xref: /openbmc/qemu/blockdev.c (revision f343346b)
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/qdict.h"
39 #include "block/throttle-groups.h"
40 #include "monitor/monitor.h"
41 #include "qemu/error-report.h"
42 #include "qemu/option.h"
43 #include "qemu/qemu-print.h"
44 #include "qemu/config-file.h"
45 #include "qapi/qapi-commands-block.h"
46 #include "qapi/qapi-commands-transaction.h"
47 #include "qapi/qapi-visit-block-core.h"
48 #include "qapi/qmp/qdict.h"
49 #include "qapi/qmp/qnum.h"
50 #include "qapi/qmp/qstring.h"
51 #include "qapi/error.h"
52 #include "qapi/qmp/qerror.h"
53 #include "qapi/qmp/qlist.h"
54 #include "qapi/qobject-output-visitor.h"
55 #include "sysemu/sysemu.h"
56 #include "sysemu/iothread.h"
57 #include "block/block_int.h"
58 #include "block/trace.h"
59 #include "sysemu/arch_init.h"
60 #include "sysemu/qtest.h"
61 #include "sysemu/runstate.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 QTAILQ_HEAD(, BlockDriverState) monitor_bdrv_states =
68     QTAILQ_HEAD_INITIALIZER(monitor_bdrv_states);
69 
70 void bdrv_set_monitor_owned(BlockDriverState *bs)
71 {
72     QTAILQ_INSERT_TAIL(&monitor_bdrv_states, bs, monitor_list);
73 }
74 
75 static const char *const if_name[IF_COUNT] = {
76     [IF_NONE] = "none",
77     [IF_IDE] = "ide",
78     [IF_SCSI] = "scsi",
79     [IF_FLOPPY] = "floppy",
80     [IF_PFLASH] = "pflash",
81     [IF_MTD] = "mtd",
82     [IF_SD] = "sd",
83     [IF_VIRTIO] = "virtio",
84     [IF_XEN] = "xen",
85 };
86 
87 static int if_max_devs[IF_COUNT] = {
88     /*
89      * Do not change these numbers!  They govern how drive option
90      * index maps to unit and bus.  That mapping is ABI.
91      *
92      * All controllers used to implement if=T drives need to support
93      * if_max_devs[T] units, for any T with if_max_devs[T] != 0.
94      * Otherwise, some index values map to "impossible" bus, unit
95      * values.
96      *
97      * For instance, if you change [IF_SCSI] to 255, -drive
98      * if=scsi,index=12 no longer means bus=1,unit=5, but
99      * bus=0,unit=12.  With an lsi53c895a controller (7 units max),
100      * the drive can't be set up.  Regression.
101      */
102     [IF_IDE] = 2,
103     [IF_SCSI] = 7,
104 };
105 
106 /**
107  * Boards may call this to offer board-by-board overrides
108  * of the default, global values.
109  */
110 void override_max_devs(BlockInterfaceType type, int max_devs)
111 {
112     BlockBackend *blk;
113     DriveInfo *dinfo;
114 
115     if (max_devs <= 0) {
116         return;
117     }
118 
119     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
120         dinfo = blk_legacy_dinfo(blk);
121         if (dinfo->type == type) {
122             fprintf(stderr, "Cannot override units-per-bus property of"
123                     " the %s interface, because a drive of that type has"
124                     " already been added.\n", if_name[type]);
125             g_assert_not_reached();
126         }
127     }
128 
129     if_max_devs[type] = max_devs;
130 }
131 
132 /*
133  * We automatically delete the drive when a device using it gets
134  * unplugged.  Questionable feature, but we can't just drop it.
135  * Device models call blockdev_mark_auto_del() to schedule the
136  * automatic deletion, and generic qdev code calls blockdev_auto_del()
137  * when deletion is actually safe.
138  */
139 void blockdev_mark_auto_del(BlockBackend *blk)
140 {
141     DriveInfo *dinfo = blk_legacy_dinfo(blk);
142     BlockJob *job;
143 
144     if (!dinfo) {
145         return;
146     }
147 
148     for (job = block_job_next(NULL); job; job = block_job_next(job)) {
149         if (block_job_has_bdrv(job, blk_bs(blk))) {
150             AioContext *aio_context = job->job.aio_context;
151             aio_context_acquire(aio_context);
152 
153             job_cancel(&job->job, false);
154 
155             aio_context_release(aio_context);
156         }
157     }
158 
159     dinfo->auto_del = 1;
160 }
161 
162 void blockdev_auto_del(BlockBackend *blk)
163 {
164     DriveInfo *dinfo = blk_legacy_dinfo(blk);
165 
166     if (dinfo && dinfo->auto_del) {
167         monitor_remove_blk(blk);
168         blk_unref(blk);
169     }
170 }
171 
172 /**
173  * Returns the current mapping of how many units per bus
174  * a particular interface can support.
175  *
176  *  A positive integer indicates n units per bus.
177  *  0 implies the mapping has not been established.
178  * -1 indicates an invalid BlockInterfaceType was given.
179  */
180 int drive_get_max_devs(BlockInterfaceType type)
181 {
182     if (type >= IF_IDE && type < IF_COUNT) {
183         return if_max_devs[type];
184     }
185 
186     return -1;
187 }
188 
189 static int drive_index_to_bus_id(BlockInterfaceType type, int index)
190 {
191     int max_devs = if_max_devs[type];
192     return max_devs ? index / max_devs : 0;
193 }
194 
195 static int drive_index_to_unit_id(BlockInterfaceType type, int index)
196 {
197     int max_devs = if_max_devs[type];
198     return max_devs ? index % max_devs : index;
199 }
200 
201 QemuOpts *drive_def(const char *optstr)
202 {
203     return qemu_opts_parse_noisily(qemu_find_opts("drive"), optstr, false);
204 }
205 
206 QemuOpts *drive_add(BlockInterfaceType type, int index, const char *file,
207                     const char *optstr)
208 {
209     QemuOpts *opts;
210 
211     opts = drive_def(optstr);
212     if (!opts) {
213         return NULL;
214     }
215     if (type != IF_DEFAULT) {
216         qemu_opt_set(opts, "if", if_name[type], &error_abort);
217     }
218     if (index >= 0) {
219         qemu_opt_set_number(opts, "index", index, &error_abort);
220     }
221     if (file)
222         qemu_opt_set(opts, "file", file, &error_abort);
223     return opts;
224 }
225 
226 DriveInfo *drive_get(BlockInterfaceType type, int bus, int unit)
227 {
228     BlockBackend *blk;
229     DriveInfo *dinfo;
230 
231     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
232         dinfo = blk_legacy_dinfo(blk);
233         if (dinfo && dinfo->type == type
234             && dinfo->bus == bus && dinfo->unit == unit) {
235             return dinfo;
236         }
237     }
238 
239     return NULL;
240 }
241 
242 void drive_mark_claimed_by_board(void)
243 {
244     BlockBackend *blk;
245     DriveInfo *dinfo;
246 
247     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
248         dinfo = blk_legacy_dinfo(blk);
249         if (dinfo && blk_get_attached_dev(blk)) {
250             dinfo->claimed_by_board = true;
251         }
252     }
253 }
254 
255 void drive_check_orphaned(void)
256 {
257     BlockBackend *blk;
258     DriveInfo *dinfo;
259     Location loc;
260     bool orphans = false;
261 
262     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
263         dinfo = blk_legacy_dinfo(blk);
264         if (dinfo->is_default || dinfo->type == IF_NONE) {
265             continue;
266         }
267         if (!blk_get_attached_dev(blk)) {
268             loc_push_none(&loc);
269             qemu_opts_loc_restore(dinfo->opts);
270             error_report("machine type does not support"
271                          " if=%s,bus=%d,unit=%d",
272                          if_name[dinfo->type], dinfo->bus, dinfo->unit);
273             loc_pop(&loc);
274             orphans = true;
275             continue;
276         }
277         if (!dinfo->claimed_by_board && dinfo->type != IF_VIRTIO) {
278             loc_push_none(&loc);
279             qemu_opts_loc_restore(dinfo->opts);
280             warn_report("bogus if=%s is deprecated, use if=none",
281                         if_name[dinfo->type]);
282             loc_pop(&loc);
283         }
284     }
285 
286     if (orphans) {
287         exit(1);
288     }
289 }
290 
291 DriveInfo *drive_get_by_index(BlockInterfaceType type, int index)
292 {
293     return drive_get(type,
294                      drive_index_to_bus_id(type, index),
295                      drive_index_to_unit_id(type, index));
296 }
297 
298 int drive_get_max_bus(BlockInterfaceType type)
299 {
300     int max_bus;
301     BlockBackend *blk;
302     DriveInfo *dinfo;
303 
304     max_bus = -1;
305     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
306         dinfo = blk_legacy_dinfo(blk);
307         if (dinfo && dinfo->type == type && dinfo->bus > max_bus) {
308             max_bus = dinfo->bus;
309         }
310     }
311     return max_bus;
312 }
313 
314 /* Get a block device.  This should only be used for single-drive devices
315    (e.g. SD/Floppy/MTD).  Multi-disk devices (scsi/ide) should use the
316    appropriate bus.  */
317 DriveInfo *drive_get_next(BlockInterfaceType type)
318 {
319     static int next_block_unit[IF_COUNT];
320 
321     return drive_get(type, 0, next_block_unit[type]++);
322 }
323 
324 static void bdrv_format_print(void *opaque, const char *name)
325 {
326     qemu_printf(" %s", name);
327 }
328 
329 typedef struct {
330     QEMUBH *bh;
331     BlockDriverState *bs;
332 } BDRVPutRefBH;
333 
334 static int parse_block_error_action(const char *buf, bool is_read, Error **errp)
335 {
336     if (!strcmp(buf, "ignore")) {
337         return BLOCKDEV_ON_ERROR_IGNORE;
338     } else if (!is_read && !strcmp(buf, "enospc")) {
339         return BLOCKDEV_ON_ERROR_ENOSPC;
340     } else if (!strcmp(buf, "stop")) {
341         return BLOCKDEV_ON_ERROR_STOP;
342     } else if (!strcmp(buf, "report")) {
343         return BLOCKDEV_ON_ERROR_REPORT;
344     } else {
345         error_setg(errp, "'%s' invalid %s error action",
346                    buf, is_read ? "read" : "write");
347         return -1;
348     }
349 }
350 
351 static bool parse_stats_intervals(BlockAcctStats *stats, QList *intervals,
352                                   Error **errp)
353 {
354     const QListEntry *entry;
355     for (entry = qlist_first(intervals); entry; entry = qlist_next(entry)) {
356         switch (qobject_type(entry->value)) {
357 
358         case QTYPE_QSTRING: {
359             unsigned long long length;
360             const char *str = qstring_get_str(qobject_to(QString,
361                                                          entry->value));
362             if (parse_uint_full(str, &length, 10) == 0 &&
363                 length > 0 && length <= UINT_MAX) {
364                 block_acct_add_interval(stats, (unsigned) length);
365             } else {
366                 error_setg(errp, "Invalid interval length: %s", str);
367                 return false;
368             }
369             break;
370         }
371 
372         case QTYPE_QNUM: {
373             int64_t length = qnum_get_int(qobject_to(QNum, entry->value));
374 
375             if (length > 0 && length <= UINT_MAX) {
376                 block_acct_add_interval(stats, (unsigned) length);
377             } else {
378                 error_setg(errp, "Invalid interval length: %" PRId64, length);
379                 return false;
380             }
381             break;
382         }
383 
384         default:
385             error_setg(errp, "The specification of stats-intervals is invalid");
386             return false;
387         }
388     }
389     return true;
390 }
391 
392 typedef enum { MEDIA_DISK, MEDIA_CDROM } DriveMediaType;
393 
394 /* All parameters but @opts are optional and may be set to NULL. */
395 static void extract_common_blockdev_options(QemuOpts *opts, int *bdrv_flags,
396     const char **throttling_group, ThrottleConfig *throttle_cfg,
397     BlockdevDetectZeroesOptions *detect_zeroes, Error **errp)
398 {
399     Error *local_error = NULL;
400     const char *aio;
401 
402     if (bdrv_flags) {
403         if (qemu_opt_get_bool(opts, "copy-on-read", false)) {
404             *bdrv_flags |= BDRV_O_COPY_ON_READ;
405         }
406 
407         if ((aio = qemu_opt_get(opts, "aio")) != NULL) {
408             if (bdrv_parse_aio(aio, bdrv_flags) < 0) {
409                 error_setg(errp, "invalid aio option");
410                 return;
411             }
412         }
413     }
414 
415     /* disk I/O throttling */
416     if (throttling_group) {
417         *throttling_group = qemu_opt_get(opts, "throttling.group");
418     }
419 
420     if (throttle_cfg) {
421         throttle_config_init(throttle_cfg);
422         throttle_cfg->buckets[THROTTLE_BPS_TOTAL].avg =
423             qemu_opt_get_number(opts, "throttling.bps-total", 0);
424         throttle_cfg->buckets[THROTTLE_BPS_READ].avg  =
425             qemu_opt_get_number(opts, "throttling.bps-read", 0);
426         throttle_cfg->buckets[THROTTLE_BPS_WRITE].avg =
427             qemu_opt_get_number(opts, "throttling.bps-write", 0);
428         throttle_cfg->buckets[THROTTLE_OPS_TOTAL].avg =
429             qemu_opt_get_number(opts, "throttling.iops-total", 0);
430         throttle_cfg->buckets[THROTTLE_OPS_READ].avg =
431             qemu_opt_get_number(opts, "throttling.iops-read", 0);
432         throttle_cfg->buckets[THROTTLE_OPS_WRITE].avg =
433             qemu_opt_get_number(opts, "throttling.iops-write", 0);
434 
435         throttle_cfg->buckets[THROTTLE_BPS_TOTAL].max =
436             qemu_opt_get_number(opts, "throttling.bps-total-max", 0);
437         throttle_cfg->buckets[THROTTLE_BPS_READ].max  =
438             qemu_opt_get_number(opts, "throttling.bps-read-max", 0);
439         throttle_cfg->buckets[THROTTLE_BPS_WRITE].max =
440             qemu_opt_get_number(opts, "throttling.bps-write-max", 0);
441         throttle_cfg->buckets[THROTTLE_OPS_TOTAL].max =
442             qemu_opt_get_number(opts, "throttling.iops-total-max", 0);
443         throttle_cfg->buckets[THROTTLE_OPS_READ].max =
444             qemu_opt_get_number(opts, "throttling.iops-read-max", 0);
445         throttle_cfg->buckets[THROTTLE_OPS_WRITE].max =
446             qemu_opt_get_number(opts, "throttling.iops-write-max", 0);
447 
448         throttle_cfg->buckets[THROTTLE_BPS_TOTAL].burst_length =
449             qemu_opt_get_number(opts, "throttling.bps-total-max-length", 1);
450         throttle_cfg->buckets[THROTTLE_BPS_READ].burst_length  =
451             qemu_opt_get_number(opts, "throttling.bps-read-max-length", 1);
452         throttle_cfg->buckets[THROTTLE_BPS_WRITE].burst_length =
453             qemu_opt_get_number(opts, "throttling.bps-write-max-length", 1);
454         throttle_cfg->buckets[THROTTLE_OPS_TOTAL].burst_length =
455             qemu_opt_get_number(opts, "throttling.iops-total-max-length", 1);
456         throttle_cfg->buckets[THROTTLE_OPS_READ].burst_length =
457             qemu_opt_get_number(opts, "throttling.iops-read-max-length", 1);
458         throttle_cfg->buckets[THROTTLE_OPS_WRITE].burst_length =
459             qemu_opt_get_number(opts, "throttling.iops-write-max-length", 1);
460 
461         throttle_cfg->op_size =
462             qemu_opt_get_number(opts, "throttling.iops-size", 0);
463 
464         if (!throttle_is_valid(throttle_cfg, errp)) {
465             return;
466         }
467     }
468 
469     if (detect_zeroes) {
470         *detect_zeroes =
471             qapi_enum_parse(&BlockdevDetectZeroesOptions_lookup,
472                             qemu_opt_get(opts, "detect-zeroes"),
473                             BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF,
474                             &local_error);
475         if (local_error) {
476             error_propagate(errp, local_error);
477             return;
478         }
479     }
480 }
481 
482 /* Takes the ownership of bs_opts */
483 static BlockBackend *blockdev_init(const char *file, QDict *bs_opts,
484                                    Error **errp)
485 {
486     const char *buf;
487     int bdrv_flags = 0;
488     int on_read_error, on_write_error;
489     bool account_invalid, account_failed;
490     bool writethrough, read_only;
491     BlockBackend *blk;
492     BlockDriverState *bs;
493     ThrottleConfig cfg;
494     int snapshot = 0;
495     Error *error = NULL;
496     QemuOpts *opts;
497     QDict *interval_dict = NULL;
498     QList *interval_list = NULL;
499     const char *id;
500     BlockdevDetectZeroesOptions detect_zeroes =
501         BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF;
502     const char *throttling_group = NULL;
503 
504     /* Check common options by copying from bs_opts to opts, all other options
505      * stay in bs_opts for processing by bdrv_open(). */
506     id = qdict_get_try_str(bs_opts, "id");
507     opts = qemu_opts_create(&qemu_common_drive_opts, id, 1, errp);
508     if (!opts) {
509         goto err_no_opts;
510     }
511 
512     if (!qemu_opts_absorb_qdict(opts, bs_opts, errp)) {
513         goto early_err;
514     }
515 
516     if (id) {
517         qdict_del(bs_opts, "id");
518     }
519 
520     /* extract parameters */
521     snapshot = qemu_opt_get_bool(opts, "snapshot", 0);
522 
523     account_invalid = qemu_opt_get_bool(opts, "stats-account-invalid", true);
524     account_failed = qemu_opt_get_bool(opts, "stats-account-failed", true);
525 
526     writethrough = !qemu_opt_get_bool(opts, BDRV_OPT_CACHE_WB, true);
527 
528     id = qemu_opts_id(opts);
529 
530     qdict_extract_subqdict(bs_opts, &interval_dict, "stats-intervals.");
531     qdict_array_split(interval_dict, &interval_list);
532 
533     if (qdict_size(interval_dict) != 0) {
534         error_setg(errp, "Invalid option stats-intervals.%s",
535                    qdict_first(interval_dict)->key);
536         goto early_err;
537     }
538 
539     extract_common_blockdev_options(opts, &bdrv_flags, &throttling_group, &cfg,
540                                     &detect_zeroes, &error);
541     if (error) {
542         error_propagate(errp, error);
543         goto early_err;
544     }
545 
546     if ((buf = qemu_opt_get(opts, "format")) != NULL) {
547         if (is_help_option(buf)) {
548             qemu_printf("Supported formats:");
549             bdrv_iterate_format(bdrv_format_print, NULL, false);
550             qemu_printf("\nSupported formats (read-only):");
551             bdrv_iterate_format(bdrv_format_print, NULL, true);
552             qemu_printf("\n");
553             goto early_err;
554         }
555 
556         if (qdict_haskey(bs_opts, "driver")) {
557             error_setg(errp, "Cannot specify both 'driver' and 'format'");
558             goto early_err;
559         }
560         qdict_put_str(bs_opts, "driver", buf);
561     }
562 
563     on_write_error = BLOCKDEV_ON_ERROR_ENOSPC;
564     if ((buf = qemu_opt_get(opts, "werror")) != NULL) {
565         on_write_error = parse_block_error_action(buf, 0, &error);
566         if (error) {
567             error_propagate(errp, error);
568             goto early_err;
569         }
570     }
571 
572     on_read_error = BLOCKDEV_ON_ERROR_REPORT;
573     if ((buf = qemu_opt_get(opts, "rerror")) != NULL) {
574         on_read_error = parse_block_error_action(buf, 1, &error);
575         if (error) {
576             error_propagate(errp, error);
577             goto early_err;
578         }
579     }
580 
581     if (snapshot) {
582         bdrv_flags |= BDRV_O_SNAPSHOT;
583     }
584 
585     read_only = qemu_opt_get_bool(opts, BDRV_OPT_READ_ONLY, false);
586 
587     /* init */
588     if ((!file || !*file) && !qdict_size(bs_opts)) {
589         BlockBackendRootState *blk_rs;
590 
591         blk = blk_new(qemu_get_aio_context(), 0, BLK_PERM_ALL);
592         blk_rs = blk_get_root_state(blk);
593         blk_rs->open_flags    = bdrv_flags;
594         blk_rs->read_only     = read_only;
595         blk_rs->detect_zeroes = detect_zeroes;
596 
597         qobject_unref(bs_opts);
598     } else {
599         if (file && !*file) {
600             file = NULL;
601         }
602 
603         /* bdrv_open() defaults to the values in bdrv_flags (for compatibility
604          * with other callers) rather than what we want as the real defaults.
605          * Apply the defaults here instead. */
606         qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_DIRECT, "off");
607         qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_NO_FLUSH, "off");
608         qdict_set_default_str(bs_opts, BDRV_OPT_READ_ONLY,
609                               read_only ? "on" : "off");
610         qdict_set_default_str(bs_opts, BDRV_OPT_AUTO_READ_ONLY, "on");
611         assert((bdrv_flags & BDRV_O_CACHE_MASK) == 0);
612 
613         if (runstate_check(RUN_STATE_INMIGRATE)) {
614             bdrv_flags |= BDRV_O_INACTIVE;
615         }
616 
617         blk = blk_new_open(file, NULL, bs_opts, bdrv_flags, errp);
618         if (!blk) {
619             goto err_no_bs_opts;
620         }
621         bs = blk_bs(blk);
622 
623         bs->detect_zeroes = detect_zeroes;
624 
625         block_acct_setup(blk_get_stats(blk), account_invalid, account_failed);
626 
627         if (!parse_stats_intervals(blk_get_stats(blk), interval_list, errp)) {
628             blk_unref(blk);
629             blk = NULL;
630             goto err_no_bs_opts;
631         }
632     }
633 
634     /* disk I/O throttling */
635     if (throttle_enabled(&cfg)) {
636         if (!throttling_group) {
637             throttling_group = id;
638         }
639         blk_io_limits_enable(blk, throttling_group);
640         blk_set_io_limits(blk, &cfg);
641     }
642 
643     blk_set_enable_write_cache(blk, !writethrough);
644     blk_set_on_error(blk, on_read_error, on_write_error);
645 
646     if (!monitor_add_blk(blk, id, errp)) {
647         blk_unref(blk);
648         blk = NULL;
649         goto err_no_bs_opts;
650     }
651 
652 err_no_bs_opts:
653     qemu_opts_del(opts);
654     qobject_unref(interval_dict);
655     qobject_unref(interval_list);
656     return blk;
657 
658 early_err:
659     qemu_opts_del(opts);
660     qobject_unref(interval_dict);
661     qobject_unref(interval_list);
662 err_no_opts:
663     qobject_unref(bs_opts);
664     return NULL;
665 }
666 
667 /* Takes the ownership of bs_opts */
668 BlockDriverState *bds_tree_init(QDict *bs_opts, Error **errp)
669 {
670     int bdrv_flags = 0;
671 
672     /* bdrv_open() defaults to the values in bdrv_flags (for compatibility
673      * with other callers) rather than what we want as the real defaults.
674      * Apply the defaults here instead. */
675     qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_DIRECT, "off");
676     qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_NO_FLUSH, "off");
677     qdict_set_default_str(bs_opts, BDRV_OPT_READ_ONLY, "off");
678 
679     if (runstate_check(RUN_STATE_INMIGRATE)) {
680         bdrv_flags |= BDRV_O_INACTIVE;
681     }
682 
683     return bdrv_open(NULL, NULL, bs_opts, bdrv_flags, errp);
684 }
685 
686 void blockdev_close_all_bdrv_states(void)
687 {
688     BlockDriverState *bs, *next_bs;
689 
690     QTAILQ_FOREACH_SAFE(bs, &monitor_bdrv_states, monitor_list, next_bs) {
691         AioContext *ctx = bdrv_get_aio_context(bs);
692 
693         aio_context_acquire(ctx);
694         bdrv_unref(bs);
695         aio_context_release(ctx);
696     }
697 }
698 
699 /* Iterates over the list of monitor-owned BlockDriverStates */
700 BlockDriverState *bdrv_next_monitor_owned(BlockDriverState *bs)
701 {
702     return bs ? QTAILQ_NEXT(bs, monitor_list)
703               : QTAILQ_FIRST(&monitor_bdrv_states);
704 }
705 
706 static bool qemu_opt_rename(QemuOpts *opts, const char *from, const char *to,
707                             Error **errp)
708 {
709     const char *value;
710 
711     value = qemu_opt_get(opts, from);
712     if (value) {
713         if (qemu_opt_find(opts, to)) {
714             error_setg(errp, "'%s' and its alias '%s' can't be used at the "
715                        "same time", to, from);
716             return false;
717         }
718     }
719 
720     /* rename all items in opts */
721     while ((value = qemu_opt_get(opts, from))) {
722         qemu_opt_set(opts, to, value, &error_abort);
723         qemu_opt_unset(opts, from);
724     }
725     return true;
726 }
727 
728 QemuOptsList qemu_legacy_drive_opts = {
729     .name = "drive",
730     .head = QTAILQ_HEAD_INITIALIZER(qemu_legacy_drive_opts.head),
731     .desc = {
732         {
733             .name = "bus",
734             .type = QEMU_OPT_NUMBER,
735             .help = "bus number",
736         },{
737             .name = "unit",
738             .type = QEMU_OPT_NUMBER,
739             .help = "unit number (i.e. lun for scsi)",
740         },{
741             .name = "index",
742             .type = QEMU_OPT_NUMBER,
743             .help = "index number",
744         },{
745             .name = "media",
746             .type = QEMU_OPT_STRING,
747             .help = "media type (disk, cdrom)",
748         },{
749             .name = "if",
750             .type = QEMU_OPT_STRING,
751             .help = "interface (ide, scsi, sd, mtd, floppy, pflash, virtio)",
752         },{
753             .name = "file",
754             .type = QEMU_OPT_STRING,
755             .help = "file name",
756         },
757 
758         /* Options that are passed on, but have special semantics with -drive */
759         {
760             .name = BDRV_OPT_READ_ONLY,
761             .type = QEMU_OPT_BOOL,
762             .help = "open drive file as read-only",
763         },{
764             .name = "rerror",
765             .type = QEMU_OPT_STRING,
766             .help = "read error action",
767         },{
768             .name = "werror",
769             .type = QEMU_OPT_STRING,
770             .help = "write error action",
771         },{
772             .name = "copy-on-read",
773             .type = QEMU_OPT_BOOL,
774             .help = "copy read data from backing file into image file",
775         },
776 
777         { /* end of list */ }
778     },
779 };
780 
781 DriveInfo *drive_new(QemuOpts *all_opts, BlockInterfaceType block_default_type,
782                      Error **errp)
783 {
784     const char *value;
785     BlockBackend *blk;
786     DriveInfo *dinfo = NULL;
787     QDict *bs_opts;
788     QemuOpts *legacy_opts;
789     DriveMediaType media = MEDIA_DISK;
790     BlockInterfaceType type;
791     int max_devs, bus_id, unit_id, index;
792     const char *werror, *rerror;
793     bool read_only = false;
794     bool copy_on_read;
795     const char *filename;
796     int i;
797 
798     /* Change legacy command line options into QMP ones */
799     static const struct {
800         const char *from;
801         const char *to;
802     } opt_renames[] = {
803         { "iops",           "throttling.iops-total" },
804         { "iops_rd",        "throttling.iops-read" },
805         { "iops_wr",        "throttling.iops-write" },
806 
807         { "bps",            "throttling.bps-total" },
808         { "bps_rd",         "throttling.bps-read" },
809         { "bps_wr",         "throttling.bps-write" },
810 
811         { "iops_max",       "throttling.iops-total-max" },
812         { "iops_rd_max",    "throttling.iops-read-max" },
813         { "iops_wr_max",    "throttling.iops-write-max" },
814 
815         { "bps_max",        "throttling.bps-total-max" },
816         { "bps_rd_max",     "throttling.bps-read-max" },
817         { "bps_wr_max",     "throttling.bps-write-max" },
818 
819         { "iops_size",      "throttling.iops-size" },
820 
821         { "group",          "throttling.group" },
822 
823         { "readonly",       BDRV_OPT_READ_ONLY },
824     };
825 
826     for (i = 0; i < ARRAY_SIZE(opt_renames); i++) {
827         if (!qemu_opt_rename(all_opts, opt_renames[i].from,
828                              opt_renames[i].to, errp)) {
829             return NULL;
830         }
831     }
832 
833     value = qemu_opt_get(all_opts, "cache");
834     if (value) {
835         int flags = 0;
836         bool writethrough;
837 
838         if (bdrv_parse_cache_mode(value, &flags, &writethrough) != 0) {
839             error_setg(errp, "invalid cache option");
840             return NULL;
841         }
842 
843         /* Specific options take precedence */
844         if (!qemu_opt_get(all_opts, BDRV_OPT_CACHE_WB)) {
845             qemu_opt_set_bool(all_opts, BDRV_OPT_CACHE_WB,
846                               !writethrough, &error_abort);
847         }
848         if (!qemu_opt_get(all_opts, BDRV_OPT_CACHE_DIRECT)) {
849             qemu_opt_set_bool(all_opts, BDRV_OPT_CACHE_DIRECT,
850                               !!(flags & BDRV_O_NOCACHE), &error_abort);
851         }
852         if (!qemu_opt_get(all_opts, BDRV_OPT_CACHE_NO_FLUSH)) {
853             qemu_opt_set_bool(all_opts, BDRV_OPT_CACHE_NO_FLUSH,
854                               !!(flags & BDRV_O_NO_FLUSH), &error_abort);
855         }
856         qemu_opt_unset(all_opts, "cache");
857     }
858 
859     /* Get a QDict for processing the options */
860     bs_opts = qdict_new();
861     qemu_opts_to_qdict(all_opts, bs_opts);
862 
863     legacy_opts = qemu_opts_create(&qemu_legacy_drive_opts, NULL, 0,
864                                    &error_abort);
865     if (!qemu_opts_absorb_qdict(legacy_opts, bs_opts, errp)) {
866         goto fail;
867     }
868 
869     /* Media type */
870     value = qemu_opt_get(legacy_opts, "media");
871     if (value) {
872         if (!strcmp(value, "disk")) {
873             media = MEDIA_DISK;
874         } else if (!strcmp(value, "cdrom")) {
875             media = MEDIA_CDROM;
876             read_only = true;
877         } else {
878             error_setg(errp, "'%s' invalid media", value);
879             goto fail;
880         }
881     }
882 
883     /* copy-on-read is disabled with a warning for read-only devices */
884     read_only |= qemu_opt_get_bool(legacy_opts, BDRV_OPT_READ_ONLY, false);
885     copy_on_read = qemu_opt_get_bool(legacy_opts, "copy-on-read", false);
886 
887     if (read_only && copy_on_read) {
888         warn_report("disabling copy-on-read on read-only drive");
889         copy_on_read = false;
890     }
891 
892     qdict_put_str(bs_opts, BDRV_OPT_READ_ONLY, read_only ? "on" : "off");
893     qdict_put_str(bs_opts, "copy-on-read", copy_on_read ? "on" : "off");
894 
895     /* Controller type */
896     value = qemu_opt_get(legacy_opts, "if");
897     if (value) {
898         for (type = 0;
899              type < IF_COUNT && strcmp(value, if_name[type]);
900              type++) {
901         }
902         if (type == IF_COUNT) {
903             error_setg(errp, "unsupported bus type '%s'", value);
904             goto fail;
905         }
906     } else {
907         type = block_default_type;
908     }
909 
910     /* Device address specified by bus/unit or index.
911      * If none was specified, try to find the first free one. */
912     bus_id  = qemu_opt_get_number(legacy_opts, "bus", 0);
913     unit_id = qemu_opt_get_number(legacy_opts, "unit", -1);
914     index   = qemu_opt_get_number(legacy_opts, "index", -1);
915 
916     max_devs = if_max_devs[type];
917 
918     if (index != -1) {
919         if (bus_id != 0 || unit_id != -1) {
920             error_setg(errp, "index cannot be used with bus and unit");
921             goto fail;
922         }
923         bus_id = drive_index_to_bus_id(type, index);
924         unit_id = drive_index_to_unit_id(type, index);
925     }
926 
927     if (unit_id == -1) {
928        unit_id = 0;
929        while (drive_get(type, bus_id, unit_id) != NULL) {
930            unit_id++;
931            if (max_devs && unit_id >= max_devs) {
932                unit_id -= max_devs;
933                bus_id++;
934            }
935        }
936     }
937 
938     if (max_devs && unit_id >= max_devs) {
939         error_setg(errp, "unit %d too big (max is %d)", unit_id, max_devs - 1);
940         goto fail;
941     }
942 
943     if (drive_get(type, bus_id, unit_id) != NULL) {
944         error_setg(errp, "drive with bus=%d, unit=%d (index=%d) exists",
945                    bus_id, unit_id, index);
946         goto fail;
947     }
948 
949     /* no id supplied -> create one */
950     if (qemu_opts_id(all_opts) == NULL) {
951         char *new_id;
952         const char *mediastr = "";
953         if (type == IF_IDE || type == IF_SCSI) {
954             mediastr = (media == MEDIA_CDROM) ? "-cd" : "-hd";
955         }
956         if (max_devs) {
957             new_id = g_strdup_printf("%s%i%s%i", if_name[type], bus_id,
958                                      mediastr, unit_id);
959         } else {
960             new_id = g_strdup_printf("%s%s%i", if_name[type],
961                                      mediastr, unit_id);
962         }
963         qdict_put_str(bs_opts, "id", new_id);
964         g_free(new_id);
965     }
966 
967     /* Add virtio block device */
968     if (type == IF_VIRTIO) {
969         QemuOpts *devopts;
970         devopts = qemu_opts_create(qemu_find_opts("device"), NULL, 0,
971                                    &error_abort);
972         if (arch_type == QEMU_ARCH_S390X) {
973             qemu_opt_set(devopts, "driver", "virtio-blk-ccw", &error_abort);
974         } else {
975             qemu_opt_set(devopts, "driver", "virtio-blk-pci", &error_abort);
976         }
977         qemu_opt_set(devopts, "drive", qdict_get_str(bs_opts, "id"),
978                      &error_abort);
979     }
980 
981     filename = qemu_opt_get(legacy_opts, "file");
982 
983     /* Check werror/rerror compatibility with if=... */
984     werror = qemu_opt_get(legacy_opts, "werror");
985     if (werror != NULL) {
986         if (type != IF_IDE && type != IF_SCSI && type != IF_VIRTIO &&
987             type != IF_NONE) {
988             error_setg(errp, "werror is not supported by this bus type");
989             goto fail;
990         }
991         qdict_put_str(bs_opts, "werror", werror);
992     }
993 
994     rerror = qemu_opt_get(legacy_opts, "rerror");
995     if (rerror != NULL) {
996         if (type != IF_IDE && type != IF_VIRTIO && type != IF_SCSI &&
997             type != IF_NONE) {
998             error_setg(errp, "rerror is not supported by this bus type");
999             goto fail;
1000         }
1001         qdict_put_str(bs_opts, "rerror", rerror);
1002     }
1003 
1004     /* Actual block device init: Functionality shared with blockdev-add */
1005     blk = blockdev_init(filename, bs_opts, errp);
1006     bs_opts = NULL;
1007     if (!blk) {
1008         goto fail;
1009     }
1010 
1011     /* Create legacy DriveInfo */
1012     dinfo = g_malloc0(sizeof(*dinfo));
1013     dinfo->opts = all_opts;
1014 
1015     dinfo->type = type;
1016     dinfo->bus = bus_id;
1017     dinfo->unit = unit_id;
1018 
1019     blk_set_legacy_dinfo(blk, dinfo);
1020 
1021     switch(type) {
1022     case IF_IDE:
1023     case IF_SCSI:
1024     case IF_XEN:
1025     case IF_NONE:
1026         dinfo->media_cd = media == MEDIA_CDROM;
1027         break;
1028     default:
1029         break;
1030     }
1031 
1032 fail:
1033     qemu_opts_del(legacy_opts);
1034     qobject_unref(bs_opts);
1035     return dinfo;
1036 }
1037 
1038 static BlockDriverState *qmp_get_root_bs(const char *name, Error **errp)
1039 {
1040     BlockDriverState *bs;
1041 
1042     bs = bdrv_lookup_bs(name, name, errp);
1043     if (bs == NULL) {
1044         return NULL;
1045     }
1046 
1047     if (!bdrv_is_root_node(bs)) {
1048         error_setg(errp, "Need a root block node");
1049         return NULL;
1050     }
1051 
1052     if (!bdrv_is_inserted(bs)) {
1053         error_setg(errp, "Device has no medium");
1054         return NULL;
1055     }
1056 
1057     return bs;
1058 }
1059 
1060 static void blockdev_do_action(TransactionAction *action, Error **errp)
1061 {
1062     TransactionActionList list;
1063 
1064     list.value = action;
1065     list.next = NULL;
1066     qmp_transaction(&list, false, NULL, errp);
1067 }
1068 
1069 void qmp_blockdev_snapshot_sync(bool has_device, const char *device,
1070                                 bool has_node_name, const char *node_name,
1071                                 const char *snapshot_file,
1072                                 bool has_snapshot_node_name,
1073                                 const char *snapshot_node_name,
1074                                 bool has_format, const char *format,
1075                                 bool has_mode, NewImageMode mode, Error **errp)
1076 {
1077     BlockdevSnapshotSync snapshot = {
1078         .has_device = has_device,
1079         .device = (char *) device,
1080         .has_node_name = has_node_name,
1081         .node_name = (char *) node_name,
1082         .snapshot_file = (char *) snapshot_file,
1083         .has_snapshot_node_name = has_snapshot_node_name,
1084         .snapshot_node_name = (char *) snapshot_node_name,
1085         .has_format = has_format,
1086         .format = (char *) format,
1087         .has_mode = has_mode,
1088         .mode = mode,
1089     };
1090     TransactionAction action = {
1091         .type = TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC,
1092         .u.blockdev_snapshot_sync.data = &snapshot,
1093     };
1094     blockdev_do_action(&action, errp);
1095 }
1096 
1097 void qmp_blockdev_snapshot(const char *node, const char *overlay,
1098                            Error **errp)
1099 {
1100     BlockdevSnapshot snapshot_data = {
1101         .node = (char *) node,
1102         .overlay = (char *) overlay
1103     };
1104     TransactionAction action = {
1105         .type = TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT,
1106         .u.blockdev_snapshot.data = &snapshot_data,
1107     };
1108     blockdev_do_action(&action, errp);
1109 }
1110 
1111 void qmp_blockdev_snapshot_internal_sync(const char *device,
1112                                          const char *name,
1113                                          Error **errp)
1114 {
1115     BlockdevSnapshotInternal snapshot = {
1116         .device = (char *) device,
1117         .name = (char *) name
1118     };
1119     TransactionAction action = {
1120         .type = TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC,
1121         .u.blockdev_snapshot_internal_sync.data = &snapshot,
1122     };
1123     blockdev_do_action(&action, errp);
1124 }
1125 
1126 SnapshotInfo *qmp_blockdev_snapshot_delete_internal_sync(const char *device,
1127                                                          bool has_id,
1128                                                          const char *id,
1129                                                          bool has_name,
1130                                                          const char *name,
1131                                                          Error **errp)
1132 {
1133     BlockDriverState *bs;
1134     AioContext *aio_context;
1135     QEMUSnapshotInfo sn;
1136     Error *local_err = NULL;
1137     SnapshotInfo *info = NULL;
1138     int ret;
1139 
1140     bs = qmp_get_root_bs(device, errp);
1141     if (!bs) {
1142         return NULL;
1143     }
1144     aio_context = bdrv_get_aio_context(bs);
1145     aio_context_acquire(aio_context);
1146 
1147     if (!has_id) {
1148         id = NULL;
1149     }
1150 
1151     if (!has_name) {
1152         name = NULL;
1153     }
1154 
1155     if (!id && !name) {
1156         error_setg(errp, "Name or id must be provided");
1157         goto out_aio_context;
1158     }
1159 
1160     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_INTERNAL_SNAPSHOT_DELETE, errp)) {
1161         goto out_aio_context;
1162     }
1163 
1164     ret = bdrv_snapshot_find_by_id_and_name(bs, id, name, &sn, &local_err);
1165     if (local_err) {
1166         error_propagate(errp, local_err);
1167         goto out_aio_context;
1168     }
1169     if (!ret) {
1170         error_setg(errp,
1171                    "Snapshot with id '%s' and name '%s' does not exist on "
1172                    "device '%s'",
1173                    STR_OR_NULL(id), STR_OR_NULL(name), device);
1174         goto out_aio_context;
1175     }
1176 
1177     bdrv_snapshot_delete(bs, id, name, &local_err);
1178     if (local_err) {
1179         error_propagate(errp, local_err);
1180         goto out_aio_context;
1181     }
1182 
1183     aio_context_release(aio_context);
1184 
1185     info = g_new0(SnapshotInfo, 1);
1186     info->id = g_strdup(sn.id_str);
1187     info->name = g_strdup(sn.name);
1188     info->date_nsec = sn.date_nsec;
1189     info->date_sec = sn.date_sec;
1190     info->vm_state_size = sn.vm_state_size;
1191     info->vm_clock_nsec = sn.vm_clock_nsec % 1000000000;
1192     info->vm_clock_sec = sn.vm_clock_nsec / 1000000000;
1193 
1194     return info;
1195 
1196 out_aio_context:
1197     aio_context_release(aio_context);
1198     return NULL;
1199 }
1200 
1201 /* New and old BlockDriverState structs for atomic group operations */
1202 
1203 typedef struct BlkActionState BlkActionState;
1204 
1205 /**
1206  * BlkActionOps:
1207  * Table of operations that define an Action.
1208  *
1209  * @instance_size: Size of state struct, in bytes.
1210  * @prepare: Prepare the work, must NOT be NULL.
1211  * @commit: Commit the changes, can be NULL.
1212  * @abort: Abort the changes on fail, can be NULL.
1213  * @clean: Clean up resources after all transaction actions have called
1214  *         commit() or abort(). Can be NULL.
1215  *
1216  * Only prepare() may fail. In a single transaction, only one of commit() or
1217  * abort() will be called. clean() will always be called if it is present.
1218  */
1219 typedef struct BlkActionOps {
1220     size_t instance_size;
1221     void (*prepare)(BlkActionState *common, Error **errp);
1222     void (*commit)(BlkActionState *common);
1223     void (*abort)(BlkActionState *common);
1224     void (*clean)(BlkActionState *common);
1225 } BlkActionOps;
1226 
1227 /**
1228  * BlkActionState:
1229  * Describes one Action's state within a Transaction.
1230  *
1231  * @action: QAPI-defined enum identifying which Action to perform.
1232  * @ops: Table of ActionOps this Action can perform.
1233  * @block_job_txn: Transaction which this action belongs to.
1234  * @entry: List membership for all Actions in this Transaction.
1235  *
1236  * This structure must be arranged as first member in a subclassed type,
1237  * assuming that the compiler will also arrange it to the same offsets as the
1238  * base class.
1239  */
1240 struct BlkActionState {
1241     TransactionAction *action;
1242     const BlkActionOps *ops;
1243     JobTxn *block_job_txn;
1244     TransactionProperties *txn_props;
1245     QTAILQ_ENTRY(BlkActionState) entry;
1246 };
1247 
1248 /* internal snapshot private data */
1249 typedef struct InternalSnapshotState {
1250     BlkActionState common;
1251     BlockDriverState *bs;
1252     QEMUSnapshotInfo sn;
1253     bool created;
1254 } InternalSnapshotState;
1255 
1256 
1257 static int action_check_completion_mode(BlkActionState *s, Error **errp)
1258 {
1259     if (s->txn_props->completion_mode != ACTION_COMPLETION_MODE_INDIVIDUAL) {
1260         error_setg(errp,
1261                    "Action '%s' does not support Transaction property "
1262                    "completion-mode = %s",
1263                    TransactionActionKind_str(s->action->type),
1264                    ActionCompletionMode_str(s->txn_props->completion_mode));
1265         return -1;
1266     }
1267     return 0;
1268 }
1269 
1270 static void internal_snapshot_prepare(BlkActionState *common,
1271                                       Error **errp)
1272 {
1273     Error *local_err = NULL;
1274     const char *device;
1275     const char *name;
1276     BlockDriverState *bs;
1277     QEMUSnapshotInfo old_sn, *sn;
1278     bool ret;
1279     qemu_timeval tv;
1280     BlockdevSnapshotInternal *internal;
1281     InternalSnapshotState *state;
1282     AioContext *aio_context;
1283     int ret1;
1284 
1285     g_assert(common->action->type ==
1286              TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC);
1287     internal = common->action->u.blockdev_snapshot_internal_sync.data;
1288     state = DO_UPCAST(InternalSnapshotState, common, common);
1289 
1290     /* 1. parse input */
1291     device = internal->device;
1292     name = internal->name;
1293 
1294     /* 2. check for validation */
1295     if (action_check_completion_mode(common, errp) < 0) {
1296         return;
1297     }
1298 
1299     bs = qmp_get_root_bs(device, errp);
1300     if (!bs) {
1301         return;
1302     }
1303 
1304     aio_context = bdrv_get_aio_context(bs);
1305     aio_context_acquire(aio_context);
1306 
1307     state->bs = bs;
1308 
1309     /* Paired with .clean() */
1310     bdrv_drained_begin(bs);
1311 
1312     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_INTERNAL_SNAPSHOT, errp)) {
1313         goto out;
1314     }
1315 
1316     if (bdrv_is_read_only(bs)) {
1317         error_setg(errp, "Device '%s' is read only", device);
1318         goto out;
1319     }
1320 
1321     if (!bdrv_can_snapshot(bs)) {
1322         error_setg(errp, "Block format '%s' used by device '%s' "
1323                    "does not support internal snapshots",
1324                    bs->drv->format_name, device);
1325         goto out;
1326     }
1327 
1328     if (!strlen(name)) {
1329         error_setg(errp, "Name is empty");
1330         goto out;
1331     }
1332 
1333     /* check whether a snapshot with name exist */
1334     ret = bdrv_snapshot_find_by_id_and_name(bs, NULL, name, &old_sn,
1335                                             &local_err);
1336     if (local_err) {
1337         error_propagate(errp, local_err);
1338         goto out;
1339     } else if (ret) {
1340         error_setg(errp,
1341                    "Snapshot with name '%s' already exists on device '%s'",
1342                    name, device);
1343         goto out;
1344     }
1345 
1346     /* 3. take the snapshot */
1347     sn = &state->sn;
1348     pstrcpy(sn->name, sizeof(sn->name), name);
1349     qemu_gettimeofday(&tv);
1350     sn->date_sec = tv.tv_sec;
1351     sn->date_nsec = tv.tv_usec * 1000;
1352     sn->vm_clock_nsec = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL);
1353 
1354     ret1 = bdrv_snapshot_create(bs, sn);
1355     if (ret1 < 0) {
1356         error_setg_errno(errp, -ret1,
1357                          "Failed to create snapshot '%s' on device '%s'",
1358                          name, device);
1359         goto out;
1360     }
1361 
1362     /* 4. succeed, mark a snapshot is created */
1363     state->created = true;
1364 
1365 out:
1366     aio_context_release(aio_context);
1367 }
1368 
1369 static void internal_snapshot_abort(BlkActionState *common)
1370 {
1371     InternalSnapshotState *state =
1372                              DO_UPCAST(InternalSnapshotState, common, common);
1373     BlockDriverState *bs = state->bs;
1374     QEMUSnapshotInfo *sn = &state->sn;
1375     AioContext *aio_context;
1376     Error *local_error = NULL;
1377 
1378     if (!state->created) {
1379         return;
1380     }
1381 
1382     aio_context = bdrv_get_aio_context(state->bs);
1383     aio_context_acquire(aio_context);
1384 
1385     if (bdrv_snapshot_delete(bs, sn->id_str, sn->name, &local_error) < 0) {
1386         error_reportf_err(local_error,
1387                           "Failed to delete snapshot with id '%s' and "
1388                           "name '%s' on device '%s' in abort: ",
1389                           sn->id_str, sn->name,
1390                           bdrv_get_device_name(bs));
1391     }
1392 
1393     aio_context_release(aio_context);
1394 }
1395 
1396 static void internal_snapshot_clean(BlkActionState *common)
1397 {
1398     InternalSnapshotState *state = DO_UPCAST(InternalSnapshotState,
1399                                              common, common);
1400     AioContext *aio_context;
1401 
1402     if (!state->bs) {
1403         return;
1404     }
1405 
1406     aio_context = bdrv_get_aio_context(state->bs);
1407     aio_context_acquire(aio_context);
1408 
1409     bdrv_drained_end(state->bs);
1410 
1411     aio_context_release(aio_context);
1412 }
1413 
1414 /* external snapshot private data */
1415 typedef struct ExternalSnapshotState {
1416     BlkActionState common;
1417     BlockDriverState *old_bs;
1418     BlockDriverState *new_bs;
1419     bool overlay_appended;
1420 } ExternalSnapshotState;
1421 
1422 static void external_snapshot_prepare(BlkActionState *common,
1423                                       Error **errp)
1424 {
1425     int flags = 0;
1426     QDict *options = NULL;
1427     Error *local_err = NULL;
1428     /* Device and node name of the image to generate the snapshot from */
1429     const char *device;
1430     const char *node_name;
1431     /* Reference to the new image (for 'blockdev-snapshot') */
1432     const char *snapshot_ref;
1433     /* File name of the new image (for 'blockdev-snapshot-sync') */
1434     const char *new_image_file;
1435     ExternalSnapshotState *state =
1436                              DO_UPCAST(ExternalSnapshotState, common, common);
1437     TransactionAction *action = common->action;
1438     AioContext *aio_context;
1439     uint64_t perm, shared;
1440 
1441     /* 'blockdev-snapshot' and 'blockdev-snapshot-sync' have similar
1442      * purpose but a different set of parameters */
1443     switch (action->type) {
1444     case TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT:
1445         {
1446             BlockdevSnapshot *s = action->u.blockdev_snapshot.data;
1447             device = s->node;
1448             node_name = s->node;
1449             new_image_file = NULL;
1450             snapshot_ref = s->overlay;
1451         }
1452         break;
1453     case TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC:
1454         {
1455             BlockdevSnapshotSync *s = action->u.blockdev_snapshot_sync.data;
1456             device = s->has_device ? s->device : NULL;
1457             node_name = s->has_node_name ? s->node_name : NULL;
1458             new_image_file = s->snapshot_file;
1459             snapshot_ref = NULL;
1460         }
1461         break;
1462     default:
1463         g_assert_not_reached();
1464     }
1465 
1466     /* start processing */
1467     if (action_check_completion_mode(common, errp) < 0) {
1468         return;
1469     }
1470 
1471     state->old_bs = bdrv_lookup_bs(device, node_name, errp);
1472     if (!state->old_bs) {
1473         return;
1474     }
1475 
1476     aio_context = bdrv_get_aio_context(state->old_bs);
1477     aio_context_acquire(aio_context);
1478 
1479     /* Paired with .clean() */
1480     bdrv_drained_begin(state->old_bs);
1481 
1482     if (!bdrv_is_inserted(state->old_bs)) {
1483         error_setg(errp, QERR_DEVICE_HAS_NO_MEDIUM, device);
1484         goto out;
1485     }
1486 
1487     if (bdrv_op_is_blocked(state->old_bs,
1488                            BLOCK_OP_TYPE_EXTERNAL_SNAPSHOT, errp)) {
1489         goto out;
1490     }
1491 
1492     if (!bdrv_is_read_only(state->old_bs)) {
1493         if (bdrv_flush(state->old_bs)) {
1494             error_setg(errp, QERR_IO_ERROR);
1495             goto out;
1496         }
1497     }
1498 
1499     if (action->type == TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC) {
1500         BlockdevSnapshotSync *s = action->u.blockdev_snapshot_sync.data;
1501         const char *format = s->has_format ? s->format : "qcow2";
1502         enum NewImageMode mode;
1503         const char *snapshot_node_name =
1504             s->has_snapshot_node_name ? s->snapshot_node_name : NULL;
1505 
1506         if (node_name && !snapshot_node_name) {
1507             error_setg(errp, "New overlay node name missing");
1508             goto out;
1509         }
1510 
1511         if (snapshot_node_name &&
1512             bdrv_lookup_bs(snapshot_node_name, snapshot_node_name, NULL)) {
1513             error_setg(errp, "New overlay node name already in use");
1514             goto out;
1515         }
1516 
1517         flags = state->old_bs->open_flags;
1518         flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_COPY_ON_READ);
1519         flags |= BDRV_O_NO_BACKING;
1520 
1521         /* create new image w/backing file */
1522         mode = s->has_mode ? s->mode : NEW_IMAGE_MODE_ABSOLUTE_PATHS;
1523         if (mode != NEW_IMAGE_MODE_EXISTING) {
1524             int64_t size = bdrv_getlength(state->old_bs);
1525             if (size < 0) {
1526                 error_setg_errno(errp, -size, "bdrv_getlength failed");
1527                 goto out;
1528             }
1529             bdrv_refresh_filename(state->old_bs);
1530             bdrv_img_create(new_image_file, format,
1531                             state->old_bs->filename,
1532                             state->old_bs->drv->format_name,
1533                             NULL, size, flags, false, &local_err);
1534             if (local_err) {
1535                 error_propagate(errp, local_err);
1536                 goto out;
1537             }
1538         }
1539 
1540         options = qdict_new();
1541         if (snapshot_node_name) {
1542             qdict_put_str(options, "node-name", snapshot_node_name);
1543         }
1544         qdict_put_str(options, "driver", format);
1545     }
1546 
1547     state->new_bs = bdrv_open(new_image_file, snapshot_ref, options, flags,
1548                               errp);
1549     /* We will manually add the backing_hd field to the bs later */
1550     if (!state->new_bs) {
1551         goto out;
1552     }
1553 
1554     /*
1555      * Allow attaching a backing file to an overlay that's already in use only
1556      * if the parents don't assume that they are already seeing a valid image.
1557      * (Specifically, allow it as a mirror target, which is write-only access.)
1558      */
1559     bdrv_get_cumulative_perm(state->new_bs, &perm, &shared);
1560     if (perm & BLK_PERM_CONSISTENT_READ) {
1561         error_setg(errp, "The overlay is already in use");
1562         goto out;
1563     }
1564 
1565     if (state->new_bs->drv->is_filter) {
1566         error_setg(errp, "Filters cannot be used as overlays");
1567         goto out;
1568     }
1569 
1570     if (bdrv_cow_child(state->new_bs)) {
1571         error_setg(errp, "The overlay already has a backing image");
1572         goto out;
1573     }
1574 
1575     if (!state->new_bs->drv->supports_backing) {
1576         error_setg(errp, "The overlay does not support backing images");
1577         goto out;
1578     }
1579 
1580     /* This removes our old bs and adds the new bs. This is an operation that
1581      * can fail, so we need to do it in .prepare; undoing it for abort is
1582      * always possible. */
1583     bdrv_ref(state->new_bs);
1584     bdrv_append(state->new_bs, state->old_bs, &local_err);
1585     if (local_err) {
1586         error_propagate(errp, local_err);
1587         goto out;
1588     }
1589     state->overlay_appended = true;
1590 
1591 out:
1592     aio_context_release(aio_context);
1593 }
1594 
1595 static void external_snapshot_commit(BlkActionState *common)
1596 {
1597     ExternalSnapshotState *state =
1598                              DO_UPCAST(ExternalSnapshotState, common, common);
1599     AioContext *aio_context;
1600 
1601     aio_context = bdrv_get_aio_context(state->old_bs);
1602     aio_context_acquire(aio_context);
1603 
1604     /* We don't need (or want) to use the transactional
1605      * bdrv_reopen_multiple() across all the entries at once, because we
1606      * don't want to abort all of them if one of them fails the reopen */
1607     if (!qatomic_read(&state->old_bs->copy_on_read)) {
1608         bdrv_reopen_set_read_only(state->old_bs, true, NULL);
1609     }
1610 
1611     aio_context_release(aio_context);
1612 }
1613 
1614 static void external_snapshot_abort(BlkActionState *common)
1615 {
1616     ExternalSnapshotState *state =
1617                              DO_UPCAST(ExternalSnapshotState, common, common);
1618     if (state->new_bs) {
1619         if (state->overlay_appended) {
1620             AioContext *aio_context;
1621             AioContext *tmp_context;
1622             int ret;
1623 
1624             aio_context = bdrv_get_aio_context(state->old_bs);
1625             aio_context_acquire(aio_context);
1626 
1627             bdrv_ref(state->old_bs);   /* we can't let bdrv_set_backind_hd()
1628                                           close state->old_bs; we need it */
1629             bdrv_set_backing_hd(state->new_bs, NULL, &error_abort);
1630 
1631             /*
1632              * The call to bdrv_set_backing_hd() above returns state->old_bs to
1633              * the main AioContext. As we're still going to be using it, return
1634              * it to the AioContext it was before.
1635              */
1636             tmp_context = bdrv_get_aio_context(state->old_bs);
1637             if (aio_context != tmp_context) {
1638                 aio_context_release(aio_context);
1639                 aio_context_acquire(tmp_context);
1640 
1641                 ret = bdrv_try_set_aio_context(state->old_bs,
1642                                                aio_context, NULL);
1643                 assert(ret == 0);
1644 
1645                 aio_context_release(tmp_context);
1646                 aio_context_acquire(aio_context);
1647             }
1648 
1649             bdrv_replace_node(state->new_bs, state->old_bs, &error_abort);
1650             bdrv_unref(state->old_bs); /* bdrv_replace_node() ref'ed old_bs */
1651 
1652             aio_context_release(aio_context);
1653         }
1654     }
1655 }
1656 
1657 static void external_snapshot_clean(BlkActionState *common)
1658 {
1659     ExternalSnapshotState *state =
1660                              DO_UPCAST(ExternalSnapshotState, common, common);
1661     AioContext *aio_context;
1662 
1663     if (!state->old_bs) {
1664         return;
1665     }
1666 
1667     aio_context = bdrv_get_aio_context(state->old_bs);
1668     aio_context_acquire(aio_context);
1669 
1670     bdrv_drained_end(state->old_bs);
1671     bdrv_unref(state->new_bs);
1672 
1673     aio_context_release(aio_context);
1674 }
1675 
1676 typedef struct DriveBackupState {
1677     BlkActionState common;
1678     BlockDriverState *bs;
1679     BlockJob *job;
1680 } DriveBackupState;
1681 
1682 static BlockJob *do_backup_common(BackupCommon *backup,
1683                                   BlockDriverState *bs,
1684                                   BlockDriverState *target_bs,
1685                                   AioContext *aio_context,
1686                                   JobTxn *txn, Error **errp);
1687 
1688 static void drive_backup_prepare(BlkActionState *common, Error **errp)
1689 {
1690     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1691     DriveBackup *backup;
1692     BlockDriverState *bs;
1693     BlockDriverState *target_bs;
1694     BlockDriverState *source = NULL;
1695     AioContext *aio_context;
1696     AioContext *old_context;
1697     QDict *options;
1698     Error *local_err = NULL;
1699     int flags;
1700     int64_t size;
1701     bool set_backing_hd = false;
1702     int ret;
1703 
1704     assert(common->action->type == TRANSACTION_ACTION_KIND_DRIVE_BACKUP);
1705     backup = common->action->u.drive_backup.data;
1706 
1707     if (!backup->has_mode) {
1708         backup->mode = NEW_IMAGE_MODE_ABSOLUTE_PATHS;
1709     }
1710 
1711     bs = bdrv_lookup_bs(backup->device, backup->device, errp);
1712     if (!bs) {
1713         return;
1714     }
1715 
1716     if (!bs->drv) {
1717         error_setg(errp, "Device has no medium");
1718         return;
1719     }
1720 
1721     aio_context = bdrv_get_aio_context(bs);
1722     aio_context_acquire(aio_context);
1723 
1724     /* Paired with .clean() */
1725     bdrv_drained_begin(bs);
1726 
1727     if (!backup->has_format) {
1728         backup->format = backup->mode == NEW_IMAGE_MODE_EXISTING ?
1729                          NULL : (char *) bs->drv->format_name;
1730     }
1731 
1732     /* Early check to avoid creating target */
1733     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_BACKUP_SOURCE, errp)) {
1734         goto out;
1735     }
1736 
1737     flags = bs->open_flags | BDRV_O_RDWR;
1738 
1739     /*
1740      * See if we have a backing HD we can use to create our new image
1741      * on top of.
1742      */
1743     if (backup->sync == MIRROR_SYNC_MODE_TOP) {
1744         /*
1745          * Backup will not replace the source by the target, so none
1746          * of the filters skipped here will be removed (in contrast to
1747          * mirror).  Therefore, we can skip all of them when looking
1748          * for the first COW relationship.
1749          */
1750         source = bdrv_cow_bs(bdrv_skip_filters(bs));
1751         if (!source) {
1752             backup->sync = MIRROR_SYNC_MODE_FULL;
1753         }
1754     }
1755     if (backup->sync == MIRROR_SYNC_MODE_NONE) {
1756         source = bs;
1757         flags |= BDRV_O_NO_BACKING;
1758         set_backing_hd = true;
1759     }
1760 
1761     size = bdrv_getlength(bs);
1762     if (size < 0) {
1763         error_setg_errno(errp, -size, "bdrv_getlength failed");
1764         goto out;
1765     }
1766 
1767     if (backup->mode != NEW_IMAGE_MODE_EXISTING) {
1768         assert(backup->format);
1769         if (source) {
1770             /* Implicit filters should not appear in the filename */
1771             BlockDriverState *explicit_backing =
1772                 bdrv_skip_implicit_filters(source);
1773 
1774             bdrv_refresh_filename(explicit_backing);
1775             bdrv_img_create(backup->target, backup->format,
1776                             explicit_backing->filename,
1777                             explicit_backing->drv->format_name, NULL,
1778                             size, flags, false, &local_err);
1779         } else {
1780             bdrv_img_create(backup->target, backup->format, NULL, NULL, NULL,
1781                             size, flags, false, &local_err);
1782         }
1783     }
1784 
1785     if (local_err) {
1786         error_propagate(errp, local_err);
1787         goto out;
1788     }
1789 
1790     options = qdict_new();
1791     qdict_put_str(options, "discard", "unmap");
1792     qdict_put_str(options, "detect-zeroes", "unmap");
1793     if (backup->format) {
1794         qdict_put_str(options, "driver", backup->format);
1795     }
1796 
1797     target_bs = bdrv_open(backup->target, NULL, options, flags, errp);
1798     if (!target_bs) {
1799         goto out;
1800     }
1801 
1802     /* Honor bdrv_try_set_aio_context() context acquisition requirements. */
1803     old_context = bdrv_get_aio_context(target_bs);
1804     aio_context_release(aio_context);
1805     aio_context_acquire(old_context);
1806 
1807     ret = bdrv_try_set_aio_context(target_bs, aio_context, errp);
1808     if (ret < 0) {
1809         bdrv_unref(target_bs);
1810         aio_context_release(old_context);
1811         return;
1812     }
1813 
1814     aio_context_release(old_context);
1815     aio_context_acquire(aio_context);
1816 
1817     if (set_backing_hd) {
1818         bdrv_set_backing_hd(target_bs, source, &local_err);
1819         if (local_err) {
1820             goto unref;
1821         }
1822     }
1823 
1824     state->bs = bs;
1825 
1826     state->job = do_backup_common(qapi_DriveBackup_base(backup),
1827                                   bs, target_bs, aio_context,
1828                                   common->block_job_txn, errp);
1829 
1830 unref:
1831     bdrv_unref(target_bs);
1832 out:
1833     aio_context_release(aio_context);
1834 }
1835 
1836 static void drive_backup_commit(BlkActionState *common)
1837 {
1838     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1839     AioContext *aio_context;
1840 
1841     aio_context = bdrv_get_aio_context(state->bs);
1842     aio_context_acquire(aio_context);
1843 
1844     assert(state->job);
1845     job_start(&state->job->job);
1846 
1847     aio_context_release(aio_context);
1848 }
1849 
1850 static void drive_backup_abort(BlkActionState *common)
1851 {
1852     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1853 
1854     if (state->job) {
1855         AioContext *aio_context;
1856 
1857         aio_context = bdrv_get_aio_context(state->bs);
1858         aio_context_acquire(aio_context);
1859 
1860         job_cancel_sync(&state->job->job);
1861 
1862         aio_context_release(aio_context);
1863     }
1864 }
1865 
1866 static void drive_backup_clean(BlkActionState *common)
1867 {
1868     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1869     AioContext *aio_context;
1870 
1871     if (!state->bs) {
1872         return;
1873     }
1874 
1875     aio_context = bdrv_get_aio_context(state->bs);
1876     aio_context_acquire(aio_context);
1877 
1878     bdrv_drained_end(state->bs);
1879 
1880     aio_context_release(aio_context);
1881 }
1882 
1883 typedef struct BlockdevBackupState {
1884     BlkActionState common;
1885     BlockDriverState *bs;
1886     BlockJob *job;
1887 } BlockdevBackupState;
1888 
1889 static void blockdev_backup_prepare(BlkActionState *common, Error **errp)
1890 {
1891     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1892     BlockdevBackup *backup;
1893     BlockDriverState *bs;
1894     BlockDriverState *target_bs;
1895     AioContext *aio_context;
1896     AioContext *old_context;
1897     int ret;
1898 
1899     assert(common->action->type == TRANSACTION_ACTION_KIND_BLOCKDEV_BACKUP);
1900     backup = common->action->u.blockdev_backup.data;
1901 
1902     bs = bdrv_lookup_bs(backup->device, backup->device, errp);
1903     if (!bs) {
1904         return;
1905     }
1906 
1907     target_bs = bdrv_lookup_bs(backup->target, backup->target, errp);
1908     if (!target_bs) {
1909         return;
1910     }
1911 
1912     /* Honor bdrv_try_set_aio_context() context acquisition requirements. */
1913     aio_context = bdrv_get_aio_context(bs);
1914     old_context = bdrv_get_aio_context(target_bs);
1915     aio_context_acquire(old_context);
1916 
1917     ret = bdrv_try_set_aio_context(target_bs, aio_context, errp);
1918     if (ret < 0) {
1919         aio_context_release(old_context);
1920         return;
1921     }
1922 
1923     aio_context_release(old_context);
1924     aio_context_acquire(aio_context);
1925     state->bs = bs;
1926 
1927     /* Paired with .clean() */
1928     bdrv_drained_begin(state->bs);
1929 
1930     state->job = do_backup_common(qapi_BlockdevBackup_base(backup),
1931                                   bs, target_bs, aio_context,
1932                                   common->block_job_txn, errp);
1933 
1934     aio_context_release(aio_context);
1935 }
1936 
1937 static void blockdev_backup_commit(BlkActionState *common)
1938 {
1939     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1940     AioContext *aio_context;
1941 
1942     aio_context = bdrv_get_aio_context(state->bs);
1943     aio_context_acquire(aio_context);
1944 
1945     assert(state->job);
1946     job_start(&state->job->job);
1947 
1948     aio_context_release(aio_context);
1949 }
1950 
1951 static void blockdev_backup_abort(BlkActionState *common)
1952 {
1953     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1954 
1955     if (state->job) {
1956         AioContext *aio_context;
1957 
1958         aio_context = bdrv_get_aio_context(state->bs);
1959         aio_context_acquire(aio_context);
1960 
1961         job_cancel_sync(&state->job->job);
1962 
1963         aio_context_release(aio_context);
1964     }
1965 }
1966 
1967 static void blockdev_backup_clean(BlkActionState *common)
1968 {
1969     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1970     AioContext *aio_context;
1971 
1972     if (!state->bs) {
1973         return;
1974     }
1975 
1976     aio_context = bdrv_get_aio_context(state->bs);
1977     aio_context_acquire(aio_context);
1978 
1979     bdrv_drained_end(state->bs);
1980 
1981     aio_context_release(aio_context);
1982 }
1983 
1984 typedef struct BlockDirtyBitmapState {
1985     BlkActionState common;
1986     BdrvDirtyBitmap *bitmap;
1987     BlockDriverState *bs;
1988     HBitmap *backup;
1989     bool prepared;
1990     bool was_enabled;
1991 } BlockDirtyBitmapState;
1992 
1993 static void block_dirty_bitmap_add_prepare(BlkActionState *common,
1994                                            Error **errp)
1995 {
1996     Error *local_err = NULL;
1997     BlockDirtyBitmapAdd *action;
1998     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
1999                                              common, common);
2000 
2001     if (action_check_completion_mode(common, errp) < 0) {
2002         return;
2003     }
2004 
2005     action = common->action->u.block_dirty_bitmap_add.data;
2006     /* AIO context taken and released within qmp_block_dirty_bitmap_add */
2007     qmp_block_dirty_bitmap_add(action->node, action->name,
2008                                action->has_granularity, action->granularity,
2009                                action->has_persistent, action->persistent,
2010                                action->has_disabled, action->disabled,
2011                                &local_err);
2012 
2013     if (!local_err) {
2014         state->prepared = true;
2015     } else {
2016         error_propagate(errp, local_err);
2017     }
2018 }
2019 
2020 static void block_dirty_bitmap_add_abort(BlkActionState *common)
2021 {
2022     BlockDirtyBitmapAdd *action;
2023     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2024                                              common, common);
2025 
2026     action = common->action->u.block_dirty_bitmap_add.data;
2027     /* Should not be able to fail: IF the bitmap was added via .prepare(),
2028      * then the node reference and bitmap name must have been valid.
2029      */
2030     if (state->prepared) {
2031         qmp_block_dirty_bitmap_remove(action->node, action->name, &error_abort);
2032     }
2033 }
2034 
2035 static void block_dirty_bitmap_clear_prepare(BlkActionState *common,
2036                                              Error **errp)
2037 {
2038     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2039                                              common, common);
2040     BlockDirtyBitmap *action;
2041 
2042     if (action_check_completion_mode(common, errp) < 0) {
2043         return;
2044     }
2045 
2046     action = common->action->u.block_dirty_bitmap_clear.data;
2047     state->bitmap = block_dirty_bitmap_lookup(action->node,
2048                                               action->name,
2049                                               &state->bs,
2050                                               errp);
2051     if (!state->bitmap) {
2052         return;
2053     }
2054 
2055     if (bdrv_dirty_bitmap_check(state->bitmap, BDRV_BITMAP_DEFAULT, errp)) {
2056         return;
2057     }
2058 
2059     bdrv_clear_dirty_bitmap(state->bitmap, &state->backup);
2060 }
2061 
2062 static void block_dirty_bitmap_restore(BlkActionState *common)
2063 {
2064     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2065                                              common, common);
2066 
2067     if (state->backup) {
2068         bdrv_restore_dirty_bitmap(state->bitmap, state->backup);
2069     }
2070 }
2071 
2072 static void block_dirty_bitmap_free_backup(BlkActionState *common)
2073 {
2074     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2075                                              common, common);
2076 
2077     hbitmap_free(state->backup);
2078 }
2079 
2080 static void block_dirty_bitmap_enable_prepare(BlkActionState *common,
2081                                               Error **errp)
2082 {
2083     BlockDirtyBitmap *action;
2084     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2085                                              common, common);
2086 
2087     if (action_check_completion_mode(common, errp) < 0) {
2088         return;
2089     }
2090 
2091     action = common->action->u.block_dirty_bitmap_enable.data;
2092     state->bitmap = block_dirty_bitmap_lookup(action->node,
2093                                               action->name,
2094                                               NULL,
2095                                               errp);
2096     if (!state->bitmap) {
2097         return;
2098     }
2099 
2100     if (bdrv_dirty_bitmap_check(state->bitmap, BDRV_BITMAP_ALLOW_RO, errp)) {
2101         return;
2102     }
2103 
2104     state->was_enabled = bdrv_dirty_bitmap_enabled(state->bitmap);
2105     bdrv_enable_dirty_bitmap(state->bitmap);
2106 }
2107 
2108 static void block_dirty_bitmap_enable_abort(BlkActionState *common)
2109 {
2110     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2111                                              common, common);
2112 
2113     if (!state->was_enabled) {
2114         bdrv_disable_dirty_bitmap(state->bitmap);
2115     }
2116 }
2117 
2118 static void block_dirty_bitmap_disable_prepare(BlkActionState *common,
2119                                                Error **errp)
2120 {
2121     BlockDirtyBitmap *action;
2122     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2123                                              common, common);
2124 
2125     if (action_check_completion_mode(common, errp) < 0) {
2126         return;
2127     }
2128 
2129     action = common->action->u.block_dirty_bitmap_disable.data;
2130     state->bitmap = block_dirty_bitmap_lookup(action->node,
2131                                               action->name,
2132                                               NULL,
2133                                               errp);
2134     if (!state->bitmap) {
2135         return;
2136     }
2137 
2138     if (bdrv_dirty_bitmap_check(state->bitmap, BDRV_BITMAP_ALLOW_RO, errp)) {
2139         return;
2140     }
2141 
2142     state->was_enabled = bdrv_dirty_bitmap_enabled(state->bitmap);
2143     bdrv_disable_dirty_bitmap(state->bitmap);
2144 }
2145 
2146 static void block_dirty_bitmap_disable_abort(BlkActionState *common)
2147 {
2148     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2149                                              common, common);
2150 
2151     if (state->was_enabled) {
2152         bdrv_enable_dirty_bitmap(state->bitmap);
2153     }
2154 }
2155 
2156 static void block_dirty_bitmap_merge_prepare(BlkActionState *common,
2157                                              Error **errp)
2158 {
2159     BlockDirtyBitmapMerge *action;
2160     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2161                                              common, common);
2162 
2163     if (action_check_completion_mode(common, errp) < 0) {
2164         return;
2165     }
2166 
2167     action = common->action->u.block_dirty_bitmap_merge.data;
2168 
2169     state->bitmap = block_dirty_bitmap_merge(action->node, action->target,
2170                                              action->bitmaps, &state->backup,
2171                                              errp);
2172 }
2173 
2174 static void block_dirty_bitmap_remove_prepare(BlkActionState *common,
2175                                               Error **errp)
2176 {
2177     BlockDirtyBitmap *action;
2178     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2179                                              common, common);
2180 
2181     if (action_check_completion_mode(common, errp) < 0) {
2182         return;
2183     }
2184 
2185     action = common->action->u.block_dirty_bitmap_remove.data;
2186 
2187     state->bitmap = block_dirty_bitmap_remove(action->node, action->name,
2188                                               false, &state->bs, errp);
2189     if (state->bitmap) {
2190         bdrv_dirty_bitmap_skip_store(state->bitmap, true);
2191         bdrv_dirty_bitmap_set_busy(state->bitmap, true);
2192     }
2193 }
2194 
2195 static void block_dirty_bitmap_remove_abort(BlkActionState *common)
2196 {
2197     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2198                                              common, common);
2199 
2200     if (state->bitmap) {
2201         bdrv_dirty_bitmap_skip_store(state->bitmap, false);
2202         bdrv_dirty_bitmap_set_busy(state->bitmap, false);
2203     }
2204 }
2205 
2206 static void block_dirty_bitmap_remove_commit(BlkActionState *common)
2207 {
2208     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2209                                              common, common);
2210 
2211     bdrv_dirty_bitmap_set_busy(state->bitmap, false);
2212     bdrv_release_dirty_bitmap(state->bitmap);
2213 }
2214 
2215 static void abort_prepare(BlkActionState *common, Error **errp)
2216 {
2217     error_setg(errp, "Transaction aborted using Abort action");
2218 }
2219 
2220 static void abort_commit(BlkActionState *common)
2221 {
2222     g_assert_not_reached(); /* this action never succeeds */
2223 }
2224 
2225 static const BlkActionOps actions[] = {
2226     [TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT] = {
2227         .instance_size = sizeof(ExternalSnapshotState),
2228         .prepare  = external_snapshot_prepare,
2229         .commit   = external_snapshot_commit,
2230         .abort = external_snapshot_abort,
2231         .clean = external_snapshot_clean,
2232     },
2233     [TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC] = {
2234         .instance_size = sizeof(ExternalSnapshotState),
2235         .prepare  = external_snapshot_prepare,
2236         .commit   = external_snapshot_commit,
2237         .abort = external_snapshot_abort,
2238         .clean = external_snapshot_clean,
2239     },
2240     [TRANSACTION_ACTION_KIND_DRIVE_BACKUP] = {
2241         .instance_size = sizeof(DriveBackupState),
2242         .prepare = drive_backup_prepare,
2243         .commit = drive_backup_commit,
2244         .abort = drive_backup_abort,
2245         .clean = drive_backup_clean,
2246     },
2247     [TRANSACTION_ACTION_KIND_BLOCKDEV_BACKUP] = {
2248         .instance_size = sizeof(BlockdevBackupState),
2249         .prepare = blockdev_backup_prepare,
2250         .commit = blockdev_backup_commit,
2251         .abort = blockdev_backup_abort,
2252         .clean = blockdev_backup_clean,
2253     },
2254     [TRANSACTION_ACTION_KIND_ABORT] = {
2255         .instance_size = sizeof(BlkActionState),
2256         .prepare = abort_prepare,
2257         .commit = abort_commit,
2258     },
2259     [TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC] = {
2260         .instance_size = sizeof(InternalSnapshotState),
2261         .prepare  = internal_snapshot_prepare,
2262         .abort = internal_snapshot_abort,
2263         .clean = internal_snapshot_clean,
2264     },
2265     [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_ADD] = {
2266         .instance_size = sizeof(BlockDirtyBitmapState),
2267         .prepare = block_dirty_bitmap_add_prepare,
2268         .abort = block_dirty_bitmap_add_abort,
2269     },
2270     [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_CLEAR] = {
2271         .instance_size = sizeof(BlockDirtyBitmapState),
2272         .prepare = block_dirty_bitmap_clear_prepare,
2273         .commit = block_dirty_bitmap_free_backup,
2274         .abort = block_dirty_bitmap_restore,
2275     },
2276     [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_ENABLE] = {
2277         .instance_size = sizeof(BlockDirtyBitmapState),
2278         .prepare = block_dirty_bitmap_enable_prepare,
2279         .abort = block_dirty_bitmap_enable_abort,
2280     },
2281     [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_DISABLE] = {
2282         .instance_size = sizeof(BlockDirtyBitmapState),
2283         .prepare = block_dirty_bitmap_disable_prepare,
2284         .abort = block_dirty_bitmap_disable_abort,
2285     },
2286     [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_MERGE] = {
2287         .instance_size = sizeof(BlockDirtyBitmapState),
2288         .prepare = block_dirty_bitmap_merge_prepare,
2289         .commit = block_dirty_bitmap_free_backup,
2290         .abort = block_dirty_bitmap_restore,
2291     },
2292     [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_REMOVE] = {
2293         .instance_size = sizeof(BlockDirtyBitmapState),
2294         .prepare = block_dirty_bitmap_remove_prepare,
2295         .commit = block_dirty_bitmap_remove_commit,
2296         .abort = block_dirty_bitmap_remove_abort,
2297     },
2298     /* Where are transactions for MIRROR, COMMIT and STREAM?
2299      * Although these blockjobs use transaction callbacks like the backup job,
2300      * these jobs do not necessarily adhere to transaction semantics.
2301      * These jobs may not fully undo all of their actions on abort, nor do they
2302      * necessarily work in transactions with more than one job in them.
2303      */
2304 };
2305 
2306 /**
2307  * Allocate a TransactionProperties structure if necessary, and fill
2308  * that structure with desired defaults if they are unset.
2309  */
2310 static TransactionProperties *get_transaction_properties(
2311     TransactionProperties *props)
2312 {
2313     if (!props) {
2314         props = g_new0(TransactionProperties, 1);
2315     }
2316 
2317     if (!props->has_completion_mode) {
2318         props->has_completion_mode = true;
2319         props->completion_mode = ACTION_COMPLETION_MODE_INDIVIDUAL;
2320     }
2321 
2322     return props;
2323 }
2324 
2325 /*
2326  * 'Atomic' group operations.  The operations are performed as a set, and if
2327  * any fail then we roll back all operations in the group.
2328  */
2329 void qmp_transaction(TransactionActionList *dev_list,
2330                      bool has_props,
2331                      struct TransactionProperties *props,
2332                      Error **errp)
2333 {
2334     TransactionActionList *dev_entry = dev_list;
2335     JobTxn *block_job_txn = NULL;
2336     BlkActionState *state, *next;
2337     Error *local_err = NULL;
2338 
2339     QTAILQ_HEAD(, BlkActionState) snap_bdrv_states;
2340     QTAILQ_INIT(&snap_bdrv_states);
2341 
2342     /* Does this transaction get canceled as a group on failure?
2343      * If not, we don't really need to make a JobTxn.
2344      */
2345     props = get_transaction_properties(props);
2346     if (props->completion_mode != ACTION_COMPLETION_MODE_INDIVIDUAL) {
2347         block_job_txn = job_txn_new();
2348     }
2349 
2350     /* drain all i/o before any operations */
2351     bdrv_drain_all();
2352 
2353     /* We don't do anything in this loop that commits us to the operations */
2354     while (NULL != dev_entry) {
2355         TransactionAction *dev_info = NULL;
2356         const BlkActionOps *ops;
2357 
2358         dev_info = dev_entry->value;
2359         dev_entry = dev_entry->next;
2360 
2361         assert(dev_info->type < ARRAY_SIZE(actions));
2362 
2363         ops = &actions[dev_info->type];
2364         assert(ops->instance_size > 0);
2365 
2366         state = g_malloc0(ops->instance_size);
2367         state->ops = ops;
2368         state->action = dev_info;
2369         state->block_job_txn = block_job_txn;
2370         state->txn_props = props;
2371         QTAILQ_INSERT_TAIL(&snap_bdrv_states, state, entry);
2372 
2373         state->ops->prepare(state, &local_err);
2374         if (local_err) {
2375             error_propagate(errp, local_err);
2376             goto delete_and_fail;
2377         }
2378     }
2379 
2380     QTAILQ_FOREACH(state, &snap_bdrv_states, entry) {
2381         if (state->ops->commit) {
2382             state->ops->commit(state);
2383         }
2384     }
2385 
2386     /* success */
2387     goto exit;
2388 
2389 delete_and_fail:
2390     /* failure, and it is all-or-none; roll back all operations */
2391     QTAILQ_FOREACH_REVERSE(state, &snap_bdrv_states, entry) {
2392         if (state->ops->abort) {
2393             state->ops->abort(state);
2394         }
2395     }
2396 exit:
2397     QTAILQ_FOREACH_SAFE(state, &snap_bdrv_states, entry, next) {
2398         if (state->ops->clean) {
2399             state->ops->clean(state);
2400         }
2401         g_free(state);
2402     }
2403     if (!has_props) {
2404         qapi_free_TransactionProperties(props);
2405     }
2406     job_txn_unref(block_job_txn);
2407 }
2408 
2409 void qmp_block_passwd(bool has_device, const char *device,
2410                       bool has_node_name, const char *node_name,
2411                       const char *password, Error **errp)
2412 {
2413     error_setg(errp,
2414                "Setting block passwords directly is no longer supported");
2415 }
2416 
2417 BlockDirtyBitmapSha256 *qmp_x_debug_block_dirty_bitmap_sha256(const char *node,
2418                                                               const char *name,
2419                                                               Error **errp)
2420 {
2421     BdrvDirtyBitmap *bitmap;
2422     BlockDriverState *bs;
2423     BlockDirtyBitmapSha256 *ret = NULL;
2424     char *sha256;
2425 
2426     bitmap = block_dirty_bitmap_lookup(node, name, &bs, errp);
2427     if (!bitmap || !bs) {
2428         return NULL;
2429     }
2430 
2431     sha256 = bdrv_dirty_bitmap_sha256(bitmap, errp);
2432     if (sha256 == NULL) {
2433         return NULL;
2434     }
2435 
2436     ret = g_new(BlockDirtyBitmapSha256, 1);
2437     ret->sha256 = sha256;
2438 
2439     return ret;
2440 }
2441 
2442 void qmp_block_resize(bool has_device, const char *device,
2443                       bool has_node_name, const char *node_name,
2444                       int64_t size, Error **errp)
2445 {
2446     Error *local_err = NULL;
2447     BlockBackend *blk = NULL;
2448     BlockDriverState *bs;
2449     AioContext *aio_context;
2450 
2451     bs = bdrv_lookup_bs(has_device ? device : NULL,
2452                         has_node_name ? node_name : NULL,
2453                         &local_err);
2454     if (local_err) {
2455         error_propagate(errp, local_err);
2456         return;
2457     }
2458 
2459     aio_context = bdrv_get_aio_context(bs);
2460     aio_context_acquire(aio_context);
2461 
2462     if (size < 0) {
2463         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "size", "a >0 size");
2464         goto out;
2465     }
2466 
2467     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_RESIZE, NULL)) {
2468         error_setg(errp, QERR_DEVICE_IN_USE, device);
2469         goto out;
2470     }
2471 
2472     blk = blk_new_with_bs(bs, BLK_PERM_RESIZE, BLK_PERM_ALL, errp);
2473     if (!blk) {
2474         goto out;
2475     }
2476 
2477     bdrv_drained_begin(bs);
2478     blk_truncate(blk, size, false, PREALLOC_MODE_OFF, 0, errp);
2479     bdrv_drained_end(bs);
2480 
2481 out:
2482     blk_unref(blk);
2483     aio_context_release(aio_context);
2484 }
2485 
2486 void qmp_block_stream(bool has_job_id, const char *job_id, const char *device,
2487                       bool has_base, const char *base,
2488                       bool has_base_node, const char *base_node,
2489                       bool has_backing_file, const char *backing_file,
2490                       bool has_speed, int64_t speed,
2491                       bool has_on_error, BlockdevOnError on_error,
2492                       bool has_auto_finalize, bool auto_finalize,
2493                       bool has_auto_dismiss, bool auto_dismiss,
2494                       Error **errp)
2495 {
2496     BlockDriverState *bs, *iter;
2497     BlockDriverState *base_bs = NULL;
2498     AioContext *aio_context;
2499     Error *local_err = NULL;
2500     const char *base_name = NULL;
2501     int job_flags = JOB_DEFAULT;
2502 
2503     if (!has_on_error) {
2504         on_error = BLOCKDEV_ON_ERROR_REPORT;
2505     }
2506 
2507     bs = bdrv_lookup_bs(device, device, errp);
2508     if (!bs) {
2509         return;
2510     }
2511 
2512     aio_context = bdrv_get_aio_context(bs);
2513     aio_context_acquire(aio_context);
2514 
2515     if (has_base && has_base_node) {
2516         error_setg(errp, "'base' and 'base-node' cannot be specified "
2517                    "at the same time");
2518         goto out;
2519     }
2520 
2521     if (has_base) {
2522         base_bs = bdrv_find_backing_image(bs, base);
2523         if (base_bs == NULL) {
2524             error_setg(errp, QERR_BASE_NOT_FOUND, base);
2525             goto out;
2526         }
2527         assert(bdrv_get_aio_context(base_bs) == aio_context);
2528         base_name = base;
2529     }
2530 
2531     if (has_base_node) {
2532         base_bs = bdrv_lookup_bs(NULL, base_node, errp);
2533         if (!base_bs) {
2534             goto out;
2535         }
2536         if (bs == base_bs || !bdrv_chain_contains(bs, base_bs)) {
2537             error_setg(errp, "Node '%s' is not a backing image of '%s'",
2538                        base_node, device);
2539             goto out;
2540         }
2541         assert(bdrv_get_aio_context(base_bs) == aio_context);
2542         bdrv_refresh_filename(base_bs);
2543         base_name = base_bs->filename;
2544     }
2545 
2546     /* Check for op blockers in the whole chain between bs and base */
2547     for (iter = bs; iter && iter != base_bs;
2548          iter = bdrv_filter_or_cow_bs(iter))
2549     {
2550         if (bdrv_op_is_blocked(iter, BLOCK_OP_TYPE_STREAM, errp)) {
2551             goto out;
2552         }
2553     }
2554 
2555     /* if we are streaming the entire chain, the result will have no backing
2556      * file, and specifying one is therefore an error */
2557     if (base_bs == NULL && has_backing_file) {
2558         error_setg(errp, "backing file specified, but streaming the "
2559                          "entire chain");
2560         goto out;
2561     }
2562 
2563     /* backing_file string overrides base bs filename */
2564     base_name = has_backing_file ? backing_file : base_name;
2565 
2566     if (has_auto_finalize && !auto_finalize) {
2567         job_flags |= JOB_MANUAL_FINALIZE;
2568     }
2569     if (has_auto_dismiss && !auto_dismiss) {
2570         job_flags |= JOB_MANUAL_DISMISS;
2571     }
2572 
2573     stream_start(has_job_id ? job_id : NULL, bs, base_bs, base_name,
2574                  job_flags, has_speed ? speed : 0, on_error, &local_err);
2575     if (local_err) {
2576         error_propagate(errp, local_err);
2577         goto out;
2578     }
2579 
2580     trace_qmp_block_stream(bs);
2581 
2582 out:
2583     aio_context_release(aio_context);
2584 }
2585 
2586 void qmp_block_commit(bool has_job_id, const char *job_id, const char *device,
2587                       bool has_base_node, const char *base_node,
2588                       bool has_base, const char *base,
2589                       bool has_top_node, const char *top_node,
2590                       bool has_top, const char *top,
2591                       bool has_backing_file, const char *backing_file,
2592                       bool has_speed, int64_t speed,
2593                       bool has_on_error, BlockdevOnError on_error,
2594                       bool has_filter_node_name, const char *filter_node_name,
2595                       bool has_auto_finalize, bool auto_finalize,
2596                       bool has_auto_dismiss, bool auto_dismiss,
2597                       Error **errp)
2598 {
2599     BlockDriverState *bs;
2600     BlockDriverState *iter;
2601     BlockDriverState *base_bs, *top_bs;
2602     AioContext *aio_context;
2603     Error *local_err = NULL;
2604     int job_flags = JOB_DEFAULT;
2605     uint64_t top_perm, top_shared;
2606 
2607     if (!has_speed) {
2608         speed = 0;
2609     }
2610     if (!has_on_error) {
2611         on_error = BLOCKDEV_ON_ERROR_REPORT;
2612     }
2613     if (!has_filter_node_name) {
2614         filter_node_name = NULL;
2615     }
2616     if (has_auto_finalize && !auto_finalize) {
2617         job_flags |= JOB_MANUAL_FINALIZE;
2618     }
2619     if (has_auto_dismiss && !auto_dismiss) {
2620         job_flags |= JOB_MANUAL_DISMISS;
2621     }
2622 
2623     /* Important Note:
2624      *  libvirt relies on the DeviceNotFound error class in order to probe for
2625      *  live commit feature versions; for this to work, we must make sure to
2626      *  perform the device lookup before any generic errors that may occur in a
2627      *  scenario in which all optional arguments are omitted. */
2628     bs = qmp_get_root_bs(device, &local_err);
2629     if (!bs) {
2630         bs = bdrv_lookup_bs(device, device, NULL);
2631         if (!bs) {
2632             error_free(local_err);
2633             error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
2634                       "Device '%s' not found", device);
2635         } else {
2636             error_propagate(errp, local_err);
2637         }
2638         return;
2639     }
2640 
2641     aio_context = bdrv_get_aio_context(bs);
2642     aio_context_acquire(aio_context);
2643 
2644     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_COMMIT_SOURCE, errp)) {
2645         goto out;
2646     }
2647 
2648     /* default top_bs is the active layer */
2649     top_bs = bs;
2650 
2651     if (has_top_node && has_top) {
2652         error_setg(errp, "'top-node' and 'top' are mutually exclusive");
2653         goto out;
2654     } else if (has_top_node) {
2655         top_bs = bdrv_lookup_bs(NULL, top_node, errp);
2656         if (top_bs == NULL) {
2657             goto out;
2658         }
2659         if (!bdrv_chain_contains(bs, top_bs)) {
2660             error_setg(errp, "'%s' is not in this backing file chain",
2661                        top_node);
2662             goto out;
2663         }
2664     } else if (has_top && top) {
2665         /* This strcmp() is just a shortcut, there is no need to
2666          * refresh @bs's filename.  If it mismatches,
2667          * bdrv_find_backing_image() will do the refresh and may still
2668          * return @bs. */
2669         if (strcmp(bs->filename, top) != 0) {
2670             top_bs = bdrv_find_backing_image(bs, top);
2671         }
2672     }
2673 
2674     if (top_bs == NULL) {
2675         error_setg(errp, "Top image file %s not found", top ? top : "NULL");
2676         goto out;
2677     }
2678 
2679     assert(bdrv_get_aio_context(top_bs) == aio_context);
2680 
2681     if (has_base_node && has_base) {
2682         error_setg(errp, "'base-node' and 'base' are mutually exclusive");
2683         goto out;
2684     } else if (has_base_node) {
2685         base_bs = bdrv_lookup_bs(NULL, base_node, errp);
2686         if (base_bs == NULL) {
2687             goto out;
2688         }
2689         if (!bdrv_chain_contains(top_bs, base_bs)) {
2690             error_setg(errp, "'%s' is not in this backing file chain",
2691                        base_node);
2692             goto out;
2693         }
2694     } else if (has_base && base) {
2695         base_bs = bdrv_find_backing_image(top_bs, base);
2696     } else {
2697         base_bs = bdrv_find_base(top_bs);
2698     }
2699 
2700     if (base_bs == NULL) {
2701         error_setg(errp, QERR_BASE_NOT_FOUND, base ? base : "NULL");
2702         goto out;
2703     }
2704 
2705     assert(bdrv_get_aio_context(base_bs) == aio_context);
2706 
2707     for (iter = top_bs; iter != bdrv_filter_or_cow_bs(base_bs);
2708          iter = bdrv_filter_or_cow_bs(iter))
2709     {
2710         if (bdrv_op_is_blocked(iter, BLOCK_OP_TYPE_COMMIT_TARGET, errp)) {
2711             goto out;
2712         }
2713     }
2714 
2715     /* Do not allow attempts to commit an image into itself */
2716     if (top_bs == base_bs) {
2717         error_setg(errp, "cannot commit an image into itself");
2718         goto out;
2719     }
2720 
2721     /*
2722      * Active commit is required if and only if someone has taken a
2723      * WRITE permission on the top node.  Historically, we have always
2724      * used active commit for top nodes, so continue that practice
2725      * lest we possibly break clients that rely on this behavior, e.g.
2726      * to later attach this node to a writing parent.
2727      * (Active commit is never really wrong.)
2728      */
2729     bdrv_get_cumulative_perm(top_bs, &top_perm, &top_shared);
2730     if (top_perm & BLK_PERM_WRITE ||
2731         bdrv_skip_filters(top_bs) == bdrv_skip_filters(bs))
2732     {
2733         if (has_backing_file) {
2734             if (bdrv_skip_filters(top_bs) == bdrv_skip_filters(bs)) {
2735                 error_setg(errp, "'backing-file' specified,"
2736                                  " but 'top' is the active layer");
2737             } else {
2738                 error_setg(errp, "'backing-file' specified, but 'top' has a "
2739                                  "writer on it");
2740             }
2741             goto out;
2742         }
2743         if (!has_job_id) {
2744             /*
2745              * Emulate here what block_job_create() does, because it
2746              * is possible that @bs != @top_bs (the block job should
2747              * be named after @bs, even if @top_bs is the actual
2748              * source)
2749              */
2750             job_id = bdrv_get_device_name(bs);
2751         }
2752         commit_active_start(job_id, top_bs, base_bs, job_flags, speed, on_error,
2753                             filter_node_name, NULL, NULL, false, &local_err);
2754     } else {
2755         BlockDriverState *overlay_bs = bdrv_find_overlay(bs, top_bs);
2756         if (bdrv_op_is_blocked(overlay_bs, BLOCK_OP_TYPE_COMMIT_TARGET, errp)) {
2757             goto out;
2758         }
2759         commit_start(has_job_id ? job_id : NULL, bs, base_bs, top_bs, job_flags,
2760                      speed, on_error, has_backing_file ? backing_file : NULL,
2761                      filter_node_name, &local_err);
2762     }
2763     if (local_err != NULL) {
2764         error_propagate(errp, local_err);
2765         goto out;
2766     }
2767 
2768 out:
2769     aio_context_release(aio_context);
2770 }
2771 
2772 /* Common QMP interface for drive-backup and blockdev-backup */
2773 static BlockJob *do_backup_common(BackupCommon *backup,
2774                                   BlockDriverState *bs,
2775                                   BlockDriverState *target_bs,
2776                                   AioContext *aio_context,
2777                                   JobTxn *txn, Error **errp)
2778 {
2779     BlockJob *job = NULL;
2780     BdrvDirtyBitmap *bmap = NULL;
2781     int job_flags = JOB_DEFAULT;
2782 
2783     if (!backup->has_speed) {
2784         backup->speed = 0;
2785     }
2786     if (!backup->has_on_source_error) {
2787         backup->on_source_error = BLOCKDEV_ON_ERROR_REPORT;
2788     }
2789     if (!backup->has_on_target_error) {
2790         backup->on_target_error = BLOCKDEV_ON_ERROR_REPORT;
2791     }
2792     if (!backup->has_job_id) {
2793         backup->job_id = NULL;
2794     }
2795     if (!backup->has_auto_finalize) {
2796         backup->auto_finalize = true;
2797     }
2798     if (!backup->has_auto_dismiss) {
2799         backup->auto_dismiss = true;
2800     }
2801     if (!backup->has_compress) {
2802         backup->compress = false;
2803     }
2804 
2805     if ((backup->sync == MIRROR_SYNC_MODE_BITMAP) ||
2806         (backup->sync == MIRROR_SYNC_MODE_INCREMENTAL)) {
2807         /* done before desugaring 'incremental' to print the right message */
2808         if (!backup->has_bitmap) {
2809             error_setg(errp, "must provide a valid bitmap name for "
2810                        "'%s' sync mode", MirrorSyncMode_str(backup->sync));
2811             return NULL;
2812         }
2813     }
2814 
2815     if (backup->sync == MIRROR_SYNC_MODE_INCREMENTAL) {
2816         if (backup->has_bitmap_mode &&
2817             backup->bitmap_mode != BITMAP_SYNC_MODE_ON_SUCCESS) {
2818             error_setg(errp, "Bitmap sync mode must be '%s' "
2819                        "when using sync mode '%s'",
2820                        BitmapSyncMode_str(BITMAP_SYNC_MODE_ON_SUCCESS),
2821                        MirrorSyncMode_str(backup->sync));
2822             return NULL;
2823         }
2824         backup->has_bitmap_mode = true;
2825         backup->sync = MIRROR_SYNC_MODE_BITMAP;
2826         backup->bitmap_mode = BITMAP_SYNC_MODE_ON_SUCCESS;
2827     }
2828 
2829     if (backup->has_bitmap) {
2830         bmap = bdrv_find_dirty_bitmap(bs, backup->bitmap);
2831         if (!bmap) {
2832             error_setg(errp, "Bitmap '%s' could not be found", backup->bitmap);
2833             return NULL;
2834         }
2835         if (!backup->has_bitmap_mode) {
2836             error_setg(errp, "Bitmap sync mode must be given "
2837                        "when providing a bitmap");
2838             return NULL;
2839         }
2840         if (bdrv_dirty_bitmap_check(bmap, BDRV_BITMAP_ALLOW_RO, errp)) {
2841             return NULL;
2842         }
2843 
2844         /* This does not produce a useful bitmap artifact: */
2845         if (backup->sync == MIRROR_SYNC_MODE_NONE) {
2846             error_setg(errp, "sync mode '%s' does not produce meaningful bitmap"
2847                        " outputs", MirrorSyncMode_str(backup->sync));
2848             return NULL;
2849         }
2850 
2851         /* If the bitmap isn't used for input or output, this is useless: */
2852         if (backup->bitmap_mode == BITMAP_SYNC_MODE_NEVER &&
2853             backup->sync != MIRROR_SYNC_MODE_BITMAP) {
2854             error_setg(errp, "Bitmap sync mode '%s' has no meaningful effect"
2855                        " when combined with sync mode '%s'",
2856                        BitmapSyncMode_str(backup->bitmap_mode),
2857                        MirrorSyncMode_str(backup->sync));
2858             return NULL;
2859         }
2860     }
2861 
2862     if (!backup->has_bitmap && backup->has_bitmap_mode) {
2863         error_setg(errp, "Cannot specify bitmap sync mode without a bitmap");
2864         return NULL;
2865     }
2866 
2867     if (!backup->auto_finalize) {
2868         job_flags |= JOB_MANUAL_FINALIZE;
2869     }
2870     if (!backup->auto_dismiss) {
2871         job_flags |= JOB_MANUAL_DISMISS;
2872     }
2873 
2874     job = backup_job_create(backup->job_id, bs, target_bs, backup->speed,
2875                             backup->sync, bmap, backup->bitmap_mode,
2876                             backup->compress,
2877                             backup->filter_node_name,
2878                             backup->on_source_error,
2879                             backup->on_target_error,
2880                             job_flags, NULL, NULL, txn, errp);
2881     return job;
2882 }
2883 
2884 void qmp_drive_backup(DriveBackup *backup, Error **errp)
2885 {
2886     TransactionAction action = {
2887         .type = TRANSACTION_ACTION_KIND_DRIVE_BACKUP,
2888         .u.drive_backup.data = backup,
2889     };
2890     blockdev_do_action(&action, errp);
2891 }
2892 
2893 BlockDeviceInfoList *qmp_query_named_block_nodes(bool has_flat,
2894                                                  bool flat,
2895                                                  Error **errp)
2896 {
2897     bool return_flat = has_flat && flat;
2898 
2899     return bdrv_named_nodes_list(return_flat, errp);
2900 }
2901 
2902 XDbgBlockGraph *qmp_x_debug_query_block_graph(Error **errp)
2903 {
2904     return bdrv_get_xdbg_block_graph(errp);
2905 }
2906 
2907 void qmp_blockdev_backup(BlockdevBackup *backup, Error **errp)
2908 {
2909     TransactionAction action = {
2910         .type = TRANSACTION_ACTION_KIND_BLOCKDEV_BACKUP,
2911         .u.blockdev_backup.data = backup,
2912     };
2913     blockdev_do_action(&action, errp);
2914 }
2915 
2916 /* Parameter check and block job starting for drive mirroring.
2917  * Caller should hold @device and @target's aio context (must be the same).
2918  **/
2919 static void blockdev_mirror_common(const char *job_id, BlockDriverState *bs,
2920                                    BlockDriverState *target,
2921                                    bool has_replaces, const char *replaces,
2922                                    enum MirrorSyncMode sync,
2923                                    BlockMirrorBackingMode backing_mode,
2924                                    bool zero_target,
2925                                    bool has_speed, int64_t speed,
2926                                    bool has_granularity, uint32_t granularity,
2927                                    bool has_buf_size, int64_t buf_size,
2928                                    bool has_on_source_error,
2929                                    BlockdevOnError on_source_error,
2930                                    bool has_on_target_error,
2931                                    BlockdevOnError on_target_error,
2932                                    bool has_unmap, bool unmap,
2933                                    bool has_filter_node_name,
2934                                    const char *filter_node_name,
2935                                    bool has_copy_mode, MirrorCopyMode copy_mode,
2936                                    bool has_auto_finalize, bool auto_finalize,
2937                                    bool has_auto_dismiss, bool auto_dismiss,
2938                                    Error **errp)
2939 {
2940     BlockDriverState *unfiltered_bs;
2941     int job_flags = JOB_DEFAULT;
2942 
2943     if (!has_speed) {
2944         speed = 0;
2945     }
2946     if (!has_on_source_error) {
2947         on_source_error = BLOCKDEV_ON_ERROR_REPORT;
2948     }
2949     if (!has_on_target_error) {
2950         on_target_error = BLOCKDEV_ON_ERROR_REPORT;
2951     }
2952     if (!has_granularity) {
2953         granularity = 0;
2954     }
2955     if (!has_buf_size) {
2956         buf_size = 0;
2957     }
2958     if (!has_unmap) {
2959         unmap = true;
2960     }
2961     if (!has_filter_node_name) {
2962         filter_node_name = NULL;
2963     }
2964     if (!has_copy_mode) {
2965         copy_mode = MIRROR_COPY_MODE_BACKGROUND;
2966     }
2967     if (has_auto_finalize && !auto_finalize) {
2968         job_flags |= JOB_MANUAL_FINALIZE;
2969     }
2970     if (has_auto_dismiss && !auto_dismiss) {
2971         job_flags |= JOB_MANUAL_DISMISS;
2972     }
2973 
2974     if (granularity != 0 && (granularity < 512 || granularity > 1048576 * 64)) {
2975         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "granularity",
2976                    "a value in range [512B, 64MB]");
2977         return;
2978     }
2979     if (granularity & (granularity - 1)) {
2980         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "granularity",
2981                    "power of 2");
2982         return;
2983     }
2984 
2985     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_MIRROR_SOURCE, errp)) {
2986         return;
2987     }
2988     if (bdrv_op_is_blocked(target, BLOCK_OP_TYPE_MIRROR_TARGET, errp)) {
2989         return;
2990     }
2991 
2992     if (!bdrv_backing_chain_next(bs) && sync == MIRROR_SYNC_MODE_TOP) {
2993         sync = MIRROR_SYNC_MODE_FULL;
2994     }
2995 
2996     if (!has_replaces) {
2997         /* We want to mirror from @bs, but keep implicit filters on top */
2998         unfiltered_bs = bdrv_skip_implicit_filters(bs);
2999         if (unfiltered_bs != bs) {
3000             replaces = unfiltered_bs->node_name;
3001             has_replaces = true;
3002         }
3003     }
3004 
3005     if (has_replaces) {
3006         BlockDriverState *to_replace_bs;
3007         AioContext *replace_aio_context;
3008         int64_t bs_size, replace_size;
3009 
3010         bs_size = bdrv_getlength(bs);
3011         if (bs_size < 0) {
3012             error_setg_errno(errp, -bs_size, "Failed to query device's size");
3013             return;
3014         }
3015 
3016         to_replace_bs = check_to_replace_node(bs, replaces, errp);
3017         if (!to_replace_bs) {
3018             return;
3019         }
3020 
3021         replace_aio_context = bdrv_get_aio_context(to_replace_bs);
3022         aio_context_acquire(replace_aio_context);
3023         replace_size = bdrv_getlength(to_replace_bs);
3024         aio_context_release(replace_aio_context);
3025 
3026         if (replace_size < 0) {
3027             error_setg_errno(errp, -replace_size,
3028                              "Failed to query the replacement node's size");
3029             return;
3030         }
3031         if (bs_size != replace_size) {
3032             error_setg(errp, "cannot replace image with a mirror image of "
3033                              "different size");
3034             return;
3035         }
3036     }
3037 
3038     /* pass the node name to replace to mirror start since it's loose coupling
3039      * and will allow to check whether the node still exist at mirror completion
3040      */
3041     mirror_start(job_id, bs, target,
3042                  has_replaces ? replaces : NULL, job_flags,
3043                  speed, granularity, buf_size, sync, backing_mode, zero_target,
3044                  on_source_error, on_target_error, unmap, filter_node_name,
3045                  copy_mode, errp);
3046 }
3047 
3048 void qmp_drive_mirror(DriveMirror *arg, Error **errp)
3049 {
3050     BlockDriverState *bs;
3051     BlockDriverState *target_backing_bs, *target_bs;
3052     AioContext *aio_context;
3053     AioContext *old_context;
3054     BlockMirrorBackingMode backing_mode;
3055     Error *local_err = NULL;
3056     QDict *options = NULL;
3057     int flags;
3058     int64_t size;
3059     const char *format = arg->format;
3060     bool zero_target;
3061     int ret;
3062 
3063     bs = qmp_get_root_bs(arg->device, errp);
3064     if (!bs) {
3065         return;
3066     }
3067 
3068     /* Early check to avoid creating target */
3069     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_MIRROR_SOURCE, errp)) {
3070         return;
3071     }
3072 
3073     aio_context = bdrv_get_aio_context(bs);
3074     aio_context_acquire(aio_context);
3075 
3076     if (!arg->has_mode) {
3077         arg->mode = NEW_IMAGE_MODE_ABSOLUTE_PATHS;
3078     }
3079 
3080     if (!arg->has_format) {
3081         format = (arg->mode == NEW_IMAGE_MODE_EXISTING
3082                   ? NULL : bs->drv->format_name);
3083     }
3084 
3085     flags = bs->open_flags | BDRV_O_RDWR;
3086     target_backing_bs = bdrv_cow_bs(bdrv_skip_filters(bs));
3087     if (!target_backing_bs && arg->sync == MIRROR_SYNC_MODE_TOP) {
3088         arg->sync = MIRROR_SYNC_MODE_FULL;
3089     }
3090     if (arg->sync == MIRROR_SYNC_MODE_NONE) {
3091         target_backing_bs = bs;
3092     }
3093 
3094     size = bdrv_getlength(bs);
3095     if (size < 0) {
3096         error_setg_errno(errp, -size, "bdrv_getlength failed");
3097         goto out;
3098     }
3099 
3100     if (arg->has_replaces) {
3101         if (!arg->has_node_name) {
3102             error_setg(errp, "a node-name must be provided when replacing a"
3103                              " named node of the graph");
3104             goto out;
3105         }
3106     }
3107 
3108     if (arg->mode == NEW_IMAGE_MODE_ABSOLUTE_PATHS) {
3109         backing_mode = MIRROR_SOURCE_BACKING_CHAIN;
3110     } else {
3111         backing_mode = MIRROR_OPEN_BACKING_CHAIN;
3112     }
3113 
3114     /* Don't open backing image in create() */
3115     flags |= BDRV_O_NO_BACKING;
3116 
3117     if ((arg->sync == MIRROR_SYNC_MODE_FULL || !target_backing_bs)
3118         && arg->mode != NEW_IMAGE_MODE_EXISTING)
3119     {
3120         /* create new image w/o backing file */
3121         assert(format);
3122         bdrv_img_create(arg->target, format,
3123                         NULL, NULL, NULL, size, flags, false, &local_err);
3124     } else {
3125         /* Implicit filters should not appear in the filename */
3126         BlockDriverState *explicit_backing =
3127             bdrv_skip_implicit_filters(target_backing_bs);
3128 
3129         switch (arg->mode) {
3130         case NEW_IMAGE_MODE_EXISTING:
3131             break;
3132         case NEW_IMAGE_MODE_ABSOLUTE_PATHS:
3133             /* create new image with backing file */
3134             bdrv_refresh_filename(explicit_backing);
3135             bdrv_img_create(arg->target, format,
3136                             explicit_backing->filename,
3137                             explicit_backing->drv->format_name,
3138                             NULL, size, flags, false, &local_err);
3139             break;
3140         default:
3141             abort();
3142         }
3143     }
3144 
3145     if (local_err) {
3146         error_propagate(errp, local_err);
3147         goto out;
3148     }
3149 
3150     options = qdict_new();
3151     if (arg->has_node_name) {
3152         qdict_put_str(options, "node-name", arg->node_name);
3153     }
3154     if (format) {
3155         qdict_put_str(options, "driver", format);
3156     }
3157 
3158     /* Mirroring takes care of copy-on-write using the source's backing
3159      * file.
3160      */
3161     target_bs = bdrv_open(arg->target, NULL, options, flags, errp);
3162     if (!target_bs) {
3163         goto out;
3164     }
3165 
3166     zero_target = (arg->sync == MIRROR_SYNC_MODE_FULL &&
3167                    (arg->mode == NEW_IMAGE_MODE_EXISTING ||
3168                     !bdrv_has_zero_init(target_bs)));
3169 
3170 
3171     /* Honor bdrv_try_set_aio_context() context acquisition requirements. */
3172     old_context = bdrv_get_aio_context(target_bs);
3173     aio_context_release(aio_context);
3174     aio_context_acquire(old_context);
3175 
3176     ret = bdrv_try_set_aio_context(target_bs, aio_context, errp);
3177     if (ret < 0) {
3178         bdrv_unref(target_bs);
3179         aio_context_release(old_context);
3180         return;
3181     }
3182 
3183     aio_context_release(old_context);
3184     aio_context_acquire(aio_context);
3185 
3186     blockdev_mirror_common(arg->has_job_id ? arg->job_id : NULL, bs, target_bs,
3187                            arg->has_replaces, arg->replaces, arg->sync,
3188                            backing_mode, zero_target,
3189                            arg->has_speed, arg->speed,
3190                            arg->has_granularity, arg->granularity,
3191                            arg->has_buf_size, arg->buf_size,
3192                            arg->has_on_source_error, arg->on_source_error,
3193                            arg->has_on_target_error, arg->on_target_error,
3194                            arg->has_unmap, arg->unmap,
3195                            false, NULL,
3196                            arg->has_copy_mode, arg->copy_mode,
3197                            arg->has_auto_finalize, arg->auto_finalize,
3198                            arg->has_auto_dismiss, arg->auto_dismiss,
3199                            errp);
3200     bdrv_unref(target_bs);
3201 out:
3202     aio_context_release(aio_context);
3203 }
3204 
3205 void qmp_blockdev_mirror(bool has_job_id, const char *job_id,
3206                          const char *device, const char *target,
3207                          bool has_replaces, const char *replaces,
3208                          MirrorSyncMode sync,
3209                          bool has_speed, int64_t speed,
3210                          bool has_granularity, uint32_t granularity,
3211                          bool has_buf_size, int64_t buf_size,
3212                          bool has_on_source_error,
3213                          BlockdevOnError on_source_error,
3214                          bool has_on_target_error,
3215                          BlockdevOnError on_target_error,
3216                          bool has_filter_node_name,
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_set_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_set_aio_context(target_bs, aio_context, 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(has_job_id ? job_id : NULL, bs, target_bs,
3258                            has_replaces, 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,
3265                            has_filter_node_name, filter_node_name,
3266                            has_copy_mode, copy_mode,
3267                            has_auto_finalize, auto_finalize,
3268                            has_auto_dismiss, auto_dismiss,
3269                            errp);
3270 out:
3271     aio_context_release(aio_context);
3272 }
3273 
3274 /* Get a block job using its ID and acquire its AioContext */
3275 static BlockJob *find_block_job(const char *id, AioContext **aio_context,
3276                                 Error **errp)
3277 {
3278     BlockJob *job;
3279 
3280     assert(id != NULL);
3281 
3282     *aio_context = NULL;
3283 
3284     job = block_job_get(id);
3285 
3286     if (!job) {
3287         error_set(errp, ERROR_CLASS_DEVICE_NOT_ACTIVE,
3288                   "Block job '%s' not found", id);
3289         return NULL;
3290     }
3291 
3292     *aio_context = blk_get_aio_context(job->blk);
3293     aio_context_acquire(*aio_context);
3294 
3295     return job;
3296 }
3297 
3298 void qmp_block_job_set_speed(const char *device, int64_t speed, Error **errp)
3299 {
3300     AioContext *aio_context;
3301     BlockJob *job = find_block_job(device, &aio_context, errp);
3302 
3303     if (!job) {
3304         return;
3305     }
3306 
3307     block_job_set_speed(job, speed, errp);
3308     aio_context_release(aio_context);
3309 }
3310 
3311 void qmp_block_job_cancel(const char *device,
3312                           bool has_force, bool force, Error **errp)
3313 {
3314     AioContext *aio_context;
3315     BlockJob *job = find_block_job(device, &aio_context, errp);
3316 
3317     if (!job) {
3318         return;
3319     }
3320 
3321     if (!has_force) {
3322         force = false;
3323     }
3324 
3325     if (job_user_paused(&job->job) && !force) {
3326         error_setg(errp, "The block job for device '%s' is currently paused",
3327                    device);
3328         goto out;
3329     }
3330 
3331     trace_qmp_block_job_cancel(job);
3332     job_user_cancel(&job->job, force, errp);
3333 out:
3334     aio_context_release(aio_context);
3335 }
3336 
3337 void qmp_block_job_pause(const char *device, Error **errp)
3338 {
3339     AioContext *aio_context;
3340     BlockJob *job = find_block_job(device, &aio_context, errp);
3341 
3342     if (!job) {
3343         return;
3344     }
3345 
3346     trace_qmp_block_job_pause(job);
3347     job_user_pause(&job->job, errp);
3348     aio_context_release(aio_context);
3349 }
3350 
3351 void qmp_block_job_resume(const char *device, Error **errp)
3352 {
3353     AioContext *aio_context;
3354     BlockJob *job = find_block_job(device, &aio_context, errp);
3355 
3356     if (!job) {
3357         return;
3358     }
3359 
3360     trace_qmp_block_job_resume(job);
3361     job_user_resume(&job->job, errp);
3362     aio_context_release(aio_context);
3363 }
3364 
3365 void qmp_block_job_complete(const char *device, Error **errp)
3366 {
3367     AioContext *aio_context;
3368     BlockJob *job = find_block_job(device, &aio_context, errp);
3369 
3370     if (!job) {
3371         return;
3372     }
3373 
3374     trace_qmp_block_job_complete(job);
3375     job_complete(&job->job, errp);
3376     aio_context_release(aio_context);
3377 }
3378 
3379 void qmp_block_job_finalize(const char *id, Error **errp)
3380 {
3381     AioContext *aio_context;
3382     BlockJob *job = find_block_job(id, &aio_context, errp);
3383 
3384     if (!job) {
3385         return;
3386     }
3387 
3388     trace_qmp_block_job_finalize(job);
3389     job_ref(&job->job);
3390     job_finalize(&job->job, errp);
3391 
3392     /*
3393      * Job's context might have changed via job_finalize (and job_txn_apply
3394      * automatically acquires the new one), so make sure we release the correct
3395      * one.
3396      */
3397     aio_context = blk_get_aio_context(job->blk);
3398     job_unref(&job->job);
3399     aio_context_release(aio_context);
3400 }
3401 
3402 void qmp_block_job_dismiss(const char *id, Error **errp)
3403 {
3404     AioContext *aio_context;
3405     BlockJob *bjob = find_block_job(id, &aio_context, errp);
3406     Job *job;
3407 
3408     if (!bjob) {
3409         return;
3410     }
3411 
3412     trace_qmp_block_job_dismiss(bjob);
3413     job = &bjob->job;
3414     job_dismiss(&job, errp);
3415     aio_context_release(aio_context);
3416 }
3417 
3418 void qmp_change_backing_file(const char *device,
3419                              const char *image_node_name,
3420                              const char *backing_file,
3421                              Error **errp)
3422 {
3423     BlockDriverState *bs = NULL;
3424     AioContext *aio_context;
3425     BlockDriverState *image_bs = NULL;
3426     Error *local_err = NULL;
3427     bool ro;
3428     int ret;
3429 
3430     bs = qmp_get_root_bs(device, errp);
3431     if (!bs) {
3432         return;
3433     }
3434 
3435     aio_context = bdrv_get_aio_context(bs);
3436     aio_context_acquire(aio_context);
3437 
3438     image_bs = bdrv_lookup_bs(NULL, image_node_name, &local_err);
3439     if (local_err) {
3440         error_propagate(errp, local_err);
3441         goto out;
3442     }
3443 
3444     if (!image_bs) {
3445         error_setg(errp, "image file not found");
3446         goto out;
3447     }
3448 
3449     if (bdrv_find_base(image_bs) == image_bs) {
3450         error_setg(errp, "not allowing backing file change on an image "
3451                          "without a backing file");
3452         goto out;
3453     }
3454 
3455     /* even though we are not necessarily operating on bs, we need it to
3456      * determine if block ops are currently prohibited on the chain */
3457     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_CHANGE, errp)) {
3458         goto out;
3459     }
3460 
3461     /* final sanity check */
3462     if (!bdrv_chain_contains(bs, image_bs)) {
3463         error_setg(errp, "'%s' and image file are not in the same chain",
3464                    device);
3465         goto out;
3466     }
3467 
3468     /* if not r/w, reopen to make r/w */
3469     ro = bdrv_is_read_only(image_bs);
3470 
3471     if (ro) {
3472         if (bdrv_reopen_set_read_only(image_bs, false, errp) != 0) {
3473             goto out;
3474         }
3475     }
3476 
3477     ret = bdrv_change_backing_file(image_bs, backing_file,
3478                                    image_bs->drv ? image_bs->drv->format_name : "",
3479                                    false);
3480 
3481     if (ret < 0) {
3482         error_setg_errno(errp, -ret, "Could not change backing file to '%s'",
3483                          backing_file);
3484         /* don't exit here, so we can try to restore open flags if
3485          * appropriate */
3486     }
3487 
3488     if (ro) {
3489         bdrv_reopen_set_read_only(image_bs, true, errp);
3490     }
3491 
3492 out:
3493     aio_context_release(aio_context);
3494 }
3495 
3496 void qmp_blockdev_add(BlockdevOptions *options, Error **errp)
3497 {
3498     BlockDriverState *bs;
3499     QObject *obj;
3500     Visitor *v = qobject_output_visitor_new(&obj);
3501     QDict *qdict;
3502 
3503     visit_type_BlockdevOptions(v, NULL, &options, &error_abort);
3504     visit_complete(v, &obj);
3505     qdict = qobject_to(QDict, obj);
3506 
3507     qdict_flatten(qdict);
3508 
3509     if (!qdict_get_try_str(qdict, "node-name")) {
3510         error_setg(errp, "'node-name' must be specified for the root node");
3511         goto fail;
3512     }
3513 
3514     bs = bds_tree_init(qdict, errp);
3515     if (!bs) {
3516         goto fail;
3517     }
3518 
3519     bdrv_set_monitor_owned(bs);
3520 
3521 fail:
3522     visit_free(v);
3523 }
3524 
3525 void qmp_x_blockdev_reopen(BlockdevOptions *options, Error **errp)
3526 {
3527     BlockDriverState *bs;
3528     AioContext *ctx;
3529     QObject *obj;
3530     Visitor *v = qobject_output_visitor_new(&obj);
3531     BlockReopenQueue *queue;
3532     QDict *qdict;
3533 
3534     /* Check for the selected node name */
3535     if (!options->has_node_name) {
3536         error_setg(errp, "Node name not specified");
3537         goto fail;
3538     }
3539 
3540     bs = bdrv_find_node(options->node_name);
3541     if (!bs) {
3542         error_setg(errp, "Cannot find node named '%s'", options->node_name);
3543         goto fail;
3544     }
3545 
3546     /* Put all options in a QDict and flatten it */
3547     visit_type_BlockdevOptions(v, NULL, &options, &error_abort);
3548     visit_complete(v, &obj);
3549     qdict = qobject_to(QDict, obj);
3550 
3551     qdict_flatten(qdict);
3552 
3553     /* Perform the reopen operation */
3554     ctx = bdrv_get_aio_context(bs);
3555     aio_context_acquire(ctx);
3556     bdrv_subtree_drained_begin(bs);
3557     queue = bdrv_reopen_queue(NULL, bs, qdict, false);
3558     bdrv_reopen_multiple(queue, errp);
3559     bdrv_subtree_drained_end(bs);
3560     aio_context_release(ctx);
3561 
3562 fail:
3563     visit_free(v);
3564 }
3565 
3566 void qmp_blockdev_del(const char *node_name, Error **errp)
3567 {
3568     AioContext *aio_context;
3569     BlockDriverState *bs;
3570 
3571     bs = bdrv_find_node(node_name);
3572     if (!bs) {
3573         error_setg(errp, "Cannot find node %s", node_name);
3574         return;
3575     }
3576     if (bdrv_has_blk(bs)) {
3577         error_setg(errp, "Node %s is in use", node_name);
3578         return;
3579     }
3580     aio_context = bdrv_get_aio_context(bs);
3581     aio_context_acquire(aio_context);
3582 
3583     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_DRIVE_DEL, errp)) {
3584         goto out;
3585     }
3586 
3587     if (!QTAILQ_IN_USE(bs, monitor_list)) {
3588         error_setg(errp, "Node %s is not owned by the monitor",
3589                    bs->node_name);
3590         goto out;
3591     }
3592 
3593     if (bs->refcnt > 1) {
3594         error_setg(errp, "Block device %s is in use",
3595                    bdrv_get_device_or_node_name(bs));
3596         goto out;
3597     }
3598 
3599     QTAILQ_REMOVE(&monitor_bdrv_states, bs, monitor_list);
3600     bdrv_unref(bs);
3601 
3602 out:
3603     aio_context_release(aio_context);
3604 }
3605 
3606 static BdrvChild *bdrv_find_child(BlockDriverState *parent_bs,
3607                                   const char *child_name)
3608 {
3609     BdrvChild *child;
3610 
3611     QLIST_FOREACH(child, &parent_bs->children, next) {
3612         if (strcmp(child->name, child_name) == 0) {
3613             return child;
3614         }
3615     }
3616 
3617     return NULL;
3618 }
3619 
3620 void qmp_x_blockdev_change(const char *parent, bool has_child,
3621                            const char *child, bool has_node,
3622                            const char *node, Error **errp)
3623 {
3624     BlockDriverState *parent_bs, *new_bs = NULL;
3625     BdrvChild *p_child;
3626 
3627     parent_bs = bdrv_lookup_bs(parent, parent, errp);
3628     if (!parent_bs) {
3629         return;
3630     }
3631 
3632     if (has_child == has_node) {
3633         if (has_child) {
3634             error_setg(errp, "The parameters child and node are in conflict");
3635         } else {
3636             error_setg(errp, "Either child or node must be specified");
3637         }
3638         return;
3639     }
3640 
3641     if (has_child) {
3642         p_child = bdrv_find_child(parent_bs, child);
3643         if (!p_child) {
3644             error_setg(errp, "Node '%s' does not have child '%s'",
3645                        parent, child);
3646             return;
3647         }
3648         bdrv_del_child(parent_bs, p_child, errp);
3649     }
3650 
3651     if (has_node) {
3652         new_bs = bdrv_find_node(node);
3653         if (!new_bs) {
3654             error_setg(errp, "Node '%s' not found", node);
3655             return;
3656         }
3657         bdrv_add_child(parent_bs, new_bs, errp);
3658     }
3659 }
3660 
3661 BlockJobInfoList *qmp_query_block_jobs(Error **errp)
3662 {
3663     BlockJobInfoList *head = NULL, **p_next = &head;
3664     BlockJob *job;
3665 
3666     for (job = block_job_next(NULL); job; job = block_job_next(job)) {
3667         BlockJobInfoList *elem;
3668         AioContext *aio_context;
3669 
3670         if (block_job_is_internal(job)) {
3671             continue;
3672         }
3673         elem = g_new0(BlockJobInfoList, 1);
3674         aio_context = blk_get_aio_context(job->blk);
3675         aio_context_acquire(aio_context);
3676         elem->value = block_job_query(job, errp);
3677         aio_context_release(aio_context);
3678         if (!elem->value) {
3679             g_free(elem);
3680             qapi_free_BlockJobInfoList(head);
3681             return NULL;
3682         }
3683         *p_next = elem;
3684         p_next = &elem->next;
3685     }
3686 
3687     return head;
3688 }
3689 
3690 void qmp_x_blockdev_set_iothread(const char *node_name, StrOrNull *iothread,
3691                                  bool has_force, bool force, Error **errp)
3692 {
3693     AioContext *old_context;
3694     AioContext *new_context;
3695     BlockDriverState *bs;
3696 
3697     bs = bdrv_find_node(node_name);
3698     if (!bs) {
3699         error_setg(errp, "Cannot find node %s", node_name);
3700         return;
3701     }
3702 
3703     /* Protects against accidents. */
3704     if (!(has_force && force) && bdrv_has_blk(bs)) {
3705         error_setg(errp, "Node %s is associated with a BlockBackend and could "
3706                          "be in use (use force=true to override this check)",
3707                          node_name);
3708         return;
3709     }
3710 
3711     if (iothread->type == QTYPE_QSTRING) {
3712         IOThread *obj = iothread_by_id(iothread->u.s);
3713         if (!obj) {
3714             error_setg(errp, "Cannot find iothread %s", iothread->u.s);
3715             return;
3716         }
3717 
3718         new_context = iothread_get_aio_context(obj);
3719     } else {
3720         new_context = qemu_get_aio_context();
3721     }
3722 
3723     old_context = bdrv_get_aio_context(bs);
3724     aio_context_acquire(old_context);
3725 
3726     bdrv_try_set_aio_context(bs, new_context, errp);
3727 
3728     aio_context_release(old_context);
3729 }
3730 
3731 QemuOptsList qemu_common_drive_opts = {
3732     .name = "drive",
3733     .head = QTAILQ_HEAD_INITIALIZER(qemu_common_drive_opts.head),
3734     .desc = {
3735         {
3736             .name = "snapshot",
3737             .type = QEMU_OPT_BOOL,
3738             .help = "enable/disable snapshot mode",
3739         },{
3740             .name = "aio",
3741             .type = QEMU_OPT_STRING,
3742             .help = "host AIO implementation (threads, native, io_uring)",
3743         },{
3744             .name = BDRV_OPT_CACHE_WB,
3745             .type = QEMU_OPT_BOOL,
3746             .help = "Enable writeback mode",
3747         },{
3748             .name = "format",
3749             .type = QEMU_OPT_STRING,
3750             .help = "disk format (raw, qcow2, ...)",
3751         },{
3752             .name = "rerror",
3753             .type = QEMU_OPT_STRING,
3754             .help = "read error action",
3755         },{
3756             .name = "werror",
3757             .type = QEMU_OPT_STRING,
3758             .help = "write error action",
3759         },{
3760             .name = BDRV_OPT_READ_ONLY,
3761             .type = QEMU_OPT_BOOL,
3762             .help = "open drive file as read-only",
3763         },
3764 
3765         THROTTLE_OPTS,
3766 
3767         {
3768             .name = "throttling.group",
3769             .type = QEMU_OPT_STRING,
3770             .help = "name of the block throttling group",
3771         },{
3772             .name = "copy-on-read",
3773             .type = QEMU_OPT_BOOL,
3774             .help = "copy read data from backing file into image file",
3775         },{
3776             .name = "detect-zeroes",
3777             .type = QEMU_OPT_STRING,
3778             .help = "try to optimize zero writes (off, on, unmap)",
3779         },{
3780             .name = "stats-account-invalid",
3781             .type = QEMU_OPT_BOOL,
3782             .help = "whether to account for invalid I/O operations "
3783                     "in the statistics",
3784         },{
3785             .name = "stats-account-failed",
3786             .type = QEMU_OPT_BOOL,
3787             .help = "whether to account for failed I/O operations "
3788                     "in the statistics",
3789         },
3790         { /* end of list */ }
3791     },
3792 };
3793 
3794 QemuOptsList qemu_drive_opts = {
3795     .name = "drive",
3796     .head = QTAILQ_HEAD_INITIALIZER(qemu_drive_opts.head),
3797     .desc = {
3798         /*
3799          * no elements => accept any params
3800          * validation will happen later
3801          */
3802         { /* end of list */ }
3803     },
3804 };
3805