xref: /openbmc/qemu/blockdev.c (revision 2068cabd)
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/runstate.h"
61 #include "sysemu/replay.h"
62 #include "qemu/cutils.h"
63 #include "qemu/help_option.h"
64 #include "qemu/main-loop.h"
65 #include "qemu/throttle-options.h"
66 
67 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 /*
243  * Check board claimed all -drive that are meant to be claimed.
244  * Fatal error if any remain unclaimed.
245  */
246 void drive_check_orphaned(void)
247 {
248     BlockBackend *blk;
249     DriveInfo *dinfo;
250     Location loc;
251     bool orphans = false;
252 
253     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
254         dinfo = blk_legacy_dinfo(blk);
255         /*
256          * Ignore default drives, because we create certain default
257          * drives unconditionally, then leave them unclaimed.  Not the
258          * users fault.
259          * Ignore IF_VIRTIO, because it gets desugared into -device,
260          * so we can leave failing to -device.
261          * Ignore IF_NONE, because leaving unclaimed IF_NONE remains
262          * available for device_add is a feature.
263          */
264         if (dinfo->is_default || dinfo->type == IF_VIRTIO
265             || dinfo->type == IF_NONE) {
266             continue;
267         }
268         if (!blk_get_attached_dev(blk)) {
269             loc_push_none(&loc);
270             qemu_opts_loc_restore(dinfo->opts);
271             error_report("machine type does not support"
272                          " if=%s,bus=%d,unit=%d",
273                          if_name[dinfo->type], dinfo->bus, dinfo->unit);
274             loc_pop(&loc);
275             orphans = true;
276         }
277     }
278 
279     if (orphans) {
280         exit(1);
281     }
282 }
283 
284 DriveInfo *drive_get_by_index(BlockInterfaceType type, int index)
285 {
286     return drive_get(type,
287                      drive_index_to_bus_id(type, index),
288                      drive_index_to_unit_id(type, index));
289 }
290 
291 int drive_get_max_bus(BlockInterfaceType type)
292 {
293     int max_bus;
294     BlockBackend *blk;
295     DriveInfo *dinfo;
296 
297     max_bus = -1;
298     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
299         dinfo = blk_legacy_dinfo(blk);
300         if (dinfo && dinfo->type == type && dinfo->bus > max_bus) {
301             max_bus = dinfo->bus;
302         }
303     }
304     return max_bus;
305 }
306 
307 /* Get a block device.  This should only be used for single-drive devices
308    (e.g. SD/Floppy/MTD).  Multi-disk devices (scsi/ide) should use the
309    appropriate bus.  */
310 DriveInfo *drive_get_next(BlockInterfaceType type)
311 {
312     static int next_block_unit[IF_COUNT];
313 
314     return drive_get(type, 0, next_block_unit[type]++);
315 }
316 
317 static void bdrv_format_print(void *opaque, const char *name)
318 {
319     qemu_printf(" %s", name);
320 }
321 
322 typedef struct {
323     QEMUBH *bh;
324     BlockDriverState *bs;
325 } BDRVPutRefBH;
326 
327 static int parse_block_error_action(const char *buf, bool is_read, Error **errp)
328 {
329     if (!strcmp(buf, "ignore")) {
330         return BLOCKDEV_ON_ERROR_IGNORE;
331     } else if (!is_read && !strcmp(buf, "enospc")) {
332         return BLOCKDEV_ON_ERROR_ENOSPC;
333     } else if (!strcmp(buf, "stop")) {
334         return BLOCKDEV_ON_ERROR_STOP;
335     } else if (!strcmp(buf, "report")) {
336         return BLOCKDEV_ON_ERROR_REPORT;
337     } else {
338         error_setg(errp, "'%s' invalid %s error action",
339                    buf, is_read ? "read" : "write");
340         return -1;
341     }
342 }
343 
344 static bool parse_stats_intervals(BlockAcctStats *stats, QList *intervals,
345                                   Error **errp)
346 {
347     const QListEntry *entry;
348     for (entry = qlist_first(intervals); entry; entry = qlist_next(entry)) {
349         switch (qobject_type(entry->value)) {
350 
351         case QTYPE_QSTRING: {
352             unsigned long long length;
353             const char *str = qstring_get_str(qobject_to(QString,
354                                                          entry->value));
355             if (parse_uint_full(str, &length, 10) == 0 &&
356                 length > 0 && length <= UINT_MAX) {
357                 block_acct_add_interval(stats, (unsigned) length);
358             } else {
359                 error_setg(errp, "Invalid interval length: %s", str);
360                 return false;
361             }
362             break;
363         }
364 
365         case QTYPE_QNUM: {
366             int64_t length = qnum_get_int(qobject_to(QNum, entry->value));
367 
368             if (length > 0 && length <= UINT_MAX) {
369                 block_acct_add_interval(stats, (unsigned) length);
370             } else {
371                 error_setg(errp, "Invalid interval length: %" PRId64, length);
372                 return false;
373             }
374             break;
375         }
376 
377         default:
378             error_setg(errp, "The specification of stats-intervals is invalid");
379             return false;
380         }
381     }
382     return true;
383 }
384 
385 typedef enum { MEDIA_DISK, MEDIA_CDROM } DriveMediaType;
386 
387 /* All parameters but @opts are optional and may be set to NULL. */
388 static void extract_common_blockdev_options(QemuOpts *opts, int *bdrv_flags,
389     const char **throttling_group, ThrottleConfig *throttle_cfg,
390     BlockdevDetectZeroesOptions *detect_zeroes, Error **errp)
391 {
392     Error *local_error = NULL;
393     const char *aio;
394 
395     if (bdrv_flags) {
396         if (qemu_opt_get_bool(opts, "copy-on-read", false)) {
397             *bdrv_flags |= BDRV_O_COPY_ON_READ;
398         }
399 
400         if ((aio = qemu_opt_get(opts, "aio")) != NULL) {
401             if (bdrv_parse_aio(aio, bdrv_flags) < 0) {
402                 error_setg(errp, "invalid aio option");
403                 return;
404             }
405         }
406     }
407 
408     /* disk I/O throttling */
409     if (throttling_group) {
410         *throttling_group = qemu_opt_get(opts, "throttling.group");
411     }
412 
413     if (throttle_cfg) {
414         throttle_config_init(throttle_cfg);
415         throttle_cfg->buckets[THROTTLE_BPS_TOTAL].avg =
416             qemu_opt_get_number(opts, "throttling.bps-total", 0);
417         throttle_cfg->buckets[THROTTLE_BPS_READ].avg  =
418             qemu_opt_get_number(opts, "throttling.bps-read", 0);
419         throttle_cfg->buckets[THROTTLE_BPS_WRITE].avg =
420             qemu_opt_get_number(opts, "throttling.bps-write", 0);
421         throttle_cfg->buckets[THROTTLE_OPS_TOTAL].avg =
422             qemu_opt_get_number(opts, "throttling.iops-total", 0);
423         throttle_cfg->buckets[THROTTLE_OPS_READ].avg =
424             qemu_opt_get_number(opts, "throttling.iops-read", 0);
425         throttle_cfg->buckets[THROTTLE_OPS_WRITE].avg =
426             qemu_opt_get_number(opts, "throttling.iops-write", 0);
427 
428         throttle_cfg->buckets[THROTTLE_BPS_TOTAL].max =
429             qemu_opt_get_number(opts, "throttling.bps-total-max", 0);
430         throttle_cfg->buckets[THROTTLE_BPS_READ].max  =
431             qemu_opt_get_number(opts, "throttling.bps-read-max", 0);
432         throttle_cfg->buckets[THROTTLE_BPS_WRITE].max =
433             qemu_opt_get_number(opts, "throttling.bps-write-max", 0);
434         throttle_cfg->buckets[THROTTLE_OPS_TOTAL].max =
435             qemu_opt_get_number(opts, "throttling.iops-total-max", 0);
436         throttle_cfg->buckets[THROTTLE_OPS_READ].max =
437             qemu_opt_get_number(opts, "throttling.iops-read-max", 0);
438         throttle_cfg->buckets[THROTTLE_OPS_WRITE].max =
439             qemu_opt_get_number(opts, "throttling.iops-write-max", 0);
440 
441         throttle_cfg->buckets[THROTTLE_BPS_TOTAL].burst_length =
442             qemu_opt_get_number(opts, "throttling.bps-total-max-length", 1);
443         throttle_cfg->buckets[THROTTLE_BPS_READ].burst_length  =
444             qemu_opt_get_number(opts, "throttling.bps-read-max-length", 1);
445         throttle_cfg->buckets[THROTTLE_BPS_WRITE].burst_length =
446             qemu_opt_get_number(opts, "throttling.bps-write-max-length", 1);
447         throttle_cfg->buckets[THROTTLE_OPS_TOTAL].burst_length =
448             qemu_opt_get_number(opts, "throttling.iops-total-max-length", 1);
449         throttle_cfg->buckets[THROTTLE_OPS_READ].burst_length =
450             qemu_opt_get_number(opts, "throttling.iops-read-max-length", 1);
451         throttle_cfg->buckets[THROTTLE_OPS_WRITE].burst_length =
452             qemu_opt_get_number(opts, "throttling.iops-write-max-length", 1);
453 
454         throttle_cfg->op_size =
455             qemu_opt_get_number(opts, "throttling.iops-size", 0);
456 
457         if (!throttle_is_valid(throttle_cfg, errp)) {
458             return;
459         }
460     }
461 
462     if (detect_zeroes) {
463         *detect_zeroes =
464             qapi_enum_parse(&BlockdevDetectZeroesOptions_lookup,
465                             qemu_opt_get(opts, "detect-zeroes"),
466                             BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF,
467                             &local_error);
468         if (local_error) {
469             error_propagate(errp, local_error);
470             return;
471         }
472     }
473 }
474 
475 /* Takes the ownership of bs_opts */
476 static BlockBackend *blockdev_init(const char *file, QDict *bs_opts,
477                                    Error **errp)
478 {
479     const char *buf;
480     int bdrv_flags = 0;
481     int on_read_error, on_write_error;
482     bool account_invalid, account_failed;
483     bool writethrough, read_only;
484     BlockBackend *blk;
485     BlockDriverState *bs;
486     ThrottleConfig cfg;
487     int snapshot = 0;
488     Error *error = NULL;
489     QemuOpts *opts;
490     QDict *interval_dict = NULL;
491     QList *interval_list = NULL;
492     const char *id;
493     BlockdevDetectZeroesOptions detect_zeroes =
494         BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF;
495     const char *throttling_group = NULL;
496 
497     /* Check common options by copying from bs_opts to opts, all other options
498      * stay in bs_opts for processing by bdrv_open(). */
499     id = qdict_get_try_str(bs_opts, "id");
500     opts = qemu_opts_create(&qemu_common_drive_opts, id, 1, errp);
501     if (!opts) {
502         goto err_no_opts;
503     }
504 
505     if (!qemu_opts_absorb_qdict(opts, bs_opts, errp)) {
506         goto early_err;
507     }
508 
509     if (id) {
510         qdict_del(bs_opts, "id");
511     }
512 
513     /* extract parameters */
514     snapshot = qemu_opt_get_bool(opts, "snapshot", 0);
515 
516     account_invalid = qemu_opt_get_bool(opts, "stats-account-invalid", true);
517     account_failed = qemu_opt_get_bool(opts, "stats-account-failed", true);
518 
519     writethrough = !qemu_opt_get_bool(opts, BDRV_OPT_CACHE_WB, true);
520 
521     id = qemu_opts_id(opts);
522 
523     qdict_extract_subqdict(bs_opts, &interval_dict, "stats-intervals.");
524     qdict_array_split(interval_dict, &interval_list);
525 
526     if (qdict_size(interval_dict) != 0) {
527         error_setg(errp, "Invalid option stats-intervals.%s",
528                    qdict_first(interval_dict)->key);
529         goto early_err;
530     }
531 
532     extract_common_blockdev_options(opts, &bdrv_flags, &throttling_group, &cfg,
533                                     &detect_zeroes, &error);
534     if (error) {
535         error_propagate(errp, error);
536         goto early_err;
537     }
538 
539     if ((buf = qemu_opt_get(opts, "format")) != NULL) {
540         if (is_help_option(buf)) {
541             qemu_printf("Supported formats:");
542             bdrv_iterate_format(bdrv_format_print, NULL, false);
543             qemu_printf("\nSupported formats (read-only):");
544             bdrv_iterate_format(bdrv_format_print, NULL, true);
545             qemu_printf("\n");
546             goto early_err;
547         }
548 
549         if (qdict_haskey(bs_opts, "driver")) {
550             error_setg(errp, "Cannot specify both 'driver' and 'format'");
551             goto early_err;
552         }
553         qdict_put_str(bs_opts, "driver", buf);
554     }
555 
556     on_write_error = BLOCKDEV_ON_ERROR_ENOSPC;
557     if ((buf = qemu_opt_get(opts, "werror")) != NULL) {
558         on_write_error = parse_block_error_action(buf, 0, &error);
559         if (error) {
560             error_propagate(errp, error);
561             goto early_err;
562         }
563     }
564 
565     on_read_error = BLOCKDEV_ON_ERROR_REPORT;
566     if ((buf = qemu_opt_get(opts, "rerror")) != NULL) {
567         on_read_error = parse_block_error_action(buf, 1, &error);
568         if (error) {
569             error_propagate(errp, error);
570             goto early_err;
571         }
572     }
573 
574     if (snapshot) {
575         bdrv_flags |= BDRV_O_SNAPSHOT;
576     }
577 
578     read_only = qemu_opt_get_bool(opts, BDRV_OPT_READ_ONLY, false);
579 
580     /* init */
581     if ((!file || !*file) && !qdict_size(bs_opts)) {
582         BlockBackendRootState *blk_rs;
583 
584         blk = blk_new(qemu_get_aio_context(), 0, BLK_PERM_ALL);
585         blk_rs = blk_get_root_state(blk);
586         blk_rs->open_flags    = bdrv_flags;
587         blk_rs->read_only     = read_only;
588         blk_rs->detect_zeroes = detect_zeroes;
589 
590         qobject_unref(bs_opts);
591     } else {
592         if (file && !*file) {
593             file = NULL;
594         }
595 
596         /* bdrv_open() defaults to the values in bdrv_flags (for compatibility
597          * with other callers) rather than what we want as the real defaults.
598          * Apply the defaults here instead. */
599         qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_DIRECT, "off");
600         qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_NO_FLUSH, "off");
601         qdict_set_default_str(bs_opts, BDRV_OPT_READ_ONLY,
602                               read_only ? "on" : "off");
603         qdict_set_default_str(bs_opts, BDRV_OPT_AUTO_READ_ONLY, "on");
604         assert((bdrv_flags & BDRV_O_CACHE_MASK) == 0);
605 
606         if (runstate_check(RUN_STATE_INMIGRATE)) {
607             bdrv_flags |= BDRV_O_INACTIVE;
608         }
609 
610         blk = blk_new_open(file, NULL, bs_opts, bdrv_flags, errp);
611         if (!blk) {
612             goto err_no_bs_opts;
613         }
614         bs = blk_bs(blk);
615 
616         bs->detect_zeroes = detect_zeroes;
617 
618         block_acct_setup(blk_get_stats(blk), account_invalid, account_failed);
619 
620         if (!parse_stats_intervals(blk_get_stats(blk), interval_list, errp)) {
621             blk_unref(blk);
622             blk = NULL;
623             goto err_no_bs_opts;
624         }
625     }
626 
627     /* disk I/O throttling */
628     if (throttle_enabled(&cfg)) {
629         if (!throttling_group) {
630             throttling_group = id;
631         }
632         blk_io_limits_enable(blk, throttling_group);
633         blk_set_io_limits(blk, &cfg);
634     }
635 
636     blk_set_enable_write_cache(blk, !writethrough);
637     blk_set_on_error(blk, on_read_error, on_write_error);
638 
639     if (!monitor_add_blk(blk, id, errp)) {
640         blk_unref(blk);
641         blk = NULL;
642         goto err_no_bs_opts;
643     }
644 
645 err_no_bs_opts:
646     qemu_opts_del(opts);
647     qobject_unref(interval_dict);
648     qobject_unref(interval_list);
649     return blk;
650 
651 early_err:
652     qemu_opts_del(opts);
653     qobject_unref(interval_dict);
654     qobject_unref(interval_list);
655 err_no_opts:
656     qobject_unref(bs_opts);
657     return NULL;
658 }
659 
660 /* Takes the ownership of bs_opts */
661 BlockDriverState *bds_tree_init(QDict *bs_opts, Error **errp)
662 {
663     int bdrv_flags = 0;
664 
665     /* bdrv_open() defaults to the values in bdrv_flags (for compatibility
666      * with other callers) rather than what we want as the real defaults.
667      * Apply the defaults here instead. */
668     qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_DIRECT, "off");
669     qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_NO_FLUSH, "off");
670     qdict_set_default_str(bs_opts, BDRV_OPT_READ_ONLY, "off");
671 
672     if (runstate_check(RUN_STATE_INMIGRATE)) {
673         bdrv_flags |= BDRV_O_INACTIVE;
674     }
675 
676     return bdrv_open(NULL, NULL, bs_opts, bdrv_flags, errp);
677 }
678 
679 void blockdev_close_all_bdrv_states(void)
680 {
681     BlockDriverState *bs, *next_bs;
682 
683     QTAILQ_FOREACH_SAFE(bs, &monitor_bdrv_states, monitor_list, next_bs) {
684         AioContext *ctx = bdrv_get_aio_context(bs);
685 
686         aio_context_acquire(ctx);
687         bdrv_unref(bs);
688         aio_context_release(ctx);
689     }
690 }
691 
692 /* Iterates over the list of monitor-owned BlockDriverStates */
693 BlockDriverState *bdrv_next_monitor_owned(BlockDriverState *bs)
694 {
695     return bs ? QTAILQ_NEXT(bs, monitor_list)
696               : QTAILQ_FIRST(&monitor_bdrv_states);
697 }
698 
699 static bool qemu_opt_rename(QemuOpts *opts, const char *from, const char *to,
700                             Error **errp)
701 {
702     const char *value;
703 
704     value = qemu_opt_get(opts, from);
705     if (value) {
706         if (qemu_opt_find(opts, to)) {
707             error_setg(errp, "'%s' and its alias '%s' can't be used at the "
708                        "same time", to, from);
709             return false;
710         }
711     }
712 
713     /* rename all items in opts */
714     while ((value = qemu_opt_get(opts, from))) {
715         qemu_opt_set(opts, to, value, &error_abort);
716         qemu_opt_unset(opts, from);
717     }
718     return true;
719 }
720 
721 QemuOptsList qemu_legacy_drive_opts = {
722     .name = "drive",
723     .head = QTAILQ_HEAD_INITIALIZER(qemu_legacy_drive_opts.head),
724     .desc = {
725         {
726             .name = "bus",
727             .type = QEMU_OPT_NUMBER,
728             .help = "bus number",
729         },{
730             .name = "unit",
731             .type = QEMU_OPT_NUMBER,
732             .help = "unit number (i.e. lun for scsi)",
733         },{
734             .name = "index",
735             .type = QEMU_OPT_NUMBER,
736             .help = "index number",
737         },{
738             .name = "media",
739             .type = QEMU_OPT_STRING,
740             .help = "media type (disk, cdrom)",
741         },{
742             .name = "if",
743             .type = QEMU_OPT_STRING,
744             .help = "interface (ide, scsi, sd, mtd, floppy, pflash, virtio)",
745         },{
746             .name = "file",
747             .type = QEMU_OPT_STRING,
748             .help = "file name",
749         },
750 
751         /* Options that are passed on, but have special semantics with -drive */
752         {
753             .name = BDRV_OPT_READ_ONLY,
754             .type = QEMU_OPT_BOOL,
755             .help = "open drive file as read-only",
756         },{
757             .name = "rerror",
758             .type = QEMU_OPT_STRING,
759             .help = "read error action",
760         },{
761             .name = "werror",
762             .type = QEMU_OPT_STRING,
763             .help = "write error action",
764         },{
765             .name = "copy-on-read",
766             .type = QEMU_OPT_BOOL,
767             .help = "copy read data from backing file into image file",
768         },
769 
770         { /* end of list */ }
771     },
772 };
773 
774 DriveInfo *drive_new(QemuOpts *all_opts, BlockInterfaceType block_default_type,
775                      Error **errp)
776 {
777     const char *value;
778     BlockBackend *blk;
779     DriveInfo *dinfo = NULL;
780     QDict *bs_opts;
781     QemuOpts *legacy_opts;
782     DriveMediaType media = MEDIA_DISK;
783     BlockInterfaceType type;
784     int max_devs, bus_id, unit_id, index;
785     const char *werror, *rerror;
786     bool read_only = false;
787     bool copy_on_read;
788     const char *filename;
789     int i;
790 
791     /* Change legacy command line options into QMP ones */
792     static const struct {
793         const char *from;
794         const char *to;
795     } opt_renames[] = {
796         { "iops",           "throttling.iops-total" },
797         { "iops_rd",        "throttling.iops-read" },
798         { "iops_wr",        "throttling.iops-write" },
799 
800         { "bps",            "throttling.bps-total" },
801         { "bps_rd",         "throttling.bps-read" },
802         { "bps_wr",         "throttling.bps-write" },
803 
804         { "iops_max",       "throttling.iops-total-max" },
805         { "iops_rd_max",    "throttling.iops-read-max" },
806         { "iops_wr_max",    "throttling.iops-write-max" },
807 
808         { "bps_max",        "throttling.bps-total-max" },
809         { "bps_rd_max",     "throttling.bps-read-max" },
810         { "bps_wr_max",     "throttling.bps-write-max" },
811 
812         { "iops_size",      "throttling.iops-size" },
813 
814         { "group",          "throttling.group" },
815 
816         { "readonly",       BDRV_OPT_READ_ONLY },
817     };
818 
819     for (i = 0; i < ARRAY_SIZE(opt_renames); i++) {
820         if (!qemu_opt_rename(all_opts, opt_renames[i].from,
821                              opt_renames[i].to, errp)) {
822             return NULL;
823         }
824     }
825 
826     value = qemu_opt_get(all_opts, "cache");
827     if (value) {
828         int flags = 0;
829         bool writethrough;
830 
831         if (bdrv_parse_cache_mode(value, &flags, &writethrough) != 0) {
832             error_setg(errp, "invalid cache option");
833             return NULL;
834         }
835 
836         /* Specific options take precedence */
837         if (!qemu_opt_get(all_opts, BDRV_OPT_CACHE_WB)) {
838             qemu_opt_set_bool(all_opts, BDRV_OPT_CACHE_WB,
839                               !writethrough, &error_abort);
840         }
841         if (!qemu_opt_get(all_opts, BDRV_OPT_CACHE_DIRECT)) {
842             qemu_opt_set_bool(all_opts, BDRV_OPT_CACHE_DIRECT,
843                               !!(flags & BDRV_O_NOCACHE), &error_abort);
844         }
845         if (!qemu_opt_get(all_opts, BDRV_OPT_CACHE_NO_FLUSH)) {
846             qemu_opt_set_bool(all_opts, BDRV_OPT_CACHE_NO_FLUSH,
847                               !!(flags & BDRV_O_NO_FLUSH), &error_abort);
848         }
849         qemu_opt_unset(all_opts, "cache");
850     }
851 
852     /* Get a QDict for processing the options */
853     bs_opts = qdict_new();
854     qemu_opts_to_qdict(all_opts, bs_opts);
855 
856     legacy_opts = qemu_opts_create(&qemu_legacy_drive_opts, NULL, 0,
857                                    &error_abort);
858     if (!qemu_opts_absorb_qdict(legacy_opts, bs_opts, errp)) {
859         goto fail;
860     }
861 
862     /* Media type */
863     value = qemu_opt_get(legacy_opts, "media");
864     if (value) {
865         if (!strcmp(value, "disk")) {
866             media = MEDIA_DISK;
867         } else if (!strcmp(value, "cdrom")) {
868             media = MEDIA_CDROM;
869             read_only = true;
870         } else {
871             error_setg(errp, "'%s' invalid media", value);
872             goto fail;
873         }
874     }
875 
876     /* copy-on-read is disabled with a warning for read-only devices */
877     read_only |= qemu_opt_get_bool(legacy_opts, BDRV_OPT_READ_ONLY, false);
878     copy_on_read = qemu_opt_get_bool(legacy_opts, "copy-on-read", false);
879 
880     if (read_only && copy_on_read) {
881         warn_report("disabling copy-on-read on read-only drive");
882         copy_on_read = false;
883     }
884 
885     qdict_put_str(bs_opts, BDRV_OPT_READ_ONLY, read_only ? "on" : "off");
886     qdict_put_str(bs_opts, "copy-on-read", copy_on_read ? "on" : "off");
887 
888     /* Controller type */
889     value = qemu_opt_get(legacy_opts, "if");
890     if (value) {
891         for (type = 0;
892              type < IF_COUNT && strcmp(value, if_name[type]);
893              type++) {
894         }
895         if (type == IF_COUNT) {
896             error_setg(errp, "unsupported bus type '%s'", value);
897             goto fail;
898         }
899     } else {
900         type = block_default_type;
901     }
902 
903     /* Device address specified by bus/unit or index.
904      * If none was specified, try to find the first free one. */
905     bus_id  = qemu_opt_get_number(legacy_opts, "bus", 0);
906     unit_id = qemu_opt_get_number(legacy_opts, "unit", -1);
907     index   = qemu_opt_get_number(legacy_opts, "index", -1);
908 
909     max_devs = if_max_devs[type];
910 
911     if (index != -1) {
912         if (bus_id != 0 || unit_id != -1) {
913             error_setg(errp, "index cannot be used with bus and unit");
914             goto fail;
915         }
916         bus_id = drive_index_to_bus_id(type, index);
917         unit_id = drive_index_to_unit_id(type, index);
918     }
919 
920     if (unit_id == -1) {
921        unit_id = 0;
922        while (drive_get(type, bus_id, unit_id) != NULL) {
923            unit_id++;
924            if (max_devs && unit_id >= max_devs) {
925                unit_id -= max_devs;
926                bus_id++;
927            }
928        }
929     }
930 
931     if (max_devs && unit_id >= max_devs) {
932         error_setg(errp, "unit %d too big (max is %d)", unit_id, max_devs - 1);
933         goto fail;
934     }
935 
936     if (drive_get(type, bus_id, unit_id) != NULL) {
937         error_setg(errp, "drive with bus=%d, unit=%d (index=%d) exists",
938                    bus_id, unit_id, index);
939         goto fail;
940     }
941 
942     /* no id supplied -> create one */
943     if (qemu_opts_id(all_opts) == NULL) {
944         char *new_id;
945         const char *mediastr = "";
946         if (type == IF_IDE || type == IF_SCSI) {
947             mediastr = (media == MEDIA_CDROM) ? "-cd" : "-hd";
948         }
949         if (max_devs) {
950             new_id = g_strdup_printf("%s%i%s%i", if_name[type], bus_id,
951                                      mediastr, unit_id);
952         } else {
953             new_id = g_strdup_printf("%s%s%i", if_name[type],
954                                      mediastr, unit_id);
955         }
956         qdict_put_str(bs_opts, "id", new_id);
957         g_free(new_id);
958     }
959 
960     /* Add virtio block device */
961     if (type == IF_VIRTIO) {
962         QemuOpts *devopts;
963         devopts = qemu_opts_create(qemu_find_opts("device"), NULL, 0,
964                                    &error_abort);
965         qemu_opt_set(devopts, "driver", "virtio-blk", &error_abort);
966         qemu_opt_set(devopts, "drive", qdict_get_str(bs_opts, "id"),
967                      &error_abort);
968     }
969 
970     filename = qemu_opt_get(legacy_opts, "file");
971 
972     /* Check werror/rerror compatibility with if=... */
973     werror = qemu_opt_get(legacy_opts, "werror");
974     if (werror != NULL) {
975         if (type != IF_IDE && type != IF_SCSI && type != IF_VIRTIO &&
976             type != IF_NONE) {
977             error_setg(errp, "werror is not supported by this bus type");
978             goto fail;
979         }
980         qdict_put_str(bs_opts, "werror", werror);
981     }
982 
983     rerror = qemu_opt_get(legacy_opts, "rerror");
984     if (rerror != NULL) {
985         if (type != IF_IDE && type != IF_VIRTIO && type != IF_SCSI &&
986             type != IF_NONE) {
987             error_setg(errp, "rerror is not supported by this bus type");
988             goto fail;
989         }
990         qdict_put_str(bs_opts, "rerror", rerror);
991     }
992 
993     /* Actual block device init: Functionality shared with blockdev-add */
994     blk = blockdev_init(filename, bs_opts, errp);
995     bs_opts = NULL;
996     if (!blk) {
997         goto fail;
998     }
999 
1000     /* Create legacy DriveInfo */
1001     dinfo = g_malloc0(sizeof(*dinfo));
1002     dinfo->opts = all_opts;
1003 
1004     dinfo->type = type;
1005     dinfo->bus = bus_id;
1006     dinfo->unit = unit_id;
1007 
1008     blk_set_legacy_dinfo(blk, dinfo);
1009 
1010     switch(type) {
1011     case IF_IDE:
1012     case IF_SCSI:
1013     case IF_XEN:
1014     case IF_NONE:
1015         dinfo->media_cd = media == MEDIA_CDROM;
1016         break;
1017     default:
1018         break;
1019     }
1020 
1021 fail:
1022     qemu_opts_del(legacy_opts);
1023     qobject_unref(bs_opts);
1024     return dinfo;
1025 }
1026 
1027 static BlockDriverState *qmp_get_root_bs(const char *name, Error **errp)
1028 {
1029     BlockDriverState *bs;
1030 
1031     bs = bdrv_lookup_bs(name, name, errp);
1032     if (bs == NULL) {
1033         return NULL;
1034     }
1035 
1036     if (!bdrv_is_root_node(bs)) {
1037         error_setg(errp, "Need a root block node");
1038         return NULL;
1039     }
1040 
1041     if (!bdrv_is_inserted(bs)) {
1042         error_setg(errp, "Device has no medium");
1043         return NULL;
1044     }
1045 
1046     return bs;
1047 }
1048 
1049 static void blockdev_do_action(TransactionAction *action, Error **errp)
1050 {
1051     TransactionActionList list;
1052 
1053     list.value = action;
1054     list.next = NULL;
1055     qmp_transaction(&list, false, NULL, errp);
1056 }
1057 
1058 void qmp_blockdev_snapshot_sync(bool has_device, const char *device,
1059                                 bool has_node_name, const char *node_name,
1060                                 const char *snapshot_file,
1061                                 bool has_snapshot_node_name,
1062                                 const char *snapshot_node_name,
1063                                 bool has_format, const char *format,
1064                                 bool has_mode, NewImageMode mode, Error **errp)
1065 {
1066     BlockdevSnapshotSync snapshot = {
1067         .has_device = has_device,
1068         .device = (char *) device,
1069         .has_node_name = has_node_name,
1070         .node_name = (char *) node_name,
1071         .snapshot_file = (char *) snapshot_file,
1072         .has_snapshot_node_name = has_snapshot_node_name,
1073         .snapshot_node_name = (char *) snapshot_node_name,
1074         .has_format = has_format,
1075         .format = (char *) format,
1076         .has_mode = has_mode,
1077         .mode = mode,
1078     };
1079     TransactionAction action = {
1080         .type = TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC,
1081         .u.blockdev_snapshot_sync.data = &snapshot,
1082     };
1083     blockdev_do_action(&action, errp);
1084 }
1085 
1086 void qmp_blockdev_snapshot(const char *node, const char *overlay,
1087                            Error **errp)
1088 {
1089     BlockdevSnapshot snapshot_data = {
1090         .node = (char *) node,
1091         .overlay = (char *) overlay
1092     };
1093     TransactionAction action = {
1094         .type = TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT,
1095         .u.blockdev_snapshot.data = &snapshot_data,
1096     };
1097     blockdev_do_action(&action, errp);
1098 }
1099 
1100 void qmp_blockdev_snapshot_internal_sync(const char *device,
1101                                          const char *name,
1102                                          Error **errp)
1103 {
1104     BlockdevSnapshotInternal snapshot = {
1105         .device = (char *) device,
1106         .name = (char *) name
1107     };
1108     TransactionAction action = {
1109         .type = TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC,
1110         .u.blockdev_snapshot_internal_sync.data = &snapshot,
1111     };
1112     blockdev_do_action(&action, errp);
1113 }
1114 
1115 SnapshotInfo *qmp_blockdev_snapshot_delete_internal_sync(const char *device,
1116                                                          bool has_id,
1117                                                          const char *id,
1118                                                          bool has_name,
1119                                                          const char *name,
1120                                                          Error **errp)
1121 {
1122     BlockDriverState *bs;
1123     AioContext *aio_context;
1124     QEMUSnapshotInfo sn;
1125     Error *local_err = NULL;
1126     SnapshotInfo *info = NULL;
1127     int ret;
1128 
1129     bs = qmp_get_root_bs(device, errp);
1130     if (!bs) {
1131         return NULL;
1132     }
1133     aio_context = bdrv_get_aio_context(bs);
1134     aio_context_acquire(aio_context);
1135 
1136     if (!has_id) {
1137         id = NULL;
1138     }
1139 
1140     if (!has_name) {
1141         name = NULL;
1142     }
1143 
1144     if (!id && !name) {
1145         error_setg(errp, "Name or id must be provided");
1146         goto out_aio_context;
1147     }
1148 
1149     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_INTERNAL_SNAPSHOT_DELETE, errp)) {
1150         goto out_aio_context;
1151     }
1152 
1153     ret = bdrv_snapshot_find_by_id_and_name(bs, id, name, &sn, &local_err);
1154     if (local_err) {
1155         error_propagate(errp, local_err);
1156         goto out_aio_context;
1157     }
1158     if (!ret) {
1159         error_setg(errp,
1160                    "Snapshot with id '%s' and name '%s' does not exist on "
1161                    "device '%s'",
1162                    STR_OR_NULL(id), STR_OR_NULL(name), device);
1163         goto out_aio_context;
1164     }
1165 
1166     bdrv_snapshot_delete(bs, id, name, &local_err);
1167     if (local_err) {
1168         error_propagate(errp, local_err);
1169         goto out_aio_context;
1170     }
1171 
1172     aio_context_release(aio_context);
1173 
1174     info = g_new0(SnapshotInfo, 1);
1175     info->id = g_strdup(sn.id_str);
1176     info->name = g_strdup(sn.name);
1177     info->date_nsec = sn.date_nsec;
1178     info->date_sec = sn.date_sec;
1179     info->vm_state_size = sn.vm_state_size;
1180     info->vm_clock_nsec = sn.vm_clock_nsec % 1000000000;
1181     info->vm_clock_sec = sn.vm_clock_nsec / 1000000000;
1182     if (sn.icount != -1ULL) {
1183         info->icount = sn.icount;
1184         info->has_icount = true;
1185     }
1186 
1187     return info;
1188 
1189 out_aio_context:
1190     aio_context_release(aio_context);
1191     return NULL;
1192 }
1193 
1194 /* New and old BlockDriverState structs for atomic group operations */
1195 
1196 typedef struct BlkActionState BlkActionState;
1197 
1198 /**
1199  * BlkActionOps:
1200  * Table of operations that define an Action.
1201  *
1202  * @instance_size: Size of state struct, in bytes.
1203  * @prepare: Prepare the work, must NOT be NULL.
1204  * @commit: Commit the changes, can be NULL.
1205  * @abort: Abort the changes on fail, can be NULL.
1206  * @clean: Clean up resources after all transaction actions have called
1207  *         commit() or abort(). Can be NULL.
1208  *
1209  * Only prepare() may fail. In a single transaction, only one of commit() or
1210  * abort() will be called. clean() will always be called if it is present.
1211  */
1212 typedef struct BlkActionOps {
1213     size_t instance_size;
1214     void (*prepare)(BlkActionState *common, Error **errp);
1215     void (*commit)(BlkActionState *common);
1216     void (*abort)(BlkActionState *common);
1217     void (*clean)(BlkActionState *common);
1218 } BlkActionOps;
1219 
1220 /**
1221  * BlkActionState:
1222  * Describes one Action's state within a Transaction.
1223  *
1224  * @action: QAPI-defined enum identifying which Action to perform.
1225  * @ops: Table of ActionOps this Action can perform.
1226  * @block_job_txn: Transaction which this action belongs to.
1227  * @entry: List membership for all Actions in this Transaction.
1228  *
1229  * This structure must be arranged as first member in a subclassed type,
1230  * assuming that the compiler will also arrange it to the same offsets as the
1231  * base class.
1232  */
1233 struct BlkActionState {
1234     TransactionAction *action;
1235     const BlkActionOps *ops;
1236     JobTxn *block_job_txn;
1237     TransactionProperties *txn_props;
1238     QTAILQ_ENTRY(BlkActionState) entry;
1239 };
1240 
1241 /* internal snapshot private data */
1242 typedef struct InternalSnapshotState {
1243     BlkActionState common;
1244     BlockDriverState *bs;
1245     QEMUSnapshotInfo sn;
1246     bool created;
1247 } InternalSnapshotState;
1248 
1249 
1250 static int action_check_completion_mode(BlkActionState *s, Error **errp)
1251 {
1252     if (s->txn_props->completion_mode != ACTION_COMPLETION_MODE_INDIVIDUAL) {
1253         error_setg(errp,
1254                    "Action '%s' does not support Transaction property "
1255                    "completion-mode = %s",
1256                    TransactionActionKind_str(s->action->type),
1257                    ActionCompletionMode_str(s->txn_props->completion_mode));
1258         return -1;
1259     }
1260     return 0;
1261 }
1262 
1263 static void internal_snapshot_prepare(BlkActionState *common,
1264                                       Error **errp)
1265 {
1266     Error *local_err = NULL;
1267     const char *device;
1268     const char *name;
1269     BlockDriverState *bs;
1270     QEMUSnapshotInfo old_sn, *sn;
1271     bool ret;
1272     qemu_timeval tv;
1273     BlockdevSnapshotInternal *internal;
1274     InternalSnapshotState *state;
1275     AioContext *aio_context;
1276     int ret1;
1277 
1278     g_assert(common->action->type ==
1279              TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC);
1280     internal = common->action->u.blockdev_snapshot_internal_sync.data;
1281     state = DO_UPCAST(InternalSnapshotState, common, common);
1282 
1283     /* 1. parse input */
1284     device = internal->device;
1285     name = internal->name;
1286 
1287     /* 2. check for validation */
1288     if (action_check_completion_mode(common, errp) < 0) {
1289         return;
1290     }
1291 
1292     bs = qmp_get_root_bs(device, errp);
1293     if (!bs) {
1294         return;
1295     }
1296 
1297     aio_context = bdrv_get_aio_context(bs);
1298     aio_context_acquire(aio_context);
1299 
1300     state->bs = bs;
1301 
1302     /* Paired with .clean() */
1303     bdrv_drained_begin(bs);
1304 
1305     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_INTERNAL_SNAPSHOT, errp)) {
1306         goto out;
1307     }
1308 
1309     if (bdrv_is_read_only(bs)) {
1310         error_setg(errp, "Device '%s' is read only", device);
1311         goto out;
1312     }
1313 
1314     if (!bdrv_can_snapshot(bs)) {
1315         error_setg(errp, "Block format '%s' used by device '%s' "
1316                    "does not support internal snapshots",
1317                    bs->drv->format_name, device);
1318         goto out;
1319     }
1320 
1321     if (!strlen(name)) {
1322         error_setg(errp, "Name is empty");
1323         goto out;
1324     }
1325 
1326     /* check whether a snapshot with name exist */
1327     ret = bdrv_snapshot_find_by_id_and_name(bs, NULL, name, &old_sn,
1328                                             &local_err);
1329     if (local_err) {
1330         error_propagate(errp, local_err);
1331         goto out;
1332     } else if (ret) {
1333         error_setg(errp,
1334                    "Snapshot with name '%s' already exists on device '%s'",
1335                    name, device);
1336         goto out;
1337     }
1338 
1339     /* 3. take the snapshot */
1340     sn = &state->sn;
1341     pstrcpy(sn->name, sizeof(sn->name), name);
1342     qemu_gettimeofday(&tv);
1343     sn->date_sec = tv.tv_sec;
1344     sn->date_nsec = tv.tv_usec * 1000;
1345     sn->vm_clock_nsec = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL);
1346     if (replay_mode != REPLAY_MODE_NONE) {
1347         sn->icount = replay_get_current_icount();
1348     } else {
1349         sn->icount = -1ULL;
1350     }
1351 
1352     ret1 = bdrv_snapshot_create(bs, sn);
1353     if (ret1 < 0) {
1354         error_setg_errno(errp, -ret1,
1355                          "Failed to create snapshot '%s' on device '%s'",
1356                          name, device);
1357         goto out;
1358     }
1359 
1360     /* 4. succeed, mark a snapshot is created */
1361     state->created = true;
1362 
1363 out:
1364     aio_context_release(aio_context);
1365 }
1366 
1367 static void internal_snapshot_abort(BlkActionState *common)
1368 {
1369     InternalSnapshotState *state =
1370                              DO_UPCAST(InternalSnapshotState, common, common);
1371     BlockDriverState *bs = state->bs;
1372     QEMUSnapshotInfo *sn = &state->sn;
1373     AioContext *aio_context;
1374     Error *local_error = NULL;
1375 
1376     if (!state->created) {
1377         return;
1378     }
1379 
1380     aio_context = bdrv_get_aio_context(state->bs);
1381     aio_context_acquire(aio_context);
1382 
1383     if (bdrv_snapshot_delete(bs, sn->id_str, sn->name, &local_error) < 0) {
1384         error_reportf_err(local_error,
1385                           "Failed to delete snapshot with id '%s' and "
1386                           "name '%s' on device '%s' in abort: ",
1387                           sn->id_str, sn->name,
1388                           bdrv_get_device_name(bs));
1389     }
1390 
1391     aio_context_release(aio_context);
1392 }
1393 
1394 static void internal_snapshot_clean(BlkActionState *common)
1395 {
1396     InternalSnapshotState *state = DO_UPCAST(InternalSnapshotState,
1397                                              common, common);
1398     AioContext *aio_context;
1399 
1400     if (!state->bs) {
1401         return;
1402     }
1403 
1404     aio_context = bdrv_get_aio_context(state->bs);
1405     aio_context_acquire(aio_context);
1406 
1407     bdrv_drained_end(state->bs);
1408 
1409     aio_context_release(aio_context);
1410 }
1411 
1412 /* external snapshot private data */
1413 typedef struct ExternalSnapshotState {
1414     BlkActionState common;
1415     BlockDriverState *old_bs;
1416     BlockDriverState *new_bs;
1417     bool overlay_appended;
1418 } ExternalSnapshotState;
1419 
1420 static void external_snapshot_prepare(BlkActionState *common,
1421                                       Error **errp)
1422 {
1423     int ret;
1424     int flags = 0;
1425     QDict *options = NULL;
1426     Error *local_err = NULL;
1427     /* Device and node name of the image to generate the snapshot from */
1428     const char *device;
1429     const char *node_name;
1430     /* Reference to the new image (for 'blockdev-snapshot') */
1431     const char *snapshot_ref;
1432     /* File name of the new image (for 'blockdev-snapshot-sync') */
1433     const char *new_image_file;
1434     ExternalSnapshotState *state =
1435                              DO_UPCAST(ExternalSnapshotState, common, common);
1436     TransactionAction *action = common->action;
1437     AioContext *aio_context;
1438     uint64_t perm, shared;
1439 
1440     /* 'blockdev-snapshot' and 'blockdev-snapshot-sync' have similar
1441      * purpose but a different set of parameters */
1442     switch (action->type) {
1443     case TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT:
1444         {
1445             BlockdevSnapshot *s = action->u.blockdev_snapshot.data;
1446             device = s->node;
1447             node_name = s->node;
1448             new_image_file = NULL;
1449             snapshot_ref = s->overlay;
1450         }
1451         break;
1452     case TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC:
1453         {
1454             BlockdevSnapshotSync *s = action->u.blockdev_snapshot_sync.data;
1455             device = s->has_device ? s->device : NULL;
1456             node_name = s->has_node_name ? s->node_name : NULL;
1457             new_image_file = s->snapshot_file;
1458             snapshot_ref = NULL;
1459         }
1460         break;
1461     default:
1462         g_assert_not_reached();
1463     }
1464 
1465     /* start processing */
1466     if (action_check_completion_mode(common, errp) < 0) {
1467         return;
1468     }
1469 
1470     state->old_bs = bdrv_lookup_bs(device, node_name, errp);
1471     if (!state->old_bs) {
1472         return;
1473     }
1474 
1475     aio_context = bdrv_get_aio_context(state->old_bs);
1476     aio_context_acquire(aio_context);
1477 
1478     /* Paired with .clean() */
1479     bdrv_drained_begin(state->old_bs);
1480 
1481     if (!bdrv_is_inserted(state->old_bs)) {
1482         error_setg(errp, QERR_DEVICE_HAS_NO_MEDIUM, device);
1483         goto out;
1484     }
1485 
1486     if (bdrv_op_is_blocked(state->old_bs,
1487                            BLOCK_OP_TYPE_EXTERNAL_SNAPSHOT, errp)) {
1488         goto out;
1489     }
1490 
1491     if (!bdrv_is_read_only(state->old_bs)) {
1492         if (bdrv_flush(state->old_bs)) {
1493             error_setg(errp, QERR_IO_ERROR);
1494             goto out;
1495         }
1496     }
1497 
1498     if (action->type == TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC) {
1499         BlockdevSnapshotSync *s = action->u.blockdev_snapshot_sync.data;
1500         const char *format = s->has_format ? s->format : "qcow2";
1501         enum NewImageMode mode;
1502         const char *snapshot_node_name =
1503             s->has_snapshot_node_name ? s->snapshot_node_name : NULL;
1504 
1505         if (node_name && !snapshot_node_name) {
1506             error_setg(errp, "New overlay node-name missing");
1507             goto out;
1508         }
1509 
1510         if (snapshot_node_name &&
1511             bdrv_lookup_bs(snapshot_node_name, snapshot_node_name, NULL)) {
1512             error_setg(errp, "New overlay node-name already in use");
1513             goto out;
1514         }
1515 
1516         flags = state->old_bs->open_flags;
1517         flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_COPY_ON_READ);
1518         flags |= BDRV_O_NO_BACKING;
1519 
1520         /* create new image w/backing file */
1521         mode = s->has_mode ? s->mode : NEW_IMAGE_MODE_ABSOLUTE_PATHS;
1522         if (mode != NEW_IMAGE_MODE_EXISTING) {
1523             int64_t size = bdrv_getlength(state->old_bs);
1524             if (size < 0) {
1525                 error_setg_errno(errp, -size, "bdrv_getlength failed");
1526                 goto out;
1527             }
1528             bdrv_refresh_filename(state->old_bs);
1529             bdrv_img_create(new_image_file, format,
1530                             state->old_bs->filename,
1531                             state->old_bs->drv->format_name,
1532                             NULL, size, flags, false, &local_err);
1533             if (local_err) {
1534                 error_propagate(errp, local_err);
1535                 goto out;
1536             }
1537         }
1538 
1539         options = qdict_new();
1540         if (snapshot_node_name) {
1541             qdict_put_str(options, "node-name", snapshot_node_name);
1542         }
1543         qdict_put_str(options, "driver", format);
1544     }
1545 
1546     state->new_bs = bdrv_open(new_image_file, snapshot_ref, options, flags,
1547                               errp);
1548     /* We will manually add the backing_hd field to the bs later */
1549     if (!state->new_bs) {
1550         goto out;
1551     }
1552 
1553     /*
1554      * Allow attaching a backing file to an overlay that's already in use only
1555      * if the parents don't assume that they are already seeing a valid image.
1556      * (Specifically, allow it as a mirror target, which is write-only access.)
1557      */
1558     bdrv_get_cumulative_perm(state->new_bs, &perm, &shared);
1559     if (perm & BLK_PERM_CONSISTENT_READ) {
1560         error_setg(errp, "The overlay is already in use");
1561         goto out;
1562     }
1563 
1564     if (state->new_bs->drv->is_filter) {
1565         error_setg(errp, "Filters cannot be used as overlays");
1566         goto out;
1567     }
1568 
1569     if (bdrv_cow_child(state->new_bs)) {
1570         error_setg(errp, "The overlay already has a backing image");
1571         goto out;
1572     }
1573 
1574     if (!state->new_bs->drv->supports_backing) {
1575         error_setg(errp, "The overlay does not support backing images");
1576         goto out;
1577     }
1578 
1579     ret = bdrv_append(state->new_bs, state->old_bs, errp);
1580     if (ret < 0) {
1581         goto out;
1582     }
1583     state->overlay_appended = true;
1584 
1585 out:
1586     aio_context_release(aio_context);
1587 }
1588 
1589 static void external_snapshot_commit(BlkActionState *common)
1590 {
1591     ExternalSnapshotState *state =
1592                              DO_UPCAST(ExternalSnapshotState, common, common);
1593     AioContext *aio_context;
1594 
1595     aio_context = bdrv_get_aio_context(state->old_bs);
1596     aio_context_acquire(aio_context);
1597 
1598     /* We don't need (or want) to use the transactional
1599      * bdrv_reopen_multiple() across all the entries at once, because we
1600      * don't want to abort all of them if one of them fails the reopen */
1601     if (!qatomic_read(&state->old_bs->copy_on_read)) {
1602         bdrv_reopen_set_read_only(state->old_bs, true, NULL);
1603     }
1604 
1605     aio_context_release(aio_context);
1606 }
1607 
1608 static void external_snapshot_abort(BlkActionState *common)
1609 {
1610     ExternalSnapshotState *state =
1611                              DO_UPCAST(ExternalSnapshotState, common, common);
1612     if (state->new_bs) {
1613         if (state->overlay_appended) {
1614             AioContext *aio_context;
1615             AioContext *tmp_context;
1616             int ret;
1617 
1618             aio_context = bdrv_get_aio_context(state->old_bs);
1619             aio_context_acquire(aio_context);
1620 
1621             bdrv_ref(state->old_bs);   /* we can't let bdrv_set_backind_hd()
1622                                           close state->old_bs; we need it */
1623             bdrv_set_backing_hd(state->new_bs, NULL, &error_abort);
1624 
1625             /*
1626              * The call to bdrv_set_backing_hd() above returns state->old_bs to
1627              * the main AioContext. As we're still going to be using it, return
1628              * it to the AioContext it was before.
1629              */
1630             tmp_context = bdrv_get_aio_context(state->old_bs);
1631             if (aio_context != tmp_context) {
1632                 aio_context_release(aio_context);
1633                 aio_context_acquire(tmp_context);
1634 
1635                 ret = bdrv_try_set_aio_context(state->old_bs,
1636                                                aio_context, NULL);
1637                 assert(ret == 0);
1638 
1639                 aio_context_release(tmp_context);
1640                 aio_context_acquire(aio_context);
1641             }
1642 
1643             bdrv_replace_node(state->new_bs, state->old_bs, &error_abort);
1644             bdrv_unref(state->old_bs); /* bdrv_replace_node() ref'ed old_bs */
1645 
1646             aio_context_release(aio_context);
1647         }
1648     }
1649 }
1650 
1651 static void external_snapshot_clean(BlkActionState *common)
1652 {
1653     ExternalSnapshotState *state =
1654                              DO_UPCAST(ExternalSnapshotState, common, common);
1655     AioContext *aio_context;
1656 
1657     if (!state->old_bs) {
1658         return;
1659     }
1660 
1661     aio_context = bdrv_get_aio_context(state->old_bs);
1662     aio_context_acquire(aio_context);
1663 
1664     bdrv_drained_end(state->old_bs);
1665     bdrv_unref(state->new_bs);
1666 
1667     aio_context_release(aio_context);
1668 }
1669 
1670 typedef struct DriveBackupState {
1671     BlkActionState common;
1672     BlockDriverState *bs;
1673     BlockJob *job;
1674 } DriveBackupState;
1675 
1676 static BlockJob *do_backup_common(BackupCommon *backup,
1677                                   BlockDriverState *bs,
1678                                   BlockDriverState *target_bs,
1679                                   AioContext *aio_context,
1680                                   JobTxn *txn, Error **errp);
1681 
1682 static void drive_backup_prepare(BlkActionState *common, Error **errp)
1683 {
1684     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1685     DriveBackup *backup;
1686     BlockDriverState *bs;
1687     BlockDriverState *target_bs;
1688     BlockDriverState *source = NULL;
1689     AioContext *aio_context;
1690     AioContext *old_context;
1691     QDict *options;
1692     Error *local_err = NULL;
1693     int flags;
1694     int64_t size;
1695     bool set_backing_hd = false;
1696     int ret;
1697 
1698     assert(common->action->type == TRANSACTION_ACTION_KIND_DRIVE_BACKUP);
1699     backup = common->action->u.drive_backup.data;
1700 
1701     if (!backup->has_mode) {
1702         backup->mode = NEW_IMAGE_MODE_ABSOLUTE_PATHS;
1703     }
1704 
1705     bs = bdrv_lookup_bs(backup->device, backup->device, errp);
1706     if (!bs) {
1707         return;
1708     }
1709 
1710     if (!bs->drv) {
1711         error_setg(errp, "Device has no medium");
1712         return;
1713     }
1714 
1715     aio_context = bdrv_get_aio_context(bs);
1716     aio_context_acquire(aio_context);
1717 
1718     /* Paired with .clean() */
1719     bdrv_drained_begin(bs);
1720 
1721     if (!backup->has_format) {
1722         backup->format = backup->mode == NEW_IMAGE_MODE_EXISTING ?
1723                          NULL : (char *) bs->drv->format_name;
1724     }
1725 
1726     /* Early check to avoid creating target */
1727     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_BACKUP_SOURCE, errp)) {
1728         goto out;
1729     }
1730 
1731     flags = bs->open_flags | BDRV_O_RDWR;
1732 
1733     /*
1734      * See if we have a backing HD we can use to create our new image
1735      * on top of.
1736      */
1737     if (backup->sync == MIRROR_SYNC_MODE_TOP) {
1738         /*
1739          * Backup will not replace the source by the target, so none
1740          * of the filters skipped here will be removed (in contrast to
1741          * mirror).  Therefore, we can skip all of them when looking
1742          * for the first COW relationship.
1743          */
1744         source = bdrv_cow_bs(bdrv_skip_filters(bs));
1745         if (!source) {
1746             backup->sync = MIRROR_SYNC_MODE_FULL;
1747         }
1748     }
1749     if (backup->sync == MIRROR_SYNC_MODE_NONE) {
1750         source = bs;
1751         flags |= BDRV_O_NO_BACKING;
1752         set_backing_hd = true;
1753     }
1754 
1755     size = bdrv_getlength(bs);
1756     if (size < 0) {
1757         error_setg_errno(errp, -size, "bdrv_getlength failed");
1758         goto out;
1759     }
1760 
1761     if (backup->mode != NEW_IMAGE_MODE_EXISTING) {
1762         assert(backup->format);
1763         if (source) {
1764             /* Implicit filters should not appear in the filename */
1765             BlockDriverState *explicit_backing =
1766                 bdrv_skip_implicit_filters(source);
1767 
1768             bdrv_refresh_filename(explicit_backing);
1769             bdrv_img_create(backup->target, backup->format,
1770                             explicit_backing->filename,
1771                             explicit_backing->drv->format_name, NULL,
1772                             size, flags, false, &local_err);
1773         } else {
1774             bdrv_img_create(backup->target, backup->format, NULL, NULL, NULL,
1775                             size, flags, false, &local_err);
1776         }
1777     }
1778 
1779     if (local_err) {
1780         error_propagate(errp, local_err);
1781         goto out;
1782     }
1783 
1784     options = qdict_new();
1785     qdict_put_str(options, "discard", "unmap");
1786     qdict_put_str(options, "detect-zeroes", "unmap");
1787     if (backup->format) {
1788         qdict_put_str(options, "driver", backup->format);
1789     }
1790 
1791     target_bs = bdrv_open(backup->target, NULL, options, flags, errp);
1792     if (!target_bs) {
1793         goto out;
1794     }
1795 
1796     /* Honor bdrv_try_set_aio_context() context acquisition requirements. */
1797     old_context = bdrv_get_aio_context(target_bs);
1798     aio_context_release(aio_context);
1799     aio_context_acquire(old_context);
1800 
1801     ret = bdrv_try_set_aio_context(target_bs, aio_context, errp);
1802     if (ret < 0) {
1803         bdrv_unref(target_bs);
1804         aio_context_release(old_context);
1805         return;
1806     }
1807 
1808     aio_context_release(old_context);
1809     aio_context_acquire(aio_context);
1810 
1811     if (set_backing_hd) {
1812         if (bdrv_set_backing_hd(target_bs, source, errp) < 0) {
1813             goto unref;
1814         }
1815     }
1816 
1817     state->bs = bs;
1818 
1819     state->job = do_backup_common(qapi_DriveBackup_base(backup),
1820                                   bs, target_bs, aio_context,
1821                                   common->block_job_txn, errp);
1822 
1823 unref:
1824     bdrv_unref(target_bs);
1825 out:
1826     aio_context_release(aio_context);
1827 }
1828 
1829 static void drive_backup_commit(BlkActionState *common)
1830 {
1831     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1832     AioContext *aio_context;
1833 
1834     aio_context = bdrv_get_aio_context(state->bs);
1835     aio_context_acquire(aio_context);
1836 
1837     assert(state->job);
1838     job_start(&state->job->job);
1839 
1840     aio_context_release(aio_context);
1841 }
1842 
1843 static void drive_backup_abort(BlkActionState *common)
1844 {
1845     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1846 
1847     if (state->job) {
1848         AioContext *aio_context;
1849 
1850         aio_context = bdrv_get_aio_context(state->bs);
1851         aio_context_acquire(aio_context);
1852 
1853         job_cancel_sync(&state->job->job);
1854 
1855         aio_context_release(aio_context);
1856     }
1857 }
1858 
1859 static void drive_backup_clean(BlkActionState *common)
1860 {
1861     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1862     AioContext *aio_context;
1863 
1864     if (!state->bs) {
1865         return;
1866     }
1867 
1868     aio_context = bdrv_get_aio_context(state->bs);
1869     aio_context_acquire(aio_context);
1870 
1871     bdrv_drained_end(state->bs);
1872 
1873     aio_context_release(aio_context);
1874 }
1875 
1876 typedef struct BlockdevBackupState {
1877     BlkActionState common;
1878     BlockDriverState *bs;
1879     BlockJob *job;
1880 } BlockdevBackupState;
1881 
1882 static void blockdev_backup_prepare(BlkActionState *common, Error **errp)
1883 {
1884     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1885     BlockdevBackup *backup;
1886     BlockDriverState *bs;
1887     BlockDriverState *target_bs;
1888     AioContext *aio_context;
1889     AioContext *old_context;
1890     int ret;
1891 
1892     assert(common->action->type == TRANSACTION_ACTION_KIND_BLOCKDEV_BACKUP);
1893     backup = common->action->u.blockdev_backup.data;
1894 
1895     bs = bdrv_lookup_bs(backup->device, backup->device, errp);
1896     if (!bs) {
1897         return;
1898     }
1899 
1900     target_bs = bdrv_lookup_bs(backup->target, backup->target, errp);
1901     if (!target_bs) {
1902         return;
1903     }
1904 
1905     /* Honor bdrv_try_set_aio_context() context acquisition requirements. */
1906     aio_context = bdrv_get_aio_context(bs);
1907     old_context = bdrv_get_aio_context(target_bs);
1908     aio_context_acquire(old_context);
1909 
1910     ret = bdrv_try_set_aio_context(target_bs, aio_context, errp);
1911     if (ret < 0) {
1912         aio_context_release(old_context);
1913         return;
1914     }
1915 
1916     aio_context_release(old_context);
1917     aio_context_acquire(aio_context);
1918     state->bs = bs;
1919 
1920     /* Paired with .clean() */
1921     bdrv_drained_begin(state->bs);
1922 
1923     state->job = do_backup_common(qapi_BlockdevBackup_base(backup),
1924                                   bs, target_bs, aio_context,
1925                                   common->block_job_txn, errp);
1926 
1927     aio_context_release(aio_context);
1928 }
1929 
1930 static void blockdev_backup_commit(BlkActionState *common)
1931 {
1932     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1933     AioContext *aio_context;
1934 
1935     aio_context = bdrv_get_aio_context(state->bs);
1936     aio_context_acquire(aio_context);
1937 
1938     assert(state->job);
1939     job_start(&state->job->job);
1940 
1941     aio_context_release(aio_context);
1942 }
1943 
1944 static void blockdev_backup_abort(BlkActionState *common)
1945 {
1946     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1947 
1948     if (state->job) {
1949         AioContext *aio_context;
1950 
1951         aio_context = bdrv_get_aio_context(state->bs);
1952         aio_context_acquire(aio_context);
1953 
1954         job_cancel_sync(&state->job->job);
1955 
1956         aio_context_release(aio_context);
1957     }
1958 }
1959 
1960 static void blockdev_backup_clean(BlkActionState *common)
1961 {
1962     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1963     AioContext *aio_context;
1964 
1965     if (!state->bs) {
1966         return;
1967     }
1968 
1969     aio_context = bdrv_get_aio_context(state->bs);
1970     aio_context_acquire(aio_context);
1971 
1972     bdrv_drained_end(state->bs);
1973 
1974     aio_context_release(aio_context);
1975 }
1976 
1977 typedef struct BlockDirtyBitmapState {
1978     BlkActionState common;
1979     BdrvDirtyBitmap *bitmap;
1980     BlockDriverState *bs;
1981     HBitmap *backup;
1982     bool prepared;
1983     bool was_enabled;
1984 } BlockDirtyBitmapState;
1985 
1986 static void block_dirty_bitmap_add_prepare(BlkActionState *common,
1987                                            Error **errp)
1988 {
1989     Error *local_err = NULL;
1990     BlockDirtyBitmapAdd *action;
1991     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
1992                                              common, common);
1993 
1994     if (action_check_completion_mode(common, errp) < 0) {
1995         return;
1996     }
1997 
1998     action = common->action->u.block_dirty_bitmap_add.data;
1999     /* AIO context taken and released within qmp_block_dirty_bitmap_add */
2000     qmp_block_dirty_bitmap_add(action->node, action->name,
2001                                action->has_granularity, action->granularity,
2002                                action->has_persistent, action->persistent,
2003                                action->has_disabled, action->disabled,
2004                                &local_err);
2005 
2006     if (!local_err) {
2007         state->prepared = true;
2008     } else {
2009         error_propagate(errp, local_err);
2010     }
2011 }
2012 
2013 static void block_dirty_bitmap_add_abort(BlkActionState *common)
2014 {
2015     BlockDirtyBitmapAdd *action;
2016     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2017                                              common, common);
2018 
2019     action = common->action->u.block_dirty_bitmap_add.data;
2020     /* Should not be able to fail: IF the bitmap was added via .prepare(),
2021      * then the node reference and bitmap name must have been valid.
2022      */
2023     if (state->prepared) {
2024         qmp_block_dirty_bitmap_remove(action->node, action->name, &error_abort);
2025     }
2026 }
2027 
2028 static void block_dirty_bitmap_clear_prepare(BlkActionState *common,
2029                                              Error **errp)
2030 {
2031     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2032                                              common, common);
2033     BlockDirtyBitmap *action;
2034 
2035     if (action_check_completion_mode(common, errp) < 0) {
2036         return;
2037     }
2038 
2039     action = common->action->u.block_dirty_bitmap_clear.data;
2040     state->bitmap = block_dirty_bitmap_lookup(action->node,
2041                                               action->name,
2042                                               &state->bs,
2043                                               errp);
2044     if (!state->bitmap) {
2045         return;
2046     }
2047 
2048     if (bdrv_dirty_bitmap_check(state->bitmap, BDRV_BITMAP_DEFAULT, errp)) {
2049         return;
2050     }
2051 
2052     bdrv_clear_dirty_bitmap(state->bitmap, &state->backup);
2053 }
2054 
2055 static void block_dirty_bitmap_restore(BlkActionState *common)
2056 {
2057     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2058                                              common, common);
2059 
2060     if (state->backup) {
2061         bdrv_restore_dirty_bitmap(state->bitmap, state->backup);
2062     }
2063 }
2064 
2065 static void block_dirty_bitmap_free_backup(BlkActionState *common)
2066 {
2067     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2068                                              common, common);
2069 
2070     hbitmap_free(state->backup);
2071 }
2072 
2073 static void block_dirty_bitmap_enable_prepare(BlkActionState *common,
2074                                               Error **errp)
2075 {
2076     BlockDirtyBitmap *action;
2077     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2078                                              common, common);
2079 
2080     if (action_check_completion_mode(common, errp) < 0) {
2081         return;
2082     }
2083 
2084     action = common->action->u.block_dirty_bitmap_enable.data;
2085     state->bitmap = block_dirty_bitmap_lookup(action->node,
2086                                               action->name,
2087                                               NULL,
2088                                               errp);
2089     if (!state->bitmap) {
2090         return;
2091     }
2092 
2093     if (bdrv_dirty_bitmap_check(state->bitmap, BDRV_BITMAP_ALLOW_RO, errp)) {
2094         return;
2095     }
2096 
2097     state->was_enabled = bdrv_dirty_bitmap_enabled(state->bitmap);
2098     bdrv_enable_dirty_bitmap(state->bitmap);
2099 }
2100 
2101 static void block_dirty_bitmap_enable_abort(BlkActionState *common)
2102 {
2103     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2104                                              common, common);
2105 
2106     if (!state->was_enabled) {
2107         bdrv_disable_dirty_bitmap(state->bitmap);
2108     }
2109 }
2110 
2111 static void block_dirty_bitmap_disable_prepare(BlkActionState *common,
2112                                                Error **errp)
2113 {
2114     BlockDirtyBitmap *action;
2115     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2116                                              common, common);
2117 
2118     if (action_check_completion_mode(common, errp) < 0) {
2119         return;
2120     }
2121 
2122     action = common->action->u.block_dirty_bitmap_disable.data;
2123     state->bitmap = block_dirty_bitmap_lookup(action->node,
2124                                               action->name,
2125                                               NULL,
2126                                               errp);
2127     if (!state->bitmap) {
2128         return;
2129     }
2130 
2131     if (bdrv_dirty_bitmap_check(state->bitmap, BDRV_BITMAP_ALLOW_RO, errp)) {
2132         return;
2133     }
2134 
2135     state->was_enabled = bdrv_dirty_bitmap_enabled(state->bitmap);
2136     bdrv_disable_dirty_bitmap(state->bitmap);
2137 }
2138 
2139 static void block_dirty_bitmap_disable_abort(BlkActionState *common)
2140 {
2141     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2142                                              common, common);
2143 
2144     if (state->was_enabled) {
2145         bdrv_enable_dirty_bitmap(state->bitmap);
2146     }
2147 }
2148 
2149 static void block_dirty_bitmap_merge_prepare(BlkActionState *common,
2150                                              Error **errp)
2151 {
2152     BlockDirtyBitmapMerge *action;
2153     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2154                                              common, common);
2155 
2156     if (action_check_completion_mode(common, errp) < 0) {
2157         return;
2158     }
2159 
2160     action = common->action->u.block_dirty_bitmap_merge.data;
2161 
2162     state->bitmap = block_dirty_bitmap_merge(action->node, action->target,
2163                                              action->bitmaps, &state->backup,
2164                                              errp);
2165 }
2166 
2167 static void block_dirty_bitmap_remove_prepare(BlkActionState *common,
2168                                               Error **errp)
2169 {
2170     BlockDirtyBitmap *action;
2171     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2172                                              common, common);
2173 
2174     if (action_check_completion_mode(common, errp) < 0) {
2175         return;
2176     }
2177 
2178     action = common->action->u.block_dirty_bitmap_remove.data;
2179 
2180     state->bitmap = block_dirty_bitmap_remove(action->node, action->name,
2181                                               false, &state->bs, errp);
2182     if (state->bitmap) {
2183         bdrv_dirty_bitmap_skip_store(state->bitmap, true);
2184         bdrv_dirty_bitmap_set_busy(state->bitmap, true);
2185     }
2186 }
2187 
2188 static void block_dirty_bitmap_remove_abort(BlkActionState *common)
2189 {
2190     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2191                                              common, common);
2192 
2193     if (state->bitmap) {
2194         bdrv_dirty_bitmap_skip_store(state->bitmap, false);
2195         bdrv_dirty_bitmap_set_busy(state->bitmap, false);
2196     }
2197 }
2198 
2199 static void block_dirty_bitmap_remove_commit(BlkActionState *common)
2200 {
2201     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2202                                              common, common);
2203 
2204     bdrv_dirty_bitmap_set_busy(state->bitmap, false);
2205     bdrv_release_dirty_bitmap(state->bitmap);
2206 }
2207 
2208 static void abort_prepare(BlkActionState *common, Error **errp)
2209 {
2210     error_setg(errp, "Transaction aborted using Abort action");
2211 }
2212 
2213 static void abort_commit(BlkActionState *common)
2214 {
2215     g_assert_not_reached(); /* this action never succeeds */
2216 }
2217 
2218 static const BlkActionOps actions[] = {
2219     [TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT] = {
2220         .instance_size = sizeof(ExternalSnapshotState),
2221         .prepare  = external_snapshot_prepare,
2222         .commit   = external_snapshot_commit,
2223         .abort = external_snapshot_abort,
2224         .clean = external_snapshot_clean,
2225     },
2226     [TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC] = {
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_DRIVE_BACKUP] = {
2234         .instance_size = sizeof(DriveBackupState),
2235         .prepare = drive_backup_prepare,
2236         .commit = drive_backup_commit,
2237         .abort = drive_backup_abort,
2238         .clean = drive_backup_clean,
2239     },
2240     [TRANSACTION_ACTION_KIND_BLOCKDEV_BACKUP] = {
2241         .instance_size = sizeof(BlockdevBackupState),
2242         .prepare = blockdev_backup_prepare,
2243         .commit = blockdev_backup_commit,
2244         .abort = blockdev_backup_abort,
2245         .clean = blockdev_backup_clean,
2246     },
2247     [TRANSACTION_ACTION_KIND_ABORT] = {
2248         .instance_size = sizeof(BlkActionState),
2249         .prepare = abort_prepare,
2250         .commit = abort_commit,
2251     },
2252     [TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC] = {
2253         .instance_size = sizeof(InternalSnapshotState),
2254         .prepare  = internal_snapshot_prepare,
2255         .abort = internal_snapshot_abort,
2256         .clean = internal_snapshot_clean,
2257     },
2258     [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_ADD] = {
2259         .instance_size = sizeof(BlockDirtyBitmapState),
2260         .prepare = block_dirty_bitmap_add_prepare,
2261         .abort = block_dirty_bitmap_add_abort,
2262     },
2263     [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_CLEAR] = {
2264         .instance_size = sizeof(BlockDirtyBitmapState),
2265         .prepare = block_dirty_bitmap_clear_prepare,
2266         .commit = block_dirty_bitmap_free_backup,
2267         .abort = block_dirty_bitmap_restore,
2268     },
2269     [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_ENABLE] = {
2270         .instance_size = sizeof(BlockDirtyBitmapState),
2271         .prepare = block_dirty_bitmap_enable_prepare,
2272         .abort = block_dirty_bitmap_enable_abort,
2273     },
2274     [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_DISABLE] = {
2275         .instance_size = sizeof(BlockDirtyBitmapState),
2276         .prepare = block_dirty_bitmap_disable_prepare,
2277         .abort = block_dirty_bitmap_disable_abort,
2278     },
2279     [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_MERGE] = {
2280         .instance_size = sizeof(BlockDirtyBitmapState),
2281         .prepare = block_dirty_bitmap_merge_prepare,
2282         .commit = block_dirty_bitmap_free_backup,
2283         .abort = block_dirty_bitmap_restore,
2284     },
2285     [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_REMOVE] = {
2286         .instance_size = sizeof(BlockDirtyBitmapState),
2287         .prepare = block_dirty_bitmap_remove_prepare,
2288         .commit = block_dirty_bitmap_remove_commit,
2289         .abort = block_dirty_bitmap_remove_abort,
2290     },
2291     /* Where are transactions for MIRROR, COMMIT and STREAM?
2292      * Although these blockjobs use transaction callbacks like the backup job,
2293      * these jobs do not necessarily adhere to transaction semantics.
2294      * These jobs may not fully undo all of their actions on abort, nor do they
2295      * necessarily work in transactions with more than one job in them.
2296      */
2297 };
2298 
2299 /**
2300  * Allocate a TransactionProperties structure if necessary, and fill
2301  * that structure with desired defaults if they are unset.
2302  */
2303 static TransactionProperties *get_transaction_properties(
2304     TransactionProperties *props)
2305 {
2306     if (!props) {
2307         props = g_new0(TransactionProperties, 1);
2308     }
2309 
2310     if (!props->has_completion_mode) {
2311         props->has_completion_mode = true;
2312         props->completion_mode = ACTION_COMPLETION_MODE_INDIVIDUAL;
2313     }
2314 
2315     return props;
2316 }
2317 
2318 /*
2319  * 'Atomic' group operations.  The operations are performed as a set, and if
2320  * any fail then we roll back all operations in the group.
2321  */
2322 void qmp_transaction(TransactionActionList *dev_list,
2323                      bool has_props,
2324                      struct TransactionProperties *props,
2325                      Error **errp)
2326 {
2327     TransactionActionList *dev_entry = dev_list;
2328     JobTxn *block_job_txn = NULL;
2329     BlkActionState *state, *next;
2330     Error *local_err = NULL;
2331 
2332     QTAILQ_HEAD(, BlkActionState) snap_bdrv_states;
2333     QTAILQ_INIT(&snap_bdrv_states);
2334 
2335     /* Does this transaction get canceled as a group on failure?
2336      * If not, we don't really need to make a JobTxn.
2337      */
2338     props = get_transaction_properties(props);
2339     if (props->completion_mode != ACTION_COMPLETION_MODE_INDIVIDUAL) {
2340         block_job_txn = job_txn_new();
2341     }
2342 
2343     /* drain all i/o before any operations */
2344     bdrv_drain_all();
2345 
2346     /* We don't do anything in this loop that commits us to the operations */
2347     while (NULL != dev_entry) {
2348         TransactionAction *dev_info = NULL;
2349         const BlkActionOps *ops;
2350 
2351         dev_info = dev_entry->value;
2352         dev_entry = dev_entry->next;
2353 
2354         assert(dev_info->type < ARRAY_SIZE(actions));
2355 
2356         ops = &actions[dev_info->type];
2357         assert(ops->instance_size > 0);
2358 
2359         state = g_malloc0(ops->instance_size);
2360         state->ops = ops;
2361         state->action = dev_info;
2362         state->block_job_txn = block_job_txn;
2363         state->txn_props = props;
2364         QTAILQ_INSERT_TAIL(&snap_bdrv_states, state, entry);
2365 
2366         state->ops->prepare(state, &local_err);
2367         if (local_err) {
2368             error_propagate(errp, local_err);
2369             goto delete_and_fail;
2370         }
2371     }
2372 
2373     QTAILQ_FOREACH(state, &snap_bdrv_states, entry) {
2374         if (state->ops->commit) {
2375             state->ops->commit(state);
2376         }
2377     }
2378 
2379     /* success */
2380     goto exit;
2381 
2382 delete_and_fail:
2383     /* failure, and it is all-or-none; roll back all operations */
2384     QTAILQ_FOREACH_REVERSE(state, &snap_bdrv_states, entry) {
2385         if (state->ops->abort) {
2386             state->ops->abort(state);
2387         }
2388     }
2389 exit:
2390     QTAILQ_FOREACH_SAFE(state, &snap_bdrv_states, entry, next) {
2391         if (state->ops->clean) {
2392             state->ops->clean(state);
2393         }
2394         g_free(state);
2395     }
2396     if (!has_props) {
2397         qapi_free_TransactionProperties(props);
2398     }
2399     job_txn_unref(block_job_txn);
2400 }
2401 
2402 BlockDirtyBitmapSha256 *qmp_x_debug_block_dirty_bitmap_sha256(const char *node,
2403                                                               const char *name,
2404                                                               Error **errp)
2405 {
2406     BdrvDirtyBitmap *bitmap;
2407     BlockDriverState *bs;
2408     BlockDirtyBitmapSha256 *ret = NULL;
2409     char *sha256;
2410 
2411     bitmap = block_dirty_bitmap_lookup(node, name, &bs, errp);
2412     if (!bitmap || !bs) {
2413         return NULL;
2414     }
2415 
2416     sha256 = bdrv_dirty_bitmap_sha256(bitmap, errp);
2417     if (sha256 == NULL) {
2418         return NULL;
2419     }
2420 
2421     ret = g_new(BlockDirtyBitmapSha256, 1);
2422     ret->sha256 = sha256;
2423 
2424     return ret;
2425 }
2426 
2427 void coroutine_fn qmp_block_resize(bool has_device, const char *device,
2428                                    bool has_node_name, const char *node_name,
2429                                    int64_t size, Error **errp)
2430 {
2431     Error *local_err = NULL;
2432     BlockBackend *blk;
2433     BlockDriverState *bs;
2434     AioContext *old_ctx;
2435 
2436     bs = bdrv_lookup_bs(has_device ? device : NULL,
2437                         has_node_name ? node_name : NULL,
2438                         &local_err);
2439     if (local_err) {
2440         error_propagate(errp, local_err);
2441         return;
2442     }
2443 
2444     if (size < 0) {
2445         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "size", "a >0 size");
2446         return;
2447     }
2448 
2449     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_RESIZE, NULL)) {
2450         error_setg(errp, QERR_DEVICE_IN_USE, device);
2451         return;
2452     }
2453 
2454     blk = blk_new_with_bs(bs, BLK_PERM_RESIZE, BLK_PERM_ALL, errp);
2455     if (!blk) {
2456         return;
2457     }
2458 
2459     bdrv_co_lock(bs);
2460     bdrv_drained_begin(bs);
2461     bdrv_co_unlock(bs);
2462 
2463     old_ctx = bdrv_co_enter(bs);
2464     blk_truncate(blk, size, false, PREALLOC_MODE_OFF, 0, errp);
2465     bdrv_co_leave(bs, old_ctx);
2466 
2467     bdrv_co_lock(bs);
2468     bdrv_drained_end(bs);
2469     blk_unref(blk);
2470     bdrv_co_unlock(bs);
2471 }
2472 
2473 void qmp_block_stream(bool has_job_id, const char *job_id, const char *device,
2474                       bool has_base, const char *base,
2475                       bool has_base_node, const char *base_node,
2476                       bool has_backing_file, const char *backing_file,
2477                       bool has_bottom, const char *bottom,
2478                       bool has_speed, int64_t speed,
2479                       bool has_on_error, BlockdevOnError on_error,
2480                       bool has_filter_node_name, const char *filter_node_name,
2481                       bool has_auto_finalize, bool auto_finalize,
2482                       bool has_auto_dismiss, bool auto_dismiss,
2483                       Error **errp)
2484 {
2485     BlockDriverState *bs, *iter, *iter_end;
2486     BlockDriverState *base_bs = NULL;
2487     BlockDriverState *bottom_bs = NULL;
2488     AioContext *aio_context;
2489     Error *local_err = NULL;
2490     int job_flags = JOB_DEFAULT;
2491 
2492     if (has_base && has_base_node) {
2493         error_setg(errp, "'base' and 'base-node' cannot be specified "
2494                    "at the same time");
2495         return;
2496     }
2497 
2498     if (has_base && has_bottom) {
2499         error_setg(errp, "'base' and 'bottom' cannot be specified "
2500                    "at the same time");
2501         return;
2502     }
2503 
2504     if (has_bottom && has_base_node) {
2505         error_setg(errp, "'bottom' and 'base-node' cannot be specified "
2506                    "at the same time");
2507         return;
2508     }
2509 
2510     if (!has_on_error) {
2511         on_error = BLOCKDEV_ON_ERROR_REPORT;
2512     }
2513 
2514     bs = bdrv_lookup_bs(device, device, errp);
2515     if (!bs) {
2516         return;
2517     }
2518 
2519     aio_context = bdrv_get_aio_context(bs);
2520     aio_context_acquire(aio_context);
2521 
2522     if (has_base) {
2523         base_bs = bdrv_find_backing_image(bs, base);
2524         if (base_bs == NULL) {
2525             error_setg(errp, "Can't find '%s' in the backing chain", base);
2526             goto out;
2527         }
2528         assert(bdrv_get_aio_context(base_bs) == aio_context);
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     }
2544 
2545     if (has_bottom) {
2546         bottom_bs = bdrv_lookup_bs(NULL, bottom, errp);
2547         if (!bottom_bs) {
2548             goto out;
2549         }
2550         if (!bottom_bs->drv) {
2551             error_setg(errp, "Node '%s' is not open", bottom);
2552             goto out;
2553         }
2554         if (bottom_bs->drv->is_filter) {
2555             error_setg(errp, "Node '%s' is a filter, use a non-filter node "
2556                        "as 'bottom'", bottom);
2557             goto out;
2558         }
2559         if (!bdrv_chain_contains(bs, bottom_bs)) {
2560             error_setg(errp, "Node '%s' is not in a chain starting from '%s'",
2561                        bottom, device);
2562             goto out;
2563         }
2564         assert(bdrv_get_aio_context(bottom_bs) == aio_context);
2565     }
2566 
2567     /*
2568      * Check for op blockers in the whole chain between bs and base (or bottom)
2569      */
2570     iter_end = has_bottom ? bdrv_filter_or_cow_bs(bottom_bs) : base_bs;
2571     for (iter = bs; iter && iter != iter_end;
2572          iter = bdrv_filter_or_cow_bs(iter))
2573     {
2574         if (bdrv_op_is_blocked(iter, BLOCK_OP_TYPE_STREAM, errp)) {
2575             goto out;
2576         }
2577     }
2578 
2579     /* if we are streaming the entire chain, the result will have no backing
2580      * file, and specifying one is therefore an error */
2581     if (base_bs == NULL && has_backing_file) {
2582         error_setg(errp, "backing file specified, but streaming the "
2583                          "entire chain");
2584         goto out;
2585     }
2586 
2587     if (has_auto_finalize && !auto_finalize) {
2588         job_flags |= JOB_MANUAL_FINALIZE;
2589     }
2590     if (has_auto_dismiss && !auto_dismiss) {
2591         job_flags |= JOB_MANUAL_DISMISS;
2592     }
2593 
2594     stream_start(has_job_id ? job_id : NULL, bs, base_bs, backing_file,
2595                  bottom_bs, job_flags, has_speed ? speed : 0, on_error,
2596                  filter_node_name, &local_err);
2597     if (local_err) {
2598         error_propagate(errp, local_err);
2599         goto out;
2600     }
2601 
2602     trace_qmp_block_stream(bs);
2603 
2604 out:
2605     aio_context_release(aio_context);
2606 }
2607 
2608 void qmp_block_commit(bool has_job_id, const char *job_id, const char *device,
2609                       bool has_base_node, const char *base_node,
2610                       bool has_base, const char *base,
2611                       bool has_top_node, const char *top_node,
2612                       bool has_top, const char *top,
2613                       bool has_backing_file, const char *backing_file,
2614                       bool has_speed, int64_t speed,
2615                       bool has_on_error, BlockdevOnError on_error,
2616                       bool has_filter_node_name, const char *filter_node_name,
2617                       bool has_auto_finalize, bool auto_finalize,
2618                       bool has_auto_dismiss, bool auto_dismiss,
2619                       Error **errp)
2620 {
2621     BlockDriverState *bs;
2622     BlockDriverState *iter;
2623     BlockDriverState *base_bs, *top_bs;
2624     AioContext *aio_context;
2625     Error *local_err = NULL;
2626     int job_flags = JOB_DEFAULT;
2627     uint64_t top_perm, top_shared;
2628 
2629     if (!has_speed) {
2630         speed = 0;
2631     }
2632     if (!has_on_error) {
2633         on_error = BLOCKDEV_ON_ERROR_REPORT;
2634     }
2635     if (!has_filter_node_name) {
2636         filter_node_name = NULL;
2637     }
2638     if (has_auto_finalize && !auto_finalize) {
2639         job_flags |= JOB_MANUAL_FINALIZE;
2640     }
2641     if (has_auto_dismiss && !auto_dismiss) {
2642         job_flags |= JOB_MANUAL_DISMISS;
2643     }
2644 
2645     /* Important Note:
2646      *  libvirt relies on the DeviceNotFound error class in order to probe for
2647      *  live commit feature versions; for this to work, we must make sure to
2648      *  perform the device lookup before any generic errors that may occur in a
2649      *  scenario in which all optional arguments are omitted. */
2650     bs = qmp_get_root_bs(device, &local_err);
2651     if (!bs) {
2652         bs = bdrv_lookup_bs(device, device, NULL);
2653         if (!bs) {
2654             error_free(local_err);
2655             error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
2656                       "Device '%s' not found", device);
2657         } else {
2658             error_propagate(errp, local_err);
2659         }
2660         return;
2661     }
2662 
2663     aio_context = bdrv_get_aio_context(bs);
2664     aio_context_acquire(aio_context);
2665 
2666     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_COMMIT_SOURCE, errp)) {
2667         goto out;
2668     }
2669 
2670     /* default top_bs is the active layer */
2671     top_bs = bs;
2672 
2673     if (has_top_node && has_top) {
2674         error_setg(errp, "'top-node' and 'top' are mutually exclusive");
2675         goto out;
2676     } else if (has_top_node) {
2677         top_bs = bdrv_lookup_bs(NULL, top_node, errp);
2678         if (top_bs == NULL) {
2679             goto out;
2680         }
2681         if (!bdrv_chain_contains(bs, top_bs)) {
2682             error_setg(errp, "'%s' is not in this backing file chain",
2683                        top_node);
2684             goto out;
2685         }
2686     } else if (has_top && top) {
2687         /* This strcmp() is just a shortcut, there is no need to
2688          * refresh @bs's filename.  If it mismatches,
2689          * bdrv_find_backing_image() will do the refresh and may still
2690          * return @bs. */
2691         if (strcmp(bs->filename, top) != 0) {
2692             top_bs = bdrv_find_backing_image(bs, top);
2693         }
2694     }
2695 
2696     if (top_bs == NULL) {
2697         error_setg(errp, "Top image file %s not found", top ? top : "NULL");
2698         goto out;
2699     }
2700 
2701     assert(bdrv_get_aio_context(top_bs) == aio_context);
2702 
2703     if (has_base_node && has_base) {
2704         error_setg(errp, "'base-node' and 'base' are mutually exclusive");
2705         goto out;
2706     } else if (has_base_node) {
2707         base_bs = bdrv_lookup_bs(NULL, base_node, errp);
2708         if (base_bs == NULL) {
2709             goto out;
2710         }
2711         if (!bdrv_chain_contains(top_bs, base_bs)) {
2712             error_setg(errp, "'%s' is not in this backing file chain",
2713                        base_node);
2714             goto out;
2715         }
2716     } else if (has_base && base) {
2717         base_bs = bdrv_find_backing_image(top_bs, base);
2718         if (base_bs == NULL) {
2719             error_setg(errp, "Can't find '%s' in the backing chain", base);
2720             goto out;
2721         }
2722     } else {
2723         base_bs = bdrv_find_base(top_bs);
2724         if (base_bs == NULL) {
2725             error_setg(errp, "There is no backimg image");
2726             goto out;
2727         }
2728     }
2729 
2730     assert(bdrv_get_aio_context(base_bs) == aio_context);
2731 
2732     for (iter = top_bs; iter != bdrv_filter_or_cow_bs(base_bs);
2733          iter = bdrv_filter_or_cow_bs(iter))
2734     {
2735         if (bdrv_op_is_blocked(iter, BLOCK_OP_TYPE_COMMIT_TARGET, errp)) {
2736             goto out;
2737         }
2738     }
2739 
2740     /* Do not allow attempts to commit an image into itself */
2741     if (top_bs == base_bs) {
2742         error_setg(errp, "cannot commit an image into itself");
2743         goto out;
2744     }
2745 
2746     /*
2747      * Active commit is required if and only if someone has taken a
2748      * WRITE permission on the top node.  Historically, we have always
2749      * used active commit for top nodes, so continue that practice
2750      * lest we possibly break clients that rely on this behavior, e.g.
2751      * to later attach this node to a writing parent.
2752      * (Active commit is never really wrong.)
2753      */
2754     bdrv_get_cumulative_perm(top_bs, &top_perm, &top_shared);
2755     if (top_perm & BLK_PERM_WRITE ||
2756         bdrv_skip_filters(top_bs) == bdrv_skip_filters(bs))
2757     {
2758         if (has_backing_file) {
2759             if (bdrv_skip_filters(top_bs) == bdrv_skip_filters(bs)) {
2760                 error_setg(errp, "'backing-file' specified,"
2761                                  " but 'top' is the active layer");
2762             } else {
2763                 error_setg(errp, "'backing-file' specified, but 'top' has a "
2764                                  "writer on it");
2765             }
2766             goto out;
2767         }
2768         if (!has_job_id) {
2769             /*
2770              * Emulate here what block_job_create() does, because it
2771              * is possible that @bs != @top_bs (the block job should
2772              * be named after @bs, even if @top_bs is the actual
2773              * source)
2774              */
2775             job_id = bdrv_get_device_name(bs);
2776         }
2777         commit_active_start(job_id, top_bs, base_bs, job_flags, speed, on_error,
2778                             filter_node_name, NULL, NULL, false, &local_err);
2779     } else {
2780         BlockDriverState *overlay_bs = bdrv_find_overlay(bs, top_bs);
2781         if (bdrv_op_is_blocked(overlay_bs, BLOCK_OP_TYPE_COMMIT_TARGET, errp)) {
2782             goto out;
2783         }
2784         commit_start(has_job_id ? job_id : NULL, bs, base_bs, top_bs, job_flags,
2785                      speed, on_error, has_backing_file ? backing_file : NULL,
2786                      filter_node_name, &local_err);
2787     }
2788     if (local_err != NULL) {
2789         error_propagate(errp, local_err);
2790         goto out;
2791     }
2792 
2793 out:
2794     aio_context_release(aio_context);
2795 }
2796 
2797 /* Common QMP interface for drive-backup and blockdev-backup */
2798 static BlockJob *do_backup_common(BackupCommon *backup,
2799                                   BlockDriverState *bs,
2800                                   BlockDriverState *target_bs,
2801                                   AioContext *aio_context,
2802                                   JobTxn *txn, Error **errp)
2803 {
2804     BlockJob *job = NULL;
2805     BdrvDirtyBitmap *bmap = NULL;
2806     BackupPerf perf = { .max_workers = 64 };
2807     int job_flags = JOB_DEFAULT;
2808 
2809     if (!backup->has_speed) {
2810         backup->speed = 0;
2811     }
2812     if (!backup->has_on_source_error) {
2813         backup->on_source_error = BLOCKDEV_ON_ERROR_REPORT;
2814     }
2815     if (!backup->has_on_target_error) {
2816         backup->on_target_error = BLOCKDEV_ON_ERROR_REPORT;
2817     }
2818     if (!backup->has_job_id) {
2819         backup->job_id = NULL;
2820     }
2821     if (!backup->has_auto_finalize) {
2822         backup->auto_finalize = true;
2823     }
2824     if (!backup->has_auto_dismiss) {
2825         backup->auto_dismiss = true;
2826     }
2827     if (!backup->has_compress) {
2828         backup->compress = false;
2829     }
2830 
2831     if (backup->x_perf) {
2832         if (backup->x_perf->has_use_copy_range) {
2833             perf.use_copy_range = backup->x_perf->use_copy_range;
2834         }
2835         if (backup->x_perf->has_max_workers) {
2836             perf.max_workers = backup->x_perf->max_workers;
2837         }
2838         if (backup->x_perf->has_max_chunk) {
2839             perf.max_chunk = backup->x_perf->max_chunk;
2840         }
2841     }
2842 
2843     if ((backup->sync == MIRROR_SYNC_MODE_BITMAP) ||
2844         (backup->sync == MIRROR_SYNC_MODE_INCREMENTAL)) {
2845         /* done before desugaring 'incremental' to print the right message */
2846         if (!backup->has_bitmap) {
2847             error_setg(errp, "must provide a valid bitmap name for "
2848                        "'%s' sync mode", MirrorSyncMode_str(backup->sync));
2849             return NULL;
2850         }
2851     }
2852 
2853     if (backup->sync == MIRROR_SYNC_MODE_INCREMENTAL) {
2854         if (backup->has_bitmap_mode &&
2855             backup->bitmap_mode != BITMAP_SYNC_MODE_ON_SUCCESS) {
2856             error_setg(errp, "Bitmap sync mode must be '%s' "
2857                        "when using sync mode '%s'",
2858                        BitmapSyncMode_str(BITMAP_SYNC_MODE_ON_SUCCESS),
2859                        MirrorSyncMode_str(backup->sync));
2860             return NULL;
2861         }
2862         backup->has_bitmap_mode = true;
2863         backup->sync = MIRROR_SYNC_MODE_BITMAP;
2864         backup->bitmap_mode = BITMAP_SYNC_MODE_ON_SUCCESS;
2865     }
2866 
2867     if (backup->has_bitmap) {
2868         bmap = bdrv_find_dirty_bitmap(bs, backup->bitmap);
2869         if (!bmap) {
2870             error_setg(errp, "Bitmap '%s' could not be found", backup->bitmap);
2871             return NULL;
2872         }
2873         if (!backup->has_bitmap_mode) {
2874             error_setg(errp, "Bitmap sync mode must be given "
2875                        "when providing a bitmap");
2876             return NULL;
2877         }
2878         if (bdrv_dirty_bitmap_check(bmap, BDRV_BITMAP_ALLOW_RO, errp)) {
2879             return NULL;
2880         }
2881 
2882         /* This does not produce a useful bitmap artifact: */
2883         if (backup->sync == MIRROR_SYNC_MODE_NONE) {
2884             error_setg(errp, "sync mode '%s' does not produce meaningful bitmap"
2885                        " outputs", MirrorSyncMode_str(backup->sync));
2886             return NULL;
2887         }
2888 
2889         /* If the bitmap isn't used for input or output, this is useless: */
2890         if (backup->bitmap_mode == BITMAP_SYNC_MODE_NEVER &&
2891             backup->sync != MIRROR_SYNC_MODE_BITMAP) {
2892             error_setg(errp, "Bitmap sync mode '%s' has no meaningful effect"
2893                        " when combined with sync mode '%s'",
2894                        BitmapSyncMode_str(backup->bitmap_mode),
2895                        MirrorSyncMode_str(backup->sync));
2896             return NULL;
2897         }
2898     }
2899 
2900     if (!backup->has_bitmap && backup->has_bitmap_mode) {
2901         error_setg(errp, "Cannot specify bitmap sync mode without a bitmap");
2902         return NULL;
2903     }
2904 
2905     if (!backup->auto_finalize) {
2906         job_flags |= JOB_MANUAL_FINALIZE;
2907     }
2908     if (!backup->auto_dismiss) {
2909         job_flags |= JOB_MANUAL_DISMISS;
2910     }
2911 
2912     job = backup_job_create(backup->job_id, bs, target_bs, backup->speed,
2913                             backup->sync, bmap, backup->bitmap_mode,
2914                             backup->compress,
2915                             backup->filter_node_name,
2916                             &perf,
2917                             backup->on_source_error,
2918                             backup->on_target_error,
2919                             job_flags, NULL, NULL, txn, errp);
2920     return job;
2921 }
2922 
2923 void qmp_drive_backup(DriveBackup *backup, Error **errp)
2924 {
2925     TransactionAction action = {
2926         .type = TRANSACTION_ACTION_KIND_DRIVE_BACKUP,
2927         .u.drive_backup.data = backup,
2928     };
2929     blockdev_do_action(&action, errp);
2930 }
2931 
2932 BlockDeviceInfoList *qmp_query_named_block_nodes(bool has_flat,
2933                                                  bool flat,
2934                                                  Error **errp)
2935 {
2936     bool return_flat = has_flat && flat;
2937 
2938     return bdrv_named_nodes_list(return_flat, errp);
2939 }
2940 
2941 XDbgBlockGraph *qmp_x_debug_query_block_graph(Error **errp)
2942 {
2943     return bdrv_get_xdbg_block_graph(errp);
2944 }
2945 
2946 void qmp_blockdev_backup(BlockdevBackup *backup, Error **errp)
2947 {
2948     TransactionAction action = {
2949         .type = TRANSACTION_ACTION_KIND_BLOCKDEV_BACKUP,
2950         .u.blockdev_backup.data = backup,
2951     };
2952     blockdev_do_action(&action, errp);
2953 }
2954 
2955 /* Parameter check and block job starting for drive mirroring.
2956  * Caller should hold @device and @target's aio context (must be the same).
2957  **/
2958 static void blockdev_mirror_common(const char *job_id, BlockDriverState *bs,
2959                                    BlockDriverState *target,
2960                                    bool has_replaces, const char *replaces,
2961                                    enum MirrorSyncMode sync,
2962                                    BlockMirrorBackingMode backing_mode,
2963                                    bool zero_target,
2964                                    bool has_speed, int64_t speed,
2965                                    bool has_granularity, uint32_t granularity,
2966                                    bool has_buf_size, int64_t buf_size,
2967                                    bool has_on_source_error,
2968                                    BlockdevOnError on_source_error,
2969                                    bool has_on_target_error,
2970                                    BlockdevOnError on_target_error,
2971                                    bool has_unmap, bool unmap,
2972                                    bool has_filter_node_name,
2973                                    const char *filter_node_name,
2974                                    bool has_copy_mode, MirrorCopyMode copy_mode,
2975                                    bool has_auto_finalize, bool auto_finalize,
2976                                    bool has_auto_dismiss, bool auto_dismiss,
2977                                    Error **errp)
2978 {
2979     BlockDriverState *unfiltered_bs;
2980     int job_flags = JOB_DEFAULT;
2981 
2982     if (!has_speed) {
2983         speed = 0;
2984     }
2985     if (!has_on_source_error) {
2986         on_source_error = BLOCKDEV_ON_ERROR_REPORT;
2987     }
2988     if (!has_on_target_error) {
2989         on_target_error = BLOCKDEV_ON_ERROR_REPORT;
2990     }
2991     if (!has_granularity) {
2992         granularity = 0;
2993     }
2994     if (!has_buf_size) {
2995         buf_size = 0;
2996     }
2997     if (!has_unmap) {
2998         unmap = true;
2999     }
3000     if (!has_filter_node_name) {
3001         filter_node_name = NULL;
3002     }
3003     if (!has_copy_mode) {
3004         copy_mode = MIRROR_COPY_MODE_BACKGROUND;
3005     }
3006     if (has_auto_finalize && !auto_finalize) {
3007         job_flags |= JOB_MANUAL_FINALIZE;
3008     }
3009     if (has_auto_dismiss && !auto_dismiss) {
3010         job_flags |= JOB_MANUAL_DISMISS;
3011     }
3012 
3013     if (granularity != 0 && (granularity < 512 || granularity > 1048576 * 64)) {
3014         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "granularity",
3015                    "a value in range [512B, 64MB]");
3016         return;
3017     }
3018     if (granularity & (granularity - 1)) {
3019         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "granularity",
3020                    "a power of 2");
3021         return;
3022     }
3023 
3024     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_MIRROR_SOURCE, errp)) {
3025         return;
3026     }
3027     if (bdrv_op_is_blocked(target, BLOCK_OP_TYPE_MIRROR_TARGET, errp)) {
3028         return;
3029     }
3030 
3031     if (!bdrv_backing_chain_next(bs) && sync == MIRROR_SYNC_MODE_TOP) {
3032         sync = MIRROR_SYNC_MODE_FULL;
3033     }
3034 
3035     if (!has_replaces) {
3036         /* We want to mirror from @bs, but keep implicit filters on top */
3037         unfiltered_bs = bdrv_skip_implicit_filters(bs);
3038         if (unfiltered_bs != bs) {
3039             replaces = unfiltered_bs->node_name;
3040             has_replaces = true;
3041         }
3042     }
3043 
3044     if (has_replaces) {
3045         BlockDriverState *to_replace_bs;
3046         AioContext *replace_aio_context;
3047         int64_t bs_size, replace_size;
3048 
3049         bs_size = bdrv_getlength(bs);
3050         if (bs_size < 0) {
3051             error_setg_errno(errp, -bs_size, "Failed to query device's size");
3052             return;
3053         }
3054 
3055         to_replace_bs = check_to_replace_node(bs, replaces, errp);
3056         if (!to_replace_bs) {
3057             return;
3058         }
3059 
3060         replace_aio_context = bdrv_get_aio_context(to_replace_bs);
3061         aio_context_acquire(replace_aio_context);
3062         replace_size = bdrv_getlength(to_replace_bs);
3063         aio_context_release(replace_aio_context);
3064 
3065         if (replace_size < 0) {
3066             error_setg_errno(errp, -replace_size,
3067                              "Failed to query the replacement node's size");
3068             return;
3069         }
3070         if (bs_size != replace_size) {
3071             error_setg(errp, "cannot replace image with a mirror image of "
3072                              "different size");
3073             return;
3074         }
3075     }
3076 
3077     /* pass the node name to replace to mirror start since it's loose coupling
3078      * and will allow to check whether the node still exist at mirror completion
3079      */
3080     mirror_start(job_id, bs, target,
3081                  has_replaces ? replaces : NULL, job_flags,
3082                  speed, granularity, buf_size, sync, backing_mode, zero_target,
3083                  on_source_error, on_target_error, unmap, filter_node_name,
3084                  copy_mode, errp);
3085 }
3086 
3087 void qmp_drive_mirror(DriveMirror *arg, Error **errp)
3088 {
3089     BlockDriverState *bs;
3090     BlockDriverState *target_backing_bs, *target_bs;
3091     AioContext *aio_context;
3092     AioContext *old_context;
3093     BlockMirrorBackingMode backing_mode;
3094     Error *local_err = NULL;
3095     QDict *options = NULL;
3096     int flags;
3097     int64_t size;
3098     const char *format = arg->format;
3099     bool zero_target;
3100     int ret;
3101 
3102     bs = qmp_get_root_bs(arg->device, errp);
3103     if (!bs) {
3104         return;
3105     }
3106 
3107     /* Early check to avoid creating target */
3108     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_MIRROR_SOURCE, errp)) {
3109         return;
3110     }
3111 
3112     aio_context = bdrv_get_aio_context(bs);
3113     aio_context_acquire(aio_context);
3114 
3115     if (!arg->has_mode) {
3116         arg->mode = NEW_IMAGE_MODE_ABSOLUTE_PATHS;
3117     }
3118 
3119     if (!arg->has_format) {
3120         format = (arg->mode == NEW_IMAGE_MODE_EXISTING
3121                   ? NULL : bs->drv->format_name);
3122     }
3123 
3124     flags = bs->open_flags | BDRV_O_RDWR;
3125     target_backing_bs = bdrv_cow_bs(bdrv_skip_filters(bs));
3126     if (!target_backing_bs && arg->sync == MIRROR_SYNC_MODE_TOP) {
3127         arg->sync = MIRROR_SYNC_MODE_FULL;
3128     }
3129     if (arg->sync == MIRROR_SYNC_MODE_NONE) {
3130         target_backing_bs = bs;
3131     }
3132 
3133     size = bdrv_getlength(bs);
3134     if (size < 0) {
3135         error_setg_errno(errp, -size, "bdrv_getlength failed");
3136         goto out;
3137     }
3138 
3139     if (arg->has_replaces) {
3140         if (!arg->has_node_name) {
3141             error_setg(errp, "a node-name must be provided when replacing a"
3142                              " named node of the graph");
3143             goto out;
3144         }
3145     }
3146 
3147     if (arg->mode == NEW_IMAGE_MODE_ABSOLUTE_PATHS) {
3148         backing_mode = MIRROR_SOURCE_BACKING_CHAIN;
3149     } else {
3150         backing_mode = MIRROR_OPEN_BACKING_CHAIN;
3151     }
3152 
3153     /* Don't open backing image in create() */
3154     flags |= BDRV_O_NO_BACKING;
3155 
3156     if ((arg->sync == MIRROR_SYNC_MODE_FULL || !target_backing_bs)
3157         && arg->mode != NEW_IMAGE_MODE_EXISTING)
3158     {
3159         /* create new image w/o backing file */
3160         assert(format);
3161         bdrv_img_create(arg->target, format,
3162                         NULL, NULL, NULL, size, flags, false, &local_err);
3163     } else {
3164         /* Implicit filters should not appear in the filename */
3165         BlockDriverState *explicit_backing =
3166             bdrv_skip_implicit_filters(target_backing_bs);
3167 
3168         switch (arg->mode) {
3169         case NEW_IMAGE_MODE_EXISTING:
3170             break;
3171         case NEW_IMAGE_MODE_ABSOLUTE_PATHS:
3172             /* create new image with backing file */
3173             bdrv_refresh_filename(explicit_backing);
3174             bdrv_img_create(arg->target, format,
3175                             explicit_backing->filename,
3176                             explicit_backing->drv->format_name,
3177                             NULL, size, flags, false, &local_err);
3178             break;
3179         default:
3180             abort();
3181         }
3182     }
3183 
3184     if (local_err) {
3185         error_propagate(errp, local_err);
3186         goto out;
3187     }
3188 
3189     options = qdict_new();
3190     if (arg->has_node_name) {
3191         qdict_put_str(options, "node-name", arg->node_name);
3192     }
3193     if (format) {
3194         qdict_put_str(options, "driver", format);
3195     }
3196 
3197     /* Mirroring takes care of copy-on-write using the source's backing
3198      * file.
3199      */
3200     target_bs = bdrv_open(arg->target, NULL, options, flags, errp);
3201     if (!target_bs) {
3202         goto out;
3203     }
3204 
3205     zero_target = (arg->sync == MIRROR_SYNC_MODE_FULL &&
3206                    (arg->mode == NEW_IMAGE_MODE_EXISTING ||
3207                     !bdrv_has_zero_init(target_bs)));
3208 
3209 
3210     /* Honor bdrv_try_set_aio_context() context acquisition requirements. */
3211     old_context = bdrv_get_aio_context(target_bs);
3212     aio_context_release(aio_context);
3213     aio_context_acquire(old_context);
3214 
3215     ret = bdrv_try_set_aio_context(target_bs, aio_context, errp);
3216     if (ret < 0) {
3217         bdrv_unref(target_bs);
3218         aio_context_release(old_context);
3219         return;
3220     }
3221 
3222     aio_context_release(old_context);
3223     aio_context_acquire(aio_context);
3224 
3225     blockdev_mirror_common(arg->has_job_id ? arg->job_id : NULL, bs, target_bs,
3226                            arg->has_replaces, arg->replaces, arg->sync,
3227                            backing_mode, zero_target,
3228                            arg->has_speed, arg->speed,
3229                            arg->has_granularity, arg->granularity,
3230                            arg->has_buf_size, arg->buf_size,
3231                            arg->has_on_source_error, arg->on_source_error,
3232                            arg->has_on_target_error, arg->on_target_error,
3233                            arg->has_unmap, arg->unmap,
3234                            false, NULL,
3235                            arg->has_copy_mode, arg->copy_mode,
3236                            arg->has_auto_finalize, arg->auto_finalize,
3237                            arg->has_auto_dismiss, arg->auto_dismiss,
3238                            errp);
3239     bdrv_unref(target_bs);
3240 out:
3241     aio_context_release(aio_context);
3242 }
3243 
3244 void qmp_blockdev_mirror(bool has_job_id, const char *job_id,
3245                          const char *device, const char *target,
3246                          bool has_replaces, const char *replaces,
3247                          MirrorSyncMode sync,
3248                          bool has_speed, int64_t speed,
3249                          bool has_granularity, uint32_t granularity,
3250                          bool has_buf_size, int64_t buf_size,
3251                          bool has_on_source_error,
3252                          BlockdevOnError on_source_error,
3253                          bool has_on_target_error,
3254                          BlockdevOnError on_target_error,
3255                          bool has_filter_node_name,
3256                          const char *filter_node_name,
3257                          bool has_copy_mode, MirrorCopyMode copy_mode,
3258                          bool has_auto_finalize, bool auto_finalize,
3259                          bool has_auto_dismiss, bool auto_dismiss,
3260                          Error **errp)
3261 {
3262     BlockDriverState *bs;
3263     BlockDriverState *target_bs;
3264     AioContext *aio_context;
3265     AioContext *old_context;
3266     BlockMirrorBackingMode backing_mode = MIRROR_LEAVE_BACKING_CHAIN;
3267     bool zero_target;
3268     int ret;
3269 
3270     bs = qmp_get_root_bs(device, errp);
3271     if (!bs) {
3272         return;
3273     }
3274 
3275     target_bs = bdrv_lookup_bs(target, target, errp);
3276     if (!target_bs) {
3277         return;
3278     }
3279 
3280     zero_target = (sync == MIRROR_SYNC_MODE_FULL);
3281 
3282     /* Honor bdrv_try_set_aio_context() context acquisition requirements. */
3283     old_context = bdrv_get_aio_context(target_bs);
3284     aio_context = bdrv_get_aio_context(bs);
3285     aio_context_acquire(old_context);
3286 
3287     ret = bdrv_try_set_aio_context(target_bs, aio_context, errp);
3288 
3289     aio_context_release(old_context);
3290     aio_context_acquire(aio_context);
3291 
3292     if (ret < 0) {
3293         goto out;
3294     }
3295 
3296     blockdev_mirror_common(has_job_id ? job_id : NULL, bs, target_bs,
3297                            has_replaces, replaces, sync, backing_mode,
3298                            zero_target, has_speed, speed,
3299                            has_granularity, granularity,
3300                            has_buf_size, buf_size,
3301                            has_on_source_error, on_source_error,
3302                            has_on_target_error, on_target_error,
3303                            true, true,
3304                            has_filter_node_name, filter_node_name,
3305                            has_copy_mode, copy_mode,
3306                            has_auto_finalize, auto_finalize,
3307                            has_auto_dismiss, auto_dismiss,
3308                            errp);
3309 out:
3310     aio_context_release(aio_context);
3311 }
3312 
3313 /* Get a block job using its ID and acquire its AioContext */
3314 static BlockJob *find_block_job(const char *id, AioContext **aio_context,
3315                                 Error **errp)
3316 {
3317     BlockJob *job;
3318 
3319     assert(id != NULL);
3320 
3321     *aio_context = NULL;
3322 
3323     job = block_job_get(id);
3324 
3325     if (!job) {
3326         error_set(errp, ERROR_CLASS_DEVICE_NOT_ACTIVE,
3327                   "Block job '%s' not found", id);
3328         return NULL;
3329     }
3330 
3331     *aio_context = blk_get_aio_context(job->blk);
3332     aio_context_acquire(*aio_context);
3333 
3334     return job;
3335 }
3336 
3337 void qmp_block_job_set_speed(const char *device, int64_t speed, 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     block_job_set_speed(job, speed, errp);
3347     aio_context_release(aio_context);
3348 }
3349 
3350 void qmp_block_job_cancel(const char *device,
3351                           bool has_force, bool force, 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     if (!has_force) {
3361         force = false;
3362     }
3363 
3364     if (job_user_paused(&job->job) && !force) {
3365         error_setg(errp, "The block job for device '%s' is currently paused",
3366                    device);
3367         goto out;
3368     }
3369 
3370     trace_qmp_block_job_cancel(job);
3371     job_user_cancel(&job->job, force, errp);
3372 out:
3373     aio_context_release(aio_context);
3374 }
3375 
3376 void qmp_block_job_pause(const char *device, Error **errp)
3377 {
3378     AioContext *aio_context;
3379     BlockJob *job = find_block_job(device, &aio_context, errp);
3380 
3381     if (!job) {
3382         return;
3383     }
3384 
3385     trace_qmp_block_job_pause(job);
3386     job_user_pause(&job->job, errp);
3387     aio_context_release(aio_context);
3388 }
3389 
3390 void qmp_block_job_resume(const char *device, Error **errp)
3391 {
3392     AioContext *aio_context;
3393     BlockJob *job = find_block_job(device, &aio_context, errp);
3394 
3395     if (!job) {
3396         return;
3397     }
3398 
3399     trace_qmp_block_job_resume(job);
3400     job_user_resume(&job->job, errp);
3401     aio_context_release(aio_context);
3402 }
3403 
3404 void qmp_block_job_complete(const char *device, Error **errp)
3405 {
3406     AioContext *aio_context;
3407     BlockJob *job = find_block_job(device, &aio_context, errp);
3408 
3409     if (!job) {
3410         return;
3411     }
3412 
3413     trace_qmp_block_job_complete(job);
3414     job_complete(&job->job, errp);
3415     aio_context_release(aio_context);
3416 }
3417 
3418 void qmp_block_job_finalize(const char *id, Error **errp)
3419 {
3420     AioContext *aio_context;
3421     BlockJob *job = find_block_job(id, &aio_context, errp);
3422 
3423     if (!job) {
3424         return;
3425     }
3426 
3427     trace_qmp_block_job_finalize(job);
3428     job_ref(&job->job);
3429     job_finalize(&job->job, errp);
3430 
3431     /*
3432      * Job's context might have changed via job_finalize (and job_txn_apply
3433      * automatically acquires the new one), so make sure we release the correct
3434      * one.
3435      */
3436     aio_context = blk_get_aio_context(job->blk);
3437     job_unref(&job->job);
3438     aio_context_release(aio_context);
3439 }
3440 
3441 void qmp_block_job_dismiss(const char *id, Error **errp)
3442 {
3443     AioContext *aio_context;
3444     BlockJob *bjob = find_block_job(id, &aio_context, errp);
3445     Job *job;
3446 
3447     if (!bjob) {
3448         return;
3449     }
3450 
3451     trace_qmp_block_job_dismiss(bjob);
3452     job = &bjob->job;
3453     job_dismiss(&job, errp);
3454     aio_context_release(aio_context);
3455 }
3456 
3457 void qmp_change_backing_file(const char *device,
3458                              const char *image_node_name,
3459                              const char *backing_file,
3460                              Error **errp)
3461 {
3462     BlockDriverState *bs = NULL;
3463     AioContext *aio_context;
3464     BlockDriverState *image_bs = NULL;
3465     Error *local_err = NULL;
3466     bool ro;
3467     int ret;
3468 
3469     bs = qmp_get_root_bs(device, errp);
3470     if (!bs) {
3471         return;
3472     }
3473 
3474     aio_context = bdrv_get_aio_context(bs);
3475     aio_context_acquire(aio_context);
3476 
3477     image_bs = bdrv_lookup_bs(NULL, image_node_name, &local_err);
3478     if (local_err) {
3479         error_propagate(errp, local_err);
3480         goto out;
3481     }
3482 
3483     if (!image_bs) {
3484         error_setg(errp, "image file not found");
3485         goto out;
3486     }
3487 
3488     if (bdrv_find_base(image_bs) == image_bs) {
3489         error_setg(errp, "not allowing backing file change on an image "
3490                          "without a backing file");
3491         goto out;
3492     }
3493 
3494     /* even though we are not necessarily operating on bs, we need it to
3495      * determine if block ops are currently prohibited on the chain */
3496     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_CHANGE, errp)) {
3497         goto out;
3498     }
3499 
3500     /* final sanity check */
3501     if (!bdrv_chain_contains(bs, image_bs)) {
3502         error_setg(errp, "'%s' and image file are not in the same chain",
3503                    device);
3504         goto out;
3505     }
3506 
3507     /* if not r/w, reopen to make r/w */
3508     ro = bdrv_is_read_only(image_bs);
3509 
3510     if (ro) {
3511         if (bdrv_reopen_set_read_only(image_bs, false, errp) != 0) {
3512             goto out;
3513         }
3514     }
3515 
3516     ret = bdrv_change_backing_file(image_bs, backing_file,
3517                                    image_bs->drv ? image_bs->drv->format_name : "",
3518                                    false);
3519 
3520     if (ret < 0) {
3521         error_setg_errno(errp, -ret, "Could not change backing file to '%s'",
3522                          backing_file);
3523         /* don't exit here, so we can try to restore open flags if
3524          * appropriate */
3525     }
3526 
3527     if (ro) {
3528         bdrv_reopen_set_read_only(image_bs, true, errp);
3529     }
3530 
3531 out:
3532     aio_context_release(aio_context);
3533 }
3534 
3535 void qmp_blockdev_add(BlockdevOptions *options, Error **errp)
3536 {
3537     BlockDriverState *bs;
3538     QObject *obj;
3539     Visitor *v = qobject_output_visitor_new(&obj);
3540     QDict *qdict;
3541 
3542     visit_type_BlockdevOptions(v, NULL, &options, &error_abort);
3543     visit_complete(v, &obj);
3544     qdict = qobject_to(QDict, obj);
3545 
3546     qdict_flatten(qdict);
3547 
3548     if (!qdict_get_try_str(qdict, "node-name")) {
3549         error_setg(errp, "'node-name' must be specified for the root node");
3550         goto fail;
3551     }
3552 
3553     bs = bds_tree_init(qdict, errp);
3554     if (!bs) {
3555         goto fail;
3556     }
3557 
3558     bdrv_set_monitor_owned(bs);
3559 
3560 fail:
3561     visit_free(v);
3562 }
3563 
3564 void qmp_x_blockdev_reopen(BlockdevOptions *options, Error **errp)
3565 {
3566     BlockDriverState *bs;
3567     AioContext *ctx;
3568     QObject *obj;
3569     Visitor *v = qobject_output_visitor_new(&obj);
3570     BlockReopenQueue *queue;
3571     QDict *qdict;
3572 
3573     /* Check for the selected node name */
3574     if (!options->has_node_name) {
3575         error_setg(errp, "node-name not specified");
3576         goto fail;
3577     }
3578 
3579     bs = bdrv_find_node(options->node_name);
3580     if (!bs) {
3581         error_setg(errp, "Failed to find node with node-name='%s'",
3582                    options->node_name);
3583         goto fail;
3584     }
3585 
3586     /* Put all options in a QDict and flatten it */
3587     visit_type_BlockdevOptions(v, NULL, &options, &error_abort);
3588     visit_complete(v, &obj);
3589     qdict = qobject_to(QDict, obj);
3590 
3591     qdict_flatten(qdict);
3592 
3593     /* Perform the reopen operation */
3594     ctx = bdrv_get_aio_context(bs);
3595     aio_context_acquire(ctx);
3596     bdrv_subtree_drained_begin(bs);
3597     queue = bdrv_reopen_queue(NULL, bs, qdict, false);
3598     bdrv_reopen_multiple(queue, errp);
3599     bdrv_subtree_drained_end(bs);
3600     aio_context_release(ctx);
3601 
3602 fail:
3603     visit_free(v);
3604 }
3605 
3606 void qmp_blockdev_del(const char *node_name, Error **errp)
3607 {
3608     AioContext *aio_context;
3609     BlockDriverState *bs;
3610 
3611     bs = bdrv_find_node(node_name);
3612     if (!bs) {
3613         error_setg(errp, "Failed to find node with node-name='%s'", node_name);
3614         return;
3615     }
3616     if (bdrv_has_blk(bs)) {
3617         error_setg(errp, "Node %s is in use", node_name);
3618         return;
3619     }
3620     aio_context = bdrv_get_aio_context(bs);
3621     aio_context_acquire(aio_context);
3622 
3623     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_DRIVE_DEL, errp)) {
3624         goto out;
3625     }
3626 
3627     if (!QTAILQ_IN_USE(bs, monitor_list)) {
3628         error_setg(errp, "Node %s is not owned by the monitor",
3629                    bs->node_name);
3630         goto out;
3631     }
3632 
3633     if (bs->refcnt > 1) {
3634         error_setg(errp, "Block device %s is in use",
3635                    bdrv_get_device_or_node_name(bs));
3636         goto out;
3637     }
3638 
3639     QTAILQ_REMOVE(&monitor_bdrv_states, bs, monitor_list);
3640     bdrv_unref(bs);
3641 
3642 out:
3643     aio_context_release(aio_context);
3644 }
3645 
3646 static BdrvChild *bdrv_find_child(BlockDriverState *parent_bs,
3647                                   const char *child_name)
3648 {
3649     BdrvChild *child;
3650 
3651     QLIST_FOREACH(child, &parent_bs->children, next) {
3652         if (strcmp(child->name, child_name) == 0) {
3653             return child;
3654         }
3655     }
3656 
3657     return NULL;
3658 }
3659 
3660 void qmp_x_blockdev_change(const char *parent, bool has_child,
3661                            const char *child, bool has_node,
3662                            const char *node, Error **errp)
3663 {
3664     BlockDriverState *parent_bs, *new_bs = NULL;
3665     BdrvChild *p_child;
3666 
3667     parent_bs = bdrv_lookup_bs(parent, parent, errp);
3668     if (!parent_bs) {
3669         return;
3670     }
3671 
3672     if (has_child == has_node) {
3673         if (has_child) {
3674             error_setg(errp, "The parameters child and node are in conflict");
3675         } else {
3676             error_setg(errp, "Either child or node must be specified");
3677         }
3678         return;
3679     }
3680 
3681     if (has_child) {
3682         p_child = bdrv_find_child(parent_bs, child);
3683         if (!p_child) {
3684             error_setg(errp, "Node '%s' does not have child '%s'",
3685                        parent, child);
3686             return;
3687         }
3688         bdrv_del_child(parent_bs, p_child, errp);
3689     }
3690 
3691     if (has_node) {
3692         new_bs = bdrv_find_node(node);
3693         if (!new_bs) {
3694             error_setg(errp, "Node '%s' not found", node);
3695             return;
3696         }
3697         bdrv_add_child(parent_bs, new_bs, errp);
3698     }
3699 }
3700 
3701 BlockJobInfoList *qmp_query_block_jobs(Error **errp)
3702 {
3703     BlockJobInfoList *head = NULL, **tail = &head;
3704     BlockJob *job;
3705 
3706     for (job = block_job_next(NULL); job; job = block_job_next(job)) {
3707         BlockJobInfo *value;
3708         AioContext *aio_context;
3709 
3710         if (block_job_is_internal(job)) {
3711             continue;
3712         }
3713         aio_context = blk_get_aio_context(job->blk);
3714         aio_context_acquire(aio_context);
3715         value = block_job_query(job, errp);
3716         aio_context_release(aio_context);
3717         if (!value) {
3718             qapi_free_BlockJobInfoList(head);
3719             return NULL;
3720         }
3721         QAPI_LIST_APPEND(tail, value);
3722     }
3723 
3724     return head;
3725 }
3726 
3727 void qmp_x_blockdev_set_iothread(const char *node_name, StrOrNull *iothread,
3728                                  bool has_force, bool force, Error **errp)
3729 {
3730     AioContext *old_context;
3731     AioContext *new_context;
3732     BlockDriverState *bs;
3733 
3734     bs = bdrv_find_node(node_name);
3735     if (!bs) {
3736         error_setg(errp, "Failed to find node with node-name='%s'", node_name);
3737         return;
3738     }
3739 
3740     /* Protects against accidents. */
3741     if (!(has_force && force) && bdrv_has_blk(bs)) {
3742         error_setg(errp, "Node %s is associated with a BlockBackend and could "
3743                          "be in use (use force=true to override this check)",
3744                          node_name);
3745         return;
3746     }
3747 
3748     if (iothread->type == QTYPE_QSTRING) {
3749         IOThread *obj = iothread_by_id(iothread->u.s);
3750         if (!obj) {
3751             error_setg(errp, "Cannot find iothread %s", iothread->u.s);
3752             return;
3753         }
3754 
3755         new_context = iothread_get_aio_context(obj);
3756     } else {
3757         new_context = qemu_get_aio_context();
3758     }
3759 
3760     old_context = bdrv_get_aio_context(bs);
3761     aio_context_acquire(old_context);
3762 
3763     bdrv_try_set_aio_context(bs, new_context, errp);
3764 
3765     aio_context_release(old_context);
3766 }
3767 
3768 QemuOptsList qemu_common_drive_opts = {
3769     .name = "drive",
3770     .head = QTAILQ_HEAD_INITIALIZER(qemu_common_drive_opts.head),
3771     .desc = {
3772         {
3773             .name = "snapshot",
3774             .type = QEMU_OPT_BOOL,
3775             .help = "enable/disable snapshot mode",
3776         },{
3777             .name = "aio",
3778             .type = QEMU_OPT_STRING,
3779             .help = "host AIO implementation (threads, native, io_uring)",
3780         },{
3781             .name = BDRV_OPT_CACHE_WB,
3782             .type = QEMU_OPT_BOOL,
3783             .help = "Enable writeback mode",
3784         },{
3785             .name = "format",
3786             .type = QEMU_OPT_STRING,
3787             .help = "disk format (raw, qcow2, ...)",
3788         },{
3789             .name = "rerror",
3790             .type = QEMU_OPT_STRING,
3791             .help = "read error action",
3792         },{
3793             .name = "werror",
3794             .type = QEMU_OPT_STRING,
3795             .help = "write error action",
3796         },{
3797             .name = BDRV_OPT_READ_ONLY,
3798             .type = QEMU_OPT_BOOL,
3799             .help = "open drive file as read-only",
3800         },
3801 
3802         THROTTLE_OPTS,
3803 
3804         {
3805             .name = "throttling.group",
3806             .type = QEMU_OPT_STRING,
3807             .help = "name of the block throttling group",
3808         },{
3809             .name = "copy-on-read",
3810             .type = QEMU_OPT_BOOL,
3811             .help = "copy read data from backing file into image file",
3812         },{
3813             .name = "detect-zeroes",
3814             .type = QEMU_OPT_STRING,
3815             .help = "try to optimize zero writes (off, on, unmap)",
3816         },{
3817             .name = "stats-account-invalid",
3818             .type = QEMU_OPT_BOOL,
3819             .help = "whether to account for invalid I/O operations "
3820                     "in the statistics",
3821         },{
3822             .name = "stats-account-failed",
3823             .type = QEMU_OPT_BOOL,
3824             .help = "whether to account for failed I/O operations "
3825                     "in the statistics",
3826         },
3827         { /* end of list */ }
3828     },
3829 };
3830 
3831 QemuOptsList qemu_drive_opts = {
3832     .name = "drive",
3833     .head = QTAILQ_HEAD_INITIALIZER(qemu_drive_opts.head),
3834     .desc = {
3835         /*
3836          * no elements => accept any params
3837          * validation will happen later
3838          */
3839         { /* end of list */ }
3840     },
3841 };
3842