errp);
1761     if (ret < 0) {
1762         bdrv_unref(target_bs);
1763         aio_context_release(old_context);
1764         return;
1765     }
1766 
1767     aio_context_release(old_context);
1768     aio_context_acquire(aio_context);
1769 
1770     if (set_backing_hd) {
1771         if (bdrv_set_backing_hd(target_bs, source, errp) < 0) {
1772             goto unref;
1773         }
1774     }
1775 
1776     state->job = do_backup_common(qapi_DriveBackup_base(backup),
1777                                   bs, target_bs, aio_context,
1778                                   block_job_txn, errp);
1779 
1780 unref:
1781     bdrv_unref(target_bs);
1782 out:
1783     aio_context_release(aio_context);
1784 }
1785 
1786 static void drive_backup_commit(void *opaque)
1787 {
1788     DriveBackupState *state = opaque;
1789     AioContext *aio_context;
1790 
1791     aio_context = bdrv_get_aio_context(state->bs);
1792     aio_context_acquire(aio_context);
1793 
1794     assert(state->job);
1795     job_start(&state->job->job);
1796 
1797     aio_context_release(aio_context);
1798 }
1799 
1800 static void drive_backup_abort(void *opaque)
1801 {
1802     DriveBackupState *state = opaque;
1803 
1804     if (state->job) {
1805         job_cancel_sync(&state->job->job, true);
1806     }
1807 }
1808 
1809 static void drive_backup_clean(void *opaque)
1810 {
1811     g_autofree DriveBackupState *state = opaque;
1812     AioContext *aio_context;
1813 
1814     if (!state->bs) {
1815         return;
1816     }
1817 
1818     aio_context = bdrv_get_aio_context(state->bs);
1819     aio_context_acquire(aio_context);
1820 
1821     bdrv_drained_end(state->bs);
1822 
1823     aio_context_release(aio_context);
1824 }
1825 
1826 typedef struct BlockdevBackupState {
1827     BlockDriverState *bs;
1828     BlockJob *job;
1829 } BlockdevBackupState;
1830 
1831 static void blockdev_backup_commit(void *opaque);
1832 static void blockdev_backup_abort(void *opaque);
1833 static void blockdev_backup_clean(void *opaque);
1834 TransactionActionDrv blockdev_backup_drv = {
1835     .commit = blockdev_backup_commit,
1836     .abort = blockdev_backup_abort,
1837     .clean = blockdev_backup_clean,
1838 };
1839 
1840 static void blockdev_backup_action(BlockdevBackup *backup,
1841                                    JobTxn *block_job_txn,
1842                                    Transaction *tran, Error **errp)
1843 {
1844     BlockdevBackupState *state = g_new0(BlockdevBackupState, 1);
1845     BlockDriverState *bs;
1846     BlockDriverState *target_bs;
1847     AioContext *aio_context;
1848     AioContext *old_context;
1849     int ret;
1850 
1851     tran_add(tran, &blockdev_backup_drv, state);
1852 
1853     bs = bdrv_lookup_bs(backup->device, backup->device, errp);
1854     if (!bs) {
1855         return;
1856     }
1857 
1858     target_bs = bdrv_lookup_bs(backup->target, backup->target, errp);
1859     if (!target_bs) {
1860         return;
1861     }
1862 
1863     /* Honor bdrv_try_change_aio_context() context acquisition requirements. */
1864     aio_context = bdrv_get_aio_context(bs);
1865     old_context = bdrv_get_aio_context(target_bs);
1866     aio_context_acquire(old_context);
1867 
1868     ret = bdrv_try_change_aio_context(target_bs, aio_context, NULL, errp);
1869     if (ret < 0) {
1870         aio_context_release(old_context);
1871         return;
1872     }
1873 
1874     aio_context_release(old_context);
1875     aio_context_acquire(aio_context);
1876     state->bs = bs;
1877 
1878     /* Paired with .clean() */
1879     bdrv_drained_begin(state->bs);
1880 
1881     state->job = do_backup_common(qapi_BlockdevBackup_base(backup),
1882                                   bs, target_bs, aio_context,
1883                                   block_job_txn, errp);
1884 
1885     aio_context_release(aio_context);
1886 }
1887 
1888 static void blockdev_backup_commit(void *opaque)
1889 {
1890     BlockdevBackupState *state = opaque;
1891     AioContext *aio_context;
1892 
1893     aio_context = bdrv_get_aio_context(state->bs);
1894     aio_context_acquire(aio_context);
1895 
1896     assert(state->job);
1897     job_start(&state->job->job);
1898 
1899     aio_context_release(aio_context);
1900 }
1901 
1902 static void blockdev_backup_abort(void *opaque)
1903 {
1904     BlockdevBackupState *state = opaque;
1905 
1906     if (state->job) {
1907         job_cancel_sync(&state->job->job, true);
1908     }
1909 }
1910 
1911 static void blockdev_backup_clean(void *opaque)
1912 {
1913     g_autofree BlockdevBackupState *state = opaque;
1914     AioContext *aio_context;
1915 
1916     if (!state->bs) {
1917         return;
1918     }
1919 
1920     aio_context = bdrv_get_aio_context(state->bs);
1921     aio_context_acquire(aio_context);
1922 
1923     bdrv_drained_end(state->bs);
1924 
1925     aio_context_release(aio_context);
1926 }
1927 
1928 typedef struct BlockDirtyBitmapState {
1929     BdrvDirtyBitmap *bitmap;
1930     BlockDriverState *bs;
1931     HBitmap *backup;
1932     bool was_enabled;
1933 } BlockDirtyBitmapState;
1934 
1935 static void block_dirty_bitmap_add_abort(void *opaque);
1936 TransactionActionDrv block_dirty_bitmap_add_drv = {
1937     .abort = block_dirty_bitmap_add_abort,
1938     .clean = g_free,
1939 };
1940 
1941 static void block_dirty_bitmap_add_action(BlockDirtyBitmapAdd *action,
1942                                           Transaction *tran, Error **errp)
1943 {
1944     Error *local_err = NULL;
1945     BlockDirtyBitmapState *state = g_new0(BlockDirtyBitmapState, 1);
1946 
1947     tran_add(tran, &block_dirty_bitmap_add_drv, state);
1948 
1949     /* AIO context taken and released within qmp_block_dirty_bitmap_add */
1950     qmp_block_dirty_bitmap_add(action->node, action->name,
1951                                action->has_granularity, action->granularity,
1952                                action->has_persistent, action->persistent,
1953                                action->has_disabled, action->disabled,
1954                                &local_err);
1955 
1956     if (!local_err) {
1957         state->bitmap = block_dirty_bitmap_lookup(action->node, action->name,
1958                                                   NULL, &error_abort);
1959     } else {
1960         error_propagate(errp, local_err);
1961     }
1962 }
1963 
1964 static void block_dirty_bitmap_add_abort(void *opaque)
1965 {
1966     BlockDirtyBitmapState *state = opaque;
1967 
1968     if (state->bitmap) {
1969         bdrv_release_dirty_bitmap(state->bitmap);
1970     }
1971 }
1972 
1973 static void block_dirty_bitmap_restore(void *opaque);
1974 static void block_dirty_bitmap_free_backup(void *opaque);
1975 TransactionActionDrv block_dirty_bitmap_clear_drv = {
1976     .abort = block_dirty_bitmap_restore,
1977     .commit = block_dirty_bitmap_free_backup,
1978     .clean = g_free,
1979 };
1980 
1981 static void block_dirty_bitmap_clear_action(BlockDirtyBitmap *action,
1982                                             Transaction *tran, Error **errp)
1983 {
1984     BlockDirtyBitmapState *state = g_new0(BlockDirtyBitmapState, 1);
1985 
1986     tran_add(tran, &block_dirty_bitmap_clear_drv, state);
1987 
1988     state->bitmap = block_dirty_bitmap_lookup(action->node,
1989                                               action->name,
1990                                               &state->bs,
1991                                               errp);
1992     if (!state->bitmap) {
1993         return;
1994     }
1995 
1996     if (bdrv_dirty_bitmap_check(state->bitmap, BDRV_BITMAP_DEFAULT, errp)) {
1997         return;
1998     }
1999 
2000     bdrv_clear_dirty_bitmap(state->bitmap, &state->backup);
2001 }
2002 
2003 static void block_dirty_bitmap_restore(void *opaque)
2004 {
2005     BlockDirtyBitmapState *state = opaque;
2006 
2007     if (state->backup) {
2008         bdrv_restore_dirty_bitmap(state->bitmap, state->backup);
2009     }
2010 }
2011 
2012 static void block_dirty_bitmap_free_backup(void *opaque)
2013 {
2014     BlockDirtyBitmapState *state = opaque;
2015 
2016     hbitmap_free(state->backup);
2017 }
2018 
2019 static void block_dirty_bitmap_enable_abort(void *opaque);
2020 TransactionActionDrv block_dirty_bitmap_enable_drv = {
2021     .abort = block_dirty_bitmap_enable_abort,
2022     .clean = g_free,
2023 };
2024 
2025 static void block_dirty_bitmap_enable_action(BlockDirtyBitmap *action,
2026                                              Transaction *tran, Error **errp)
2027 {
2028     BlockDirtyBitmapState *state = g_new0(BlockDirtyBitmapState, 1);
2029 
2030     tran_add(tran, &block_dirty_bitmap_enable_drv, state);
2031 
2032     state->bitmap = block_dirty_bitmap_lookup(action->node,
2033                                               action->name,
2034                                               NULL,
2035                                               errp);
2036     if (!state->bitmap) {
2037         return;
2038     }
2039 
2040     if (bdrv_dirty_bitmap_check(state->bitmap, BDRV_BITMAP_ALLOW_RO, errp)) {
2041         return;
2042     }
2043 
2044     state->was_enabled = bdrv_dirty_bitmap_enabled(state->bitmap);
2045     bdrv_enable_dirty_bitmap(state->bitmap);
2046 }
2047 
2048 static void block_dirty_bitmap_enable_abort(void *opaque)
2049 {
2050     BlockDirtyBitmapState *state = opaque;
2051 
2052     if (!state->was_enabled) {
2053         bdrv_disable_dirty_bitmap(state->bitmap);
2054     }
2055 }
2056 
2057 static void block_dirty_bitmap_disable_abort(void *opaque);
2058 TransactionActionDrv block_dirty_bitmap_disable_drv = {
2059     .abort = block_dirty_bitmap_disable_abort,
2060     .clean = g_free,
2061 };
2062 
2063 static void block_dirty_bitmap_disable_action(BlockDirtyBitmap *action,
2064                                               Transaction *tran, Error **errp)
2065 {
2066     BlockDirtyBitmapState *state = g_new0(BlockDirtyBitmapState, 1);
2067 
2068     tran_add(tran, &block_dirty_bitmap_disable_drv, state);
2069 
2070     state->bitmap = block_dirty_bitmap_lookup(action->node,
2071                                               action->name,
2072                                               NULL,
2073                                               errp);
2074     if (!state->bitmap) {
2075         return;
2076     }
2077 
2078     if (bdrv_dirty_bitmap_check(state->bitmap, BDRV_BITMAP_ALLOW_RO, errp)) {
2079         return;
2080     }
2081 
2082     state->was_enabled = bdrv_dirty_bitmap_enabled(state->bitmap);
2083     bdrv_disable_dirty_bitmap(state->bitmap);
2084 }
2085 
2086 static void block_dirty_bitmap_disable_abort(void *opaque)
2087 {
2088     BlockDirtyBitmapState *state = opaque;
2089 
2090     if (state->was_enabled) {
2091         bdrv_enable_dirty_bitmap(state->bitmap);
2092     }
2093 }
2094 
2095 TransactionActionDrv block_dirty_bitmap_merge_drv = {
2096     .commit = block_dirty_bitmap_free_backup,
2097     .abort = block_dirty_bitmap_restore,
2098     .clean = g_free,
2099 };
2100 
2101 static void block_dirty_bitmap_merge_action(BlockDirtyBitmapMerge *action,
2102                                             Transaction *tran, Error **errp)
2103 {
2104     BlockDirtyBitmapState *state = g_new0(BlockDirtyBitmapState, 1);
2105 
2106     tran_add(tran, &block_dirty_bitmap_merge_drv, state);
2107 
2108     state->bitmap = block_dirty_bitmap_merge(action->node, action->target,
2109                                              action->bitmaps, &state->backup,
2110                                              errp);
2111 }
2112 
2113 static void block_dirty_bitmap_remove_commit(void *opaque);
2114 static void block_dirty_bitmap_remove_abort(void *opaque);
2115 TransactionActionDrv block_dirty_bitmap_remove_drv = {
2116     .commit = block_dirty_bitmap_remove_commit,
2117     .abort = block_dirty_bitmap_remove_abort,
2118     .clean = g_free,
2119 };
2120 
2121 static void block_dirty_bitmap_remove_action(BlockDirtyBitmap *action,
2122                                              Transaction *tran, Error **errp)
2123 {
2124     BlockDirtyBitmapState *state = g_new0(BlockDirtyBitmapState, 1);
2125 
2126     tran_add(tran, &block_dirty_bitmap_remove_drv, state);
2127 
2128 
2129     state->bitmap = block_dirty_bitmap_remove(action->node, action->name,
2130                                               false, &state->bs, errp);
2131     if (state->bitmap) {
2132         bdrv_dirty_bitmap_skip_store(state->bitmap, true);
2133         bdrv_dirty_bitmap_set_busy(state->bitmap, true);
2134     }
2135 }
2136 
2137 static void block_dirty_bitmap_remove_abort(void *opaque)
2138 {
2139     BlockDirtyBitmapState *state = opaque;
2140 
2141     if (state->bitmap) {
2142         bdrv_dirty_bitmap_skip_store(state->bitmap, false);
2143         bdrv_dirty_bitmap_set_busy(state->bitmap, false);
2144     }
2145 }
2146 
2147 static void block_dirty_bitmap_remove_commit(void *opaque)
2148 {
2149     BlockDirtyBitmapState *state = opaque;
2150 
2151     bdrv_dirty_bitmap_set_busy(state->bitmap, false);
2152     bdrv_release_dirty_bitmap(state->bitmap);
2153 }
2154 
2155 static void abort_commit(void *opaque);
2156 TransactionActionDrv abort_drv = {
2157     .commit = abort_commit,
2158 };
2159 
2160 static void abort_action(Transaction *tran, Error **errp)
2161 {
2162     tran_add(tran, &abort_drv, NULL);
2163     error_setg(errp, "Transaction aborted using Abort action");
2164 }
2165 
2166 static void abort_commit(void *opaque)
2167 {
2168     g_assert_not_reached(); /* this action never succeeds */
2169 }
2170 
2171 static void transaction_action(TransactionAction *act, JobTxn *block_job_txn,
2172                                Transaction *tran, Error **errp)
2173 {
2174     switch (act->type) {
2175     case TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT:
2176     case TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC:
2177         external_snapshot_action(act, tran, errp);
2178         return;
2179     case TRANSACTION_ACTION_KIND_DRIVE_BACKUP:
2180         drive_backup_action(act->u.drive_backup.data,
2181                             block_job_txn, tran, errp);
2182         return;
2183     case TRANSACTION_ACTION_KIND_BLOCKDEV_BACKUP:
2184         blockdev_backup_action(act->u.blockdev_backup.data,
2185                                block_job_txn, tran, errp);
2186         return;
2187     case TRANSACTION_ACTION_KIND_ABORT:
2188         abort_action(tran, errp);
2189         return;
2190     case TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC:
2191         internal_snapshot_action(act->u.blockdev_snapshot_internal_sync.data,
2192                                  tran, errp);
2193         return;
2194     case TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_ADD:
2195         block_dirty_bitmap_add_action(act->u.block_dirty_bitmap_add.data,
2196                                       tran, errp);
2197         return;
2198     case TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_CLEAR:
2199         block_dirty_bitmap_clear_action(act->u.block_dirty_bitmap_clear.data,
2200                                         tran, errp);
2201         return;
2202     case TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_ENABLE:
2203         block_dirty_bitmap_enable_action(act->u.block_dirty_bitmap_enable.data,
2204                                          tran, errp);
2205         return;
2206     case TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_DISABLE:
2207         block_dirty_bitmap_disable_action(
2208                 act->u.block_dirty_bitmap_disable.data, tran, errp);
2209         return;
2210     case TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_MERGE:
2211         block_dirty_bitmap_merge_action(act->u.block_dirty_bitmap_merge.data,
2212                                         tran, errp);
2213         return;
2214     case TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_REMOVE:
2215         block_dirty_bitmap_remove_action(act->u.block_dirty_bitmap_remove.data,
2216                                          tran, errp);
2217         return;
2218     /*
2219      * Where are transactions for MIRROR, COMMIT and STREAM?
2220      * Although these blockjobs use transaction callbacks like the backup job,
2221      * these jobs do not necessarily adhere to transaction semantics.
2222      * These jobs may not fully undo all of their actions on abort, nor do they
2223      * necessarily work in transactions with more than one job in them.
2224      */
2225     case TRANSACTION_ACTION_KIND__MAX:
2226     default:
2227         g_assert_not_reached();
2228     };
2229 }
2230 
2231 
2232 /*
2233  * 'Atomic' group operations.  The operations are performed as a set, and if
2234  * any fail then we roll back all operations in the group.
2235  *
2236  * Always run under BQL.
2237  */
2238 void qmp_transaction(TransactionActionList *actions,
2239                      struct TransactionProperties *properties,
2240                      Error **errp)
2241 {
2242     TransactionActionList *act;
2243     JobTxn *block_job_txn = NULL;
2244     Error *local_err = NULL;
2245     Transaction *tran;
2246     ActionCompletionMode comp_mode =
2247         properties ? properties->completion_mode :
2248         ACTION_COMPLETION_MODE_INDIVIDUAL;
2249 
2250     GLOBAL_STATE_CODE();
2251 
2252     /* Does this transaction get canceled as a group on failure?
2253      * If not, we don't really need to make a JobTxn.
2254      */
2255     if (comp_mode != ACTION_COMPLETION_MODE_INDIVIDUAL) {
2256         for (act = actions; act; act = act->next) {
2257             TransactionActionKind type = act->value->type;
2258 
2259             if (type != TRANSACTION_ACTION_KIND_BLOCKDEV_BACKUP &&
2260                 type != TRANSACTION_ACTION_KIND_DRIVE_BACKUP)
2261             {
2262                 error_setg(errp,
2263                            "Action '%s' does not support transaction property "
2264                            "completion-mode = %s",
2265                            TransactionActionKind_str(type),
2266                            ActionCompletionMode_str(comp_mode));
2267                 return;
2268             }
2269         }
2270 
2271         block_job_txn = job_txn_new();
2272     }
2273 
2274     /* drain all i/o before any operations */
2275     bdrv_drain_all();
2276 
2277     tran = tran_new();
2278 
2279     /* We don't do anything in this loop that commits us to the operations */
2280     for (act = actions; act; act = act->next) {
2281         transaction_action(act->value, block_job_txn, tran, &local_err);
2282         if (local_err) {
2283             error_propagate(errp, local_err);
2284             goto delete_and_fail;
2285         }
2286     }
2287 
2288     tran_commit(tran);
2289 
2290     /* success */
2291     goto exit;
2292 
2293 delete_and_fail:
2294     /* failure, and it is all-or-none; roll back all operations */
2295     tran_abort(tran);
2296 exit:
2297     job_txn_unref(block_job_txn);
2298 }
2299 
2300 BlockDirtyBitmapSha256 *qmp_x_debug_block_dirty_bitmap_sha256(const char *node,
2301                                                               const char *name,
2302                                                               Error **errp)
2303 {
2304     BdrvDirtyBitmap *bitmap;
2305     BlockDriverState *bs;
2306     BlockDirtyBitmapSha256 *ret = NULL;
2307     char *sha256;
2308 
2309     bitmap = block_dirty_bitmap_lookup(node, name, &bs, errp);
2310     if (!bitmap || !bs) {
2311         return NULL;
2312     }
2313 
2314     sha256 = bdrv_dirty_bitmap_sha256(bitmap, errp);
2315     if (sha256 == NULL) {
2316         return NULL;
2317     }
2318 
2319     ret = g_new(BlockDirtyBitmapSha256, 1);
2320     ret->sha256 = sha256;
2321 
2322     return ret;
2323 }
2324 
2325 void coroutine_fn qmp_block_resize(const char *device, const char *node_name,
2326                                    int64_t size, Error **errp)
2327 {
2328     Error *local_err = NULL;
2329     BlockBackend *blk;
2330     BlockDriverState *bs;
2331     AioContext *old_ctx;
2332 
2333     bs = bdrv_lookup_bs(device, node_name, &local_err);
2334     if (local_err) {
2335         error_propagate(errp, local_err);
2336         return;
2337     }
2338 
2339     if (size < 0) {
2340         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "size", "a >0 size");
2341         return;
2342     }
2343 
2344     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_RESIZE, NULL)) {
2345         error_setg(errp, QERR_DEVICE_IN_USE, device);
2346         return;
2347     }
2348 
2349     blk = blk_co_new_with_bs(bs, BLK_PERM_RESIZE, BLK_PERM_ALL, errp);
2350     if (!blk) {
2351         return;
2352     }
2353 
2354     bdrv_co_lock(bs);
2355     bdrv_drained_begin(bs);
2356     bdrv_co_unlock(bs);
2357 
2358     old_ctx = bdrv_co_enter(bs);
2359     blk_co_truncate(blk, size, false, PREALLOC_MODE_OFF, 0, errp);
2360     bdrv_co_leave(bs, old_ctx);
2361 
2362     bdrv_co_lock(bs);
2363     bdrv_drained_end(bs);
2364     blk_co_unref(blk);
2365     bdrv_co_unlock(bs);
2366 }
2367 
2368 void qmp_block_stream(const char *job_id, const char *device,
2369                       const char *base,
2370                       const char *base_node,
2371                       const char *backing_file,
2372                       const char *bottom,
2373                       bool has_speed, int64_t speed,
2374                       bool has_on_error, BlockdevOnError on_error,
2375                       const char *filter_node_name,
2376                       bool has_auto_finalize, bool auto_finalize,
2377                       bool has_auto_dismiss, bool auto_dismiss,
2378                       Error **errp)
2379 {
2380     BlockDriverState *bs, *iter, *iter_end;
2381     BlockDriverState *base_bs = NULL;
2382     BlockDriverState *bottom_bs = NULL;
2383     AioContext *aio_context;
2384     Error *local_err = NULL;
2385     int job_flags = JOB_DEFAULT;
2386 
2387     if (base && base_node) {
2388         error_setg(errp, "'base' and 'base-node' cannot be specified "
2389                    "at the same time");
2390         return;
2391     }
2392 
2393     if (base && bottom) {
2394         error_setg(errp, "'base' and 'bottom' cannot be specified "
2395                    "at the same time");
2396         return;
2397     }
2398 
2399     if (bottom && base_node) {
2400         error_setg(errp, "'bottom' and 'base-node' cannot be specified "
2401                    "at the same time");
2402         return;
2403     }
2404 
2405     if (!has_on_error) {
2406         on_error = BLOCKDEV_ON_ERROR_REPORT;
2407     }
2408 
2409     bs = bdrv_lookup_bs(device, device, errp);
2410     if (!bs) {
2411         return;
2412     }
2413 
2414     aio_context = bdrv_get_aio_context(bs);
2415     aio_context_acquire(aio_context);
2416 
2417     if (base) {
2418         base_bs = bdrv_find_backing_image(bs, base);
2419         if (base_bs == NULL) {
2420             error_setg(errp, "Can't find '%s' in the backing chain", base);
2421             goto out;
2422         }
2423         assert(bdrv_get_aio_context(base_bs) == aio_context);
2424     }
2425 
2426     if (base_node) {
2427         base_bs = bdrv_lookup_bs(NULL, base_node, errp);
2428         if (!base_bs) {
2429             goto out;
2430         }
2431         if (bs == base_bs || !bdrv_chain_contains(bs, base_bs)) {
2432             error_setg(errp, "Node '%s' is not a backing image of '%s'",
2433                        base_node, device);
2434             goto out;
2435         }
2436         assert(bdrv_get_aio_context(base_bs) == aio_context);
2437         bdrv_refresh_filename(base_bs);
2438     }
2439 
2440     if (bottom) {
2441         bottom_bs = bdrv_lookup_bs(NULL, bottom, errp);
2442         if (!bottom_bs) {
2443             goto out;
2444         }
2445         if (!bottom_bs->drv) {
2446             error_setg(errp, "Node '%s' is not open", bottom);
2447             goto out;
2448         }
2449         if (bottom_bs->drv->is_filter) {
2450             error_setg(errp, "Node '%s' is a filter, use a non-filter node "
2451                        "as 'bottom'", bottom);
2452             goto out;
2453         }
2454         if (!bdrv_chain_contains(bs, bottom_bs)) {
2455             error_setg(errp, "Node '%s' is not in a chain starting from '%s'",
2456                        bottom, device);
2457             goto out;
2458         }
2459         assert(bdrv_get_aio_context(bottom_bs) == aio_context);
2460     }
2461 
2462     /*
2463      * Check for op blockers in the whole chain between bs and base (or bottom)
2464      */
2465     iter_end = bottom ? bdrv_filter_or_cow_bs(bottom_bs) : base_bs;
2466     for (iter = bs; iter && iter != iter_end;
2467          iter = bdrv_filter_or_cow_bs(iter))
2468     {
2469         if (bdrv_op_is_blocked(iter, BLOCK_OP_TYPE_STREAM, errp)) {
2470             goto out;
2471         }
2472     }
2473 
2474     /* if we are streaming the entire chain, the result will have no backing
2475      * file, and specifying one is therefore an error */
2476     if (!base_bs && backing_file) {
2477         error_setg(errp, "backing file specified, but streaming the "
2478                          "entire chain");
2479         goto out;
2480     }
2481 
2482     if (has_auto_finalize && !auto_finalize) {
2483         job_flags |= JOB_MANUAL_FINALIZE;
2484     }
2485     if (has_auto_dismiss && !auto_dismiss) {
2486         job_flags |= JOB_MANUAL_DISMISS;
2487     }
2488 
2489     stream_start(job_id, bs, base_bs, backing_file,
2490                  bottom_bs, job_flags, has_speed ? speed : 0, on_error,
2491                  filter_node_name, &local_err);
2492     if (local_err) {
2493         error_propagate(errp, local_err);
2494         goto out;
2495     }
2496 
2497     trace_qmp_block_stream(bs);
2498 
2499 out:
2500     aio_context_release(aio_context);
2501 }
2502 
2503 void qmp_block_commit(const char *job_id, const char *device,
2504                       const char *base_node,
2505                       const char *base,
2506                       const char *top_node,
2507                       const char *top,
2508                       const char *backing_file,
2509                       bool has_speed, int64_t speed,
2510                       bool has_on_error, BlockdevOnError on_error,
2511                       const char *filter_node_name,
2512                       bool has_auto_finalize, bool auto_finalize,
2513                       bool has_auto_dismiss, bool auto_dismiss,
2514                       Error **errp)
2515 {
2516     BlockDriverState *bs;
2517     BlockDriverState *iter;
2518     BlockDriverState *base_bs, *top_bs;
2519     AioContext *aio_context;
2520     Error *local_err = NULL;
2521     int job_flags = JOB_DEFAULT;
2522     uint64_t top_perm, top_shared;
2523 
2524     if (!has_speed) {
2525         speed = 0;
2526     }
2527     if (!has_on_error) {
2528         on_error = BLOCKDEV_ON_ERROR_REPORT;
2529     }
2530     if (has_auto_finalize && !auto_finalize) {
2531         job_flags |= JOB_MANUAL_FINALIZE;
2532     }
2533     if (has_auto_dismiss && !auto_dismiss) {
2534         job_flags |= JOB_MANUAL_DISMISS;
2535     }
2536 
2537     /* Important Note:
2538      *  libvirt relies on the DeviceNotFound error class in order to probe for
2539      *  live commit feature versions; for this to work, we must make sure to
2540      *  perform the device lookup before any generic errors that may occur in a
2541      *  scenario in which all optional arguments are omitted. */
2542     bs = qmp_get_root_bs(device, &local_err);
2543     if (!bs) {
2544         bs = bdrv_lookup_bs(device, device, NULL);
2545         if (!bs) {
2546             error_free(local_err);
2547             error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
2548                       "Device '%s' not found", device);
2549         } else {
2550             error_propagate(errp, local_err);
2551         }
2552         return;
2553     }
2554 
2555     aio_context = bdrv_get_aio_context(bs);
2556     aio_context_acquire(aio_context);
2557 
2558     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_COMMIT_SOURCE, errp)) {
2559         goto out;
2560     }
2561 
2562     /* default top_bs is the active layer */
2563     top_bs = bs;
2564 
2565     if (top_node && top) {
2566         error_setg(errp, "'top-node' and 'top' are mutually exclusive");
2567         goto out;
2568     } else if (top_node) {
2569         top_bs = bdrv_lookup_bs(NULL, top_node, errp);
2570         if (top_bs == NULL) {
2571             goto out;
2572         }
2573         if (!bdrv_chain_contains(bs, top_bs)) {
2574             error_setg(errp, "'%s' is not in this backing file chain",
2575                        top_node);
2576             goto out;
2577         }
2578     } else if (top) {
2579         /* This strcmp() is just a shortcut, there is no need to
2580          * refresh @bs's filename.  If it mismatches,
2581          * bdrv_find_backing_image() will do the refresh and may still
2582          * return @bs. */
2583         if (strcmp(bs->filename, top) != 0) {
2584             top_bs = bdrv_find_backing_image(bs, top);
2585         }
2586     }
2587 
2588     if (top_bs == NULL) {
2589         error_setg(errp, "Top image file %s not found", top ? top : "NULL");
2590         goto out;
2591     }
2592 
2593     assert(bdrv_get_aio_context(top_bs) == aio_context);
2594 
2595     if (base_node && base) {
2596         error_setg(errp, "'base-node' and 'base' are mutually exclusive");
2597         goto out;
2598     } else if (base_node) {
2599         base_bs = bdrv_lookup_bs(NULL, base_node, errp);
2600         if (base_bs == NULL) {
2601             goto out;
2602         }
2603         if (!bdrv_chain_contains(top_bs, base_bs)) {
2604             error_setg(errp, "'%s' is not in this backing file chain",
2605                        base_node);
2606             goto out;
2607         }
2608     } else if (base) {
2609         base_bs = bdrv_find_backing_image(top_bs, base);
2610         if (base_bs == NULL) {
2611             error_setg(errp, "Can't find '%s' in the backing chain", base);
2612             goto out;
2613         }
2614     } else {
2615         base_bs = bdrv_find_base(top_bs);
2616         if (base_bs == NULL) {
2617             error_setg(errp, "There is no backimg image");
2618             goto out;
2619         }
2620     }
2621 
2622     assert(bdrv_get_aio_context(base_bs) == aio_context);
2623 
2624     for (iter = top_bs; iter != bdrv_filter_or_cow_bs(base_bs);
2625          iter = bdrv_filter_or_cow_bs(iter))
2626     {
2627         if (bdrv_op_is_blocked(iter, BLOCK_OP_TYPE_COMMIT_TARGET, errp)) {
2628             goto out;
2629         }
2630     }
2631 
2632     /* Do not allow attempts to commit an image into itself */
2633     if (top_bs == base_bs) {
2634         error_setg(errp, "cannot commit an image into itself");
2635         goto out;
2636     }
2637 
2638     /*
2639      * Active commit is required if and only if someone has taken a
2640      * WRITE permission on the top node.  Historically, we have always
2641      * used active commit for top nodes, so continue that practice
2642      * lest we possibly break clients that rely on this behavior, e.g.
2643      * to later attach this node to a writing parent.
2644      * (Active commit is never really wrong.)
2645      */
2646     bdrv_get_cumulative_perm(top_bs, &top_perm, &top_shared);
2647     if (top_perm & BLK_PERM_WRITE ||
2648         bdrv_skip_filters(top_bs) == bdrv_skip_filters(bs))
2649     {
2650         if (backing_file) {
2651             if (bdrv_skip_filters(top_bs) == bdrv_skip_filters(bs)) {
2652                 error_setg(errp, "'backing-file' specified,"
2653                                  " but 'top' is the active layer");
2654             } else {
2655                 error_setg(errp, "'backing-file' specified, but 'top' has a "
2656                                  "writer on it");
2657             }
2658             goto out;
2659         }
2660         if (!job_id) {
2661             /*
2662              * Emulate here what block_job_create() does, because it
2663              * is possible that @bs != @top_bs (the block job should
2664              * be named after @bs, even if @top_bs is the actual
2665              * source)
2666              */
2667             job_id = bdrv_get_device_name(bs);
2668         }
2669         commit_active_start(job_id, top_bs, base_bs, job_flags, speed, on_error,
2670                             filter_node_name, NULL, NULL, false, &local_err);
2671     } else {
2672         BlockDriverState *overlay_bs = bdrv_find_overlay(bs, top_bs);
2673         if (bdrv_op_is_blocked(overlay_bs, BLOCK_OP_TYPE_COMMIT_TARGET, errp)) {
2674             goto out;
2675         }
2676         commit_start(job_id, bs, base_bs, top_bs, job_flags,
2677                      speed, on_error, backing_file,
2678                      filter_node_name, &local_err);
2679     }
2680     if (local_err != NULL) {
2681         error_propagate(errp, local_err);
2682         goto out;
2683     }
2684 
2685 out:
2686     aio_context_release(aio_context);
2687 }
2688 
2689 /* Common QMP interface for drive-backup and blockdev-backup */
2690 static BlockJob *do_backup_common(BackupCommon *backup,
2691                                   BlockDriverState *bs,
2692                                   BlockDriverState *target_bs,
2693                                   AioContext *aio_context,
2694                                   JobTxn *txn, Error **errp)
2695 {
2696     BlockJob *job = NULL;
2697     BdrvDirtyBitmap *bmap = NULL;
2698     BackupPerf perf = { .max_workers = 64 };
2699     int job_flags = JOB_DEFAULT;
2700 
2701     if (!backup->has_speed) {
2702         backup->speed = 0;
2703     }
2704     if (!backup->has_on_source_error) {
2705         backup->on_source_error = BLOCKDEV_ON_ERROR_REPORT;
2706     }
2707     if (!backup->has_on_target_error) {
2708         backup->on_target_error = BLOCKDEV_ON_ERROR_REPORT;
2709     }
2710     if (!backup->has_auto_finalize) {
2711         backup->auto_finalize = true;
2712     }
2713     if (!backup->has_auto_dismiss) {
2714         backup->auto_dismiss = true;
2715     }
2716     if (!backup->has_compress) {
2717         backup->compress = false;
2718     }
2719 
2720     if (backup->x_perf) {
2721         if (backup->x_perf->has_use_copy_range) {
2722             perf.use_copy_range = backup->x_perf->use_copy_range;
2723         }
2724         if (backup->x_perf->has_max_workers) {
2725             perf.max_workers = backup->x_perf->max_workers;
2726         }
2727         if (backup->x_perf->has_max_chunk) {
2728             perf.max_chunk = backup->x_perf->max_chunk;
2729         }
2730     }
2731 
2732     if ((backup->sync == MIRROR_SYNC_MODE_BITMAP) ||
2733         (backup->sync == MIRROR_SYNC_MODE_INCREMENTAL)) {
2734         /* done before desugaring 'incremental' to print the right message */
2735         if (!backup->bitmap) {
2736             error_setg(errp, "must provide a valid bitmap name for "
2737                        "'%s' sync mode", MirrorSyncMode_str(backup->sync));
2738             return NULL;
2739         }
2740     }
2741 
2742     if (backup->sync == MIRROR_SYNC_MODE_INCREMENTAL) {
2743         if (backup->has_bitmap_mode &&
2744             backup->bitmap_mode != BITMAP_SYNC_MODE_ON_SUCCESS) {
2745             error_setg(errp, "Bitmap sync mode must be '%s' "
2746                        "when using sync mode '%s'",
2747                        BitmapSyncMode_str(BITMAP_SYNC_MODE_ON_SUCCESS),
2748                        MirrorSyncMode_str(backup->sync));
2749             return NULL;
2750         }
2751         backup->has_bitmap_mode = true;
2752         backup->sync = MIRROR_SYNC_MODE_BITMAP;
2753         backup->bitmap_mode = BITMAP_SYNC_MODE_ON_SUCCESS;
2754     }
2755 
2756     if (backup->bitmap) {
2757         bmap = bdrv_find_dirty_bitmap(bs, backup->bitmap);
2758         if (!bmap) {
2759             error_setg(errp, "Bitmap '%s' could not be found", backup->bitmap);
2760             return NULL;
2761         }
2762         if (!backup->has_bitmap_mode) {
2763             error_setg(errp, "Bitmap sync mode must be given "
2764                        "when providing a bitmap");
2765             return NULL;
2766         }
2767         if (bdrv_dirty_bitmap_check(bmap, BDRV_BITMAP_ALLOW_RO, errp)) {
2768             return NULL;
2769         }
2770 
2771         /* This does not produce a useful bitmap artifact: */
2772         if (backup->sync == MIRROR_SYNC_MODE_NONE) {
2773             error_setg(errp, "sync mode '%s' does not produce meaningful bitmap"
2774                        " outputs", MirrorSyncMode_str(backup->sync));
2775             return NULL;
2776         }
2777 
2778         /* If the bitmap isn't used for input or output, this is useless: */
2779         if (backup->bitmap_mode == BITMAP_SYNC_MODE_NEVER &&
2780             backup->sync != MIRROR_SYNC_MODE_BITMAP) {
2781             error_setg(errp, "Bitmap sync mode '%s' has no meaningful effect"
2782                        " when combined with sync mode '%s'",
2783                        BitmapSyncMode_str(backup->bitmap_mode),
2784                        MirrorSyncMode_str(backup->sync));
2785             return NULL;
2786         }
2787     }
2788 
2789     if (!backup->bitmap && backup->has_bitmap_mode) {
2790         error_setg(errp, "Cannot specify bitmap sync mode without a bitmap");
2791         return NULL;
2792     }
2793 
2794     if (!backup->auto_finalize) {
2795         job_flags |= JOB_MANUAL_FINALIZE;
2796     }
2797     if (!backup->auto_dismiss) {
2798         job_flags |= JOB_MANUAL_DISMISS;
2799     }
2800 
2801     job = backup_job_create(backup->job_id, bs, target_bs, backup->speed,
2802                             backup->sync, bmap, backup->bitmap_mode,
2803                             backup->compress,
2804                             backup->filter_node_name,
2805                             &perf,
2806                             backup->on_source_error,
2807                             backup->on_target_error,
2808                             job_flags, NULL, NULL, txn, errp);
2809     return job;
2810 }
2811 
2812 void qmp_drive_backup(DriveBackup *backup, Error **errp)
2813 {
2814     TransactionAction action = {
2815         .type = TRANSACTION_ACTION_KIND_DRIVE_BACKUP,
2816         .u.drive_backup.data = backup,
2817     };
2818     blockdev_do_action(&action, errp);
2819 }
2820 
2821 BlockDeviceInfoList *qmp_query_named_block_nodes(bool has_flat,
2822                                                  bool flat,
2823                                                  Error **errp)
2824 {
2825     bool return_flat = has_flat && flat;
2826 
2827     return bdrv_named_nodes_list(return_flat, errp);
2828 }
2829 
2830 XDbgBlockGraph *qmp_x_debug_query_block_graph(Error **errp)
2831 {
2832     return bdrv_get_xdbg_block_graph(errp);
2833 }
2834 
2835 void qmp_blockdev_backup(BlockdevBackup *backup, Error **errp)
2836 {
2837     TransactionAction action = {
2838         .type = TRANSACTION_ACTION_KIND_BLOCKDEV_BACKUP,
2839         .u.blockdev_backup.data = backup,
2840     };
2841     blockdev_do_action(&action, errp);
2842 }
2843 
2844 /* Parameter check and block job starting for drive mirroring.
2845  * Caller should hold @device and @target's aio context (must be the same).
2846  **/
2847 static void blockdev_mirror_common(const char *job_id, BlockDriverState *bs,
2848                                    BlockDriverState *target,
2849                                    const char *replaces,
2850                                    enum MirrorSyncMode sync,
2851                                    BlockMirrorBackingMode backing_mode,
2852                                    bool zero_target,
2853                                    bool has_speed, int64_t speed,
2854                                    bool has_granularity, uint32_t granularity,
2855                                    bool has_buf_size, int64_t buf_size,
2856                                    bool has_on_source_error,
2857                                    BlockdevOnError on_source_error,
2858                                    bool has_on_target_error,
2859                                    BlockdevOnError on_target_error,
2860                                    bool has_unmap, bool unmap,
2861                                    const char *filter_node_name,
2862                                    bool has_copy_mode, MirrorCopyMode copy_mode,
2863                                    bool has_auto_finalize, bool auto_finalize,
2864                                    bool has_auto_dismiss, bool auto_dismiss,
2865                                    Error **errp)
2866 {
2867     BlockDriverState *unfiltered_bs;
2868     int job_flags = JOB_DEFAULT;
2869 
2870     GLOBAL_STATE_CODE();
2871     GRAPH_RDLOCK_GUARD_MAINLOOP();
2872 
2873     if (!has_speed) {
2874         speed = 0;
2875     }
2876     if (!has_on_source_error) {
2877         on_source_error = BLOCKDEV_ON_ERROR_REPORT;
2878     }
2879     if (!has_on_target_error) {
2880         on_target_error = BLOCKDEV_ON_ERROR_REPORT;
2881     }
2882     if (!has_granularity) {
2883         granularity = 0;
2884     }
2885     if (!has_buf_size) {
2886         buf_size = 0;
2887     }
2888     if (!has_unmap) {
2889         unmap = true;
2890     }
2891     if (!has_copy_mode) {
2892         copy_mode = MIRROR_COPY_MODE_BACKGROUND;
2893     }
2894     if (has_auto_finalize && !auto_finalize) {
2895         job_flags |= JOB_MANUAL_FINALIZE;
2896     }
2897     if (has_auto_dismiss && !auto_dismiss) {
2898         job_flags |= JOB_MANUAL_DISMISS;
2899     }
2900 
2901     if (granularity != 0 && (granularity < 512 || granularity > 1048576 * 64)) {
2902         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "granularity",
2903                    "a value in range [512B, 64MB]");
2904         return;
2905     }
2906     if (granularity & (granularity - 1)) {
2907         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "granularity",
2908                    "a power of 2");
2909         return;
2910     }
2911 
2912     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_MIRROR_SOURCE, errp)) {
2913         return;
2914     }
2915     if (bdrv_op_is_blocked(target, BLOCK_OP_TYPE_MIRROR_TARGET, errp)) {
2916         return;
2917     }
2918 
2919     if (!bdrv_backing_chain_next(bs) && sync == MIRROR_SYNC_MODE_TOP) {
2920         sync = MIRROR_SYNC_MODE_FULL;
2921     }
2922 
2923     if (!replaces) {
2924         /* We want to mirror from @bs, but keep implicit filters on top */
2925         unfiltered_bs = bdrv_skip_implicit_filters(bs);
2926         if (unfiltered_bs != bs) {
2927             replaces = unfiltered_bs->node_name;
2928         }
2929     }
2930 
2931     if (replaces) {
2932         BlockDriverState *to_replace_bs;
2933         AioContext *replace_aio_context;
2934         int64_t bs_size, replace_size;
2935 
2936         bs_size = bdrv_getlength(bs);
2937         if (bs_size < 0) {
2938             error_setg_errno(errp, -bs_size, "Failed to query device's size");
2939             return;
2940         }
2941 
2942         to_replace_bs = check_to_replace_node(bs, replaces, errp);
2943         if (!to_replace_bs) {
2944             return;
2945         }
2946 
2947         replace_aio_context = bdrv_get_aio_context(to_replace_bs);
2948         aio_context_acquire(replace_aio_context);
2949         replace_size = bdrv_getlength(to_replace_bs);
2950         aio_context_release(replace_aio_context);
2951 
2952         if (replace_size < 0) {
2953             error_setg_errno(errp, -replace_size,
2954                              "Failed to query the replacement node's size");
2955             return;
2956         }
2957         if (bs_size != replace_size) {
2958             error_setg(errp, "cannot replace image with a mirror image of "
2959                              "different size");
2960             return;
2961         }
2962     }
2963 
2964     /* pass the node name to replace to mirror start since it's loose coupling
2965      * and will allow to check whether the node still exist at mirror completion
2966      */
2967     mirror_start(job_id, bs, target,
2968                  replaces, job_flags,
2969                  speed, granularity, buf_size, sync, backing_mode, zero_target,
2970                  on_source_error, on_target_error, unmap, filter_node_name,
2971                  copy_mode, errp);
2972 }
2973 
2974 void qmp_drive_mirror(DriveMirror *arg, Error **errp)
2975 {
2976     BlockDriverState *bs;
2977     BlockDriverState *target_backing_bs, *target_bs;
2978     AioContext *aio_context;
2979     AioContext *old_context;
2980     BlockMirrorBackingMode backing_mode;
2981     Error *local_err = NULL;
2982     QDict *options = NULL;
2983     int flags;
2984     int64_t size;
2985     const char *format = arg->format;
2986     bool zero_target;
2987     int ret;
2988 
2989     bs = qmp_get_root_bs(arg->device, errp);
2990     if (!bs) {
2991         return;
2992     }
2993 
2994     /* Early check to avoid creating target */
2995     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_MIRROR_SOURCE, errp)) {
2996         return;
2997     }
2998 
2999     aio_context = bdrv_get_aio_context(bs);
3000     aio_context_acquire(aio_context);
3001 
3002     if (!arg->has_mode) {
3003         arg->mode = NEW_IMAGE_MODE_ABSOLUTE_PATHS;
3004     }
3005 
3006     if (!arg->format) {
3007         format = (arg->mode == NEW_IMAGE_MODE_EXISTING
3008                   ? NULL : bs->drv->format_name);
3009     }
3010 
3011     flags = bs->open_flags | BDRV_O_RDWR;
3012     target_backing_bs = bdrv_cow_bs(bdrv_skip_filters(bs));
3013     if (!target_backing_bs && arg->sync == MIRROR_SYNC_MODE_TOP) {
3014         arg->sync = MIRROR_SYNC_MODE_FULL;
3015     }
3016     if (arg->sync == MIRROR_SYNC_MODE_NONE) {
3017         target_backing_bs = bs;
3018     }
3019 
3020     size = bdrv_getlength(bs);
3021     if (size < 0) {
3022         error_setg_errno(errp, -size, "bdrv_getlength failed");
3023         goto out;
3024     }
3025 
3026     if (arg->replaces) {
3027         if (!arg->node_name) {
3028             error_setg(errp, "a node-name must be provided when replacing a"
3029                              " named node of the graph");
3030             goto out;
3031         }
3032     }
3033 
3034     if (arg->mode == NEW_IMAGE_MODE_ABSOLUTE_PATHS) {
3035         backing_mode = MIRROR_SOURCE_BACKING_CHAIN;
3036     } else {
3037         backing_mode = MIRROR_OPEN_BACKING_CHAIN;
3038     }
3039 
3040     /* Don't open backing image in create() */
3041     flags |= BDRV_O_NO_BACKING;
3042 
3043     if ((arg->sync == MIRROR_SYNC_MODE_FULL || !target_backing_bs)
3044         && arg->mode != NEW_IMAGE_MODE_EXISTING)
3045     {
3046         /* create new image w/o backing file */
3047         assert(format);
3048         bdrv_img_create(arg->target, format,
3049                         NULL, NULL, NULL, size, flags, false, &local_err);
3050     } else {
3051         /* Implicit filters should not appear in the filename */
3052         BlockDriverState *explicit_backing =
3053             bdrv_skip_implicit_filters(target_backing_bs);
3054 
3055         switch (arg->mode) {
3056         case NEW_IMAGE_MODE_EXISTING:
3057             break;
3058         case NEW_IMAGE_MODE_ABSOLUTE_PATHS:
3059             /* create new image with backing file */
3060             bdrv_refresh_filename(explicit_backing);
3061             bdrv_img_create(arg->target, format,
3062                             explicit_backing->filename,
3063                             explicit_backing->drv->format_name,
3064                             NULL, size, flags, false, &local_err);
3065             break;
3066         default:
3067             abort();
3068         }
3069     }
3070 
3071     if (local_err) {
3072         error_propagate(errp, local_err);
3073         goto out;
3074     }
3075 
3076     options = qdict_new();
3077     if (arg->node_name) {
3078         qdict_put_str(options, "node-name", arg->node_name);
3079     }
3080     if (format) {
3081         qdict_put_str(options, "driver", format);
3082     }
3083     aio_context_release(aio_context);
3084 
3085     /* Mirroring takes care of copy-on-write using the source's backing
3086      * file.
3087      */
3088     aio_context_acquire(qemu_get_aio_context());
3089     target_bs = bdrv_open(arg->target, NULL, options, flags, errp);
3090     aio_context_release(qemu_get_aio_context());
3091 
3092     if (!target_bs) {
3093         return;
3094     }
3095 
3096     zero_target = (arg->sync == MIRROR_SYNC_MODE_FULL &&
3097                    (arg->mode == NEW_IMAGE_MODE_EXISTING ||
3098                     !bdrv_has_zero_init(target_bs)));
3099 
3100 
3101     /* Honor bdrv_try_change_aio_context() context acquisition requirements. */
3102     old_context = bdrv_get_aio_context(target_bs);
3103     aio_context_acquire(old_context);
3104 
3105     ret = bdrv_try_change_aio_context(target_bs, aio_context, NULL, errp);
3106     if (ret < 0) {
3107         bdrv_unref(target_bs);
3108         aio_context_release(old_context);
3109         return;
3110     }
3111 
3112     aio_context_release(old_context);
3113     aio_context_acquire(aio_context);
3114 
3115     blockdev_mirror_common(arg->job_id, bs, target_bs,
3116                            arg->replaces, arg->sync,
3117                            backing_mode, zero_target,
3118                            arg->has_speed, arg->speed,
3119                            arg->has_granularity, arg->granularity,
3120                            arg->has_buf_size, arg->buf_size,
3121                            arg->has_on_source_error, arg->on_source_error,
3122                            arg->has_on_target_error, arg->on_target_error,
3123                            arg->has_unmap, arg->unmap,
3124                            NULL,
3125                            arg->has_copy_mode, arg->copy_mode,
3126                            arg->has_auto_finalize, arg->auto_finalize,
3127                            arg->has_auto_dismiss, arg->auto_dismiss,
3128                            errp);
3129     bdrv_unref(target_bs);
3130 out:
3131     aio_context_release(aio_context);
3132 }
3133 
3134 void qmp_blockdev_mirror(const char *job_id,
3135                          const char *device, const char *target,
3136                          const char *replaces,
3137                          MirrorSyncMode sync,
3138                          bool has_speed, int64_t speed,
3139                          bool has_granularity, uint32_t granularity,
3140                          bool has_buf_size, int64_t buf_size,
3141                          bool has_on_source_error,
3142                          BlockdevOnError on_source_error,
3143                          bool has_on_target_error,
3144                          BlockdevOnError on_target_error,
3145                          const char *filter_node_name,
3146                          bool has_copy_mode, MirrorCopyMode copy_mode,
3147                          bool has_auto_finalize, bool auto_finalize,
3148                          bool has_auto_dismiss, bool auto_dismiss,
3149                          Error **errp)
3150 {
3151     BlockDriverState *bs;
3152     BlockDriverState *target_bs;
3153     AioContext *aio_context;
3154     AioContext *old_context;
3155     BlockMirrorBackingMode backing_mode = MIRROR_LEAVE_BACKING_CHAIN;
3156     bool zero_target;
3157     int ret;
3158 
3159     bs = qmp_get_root_bs(device, errp);
3160     if (!bs) {
3161         return;
3162     }
3163 
3164     target_bs = bdrv_lookup_bs(target, target, errp);
3165     if (!target_bs) {
3166         return;
3167     }
3168 
3169     zero_target = (sync == MIRROR_SYNC_MODE_FULL);
3170 
3171     /* Honor bdrv_try_change_aio_context() context acquisition requirements. */
3172     old_context = bdrv_get_aio_context(target_bs);
3173     aio_context = bdrv_get_aio_context(bs);
3174     aio_context_acquire(old_context);
3175 
3176     ret = bdrv_try_change_aio_context(target_bs, aio_context, NULL, errp);
3177 
3178     aio_context_release(old_context);
3179     aio_context_acquire(aio_context);
3180 
3181     if (ret < 0) {
3182         goto out;
3183     }
3184 
3185     blockdev_mirror_common(job_id, bs, target_bs,
3186                            replaces, sync, backing_mode,
3187                            zero_target, has_speed, speed,
3188                            has_granularity, granularity,
3189                            has_buf_size, buf_size,
3190                            has_on_source_error, on_source_error,
3191                            has_on_target_error, on_target_error,
3192                            true, true, filter_node_name,
3193                            has_copy_mode, copy_mode,
3194                            has_auto_finalize, auto_finalize,
3195                            has_auto_dismiss, auto_dismiss,
3196                            errp);
3197 out:
3198     aio_context_release(aio_context);
3199 }
3200 
3201 /*
3202  * Get a block job using its ID. Called with job_mutex held.
3203  */
3204 static BlockJob *find_block_job_locked(const char *id, Error **errp)
3205 {
3206     BlockJob *job;
3207 
3208     assert(id != NULL);
3209 
3210     job = block_job_get_locked(id);
3211 
3212     if (!job) {
3213         error_set(errp, ERROR_CLASS_DEVICE_NOT_ACTIVE,
3214                   "Block job '%s' not found", id);
3215         return NULL;
3216     }
3217 
3218     return job;
3219 }
3220 
3221 void qmp_block_job_set_speed(const char *device, int64_t speed, Error **errp)
3222 {
3223     BlockJob *job;
3224 
3225     JOB_LOCK_GUARD();
3226     job = find_block_job_locked(device, errp);
3227 
3228     if (!job) {
3229         return;
3230     }
3231 
3232     block_job_set_speed_locked(job, speed, errp);
3233 }
3234 
3235 void qmp_block_job_cancel(const char *device,
3236                           bool has_force, bool force, Error **errp)
3237 {
3238     BlockJob *job;
3239 
3240     JOB_LOCK_GUARD();
3241     job = find_block_job_locked(device, errp);
3242 
3243     if (!job) {
3244         return;
3245     }
3246 
3247     if (!has_force) {
3248         force = false;
3249     }
3250 
3251     if (job_user_paused_locked(&job->job) && !force) {
3252         error_setg(errp, "The block job for device '%s' is currently paused",
3253                    device);
3254         return;
3255     }
3256 
3257     trace_qmp_block_job_cancel(job);
3258     job_user_cancel_locked(&job->job, force, errp);
3259 }
3260 
3261 void qmp_block_job_pause(const char *device, Error **errp)
3262 {
3263     BlockJob *job;
3264 
3265     JOB_LOCK_GUARD();
3266     job = find_block_job_locked(device, errp);
3267 
3268     if (!job) {
3269         return;
3270     }
3271 
3272     trace_qmp_block_job_pause(job);
3273     job_user_pause_locked(&job->job, errp);
3274 }
3275 
3276 void qmp_block_job_resume(const char *device, Error **errp)
3277 {
3278     BlockJob *job;
3279 
3280     JOB_LOCK_GUARD();
3281     job = find_block_job_locked(device, errp);
3282 
3283     if (!job) {
3284         return;
3285     }
3286 
3287     trace_qmp_block_job_resume(job);
3288     job_user_resume_locked(&job->job, errp);
3289 }
3290 
3291 void qmp_block_job_complete(const char *device, Error **errp)
3292 {
3293     BlockJob *job;
3294 
3295     JOB_LOCK_GUARD();
3296     job = find_block_job_locked(device, errp);
3297 
3298     if (!job) {
3299         return;
3300     }
3301 
3302     trace_qmp_block_job_complete(job);
3303     job_complete_locked(&job->job, errp);
3304 }
3305 
3306 void qmp_block_job_finalize(const char *id, Error **errp)
3307 {
3308     BlockJob *job;
3309 
3310     JOB_LOCK_GUARD();
3311     job = find_block_job_locked(id, errp);
3312 
3313     if (!job) {
3314         return;
3315     }
3316 
3317     trace_qmp_block_job_finalize(job);
3318     job_ref_locked(&job->job);
3319     job_finalize_locked(&job->job, errp);
3320 
3321     job_unref_locked(&job->job);
3322 }
3323 
3324 void qmp_block_job_dismiss(const char *id, Error **errp)
3325 {
3326     BlockJob *bjob;
3327     Job *job;
3328 
3329     JOB_LOCK_GUARD();
3330     bjob = find_block_job_locked(id, errp);
3331 
3332     if (!bjob) {
3333         return;
3334     }
3335 
3336     trace_qmp_block_job_dismiss(bjob);
3337     job = &bjob->job;
3338     job_dismiss_locked(&job, errp);
3339 }
3340 
3341 void qmp_change_backing_file(const char *device,
3342                              const char *image_node_name,
3343                              const char *backing_file,
3344                              Error **errp)
3345 {
3346     BlockDriverState *bs = NULL;
3347     AioContext *aio_context;
3348     BlockDriverState *image_bs = NULL;
3349     Error *local_err = NULL;
3350     bool ro;
3351     int ret;
3352 
3353     bs = qmp_get_root_bs(device, errp);
3354     if (!bs) {
3355         return;
3356     }
3357 
3358     aio_context = bdrv_get_aio_context(bs);
3359     aio_context_acquire(aio_context);
3360 
3361     image_bs = bdrv_lookup_bs(NULL, image_node_name, &local_err);
3362     if (local_err) {
3363         error_propagate(errp, local_err);
3364         goto out;
3365     }
3366 
3367     if (!image_bs) {
3368         error_setg(errp, "image file not found");
3369         goto out;
3370     }
3371 
3372     if (bdrv_find_base(image_bs) == image_bs) {
3373         error_setg(errp, "not allowing backing file change on an image "
3374                          "without a backing file");
3375         goto out;
3376     }
3377 
3378     /* even though we are not necessarily operating on bs, we need it to
3379      * determine if block ops are currently prohibited on the chain */
3380     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_CHANGE, errp)) {
3381         goto out;
3382     }
3383 
3384     /* final sanity check */
3385     if (!bdrv_chain_contains(bs, image_bs)) {
3386         error_setg(errp, "'%s' and image file are not in the same chain",
3387                    device);
3388         goto out;
3389     }
3390 
3391     /* if not r/w, reopen to make r/w */
3392     ro = bdrv_is_read_only(image_bs);
3393 
3394     if (ro) {
3395         if (bdrv_reopen_set_read_only(image_bs, false, errp) != 0) {
3396             goto out;
3397         }
3398     }
3399 
3400     ret = bdrv_change_backing_file(image_bs, backing_file,
3401                                    image_bs->drv ? image_bs->drv->format_name : "",
3402                                    false);
3403 
3404     if (ret < 0) {
3405         error_setg_errno(errp, -ret, "Could not change backing file to '%s'",
3406                          backing_file);
3407         /* don't exit here, so we can try to restore open flags if
3408          * appropriate */
3409     }
3410 
3411     if (ro) {
3412         bdrv_reopen_set_read_only(image_bs, true, errp);
3413     }
3414 
3415 out:
3416     aio_context_release(aio_context);
3417 }
3418 
3419 void qmp_blockdev_add(BlockdevOptions *options, Error **errp)
3420 {
3421     BlockDriverState *bs;
3422     QObject *obj;
3423     Visitor *v = qobject_output_visitor_new(&obj);
3424     QDict *qdict;
3425 
3426     visit_type_BlockdevOptions(v, NULL, &options, &error_abort);
3427     visit_complete(v, &obj);
3428     qdict = qobject_to(QDict, obj);
3429 
3430     qdict_flatten(qdict);
3431 
3432     if (!qdict_get_try_str(qdict, "node-name")) {
3433         error_setg(errp, "'node-name' must be specified for the root node");
3434         goto fail;
3435     }
3436 
3437     bs = bds_tree_init(qdict, errp);
3438     if (!bs) {
3439         goto fail;
3440     }
3441 
3442     bdrv_set_monitor_owned(bs);
3443 
3444 fail:
3445     visit_free(v);
3446 }
3447 
3448 void qmp_blockdev_reopen(BlockdevOptionsList *reopen_list, Error **errp)
3449 {
3450     BlockReopenQueue *queue = NULL;
3451 
3452     /* Add each one of the BDS that we want to reopen to the queue */
3453     for (; reopen_list != NULL; reopen_list = reopen_list->next) {
3454         BlockdevOptions *options = reopen_list->value;
3455         BlockDriverState *bs;
3456         AioContext *ctx;
3457         QObject *obj;
3458         Visitor *v;
3459         QDict *qdict;
3460 
3461         /* Check for the selected node name */
3462         if (!options->node_name) {
3463             error_setg(errp, "node-name not specified");
3464             goto fail;
3465         }
3466 
3467         bs = bdrv_find_node(options->node_name);
3468         if (!bs) {
3469             error_setg(errp, "Failed to find node with node-name='%s'",
3470                        options->node_name);
3471             goto fail;
3472         }
3473 
3474         /* Put all options in a QDict and flatten it */
3475         v = qobject_output_visitor_new(&obj);
3476         visit_type_BlockdevOptions(v, NULL, &options, &error_abort);
3477         visit_complete(v, &obj);
3478         visit_free(v);
3479 
3480         qdict = qobject_to(QDict, obj);
3481 
3482         qdict_flatten(qdict);
3483 
3484         ctx = bdrv_get_aio_context(bs);
3485         aio_context_acquire(ctx);
3486 
3487         queue = bdrv_reopen_queue(queue, bs, qdict, false);
3488 
3489         aio_context_release(ctx);
3490     }
3491 
3492     /* Perform the reopen operation */
3493     bdrv_reopen_multiple(queue, errp);
3494     queue = NULL;
3495 
3496 fail:
3497     bdrv_reopen_queue_free(queue);
3498 }
3499 
3500 void qmp_blockdev_del(const char *node_name, Error **errp)
3501 {
3502     AioContext *aio_context;
3503     BlockDriverState *bs;
3504 
3505     GLOBAL_STATE_CODE();
3506 
3507     bs = bdrv_find_node(node_name);
3508     if (!bs) {
3509         error_setg(errp, "Failed to find node with node-name='%s'", node_name);
3510         return;
3511     }
3512     if (bdrv_has_blk(bs)) {
3513         error_setg(errp, "Node %s is in use", node_name);
3514         return;
3515     }
3516     aio_context = bdrv_get_aio_context(bs);
3517     aio_context_acquire(aio_context);
3518 
3519     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_DRIVE_DEL, errp)) {
3520         goto out;
3521     }
3522 
3523     if (!QTAILQ_IN_USE(bs, monitor_list)) {
3524         error_setg(errp, "Node %s is not owned by the monitor",
3525                    bs->node_name);
3526         goto out;
3527     }
3528 
3529     if (bs->refcnt > 1) {
3530         error_setg(errp, "Block device %s is in use",
3531                    bdrv_get_device_or_node_name(bs));
3532         goto out;
3533     }
3534 
3535     QTAILQ_REMOVE(&monitor_bdrv_states, bs, monitor_list);
3536     bdrv_unref(bs);
3537 
3538 out:
3539     aio_context_release(aio_context);
3540 }
3541 
3542 static BdrvChild *bdrv_find_child(BlockDriverState *parent_bs,
3543                                   const char *child_name)
3544 {
3545     BdrvChild *child;
3546 
3547     QLIST_FOREACH(child, &parent_bs->children, next) {
3548         if (strcmp(child->name, child_name) == 0) {
3549             return child;
3550         }
3551     }
3552 
3553     return NULL;
3554 }
3555 
3556 void qmp_x_blockdev_change(const char *parent, const char *child,
3557                            const char *node, Error **errp)
3558 {
3559     BlockDriverState *parent_bs, *new_bs = NULL;
3560     BdrvChild *p_child;
3561 
3562     parent_bs = bdrv_lookup_bs(parent, parent, errp);
3563     if (!parent_bs) {
3564         return;
3565     }
3566 
3567     if (!child == !node) {
3568         if (child) {
3569             error_setg(errp, "The parameters child and node are in conflict");
3570         } else {
3571             error_setg(errp, "Either child or node must be specified");
3572         }
3573         return;
3574     }
3575 
3576     if (child) {
3577         p_child = bdrv_find_child(parent_bs, child);
3578         if (!p_child) {
3579             error_setg(errp, "Node '%s' does not have child '%s'",
3580                        parent, child);
3581             return;
3582         }
3583         bdrv_del_child(parent_bs, p_child, errp);
3584     }
3585 
3586     if (node) {
3587         new_bs = bdrv_find_node(node);
3588         if (!new_bs) {
3589             error_setg(errp, "Node '%s' not found", node);
3590             return;
3591         }
3592         bdrv_add_child(parent_bs, new_bs, errp);
3593     }
3594 }
3595 
3596 BlockJobInfoList *qmp_query_block_jobs(Error **errp)
3597 {
3598     BlockJobInfoList *head = NULL, **tail = &head;
3599     BlockJob *job;
3600 
3601     JOB_LOCK_GUARD();
3602 
3603     for (job = block_job_next_locked(NULL); job;
3604          job = block_job_next_locked(job)) {
3605         BlockJobInfo *value;
3606 
3607         if (block_job_is_internal(job)) {
3608             continue;
3609         }
3610         value = block_job_query_locked(job, errp);
3611         if (!value) {
3612             qapi_free_BlockJobInfoList(head);
3613             return NULL;
3614         }
3615         QAPI_LIST_APPEND(tail, value);
3616     }
3617 
3618     return head;
3619 }
3620 
3621 void qmp_x_blockdev_set_iothread(const char *node_name, StrOrNull *iothread,
3622                                  bool has_force, bool force, Error **errp)
3623 {
3624     AioContext *old_context;
3625     AioContext *new_context;
3626     BlockDriverState *bs;
3627 
3628     bs = bdrv_find_node(node_name);
3629     if (!bs) {
3630         error_setg(errp, "Failed to find node with node-name='%s'", node_name);
3631         return;
3632     }
3633 
3634     /* Protects against accidents. */
3635     if (!(has_force && force) && bdrv_has_blk(bs)) {
3636         error_setg(errp, "Node %s is associated with a BlockBackend and could "
3637                          "be in use (use force=true to override this check)",
3638                          node_name);
3639         return;
3640     }
3641 
3642     if (iothread->type == QTYPE_QSTRING) {
3643         IOThread *obj = iothread_by_id(iothread->u.s);
3644         if (!obj) {
3645             error_setg(errp, "Cannot find iothread %s", iothread->u.s);
3646             return;
3647         }
3648 
3649         new_context = iothread_get_aio_context(obj);
3650     } else {
3651         new_context = qemu_get_aio_context();
3652     }
3653 
3654     old_context = bdrv_get_aio_context(bs);
3655     aio_context_acquire(old_context);
3656 
3657     bdrv_try_change_aio_context(bs, new_context, NULL, errp);
3658 
3659     aio_context_release(old_context);
3660 }
3661 
3662 QemuOptsList qemu_common_drive_opts = {
3663     .name = "drive",
3664     .head = QTAILQ_HEAD_INITIALIZER(qemu_common_drive_opts.head),
3665     .desc = {
3666         {
3667             .name = "snapshot",
3668             .type = QEMU_OPT_BOOL,
3669             .help = "enable/disable snapshot mode",
3670         },{
3671             .name = "aio",
3672             .type = QEMU_OPT_STRING,
3673             .help = "host AIO implementation (threads, native, io_uring)",
3674         },{
3675             .name = BDRV_OPT_CACHE_WB,
3676             .type = QEMU_OPT_BOOL,
3677             .help = "Enable writeback mode",
3678         },{
3679             .name = "format",
3680             .type = QEMU_OPT_STRING,
3681             .help = "disk format (raw, qcow2, ...)",
3682         },{
3683             .name = "rerror",
3684             .type = QEMU_OPT_STRING,
3685             .help = "read error action",
3686         },{
3687             .name = "werror",
3688             .type = QEMU_OPT_STRING,
3689             .help = "write error action",
3690         },{
3691             .name = BDRV_OPT_READ_ONLY,
3692             .type = QEMU_OPT_BOOL,
3693             .help = "open drive file as read-only",
3694         },
3695 
3696         THROTTLE_OPTS,
3697 
3698         {
3699             .name = "throttling.group",
3700             .type = QEMU_OPT_STRING,
3701             .help = "name of the block throttling group",
3702         },{
3703             .name = "copy-on-read",
3704             .type = QEMU_OPT_BOOL,
3705             .help = "copy read data from backing file into image file",
3706         },{
3707             .name = "detect-zeroes",
3708             .type = QEMU_OPT_STRING,
3709             .help = "try to optimize zero writes (off, on, unmap)",
3710         },{
3711             .name = "stats-account-invalid",
3712             .type = QEMU_OPT_BOOL,
3713             .help = "whether to account for invalid I/O operations "
3714                     "in the statistics",
3715         },{
3716             .name = "stats-account-failed",
3717             .type = QEMU_OPT_BOOL,
3718             .help = "whether to account for failed I/O operations "
3719                     "in the statistics",
3720         },
3721         { /* end of list */ }
3722     },
3723 };
3724 
3725 QemuOptsList qemu_drive_opts = {
3726     .name = "drive",
3727     .head = QTAILQ_HEAD_INITIALIZER(qemu_drive_opts.head),
3728     .desc = {
3729         /*
3730          * no elements => accept any params
3731          * validation will happen later
3732          */
3733         { /* end of list */ }
3734     },
3735 };
3736