xref: /openbmc/qemu/blockdev.c (revision 0a0474b065b97e9a25c2723bde87afbc77ba4bd9)
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 "system/block-backend.h"
35 #include "system/blockdev.h"
36 #include "hw/block/block.h"
37 #include "block/blockjob.h"
38 #include "block/dirty-bitmap.h"
39 #include "block/qdict.h"
40 #include "block/throttle-groups.h"
41 #include "monitor/monitor.h"
42 #include "qemu/error-report.h"
43 #include "qemu/option.h"
44 #include "qemu/qemu-print.h"
45 #include "qemu/config-file.h"
46 #include "qapi/qapi-commands-block.h"
47 #include "qapi/qapi-commands-transaction.h"
48 #include "qapi/qapi-visit-block-core.h"
49 #include "qobject/qdict.h"
50 #include "qobject/qnum.h"
51 #include "qobject/qstring.h"
52 #include "qapi/error.h"
53 #include "qapi/qmp/qerror.h"
54 #include "qobject/qlist.h"
55 #include "qapi/qobject-output-visitor.h"
56 #include "system/system.h"
57 #include "system/iothread.h"
58 #include "block/block_int.h"
59 #include "block/trace.h"
60 #include "system/runstate.h"
61 #include "system/replay.h"
62 #include "qemu/cutils.h"
63 #include "qemu/help_option.h"
64 #include "qemu/main-loop.h"
65 #include "qemu/throttle-options.h"
66 
67 /* Protected by BQL */
68 QTAILQ_HEAD(, BlockDriverState) monitor_bdrv_states =
69     QTAILQ_HEAD_INITIALIZER(monitor_bdrv_states);
70 
71 void bdrv_set_monitor_owned(BlockDriverState *bs)
72 {
73     GLOBAL_STATE_CODE();
74     QTAILQ_INSERT_TAIL(&monitor_bdrv_states, bs, monitor_list);
75 }
76 
77 static const char *const if_name[IF_COUNT] = {
78     [IF_NONE] = "none",
79     [IF_IDE] = "ide",
80     [IF_SCSI] = "scsi",
81     [IF_FLOPPY] = "floppy",
82     [IF_PFLASH] = "pflash",
83     [IF_MTD] = "mtd",
84     [IF_SD] = "sd",
85     [IF_VIRTIO] = "virtio",
86     [IF_XEN] = "xen",
87 };
88 
89 static int if_max_devs[IF_COUNT] = {
90     /*
91      * Do not change these numbers!  They govern how drive option
92      * index maps to unit and bus.  That mapping is ABI.
93      *
94      * All controllers used to implement if=T drives need to support
95      * if_max_devs[T] units, for any T with if_max_devs[T] != 0.
96      * Otherwise, some index values map to "impossible" bus, unit
97      * values.
98      *
99      * For instance, if you change [IF_SCSI] to 255, -drive
100      * if=scsi,index=12 no longer means bus=1,unit=5, but
101      * bus=0,unit=12.  With an lsi53c895a controller (7 units max),
102      * the drive can't be set up.  Regression.
103      */
104     [IF_IDE] = 2,
105     [IF_SCSI] = 7,
106 };
107 
108 /**
109  * Boards may call this to offer board-by-board overrides
110  * of the default, global values.
111  */
112 void override_max_devs(BlockInterfaceType type, int max_devs)
113 {
114     BlockBackend *blk;
115     DriveInfo *dinfo;
116 
117     GLOBAL_STATE_CODE();
118 
119     if (max_devs <= 0) {
120         return;
121     }
122 
123     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
124         dinfo = blk_legacy_dinfo(blk);
125         if (dinfo->type == type) {
126             fprintf(stderr, "Cannot override units-per-bus property of"
127                     " the %s interface, because a drive of that type has"
128                     " already been added.\n", if_name[type]);
129             g_assert_not_reached();
130         }
131     }
132 
133     if_max_devs[type] = max_devs;
134 }
135 
136 /*
137  * We automatically delete the drive when a device using it gets
138  * unplugged.  Questionable feature, but we can't just drop it.
139  * Device models call blockdev_mark_auto_del() to schedule the
140  * automatic deletion, and generic qdev code calls blockdev_auto_del()
141  * when deletion is actually safe.
142  */
143 void blockdev_mark_auto_del(BlockBackend *blk)
144 {
145     DriveInfo *dinfo = blk_legacy_dinfo(blk);
146     BlockJob *job;
147 
148     GLOBAL_STATE_CODE();
149 
150     if (!dinfo) {
151         return;
152     }
153 
154     JOB_LOCK_GUARD();
155 
156     do {
157         job = block_job_next_locked(NULL);
158         while (job && (job->job.cancelled ||
159                        job->job.deferred_to_main_loop ||
160                        !block_job_has_bdrv(job, blk_bs(blk))))
161         {
162             job = block_job_next_locked(job);
163         }
164         if (job) {
165             /*
166              * This drops the job lock temporarily and polls, so we need to
167              * restart processing the list from the start after this.
168              */
169             job_cancel_locked(&job->job, false);
170         }
171     } while (job);
172 
173     dinfo->auto_del = 1;
174 }
175 
176 void blockdev_auto_del(BlockBackend *blk)
177 {
178     DriveInfo *dinfo = blk_legacy_dinfo(blk);
179     GLOBAL_STATE_CODE();
180 
181     if (dinfo && dinfo->auto_del) {
182         monitor_remove_blk(blk);
183         blk_unref(blk);
184     }
185 }
186 
187 static int drive_index_to_bus_id(BlockInterfaceType type, int index)
188 {
189     int max_devs = if_max_devs[type];
190     return max_devs ? index / max_devs : 0;
191 }
192 
193 static int drive_index_to_unit_id(BlockInterfaceType type, int index)
194 {
195     int max_devs = if_max_devs[type];
196     return max_devs ? index % max_devs : index;
197 }
198 
199 QemuOpts *drive_add(BlockInterfaceType type, int index, const char *file,
200                     const char *optstr)
201 {
202     QemuOpts *opts;
203 
204     GLOBAL_STATE_CODE();
205 
206     opts = qemu_opts_parse_noisily(qemu_find_opts("drive"), optstr, false);
207     if (!opts) {
208         return NULL;
209     }
210     if (type != IF_DEFAULT) {
211         qemu_opt_set(opts, "if", if_name[type], &error_abort);
212     }
213     if (index >= 0) {
214         qemu_opt_set_number(opts, "index", index, &error_abort);
215     }
216     if (file)
217         qemu_opt_set(opts, "file", file, &error_abort);
218     return opts;
219 }
220 
221 DriveInfo *drive_get(BlockInterfaceType type, int bus, int unit)
222 {
223     BlockBackend *blk;
224     DriveInfo *dinfo;
225 
226     GLOBAL_STATE_CODE();
227 
228     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
229         dinfo = blk_legacy_dinfo(blk);
230         if (dinfo && dinfo->type == type
231             && dinfo->bus == bus && dinfo->unit == unit) {
232             return dinfo;
233         }
234     }
235 
236     return NULL;
237 }
238 
239 /*
240  * Check board claimed all -drive that are meant to be claimed.
241  * Fatal error if any remain unclaimed.
242  */
243 void drive_check_orphaned(void)
244 {
245     BlockBackend *blk;
246     DriveInfo *dinfo;
247     Location loc;
248     bool orphans = false;
249 
250     GLOBAL_STATE_CODE();
251 
252     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
253         dinfo = blk_legacy_dinfo(blk);
254         /*
255          * Ignore default drives, because we create certain default
256          * drives unconditionally, then leave them unclaimed.  Not the
257          * users fault.
258          * Ignore IF_VIRTIO or IF_XEN, because it gets desugared into
259          * -device, so we can leave failing to -device.
260          * Ignore IF_NONE, because leaving unclaimed IF_NONE remains
261          * available for device_add is a feature.
262          */
263         if (dinfo->is_default || dinfo->type == IF_VIRTIO
264             || dinfo->type == IF_XEN || dinfo->type == IF_NONE) {
265             continue;
266         }
267         if (!blk_get_attached_dev(blk)) {
268             loc_push_none(&loc);
269             qemu_opts_loc_restore(dinfo->opts);
270             error_report("machine type does not support"
271                          " if=%s,bus=%d,unit=%d",
272                          if_name[dinfo->type], dinfo->bus, dinfo->unit);
273             loc_pop(&loc);
274             orphans = true;
275         }
276     }
277 
278     if (orphans) {
279         exit(1);
280     }
281 }
282 
283 DriveInfo *drive_get_by_index(BlockInterfaceType type, int index)
284 {
285     GLOBAL_STATE_CODE();
286     return drive_get(type,
287                      drive_index_to_bus_id(type, index),
288                      drive_index_to_unit_id(type, index));
289 }
290 
291 int drive_get_max_bus(BlockInterfaceType type)
292 {
293     int max_bus;
294     BlockBackend *blk;
295     DriveInfo *dinfo;
296 
297     GLOBAL_STATE_CODE();
298 
299     max_bus = -1;
300     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
301         dinfo = blk_legacy_dinfo(blk);
302         if (dinfo && dinfo->type == type && dinfo->bus > max_bus) {
303             max_bus = dinfo->bus;
304         }
305     }
306     return max_bus;
307 }
308 
309 static void bdrv_format_print(void *opaque, const char *name)
310 {
311     qemu_printf(" %s", name);
312 }
313 
314 typedef struct {
315     QEMUBH *bh;
316     BlockDriverState *bs;
317 } BDRVPutRefBH;
318 
319 static int parse_block_error_action(const char *buf, bool is_read, Error **errp)
320 {
321     if (!strcmp(buf, "ignore")) {
322         return BLOCKDEV_ON_ERROR_IGNORE;
323     } else if (!is_read && !strcmp(buf, "enospc")) {
324         return BLOCKDEV_ON_ERROR_ENOSPC;
325     } else if (!strcmp(buf, "stop")) {
326         return BLOCKDEV_ON_ERROR_STOP;
327     } else if (!strcmp(buf, "report")) {
328         return BLOCKDEV_ON_ERROR_REPORT;
329     } else {
330         error_setg(errp, "'%s' invalid %s error action",
331                    buf, is_read ? "read" : "write");
332         return -1;
333     }
334 }
335 
336 static bool parse_stats_intervals(BlockAcctStats *stats, QList *intervals,
337                                   Error **errp)
338 {
339     const QListEntry *entry;
340     for (entry = qlist_first(intervals); entry; entry = qlist_next(entry)) {
341         switch (qobject_type(entry->value)) {
342 
343         case QTYPE_QSTRING: {
344             uint64_t length;
345             const char *str = qstring_get_str(qobject_to(QString,
346                                                          entry->value));
347             if (parse_uint_full(str, 10, &length) == 0 &&
348                 length > 0 && length <= UINT_MAX) {
349                 block_acct_add_interval(stats, (unsigned) length);
350             } else {
351                 error_setg(errp, "Invalid interval length: %s", str);
352                 return false;
353             }
354             break;
355         }
356 
357         case QTYPE_QNUM: {
358             int64_t length = qnum_get_int(qobject_to(QNum, entry->value));
359 
360             if (length > 0 && length <= UINT_MAX) {
361                 block_acct_add_interval(stats, (unsigned) length);
362             } else {
363                 error_setg(errp, "Invalid interval length: %" PRId64, length);
364                 return false;
365             }
366             break;
367         }
368 
369         default:
370             error_setg(errp, "The specification of stats-intervals is invalid");
371             return false;
372         }
373     }
374     return true;
375 }
376 
377 typedef enum { MEDIA_DISK, MEDIA_CDROM } DriveMediaType;
378 
379 /* All parameters but @opts are optional and may be set to NULL. */
380 static void extract_common_blockdev_options(QemuOpts *opts, int *bdrv_flags,
381     const char **throttling_group, ThrottleConfig *throttle_cfg,
382     BlockdevDetectZeroesOptions *detect_zeroes, Error **errp)
383 {
384     Error *local_error = NULL;
385     const char *aio;
386 
387     if (bdrv_flags) {
388         if (qemu_opt_get_bool(opts, "copy-on-read", false)) {
389             *bdrv_flags |= BDRV_O_COPY_ON_READ;
390         }
391 
392         if ((aio = qemu_opt_get(opts, "aio")) != NULL) {
393             if (bdrv_parse_aio(aio, bdrv_flags) < 0) {
394                 error_setg(errp, "invalid aio option");
395                 return;
396             }
397         }
398     }
399 
400     /* disk I/O throttling */
401     if (throttling_group) {
402         *throttling_group = qemu_opt_get(opts, "throttling.group");
403     }
404 
405     if (throttle_cfg) {
406         throttle_config_init(throttle_cfg);
407         throttle_cfg->buckets[THROTTLE_BPS_TOTAL].avg =
408             qemu_opt_get_number(opts, "throttling.bps-total", 0);
409         throttle_cfg->buckets[THROTTLE_BPS_READ].avg  =
410             qemu_opt_get_number(opts, "throttling.bps-read", 0);
411         throttle_cfg->buckets[THROTTLE_BPS_WRITE].avg =
412             qemu_opt_get_number(opts, "throttling.bps-write", 0);
413         throttle_cfg->buckets[THROTTLE_OPS_TOTAL].avg =
414             qemu_opt_get_number(opts, "throttling.iops-total", 0);
415         throttle_cfg->buckets[THROTTLE_OPS_READ].avg =
416             qemu_opt_get_number(opts, "throttling.iops-read", 0);
417         throttle_cfg->buckets[THROTTLE_OPS_WRITE].avg =
418             qemu_opt_get_number(opts, "throttling.iops-write", 0);
419 
420         throttle_cfg->buckets[THROTTLE_BPS_TOTAL].max =
421             qemu_opt_get_number(opts, "throttling.bps-total-max", 0);
422         throttle_cfg->buckets[THROTTLE_BPS_READ].max  =
423             qemu_opt_get_number(opts, "throttling.bps-read-max", 0);
424         throttle_cfg->buckets[THROTTLE_BPS_WRITE].max =
425             qemu_opt_get_number(opts, "throttling.bps-write-max", 0);
426         throttle_cfg->buckets[THROTTLE_OPS_TOTAL].max =
427             qemu_opt_get_number(opts, "throttling.iops-total-max", 0);
428         throttle_cfg->buckets[THROTTLE_OPS_READ].max =
429             qemu_opt_get_number(opts, "throttling.iops-read-max", 0);
430         throttle_cfg->buckets[THROTTLE_OPS_WRITE].max =
431             qemu_opt_get_number(opts, "throttling.iops-write-max", 0);
432 
433         throttle_cfg->buckets[THROTTLE_BPS_TOTAL].burst_length =
434             qemu_opt_get_number(opts, "throttling.bps-total-max-length", 1);
435         throttle_cfg->buckets[THROTTLE_BPS_READ].burst_length  =
436             qemu_opt_get_number(opts, "throttling.bps-read-max-length", 1);
437         throttle_cfg->buckets[THROTTLE_BPS_WRITE].burst_length =
438             qemu_opt_get_number(opts, "throttling.bps-write-max-length", 1);
439         throttle_cfg->buckets[THROTTLE_OPS_TOTAL].burst_length =
440             qemu_opt_get_number(opts, "throttling.iops-total-max-length", 1);
441         throttle_cfg->buckets[THROTTLE_OPS_READ].burst_length =
442             qemu_opt_get_number(opts, "throttling.iops-read-max-length", 1);
443         throttle_cfg->buckets[THROTTLE_OPS_WRITE].burst_length =
444             qemu_opt_get_number(opts, "throttling.iops-write-max-length", 1);
445 
446         throttle_cfg->op_size =
447             qemu_opt_get_number(opts, "throttling.iops-size", 0);
448 
449         if (!throttle_is_valid(throttle_cfg, errp)) {
450             return;
451         }
452     }
453 
454     if (detect_zeroes) {
455         *detect_zeroes =
456             qapi_enum_parse(&BlockdevDetectZeroesOptions_lookup,
457                             qemu_opt_get(opts, "detect-zeroes"),
458                             BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF,
459                             &local_error);
460         if (local_error) {
461             error_propagate(errp, local_error);
462             return;
463         }
464     }
465 }
466 
467 static OnOffAuto account_get_opt(QemuOpts *opts, const char *name)
468 {
469     if (!qemu_opt_find(opts, name)) {
470         return ON_OFF_AUTO_AUTO;
471     }
472     if (qemu_opt_get_bool(opts, name, true)) {
473         return ON_OFF_AUTO_ON;
474     }
475     return ON_OFF_AUTO_OFF;
476 }
477 
478 /* Takes the ownership of bs_opts */
479 static BlockBackend *blockdev_init(const char *file, QDict *bs_opts,
480                                    Error **errp)
481 {
482     const char *buf;
483     int bdrv_flags = 0;
484     int on_read_error, on_write_error;
485     OnOffAuto account_invalid, account_failed;
486     bool writethrough, read_only;
487     BlockBackend *blk;
488     BlockDriverState *bs;
489     ThrottleConfig cfg;
490     int snapshot = 0;
491     Error *error = NULL;
492     QemuOpts *opts;
493     QDict *interval_dict = NULL;
494     QList *interval_list = NULL;
495     const char *id;
496     BlockdevDetectZeroesOptions detect_zeroes =
497         BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF;
498     const char *throttling_group = NULL;
499 
500     /* Check common options by copying from bs_opts to opts, all other options
501      * stay in bs_opts for processing by bdrv_open(). */
502     id = qdict_get_try_str(bs_opts, "id");
503     opts = qemu_opts_create(&qemu_common_drive_opts, id, 1, errp);
504     if (!opts) {
505         goto err_no_opts;
506     }
507 
508     if (!qemu_opts_absorb_qdict(opts, bs_opts, errp)) {
509         goto early_err;
510     }
511 
512     if (id) {
513         qdict_del(bs_opts, "id");
514     }
515 
516     /* extract parameters */
517     snapshot = qemu_opt_get_bool(opts, "snapshot", 0);
518 
519     account_invalid = account_get_opt(opts, "stats-account-invalid");
520     account_failed = account_get_opt(opts, "stats-account-failed");
521 
522     writethrough = !qemu_opt_get_bool(opts, BDRV_OPT_CACHE_WB, true);
523 
524     id = qemu_opts_id(opts);
525 
526     qdict_extract_subqdict(bs_opts, &interval_dict, "stats-intervals.");
527     qdict_array_split(interval_dict, &interval_list);
528 
529     if (qdict_size(interval_dict) != 0) {
530         error_setg(errp, "Invalid option stats-intervals.%s",
531                    qdict_first(interval_dict)->key);
532         goto early_err;
533     }
534 
535     extract_common_blockdev_options(opts, &bdrv_flags, &throttling_group, &cfg,
536                                     &detect_zeroes, &error);
537     if (error) {
538         error_propagate(errp, error);
539         goto early_err;
540     }
541 
542     if ((buf = qemu_opt_get(opts, "format")) != NULL) {
543         if (is_help_option(buf)) {
544             qemu_printf("Supported formats:");
545             bdrv_iterate_format(bdrv_format_print, NULL, false);
546             qemu_printf("\nSupported formats (read-only):");
547             bdrv_iterate_format(bdrv_format_print, NULL, true);
548             qemu_printf("\n");
549             goto early_err;
550         }
551 
552         if (qdict_haskey(bs_opts, "driver")) {
553             error_setg(errp, "Cannot specify both 'driver' and 'format'");
554             goto early_err;
555         }
556         qdict_put_str(bs_opts, "driver", buf);
557     }
558 
559     on_write_error = BLOCKDEV_ON_ERROR_ENOSPC;
560     if ((buf = qemu_opt_get(opts, "werror")) != NULL) {
561         on_write_error = parse_block_error_action(buf, 0, &error);
562         if (error) {
563             error_propagate(errp, error);
564             goto early_err;
565         }
566     }
567 
568     on_read_error = BLOCKDEV_ON_ERROR_REPORT;
569     if ((buf = qemu_opt_get(opts, "rerror")) != NULL) {
570         on_read_error = parse_block_error_action(buf, 1, &error);
571         if (error) {
572             error_propagate(errp, error);
573             goto early_err;
574         }
575     }
576 
577     if (snapshot) {
578         bdrv_flags |= BDRV_O_SNAPSHOT;
579     }
580 
581     read_only = qemu_opt_get_bool(opts, BDRV_OPT_READ_ONLY, false);
582 
583     /* init */
584     if ((!file || !*file) && !qdict_size(bs_opts)) {
585         BlockBackendRootState *blk_rs;
586 
587         blk = blk_new(qemu_get_aio_context(), 0, BLK_PERM_ALL);
588         blk_rs = blk_get_root_state(blk);
589         blk_rs->open_flags    = bdrv_flags | (read_only ? 0 : BDRV_O_RDWR);
590         blk_rs->detect_zeroes = detect_zeroes;
591 
592         qobject_unref(bs_opts);
593     } else {
594         if (file && !*file) {
595             file = NULL;
596         }
597 
598         /* bdrv_open() defaults to the values in bdrv_flags (for compatibility
599          * with other callers) rather than what we want as the real defaults.
600          * Apply the defaults here instead. */
601         qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_DIRECT, "off");
602         qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_NO_FLUSH, "off");
603         qdict_set_default_str(bs_opts, BDRV_OPT_READ_ONLY,
604                               read_only ? "on" : "off");
605         qdict_set_default_str(bs_opts, BDRV_OPT_AUTO_READ_ONLY, "on");
606         assert((bdrv_flags & BDRV_O_CACHE_MASK) == 0);
607 
608         if (runstate_check(RUN_STATE_INMIGRATE)) {
609             bdrv_flags |= BDRV_O_INACTIVE;
610         }
611 
612         blk = blk_new_open(file, NULL, bs_opts, bdrv_flags, errp);
613         if (!blk) {
614             goto err_no_bs_opts;
615         }
616         bs = blk_bs(blk);
617 
618         bs->detect_zeroes = detect_zeroes;
619 
620         block_acct_setup(blk_get_stats(blk), account_invalid, account_failed);
621 
622         if (!parse_stats_intervals(blk_get_stats(blk), interval_list, errp)) {
623             blk_unref(blk);
624             blk = NULL;
625             goto err_no_bs_opts;
626         }
627     }
628 
629     /* disk I/O throttling */
630     if (throttle_enabled(&cfg)) {
631         if (!throttling_group) {
632             throttling_group = id;
633         }
634         blk_io_limits_enable(blk, throttling_group);
635         blk_set_io_limits(blk, &cfg);
636     }
637 
638     blk_set_enable_write_cache(blk, !writethrough);
639     blk_set_on_error(blk, on_read_error, on_write_error);
640 
641     if (!monitor_add_blk(blk, id, errp)) {
642         blk_unref(blk);
643         blk = NULL;
644         goto err_no_bs_opts;
645     }
646 
647 err_no_bs_opts:
648     qemu_opts_del(opts);
649     qobject_unref(interval_dict);
650     qobject_unref(interval_list);
651     return blk;
652 
653 early_err:
654     qemu_opts_del(opts);
655     qobject_unref(interval_dict);
656     qobject_unref(interval_list);
657 err_no_opts:
658     qobject_unref(bs_opts);
659     return NULL;
660 }
661 
662 /* Takes the ownership of bs_opts */
663 BlockDriverState *bds_tree_init(QDict *bs_opts, Error **errp)
664 {
665     int bdrv_flags = 0;
666 
667     GLOBAL_STATE_CODE();
668     /* bdrv_open() defaults to the values in bdrv_flags (for compatibility
669      * with other callers) rather than what we want as the real defaults.
670      * Apply the defaults here instead. */
671     qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_DIRECT, "off");
672     qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_NO_FLUSH, "off");
673     qdict_set_default_str(bs_opts, BDRV_OPT_READ_ONLY, "off");
674 
675     if (runstate_check(RUN_STATE_INMIGRATE)) {
676         bdrv_flags |= BDRV_O_INACTIVE;
677     }
678 
679     return bdrv_open(NULL, NULL, bs_opts, bdrv_flags, errp);
680 }
681 
682 void blockdev_close_all_bdrv_states(void)
683 {
684     BlockDriverState *bs, *next_bs;
685 
686     GLOBAL_STATE_CODE();
687     QTAILQ_FOREACH_SAFE(bs, &monitor_bdrv_states, monitor_list, next_bs) {
688         bdrv_unref(bs);
689     }
690 }
691 
692 /* Iterates over the list of monitor-owned BlockDriverStates */
693 BlockDriverState *bdrv_next_monitor_owned(BlockDriverState *bs)
694 {
695     GLOBAL_STATE_CODE();
696     return bs ? QTAILQ_NEXT(bs, monitor_list)
697               : QTAILQ_FIRST(&monitor_bdrv_states);
698 }
699 
700 static bool qemu_opt_rename(QemuOpts *opts, const char *from, const char *to,
701                             Error **errp)
702 {
703     const char *value;
704 
705     value = qemu_opt_get(opts, from);
706     if (value) {
707         if (qemu_opt_find(opts, to)) {
708             error_setg(errp, "'%s' and its alias '%s' can't be used at the "
709                        "same time", to, from);
710             return false;
711         }
712     }
713 
714     /* rename all items in opts */
715     while ((value = qemu_opt_get(opts, from))) {
716         qemu_opt_set(opts, to, value, &error_abort);
717         qemu_opt_unset(opts, from);
718     }
719     return true;
720 }
721 
722 QemuOptsList qemu_legacy_drive_opts = {
723     .name = "drive",
724     .head = QTAILQ_HEAD_INITIALIZER(qemu_legacy_drive_opts.head),
725     .desc = {
726         {
727             .name = "bus",
728             .type = QEMU_OPT_NUMBER,
729             .help = "bus number",
730         },{
731             .name = "unit",
732             .type = QEMU_OPT_NUMBER,
733             .help = "unit number (i.e. lun for scsi)",
734         },{
735             .name = "index",
736             .type = QEMU_OPT_NUMBER,
737             .help = "index number",
738         },{
739             .name = "media",
740             .type = QEMU_OPT_STRING,
741             .help = "media type (disk, cdrom)",
742         },{
743             .name = "if",
744             .type = QEMU_OPT_STRING,
745             .help = "interface (ide, scsi, sd, mtd, floppy, pflash, virtio)",
746         },{
747             .name = "file",
748             .type = QEMU_OPT_STRING,
749             .help = "file name",
750         },
751 
752         /* Options that are passed on, but have special semantics with -drive */
753         {
754             .name = BDRV_OPT_READ_ONLY,
755             .type = QEMU_OPT_BOOL,
756             .help = "open drive file as read-only",
757         },{
758             .name = "rerror",
759             .type = QEMU_OPT_STRING,
760             .help = "read error action",
761         },{
762             .name = "werror",
763             .type = QEMU_OPT_STRING,
764             .help = "write error action",
765         },{
766             .name = "copy-on-read",
767             .type = QEMU_OPT_BOOL,
768             .help = "copy read data from backing file into image file",
769         },
770 
771         { /* end of list */ }
772     },
773 };
774 
775 DriveInfo *drive_new(QemuOpts *all_opts, BlockInterfaceType block_default_type,
776                      Error **errp)
777 {
778     const char *value;
779     BlockBackend *blk;
780     DriveInfo *dinfo = NULL;
781     QDict *bs_opts;
782     QemuOpts *legacy_opts;
783     DriveMediaType media = MEDIA_DISK;
784     BlockInterfaceType type;
785     int max_devs, bus_id, unit_id, index;
786     const char *werror, *rerror;
787     bool read_only = false;
788     bool copy_on_read;
789     const char *filename;
790     int i;
791 
792     GLOBAL_STATE_CODE();
793 
794     /* Change legacy command line options into QMP ones */
795     static const struct {
796         const char *from;
797         const char *to;
798     } opt_renames[] = {
799         { "iops",           "throttling.iops-total" },
800         { "iops_rd",        "throttling.iops-read" },
801         { "iops_wr",        "throttling.iops-write" },
802 
803         { "bps",            "throttling.bps-total" },
804         { "bps_rd",         "throttling.bps-read" },
805         { "bps_wr",         "throttling.bps-write" },
806 
807         { "iops_max",       "throttling.iops-total-max" },
808         { "iops_rd_max",    "throttling.iops-read-max" },
809         { "iops_wr_max",    "throttling.iops-write-max" },
810 
811         { "bps_max",        "throttling.bps-total-max" },
812         { "bps_rd_max",     "throttling.bps-read-max" },
813         { "bps_wr_max",     "throttling.bps-write-max" },
814 
815         { "iops_size",      "throttling.iops-size" },
816 
817         { "group",          "throttling.group" },
818 
819         { "readonly",       BDRV_OPT_READ_ONLY },
820     };
821 
822     for (i = 0; i < ARRAY_SIZE(opt_renames); i++) {
823         if (!qemu_opt_rename(all_opts, opt_renames[i].from,
824                              opt_renames[i].to, errp)) {
825             return NULL;
826         }
827     }
828 
829     value = qemu_opt_get(all_opts, "cache");
830     if (value) {
831         int flags = 0;
832         bool writethrough;
833 
834         if (bdrv_parse_cache_mode(value, &flags, &writethrough) != 0) {
835             error_setg(errp, "invalid cache option");
836             return NULL;
837         }
838 
839         /* Specific options take precedence */
840         if (!qemu_opt_get(all_opts, BDRV_OPT_CACHE_WB)) {
841             qemu_opt_set_bool(all_opts, BDRV_OPT_CACHE_WB,
842                               !writethrough, &error_abort);
843         }
844         if (!qemu_opt_get(all_opts, BDRV_OPT_CACHE_DIRECT)) {
845             qemu_opt_set_bool(all_opts, BDRV_OPT_CACHE_DIRECT,
846                               !!(flags & BDRV_O_NOCACHE), &error_abort);
847         }
848         if (!qemu_opt_get(all_opts, BDRV_OPT_CACHE_NO_FLUSH)) {
849             qemu_opt_set_bool(all_opts, BDRV_OPT_CACHE_NO_FLUSH,
850                               !!(flags & BDRV_O_NO_FLUSH), &error_abort);
851         }
852         qemu_opt_unset(all_opts, "cache");
853     }
854 
855     /* Get a QDict for processing the options */
856     bs_opts = qdict_new();
857     qemu_opts_to_qdict(all_opts, bs_opts);
858 
859     legacy_opts = qemu_opts_create(&qemu_legacy_drive_opts, NULL, 0,
860                                    &error_abort);
861     if (!qemu_opts_absorb_qdict(legacy_opts, bs_opts, errp)) {
862         goto fail;
863     }
864 
865     /* Media type */
866     value = qemu_opt_get(legacy_opts, "media");
867     if (value) {
868         if (!strcmp(value, "disk")) {
869             media = MEDIA_DISK;
870         } else if (!strcmp(value, "cdrom")) {
871             media = MEDIA_CDROM;
872             read_only = true;
873         } else {
874             error_setg(errp, "'%s' invalid media", value);
875             goto fail;
876         }
877     }
878 
879     /* copy-on-read is disabled with a warning for read-only devices */
880     read_only |= qemu_opt_get_bool(legacy_opts, BDRV_OPT_READ_ONLY, false);
881     copy_on_read = qemu_opt_get_bool(legacy_opts, "copy-on-read", false);
882 
883     if (read_only && copy_on_read) {
884         warn_report("disabling copy-on-read on read-only drive");
885         copy_on_read = false;
886     }
887 
888     qdict_put_str(bs_opts, BDRV_OPT_READ_ONLY, read_only ? "on" : "off");
889     qdict_put_str(bs_opts, "copy-on-read", copy_on_read ? "on" : "off");
890 
891     /* Controller type */
892     value = qemu_opt_get(legacy_opts, "if");
893     if (value) {
894         for (type = 0;
895              type < IF_COUNT && strcmp(value, if_name[type]);
896              type++) {
897         }
898         if (type == IF_COUNT) {
899             error_setg(errp, "unsupported bus type '%s'", value);
900             goto fail;
901         }
902     } else {
903         type = block_default_type;
904     }
905 
906     /* Device address specified by bus/unit or index.
907      * If none was specified, try to find the first free one. */
908     bus_id  = qemu_opt_get_number(legacy_opts, "bus", 0);
909     unit_id = qemu_opt_get_number(legacy_opts, "unit", -1);
910     index   = qemu_opt_get_number(legacy_opts, "index", -1);
911 
912     max_devs = if_max_devs[type];
913 
914     if (index != -1) {
915         if (bus_id != 0 || unit_id != -1) {
916             error_setg(errp, "index cannot be used with bus and unit");
917             goto fail;
918         }
919         bus_id = drive_index_to_bus_id(type, index);
920         unit_id = drive_index_to_unit_id(type, index);
921     }
922 
923     if (unit_id == -1) {
924        unit_id = 0;
925        while (drive_get(type, bus_id, unit_id) != NULL) {
926            unit_id++;
927            if (max_devs && unit_id >= max_devs) {
928                unit_id -= max_devs;
929                bus_id++;
930            }
931        }
932     }
933 
934     if (max_devs && unit_id >= max_devs) {
935         error_setg(errp, "unit %d too big (max is %d)", unit_id, max_devs - 1);
936         goto fail;
937     }
938 
939     if (drive_get(type, bus_id, unit_id) != NULL) {
940         error_setg(errp, "drive with bus=%d, unit=%d (index=%d) exists",
941                    bus_id, unit_id, index);
942         goto fail;
943     }
944 
945     /* no id supplied -> create one */
946     if (qemu_opts_id(all_opts) == NULL) {
947         char *new_id;
948         const char *mediastr = "";
949         if (type == IF_IDE || type == IF_SCSI) {
950             mediastr = (media == MEDIA_CDROM) ? "-cd" : "-hd";
951         }
952         if (max_devs) {
953             new_id = g_strdup_printf("%s%i%s%i", if_name[type], bus_id,
954                                      mediastr, unit_id);
955         } else {
956             new_id = g_strdup_printf("%s%s%i", if_name[type],
957                                      mediastr, unit_id);
958         }
959         qdict_put_str(bs_opts, "id", new_id);
960         g_free(new_id);
961     }
962 
963     /* Add virtio block device */
964     if (type == IF_VIRTIO) {
965         QemuOpts *devopts;
966         devopts = qemu_opts_create(qemu_find_opts("device"), NULL, 0,
967                                    &error_abort);
968         qemu_opt_set(devopts, "driver", "virtio-blk", &error_abort);
969         qemu_opt_set(devopts, "drive", qdict_get_str(bs_opts, "id"),
970                      &error_abort);
971     } else if (type == IF_XEN) {
972         QemuOpts *devopts;
973         devopts = qemu_opts_create(qemu_find_opts("device"), NULL, 0,
974                                    &error_abort);
975         qemu_opt_set(devopts, "driver",
976                      (media == MEDIA_CDROM) ? "xen-cdrom" : "xen-disk",
977                      &error_abort);
978         qemu_opt_set(devopts, "drive", qdict_get_str(bs_opts, "id"),
979                      &error_abort);
980     }
981 
982     filename = qemu_opt_get(legacy_opts, "file");
983 
984     /* Check werror/rerror compatibility with if=... */
985     werror = qemu_opt_get(legacy_opts, "werror");
986     if (werror != NULL) {
987         if (type != IF_IDE && type != IF_SCSI && type != IF_VIRTIO &&
988             type != IF_NONE) {
989             error_setg(errp, "werror is not supported by this bus type");
990             goto fail;
991         }
992         qdict_put_str(bs_opts, "werror", werror);
993     }
994 
995     rerror = qemu_opt_get(legacy_opts, "rerror");
996     if (rerror != NULL) {
997         if (type != IF_IDE && type != IF_VIRTIO && type != IF_SCSI &&
998             type != IF_NONE) {
999             error_setg(errp, "rerror is not supported by this bus type");
1000             goto fail;
1001         }
1002         qdict_put_str(bs_opts, "rerror", rerror);
1003     }
1004 
1005     /* Actual block device init: Functionality shared with blockdev-add */
1006     blk = blockdev_init(filename, bs_opts, errp);
1007     bs_opts = NULL;
1008     if (!blk) {
1009         goto fail;
1010     }
1011 
1012     /* Create legacy DriveInfo */
1013     dinfo = g_malloc0(sizeof(*dinfo));
1014     dinfo->opts = all_opts;
1015 
1016     dinfo->type = type;
1017     dinfo->bus = bus_id;
1018     dinfo->unit = unit_id;
1019 
1020     blk_set_legacy_dinfo(blk, dinfo);
1021 
1022     switch(type) {
1023     case IF_IDE:
1024     case IF_SCSI:
1025     case IF_XEN:
1026     case IF_NONE:
1027         dinfo->media_cd = media == MEDIA_CDROM;
1028         break;
1029     default:
1030         break;
1031     }
1032 
1033 fail:
1034     qemu_opts_del(legacy_opts);
1035     qobject_unref(bs_opts);
1036     return dinfo;
1037 }
1038 
1039 static BlockDriverState *qmp_get_root_bs(const char *name, Error **errp)
1040 {
1041     BlockDriverState *bs;
1042 
1043     GRAPH_RDLOCK_GUARD_MAINLOOP();
1044 
1045     bs = bdrv_lookup_bs(name, name, errp);
1046     if (bs == NULL) {
1047         return NULL;
1048     }
1049 
1050     if (!bdrv_is_root_node(bs)) {
1051         error_setg(errp, "Need a root block node");
1052         return NULL;
1053     }
1054 
1055     if (!bdrv_is_inserted(bs)) {
1056         error_setg(errp, "Device has no medium");
1057         bs = NULL;
1058     }
1059 
1060     return bs;
1061 }
1062 
1063 static void blockdev_do_action(TransactionAction *action, Error **errp)
1064 {
1065     TransactionActionList list;
1066 
1067     list.value = action;
1068     list.next = NULL;
1069     qmp_transaction(&list, NULL, errp);
1070 }
1071 
1072 void qmp_blockdev_snapshot_sync(const char *device, const char *node_name,
1073                                 const char *snapshot_file,
1074                                 const char *snapshot_node_name,
1075                                 const char *format,
1076                                 bool has_mode, NewImageMode mode, Error **errp)
1077 {
1078     BlockdevSnapshotSync snapshot = {
1079         .device = (char *) device,
1080         .node_name = (char *) node_name,
1081         .snapshot_file = (char *) snapshot_file,
1082         .snapshot_node_name = (char *) snapshot_node_name,
1083         .format = (char *) format,
1084         .has_mode = has_mode,
1085         .mode = mode,
1086     };
1087     TransactionAction action = {
1088         .type = TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC,
1089         .u.blockdev_snapshot_sync.data = &snapshot,
1090     };
1091     blockdev_do_action(&action, errp);
1092 }
1093 
1094 void qmp_blockdev_snapshot(const char *node, const char *overlay,
1095                            Error **errp)
1096 {
1097     BlockdevSnapshot snapshot_data = {
1098         .node = (char *) node,
1099         .overlay = (char *) overlay
1100     };
1101     TransactionAction action = {
1102         .type = TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT,
1103         .u.blockdev_snapshot.data = &snapshot_data,
1104     };
1105     blockdev_do_action(&action, errp);
1106 }
1107 
1108 void qmp_blockdev_snapshot_internal_sync(const char *device,
1109                                          const char *name,
1110                                          Error **errp)
1111 {
1112     BlockdevSnapshotInternal snapshot = {
1113         .device = (char *) device,
1114         .name = (char *) name
1115     };
1116     TransactionAction action = {
1117         .type = TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC,
1118         .u.blockdev_snapshot_internal_sync.data = &snapshot,
1119     };
1120     blockdev_do_action(&action, errp);
1121 }
1122 
1123 SnapshotInfo *qmp_blockdev_snapshot_delete_internal_sync(const char *device,
1124                                                          const char *id,
1125                                                          const char *name,
1126                                                          Error **errp)
1127 {
1128     BlockDriverState *bs;
1129     QEMUSnapshotInfo sn;
1130     Error *local_err = NULL;
1131     SnapshotInfo *info = NULL;
1132     int ret;
1133 
1134     GLOBAL_STATE_CODE();
1135 
1136     bdrv_drain_all_begin();
1137     bdrv_graph_rdlock_main_loop();
1138 
1139     bs = qmp_get_root_bs(device, errp);
1140     if (!bs) {
1141         goto error;
1142     }
1143 
1144     if (!id && !name) {
1145         error_setg(errp, "Name or id must be provided");
1146         goto error;
1147     }
1148 
1149     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_INTERNAL_SNAPSHOT_DELETE, errp)) {
1150         goto error;
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 error;
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 error;
1164     }
1165 
1166     bdrv_snapshot_delete(bs, id, name, &local_err);
1167     if (local_err) {
1168         error_propagate(errp, local_err);
1169         goto error;
1170     }
1171 
1172     info = g_new0(SnapshotInfo, 1);
1173     info->id = g_strdup(sn.id_str);
1174     info->name = g_strdup(sn.name);
1175     info->date_nsec = sn.date_nsec;
1176     info->date_sec = sn.date_sec;
1177     info->vm_state_size = sn.vm_state_size;
1178     info->vm_clock_nsec = sn.vm_clock_nsec % 1000000000;
1179     info->vm_clock_sec = sn.vm_clock_nsec / 1000000000;
1180     if (sn.icount != -1ULL) {
1181         info->icount = sn.icount;
1182         info->has_icount = true;
1183     }
1184 
1185 error:
1186     bdrv_graph_rdunlock_main_loop();
1187     bdrv_drain_all_end();
1188     return info;
1189 }
1190 
1191 /* internal snapshot private data */
1192 typedef struct InternalSnapshotState {
1193     BlockDriverState *bs;
1194     QEMUSnapshotInfo sn;
1195     bool created;
1196 } InternalSnapshotState;
1197 
1198 static void internal_snapshot_abort(void *opaque);
1199 static void internal_snapshot_clean(void *opaque);
1200 TransactionActionDrv internal_snapshot_drv = {
1201     .abort = internal_snapshot_abort,
1202     .clean = internal_snapshot_clean,
1203 };
1204 
1205 static void internal_snapshot_action(BlockdevSnapshotInternal *internal,
1206                                      Transaction *tran, Error **errp)
1207 {
1208     Error *local_err = NULL;
1209     const char *device;
1210     const char *name;
1211     BlockDriverState *bs, *check_bs;
1212     QEMUSnapshotInfo old_sn, *sn;
1213     bool ret;
1214     int64_t rt;
1215     InternalSnapshotState *state = g_new0(InternalSnapshotState, 1);
1216     int ret1;
1217 
1218     GLOBAL_STATE_CODE();
1219     bdrv_graph_rdlock_main_loop();
1220 
1221     tran_add(tran, &internal_snapshot_drv, state);
1222 
1223     device = internal->device;
1224     name = internal->name;
1225 
1226     bs = qmp_get_root_bs(device, errp);
1227     if (!bs) {
1228         bdrv_graph_rdunlock_main_loop();
1229         return;
1230     }
1231 
1232     state->bs = bs;
1233 
1234     /* Need to drain while unlocked. */
1235     bdrv_graph_rdunlock_main_loop();
1236     /* Paired with .clean() */
1237     bdrv_drained_begin(bs);
1238 
1239     GRAPH_RDLOCK_GUARD_MAINLOOP();
1240 
1241     /* Make sure the root bs did not change with the drain. */
1242     check_bs = qmp_get_root_bs(device, errp);
1243     if (bs != check_bs) {
1244         if (check_bs) {
1245             error_setg(errp, "Block node of device '%s' unexpectedly changed",
1246                        device);
1247         } /* else errp is already set */
1248         return;
1249     }
1250 
1251     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_INTERNAL_SNAPSHOT, errp)) {
1252         return;
1253     }
1254 
1255     if (bdrv_is_read_only(bs)) {
1256         error_setg(errp, "Device '%s' is read only", device);
1257         return;
1258     }
1259 
1260     if (!bdrv_can_snapshot(bs)) {
1261         error_setg(errp, "Block format '%s' used by device '%s' "
1262                    "does not support internal snapshots",
1263                    bs->drv->format_name, device);
1264         return;
1265     }
1266 
1267     if (!strlen(name)) {
1268         error_setg(errp, "Name is empty");
1269         return;
1270     }
1271 
1272     /* check whether a snapshot with name exist */
1273     ret = bdrv_snapshot_find_by_id_and_name(bs, NULL, name, &old_sn,
1274                                             &local_err);
1275     if (local_err) {
1276         error_propagate(errp, local_err);
1277         return;
1278     } else if (ret) {
1279         error_setg(errp,
1280                    "Snapshot with name '%s' already exists on device '%s'",
1281                    name, device);
1282         return;
1283     }
1284 
1285     /* 3. take the snapshot */
1286     sn = &state->sn;
1287     pstrcpy(sn->name, sizeof(sn->name), name);
1288     rt = g_get_real_time();
1289     sn->date_sec = rt / G_USEC_PER_SEC;
1290     sn->date_nsec = (rt % G_USEC_PER_SEC) * 1000;
1291     sn->vm_clock_nsec = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL);
1292     if (replay_mode != REPLAY_MODE_NONE) {
1293         sn->icount = replay_get_current_icount();
1294     } else {
1295         sn->icount = -1ULL;
1296     }
1297 
1298     ret1 = bdrv_snapshot_create(bs, sn);
1299     if (ret1 < 0) {
1300         error_setg_errno(errp, -ret1,
1301                          "Failed to create snapshot '%s' on device '%s'",
1302                          name, device);
1303         return;
1304     }
1305 
1306     /* 4. succeed, mark a snapshot is created */
1307     state->created = true;
1308 }
1309 
1310 static void internal_snapshot_abort(void *opaque)
1311 {
1312     InternalSnapshotState *state = opaque;
1313     BlockDriverState *bs = state->bs;
1314     QEMUSnapshotInfo *sn = &state->sn;
1315     Error *local_error = NULL;
1316 
1317     GLOBAL_STATE_CODE();
1318 
1319     if (!state->created) {
1320         return;
1321     }
1322 
1323     bdrv_drain_all_begin();
1324     bdrv_graph_rdlock_main_loop();
1325 
1326     if (bdrv_snapshot_delete(bs, sn->id_str, sn->name, &local_error) < 0) {
1327         error_reportf_err(local_error,
1328                           "Failed to delete snapshot with id '%s' and "
1329                           "name '%s' on device '%s' in abort: ",
1330                           sn->id_str, sn->name,
1331                           bdrv_get_device_name(bs));
1332     }
1333     bdrv_graph_rdunlock_main_loop();
1334     bdrv_drain_all_end();
1335 }
1336 
1337 static void internal_snapshot_clean(void *opaque)
1338 {
1339     g_autofree InternalSnapshotState *state = opaque;
1340 
1341     if (!state->bs) {
1342         return;
1343     }
1344 
1345     bdrv_drained_end(state->bs);
1346 }
1347 
1348 /* external snapshot private data */
1349 typedef struct ExternalSnapshotState {
1350     BlockDriverState *old_bs;
1351     BlockDriverState *new_bs;
1352     bool overlay_appended;
1353 } ExternalSnapshotState;
1354 
1355 static void external_snapshot_commit(void *opaque);
1356 static void external_snapshot_abort(void *opaque);
1357 static void external_snapshot_clean(void *opaque);
1358 TransactionActionDrv external_snapshot_drv = {
1359     .commit = external_snapshot_commit,
1360     .abort = external_snapshot_abort,
1361     .clean = external_snapshot_clean,
1362 };
1363 
1364 static void external_snapshot_action(TransactionAction *action,
1365                                      Transaction *tran, Error **errp)
1366 {
1367     int ret;
1368     int flags = 0;
1369     QDict *options = NULL;
1370     Error *local_err = NULL;
1371     /* Device and node name of the image to generate the snapshot from */
1372     const char *device;
1373     const char *node_name;
1374     /* Reference to the new image (for 'blockdev-snapshot') */
1375     const char *snapshot_ref;
1376     /* File name of the new image (for 'blockdev-snapshot-sync') */
1377     const char *new_image_file;
1378     ExternalSnapshotState *state = g_new0(ExternalSnapshotState, 1);
1379     uint64_t perm, shared;
1380     BlockDriverState *check_bs;
1381 
1382     /* TODO We'll eventually have to take a writer lock in this function */
1383     bdrv_graph_rdlock_main_loop();
1384 
1385     tran_add(tran, &external_snapshot_drv, state);
1386 
1387     /* 'blockdev-snapshot' and 'blockdev-snapshot-sync' have similar
1388      * purpose but a different set of parameters */
1389     switch (action->type) {
1390     case TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT:
1391         {
1392             BlockdevSnapshot *s = action->u.blockdev_snapshot.data;
1393             device = s->node;
1394             node_name = s->node;
1395             new_image_file = NULL;
1396             snapshot_ref = s->overlay;
1397         }
1398         break;
1399     case TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC:
1400         {
1401             BlockdevSnapshotSync *s = action->u.blockdev_snapshot_sync.data;
1402             device = s->device;
1403             node_name = s->node_name;
1404             new_image_file = s->snapshot_file;
1405             snapshot_ref = NULL;
1406         }
1407         break;
1408     default:
1409         g_assert_not_reached();
1410     }
1411 
1412     /* start processing */
1413 
1414     state->old_bs = bdrv_lookup_bs(device, node_name, errp);
1415     if (!state->old_bs) {
1416         bdrv_graph_rdunlock_main_loop();
1417         return;
1418     }
1419 
1420     /* Need to drain while unlocked. */
1421     bdrv_graph_rdunlock_main_loop();
1422     /* Paired with .clean() */
1423     bdrv_drained_begin(state->old_bs);
1424     GRAPH_RDLOCK_GUARD_MAINLOOP();
1425 
1426     /* Make sure the associated bs did not change with the drain. */
1427     check_bs = bdrv_lookup_bs(device, node_name, errp);
1428     if (state->old_bs != check_bs) {
1429         if (check_bs) {
1430             error_setg(errp, "Block node of device '%s' unexpectedly changed",
1431                        device);
1432         } /* else errp is already set */
1433         return;
1434     }
1435 
1436     if (!bdrv_is_inserted(state->old_bs)) {
1437         error_setg(errp, "Device '%s' has no medium",
1438                    bdrv_get_device_or_node_name(state->old_bs));
1439         return;
1440     }
1441 
1442     if (bdrv_op_is_blocked(state->old_bs,
1443                            BLOCK_OP_TYPE_EXTERNAL_SNAPSHOT, errp)) {
1444         return;
1445     }
1446 
1447     if (!bdrv_is_read_only(state->old_bs)) {
1448         ret = bdrv_flush(state->old_bs);
1449         if (ret < 0) {
1450             error_setg_errno(errp, -ret, "Write to node '%s' failed",
1451                              bdrv_get_device_or_node_name(state->old_bs));
1452             return;
1453         }
1454     }
1455 
1456     if (action->type == TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC) {
1457         BlockdevSnapshotSync *s = action->u.blockdev_snapshot_sync.data;
1458         const char *format = s->format ?: "qcow2";
1459         enum NewImageMode mode;
1460         const char *snapshot_node_name = s->snapshot_node_name;
1461 
1462         if (node_name && !snapshot_node_name) {
1463             error_setg(errp, "New overlay node-name missing");
1464             return;
1465         }
1466 
1467         if (snapshot_node_name &&
1468             bdrv_lookup_bs(snapshot_node_name, snapshot_node_name, NULL)) {
1469             error_setg(errp, "New overlay node-name already in use");
1470             return;
1471         }
1472 
1473         flags = state->old_bs->open_flags;
1474         flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_COPY_ON_READ);
1475         flags |= BDRV_O_NO_BACKING;
1476 
1477         /* create new image w/backing file */
1478         mode = s->has_mode ? s->mode : NEW_IMAGE_MODE_ABSOLUTE_PATHS;
1479         if (mode != NEW_IMAGE_MODE_EXISTING) {
1480             int64_t size = bdrv_getlength(state->old_bs);
1481             if (size < 0) {
1482                 error_setg_errno(errp, -size, "bdrv_getlength failed");
1483                 return;
1484             }
1485             bdrv_refresh_filename(state->old_bs);
1486 
1487             bdrv_img_create(new_image_file, format,
1488                             state->old_bs->filename,
1489                             state->old_bs->drv->format_name,
1490                             NULL, size, flags, false, &local_err);
1491 
1492             if (local_err) {
1493                 error_propagate(errp, local_err);
1494                 return;
1495             }
1496         }
1497 
1498         options = qdict_new();
1499         if (snapshot_node_name) {
1500             qdict_put_str(options, "node-name", snapshot_node_name);
1501         }
1502         qdict_put_str(options, "driver", format);
1503     }
1504 
1505     state->new_bs = bdrv_open(new_image_file, snapshot_ref, options, flags,
1506                               errp);
1507 
1508     /* We will manually add the backing_hd field to the bs later */
1509     if (!state->new_bs) {
1510         return;
1511     }
1512 
1513     /*
1514      * Allow attaching a backing file to an overlay that's already in use only
1515      * if the parents don't assume that they are already seeing a valid image.
1516      * (Specifically, allow it as a mirror target, which is write-only access.)
1517      */
1518     bdrv_get_cumulative_perm(state->new_bs, &perm, &shared);
1519     if (perm & BLK_PERM_CONSISTENT_READ) {
1520         error_setg(errp, "The overlay is already in use");
1521         return;
1522     }
1523 
1524     if (state->new_bs->drv->is_filter) {
1525         error_setg(errp, "Filters cannot be used as overlays");
1526         return;
1527     }
1528 
1529     if (bdrv_cow_child(state->new_bs)) {
1530         error_setg(errp, "The overlay already has a backing image");
1531         return;
1532     }
1533 
1534     if (!state->new_bs->drv->supports_backing) {
1535         error_setg(errp, "The overlay does not support backing images");
1536         return;
1537     }
1538 
1539     /*
1540      * Older QEMU versions have allowed adding an active parent node to an
1541      * inactive child node. This is unsafe in the general case, but there is an
1542      * important use case, which is taking a VM snapshot with migration to file
1543      * and then adding an external snapshot while the VM is still stopped and
1544      * images are inactive. Requiring the user to explicitly create the overlay
1545      * as inactive would break compatibility, so just do it automatically here
1546      * to keep this working.
1547      */
1548     if (bdrv_is_inactive(state->old_bs) && !bdrv_is_inactive(state->new_bs)) {
1549         ret = bdrv_inactivate(state->new_bs, errp);
1550         if (ret < 0) {
1551             return;
1552         }
1553     }
1554 
1555     ret = bdrv_append(state->new_bs, state->old_bs, errp);
1556     if (ret < 0) {
1557         return;
1558     }
1559     state->overlay_appended = true;
1560 }
1561 
1562 static void external_snapshot_commit(void *opaque)
1563 {
1564     ExternalSnapshotState *state = opaque;
1565 
1566     /* We don't need (or want) to use the transactional
1567      * bdrv_reopen_multiple() across all the entries at once, because we
1568      * don't want to abort all of them if one of them fails the reopen */
1569     if (!qatomic_read(&state->old_bs->copy_on_read)) {
1570         bdrv_reopen_set_read_only(state->old_bs, true, NULL);
1571     }
1572 }
1573 
1574 static void external_snapshot_abort(void *opaque)
1575 {
1576     ExternalSnapshotState *state = opaque;
1577     if (state->new_bs) {
1578         if (state->overlay_appended) {
1579             AioContext *aio_context;
1580             AioContext *tmp_context;
1581             int ret;
1582 
1583             bdrv_graph_wrlock_drained();
1584 
1585             aio_context = bdrv_get_aio_context(state->old_bs);
1586 
1587             /*
1588              * Note that state->old_bs would not disappear during the
1589              * write-locked section, because the unref from
1590              * bdrv_set_backing_hd() only happens at the end of the write-locked
1591              * section. However, just be explicit about keeping a reference and
1592              * don't rely on that implicit detail.
1593              */
1594             bdrv_ref(state->old_bs);
1595             bdrv_set_backing_hd(state->new_bs, NULL, &error_abort);
1596 
1597             /*
1598              * The call to bdrv_set_backing_hd() above returns state->old_bs to
1599              * the main AioContext. As we're still going to be using it, return
1600              * it to the AioContext it was before.
1601              */
1602             tmp_context = bdrv_get_aio_context(state->old_bs);
1603             if (aio_context != tmp_context) {
1604                 ret = bdrv_try_change_aio_context_locked(state->old_bs,
1605                                                          aio_context, NULL,
1606                                                          NULL);
1607                 assert(ret == 0);
1608             }
1609 
1610             bdrv_replace_node(state->new_bs, state->old_bs, &error_abort);
1611             bdrv_graph_wrunlock();
1612 
1613             bdrv_unref(state->old_bs); /* bdrv_replace_node() ref'ed old_bs */
1614         }
1615     }
1616 }
1617 
1618 static void external_snapshot_clean(void *opaque)
1619 {
1620     g_autofree ExternalSnapshotState *state = opaque;
1621 
1622     if (!state->old_bs) {
1623         return;
1624     }
1625 
1626     bdrv_drained_end(state->old_bs);
1627     bdrv_unref(state->new_bs);
1628 }
1629 
1630 typedef struct DriveBackupState {
1631     BlockDriverState *bs;
1632     BlockJob *job;
1633 } DriveBackupState;
1634 
1635 static BlockJob *do_backup_common(BackupCommon *backup,
1636                                   BlockDriverState *bs,
1637                                   BlockDriverState *target_bs,
1638                                   AioContext *aio_context,
1639                                   JobTxn *txn, Error **errp);
1640 
1641 static void drive_backup_commit(void *opaque);
1642 static void drive_backup_abort(void *opaque);
1643 static void drive_backup_clean(void *opaque);
1644 TransactionActionDrv drive_backup_drv = {
1645     .commit = drive_backup_commit,
1646     .abort = drive_backup_abort,
1647     .clean = drive_backup_clean,
1648 };
1649 
1650 static void drive_backup_action(DriveBackup *backup,
1651                                 JobTxn *block_job_txn,
1652                                 Transaction *tran, Error **errp)
1653 {
1654     DriveBackupState *state = g_new0(DriveBackupState, 1);
1655     BlockDriverState *bs;
1656     BlockDriverState *target_bs;
1657     BlockDriverState *source = NULL;
1658     AioContext *aio_context;
1659     const char *format;
1660     QDict *options;
1661     Error *local_err = NULL;
1662     int flags;
1663     int64_t size;
1664     bool set_backing_hd = false;
1665     int ret;
1666 
1667     GLOBAL_STATE_CODE();
1668 
1669     tran_add(tran, &drive_backup_drv, state);
1670 
1671     if (!backup->has_mode) {
1672         backup->mode = NEW_IMAGE_MODE_ABSOLUTE_PATHS;
1673     }
1674 
1675     bs = bdrv_lookup_bs(backup->device, backup->device, errp);
1676     if (!bs) {
1677         return;
1678     }
1679 
1680     if (!bs->drv) {
1681         error_setg(errp, "Device has no medium");
1682         return;
1683     }
1684 
1685     aio_context = bdrv_get_aio_context(bs);
1686 
1687     state->bs = bs;
1688     /* Paired with .clean() */
1689     bdrv_drained_begin(bs);
1690 
1691     format = backup->format;
1692     if (!format && backup->mode != NEW_IMAGE_MODE_EXISTING) {
1693         format = bs->drv->format_name;
1694     }
1695 
1696     /* Early check to avoid creating target */
1697     bdrv_graph_rdlock_main_loop();
1698     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_BACKUP_SOURCE, errp)) {
1699         bdrv_graph_rdunlock_main_loop();
1700         return;
1701     }
1702 
1703     flags = bs->open_flags | BDRV_O_RDWR;
1704 
1705     /*
1706      * See if we have a backing HD we can use to create our new image
1707      * on top of.
1708      */
1709     if (backup->sync == MIRROR_SYNC_MODE_TOP) {
1710         /*
1711          * Backup will not replace the source by the target, so none
1712          * of the filters skipped here will be removed (in contrast to
1713          * mirror).  Therefore, we can skip all of them when looking
1714          * for the first COW relationship.
1715          */
1716         source = bdrv_cow_bs(bdrv_skip_filters(bs));
1717         if (!source) {
1718             backup->sync = MIRROR_SYNC_MODE_FULL;
1719         }
1720     }
1721     if (backup->sync == MIRROR_SYNC_MODE_NONE) {
1722         source = bs;
1723         flags |= BDRV_O_NO_BACKING;
1724         set_backing_hd = true;
1725     }
1726     bdrv_graph_rdunlock_main_loop();
1727 
1728     size = bdrv_getlength(bs);
1729     if (size < 0) {
1730         error_setg_errno(errp, -size, "bdrv_getlength failed");
1731         return;
1732     }
1733 
1734     if (backup->mode != NEW_IMAGE_MODE_EXISTING) {
1735         assert(format);
1736         if (source) {
1737             /* Implicit filters should not appear in the filename */
1738             BlockDriverState *explicit_backing;
1739 
1740             bdrv_graph_rdlock_main_loop();
1741             explicit_backing = bdrv_skip_implicit_filters(source);
1742             bdrv_refresh_filename(explicit_backing);
1743             bdrv_graph_rdunlock_main_loop();
1744 
1745             bdrv_img_create(backup->target, format,
1746                             explicit_backing->filename,
1747                             explicit_backing->drv->format_name, NULL,
1748                             size, flags, false, &local_err);
1749         } else {
1750             bdrv_img_create(backup->target, format, NULL, NULL, NULL,
1751                             size, flags, false, &local_err);
1752         }
1753     }
1754 
1755     if (local_err) {
1756         error_propagate(errp, local_err);
1757         return;
1758     }
1759 
1760     options = qdict_new();
1761     qdict_put_str(options, "discard", "unmap");
1762     qdict_put_str(options, "detect-zeroes", "unmap");
1763     if (format) {
1764         qdict_put_str(options, "driver", format);
1765     }
1766 
1767     target_bs = bdrv_open(backup->target, NULL, options, flags, errp);
1768     if (!target_bs) {
1769         return;
1770     }
1771 
1772     ret = bdrv_try_change_aio_context(target_bs, aio_context, NULL, errp);
1773     if (ret < 0) {
1774         bdrv_unref(target_bs);
1775         return;
1776     }
1777 
1778     if (set_backing_hd) {
1779         bdrv_graph_wrlock_drained();
1780         ret = bdrv_set_backing_hd(target_bs, source, errp);
1781         bdrv_graph_wrunlock();
1782         if (ret < 0) {
1783             goto unref;
1784         }
1785     }
1786 
1787     state->job = do_backup_common(qapi_DriveBackup_base(backup),
1788                                   bs, target_bs, aio_context,
1789                                   block_job_txn, errp);
1790 
1791 unref:
1792     bdrv_unref(target_bs);
1793 }
1794 
1795 static void drive_backup_commit(void *opaque)
1796 {
1797     DriveBackupState *state = opaque;
1798 
1799     assert(state->job);
1800     job_start(&state->job->job);
1801 }
1802 
1803 static void drive_backup_abort(void *opaque)
1804 {
1805     DriveBackupState *state = opaque;
1806 
1807     if (state->job) {
1808         job_cancel_sync(&state->job->job, true);
1809     }
1810 }
1811 
1812 static void drive_backup_clean(void *opaque)
1813 {
1814     g_autofree DriveBackupState *state = opaque;
1815 
1816     if (!state->bs) {
1817         return;
1818     }
1819 
1820     bdrv_drained_end(state->bs);
1821 }
1822 
1823 typedef struct BlockdevBackupState {
1824     BlockDriverState *bs;
1825     BlockJob *job;
1826 } BlockdevBackupState;
1827 
1828 static void blockdev_backup_commit(void *opaque);
1829 static void blockdev_backup_abort(void *opaque);
1830 static void blockdev_backup_clean(void *opaque);
1831 TransactionActionDrv blockdev_backup_drv = {
1832     .commit = blockdev_backup_commit,
1833     .abort = blockdev_backup_abort,
1834     .clean = blockdev_backup_clean,
1835 };
1836 
1837 static void blockdev_backup_action(BlockdevBackup *backup,
1838                                    JobTxn *block_job_txn,
1839                                    Transaction *tran, Error **errp)
1840 {
1841     BlockdevBackupState *state = g_new0(BlockdevBackupState, 1);
1842     BlockDriverState *bs;
1843     BlockDriverState *target_bs;
1844     AioContext *aio_context;
1845     int ret;
1846 
1847     tran_add(tran, &blockdev_backup_drv, state);
1848 
1849     bs = bdrv_lookup_bs(backup->device, backup->device, errp);
1850     if (!bs) {
1851         return;
1852     }
1853 
1854     target_bs = bdrv_lookup_bs(backup->target, backup->target, errp);
1855     if (!target_bs) {
1856         return;
1857     }
1858 
1859     /* Honor bdrv_try_change_aio_context() context acquisition requirements. */
1860     aio_context = bdrv_get_aio_context(bs);
1861 
1862     ret = bdrv_try_change_aio_context(target_bs, aio_context, NULL, errp);
1863     if (ret < 0) {
1864         return;
1865     }
1866 
1867     state->bs = bs;
1868 
1869     /* Paired with .clean() */
1870     bdrv_drained_begin(state->bs);
1871 
1872     state->job = do_backup_common(qapi_BlockdevBackup_base(backup),
1873                                   bs, target_bs, aio_context,
1874                                   block_job_txn, errp);
1875 }
1876 
1877 static void blockdev_backup_commit(void *opaque)
1878 {
1879     BlockdevBackupState *state = opaque;
1880 
1881     assert(state->job);
1882     job_start(&state->job->job);
1883 }
1884 
1885 static void blockdev_backup_abort(void *opaque)
1886 {
1887     BlockdevBackupState *state = opaque;
1888 
1889     if (state->job) {
1890         job_cancel_sync(&state->job->job, true);
1891     }
1892 }
1893 
1894 static void blockdev_backup_clean(void *opaque)
1895 {
1896     g_autofree BlockdevBackupState *state = opaque;
1897 
1898     if (!state->bs) {
1899         return;
1900     }
1901 
1902     bdrv_drained_end(state->bs);
1903 }
1904 
1905 typedef struct BlockDirtyBitmapState {
1906     BdrvDirtyBitmap *bitmap;
1907     BlockDriverState *bs;
1908     HBitmap *backup;
1909     bool was_enabled;
1910 } BlockDirtyBitmapState;
1911 
1912 static void block_dirty_bitmap_add_abort(void *opaque);
1913 TransactionActionDrv block_dirty_bitmap_add_drv = {
1914     .abort = block_dirty_bitmap_add_abort,
1915     .clean = g_free,
1916 };
1917 
1918 static void block_dirty_bitmap_add_action(BlockDirtyBitmapAdd *action,
1919                                           Transaction *tran, Error **errp)
1920 {
1921     Error *local_err = NULL;
1922     BlockDirtyBitmapState *state = g_new0(BlockDirtyBitmapState, 1);
1923 
1924     tran_add(tran, &block_dirty_bitmap_add_drv, state);
1925 
1926     /* AIO context taken and released within qmp_block_dirty_bitmap_add */
1927     qmp_block_dirty_bitmap_add(action->node, action->name,
1928                                action->has_granularity, action->granularity,
1929                                action->has_persistent, action->persistent,
1930                                action->has_disabled, action->disabled,
1931                                &local_err);
1932 
1933     if (!local_err) {
1934         state->bitmap = block_dirty_bitmap_lookup(action->node, action->name,
1935                                                   NULL, &error_abort);
1936     } else {
1937         error_propagate(errp, local_err);
1938     }
1939 }
1940 
1941 static void block_dirty_bitmap_add_abort(void *opaque)
1942 {
1943     BlockDirtyBitmapState *state = opaque;
1944 
1945     if (state->bitmap) {
1946         bdrv_release_dirty_bitmap(state->bitmap);
1947     }
1948 }
1949 
1950 static void block_dirty_bitmap_restore(void *opaque);
1951 static void block_dirty_bitmap_free_backup(void *opaque);
1952 TransactionActionDrv block_dirty_bitmap_clear_drv = {
1953     .abort = block_dirty_bitmap_restore,
1954     .commit = block_dirty_bitmap_free_backup,
1955     .clean = g_free,
1956 };
1957 
1958 static void block_dirty_bitmap_clear_action(BlockDirtyBitmap *action,
1959                                             Transaction *tran, Error **errp)
1960 {
1961     BlockDirtyBitmapState *state = g_new0(BlockDirtyBitmapState, 1);
1962 
1963     tran_add(tran, &block_dirty_bitmap_clear_drv, state);
1964 
1965     state->bitmap = block_dirty_bitmap_lookup(action->node,
1966                                               action->name,
1967                                               &state->bs,
1968                                               errp);
1969     if (!state->bitmap) {
1970         return;
1971     }
1972 
1973     if (bdrv_dirty_bitmap_check(state->bitmap, BDRV_BITMAP_DEFAULT, errp)) {
1974         return;
1975     }
1976 
1977     bdrv_clear_dirty_bitmap(state->bitmap, &state->backup);
1978 }
1979 
1980 static void block_dirty_bitmap_restore(void *opaque)
1981 {
1982     BlockDirtyBitmapState *state = opaque;
1983 
1984     if (state->backup) {
1985         bdrv_restore_dirty_bitmap(state->bitmap, state->backup);
1986     }
1987 }
1988 
1989 static void block_dirty_bitmap_free_backup(void *opaque)
1990 {
1991     BlockDirtyBitmapState *state = opaque;
1992 
1993     hbitmap_free(state->backup);
1994 }
1995 
1996 static void block_dirty_bitmap_enable_abort(void *opaque);
1997 TransactionActionDrv block_dirty_bitmap_enable_drv = {
1998     .abort = block_dirty_bitmap_enable_abort,
1999     .clean = g_free,
2000 };
2001 
2002 static void block_dirty_bitmap_enable_action(BlockDirtyBitmap *action,
2003                                              Transaction *tran, Error **errp)
2004 {
2005     BlockDirtyBitmapState *state = g_new0(BlockDirtyBitmapState, 1);
2006 
2007     tran_add(tran, &block_dirty_bitmap_enable_drv, state);
2008 
2009     state->bitmap = block_dirty_bitmap_lookup(action->node,
2010                                               action->name,
2011                                               NULL,
2012                                               errp);
2013     if (!state->bitmap) {
2014         return;
2015     }
2016 
2017     if (bdrv_dirty_bitmap_check(state->bitmap, BDRV_BITMAP_ALLOW_RO, errp)) {
2018         return;
2019     }
2020 
2021     state->was_enabled = bdrv_dirty_bitmap_enabled(state->bitmap);
2022     bdrv_enable_dirty_bitmap(state->bitmap);
2023 }
2024 
2025 static void block_dirty_bitmap_enable_abort(void *opaque)
2026 {
2027     BlockDirtyBitmapState *state = opaque;
2028 
2029     if (!state->was_enabled) {
2030         bdrv_disable_dirty_bitmap(state->bitmap);
2031     }
2032 }
2033 
2034 static void block_dirty_bitmap_disable_abort(void *opaque);
2035 TransactionActionDrv block_dirty_bitmap_disable_drv = {
2036     .abort = block_dirty_bitmap_disable_abort,
2037     .clean = g_free,
2038 };
2039 
2040 static void block_dirty_bitmap_disable_action(BlockDirtyBitmap *action,
2041                                               Transaction *tran, Error **errp)
2042 {
2043     BlockDirtyBitmapState *state = g_new0(BlockDirtyBitmapState, 1);
2044 
2045     tran_add(tran, &block_dirty_bitmap_disable_drv, state);
2046 
2047     state->bitmap = block_dirty_bitmap_lookup(action->node,
2048                                               action->name,
2049                                               NULL,
2050                                               errp);
2051     if (!state->bitmap) {
2052         return;
2053     }
2054 
2055     if (bdrv_dirty_bitmap_check(state->bitmap, BDRV_BITMAP_ALLOW_RO, errp)) {
2056         return;
2057     }
2058 
2059     state->was_enabled = bdrv_dirty_bitmap_enabled(state->bitmap);
2060     bdrv_disable_dirty_bitmap(state->bitmap);
2061 }
2062 
2063 static void block_dirty_bitmap_disable_abort(void *opaque)
2064 {
2065     BlockDirtyBitmapState *state = opaque;
2066 
2067     if (state->was_enabled) {
2068         bdrv_enable_dirty_bitmap(state->bitmap);
2069     }
2070 }
2071 
2072 TransactionActionDrv block_dirty_bitmap_merge_drv = {
2073     .commit = block_dirty_bitmap_free_backup,
2074     .abort = block_dirty_bitmap_restore,
2075     .clean = g_free,
2076 };
2077 
2078 static void block_dirty_bitmap_merge_action(BlockDirtyBitmapMerge *action,
2079                                             Transaction *tran, Error **errp)
2080 {
2081     BlockDirtyBitmapState *state = g_new0(BlockDirtyBitmapState, 1);
2082 
2083     tran_add(tran, &block_dirty_bitmap_merge_drv, state);
2084 
2085     state->bitmap = block_dirty_bitmap_merge(action->node, action->target,
2086                                              action->bitmaps, &state->backup,
2087                                              errp);
2088 }
2089 
2090 static void block_dirty_bitmap_remove_commit(void *opaque);
2091 static void block_dirty_bitmap_remove_abort(void *opaque);
2092 TransactionActionDrv block_dirty_bitmap_remove_drv = {
2093     .commit = block_dirty_bitmap_remove_commit,
2094     .abort = block_dirty_bitmap_remove_abort,
2095     .clean = g_free,
2096 };
2097 
2098 static void block_dirty_bitmap_remove_action(BlockDirtyBitmap *action,
2099                                              Transaction *tran, Error **errp)
2100 {
2101     BlockDirtyBitmapState *state = g_new0(BlockDirtyBitmapState, 1);
2102 
2103     tran_add(tran, &block_dirty_bitmap_remove_drv, state);
2104 
2105 
2106     state->bitmap = block_dirty_bitmap_remove(action->node, action->name,
2107                                               false, &state->bs, errp);
2108     if (state->bitmap) {
2109         bdrv_dirty_bitmap_skip_store(state->bitmap, true);
2110         bdrv_dirty_bitmap_set_busy(state->bitmap, true);
2111     }
2112 }
2113 
2114 static void block_dirty_bitmap_remove_abort(void *opaque)
2115 {
2116     BlockDirtyBitmapState *state = opaque;
2117 
2118     if (state->bitmap) {
2119         bdrv_dirty_bitmap_skip_store(state->bitmap, false);
2120         bdrv_dirty_bitmap_set_busy(state->bitmap, false);
2121     }
2122 }
2123 
2124 static void block_dirty_bitmap_remove_commit(void *opaque)
2125 {
2126     BlockDirtyBitmapState *state = opaque;
2127 
2128     bdrv_dirty_bitmap_set_busy(state->bitmap, false);
2129     bdrv_release_dirty_bitmap(state->bitmap);
2130 }
2131 
2132 static void abort_commit(void *opaque);
2133 TransactionActionDrv abort_drv = {
2134     .commit = abort_commit,
2135 };
2136 
2137 static void abort_action(Transaction *tran, Error **errp)
2138 {
2139     tran_add(tran, &abort_drv, NULL);
2140     error_setg(errp, "Transaction aborted using Abort action");
2141 }
2142 
2143 static void abort_commit(void *opaque)
2144 {
2145     g_assert_not_reached(); /* this action never succeeds */
2146 }
2147 
2148 static void transaction_action(TransactionAction *act, JobTxn *block_job_txn,
2149                                Transaction *tran, Error **errp)
2150 {
2151     switch (act->type) {
2152     case TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT:
2153     case TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC:
2154         external_snapshot_action(act, tran, errp);
2155         return;
2156     case TRANSACTION_ACTION_KIND_DRIVE_BACKUP:
2157         drive_backup_action(act->u.drive_backup.data,
2158                             block_job_txn, tran, errp);
2159         return;
2160     case TRANSACTION_ACTION_KIND_BLOCKDEV_BACKUP:
2161         blockdev_backup_action(act->u.blockdev_backup.data,
2162                                block_job_txn, tran, errp);
2163         return;
2164     case TRANSACTION_ACTION_KIND_ABORT:
2165         abort_action(tran, errp);
2166         return;
2167     case TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC:
2168         internal_snapshot_action(act->u.blockdev_snapshot_internal_sync.data,
2169                                  tran, errp);
2170         return;
2171     case TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_ADD:
2172         block_dirty_bitmap_add_action(act->u.block_dirty_bitmap_add.data,
2173                                       tran, errp);
2174         return;
2175     case TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_CLEAR:
2176         block_dirty_bitmap_clear_action(act->u.block_dirty_bitmap_clear.data,
2177                                         tran, errp);
2178         return;
2179     case TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_ENABLE:
2180         block_dirty_bitmap_enable_action(act->u.block_dirty_bitmap_enable.data,
2181                                          tran, errp);
2182         return;
2183     case TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_DISABLE:
2184         block_dirty_bitmap_disable_action(
2185                 act->u.block_dirty_bitmap_disable.data, tran, errp);
2186         return;
2187     case TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_MERGE:
2188         block_dirty_bitmap_merge_action(act->u.block_dirty_bitmap_merge.data,
2189                                         tran, errp);
2190         return;
2191     case TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_REMOVE:
2192         block_dirty_bitmap_remove_action(act->u.block_dirty_bitmap_remove.data,
2193                                          tran, errp);
2194         return;
2195     /*
2196      * Where are transactions for MIRROR, COMMIT and STREAM?
2197      * Although these blockjobs use transaction callbacks like the backup job,
2198      * these jobs do not necessarily adhere to transaction semantics.
2199      * These jobs may not fully undo all of their actions on abort, nor do they
2200      * necessarily work in transactions with more than one job in them.
2201      */
2202     case TRANSACTION_ACTION_KIND__MAX:
2203     default:
2204         g_assert_not_reached();
2205     };
2206 }
2207 
2208 
2209 /*
2210  * 'Atomic' group operations.  The operations are performed as a set, and if
2211  * any fail then we roll back all operations in the group.
2212  *
2213  * Always run under BQL.
2214  */
2215 void qmp_transaction(TransactionActionList *actions,
2216                      struct TransactionProperties *properties,
2217                      Error **errp)
2218 {
2219     TransactionActionList *act;
2220     JobTxn *block_job_txn = NULL;
2221     Error *local_err = NULL;
2222     Transaction *tran;
2223     ActionCompletionMode comp_mode =
2224         properties ? properties->completion_mode :
2225         ACTION_COMPLETION_MODE_INDIVIDUAL;
2226 
2227     GLOBAL_STATE_CODE();
2228 
2229     /* Does this transaction get canceled as a group on failure?
2230      * If not, we don't really need to make a JobTxn.
2231      */
2232     if (comp_mode != ACTION_COMPLETION_MODE_INDIVIDUAL) {
2233         for (act = actions; act; act = act->next) {
2234             TransactionActionKind type = act->value->type;
2235 
2236             if (type != TRANSACTION_ACTION_KIND_BLOCKDEV_BACKUP &&
2237                 type != TRANSACTION_ACTION_KIND_DRIVE_BACKUP)
2238             {
2239                 error_setg(errp,
2240                            "Action '%s' does not support transaction property "
2241                            "completion-mode = %s",
2242                            TransactionActionKind_str(type),
2243                            ActionCompletionMode_str(comp_mode));
2244                 return;
2245             }
2246         }
2247 
2248         block_job_txn = job_txn_new();
2249     }
2250 
2251     /* drain all i/o before any operations */
2252     bdrv_drain_all();
2253 
2254     tran = tran_new();
2255 
2256     /* We don't do anything in this loop that commits us to the operations */
2257     for (act = actions; act; act = act->next) {
2258         transaction_action(act->value, block_job_txn, tran, &local_err);
2259         if (local_err) {
2260             error_propagate(errp, local_err);
2261             goto delete_and_fail;
2262         }
2263     }
2264 
2265     tran_commit(tran);
2266 
2267     /* success */
2268     goto exit;
2269 
2270 delete_and_fail:
2271     /* failure, and it is all-or-none; roll back all operations */
2272     tran_abort(tran);
2273 exit:
2274     job_txn_unref(block_job_txn);
2275 }
2276 
2277 BlockDirtyBitmapSha256 *qmp_x_debug_block_dirty_bitmap_sha256(const char *node,
2278                                                               const char *name,
2279                                                               Error **errp)
2280 {
2281     BdrvDirtyBitmap *bitmap;
2282     BlockDriverState *bs;
2283     BlockDirtyBitmapSha256 *ret = NULL;
2284     char *sha256;
2285 
2286     bitmap = block_dirty_bitmap_lookup(node, name, &bs, errp);
2287     if (!bitmap || !bs) {
2288         return NULL;
2289     }
2290 
2291     sha256 = bdrv_dirty_bitmap_sha256(bitmap, errp);
2292     if (sha256 == NULL) {
2293         return NULL;
2294     }
2295 
2296     ret = g_new(BlockDirtyBitmapSha256, 1);
2297     ret->sha256 = sha256;
2298 
2299     return ret;
2300 }
2301 
2302 void coroutine_fn qmp_block_resize(const char *device, const char *node_name,
2303                                    int64_t size, Error **errp)
2304 {
2305     Error *local_err = NULL;
2306     BlockBackend *blk;
2307     BlockDriverState *bs;
2308     AioContext *old_ctx;
2309 
2310     bs = bdrv_lookup_bs(device, node_name, &local_err);
2311     if (local_err) {
2312         error_propagate(errp, local_err);
2313         return;
2314     }
2315 
2316     if (size < 0) {
2317         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "size", "a >0 size");
2318         return;
2319     }
2320 
2321     bdrv_graph_co_rdlock();
2322     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_RESIZE, errp)) {
2323         bdrv_graph_co_rdunlock();
2324         return;
2325     }
2326     bdrv_graph_co_rdunlock();
2327 
2328     blk = blk_co_new_with_bs(bs, BLK_PERM_RESIZE, BLK_PERM_ALL, errp);
2329     if (!blk) {
2330         return;
2331     }
2332 
2333     bdrv_drained_begin(bs);
2334 
2335     old_ctx = bdrv_co_enter(bs);
2336     blk_co_truncate(blk, size, false, PREALLOC_MODE_OFF, 0, errp);
2337     bdrv_co_leave(bs, old_ctx);
2338 
2339     bdrv_drained_end(bs);
2340     blk_co_unref(blk);
2341 }
2342 
2343 void qmp_block_stream(const char *job_id, const char *device,
2344                       const char *base,
2345                       const char *base_node,
2346                       const char *backing_file,
2347                       bool has_backing_mask_protocol,
2348                       bool backing_mask_protocol,
2349                       const char *bottom,
2350                       bool has_speed, int64_t speed,
2351                       bool has_on_error, BlockdevOnError on_error,
2352                       const char *filter_node_name,
2353                       bool has_auto_finalize, bool auto_finalize,
2354                       bool has_auto_dismiss, bool auto_dismiss,
2355                       Error **errp)
2356 {
2357     BlockDriverState *bs, *iter, *iter_end;
2358     BlockDriverState *base_bs = NULL;
2359     BlockDriverState *bottom_bs = NULL;
2360     AioContext *aio_context;
2361     Error *local_err = NULL;
2362     int job_flags = JOB_DEFAULT;
2363 
2364     GLOBAL_STATE_CODE();
2365 
2366     if (base && base_node) {
2367         error_setg(errp, "'base' and 'base-node' cannot be specified "
2368                    "at the same time");
2369         return;
2370     }
2371 
2372     if (base && bottom) {
2373         error_setg(errp, "'base' and 'bottom' cannot be specified "
2374                    "at the same time");
2375         return;
2376     }
2377 
2378     if (bottom && base_node) {
2379         error_setg(errp, "'bottom' and 'base-node' cannot be specified "
2380                    "at the same time");
2381         return;
2382     }
2383 
2384     if (!has_backing_mask_protocol) {
2385         backing_mask_protocol = false;
2386     }
2387 
2388     if (!has_on_error) {
2389         on_error = BLOCKDEV_ON_ERROR_REPORT;
2390     }
2391 
2392     bs = bdrv_lookup_bs(device, device, errp);
2393     if (!bs) {
2394         return;
2395     }
2396 
2397     aio_context = bdrv_get_aio_context(bs);
2398 
2399     bdrv_graph_rdlock_main_loop();
2400     if (base) {
2401         base_bs = bdrv_find_backing_image(bs, base);
2402         if (base_bs == NULL) {
2403             error_setg(errp, "Can't find '%s' in the backing chain", base);
2404             goto out_rdlock;
2405         }
2406         assert(bdrv_get_aio_context(base_bs) == aio_context);
2407     }
2408 
2409     if (base_node) {
2410         base_bs = bdrv_lookup_bs(NULL, base_node, errp);
2411         if (!base_bs) {
2412             goto out_rdlock;
2413         }
2414         if (bs == base_bs || !bdrv_chain_contains(bs, base_bs)) {
2415             error_setg(errp, "Node '%s' is not a backing image of '%s'",
2416                        base_node, device);
2417             goto out_rdlock;
2418         }
2419         assert(bdrv_get_aio_context(base_bs) == aio_context);
2420 
2421         bdrv_refresh_filename(base_bs);
2422     }
2423 
2424     if (bottom) {
2425         bottom_bs = bdrv_lookup_bs(NULL, bottom, errp);
2426         if (!bottom_bs) {
2427             goto out_rdlock;
2428         }
2429         if (!bottom_bs->drv) {
2430             error_setg(errp, "Node '%s' is not open", bottom);
2431             goto out_rdlock;
2432         }
2433         if (bottom_bs->drv->is_filter) {
2434             error_setg(errp, "Node '%s' is a filter, use a non-filter node "
2435                        "as 'bottom'", bottom);
2436             goto out_rdlock;
2437         }
2438         if (!bdrv_chain_contains(bs, bottom_bs)) {
2439             error_setg(errp, "Node '%s' is not in a chain starting from '%s'",
2440                        bottom, device);
2441             goto out_rdlock;
2442         }
2443         assert(bdrv_get_aio_context(bottom_bs) == aio_context);
2444     }
2445 
2446     /*
2447      * Check for op blockers in the whole chain between bs and base (or bottom)
2448      */
2449     iter_end = bottom ? bdrv_filter_or_cow_bs(bottom_bs) : base_bs;
2450     for (iter = bs; iter && iter != iter_end;
2451          iter = bdrv_filter_or_cow_bs(iter))
2452     {
2453         if (bdrv_op_is_blocked(iter, BLOCK_OP_TYPE_STREAM, errp)) {
2454             goto out_rdlock;
2455         }
2456     }
2457     bdrv_graph_rdunlock_main_loop();
2458 
2459     /* if we are streaming the entire chain, the result will have no backing
2460      * file, and specifying one is therefore an error */
2461     if (!base_bs && backing_file) {
2462         error_setg(errp, "backing file specified, but streaming the "
2463                          "entire chain");
2464         return;
2465     }
2466 
2467     if (has_auto_finalize && !auto_finalize) {
2468         job_flags |= JOB_MANUAL_FINALIZE;
2469     }
2470     if (has_auto_dismiss && !auto_dismiss) {
2471         job_flags |= JOB_MANUAL_DISMISS;
2472     }
2473 
2474     stream_start(job_id, bs, base_bs, backing_file,
2475                  backing_mask_protocol,
2476                  bottom_bs, job_flags, has_speed ? speed : 0, on_error,
2477                  filter_node_name, &local_err);
2478     if (local_err) {
2479         error_propagate(errp, local_err);
2480         return;
2481     }
2482 
2483     trace_qmp_block_stream(bs);
2484     return;
2485 
2486 out_rdlock:
2487     bdrv_graph_rdunlock_main_loop();
2488 }
2489 
2490 void qmp_block_commit(const char *job_id, const char *device,
2491                       const char *base_node,
2492                       const char *base,
2493                       const char *top_node,
2494                       const char *top,
2495                       const char *backing_file,
2496                       bool has_backing_mask_protocol,
2497                       bool backing_mask_protocol,
2498                       bool has_speed, int64_t speed,
2499                       bool has_on_error, BlockdevOnError on_error,
2500                       const char *filter_node_name,
2501                       bool has_auto_finalize, bool auto_finalize,
2502                       bool has_auto_dismiss, bool auto_dismiss,
2503                       Error **errp)
2504 {
2505     BlockDriverState *bs;
2506     BlockDriverState *iter;
2507     BlockDriverState *base_bs, *top_bs;
2508     AioContext *aio_context;
2509     Error *local_err = NULL;
2510     int job_flags = JOB_DEFAULT;
2511     uint64_t top_perm, top_shared;
2512 
2513     /* TODO We'll eventually have to take a writer lock in this function */
2514     GRAPH_RDLOCK_GUARD_MAINLOOP();
2515 
2516     if (!has_speed) {
2517         speed = 0;
2518     }
2519     if (!has_on_error) {
2520         on_error = BLOCKDEV_ON_ERROR_REPORT;
2521     }
2522     if (has_auto_finalize && !auto_finalize) {
2523         job_flags |= JOB_MANUAL_FINALIZE;
2524     }
2525     if (has_auto_dismiss && !auto_dismiss) {
2526         job_flags |= JOB_MANUAL_DISMISS;
2527     }
2528     if (!has_backing_mask_protocol) {
2529         backing_mask_protocol = false;
2530     }
2531 
2532     /* Important Note:
2533      *  libvirt relies on the DeviceNotFound error class in order to probe for
2534      *  live commit feature versions; for this to work, we must make sure to
2535      *  perform the device lookup before any generic errors that may occur in a
2536      *  scenario in which all optional arguments are omitted. */
2537     bs = qmp_get_root_bs(device, &local_err);
2538     if (!bs) {
2539         bs = bdrv_lookup_bs(device, device, NULL);
2540         if (!bs) {
2541             error_free(local_err);
2542             error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
2543                       "Device '%s' not found", device);
2544         } else {
2545             error_propagate(errp, local_err);
2546         }
2547         return;
2548     }
2549 
2550     aio_context = bdrv_get_aio_context(bs);
2551 
2552     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_COMMIT_SOURCE, errp)) {
2553         return;
2554     }
2555 
2556     /* default top_bs is the active layer */
2557     top_bs = bs;
2558 
2559     if (top_node && top) {
2560         error_setg(errp, "'top-node' and 'top' are mutually exclusive");
2561         return;
2562     } else if (top_node) {
2563         top_bs = bdrv_lookup_bs(NULL, top_node, errp);
2564         if (top_bs == NULL) {
2565             return;
2566         }
2567         if (!bdrv_chain_contains(bs, top_bs)) {
2568             error_setg(errp, "'%s' is not in this backing file chain",
2569                        top_node);
2570             return;
2571         }
2572     } else if (top) {
2573         /* This strcmp() is just a shortcut, there is no need to
2574          * refresh @bs's filename.  If it mismatches,
2575          * bdrv_find_backing_image() will do the refresh and may still
2576          * return @bs. */
2577         if (strcmp(bs->filename, top) != 0) {
2578             top_bs = bdrv_find_backing_image(bs, top);
2579         }
2580     }
2581 
2582     if (top_bs == NULL) {
2583         error_setg(errp, "Top image file %s not found", top ? top : "NULL");
2584         return;
2585     }
2586 
2587     assert(bdrv_get_aio_context(top_bs) == aio_context);
2588 
2589     if (base_node && base) {
2590         error_setg(errp, "'base-node' and 'base' are mutually exclusive");
2591         return;
2592     } else if (base_node) {
2593         base_bs = bdrv_lookup_bs(NULL, base_node, errp);
2594         if (base_bs == NULL) {
2595             return;
2596         }
2597         if (!bdrv_chain_contains(top_bs, base_bs)) {
2598             error_setg(errp, "'%s' is not in this backing file chain",
2599                        base_node);
2600             return;
2601         }
2602     } else if (base) {
2603         base_bs = bdrv_find_backing_image(top_bs, base);
2604         if (base_bs == NULL) {
2605             error_setg(errp, "Can't find '%s' in the backing chain", base);
2606             return;
2607         }
2608     } else {
2609         base_bs = bdrv_find_base(top_bs);
2610         if (base_bs == NULL) {
2611             error_setg(errp, "There is no backimg image");
2612             return;
2613         }
2614     }
2615 
2616     assert(bdrv_get_aio_context(base_bs) == aio_context);
2617 
2618     for (iter = top_bs; iter != bdrv_filter_or_cow_bs(base_bs);
2619          iter = bdrv_filter_or_cow_bs(iter))
2620     {
2621         if (bdrv_op_is_blocked(iter, BLOCK_OP_TYPE_COMMIT_TARGET, errp)) {
2622             return;
2623         }
2624     }
2625 
2626     /* Do not allow attempts to commit an image into itself */
2627     if (top_bs == base_bs) {
2628         error_setg(errp, "cannot commit an image into itself");
2629         return;
2630     }
2631 
2632     /*
2633      * Active commit is required if and only if someone has taken a
2634      * WRITE permission on the top node.  Historically, we have always
2635      * used active commit for top nodes, so continue that practice
2636      * lest we possibly break clients that rely on this behavior, e.g.
2637      * to later attach this node to a writing parent.
2638      * (Active commit is never really wrong.)
2639      */
2640     bdrv_get_cumulative_perm(top_bs, &top_perm, &top_shared);
2641     if (top_perm & BLK_PERM_WRITE ||
2642         bdrv_skip_filters(top_bs) == bdrv_skip_filters(bs))
2643     {
2644         if (backing_file) {
2645             if (bdrv_skip_filters(top_bs) == bdrv_skip_filters(bs)) {
2646                 error_setg(errp, "'backing-file' specified,"
2647                                  " but 'top' is the active layer");
2648             } else {
2649                 error_setg(errp, "'backing-file' specified, but 'top' has a "
2650                                  "writer on it");
2651             }
2652             return;
2653         }
2654         if (!job_id) {
2655             /*
2656              * Emulate here what block_job_create() does, because it
2657              * is possible that @bs != @top_bs (the block job should
2658              * be named after @bs, even if @top_bs is the actual
2659              * source)
2660              */
2661             job_id = bdrv_get_device_name(bs);
2662         }
2663         commit_active_start(job_id, top_bs, base_bs, job_flags, speed, on_error,
2664                             filter_node_name, NULL, NULL, false, &local_err);
2665     } else {
2666         BlockDriverState *overlay_bs = bdrv_find_overlay(bs, top_bs);
2667         if (bdrv_op_is_blocked(overlay_bs, BLOCK_OP_TYPE_COMMIT_TARGET, errp)) {
2668             return;
2669         }
2670         commit_start(job_id, bs, base_bs, top_bs, job_flags,
2671                      speed, on_error, backing_file,
2672                      backing_mask_protocol,
2673                      filter_node_name, &local_err);
2674     }
2675     if (local_err != NULL) {
2676         error_propagate(errp, local_err);
2677         return;
2678     }
2679 }
2680 
2681 /* Common QMP interface for drive-backup and blockdev-backup */
2682 static BlockJob *do_backup_common(BackupCommon *backup,
2683                                   BlockDriverState *bs,
2684                                   BlockDriverState *target_bs,
2685                                   AioContext *aio_context,
2686                                   JobTxn *txn, Error **errp)
2687 {
2688     BlockJob *job = NULL;
2689     BdrvDirtyBitmap *bmap = NULL;
2690     BackupPerf perf = { .max_workers = 64 };
2691     int job_flags = JOB_DEFAULT;
2692     OnCbwError on_cbw_error = ON_CBW_ERROR_BREAK_GUEST_WRITE;
2693 
2694     if (!backup->has_speed) {
2695         backup->speed = 0;
2696     }
2697     if (!backup->has_on_source_error) {
2698         backup->on_source_error = BLOCKDEV_ON_ERROR_REPORT;
2699     }
2700     if (!backup->has_on_target_error) {
2701         backup->on_target_error = BLOCKDEV_ON_ERROR_REPORT;
2702     }
2703     if (!backup->has_auto_finalize) {
2704         backup->auto_finalize = true;
2705     }
2706     if (!backup->has_auto_dismiss) {
2707         backup->auto_dismiss = true;
2708     }
2709     if (!backup->has_compress) {
2710         backup->compress = false;
2711     }
2712 
2713     if (backup->x_perf) {
2714         if (backup->x_perf->has_use_copy_range) {
2715             perf.use_copy_range = backup->x_perf->use_copy_range;
2716         }
2717         if (backup->x_perf->has_max_workers) {
2718             perf.max_workers = backup->x_perf->max_workers;
2719         }
2720         if (backup->x_perf->has_max_chunk) {
2721             perf.max_chunk = backup->x_perf->max_chunk;
2722         }
2723         if (backup->x_perf->has_min_cluster_size) {
2724             perf.min_cluster_size = backup->x_perf->min_cluster_size;
2725         }
2726     }
2727 
2728     if ((backup->sync == MIRROR_SYNC_MODE_BITMAP) ||
2729         (backup->sync == MIRROR_SYNC_MODE_INCREMENTAL)) {
2730         /* done before desugaring 'incremental' to print the right message */
2731         if (!backup->bitmap) {
2732             error_setg(errp, "must provide a valid bitmap name for "
2733                        "'%s' sync mode", MirrorSyncMode_str(backup->sync));
2734             return NULL;
2735         }
2736     }
2737 
2738     if (backup->sync == MIRROR_SYNC_MODE_INCREMENTAL) {
2739         if (backup->has_bitmap_mode &&
2740             backup->bitmap_mode != BITMAP_SYNC_MODE_ON_SUCCESS) {
2741             error_setg(errp, "Bitmap sync mode must be '%s' "
2742                        "when using sync mode '%s'",
2743                        BitmapSyncMode_str(BITMAP_SYNC_MODE_ON_SUCCESS),
2744                        MirrorSyncMode_str(backup->sync));
2745             return NULL;
2746         }
2747         backup->has_bitmap_mode = true;
2748         backup->sync = MIRROR_SYNC_MODE_BITMAP;
2749         backup->bitmap_mode = BITMAP_SYNC_MODE_ON_SUCCESS;
2750     }
2751 
2752     if (backup->bitmap) {
2753         bmap = bdrv_find_dirty_bitmap(bs, backup->bitmap);
2754         if (!bmap) {
2755             error_setg(errp, "Bitmap '%s' could not be found", backup->bitmap);
2756             return NULL;
2757         }
2758         if (!backup->has_bitmap_mode) {
2759             error_setg(errp, "Bitmap sync mode must be given "
2760                        "when providing a bitmap");
2761             return NULL;
2762         }
2763         if (bdrv_dirty_bitmap_check(bmap, BDRV_BITMAP_ALLOW_RO, errp)) {
2764             return NULL;
2765         }
2766 
2767         /* This does not produce a useful bitmap artifact: */
2768         if (backup->sync == MIRROR_SYNC_MODE_NONE) {
2769             error_setg(errp, "sync mode '%s' does not produce meaningful bitmap"
2770                        " outputs", MirrorSyncMode_str(backup->sync));
2771             return NULL;
2772         }
2773 
2774         /* If the bitmap isn't used for input or output, this is useless: */
2775         if (backup->bitmap_mode == BITMAP_SYNC_MODE_NEVER &&
2776             backup->sync != MIRROR_SYNC_MODE_BITMAP) {
2777             error_setg(errp, "Bitmap sync mode '%s' has no meaningful effect"
2778                        " when combined with sync mode '%s'",
2779                        BitmapSyncMode_str(backup->bitmap_mode),
2780                        MirrorSyncMode_str(backup->sync));
2781             return NULL;
2782         }
2783     }
2784 
2785     if (!backup->bitmap && backup->has_bitmap_mode) {
2786         error_setg(errp, "Cannot specify bitmap sync mode without a bitmap");
2787         return NULL;
2788     }
2789 
2790     if (!backup->auto_finalize) {
2791         job_flags |= JOB_MANUAL_FINALIZE;
2792     }
2793     if (!backup->auto_dismiss) {
2794         job_flags |= JOB_MANUAL_DISMISS;
2795     }
2796 
2797     if (backup->has_on_cbw_error) {
2798         on_cbw_error = backup->on_cbw_error;
2799     }
2800 
2801     job = backup_job_create(backup->job_id, bs, target_bs, backup->speed,
2802                             backup->sync, bmap, backup->bitmap_mode,
2803                             backup->compress, backup->discard_source,
2804                             backup->filter_node_name,
2805                             &perf,
2806                             backup->on_source_error,
2807                             backup->on_target_error,
2808                             on_cbw_error,
2809                             job_flags, NULL, NULL, txn, errp);
2810     return job;
2811 }
2812 
2813 void qmp_drive_backup(DriveBackup *backup, Error **errp)
2814 {
2815     TransactionAction action = {
2816         .type = TRANSACTION_ACTION_KIND_DRIVE_BACKUP,
2817         .u.drive_backup.data = backup,
2818     };
2819     blockdev_do_action(&action, errp);
2820 }
2821 
2822 BlockDeviceInfoList *qmp_query_named_block_nodes(bool has_flat,
2823                                                  bool flat,
2824                                                  Error **errp)
2825 {
2826     bool return_flat = has_flat && flat;
2827 
2828     return bdrv_named_nodes_list(return_flat, errp);
2829 }
2830 
2831 XDbgBlockGraph *qmp_x_debug_query_block_graph(Error **errp)
2832 {
2833     GRAPH_RDLOCK_GUARD_MAINLOOP();
2834 
2835     return bdrv_get_xdbg_block_graph(errp);
2836 }
2837 
2838 void qmp_blockdev_backup(BlockdevBackup *backup, Error **errp)
2839 {
2840     TransactionAction action = {
2841         .type = TRANSACTION_ACTION_KIND_BLOCKDEV_BACKUP,
2842         .u.blockdev_backup.data = backup,
2843     };
2844     blockdev_do_action(&action, errp);
2845 }
2846 
2847 /* Parameter check and block job starting for drive mirroring.
2848  * Caller should hold @device and @target's aio context (must be the same).
2849  **/
2850 static void blockdev_mirror_common(const char *job_id, BlockDriverState *bs,
2851                                    BlockDriverState *target,
2852                                    const char *replaces,
2853                                    enum MirrorSyncMode sync,
2854                                    BlockMirrorBackingMode backing_mode,
2855                                    bool target_is_zero,
2856                                    bool has_speed, int64_t speed,
2857                                    bool has_granularity, uint32_t granularity,
2858                                    bool has_buf_size, int64_t buf_size,
2859                                    bool has_on_source_error,
2860                                    BlockdevOnError on_source_error,
2861                                    bool has_on_target_error,
2862                                    BlockdevOnError on_target_error,
2863                                    bool has_unmap, bool unmap,
2864                                    const char *filter_node_name,
2865                                    bool has_copy_mode, MirrorCopyMode copy_mode,
2866                                    bool has_auto_finalize, bool auto_finalize,
2867                                    bool has_auto_dismiss, bool auto_dismiss,
2868                                    Error **errp)
2869 {
2870     BlockDriverState *unfiltered_bs;
2871     int job_flags = JOB_DEFAULT;
2872 
2873     GLOBAL_STATE_CODE();
2874     GRAPH_RDLOCK_GUARD_MAINLOOP();
2875 
2876     if (!has_speed) {
2877         speed = 0;
2878     }
2879     if (!has_on_source_error) {
2880         on_source_error = BLOCKDEV_ON_ERROR_REPORT;
2881     }
2882     if (!has_on_target_error) {
2883         on_target_error = BLOCKDEV_ON_ERROR_REPORT;
2884     }
2885     if (!has_granularity) {
2886         granularity = 0;
2887     }
2888     if (!has_buf_size) {
2889         buf_size = 0;
2890     }
2891     if (!has_unmap) {
2892         unmap = true;
2893     }
2894     if (!has_copy_mode) {
2895         copy_mode = MIRROR_COPY_MODE_BACKGROUND;
2896     }
2897     if (has_auto_finalize && !auto_finalize) {
2898         job_flags |= JOB_MANUAL_FINALIZE;
2899     }
2900     if (has_auto_dismiss && !auto_dismiss) {
2901         job_flags |= JOB_MANUAL_DISMISS;
2902     }
2903 
2904     if (granularity != 0 && (granularity < 512 || granularity > 1048576 * 64)) {
2905         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "granularity",
2906                    "a value in range [512B, 64MB]");
2907         return;
2908     }
2909     if (granularity & (granularity - 1)) {
2910         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "granularity",
2911                    "a power of 2");
2912         return;
2913     }
2914 
2915     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_MIRROR_SOURCE, errp)) {
2916         return;
2917     }
2918     if (bdrv_op_is_blocked(target, BLOCK_OP_TYPE_MIRROR_TARGET, errp)) {
2919         return;
2920     }
2921 
2922     if (!replaces) {
2923         /* We want to mirror from @bs, but keep implicit filters on top */
2924         unfiltered_bs = bdrv_skip_implicit_filters(bs);
2925         if (unfiltered_bs != bs) {
2926             replaces = unfiltered_bs->node_name;
2927         }
2928     }
2929 
2930     if (replaces) {
2931         BlockDriverState *to_replace_bs;
2932         int64_t bs_size, replace_size;
2933 
2934         bs_size = bdrv_getlength(bs);
2935         if (bs_size < 0) {
2936             error_setg_errno(errp, -bs_size, "Failed to query device's size");
2937             return;
2938         }
2939 
2940         to_replace_bs = check_to_replace_node(bs, replaces, errp);
2941         if (!to_replace_bs) {
2942             return;
2943         }
2944 
2945         replace_size = bdrv_getlength(to_replace_bs);
2946 
2947         if (replace_size < 0) {
2948             error_setg_errno(errp, -replace_size,
2949                              "Failed to query the replacement node's size");
2950             return;
2951         }
2952         if (bs_size != replace_size) {
2953             error_setg(errp, "cannot replace image with a mirror image of "
2954                              "different size");
2955             return;
2956         }
2957     }
2958 
2959     /* pass the node name to replace to mirror start since it's loose coupling
2960      * and will allow to check whether the node still exist at mirror completion
2961      */
2962     mirror_start(job_id, bs, target, replaces, job_flags,
2963                  speed, granularity, buf_size, sync, backing_mode,
2964                  target_is_zero, on_source_error, on_target_error, unmap,
2965                  filter_node_name, copy_mode, errp);
2966 }
2967 
2968 void qmp_drive_mirror(DriveMirror *arg, Error **errp)
2969 {
2970     BlockDriverState *bs;
2971     BlockDriverState *target_backing_bs, *target_bs;
2972     AioContext *aio_context;
2973     BlockMirrorBackingMode backing_mode;
2974     Error *local_err = NULL;
2975     QDict *options = NULL;
2976     int flags;
2977     int64_t size;
2978     const char *format = arg->format;
2979     bool target_is_zero;
2980     int ret;
2981 
2982     bs = qmp_get_root_bs(arg->device, errp);
2983     if (!bs) {
2984         return;
2985     }
2986 
2987     /* Early check to avoid creating target */
2988     bdrv_graph_rdlock_main_loop();
2989     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_MIRROR_SOURCE, errp)) {
2990         bdrv_graph_rdunlock_main_loop();
2991         return;
2992     }
2993 
2994     aio_context = bdrv_get_aio_context(bs);
2995 
2996     if (!arg->has_mode) {
2997         arg->mode = NEW_IMAGE_MODE_ABSOLUTE_PATHS;
2998     }
2999 
3000     if (!arg->format) {
3001         format = (arg->mode == NEW_IMAGE_MODE_EXISTING
3002                   ? NULL : bs->drv->format_name);
3003     }
3004 
3005     flags = bs->open_flags | BDRV_O_RDWR;
3006     target_backing_bs = bdrv_cow_bs(bdrv_skip_filters(bs));
3007     if (!target_backing_bs && arg->sync == MIRROR_SYNC_MODE_TOP) {
3008         arg->sync = MIRROR_SYNC_MODE_FULL;
3009     }
3010     if (arg->sync == MIRROR_SYNC_MODE_NONE) {
3011         target_backing_bs = bs;
3012     }
3013     bdrv_graph_rdunlock_main_loop();
3014 
3015     size = bdrv_getlength(bs);
3016     if (size < 0) {
3017         error_setg_errno(errp, -size, "bdrv_getlength failed");
3018         return;
3019     }
3020 
3021     if (arg->replaces) {
3022         if (!arg->node_name) {
3023             error_setg(errp, "a node-name must be provided when replacing a"
3024                              " named node of the graph");
3025             return;
3026         }
3027     }
3028 
3029     if (arg->mode == NEW_IMAGE_MODE_ABSOLUTE_PATHS) {
3030         backing_mode = MIRROR_SOURCE_BACKING_CHAIN;
3031     } else {
3032         backing_mode = MIRROR_OPEN_BACKING_CHAIN;
3033     }
3034 
3035     /* Don't open backing image in create() */
3036     flags |= BDRV_O_NO_BACKING;
3037 
3038     if ((arg->sync == MIRROR_SYNC_MODE_FULL || !target_backing_bs)
3039         && arg->mode != NEW_IMAGE_MODE_EXISTING)
3040     {
3041         /* create new image w/o backing file */
3042         assert(format);
3043         bdrv_img_create(arg->target, format,
3044                         NULL, NULL, NULL, size, flags, false, &local_err);
3045     } else {
3046         BlockDriverState *explicit_backing;
3047 
3048         switch (arg->mode) {
3049         case NEW_IMAGE_MODE_EXISTING:
3050             break;
3051         case NEW_IMAGE_MODE_ABSOLUTE_PATHS:
3052             /*
3053              * Create new image with backing file.
3054              * Implicit filters should not appear in the filename.
3055              */
3056             bdrv_graph_rdlock_main_loop();
3057             explicit_backing = bdrv_skip_implicit_filters(target_backing_bs);
3058             bdrv_refresh_filename(explicit_backing);
3059             bdrv_graph_rdunlock_main_loop();
3060 
3061             bdrv_img_create(arg->target, format,
3062                             explicit_backing->filename,
3063                             explicit_backing->drv->format_name,
3064                             NULL, size, flags, false, &local_err);
3065             break;
3066         default:
3067             abort();
3068         }
3069     }
3070 
3071     if (local_err) {
3072         error_propagate(errp, local_err);
3073         return;
3074     }
3075 
3076     options = qdict_new();
3077     if (arg->node_name) {
3078         qdict_put_str(options, "node-name", arg->node_name);
3079     }
3080     if (format) {
3081         qdict_put_str(options, "driver", format);
3082     }
3083 
3084     /* Mirroring takes care of copy-on-write using the source's backing
3085      * file.
3086      */
3087     target_bs = bdrv_open(arg->target, NULL, options, flags, errp);
3088     if (!target_bs) {
3089         return;
3090     }
3091 
3092     bdrv_graph_rdlock_main_loop();
3093     target_is_zero = (arg->mode != NEW_IMAGE_MODE_EXISTING &&
3094                       bdrv_has_zero_init(target_bs));
3095     bdrv_graph_rdunlock_main_loop();
3096 
3097 
3098     ret = bdrv_try_change_aio_context(target_bs, aio_context, NULL, errp);
3099     if (ret < 0) {
3100         bdrv_unref(target_bs);
3101         return;
3102     }
3103 
3104     blockdev_mirror_common(arg->job_id, bs, target_bs,
3105                            arg->replaces, arg->sync,
3106                            backing_mode, target_is_zero,
3107                            arg->has_speed, arg->speed,
3108                            arg->has_granularity, arg->granularity,
3109                            arg->has_buf_size, arg->buf_size,
3110                            arg->has_on_source_error, arg->on_source_error,
3111                            arg->has_on_target_error, arg->on_target_error,
3112                            arg->has_unmap, arg->unmap,
3113                            NULL,
3114                            arg->has_copy_mode, arg->copy_mode,
3115                            arg->has_auto_finalize, arg->auto_finalize,
3116                            arg->has_auto_dismiss, arg->auto_dismiss,
3117                            errp);
3118     bdrv_unref(target_bs);
3119 }
3120 
3121 void qmp_blockdev_mirror(const char *job_id,
3122                          const char *device, const char *target,
3123                          const char *replaces,
3124                          MirrorSyncMode sync,
3125                          bool has_speed, int64_t speed,
3126                          bool has_granularity, uint32_t granularity,
3127                          bool has_buf_size, int64_t buf_size,
3128                          bool has_on_source_error,
3129                          BlockdevOnError on_source_error,
3130                          bool has_on_target_error,
3131                          BlockdevOnError on_target_error,
3132                          const char *filter_node_name,
3133                          bool has_copy_mode, MirrorCopyMode copy_mode,
3134                          bool has_auto_finalize, bool auto_finalize,
3135                          bool has_auto_dismiss, bool auto_dismiss,
3136                          bool has_target_is_zero, bool target_is_zero,
3137                          Error **errp)
3138 {
3139     BlockDriverState *bs;
3140     BlockDriverState *target_bs;
3141     AioContext *aio_context;
3142     BlockMirrorBackingMode backing_mode = MIRROR_LEAVE_BACKING_CHAIN;
3143     int ret;
3144 
3145     bs = qmp_get_root_bs(device, errp);
3146     if (!bs) {
3147         return;
3148     }
3149 
3150     target_bs = bdrv_lookup_bs(target, target, errp);
3151     if (!target_bs) {
3152         return;
3153     }
3154 
3155     aio_context = bdrv_get_aio_context(bs);
3156 
3157     ret = bdrv_try_change_aio_context(target_bs, aio_context, NULL, errp);
3158     if (ret < 0) {
3159         return;
3160     }
3161 
3162     blockdev_mirror_common(job_id, bs, target_bs,
3163                            replaces, sync, backing_mode,
3164                            has_target_is_zero && target_is_zero,
3165                            has_speed, speed,
3166                            has_granularity, granularity,
3167                            has_buf_size, buf_size,
3168                            has_on_source_error, on_source_error,
3169                            has_on_target_error, on_target_error,
3170                            true, true, filter_node_name,
3171                            has_copy_mode, copy_mode,
3172                            has_auto_finalize, auto_finalize,
3173                            has_auto_dismiss, auto_dismiss,
3174                            errp);
3175 }
3176 
3177 /*
3178  * Get a block job using its ID. Called with job_mutex held.
3179  */
3180 static BlockJob *find_block_job_locked(const char *id, Error **errp)
3181 {
3182     BlockJob *job;
3183 
3184     assert(id != NULL);
3185 
3186     job = block_job_get_locked(id);
3187 
3188     if (!job) {
3189         error_set(errp, ERROR_CLASS_DEVICE_NOT_ACTIVE,
3190                   "Block job '%s' not found", id);
3191         return NULL;
3192     }
3193 
3194     return job;
3195 }
3196 
3197 void qmp_block_job_set_speed(const char *device, int64_t speed, Error **errp)
3198 {
3199     BlockJob *job;
3200 
3201     JOB_LOCK_GUARD();
3202     job = find_block_job_locked(device, errp);
3203 
3204     if (!job) {
3205         return;
3206     }
3207 
3208     block_job_set_speed_locked(job, speed, errp);
3209 }
3210 
3211 void qmp_block_job_cancel(const char *device,
3212                           bool has_force, bool force, Error **errp)
3213 {
3214     BlockJob *job;
3215 
3216     JOB_LOCK_GUARD();
3217     job = find_block_job_locked(device, errp);
3218 
3219     if (!job) {
3220         return;
3221     }
3222 
3223     if (!has_force) {
3224         force = false;
3225     }
3226 
3227     if (job_user_paused_locked(&job->job) && !force) {
3228         error_setg(errp, "The block job for device '%s' is currently paused",
3229                    device);
3230         return;
3231     }
3232 
3233     trace_qmp_block_job_cancel(job);
3234     job_user_cancel_locked(&job->job, force, errp);
3235 }
3236 
3237 void qmp_block_job_pause(const char *device, Error **errp)
3238 {
3239     BlockJob *job;
3240 
3241     JOB_LOCK_GUARD();
3242     job = find_block_job_locked(device, errp);
3243 
3244     if (!job) {
3245         return;
3246     }
3247 
3248     trace_qmp_block_job_pause(job);
3249     job_user_pause_locked(&job->job, errp);
3250 }
3251 
3252 void qmp_block_job_resume(const char *device, Error **errp)
3253 {
3254     BlockJob *job;
3255 
3256     JOB_LOCK_GUARD();
3257     job = find_block_job_locked(device, errp);
3258 
3259     if (!job) {
3260         return;
3261     }
3262 
3263     trace_qmp_block_job_resume(job);
3264     job_user_resume_locked(&job->job, errp);
3265 }
3266 
3267 void qmp_block_job_complete(const char *device, Error **errp)
3268 {
3269     BlockJob *job;
3270 
3271     JOB_LOCK_GUARD();
3272     job = find_block_job_locked(device, errp);
3273 
3274     if (!job) {
3275         return;
3276     }
3277 
3278     trace_qmp_block_job_complete(job);
3279     job_complete_locked(&job->job, errp);
3280 }
3281 
3282 void qmp_block_job_finalize(const char *id, Error **errp)
3283 {
3284     BlockJob *job;
3285 
3286     JOB_LOCK_GUARD();
3287     job = find_block_job_locked(id, errp);
3288 
3289     if (!job) {
3290         return;
3291     }
3292 
3293     trace_qmp_block_job_finalize(job);
3294     job_ref_locked(&job->job);
3295     job_finalize_locked(&job->job, errp);
3296 
3297     job_unref_locked(&job->job);
3298 }
3299 
3300 void qmp_block_job_dismiss(const char *id, Error **errp)
3301 {
3302     BlockJob *bjob;
3303     Job *job;
3304 
3305     JOB_LOCK_GUARD();
3306     bjob = find_block_job_locked(id, errp);
3307 
3308     if (!bjob) {
3309         return;
3310     }
3311 
3312     trace_qmp_block_job_dismiss(bjob);
3313     job = &bjob->job;
3314     job_dismiss_locked(&job, errp);
3315 }
3316 
3317 void qmp_block_job_change(BlockJobChangeOptions *opts, Error **errp)
3318 {
3319     BlockJob *job;
3320 
3321     JOB_LOCK_GUARD();
3322     job = find_block_job_locked(opts->id, errp);
3323 
3324     if (!job) {
3325         return;
3326     }
3327 
3328     block_job_change_locked(job, opts, errp);
3329 }
3330 
3331 void qmp_change_backing_file(const char *device,
3332                              const char *image_node_name,
3333                              const char *backing_file,
3334                              Error **errp)
3335 {
3336     BlockDriverState *bs = NULL;
3337     BlockDriverState *image_bs = NULL;
3338     Error *local_err = NULL;
3339     bool ro;
3340     int ret;
3341 
3342     bs = qmp_get_root_bs(device, errp);
3343     if (!bs) {
3344         return;
3345     }
3346 
3347     bdrv_graph_rdlock_main_loop();
3348 
3349     image_bs = bdrv_lookup_bs(NULL, image_node_name, &local_err);
3350     if (local_err) {
3351         error_propagate(errp, local_err);
3352         goto out_rdlock;
3353     }
3354 
3355     if (!image_bs) {
3356         error_setg(errp, "image file not found");
3357         goto out_rdlock;
3358     }
3359 
3360     if (bdrv_find_base(image_bs) == image_bs) {
3361         error_setg(errp, "not allowing backing file change on an image "
3362                          "without a backing file");
3363         goto out_rdlock;
3364     }
3365 
3366     /* even though we are not necessarily operating on bs, we need it to
3367      * determine if block ops are currently prohibited on the chain */
3368     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_CHANGE, errp)) {
3369         goto out_rdlock;
3370     }
3371 
3372     /* final sanity check */
3373     if (!bdrv_chain_contains(bs, image_bs)) {
3374         error_setg(errp, "'%s' and image file are not in the same chain",
3375                    device);
3376         goto out_rdlock;
3377     }
3378     bdrv_graph_rdunlock_main_loop();
3379 
3380     /* if not r/w, reopen to make r/w */
3381     ro = bdrv_is_read_only(image_bs);
3382 
3383     if (ro) {
3384         if (bdrv_reopen_set_read_only(image_bs, false, errp) != 0) {
3385             return;
3386         }
3387     }
3388 
3389     ret = bdrv_change_backing_file(image_bs, backing_file,
3390                                    image_bs->drv ? image_bs->drv->format_name : "",
3391                                    false);
3392 
3393     if (ret < 0) {
3394         error_setg_errno(errp, -ret, "Could not change backing file to '%s'",
3395                          backing_file);
3396         /* don't exit here, so we can try to restore open flags if
3397          * appropriate */
3398     }
3399 
3400     if (ro) {
3401         bdrv_reopen_set_read_only(image_bs, true, errp);
3402     }
3403     return;
3404 
3405 out_rdlock:
3406     bdrv_graph_rdunlock_main_loop();
3407 }
3408 
3409 void qmp_blockdev_add(BlockdevOptions *options, Error **errp)
3410 {
3411     BlockDriverState *bs;
3412     QObject *obj;
3413     Visitor *v = qobject_output_visitor_new(&obj);
3414     QDict *qdict;
3415 
3416     visit_type_BlockdevOptions(v, NULL, &options, &error_abort);
3417     visit_complete(v, &obj);
3418     qdict = qobject_to(QDict, obj);
3419 
3420     qdict_flatten(qdict);
3421 
3422     if (!qdict_get_try_str(qdict, "node-name")) {
3423         error_setg(errp, "'node-name' must be specified for the root node");
3424         goto fail;
3425     }
3426 
3427     bs = bds_tree_init(qdict, errp);
3428     if (!bs) {
3429         goto fail;
3430     }
3431 
3432     bdrv_set_monitor_owned(bs);
3433 
3434 fail:
3435     visit_free(v);
3436 }
3437 
3438 void qmp_blockdev_reopen(BlockdevOptionsList *reopen_list, Error **errp)
3439 {
3440     BlockReopenQueue *queue = NULL;
3441 
3442     /* Add each one of the BDS that we want to reopen to the queue */
3443     for (; reopen_list != NULL; reopen_list = reopen_list->next) {
3444         BlockdevOptions *options = reopen_list->value;
3445         BlockDriverState *bs;
3446         QObject *obj;
3447         Visitor *v;
3448         QDict *qdict;
3449 
3450         /* Check for the selected node name */
3451         if (!options->node_name) {
3452             error_setg(errp, "node-name not specified");
3453             goto fail;
3454         }
3455 
3456         bs = bdrv_find_node(options->node_name);
3457         if (!bs) {
3458             error_setg(errp, "Failed to find node with node-name='%s'",
3459                        options->node_name);
3460             goto fail;
3461         }
3462 
3463         /* Put all options in a QDict and flatten it */
3464         v = qobject_output_visitor_new(&obj);
3465         visit_type_BlockdevOptions(v, NULL, &options, &error_abort);
3466         visit_complete(v, &obj);
3467         visit_free(v);
3468 
3469         qdict = qobject_to(QDict, obj);
3470 
3471         qdict_flatten(qdict);
3472 
3473         queue = bdrv_reopen_queue(queue, bs, qdict, false);
3474     }
3475 
3476     /* Perform the reopen operation */
3477     bdrv_reopen_multiple(queue, errp);
3478     queue = NULL;
3479 
3480 fail:
3481     bdrv_reopen_queue_free(queue);
3482 }
3483 
3484 void qmp_blockdev_del(const char *node_name, Error **errp)
3485 {
3486     BlockDriverState *bs;
3487 
3488     GLOBAL_STATE_CODE();
3489     GRAPH_RDLOCK_GUARD_MAINLOOP();
3490 
3491     bs = bdrv_find_node(node_name);
3492     if (!bs) {
3493         error_setg(errp, "Failed to find node with node-name='%s'", node_name);
3494         return;
3495     }
3496     if (bdrv_has_blk(bs)) {
3497         error_setg(errp, "Node %s is in use", node_name);
3498         return;
3499     }
3500 
3501     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_DRIVE_DEL, errp)) {
3502         return;
3503     }
3504 
3505     if (!QTAILQ_IN_USE(bs, monitor_list)) {
3506         error_setg(errp, "Node %s is not owned by the monitor",
3507                    bs->node_name);
3508         return;
3509     }
3510 
3511     if (bs->refcnt > 1) {
3512         error_setg(errp, "Block device %s is in use",
3513                    bdrv_get_device_or_node_name(bs));
3514         return;
3515     }
3516 
3517     QTAILQ_REMOVE(&monitor_bdrv_states, bs, monitor_list);
3518     bdrv_unref(bs);
3519 }
3520 
3521 void qmp_blockdev_set_active(const char *node_name, bool active, Error **errp)
3522 {
3523     int ret;
3524 
3525     GLOBAL_STATE_CODE();
3526     GRAPH_RDLOCK_GUARD_MAINLOOP();
3527 
3528     if (!node_name) {
3529         if (active) {
3530             bdrv_activate_all(errp);
3531         } else {
3532             ret = bdrv_inactivate_all();
3533             if (ret < 0) {
3534                 error_setg_errno(errp, -ret, "Failed to inactivate all nodes");
3535             }
3536         }
3537     } else {
3538         BlockDriverState *bs = bdrv_find_node(node_name);
3539         if (!bs) {
3540             error_setg(errp, "Failed to find node with node-name='%s'",
3541                        node_name);
3542             return;
3543         }
3544 
3545         if (active) {
3546             bdrv_activate(bs, errp);
3547         } else {
3548             bdrv_inactivate(bs, errp);
3549         }
3550     }
3551 }
3552 
3553 static BdrvChild * GRAPH_RDLOCK
3554 bdrv_find_child(BlockDriverState *parent_bs, const char *child_name)
3555 {
3556     BdrvChild *child;
3557 
3558     QLIST_FOREACH(child, &parent_bs->children, next) {
3559         if (strcmp(child->name, child_name) == 0) {
3560             return child;
3561         }
3562     }
3563 
3564     return NULL;
3565 }
3566 
3567 void qmp_x_blockdev_change(const char *parent, const char *child,
3568                            const char *node, Error **errp)
3569 {
3570     BlockDriverState *parent_bs, *new_bs = NULL;
3571     BdrvChild *p_child;
3572 
3573     bdrv_graph_wrlock_drained();
3574 
3575     parent_bs = bdrv_lookup_bs(parent, parent, errp);
3576     if (!parent_bs) {
3577         goto out;
3578     }
3579 
3580     if (!child == !node) {
3581         if (child) {
3582             error_setg(errp, "The parameters child and node are in conflict");
3583         } else {
3584             error_setg(errp, "Either child or node must be specified");
3585         }
3586         goto out;
3587     }
3588 
3589     if (child) {
3590         p_child = bdrv_find_child(parent_bs, child);
3591         if (!p_child) {
3592             error_setg(errp, "Node '%s' does not have child '%s'",
3593                        parent, child);
3594             goto out;
3595         }
3596         bdrv_del_child(parent_bs, p_child, errp);
3597     }
3598 
3599     if (node) {
3600         new_bs = bdrv_find_node(node);
3601         if (!new_bs) {
3602             error_setg(errp, "Node '%s' not found", node);
3603             goto out;
3604         }
3605         bdrv_add_child(parent_bs, new_bs, errp);
3606     }
3607 
3608 out:
3609     bdrv_graph_wrunlock();
3610 }
3611 
3612 BlockJobInfoList *qmp_query_block_jobs(Error **errp)
3613 {
3614     BlockJobInfoList *head = NULL, **tail = &head;
3615     BlockJob *job;
3616 
3617     JOB_LOCK_GUARD();
3618 
3619     for (job = block_job_next_locked(NULL); job;
3620          job = block_job_next_locked(job)) {
3621         BlockJobInfo *value;
3622 
3623         if (block_job_is_internal(job)) {
3624             continue;
3625         }
3626         value = block_job_query_locked(job, errp);
3627         if (!value) {
3628             qapi_free_BlockJobInfoList(head);
3629             return NULL;
3630         }
3631         QAPI_LIST_APPEND(tail, value);
3632     }
3633 
3634     return head;
3635 }
3636 
3637 void qmp_x_blockdev_set_iothread(const char *node_name, StrOrNull *iothread,
3638                                  bool has_force, bool force, Error **errp)
3639 {
3640     AioContext *new_context;
3641     BlockDriverState *bs;
3642 
3643     bdrv_drain_all_begin();
3644     bdrv_graph_rdlock_main_loop();
3645 
3646     bs = bdrv_find_node(node_name);
3647     if (!bs) {
3648         error_setg(errp, "Failed to find node with node-name='%s'", node_name);
3649         goto out;
3650     }
3651 
3652     /* Protects against accidents. */
3653     if (!(has_force && force) && bdrv_has_blk(bs)) {
3654         error_setg(errp, "Node %s is associated with a BlockBackend and could "
3655                          "be in use (use force=true to override this check)",
3656                          node_name);
3657         goto out;
3658     }
3659 
3660     if (iothread->type == QTYPE_QSTRING) {
3661         IOThread *obj = iothread_by_id(iothread->u.s);
3662         if (!obj) {
3663             error_setg(errp, "Cannot find iothread %s", iothread->u.s);
3664             goto out;
3665         }
3666 
3667         new_context = iothread_get_aio_context(obj);
3668     } else {
3669         new_context = qemu_get_aio_context();
3670     }
3671 
3672     bdrv_try_change_aio_context_locked(bs, new_context, NULL, errp);
3673 
3674 out:
3675     bdrv_graph_rdunlock_main_loop();
3676     bdrv_drain_all_end();
3677 }
3678 
3679 QemuOptsList qemu_common_drive_opts = {
3680     .name = "drive",
3681     .head = QTAILQ_HEAD_INITIALIZER(qemu_common_drive_opts.head),
3682     .desc = {
3683         {
3684             .name = "snapshot",
3685             .type = QEMU_OPT_BOOL,
3686             .help = "enable/disable snapshot mode",
3687         },{
3688             .name = "aio",
3689             .type = QEMU_OPT_STRING,
3690             .help = "host AIO implementation (threads, native, io_uring)",
3691         },{
3692             .name = BDRV_OPT_CACHE_WB,
3693             .type = QEMU_OPT_BOOL,
3694             .help = "Enable writeback mode",
3695         },{
3696             .name = "format",
3697             .type = QEMU_OPT_STRING,
3698             .help = "disk format (raw, qcow2, ...)",
3699         },{
3700             .name = "rerror",
3701             .type = QEMU_OPT_STRING,
3702             .help = "read error action",
3703         },{
3704             .name = "werror",
3705             .type = QEMU_OPT_STRING,
3706             .help = "write error action",
3707         },{
3708             .name = BDRV_OPT_READ_ONLY,
3709             .type = QEMU_OPT_BOOL,
3710             .help = "open drive file as read-only",
3711         },
3712 
3713         THROTTLE_OPTS,
3714 
3715         {
3716             .name = "throttling.group",
3717             .type = QEMU_OPT_STRING,
3718             .help = "name of the block throttling group",
3719         },{
3720             .name = "copy-on-read",
3721             .type = QEMU_OPT_BOOL,
3722             .help = "copy read data from backing file into image file",
3723         },{
3724             .name = "detect-zeroes",
3725             .type = QEMU_OPT_STRING,
3726             .help = "try to optimize zero writes (off, on, unmap)",
3727         },{
3728             .name = "stats-account-invalid",
3729             .type = QEMU_OPT_BOOL,
3730             .help = "whether to account for invalid I/O operations "
3731                     "in the statistics",
3732         },{
3733             .name = "stats-account-failed",
3734             .type = QEMU_OPT_BOOL,
3735             .help = "whether to account for failed I/O operations "
3736                     "in the statistics",
3737         },
3738         { /* end of list */ }
3739     },
3740 };
3741 
3742 QemuOptsList qemu_drive_opts = {
3743     .name = "drive",
3744     .head = QTAILQ_HEAD_INITIALIZER(qemu_drive_opts.head),
3745     .desc = {
3746         /*
3747          * no elements => accept any params
3748          * validation will happen later
3749          */
3750         { /* end of list */ }
3751     },
3752 };
3753