xref: /openbmc/qemu/blockdev.c (revision ffe98631)
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     for (job = block_job_next_locked(NULL); job;
157          job = block_job_next_locked(job)) {
158         if (block_job_has_bdrv(job, blk_bs(blk))) {
159             job_cancel_locked(&job->job, false);
160         }
161     }
162 
163     dinfo->auto_del = 1;
164 }
165 
166 void blockdev_auto_del(BlockBackend *blk)
167 {
168     DriveInfo *dinfo = blk_legacy_dinfo(blk);
169     GLOBAL_STATE_CODE();
170 
171     if (dinfo && dinfo->auto_del) {
172         monitor_remove_blk(blk);
173         blk_unref(blk);
174     }
175 }
176 
177 static int drive_index_to_bus_id(BlockInterfaceType type, int index)
178 {
179     int max_devs = if_max_devs[type];
180     return max_devs ? index / max_devs : 0;
181 }
182 
183 static int drive_index_to_unit_id(BlockInterfaceType type, int index)
184 {
185     int max_devs = if_max_devs[type];
186     return max_devs ? index % max_devs : index;
187 }
188 
189 QemuOpts *drive_add(BlockInterfaceType type, int index, const char *file,
190                     const char *optstr)
191 {
192     QemuOpts *opts;
193 
194     GLOBAL_STATE_CODE();
195 
196     opts = qemu_opts_parse_noisily(qemu_find_opts("drive"), optstr, false);
197     if (!opts) {
198         return NULL;
199     }
200     if (type != IF_DEFAULT) {
201         qemu_opt_set(opts, "if", if_name[type], &error_abort);
202     }
203     if (index >= 0) {
204         qemu_opt_set_number(opts, "index", index, &error_abort);
205     }
206     if (file)
207         qemu_opt_set(opts, "file", file, &error_abort);
208     return opts;
209 }
210 
211 DriveInfo *drive_get(BlockInterfaceType type, int bus, int unit)
212 {
213     BlockBackend *blk;
214     DriveInfo *dinfo;
215 
216     GLOBAL_STATE_CODE();
217 
218     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
219         dinfo = blk_legacy_dinfo(blk);
220         if (dinfo && dinfo->type == type
221             && dinfo->bus == bus && dinfo->unit == unit) {
222             return dinfo;
223         }
224     }
225 
226     return NULL;
227 }
228 
229 /*
230  * Check board claimed all -drive that are meant to be claimed.
231  * Fatal error if any remain unclaimed.
232  */
233 void drive_check_orphaned(void)
234 {
235     BlockBackend *blk;
236     DriveInfo *dinfo;
237     Location loc;
238     bool orphans = false;
239 
240     GLOBAL_STATE_CODE();
241 
242     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
243         dinfo = blk_legacy_dinfo(blk);
244         /*
245          * Ignore default drives, because we create certain default
246          * drives unconditionally, then leave them unclaimed.  Not the
247          * users fault.
248          * Ignore IF_VIRTIO, because it gets desugared into -device,
249          * so we can leave failing to -device.
250          * Ignore IF_NONE, because leaving unclaimed IF_NONE remains
251          * available for device_add is a feature.
252          */
253         if (dinfo->is_default || dinfo->type == IF_VIRTIO
254             || dinfo->type == IF_NONE) {
255             continue;
256         }
257         if (!blk_get_attached_dev(blk)) {
258             loc_push_none(&loc);
259             qemu_opts_loc_restore(dinfo->opts);
260             error_report("machine type does not support"
261                          " if=%s,bus=%d,unit=%d",
262                          if_name[dinfo->type], dinfo->bus, dinfo->unit);
263             loc_pop(&loc);
264             orphans = true;
265         }
266     }
267 
268     if (orphans) {
269         exit(1);
270     }
271 }
272 
273 DriveInfo *drive_get_by_index(BlockInterfaceType type, int index)
274 {
275     GLOBAL_STATE_CODE();
276     return drive_get(type,
277                      drive_index_to_bus_id(type, index),
278                      drive_index_to_unit_id(type, index));
279 }
280 
281 int drive_get_max_bus(BlockInterfaceType type)
282 {
283     int max_bus;
284     BlockBackend *blk;
285     DriveInfo *dinfo;
286 
287     GLOBAL_STATE_CODE();
288 
289     max_bus = -1;
290     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
291         dinfo = blk_legacy_dinfo(blk);
292         if (dinfo && dinfo->type == type && dinfo->bus > max_bus) {
293             max_bus = dinfo->bus;
294         }
295     }
296     return max_bus;
297 }
298 
299 static void bdrv_format_print(void *opaque, const char *name)
300 {
301     qemu_printf(" %s", name);
302 }
303 
304 typedef struct {
305     QEMUBH *bh;
306     BlockDriverState *bs;
307 } BDRVPutRefBH;
308 
309 static int parse_block_error_action(const char *buf, bool is_read, Error **errp)
310 {
311     if (!strcmp(buf, "ignore")) {
312         return BLOCKDEV_ON_ERROR_IGNORE;
313     } else if (!is_read && !strcmp(buf, "enospc")) {
314         return BLOCKDEV_ON_ERROR_ENOSPC;
315     } else if (!strcmp(buf, "stop")) {
316         return BLOCKDEV_ON_ERROR_STOP;
317     } else if (!strcmp(buf, "report")) {
318         return BLOCKDEV_ON_ERROR_REPORT;
319     } else {
320         error_setg(errp, "'%s' invalid %s error action",
321                    buf, is_read ? "read" : "write");
322         return -1;
323     }
324 }
325 
326 static bool parse_stats_intervals(BlockAcctStats *stats, QList *intervals,
327                                   Error **errp)
328 {
329     const QListEntry *entry;
330     for (entry = qlist_first(intervals); entry; entry = qlist_next(entry)) {
331         switch (qobject_type(entry->value)) {
332 
333         case QTYPE_QSTRING: {
334             unsigned long long length;
335             const char *str = qstring_get_str(qobject_to(QString,
336                                                          entry->value));
337             if (parse_uint_full(str, &length, 10) == 0 &&
338                 length > 0 && length <= UINT_MAX) {
339                 block_acct_add_interval(stats, (unsigned) length);
340             } else {
341                 error_setg(errp, "Invalid interval length: %s", str);
342                 return false;
343             }
344             break;
345         }
346 
347         case QTYPE_QNUM: {
348             int64_t length = qnum_get_int(qobject_to(QNum, entry->value));
349 
350             if (length > 0 && length <= UINT_MAX) {
351                 block_acct_add_interval(stats, (unsigned) length);
352             } else {
353                 error_setg(errp, "Invalid interval length: %" PRId64, length);
354                 return false;
355             }
356             break;
357         }
358 
359         default:
360             error_setg(errp, "The specification of stats-intervals is invalid");
361             return false;
362         }
363     }
364     return true;
365 }
366 
367 typedef enum { MEDIA_DISK, MEDIA_CDROM } DriveMediaType;
368 
369 /* All parameters but @opts are optional and may be set to NULL. */
370 static void extract_common_blockdev_options(QemuOpts *opts, int *bdrv_flags,
371     const char **throttling_group, ThrottleConfig *throttle_cfg,
372     BlockdevDetectZeroesOptions *detect_zeroes, Error **errp)
373 {
374     Error *local_error = NULL;
375     const char *aio;
376 
377     if (bdrv_flags) {
378         if (qemu_opt_get_bool(opts, "copy-on-read", false)) {
379             *bdrv_flags |= BDRV_O_COPY_ON_READ;
380         }
381 
382         if ((aio = qemu_opt_get(opts, "aio")) != NULL) {
383             if (bdrv_parse_aio(aio, bdrv_flags) < 0) {
384                 error_setg(errp, "invalid aio option");
385                 return;
386             }
387         }
388     }
389 
390     /* disk I/O throttling */
391     if (throttling_group) {
392         *throttling_group = qemu_opt_get(opts, "throttling.group");
393     }
394 
395     if (throttle_cfg) {
396         throttle_config_init(throttle_cfg);
397         throttle_cfg->buckets[THROTTLE_BPS_TOTAL].avg =
398             qemu_opt_get_number(opts, "throttling.bps-total", 0);
399         throttle_cfg->buckets[THROTTLE_BPS_READ].avg  =
400             qemu_opt_get_number(opts, "throttling.bps-read", 0);
401         throttle_cfg->buckets[THROTTLE_BPS_WRITE].avg =
402             qemu_opt_get_number(opts, "throttling.bps-write", 0);
403         throttle_cfg->buckets[THROTTLE_OPS_TOTAL].avg =
404             qemu_opt_get_number(opts, "throttling.iops-total", 0);
405         throttle_cfg->buckets[THROTTLE_OPS_READ].avg =
406             qemu_opt_get_number(opts, "throttling.iops-read", 0);
407         throttle_cfg->buckets[THROTTLE_OPS_WRITE].avg =
408             qemu_opt_get_number(opts, "throttling.iops-write", 0);
409 
410         throttle_cfg->buckets[THROTTLE_BPS_TOTAL].max =
411             qemu_opt_get_number(opts, "throttling.bps-total-max", 0);
412         throttle_cfg->buckets[THROTTLE_BPS_READ].max  =
413             qemu_opt_get_number(opts, "throttling.bps-read-max", 0);
414         throttle_cfg->buckets[THROTTLE_BPS_WRITE].max =
415             qemu_opt_get_number(opts, "throttling.bps-write-max", 0);
416         throttle_cfg->buckets[THROTTLE_OPS_TOTAL].max =
417             qemu_opt_get_number(opts, "throttling.iops-total-max", 0);
418         throttle_cfg->buckets[THROTTLE_OPS_READ].max =
419             qemu_opt_get_number(opts, "throttling.iops-read-max", 0);
420         throttle_cfg->buckets[THROTTLE_OPS_WRITE].max =
421             qemu_opt_get_number(opts, "throttling.iops-write-max", 0);
422 
423         throttle_cfg->buckets[THROTTLE_BPS_TOTAL].burst_length =
424             qemu_opt_get_number(opts, "throttling.bps-total-max-length", 1);
425         throttle_cfg->buckets[THROTTLE_BPS_READ].burst_length  =
426             qemu_opt_get_number(opts, "throttling.bps-read-max-length", 1);
427         throttle_cfg->buckets[THROTTLE_BPS_WRITE].burst_length =
428             qemu_opt_get_number(opts, "throttling.bps-write-max-length", 1);
429         throttle_cfg->buckets[THROTTLE_OPS_TOTAL].burst_length =
430             qemu_opt_get_number(opts, "throttling.iops-total-max-length", 1);
431         throttle_cfg->buckets[THROTTLE_OPS_READ].burst_length =
432             qemu_opt_get_number(opts, "throttling.iops-read-max-length", 1);
433         throttle_cfg->buckets[THROTTLE_OPS_WRITE].burst_length =
434             qemu_opt_get_number(opts, "throttling.iops-write-max-length", 1);
435 
436         throttle_cfg->op_size =
437             qemu_opt_get_number(opts, "throttling.iops-size", 0);
438 
439         if (!throttle_is_valid(throttle_cfg, errp)) {
440             return;
441         }
442     }
443 
444     if (detect_zeroes) {
445         *detect_zeroes =
446             qapi_enum_parse(&BlockdevDetectZeroesOptions_lookup,
447                             qemu_opt_get(opts, "detect-zeroes"),
448                             BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF,
449                             &local_error);
450         if (local_error) {
451             error_propagate(errp, local_error);
452             return;
453         }
454     }
455 }
456 
457 static OnOffAuto account_get_opt(QemuOpts *opts, const char *name)
458 {
459     if (!qemu_opt_find(opts, name)) {
460         return ON_OFF_AUTO_AUTO;
461     }
462     if (qemu_opt_get_bool(opts, name, true)) {
463         return ON_OFF_AUTO_ON;
464     }
465     return ON_OFF_AUTO_OFF;
466 }
467 
468 /* Takes the ownership of bs_opts */
469 static BlockBackend *blockdev_init(const char *file, QDict *bs_opts,
470                                    Error **errp)
471 {
472     const char *buf;
473     int bdrv_flags = 0;
474     int on_read_error, on_write_error;
475     OnOffAuto account_invalid, account_failed;
476     bool writethrough, read_only;
477     BlockBackend *blk;
478     BlockDriverState *bs;
479     ThrottleConfig cfg;
480     int snapshot = 0;
481     Error *error = NULL;
482     QemuOpts *opts;
483     QDict *interval_dict = NULL;
484     QList *interval_list = NULL;
485     const char *id;
486     BlockdevDetectZeroesOptions detect_zeroes =
487         BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF;
488     const char *throttling_group = NULL;
489 
490     /* Check common options by copying from bs_opts to opts, all other options
491      * stay in bs_opts for processing by bdrv_open(). */
492     id = qdict_get_try_str(bs_opts, "id");
493     opts = qemu_opts_create(&qemu_common_drive_opts, id, 1, errp);
494     if (!opts) {
495         goto err_no_opts;
496     }
497 
498     if (!qemu_opts_absorb_qdict(opts, bs_opts, errp)) {
499         goto early_err;
500     }
501 
502     if (id) {
503         qdict_del(bs_opts, "id");
504     }
505 
506     /* extract parameters */
507     snapshot = qemu_opt_get_bool(opts, "snapshot", 0);
508 
509     account_invalid = account_get_opt(opts, "stats-account-invalid");
510     account_failed = account_get_opt(opts, "stats-account-failed");
511 
512     writethrough = !qemu_opt_get_bool(opts, BDRV_OPT_CACHE_WB, true);
513 
514     id = qemu_opts_id(opts);
515 
516     qdict_extract_subqdict(bs_opts, &interval_dict, "stats-intervals.");
517     qdict_array_split(interval_dict, &interval_list);
518 
519     if (qdict_size(interval_dict) != 0) {
520         error_setg(errp, "Invalid option stats-intervals.%s",
521                    qdict_first(interval_dict)->key);
522         goto early_err;
523     }
524 
525     extract_common_blockdev_options(opts, &bdrv_flags, &throttling_group, &cfg,
526                                     &detect_zeroes, &error);
527     if (error) {
528         error_propagate(errp, error);
529         goto early_err;
530     }
531 
532     if ((buf = qemu_opt_get(opts, "format")) != NULL) {
533         if (is_help_option(buf)) {
534             qemu_printf("Supported formats:");
535             bdrv_iterate_format(bdrv_format_print, NULL, false);
536             qemu_printf("\nSupported formats (read-only):");
537             bdrv_iterate_format(bdrv_format_print, NULL, true);
538             qemu_printf("\n");
539             goto early_err;
540         }
541 
542         if (qdict_haskey(bs_opts, "driver")) {
543             error_setg(errp, "Cannot specify both 'driver' and 'format'");
544             goto early_err;
545         }
546         qdict_put_str(bs_opts, "driver", buf);
547     }
548 
549     on_write_error = BLOCKDEV_ON_ERROR_ENOSPC;
550     if ((buf = qemu_opt_get(opts, "werror")) != NULL) {
551         on_write_error = parse_block_error_action(buf, 0, &error);
552         if (error) {
553             error_propagate(errp, error);
554             goto early_err;
555         }
556     }
557 
558     on_read_error = BLOCKDEV_ON_ERROR_REPORT;
559     if ((buf = qemu_opt_get(opts, "rerror")) != NULL) {
560         on_read_error = parse_block_error_action(buf, 1, &error);
561         if (error) {
562             error_propagate(errp, error);
563             goto early_err;
564         }
565     }
566 
567     if (snapshot) {
568         bdrv_flags |= BDRV_O_SNAPSHOT;
569     }
570 
571     read_only = qemu_opt_get_bool(opts, BDRV_OPT_READ_ONLY, false);
572 
573     /* init */
574     if ((!file || !*file) && !qdict_size(bs_opts)) {
575         BlockBackendRootState *blk_rs;
576 
577         blk = blk_new(qemu_get_aio_context(), 0, BLK_PERM_ALL);
578         blk_rs = blk_get_root_state(blk);
579         blk_rs->open_flags    = bdrv_flags | (read_only ? 0 : BDRV_O_RDWR);
580         blk_rs->detect_zeroes = detect_zeroes;
581 
582         qobject_unref(bs_opts);
583     } else {
584         if (file && !*file) {
585             file = NULL;
586         }
587 
588         /* bdrv_open() defaults to the values in bdrv_flags (for compatibility
589          * with other callers) rather than what we want as the real defaults.
590          * Apply the defaults here instead. */
591         qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_DIRECT, "off");
592         qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_NO_FLUSH, "off");
593         qdict_set_default_str(bs_opts, BDRV_OPT_READ_ONLY,
594                               read_only ? "on" : "off");
595         qdict_set_default_str(bs_opts, BDRV_OPT_AUTO_READ_ONLY, "on");
596         assert((bdrv_flags & BDRV_O_CACHE_MASK) == 0);
597 
598         if (runstate_check(RUN_STATE_INMIGRATE)) {
599             bdrv_flags |= BDRV_O_INACTIVE;
600         }
601 
602         blk = blk_new_open(file, NULL, bs_opts, bdrv_flags, errp);
603         if (!blk) {
604             goto err_no_bs_opts;
605         }
606         bs = blk_bs(blk);
607 
608         bs->detect_zeroes = detect_zeroes;
609 
610         block_acct_setup(blk_get_stats(blk), account_invalid, account_failed);
611 
612         if (!parse_stats_intervals(blk_get_stats(blk), interval_list, errp)) {
613             blk_unref(blk);
614             blk = NULL;
615             goto err_no_bs_opts;
616         }
617     }
618 
619     /* disk I/O throttling */
620     if (throttle_enabled(&cfg)) {
621         if (!throttling_group) {
622             throttling_group = id;
623         }
624         blk_io_limits_enable(blk, throttling_group);
625         blk_set_io_limits(blk, &cfg);
626     }
627 
628     blk_set_enable_write_cache(blk, !writethrough);
629     blk_set_on_error(blk, on_read_error, on_write_error);
630 
631     if (!monitor_add_blk(blk, id, errp)) {
632         blk_unref(blk);
633         blk = NULL;
634         goto err_no_bs_opts;
635     }
636 
637 err_no_bs_opts:
638     qemu_opts_del(opts);
639     qobject_unref(interval_dict);
640     qobject_unref(interval_list);
641     return blk;
642 
643 early_err:
644     qemu_opts_del(opts);
645     qobject_unref(interval_dict);
646     qobject_unref(interval_list);
647 err_no_opts:
648     qobject_unref(bs_opts);
649     return NULL;
650 }
651 
652 /* Takes the ownership of bs_opts */
653 BlockDriverState *bds_tree_init(QDict *bs_opts, Error **errp)
654 {
655     int bdrv_flags = 0;
656 
657     GLOBAL_STATE_CODE();
658     /* bdrv_open() defaults to the values in bdrv_flags (for compatibility
659      * with other callers) rather than what we want as the real defaults.
660      * Apply the defaults here instead. */
661     qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_DIRECT, "off");
662     qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_NO_FLUSH, "off");
663     qdict_set_default_str(bs_opts, BDRV_OPT_READ_ONLY, "off");
664 
665     if (runstate_check(RUN_STATE_INMIGRATE)) {
666         bdrv_flags |= BDRV_O_INACTIVE;
667     }
668 
669     return bdrv_open(NULL, NULL, bs_opts, bdrv_flags, errp);
670 }
671 
672 void blockdev_close_all_bdrv_states(void)
673 {
674     BlockDriverState *bs, *next_bs;
675 
676     GLOBAL_STATE_CODE();
677     QTAILQ_FOREACH_SAFE(bs, &monitor_bdrv_states, monitor_list, next_bs) {
678         AioContext *ctx = bdrv_get_aio_context(bs);
679 
680         aio_context_acquire(ctx);
681         bdrv_unref(bs);
682         aio_context_release(ctx);
683     }
684 }
685 
686 /* Iterates over the list of monitor-owned BlockDriverStates */
687 BlockDriverState *bdrv_next_monitor_owned(BlockDriverState *bs)
688 {
689     GLOBAL_STATE_CODE();
690     return bs ? QTAILQ_NEXT(bs, monitor_list)
691               : QTAILQ_FIRST(&monitor_bdrv_states);
692 }
693 
694 static bool qemu_opt_rename(QemuOpts *opts, const char *from, const char *to,
695                             Error **errp)
696 {
697     const char *value;
698 
699     value = qemu_opt_get(opts, from);
700     if (value) {
701         if (qemu_opt_find(opts, to)) {
702             error_setg(errp, "'%s' and its alias '%s' can't be used at the "
703                        "same time", to, from);
704             return false;
705         }
706     }
707 
708     /* rename all items in opts */
709     while ((value = qemu_opt_get(opts, from))) {
710         qemu_opt_set(opts, to, value, &error_abort);
711         qemu_opt_unset(opts, from);
712     }
713     return true;
714 }
715 
716 QemuOptsList qemu_legacy_drive_opts = {
717     .name = "drive",
718     .head = QTAILQ_HEAD_INITIALIZER(qemu_legacy_drive_opts.head),
719     .desc = {
720         {
721             .name = "bus",
722             .type = QEMU_OPT_NUMBER,
723             .help = "bus number",
724         },{
725             .name = "unit",
726             .type = QEMU_OPT_NUMBER,
727             .help = "unit number (i.e. lun for scsi)",
728         },{
729             .name = "index",
730             .type = QEMU_OPT_NUMBER,
731             .help = "index number",
732         },{
733             .name = "media",
734             .type = QEMU_OPT_STRING,
735             .help = "media type (disk, cdrom)",
736         },{
737             .name = "if",
738             .type = QEMU_OPT_STRING,
739             .help = "interface (ide, scsi, sd, mtd, floppy, pflash, virtio)",
740         },{
741             .name = "file",
742             .type = QEMU_OPT_STRING,
743             .help = "file name",
744         },
745 
746         /* Options that are passed on, but have special semantics with -drive */
747         {
748             .name = BDRV_OPT_READ_ONLY,
749             .type = QEMU_OPT_BOOL,
750             .help = "open drive file as read-only",
751         },{
752             .name = "rerror",
753             .type = QEMU_OPT_STRING,
754             .help = "read error action",
755         },{
756             .name = "werror",
757             .type = QEMU_OPT_STRING,
758             .help = "write error action",
759         },{
760             .name = "copy-on-read",
761             .type = QEMU_OPT_BOOL,
762             .help = "copy read data from backing file into image file",
763         },
764 
765         { /* end of list */ }
766     },
767 };
768 
769 DriveInfo *drive_new(QemuOpts *all_opts, BlockInterfaceType block_default_type,
770                      Error **errp)
771 {
772     const char *value;
773     BlockBackend *blk;
774     DriveInfo *dinfo = NULL;
775     QDict *bs_opts;
776     QemuOpts *legacy_opts;
777     DriveMediaType media = MEDIA_DISK;
778     BlockInterfaceType type;
779     int max_devs, bus_id, unit_id, index;
780     const char *werror, *rerror;
781     bool read_only = false;
782     bool copy_on_read;
783     const char *filename;
784     int i;
785 
786     GLOBAL_STATE_CODE();
787 
788     /* Change legacy command line options into QMP ones */
789     static const struct {
790         const char *from;
791         const char *to;
792     } opt_renames[] = {
793         { "iops",           "throttling.iops-total" },
794         { "iops_rd",        "throttling.iops-read" },
795         { "iops_wr",        "throttling.iops-write" },
796 
797         { "bps",            "throttling.bps-total" },
798         { "bps_rd",         "throttling.bps-read" },
799         { "bps_wr",         "throttling.bps-write" },
800 
801         { "iops_max",       "throttling.iops-total-max" },
802         { "iops_rd_max",    "throttling.iops-read-max" },
803         { "iops_wr_max",    "throttling.iops-write-max" },
804 
805         { "bps_max",        "throttling.bps-total-max" },
806         { "bps_rd_max",     "throttling.bps-read-max" },
807         { "bps_wr_max",     "throttling.bps-write-max" },
808 
809         { "iops_size",      "throttling.iops-size" },
810 
811         { "group",          "throttling.group" },
812 
813         { "readonly",       BDRV_OPT_READ_ONLY },
814     };
815 
816     for (i = 0; i < ARRAY_SIZE(opt_renames); i++) {
817         if (!qemu_opt_rename(all_opts, opt_renames[i].from,
818                              opt_renames[i].to, errp)) {
819             return NULL;
820         }
821     }
822 
823     value = qemu_opt_get(all_opts, "cache");
824     if (value) {
825         int flags = 0;
826         bool writethrough;
827 
828         if (bdrv_parse_cache_mode(value, &flags, &writethrough) != 0) {
829             error_setg(errp, "invalid cache option");
830             return NULL;
831         }
832 
833         /* Specific options take precedence */
834         if (!qemu_opt_get(all_opts, BDRV_OPT_CACHE_WB)) {
835             qemu_opt_set_bool(all_opts, BDRV_OPT_CACHE_WB,
836                               !writethrough, &error_abort);
837         }
838         if (!qemu_opt_get(all_opts, BDRV_OPT_CACHE_DIRECT)) {
839             qemu_opt_set_bool(all_opts, BDRV_OPT_CACHE_DIRECT,
840                               !!(flags & BDRV_O_NOCACHE), &error_abort);
841         }
842         if (!qemu_opt_get(all_opts, BDRV_OPT_CACHE_NO_FLUSH)) {
843             qemu_opt_set_bool(all_opts, BDRV_OPT_CACHE_NO_FLUSH,
844                               !!(flags & BDRV_O_NO_FLUSH), &error_abort);
845         }
846         qemu_opt_unset(all_opts, "cache");
847     }
848 
849     /* Get a QDict for processing the options */
850     bs_opts = qdict_new();
851     qemu_opts_to_qdict(all_opts, bs_opts);
852 
853     legacy_opts = qemu_opts_create(&qemu_legacy_drive_opts, NULL, 0,
854                                    &error_abort);
855     if (!qemu_opts_absorb_qdict(legacy_opts, bs_opts, errp)) {
856         goto fail;
857     }
858 
859     /* Media type */
860     value = qemu_opt_get(legacy_opts, "media");
861     if (value) {
862         if (!strcmp(value, "disk")) {
863             media = MEDIA_DISK;
864         } else if (!strcmp(value, "cdrom")) {
865             media = MEDIA_CDROM;
866             read_only = true;
867         } else {
868             error_setg(errp, "'%s' invalid media", value);
869             goto fail;
870         }
871     }
872 
873     /* copy-on-read is disabled with a warning for read-only devices */
874     read_only |= qemu_opt_get_bool(legacy_opts, BDRV_OPT_READ_ONLY, false);
875     copy_on_read = qemu_opt_get_bool(legacy_opts, "copy-on-read", false);
876 
877     if (read_only && copy_on_read) {
878         warn_report("disabling copy-on-read on read-only drive");
879         copy_on_read = false;
880     }
881 
882     qdict_put_str(bs_opts, BDRV_OPT_READ_ONLY, read_only ? "on" : "off");
883     qdict_put_str(bs_opts, "copy-on-read", copy_on_read ? "on" : "off");
884 
885     /* Controller type */
886     value = qemu_opt_get(legacy_opts, "if");
887     if (value) {
888         for (type = 0;
889              type < IF_COUNT && strcmp(value, if_name[type]);
890              type++) {
891         }
892         if (type == IF_COUNT) {
893             error_setg(errp, "unsupported bus type '%s'", value);
894             goto fail;
895         }
896     } else {
897         type = block_default_type;
898     }
899 
900     /* Device address specified by bus/unit or index.
901      * If none was specified, try to find the first free one. */
902     bus_id  = qemu_opt_get_number(legacy_opts, "bus", 0);
903     unit_id = qemu_opt_get_number(legacy_opts, "unit", -1);
904     index   = qemu_opt_get_number(legacy_opts, "index", -1);
905 
906     max_devs = if_max_devs[type];
907 
908     if (index != -1) {
909         if (bus_id != 0 || unit_id != -1) {
910             error_setg(errp, "index cannot be used with bus and unit");
911             goto fail;
912         }
913         bus_id = drive_index_to_bus_id(type, index);
914         unit_id = drive_index_to_unit_id(type, index);
915     }
916 
917     if (unit_id == -1) {
918        unit_id = 0;
919        while (drive_get(type, bus_id, unit_id) != NULL) {
920            unit_id++;
921            if (max_devs && unit_id >= max_devs) {
922                unit_id -= max_devs;
923                bus_id++;
924            }
925        }
926     }
927 
928     if (max_devs && unit_id >= max_devs) {
929         error_setg(errp, "unit %d too big (max is %d)", unit_id, max_devs - 1);
930         goto fail;
931     }
932 
933     if (drive_get(type, bus_id, unit_id) != NULL) {
934         error_setg(errp, "drive with bus=%d, unit=%d (index=%d) exists",
935                    bus_id, unit_id, index);
936         goto fail;
937     }
938 
939     /* no id supplied -> create one */
940     if (qemu_opts_id(all_opts) == NULL) {
941         char *new_id;
942         const char *mediastr = "";
943         if (type == IF_IDE || type == IF_SCSI) {
944             mediastr = (media == MEDIA_CDROM) ? "-cd" : "-hd";
945         }
946         if (max_devs) {
947             new_id = g_strdup_printf("%s%i%s%i", if_name[type], bus_id,
948                                      mediastr, unit_id);
949         } else {
950             new_id = g_strdup_printf("%s%s%i", if_name[type],
951                                      mediastr, unit_id);
952         }
953         qdict_put_str(bs_opts, "id", new_id);
954         g_free(new_id);
955     }
956 
957     /* Add virtio block device */
958     if (type == IF_VIRTIO) {
959         QemuOpts *devopts;
960         devopts = qemu_opts_create(qemu_find_opts("device"), NULL, 0,
961                                    &error_abort);
962         qemu_opt_set(devopts, "driver", "virtio-blk", &error_abort);
963         qemu_opt_set(devopts, "drive", qdict_get_str(bs_opts, "id"),
964                      &error_abort);
965     }
966 
967     filename = qemu_opt_get(legacy_opts, "file");
968 
969     /* Check werror/rerror compatibility with if=... */
970     werror = qemu_opt_get(legacy_opts, "werror");
971     if (werror != NULL) {
972         if (type != IF_IDE && type != IF_SCSI && type != IF_VIRTIO &&
973             type != IF_NONE) {
974             error_setg(errp, "werror is not supported by this bus type");
975             goto fail;
976         }
977         qdict_put_str(bs_opts, "werror", werror);
978     }
979 
980     rerror = qemu_opt_get(legacy_opts, "rerror");
981     if (rerror != NULL) {
982         if (type != IF_IDE && type != IF_VIRTIO && type != IF_SCSI &&
983             type != IF_NONE) {
984             error_setg(errp, "rerror is not supported by this bus type");
985             goto fail;
986         }
987         qdict_put_str(bs_opts, "rerror", rerror);
988     }
989 
990     /* Actual block device init: Functionality shared with blockdev-add */
991     blk = blockdev_init(filename, bs_opts, errp);
992     bs_opts = NULL;
993     if (!blk) {
994         goto fail;
995     }
996 
997     /* Create legacy DriveInfo */
998     dinfo = g_malloc0(sizeof(*dinfo));
999     dinfo->opts = all_opts;
1000 
1001     dinfo->type = type;
1002     dinfo->bus = bus_id;
1003     dinfo->unit = unit_id;
1004 
1005     blk_set_legacy_dinfo(blk, dinfo);
1006 
1007     switch(type) {
1008     case IF_IDE:
1009     case IF_SCSI:
1010     case IF_XEN:
1011     case IF_NONE:
1012         dinfo->media_cd = media == MEDIA_CDROM;
1013         break;
1014     default:
1015         break;
1016     }
1017 
1018 fail:
1019     qemu_opts_del(legacy_opts);
1020     qobject_unref(bs_opts);
1021     return dinfo;
1022 }
1023 
1024 static BlockDriverState *qmp_get_root_bs(const char *name, Error **errp)
1025 {
1026     BlockDriverState *bs;
1027 
1028     bs = bdrv_lookup_bs(name, name, errp);
1029     if (bs == NULL) {
1030         return NULL;
1031     }
1032 
1033     if (!bdrv_is_root_node(bs)) {
1034         error_setg(errp, "Need a root block node");
1035         return NULL;
1036     }
1037 
1038     if (!bdrv_is_inserted(bs)) {
1039         error_setg(errp, "Device has no medium");
1040         return NULL;
1041     }
1042 
1043     return bs;
1044 }
1045 
1046 static void blockdev_do_action(TransactionAction *action, Error **errp)
1047 {
1048     TransactionActionList list;
1049 
1050     list.value = action;
1051     list.next = NULL;
1052     qmp_transaction(&list, NULL, errp);
1053 }
1054 
1055 void qmp_blockdev_snapshot_sync(const char *device, const char *node_name,
1056                                 const char *snapshot_file,
1057                                 const char *snapshot_node_name,
1058                                 const char *format,
1059                                 bool has_mode, NewImageMode mode, Error **errp)
1060 {
1061     BlockdevSnapshotSync snapshot = {
1062         .device = (char *) device,
1063         .node_name = (char *) node_name,
1064         .snapshot_file = (char *) snapshot_file,
1065         .snapshot_node_name = (char *) snapshot_node_name,
1066         .format = (char *) format,
1067         .has_mode = has_mode,
1068         .mode = mode,
1069     };
1070     TransactionAction action = {
1071         .type = TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC,
1072         .u.blockdev_snapshot_sync.data = &snapshot,
1073     };
1074     blockdev_do_action(&action, errp);
1075 }
1076 
1077 void qmp_blockdev_snapshot(const char *node, const char *overlay,
1078                            Error **errp)
1079 {
1080     BlockdevSnapshot snapshot_data = {
1081         .node = (char *) node,
1082         .overlay = (char *) overlay
1083     };
1084     TransactionAction action = {
1085         .type = TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT,
1086         .u.blockdev_snapshot.data = &snapshot_data,
1087     };
1088     blockdev_do_action(&action, errp);
1089 }
1090 
1091 void qmp_blockdev_snapshot_internal_sync(const char *device,
1092                                          const char *name,
1093                                          Error **errp)
1094 {
1095     BlockdevSnapshotInternal snapshot = {
1096         .device = (char *) device,
1097         .name = (char *) name
1098     };
1099     TransactionAction action = {
1100         .type = TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC,
1101         .u.blockdev_snapshot_internal_sync.data = &snapshot,
1102     };
1103     blockdev_do_action(&action, errp);
1104 }
1105 
1106 SnapshotInfo *qmp_blockdev_snapshot_delete_internal_sync(const char *device,
1107                                                          const char *id,
1108                                                          const char *name,
1109                                                          Error **errp)
1110 {
1111     BlockDriverState *bs;
1112     AioContext *aio_context;
1113     QEMUSnapshotInfo sn;
1114     Error *local_err = NULL;
1115     SnapshotInfo *info = NULL;
1116     int ret;
1117 
1118     bs = qmp_get_root_bs(device, errp);
1119     if (!bs) {
1120         return NULL;
1121     }
1122     aio_context = bdrv_get_aio_context(bs);
1123     aio_context_acquire(aio_context);
1124 
1125     if (!id && !name) {
1126         error_setg(errp, "Name or id must be provided");
1127         goto out_aio_context;
1128     }
1129 
1130     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_INTERNAL_SNAPSHOT_DELETE, errp)) {
1131         goto out_aio_context;
1132     }
1133 
1134     ret = bdrv_snapshot_find_by_id_and_name(bs, id, name, &sn, &local_err);
1135     if (local_err) {
1136         error_propagate(errp, local_err);
1137         goto out_aio_context;
1138     }
1139     if (!ret) {
1140         error_setg(errp,
1141                    "Snapshot with id '%s' and name '%s' does not exist on "
1142                    "device '%s'",
1143                    STR_OR_NULL(id), STR_OR_NULL(name), device);
1144         goto out_aio_context;
1145     }
1146 
1147     bdrv_snapshot_delete(bs, id, name, &local_err);
1148     if (local_err) {
1149         error_propagate(errp, local_err);
1150         goto out_aio_context;
1151     }
1152 
1153     aio_context_release(aio_context);
1154 
1155     info = g_new0(SnapshotInfo, 1);
1156     info->id = g_strdup(sn.id_str);
1157     info->name = g_strdup(sn.name);
1158     info->date_nsec = sn.date_nsec;
1159     info->date_sec = sn.date_sec;
1160     info->vm_state_size = sn.vm_state_size;
1161     info->vm_clock_nsec = sn.vm_clock_nsec % 1000000000;
1162     info->vm_clock_sec = sn.vm_clock_nsec / 1000000000;
1163     if (sn.icount != -1ULL) {
1164         info->icount = sn.icount;
1165         info->has_icount = true;
1166     }
1167 
1168     return info;
1169 
1170 out_aio_context:
1171     aio_context_release(aio_context);
1172     return NULL;
1173 }
1174 
1175 /* New and old BlockDriverState structs for atomic group operations */
1176 
1177 typedef struct BlkActionState BlkActionState;
1178 
1179 /**
1180  * BlkActionOps:
1181  * Table of operations that define an Action.
1182  *
1183  * @instance_size: Size of state struct, in bytes.
1184  * @prepare: Prepare the work, must NOT be NULL.
1185  * @commit: Commit the changes, can be NULL.
1186  * @abort: Abort the changes on fail, can be NULL.
1187  * @clean: Clean up resources after all transaction actions have called
1188  *         commit() or abort(). Can be NULL.
1189  *
1190  * Only prepare() may fail. In a single transaction, only one of commit() or
1191  * abort() will be called. clean() will always be called if it is present.
1192  *
1193  * Always run under BQL.
1194  */
1195 typedef struct BlkActionOps {
1196     size_t instance_size;
1197     void (*prepare)(BlkActionState *common, Error **errp);
1198     void (*commit)(BlkActionState *common);
1199     void (*abort)(BlkActionState *common);
1200     void (*clean)(BlkActionState *common);
1201 } BlkActionOps;
1202 
1203 /**
1204  * BlkActionState:
1205  * Describes one Action's state within a Transaction.
1206  *
1207  * @action: QAPI-defined enum identifying which Action to perform.
1208  * @ops: Table of ActionOps this Action can perform.
1209  * @block_job_txn: Transaction which this action belongs to.
1210  * @entry: List membership for all Actions in this Transaction.
1211  *
1212  * This structure must be arranged as first member in a subclassed type,
1213  * assuming that the compiler will also arrange it to the same offsets as the
1214  * base class.
1215  */
1216 struct BlkActionState {
1217     TransactionAction *action;
1218     const BlkActionOps *ops;
1219     JobTxn *block_job_txn;
1220     TransactionProperties *txn_props;
1221     QTAILQ_ENTRY(BlkActionState) entry;
1222 };
1223 
1224 /* internal snapshot private data */
1225 typedef struct InternalSnapshotState {
1226     BlkActionState common;
1227     BlockDriverState *bs;
1228     QEMUSnapshotInfo sn;
1229     bool created;
1230 } InternalSnapshotState;
1231 
1232 
1233 static int action_check_completion_mode(BlkActionState *s, Error **errp)
1234 {
1235     if (s->txn_props->completion_mode != ACTION_COMPLETION_MODE_INDIVIDUAL) {
1236         error_setg(errp,
1237                    "Action '%s' does not support Transaction property "
1238                    "completion-mode = %s",
1239                    TransactionActionKind_str(s->action->type),
1240                    ActionCompletionMode_str(s->txn_props->completion_mode));
1241         return -1;
1242     }
1243     return 0;
1244 }
1245 
1246 static void internal_snapshot_prepare(BlkActionState *common,
1247                                       Error **errp)
1248 {
1249     Error *local_err = NULL;
1250     const char *device;
1251     const char *name;
1252     BlockDriverState *bs;
1253     QEMUSnapshotInfo old_sn, *sn;
1254     bool ret;
1255     int64_t rt;
1256     BlockdevSnapshotInternal *internal;
1257     InternalSnapshotState *state;
1258     AioContext *aio_context;
1259     int ret1;
1260 
1261     g_assert(common->action->type ==
1262              TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC);
1263     internal = common->action->u.blockdev_snapshot_internal_sync.data;
1264     state = DO_UPCAST(InternalSnapshotState, common, common);
1265 
1266     /* 1. parse input */
1267     device = internal->device;
1268     name = internal->name;
1269 
1270     /* 2. check for validation */
1271     if (action_check_completion_mode(common, errp) < 0) {
1272         return;
1273     }
1274 
1275     bs = qmp_get_root_bs(device, errp);
1276     if (!bs) {
1277         return;
1278     }
1279 
1280     aio_context = bdrv_get_aio_context(bs);
1281     aio_context_acquire(aio_context);
1282 
1283     state->bs = bs;
1284 
1285     /* Paired with .clean() */
1286     bdrv_drained_begin(bs);
1287 
1288     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_INTERNAL_SNAPSHOT, errp)) {
1289         goto out;
1290     }
1291 
1292     if (bdrv_is_read_only(bs)) {
1293         error_setg(errp, "Device '%s' is read only", device);
1294         goto out;
1295     }
1296 
1297     if (!bdrv_can_snapshot(bs)) {
1298         error_setg(errp, "Block format '%s' used by device '%s' "
1299                    "does not support internal snapshots",
1300                    bs->drv->format_name, device);
1301         goto out;
1302     }
1303 
1304     if (!strlen(name)) {
1305         error_setg(errp, "Name is empty");
1306         goto out;
1307     }
1308 
1309     /* check whether a snapshot with name exist */
1310     ret = bdrv_snapshot_find_by_id_and_name(bs, NULL, name, &old_sn,
1311                                             &local_err);
1312     if (local_err) {
1313         error_propagate(errp, local_err);
1314         goto out;
1315     } else if (ret) {
1316         error_setg(errp,
1317                    "Snapshot with name '%s' already exists on device '%s'",
1318                    name, device);
1319         goto out;
1320     }
1321 
1322     /* 3. take the snapshot */
1323     sn = &state->sn;
1324     pstrcpy(sn->name, sizeof(sn->name), name);
1325     rt = g_get_real_time();
1326     sn->date_sec = rt / G_USEC_PER_SEC;
1327     sn->date_nsec = (rt % G_USEC_PER_SEC) * 1000;
1328     sn->vm_clock_nsec = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL);
1329     if (replay_mode != REPLAY_MODE_NONE) {
1330         sn->icount = replay_get_current_icount();
1331     } else {
1332         sn->icount = -1ULL;
1333     }
1334 
1335     ret1 = bdrv_snapshot_create(bs, sn);
1336     if (ret1 < 0) {
1337         error_setg_errno(errp, -ret1,
1338                          "Failed to create snapshot '%s' on device '%s'",
1339                          name, device);
1340         goto out;
1341     }
1342 
1343     /* 4. succeed, mark a snapshot is created */
1344     state->created = true;
1345 
1346 out:
1347     aio_context_release(aio_context);
1348 }
1349 
1350 static void internal_snapshot_abort(BlkActionState *common)
1351 {
1352     InternalSnapshotState *state =
1353                              DO_UPCAST(InternalSnapshotState, common, common);
1354     BlockDriverState *bs = state->bs;
1355     QEMUSnapshotInfo *sn = &state->sn;
1356     AioContext *aio_context;
1357     Error *local_error = NULL;
1358 
1359     if (!state->created) {
1360         return;
1361     }
1362 
1363     aio_context = bdrv_get_aio_context(state->bs);
1364     aio_context_acquire(aio_context);
1365 
1366     if (bdrv_snapshot_delete(bs, sn->id_str, sn->name, &local_error) < 0) {
1367         error_reportf_err(local_error,
1368                           "Failed to delete snapshot with id '%s' and "
1369                           "name '%s' on device '%s' in abort: ",
1370                           sn->id_str, sn->name,
1371                           bdrv_get_device_name(bs));
1372     }
1373 
1374     aio_context_release(aio_context);
1375 }
1376 
1377 static void internal_snapshot_clean(BlkActionState *common)
1378 {
1379     InternalSnapshotState *state = DO_UPCAST(InternalSnapshotState,
1380                                              common, common);
1381     AioContext *aio_context;
1382 
1383     if (!state->bs) {
1384         return;
1385     }
1386 
1387     aio_context = bdrv_get_aio_context(state->bs);
1388     aio_context_acquire(aio_context);
1389 
1390     bdrv_drained_end(state->bs);
1391 
1392     aio_context_release(aio_context);
1393 }
1394 
1395 /* external snapshot private data */
1396 typedef struct ExternalSnapshotState {
1397     BlkActionState common;
1398     BlockDriverState *old_bs;
1399     BlockDriverState *new_bs;
1400     bool overlay_appended;
1401 } ExternalSnapshotState;
1402 
1403 static void external_snapshot_prepare(BlkActionState *common,
1404                                       Error **errp)
1405 {
1406     int ret;
1407     int flags = 0;
1408     QDict *options = NULL;
1409     Error *local_err = NULL;
1410     /* Device and node name of the image to generate the snapshot from */
1411     const char *device;
1412     const char *node_name;
1413     /* Reference to the new image (for 'blockdev-snapshot') */
1414     const char *snapshot_ref;
1415     /* File name of the new image (for 'blockdev-snapshot-sync') */
1416     const char *new_image_file;
1417     ExternalSnapshotState *state =
1418                              DO_UPCAST(ExternalSnapshotState, common, common);
1419     TransactionAction *action = common->action;
1420     AioContext *aio_context;
1421     uint64_t perm, shared;
1422 
1423     /* 'blockdev-snapshot' and 'blockdev-snapshot-sync' have similar
1424      * purpose but a different set of parameters */
1425     switch (action->type) {
1426     case TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT:
1427         {
1428             BlockdevSnapshot *s = action->u.blockdev_snapshot.data;
1429             device = s->node;
1430             node_name = s->node;
1431             new_image_file = NULL;
1432             snapshot_ref = s->overlay;
1433         }
1434         break;
1435     case TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC:
1436         {
1437             BlockdevSnapshotSync *s = action->u.blockdev_snapshot_sync.data;
1438             device = s->device;
1439             node_name = s->node_name;
1440             new_image_file = s->snapshot_file;
1441             snapshot_ref = NULL;
1442         }
1443         break;
1444     default:
1445         g_assert_not_reached();
1446     }
1447 
1448     /* start processing */
1449     if (action_check_completion_mode(common, errp) < 0) {
1450         return;
1451     }
1452 
1453     state->old_bs = bdrv_lookup_bs(device, node_name, errp);
1454     if (!state->old_bs) {
1455         return;
1456     }
1457 
1458     aio_context = bdrv_get_aio_context(state->old_bs);
1459     aio_context_acquire(aio_context);
1460 
1461     /* Paired with .clean() */
1462     bdrv_drained_begin(state->old_bs);
1463 
1464     if (!bdrv_is_inserted(state->old_bs)) {
1465         error_setg(errp, QERR_DEVICE_HAS_NO_MEDIUM, device);
1466         goto out;
1467     }
1468 
1469     if (bdrv_op_is_blocked(state->old_bs,
1470                            BLOCK_OP_TYPE_EXTERNAL_SNAPSHOT, errp)) {
1471         goto out;
1472     }
1473 
1474     if (!bdrv_is_read_only(state->old_bs)) {
1475         if (bdrv_flush(state->old_bs)) {
1476             error_setg(errp, QERR_IO_ERROR);
1477             goto out;
1478         }
1479     }
1480 
1481     if (action->type == TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC) {
1482         BlockdevSnapshotSync *s = action->u.blockdev_snapshot_sync.data;
1483         const char *format = s->format ?: "qcow2";
1484         enum NewImageMode mode;
1485         const char *snapshot_node_name = s->snapshot_node_name;
1486 
1487         if (node_name && !snapshot_node_name) {
1488             error_setg(errp, "New overlay node-name missing");
1489             goto out;
1490         }
1491 
1492         if (snapshot_node_name &&
1493             bdrv_lookup_bs(snapshot_node_name, snapshot_node_name, NULL)) {
1494             error_setg(errp, "New overlay node-name already in use");
1495             goto out;
1496         }
1497 
1498         flags = state->old_bs->open_flags;
1499         flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_COPY_ON_READ);
1500         flags |= BDRV_O_NO_BACKING;
1501 
1502         /* create new image w/backing file */
1503         mode = s->has_mode ? s->mode : NEW_IMAGE_MODE_ABSOLUTE_PATHS;
1504         if (mode != NEW_IMAGE_MODE_EXISTING) {
1505             int64_t size = bdrv_getlength(state->old_bs);
1506             if (size < 0) {
1507                 error_setg_errno(errp, -size, "bdrv_getlength failed");
1508                 goto out;
1509             }
1510             bdrv_refresh_filename(state->old_bs);
1511 
1512             aio_context_release(aio_context);
1513             bdrv_img_create(new_image_file, format,
1514                             state->old_bs->filename,
1515                             state->old_bs->drv->format_name,
1516                             NULL, size, flags, false, &local_err);
1517             aio_context_acquire(aio_context);
1518 
1519             if (local_err) {
1520                 error_propagate(errp, local_err);
1521                 goto out;
1522             }
1523         }
1524 
1525         options = qdict_new();
1526         if (snapshot_node_name) {
1527             qdict_put_str(options, "node-name", snapshot_node_name);
1528         }
1529         qdict_put_str(options, "driver", format);
1530     }
1531 
1532     state->new_bs = bdrv_open(new_image_file, snapshot_ref, options, flags,
1533                               errp);
1534     /* We will manually add the backing_hd field to the bs later */
1535     if (!state->new_bs) {
1536         goto out;
1537     }
1538 
1539     /*
1540      * Allow attaching a backing file to an overlay that's already in use only
1541      * if the parents don't assume that they are already seeing a valid image.
1542      * (Specifically, allow it as a mirror target, which is write-only access.)
1543      */
1544     bdrv_get_cumulative_perm(state->new_bs, &perm, &shared);
1545     if (perm & BLK_PERM_CONSISTENT_READ) {
1546         error_setg(errp, "The overlay is already in use");
1547         goto out;
1548     }
1549 
1550     if (state->new_bs->drv->is_filter) {
1551         error_setg(errp, "Filters cannot be used as overlays");
1552         goto out;
1553     }
1554 
1555     if (bdrv_cow_child(state->new_bs)) {
1556         error_setg(errp, "The overlay already has a backing image");
1557         goto out;
1558     }
1559 
1560     if (!state->new_bs->drv->supports_backing) {
1561         error_setg(errp, "The overlay does not support backing images");
1562         goto out;
1563     }
1564 
1565     ret = bdrv_append(state->new_bs, state->old_bs, errp);
1566     if (ret < 0) {
1567         goto out;
1568     }
1569     state->overlay_appended = true;
1570 
1571 out:
1572     aio_context_release(aio_context);
1573 }
1574 
1575 static void external_snapshot_commit(BlkActionState *common)
1576 {
1577     ExternalSnapshotState *state =
1578                              DO_UPCAST(ExternalSnapshotState, common, common);
1579     AioContext *aio_context;
1580 
1581     aio_context = bdrv_get_aio_context(state->old_bs);
1582     aio_context_acquire(aio_context);
1583 
1584     /* We don't need (or want) to use the transactional
1585      * bdrv_reopen_multiple() across all the entries at once, because we
1586      * don't want to abort all of them if one of them fails the reopen */
1587     if (!qatomic_read(&state->old_bs->copy_on_read)) {
1588         bdrv_reopen_set_read_only(state->old_bs, true, NULL);
1589     }
1590 
1591     aio_context_release(aio_context);
1592 }
1593 
1594 static void external_snapshot_abort(BlkActionState *common)
1595 {
1596     ExternalSnapshotState *state =
1597                              DO_UPCAST(ExternalSnapshotState, common, common);
1598     if (state->new_bs) {
1599         if (state->overlay_appended) {
1600             AioContext *aio_context;
1601             AioContext *tmp_context;
1602             int ret;
1603 
1604             aio_context = bdrv_get_aio_context(state->old_bs);
1605             aio_context_acquire(aio_context);
1606 
1607             bdrv_ref(state->old_bs);   /* we can't let bdrv_set_backind_hd()
1608                                           close state->old_bs; we need it */
1609             bdrv_set_backing_hd(state->new_bs, NULL, &error_abort);
1610 
1611             /*
1612              * The call to bdrv_set_backing_hd() above returns state->old_bs to
1613              * the main AioContext. As we're still going to be using it, return
1614              * it to the AioContext it was before.
1615              */
1616             tmp_context = bdrv_get_aio_context(state->old_bs);
1617             if (aio_context != tmp_context) {
1618                 aio_context_release(aio_context);
1619                 aio_context_acquire(tmp_context);
1620 
1621                 ret = bdrv_try_change_aio_context(state->old_bs,
1622                                                   aio_context, NULL, NULL);
1623                 assert(ret == 0);
1624 
1625                 aio_context_release(tmp_context);
1626                 aio_context_acquire(aio_context);
1627             }
1628 
1629             bdrv_replace_node(state->new_bs, state->old_bs, &error_abort);
1630             bdrv_unref(state->old_bs); /* bdrv_replace_node() ref'ed old_bs */
1631 
1632             aio_context_release(aio_context);
1633         }
1634     }
1635 }
1636 
1637 static void external_snapshot_clean(BlkActionState *common)
1638 {
1639     ExternalSnapshotState *state =
1640                              DO_UPCAST(ExternalSnapshotState, common, common);
1641     AioContext *aio_context;
1642 
1643     if (!state->old_bs) {
1644         return;
1645     }
1646 
1647     aio_context = bdrv_get_aio_context(state->old_bs);
1648     aio_context_acquire(aio_context);
1649 
1650     bdrv_drained_end(state->old_bs);
1651     bdrv_unref(state->new_bs);
1652 
1653     aio_context_release(aio_context);
1654 }
1655 
1656 typedef struct DriveBackupState {
1657     BlkActionState common;
1658     BlockDriverState *bs;
1659     BlockJob *job;
1660 } DriveBackupState;
1661 
1662 static BlockJob *do_backup_common(BackupCommon *backup,
1663                                   BlockDriverState *bs,
1664                                   BlockDriverState *target_bs,
1665                                   AioContext *aio_context,
1666                                   JobTxn *txn, Error **errp);
1667 
1668 static void drive_backup_prepare(BlkActionState *common, Error **errp)
1669 {
1670     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1671     DriveBackup *backup;
1672     BlockDriverState *bs;
1673     BlockDriverState *target_bs;
1674     BlockDriverState *source = NULL;
1675     AioContext *aio_context;
1676     AioContext *old_context;
1677     const char *format;
1678     QDict *options;
1679     Error *local_err = NULL;
1680     int flags;
1681     int64_t size;
1682     bool set_backing_hd = false;
1683     int ret;
1684 
1685     assert(common->action->type == TRANSACTION_ACTION_KIND_DRIVE_BACKUP);
1686     backup = common->action->u.drive_backup.data;
1687 
1688     if (!backup->has_mode) {
1689         backup->mode = NEW_IMAGE_MODE_ABSOLUTE_PATHS;
1690     }
1691 
1692     bs = bdrv_lookup_bs(backup->device, backup->device, errp);
1693     if (!bs) {
1694         return;
1695     }
1696 
1697     if (!bs->drv) {
1698         error_setg(errp, "Device has no medium");
1699         return;
1700     }
1701 
1702     aio_context = bdrv_get_aio_context(bs);
1703     aio_context_acquire(aio_context);
1704 
1705     state->bs = bs;
1706     /* Paired with .clean() */
1707     bdrv_drained_begin(bs);
1708 
1709     format = backup->format;
1710     if (!format && backup->mode != NEW_IMAGE_MODE_EXISTING) {
1711         format = bs->drv->format_name;
1712     }
1713 
1714     /* Early check to avoid creating target */
1715     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_BACKUP_SOURCE, errp)) {
1716         goto out;
1717     }
1718 
1719     flags = bs->open_flags | BDRV_O_RDWR;
1720 
1721     /*
1722      * See if we have a backing HD we can use to create our new image
1723      * on top of.
1724      */
1725     if (backup->sync == MIRROR_SYNC_MODE_TOP) {
1726         /*
1727          * Backup will not replace the source by the target, so none
1728          * of the filters skipped here will be removed (in contrast to
1729          * mirror).  Therefore, we can skip all of them when looking
1730          * for the first COW relationship.
1731          */
1732         source = bdrv_cow_bs(bdrv_skip_filters(bs));
1733         if (!source) {
1734             backup->sync = MIRROR_SYNC_MODE_FULL;
1735         }
1736     }
1737     if (backup->sync == MIRROR_SYNC_MODE_NONE) {
1738         source = bs;
1739         flags |= BDRV_O_NO_BACKING;
1740         set_backing_hd = true;
1741     }
1742 
1743     size = bdrv_getlength(bs);
1744     if (size < 0) {
1745         error_setg_errno(errp, -size, "bdrv_getlength failed");
1746         goto out;
1747     }
1748 
1749     if (backup->mode != NEW_IMAGE_MODE_EXISTING) {
1750         assert(format);
1751         if (source) {
1752             /* Implicit filters should not appear in the filename */
1753             BlockDriverState *explicit_backing =
1754                 bdrv_skip_implicit_filters(source);
1755 
1756             bdrv_refresh_filename(explicit_backing);
1757             bdrv_img_create(backup->target, format,
1758                             explicit_backing->filename,
1759                             explicit_backing->drv->format_name, NULL,
1760                             size, flags, false, &local_err);
1761         } else {
1762             bdrv_img_create(backup->target, format, NULL, NULL, NULL,
1763                             size, flags, false, &local_err);
1764         }
1765     }
1766 
1767     if (local_err) {
1768         error_propagate(errp, local_err);
1769         goto out;
1770     }
1771 
1772     options = qdict_new();
1773     qdict_put_str(options, "discard", "unmap");
1774     qdict_put_str(options, "detect-zeroes", "unmap");
1775     if (format) {
1776         qdict_put_str(options, "driver", format);
1777     }
1778 
1779     target_bs = bdrv_open(backup->target, NULL, options, flags, errp);
1780     if (!target_bs) {
1781         goto out;
1782     }
1783 
1784     /* Honor bdrv_try_change_aio_context() context acquisition requirements. */
1785     old_context = bdrv_get_aio_context(target_bs);
1786     aio_context_release(aio_context);
1787     aio_context_acquire(old_context);
1788 
1789     ret = bdrv_try_change_aio_context(target_bs, aio_context, NULL, errp);
1790     if (ret < 0) {
1791         bdrv_unref(target_bs);
1792         aio_context_release(old_context);
1793         return;
1794     }
1795 
1796     aio_context_release(old_context);
1797     aio_context_acquire(aio_context);
1798 
1799     if (set_backing_hd) {
1800         if (bdrv_set_backing_hd(target_bs, source, errp) < 0) {
1801             goto unref;
1802         }
1803     }
1804 
1805     state->job = do_backup_common(qapi_DriveBackup_base(backup),
1806                                   bs, target_bs, aio_context,
1807                                   common->block_job_txn, errp);
1808 
1809 unref:
1810     bdrv_unref(target_bs);
1811 out:
1812     aio_context_release(aio_context);
1813 }
1814 
1815 static void drive_backup_commit(BlkActionState *common)
1816 {
1817     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1818     AioContext *aio_context;
1819 
1820     aio_context = bdrv_get_aio_context(state->bs);
1821     aio_context_acquire(aio_context);
1822 
1823     assert(state->job);
1824     job_start(&state->job->job);
1825 
1826     aio_context_release(aio_context);
1827 }
1828 
1829 static void drive_backup_abort(BlkActionState *common)
1830 {
1831     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1832 
1833     if (state->job) {
1834         job_cancel_sync(&state->job->job, true);
1835     }
1836 }
1837 
1838 static void drive_backup_clean(BlkActionState *common)
1839 {
1840     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1841     AioContext *aio_context;
1842 
1843     if (!state->bs) {
1844         return;
1845     }
1846 
1847     aio_context = bdrv_get_aio_context(state->bs);
1848     aio_context_acquire(aio_context);
1849 
1850     bdrv_drained_end(state->bs);
1851 
1852     aio_context_release(aio_context);
1853 }
1854 
1855 typedef struct BlockdevBackupState {
1856     BlkActionState common;
1857     BlockDriverState *bs;
1858     BlockJob *job;
1859 } BlockdevBackupState;
1860 
1861 static void blockdev_backup_prepare(BlkActionState *common, Error **errp)
1862 {
1863     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1864     BlockdevBackup *backup;
1865     BlockDriverState *bs;
1866     BlockDriverState *target_bs;
1867     AioContext *aio_context;
1868     AioContext *old_context;
1869     int ret;
1870 
1871     assert(common->action->type == TRANSACTION_ACTION_KIND_BLOCKDEV_BACKUP);
1872     backup = common->action->u.blockdev_backup.data;
1873 
1874     bs = bdrv_lookup_bs(backup->device, backup->device, errp);
1875     if (!bs) {
1876         return;
1877     }
1878 
1879     target_bs = bdrv_lookup_bs(backup->target, backup->target, errp);
1880     if (!target_bs) {
1881         return;
1882     }
1883 
1884     /* Honor bdrv_try_change_aio_context() context acquisition requirements. */
1885     aio_context = bdrv_get_aio_context(bs);
1886     old_context = bdrv_get_aio_context(target_bs);
1887     aio_context_acquire(old_context);
1888 
1889     ret = bdrv_try_change_aio_context(target_bs, aio_context, NULL, errp);
1890     if (ret < 0) {
1891         aio_context_release(old_context);
1892         return;
1893     }
1894 
1895     aio_context_release(old_context);
1896     aio_context_acquire(aio_context);
1897     state->bs = bs;
1898 
1899     /* Paired with .clean() */
1900     bdrv_drained_begin(state->bs);
1901 
1902     state->job = do_backup_common(qapi_BlockdevBackup_base(backup),
1903                                   bs, target_bs, aio_context,
1904                                   common->block_job_txn, errp);
1905 
1906     aio_context_release(aio_context);
1907 }
1908 
1909 static void blockdev_backup_commit(BlkActionState *common)
1910 {
1911     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1912     AioContext *aio_context;
1913 
1914     aio_context = bdrv_get_aio_context(state->bs);
1915     aio_context_acquire(aio_context);
1916 
1917     assert(state->job);
1918     job_start(&state->job->job);
1919 
1920     aio_context_release(aio_context);
1921 }
1922 
1923 static void blockdev_backup_abort(BlkActionState *common)
1924 {
1925     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1926 
1927     if (state->job) {
1928         job_cancel_sync(&state->job->job, true);
1929     }
1930 }
1931 
1932 static void blockdev_backup_clean(BlkActionState *common)
1933 {
1934     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1935     AioContext *aio_context;
1936 
1937     if (!state->bs) {
1938         return;
1939     }
1940 
1941     aio_context = bdrv_get_aio_context(state->bs);
1942     aio_context_acquire(aio_context);
1943 
1944     bdrv_drained_end(state->bs);
1945 
1946     aio_context_release(aio_context);
1947 }
1948 
1949 typedef struct BlockDirtyBitmapState {
1950     BlkActionState common;
1951     BdrvDirtyBitmap *bitmap;
1952     BlockDriverState *bs;
1953     HBitmap *backup;
1954     bool prepared;
1955     bool was_enabled;
1956 } BlockDirtyBitmapState;
1957 
1958 static void block_dirty_bitmap_add_prepare(BlkActionState *common,
1959                                            Error **errp)
1960 {
1961     Error *local_err = NULL;
1962     BlockDirtyBitmapAdd *action;
1963     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
1964                                              common, common);
1965 
1966     if (action_check_completion_mode(common, errp) < 0) {
1967         return;
1968     }
1969 
1970     action = common->action->u.block_dirty_bitmap_add.data;
1971     /* AIO context taken and released within qmp_block_dirty_bitmap_add */
1972     qmp_block_dirty_bitmap_add(action->node, action->name,
1973                                action->has_granularity, action->granularity,
1974                                action->has_persistent, action->persistent,
1975                                action->has_disabled, action->disabled,
1976                                &local_err);
1977 
1978     if (!local_err) {
1979         state->prepared = true;
1980     } else {
1981         error_propagate(errp, local_err);
1982     }
1983 }
1984 
1985 static void block_dirty_bitmap_add_abort(BlkActionState *common)
1986 {
1987     BlockDirtyBitmapAdd *action;
1988     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
1989                                              common, common);
1990 
1991     action = common->action->u.block_dirty_bitmap_add.data;
1992     /* Should not be able to fail: IF the bitmap was added via .prepare(),
1993      * then the node reference and bitmap name must have been valid.
1994      */
1995     if (state->prepared) {
1996         qmp_block_dirty_bitmap_remove(action->node, action->name, &error_abort);
1997     }
1998 }
1999 
2000 static void block_dirty_bitmap_clear_prepare(BlkActionState *common,
2001                                              Error **errp)
2002 {
2003     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2004                                              common, common);
2005     BlockDirtyBitmap *action;
2006 
2007     if (action_check_completion_mode(common, errp) < 0) {
2008         return;
2009     }
2010 
2011     action = common->action->u.block_dirty_bitmap_clear.data;
2012     state->bitmap = block_dirty_bitmap_lookup(action->node,
2013                                               action->name,
2014                                               &state->bs,
2015                                               errp);
2016     if (!state->bitmap) {
2017         return;
2018     }
2019 
2020     if (bdrv_dirty_bitmap_check(state->bitmap, BDRV_BITMAP_DEFAULT, errp)) {
2021         return;
2022     }
2023 
2024     bdrv_clear_dirty_bitmap(state->bitmap, &state->backup);
2025 }
2026 
2027 static void block_dirty_bitmap_restore(BlkActionState *common)
2028 {
2029     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2030                                              common, common);
2031 
2032     if (state->backup) {
2033         bdrv_restore_dirty_bitmap(state->bitmap, state->backup);
2034     }
2035 }
2036 
2037 static void block_dirty_bitmap_free_backup(BlkActionState *common)
2038 {
2039     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2040                                              common, common);
2041 
2042     hbitmap_free(state->backup);
2043 }
2044 
2045 static void block_dirty_bitmap_enable_prepare(BlkActionState *common,
2046                                               Error **errp)
2047 {
2048     BlockDirtyBitmap *action;
2049     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2050                                              common, common);
2051 
2052     if (action_check_completion_mode(common, errp) < 0) {
2053         return;
2054     }
2055 
2056     action = common->action->u.block_dirty_bitmap_enable.data;
2057     state->bitmap = block_dirty_bitmap_lookup(action->node,
2058                                               action->name,
2059                                               NULL,
2060                                               errp);
2061     if (!state->bitmap) {
2062         return;
2063     }
2064 
2065     if (bdrv_dirty_bitmap_check(state->bitmap, BDRV_BITMAP_ALLOW_RO, errp)) {
2066         return;
2067     }
2068 
2069     state->was_enabled = bdrv_dirty_bitmap_enabled(state->bitmap);
2070     bdrv_enable_dirty_bitmap(state->bitmap);
2071 }
2072 
2073 static void block_dirty_bitmap_enable_abort(BlkActionState *common)
2074 {
2075     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2076                                              common, common);
2077 
2078     if (!state->was_enabled) {
2079         bdrv_disable_dirty_bitmap(state->bitmap);
2080     }
2081 }
2082 
2083 static void block_dirty_bitmap_disable_prepare(BlkActionState *common,
2084                                                Error **errp)
2085 {
2086     BlockDirtyBitmap *action;
2087     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2088                                              common, common);
2089 
2090     if (action_check_completion_mode(common, errp) < 0) {
2091         return;
2092     }
2093 
2094     action = common->action->u.block_dirty_bitmap_disable.data;
2095     state->bitmap = block_dirty_bitmap_lookup(action->node,
2096                                               action->name,
2097                                               NULL,
2098                                               errp);
2099     if (!state->bitmap) {
2100         return;
2101     }
2102 
2103     if (bdrv_dirty_bitmap_check(state->bitmap, BDRV_BITMAP_ALLOW_RO, errp)) {
2104         return;
2105     }
2106 
2107     state->was_enabled = bdrv_dirty_bitmap_enabled(state->bitmap);
2108     bdrv_disable_dirty_bitmap(state->bitmap);
2109 }
2110 
2111 static void block_dirty_bitmap_disable_abort(BlkActionState *common)
2112 {
2113     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2114                                              common, common);
2115 
2116     if (state->was_enabled) {
2117         bdrv_enable_dirty_bitmap(state->bitmap);
2118     }
2119 }
2120 
2121 static void block_dirty_bitmap_merge_prepare(BlkActionState *common,
2122                                              Error **errp)
2123 {
2124     BlockDirtyBitmapMerge *action;
2125     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2126                                              common, common);
2127 
2128     if (action_check_completion_mode(common, errp) < 0) {
2129         return;
2130     }
2131 
2132     action = common->action->u.block_dirty_bitmap_merge.data;
2133 
2134     state->bitmap = block_dirty_bitmap_merge(action->node, action->target,
2135                                              action->bitmaps, &state->backup,
2136                                              errp);
2137 }
2138 
2139 static void block_dirty_bitmap_remove_prepare(BlkActionState *common,
2140                                               Error **errp)
2141 {
2142     BlockDirtyBitmap *action;
2143     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2144                                              common, common);
2145 
2146     if (action_check_completion_mode(common, errp) < 0) {
2147         return;
2148     }
2149 
2150     action = common->action->u.block_dirty_bitmap_remove.data;
2151 
2152     state->bitmap = block_dirty_bitmap_remove(action->node, action->name,
2153                                               false, &state->bs, errp);
2154     if (state->bitmap) {
2155         bdrv_dirty_bitmap_skip_store(state->bitmap, true);
2156         bdrv_dirty_bitmap_set_busy(state->bitmap, true);
2157     }
2158 }
2159 
2160 static void block_dirty_bitmap_remove_abort(BlkActionState *common)
2161 {
2162     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2163                                              common, common);
2164 
2165     if (state->bitmap) {
2166         bdrv_dirty_bitmap_skip_store(state->bitmap, false);
2167         bdrv_dirty_bitmap_set_busy(state->bitmap, false);
2168     }
2169 }
2170 
2171 static void block_dirty_bitmap_remove_commit(BlkActionState *common)
2172 {
2173     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2174                                              common, common);
2175 
2176     bdrv_dirty_bitmap_set_busy(state->bitmap, false);
2177     bdrv_release_dirty_bitmap(state->bitmap);
2178 }
2179 
2180 static void abort_prepare(BlkActionState *common, Error **errp)
2181 {
2182     error_setg(errp, "Transaction aborted using Abort action");
2183 }
2184 
2185 static void abort_commit(BlkActionState *common)
2186 {
2187     g_assert_not_reached(); /* this action never succeeds */
2188 }
2189 
2190 static const BlkActionOps actions[] = {
2191     [TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT] = {
2192         .instance_size = sizeof(ExternalSnapshotState),
2193         .prepare  = external_snapshot_prepare,
2194         .commit   = external_snapshot_commit,
2195         .abort = external_snapshot_abort,
2196         .clean = external_snapshot_clean,
2197     },
2198     [TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC] = {
2199         .instance_size = sizeof(ExternalSnapshotState),
2200         .prepare  = external_snapshot_prepare,
2201         .commit   = external_snapshot_commit,
2202         .abort = external_snapshot_abort,
2203         .clean = external_snapshot_clean,
2204     },
2205     [TRANSACTION_ACTION_KIND_DRIVE_BACKUP] = {
2206         .instance_size = sizeof(DriveBackupState),
2207         .prepare = drive_backup_prepare,
2208         .commit = drive_backup_commit,
2209         .abort = drive_backup_abort,
2210         .clean = drive_backup_clean,
2211     },
2212     [TRANSACTION_ACTION_KIND_BLOCKDEV_BACKUP] = {
2213         .instance_size = sizeof(BlockdevBackupState),
2214         .prepare = blockdev_backup_prepare,
2215         .commit = blockdev_backup_commit,
2216         .abort = blockdev_backup_abort,
2217         .clean = blockdev_backup_clean,
2218     },
2219     [TRANSACTION_ACTION_KIND_ABORT] = {
2220         .instance_size = sizeof(BlkActionState),
2221         .prepare = abort_prepare,
2222         .commit = abort_commit,
2223     },
2224     [TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC] = {
2225         .instance_size = sizeof(InternalSnapshotState),
2226         .prepare  = internal_snapshot_prepare,
2227         .abort = internal_snapshot_abort,
2228         .clean = internal_snapshot_clean,
2229     },
2230     [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_ADD] = {
2231         .instance_size = sizeof(BlockDirtyBitmapState),
2232         .prepare = block_dirty_bitmap_add_prepare,
2233         .abort = block_dirty_bitmap_add_abort,
2234     },
2235     [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_CLEAR] = {
2236         .instance_size = sizeof(BlockDirtyBitmapState),
2237         .prepare = block_dirty_bitmap_clear_prepare,
2238         .commit = block_dirty_bitmap_free_backup,
2239         .abort = block_dirty_bitmap_restore,
2240     },
2241     [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_ENABLE] = {
2242         .instance_size = sizeof(BlockDirtyBitmapState),
2243         .prepare = block_dirty_bitmap_enable_prepare,
2244         .abort = block_dirty_bitmap_enable_abort,
2245     },
2246     [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_DISABLE] = {
2247         .instance_size = sizeof(BlockDirtyBitmapState),
2248         .prepare = block_dirty_bitmap_disable_prepare,
2249         .abort = block_dirty_bitmap_disable_abort,
2250     },
2251     [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_MERGE] = {
2252         .instance_size = sizeof(BlockDirtyBitmapState),
2253         .prepare = block_dirty_bitmap_merge_prepare,
2254         .commit = block_dirty_bitmap_free_backup,
2255         .abort = block_dirty_bitmap_restore,
2256     },
2257     [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_REMOVE] = {
2258         .instance_size = sizeof(BlockDirtyBitmapState),
2259         .prepare = block_dirty_bitmap_remove_prepare,
2260         .commit = block_dirty_bitmap_remove_commit,
2261         .abort = block_dirty_bitmap_remove_abort,
2262     },
2263     /* Where are transactions for MIRROR, COMMIT and STREAM?
2264      * Although these blockjobs use transaction callbacks like the backup job,
2265      * these jobs do not necessarily adhere to transaction semantics.
2266      * These jobs may not fully undo all of their actions on abort, nor do they
2267      * necessarily work in transactions with more than one job in them.
2268      */
2269 };
2270 
2271 /**
2272  * Allocate a TransactionProperties structure if necessary, and fill
2273  * that structure with desired defaults if they are unset.
2274  */
2275 static TransactionProperties *get_transaction_properties(
2276     TransactionProperties *props)
2277 {
2278     if (!props) {
2279         props = g_new0(TransactionProperties, 1);
2280     }
2281 
2282     if (!props->has_completion_mode) {
2283         props->has_completion_mode = true;
2284         props->completion_mode = ACTION_COMPLETION_MODE_INDIVIDUAL;
2285     }
2286 
2287     return props;
2288 }
2289 
2290 /*
2291  * 'Atomic' group operations.  The operations are performed as a set, and if
2292  * any fail then we roll back all operations in the group.
2293  *
2294  * Always run under BQL.
2295  */
2296 void qmp_transaction(TransactionActionList *dev_list,
2297                      struct TransactionProperties *props,
2298                      Error **errp)
2299 {
2300     TransactionActionList *dev_entry = dev_list;
2301     bool has_props = !!props;
2302     JobTxn *block_job_txn = NULL;
2303     BlkActionState *state, *next;
2304     Error *local_err = NULL;
2305 
2306     GLOBAL_STATE_CODE();
2307 
2308     QTAILQ_HEAD(, BlkActionState) snap_bdrv_states;
2309     QTAILQ_INIT(&snap_bdrv_states);
2310 
2311     /* Does this transaction get canceled as a group on failure?
2312      * If not, we don't really need to make a JobTxn.
2313      */
2314     props = get_transaction_properties(props);
2315     if (props->completion_mode != ACTION_COMPLETION_MODE_INDIVIDUAL) {
2316         block_job_txn = job_txn_new();
2317     }
2318 
2319     /* drain all i/o before any operations */
2320     bdrv_drain_all();
2321 
2322     /* We don't do anything in this loop that commits us to the operations */
2323     while (NULL != dev_entry) {
2324         TransactionAction *dev_info = NULL;
2325         const BlkActionOps *ops;
2326 
2327         dev_info = dev_entry->value;
2328         dev_entry = dev_entry->next;
2329 
2330         assert(dev_info->type < ARRAY_SIZE(actions));
2331 
2332         ops = &actions[dev_info->type];
2333         assert(ops->instance_size > 0);
2334 
2335         state = g_malloc0(ops->instance_size);
2336         state->ops = ops;
2337         state->action = dev_info;
2338         state->block_job_txn = block_job_txn;
2339         state->txn_props = props;
2340         QTAILQ_INSERT_TAIL(&snap_bdrv_states, state, entry);
2341 
2342         state->ops->prepare(state, &local_err);
2343         if (local_err) {
2344             error_propagate(errp, local_err);
2345             goto delete_and_fail;
2346         }
2347     }
2348 
2349     QTAILQ_FOREACH(state, &snap_bdrv_states, entry) {
2350         if (state->ops->commit) {
2351             state->ops->commit(state);
2352         }
2353     }
2354 
2355     /* success */
2356     goto exit;
2357 
2358 delete_and_fail:
2359     /* failure, and it is all-or-none; roll back all operations */
2360     QTAILQ_FOREACH_REVERSE(state, &snap_bdrv_states, entry) {
2361         if (state->ops->abort) {
2362             state->ops->abort(state);
2363         }
2364     }
2365 exit:
2366     QTAILQ_FOREACH_SAFE(state, &snap_bdrv_states, entry, next) {
2367         if (state->ops->clean) {
2368             state->ops->clean(state);
2369         }
2370         g_free(state);
2371     }
2372     if (!has_props) {
2373         qapi_free_TransactionProperties(props);
2374     }
2375     job_txn_unref(block_job_txn);
2376 }
2377 
2378 BlockDirtyBitmapSha256 *qmp_x_debug_block_dirty_bitmap_sha256(const char *node,
2379                                                               const char *name,
2380                                                               Error **errp)
2381 {
2382     BdrvDirtyBitmap *bitmap;
2383     BlockDriverState *bs;
2384     BlockDirtyBitmapSha256 *ret = NULL;
2385     char *sha256;
2386 
2387     bitmap = block_dirty_bitmap_lookup(node, name, &bs, errp);
2388     if (!bitmap || !bs) {
2389         return NULL;
2390     }
2391 
2392     sha256 = bdrv_dirty_bitmap_sha256(bitmap, errp);
2393     if (sha256 == NULL) {
2394         return NULL;
2395     }
2396 
2397     ret = g_new(BlockDirtyBitmapSha256, 1);
2398     ret->sha256 = sha256;
2399 
2400     return ret;
2401 }
2402 
2403 void coroutine_fn qmp_block_resize(const char *device, const char *node_name,
2404                                    int64_t size, Error **errp)
2405 {
2406     Error *local_err = NULL;
2407     BlockBackend *blk;
2408     BlockDriverState *bs;
2409     AioContext *old_ctx;
2410 
2411     bs = bdrv_lookup_bs(device, node_name, &local_err);
2412     if (local_err) {
2413         error_propagate(errp, local_err);
2414         return;
2415     }
2416 
2417     if (size < 0) {
2418         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "size", "a >0 size");
2419         return;
2420     }
2421 
2422     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_RESIZE, NULL)) {
2423         error_setg(errp, QERR_DEVICE_IN_USE, device);
2424         return;
2425     }
2426 
2427     blk = blk_new_with_bs(bs, BLK_PERM_RESIZE, BLK_PERM_ALL, errp);
2428     if (!blk) {
2429         return;
2430     }
2431 
2432     bdrv_co_lock(bs);
2433     bdrv_drained_begin(bs);
2434     bdrv_co_unlock(bs);
2435 
2436     old_ctx = bdrv_co_enter(bs);
2437     blk_co_truncate(blk, size, false, PREALLOC_MODE_OFF, 0, errp);
2438     bdrv_co_leave(bs, old_ctx);
2439 
2440     bdrv_co_lock(bs);
2441     bdrv_drained_end(bs);
2442     blk_unref(blk);
2443     bdrv_co_unlock(bs);
2444 }
2445 
2446 void qmp_block_stream(const char *job_id, const char *device,
2447                       const char *base,
2448                       const char *base_node,
2449                       const char *backing_file,
2450                       const char *bottom,
2451                       bool has_speed, int64_t speed,
2452                       bool has_on_error, BlockdevOnError on_error,
2453                       const char *filter_node_name,
2454                       bool has_auto_finalize, bool auto_finalize,
2455                       bool has_auto_dismiss, bool auto_dismiss,
2456                       Error **errp)
2457 {
2458     BlockDriverState *bs, *iter, *iter_end;
2459     BlockDriverState *base_bs = NULL;
2460     BlockDriverState *bottom_bs = NULL;
2461     AioContext *aio_context;
2462     Error *local_err = NULL;
2463     int job_flags = JOB_DEFAULT;
2464 
2465     if (base && base_node) {
2466         error_setg(errp, "'base' and 'base-node' cannot be specified "
2467                    "at the same time");
2468         return;
2469     }
2470 
2471     if (base && bottom) {
2472         error_setg(errp, "'base' and 'bottom' cannot be specified "
2473                    "at the same time");
2474         return;
2475     }
2476 
2477     if (bottom && base_node) {
2478         error_setg(errp, "'bottom' and 'base-node' cannot be specified "
2479                    "at the same time");
2480         return;
2481     }
2482 
2483     if (!has_on_error) {
2484         on_error = BLOCKDEV_ON_ERROR_REPORT;
2485     }
2486 
2487     bs = bdrv_lookup_bs(device, device, errp);
2488     if (!bs) {
2489         return;
2490     }
2491 
2492     aio_context = bdrv_get_aio_context(bs);
2493     aio_context_acquire(aio_context);
2494 
2495     if (base) {
2496         base_bs = bdrv_find_backing_image(bs, base);
2497         if (base_bs == NULL) {
2498             error_setg(errp, "Can't find '%s' in the backing chain", base);
2499             goto out;
2500         }
2501         assert(bdrv_get_aio_context(base_bs) == aio_context);
2502     }
2503 
2504     if (base_node) {
2505         base_bs = bdrv_lookup_bs(NULL, base_node, errp);
2506         if (!base_bs) {
2507             goto out;
2508         }
2509         if (bs == base_bs || !bdrv_chain_contains(bs, base_bs)) {
2510             error_setg(errp, "Node '%s' is not a backing image of '%s'",
2511                        base_node, device);
2512             goto out;
2513         }
2514         assert(bdrv_get_aio_context(base_bs) == aio_context);
2515         bdrv_refresh_filename(base_bs);
2516     }
2517 
2518     if (bottom) {
2519         bottom_bs = bdrv_lookup_bs(NULL, bottom, errp);
2520         if (!bottom_bs) {
2521             goto out;
2522         }
2523         if (!bottom_bs->drv) {
2524             error_setg(errp, "Node '%s' is not open", bottom);
2525             goto out;
2526         }
2527         if (bottom_bs->drv->is_filter) {
2528             error_setg(errp, "Node '%s' is a filter, use a non-filter node "
2529                        "as 'bottom'", bottom);
2530             goto out;
2531         }
2532         if (!bdrv_chain_contains(bs, bottom_bs)) {
2533             error_setg(errp, "Node '%s' is not in a chain starting from '%s'",
2534                        bottom, device);
2535             goto out;
2536         }
2537         assert(bdrv_get_aio_context(bottom_bs) == aio_context);
2538     }
2539 
2540     /*
2541      * Check for op blockers in the whole chain between bs and base (or bottom)
2542      */
2543     iter_end = bottom ? bdrv_filter_or_cow_bs(bottom_bs) : base_bs;
2544     for (iter = bs; iter && iter != iter_end;
2545          iter = bdrv_filter_or_cow_bs(iter))
2546     {
2547         if (bdrv_op_is_blocked(iter, BLOCK_OP_TYPE_STREAM, errp)) {
2548             goto out;
2549         }
2550     }
2551 
2552     /* if we are streaming the entire chain, the result will have no backing
2553      * file, and specifying one is therefore an error */
2554     if (!base_bs && backing_file) {
2555         error_setg(errp, "backing file specified, but streaming the "
2556                          "entire chain");
2557         goto out;
2558     }
2559 
2560     if (has_auto_finalize && !auto_finalize) {
2561         job_flags |= JOB_MANUAL_FINALIZE;
2562     }
2563     if (has_auto_dismiss && !auto_dismiss) {
2564         job_flags |= JOB_MANUAL_DISMISS;
2565     }
2566 
2567     stream_start(job_id, bs, base_bs, backing_file,
2568                  bottom_bs, job_flags, has_speed ? speed : 0, on_error,
2569                  filter_node_name, &local_err);
2570     if (local_err) {
2571         error_propagate(errp, local_err);
2572         goto out;
2573     }
2574 
2575     trace_qmp_block_stream(bs);
2576 
2577 out:
2578     aio_context_release(aio_context);
2579 }
2580 
2581 void qmp_block_commit(const char *job_id, const char *device,
2582                       const char *base_node,
2583                       const char *base,
2584                       const char *top_node,
2585                       const char *top,
2586                       const char *backing_file,
2587                       bool has_speed, int64_t speed,
2588                       bool has_on_error, BlockdevOnError on_error,
2589                       const char *filter_node_name,
2590                       bool has_auto_finalize, bool auto_finalize,
2591                       bool has_auto_dismiss, bool auto_dismiss,
2592                       Error **errp)
2593 {
2594     BlockDriverState *bs;
2595     BlockDriverState *iter;
2596     BlockDriverState *base_bs, *top_bs;
2597     AioContext *aio_context;
2598     Error *local_err = NULL;
2599     int job_flags = JOB_DEFAULT;
2600     uint64_t top_perm, top_shared;
2601 
2602     if (!has_speed) {
2603         speed = 0;
2604     }
2605     if (!has_on_error) {
2606         on_error = BLOCKDEV_ON_ERROR_REPORT;
2607     }
2608     if (has_auto_finalize && !auto_finalize) {
2609         job_flags |= JOB_MANUAL_FINALIZE;
2610     }
2611     if (has_auto_dismiss && !auto_dismiss) {
2612         job_flags |= JOB_MANUAL_DISMISS;
2613     }
2614 
2615     /* Important Note:
2616      *  libvirt relies on the DeviceNotFound error class in order to probe for
2617      *  live commit feature versions; for this to work, we must make sure to
2618      *  perform the device lookup before any generic errors that may occur in a
2619      *  scenario in which all optional arguments are omitted. */
2620     bs = qmp_get_root_bs(device, &local_err);
2621     if (!bs) {
2622         bs = bdrv_lookup_bs(device, device, NULL);
2623         if (!bs) {
2624             error_free(local_err);
2625             error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
2626                       "Device '%s' not found", device);
2627         } else {
2628             error_propagate(errp, local_err);
2629         }
2630         return;
2631     }
2632 
2633     aio_context = bdrv_get_aio_context(bs);
2634     aio_context_acquire(aio_context);
2635 
2636     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_COMMIT_SOURCE, errp)) {
2637         goto out;
2638     }
2639 
2640     /* default top_bs is the active layer */
2641     top_bs = bs;
2642 
2643     if (top_node && top) {
2644         error_setg(errp, "'top-node' and 'top' are mutually exclusive");
2645         goto out;
2646     } else if (top_node) {
2647         top_bs = bdrv_lookup_bs(NULL, top_node, errp);
2648         if (top_bs == NULL) {
2649             goto out;
2650         }
2651         if (!bdrv_chain_contains(bs, top_bs)) {
2652             error_setg(errp, "'%s' is not in this backing file chain",
2653                        top_node);
2654             goto out;
2655         }
2656     } else if (top) {
2657         /* This strcmp() is just a shortcut, there is no need to
2658          * refresh @bs's filename.  If it mismatches,
2659          * bdrv_find_backing_image() will do the refresh and may still
2660          * return @bs. */
2661         if (strcmp(bs->filename, top) != 0) {
2662             top_bs = bdrv_find_backing_image(bs, top);
2663         }
2664     }
2665 
2666     if (top_bs == NULL) {
2667         error_setg(errp, "Top image file %s not found", top ? top : "NULL");
2668         goto out;
2669     }
2670 
2671     assert(bdrv_get_aio_context(top_bs) == aio_context);
2672 
2673     if (base_node && base) {
2674         error_setg(errp, "'base-node' and 'base' are mutually exclusive");
2675         goto out;
2676     } else if (base_node) {
2677         base_bs = bdrv_lookup_bs(NULL, base_node, errp);
2678         if (base_bs == NULL) {
2679             goto out;
2680         }
2681         if (!bdrv_chain_contains(top_bs, base_bs)) {
2682             error_setg(errp, "'%s' is not in this backing file chain",
2683                        base_node);
2684             goto out;
2685         }
2686     } else if (base) {
2687         base_bs = bdrv_find_backing_image(top_bs, base);
2688         if (base_bs == NULL) {
2689             error_setg(errp, "Can't find '%s' in the backing chain", base);
2690             goto out;
2691         }
2692     } else {
2693         base_bs = bdrv_find_base(top_bs);
2694         if (base_bs == NULL) {
2695             error_setg(errp, "There is no backimg image");
2696             goto out;
2697         }
2698     }
2699 
2700     assert(bdrv_get_aio_context(base_bs) == aio_context);
2701 
2702     for (iter = top_bs; iter != bdrv_filter_or_cow_bs(base_bs);
2703          iter = bdrv_filter_or_cow_bs(iter))
2704     {
2705         if (bdrv_op_is_blocked(iter, BLOCK_OP_TYPE_COMMIT_TARGET, errp)) {
2706             goto out;
2707         }
2708     }
2709 
2710     /* Do not allow attempts to commit an image into itself */
2711     if (top_bs == base_bs) {
2712         error_setg(errp, "cannot commit an image into itself");
2713         goto out;
2714     }
2715 
2716     /*
2717      * Active commit is required if and only if someone has taken a
2718      * WRITE permission on the top node.  Historically, we have always
2719      * used active commit for top nodes, so continue that practice
2720      * lest we possibly break clients that rely on this behavior, e.g.
2721      * to later attach this node to a writing parent.
2722      * (Active commit is never really wrong.)
2723      */
2724     bdrv_get_cumulative_perm(top_bs, &top_perm, &top_shared);
2725     if (top_perm & BLK_PERM_WRITE ||
2726         bdrv_skip_filters(top_bs) == bdrv_skip_filters(bs))
2727     {
2728         if (backing_file) {
2729             if (bdrv_skip_filters(top_bs) == bdrv_skip_filters(bs)) {
2730                 error_setg(errp, "'backing-file' specified,"
2731                                  " but 'top' is the active layer");
2732             } else {
2733                 error_setg(errp, "'backing-file' specified, but 'top' has a "
2734                                  "writer on it");
2735             }
2736             goto out;
2737         }
2738         if (!job_id) {
2739             /*
2740              * Emulate here what block_job_create() does, because it
2741              * is possible that @bs != @top_bs (the block job should
2742              * be named after @bs, even if @top_bs is the actual
2743              * source)
2744              */
2745             job_id = bdrv_get_device_name(bs);
2746         }
2747         commit_active_start(job_id, top_bs, base_bs, job_flags, speed, on_error,
2748                             filter_node_name, NULL, NULL, false, &local_err);
2749     } else {
2750         BlockDriverState *overlay_bs = bdrv_find_overlay(bs, top_bs);
2751         if (bdrv_op_is_blocked(overlay_bs, BLOCK_OP_TYPE_COMMIT_TARGET, errp)) {
2752             goto out;
2753         }
2754         commit_start(job_id, bs, base_bs, top_bs, job_flags,
2755                      speed, on_error, backing_file,
2756                      filter_node_name, &local_err);
2757     }
2758     if (local_err != NULL) {
2759         error_propagate(errp, local_err);
2760         goto out;
2761     }
2762 
2763 out:
2764     aio_context_release(aio_context);
2765 }
2766 
2767 /* Common QMP interface for drive-backup and blockdev-backup */
2768 static BlockJob *do_backup_common(BackupCommon *backup,
2769                                   BlockDriverState *bs,
2770                                   BlockDriverState *target_bs,
2771                                   AioContext *aio_context,
2772                                   JobTxn *txn, Error **errp)
2773 {
2774     BlockJob *job = NULL;
2775     BdrvDirtyBitmap *bmap = NULL;
2776     BackupPerf perf = { .max_workers = 64 };
2777     int job_flags = JOB_DEFAULT;
2778 
2779     if (!backup->has_speed) {
2780         backup->speed = 0;
2781     }
2782     if (!backup->has_on_source_error) {
2783         backup->on_source_error = BLOCKDEV_ON_ERROR_REPORT;
2784     }
2785     if (!backup->has_on_target_error) {
2786         backup->on_target_error = BLOCKDEV_ON_ERROR_REPORT;
2787     }
2788     if (!backup->has_auto_finalize) {
2789         backup->auto_finalize = true;
2790     }
2791     if (!backup->has_auto_dismiss) {
2792         backup->auto_dismiss = true;
2793     }
2794     if (!backup->has_compress) {
2795         backup->compress = false;
2796     }
2797 
2798     if (backup->x_perf) {
2799         if (backup->x_perf->has_use_copy_range) {
2800             perf.use_copy_range = backup->x_perf->use_copy_range;
2801         }
2802         if (backup->x_perf->has_max_workers) {
2803             perf.max_workers = backup->x_perf->max_workers;
2804         }
2805         if (backup->x_perf->has_max_chunk) {
2806             perf.max_chunk = backup->x_perf->max_chunk;
2807         }
2808     }
2809 
2810     if ((backup->sync == MIRROR_SYNC_MODE_BITMAP) ||
2811         (backup->sync == MIRROR_SYNC_MODE_INCREMENTAL)) {
2812         /* done before desugaring 'incremental' to print the right message */
2813         if (!backup->bitmap) {
2814             error_setg(errp, "must provide a valid bitmap name for "
2815                        "'%s' sync mode", MirrorSyncMode_str(backup->sync));
2816             return NULL;
2817         }
2818     }
2819 
2820     if (backup->sync == MIRROR_SYNC_MODE_INCREMENTAL) {
2821         if (backup->has_bitmap_mode &&
2822             backup->bitmap_mode != BITMAP_SYNC_MODE_ON_SUCCESS) {
2823             error_setg(errp, "Bitmap sync mode must be '%s' "
2824                        "when using sync mode '%s'",
2825                        BitmapSyncMode_str(BITMAP_SYNC_MODE_ON_SUCCESS),
2826                        MirrorSyncMode_str(backup->sync));
2827             return NULL;
2828         }
2829         backup->has_bitmap_mode = true;
2830         backup->sync = MIRROR_SYNC_MODE_BITMAP;
2831         backup->bitmap_mode = BITMAP_SYNC_MODE_ON_SUCCESS;
2832     }
2833 
2834     if (backup->bitmap) {
2835         bmap = bdrv_find_dirty_bitmap(bs, backup->bitmap);
2836         if (!bmap) {
2837             error_setg(errp, "Bitmap '%s' could not be found", backup->bitmap);
2838             return NULL;
2839         }
2840         if (!backup->has_bitmap_mode) {
2841             error_setg(errp, "Bitmap sync mode must be given "
2842                        "when providing a bitmap");
2843             return NULL;
2844         }
2845         if (bdrv_dirty_bitmap_check(bmap, BDRV_BITMAP_ALLOW_RO, errp)) {
2846             return NULL;
2847         }
2848 
2849         /* This does not produce a useful bitmap artifact: */
2850         if (backup->sync == MIRROR_SYNC_MODE_NONE) {
2851             error_setg(errp, "sync mode '%s' does not produce meaningful bitmap"
2852                        " outputs", MirrorSyncMode_str(backup->sync));
2853             return NULL;
2854         }
2855 
2856         /* If the bitmap isn't used for input or output, this is useless: */
2857         if (backup->bitmap_mode == BITMAP_SYNC_MODE_NEVER &&
2858             backup->sync != MIRROR_SYNC_MODE_BITMAP) {
2859             error_setg(errp, "Bitmap sync mode '%s' has no meaningful effect"
2860                        " when combined with sync mode '%s'",
2861                        BitmapSyncMode_str(backup->bitmap_mode),
2862                        MirrorSyncMode_str(backup->sync));
2863             return NULL;
2864         }
2865     }
2866 
2867     if (!backup->bitmap && backup->has_bitmap_mode) {
2868         error_setg(errp, "Cannot specify bitmap sync mode without a bitmap");
2869         return NULL;
2870     }
2871 
2872     if (!backup->auto_finalize) {
2873         job_flags |= JOB_MANUAL_FINALIZE;
2874     }
2875     if (!backup->auto_dismiss) {
2876         job_flags |= JOB_MANUAL_DISMISS;
2877     }
2878 
2879     job = backup_job_create(backup->job_id, bs, target_bs, backup->speed,
2880                             backup->sync, bmap, backup->bitmap_mode,
2881                             backup->compress,
2882                             backup->filter_node_name,
2883                             &perf,
2884                             backup->on_source_error,
2885                             backup->on_target_error,
2886                             job_flags, NULL, NULL, txn, errp);
2887     return job;
2888 }
2889 
2890 void qmp_drive_backup(DriveBackup *backup, Error **errp)
2891 {
2892     TransactionAction action = {
2893         .type = TRANSACTION_ACTION_KIND_DRIVE_BACKUP,
2894         .u.drive_backup.data = backup,
2895     };
2896     blockdev_do_action(&action, errp);
2897 }
2898 
2899 BlockDeviceInfoList *qmp_query_named_block_nodes(bool has_flat,
2900                                                  bool flat,
2901                                                  Error **errp)
2902 {
2903     bool return_flat = has_flat && flat;
2904 
2905     return bdrv_named_nodes_list(return_flat, errp);
2906 }
2907 
2908 XDbgBlockGraph *qmp_x_debug_query_block_graph(Error **errp)
2909 {
2910     return bdrv_get_xdbg_block_graph(errp);
2911 }
2912 
2913 void qmp_blockdev_backup(BlockdevBackup *backup, Error **errp)
2914 {
2915     TransactionAction action = {
2916         .type = TRANSACTION_ACTION_KIND_BLOCKDEV_BACKUP,
2917         .u.blockdev_backup.data = backup,
2918     };
2919     blockdev_do_action(&action, errp);
2920 }
2921 
2922 /* Parameter check and block job starting for drive mirroring.
2923  * Caller should hold @device and @target's aio context (must be the same).
2924  **/
2925 static void blockdev_mirror_common(const char *job_id, BlockDriverState *bs,
2926                                    BlockDriverState *target,
2927                                    const char *replaces,
2928                                    enum MirrorSyncMode sync,
2929                                    BlockMirrorBackingMode backing_mode,
2930                                    bool zero_target,
2931                                    bool has_speed, int64_t speed,
2932                                    bool has_granularity, uint32_t granularity,
2933                                    bool has_buf_size, int64_t buf_size,
2934                                    bool has_on_source_error,
2935                                    BlockdevOnError on_source_error,
2936                                    bool has_on_target_error,
2937                                    BlockdevOnError on_target_error,
2938                                    bool has_unmap, bool unmap,
2939                                    const char *filter_node_name,
2940                                    bool has_copy_mode, MirrorCopyMode copy_mode,
2941                                    bool has_auto_finalize, bool auto_finalize,
2942                                    bool has_auto_dismiss, bool auto_dismiss,
2943                                    Error **errp)
2944 {
2945     BlockDriverState *unfiltered_bs;
2946     int job_flags = JOB_DEFAULT;
2947 
2948     if (!has_speed) {
2949         speed = 0;
2950     }
2951     if (!has_on_source_error) {
2952         on_source_error = BLOCKDEV_ON_ERROR_REPORT;
2953     }
2954     if (!has_on_target_error) {
2955         on_target_error = BLOCKDEV_ON_ERROR_REPORT;
2956     }
2957     if (!has_granularity) {
2958         granularity = 0;
2959     }
2960     if (!has_buf_size) {
2961         buf_size = 0;
2962     }
2963     if (!has_unmap) {
2964         unmap = true;
2965     }
2966     if (!has_copy_mode) {
2967         copy_mode = MIRROR_COPY_MODE_BACKGROUND;
2968     }
2969     if (has_auto_finalize && !auto_finalize) {
2970         job_flags |= JOB_MANUAL_FINALIZE;
2971     }
2972     if (has_auto_dismiss && !auto_dismiss) {
2973         job_flags |= JOB_MANUAL_DISMISS;
2974     }
2975 
2976     if (granularity != 0 && (granularity < 512 || granularity > 1048576 * 64)) {
2977         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "granularity",
2978                    "a value in range [512B, 64MB]");
2979         return;
2980     }
2981     if (granularity & (granularity - 1)) {
2982         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "granularity",
2983                    "a power of 2");
2984         return;
2985     }
2986 
2987     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_MIRROR_SOURCE, errp)) {
2988         return;
2989     }
2990     if (bdrv_op_is_blocked(target, BLOCK_OP_TYPE_MIRROR_TARGET, errp)) {
2991         return;
2992     }
2993 
2994     if (!bdrv_backing_chain_next(bs) && sync == MIRROR_SYNC_MODE_TOP) {
2995         sync = MIRROR_SYNC_MODE_FULL;
2996     }
2997 
2998     if (!replaces) {
2999         /* We want to mirror from @bs, but keep implicit filters on top */
3000         unfiltered_bs = bdrv_skip_implicit_filters(bs);
3001         if (unfiltered_bs != bs) {
3002             replaces = unfiltered_bs->node_name;
3003         }
3004     }
3005 
3006     if (replaces) {
3007         BlockDriverState *to_replace_bs;
3008         AioContext *replace_aio_context;
3009         int64_t bs_size, replace_size;
3010 
3011         bs_size = bdrv_getlength(bs);
3012         if (bs_size < 0) {
3013             error_setg_errno(errp, -bs_size, "Failed to query device's size");
3014             return;
3015         }
3016 
3017         to_replace_bs = check_to_replace_node(bs, replaces, errp);
3018         if (!to_replace_bs) {
3019             return;
3020         }
3021 
3022         replace_aio_context = bdrv_get_aio_context(to_replace_bs);
3023         aio_context_acquire(replace_aio_context);
3024         replace_size = bdrv_getlength(to_replace_bs);
3025         aio_context_release(replace_aio_context);
3026 
3027         if (replace_size < 0) {
3028             error_setg_errno(errp, -replace_size,
3029                              "Failed to query the replacement node's size");
3030             return;
3031         }
3032         if (bs_size != replace_size) {
3033             error_setg(errp, "cannot replace image with a mirror image of "
3034                              "different size");
3035             return;
3036         }
3037     }
3038 
3039     /* pass the node name to replace to mirror start since it's loose coupling
3040      * and will allow to check whether the node still exist at mirror completion
3041      */
3042     mirror_start(job_id, bs, target,
3043                  replaces, job_flags,
3044                  speed, granularity, buf_size, sync, backing_mode, zero_target,
3045                  on_source_error, on_target_error, unmap, filter_node_name,
3046                  copy_mode, errp);
3047 }
3048 
3049 void qmp_drive_mirror(DriveMirror *arg, Error **errp)
3050 {
3051     BlockDriverState *bs;
3052     BlockDriverState *target_backing_bs, *target_bs;
3053     AioContext *aio_context;
3054     AioContext *old_context;
3055     BlockMirrorBackingMode backing_mode;
3056     Error *local_err = NULL;
3057     QDict *options = NULL;
3058     int flags;
3059     int64_t size;
3060     const char *format = arg->format;
3061     bool zero_target;
3062     int ret;
3063 
3064     bs = qmp_get_root_bs(arg->device, errp);
3065     if (!bs) {
3066         return;
3067     }
3068 
3069     /* Early check to avoid creating target */
3070     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_MIRROR_SOURCE, errp)) {
3071         return;
3072     }
3073 
3074     aio_context = bdrv_get_aio_context(bs);
3075     aio_context_acquire(aio_context);
3076 
3077     if (!arg->has_mode) {
3078         arg->mode = NEW_IMAGE_MODE_ABSOLUTE_PATHS;
3079     }
3080 
3081     if (!arg->format) {
3082         format = (arg->mode == NEW_IMAGE_MODE_EXISTING
3083                   ? NULL : bs->drv->format_name);
3084     }
3085 
3086     flags = bs->open_flags | BDRV_O_RDWR;
3087     target_backing_bs = bdrv_cow_bs(bdrv_skip_filters(bs));
3088     if (!target_backing_bs && arg->sync == MIRROR_SYNC_MODE_TOP) {
3089         arg->sync = MIRROR_SYNC_MODE_FULL;
3090     }
3091     if (arg->sync == MIRROR_SYNC_MODE_NONE) {
3092         target_backing_bs = bs;
3093     }
3094 
3095     size = bdrv_getlength(bs);
3096     if (size < 0) {
3097         error_setg_errno(errp, -size, "bdrv_getlength failed");
3098         goto out;
3099     }
3100 
3101     if (arg->replaces) {
3102         if (!arg->node_name) {
3103             error_setg(errp, "a node-name must be provided when replacing a"
3104                              " named node of the graph");
3105             goto out;
3106         }
3107     }
3108 
3109     if (arg->mode == NEW_IMAGE_MODE_ABSOLUTE_PATHS) {
3110         backing_mode = MIRROR_SOURCE_BACKING_CHAIN;
3111     } else {
3112         backing_mode = MIRROR_OPEN_BACKING_CHAIN;
3113     }
3114 
3115     /* Don't open backing image in create() */
3116     flags |= BDRV_O_NO_BACKING;
3117 
3118     if ((arg->sync == MIRROR_SYNC_MODE_FULL || !target_backing_bs)
3119         && arg->mode != NEW_IMAGE_MODE_EXISTING)
3120     {
3121         /* create new image w/o backing file */
3122         assert(format);
3123         bdrv_img_create(arg->target, format,
3124                         NULL, NULL, NULL, size, flags, false, &local_err);
3125     } else {
3126         /* Implicit filters should not appear in the filename */
3127         BlockDriverState *explicit_backing =
3128             bdrv_skip_implicit_filters(target_backing_bs);
3129 
3130         switch (arg->mode) {
3131         case NEW_IMAGE_MODE_EXISTING:
3132             break;
3133         case NEW_IMAGE_MODE_ABSOLUTE_PATHS:
3134             /* create new image with backing file */
3135             bdrv_refresh_filename(explicit_backing);
3136             bdrv_img_create(arg->target, format,
3137                             explicit_backing->filename,
3138                             explicit_backing->drv->format_name,
3139                             NULL, size, flags, false, &local_err);
3140             break;
3141         default:
3142             abort();
3143         }
3144     }
3145 
3146     if (local_err) {
3147         error_propagate(errp, local_err);
3148         goto out;
3149     }
3150 
3151     options = qdict_new();
3152     if (arg->node_name) {
3153         qdict_put_str(options, "node-name", arg->node_name);
3154     }
3155     if (format) {
3156         qdict_put_str(options, "driver", format);
3157     }
3158 
3159     /* Mirroring takes care of copy-on-write using the source's backing
3160      * file.
3161      */
3162     target_bs = bdrv_open(arg->target, NULL, options, flags, errp);
3163     if (!target_bs) {
3164         goto out;
3165     }
3166 
3167     zero_target = (arg->sync == MIRROR_SYNC_MODE_FULL &&
3168                    (arg->mode == NEW_IMAGE_MODE_EXISTING ||
3169                     !bdrv_has_zero_init(target_bs)));
3170 
3171 
3172     /* Honor bdrv_try_change_aio_context() context acquisition requirements. */
3173     old_context = bdrv_get_aio_context(target_bs);
3174     aio_context_release(aio_context);
3175     aio_context_acquire(old_context);
3176 
3177     ret = bdrv_try_change_aio_context(target_bs, aio_context, NULL, errp);
3178     if (ret < 0) {
3179         bdrv_unref(target_bs);
3180         aio_context_release(old_context);
3181         return;
3182     }
3183 
3184     aio_context_release(old_context);
3185     aio_context_acquire(aio_context);
3186 
3187     blockdev_mirror_common(arg->job_id, bs, target_bs,
3188                            arg->replaces, arg->sync,
3189                            backing_mode, zero_target,
3190                            arg->has_speed, arg->speed,
3191                            arg->has_granularity, arg->granularity,
3192                            arg->has_buf_size, arg->buf_size,
3193                            arg->has_on_source_error, arg->on_source_error,
3194                            arg->has_on_target_error, arg->on_target_error,
3195                            arg->has_unmap, arg->unmap,
3196                            NULL,
3197                            arg->has_copy_mode, arg->copy_mode,
3198                            arg->has_auto_finalize, arg->auto_finalize,
3199                            arg->has_auto_dismiss, arg->auto_dismiss,
3200                            errp);
3201     bdrv_unref(target_bs);
3202 out:
3203     aio_context_release(aio_context);
3204 }
3205 
3206 void qmp_blockdev_mirror(const char *job_id,
3207                          const char *device, const char *target,
3208                          const char *replaces,
3209                          MirrorSyncMode sync,
3210                          bool has_speed, int64_t speed,
3211                          bool has_granularity, uint32_t granularity,
3212                          bool has_buf_size, int64_t buf_size,
3213                          bool has_on_source_error,
3214                          BlockdevOnError on_source_error,
3215                          bool has_on_target_error,
3216                          BlockdevOnError on_target_error,
3217                          const char *filter_node_name,
3218                          bool has_copy_mode, MirrorCopyMode copy_mode,
3219                          bool has_auto_finalize, bool auto_finalize,
3220                          bool has_auto_dismiss, bool auto_dismiss,
3221                          Error **errp)
3222 {
3223     BlockDriverState *bs;
3224     BlockDriverState *target_bs;
3225     AioContext *aio_context;
3226     AioContext *old_context;
3227     BlockMirrorBackingMode backing_mode = MIRROR_LEAVE_BACKING_CHAIN;
3228     bool zero_target;
3229     int ret;
3230 
3231     bs = qmp_get_root_bs(device, errp);
3232     if (!bs) {
3233         return;
3234     }
3235 
3236     target_bs = bdrv_lookup_bs(target, target, errp);
3237     if (!target_bs) {
3238         return;
3239     }
3240 
3241     zero_target = (sync == MIRROR_SYNC_MODE_FULL);
3242 
3243     /* Honor bdrv_try_change_aio_context() context acquisition requirements. */
3244     old_context = bdrv_get_aio_context(target_bs);
3245     aio_context = bdrv_get_aio_context(bs);
3246     aio_context_acquire(old_context);
3247 
3248     ret = bdrv_try_change_aio_context(target_bs, aio_context, NULL, errp);
3249 
3250     aio_context_release(old_context);
3251     aio_context_acquire(aio_context);
3252 
3253     if (ret < 0) {
3254         goto out;
3255     }
3256 
3257     blockdev_mirror_common(job_id, bs, target_bs,
3258                            replaces, sync, backing_mode,
3259                            zero_target, has_speed, speed,
3260                            has_granularity, granularity,
3261                            has_buf_size, buf_size,
3262                            has_on_source_error, on_source_error,
3263                            has_on_target_error, on_target_error,
3264                            true, true, filter_node_name,
3265                            has_copy_mode, copy_mode,
3266                            has_auto_finalize, auto_finalize,
3267                            has_auto_dismiss, auto_dismiss,
3268                            errp);
3269 out:
3270     aio_context_release(aio_context);
3271 }
3272 
3273 /*
3274  * Get a block job using its ID. Called with job_mutex held.
3275  */
3276 static BlockJob *find_block_job_locked(const char *id, Error **errp)
3277 {
3278     BlockJob *job;
3279 
3280     assert(id != NULL);
3281 
3282     job = block_job_get_locked(id);
3283 
3284     if (!job) {
3285         error_set(errp, ERROR_CLASS_DEVICE_NOT_ACTIVE,
3286                   "Block job '%s' not found", id);
3287         return NULL;
3288     }
3289 
3290     return job;
3291 }
3292 
3293 void qmp_block_job_set_speed(const char *device, int64_t speed, Error **errp)
3294 {
3295     BlockJob *job;
3296 
3297     JOB_LOCK_GUARD();
3298     job = find_block_job_locked(device, errp);
3299 
3300     if (!job) {
3301         return;
3302     }
3303 
3304     block_job_set_speed_locked(job, speed, errp);
3305 }
3306 
3307 void qmp_block_job_cancel(const char *device,
3308                           bool has_force, bool force, Error **errp)
3309 {
3310     BlockJob *job;
3311 
3312     JOB_LOCK_GUARD();
3313     job = find_block_job_locked(device, errp);
3314 
3315     if (!job) {
3316         return;
3317     }
3318 
3319     if (!has_force) {
3320         force = false;
3321     }
3322 
3323     if (job_user_paused_locked(&job->job) && !force) {
3324         error_setg(errp, "The block job for device '%s' is currently paused",
3325                    device);
3326         return;
3327     }
3328 
3329     trace_qmp_block_job_cancel(job);
3330     job_user_cancel_locked(&job->job, force, errp);
3331 }
3332 
3333 void qmp_block_job_pause(const char *device, Error **errp)
3334 {
3335     BlockJob *job;
3336 
3337     JOB_LOCK_GUARD();
3338     job = find_block_job_locked(device, errp);
3339 
3340     if (!job) {
3341         return;
3342     }
3343 
3344     trace_qmp_block_job_pause(job);
3345     job_user_pause_locked(&job->job, errp);
3346 }
3347 
3348 void qmp_block_job_resume(const char *device, Error **errp)
3349 {
3350     BlockJob *job;
3351 
3352     JOB_LOCK_GUARD();
3353     job = find_block_job_locked(device, errp);
3354 
3355     if (!job) {
3356         return;
3357     }
3358 
3359     trace_qmp_block_job_resume(job);
3360     job_user_resume_locked(&job->job, errp);
3361 }
3362 
3363 void qmp_block_job_complete(const char *device, Error **errp)
3364 {
3365     BlockJob *job;
3366 
3367     JOB_LOCK_GUARD();
3368     job = find_block_job_locked(device, errp);
3369 
3370     if (!job) {
3371         return;
3372     }
3373 
3374     trace_qmp_block_job_complete(job);
3375     job_complete_locked(&job->job, errp);
3376 }
3377 
3378 void qmp_block_job_finalize(const char *id, Error **errp)
3379 {
3380     BlockJob *job;
3381 
3382     JOB_LOCK_GUARD();
3383     job = find_block_job_locked(id, errp);
3384 
3385     if (!job) {
3386         return;
3387     }
3388 
3389     trace_qmp_block_job_finalize(job);
3390     job_ref_locked(&job->job);
3391     job_finalize_locked(&job->job, errp);
3392 
3393     job_unref_locked(&job->job);
3394 }
3395 
3396 void qmp_block_job_dismiss(const char *id, Error **errp)
3397 {
3398     BlockJob *bjob;
3399     Job *job;
3400 
3401     JOB_LOCK_GUARD();
3402     bjob = find_block_job_locked(id, errp);
3403 
3404     if (!bjob) {
3405         return;
3406     }
3407 
3408     trace_qmp_block_job_dismiss(bjob);
3409     job = &bjob->job;
3410     job_dismiss_locked(&job, errp);
3411 }
3412 
3413 void qmp_change_backing_file(const char *device,
3414                              const char *image_node_name,
3415                              const char *backing_file,
3416                              Error **errp)
3417 {
3418     BlockDriverState *bs = NULL;
3419     AioContext *aio_context;
3420     BlockDriverState *image_bs = NULL;
3421     Error *local_err = NULL;
3422     bool ro;
3423     int ret;
3424 
3425     bs = qmp_get_root_bs(device, errp);
3426     if (!bs) {
3427         return;
3428     }
3429 
3430     aio_context = bdrv_get_aio_context(bs);
3431     aio_context_acquire(aio_context);
3432 
3433     image_bs = bdrv_lookup_bs(NULL, image_node_name, &local_err);
3434     if (local_err) {
3435         error_propagate(errp, local_err);
3436         goto out;
3437     }
3438 
3439     if (!image_bs) {
3440         error_setg(errp, "image file not found");
3441         goto out;
3442     }
3443 
3444     if (bdrv_find_base(image_bs) == image_bs) {
3445         error_setg(errp, "not allowing backing file change on an image "
3446                          "without a backing file");
3447         goto out;
3448     }
3449 
3450     /* even though we are not necessarily operating on bs, we need it to
3451      * determine if block ops are currently prohibited on the chain */
3452     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_CHANGE, errp)) {
3453         goto out;
3454     }
3455 
3456     /* final sanity check */
3457     if (!bdrv_chain_contains(bs, image_bs)) {
3458         error_setg(errp, "'%s' and image file are not in the same chain",
3459                    device);
3460         goto out;
3461     }
3462 
3463     /* if not r/w, reopen to make r/w */
3464     ro = bdrv_is_read_only(image_bs);
3465 
3466     if (ro) {
3467         if (bdrv_reopen_set_read_only(image_bs, false, errp) != 0) {
3468             goto out;
3469         }
3470     }
3471 
3472     ret = bdrv_change_backing_file(image_bs, backing_file,
3473                                    image_bs->drv ? image_bs->drv->format_name : "",
3474                                    false);
3475 
3476     if (ret < 0) {
3477         error_setg_errno(errp, -ret, "Could not change backing file to '%s'",
3478                          backing_file);
3479         /* don't exit here, so we can try to restore open flags if
3480          * appropriate */
3481     }
3482 
3483     if (ro) {
3484         bdrv_reopen_set_read_only(image_bs, true, errp);
3485     }
3486 
3487 out:
3488     aio_context_release(aio_context);
3489 }
3490 
3491 void qmp_blockdev_add(BlockdevOptions *options, Error **errp)
3492 {
3493     BlockDriverState *bs;
3494     QObject *obj;
3495     Visitor *v = qobject_output_visitor_new(&obj);
3496     QDict *qdict;
3497 
3498     visit_type_BlockdevOptions(v, NULL, &options, &error_abort);
3499     visit_complete(v, &obj);
3500     qdict = qobject_to(QDict, obj);
3501 
3502     qdict_flatten(qdict);
3503 
3504     if (!qdict_get_try_str(qdict, "node-name")) {
3505         error_setg(errp, "'node-name' must be specified for the root node");
3506         goto fail;
3507     }
3508 
3509     bs = bds_tree_init(qdict, errp);
3510     if (!bs) {
3511         goto fail;
3512     }
3513 
3514     bdrv_set_monitor_owned(bs);
3515 
3516 fail:
3517     visit_free(v);
3518 }
3519 
3520 void qmp_blockdev_reopen(BlockdevOptionsList *reopen_list, Error **errp)
3521 {
3522     BlockReopenQueue *queue = NULL;
3523 
3524     /* Add each one of the BDS that we want to reopen to the queue */
3525     for (; reopen_list != NULL; reopen_list = reopen_list->next) {
3526         BlockdevOptions *options = reopen_list->value;
3527         BlockDriverState *bs;
3528         AioContext *ctx;
3529         QObject *obj;
3530         Visitor *v;
3531         QDict *qdict;
3532 
3533         /* Check for the selected node name */
3534         if (!options->node_name) {
3535             error_setg(errp, "node-name not specified");
3536             goto fail;
3537         }
3538 
3539         bs = bdrv_find_node(options->node_name);
3540         if (!bs) {
3541             error_setg(errp, "Failed to find node with node-name='%s'",
3542                        options->node_name);
3543             goto fail;
3544         }
3545 
3546         /* Put all options in a QDict and flatten it */
3547         v = qobject_output_visitor_new(&obj);
3548         visit_type_BlockdevOptions(v, NULL, &options, &error_abort);
3549         visit_complete(v, &obj);
3550         visit_free(v);
3551 
3552         qdict = qobject_to(QDict, obj);
3553 
3554         qdict_flatten(qdict);
3555 
3556         ctx = bdrv_get_aio_context(bs);
3557         aio_context_acquire(ctx);
3558 
3559         queue = bdrv_reopen_queue(queue, bs, qdict, false);
3560 
3561         aio_context_release(ctx);
3562     }
3563 
3564     /* Perform the reopen operation */
3565     bdrv_reopen_multiple(queue, errp);
3566     queue = NULL;
3567 
3568 fail:
3569     bdrv_reopen_queue_free(queue);
3570 }
3571 
3572 void qmp_blockdev_del(const char *node_name, Error **errp)
3573 {
3574     AioContext *aio_context;
3575     BlockDriverState *bs;
3576 
3577     GLOBAL_STATE_CODE();
3578 
3579     bs = bdrv_find_node(node_name);
3580     if (!bs) {
3581         error_setg(errp, "Failed to find node with node-name='%s'", node_name);
3582         return;
3583     }
3584     if (bdrv_has_blk(bs)) {
3585         error_setg(errp, "Node %s is in use", node_name);
3586         return;
3587     }
3588     aio_context = bdrv_get_aio_context(bs);
3589     aio_context_acquire(aio_context);
3590 
3591     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_DRIVE_DEL, errp)) {
3592         goto out;
3593     }
3594 
3595     if (!QTAILQ_IN_USE(bs, monitor_list)) {
3596         error_setg(errp, "Node %s is not owned by the monitor",
3597                    bs->node_name);
3598         goto out;
3599     }
3600 
3601     if (bs->refcnt > 1) {
3602         error_setg(errp, "Block device %s is in use",
3603                    bdrv_get_device_or_node_name(bs));
3604         goto out;
3605     }
3606 
3607     QTAILQ_REMOVE(&monitor_bdrv_states, bs, monitor_list);
3608     bdrv_unref(bs);
3609 
3610 out:
3611     aio_context_release(aio_context);
3612 }
3613 
3614 static BdrvChild *bdrv_find_child(BlockDriverState *parent_bs,
3615                                   const char *child_name)
3616 {
3617     BdrvChild *child;
3618 
3619     QLIST_FOREACH(child, &parent_bs->children, next) {
3620         if (strcmp(child->name, child_name) == 0) {
3621             return child;
3622         }
3623     }
3624 
3625     return NULL;
3626 }
3627 
3628 void qmp_x_blockdev_change(const char *parent, const char *child,
3629                            const char *node, Error **errp)
3630 {
3631     BlockDriverState *parent_bs, *new_bs = NULL;
3632     BdrvChild *p_child;
3633 
3634     parent_bs = bdrv_lookup_bs(parent, parent, errp);
3635     if (!parent_bs) {
3636         return;
3637     }
3638 
3639     if (!child == !node) {
3640         if (child) {
3641             error_setg(errp, "The parameters child and node are in conflict");
3642         } else {
3643             error_setg(errp, "Either child or node must be specified");
3644         }
3645         return;
3646     }
3647 
3648     if (child) {
3649         p_child = bdrv_find_child(parent_bs, child);
3650         if (!p_child) {
3651             error_setg(errp, "Node '%s' does not have child '%s'",
3652                        parent, child);
3653             return;
3654         }
3655         bdrv_del_child(parent_bs, p_child, errp);
3656     }
3657 
3658     if (node) {
3659         new_bs = bdrv_find_node(node);
3660         if (!new_bs) {
3661             error_setg(errp, "Node '%s' not found", node);
3662             return;
3663         }
3664         bdrv_add_child(parent_bs, new_bs, errp);
3665     }
3666 }
3667 
3668 BlockJobInfoList *qmp_query_block_jobs(Error **errp)
3669 {
3670     BlockJobInfoList *head = NULL, **tail = &head;
3671     BlockJob *job;
3672 
3673     JOB_LOCK_GUARD();
3674 
3675     for (job = block_job_next_locked(NULL); job;
3676          job = block_job_next_locked(job)) {
3677         BlockJobInfo *value;
3678 
3679         if (block_job_is_internal(job)) {
3680             continue;
3681         }
3682         value = block_job_query_locked(job, errp);
3683         if (!value) {
3684             qapi_free_BlockJobInfoList(head);
3685             return NULL;
3686         }
3687         QAPI_LIST_APPEND(tail, value);
3688     }
3689 
3690     return head;
3691 }
3692 
3693 void qmp_x_blockdev_set_iothread(const char *node_name, StrOrNull *iothread,
3694                                  bool has_force, bool force, Error **errp)
3695 {
3696     AioContext *old_context;
3697     AioContext *new_context;
3698     BlockDriverState *bs;
3699 
3700     bs = bdrv_find_node(node_name);
3701     if (!bs) {
3702         error_setg(errp, "Failed to find node with node-name='%s'", node_name);
3703         return;
3704     }
3705 
3706     /* Protects against accidents. */
3707     if (!(has_force && force) && bdrv_has_blk(bs)) {
3708         error_setg(errp, "Node %s is associated with a BlockBackend and could "
3709                          "be in use (use force=true to override this check)",
3710                          node_name);
3711         return;
3712     }
3713 
3714     if (iothread->type == QTYPE_QSTRING) {
3715         IOThread *obj = iothread_by_id(iothread->u.s);
3716         if (!obj) {
3717             error_setg(errp, "Cannot find iothread %s", iothread->u.s);
3718             return;
3719         }
3720 
3721         new_context = iothread_get_aio_context(obj);
3722     } else {
3723         new_context = qemu_get_aio_context();
3724     }
3725 
3726     old_context = bdrv_get_aio_context(bs);
3727     aio_context_acquire(old_context);
3728 
3729     bdrv_try_change_aio_context(bs, new_context, NULL, errp);
3730 
3731     aio_context_release(old_context);
3732 }
3733 
3734 QemuOptsList qemu_common_drive_opts = {
3735     .name = "drive",
3736     .head = QTAILQ_HEAD_INITIALIZER(qemu_common_drive_opts.head),
3737     .desc = {
3738         {
3739             .name = "snapshot",
3740             .type = QEMU_OPT_BOOL,
3741             .help = "enable/disable snapshot mode",
3742         },{
3743             .name = "aio",
3744             .type = QEMU_OPT_STRING,
3745             .help = "host AIO implementation (threads, native, io_uring)",
3746         },{
3747             .name = BDRV_OPT_CACHE_WB,
3748             .type = QEMU_OPT_BOOL,
3749             .help = "Enable writeback mode",
3750         },{
3751             .name = "format",
3752             .type = QEMU_OPT_STRING,
3753             .help = "disk format (raw, qcow2, ...)",
3754         },{
3755             .name = "rerror",
3756             .type = QEMU_OPT_STRING,
3757             .help = "read error action",
3758         },{
3759             .name = "werror",
3760             .type = QEMU_OPT_STRING,
3761             .help = "write error action",
3762         },{
3763             .name = BDRV_OPT_READ_ONLY,
3764             .type = QEMU_OPT_BOOL,
3765             .help = "open drive file as read-only",
3766         },
3767 
3768         THROTTLE_OPTS,
3769 
3770         {
3771             .name = "throttling.group",
3772             .type = QEMU_OPT_STRING,
3773             .help = "name of the block throttling group",
3774         },{
3775             .name = "copy-on-read",
3776             .type = QEMU_OPT_BOOL,
3777             .help = "copy read data from backing file into image file",
3778         },{
3779             .name = "detect-zeroes",
3780             .type = QEMU_OPT_STRING,
3781             .help = "try to optimize zero writes (off, on, unmap)",
3782         },{
3783             .name = "stats-account-invalid",
3784             .type = QEMU_OPT_BOOL,
3785             .help = "whether to account for invalid I/O operations "
3786                     "in the statistics",
3787         },{
3788             .name = "stats-account-failed",
3789             .type = QEMU_OPT_BOOL,
3790             .help = "whether to account for failed I/O operations "
3791                     "in the statistics",
3792         },
3793         { /* end of list */ }
3794     },
3795 };
3796 
3797 QemuOptsList qemu_drive_opts = {
3798     .name = "drive",
3799     .head = QTAILQ_HEAD_INITIALIZER(qemu_drive_opts.head),
3800     .desc = {
3801         /*
3802          * no elements => accept any params
3803          * validation will happen later
3804          */
3805         { /* end of list */ }
3806     },
3807 };
3808